xref: /freebsd/sys/netinet6/nd6.c (revision c94c8223bd444fa38ded3797060110c590f422f4)
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/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/time.h>
46 #include <sys/kernel.h>
47 #include <sys/protosw.h>
48 #include <sys/errno.h>
49 #include <sys/syslog.h>
50 #include <sys/lock.h>
51 #include <sys/rwlock.h>
52 #include <sys/queue.h>
53 #include <sys/sdt.h>
54 #include <sys/sysctl.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_arc.h>
59 #include <net/if_dl.h>
60 #include <net/if_types.h>
61 #include <net/iso88025.h>
62 #include <net/fddi.h>
63 #include <net/route.h>
64 #include <net/vnet.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/in_kdtrace.h>
68 #include <net/if_llatbl.h>
69 #include <netinet/if_ether.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/in6_ifattach.h>
76 #include <netinet/icmp6.h>
77 #include <netinet6/send.h>
78 
79 #include <sys/limits.h>
80 
81 #include <security/mac/mac_framework.h>
82 
83 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
84 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
85 
86 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
87 
88 /* timer values */
89 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
90 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
91 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
92 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
93 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
94 					 * local traffic */
95 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
96 					 * collection timer */
97 
98 /* preventing too many loops in ND option parsing */
99 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
100 
101 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
102 					 * layer hints */
103 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
104 					 * ND entries */
105 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
106 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
107 
108 #ifdef ND6_DEBUG
109 VNET_DEFINE(int, nd6_debug) = 1;
110 #else
111 VNET_DEFINE(int, nd6_debug) = 0;
112 #endif
113 
114 static eventhandler_tag lle_event_eh, iflladdr_event_eh;
115 
116 VNET_DEFINE(struct nd_drhead, nd_defrouter);
117 VNET_DEFINE(struct nd_prhead, nd_prefix);
118 VNET_DEFINE(struct rwlock, nd6_lock);
119 
120 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
121 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
122 
123 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
124 
125 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
126 	struct ifnet *);
127 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
128 static void nd6_slowtimo(void *);
129 static int regen_tmpaddr(struct in6_ifaddr *);
130 static void nd6_free(struct llentry *, int);
131 static void nd6_free_redirect(const struct llentry *);
132 static void nd6_llinfo_timer(void *);
133 static void nd6_llinfo_settimer_locked(struct llentry *, long);
134 static void clear_llinfo_pqueue(struct llentry *);
135 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
136 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
137     const struct sockaddr_in6 *, u_char *, uint32_t *);
138 static int nd6_need_cache(struct ifnet *);
139 
140 
141 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
142 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
143 
144 VNET_DEFINE(struct callout, nd6_timer_ch);
145 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
146 
147 static void
148 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
149 {
150 	struct rt_addrinfo rtinfo;
151 	struct sockaddr_in6 dst;
152 	struct sockaddr_dl gw;
153 	struct ifnet *ifp;
154 	int type;
155 
156 	LLE_WLOCK_ASSERT(lle);
157 
158 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
159 		return;
160 
161 	switch (evt) {
162 	case LLENTRY_RESOLVED:
163 		type = RTM_ADD;
164 		KASSERT(lle->la_flags & LLE_VALID,
165 		    ("%s: %p resolved but not valid?", __func__, lle));
166 		break;
167 	case LLENTRY_EXPIRED:
168 		type = RTM_DELETE;
169 		break;
170 	default:
171 		return;
172 	}
173 
174 	ifp = lltable_get_ifp(lle->lle_tbl);
175 
176 	bzero(&dst, sizeof(dst));
177 	bzero(&gw, sizeof(gw));
178 	bzero(&rtinfo, sizeof(rtinfo));
179 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
180 	dst.sin6_scope_id = in6_getscopezone(ifp,
181 	    in6_addrscope(&dst.sin6_addr));
182 	gw.sdl_len = sizeof(struct sockaddr_dl);
183 	gw.sdl_family = AF_LINK;
184 	gw.sdl_alen = ifp->if_addrlen;
185 	gw.sdl_index = ifp->if_index;
186 	gw.sdl_type = ifp->if_type;
187 	if (evt == LLENTRY_RESOLVED)
188 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
189 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
190 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
191 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
192 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
193 	    type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB);
194 }
195 
196 /*
197  * A handler for interface link layer address change event.
198  */
199 static void
200 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
201 {
202 
203 	lltable_update_ifaddr(LLTABLE6(ifp));
204 }
205 
206 void
207 nd6_init(void)
208 {
209 
210 	rw_init(&V_nd6_lock, "nd6");
211 
212 	LIST_INIT(&V_nd_prefix);
213 
214 	/* initialization of the default router list */
215 	TAILQ_INIT(&V_nd_defrouter);
216 
217 	/* Start timers. */
218 	callout_init(&V_nd6_slowtimo_ch, 0);
219 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
220 	    nd6_slowtimo, curvnet);
221 
222 	callout_init(&V_nd6_timer_ch, 0);
223 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
224 
225 	nd6_dad_init();
226 	if (IS_DEFAULT_VNET(curvnet)) {
227 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
228 		    NULL, EVENTHANDLER_PRI_ANY);
229 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
230 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
231 	}
232 }
233 
234 #ifdef VIMAGE
235 void
236 nd6_destroy()
237 {
238 
239 	callout_drain(&V_nd6_slowtimo_ch);
240 	callout_drain(&V_nd6_timer_ch);
241 	if (IS_DEFAULT_VNET(curvnet)) {
242 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
243 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
244 	}
245 	rw_destroy(&V_nd6_lock);
246 }
247 #endif
248 
249 struct nd_ifinfo *
250 nd6_ifattach(struct ifnet *ifp)
251 {
252 	struct nd_ifinfo *nd;
253 
254 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
255 	nd->initialized = 1;
256 
257 	nd->chlim = IPV6_DEFHLIM;
258 	nd->basereachable = REACHABLE_TIME;
259 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
260 	nd->retrans = RETRANS_TIMER;
261 
262 	nd->flags = ND6_IFF_PERFORMNUD;
263 
264 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
265 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
266 	 * default regardless of the V_ip6_auto_linklocal configuration to
267 	 * give a reasonable default behavior.
268 	 */
269 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
270 	    (ifp->if_flags & IFF_LOOPBACK))
271 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
272 	/*
273 	 * A loopback interface does not need to accept RTADV.
274 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
275 	 * default regardless of the V_ip6_accept_rtadv configuration to
276 	 * prevent the interface from accepting RA messages arrived
277 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
278 	 */
279 	if (V_ip6_accept_rtadv &&
280 	    !(ifp->if_flags & IFF_LOOPBACK) &&
281 	    (ifp->if_type != IFT_BRIDGE))
282 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
283 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
284 		nd->flags |= ND6_IFF_NO_RADR;
285 
286 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
287 	nd6_setmtu0(ifp, nd);
288 
289 	return nd;
290 }
291 
292 void
293 nd6_ifdetach(struct nd_ifinfo *nd)
294 {
295 
296 	free(nd, M_IP6NDP);
297 }
298 
299 /*
300  * Reset ND level link MTU. This function is called when the physical MTU
301  * changes, which means we might have to adjust the ND level MTU.
302  */
303 void
304 nd6_setmtu(struct ifnet *ifp)
305 {
306 	if (ifp->if_afdata[AF_INET6] == NULL)
307 		return;
308 
309 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
310 }
311 
312 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
313 void
314 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
315 {
316 	u_int32_t omaxmtu;
317 
318 	omaxmtu = ndi->maxmtu;
319 
320 	switch (ifp->if_type) {
321 	case IFT_ARCNET:
322 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
323 		break;
324 	case IFT_FDDI:
325 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
326 		break;
327 	case IFT_ISO88025:
328 		 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu);
329 		 break;
330 	default:
331 		ndi->maxmtu = ifp->if_mtu;
332 		break;
333 	}
334 
335 	/*
336 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
337 	 * undesirable situation.  We thus notify the operator of the change
338 	 * explicitly.  The check for omaxmtu is necessary to restrict the
339 	 * log to the case of changing the MTU, not initializing it.
340 	 */
341 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
342 		log(LOG_NOTICE, "nd6_setmtu0: "
343 		    "new link MTU on %s (%lu) is too small for IPv6\n",
344 		    if_name(ifp), (unsigned long)ndi->maxmtu);
345 	}
346 
347 	if (ndi->maxmtu > V_in6_maxmtu)
348 		in6_setmaxmtu(); /* check all interfaces just in case */
349 
350 }
351 
352 void
353 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
354 {
355 
356 	bzero(ndopts, sizeof(*ndopts));
357 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
358 	ndopts->nd_opts_last
359 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
360 
361 	if (icmp6len == 0) {
362 		ndopts->nd_opts_done = 1;
363 		ndopts->nd_opts_search = NULL;
364 	}
365 }
366 
367 /*
368  * Take one ND option.
369  */
370 struct nd_opt_hdr *
371 nd6_option(union nd_opts *ndopts)
372 {
373 	struct nd_opt_hdr *nd_opt;
374 	int olen;
375 
376 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
377 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
378 	    __func__));
379 	if (ndopts->nd_opts_search == NULL)
380 		return NULL;
381 	if (ndopts->nd_opts_done)
382 		return NULL;
383 
384 	nd_opt = ndopts->nd_opts_search;
385 
386 	/* make sure nd_opt_len is inside the buffer */
387 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
388 		bzero(ndopts, sizeof(*ndopts));
389 		return NULL;
390 	}
391 
392 	olen = nd_opt->nd_opt_len << 3;
393 	if (olen == 0) {
394 		/*
395 		 * Message validation requires that all included
396 		 * options have a length that is greater than zero.
397 		 */
398 		bzero(ndopts, sizeof(*ndopts));
399 		return NULL;
400 	}
401 
402 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
403 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
404 		/* option overruns the end of buffer, invalid */
405 		bzero(ndopts, sizeof(*ndopts));
406 		return NULL;
407 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
408 		/* reached the end of options chain */
409 		ndopts->nd_opts_done = 1;
410 		ndopts->nd_opts_search = NULL;
411 	}
412 	return nd_opt;
413 }
414 
415 /*
416  * Parse multiple ND options.
417  * This function is much easier to use, for ND routines that do not need
418  * multiple options of the same type.
419  */
420 int
421 nd6_options(union nd_opts *ndopts)
422 {
423 	struct nd_opt_hdr *nd_opt;
424 	int i = 0;
425 
426 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
427 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
428 	    __func__));
429 	if (ndopts->nd_opts_search == NULL)
430 		return 0;
431 
432 	while (1) {
433 		nd_opt = nd6_option(ndopts);
434 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
435 			/*
436 			 * Message validation requires that all included
437 			 * options have a length that is greater than zero.
438 			 */
439 			ICMP6STAT_INC(icp6s_nd_badopt);
440 			bzero(ndopts, sizeof(*ndopts));
441 			return -1;
442 		}
443 
444 		if (nd_opt == NULL)
445 			goto skip1;
446 
447 		switch (nd_opt->nd_opt_type) {
448 		case ND_OPT_SOURCE_LINKADDR:
449 		case ND_OPT_TARGET_LINKADDR:
450 		case ND_OPT_MTU:
451 		case ND_OPT_REDIRECTED_HEADER:
452 		case ND_OPT_NONCE:
453 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
454 				nd6log((LOG_INFO,
455 				    "duplicated ND6 option found (type=%d)\n",
456 				    nd_opt->nd_opt_type));
457 				/* XXX bark? */
458 			} else {
459 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
460 					= nd_opt;
461 			}
462 			break;
463 		case ND_OPT_PREFIX_INFORMATION:
464 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
465 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
466 					= nd_opt;
467 			}
468 			ndopts->nd_opts_pi_end =
469 				(struct nd_opt_prefix_info *)nd_opt;
470 			break;
471 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
472 		case ND_OPT_RDNSS:	/* RFC 6106 */
473 		case ND_OPT_DNSSL:	/* RFC 6106 */
474 			/*
475 			 * Silently ignore options we know and do not care about
476 			 * in the kernel.
477 			 */
478 			break;
479 		default:
480 			/*
481 			 * Unknown options must be silently ignored,
482 			 * to accomodate future extension to the protocol.
483 			 */
484 			nd6log((LOG_DEBUG,
485 			    "nd6_options: unsupported option %d - "
486 			    "option ignored\n", nd_opt->nd_opt_type));
487 		}
488 
489 skip1:
490 		i++;
491 		if (i > V_nd6_maxndopt) {
492 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
493 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
494 			break;
495 		}
496 
497 		if (ndopts->nd_opts_done)
498 			break;
499 	}
500 
501 	return 0;
502 }
503 
504 /*
505  * ND6 timer routine to handle ND6 entries
506  */
507 static void
508 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
509 {
510 	int canceled;
511 
512 	LLE_WLOCK_ASSERT(ln);
513 
514 	if (tick < 0) {
515 		ln->la_expire = 0;
516 		ln->ln_ntick = 0;
517 		canceled = callout_stop(&ln->lle_timer);
518 	} else {
519 		ln->la_expire = time_uptime + tick / hz;
520 		LLE_ADDREF(ln);
521 		if (tick > INT_MAX) {
522 			ln->ln_ntick = tick - INT_MAX;
523 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
524 			    nd6_llinfo_timer, ln);
525 		} else {
526 			ln->ln_ntick = 0;
527 			canceled = callout_reset(&ln->lle_timer, tick,
528 			    nd6_llinfo_timer, ln);
529 		}
530 	}
531 	if (canceled > 0)
532 		LLE_REMREF(ln);
533 }
534 
535 /*
536  * Gets source address of the first packet in hold queue
537  * and stores it in @src.
538  * Returns pointer to @src (if hold queue is not empty) or NULL.
539  *
540  * Set noinline to be dtrace-friendly
541  */
542 static __noinline struct in6_addr *
543 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
544 {
545 	struct ip6_hdr hdr;
546 	struct mbuf *m;
547 
548 	if (ln->la_hold == NULL)
549 		return (NULL);
550 
551 	/*
552 	 * assume every packet in la_hold has the same IP header
553 	 */
554 	m = ln->la_hold;
555 	if (sizeof(hdr) > m->m_len)
556 		return (NULL);
557 
558 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
559 	*src = hdr.ip6_src;
560 
561 	return (src);
562 }
563 
564 /*
565  * Checks if we need to switch from STALE state.
566  *
567  * RFC 4861 requires switching from STALE to DELAY state
568  * on first packet matching entry, waiting V_nd6_delay and
569  * transition to PROBE state (if upper layer confirmation was
570  * not received).
571  *
572  * This code performs a bit differently:
573  * On packet hit we don't change state (but desired state
574  * can be guessed by control plane). However, after V_nd6_delay
575  * seconds code will transition to PROBE state (so DELAY state
576  * is kinda skipped in most situations).
577  *
578  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
579  * we perform the following upon entering STALE state:
580  *
581  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
582  * if packet was transmitted at the start of given interval, we
583  * would be able to switch to PROBE state in V_nd6_delay seconds
584  * as user expects.
585  *
586  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
587  * lle in STALE state (remaining timer value stored in lle_remtime).
588  *
589  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
590  * seconds ago.
591  *
592  * Returns non-zero value if the entry is still STALE (storing
593  * the next timer interval in @pdelay).
594  *
595  * Returns zero value if original timer expired or we need to switch to
596  * PROBE (store that in @do_switch variable).
597  */
598 static int
599 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
600 {
601 	int nd_delay, nd_gctimer, r_skip_req;
602 	time_t lle_hittime;
603 	long delay;
604 
605 	*do_switch = 0;
606 	nd_gctimer = V_nd6_gctimer;
607 	nd_delay = V_nd6_delay;
608 
609 	LLE_REQ_LOCK(lle);
610 	r_skip_req = lle->r_skip_req;
611 	lle_hittime = lle->lle_hittime;
612 	LLE_REQ_UNLOCK(lle);
613 
614 	if (r_skip_req > 0) {
615 
616 		/*
617 		 * Nonzero r_skip_req value was set upon entering
618 		 * STALE state. Since value was not changed, no
619 		 * packets were passed using this lle. Ask for
620 		 * timer reschedule and keep STALE state.
621 		 */
622 		delay = (long)(MIN(nd_gctimer, nd_delay));
623 		delay *= hz;
624 		if (lle->lle_remtime > delay)
625 			lle->lle_remtime -= delay;
626 		else {
627 			delay = lle->lle_remtime;
628 			lle->lle_remtime = 0;
629 		}
630 
631 		if (delay == 0) {
632 
633 			/*
634 			 * The original ng6_gctime timeout ended,
635 			 * no more rescheduling.
636 			 */
637 			return (0);
638 		}
639 
640 		*pdelay = delay;
641 		return (1);
642 	}
643 
644 	/*
645 	 * Packet received. Verify timestamp
646 	 */
647 	delay = (long)(time_uptime - lle_hittime);
648 	if (delay < nd_delay) {
649 
650 		/*
651 		 * V_nd6_delay still not passed since the first
652 		 * hit in STALE state.
653 		 * Reshedule timer and return.
654 		 */
655 		*pdelay = (long)(nd_delay - delay) * hz;
656 		return (1);
657 	}
658 
659 	/* Request switching to probe */
660 	*do_switch = 1;
661 	return (0);
662 }
663 
664 
665 /*
666  * Switch @lle state to new state optionally arming timers.
667  *
668  * Set noinline to be dtrace-friendly
669  */
670 __noinline void
671 nd6_llinfo_setstate(struct llentry *lle, int newstate)
672 {
673 	struct ifnet *ifp;
674 	int nd_gctimer, nd_delay;
675 	long delay, remtime;
676 
677 	delay = 0;
678 	remtime = 0;
679 
680 	switch (newstate) {
681 	case ND6_LLINFO_INCOMPLETE:
682 		ifp = lle->lle_tbl->llt_ifp;
683 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
684 		break;
685 	case ND6_LLINFO_REACHABLE:
686 		if (!ND6_LLINFO_PERMANENT(lle)) {
687 			ifp = lle->lle_tbl->llt_ifp;
688 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
689 		}
690 		break;
691 	case ND6_LLINFO_STALE:
692 
693 		/*
694 		 * Notify fast path that we want to know if any packet
695 		 * is transmitted by setting r_skip_req.
696 		 */
697 		LLE_REQ_LOCK(lle);
698 		lle->r_skip_req = 1;
699 		LLE_REQ_UNLOCK(lle);
700 		nd_delay = V_nd6_delay;
701 		nd_gctimer = V_nd6_gctimer;
702 
703 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
704 		remtime = (long)nd_gctimer * hz - delay;
705 		break;
706 	case ND6_LLINFO_DELAY:
707 		lle->la_asked = 0;
708 		delay = (long)V_nd6_delay * hz;
709 		break;
710 	}
711 
712 	if (delay > 0)
713 		nd6_llinfo_settimer_locked(lle, delay);
714 
715 	lle->lle_remtime = remtime;
716 	lle->ln_state = newstate;
717 }
718 
719 /*
720  * Timer-dependent part of nd state machine.
721  *
722  * Set noinline to be dtrace-friendly
723  */
724 static __noinline void
725 nd6_llinfo_timer(void *arg)
726 {
727 	struct llentry *ln;
728 	struct in6_addr *dst, *pdst, *psrc, src;
729 	struct ifnet *ifp;
730 	struct nd_ifinfo *ndi = NULL;
731 	int do_switch, send_ns;
732 	long delay;
733 
734 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
735 	ln = (struct llentry *)arg;
736 	LLE_WLOCK(ln);
737 	if (callout_pending(&ln->lle_timer)) {
738 		/*
739 		 * Here we are a bit odd here in the treatment of
740 		 * active/pending. If the pending bit is set, it got
741 		 * rescheduled before I ran. The active
742 		 * bit we ignore, since if it was stopped
743 		 * in ll_tablefree() and was currently running
744 		 * it would have return 0 so the code would
745 		 * not have deleted it since the callout could
746 		 * not be stopped so we want to go through
747 		 * with the delete here now. If the callout
748 		 * was restarted, the pending bit will be back on and
749 		 * we just want to bail since the callout_reset would
750 		 * return 1 and our reference would have been removed
751 		 * by nd6_llinfo_settimer_locked above since canceled
752 		 * would have been 1.
753 		 */
754 		LLE_WUNLOCK(ln);
755 		return;
756 	}
757 	ifp = ln->lle_tbl->llt_ifp;
758 	CURVNET_SET(ifp->if_vnet);
759 	ndi = ND_IFINFO(ifp);
760 	send_ns = 0;
761 	dst = &ln->r_l3addr.addr6;
762 	pdst = dst;
763 
764 	if (ln->ln_ntick > 0) {
765 		if (ln->ln_ntick > INT_MAX) {
766 			ln->ln_ntick -= INT_MAX;
767 			nd6_llinfo_settimer_locked(ln, INT_MAX);
768 		} else {
769 			ln->ln_ntick = 0;
770 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
771 		}
772 		goto done;
773 	}
774 
775 	if (ln->la_flags & LLE_STATIC) {
776 		goto done;
777 	}
778 
779 	if (ln->la_flags & LLE_DELETED) {
780 		nd6_free(ln, 0);
781 		ln = NULL;
782 		goto done;
783 	}
784 
785 	switch (ln->ln_state) {
786 	case ND6_LLINFO_INCOMPLETE:
787 		if (ln->la_asked < V_nd6_mmaxtries) {
788 			ln->la_asked++;
789 			send_ns = 1;
790 			/* Send NS to multicast address */
791 			pdst = NULL;
792 		} else {
793 			struct mbuf *m = ln->la_hold;
794 			if (m) {
795 				struct mbuf *m0;
796 
797 				/*
798 				 * assuming every packet in la_hold has the
799 				 * same IP header.  Send error after unlock.
800 				 */
801 				m0 = m->m_nextpkt;
802 				m->m_nextpkt = NULL;
803 				ln->la_hold = m0;
804 				clear_llinfo_pqueue(ln);
805 			}
806 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT);
807 			nd6_free(ln, 0);
808 			ln = NULL;
809 			if (m != NULL)
810 				icmp6_error2(m, ICMP6_DST_UNREACH,
811 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
812 		}
813 		break;
814 	case ND6_LLINFO_REACHABLE:
815 		if (!ND6_LLINFO_PERMANENT(ln))
816 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
817 		break;
818 
819 	case ND6_LLINFO_STALE:
820 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
821 
822 			/*
823 			 * No packet has used this entry and GC timeout
824 			 * has not been passed. Reshedule timer and
825 			 * return.
826 			 */
827 			nd6_llinfo_settimer_locked(ln, delay);
828 			break;
829 		}
830 
831 		if (do_switch == 0) {
832 
833 			/*
834 			 * GC timer has ended and entry hasn't been used.
835 			 * Run Garbage collector (RFC 4861, 5.3)
836 			 */
837 			if (!ND6_LLINFO_PERMANENT(ln)) {
838 				EVENTHANDLER_INVOKE(lle_event, ln,
839 				    LLENTRY_EXPIRED);
840 				nd6_free(ln, 1);
841 				ln = NULL;
842 			}
843 			break;
844 		}
845 
846 		/* Entry has been used AND delay timer has ended. */
847 
848 		/* FALLTHROUGH */
849 
850 	case ND6_LLINFO_DELAY:
851 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
852 			/* We need NUD */
853 			ln->la_asked = 1;
854 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
855 			send_ns = 1;
856 		} else
857 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
858 		break;
859 	case ND6_LLINFO_PROBE:
860 		if (ln->la_asked < V_nd6_umaxtries) {
861 			ln->la_asked++;
862 			send_ns = 1;
863 		} else {
864 			EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
865 			nd6_free(ln, 0);
866 			ln = NULL;
867 		}
868 		break;
869 	default:
870 		panic("%s: paths in a dark night can be confusing: %d",
871 		    __func__, ln->ln_state);
872 	}
873 done:
874 	if (send_ns != 0) {
875 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
876 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
877 		LLE_FREE_LOCKED(ln);
878 		ln = NULL;
879 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
880 	}
881 
882 	if (ln != NULL)
883 		LLE_FREE_LOCKED(ln);
884 	CURVNET_RESTORE();
885 }
886 
887 
888 /*
889  * ND6 timer routine to expire default route list and prefix list
890  */
891 void
892 nd6_timer(void *arg)
893 {
894 	CURVNET_SET((struct vnet *) arg);
895 	struct nd_drhead drq;
896 	struct nd_defrouter *dr, *ndr;
897 	struct nd_prefix *pr, *npr;
898 	struct in6_ifaddr *ia6, *nia6;
899 
900 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
901 	    nd6_timer, curvnet);
902 
903 	TAILQ_INIT(&drq);
904 
905 	/* expire default router list */
906 	ND6_WLOCK();
907 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
908 		if (dr->expire && dr->expire < time_uptime)
909 			defrouter_unlink(dr, &drq);
910 	ND6_WUNLOCK();
911 
912 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
913 		TAILQ_REMOVE(&drq, dr, dr_entry);
914 		defrouter_del(dr);
915 	}
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 	TAILQ_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 			if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
997 			    (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
998 				== 0 ||
999 			    (ND_IFINFO(ia6->ia_ifp)->flags &
1000 				ND6_IFF_IFDISABLED) != 0) {
1001 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1002 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1003 			}
1004 			/*
1005 			 * A new RA might have made a deprecated address
1006 			 * preferred.
1007 			 */
1008 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1009 		}
1010 	}
1011 
1012 	/* expire prefix list */
1013 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1014 		/*
1015 		 * check prefix lifetime.
1016 		 * since pltime is just for autoconf, pltime processing for
1017 		 * prefix is not necessary.
1018 		 */
1019 		if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME &&
1020 		    time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) {
1021 
1022 			/*
1023 			 * address expiration and prefix expiration are
1024 			 * separate.  NEVER perform in6_purgeaddr here.
1025 			 */
1026 			prelist_remove(pr);
1027 		}
1028 	}
1029 	CURVNET_RESTORE();
1030 }
1031 
1032 /*
1033  * ia6 - deprecated/invalidated temporary address
1034  */
1035 static int
1036 regen_tmpaddr(struct in6_ifaddr *ia6)
1037 {
1038 	struct ifaddr *ifa;
1039 	struct ifnet *ifp;
1040 	struct in6_ifaddr *public_ifa6 = NULL;
1041 
1042 	ifp = ia6->ia_ifa.ifa_ifp;
1043 	IF_ADDR_RLOCK(ifp);
1044 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1045 		struct in6_ifaddr *it6;
1046 
1047 		if (ifa->ifa_addr->sa_family != AF_INET6)
1048 			continue;
1049 
1050 		it6 = (struct in6_ifaddr *)ifa;
1051 
1052 		/* ignore no autoconf addresses. */
1053 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1054 			continue;
1055 
1056 		/* ignore autoconf addresses with different prefixes. */
1057 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1058 			continue;
1059 
1060 		/*
1061 		 * Now we are looking at an autoconf address with the same
1062 		 * prefix as ours.  If the address is temporary and is still
1063 		 * preferred, do not create another one.  It would be rare, but
1064 		 * could happen, for example, when we resume a laptop PC after
1065 		 * a long period.
1066 		 */
1067 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1068 		    !IFA6_IS_DEPRECATED(it6)) {
1069 			public_ifa6 = NULL;
1070 			break;
1071 		}
1072 
1073 		/*
1074 		 * This is a public autoconf address that has the same prefix
1075 		 * as ours.  If it is preferred, keep it.  We can't break the
1076 		 * loop here, because there may be a still-preferred temporary
1077 		 * address with the prefix.
1078 		 */
1079 		if (!IFA6_IS_DEPRECATED(it6))
1080 			public_ifa6 = it6;
1081 	}
1082 	if (public_ifa6 != NULL)
1083 		ifa_ref(&public_ifa6->ia_ifa);
1084 	IF_ADDR_RUNLOCK(ifp);
1085 
1086 	if (public_ifa6 != NULL) {
1087 		int e;
1088 
1089 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1090 			ifa_free(&public_ifa6->ia_ifa);
1091 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1092 			    " tmp addr,errno=%d\n", e);
1093 			return (-1);
1094 		}
1095 		ifa_free(&public_ifa6->ia_ifa);
1096 		return (0);
1097 	}
1098 
1099 	return (-1);
1100 }
1101 
1102 /*
1103  * Nuke neighbor cache/prefix/default router management table, right before
1104  * ifp goes away.
1105  */
1106 void
1107 nd6_purge(struct ifnet *ifp)
1108 {
1109 	struct nd_drhead drq;
1110 	struct nd_defrouter *dr, *ndr;
1111 	struct nd_prefix *pr, *npr;
1112 
1113 	TAILQ_INIT(&drq);
1114 
1115 	/*
1116 	 * Nuke default router list entries toward ifp.
1117 	 * We defer removal of default router list entries that is installed
1118 	 * in the routing table, in order to keep additional side effects as
1119 	 * small as possible.
1120 	 */
1121 	ND6_WLOCK();
1122 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1123 		if (dr->installed)
1124 			continue;
1125 		if (dr->ifp == ifp)
1126 			defrouter_unlink(dr, &drq);
1127 	}
1128 
1129 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1130 		if (!dr->installed)
1131 			continue;
1132 		if (dr->ifp == ifp)
1133 			defrouter_unlink(dr, &drq);
1134 	}
1135 	ND6_WUNLOCK();
1136 
1137 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1138 		TAILQ_REMOVE(&drq, dr, dr_entry);
1139 		defrouter_del(dr);
1140 	}
1141 
1142 	/* Nuke prefix list entries toward ifp */
1143 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1144 		if (pr->ndpr_ifp == ifp) {
1145 			/*
1146 			 * Because if_detach() does *not* release prefixes
1147 			 * while purging addresses the reference count will
1148 			 * still be above zero. We therefore reset it to
1149 			 * make sure that the prefix really gets purged.
1150 			 */
1151 			pr->ndpr_refcnt = 0;
1152 
1153 			/*
1154 			 * Previously, pr->ndpr_addr is removed as well,
1155 			 * but I strongly believe we don't have to do it.
1156 			 * nd6_purge() is only called from in6_ifdetach(),
1157 			 * which removes all the associated interface addresses
1158 			 * by itself.
1159 			 * (jinmei@kame.net 20010129)
1160 			 */
1161 			prelist_remove(pr);
1162 		}
1163 	}
1164 
1165 	/* cancel default outgoing interface setting */
1166 	if (V_nd6_defifindex == ifp->if_index)
1167 		nd6_setdefaultiface(0);
1168 
1169 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1170 		/* Refresh default router list. */
1171 		defrouter_select();
1172 	}
1173 
1174 	/* XXXXX
1175 	 * We do not nuke the neighbor cache entries here any more
1176 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
1177 	 * nd6_purge() is invoked by in6_ifdetach() which is called
1178 	 * from if_detach() where everything gets purged. So let
1179 	 * in6_domifdetach() do the actual L2 table purging work.
1180 	 */
1181 }
1182 
1183 /*
1184  * the caller acquires and releases the lock on the lltbls
1185  * Returns the llentry locked
1186  */
1187 struct llentry *
1188 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1189 {
1190 	struct sockaddr_in6 sin6;
1191 	struct llentry *ln;
1192 
1193 	bzero(&sin6, sizeof(sin6));
1194 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1195 	sin6.sin6_family = AF_INET6;
1196 	sin6.sin6_addr = *addr6;
1197 
1198 	IF_AFDATA_LOCK_ASSERT(ifp);
1199 
1200 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1201 
1202 	return (ln);
1203 }
1204 
1205 struct llentry *
1206 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1207 {
1208 	struct sockaddr_in6 sin6;
1209 	struct llentry *ln;
1210 
1211 	bzero(&sin6, sizeof(sin6));
1212 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1213 	sin6.sin6_family = AF_INET6;
1214 	sin6.sin6_addr = *addr6;
1215 
1216 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1217 	if (ln != NULL)
1218 		ln->ln_state = ND6_LLINFO_NOSTATE;
1219 
1220 	return (ln);
1221 }
1222 
1223 /*
1224  * Test whether a given IPv6 address is a neighbor or not, ignoring
1225  * the actual neighbor cache.  The neighbor cache is ignored in order
1226  * to not reenter the routing code from within itself.
1227  */
1228 static int
1229 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1230 {
1231 	struct nd_prefix *pr;
1232 	struct ifaddr *dstaddr;
1233 	struct rt_addrinfo info;
1234 	struct sockaddr_in6 rt_key;
1235 	struct sockaddr *dst6;
1236 	int fibnum;
1237 
1238 	/*
1239 	 * A link-local address is always a neighbor.
1240 	 * XXX: a link does not necessarily specify a single interface.
1241 	 */
1242 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1243 		struct sockaddr_in6 sin6_copy;
1244 		u_int32_t zone;
1245 
1246 		/*
1247 		 * We need sin6_copy since sa6_recoverscope() may modify the
1248 		 * content (XXX).
1249 		 */
1250 		sin6_copy = *addr;
1251 		if (sa6_recoverscope(&sin6_copy))
1252 			return (0); /* XXX: should be impossible */
1253 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1254 			return (0);
1255 		if (sin6_copy.sin6_scope_id == zone)
1256 			return (1);
1257 		else
1258 			return (0);
1259 	}
1260 
1261 	bzero(&rt_key, sizeof(rt_key));
1262 	bzero(&info, sizeof(info));
1263 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1264 
1265 	/* Always use the default FIB here. XXME - why? */
1266 	fibnum = RT_DEFAULT_FIB;
1267 
1268 	/*
1269 	 * If the address matches one of our addresses,
1270 	 * it should be a neighbor.
1271 	 * If the address matches one of our on-link prefixes, it should be a
1272 	 * neighbor.
1273 	 */
1274 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1275 		if (pr->ndpr_ifp != ifp)
1276 			continue;
1277 
1278 		if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) {
1279 
1280 			/* Always use the default FIB here. */
1281 			dst6 = (struct sockaddr *)&pr->ndpr_prefix;
1282 
1283 			/* Restore length field before retrying lookup */
1284 			rt_key.sin6_len = sizeof(rt_key);
1285 			if (rib_lookup_info(fibnum, dst6, 0, 0, &info) != 0)
1286 				continue;
1287 			/*
1288 			 * This is the case where multiple interfaces
1289 			 * have the same prefix, but only one is installed
1290 			 * into the routing table and that prefix entry
1291 			 * is not the one being examined here. In the case
1292 			 * where RADIX_MPATH is enabled, multiple route
1293 			 * entries (of the same rt_key value) will be
1294 			 * installed because the interface addresses all
1295 			 * differ.
1296 			 */
1297 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1298 			       &rt_key.sin6_addr))
1299 				continue;
1300 		}
1301 
1302 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1303 		    &addr->sin6_addr, &pr->ndpr_mask))
1304 			return (1);
1305 	}
1306 
1307 	/*
1308 	 * If the address is assigned on the node of the other side of
1309 	 * a p2p interface, the address should be a neighbor.
1310 	 */
1311 	dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS);
1312 	if (dstaddr != NULL) {
1313 		if (dstaddr->ifa_ifp == ifp) {
1314 			ifa_free(dstaddr);
1315 			return (1);
1316 		}
1317 		ifa_free(dstaddr);
1318 	}
1319 
1320 	/*
1321 	 * If the default router list is empty, all addresses are regarded
1322 	 * as on-link, and thus, as a neighbor.
1323 	 */
1324 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1325 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1326 	    V_nd6_defifindex == ifp->if_index) {
1327 		return (1);
1328 	}
1329 
1330 	return (0);
1331 }
1332 
1333 
1334 /*
1335  * Detect if a given IPv6 address identifies a neighbor on a given link.
1336  * XXX: should take care of the destination of a p2p link?
1337  */
1338 int
1339 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1340 {
1341 	struct llentry *lle;
1342 	int rc = 0;
1343 
1344 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1345 	if (nd6_is_new_addr_neighbor(addr, ifp))
1346 		return (1);
1347 
1348 	/*
1349 	 * Even if the address matches none of our addresses, it might be
1350 	 * in the neighbor cache.
1351 	 */
1352 	IF_AFDATA_RLOCK(ifp);
1353 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1354 		LLE_RUNLOCK(lle);
1355 		rc = 1;
1356 	}
1357 	IF_AFDATA_RUNLOCK(ifp);
1358 	return (rc);
1359 }
1360 
1361 /*
1362  * Free an nd6 llinfo entry.
1363  * Since the function would cause significant changes in the kernel, DO NOT
1364  * make it global, unless you have a strong reason for the change, and are sure
1365  * that the change is safe.
1366  *
1367  * Set noinline to be dtrace-friendly
1368  */
1369 static __noinline void
1370 nd6_free(struct llentry *ln, int gc)
1371 {
1372 	struct nd_defrouter *dr;
1373 	struct ifnet *ifp;
1374 
1375 	LLE_WLOCK_ASSERT(ln);
1376 
1377 	/*
1378 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1379 	 * even though it is not harmful, it was not really necessary.
1380 	 */
1381 
1382 	/* cancel timer */
1383 	nd6_llinfo_settimer_locked(ln, -1);
1384 
1385 	dr = NULL;
1386 	ifp = ln->lle_tbl->llt_ifp;
1387 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1388 		dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp);
1389 
1390 		if (dr != NULL && dr->expire &&
1391 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1392 			/*
1393 			 * If the reason for the deletion is just garbage
1394 			 * collection, and the neighbor is an active default
1395 			 * router, do not delete it.  Instead, reset the GC
1396 			 * timer using the router's lifetime.
1397 			 * Simply deleting the entry would affect default
1398 			 * router selection, which is not necessarily a good
1399 			 * thing, especially when we're using router preference
1400 			 * values.
1401 			 * XXX: the check for ln_state would be redundant,
1402 			 *      but we intentionally keep it just in case.
1403 			 */
1404 			if (dr->expire > time_uptime)
1405 				nd6_llinfo_settimer_locked(ln,
1406 				    (dr->expire - time_uptime) * hz);
1407 			else
1408 				nd6_llinfo_settimer_locked(ln,
1409 				    (long)V_nd6_gctimer * hz);
1410 
1411 			LLE_REMREF(ln);
1412 			LLE_WUNLOCK(ln);
1413 			defrouter_rele(dr);
1414 			return;
1415 		}
1416 
1417 		if (dr) {
1418 			/*
1419 			 * Unreachablity of a router might affect the default
1420 			 * router selection and on-link detection of advertised
1421 			 * prefixes.
1422 			 */
1423 
1424 			/*
1425 			 * Temporarily fake the state to choose a new default
1426 			 * router and to perform on-link determination of
1427 			 * prefixes correctly.
1428 			 * Below the state will be set correctly,
1429 			 * or the entry itself will be deleted.
1430 			 */
1431 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1432 		}
1433 
1434 		if (ln->ln_router || dr) {
1435 
1436 			/*
1437 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1438 			 * rnh and for the calls to pfxlist_onlink_check() and
1439 			 * defrouter_select() in the block further down for calls
1440 			 * into nd6_lookup().  We still hold a ref.
1441 			 */
1442 			LLE_WUNLOCK(ln);
1443 
1444 			/*
1445 			 * rt6_flush must be called whether or not the neighbor
1446 			 * is in the Default Router List.
1447 			 * See a corresponding comment in nd6_na_input().
1448 			 */
1449 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1450 		}
1451 
1452 		if (dr) {
1453 			/*
1454 			 * Since defrouter_select() does not affect the
1455 			 * on-link determination and MIP6 needs the check
1456 			 * before the default router selection, we perform
1457 			 * the check now.
1458 			 */
1459 			pfxlist_onlink_check();
1460 
1461 			/*
1462 			 * Refresh default router list.
1463 			 */
1464 			defrouter_select();
1465 		}
1466 
1467 		/*
1468 		 * If this entry was added by an on-link redirect, remove the
1469 		 * corresponding host route.
1470 		 */
1471 		if (ln->la_flags & LLE_REDIRECT)
1472 			nd6_free_redirect(ln);
1473 
1474 		if (ln->ln_router || dr)
1475 			LLE_WLOCK(ln);
1476 	}
1477 
1478 	/*
1479 	 * Save to unlock. We still hold an extra reference and will not
1480 	 * free(9) in llentry_free() if someone else holds one as well.
1481 	 */
1482 	LLE_WUNLOCK(ln);
1483 	IF_AFDATA_LOCK(ifp);
1484 	LLE_WLOCK(ln);
1485 	/* Guard against race with other llentry_free(). */
1486 	if (ln->la_flags & LLE_LINKED) {
1487 		/* Remove callout reference */
1488 		LLE_REMREF(ln);
1489 		lltable_unlink_entry(ln->lle_tbl, ln);
1490 	}
1491 	IF_AFDATA_UNLOCK(ifp);
1492 
1493 	llentry_free(ln);
1494 	if (dr != NULL)
1495 		defrouter_rele(dr);
1496 }
1497 
1498 static int
1499 nd6_isdynrte(const struct rtentry *rt, void *xap)
1500 {
1501 
1502 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1503 		return (1);
1504 
1505 	return (0);
1506 }
1507 /*
1508  * Remove the rtentry for the given llentry,
1509  * both of which were installed by a redirect.
1510  */
1511 static void
1512 nd6_free_redirect(const struct llentry *ln)
1513 {
1514 	int fibnum;
1515 	struct sockaddr_in6 sin6;
1516 	struct rt_addrinfo info;
1517 
1518 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1519 	memset(&info, 0, sizeof(info));
1520 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1521 	info.rti_filter = nd6_isdynrte;
1522 
1523 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1524 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1525 }
1526 
1527 /*
1528  * Rejuvenate this function for routing operations related
1529  * processing.
1530  */
1531 void
1532 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1533 {
1534 	struct sockaddr_in6 *gateway;
1535 	struct nd_defrouter *dr;
1536 	struct ifnet *ifp;
1537 
1538 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1539 	ifp = rt->rt_ifp;
1540 
1541 	switch (req) {
1542 	case RTM_ADD:
1543 		break;
1544 
1545 	case RTM_DELETE:
1546 		if (!ifp)
1547 			return;
1548 		/*
1549 		 * Only indirect routes are interesting.
1550 		 */
1551 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1552 			return;
1553 		/*
1554 		 * check for default route
1555 		 */
1556 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1557 		    &SIN6(rt_key(rt))->sin6_addr)) {
1558 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1559 			if (dr != NULL) {
1560 				dr->installed = 0;
1561 				defrouter_rele(dr);
1562 			}
1563 		}
1564 		break;
1565 	}
1566 }
1567 
1568 
1569 int
1570 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1571 {
1572 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1573 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1574 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1575 	int error = 0;
1576 
1577 	if (ifp->if_afdata[AF_INET6] == NULL)
1578 		return (EPFNOSUPPORT);
1579 	switch (cmd) {
1580 	case OSIOCGIFINFO_IN6:
1581 #define ND	ndi->ndi
1582 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1583 		bzero(&ND, sizeof(ND));
1584 		ND.linkmtu = IN6_LINKMTU(ifp);
1585 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1586 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1587 		ND.reachable = ND_IFINFO(ifp)->reachable;
1588 		ND.retrans = ND_IFINFO(ifp)->retrans;
1589 		ND.flags = ND_IFINFO(ifp)->flags;
1590 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1591 		ND.chlim = ND_IFINFO(ifp)->chlim;
1592 		break;
1593 	case SIOCGIFINFO_IN6:
1594 		ND = *ND_IFINFO(ifp);
1595 		break;
1596 	case SIOCSIFINFO_IN6:
1597 		/*
1598 		 * used to change host variables from userland.
1599 		 * intented for a use on router to reflect RA configurations.
1600 		 */
1601 		/* 0 means 'unspecified' */
1602 		if (ND.linkmtu != 0) {
1603 			if (ND.linkmtu < IPV6_MMTU ||
1604 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1605 				error = EINVAL;
1606 				break;
1607 			}
1608 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1609 		}
1610 
1611 		if (ND.basereachable != 0) {
1612 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1613 
1614 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1615 			if (ND.basereachable != obasereachable)
1616 				ND_IFINFO(ifp)->reachable =
1617 				    ND_COMPUTE_RTIME(ND.basereachable);
1618 		}
1619 		if (ND.retrans != 0)
1620 			ND_IFINFO(ifp)->retrans = ND.retrans;
1621 		if (ND.chlim != 0)
1622 			ND_IFINFO(ifp)->chlim = ND.chlim;
1623 		/* FALLTHROUGH */
1624 	case SIOCSIFINFO_FLAGS:
1625 	{
1626 		struct ifaddr *ifa;
1627 		struct in6_ifaddr *ia;
1628 
1629 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1630 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1631 			/* ifdisabled 1->0 transision */
1632 
1633 			/*
1634 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1635 			 * has an link-local address with IN6_IFF_DUPLICATED,
1636 			 * do not clear ND6_IFF_IFDISABLED.
1637 			 * See RFC 4862, Section 5.4.5.
1638 			 */
1639 			IF_ADDR_RLOCK(ifp);
1640 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1641 				if (ifa->ifa_addr->sa_family != AF_INET6)
1642 					continue;
1643 				ia = (struct in6_ifaddr *)ifa;
1644 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1645 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1646 					break;
1647 			}
1648 			IF_ADDR_RUNLOCK(ifp);
1649 
1650 			if (ifa != NULL) {
1651 				/* LLA is duplicated. */
1652 				ND.flags |= ND6_IFF_IFDISABLED;
1653 				log(LOG_ERR, "Cannot enable an interface"
1654 				    " with a link-local address marked"
1655 				    " duplicate.\n");
1656 			} else {
1657 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1658 				if (ifp->if_flags & IFF_UP)
1659 					in6_if_up(ifp);
1660 			}
1661 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1662 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1663 			/* ifdisabled 0->1 transision */
1664 			/* Mark all IPv6 address as tentative. */
1665 
1666 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1667 			if (V_ip6_dad_count > 0 &&
1668 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1669 				IF_ADDR_RLOCK(ifp);
1670 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1671 				    ifa_link) {
1672 					if (ifa->ifa_addr->sa_family !=
1673 					    AF_INET6)
1674 						continue;
1675 					ia = (struct in6_ifaddr *)ifa;
1676 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1677 				}
1678 				IF_ADDR_RUNLOCK(ifp);
1679 			}
1680 		}
1681 
1682 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1683 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1684 				/* auto_linklocal 0->1 transision */
1685 
1686 				/* If no link-local address on ifp, configure */
1687 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1688 				in6_ifattach(ifp, NULL);
1689 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1690 			    ifp->if_flags & IFF_UP) {
1691 				/*
1692 				 * When the IF already has
1693 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1694 				 * address is assigned, and IFF_UP, try to
1695 				 * assign one.
1696 				 */
1697 				IF_ADDR_RLOCK(ifp);
1698 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1699 				    ifa_link) {
1700 					if (ifa->ifa_addr->sa_family !=
1701 					    AF_INET6)
1702 						continue;
1703 					ia = (struct in6_ifaddr *)ifa;
1704 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1705 						break;
1706 				}
1707 				IF_ADDR_RUNLOCK(ifp);
1708 				if (ifa != NULL)
1709 					/* No LLA is configured. */
1710 					in6_ifattach(ifp, NULL);
1711 			}
1712 		}
1713 	}
1714 		ND_IFINFO(ifp)->flags = ND.flags;
1715 		break;
1716 #undef ND
1717 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1718 		/* sync kernel routing table with the default router list */
1719 		defrouter_reset();
1720 		defrouter_select();
1721 		break;
1722 	case SIOCSPFXFLUSH_IN6:
1723 	{
1724 		/* flush all the prefix advertised by routers */
1725 		struct nd_prefix *pr, *next;
1726 
1727 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1728 			struct in6_ifaddr *ia, *ia_next;
1729 
1730 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1731 				continue; /* XXX */
1732 
1733 			/* do we really have to remove addresses as well? */
1734 			/* XXXRW: in6_ifaddrhead locking. */
1735 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1736 			    ia_next) {
1737 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1738 					continue;
1739 
1740 				if (ia->ia6_ndpr == pr)
1741 					in6_purgeaddr(&ia->ia_ifa);
1742 			}
1743 			prelist_remove(pr);
1744 		}
1745 		break;
1746 	}
1747 	case SIOCSRTRFLUSH_IN6:
1748 	{
1749 		/* flush all the default routers */
1750 		struct nd_drhead drq;
1751 		struct nd_defrouter *dr;
1752 
1753 		TAILQ_INIT(&drq);
1754 
1755 		defrouter_reset();
1756 
1757 		ND6_WLOCK();
1758 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1759 			defrouter_unlink(dr, &drq);
1760 		ND6_WUNLOCK();
1761 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1762 			TAILQ_REMOVE(&drq, dr, dr_entry);
1763 			defrouter_del(dr);
1764 		}
1765 
1766 		defrouter_select();
1767 		break;
1768 	}
1769 	case SIOCGNBRINFO_IN6:
1770 	{
1771 		struct llentry *ln;
1772 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1773 
1774 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1775 			return (error);
1776 
1777 		IF_AFDATA_RLOCK(ifp);
1778 		ln = nd6_lookup(&nb_addr, 0, ifp);
1779 		IF_AFDATA_RUNLOCK(ifp);
1780 
1781 		if (ln == NULL) {
1782 			error = EINVAL;
1783 			break;
1784 		}
1785 		nbi->state = ln->ln_state;
1786 		nbi->asked = ln->la_asked;
1787 		nbi->isrouter = ln->ln_router;
1788 		if (ln->la_expire == 0)
1789 			nbi->expire = 0;
1790 		else
1791 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1792 			    (time_second - time_uptime);
1793 		LLE_RUNLOCK(ln);
1794 		break;
1795 	}
1796 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1797 		ndif->ifindex = V_nd6_defifindex;
1798 		break;
1799 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1800 		return (nd6_setdefaultiface(ndif->ifindex));
1801 	}
1802 	return (error);
1803 }
1804 
1805 /*
1806  * Calculates new isRouter value based on provided parameters and
1807  * returns it.
1808  */
1809 static int
1810 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1811     int ln_router)
1812 {
1813 
1814 	/*
1815 	 * ICMP6 type dependent behavior.
1816 	 *
1817 	 * NS: clear IsRouter if new entry
1818 	 * RS: clear IsRouter
1819 	 * RA: set IsRouter if there's lladdr
1820 	 * redir: clear IsRouter if new entry
1821 	 *
1822 	 * RA case, (1):
1823 	 * The spec says that we must set IsRouter in the following cases:
1824 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1825 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1826 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1827 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1828 	 * neighbor cache, this is similar to (6).
1829 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1830 	 *
1831 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1832 	 *							D R
1833 	 *	0	n	n	(1)	c   ?     s
1834 	 *	0	y	n	(2)	c   s     s
1835 	 *	0	n	y	(3)	c   s     s
1836 	 *	0	y	y	(4)	c   s     s
1837 	 *	0	y	y	(5)	c   s     s
1838 	 *	1	--	n	(6) c	c	c s
1839 	 *	1	--	y	(7) c	c   s	c s
1840 	 *
1841 	 *					(c=clear s=set)
1842 	 */
1843 	switch (type & 0xff) {
1844 	case ND_NEIGHBOR_SOLICIT:
1845 		/*
1846 		 * New entry must have is_router flag cleared.
1847 		 */
1848 		if (is_new)					/* (6-7) */
1849 			ln_router = 0;
1850 		break;
1851 	case ND_REDIRECT:
1852 		/*
1853 		 * If the icmp is a redirect to a better router, always set the
1854 		 * is_router flag.  Otherwise, if the entry is newly created,
1855 		 * clear the flag.  [RFC 2461, sec 8.3]
1856 		 */
1857 		if (code == ND_REDIRECT_ROUTER)
1858 			ln_router = 1;
1859 		else {
1860 			if (is_new)				/* (6-7) */
1861 				ln_router = 0;
1862 		}
1863 		break;
1864 	case ND_ROUTER_SOLICIT:
1865 		/*
1866 		 * is_router flag must always be cleared.
1867 		 */
1868 		ln_router = 0;
1869 		break;
1870 	case ND_ROUTER_ADVERT:
1871 		/*
1872 		 * Mark an entry with lladdr as a router.
1873 		 */
1874 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1875 		    (is_new && new_addr)) {			/* (7) */
1876 			ln_router = 1;
1877 		}
1878 		break;
1879 	}
1880 
1881 	return (ln_router);
1882 }
1883 
1884 /*
1885  * Create neighbor cache entry and cache link-layer address,
1886  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1887  *
1888  * type - ICMP6 type
1889  * code - type dependent information
1890  *
1891  */
1892 void
1893 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1894     int lladdrlen, int type, int code)
1895 {
1896 	struct llentry *ln = NULL, *ln_tmp;
1897 	int is_newentry;
1898 	int do_update;
1899 	int olladdr;
1900 	int llchange;
1901 	int flags;
1902 	uint16_t router = 0;
1903 	struct sockaddr_in6 sin6;
1904 	struct mbuf *chain = NULL;
1905 	u_char linkhdr[LLE_MAX_LINKHDR];
1906 	size_t linkhdrsize;
1907 	int lladdr_off;
1908 
1909 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1910 
1911 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1912 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1913 
1914 	/* nothing must be updated for unspecified address */
1915 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1916 		return;
1917 
1918 	/*
1919 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1920 	 * the caller.
1921 	 *
1922 	 * XXX If the link does not have link-layer adderss, what should
1923 	 * we do? (ifp->if_addrlen == 0)
1924 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1925 	 * description on it in NS section (RFC 2461 7.2.3).
1926 	 */
1927 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1928 	IF_AFDATA_RLOCK(ifp);
1929 	ln = nd6_lookup(from, flags, ifp);
1930 	IF_AFDATA_RUNLOCK(ifp);
1931 	is_newentry = 0;
1932 	if (ln == NULL) {
1933 		flags |= LLE_EXCLUSIVE;
1934 		ln = nd6_alloc(from, 0, ifp);
1935 		if (ln == NULL)
1936 			return;
1937 
1938 		/*
1939 		 * Since we already know all the data for the new entry,
1940 		 * fill it before insertion.
1941 		 */
1942 		if (lladdr != NULL) {
1943 			linkhdrsize = sizeof(linkhdr);
1944 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1945 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1946 				return;
1947 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1948 			    lladdr_off);
1949 		}
1950 
1951 		IF_AFDATA_WLOCK(ifp);
1952 		LLE_WLOCK(ln);
1953 		/* Prefer any existing lle over newly-created one */
1954 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1955 		if (ln_tmp == NULL)
1956 			lltable_link_entry(LLTABLE6(ifp), ln);
1957 		IF_AFDATA_WUNLOCK(ifp);
1958 		if (ln_tmp == NULL) {
1959 			/* No existing lle, mark as new entry (6,7) */
1960 			is_newentry = 1;
1961 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1962 			if (lladdr != NULL)	/* (7) */
1963 				EVENTHANDLER_INVOKE(lle_event, ln,
1964 				    LLENTRY_RESOLVED);
1965 		} else {
1966 			lltable_free_entry(LLTABLE6(ifp), ln);
1967 			ln = ln_tmp;
1968 			ln_tmp = NULL;
1969 		}
1970 	}
1971 	/* do nothing if static ndp is set */
1972 	if ((ln->la_flags & LLE_STATIC)) {
1973 		if (flags & LLE_EXCLUSIVE)
1974 			LLE_WUNLOCK(ln);
1975 		else
1976 			LLE_RUNLOCK(ln);
1977 		return;
1978 	}
1979 
1980 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1981 	if (olladdr && lladdr) {
1982 		llchange = bcmp(lladdr, ln->ll_addr,
1983 		    ifp->if_addrlen);
1984 	} else if (!olladdr && lladdr)
1985 		llchange = 1;
1986 	else
1987 		llchange = 0;
1988 
1989 	/*
1990 	 * newentry olladdr  lladdr  llchange	(*=record)
1991 	 *	0	n	n	--	(1)
1992 	 *	0	y	n	--	(2)
1993 	 *	0	n	y	y	(3) * STALE
1994 	 *	0	y	y	n	(4) *
1995 	 *	0	y	y	y	(5) * STALE
1996 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1997 	 *	1	--	y	--	(7) * STALE
1998 	 */
1999 
2000 	do_update = 0;
2001 	if (is_newentry == 0 && llchange != 0) {
2002 		do_update = 1;	/* (3,5) */
2003 
2004 		/*
2005 		 * Record source link-layer address
2006 		 * XXX is it dependent to ifp->if_type?
2007 		 */
2008 		linkhdrsize = sizeof(linkhdr);
2009 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2010 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2011 			return;
2012 
2013 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2014 		    lladdr_off) == 0) {
2015 			/* Entry was deleted */
2016 			return;
2017 		}
2018 
2019 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2020 
2021 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2022 
2023 		if (ln->la_hold != NULL)
2024 			nd6_grab_holdchain(ln, &chain, &sin6);
2025 	}
2026 
2027 	/* Calculates new router status */
2028 	router = nd6_is_router(type, code, is_newentry, olladdr,
2029 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2030 
2031 	ln->ln_router = router;
2032 	/* Mark non-router redirects with special flag */
2033 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2034 		ln->la_flags |= LLE_REDIRECT;
2035 
2036 	if (flags & LLE_EXCLUSIVE)
2037 		LLE_WUNLOCK(ln);
2038 	else
2039 		LLE_RUNLOCK(ln);
2040 
2041 	if (chain != NULL)
2042 		nd6_flush_holdchain(ifp, ifp, chain, &sin6);
2043 
2044 	/*
2045 	 * When the link-layer address of a router changes, select the
2046 	 * best router again.  In particular, when the neighbor entry is newly
2047 	 * created, it might affect the selection policy.
2048 	 * Question: can we restrict the first condition to the "is_newentry"
2049 	 * case?
2050 	 * XXX: when we hear an RA from a new router with the link-layer
2051 	 * address option, defrouter_select() is called twice, since
2052 	 * defrtrlist_update called the function as well.  However, I believe
2053 	 * we can compromise the overhead, since it only happens the first
2054 	 * time.
2055 	 * XXX: although defrouter_select() should not have a bad effect
2056 	 * for those are not autoconfigured hosts, we explicitly avoid such
2057 	 * cases for safety.
2058 	 */
2059 	if ((do_update || is_newentry) && router &&
2060 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2061 		/*
2062 		 * guaranteed recursion
2063 		 */
2064 		defrouter_select();
2065 	}
2066 }
2067 
2068 static void
2069 nd6_slowtimo(void *arg)
2070 {
2071 	CURVNET_SET((struct vnet *) arg);
2072 	struct nd_ifinfo *nd6if;
2073 	struct ifnet *ifp;
2074 
2075 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2076 	    nd6_slowtimo, curvnet);
2077 	IFNET_RLOCK_NOSLEEP();
2078 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2079 		if (ifp->if_afdata[AF_INET6] == NULL)
2080 			continue;
2081 		nd6if = ND_IFINFO(ifp);
2082 		if (nd6if->basereachable && /* already initialized */
2083 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2084 			/*
2085 			 * Since reachable time rarely changes by router
2086 			 * advertisements, we SHOULD insure that a new random
2087 			 * value gets recomputed at least once every few hours.
2088 			 * (RFC 2461, 6.3.4)
2089 			 */
2090 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2091 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2092 		}
2093 	}
2094 	IFNET_RUNLOCK_NOSLEEP();
2095 	CURVNET_RESTORE();
2096 }
2097 
2098 void
2099 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2100     struct sockaddr_in6 *sin6)
2101 {
2102 
2103 	LLE_WLOCK_ASSERT(ln);
2104 
2105 	*chain = ln->la_hold;
2106 	ln->la_hold = NULL;
2107 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2108 
2109 	if (ln->ln_state == ND6_LLINFO_STALE) {
2110 
2111 		/*
2112 		 * The first time we send a packet to a
2113 		 * neighbor whose entry is STALE, we have
2114 		 * to change the state to DELAY and a sets
2115 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2116 		 * seconds to ensure do neighbor unreachability
2117 		 * detection on expiration.
2118 		 * (RFC 2461 7.3.3)
2119 		 */
2120 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2121 	}
2122 }
2123 
2124 int
2125 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2126     struct sockaddr_in6 *dst, struct route *ro)
2127 {
2128 	int error;
2129 	int ip6len;
2130 	struct ip6_hdr *ip6;
2131 	struct m_tag *mtag;
2132 
2133 #ifdef MAC
2134 	mac_netinet6_nd6_send(ifp, m);
2135 #endif
2136 
2137 	/*
2138 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2139 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2140 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2141 	 * to be diverted to user space.  When re-injected into the kernel,
2142 	 * send_output() will directly dispatch them to the outgoing interface.
2143 	 */
2144 	if (send_sendso_input_hook != NULL) {
2145 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2146 		if (mtag != NULL) {
2147 			ip6 = mtod(m, struct ip6_hdr *);
2148 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2149 			/* Use the SEND socket */
2150 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2151 			    ip6len);
2152 			/* -1 == no app on SEND socket */
2153 			if (error == 0 || error != -1)
2154 			    return (error);
2155 		}
2156 	}
2157 
2158 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2159 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2160 	    mtod(m, struct ip6_hdr *));
2161 
2162 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2163 		origifp = ifp;
2164 
2165 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2166 	return (error);
2167 }
2168 
2169 /*
2170  * Lookup link headerfor @sa_dst address. Stores found
2171  * data in @desten buffer. Copy of lle ln_flags can be also
2172  * saved in @pflags if @pflags is non-NULL.
2173  *
2174  * If destination LLE does not exists or lle state modification
2175  * is required, call "slow" version.
2176  *
2177  * Return values:
2178  * - 0 on success (address copied to buffer).
2179  * - EWOULDBLOCK (no local error, but address is still unresolved)
2180  * - other errors (alloc failure, etc)
2181  */
2182 int
2183 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2184     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags)
2185 {
2186 	struct llentry *ln = NULL;
2187 	const struct sockaddr_in6 *dst6;
2188 
2189 	if (pflags != NULL)
2190 		*pflags = 0;
2191 
2192 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2193 
2194 	/* discard the packet if IPv6 operation is disabled on the interface */
2195 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2196 		m_freem(m);
2197 		return (ENETDOWN); /* better error? */
2198 	}
2199 
2200 	if (m != NULL && m->m_flags & M_MCAST) {
2201 		switch (ifp->if_type) {
2202 		case IFT_ETHER:
2203 		case IFT_FDDI:
2204 		case IFT_L2VLAN:
2205 		case IFT_IEEE80211:
2206 		case IFT_BRIDGE:
2207 		case IFT_ISO88025:
2208 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2209 						 desten);
2210 			return (0);
2211 		default:
2212 			m_freem(m);
2213 			return (EAFNOSUPPORT);
2214 		}
2215 	}
2216 
2217 	IF_AFDATA_RLOCK(ifp);
2218 	ln = nd6_lookup(&dst6->sin6_addr, LLE_UNLOCKED, ifp);
2219 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2220 		/* Entry found, let's copy lle info */
2221 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2222 		if (pflags != NULL)
2223 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2224 		/* Check if we have feedback request from nd6 timer */
2225 		if (ln->r_skip_req != 0) {
2226 			LLE_REQ_LOCK(ln);
2227 			ln->r_skip_req = 0; /* Notify that entry was used */
2228 			ln->lle_hittime = time_uptime;
2229 			LLE_REQ_UNLOCK(ln);
2230 		}
2231 		IF_AFDATA_RUNLOCK(ifp);
2232 		return (0);
2233 	}
2234 	IF_AFDATA_RUNLOCK(ifp);
2235 
2236 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags));
2237 }
2238 
2239 
2240 /*
2241  * Do L2 address resolution for @sa_dst address. Stores found
2242  * address in @desten buffer. Copy of lle ln_flags can be also
2243  * saved in @pflags if @pflags is non-NULL.
2244  *
2245  * Heavy version.
2246  * Function assume that destination LLE does not exist,
2247  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2248  *
2249  * Set noinline to be dtrace-friendly
2250  */
2251 static __noinline int
2252 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2253     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags)
2254 {
2255 	struct llentry *lle = NULL, *lle_tmp;
2256 	struct in6_addr *psrc, src;
2257 	int send_ns, ll_len;
2258 	char *lladdr;
2259 
2260 	/*
2261 	 * Address resolution or Neighbor Unreachability Detection
2262 	 * for the next hop.
2263 	 * At this point, the destination of the packet must be a unicast
2264 	 * or an anycast address(i.e. not a multicast).
2265 	 */
2266 	if (lle == NULL) {
2267 		IF_AFDATA_RLOCK(ifp);
2268 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2269 		IF_AFDATA_RUNLOCK(ifp);
2270 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2271 			/*
2272 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2273 			 * the condition below is not very efficient.  But we believe
2274 			 * it is tolerable, because this should be a rare case.
2275 			 */
2276 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2277 			if (lle == NULL) {
2278 				char ip6buf[INET6_ADDRSTRLEN];
2279 				log(LOG_DEBUG,
2280 				    "nd6_output: can't allocate llinfo for %s "
2281 				    "(ln=%p)\n",
2282 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2283 				m_freem(m);
2284 				return (ENOBUFS);
2285 			}
2286 
2287 			IF_AFDATA_WLOCK(ifp);
2288 			LLE_WLOCK(lle);
2289 			/* Prefer any existing entry over newly-created one */
2290 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2291 			if (lle_tmp == NULL)
2292 				lltable_link_entry(LLTABLE6(ifp), lle);
2293 			IF_AFDATA_WUNLOCK(ifp);
2294 			if (lle_tmp != NULL) {
2295 				lltable_free_entry(LLTABLE6(ifp), lle);
2296 				lle = lle_tmp;
2297 				lle_tmp = NULL;
2298 			}
2299 		}
2300 	}
2301 	if (lle == NULL) {
2302 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2303 			m_freem(m);
2304 			return (ENOBUFS);
2305 		}
2306 
2307 		if (m != NULL)
2308 			m_freem(m);
2309 		return (ENOBUFS);
2310 	}
2311 
2312 	LLE_WLOCK_ASSERT(lle);
2313 
2314 	/*
2315 	 * The first time we send a packet to a neighbor whose entry is
2316 	 * STALE, we have to change the state to DELAY and a sets a timer to
2317 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2318 	 * neighbor unreachability detection on expiration.
2319 	 * (RFC 2461 7.3.3)
2320 	 */
2321 	if (lle->ln_state == ND6_LLINFO_STALE)
2322 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2323 
2324 	/*
2325 	 * If the neighbor cache entry has a state other than INCOMPLETE
2326 	 * (i.e. its link-layer address is already resolved), just
2327 	 * send the packet.
2328 	 */
2329 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2330 		if (flags & LLE_ADDRONLY) {
2331 			lladdr = lle->ll_addr;
2332 			ll_len = ifp->if_addrlen;
2333 		} else {
2334 			lladdr = lle->r_linkdata;
2335 			ll_len = lle->r_hdrlen;
2336 		}
2337 		bcopy(lladdr, desten, ll_len);
2338 		if (pflags != NULL)
2339 			*pflags = lle->la_flags;
2340 		LLE_WUNLOCK(lle);
2341 		return (0);
2342 	}
2343 
2344 	/*
2345 	 * There is a neighbor cache entry, but no ethernet address
2346 	 * response yet.  Append this latest packet to the end of the
2347 	 * packet queue in the mbuf, unless the number of the packet
2348 	 * does not exceed nd6_maxqueuelen.  When it exceeds nd6_maxqueuelen,
2349 	 * the oldest packet in the queue will be removed.
2350 	 */
2351 
2352 	if (lle->la_hold != NULL) {
2353 		struct mbuf *m_hold;
2354 		int i;
2355 
2356 		i = 0;
2357 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2358 			i++;
2359 			if (m_hold->m_nextpkt == NULL) {
2360 				m_hold->m_nextpkt = m;
2361 				break;
2362 			}
2363 		}
2364 		while (i >= V_nd6_maxqueuelen) {
2365 			m_hold = lle->la_hold;
2366 			lle->la_hold = lle->la_hold->m_nextpkt;
2367 			m_freem(m_hold);
2368 			i--;
2369 		}
2370 	} else {
2371 		lle->la_hold = m;
2372 	}
2373 
2374 	/*
2375 	 * If there has been no NS for the neighbor after entering the
2376 	 * INCOMPLETE state, send the first solicitation.
2377 	 * Note that for newly-created lle la_asked will be 0,
2378 	 * so we will transition from ND6_LLINFO_NOSTATE to
2379 	 * ND6_LLINFO_INCOMPLETE state here.
2380 	 */
2381 	psrc = NULL;
2382 	send_ns = 0;
2383 	if (lle->la_asked == 0) {
2384 		lle->la_asked++;
2385 		send_ns = 1;
2386 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2387 
2388 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2389 	}
2390 	LLE_WUNLOCK(lle);
2391 	if (send_ns != 0)
2392 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2393 
2394 	return (EWOULDBLOCK);
2395 }
2396 
2397 /*
2398  * Do L2 address resolution for @sa_dst address. Stores found
2399  * address in @desten buffer. Copy of lle ln_flags can be also
2400  * saved in @pflags if @pflags is non-NULL.
2401  *
2402  * Return values:
2403  * - 0 on success (address copied to buffer).
2404  * - EWOULDBLOCK (no local error, but address is still unresolved)
2405  * - other errors (alloc failure, etc)
2406  */
2407 int
2408 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2409     char *desten, uint32_t *pflags)
2410 {
2411 	int error;
2412 
2413 	flags |= LLE_ADDRONLY;
2414 	error = nd6_resolve_slow(ifp, flags, NULL,
2415 	    (const struct sockaddr_in6 *)dst, desten, pflags);
2416 	return (error);
2417 }
2418 
2419 int
2420 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain,
2421     struct sockaddr_in6 *dst)
2422 {
2423 	struct mbuf *m, *m_head;
2424 	struct ifnet *outifp;
2425 	int error = 0;
2426 
2427 	m_head = chain;
2428 	if ((ifp->if_flags & IFF_LOOPBACK) != 0)
2429 		outifp = origifp;
2430 	else
2431 		outifp = ifp;
2432 
2433 	while (m_head) {
2434 		m = m_head;
2435 		m_head = m_head->m_nextpkt;
2436 		error = nd6_output_ifp(ifp, origifp, m, dst, NULL);
2437 	}
2438 
2439 	/*
2440 	 * XXX
2441 	 * note that intermediate errors are blindly ignored
2442 	 */
2443 	return (error);
2444 }
2445 
2446 static int
2447 nd6_need_cache(struct ifnet *ifp)
2448 {
2449 	/*
2450 	 * XXX: we currently do not make neighbor cache on any interface
2451 	 * other than ARCnet, Ethernet, FDDI and GIF.
2452 	 *
2453 	 * RFC2893 says:
2454 	 * - unidirectional tunnels needs no ND
2455 	 */
2456 	switch (ifp->if_type) {
2457 	case IFT_ARCNET:
2458 	case IFT_ETHER:
2459 	case IFT_FDDI:
2460 	case IFT_IEEE1394:
2461 	case IFT_L2VLAN:
2462 	case IFT_IEEE80211:
2463 	case IFT_INFINIBAND:
2464 	case IFT_BRIDGE:
2465 	case IFT_PROPVIRTUAL:
2466 		return (1);
2467 	default:
2468 		return (0);
2469 	}
2470 }
2471 
2472 /*
2473  * Add pernament ND6 link-layer record for given
2474  * interface address.
2475  *
2476  * Very similar to IPv4 arp_ifinit(), but:
2477  * 1) IPv6 DAD is performed in different place
2478  * 2) It is called by IPv6 protocol stack in contrast to
2479  * arp_ifinit() which is typically called in SIOCSIFADDR
2480  * driver ioctl handler.
2481  *
2482  */
2483 int
2484 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2485 {
2486 	struct ifnet *ifp;
2487 	struct llentry *ln, *ln_tmp;
2488 	struct sockaddr *dst;
2489 
2490 	ifp = ia->ia_ifa.ifa_ifp;
2491 	if (nd6_need_cache(ifp) == 0)
2492 		return (0);
2493 
2494 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2495 	dst = (struct sockaddr *)&ia->ia_addr;
2496 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2497 	if (ln == NULL)
2498 		return (ENOBUFS);
2499 
2500 	IF_AFDATA_WLOCK(ifp);
2501 	LLE_WLOCK(ln);
2502 	/* Unlink any entry if exists */
2503 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2504 	if (ln_tmp != NULL)
2505 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2506 	lltable_link_entry(LLTABLE6(ifp), ln);
2507 	IF_AFDATA_WUNLOCK(ifp);
2508 
2509 	if (ln_tmp != NULL)
2510 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2511 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2512 
2513 	LLE_WUNLOCK(ln);
2514 	if (ln_tmp != NULL)
2515 		llentry_free(ln_tmp);
2516 
2517 	return (0);
2518 }
2519 
2520 /*
2521  * Removes either all lle entries for given @ia, or lle
2522  * corresponding to @ia address.
2523  */
2524 void
2525 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2526 {
2527 	struct sockaddr_in6 mask, addr;
2528 	struct sockaddr *saddr, *smask;
2529 	struct ifnet *ifp;
2530 
2531 	ifp = ia->ia_ifa.ifa_ifp;
2532 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2533 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2534 	saddr = (struct sockaddr *)&addr;
2535 	smask = (struct sockaddr *)&mask;
2536 
2537 	if (all != 0)
2538 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2539 	else
2540 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2541 }
2542 
2543 static void
2544 clear_llinfo_pqueue(struct llentry *ln)
2545 {
2546 	struct mbuf *m_hold, *m_hold_next;
2547 
2548 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2549 		m_hold_next = m_hold->m_nextpkt;
2550 		m_freem(m_hold);
2551 	}
2552 
2553 	ln->la_hold = NULL;
2554 }
2555 
2556 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2557 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2558 #ifdef SYSCTL_DECL
2559 SYSCTL_DECL(_net_inet6_icmp6);
2560 #endif
2561 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2562 	CTLFLAG_RD, nd6_sysctl_drlist, "");
2563 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2564 	CTLFLAG_RD, nd6_sysctl_prlist, "");
2565 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2566 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2567 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2568 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2569 
2570 static int
2571 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2572 {
2573 	struct in6_defrouter d;
2574 	struct nd_defrouter *dr;
2575 	int error;
2576 
2577 	if (req->newptr != NULL)
2578 		return (EPERM);
2579 
2580 	error = sysctl_wire_old_buffer(req, 0);
2581 	if (error != 0)
2582 		return (error);
2583 
2584 	bzero(&d, sizeof(d));
2585 	d.rtaddr.sin6_family = AF_INET6;
2586 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2587 
2588 	ND6_RLOCK();
2589 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2590 		d.rtaddr.sin6_addr = dr->rtaddr;
2591 		error = sa6_recoverscope(&d.rtaddr);
2592 		if (error != 0)
2593 			break;
2594 		d.flags = dr->raflags;
2595 		d.rtlifetime = dr->rtlifetime;
2596 		d.expire = dr->expire + (time_second - time_uptime);
2597 		d.if_index = dr->ifp->if_index;
2598 		error = SYSCTL_OUT(req, &d, sizeof(d));
2599 		if (error != 0)
2600 			break;
2601 	}
2602 	ND6_RUNLOCK();
2603 	return (error);
2604 }
2605 
2606 static int
2607 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2608 {
2609 	struct in6_prefix p;
2610 	struct sockaddr_in6 s6;
2611 	struct nd_prefix *pr;
2612 	struct nd_pfxrouter *pfr;
2613 	time_t maxexpire;
2614 	int error;
2615 	char ip6buf[INET6_ADDRSTRLEN];
2616 
2617 	if (req->newptr)
2618 		return (EPERM);
2619 
2620 	bzero(&p, sizeof(p));
2621 	p.origin = PR_ORIG_RA;
2622 	bzero(&s6, sizeof(s6));
2623 	s6.sin6_family = AF_INET6;
2624 	s6.sin6_len = sizeof(s6);
2625 
2626 	/*
2627 	 * XXX locking
2628 	 */
2629 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2630 		p.prefix = pr->ndpr_prefix;
2631 		if (sa6_recoverscope(&p.prefix)) {
2632 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2633 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2634 			/* XXX: press on... */
2635 		}
2636 		p.raflags = pr->ndpr_raf;
2637 		p.prefixlen = pr->ndpr_plen;
2638 		p.vltime = pr->ndpr_vltime;
2639 		p.pltime = pr->ndpr_pltime;
2640 		p.if_index = pr->ndpr_ifp->if_index;
2641 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2642 			p.expire = 0;
2643 		else {
2644 			/* XXX: we assume time_t is signed. */
2645 			maxexpire = (-1) &
2646 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2647 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2648 				p.expire = pr->ndpr_lastupdate +
2649 				    pr->ndpr_vltime +
2650 				    (time_second - time_uptime);
2651 			else
2652 				p.expire = maxexpire;
2653 		}
2654 		p.refcnt = pr->ndpr_refcnt;
2655 		p.flags = pr->ndpr_stateflags;
2656 		p.advrtrs = 0;
2657 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2658 			p.advrtrs++;
2659 		error = SYSCTL_OUT(req, &p, sizeof(p));
2660 		if (error != 0)
2661 			return (error);
2662 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2663 			s6.sin6_addr = pfr->router->rtaddr;
2664 			if (sa6_recoverscope(&s6))
2665 				log(LOG_ERR,
2666 				    "scope error in prefix list (%s)\n",
2667 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2668 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2669 			if (error != 0)
2670 				return (error);
2671 		}
2672 	}
2673 	return (0);
2674 }
2675