xref: /freebsd/sys/netinet6/nd6.c (revision cd8537910406e68d4719136a5b0cf6d23bb1b23b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_route.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/eventhandler.h>
44 #include <sys/callout.h>
45 #include <sys/lock.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/mutex.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/time.h>
52 #include <sys/kernel.h>
53 #include <sys/protosw.h>
54 #include <sys/errno.h>
55 #include <sys/syslog.h>
56 #include <sys/rwlock.h>
57 #include <sys/queue.h>
58 #include <sys/sdt.h>
59 #include <sys/sysctl.h>
60 
61 #include <net/if.h>
62 #include <net/if_var.h>
63 #include <net/if_dl.h>
64 #include <net/if_types.h>
65 #include <net/route.h>
66 #include <net/route/route_ctl.h>
67 #include <net/route/nhop.h>
68 #include <net/vnet.h>
69 
70 #include <netinet/in.h>
71 #include <netinet/in_kdtrace.h>
72 #include <net/if_llatbl.h>
73 #include <netinet/if_ether.h>
74 #include <netinet6/in6_var.h>
75 #include <netinet/ip6.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
79 #include <netinet6/in6_ifattach.h>
80 #include <netinet/icmp6.h>
81 #include <netinet6/send.h>
82 
83 #include <sys/limits.h>
84 
85 #include <security/mac/mac_framework.h>
86 
87 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
88 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
89 
90 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
91 
92 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
93 
94 /* timer values */
95 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
96 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
97 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
98 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
99 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
100 					 * local traffic */
101 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
102 					 * collection timer */
103 
104 /* preventing too many loops in ND option parsing */
105 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
106 
107 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
108 					 * layer hints */
109 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
110 					 * ND entries */
111 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
112 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
113 
114 #ifdef ND6_DEBUG
115 VNET_DEFINE(int, nd6_debug) = 1;
116 #else
117 VNET_DEFINE(int, nd6_debug) = 0;
118 #endif
119 
120 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
121 
122 VNET_DEFINE(struct nd_prhead, nd_prefix);
123 VNET_DEFINE(struct rwlock, nd6_lock);
124 VNET_DEFINE(uint64_t, nd6_list_genid);
125 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
126 
127 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
128 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
129 
130 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
131 
132 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
133 	struct ifnet *);
134 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
135 static void nd6_slowtimo(void *);
136 static int regen_tmpaddr(struct in6_ifaddr *);
137 static void nd6_free(struct llentry **, int);
138 static void nd6_free_redirect(const struct llentry *);
139 static void nd6_llinfo_timer(void *);
140 static void nd6_llinfo_settimer_locked(struct llentry *, long);
141 static void clear_llinfo_pqueue(struct llentry *);
142 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
143     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
144 static int nd6_need_cache(struct ifnet *);
145 
146 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
147 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
148 
149 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
150 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
151 
152 SYSCTL_DECL(_net_inet6_icmp6);
153 
154 static void
155 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
156 {
157 	struct rt_addrinfo rtinfo;
158 	struct sockaddr_in6 dst;
159 	struct sockaddr_dl gw;
160 	struct ifnet *ifp;
161 	int type;
162 	int fibnum;
163 
164 	LLE_WLOCK_ASSERT(lle);
165 
166 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
167 		return;
168 
169 	switch (evt) {
170 	case LLENTRY_RESOLVED:
171 		type = RTM_ADD;
172 		KASSERT(lle->la_flags & LLE_VALID,
173 		    ("%s: %p resolved but not valid?", __func__, lle));
174 		break;
175 	case LLENTRY_EXPIRED:
176 		type = RTM_DELETE;
177 		break;
178 	default:
179 		return;
180 	}
181 
182 	ifp = lltable_get_ifp(lle->lle_tbl);
183 
184 	bzero(&dst, sizeof(dst));
185 	bzero(&gw, sizeof(gw));
186 	bzero(&rtinfo, sizeof(rtinfo));
187 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
188 	dst.sin6_scope_id = in6_getscopezone(ifp,
189 	    in6_addrscope(&dst.sin6_addr));
190 	gw.sdl_len = sizeof(struct sockaddr_dl);
191 	gw.sdl_family = AF_LINK;
192 	gw.sdl_alen = ifp->if_addrlen;
193 	gw.sdl_index = ifp->if_index;
194 	gw.sdl_type = ifp->if_type;
195 	if (evt == LLENTRY_RESOLVED)
196 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
197 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
198 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
199 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
200 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
201 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
202 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
203 }
204 
205 /*
206  * A handler for interface link layer address change event.
207  */
208 static void
209 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
210 {
211 
212 	lltable_update_ifaddr(LLTABLE6(ifp));
213 }
214 
215 void
216 nd6_init(void)
217 {
218 
219 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
220 	rw_init(&V_nd6_lock, "nd6 list");
221 
222 	LIST_INIT(&V_nd_prefix);
223 	nd6_defrouter_init();
224 
225 	/* Start timers. */
226 	callout_init(&V_nd6_slowtimo_ch, 0);
227 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
228 	    nd6_slowtimo, curvnet);
229 
230 	callout_init(&V_nd6_timer_ch, 0);
231 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
232 
233 	nd6_dad_init();
234 	if (IS_DEFAULT_VNET(curvnet)) {
235 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
236 		    NULL, EVENTHANDLER_PRI_ANY);
237 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
238 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
239 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
240 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
241 	}
242 }
243 
244 #ifdef VIMAGE
245 void
246 nd6_destroy()
247 {
248 
249 	callout_drain(&V_nd6_slowtimo_ch);
250 	callout_drain(&V_nd6_timer_ch);
251 	if (IS_DEFAULT_VNET(curvnet)) {
252 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
253 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
254 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
255 	}
256 	rw_destroy(&V_nd6_lock);
257 	mtx_destroy(&V_nd6_onlink_mtx);
258 }
259 #endif
260 
261 struct nd_ifinfo *
262 nd6_ifattach(struct ifnet *ifp)
263 {
264 	struct nd_ifinfo *nd;
265 
266 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
267 	nd->initialized = 1;
268 
269 	nd->chlim = IPV6_DEFHLIM;
270 	nd->basereachable = REACHABLE_TIME;
271 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
272 	nd->retrans = RETRANS_TIMER;
273 
274 	nd->flags = ND6_IFF_PERFORMNUD;
275 
276 	/* Set IPv6 disabled on all interfaces but loopback by default. */
277 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
278 		nd->flags |= ND6_IFF_IFDISABLED;
279 
280 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
281 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
282 	 * default regardless of the V_ip6_auto_linklocal configuration to
283 	 * give a reasonable default behavior.
284 	 */
285 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
286 	    (ifp->if_flags & IFF_LOOPBACK))
287 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
288 	/*
289 	 * A loopback interface does not need to accept RTADV.
290 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
291 	 * default regardless of the V_ip6_accept_rtadv configuration to
292 	 * prevent the interface from accepting RA messages arrived
293 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
294 	 */
295 	if (V_ip6_accept_rtadv &&
296 	    !(ifp->if_flags & IFF_LOOPBACK) &&
297 	    (ifp->if_type != IFT_BRIDGE)) {
298 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
299 			/* If we globally accept rtadv, assume IPv6 on. */
300 			nd->flags &= ~ND6_IFF_IFDISABLED;
301 	}
302 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
303 		nd->flags |= ND6_IFF_NO_RADR;
304 
305 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
306 	nd6_setmtu0(ifp, nd);
307 
308 	return nd;
309 }
310 
311 void
312 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
313 {
314 	struct epoch_tracker et;
315 	struct ifaddr *ifa, *next;
316 
317 	NET_EPOCH_ENTER(et);
318 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
319 		if (ifa->ifa_addr->sa_family != AF_INET6)
320 			continue;
321 
322 		/* stop DAD processing */
323 		nd6_dad_stop(ifa);
324 	}
325 	NET_EPOCH_EXIT(et);
326 
327 	free(nd, M_IP6NDP);
328 }
329 
330 /*
331  * Reset ND level link MTU. This function is called when the physical MTU
332  * changes, which means we might have to adjust the ND level MTU.
333  */
334 void
335 nd6_setmtu(struct ifnet *ifp)
336 {
337 	if (ifp->if_afdata[AF_INET6] == NULL)
338 		return;
339 
340 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
341 }
342 
343 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
344 void
345 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
346 {
347 	u_int32_t omaxmtu;
348 
349 	omaxmtu = ndi->maxmtu;
350 	ndi->maxmtu = ifp->if_mtu;
351 
352 	/*
353 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
354 	 * undesirable situation.  We thus notify the operator of the change
355 	 * explicitly.  The check for omaxmtu is necessary to restrict the
356 	 * log to the case of changing the MTU, not initializing it.
357 	 */
358 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
359 		log(LOG_NOTICE, "nd6_setmtu0: "
360 		    "new link MTU on %s (%lu) is too small for IPv6\n",
361 		    if_name(ifp), (unsigned long)ndi->maxmtu);
362 	}
363 
364 	if (ndi->maxmtu > V_in6_maxmtu)
365 		in6_setmaxmtu(); /* check all interfaces just in case */
366 
367 }
368 
369 void
370 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
371 {
372 
373 	bzero(ndopts, sizeof(*ndopts));
374 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
375 	ndopts->nd_opts_last
376 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
377 
378 	if (icmp6len == 0) {
379 		ndopts->nd_opts_done = 1;
380 		ndopts->nd_opts_search = NULL;
381 	}
382 }
383 
384 /*
385  * Take one ND option.
386  */
387 struct nd_opt_hdr *
388 nd6_option(union nd_opts *ndopts)
389 {
390 	struct nd_opt_hdr *nd_opt;
391 	int olen;
392 
393 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
394 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
395 	    __func__));
396 	if (ndopts->nd_opts_search == NULL)
397 		return NULL;
398 	if (ndopts->nd_opts_done)
399 		return NULL;
400 
401 	nd_opt = ndopts->nd_opts_search;
402 
403 	/* make sure nd_opt_len is inside the buffer */
404 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
405 		bzero(ndopts, sizeof(*ndopts));
406 		return NULL;
407 	}
408 
409 	olen = nd_opt->nd_opt_len << 3;
410 	if (olen == 0) {
411 		/*
412 		 * Message validation requires that all included
413 		 * options have a length that is greater than zero.
414 		 */
415 		bzero(ndopts, sizeof(*ndopts));
416 		return NULL;
417 	}
418 
419 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
420 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
421 		/* option overruns the end of buffer, invalid */
422 		bzero(ndopts, sizeof(*ndopts));
423 		return NULL;
424 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
425 		/* reached the end of options chain */
426 		ndopts->nd_opts_done = 1;
427 		ndopts->nd_opts_search = NULL;
428 	}
429 	return nd_opt;
430 }
431 
432 /*
433  * Parse multiple ND options.
434  * This function is much easier to use, for ND routines that do not need
435  * multiple options of the same type.
436  */
437 int
438 nd6_options(union nd_opts *ndopts)
439 {
440 	struct nd_opt_hdr *nd_opt;
441 	int i = 0;
442 
443 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
444 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
445 	    __func__));
446 	if (ndopts->nd_opts_search == NULL)
447 		return 0;
448 
449 	while (1) {
450 		nd_opt = nd6_option(ndopts);
451 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
452 			/*
453 			 * Message validation requires that all included
454 			 * options have a length that is greater than zero.
455 			 */
456 			ICMP6STAT_INC(icp6s_nd_badopt);
457 			bzero(ndopts, sizeof(*ndopts));
458 			return -1;
459 		}
460 
461 		if (nd_opt == NULL)
462 			goto skip1;
463 
464 		switch (nd_opt->nd_opt_type) {
465 		case ND_OPT_SOURCE_LINKADDR:
466 		case ND_OPT_TARGET_LINKADDR:
467 		case ND_OPT_MTU:
468 		case ND_OPT_REDIRECTED_HEADER:
469 		case ND_OPT_NONCE:
470 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
471 				nd6log((LOG_INFO,
472 				    "duplicated ND6 option found (type=%d)\n",
473 				    nd_opt->nd_opt_type));
474 				/* XXX bark? */
475 			} else {
476 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
477 					= nd_opt;
478 			}
479 			break;
480 		case ND_OPT_PREFIX_INFORMATION:
481 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
482 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
483 					= nd_opt;
484 			}
485 			ndopts->nd_opts_pi_end =
486 				(struct nd_opt_prefix_info *)nd_opt;
487 			break;
488 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
489 		case ND_OPT_RDNSS:	/* RFC 6106 */
490 		case ND_OPT_DNSSL:	/* RFC 6106 */
491 			/*
492 			 * Silently ignore options we know and do not care about
493 			 * in the kernel.
494 			 */
495 			break;
496 		default:
497 			/*
498 			 * Unknown options must be silently ignored,
499 			 * to accommodate future extension to the protocol.
500 			 */
501 			nd6log((LOG_DEBUG,
502 			    "nd6_options: unsupported option %d - "
503 			    "option ignored\n", nd_opt->nd_opt_type));
504 		}
505 
506 skip1:
507 		i++;
508 		if (i > V_nd6_maxndopt) {
509 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
510 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
511 			break;
512 		}
513 
514 		if (ndopts->nd_opts_done)
515 			break;
516 	}
517 
518 	return 0;
519 }
520 
521 /*
522  * ND6 timer routine to handle ND6 entries
523  */
524 static void
525 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
526 {
527 	int canceled;
528 
529 	LLE_WLOCK_ASSERT(ln);
530 
531 	if (tick < 0) {
532 		ln->la_expire = 0;
533 		ln->ln_ntick = 0;
534 		canceled = callout_stop(&ln->lle_timer);
535 	} else {
536 		ln->la_expire = time_uptime + tick / hz;
537 		LLE_ADDREF(ln);
538 		if (tick > INT_MAX) {
539 			ln->ln_ntick = tick - INT_MAX;
540 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
541 			    nd6_llinfo_timer, ln);
542 		} else {
543 			ln->ln_ntick = 0;
544 			canceled = callout_reset(&ln->lle_timer, tick,
545 			    nd6_llinfo_timer, ln);
546 		}
547 	}
548 	if (canceled > 0)
549 		LLE_REMREF(ln);
550 }
551 
552 /*
553  * Gets source address of the first packet in hold queue
554  * and stores it in @src.
555  * Returns pointer to @src (if hold queue is not empty) or NULL.
556  *
557  * Set noinline to be dtrace-friendly
558  */
559 static __noinline struct in6_addr *
560 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
561 {
562 	struct ip6_hdr hdr;
563 	struct mbuf *m;
564 
565 	if (ln->la_hold == NULL)
566 		return (NULL);
567 
568 	/*
569 	 * assume every packet in la_hold has the same IP header
570 	 */
571 	m = ln->la_hold;
572 	if (sizeof(hdr) > m->m_len)
573 		return (NULL);
574 
575 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
576 	*src = hdr.ip6_src;
577 
578 	return (src);
579 }
580 
581 /*
582  * Checks if we need to switch from STALE state.
583  *
584  * RFC 4861 requires switching from STALE to DELAY state
585  * on first packet matching entry, waiting V_nd6_delay and
586  * transition to PROBE state (if upper layer confirmation was
587  * not received).
588  *
589  * This code performs a bit differently:
590  * On packet hit we don't change state (but desired state
591  * can be guessed by control plane). However, after V_nd6_delay
592  * seconds code will transition to PROBE state (so DELAY state
593  * is kinda skipped in most situations).
594  *
595  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
596  * we perform the following upon entering STALE state:
597  *
598  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
599  * if packet was transmitted at the start of given interval, we
600  * would be able to switch to PROBE state in V_nd6_delay seconds
601  * as user expects.
602  *
603  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
604  * lle in STALE state (remaining timer value stored in lle_remtime).
605  *
606  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
607  * seconds ago.
608  *
609  * Returns non-zero value if the entry is still STALE (storing
610  * the next timer interval in @pdelay).
611  *
612  * Returns zero value if original timer expired or we need to switch to
613  * PROBE (store that in @do_switch variable).
614  */
615 static int
616 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
617 {
618 	int nd_delay, nd_gctimer, r_skip_req;
619 	time_t lle_hittime;
620 	long delay;
621 
622 	*do_switch = 0;
623 	nd_gctimer = V_nd6_gctimer;
624 	nd_delay = V_nd6_delay;
625 
626 	LLE_REQ_LOCK(lle);
627 	r_skip_req = lle->r_skip_req;
628 	lle_hittime = lle->lle_hittime;
629 	LLE_REQ_UNLOCK(lle);
630 
631 	if (r_skip_req > 0) {
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 			 * The original ng6_gctime timeout ended,
650 			 * no more rescheduling.
651 			 */
652 			return (0);
653 		}
654 
655 		*pdelay = delay;
656 		return (1);
657 	}
658 
659 	/*
660 	 * Packet received. Verify timestamp
661 	 */
662 	delay = (long)(time_uptime - lle_hittime);
663 	if (delay < nd_delay) {
664 		/*
665 		 * V_nd6_delay still not passed since the first
666 		 * hit in STALE state.
667 		 * Reshedule timer and return.
668 		 */
669 		*pdelay = (long)(nd_delay - delay) * hz;
670 		return (1);
671 	}
672 
673 	/* Request switching to probe */
674 	*do_switch = 1;
675 	return (0);
676 }
677 
678 /*
679  * Switch @lle state to new state optionally arming timers.
680  *
681  * Set noinline to be dtrace-friendly
682  */
683 __noinline void
684 nd6_llinfo_setstate(struct llentry *lle, int newstate)
685 {
686 	struct ifnet *ifp;
687 	int nd_gctimer, nd_delay;
688 	long delay, remtime;
689 
690 	delay = 0;
691 	remtime = 0;
692 
693 	switch (newstate) {
694 	case ND6_LLINFO_INCOMPLETE:
695 		ifp = lle->lle_tbl->llt_ifp;
696 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
697 		break;
698 	case ND6_LLINFO_REACHABLE:
699 		if (!ND6_LLINFO_PERMANENT(lle)) {
700 			ifp = lle->lle_tbl->llt_ifp;
701 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
702 		}
703 		break;
704 	case ND6_LLINFO_STALE:
705 
706 		/*
707 		 * Notify fast path that we want to know if any packet
708 		 * is transmitted by setting r_skip_req.
709 		 */
710 		LLE_REQ_LOCK(lle);
711 		lle->r_skip_req = 1;
712 		LLE_REQ_UNLOCK(lle);
713 		nd_delay = V_nd6_delay;
714 		nd_gctimer = V_nd6_gctimer;
715 
716 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
717 		remtime = (long)nd_gctimer * hz - delay;
718 		break;
719 	case ND6_LLINFO_DELAY:
720 		lle->la_asked = 0;
721 		delay = (long)V_nd6_delay * hz;
722 		break;
723 	}
724 
725 	if (delay > 0)
726 		nd6_llinfo_settimer_locked(lle, delay);
727 
728 	lle->lle_remtime = remtime;
729 	lle->ln_state = newstate;
730 }
731 
732 /*
733  * Timer-dependent part of nd state machine.
734  *
735  * Set noinline to be dtrace-friendly
736  */
737 static __noinline void
738 nd6_llinfo_timer(void *arg)
739 {
740 	struct epoch_tracker et;
741 	struct llentry *ln;
742 	struct in6_addr *dst, *pdst, *psrc, src;
743 	struct ifnet *ifp;
744 	struct nd_ifinfo *ndi;
745 	int do_switch, send_ns;
746 	long delay;
747 
748 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
749 	ln = (struct llentry *)arg;
750 	ifp = lltable_get_ifp(ln->lle_tbl);
751 	CURVNET_SET(ifp->if_vnet);
752 
753 	ND6_RLOCK();
754 	LLE_WLOCK(ln);
755 	if (callout_pending(&ln->lle_timer)) {
756 		/*
757 		 * Here we are a bit odd here in the treatment of
758 		 * active/pending. If the pending bit is set, it got
759 		 * rescheduled before I ran. The active
760 		 * bit we ignore, since if it was stopped
761 		 * in ll_tablefree() and was currently running
762 		 * it would have return 0 so the code would
763 		 * not have deleted it since the callout could
764 		 * not be stopped so we want to go through
765 		 * with the delete here now. If the callout
766 		 * was restarted, the pending bit will be back on and
767 		 * we just want to bail since the callout_reset would
768 		 * return 1 and our reference would have been removed
769 		 * by nd6_llinfo_settimer_locked above since canceled
770 		 * would have been 1.
771 		 */
772 		LLE_WUNLOCK(ln);
773 		ND6_RUNLOCK();
774 		CURVNET_RESTORE();
775 		return;
776 	}
777 	NET_EPOCH_ENTER(et);
778 	ndi = ND_IFINFO(ifp);
779 	send_ns = 0;
780 	dst = &ln->r_l3addr.addr6;
781 	pdst = dst;
782 
783 	if (ln->ln_ntick > 0) {
784 		if (ln->ln_ntick > INT_MAX) {
785 			ln->ln_ntick -= INT_MAX;
786 			nd6_llinfo_settimer_locked(ln, INT_MAX);
787 		} else {
788 			ln->ln_ntick = 0;
789 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
790 		}
791 		goto done;
792 	}
793 
794 	if (ln->la_flags & LLE_STATIC) {
795 		goto done;
796 	}
797 
798 	if (ln->la_flags & LLE_DELETED) {
799 		nd6_free(&ln, 0);
800 		goto done;
801 	}
802 
803 	switch (ln->ln_state) {
804 	case ND6_LLINFO_INCOMPLETE:
805 		if (ln->la_asked < V_nd6_mmaxtries) {
806 			ln->la_asked++;
807 			send_ns = 1;
808 			/* Send NS to multicast address */
809 			pdst = NULL;
810 		} else {
811 			struct mbuf *m = ln->la_hold;
812 			if (m) {
813 				struct mbuf *m0;
814 
815 				/*
816 				 * assuming every packet in la_hold has the
817 				 * same IP header.  Send error after unlock.
818 				 */
819 				m0 = m->m_nextpkt;
820 				m->m_nextpkt = NULL;
821 				ln->la_hold = m0;
822 				clear_llinfo_pqueue(ln);
823 			}
824 			nd6_free(&ln, 0);
825 			if (m != NULL) {
826 				struct mbuf *n = m;
827 
828 				/*
829 				 * if there are any ummapped mbufs, we
830 				 * must free them, rather than using
831 				 * them for an ICMP, as they cannot be
832 				 * checksummed.
833 				 */
834 				while ((n = n->m_next) != NULL) {
835 					if (n->m_flags & M_EXTPG)
836 						break;
837 				}
838 				if (n != NULL) {
839 					m_freem(m);
840 					m = NULL;
841 				} else {
842 					icmp6_error2(m, ICMP6_DST_UNREACH,
843 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
844 				}
845 			}
846 		}
847 		break;
848 	case ND6_LLINFO_REACHABLE:
849 		if (!ND6_LLINFO_PERMANENT(ln))
850 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
851 		break;
852 
853 	case ND6_LLINFO_STALE:
854 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
855 			/*
856 			 * No packet has used this entry and GC timeout
857 			 * has not been passed. Reshedule timer and
858 			 * return.
859 			 */
860 			nd6_llinfo_settimer_locked(ln, delay);
861 			break;
862 		}
863 
864 		if (do_switch == 0) {
865 			/*
866 			 * GC timer has ended and entry hasn't been used.
867 			 * Run Garbage collector (RFC 4861, 5.3)
868 			 */
869 			if (!ND6_LLINFO_PERMANENT(ln))
870 				nd6_free(&ln, 1);
871 			break;
872 		}
873 
874 		/* Entry has been used AND delay timer has ended. */
875 
876 		/* FALLTHROUGH */
877 
878 	case ND6_LLINFO_DELAY:
879 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
880 			/* We need NUD */
881 			ln->la_asked = 1;
882 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
883 			send_ns = 1;
884 		} else
885 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
886 		break;
887 	case ND6_LLINFO_PROBE:
888 		if (ln->la_asked < V_nd6_umaxtries) {
889 			ln->la_asked++;
890 			send_ns = 1;
891 		} else {
892 			nd6_free(&ln, 0);
893 		}
894 		break;
895 	default:
896 		panic("%s: paths in a dark night can be confusing: %d",
897 		    __func__, ln->ln_state);
898 	}
899 done:
900 	if (ln != NULL)
901 		ND6_RUNLOCK();
902 	if (send_ns != 0) {
903 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
904 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
905 		LLE_FREE_LOCKED(ln);
906 		ln = NULL;
907 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
908 	}
909 
910 	if (ln != NULL)
911 		LLE_FREE_LOCKED(ln);
912 	NET_EPOCH_EXIT(et);
913 	CURVNET_RESTORE();
914 }
915 
916 /*
917  * ND6 timer routine to expire default route list and prefix list
918  */
919 void
920 nd6_timer(void *arg)
921 {
922 	CURVNET_SET((struct vnet *) arg);
923 	struct epoch_tracker et;
924 	struct nd_prhead prl;
925 	struct nd_prefix *pr, *npr;
926 	struct ifnet *ifp;
927 	struct in6_ifaddr *ia6, *nia6;
928 	uint64_t genid;
929 
930 	LIST_INIT(&prl);
931 
932 	NET_EPOCH_ENTER(et);
933 	nd6_defrouter_timer();
934 
935 	/*
936 	 * expire interface addresses.
937 	 * in the past the loop was inside prefix expiry processing.
938 	 * However, from a stricter speci-confrmance standpoint, we should
939 	 * rather separate address lifetimes and prefix lifetimes.
940 	 *
941 	 * XXXRW: in6_ifaddrhead locking.
942 	 */
943   addrloop:
944 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
945 		/* check address lifetime */
946 		if (IFA6_IS_INVALID(ia6)) {
947 			int regen = 0;
948 
949 			/*
950 			 * If the expiring address is temporary, try
951 			 * regenerating a new one.  This would be useful when
952 			 * we suspended a laptop PC, then turned it on after a
953 			 * period that could invalidate all temporary
954 			 * addresses.  Although we may have to restart the
955 			 * loop (see below), it must be after purging the
956 			 * address.  Otherwise, we'd see an infinite loop of
957 			 * regeneration.
958 			 */
959 			if (V_ip6_use_tempaddr &&
960 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
961 				if (regen_tmpaddr(ia6) == 0)
962 					regen = 1;
963 			}
964 
965 			in6_purgeaddr(&ia6->ia_ifa);
966 
967 			if (regen)
968 				goto addrloop; /* XXX: see below */
969 		} else if (IFA6_IS_DEPRECATED(ia6)) {
970 			int oldflags = ia6->ia6_flags;
971 
972 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
973 
974 			/*
975 			 * If a temporary address has just become deprecated,
976 			 * regenerate a new one if possible.
977 			 */
978 			if (V_ip6_use_tempaddr &&
979 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
980 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
981 				if (regen_tmpaddr(ia6) == 0) {
982 					/*
983 					 * A new temporary address is
984 					 * generated.
985 					 * XXX: this means the address chain
986 					 * has changed while we are still in
987 					 * the loop.  Although the change
988 					 * would not cause disaster (because
989 					 * it's not a deletion, but an
990 					 * addition,) we'd rather restart the
991 					 * loop just for safety.  Or does this
992 					 * significantly reduce performance??
993 					 */
994 					goto addrloop;
995 				}
996 			}
997 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
998 			/*
999 			 * Schedule DAD for a tentative address.  This happens
1000 			 * if the interface was down or not running
1001 			 * when the address was configured.
1002 			 */
1003 			int delay;
1004 
1005 			delay = arc4random() %
1006 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1007 			nd6_dad_start((struct ifaddr *)ia6, delay);
1008 		} else {
1009 			/*
1010 			 * Check status of the interface.  If it is down,
1011 			 * mark the address as tentative for future DAD.
1012 			 */
1013 			ifp = ia6->ia_ifp;
1014 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1015 			    ((ifp->if_flags & IFF_UP) == 0 ||
1016 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1017 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1018 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1019 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1020 			}
1021 
1022 			/*
1023 			 * A new RA might have made a deprecated address
1024 			 * preferred.
1025 			 */
1026 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1027 		}
1028 	}
1029 	NET_EPOCH_EXIT(et);
1030 
1031 	ND6_WLOCK();
1032 restart:
1033 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1034 		/*
1035 		 * Expire prefixes. Since the pltime is only used for
1036 		 * autoconfigured addresses, pltime processing for prefixes is
1037 		 * not necessary.
1038 		 *
1039 		 * Only unlink after all derived addresses have expired. This
1040 		 * may not occur until two hours after the prefix has expired
1041 		 * per RFC 4862. If the prefix expires before its derived
1042 		 * addresses, mark it off-link. This will be done automatically
1043 		 * after unlinking if no address references remain.
1044 		 */
1045 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1046 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1047 			continue;
1048 
1049 		if (pr->ndpr_addrcnt == 0) {
1050 			nd6_prefix_unlink(pr, &prl);
1051 			continue;
1052 		}
1053 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1054 			genid = V_nd6_list_genid;
1055 			nd6_prefix_ref(pr);
1056 			ND6_WUNLOCK();
1057 			ND6_ONLINK_LOCK();
1058 			(void)nd6_prefix_offlink(pr);
1059 			ND6_ONLINK_UNLOCK();
1060 			ND6_WLOCK();
1061 			nd6_prefix_rele(pr);
1062 			if (genid != V_nd6_list_genid)
1063 				goto restart;
1064 		}
1065 	}
1066 	ND6_WUNLOCK();
1067 
1068 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1069 		LIST_REMOVE(pr, ndpr_entry);
1070 		nd6_prefix_del(pr);
1071 	}
1072 
1073 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1074 	    nd6_timer, curvnet);
1075 
1076 	CURVNET_RESTORE();
1077 }
1078 
1079 /*
1080  * ia6 - deprecated/invalidated temporary address
1081  */
1082 static int
1083 regen_tmpaddr(struct in6_ifaddr *ia6)
1084 {
1085 	struct ifaddr *ifa;
1086 	struct ifnet *ifp;
1087 	struct in6_ifaddr *public_ifa6 = NULL;
1088 
1089 	NET_EPOCH_ASSERT();
1090 
1091 	ifp = ia6->ia_ifa.ifa_ifp;
1092 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1093 		struct in6_ifaddr *it6;
1094 
1095 		if (ifa->ifa_addr->sa_family != AF_INET6)
1096 			continue;
1097 
1098 		it6 = (struct in6_ifaddr *)ifa;
1099 
1100 		/* ignore no autoconf addresses. */
1101 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1102 			continue;
1103 
1104 		/* ignore autoconf addresses with different prefixes. */
1105 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1106 			continue;
1107 
1108 		/*
1109 		 * Now we are looking at an autoconf address with the same
1110 		 * prefix as ours.  If the address is temporary and is still
1111 		 * preferred, do not create another one.  It would be rare, but
1112 		 * could happen, for example, when we resume a laptop PC after
1113 		 * a long period.
1114 		 */
1115 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1116 		    !IFA6_IS_DEPRECATED(it6)) {
1117 			public_ifa6 = NULL;
1118 			break;
1119 		}
1120 
1121 		/*
1122 		 * This is a public autoconf address that has the same prefix
1123 		 * as ours.  If it is preferred, keep it.  We can't break the
1124 		 * loop here, because there may be a still-preferred temporary
1125 		 * address with the prefix.
1126 		 */
1127 		if (!IFA6_IS_DEPRECATED(it6))
1128 			public_ifa6 = it6;
1129 	}
1130 	if (public_ifa6 != NULL)
1131 		ifa_ref(&public_ifa6->ia_ifa);
1132 
1133 	if (public_ifa6 != NULL) {
1134 		int e;
1135 
1136 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1137 			ifa_free(&public_ifa6->ia_ifa);
1138 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1139 			    " tmp addr,errno=%d\n", e);
1140 			return (-1);
1141 		}
1142 		ifa_free(&public_ifa6->ia_ifa);
1143 		return (0);
1144 	}
1145 
1146 	return (-1);
1147 }
1148 
1149 /*
1150  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1151  * cache entries are freed in in6_domifdetach().
1152  */
1153 void
1154 nd6_purge(struct ifnet *ifp)
1155 {
1156 	struct nd_prhead prl;
1157 	struct nd_prefix *pr, *npr;
1158 
1159 	LIST_INIT(&prl);
1160 
1161 	/* Purge default router list entries toward ifp. */
1162 	nd6_defrouter_purge(ifp);
1163 
1164 	ND6_WLOCK();
1165 	/*
1166 	 * Remove prefixes on ifp. We should have already removed addresses on
1167 	 * this interface, so no addresses should be referencing these prefixes.
1168 	 */
1169 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1170 		if (pr->ndpr_ifp == ifp)
1171 			nd6_prefix_unlink(pr, &prl);
1172 	}
1173 	ND6_WUNLOCK();
1174 
1175 	/* Delete the unlinked prefix objects. */
1176 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1177 		LIST_REMOVE(pr, ndpr_entry);
1178 		nd6_prefix_del(pr);
1179 	}
1180 
1181 	/* cancel default outgoing interface setting */
1182 	if (V_nd6_defifindex == ifp->if_index)
1183 		nd6_setdefaultiface(0);
1184 
1185 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1186 		/* Refresh default router list. */
1187 		defrouter_select_fib(ifp->if_fib);
1188 	}
1189 }
1190 
1191 /*
1192  * the caller acquires and releases the lock on the lltbls
1193  * Returns the llentry locked
1194  */
1195 struct llentry *
1196 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1197 {
1198 	struct sockaddr_in6 sin6;
1199 	struct llentry *ln;
1200 
1201 	bzero(&sin6, sizeof(sin6));
1202 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1203 	sin6.sin6_family = AF_INET6;
1204 	sin6.sin6_addr = *addr6;
1205 
1206 	IF_AFDATA_LOCK_ASSERT(ifp);
1207 
1208 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1209 
1210 	return (ln);
1211 }
1212 
1213 static struct llentry *
1214 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1215 {
1216 	struct sockaddr_in6 sin6;
1217 	struct llentry *ln;
1218 
1219 	bzero(&sin6, sizeof(sin6));
1220 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1221 	sin6.sin6_family = AF_INET6;
1222 	sin6.sin6_addr = *addr6;
1223 
1224 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1225 	if (ln != NULL)
1226 		ln->ln_state = ND6_LLINFO_NOSTATE;
1227 
1228 	return (ln);
1229 }
1230 
1231 /*
1232  * Test whether a given IPv6 address is a neighbor or not, ignoring
1233  * the actual neighbor cache.  The neighbor cache is ignored in order
1234  * to not reenter the routing code from within itself.
1235  */
1236 static int
1237 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1238 {
1239 	struct nd_prefix *pr;
1240 	struct ifaddr *ifa;
1241 	struct rt_addrinfo info;
1242 	struct sockaddr_in6 rt_key;
1243 	const struct sockaddr *dst6;
1244 	uint64_t genid;
1245 	int error, fibnum;
1246 
1247 	/*
1248 	 * A link-local address is always a neighbor.
1249 	 * XXX: a link does not necessarily specify a single interface.
1250 	 */
1251 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1252 		struct sockaddr_in6 sin6_copy;
1253 		u_int32_t zone;
1254 
1255 		/*
1256 		 * We need sin6_copy since sa6_recoverscope() may modify the
1257 		 * content (XXX).
1258 		 */
1259 		sin6_copy = *addr;
1260 		if (sa6_recoverscope(&sin6_copy))
1261 			return (0); /* XXX: should be impossible */
1262 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1263 			return (0);
1264 		if (sin6_copy.sin6_scope_id == zone)
1265 			return (1);
1266 		else
1267 			return (0);
1268 	}
1269 
1270 	bzero(&rt_key, sizeof(rt_key));
1271 	bzero(&info, sizeof(info));
1272 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1273 
1274 	/*
1275 	 * If the address matches one of our addresses,
1276 	 * it should be a neighbor.
1277 	 * If the address matches one of our on-link prefixes, it should be a
1278 	 * neighbor.
1279 	 */
1280 	ND6_RLOCK();
1281 restart:
1282 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1283 		if (pr->ndpr_ifp != ifp)
1284 			continue;
1285 
1286 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1287 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1288 
1289 			/*
1290 			 * We only need to check all FIBs if add_addr_allfibs
1291 			 * is unset. If set, checking any FIB will suffice.
1292 			 */
1293 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1294 			for (; fibnum < rt_numfibs; fibnum++) {
1295 				genid = V_nd6_list_genid;
1296 				ND6_RUNLOCK();
1297 
1298 				/*
1299 				 * Restore length field before
1300 				 * retrying lookup
1301 				 */
1302 				rt_key.sin6_len = sizeof(rt_key);
1303 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1304 						        &info);
1305 
1306 				ND6_RLOCK();
1307 				if (genid != V_nd6_list_genid)
1308 					goto restart;
1309 				if (error == 0)
1310 					break;
1311 			}
1312 			if (error != 0)
1313 				continue;
1314 
1315 			/*
1316 			 * This is the case where multiple interfaces
1317 			 * have the same prefix, but only one is installed
1318 			 * into the routing table and that prefix entry
1319 			 * is not the one being examined here. In the case
1320 			 * where RADIX_MPATH is enabled, multiple route
1321 			 * entries (of the same rt_key value) will be
1322 			 * installed because the interface addresses all
1323 			 * differ.
1324 			 */
1325 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1326 			    &rt_key.sin6_addr))
1327 				continue;
1328 		}
1329 
1330 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1331 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1332 			ND6_RUNLOCK();
1333 			return (1);
1334 		}
1335 	}
1336 	ND6_RUNLOCK();
1337 
1338 	/*
1339 	 * If the address is assigned on the node of the other side of
1340 	 * a p2p interface, the address should be a neighbor.
1341 	 */
1342 	if (ifp->if_flags & IFF_POINTOPOINT) {
1343 		struct epoch_tracker et;
1344 
1345 		NET_EPOCH_ENTER(et);
1346 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1347 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1348 				continue;
1349 			if (ifa->ifa_dstaddr != NULL &&
1350 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1351 				NET_EPOCH_EXIT(et);
1352 				return 1;
1353 			}
1354 		}
1355 		NET_EPOCH_EXIT(et);
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 	    nd6_defrouter_list_empty() &&
1364 	    V_nd6_defifindex == ifp->if_index) {
1365 		return (1);
1366 	}
1367 
1368 	return (0);
1369 }
1370 
1371 /*
1372  * Detect if a given IPv6 address identifies a neighbor on a given link.
1373  * XXX: should take care of the destination of a p2p link?
1374  */
1375 int
1376 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1377 {
1378 	struct llentry *lle;
1379 	int rc = 0;
1380 
1381 	NET_EPOCH_ASSERT();
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 ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1391 		LLE_RUNLOCK(lle);
1392 		rc = 1;
1393 	}
1394 	return (rc);
1395 }
1396 
1397 /*
1398  * Free an nd6 llinfo entry.
1399  * Since the function would cause significant changes in the kernel, DO NOT
1400  * make it global, unless you have a strong reason for the change, and are sure
1401  * that the change is safe.
1402  *
1403  * Set noinline to be dtrace-friendly
1404  */
1405 static __noinline void
1406 nd6_free(struct llentry **lnp, int gc)
1407 {
1408 	struct ifnet *ifp;
1409 	struct llentry *ln;
1410 	struct nd_defrouter *dr;
1411 
1412 	ln = *lnp;
1413 	*lnp = NULL;
1414 
1415 	LLE_WLOCK_ASSERT(ln);
1416 	ND6_RLOCK_ASSERT();
1417 
1418 	ifp = lltable_get_ifp(ln->lle_tbl);
1419 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1420 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1421 	else
1422 		dr = NULL;
1423 	ND6_RUNLOCK();
1424 
1425 	if ((ln->la_flags & LLE_DELETED) == 0)
1426 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1427 
1428 	/*
1429 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1430 	 * even though it is not harmful, it was not really necessary.
1431 	 */
1432 
1433 	/* cancel timer */
1434 	nd6_llinfo_settimer_locked(ln, -1);
1435 
1436 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1437 		if (dr != NULL && dr->expire &&
1438 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1439 			/*
1440 			 * If the reason for the deletion is just garbage
1441 			 * collection, and the neighbor is an active default
1442 			 * router, do not delete it.  Instead, reset the GC
1443 			 * timer using the router's lifetime.
1444 			 * Simply deleting the entry would affect default
1445 			 * router selection, which is not necessarily a good
1446 			 * thing, especially when we're using router preference
1447 			 * values.
1448 			 * XXX: the check for ln_state would be redundant,
1449 			 *      but we intentionally keep it just in case.
1450 			 */
1451 			if (dr->expire > time_uptime)
1452 				nd6_llinfo_settimer_locked(ln,
1453 				    (dr->expire - time_uptime) * hz);
1454 			else
1455 				nd6_llinfo_settimer_locked(ln,
1456 				    (long)V_nd6_gctimer * hz);
1457 
1458 			LLE_REMREF(ln);
1459 			LLE_WUNLOCK(ln);
1460 			defrouter_rele(dr);
1461 			return;
1462 		}
1463 
1464 		if (dr) {
1465 			/*
1466 			 * Unreachablity of a router might affect the default
1467 			 * router selection and on-link detection of advertised
1468 			 * prefixes.
1469 			 */
1470 
1471 			/*
1472 			 * Temporarily fake the state to choose a new default
1473 			 * router and to perform on-link determination of
1474 			 * prefixes correctly.
1475 			 * Below the state will be set correctly,
1476 			 * or the entry itself will be deleted.
1477 			 */
1478 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1479 		}
1480 
1481 		if (ln->ln_router || dr) {
1482 			/*
1483 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1484 			 * rnh and for the calls to pfxlist_onlink_check() and
1485 			 * defrouter_select_fib() in the block further down for calls
1486 			 * into nd6_lookup().  We still hold a ref.
1487 			 */
1488 			LLE_WUNLOCK(ln);
1489 
1490 			/*
1491 			 * rt6_flush must be called whether or not the neighbor
1492 			 * is in the Default Router List.
1493 			 * See a corresponding comment in nd6_na_input().
1494 			 */
1495 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1496 		}
1497 
1498 		if (dr) {
1499 			/*
1500 			 * Since defrouter_select_fib() does not affect the
1501 			 * on-link determination and MIP6 needs the check
1502 			 * before the default router selection, we perform
1503 			 * the check now.
1504 			 */
1505 			pfxlist_onlink_check();
1506 
1507 			/*
1508 			 * Refresh default router list.
1509 			 */
1510 			defrouter_select_fib(dr->ifp->if_fib);
1511 		}
1512 
1513 		/*
1514 		 * If this entry was added by an on-link redirect, remove the
1515 		 * corresponding host route.
1516 		 */
1517 		if (ln->la_flags & LLE_REDIRECT)
1518 			nd6_free_redirect(ln);
1519 
1520 		if (ln->ln_router || dr)
1521 			LLE_WLOCK(ln);
1522 	}
1523 
1524 	/*
1525 	 * Save to unlock. We still hold an extra reference and will not
1526 	 * free(9) in llentry_free() if someone else holds one as well.
1527 	 */
1528 	LLE_WUNLOCK(ln);
1529 	IF_AFDATA_LOCK(ifp);
1530 	LLE_WLOCK(ln);
1531 	/* Guard against race with other llentry_free(). */
1532 	if (ln->la_flags & LLE_LINKED) {
1533 		/* Remove callout reference */
1534 		LLE_REMREF(ln);
1535 		lltable_unlink_entry(ln->lle_tbl, ln);
1536 	}
1537 	IF_AFDATA_UNLOCK(ifp);
1538 
1539 	llentry_free(ln);
1540 	if (dr != NULL)
1541 		defrouter_rele(dr);
1542 }
1543 
1544 static int
1545 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1546 {
1547 
1548 	if (nh->nh_flags & NHF_REDIRECT)
1549 		return (1);
1550 
1551 	return (0);
1552 }
1553 
1554 /*
1555  * Remove the rtentry for the given llentry,
1556  * both of which were installed by a redirect.
1557  */
1558 static void
1559 nd6_free_redirect(const struct llentry *ln)
1560 {
1561 	int fibnum;
1562 	struct sockaddr_in6 sin6;
1563 	struct rt_addrinfo info;
1564 	struct rib_cmd_info rc;
1565 	struct epoch_tracker et;
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 	NET_EPOCH_ENTER(et);
1573 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1574 		rib_action(fibnum, RTM_DELETE, &info, &rc);
1575 	NET_EPOCH_EXIT(et);
1576 }
1577 
1578 /*
1579  * Updates status of the default router route.
1580  */
1581 static void
1582 check_release_defrouter(struct rib_cmd_info *rc, void *_cbdata)
1583 {
1584 	struct nd_defrouter *dr;
1585 	struct nhop_object *nh;
1586 
1587 	nh = rc->rc_nh_old;
1588 
1589 	if ((nh != NULL) && (nh->nh_flags & NHF_DEFAULT)) {
1590 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1591 		if (dr != NULL) {
1592 			dr->installed = 0;
1593 			defrouter_rele(dr);
1594 		}
1595 	}
1596 }
1597 
1598 void
1599 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1600 {
1601 
1602 #ifdef ROUTE_MPATH
1603 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1604 #else
1605 	check_release_defrouter(rc, NULL);
1606 #endif
1607 }
1608 
1609 int
1610 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1611 {
1612 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1613 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1614 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1615 	struct epoch_tracker et;
1616 	int error = 0;
1617 
1618 	if (ifp->if_afdata[AF_INET6] == NULL)
1619 		return (EPFNOSUPPORT);
1620 	switch (cmd) {
1621 	case OSIOCGIFINFO_IN6:
1622 #define ND	ndi->ndi
1623 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1624 		bzero(&ND, sizeof(ND));
1625 		ND.linkmtu = IN6_LINKMTU(ifp);
1626 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1627 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1628 		ND.reachable = ND_IFINFO(ifp)->reachable;
1629 		ND.retrans = ND_IFINFO(ifp)->retrans;
1630 		ND.flags = ND_IFINFO(ifp)->flags;
1631 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1632 		ND.chlim = ND_IFINFO(ifp)->chlim;
1633 		break;
1634 	case SIOCGIFINFO_IN6:
1635 		ND = *ND_IFINFO(ifp);
1636 		break;
1637 	case SIOCSIFINFO_IN6:
1638 		/*
1639 		 * used to change host variables from userland.
1640 		 * intended for a use on router to reflect RA configurations.
1641 		 */
1642 		/* 0 means 'unspecified' */
1643 		if (ND.linkmtu != 0) {
1644 			if (ND.linkmtu < IPV6_MMTU ||
1645 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1646 				error = EINVAL;
1647 				break;
1648 			}
1649 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1650 		}
1651 
1652 		if (ND.basereachable != 0) {
1653 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1654 
1655 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1656 			if (ND.basereachable != obasereachable)
1657 				ND_IFINFO(ifp)->reachable =
1658 				    ND_COMPUTE_RTIME(ND.basereachable);
1659 		}
1660 		if (ND.retrans != 0)
1661 			ND_IFINFO(ifp)->retrans = ND.retrans;
1662 		if (ND.chlim != 0)
1663 			ND_IFINFO(ifp)->chlim = ND.chlim;
1664 		/* FALLTHROUGH */
1665 	case SIOCSIFINFO_FLAGS:
1666 	{
1667 		struct ifaddr *ifa;
1668 		struct in6_ifaddr *ia;
1669 
1670 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1671 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1672 			/* ifdisabled 1->0 transision */
1673 
1674 			/*
1675 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1676 			 * has an link-local address with IN6_IFF_DUPLICATED,
1677 			 * do not clear ND6_IFF_IFDISABLED.
1678 			 * See RFC 4862, Section 5.4.5.
1679 			 */
1680 			NET_EPOCH_ENTER(et);
1681 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1682 				if (ifa->ifa_addr->sa_family != AF_INET6)
1683 					continue;
1684 				ia = (struct in6_ifaddr *)ifa;
1685 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1686 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1687 					break;
1688 			}
1689 			NET_EPOCH_EXIT(et);
1690 
1691 			if (ifa != NULL) {
1692 				/* LLA is duplicated. */
1693 				ND.flags |= ND6_IFF_IFDISABLED;
1694 				log(LOG_ERR, "Cannot enable an interface"
1695 				    " with a link-local address marked"
1696 				    " duplicate.\n");
1697 			} else {
1698 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1699 				if (ifp->if_flags & IFF_UP)
1700 					in6_if_up(ifp);
1701 			}
1702 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1703 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1704 			/* ifdisabled 0->1 transision */
1705 			/* Mark all IPv6 address as tentative. */
1706 
1707 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1708 			if (V_ip6_dad_count > 0 &&
1709 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1710 				NET_EPOCH_ENTER(et);
1711 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1712 				    ifa_link) {
1713 					if (ifa->ifa_addr->sa_family !=
1714 					    AF_INET6)
1715 						continue;
1716 					ia = (struct in6_ifaddr *)ifa;
1717 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1718 				}
1719 				NET_EPOCH_EXIT(et);
1720 			}
1721 		}
1722 
1723 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1724 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1725 				/* auto_linklocal 0->1 transision */
1726 
1727 				/* If no link-local address on ifp, configure */
1728 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1729 				in6_ifattach(ifp, NULL);
1730 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1731 			    ifp->if_flags & IFF_UP) {
1732 				/*
1733 				 * When the IF already has
1734 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1735 				 * address is assigned, and IFF_UP, try to
1736 				 * assign one.
1737 				 */
1738 				NET_EPOCH_ENTER(et);
1739 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1740 				    ifa_link) {
1741 					if (ifa->ifa_addr->sa_family !=
1742 					    AF_INET6)
1743 						continue;
1744 					ia = (struct in6_ifaddr *)ifa;
1745 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1746 						break;
1747 				}
1748 				NET_EPOCH_EXIT(et);
1749 				if (ifa != NULL)
1750 					/* No LLA is configured. */
1751 					in6_ifattach(ifp, NULL);
1752 			}
1753 		}
1754 		ND_IFINFO(ifp)->flags = ND.flags;
1755 		break;
1756 	}
1757 #undef ND
1758 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1759 		/* sync kernel routing table with the default router list */
1760 		defrouter_reset();
1761 		defrouter_select_fib(RT_ALL_FIBS);
1762 		break;
1763 	case SIOCSPFXFLUSH_IN6:
1764 	{
1765 		/* flush all the prefix advertised by routers */
1766 		struct in6_ifaddr *ia, *ia_next;
1767 		struct nd_prefix *pr, *next;
1768 		struct nd_prhead prl;
1769 
1770 		LIST_INIT(&prl);
1771 
1772 		ND6_WLOCK();
1773 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1774 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1775 				continue; /* XXX */
1776 			nd6_prefix_unlink(pr, &prl);
1777 		}
1778 		ND6_WUNLOCK();
1779 
1780 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1781 			LIST_REMOVE(pr, ndpr_entry);
1782 			/* XXXRW: in6_ifaddrhead locking. */
1783 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1784 			    ia_next) {
1785 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1786 					continue;
1787 
1788 				if (ia->ia6_ndpr == pr)
1789 					in6_purgeaddr(&ia->ia_ifa);
1790 			}
1791 			nd6_prefix_del(pr);
1792 		}
1793 		break;
1794 	}
1795 	case SIOCSRTRFLUSH_IN6:
1796 	{
1797 		/* flush all the default routers */
1798 
1799 		defrouter_reset();
1800 		nd6_defrouter_flush_all();
1801 		defrouter_select_fib(RT_ALL_FIBS);
1802 		break;
1803 	}
1804 	case SIOCGNBRINFO_IN6:
1805 	{
1806 		struct llentry *ln;
1807 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1808 
1809 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1810 			return (error);
1811 
1812 		NET_EPOCH_ENTER(et);
1813 		ln = nd6_lookup(&nb_addr, 0, ifp);
1814 		NET_EPOCH_EXIT(et);
1815 
1816 		if (ln == NULL) {
1817 			error = EINVAL;
1818 			break;
1819 		}
1820 		nbi->state = ln->ln_state;
1821 		nbi->asked = ln->la_asked;
1822 		nbi->isrouter = ln->ln_router;
1823 		if (ln->la_expire == 0)
1824 			nbi->expire = 0;
1825 		else
1826 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1827 			    (time_second - time_uptime);
1828 		LLE_RUNLOCK(ln);
1829 		break;
1830 	}
1831 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1832 		ndif->ifindex = V_nd6_defifindex;
1833 		break;
1834 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1835 		return (nd6_setdefaultiface(ndif->ifindex));
1836 	}
1837 	return (error);
1838 }
1839 
1840 /*
1841  * Calculates new isRouter value based on provided parameters and
1842  * returns it.
1843  */
1844 static int
1845 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1846     int ln_router)
1847 {
1848 
1849 	/*
1850 	 * ICMP6 type dependent behavior.
1851 	 *
1852 	 * NS: clear IsRouter if new entry
1853 	 * RS: clear IsRouter
1854 	 * RA: set IsRouter if there's lladdr
1855 	 * redir: clear IsRouter if new entry
1856 	 *
1857 	 * RA case, (1):
1858 	 * The spec says that we must set IsRouter in the following cases:
1859 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1860 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1861 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1862 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1863 	 * neighbor cache, this is similar to (6).
1864 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1865 	 *
1866 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1867 	 *							D R
1868 	 *	0	n	n	(1)	c   ?     s
1869 	 *	0	y	n	(2)	c   s     s
1870 	 *	0	n	y	(3)	c   s     s
1871 	 *	0	y	y	(4)	c   s     s
1872 	 *	0	y	y	(5)	c   s     s
1873 	 *	1	--	n	(6) c	c	c s
1874 	 *	1	--	y	(7) c	c   s	c s
1875 	 *
1876 	 *					(c=clear s=set)
1877 	 */
1878 	switch (type & 0xff) {
1879 	case ND_NEIGHBOR_SOLICIT:
1880 		/*
1881 		 * New entry must have is_router flag cleared.
1882 		 */
1883 		if (is_new)					/* (6-7) */
1884 			ln_router = 0;
1885 		break;
1886 	case ND_REDIRECT:
1887 		/*
1888 		 * If the icmp is a redirect to a better router, always set the
1889 		 * is_router flag.  Otherwise, if the entry is newly created,
1890 		 * clear the flag.  [RFC 2461, sec 8.3]
1891 		 */
1892 		if (code == ND_REDIRECT_ROUTER)
1893 			ln_router = 1;
1894 		else {
1895 			if (is_new)				/* (6-7) */
1896 				ln_router = 0;
1897 		}
1898 		break;
1899 	case ND_ROUTER_SOLICIT:
1900 		/*
1901 		 * is_router flag must always be cleared.
1902 		 */
1903 		ln_router = 0;
1904 		break;
1905 	case ND_ROUTER_ADVERT:
1906 		/*
1907 		 * Mark an entry with lladdr as a router.
1908 		 */
1909 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1910 		    (is_new && new_addr)) {			/* (7) */
1911 			ln_router = 1;
1912 		}
1913 		break;
1914 	}
1915 
1916 	return (ln_router);
1917 }
1918 
1919 /*
1920  * Create neighbor cache entry and cache link-layer address,
1921  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1922  *
1923  * type - ICMP6 type
1924  * code - type dependent information
1925  *
1926  */
1927 void
1928 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1929     int lladdrlen, int type, int code)
1930 {
1931 	struct llentry *ln = NULL, *ln_tmp;
1932 	int is_newentry;
1933 	int do_update;
1934 	int olladdr;
1935 	int llchange;
1936 	int flags;
1937 	uint16_t router = 0;
1938 	struct sockaddr_in6 sin6;
1939 	struct mbuf *chain = NULL;
1940 	u_char linkhdr[LLE_MAX_LINKHDR];
1941 	size_t linkhdrsize;
1942 	int lladdr_off;
1943 
1944 	NET_EPOCH_ASSERT();
1945 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1946 
1947 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1948 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1949 
1950 	/* nothing must be updated for unspecified address */
1951 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1952 		return;
1953 
1954 	/*
1955 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1956 	 * the caller.
1957 	 *
1958 	 * XXX If the link does not have link-layer adderss, what should
1959 	 * we do? (ifp->if_addrlen == 0)
1960 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1961 	 * description on it in NS section (RFC 2461 7.2.3).
1962 	 */
1963 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1964 	ln = nd6_lookup(from, flags, ifp);
1965 	is_newentry = 0;
1966 	if (ln == NULL) {
1967 		flags |= LLE_EXCLUSIVE;
1968 		ln = nd6_alloc(from, 0, ifp);
1969 		if (ln == NULL)
1970 			return;
1971 
1972 		/*
1973 		 * Since we already know all the data for the new entry,
1974 		 * fill it before insertion.
1975 		 */
1976 		if (lladdr != NULL) {
1977 			linkhdrsize = sizeof(linkhdr);
1978 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1979 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1980 				return;
1981 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1982 			    lladdr_off);
1983 		}
1984 
1985 		IF_AFDATA_WLOCK(ifp);
1986 		LLE_WLOCK(ln);
1987 		/* Prefer any existing lle over newly-created one */
1988 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1989 		if (ln_tmp == NULL)
1990 			lltable_link_entry(LLTABLE6(ifp), ln);
1991 		IF_AFDATA_WUNLOCK(ifp);
1992 		if (ln_tmp == NULL) {
1993 			/* No existing lle, mark as new entry (6,7) */
1994 			is_newentry = 1;
1995 			if (lladdr != NULL) {	/* (7) */
1996 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1997 				EVENTHANDLER_INVOKE(lle_event, ln,
1998 				    LLENTRY_RESOLVED);
1999 			}
2000 		} else {
2001 			lltable_free_entry(LLTABLE6(ifp), ln);
2002 			ln = ln_tmp;
2003 			ln_tmp = NULL;
2004 		}
2005 	}
2006 	/* do nothing if static ndp is set */
2007 	if ((ln->la_flags & LLE_STATIC)) {
2008 		if (flags & LLE_EXCLUSIVE)
2009 			LLE_WUNLOCK(ln);
2010 		else
2011 			LLE_RUNLOCK(ln);
2012 		return;
2013 	}
2014 
2015 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2016 	if (olladdr && lladdr) {
2017 		llchange = bcmp(lladdr, ln->ll_addr,
2018 		    ifp->if_addrlen);
2019 	} else if (!olladdr && lladdr)
2020 		llchange = 1;
2021 	else
2022 		llchange = 0;
2023 
2024 	/*
2025 	 * newentry olladdr  lladdr  llchange	(*=record)
2026 	 *	0	n	n	--	(1)
2027 	 *	0	y	n	--	(2)
2028 	 *	0	n	y	y	(3) * STALE
2029 	 *	0	y	y	n	(4) *
2030 	 *	0	y	y	y	(5) * STALE
2031 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2032 	 *	1	--	y	--	(7) * STALE
2033 	 */
2034 
2035 	do_update = 0;
2036 	if (is_newentry == 0 && llchange != 0) {
2037 		do_update = 1;	/* (3,5) */
2038 
2039 		/*
2040 		 * Record source link-layer address
2041 		 * XXX is it dependent to ifp->if_type?
2042 		 */
2043 		linkhdrsize = sizeof(linkhdr);
2044 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2045 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2046 			return;
2047 
2048 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2049 		    lladdr_off) == 0) {
2050 			/* Entry was deleted */
2051 			return;
2052 		}
2053 
2054 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2055 
2056 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2057 
2058 		if (ln->la_hold != NULL)
2059 			nd6_grab_holdchain(ln, &chain, &sin6);
2060 	}
2061 
2062 	/* Calculates new router status */
2063 	router = nd6_is_router(type, code, is_newentry, olladdr,
2064 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2065 
2066 	ln->ln_router = router;
2067 	/* Mark non-router redirects with special flag */
2068 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2069 		ln->la_flags |= LLE_REDIRECT;
2070 
2071 	if (flags & LLE_EXCLUSIVE)
2072 		LLE_WUNLOCK(ln);
2073 	else
2074 		LLE_RUNLOCK(ln);
2075 
2076 	if (chain != NULL)
2077 		nd6_flush_holdchain(ifp, chain, &sin6);
2078 
2079 	/*
2080 	 * When the link-layer address of a router changes, select the
2081 	 * best router again.  In particular, when the neighbor entry is newly
2082 	 * created, it might affect the selection policy.
2083 	 * Question: can we restrict the first condition to the "is_newentry"
2084 	 * case?
2085 	 * XXX: when we hear an RA from a new router with the link-layer
2086 	 * address option, defrouter_select_fib() is called twice, since
2087 	 * defrtrlist_update called the function as well.  However, I believe
2088 	 * we can compromise the overhead, since it only happens the first
2089 	 * time.
2090 	 * XXX: although defrouter_select_fib() should not have a bad effect
2091 	 * for those are not autoconfigured hosts, we explicitly avoid such
2092 	 * cases for safety.
2093 	 */
2094 	if ((do_update || is_newentry) && router &&
2095 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2096 		/*
2097 		 * guaranteed recursion
2098 		 */
2099 		defrouter_select_fib(ifp->if_fib);
2100 	}
2101 }
2102 
2103 static void
2104 nd6_slowtimo(void *arg)
2105 {
2106 	struct epoch_tracker et;
2107 	CURVNET_SET((struct vnet *) arg);
2108 	struct nd_ifinfo *nd6if;
2109 	struct ifnet *ifp;
2110 
2111 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2112 	    nd6_slowtimo, curvnet);
2113 	NET_EPOCH_ENTER(et);
2114 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2115 		if (ifp->if_afdata[AF_INET6] == NULL)
2116 			continue;
2117 		nd6if = ND_IFINFO(ifp);
2118 		if (nd6if->basereachable && /* already initialized */
2119 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2120 			/*
2121 			 * Since reachable time rarely changes by router
2122 			 * advertisements, we SHOULD insure that a new random
2123 			 * value gets recomputed at least once every few hours.
2124 			 * (RFC 2461, 6.3.4)
2125 			 */
2126 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2127 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2128 		}
2129 	}
2130 	NET_EPOCH_EXIT(et);
2131 	CURVNET_RESTORE();
2132 }
2133 
2134 void
2135 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2136     struct sockaddr_in6 *sin6)
2137 {
2138 
2139 	LLE_WLOCK_ASSERT(ln);
2140 
2141 	*chain = ln->la_hold;
2142 	ln->la_hold = NULL;
2143 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2144 
2145 	if (ln->ln_state == ND6_LLINFO_STALE) {
2146 		/*
2147 		 * The first time we send a packet to a
2148 		 * neighbor whose entry is STALE, we have
2149 		 * to change the state to DELAY and a sets
2150 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2151 		 * seconds to ensure do neighbor unreachability
2152 		 * detection on expiration.
2153 		 * (RFC 2461 7.3.3)
2154 		 */
2155 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2156 	}
2157 }
2158 
2159 int
2160 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2161     struct sockaddr_in6 *dst, struct route *ro)
2162 {
2163 	int error;
2164 	int ip6len;
2165 	struct ip6_hdr *ip6;
2166 	struct m_tag *mtag;
2167 
2168 #ifdef MAC
2169 	mac_netinet6_nd6_send(ifp, m);
2170 #endif
2171 
2172 	/*
2173 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2174 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2175 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2176 	 * to be diverted to user space.  When re-injected into the kernel,
2177 	 * send_output() will directly dispatch them to the outgoing interface.
2178 	 */
2179 	if (send_sendso_input_hook != NULL) {
2180 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2181 		if (mtag != NULL) {
2182 			ip6 = mtod(m, struct ip6_hdr *);
2183 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2184 			/* Use the SEND socket */
2185 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2186 			    ip6len);
2187 			/* -1 == no app on SEND socket */
2188 			if (error == 0 || error != -1)
2189 			    return (error);
2190 		}
2191 	}
2192 
2193 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2194 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2195 	    mtod(m, struct ip6_hdr *));
2196 
2197 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2198 		origifp = ifp;
2199 
2200 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2201 	return (error);
2202 }
2203 
2204 /*
2205  * Lookup link headerfor @sa_dst address. Stores found
2206  * data in @desten buffer. Copy of lle ln_flags can be also
2207  * saved in @pflags if @pflags is non-NULL.
2208  *
2209  * If destination LLE does not exists or lle state modification
2210  * is required, call "slow" version.
2211  *
2212  * Return values:
2213  * - 0 on success (address copied to buffer).
2214  * - EWOULDBLOCK (no local error, but address is still unresolved)
2215  * - other errors (alloc failure, etc)
2216  */
2217 int
2218 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2219     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2220     struct llentry **plle)
2221 {
2222 	struct llentry *ln = NULL;
2223 	const struct sockaddr_in6 *dst6;
2224 
2225 	NET_EPOCH_ASSERT();
2226 
2227 	if (pflags != NULL)
2228 		*pflags = 0;
2229 
2230 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2231 
2232 	/* discard the packet if IPv6 operation is disabled on the interface */
2233 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2234 		m_freem(m);
2235 		return (ENETDOWN); /* better error? */
2236 	}
2237 
2238 	if (m != NULL && m->m_flags & M_MCAST) {
2239 		switch (ifp->if_type) {
2240 		case IFT_ETHER:
2241 		case IFT_L2VLAN:
2242 		case IFT_BRIDGE:
2243 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2244 						 desten);
2245 			return (0);
2246 		default:
2247 			m_freem(m);
2248 			return (EAFNOSUPPORT);
2249 		}
2250 	}
2251 
2252 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2253 	    ifp);
2254 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2255 		/* Entry found, let's copy lle info */
2256 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2257 		if (pflags != NULL)
2258 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2259 		/* Check if we have feedback request from nd6 timer */
2260 		if (ln->r_skip_req != 0) {
2261 			LLE_REQ_LOCK(ln);
2262 			ln->r_skip_req = 0; /* Notify that entry was used */
2263 			ln->lle_hittime = time_uptime;
2264 			LLE_REQ_UNLOCK(ln);
2265 		}
2266 		if (plle) {
2267 			LLE_ADDREF(ln);
2268 			*plle = ln;
2269 			LLE_WUNLOCK(ln);
2270 		}
2271 		return (0);
2272 	} else if (plle && ln)
2273 		LLE_WUNLOCK(ln);
2274 
2275 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2276 }
2277 
2278 /*
2279  * Do L2 address resolution for @sa_dst address. Stores found
2280  * address in @desten buffer. Copy of lle ln_flags can be also
2281  * saved in @pflags if @pflags is non-NULL.
2282  *
2283  * Heavy version.
2284  * Function assume that destination LLE does not exist,
2285  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2286  *
2287  * Set noinline to be dtrace-friendly
2288  */
2289 static __noinline int
2290 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2291     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2292     struct llentry **plle)
2293 {
2294 	struct llentry *lle = NULL, *lle_tmp;
2295 	struct in6_addr *psrc, src;
2296 	int send_ns, ll_len;
2297 	char *lladdr;
2298 
2299 	NET_EPOCH_ASSERT();
2300 
2301 	/*
2302 	 * Address resolution or Neighbor Unreachability Detection
2303 	 * for the next hop.
2304 	 * At this point, the destination of the packet must be a unicast
2305 	 * or an anycast address(i.e. not a multicast).
2306 	 */
2307 	if (lle == NULL) {
2308 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2309 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2310 			/*
2311 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2312 			 * the condition below is not very efficient.  But we believe
2313 			 * it is tolerable, because this should be a rare case.
2314 			 */
2315 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2316 			if (lle == NULL) {
2317 				char ip6buf[INET6_ADDRSTRLEN];
2318 				log(LOG_DEBUG,
2319 				    "nd6_output: can't allocate llinfo for %s "
2320 				    "(ln=%p)\n",
2321 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2322 				m_freem(m);
2323 				return (ENOBUFS);
2324 			}
2325 
2326 			IF_AFDATA_WLOCK(ifp);
2327 			LLE_WLOCK(lle);
2328 			/* Prefer any existing entry over newly-created one */
2329 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2330 			if (lle_tmp == NULL)
2331 				lltable_link_entry(LLTABLE6(ifp), lle);
2332 			IF_AFDATA_WUNLOCK(ifp);
2333 			if (lle_tmp != NULL) {
2334 				lltable_free_entry(LLTABLE6(ifp), lle);
2335 				lle = lle_tmp;
2336 				lle_tmp = NULL;
2337 			}
2338 		}
2339 	}
2340 	if (lle == NULL) {
2341 		m_freem(m);
2342 		return (ENOBUFS);
2343 	}
2344 
2345 	LLE_WLOCK_ASSERT(lle);
2346 
2347 	/*
2348 	 * The first time we send a packet to a neighbor whose entry is
2349 	 * STALE, we have to change the state to DELAY and a sets a timer to
2350 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2351 	 * neighbor unreachability detection on expiration.
2352 	 * (RFC 2461 7.3.3)
2353 	 */
2354 	if (lle->ln_state == ND6_LLINFO_STALE)
2355 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2356 
2357 	/*
2358 	 * If the neighbor cache entry has a state other than INCOMPLETE
2359 	 * (i.e. its link-layer address is already resolved), just
2360 	 * send the packet.
2361 	 */
2362 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2363 		if (flags & LLE_ADDRONLY) {
2364 			lladdr = lle->ll_addr;
2365 			ll_len = ifp->if_addrlen;
2366 		} else {
2367 			lladdr = lle->r_linkdata;
2368 			ll_len = lle->r_hdrlen;
2369 		}
2370 		bcopy(lladdr, desten, ll_len);
2371 		if (pflags != NULL)
2372 			*pflags = lle->la_flags;
2373 		if (plle) {
2374 			LLE_ADDREF(lle);
2375 			*plle = lle;
2376 		}
2377 		LLE_WUNLOCK(lle);
2378 		return (0);
2379 	}
2380 
2381 	/*
2382 	 * There is a neighbor cache entry, but no ethernet address
2383 	 * response yet.  Append this latest packet to the end of the
2384 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2385 	 * the oldest packet in the queue will be removed.
2386 	 */
2387 
2388 	if (lle->la_hold != NULL) {
2389 		struct mbuf *m_hold;
2390 		int i;
2391 
2392 		i = 0;
2393 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2394 			i++;
2395 			if (m_hold->m_nextpkt == NULL) {
2396 				m_hold->m_nextpkt = m;
2397 				break;
2398 			}
2399 		}
2400 		while (i >= V_nd6_maxqueuelen) {
2401 			m_hold = lle->la_hold;
2402 			lle->la_hold = lle->la_hold->m_nextpkt;
2403 			m_freem(m_hold);
2404 			i--;
2405 		}
2406 	} else {
2407 		lle->la_hold = m;
2408 	}
2409 
2410 	/*
2411 	 * If there has been no NS for the neighbor after entering the
2412 	 * INCOMPLETE state, send the first solicitation.
2413 	 * Note that for newly-created lle la_asked will be 0,
2414 	 * so we will transition from ND6_LLINFO_NOSTATE to
2415 	 * ND6_LLINFO_INCOMPLETE state here.
2416 	 */
2417 	psrc = NULL;
2418 	send_ns = 0;
2419 	if (lle->la_asked == 0) {
2420 		lle->la_asked++;
2421 		send_ns = 1;
2422 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2423 
2424 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2425 	}
2426 	LLE_WUNLOCK(lle);
2427 	if (send_ns != 0)
2428 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2429 
2430 	return (EWOULDBLOCK);
2431 }
2432 
2433 /*
2434  * Do L2 address resolution for @sa_dst address. Stores found
2435  * address in @desten buffer. Copy of lle ln_flags can be also
2436  * saved in @pflags if @pflags is non-NULL.
2437  *
2438  * Return values:
2439  * - 0 on success (address copied to buffer).
2440  * - EWOULDBLOCK (no local error, but address is still unresolved)
2441  * - other errors (alloc failure, etc)
2442  */
2443 int
2444 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2445     char *desten, uint32_t *pflags)
2446 {
2447 	int error;
2448 
2449 	flags |= LLE_ADDRONLY;
2450 	error = nd6_resolve_slow(ifp, flags, NULL,
2451 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2452 	return (error);
2453 }
2454 
2455 int
2456 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2457     struct sockaddr_in6 *dst)
2458 {
2459 	struct mbuf *m, *m_head;
2460 	int error = 0;
2461 
2462 	m_head = chain;
2463 
2464 	while (m_head) {
2465 		m = m_head;
2466 		m_head = m_head->m_nextpkt;
2467 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2468 	}
2469 
2470 	/*
2471 	 * XXX
2472 	 * note that intermediate errors are blindly ignored
2473 	 */
2474 	return (error);
2475 }
2476 
2477 static int
2478 nd6_need_cache(struct ifnet *ifp)
2479 {
2480 	/*
2481 	 * XXX: we currently do not make neighbor cache on any interface
2482 	 * other than Ethernet and GIF.
2483 	 *
2484 	 * RFC2893 says:
2485 	 * - unidirectional tunnels needs no ND
2486 	 */
2487 	switch (ifp->if_type) {
2488 	case IFT_ETHER:
2489 	case IFT_IEEE1394:
2490 	case IFT_L2VLAN:
2491 	case IFT_INFINIBAND:
2492 	case IFT_BRIDGE:
2493 	case IFT_PROPVIRTUAL:
2494 		return (1);
2495 	default:
2496 		return (0);
2497 	}
2498 }
2499 
2500 /*
2501  * Add pernament ND6 link-layer record for given
2502  * interface address.
2503  *
2504  * Very similar to IPv4 arp_ifinit(), but:
2505  * 1) IPv6 DAD is performed in different place
2506  * 2) It is called by IPv6 protocol stack in contrast to
2507  * arp_ifinit() which is typically called in SIOCSIFADDR
2508  * driver ioctl handler.
2509  *
2510  */
2511 int
2512 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2513 {
2514 	struct ifnet *ifp;
2515 	struct llentry *ln, *ln_tmp;
2516 	struct sockaddr *dst;
2517 
2518 	ifp = ia->ia_ifa.ifa_ifp;
2519 	if (nd6_need_cache(ifp) == 0)
2520 		return (0);
2521 
2522 	dst = (struct sockaddr *)&ia->ia_addr;
2523 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2524 	if (ln == NULL)
2525 		return (ENOBUFS);
2526 
2527 	IF_AFDATA_WLOCK(ifp);
2528 	LLE_WLOCK(ln);
2529 	/* Unlink any entry if exists */
2530 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2531 	if (ln_tmp != NULL)
2532 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2533 	lltable_link_entry(LLTABLE6(ifp), ln);
2534 	IF_AFDATA_WUNLOCK(ifp);
2535 
2536 	if (ln_tmp != NULL)
2537 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2538 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2539 
2540 	LLE_WUNLOCK(ln);
2541 	if (ln_tmp != NULL)
2542 		llentry_free(ln_tmp);
2543 
2544 	return (0);
2545 }
2546 
2547 /*
2548  * Removes either all lle entries for given @ia, or lle
2549  * corresponding to @ia address.
2550  */
2551 void
2552 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2553 {
2554 	struct sockaddr_in6 mask, addr;
2555 	struct sockaddr *saddr, *smask;
2556 	struct ifnet *ifp;
2557 
2558 	ifp = ia->ia_ifa.ifa_ifp;
2559 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2560 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2561 	saddr = (struct sockaddr *)&addr;
2562 	smask = (struct sockaddr *)&mask;
2563 
2564 	if (all != 0)
2565 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2566 	else
2567 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2568 }
2569 
2570 static void
2571 clear_llinfo_pqueue(struct llentry *ln)
2572 {
2573 	struct mbuf *m_hold, *m_hold_next;
2574 
2575 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2576 		m_hold_next = m_hold->m_nextpkt;
2577 		m_freem(m_hold);
2578 	}
2579 
2580 	ln->la_hold = NULL;
2581 }
2582 
2583 static int
2584 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2585 {
2586 	struct in6_prefix p;
2587 	struct sockaddr_in6 s6;
2588 	struct nd_prefix *pr;
2589 	struct nd_pfxrouter *pfr;
2590 	time_t maxexpire;
2591 	int error;
2592 	char ip6buf[INET6_ADDRSTRLEN];
2593 
2594 	if (req->newptr)
2595 		return (EPERM);
2596 
2597 	error = sysctl_wire_old_buffer(req, 0);
2598 	if (error != 0)
2599 		return (error);
2600 
2601 	bzero(&p, sizeof(p));
2602 	p.origin = PR_ORIG_RA;
2603 	bzero(&s6, sizeof(s6));
2604 	s6.sin6_family = AF_INET6;
2605 	s6.sin6_len = sizeof(s6);
2606 
2607 	ND6_RLOCK();
2608 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2609 		p.prefix = pr->ndpr_prefix;
2610 		if (sa6_recoverscope(&p.prefix)) {
2611 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2612 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2613 			/* XXX: press on... */
2614 		}
2615 		p.raflags = pr->ndpr_raf;
2616 		p.prefixlen = pr->ndpr_plen;
2617 		p.vltime = pr->ndpr_vltime;
2618 		p.pltime = pr->ndpr_pltime;
2619 		p.if_index = pr->ndpr_ifp->if_index;
2620 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2621 			p.expire = 0;
2622 		else {
2623 			/* XXX: we assume time_t is signed. */
2624 			maxexpire = (-1) &
2625 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2626 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2627 				p.expire = pr->ndpr_lastupdate +
2628 				    pr->ndpr_vltime +
2629 				    (time_second - time_uptime);
2630 			else
2631 				p.expire = maxexpire;
2632 		}
2633 		p.refcnt = pr->ndpr_addrcnt;
2634 		p.flags = pr->ndpr_stateflags;
2635 		p.advrtrs = 0;
2636 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2637 			p.advrtrs++;
2638 		error = SYSCTL_OUT(req, &p, sizeof(p));
2639 		if (error != 0)
2640 			break;
2641 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2642 			s6.sin6_addr = pfr->router->rtaddr;
2643 			if (sa6_recoverscope(&s6))
2644 				log(LOG_ERR,
2645 				    "scope error in prefix list (%s)\n",
2646 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2647 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2648 			if (error != 0)
2649 				goto out;
2650 		}
2651 	}
2652 out:
2653 	ND6_RUNLOCK();
2654 	return (error);
2655 }
2656 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2657 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2658 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2659 	"NDP prefix list");
2660 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2661 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2662 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2663 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2664