xref: /freebsd/sys/netinet6/nd6.c (revision 63d1fd5970ec814904aa0f4580b10a0d302d08b2)
1 /*-
2  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. Neither the name of the project nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/callout.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/mutex.h>
45 #include <sys/socket.h>
46 #include <sys/sockio.h>
47 #include <sys/time.h>
48 #include <sys/kernel.h>
49 #include <sys/protosw.h>
50 #include <sys/errno.h>
51 #include <sys/syslog.h>
52 #include <sys/rwlock.h>
53 #include <sys/queue.h>
54 #include <sys/sdt.h>
55 #include <sys/sysctl.h>
56 
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_arc.h>
60 #include <net/if_dl.h>
61 #include <net/if_types.h>
62 #include <net/iso88025.h>
63 #include <net/fddi.h>
64 #include <net/route.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/in_kdtrace.h>
69 #include <net/if_llatbl.h>
70 #include <netinet/if_ether.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet/ip6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/scope6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet6/in6_ifattach.h>
77 #include <netinet/icmp6.h>
78 #include <netinet6/send.h>
79 
80 #include <sys/limits.h>
81 
82 #include <security/mac/mac_framework.h>
83 
84 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
85 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
86 
87 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
88 
89 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
90 
91 /* timer values */
92 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
93 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
94 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
95 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
96 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
97 					 * local traffic */
98 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
99 					 * collection timer */
100 
101 /* preventing too many loops in ND option parsing */
102 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
103 
104 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
105 					 * layer hints */
106 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
107 					 * ND entries */
108 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
109 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
110 
111 #ifdef ND6_DEBUG
112 VNET_DEFINE(int, nd6_debug) = 1;
113 #else
114 VNET_DEFINE(int, nd6_debug) = 0;
115 #endif
116 
117 static eventhandler_tag lle_event_eh, iflladdr_event_eh;
118 
119 VNET_DEFINE(struct nd_drhead, nd_defrouter);
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 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
147 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
148 
149 VNET_DEFINE(struct callout, nd6_timer_ch);
150 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
151 
152 static void
153 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
154 {
155 	struct rt_addrinfo rtinfo;
156 	struct sockaddr_in6 dst;
157 	struct sockaddr_dl gw;
158 	struct ifnet *ifp;
159 	int type;
160 
161 	LLE_WLOCK_ASSERT(lle);
162 
163 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
164 		return;
165 
166 	switch (evt) {
167 	case LLENTRY_RESOLVED:
168 		type = RTM_ADD;
169 		KASSERT(lle->la_flags & LLE_VALID,
170 		    ("%s: %p resolved but not valid?", __func__, lle));
171 		break;
172 	case LLENTRY_EXPIRED:
173 		type = RTM_DELETE;
174 		break;
175 	default:
176 		return;
177 	}
178 
179 	ifp = lltable_get_ifp(lle->lle_tbl);
180 
181 	bzero(&dst, sizeof(dst));
182 	bzero(&gw, sizeof(gw));
183 	bzero(&rtinfo, sizeof(rtinfo));
184 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
185 	dst.sin6_scope_id = in6_getscopezone(ifp,
186 	    in6_addrscope(&dst.sin6_addr));
187 	gw.sdl_len = sizeof(struct sockaddr_dl);
188 	gw.sdl_family = AF_LINK;
189 	gw.sdl_alen = ifp->if_addrlen;
190 	gw.sdl_index = ifp->if_index;
191 	gw.sdl_type = ifp->if_type;
192 	if (evt == LLENTRY_RESOLVED)
193 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
194 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
195 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
196 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
197 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
198 	    type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
199 }
200 
201 /*
202  * A handler for interface link layer address change event.
203  */
204 static void
205 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
206 {
207 
208 	lltable_update_ifaddr(LLTABLE6(ifp));
209 }
210 
211 void
212 nd6_init(void)
213 {
214 
215 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
216 	rw_init(&V_nd6_lock, "nd6 list");
217 
218 	LIST_INIT(&V_nd_prefix);
219 	TAILQ_INIT(&V_nd_defrouter);
220 
221 	/* Start timers. */
222 	callout_init(&V_nd6_slowtimo_ch, 0);
223 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
224 	    nd6_slowtimo, curvnet);
225 
226 	callout_init(&V_nd6_timer_ch, 0);
227 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
228 
229 	nd6_dad_init();
230 	if (IS_DEFAULT_VNET(curvnet)) {
231 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
232 		    NULL, EVENTHANDLER_PRI_ANY);
233 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
234 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
235 	}
236 }
237 
238 #ifdef VIMAGE
239 void
240 nd6_destroy()
241 {
242 
243 	callout_drain(&V_nd6_slowtimo_ch);
244 	callout_drain(&V_nd6_timer_ch);
245 	if (IS_DEFAULT_VNET(curvnet)) {
246 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
247 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
248 	}
249 	rw_destroy(&V_nd6_lock);
250 	mtx_destroy(&V_nd6_onlink_mtx);
251 }
252 #endif
253 
254 struct nd_ifinfo *
255 nd6_ifattach(struct ifnet *ifp)
256 {
257 	struct nd_ifinfo *nd;
258 
259 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
260 	nd->initialized = 1;
261 
262 	nd->chlim = IPV6_DEFHLIM;
263 	nd->basereachable = REACHABLE_TIME;
264 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
265 	nd->retrans = RETRANS_TIMER;
266 
267 	nd->flags = ND6_IFF_PERFORMNUD;
268 
269 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
270 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
271 	 * default regardless of the V_ip6_auto_linklocal configuration to
272 	 * give a reasonable default behavior.
273 	 */
274 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
275 	    (ifp->if_flags & IFF_LOOPBACK))
276 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
277 	/*
278 	 * A loopback interface does not need to accept RTADV.
279 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
280 	 * default regardless of the V_ip6_accept_rtadv configuration to
281 	 * prevent the interface from accepting RA messages arrived
282 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
283 	 */
284 	if (V_ip6_accept_rtadv &&
285 	    !(ifp->if_flags & IFF_LOOPBACK) &&
286 	    (ifp->if_type != IFT_BRIDGE))
287 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
288 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
289 		nd->flags |= ND6_IFF_NO_RADR;
290 
291 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
292 	nd6_setmtu0(ifp, nd);
293 
294 	return nd;
295 }
296 
297 void
298 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
299 {
300 	struct ifaddr *ifa, *next;
301 
302 	IF_ADDR_RLOCK(ifp);
303 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
304 		if (ifa->ifa_addr->sa_family != AF_INET6)
305 			continue;
306 
307 		/* stop DAD processing */
308 		nd6_dad_stop(ifa);
309 	}
310 	IF_ADDR_RUNLOCK(ifp);
311 
312 	free(nd, M_IP6NDP);
313 }
314 
315 /*
316  * Reset ND level link MTU. This function is called when the physical MTU
317  * changes, which means we might have to adjust the ND level MTU.
318  */
319 void
320 nd6_setmtu(struct ifnet *ifp)
321 {
322 	if (ifp->if_afdata[AF_INET6] == NULL)
323 		return;
324 
325 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
326 }
327 
328 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
329 void
330 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
331 {
332 	u_int32_t omaxmtu;
333 
334 	omaxmtu = ndi->maxmtu;
335 
336 	switch (ifp->if_type) {
337 	case IFT_ARCNET:
338 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
339 		break;
340 	case IFT_FDDI:
341 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
342 		break;
343 	case IFT_ISO88025:
344 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
345 		 break;
346 	default:
347 		ndi->maxmtu = ifp->if_mtu;
348 		break;
349 	}
350 
351 	/*
352 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
353 	 * undesirable situation.  We thus notify the operator of the change
354 	 * explicitly.  The check for omaxmtu is necessary to restrict the
355 	 * log to the case of changing the MTU, not initializing it.
356 	 */
357 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
358 		log(LOG_NOTICE, "nd6_setmtu0: "
359 		    "new link MTU on %s (%lu) is too small for IPv6\n",
360 		    if_name(ifp), (unsigned long)ndi->maxmtu);
361 	}
362 
363 	if (ndi->maxmtu > V_in6_maxmtu)
364 		in6_setmaxmtu(); /* check all interfaces just in case */
365 
366 }
367 
368 void
369 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
370 {
371 
372 	bzero(ndopts, sizeof(*ndopts));
373 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
374 	ndopts->nd_opts_last
375 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
376 
377 	if (icmp6len == 0) {
378 		ndopts->nd_opts_done = 1;
379 		ndopts->nd_opts_search = NULL;
380 	}
381 }
382 
383 /*
384  * Take one ND option.
385  */
386 struct nd_opt_hdr *
387 nd6_option(union nd_opts *ndopts)
388 {
389 	struct nd_opt_hdr *nd_opt;
390 	int olen;
391 
392 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
393 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
394 	    __func__));
395 	if (ndopts->nd_opts_search == NULL)
396 		return NULL;
397 	if (ndopts->nd_opts_done)
398 		return NULL;
399 
400 	nd_opt = ndopts->nd_opts_search;
401 
402 	/* make sure nd_opt_len is inside the buffer */
403 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
404 		bzero(ndopts, sizeof(*ndopts));
405 		return NULL;
406 	}
407 
408 	olen = nd_opt->nd_opt_len << 3;
409 	if (olen == 0) {
410 		/*
411 		 * Message validation requires that all included
412 		 * options have a length that is greater than zero.
413 		 */
414 		bzero(ndopts, sizeof(*ndopts));
415 		return NULL;
416 	}
417 
418 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
419 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
420 		/* option overruns the end of buffer, invalid */
421 		bzero(ndopts, sizeof(*ndopts));
422 		return NULL;
423 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
424 		/* reached the end of options chain */
425 		ndopts->nd_opts_done = 1;
426 		ndopts->nd_opts_search = NULL;
427 	}
428 	return nd_opt;
429 }
430 
431 /*
432  * Parse multiple ND options.
433  * This function is much easier to use, for ND routines that do not need
434  * multiple options of the same type.
435  */
436 int
437 nd6_options(union nd_opts *ndopts)
438 {
439 	struct nd_opt_hdr *nd_opt;
440 	int i = 0;
441 
442 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
443 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
444 	    __func__));
445 	if (ndopts->nd_opts_search == NULL)
446 		return 0;
447 
448 	while (1) {
449 		nd_opt = nd6_option(ndopts);
450 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
451 			/*
452 			 * Message validation requires that all included
453 			 * options have a length that is greater than zero.
454 			 */
455 			ICMP6STAT_INC(icp6s_nd_badopt);
456 			bzero(ndopts, sizeof(*ndopts));
457 			return -1;
458 		}
459 
460 		if (nd_opt == NULL)
461 			goto skip1;
462 
463 		switch (nd_opt->nd_opt_type) {
464 		case ND_OPT_SOURCE_LINKADDR:
465 		case ND_OPT_TARGET_LINKADDR:
466 		case ND_OPT_MTU:
467 		case ND_OPT_REDIRECTED_HEADER:
468 		case ND_OPT_NONCE:
469 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
470 				nd6log((LOG_INFO,
471 				    "duplicated ND6 option found (type=%d)\n",
472 				    nd_opt->nd_opt_type));
473 				/* XXX bark? */
474 			} else {
475 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
476 					= nd_opt;
477 			}
478 			break;
479 		case ND_OPT_PREFIX_INFORMATION:
480 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
481 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
482 					= nd_opt;
483 			}
484 			ndopts->nd_opts_pi_end =
485 				(struct nd_opt_prefix_info *)nd_opt;
486 			break;
487 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
488 		case ND_OPT_RDNSS:	/* RFC 6106 */
489 		case ND_OPT_DNSSL:	/* RFC 6106 */
490 			/*
491 			 * Silently ignore options we know and do not care about
492 			 * in the kernel.
493 			 */
494 			break;
495 		default:
496 			/*
497 			 * Unknown options must be silently ignored,
498 			 * to accommodate future extension to the protocol.
499 			 */
500 			nd6log((LOG_DEBUG,
501 			    "nd6_options: unsupported option %d - "
502 			    "option ignored\n", nd_opt->nd_opt_type));
503 		}
504 
505 skip1:
506 		i++;
507 		if (i > V_nd6_maxndopt) {
508 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
509 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
510 			break;
511 		}
512 
513 		if (ndopts->nd_opts_done)
514 			break;
515 	}
516 
517 	return 0;
518 }
519 
520 /*
521  * ND6 timer routine to handle ND6 entries
522  */
523 static void
524 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
525 {
526 	int canceled;
527 
528 	LLE_WLOCK_ASSERT(ln);
529 
530 	if (tick < 0) {
531 		ln->la_expire = 0;
532 		ln->ln_ntick = 0;
533 		canceled = callout_stop(&ln->lle_timer);
534 	} else {
535 		ln->la_expire = time_uptime + tick / hz;
536 		LLE_ADDREF(ln);
537 		if (tick > INT_MAX) {
538 			ln->ln_ntick = tick - INT_MAX;
539 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
540 			    nd6_llinfo_timer, ln);
541 		} else {
542 			ln->ln_ntick = 0;
543 			canceled = callout_reset(&ln->lle_timer, tick,
544 			    nd6_llinfo_timer, ln);
545 		}
546 	}
547 	if (canceled > 0)
548 		LLE_REMREF(ln);
549 }
550 
551 /*
552  * Gets source address of the first packet in hold queue
553  * and stores it in @src.
554  * Returns pointer to @src (if hold queue is not empty) or NULL.
555  *
556  * Set noinline to be dtrace-friendly
557  */
558 static __noinline struct in6_addr *
559 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
560 {
561 	struct ip6_hdr hdr;
562 	struct mbuf *m;
563 
564 	if (ln->la_hold == NULL)
565 		return (NULL);
566 
567 	/*
568 	 * assume every packet in la_hold has the same IP header
569 	 */
570 	m = ln->la_hold;
571 	if (sizeof(hdr) > m->m_len)
572 		return (NULL);
573 
574 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
575 	*src = hdr.ip6_src;
576 
577 	return (src);
578 }
579 
580 /*
581  * Checks if we need to switch from STALE state.
582  *
583  * RFC 4861 requires switching from STALE to DELAY state
584  * on first packet matching entry, waiting V_nd6_delay and
585  * transition to PROBE state (if upper layer confirmation was
586  * not received).
587  *
588  * This code performs a bit differently:
589  * On packet hit we don't change state (but desired state
590  * can be guessed by control plane). However, after V_nd6_delay
591  * seconds code will transition to PROBE state (so DELAY state
592  * is kinda skipped in most situations).
593  *
594  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
595  * we perform the following upon entering STALE state:
596  *
597  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
598  * if packet was transmitted at the start of given interval, we
599  * would be able to switch to PROBE state in V_nd6_delay seconds
600  * as user expects.
601  *
602  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
603  * lle in STALE state (remaining timer value stored in lle_remtime).
604  *
605  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
606  * seconds ago.
607  *
608  * Returns non-zero value if the entry is still STALE (storing
609  * the next timer interval in @pdelay).
610  *
611  * Returns zero value if original timer expired or we need to switch to
612  * PROBE (store that in @do_switch variable).
613  */
614 static int
615 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
616 {
617 	int nd_delay, nd_gctimer, r_skip_req;
618 	time_t lle_hittime;
619 	long delay;
620 
621 	*do_switch = 0;
622 	nd_gctimer = V_nd6_gctimer;
623 	nd_delay = V_nd6_delay;
624 
625 	LLE_REQ_LOCK(lle);
626 	r_skip_req = lle->r_skip_req;
627 	lle_hittime = lle->lle_hittime;
628 	LLE_REQ_UNLOCK(lle);
629 
630 	if (r_skip_req > 0) {
631 
632 		/*
633 		 * Nonzero r_skip_req value was set upon entering
634 		 * STALE state. Since value was not changed, no
635 		 * packets were passed using this lle. Ask for
636 		 * timer reschedule and keep STALE state.
637 		 */
638 		delay = (long)(MIN(nd_gctimer, nd_delay));
639 		delay *= hz;
640 		if (lle->lle_remtime > delay)
641 			lle->lle_remtime -= delay;
642 		else {
643 			delay = lle->lle_remtime;
644 			lle->lle_remtime = 0;
645 		}
646 
647 		if (delay == 0) {
648 
649 			/*
650 			 * The original ng6_gctime timeout ended,
651 			 * no more rescheduling.
652 			 */
653 			return (0);
654 		}
655 
656 		*pdelay = delay;
657 		return (1);
658 	}
659 
660 	/*
661 	 * Packet received. Verify timestamp
662 	 */
663 	delay = (long)(time_uptime - lle_hittime);
664 	if (delay < nd_delay) {
665 
666 		/*
667 		 * V_nd6_delay still not passed since the first
668 		 * hit in STALE state.
669 		 * Reshedule timer and return.
670 		 */
671 		*pdelay = (long)(nd_delay - delay) * hz;
672 		return (1);
673 	}
674 
675 	/* Request switching to probe */
676 	*do_switch = 1;
677 	return (0);
678 }
679 
680 
681 /*
682  * Switch @lle state to new state optionally arming timers.
683  *
684  * Set noinline to be dtrace-friendly
685  */
686 __noinline void
687 nd6_llinfo_setstate(struct llentry *lle, int newstate)
688 {
689 	struct ifnet *ifp;
690 	int nd_gctimer, nd_delay;
691 	long delay, remtime;
692 
693 	delay = 0;
694 	remtime = 0;
695 
696 	switch (newstate) {
697 	case ND6_LLINFO_INCOMPLETE:
698 		ifp = lle->lle_tbl->llt_ifp;
699 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
700 		break;
701 	case ND6_LLINFO_REACHABLE:
702 		if (!ND6_LLINFO_PERMANENT(lle)) {
703 			ifp = lle->lle_tbl->llt_ifp;
704 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
705 		}
706 		break;
707 	case ND6_LLINFO_STALE:
708 
709 		/*
710 		 * Notify fast path that we want to know if any packet
711 		 * is transmitted by setting r_skip_req.
712 		 */
713 		LLE_REQ_LOCK(lle);
714 		lle->r_skip_req = 1;
715 		LLE_REQ_UNLOCK(lle);
716 		nd_delay = V_nd6_delay;
717 		nd_gctimer = V_nd6_gctimer;
718 
719 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
720 		remtime = (long)nd_gctimer * hz - delay;
721 		break;
722 	case ND6_LLINFO_DELAY:
723 		lle->la_asked = 0;
724 		delay = (long)V_nd6_delay * hz;
725 		break;
726 	}
727 
728 	if (delay > 0)
729 		nd6_llinfo_settimer_locked(lle, delay);
730 
731 	lle->lle_remtime = remtime;
732 	lle->ln_state = newstate;
733 }
734 
735 /*
736  * Timer-dependent part of nd state machine.
737  *
738  * Set noinline to be dtrace-friendly
739  */
740 static __noinline void
741 nd6_llinfo_timer(void *arg)
742 {
743 	struct llentry *ln;
744 	struct in6_addr *dst, *pdst, *psrc, src;
745 	struct ifnet *ifp;
746 	struct nd_ifinfo *ndi;
747 	int do_switch, send_ns;
748 	long delay;
749 
750 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
751 	ln = (struct llentry *)arg;
752 	ifp = lltable_get_ifp(ln->lle_tbl);
753 	CURVNET_SET(ifp->if_vnet);
754 
755 	ND6_RLOCK();
756 	LLE_WLOCK(ln);
757 	if (callout_pending(&ln->lle_timer)) {
758 		/*
759 		 * Here we are a bit odd here in the treatment of
760 		 * active/pending. If the pending bit is set, it got
761 		 * rescheduled before I ran. The active
762 		 * bit we ignore, since if it was stopped
763 		 * in ll_tablefree() and was currently running
764 		 * it would have return 0 so the code would
765 		 * not have deleted it since the callout could
766 		 * not be stopped so we want to go through
767 		 * with the delete here now. If the callout
768 		 * was restarted, the pending bit will be back on and
769 		 * we just want to bail since the callout_reset would
770 		 * return 1 and our reference would have been removed
771 		 * by nd6_llinfo_settimer_locked above since canceled
772 		 * would have been 1.
773 		 */
774 		LLE_WUNLOCK(ln);
775 		ND6_RUNLOCK();
776 		CURVNET_RESTORE();
777 		return;
778 	}
779 	ndi = ND_IFINFO(ifp);
780 	send_ns = 0;
781 	dst = &ln->r_l3addr.addr6;
782 	pdst = dst;
783 
784 	if (ln->ln_ntick > 0) {
785 		if (ln->ln_ntick > INT_MAX) {
786 			ln->ln_ntick -= INT_MAX;
787 			nd6_llinfo_settimer_locked(ln, INT_MAX);
788 		} else {
789 			ln->ln_ntick = 0;
790 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
791 		}
792 		goto done;
793 	}
794 
795 	if (ln->la_flags & LLE_STATIC) {
796 		goto done;
797 	}
798 
799 	if (ln->la_flags & LLE_DELETED) {
800 		nd6_free(&ln, 0);
801 		goto done;
802 	}
803 
804 	switch (ln->ln_state) {
805 	case ND6_LLINFO_INCOMPLETE:
806 		if (ln->la_asked < V_nd6_mmaxtries) {
807 			ln->la_asked++;
808 			send_ns = 1;
809 			/* Send NS to multicast address */
810 			pdst = NULL;
811 		} else {
812 			struct mbuf *m = ln->la_hold;
813 			if (m) {
814 				struct mbuf *m0;
815 
816 				/*
817 				 * assuming every packet in la_hold has the
818 				 * same IP header.  Send error after unlock.
819 				 */
820 				m0 = m->m_nextpkt;
821 				m->m_nextpkt = NULL;
822 				ln->la_hold = m0;
823 				clear_llinfo_pqueue(ln);
824 			}
825 			nd6_free(&ln, 0);
826 			if (m != NULL)
827 				icmp6_error2(m, ICMP6_DST_UNREACH,
828 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
829 		}
830 		break;
831 	case ND6_LLINFO_REACHABLE:
832 		if (!ND6_LLINFO_PERMANENT(ln))
833 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
834 		break;
835 
836 	case ND6_LLINFO_STALE:
837 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
838 
839 			/*
840 			 * No packet has used this entry and GC timeout
841 			 * has not been passed. Reshedule timer and
842 			 * return.
843 			 */
844 			nd6_llinfo_settimer_locked(ln, delay);
845 			break;
846 		}
847 
848 		if (do_switch == 0) {
849 
850 			/*
851 			 * GC timer has ended and entry hasn't been used.
852 			 * Run Garbage collector (RFC 4861, 5.3)
853 			 */
854 			if (!ND6_LLINFO_PERMANENT(ln))
855 				nd6_free(&ln, 1);
856 			break;
857 		}
858 
859 		/* Entry has been used AND delay timer has ended. */
860 
861 		/* FALLTHROUGH */
862 
863 	case ND6_LLINFO_DELAY:
864 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
865 			/* We need NUD */
866 			ln->la_asked = 1;
867 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
868 			send_ns = 1;
869 		} else
870 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
871 		break;
872 	case ND6_LLINFO_PROBE:
873 		if (ln->la_asked < V_nd6_umaxtries) {
874 			ln->la_asked++;
875 			send_ns = 1;
876 		} else {
877 			nd6_free(&ln, 0);
878 		}
879 		break;
880 	default:
881 		panic("%s: paths in a dark night can be confusing: %d",
882 		    __func__, ln->ln_state);
883 	}
884 done:
885 	if (ln != NULL)
886 		ND6_RUNLOCK();
887 	if (send_ns != 0) {
888 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
889 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
890 		LLE_FREE_LOCKED(ln);
891 		ln = NULL;
892 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
893 	}
894 
895 	if (ln != NULL)
896 		LLE_FREE_LOCKED(ln);
897 	CURVNET_RESTORE();
898 }
899 
900 
901 /*
902  * ND6 timer routine to expire default route list and prefix list
903  */
904 void
905 nd6_timer(void *arg)
906 {
907 	CURVNET_SET((struct vnet *) arg);
908 	struct nd_drhead drq;
909 	struct nd_prhead prl;
910 	struct nd_defrouter *dr, *ndr;
911 	struct nd_prefix *pr, *npr;
912 	struct in6_ifaddr *ia6, *nia6;
913 	uint64_t genid;
914 
915 	TAILQ_INIT(&drq);
916 	LIST_INIT(&prl);
917 
918 	ND6_WLOCK();
919 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
920 		if (dr->expire && dr->expire < time_uptime)
921 			defrouter_unlink(dr, &drq);
922 	ND6_WUNLOCK();
923 
924 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
925 		TAILQ_REMOVE(&drq, dr, dr_entry);
926 		defrouter_del(dr);
927 	}
928 
929 	/*
930 	 * expire interface addresses.
931 	 * in the past the loop was inside prefix expiry processing.
932 	 * However, from a stricter speci-confrmance standpoint, we should
933 	 * rather separate address lifetimes and prefix lifetimes.
934 	 *
935 	 * XXXRW: in6_ifaddrhead locking.
936 	 */
937   addrloop:
938 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
939 		/* check address lifetime */
940 		if (IFA6_IS_INVALID(ia6)) {
941 			int regen = 0;
942 
943 			/*
944 			 * If the expiring address is temporary, try
945 			 * regenerating a new one.  This would be useful when
946 			 * we suspended a laptop PC, then turned it on after a
947 			 * period that could invalidate all temporary
948 			 * addresses.  Although we may have to restart the
949 			 * loop (see below), it must be after purging the
950 			 * address.  Otherwise, we'd see an infinite loop of
951 			 * regeneration.
952 			 */
953 			if (V_ip6_use_tempaddr &&
954 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
955 				if (regen_tmpaddr(ia6) == 0)
956 					regen = 1;
957 			}
958 
959 			in6_purgeaddr(&ia6->ia_ifa);
960 
961 			if (regen)
962 				goto addrloop; /* XXX: see below */
963 		} else if (IFA6_IS_DEPRECATED(ia6)) {
964 			int oldflags = ia6->ia6_flags;
965 
966 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
967 
968 			/*
969 			 * If a temporary address has just become deprecated,
970 			 * regenerate a new one if possible.
971 			 */
972 			if (V_ip6_use_tempaddr &&
973 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
974 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
975 
976 				if (regen_tmpaddr(ia6) == 0) {
977 					/*
978 					 * A new temporary address is
979 					 * generated.
980 					 * XXX: this means the address chain
981 					 * has changed while we are still in
982 					 * the loop.  Although the change
983 					 * would not cause disaster (because
984 					 * it's not a deletion, but an
985 					 * addition,) we'd rather restart the
986 					 * loop just for safety.  Or does this
987 					 * significantly reduce performance??
988 					 */
989 					goto addrloop;
990 				}
991 			}
992 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
993 			/*
994 			 * Schedule DAD for a tentative address.  This happens
995 			 * if the interface was down or not running
996 			 * when the address was configured.
997 			 */
998 			int delay;
999 
1000 			delay = arc4random() %
1001 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1002 			nd6_dad_start((struct ifaddr *)ia6, delay);
1003 		} else {
1004 			/*
1005 			 * Check status of the interface.  If it is down,
1006 			 * mark the address as tentative for future DAD.
1007 			 */
1008 			if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
1009 			    (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
1010 				== 0 ||
1011 			    (ND_IFINFO(ia6->ia_ifp)->flags &
1012 				ND6_IFF_IFDISABLED) != 0) {
1013 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1014 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1015 			}
1016 			/*
1017 			 * A new RA might have made a deprecated address
1018 			 * preferred.
1019 			 */
1020 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1021 		}
1022 	}
1023 
1024 	ND6_WLOCK();
1025 restart:
1026 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1027 		/*
1028 		 * Expire prefixes. Since the pltime is only used for
1029 		 * autoconfigured addresses, pltime processing for prefixes is
1030 		 * not necessary.
1031 		 *
1032 		 * Only unlink after all derived addresses have expired. This
1033 		 * may not occur until two hours after the prefix has expired
1034 		 * per RFC 4862. If the prefix expires before its derived
1035 		 * addresses, mark it off-link. This will be done automatically
1036 		 * after unlinking if no address references remain.
1037 		 */
1038 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1039 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1040 			continue;
1041 
1042 		if (pr->ndpr_addrcnt == 0) {
1043 			nd6_prefix_unlink(pr, &prl);
1044 			continue;
1045 		}
1046 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1047 			genid = V_nd6_list_genid;
1048 			nd6_prefix_ref(pr);
1049 			ND6_WUNLOCK();
1050 			ND6_ONLINK_LOCK();
1051 			(void)nd6_prefix_offlink(pr);
1052 			ND6_ONLINK_UNLOCK();
1053 			ND6_WLOCK();
1054 			nd6_prefix_rele(pr);
1055 			if (genid != V_nd6_list_genid)
1056 				goto restart;
1057 		}
1058 	}
1059 	ND6_WUNLOCK();
1060 
1061 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1062 		LIST_REMOVE(pr, ndpr_entry);
1063 		nd6_prefix_del(pr);
1064 	}
1065 
1066 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1067 	    nd6_timer, curvnet);
1068 
1069 	CURVNET_RESTORE();
1070 }
1071 
1072 /*
1073  * ia6 - deprecated/invalidated temporary address
1074  */
1075 static int
1076 regen_tmpaddr(struct in6_ifaddr *ia6)
1077 {
1078 	struct ifaddr *ifa;
1079 	struct ifnet *ifp;
1080 	struct in6_ifaddr *public_ifa6 = NULL;
1081 
1082 	ifp = ia6->ia_ifa.ifa_ifp;
1083 	IF_ADDR_RLOCK(ifp);
1084 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1085 		struct in6_ifaddr *it6;
1086 
1087 		if (ifa->ifa_addr->sa_family != AF_INET6)
1088 			continue;
1089 
1090 		it6 = (struct in6_ifaddr *)ifa;
1091 
1092 		/* ignore no autoconf addresses. */
1093 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1094 			continue;
1095 
1096 		/* ignore autoconf addresses with different prefixes. */
1097 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1098 			continue;
1099 
1100 		/*
1101 		 * Now we are looking at an autoconf address with the same
1102 		 * prefix as ours.  If the address is temporary and is still
1103 		 * preferred, do not create another one.  It would be rare, but
1104 		 * could happen, for example, when we resume a laptop PC after
1105 		 * a long period.
1106 		 */
1107 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1108 		    !IFA6_IS_DEPRECATED(it6)) {
1109 			public_ifa6 = NULL;
1110 			break;
1111 		}
1112 
1113 		/*
1114 		 * This is a public autoconf address that has the same prefix
1115 		 * as ours.  If it is preferred, keep it.  We can't break the
1116 		 * loop here, because there may be a still-preferred temporary
1117 		 * address with the prefix.
1118 		 */
1119 		if (!IFA6_IS_DEPRECATED(it6))
1120 			public_ifa6 = it6;
1121 	}
1122 	if (public_ifa6 != NULL)
1123 		ifa_ref(&public_ifa6->ia_ifa);
1124 	IF_ADDR_RUNLOCK(ifp);
1125 
1126 	if (public_ifa6 != NULL) {
1127 		int e;
1128 
1129 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1130 			ifa_free(&public_ifa6->ia_ifa);
1131 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1132 			    " tmp addr,errno=%d\n", e);
1133 			return (-1);
1134 		}
1135 		ifa_free(&public_ifa6->ia_ifa);
1136 		return (0);
1137 	}
1138 
1139 	return (-1);
1140 }
1141 
1142 /*
1143  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1144  * cache entries are freed in in6_domifdetach().
1145  */
1146 void
1147 nd6_purge(struct ifnet *ifp)
1148 {
1149 	struct nd_drhead drq;
1150 	struct nd_prhead prl;
1151 	struct nd_defrouter *dr, *ndr;
1152 	struct nd_prefix *pr, *npr;
1153 
1154 	TAILQ_INIT(&drq);
1155 	LIST_INIT(&prl);
1156 
1157 	/*
1158 	 * Nuke default router list entries toward ifp.
1159 	 * We defer removal of default router list entries that is installed
1160 	 * in the routing table, in order to keep additional side effects as
1161 	 * small as possible.
1162 	 */
1163 	ND6_WLOCK();
1164 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1165 		if (dr->installed)
1166 			continue;
1167 		if (dr->ifp == ifp)
1168 			defrouter_unlink(dr, &drq);
1169 	}
1170 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1171 		if (!dr->installed)
1172 			continue;
1173 		if (dr->ifp == ifp)
1174 			defrouter_unlink(dr, &drq);
1175 	}
1176 
1177 	/*
1178 	 * Remove prefixes on ifp. We should have already removed addresses on
1179 	 * this interface, so no addresses should be referencing these prefixes.
1180 	 */
1181 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1182 		if (pr->ndpr_ifp == ifp)
1183 			nd6_prefix_unlink(pr, &prl);
1184 	}
1185 	ND6_WUNLOCK();
1186 
1187 	/* Delete the unlinked router and prefix objects. */
1188 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1189 		TAILQ_REMOVE(&drq, dr, dr_entry);
1190 		defrouter_del(dr);
1191 	}
1192 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1193 		LIST_REMOVE(pr, ndpr_entry);
1194 		nd6_prefix_del(pr);
1195 	}
1196 
1197 	/* cancel default outgoing interface setting */
1198 	if (V_nd6_defifindex == ifp->if_index)
1199 		nd6_setdefaultiface(0);
1200 
1201 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1202 		/* Refresh default router list. */
1203 		defrouter_select();
1204 	}
1205 }
1206 
1207 /*
1208  * the caller acquires and releases the lock on the lltbls
1209  * Returns the llentry locked
1210  */
1211 struct llentry *
1212 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1213 {
1214 	struct sockaddr_in6 sin6;
1215 	struct llentry *ln;
1216 
1217 	bzero(&sin6, sizeof(sin6));
1218 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1219 	sin6.sin6_family = AF_INET6;
1220 	sin6.sin6_addr = *addr6;
1221 
1222 	IF_AFDATA_LOCK_ASSERT(ifp);
1223 
1224 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1225 
1226 	return (ln);
1227 }
1228 
1229 struct llentry *
1230 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1231 {
1232 	struct sockaddr_in6 sin6;
1233 	struct llentry *ln;
1234 
1235 	bzero(&sin6, sizeof(sin6));
1236 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1237 	sin6.sin6_family = AF_INET6;
1238 	sin6.sin6_addr = *addr6;
1239 
1240 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1241 	if (ln != NULL)
1242 		ln->ln_state = ND6_LLINFO_NOSTATE;
1243 
1244 	return (ln);
1245 }
1246 
1247 /*
1248  * Test whether a given IPv6 address is a neighbor or not, ignoring
1249  * the actual neighbor cache.  The neighbor cache is ignored in order
1250  * to not reenter the routing code from within itself.
1251  */
1252 static int
1253 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1254 {
1255 	struct nd_prefix *pr;
1256 	struct ifaddr *dstaddr;
1257 	struct rt_addrinfo info;
1258 	struct sockaddr_in6 rt_key;
1259 	const struct sockaddr *dst6;
1260 	uint64_t genid;
1261 	int error, fibnum;
1262 
1263 	/*
1264 	 * A link-local address is always a neighbor.
1265 	 * XXX: a link does not necessarily specify a single interface.
1266 	 */
1267 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1268 		struct sockaddr_in6 sin6_copy;
1269 		u_int32_t zone;
1270 
1271 		/*
1272 		 * We need sin6_copy since sa6_recoverscope() may modify the
1273 		 * content (XXX).
1274 		 */
1275 		sin6_copy = *addr;
1276 		if (sa6_recoverscope(&sin6_copy))
1277 			return (0); /* XXX: should be impossible */
1278 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1279 			return (0);
1280 		if (sin6_copy.sin6_scope_id == zone)
1281 			return (1);
1282 		else
1283 			return (0);
1284 	}
1285 
1286 	bzero(&rt_key, sizeof(rt_key));
1287 	bzero(&info, sizeof(info));
1288 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1289 
1290 	/* Always use the default FIB here. XXME - why? */
1291 	fibnum = RT_DEFAULT_FIB;
1292 
1293 	/*
1294 	 * If the address matches one of our addresses,
1295 	 * it should be a neighbor.
1296 	 * If the address matches one of our on-link prefixes, it should be a
1297 	 * neighbor.
1298 	 */
1299 	ND6_RLOCK();
1300 restart:
1301 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1302 		if (pr->ndpr_ifp != ifp)
1303 			continue;
1304 
1305 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1306 			/* Always use the default FIB here. */
1307 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1308 
1309 			genid = V_nd6_list_genid;
1310 			ND6_RUNLOCK();
1311 
1312 			/* Restore length field before retrying lookup */
1313 			rt_key.sin6_len = sizeof(rt_key);
1314 			error = rib_lookup_info(fibnum, dst6, 0, 0, &info);
1315 
1316 			ND6_RLOCK();
1317 			if (genid != V_nd6_list_genid)
1318 				goto restart;
1319 			if (error != 0)
1320 				continue;
1321 
1322 			/*
1323 			 * This is the case where multiple interfaces
1324 			 * have the same prefix, but only one is installed
1325 			 * into the routing table and that prefix entry
1326 			 * is not the one being examined here. In the case
1327 			 * where RADIX_MPATH is enabled, multiple route
1328 			 * entries (of the same rt_key value) will be
1329 			 * installed because the interface addresses all
1330 			 * differ.
1331 			 */
1332 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1333 			    &rt_key.sin6_addr))
1334 				continue;
1335 		}
1336 
1337 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1338 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1339 			ND6_RUNLOCK();
1340 			return (1);
1341 		}
1342 	}
1343 	ND6_RUNLOCK();
1344 
1345 	/*
1346 	 * If the address is assigned on the node of the other side of
1347 	 * a p2p interface, the address should be a neighbor.
1348 	 */
1349 	dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1350 	if (dstaddr != NULL) {
1351 		if (dstaddr->ifa_ifp == ifp) {
1352 			ifa_free(dstaddr);
1353 			return (1);
1354 		}
1355 		ifa_free(dstaddr);
1356 	}
1357 
1358 	/*
1359 	 * If the default router list is empty, all addresses are regarded
1360 	 * as on-link, and thus, as a neighbor.
1361 	 */
1362 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1363 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1364 	    V_nd6_defifindex == ifp->if_index) {
1365 		return (1);
1366 	}
1367 
1368 	return (0);
1369 }
1370 
1371 
1372 /*
1373  * Detect if a given IPv6 address identifies a neighbor on a given link.
1374  * XXX: should take care of the destination of a p2p link?
1375  */
1376 int
1377 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1378 {
1379 	struct llentry *lle;
1380 	int rc = 0;
1381 
1382 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1383 	if (nd6_is_new_addr_neighbor(addr, ifp))
1384 		return (1);
1385 
1386 	/*
1387 	 * Even if the address matches none of our addresses, it might be
1388 	 * in the neighbor cache.
1389 	 */
1390 	IF_AFDATA_RLOCK(ifp);
1391 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1392 		LLE_RUNLOCK(lle);
1393 		rc = 1;
1394 	}
1395 	IF_AFDATA_RUNLOCK(ifp);
1396 	return (rc);
1397 }
1398 
1399 /*
1400  * Free an nd6 llinfo entry.
1401  * Since the function would cause significant changes in the kernel, DO NOT
1402  * make it global, unless you have a strong reason for the change, and are sure
1403  * that the change is safe.
1404  *
1405  * Set noinline to be dtrace-friendly
1406  */
1407 static __noinline void
1408 nd6_free(struct llentry **lnp, int gc)
1409 {
1410 	struct ifnet *ifp;
1411 	struct llentry *ln;
1412 	struct nd_defrouter *dr;
1413 
1414 	ln = *lnp;
1415 	*lnp = NULL;
1416 
1417 	LLE_WLOCK_ASSERT(ln);
1418 	ND6_RLOCK_ASSERT();
1419 
1420 	ifp = lltable_get_ifp(ln->lle_tbl);
1421 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1422 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1423 	else
1424 		dr = NULL;
1425 	ND6_RUNLOCK();
1426 
1427 	if ((ln->la_flags & LLE_DELETED) == 0)
1428 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1429 
1430 	/*
1431 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1432 	 * even though it is not harmful, it was not really necessary.
1433 	 */
1434 
1435 	/* cancel timer */
1436 	nd6_llinfo_settimer_locked(ln, -1);
1437 
1438 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1439 		if (dr != NULL && dr->expire &&
1440 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1441 			/*
1442 			 * If the reason for the deletion is just garbage
1443 			 * collection, and the neighbor is an active default
1444 			 * router, do not delete it.  Instead, reset the GC
1445 			 * timer using the router's lifetime.
1446 			 * Simply deleting the entry would affect default
1447 			 * router selection, which is not necessarily a good
1448 			 * thing, especially when we're using router preference
1449 			 * values.
1450 			 * XXX: the check for ln_state would be redundant,
1451 			 *      but we intentionally keep it just in case.
1452 			 */
1453 			if (dr->expire > time_uptime)
1454 				nd6_llinfo_settimer_locked(ln,
1455 				    (dr->expire - time_uptime) * hz);
1456 			else
1457 				nd6_llinfo_settimer_locked(ln,
1458 				    (long)V_nd6_gctimer * hz);
1459 
1460 			LLE_REMREF(ln);
1461 			LLE_WUNLOCK(ln);
1462 			defrouter_rele(dr);
1463 			return;
1464 		}
1465 
1466 		if (dr) {
1467 			/*
1468 			 * Unreachablity of a router might affect the default
1469 			 * router selection and on-link detection of advertised
1470 			 * prefixes.
1471 			 */
1472 
1473 			/*
1474 			 * Temporarily fake the state to choose a new default
1475 			 * router and to perform on-link determination of
1476 			 * prefixes correctly.
1477 			 * Below the state will be set correctly,
1478 			 * or the entry itself will be deleted.
1479 			 */
1480 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1481 		}
1482 
1483 		if (ln->ln_router || dr) {
1484 
1485 			/*
1486 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1487 			 * rnh and for the calls to pfxlist_onlink_check() and
1488 			 * defrouter_select() in the block further down for calls
1489 			 * into nd6_lookup().  We still hold a ref.
1490 			 */
1491 			LLE_WUNLOCK(ln);
1492 
1493 			/*
1494 			 * rt6_flush must be called whether or not the neighbor
1495 			 * is in the Default Router List.
1496 			 * See a corresponding comment in nd6_na_input().
1497 			 */
1498 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1499 		}
1500 
1501 		if (dr) {
1502 			/*
1503 			 * Since defrouter_select() does not affect the
1504 			 * on-link determination and MIP6 needs the check
1505 			 * before the default router selection, we perform
1506 			 * the check now.
1507 			 */
1508 			pfxlist_onlink_check();
1509 
1510 			/*
1511 			 * Refresh default router list.
1512 			 */
1513 			defrouter_select();
1514 		}
1515 
1516 		/*
1517 		 * If this entry was added by an on-link redirect, remove the
1518 		 * corresponding host route.
1519 		 */
1520 		if (ln->la_flags & LLE_REDIRECT)
1521 			nd6_free_redirect(ln);
1522 
1523 		if (ln->ln_router || dr)
1524 			LLE_WLOCK(ln);
1525 	}
1526 
1527 	/*
1528 	 * Save to unlock. We still hold an extra reference and will not
1529 	 * free(9) in llentry_free() if someone else holds one as well.
1530 	 */
1531 	LLE_WUNLOCK(ln);
1532 	IF_AFDATA_LOCK(ifp);
1533 	LLE_WLOCK(ln);
1534 	/* Guard against race with other llentry_free(). */
1535 	if (ln->la_flags & LLE_LINKED) {
1536 		/* Remove callout reference */
1537 		LLE_REMREF(ln);
1538 		lltable_unlink_entry(ln->lle_tbl, ln);
1539 	}
1540 	IF_AFDATA_UNLOCK(ifp);
1541 
1542 	llentry_free(ln);
1543 	if (dr != NULL)
1544 		defrouter_rele(dr);
1545 }
1546 
1547 static int
1548 nd6_isdynrte(const struct rtentry *rt, void *xap)
1549 {
1550 
1551 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1552 		return (1);
1553 
1554 	return (0);
1555 }
1556 /*
1557  * Remove the rtentry for the given llentry,
1558  * both of which were installed by a redirect.
1559  */
1560 static void
1561 nd6_free_redirect(const struct llentry *ln)
1562 {
1563 	int fibnum;
1564 	struct sockaddr_in6 sin6;
1565 	struct rt_addrinfo info;
1566 
1567 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1568 	memset(&info, 0, sizeof(info));
1569 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1570 	info.rti_filter = nd6_isdynrte;
1571 
1572 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1573 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1574 }
1575 
1576 /*
1577  * Rejuvenate this function for routing operations related
1578  * processing.
1579  */
1580 void
1581 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1582 {
1583 	struct sockaddr_in6 *gateway;
1584 	struct nd_defrouter *dr;
1585 	struct ifnet *ifp;
1586 
1587 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1588 	ifp = rt->rt_ifp;
1589 
1590 	switch (req) {
1591 	case RTM_ADD:
1592 		break;
1593 
1594 	case RTM_DELETE:
1595 		if (!ifp)
1596 			return;
1597 		/*
1598 		 * Only indirect routes are interesting.
1599 		 */
1600 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1601 			return;
1602 		/*
1603 		 * check for default route
1604 		 */
1605 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1606 		    &SIN6(rt_key(rt))->sin6_addr)) {
1607 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1608 			if (dr != NULL) {
1609 				dr->installed = 0;
1610 				defrouter_rele(dr);
1611 			}
1612 		}
1613 		break;
1614 	}
1615 }
1616 
1617 
1618 int
1619 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1620 {
1621 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1622 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1623 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1624 	int error = 0;
1625 
1626 	if (ifp->if_afdata[AF_INET6] == NULL)
1627 		return (EPFNOSUPPORT);
1628 	switch (cmd) {
1629 	case OSIOCGIFINFO_IN6:
1630 #define ND	ndi->ndi
1631 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1632 		bzero(&ND, sizeof(ND));
1633 		ND.linkmtu = IN6_LINKMTU(ifp);
1634 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1635 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1636 		ND.reachable = ND_IFINFO(ifp)->reachable;
1637 		ND.retrans = ND_IFINFO(ifp)->retrans;
1638 		ND.flags = ND_IFINFO(ifp)->flags;
1639 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1640 		ND.chlim = ND_IFINFO(ifp)->chlim;
1641 		break;
1642 	case SIOCGIFINFO_IN6:
1643 		ND = *ND_IFINFO(ifp);
1644 		break;
1645 	case SIOCSIFINFO_IN6:
1646 		/*
1647 		 * used to change host variables from userland.
1648 		 * intended for a use on router to reflect RA configurations.
1649 		 */
1650 		/* 0 means 'unspecified' */
1651 		if (ND.linkmtu != 0) {
1652 			if (ND.linkmtu < IPV6_MMTU ||
1653 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1654 				error = EINVAL;
1655 				break;
1656 			}
1657 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1658 		}
1659 
1660 		if (ND.basereachable != 0) {
1661 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1662 
1663 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1664 			if (ND.basereachable != obasereachable)
1665 				ND_IFINFO(ifp)->reachable =
1666 				    ND_COMPUTE_RTIME(ND.basereachable);
1667 		}
1668 		if (ND.retrans != 0)
1669 			ND_IFINFO(ifp)->retrans = ND.retrans;
1670 		if (ND.chlim != 0)
1671 			ND_IFINFO(ifp)->chlim = ND.chlim;
1672 		/* FALLTHROUGH */
1673 	case SIOCSIFINFO_FLAGS:
1674 	{
1675 		struct ifaddr *ifa;
1676 		struct in6_ifaddr *ia;
1677 
1678 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1679 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1680 			/* ifdisabled 1->0 transision */
1681 
1682 			/*
1683 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1684 			 * has an link-local address with IN6_IFF_DUPLICATED,
1685 			 * do not clear ND6_IFF_IFDISABLED.
1686 			 * See RFC 4862, Section 5.4.5.
1687 			 */
1688 			IF_ADDR_RLOCK(ifp);
1689 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1690 				if (ifa->ifa_addr->sa_family != AF_INET6)
1691 					continue;
1692 				ia = (struct in6_ifaddr *)ifa;
1693 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1694 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1695 					break;
1696 			}
1697 			IF_ADDR_RUNLOCK(ifp);
1698 
1699 			if (ifa != NULL) {
1700 				/* LLA is duplicated. */
1701 				ND.flags |= ND6_IFF_IFDISABLED;
1702 				log(LOG_ERR, "Cannot enable an interface"
1703 				    " with a link-local address marked"
1704 				    " duplicate.\n");
1705 			} else {
1706 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1707 				if (ifp->if_flags & IFF_UP)
1708 					in6_if_up(ifp);
1709 			}
1710 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1711 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1712 			/* ifdisabled 0->1 transision */
1713 			/* Mark all IPv6 address as tentative. */
1714 
1715 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1716 			if (V_ip6_dad_count > 0 &&
1717 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1718 				IF_ADDR_RLOCK(ifp);
1719 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1720 				    ifa_link) {
1721 					if (ifa->ifa_addr->sa_family !=
1722 					    AF_INET6)
1723 						continue;
1724 					ia = (struct in6_ifaddr *)ifa;
1725 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1726 				}
1727 				IF_ADDR_RUNLOCK(ifp);
1728 			}
1729 		}
1730 
1731 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1732 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1733 				/* auto_linklocal 0->1 transision */
1734 
1735 				/* If no link-local address on ifp, configure */
1736 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1737 				in6_ifattach(ifp, NULL);
1738 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1739 			    ifp->if_flags & IFF_UP) {
1740 				/*
1741 				 * When the IF already has
1742 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1743 				 * address is assigned, and IFF_UP, try to
1744 				 * assign one.
1745 				 */
1746 				IF_ADDR_RLOCK(ifp);
1747 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1748 				    ifa_link) {
1749 					if (ifa->ifa_addr->sa_family !=
1750 					    AF_INET6)
1751 						continue;
1752 					ia = (struct in6_ifaddr *)ifa;
1753 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1754 						break;
1755 				}
1756 				IF_ADDR_RUNLOCK(ifp);
1757 				if (ifa != NULL)
1758 					/* No LLA is configured. */
1759 					in6_ifattach(ifp, NULL);
1760 			}
1761 		}
1762 	}
1763 		ND_IFINFO(ifp)->flags = ND.flags;
1764 		break;
1765 #undef ND
1766 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1767 		/* sync kernel routing table with the default router list */
1768 		defrouter_reset();
1769 		defrouter_select();
1770 		break;
1771 	case SIOCSPFXFLUSH_IN6:
1772 	{
1773 		/* flush all the prefix advertised by routers */
1774 		struct in6_ifaddr *ia, *ia_next;
1775 		struct nd_prefix *pr, *next;
1776 		struct nd_prhead prl;
1777 
1778 		LIST_INIT(&prl);
1779 
1780 		ND6_WLOCK();
1781 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1782 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1783 				continue; /* XXX */
1784 			nd6_prefix_unlink(pr, &prl);
1785 		}
1786 		ND6_WUNLOCK();
1787 
1788 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1789 			LIST_REMOVE(pr, ndpr_entry);
1790 			/* XXXRW: in6_ifaddrhead locking. */
1791 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1792 			    ia_next) {
1793 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1794 					continue;
1795 
1796 				if (ia->ia6_ndpr == pr)
1797 					in6_purgeaddr(&ia->ia_ifa);
1798 			}
1799 			nd6_prefix_del(pr);
1800 		}
1801 		break;
1802 	}
1803 	case SIOCSRTRFLUSH_IN6:
1804 	{
1805 		/* flush all the default routers */
1806 		struct nd_drhead drq;
1807 		struct nd_defrouter *dr;
1808 
1809 		TAILQ_INIT(&drq);
1810 
1811 		defrouter_reset();
1812 
1813 		ND6_WLOCK();
1814 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1815 			defrouter_unlink(dr, &drq);
1816 		ND6_WUNLOCK();
1817 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1818 			TAILQ_REMOVE(&drq, dr, dr_entry);
1819 			defrouter_del(dr);
1820 		}
1821 
1822 		defrouter_select();
1823 		break;
1824 	}
1825 	case SIOCGNBRINFO_IN6:
1826 	{
1827 		struct llentry *ln;
1828 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1829 
1830 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1831 			return (error);
1832 
1833 		IF_AFDATA_RLOCK(ifp);
1834 		ln = nd6_lookup(&nb_addr, 0, ifp);
1835 		IF_AFDATA_RUNLOCK(ifp);
1836 
1837 		if (ln == NULL) {
1838 			error = EINVAL;
1839 			break;
1840 		}
1841 		nbi->state = ln->ln_state;
1842 		nbi->asked = ln->la_asked;
1843 		nbi->isrouter = ln->ln_router;
1844 		if (ln->la_expire == 0)
1845 			nbi->expire = 0;
1846 		else
1847 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1848 			    (time_second - time_uptime);
1849 		LLE_RUNLOCK(ln);
1850 		break;
1851 	}
1852 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1853 		ndif->ifindex = V_nd6_defifindex;
1854 		break;
1855 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1856 		return (nd6_setdefaultiface(ndif->ifindex));
1857 	}
1858 	return (error);
1859 }
1860 
1861 /*
1862  * Calculates new isRouter value based on provided parameters and
1863  * returns it.
1864  */
1865 static int
1866 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1867     int ln_router)
1868 {
1869 
1870 	/*
1871 	 * ICMP6 type dependent behavior.
1872 	 *
1873 	 * NS: clear IsRouter if new entry
1874 	 * RS: clear IsRouter
1875 	 * RA: set IsRouter if there's lladdr
1876 	 * redir: clear IsRouter if new entry
1877 	 *
1878 	 * RA case, (1):
1879 	 * The spec says that we must set IsRouter in the following cases:
1880 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1881 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1882 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1883 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1884 	 * neighbor cache, this is similar to (6).
1885 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1886 	 *
1887 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1888 	 *							D R
1889 	 *	0	n	n	(1)	c   ?     s
1890 	 *	0	y	n	(2)	c   s     s
1891 	 *	0	n	y	(3)	c   s     s
1892 	 *	0	y	y	(4)	c   s     s
1893 	 *	0	y	y	(5)	c   s     s
1894 	 *	1	--	n	(6) c	c	c s
1895 	 *	1	--	y	(7) c	c   s	c s
1896 	 *
1897 	 *					(c=clear s=set)
1898 	 */
1899 	switch (type & 0xff) {
1900 	case ND_NEIGHBOR_SOLICIT:
1901 		/*
1902 		 * New entry must have is_router flag cleared.
1903 		 */
1904 		if (is_new)					/* (6-7) */
1905 			ln_router = 0;
1906 		break;
1907 	case ND_REDIRECT:
1908 		/*
1909 		 * If the icmp is a redirect to a better router, always set the
1910 		 * is_router flag.  Otherwise, if the entry is newly created,
1911 		 * clear the flag.  [RFC 2461, sec 8.3]
1912 		 */
1913 		if (code == ND_REDIRECT_ROUTER)
1914 			ln_router = 1;
1915 		else {
1916 			if (is_new)				/* (6-7) */
1917 				ln_router = 0;
1918 		}
1919 		break;
1920 	case ND_ROUTER_SOLICIT:
1921 		/*
1922 		 * is_router flag must always be cleared.
1923 		 */
1924 		ln_router = 0;
1925 		break;
1926 	case ND_ROUTER_ADVERT:
1927 		/*
1928 		 * Mark an entry with lladdr as a router.
1929 		 */
1930 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1931 		    (is_new && new_addr)) {			/* (7) */
1932 			ln_router = 1;
1933 		}
1934 		break;
1935 	}
1936 
1937 	return (ln_router);
1938 }
1939 
1940 /*
1941  * Create neighbor cache entry and cache link-layer address,
1942  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1943  *
1944  * type - ICMP6 type
1945  * code - type dependent information
1946  *
1947  */
1948 void
1949 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1950     int lladdrlen, int type, int code)
1951 {
1952 	struct llentry *ln = NULL, *ln_tmp;
1953 	int is_newentry;
1954 	int do_update;
1955 	int olladdr;
1956 	int llchange;
1957 	int flags;
1958 	uint16_t router = 0;
1959 	struct sockaddr_in6 sin6;
1960 	struct mbuf *chain = NULL;
1961 	u_char linkhdr[LLE_MAX_LINKHDR];
1962 	size_t linkhdrsize;
1963 	int lladdr_off;
1964 
1965 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1966 
1967 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1968 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1969 
1970 	/* nothing must be updated for unspecified address */
1971 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1972 		return;
1973 
1974 	/*
1975 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1976 	 * the caller.
1977 	 *
1978 	 * XXX If the link does not have link-layer adderss, what should
1979 	 * we do? (ifp->if_addrlen == 0)
1980 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1981 	 * description on it in NS section (RFC 2461 7.2.3).
1982 	 */
1983 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1984 	IF_AFDATA_RLOCK(ifp);
1985 	ln = nd6_lookup(from, flags, ifp);
1986 	IF_AFDATA_RUNLOCK(ifp);
1987 	is_newentry = 0;
1988 	if (ln == NULL) {
1989 		flags |= LLE_EXCLUSIVE;
1990 		ln = nd6_alloc(from, 0, ifp);
1991 		if (ln == NULL)
1992 			return;
1993 
1994 		/*
1995 		 * Since we already know all the data for the new entry,
1996 		 * fill it before insertion.
1997 		 */
1998 		if (lladdr != NULL) {
1999 			linkhdrsize = sizeof(linkhdr);
2000 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2001 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2002 				return;
2003 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2004 			    lladdr_off);
2005 		}
2006 
2007 		IF_AFDATA_WLOCK(ifp);
2008 		LLE_WLOCK(ln);
2009 		/* Prefer any existing lle over newly-created one */
2010 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
2011 		if (ln_tmp == NULL)
2012 			lltable_link_entry(LLTABLE6(ifp), ln);
2013 		IF_AFDATA_WUNLOCK(ifp);
2014 		if (ln_tmp == NULL) {
2015 			/* No existing lle, mark as new entry (6,7) */
2016 			is_newentry = 1;
2017 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2018 			if (lladdr != NULL)	/* (7) */
2019 				EVENTHANDLER_INVOKE(lle_event, ln,
2020 				    LLENTRY_RESOLVED);
2021 		} else {
2022 			lltable_free_entry(LLTABLE6(ifp), ln);
2023 			ln = ln_tmp;
2024 			ln_tmp = NULL;
2025 		}
2026 	}
2027 	/* do nothing if static ndp is set */
2028 	if ((ln->la_flags & LLE_STATIC)) {
2029 		if (flags & LLE_EXCLUSIVE)
2030 			LLE_WUNLOCK(ln);
2031 		else
2032 			LLE_RUNLOCK(ln);
2033 		return;
2034 	}
2035 
2036 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2037 	if (olladdr && lladdr) {
2038 		llchange = bcmp(lladdr, ln->ll_addr,
2039 		    ifp->if_addrlen);
2040 	} else if (!olladdr && lladdr)
2041 		llchange = 1;
2042 	else
2043 		llchange = 0;
2044 
2045 	/*
2046 	 * newentry olladdr  lladdr  llchange	(*=record)
2047 	 *	0	n	n	--	(1)
2048 	 *	0	y	n	--	(2)
2049 	 *	0	n	y	y	(3) * STALE
2050 	 *	0	y	y	n	(4) *
2051 	 *	0	y	y	y	(5) * STALE
2052 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2053 	 *	1	--	y	--	(7) * STALE
2054 	 */
2055 
2056 	do_update = 0;
2057 	if (is_newentry == 0 && llchange != 0) {
2058 		do_update = 1;	/* (3,5) */
2059 
2060 		/*
2061 		 * Record source link-layer address
2062 		 * XXX is it dependent to ifp->if_type?
2063 		 */
2064 		linkhdrsize = sizeof(linkhdr);
2065 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2066 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2067 			return;
2068 
2069 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2070 		    lladdr_off) == 0) {
2071 			/* Entry was deleted */
2072 			return;
2073 		}
2074 
2075 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2076 
2077 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2078 
2079 		if (ln->la_hold != NULL)
2080 			nd6_grab_holdchain(ln, &chain, &sin6);
2081 	}
2082 
2083 	/* Calculates new router status */
2084 	router = nd6_is_router(type, code, is_newentry, olladdr,
2085 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2086 
2087 	ln->ln_router = router;
2088 	/* Mark non-router redirects with special flag */
2089 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2090 		ln->la_flags |= LLE_REDIRECT;
2091 
2092 	if (flags & LLE_EXCLUSIVE)
2093 		LLE_WUNLOCK(ln);
2094 	else
2095 		LLE_RUNLOCK(ln);
2096 
2097 	if (chain != NULL)
2098 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
2099 
2100 	/*
2101 	 * When the link-layer address of a router changes, select the
2102 	 * best router again.  In particular, when the neighbor entry is newly
2103 	 * created, it might affect the selection policy.
2104 	 * Question: can we restrict the first condition to the "is_newentry"
2105 	 * case?
2106 	 * XXX: when we hear an RA from a new router with the link-layer
2107 	 * address option, defrouter_select() is called twice, since
2108 	 * defrtrlist_update called the function as well.  However, I believe
2109 	 * we can compromise the overhead, since it only happens the first
2110 	 * time.
2111 	 * XXX: although defrouter_select() should not have a bad effect
2112 	 * for those are not autoconfigured hosts, we explicitly avoid such
2113 	 * cases for safety.
2114 	 */
2115 	if ((do_update || is_newentry) && router &&
2116 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2117 		/*
2118 		 * guaranteed recursion
2119 		 */
2120 		defrouter_select();
2121 	}
2122 }
2123 
2124 static void
2125 nd6_slowtimo(void *arg)
2126 {
2127 	CURVNET_SET((struct vnet *) arg);
2128 	struct nd_ifinfo *nd6if;
2129 	struct ifnet *ifp;
2130 
2131 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2132 	    nd6_slowtimo, curvnet);
2133 	IFNET_RLOCK_NOSLEEP();
2134 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2135 		if (ifp->if_afdata[AF_INET6] == NULL)
2136 			continue;
2137 		nd6if = ND_IFINFO(ifp);
2138 		if (nd6if->basereachable && /* already initialized */
2139 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2140 			/*
2141 			 * Since reachable time rarely changes by router
2142 			 * advertisements, we SHOULD insure that a new random
2143 			 * value gets recomputed at least once every few hours.
2144 			 * (RFC 2461, 6.3.4)
2145 			 */
2146 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2147 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2148 		}
2149 	}
2150 	IFNET_RUNLOCK_NOSLEEP();
2151 	CURVNET_RESTORE();
2152 }
2153 
2154 void
2155 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2156     struct sockaddr_in6 *sin6)
2157 {
2158 
2159 	LLE_WLOCK_ASSERT(ln);
2160 
2161 	*chain = ln->la_hold;
2162 	ln->la_hold = NULL;
2163 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2164 
2165 	if (ln->ln_state == ND6_LLINFO_STALE) {
2166 
2167 		/*
2168 		 * The first time we send a packet to a
2169 		 * neighbor whose entry is STALE, we have
2170 		 * to change the state to DELAY and a sets
2171 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2172 		 * seconds to ensure do neighbor unreachability
2173 		 * detection on expiration.
2174 		 * (RFC 2461 7.3.3)
2175 		 */
2176 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2177 	}
2178 }
2179 
2180 int
2181 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2182     struct sockaddr_in6 *dst, struct route *ro)
2183 {
2184 	int error;
2185 	int ip6len;
2186 	struct ip6_hdr *ip6;
2187 	struct m_tag *mtag;
2188 
2189 #ifdef MAC
2190 	mac_netinet6_nd6_send(ifp, m);
2191 #endif
2192 
2193 	/*
2194 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2195 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2196 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2197 	 * to be diverted to user space.  When re-injected into the kernel,
2198 	 * send_output() will directly dispatch them to the outgoing interface.
2199 	 */
2200 	if (send_sendso_input_hook != NULL) {
2201 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2202 		if (mtag != NULL) {
2203 			ip6 = mtod(m, struct ip6_hdr *);
2204 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2205 			/* Use the SEND socket */
2206 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2207 			    ip6len);
2208 			/* -1 == no app on SEND socket */
2209 			if (error == 0 || error != -1)
2210 			    return (error);
2211 		}
2212 	}
2213 
2214 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2215 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2216 	    mtod(m, struct ip6_hdr *));
2217 
2218 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2219 		origifp = ifp;
2220 
2221 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2222 	return (error);
2223 }
2224 
2225 /*
2226  * Lookup link headerfor @sa_dst address. Stores found
2227  * data in @desten buffer. Copy of lle ln_flags can be also
2228  * saved in @pflags if @pflags is non-NULL.
2229  *
2230  * If destination LLE does not exists or lle state modification
2231  * is required, call "slow" version.
2232  *
2233  * Return values:
2234  * - 0 on success (address copied to buffer).
2235  * - EWOULDBLOCK (no local error, but address is still unresolved)
2236  * - other errors (alloc failure, etc)
2237  */
2238 int
2239 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2240     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2241     struct llentry **plle)
2242 {
2243 	struct llentry *ln = NULL;
2244 	const struct sockaddr_in6 *dst6;
2245 
2246 	if (pflags != NULL)
2247 		*pflags = 0;
2248 
2249 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2250 
2251 	/* discard the packet if IPv6 operation is disabled on the interface */
2252 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2253 		m_freem(m);
2254 		return (ENETDOWN); /* better error? */
2255 	}
2256 
2257 	if (m != NULL && m->m_flags & M_MCAST) {
2258 		switch (ifp->if_type) {
2259 		case IFT_ETHER:
2260 		case IFT_FDDI:
2261 		case IFT_L2VLAN:
2262 		case IFT_IEEE80211:
2263 		case IFT_BRIDGE:
2264 		case IFT_ISO88025:
2265 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2266 						 desten);
2267 			return (0);
2268 		default:
2269 			m_freem(m);
2270 			return (EAFNOSUPPORT);
2271 		}
2272 	}
2273 
2274 	IF_AFDATA_RLOCK(ifp);
2275 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2276 	    ifp);
2277 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2278 		/* Entry found, let's copy lle info */
2279 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2280 		if (pflags != NULL)
2281 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2282 		/* Check if we have feedback request from nd6 timer */
2283 		if (ln->r_skip_req != 0) {
2284 			LLE_REQ_LOCK(ln);
2285 			ln->r_skip_req = 0; /* Notify that entry was used */
2286 			ln->lle_hittime = time_uptime;
2287 			LLE_REQ_UNLOCK(ln);
2288 		}
2289 		if (plle) {
2290 			LLE_ADDREF(ln);
2291 			*plle = ln;
2292 			LLE_WUNLOCK(ln);
2293 		}
2294 		IF_AFDATA_RUNLOCK(ifp);
2295 		return (0);
2296 	} else if (plle && ln)
2297 		LLE_WUNLOCK(ln);
2298 	IF_AFDATA_RUNLOCK(ifp);
2299 
2300 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2301 }
2302 
2303 
2304 /*
2305  * Do L2 address resolution for @sa_dst address. Stores found
2306  * address in @desten buffer. Copy of lle ln_flags can be also
2307  * saved in @pflags if @pflags is non-NULL.
2308  *
2309  * Heavy version.
2310  * Function assume that destination LLE does not exist,
2311  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2312  *
2313  * Set noinline to be dtrace-friendly
2314  */
2315 static __noinline int
2316 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2317     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2318     struct llentry **plle)
2319 {
2320 	struct llentry *lle = NULL, *lle_tmp;
2321 	struct in6_addr *psrc, src;
2322 	int send_ns, ll_len;
2323 	char *lladdr;
2324 
2325 	/*
2326 	 * Address resolution or Neighbor Unreachability Detection
2327 	 * for the next hop.
2328 	 * At this point, the destination of the packet must be a unicast
2329 	 * or an anycast address(i.e. not a multicast).
2330 	 */
2331 	if (lle == NULL) {
2332 		IF_AFDATA_RLOCK(ifp);
2333 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2334 		IF_AFDATA_RUNLOCK(ifp);
2335 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2336 			/*
2337 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2338 			 * the condition below is not very efficient.  But we believe
2339 			 * it is tolerable, because this should be a rare case.
2340 			 */
2341 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2342 			if (lle == NULL) {
2343 				char ip6buf[INET6_ADDRSTRLEN];
2344 				log(LOG_DEBUG,
2345 				    "nd6_output: can't allocate llinfo for %s "
2346 				    "(ln=%p)\n",
2347 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2348 				m_freem(m);
2349 				return (ENOBUFS);
2350 			}
2351 
2352 			IF_AFDATA_WLOCK(ifp);
2353 			LLE_WLOCK(lle);
2354 			/* Prefer any existing entry over newly-created one */
2355 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2356 			if (lle_tmp == NULL)
2357 				lltable_link_entry(LLTABLE6(ifp), lle);
2358 			IF_AFDATA_WUNLOCK(ifp);
2359 			if (lle_tmp != NULL) {
2360 				lltable_free_entry(LLTABLE6(ifp), lle);
2361 				lle = lle_tmp;
2362 				lle_tmp = NULL;
2363 			}
2364 		}
2365 	}
2366 	if (lle == NULL) {
2367 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2368 			m_freem(m);
2369 			return (ENOBUFS);
2370 		}
2371 
2372 		if (m != NULL)
2373 			m_freem(m);
2374 		return (ENOBUFS);
2375 	}
2376 
2377 	LLE_WLOCK_ASSERT(lle);
2378 
2379 	/*
2380 	 * The first time we send a packet to a neighbor whose entry is
2381 	 * STALE, we have to change the state to DELAY and a sets a timer to
2382 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2383 	 * neighbor unreachability detection on expiration.
2384 	 * (RFC 2461 7.3.3)
2385 	 */
2386 	if (lle->ln_state == ND6_LLINFO_STALE)
2387 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2388 
2389 	/*
2390 	 * If the neighbor cache entry has a state other than INCOMPLETE
2391 	 * (i.e. its link-layer address is already resolved), just
2392 	 * send the packet.
2393 	 */
2394 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2395 		if (flags & LLE_ADDRONLY) {
2396 			lladdr = lle->ll_addr;
2397 			ll_len = ifp->if_addrlen;
2398 		} else {
2399 			lladdr = lle->r_linkdata;
2400 			ll_len = lle->r_hdrlen;
2401 		}
2402 		bcopy(lladdr, desten, ll_len);
2403 		if (pflags != NULL)
2404 			*pflags = lle->la_flags;
2405 		if (plle) {
2406 			LLE_ADDREF(lle);
2407 			*plle = lle;
2408 		}
2409 		LLE_WUNLOCK(lle);
2410 		return (0);
2411 	}
2412 
2413 	/*
2414 	 * There is a neighbor cache entry, but no ethernet address
2415 	 * response yet.  Append this latest packet to the end of the
2416 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2417 	 * the oldest packet in the queue will be removed.
2418 	 */
2419 
2420 	if (lle->la_hold != NULL) {
2421 		struct mbuf *m_hold;
2422 		int i;
2423 
2424 		i = 0;
2425 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2426 			i++;
2427 			if (m_hold->m_nextpkt == NULL) {
2428 				m_hold->m_nextpkt = m;
2429 				break;
2430 			}
2431 		}
2432 		while (i >= V_nd6_maxqueuelen) {
2433 			m_hold = lle->la_hold;
2434 			lle->la_hold = lle->la_hold->m_nextpkt;
2435 			m_freem(m_hold);
2436 			i--;
2437 		}
2438 	} else {
2439 		lle->la_hold = m;
2440 	}
2441 
2442 	/*
2443 	 * If there has been no NS for the neighbor after entering the
2444 	 * INCOMPLETE state, send the first solicitation.
2445 	 * Note that for newly-created lle la_asked will be 0,
2446 	 * so we will transition from ND6_LLINFO_NOSTATE to
2447 	 * ND6_LLINFO_INCOMPLETE state here.
2448 	 */
2449 	psrc = NULL;
2450 	send_ns = 0;
2451 	if (lle->la_asked == 0) {
2452 		lle->la_asked++;
2453 		send_ns = 1;
2454 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2455 
2456 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2457 	}
2458 	LLE_WUNLOCK(lle);
2459 	if (send_ns != 0)
2460 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2461 
2462 	return (EWOULDBLOCK);
2463 }
2464 
2465 /*
2466  * Do L2 address resolution for @sa_dst address. Stores found
2467  * address in @desten buffer. Copy of lle ln_flags can be also
2468  * saved in @pflags if @pflags is non-NULL.
2469  *
2470  * Return values:
2471  * - 0 on success (address copied to buffer).
2472  * - EWOULDBLOCK (no local error, but address is still unresolved)
2473  * - other errors (alloc failure, etc)
2474  */
2475 int
2476 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2477     char *desten, uint32_t *pflags)
2478 {
2479 	int error;
2480 
2481 	flags |= LLE_ADDRONLY;
2482 	error = nd6_resolve_slow(ifp, flags, NULL,
2483 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2484 	return (error);
2485 }
2486 
2487 int
2488 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2489     struct sockaddr_in6 *dst)
2490 {
2491 	struct mbuf *m, *m_head;
2492 	struct ifnet *outifp;
2493 	int error = 0;
2494 
2495 	m_head = chain;
2496 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2497 		outifp = origifp;
2498 	else
2499 		outifp = ifp;
2500 
2501 	while (m_head) {
2502 		m = m_head;
2503 		m_head = m_head->m_nextpkt;
2504 		error = nd6_output_ifp(ifp, origifp, m, dst, NULL);
2505 	}
2506 
2507 	/*
2508 	 * XXX
2509 	 * note that intermediate errors are blindly ignored
2510 	 */
2511 	return (error);
2512 }
2513 
2514 static int
2515 nd6_need_cache(struct ifnet *ifp)
2516 {
2517 	/*
2518 	 * XXX: we currently do not make neighbor cache on any interface
2519 	 * other than ARCnet, Ethernet, FDDI and GIF.
2520 	 *
2521 	 * RFC2893 says:
2522 	 * - unidirectional tunnels needs no ND
2523 	 */
2524 	switch (ifp->if_type) {
2525 	case IFT_ARCNET:
2526 	case IFT_ETHER:
2527 	case IFT_FDDI:
2528 	case IFT_IEEE1394:
2529 	case IFT_L2VLAN:
2530 	case IFT_IEEE80211:
2531 	case IFT_INFINIBAND:
2532 	case IFT_BRIDGE:
2533 	case IFT_PROPVIRTUAL:
2534 		return (1);
2535 	default:
2536 		return (0);
2537 	}
2538 }
2539 
2540 /*
2541  * Add pernament ND6 link-layer record for given
2542  * interface address.
2543  *
2544  * Very similar to IPv4 arp_ifinit(), but:
2545  * 1) IPv6 DAD is performed in different place
2546  * 2) It is called by IPv6 protocol stack in contrast to
2547  * arp_ifinit() which is typically called in SIOCSIFADDR
2548  * driver ioctl handler.
2549  *
2550  */
2551 int
2552 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2553 {
2554 	struct ifnet *ifp;
2555 	struct llentry *ln, *ln_tmp;
2556 	struct sockaddr *dst;
2557 
2558 	ifp = ia->ia_ifa.ifa_ifp;
2559 	if (nd6_need_cache(ifp) == 0)
2560 		return (0);
2561 
2562 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2563 	dst = (struct sockaddr *)&ia->ia_addr;
2564 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2565 	if (ln == NULL)
2566 		return (ENOBUFS);
2567 
2568 	IF_AFDATA_WLOCK(ifp);
2569 	LLE_WLOCK(ln);
2570 	/* Unlink any entry if exists */
2571 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2572 	if (ln_tmp != NULL)
2573 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2574 	lltable_link_entry(LLTABLE6(ifp), ln);
2575 	IF_AFDATA_WUNLOCK(ifp);
2576 
2577 	if (ln_tmp != NULL)
2578 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2579 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2580 
2581 	LLE_WUNLOCK(ln);
2582 	if (ln_tmp != NULL)
2583 		llentry_free(ln_tmp);
2584 
2585 	return (0);
2586 }
2587 
2588 /*
2589  * Removes either all lle entries for given @ia, or lle
2590  * corresponding to @ia address.
2591  */
2592 void
2593 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2594 {
2595 	struct sockaddr_in6 mask, addr;
2596 	struct sockaddr *saddr, *smask;
2597 	struct ifnet *ifp;
2598 
2599 	ifp = ia->ia_ifa.ifa_ifp;
2600 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2601 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2602 	saddr = (struct sockaddr *)&addr;
2603 	smask = (struct sockaddr *)&mask;
2604 
2605 	if (all != 0)
2606 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2607 	else
2608 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2609 }
2610 
2611 static void
2612 clear_llinfo_pqueue(struct llentry *ln)
2613 {
2614 	struct mbuf *m_hold, *m_hold_next;
2615 
2616 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2617 		m_hold_next = m_hold->m_nextpkt;
2618 		m_freem(m_hold);
2619 	}
2620 
2621 	ln->la_hold = NULL;
2622 }
2623 
2624 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2625 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2626 
2627 SYSCTL_DECL(_net_inet6_icmp6);
2628 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2629 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2630 	NULL, 0, nd6_sysctl_drlist, "S,in6_defrouter",
2631 	"NDP default router list");
2632 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2633 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2634 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2635 	"NDP prefix list");
2636 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2637 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2638 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2639 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2640 
2641 static int
2642 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2643 {
2644 	struct in6_defrouter d;
2645 	struct nd_defrouter *dr;
2646 	int error;
2647 
2648 	if (req->newptr != NULL)
2649 		return (EPERM);
2650 
2651 	error = sysctl_wire_old_buffer(req, 0);
2652 	if (error != 0)
2653 		return (error);
2654 
2655 	bzero(&d, sizeof(d));
2656 	d.rtaddr.sin6_family = AF_INET6;
2657 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2658 
2659 	ND6_RLOCK();
2660 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2661 		d.rtaddr.sin6_addr = dr->rtaddr;
2662 		error = sa6_recoverscope(&d.rtaddr);
2663 		if (error != 0)
2664 			break;
2665 		d.flags = dr->raflags;
2666 		d.rtlifetime = dr->rtlifetime;
2667 		d.expire = dr->expire + (time_second - time_uptime);
2668 		d.if_index = dr->ifp->if_index;
2669 		error = SYSCTL_OUT(req, &d, sizeof(d));
2670 		if (error != 0)
2671 			break;
2672 	}
2673 	ND6_RUNLOCK();
2674 	return (error);
2675 }
2676 
2677 static int
2678 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2679 {
2680 	struct in6_prefix p;
2681 	struct sockaddr_in6 s6;
2682 	struct nd_prefix *pr;
2683 	struct nd_pfxrouter *pfr;
2684 	time_t maxexpire;
2685 	int error;
2686 	char ip6buf[INET6_ADDRSTRLEN];
2687 
2688 	if (req->newptr)
2689 		return (EPERM);
2690 
2691 	error = sysctl_wire_old_buffer(req, 0);
2692 	if (error != 0)
2693 		return (error);
2694 
2695 	bzero(&p, sizeof(p));
2696 	p.origin = PR_ORIG_RA;
2697 	bzero(&s6, sizeof(s6));
2698 	s6.sin6_family = AF_INET6;
2699 	s6.sin6_len = sizeof(s6);
2700 
2701 	ND6_RLOCK();
2702 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2703 		p.prefix = pr->ndpr_prefix;
2704 		if (sa6_recoverscope(&p.prefix)) {
2705 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2706 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2707 			/* XXX: press on... */
2708 		}
2709 		p.raflags = pr->ndpr_raf;
2710 		p.prefixlen = pr->ndpr_plen;
2711 		p.vltime = pr->ndpr_vltime;
2712 		p.pltime = pr->ndpr_pltime;
2713 		p.if_index = pr->ndpr_ifp->if_index;
2714 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2715 			p.expire = 0;
2716 		else {
2717 			/* XXX: we assume time_t is signed. */
2718 			maxexpire = (-1) &
2719 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2720 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2721 				p.expire = pr->ndpr_lastupdate +
2722 				    pr->ndpr_vltime +
2723 				    (time_second - time_uptime);
2724 			else
2725 				p.expire = maxexpire;
2726 		}
2727 		p.refcnt = pr->ndpr_addrcnt;
2728 		p.flags = pr->ndpr_stateflags;
2729 		p.advrtrs = 0;
2730 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2731 			p.advrtrs++;
2732 		error = SYSCTL_OUT(req, &p, sizeof(p));
2733 		if (error != 0)
2734 			break;
2735 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2736 			s6.sin6_addr = pfr->router->rtaddr;
2737 			if (sa6_recoverscope(&s6))
2738 				log(LOG_ERR,
2739 				    "scope error in prefix list (%s)\n",
2740 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2741 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2742 			if (error != 0)
2743 				goto out;
2744 		}
2745 	}
2746 out:
2747 	ND6_RUNLOCK();
2748 	return (error);
2749 }
2750