xref: /freebsd/sys/netinet6/nd6.c (revision 6469bdcdb6a5968dc7edfcfb495d427b4bfdb3dd)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/callout.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/mutex.h>
47 #include <sys/socket.h>
48 #include <sys/sockio.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/protosw.h>
52 #include <sys/errno.h>
53 #include <sys/syslog.h>
54 #include <sys/rwlock.h>
55 #include <sys/queue.h>
56 #include <sys/sdt.h>
57 #include <sys/sysctl.h>
58 
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_arc.h>
62 #include <net/if_dl.h>
63 #include <net/if_types.h>
64 #include <net/fddi.h>
65 #include <net/route.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_kdtrace.h>
70 #include <net/if_llatbl.h>
71 #include <netinet/if_ether.h>
72 #include <netinet6/in6_var.h>
73 #include <netinet/ip6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/scope6_var.h>
76 #include <netinet6/nd6.h>
77 #include <netinet6/in6_ifattach.h>
78 #include <netinet/icmp6.h>
79 #include <netinet6/send.h>
80 
81 #include <sys/limits.h>
82 
83 #include <security/mac/mac_framework.h>
84 
85 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
86 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
87 
88 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
89 
90 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
91 
92 /* timer values */
93 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
94 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
95 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
96 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
97 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
98 					 * local traffic */
99 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
100 					 * collection timer */
101 
102 /* preventing too many loops in ND option parsing */
103 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
104 
105 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
106 					 * layer hints */
107 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
108 					 * ND entries */
109 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
110 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
111 
112 #ifdef ND6_DEBUG
113 VNET_DEFINE(int, nd6_debug) = 1;
114 #else
115 VNET_DEFINE(int, nd6_debug) = 0;
116 #endif
117 
118 static eventhandler_tag lle_event_eh, iflladdr_event_eh;
119 
120 VNET_DEFINE(struct nd_drhead, nd_defrouter);
121 VNET_DEFINE(struct nd_prhead, nd_prefix);
122 VNET_DEFINE(struct rwlock, nd6_lock);
123 VNET_DEFINE(uint64_t, nd6_list_genid);
124 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
125 
126 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
127 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
128 
129 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
130 
131 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
132 	struct ifnet *);
133 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
134 static void nd6_slowtimo(void *);
135 static int regen_tmpaddr(struct in6_ifaddr *);
136 static void nd6_free(struct llentry **, int);
137 static void nd6_free_redirect(const struct llentry *);
138 static void nd6_llinfo_timer(void *);
139 static void nd6_llinfo_settimer_locked(struct llentry *, long);
140 static void clear_llinfo_pqueue(struct llentry *);
141 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
142 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
143     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
144 static int nd6_need_cache(struct ifnet *);
145 
146 
147 static VNET_DEFINE(struct callout, nd6_slowtimo_ch);
148 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
149 
150 VNET_DEFINE(struct callout, nd6_timer_ch);
151 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
152 
153 static void
154 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
155 {
156 	struct rt_addrinfo rtinfo;
157 	struct sockaddr_in6 dst;
158 	struct sockaddr_dl gw;
159 	struct ifnet *ifp;
160 	int type;
161 	int fibnum;
162 
163 	LLE_WLOCK_ASSERT(lle);
164 
165 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
166 		return;
167 
168 	switch (evt) {
169 	case LLENTRY_RESOLVED:
170 		type = RTM_ADD;
171 		KASSERT(lle->la_flags & LLE_VALID,
172 		    ("%s: %p resolved but not valid?", __func__, lle));
173 		break;
174 	case LLENTRY_EXPIRED:
175 		type = RTM_DELETE;
176 		break;
177 	default:
178 		return;
179 	}
180 
181 	ifp = lltable_get_ifp(lle->lle_tbl);
182 
183 	bzero(&dst, sizeof(dst));
184 	bzero(&gw, sizeof(gw));
185 	bzero(&rtinfo, sizeof(rtinfo));
186 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
187 	dst.sin6_scope_id = in6_getscopezone(ifp,
188 	    in6_addrscope(&dst.sin6_addr));
189 	gw.sdl_len = sizeof(struct sockaddr_dl);
190 	gw.sdl_family = AF_LINK;
191 	gw.sdl_alen = ifp->if_addrlen;
192 	gw.sdl_index = ifp->if_index;
193 	gw.sdl_type = ifp->if_type;
194 	if (evt == LLENTRY_RESOLVED)
195 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
196 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
197 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
198 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
199 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
200 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
201 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
202 }
203 
204 /*
205  * A handler for interface link layer address change event.
206  */
207 static void
208 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
209 {
210 
211 	lltable_update_ifaddr(LLTABLE6(ifp));
212 }
213 
214 void
215 nd6_init(void)
216 {
217 
218 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
219 	rw_init(&V_nd6_lock, "nd6 list");
220 
221 	LIST_INIT(&V_nd_prefix);
222 	TAILQ_INIT(&V_nd_defrouter);
223 
224 	/* Start timers. */
225 	callout_init(&V_nd6_slowtimo_ch, 0);
226 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
227 	    nd6_slowtimo, curvnet);
228 
229 	callout_init(&V_nd6_timer_ch, 0);
230 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
231 
232 	nd6_dad_init();
233 	if (IS_DEFAULT_VNET(curvnet)) {
234 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
235 		    NULL, EVENTHANDLER_PRI_ANY);
236 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
237 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
238 	}
239 }
240 
241 #ifdef VIMAGE
242 void
243 nd6_destroy()
244 {
245 
246 	callout_drain(&V_nd6_slowtimo_ch);
247 	callout_drain(&V_nd6_timer_ch);
248 	if (IS_DEFAULT_VNET(curvnet)) {
249 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
250 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
251 	}
252 	rw_destroy(&V_nd6_lock);
253 	mtx_destroy(&V_nd6_onlink_mtx);
254 }
255 #endif
256 
257 struct nd_ifinfo *
258 nd6_ifattach(struct ifnet *ifp)
259 {
260 	struct nd_ifinfo *nd;
261 
262 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
263 	nd->initialized = 1;
264 
265 	nd->chlim = IPV6_DEFHLIM;
266 	nd->basereachable = REACHABLE_TIME;
267 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
268 	nd->retrans = RETRANS_TIMER;
269 
270 	nd->flags = ND6_IFF_PERFORMNUD;
271 
272 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
273 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
274 	 * default regardless of the V_ip6_auto_linklocal configuration to
275 	 * give a reasonable default behavior.
276 	 */
277 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
278 	    (ifp->if_flags & IFF_LOOPBACK))
279 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
280 	/*
281 	 * A loopback interface does not need to accept RTADV.
282 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
283 	 * default regardless of the V_ip6_accept_rtadv configuration to
284 	 * prevent the interface from accepting RA messages arrived
285 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
286 	 */
287 	if (V_ip6_accept_rtadv &&
288 	    !(ifp->if_flags & IFF_LOOPBACK) &&
289 	    (ifp->if_type != IFT_BRIDGE))
290 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
291 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
292 		nd->flags |= ND6_IFF_NO_RADR;
293 
294 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
295 	nd6_setmtu0(ifp, nd);
296 
297 	return nd;
298 }
299 
300 void
301 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
302 {
303 	struct ifaddr *ifa, *next;
304 
305 	IF_ADDR_RLOCK(ifp);
306 	TAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
307 		if (ifa->ifa_addr->sa_family != AF_INET6)
308 			continue;
309 
310 		/* stop DAD processing */
311 		nd6_dad_stop(ifa);
312 	}
313 	IF_ADDR_RUNLOCK(ifp);
314 
315 	free(nd, M_IP6NDP);
316 }
317 
318 /*
319  * Reset ND level link MTU. This function is called when the physical MTU
320  * changes, which means we might have to adjust the ND level MTU.
321  */
322 void
323 nd6_setmtu(struct ifnet *ifp)
324 {
325 	if (ifp->if_afdata[AF_INET6] == NULL)
326 		return;
327 
328 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
329 }
330 
331 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
332 void
333 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
334 {
335 	u_int32_t omaxmtu;
336 
337 	omaxmtu = ndi->maxmtu;
338 
339 	switch (ifp->if_type) {
340 	case IFT_ARCNET:
341 		ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */
342 		break;
343 	case IFT_FDDI:
344 		ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */
345 		break;
346 	default:
347 		ndi->maxmtu = ifp->if_mtu;
348 		break;
349 	}
350 
351 	/*
352 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
353 	 * undesirable situation.  We thus notify the operator of the change
354 	 * explicitly.  The check for omaxmtu is necessary to restrict the
355 	 * log to the case of changing the MTU, not initializing it.
356 	 */
357 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
358 		log(LOG_NOTICE, "nd6_setmtu0: "
359 		    "new link MTU on %s (%lu) is too small for IPv6\n",
360 		    if_name(ifp), (unsigned long)ndi->maxmtu);
361 	}
362 
363 	if (ndi->maxmtu > V_in6_maxmtu)
364 		in6_setmaxmtu(); /* check all interfaces just in case */
365 
366 }
367 
368 void
369 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
370 {
371 
372 	bzero(ndopts, sizeof(*ndopts));
373 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
374 	ndopts->nd_opts_last
375 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
376 
377 	if (icmp6len == 0) {
378 		ndopts->nd_opts_done = 1;
379 		ndopts->nd_opts_search = NULL;
380 	}
381 }
382 
383 /*
384  * Take one ND option.
385  */
386 struct nd_opt_hdr *
387 nd6_option(union nd_opts *ndopts)
388 {
389 	struct nd_opt_hdr *nd_opt;
390 	int olen;
391 
392 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
393 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
394 	    __func__));
395 	if (ndopts->nd_opts_search == NULL)
396 		return NULL;
397 	if (ndopts->nd_opts_done)
398 		return NULL;
399 
400 	nd_opt = ndopts->nd_opts_search;
401 
402 	/* make sure nd_opt_len is inside the buffer */
403 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
404 		bzero(ndopts, sizeof(*ndopts));
405 		return NULL;
406 	}
407 
408 	olen = nd_opt->nd_opt_len << 3;
409 	if (olen == 0) {
410 		/*
411 		 * Message validation requires that all included
412 		 * options have a length that is greater than zero.
413 		 */
414 		bzero(ndopts, sizeof(*ndopts));
415 		return NULL;
416 	}
417 
418 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
419 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
420 		/* option overruns the end of buffer, invalid */
421 		bzero(ndopts, sizeof(*ndopts));
422 		return NULL;
423 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
424 		/* reached the end of options chain */
425 		ndopts->nd_opts_done = 1;
426 		ndopts->nd_opts_search = NULL;
427 	}
428 	return nd_opt;
429 }
430 
431 /*
432  * Parse multiple ND options.
433  * This function is much easier to use, for ND routines that do not need
434  * multiple options of the same type.
435  */
436 int
437 nd6_options(union nd_opts *ndopts)
438 {
439 	struct nd_opt_hdr *nd_opt;
440 	int i = 0;
441 
442 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
443 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
444 	    __func__));
445 	if (ndopts->nd_opts_search == NULL)
446 		return 0;
447 
448 	while (1) {
449 		nd_opt = nd6_option(ndopts);
450 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
451 			/*
452 			 * Message validation requires that all included
453 			 * options have a length that is greater than zero.
454 			 */
455 			ICMP6STAT_INC(icp6s_nd_badopt);
456 			bzero(ndopts, sizeof(*ndopts));
457 			return -1;
458 		}
459 
460 		if (nd_opt == NULL)
461 			goto skip1;
462 
463 		switch (nd_opt->nd_opt_type) {
464 		case ND_OPT_SOURCE_LINKADDR:
465 		case ND_OPT_TARGET_LINKADDR:
466 		case ND_OPT_MTU:
467 		case ND_OPT_REDIRECTED_HEADER:
468 		case ND_OPT_NONCE:
469 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
470 				nd6log((LOG_INFO,
471 				    "duplicated ND6 option found (type=%d)\n",
472 				    nd_opt->nd_opt_type));
473 				/* XXX bark? */
474 			} else {
475 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
476 					= nd_opt;
477 			}
478 			break;
479 		case ND_OPT_PREFIX_INFORMATION:
480 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
481 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
482 					= nd_opt;
483 			}
484 			ndopts->nd_opts_pi_end =
485 				(struct nd_opt_prefix_info *)nd_opt;
486 			break;
487 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
488 		case ND_OPT_RDNSS:	/* RFC 6106 */
489 		case ND_OPT_DNSSL:	/* RFC 6106 */
490 			/*
491 			 * Silently ignore options we know and do not care about
492 			 * in the kernel.
493 			 */
494 			break;
495 		default:
496 			/*
497 			 * Unknown options must be silently ignored,
498 			 * to accommodate future extension to the protocol.
499 			 */
500 			nd6log((LOG_DEBUG,
501 			    "nd6_options: unsupported option %d - "
502 			    "option ignored\n", nd_opt->nd_opt_type));
503 		}
504 
505 skip1:
506 		i++;
507 		if (i > V_nd6_maxndopt) {
508 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
509 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
510 			break;
511 		}
512 
513 		if (ndopts->nd_opts_done)
514 			break;
515 	}
516 
517 	return 0;
518 }
519 
520 /*
521  * ND6 timer routine to handle ND6 entries
522  */
523 static void
524 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
525 {
526 	int canceled;
527 
528 	LLE_WLOCK_ASSERT(ln);
529 
530 	if (tick < 0) {
531 		ln->la_expire = 0;
532 		ln->ln_ntick = 0;
533 		canceled = callout_stop(&ln->lle_timer);
534 	} else {
535 		ln->la_expire = time_uptime + tick / hz;
536 		LLE_ADDREF(ln);
537 		if (tick > INT_MAX) {
538 			ln->ln_ntick = tick - INT_MAX;
539 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
540 			    nd6_llinfo_timer, ln);
541 		} else {
542 			ln->ln_ntick = 0;
543 			canceled = callout_reset(&ln->lle_timer, tick,
544 			    nd6_llinfo_timer, ln);
545 		}
546 	}
547 	if (canceled > 0)
548 		LLE_REMREF(ln);
549 }
550 
551 /*
552  * Gets source address of the first packet in hold queue
553  * and stores it in @src.
554  * Returns pointer to @src (if hold queue is not empty) or NULL.
555  *
556  * Set noinline to be dtrace-friendly
557  */
558 static __noinline struct in6_addr *
559 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
560 {
561 	struct ip6_hdr hdr;
562 	struct mbuf *m;
563 
564 	if (ln->la_hold == NULL)
565 		return (NULL);
566 
567 	/*
568 	 * assume every packet in la_hold has the same IP header
569 	 */
570 	m = ln->la_hold;
571 	if (sizeof(hdr) > m->m_len)
572 		return (NULL);
573 
574 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
575 	*src = hdr.ip6_src;
576 
577 	return (src);
578 }
579 
580 /*
581  * Checks if we need to switch from STALE state.
582  *
583  * RFC 4861 requires switching from STALE to DELAY state
584  * on first packet matching entry, waiting V_nd6_delay and
585  * transition to PROBE state (if upper layer confirmation was
586  * not received).
587  *
588  * This code performs a bit differently:
589  * On packet hit we don't change state (but desired state
590  * can be guessed by control plane). However, after V_nd6_delay
591  * seconds code will transition to PROBE state (so DELAY state
592  * is kinda skipped in most situations).
593  *
594  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
595  * we perform the following upon entering STALE state:
596  *
597  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
598  * if packet was transmitted at the start of given interval, we
599  * would be able to switch to PROBE state in V_nd6_delay seconds
600  * as user expects.
601  *
602  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
603  * lle in STALE state (remaining timer value stored in lle_remtime).
604  *
605  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
606  * seconds ago.
607  *
608  * Returns non-zero value if the entry is still STALE (storing
609  * the next timer interval in @pdelay).
610  *
611  * Returns zero value if original timer expired or we need to switch to
612  * PROBE (store that in @do_switch variable).
613  */
614 static int
615 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
616 {
617 	int nd_delay, nd_gctimer, r_skip_req;
618 	time_t lle_hittime;
619 	long delay;
620 
621 	*do_switch = 0;
622 	nd_gctimer = V_nd6_gctimer;
623 	nd_delay = V_nd6_delay;
624 
625 	LLE_REQ_LOCK(lle);
626 	r_skip_req = lle->r_skip_req;
627 	lle_hittime = lle->lle_hittime;
628 	LLE_REQ_UNLOCK(lle);
629 
630 	if (r_skip_req > 0) {
631 
632 		/*
633 		 * Nonzero r_skip_req value was set upon entering
634 		 * STALE state. Since value was not changed, no
635 		 * packets were passed using this lle. Ask for
636 		 * timer reschedule and keep STALE state.
637 		 */
638 		delay = (long)(MIN(nd_gctimer, nd_delay));
639 		delay *= hz;
640 		if (lle->lle_remtime > delay)
641 			lle->lle_remtime -= delay;
642 		else {
643 			delay = lle->lle_remtime;
644 			lle->lle_remtime = 0;
645 		}
646 
647 		if (delay == 0) {
648 
649 			/*
650 			 * The original ng6_gctime timeout ended,
651 			 * no more rescheduling.
652 			 */
653 			return (0);
654 		}
655 
656 		*pdelay = delay;
657 		return (1);
658 	}
659 
660 	/*
661 	 * Packet received. Verify timestamp
662 	 */
663 	delay = (long)(time_uptime - lle_hittime);
664 	if (delay < nd_delay) {
665 
666 		/*
667 		 * V_nd6_delay still not passed since the first
668 		 * hit in STALE state.
669 		 * Reshedule timer and return.
670 		 */
671 		*pdelay = (long)(nd_delay - delay) * hz;
672 		return (1);
673 	}
674 
675 	/* Request switching to probe */
676 	*do_switch = 1;
677 	return (0);
678 }
679 
680 
681 /*
682  * Switch @lle state to new state optionally arming timers.
683  *
684  * Set noinline to be dtrace-friendly
685  */
686 __noinline void
687 nd6_llinfo_setstate(struct llentry *lle, int newstate)
688 {
689 	struct ifnet *ifp;
690 	int nd_gctimer, nd_delay;
691 	long delay, remtime;
692 
693 	delay = 0;
694 	remtime = 0;
695 
696 	switch (newstate) {
697 	case ND6_LLINFO_INCOMPLETE:
698 		ifp = lle->lle_tbl->llt_ifp;
699 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
700 		break;
701 	case ND6_LLINFO_REACHABLE:
702 		if (!ND6_LLINFO_PERMANENT(lle)) {
703 			ifp = lle->lle_tbl->llt_ifp;
704 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
705 		}
706 		break;
707 	case ND6_LLINFO_STALE:
708 
709 		/*
710 		 * Notify fast path that we want to know if any packet
711 		 * is transmitted by setting r_skip_req.
712 		 */
713 		LLE_REQ_LOCK(lle);
714 		lle->r_skip_req = 1;
715 		LLE_REQ_UNLOCK(lle);
716 		nd_delay = V_nd6_delay;
717 		nd_gctimer = V_nd6_gctimer;
718 
719 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
720 		remtime = (long)nd_gctimer * hz - delay;
721 		break;
722 	case ND6_LLINFO_DELAY:
723 		lle->la_asked = 0;
724 		delay = (long)V_nd6_delay * hz;
725 		break;
726 	}
727 
728 	if (delay > 0)
729 		nd6_llinfo_settimer_locked(lle, delay);
730 
731 	lle->lle_remtime = remtime;
732 	lle->ln_state = newstate;
733 }
734 
735 /*
736  * Timer-dependent part of nd state machine.
737  *
738  * Set noinline to be dtrace-friendly
739  */
740 static __noinline void
741 nd6_llinfo_timer(void *arg)
742 {
743 	struct llentry *ln;
744 	struct in6_addr *dst, *pdst, *psrc, src;
745 	struct ifnet *ifp;
746 	struct nd_ifinfo *ndi;
747 	int do_switch, send_ns;
748 	long delay;
749 
750 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
751 	ln = (struct llentry *)arg;
752 	ifp = lltable_get_ifp(ln->lle_tbl);
753 	CURVNET_SET(ifp->if_vnet);
754 
755 	ND6_RLOCK();
756 	LLE_WLOCK(ln);
757 	if (callout_pending(&ln->lle_timer)) {
758 		/*
759 		 * Here we are a bit odd here in the treatment of
760 		 * active/pending. If the pending bit is set, it got
761 		 * rescheduled before I ran. The active
762 		 * bit we ignore, since if it was stopped
763 		 * in ll_tablefree() and was currently running
764 		 * it would have return 0 so the code would
765 		 * not have deleted it since the callout could
766 		 * not be stopped so we want to go through
767 		 * with the delete here now. If the callout
768 		 * was restarted, the pending bit will be back on and
769 		 * we just want to bail since the callout_reset would
770 		 * return 1 and our reference would have been removed
771 		 * by nd6_llinfo_settimer_locked above since canceled
772 		 * would have been 1.
773 		 */
774 		LLE_WUNLOCK(ln);
775 		ND6_RUNLOCK();
776 		CURVNET_RESTORE();
777 		return;
778 	}
779 	ndi = ND_IFINFO(ifp);
780 	send_ns = 0;
781 	dst = &ln->r_l3addr.addr6;
782 	pdst = dst;
783 
784 	if (ln->ln_ntick > 0) {
785 		if (ln->ln_ntick > INT_MAX) {
786 			ln->ln_ntick -= INT_MAX;
787 			nd6_llinfo_settimer_locked(ln, INT_MAX);
788 		} else {
789 			ln->ln_ntick = 0;
790 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
791 		}
792 		goto done;
793 	}
794 
795 	if (ln->la_flags & LLE_STATIC) {
796 		goto done;
797 	}
798 
799 	if (ln->la_flags & LLE_DELETED) {
800 		nd6_free(&ln, 0);
801 		goto done;
802 	}
803 
804 	switch (ln->ln_state) {
805 	case ND6_LLINFO_INCOMPLETE:
806 		if (ln->la_asked < V_nd6_mmaxtries) {
807 			ln->la_asked++;
808 			send_ns = 1;
809 			/* Send NS to multicast address */
810 			pdst = NULL;
811 		} else {
812 			struct mbuf *m = ln->la_hold;
813 			if (m) {
814 				struct mbuf *m0;
815 
816 				/*
817 				 * assuming every packet in la_hold has the
818 				 * same IP header.  Send error after unlock.
819 				 */
820 				m0 = m->m_nextpkt;
821 				m->m_nextpkt = NULL;
822 				ln->la_hold = m0;
823 				clear_llinfo_pqueue(ln);
824 			}
825 			nd6_free(&ln, 0);
826 			if (m != NULL)
827 				icmp6_error2(m, ICMP6_DST_UNREACH,
828 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
829 		}
830 		break;
831 	case ND6_LLINFO_REACHABLE:
832 		if (!ND6_LLINFO_PERMANENT(ln))
833 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
834 		break;
835 
836 	case ND6_LLINFO_STALE:
837 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
838 
839 			/*
840 			 * No packet has used this entry and GC timeout
841 			 * has not been passed. Reshedule timer and
842 			 * return.
843 			 */
844 			nd6_llinfo_settimer_locked(ln, delay);
845 			break;
846 		}
847 
848 		if (do_switch == 0) {
849 
850 			/*
851 			 * GC timer has ended and entry hasn't been used.
852 			 * Run Garbage collector (RFC 4861, 5.3)
853 			 */
854 			if (!ND6_LLINFO_PERMANENT(ln))
855 				nd6_free(&ln, 1);
856 			break;
857 		}
858 
859 		/* Entry has been used AND delay timer has ended. */
860 
861 		/* FALLTHROUGH */
862 
863 	case ND6_LLINFO_DELAY:
864 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
865 			/* We need NUD */
866 			ln->la_asked = 1;
867 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
868 			send_ns = 1;
869 		} else
870 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
871 		break;
872 	case ND6_LLINFO_PROBE:
873 		if (ln->la_asked < V_nd6_umaxtries) {
874 			ln->la_asked++;
875 			send_ns = 1;
876 		} else {
877 			nd6_free(&ln, 0);
878 		}
879 		break;
880 	default:
881 		panic("%s: paths in a dark night can be confusing: %d",
882 		    __func__, ln->ln_state);
883 	}
884 done:
885 	if (ln != NULL)
886 		ND6_RUNLOCK();
887 	if (send_ns != 0) {
888 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
889 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
890 		LLE_FREE_LOCKED(ln);
891 		ln = NULL;
892 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
893 	}
894 
895 	if (ln != NULL)
896 		LLE_FREE_LOCKED(ln);
897 	CURVNET_RESTORE();
898 }
899 
900 
901 /*
902  * ND6 timer routine to expire default route list and prefix list
903  */
904 void
905 nd6_timer(void *arg)
906 {
907 	CURVNET_SET((struct vnet *) arg);
908 	struct nd_drhead drq;
909 	struct nd_prhead prl;
910 	struct nd_defrouter *dr, *ndr;
911 	struct nd_prefix *pr, *npr;
912 	struct in6_ifaddr *ia6, *nia6;
913 	uint64_t genid;
914 
915 	TAILQ_INIT(&drq);
916 	LIST_INIT(&prl);
917 
918 	ND6_WLOCK();
919 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
920 		if (dr->expire && dr->expire < time_uptime)
921 			defrouter_unlink(dr, &drq);
922 	ND6_WUNLOCK();
923 
924 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
925 		TAILQ_REMOVE(&drq, dr, dr_entry);
926 		defrouter_del(dr);
927 	}
928 
929 	/*
930 	 * expire interface addresses.
931 	 * in the past the loop was inside prefix expiry processing.
932 	 * However, from a stricter speci-confrmance standpoint, we should
933 	 * rather separate address lifetimes and prefix lifetimes.
934 	 *
935 	 * XXXRW: in6_ifaddrhead locking.
936 	 */
937   addrloop:
938 	TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
939 		/* check address lifetime */
940 		if (IFA6_IS_INVALID(ia6)) {
941 			int regen = 0;
942 
943 			/*
944 			 * If the expiring address is temporary, try
945 			 * regenerating a new one.  This would be useful when
946 			 * we suspended a laptop PC, then turned it on after a
947 			 * period that could invalidate all temporary
948 			 * addresses.  Although we may have to restart the
949 			 * loop (see below), it must be after purging the
950 			 * address.  Otherwise, we'd see an infinite loop of
951 			 * regeneration.
952 			 */
953 			if (V_ip6_use_tempaddr &&
954 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
955 				if (regen_tmpaddr(ia6) == 0)
956 					regen = 1;
957 			}
958 
959 			in6_purgeaddr(&ia6->ia_ifa);
960 
961 			if (regen)
962 				goto addrloop; /* XXX: see below */
963 		} else if (IFA6_IS_DEPRECATED(ia6)) {
964 			int oldflags = ia6->ia6_flags;
965 
966 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
967 
968 			/*
969 			 * If a temporary address has just become deprecated,
970 			 * regenerate a new one if possible.
971 			 */
972 			if (V_ip6_use_tempaddr &&
973 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
974 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
975 
976 				if (regen_tmpaddr(ia6) == 0) {
977 					/*
978 					 * A new temporary address is
979 					 * generated.
980 					 * XXX: this means the address chain
981 					 * has changed while we are still in
982 					 * the loop.  Although the change
983 					 * would not cause disaster (because
984 					 * it's not a deletion, but an
985 					 * addition,) we'd rather restart the
986 					 * loop just for safety.  Or does this
987 					 * significantly reduce performance??
988 					 */
989 					goto addrloop;
990 				}
991 			}
992 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
993 			/*
994 			 * Schedule DAD for a tentative address.  This happens
995 			 * if the interface was down or not running
996 			 * when the address was configured.
997 			 */
998 			int delay;
999 
1000 			delay = arc4random() %
1001 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1002 			nd6_dad_start((struct ifaddr *)ia6, delay);
1003 		} else {
1004 			/*
1005 			 * Check status of the interface.  If it is down,
1006 			 * mark the address as tentative for future DAD.
1007 			 */
1008 			if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 ||
1009 			    (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING)
1010 				== 0 ||
1011 			    (ND_IFINFO(ia6->ia_ifp)->flags &
1012 				ND6_IFF_IFDISABLED) != 0) {
1013 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1014 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1015 			}
1016 			/*
1017 			 * A new RA might have made a deprecated address
1018 			 * preferred.
1019 			 */
1020 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1021 		}
1022 	}
1023 
1024 	ND6_WLOCK();
1025 restart:
1026 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1027 		/*
1028 		 * Expire prefixes. Since the pltime is only used for
1029 		 * autoconfigured addresses, pltime processing for prefixes is
1030 		 * not necessary.
1031 		 *
1032 		 * Only unlink after all derived addresses have expired. This
1033 		 * may not occur until two hours after the prefix has expired
1034 		 * per RFC 4862. If the prefix expires before its derived
1035 		 * addresses, mark it off-link. This will be done automatically
1036 		 * after unlinking if no address references remain.
1037 		 */
1038 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1039 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1040 			continue;
1041 
1042 		if (pr->ndpr_addrcnt == 0) {
1043 			nd6_prefix_unlink(pr, &prl);
1044 			continue;
1045 		}
1046 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1047 			genid = V_nd6_list_genid;
1048 			nd6_prefix_ref(pr);
1049 			ND6_WUNLOCK();
1050 			ND6_ONLINK_LOCK();
1051 			(void)nd6_prefix_offlink(pr);
1052 			ND6_ONLINK_UNLOCK();
1053 			ND6_WLOCK();
1054 			nd6_prefix_rele(pr);
1055 			if (genid != V_nd6_list_genid)
1056 				goto restart;
1057 		}
1058 	}
1059 	ND6_WUNLOCK();
1060 
1061 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1062 		LIST_REMOVE(pr, ndpr_entry);
1063 		nd6_prefix_del(pr);
1064 	}
1065 
1066 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1067 	    nd6_timer, curvnet);
1068 
1069 	CURVNET_RESTORE();
1070 }
1071 
1072 /*
1073  * ia6 - deprecated/invalidated temporary address
1074  */
1075 static int
1076 regen_tmpaddr(struct in6_ifaddr *ia6)
1077 {
1078 	struct ifaddr *ifa;
1079 	struct ifnet *ifp;
1080 	struct in6_ifaddr *public_ifa6 = NULL;
1081 
1082 	ifp = ia6->ia_ifa.ifa_ifp;
1083 	IF_ADDR_RLOCK(ifp);
1084 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1085 		struct in6_ifaddr *it6;
1086 
1087 		if (ifa->ifa_addr->sa_family != AF_INET6)
1088 			continue;
1089 
1090 		it6 = (struct in6_ifaddr *)ifa;
1091 
1092 		/* ignore no autoconf addresses. */
1093 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1094 			continue;
1095 
1096 		/* ignore autoconf addresses with different prefixes. */
1097 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1098 			continue;
1099 
1100 		/*
1101 		 * Now we are looking at an autoconf address with the same
1102 		 * prefix as ours.  If the address is temporary and is still
1103 		 * preferred, do not create another one.  It would be rare, but
1104 		 * could happen, for example, when we resume a laptop PC after
1105 		 * a long period.
1106 		 */
1107 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1108 		    !IFA6_IS_DEPRECATED(it6)) {
1109 			public_ifa6 = NULL;
1110 			break;
1111 		}
1112 
1113 		/*
1114 		 * This is a public autoconf address that has the same prefix
1115 		 * as ours.  If it is preferred, keep it.  We can't break the
1116 		 * loop here, because there may be a still-preferred temporary
1117 		 * address with the prefix.
1118 		 */
1119 		if (!IFA6_IS_DEPRECATED(it6))
1120 			public_ifa6 = it6;
1121 	}
1122 	if (public_ifa6 != NULL)
1123 		ifa_ref(&public_ifa6->ia_ifa);
1124 	IF_ADDR_RUNLOCK(ifp);
1125 
1126 	if (public_ifa6 != NULL) {
1127 		int e;
1128 
1129 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1130 			ifa_free(&public_ifa6->ia_ifa);
1131 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1132 			    " tmp addr,errno=%d\n", e);
1133 			return (-1);
1134 		}
1135 		ifa_free(&public_ifa6->ia_ifa);
1136 		return (0);
1137 	}
1138 
1139 	return (-1);
1140 }
1141 
1142 /*
1143  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1144  * cache entries are freed in in6_domifdetach().
1145  */
1146 void
1147 nd6_purge(struct ifnet *ifp)
1148 {
1149 	struct nd_drhead drq;
1150 	struct nd_prhead prl;
1151 	struct nd_defrouter *dr, *ndr;
1152 	struct nd_prefix *pr, *npr;
1153 
1154 	TAILQ_INIT(&drq);
1155 	LIST_INIT(&prl);
1156 
1157 	/*
1158 	 * Nuke default router list entries toward ifp.
1159 	 * We defer removal of default router list entries that is installed
1160 	 * in the routing table, in order to keep additional side effects as
1161 	 * small as possible.
1162 	 */
1163 	ND6_WLOCK();
1164 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1165 		if (dr->installed)
1166 			continue;
1167 		if (dr->ifp == ifp)
1168 			defrouter_unlink(dr, &drq);
1169 	}
1170 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1171 		if (!dr->installed)
1172 			continue;
1173 		if (dr->ifp == ifp)
1174 			defrouter_unlink(dr, &drq);
1175 	}
1176 
1177 	/*
1178 	 * Remove prefixes on ifp. We should have already removed addresses on
1179 	 * this interface, so no addresses should be referencing these prefixes.
1180 	 */
1181 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1182 		if (pr->ndpr_ifp == ifp)
1183 			nd6_prefix_unlink(pr, &prl);
1184 	}
1185 	ND6_WUNLOCK();
1186 
1187 	/* Delete the unlinked router and prefix objects. */
1188 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1189 		TAILQ_REMOVE(&drq, dr, dr_entry);
1190 		defrouter_del(dr);
1191 	}
1192 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1193 		LIST_REMOVE(pr, ndpr_entry);
1194 		nd6_prefix_del(pr);
1195 	}
1196 
1197 	/* cancel default outgoing interface setting */
1198 	if (V_nd6_defifindex == ifp->if_index)
1199 		nd6_setdefaultiface(0);
1200 
1201 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1202 		/* Refresh default router list. */
1203 		defrouter_select_fib(ifp->if_fib);
1204 	}
1205 }
1206 
1207 /*
1208  * the caller acquires and releases the lock on the lltbls
1209  * Returns the llentry locked
1210  */
1211 struct llentry *
1212 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1213 {
1214 	struct sockaddr_in6 sin6;
1215 	struct llentry *ln;
1216 
1217 	bzero(&sin6, sizeof(sin6));
1218 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1219 	sin6.sin6_family = AF_INET6;
1220 	sin6.sin6_addr = *addr6;
1221 
1222 	IF_AFDATA_LOCK_ASSERT(ifp);
1223 
1224 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1225 
1226 	return (ln);
1227 }
1228 
1229 struct llentry *
1230 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1231 {
1232 	struct sockaddr_in6 sin6;
1233 	struct llentry *ln;
1234 
1235 	bzero(&sin6, sizeof(sin6));
1236 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1237 	sin6.sin6_family = AF_INET6;
1238 	sin6.sin6_addr = *addr6;
1239 
1240 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1241 	if (ln != NULL)
1242 		ln->ln_state = ND6_LLINFO_NOSTATE;
1243 
1244 	return (ln);
1245 }
1246 
1247 /*
1248  * Test whether a given IPv6 address is a neighbor or not, ignoring
1249  * the actual neighbor cache.  The neighbor cache is ignored in order
1250  * to not reenter the routing code from within itself.
1251  */
1252 static int
1253 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1254 {
1255 	struct nd_prefix *pr;
1256 	struct ifaddr *ifa;
1257 	struct rt_addrinfo info;
1258 	struct sockaddr_in6 rt_key;
1259 	const struct sockaddr *dst6;
1260 	uint64_t genid;
1261 	int error, fibnum;
1262 
1263 	/*
1264 	 * A link-local address is always a neighbor.
1265 	 * XXX: a link does not necessarily specify a single interface.
1266 	 */
1267 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1268 		struct sockaddr_in6 sin6_copy;
1269 		u_int32_t zone;
1270 
1271 		/*
1272 		 * We need sin6_copy since sa6_recoverscope() may modify the
1273 		 * content (XXX).
1274 		 */
1275 		sin6_copy = *addr;
1276 		if (sa6_recoverscope(&sin6_copy))
1277 			return (0); /* XXX: should be impossible */
1278 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1279 			return (0);
1280 		if (sin6_copy.sin6_scope_id == zone)
1281 			return (1);
1282 		else
1283 			return (0);
1284 	}
1285 
1286 	bzero(&rt_key, sizeof(rt_key));
1287 	bzero(&info, sizeof(info));
1288 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1289 
1290 	/*
1291 	 * If the address matches one of our addresses,
1292 	 * it should be a neighbor.
1293 	 * If the address matches one of our on-link prefixes, it should be a
1294 	 * neighbor.
1295 	 */
1296 	ND6_RLOCK();
1297 restart:
1298 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1299 		if (pr->ndpr_ifp != ifp)
1300 			continue;
1301 
1302 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1303 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1304 
1305 			/*
1306 			 * We only need to check all FIBs if add_addr_allfibs
1307 			 * is unset. If set, checking any FIB will suffice.
1308 			 */
1309 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1310 			for (; fibnum < rt_numfibs; fibnum++) {
1311 				genid = V_nd6_list_genid;
1312 				ND6_RUNLOCK();
1313 
1314 				/*
1315 				 * Restore length field before
1316 				 * retrying lookup
1317 				 */
1318 				rt_key.sin6_len = sizeof(rt_key);
1319 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1320 						        &info);
1321 
1322 				ND6_RLOCK();
1323 				if (genid != V_nd6_list_genid)
1324 					goto restart;
1325 				if (error == 0)
1326 					break;
1327 			}
1328 			if (error != 0)
1329 				continue;
1330 
1331 			/*
1332 			 * This is the case where multiple interfaces
1333 			 * have the same prefix, but only one is installed
1334 			 * into the routing table and that prefix entry
1335 			 * is not the one being examined here. In the case
1336 			 * where RADIX_MPATH is enabled, multiple route
1337 			 * entries (of the same rt_key value) will be
1338 			 * installed because the interface addresses all
1339 			 * differ.
1340 			 */
1341 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1342 			    &rt_key.sin6_addr))
1343 				continue;
1344 		}
1345 
1346 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1347 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1348 			ND6_RUNLOCK();
1349 			return (1);
1350 		}
1351 	}
1352 	ND6_RUNLOCK();
1353 
1354 	/*
1355 	 * If the address is assigned on the node of the other side of
1356 	 * a p2p interface, the address should be a neighbor.
1357 	 */
1358 	if (ifp->if_flags & IFF_POINTOPOINT) {
1359 		IF_ADDR_RLOCK(ifp);
1360 		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1361 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1362 				continue;
1363 			if (ifa->ifa_dstaddr != NULL &&
1364 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1365 				IF_ADDR_RUNLOCK(ifp);
1366 				return 1;
1367 			}
1368 		}
1369 		IF_ADDR_RUNLOCK(ifp);
1370 	}
1371 
1372 	/*
1373 	 * If the default router list is empty, all addresses are regarded
1374 	 * as on-link, and thus, as a neighbor.
1375 	 */
1376 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1377 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1378 	    V_nd6_defifindex == ifp->if_index) {
1379 		return (1);
1380 	}
1381 
1382 	return (0);
1383 }
1384 
1385 
1386 /*
1387  * Detect if a given IPv6 address identifies a neighbor on a given link.
1388  * XXX: should take care of the destination of a p2p link?
1389  */
1390 int
1391 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1392 {
1393 	struct llentry *lle;
1394 	int rc = 0;
1395 
1396 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1397 	if (nd6_is_new_addr_neighbor(addr, ifp))
1398 		return (1);
1399 
1400 	/*
1401 	 * Even if the address matches none of our addresses, it might be
1402 	 * in the neighbor cache.
1403 	 */
1404 	IF_AFDATA_RLOCK(ifp);
1405 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1406 		LLE_RUNLOCK(lle);
1407 		rc = 1;
1408 	}
1409 	IF_AFDATA_RUNLOCK(ifp);
1410 	return (rc);
1411 }
1412 
1413 /*
1414  * Free an nd6 llinfo entry.
1415  * Since the function would cause significant changes in the kernel, DO NOT
1416  * make it global, unless you have a strong reason for the change, and are sure
1417  * that the change is safe.
1418  *
1419  * Set noinline to be dtrace-friendly
1420  */
1421 static __noinline void
1422 nd6_free(struct llentry **lnp, int gc)
1423 {
1424 	struct ifnet *ifp;
1425 	struct llentry *ln;
1426 	struct nd_defrouter *dr;
1427 
1428 	ln = *lnp;
1429 	*lnp = NULL;
1430 
1431 	LLE_WLOCK_ASSERT(ln);
1432 	ND6_RLOCK_ASSERT();
1433 
1434 	ifp = lltable_get_ifp(ln->lle_tbl);
1435 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1436 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1437 	else
1438 		dr = NULL;
1439 	ND6_RUNLOCK();
1440 
1441 	if ((ln->la_flags & LLE_DELETED) == 0)
1442 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1443 
1444 	/*
1445 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1446 	 * even though it is not harmful, it was not really necessary.
1447 	 */
1448 
1449 	/* cancel timer */
1450 	nd6_llinfo_settimer_locked(ln, -1);
1451 
1452 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1453 		if (dr != NULL && dr->expire &&
1454 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1455 			/*
1456 			 * If the reason for the deletion is just garbage
1457 			 * collection, and the neighbor is an active default
1458 			 * router, do not delete it.  Instead, reset the GC
1459 			 * timer using the router's lifetime.
1460 			 * Simply deleting the entry would affect default
1461 			 * router selection, which is not necessarily a good
1462 			 * thing, especially when we're using router preference
1463 			 * values.
1464 			 * XXX: the check for ln_state would be redundant,
1465 			 *      but we intentionally keep it just in case.
1466 			 */
1467 			if (dr->expire > time_uptime)
1468 				nd6_llinfo_settimer_locked(ln,
1469 				    (dr->expire - time_uptime) * hz);
1470 			else
1471 				nd6_llinfo_settimer_locked(ln,
1472 				    (long)V_nd6_gctimer * hz);
1473 
1474 			LLE_REMREF(ln);
1475 			LLE_WUNLOCK(ln);
1476 			defrouter_rele(dr);
1477 			return;
1478 		}
1479 
1480 		if (dr) {
1481 			/*
1482 			 * Unreachablity of a router might affect the default
1483 			 * router selection and on-link detection of advertised
1484 			 * prefixes.
1485 			 */
1486 
1487 			/*
1488 			 * Temporarily fake the state to choose a new default
1489 			 * router and to perform on-link determination of
1490 			 * prefixes correctly.
1491 			 * Below the state will be set correctly,
1492 			 * or the entry itself will be deleted.
1493 			 */
1494 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1495 		}
1496 
1497 		if (ln->ln_router || dr) {
1498 
1499 			/*
1500 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1501 			 * rnh and for the calls to pfxlist_onlink_check() and
1502 			 * defrouter_select_fib() in the block further down for calls
1503 			 * into nd6_lookup().  We still hold a ref.
1504 			 */
1505 			LLE_WUNLOCK(ln);
1506 
1507 			/*
1508 			 * rt6_flush must be called whether or not the neighbor
1509 			 * is in the Default Router List.
1510 			 * See a corresponding comment in nd6_na_input().
1511 			 */
1512 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1513 		}
1514 
1515 		if (dr) {
1516 			/*
1517 			 * Since defrouter_select_fib() does not affect the
1518 			 * on-link determination and MIP6 needs the check
1519 			 * before the default router selection, we perform
1520 			 * the check now.
1521 			 */
1522 			pfxlist_onlink_check();
1523 
1524 			/*
1525 			 * Refresh default router list.
1526 			 */
1527 			defrouter_select_fib(dr->ifp->if_fib);
1528 		}
1529 
1530 		/*
1531 		 * If this entry was added by an on-link redirect, remove the
1532 		 * corresponding host route.
1533 		 */
1534 		if (ln->la_flags & LLE_REDIRECT)
1535 			nd6_free_redirect(ln);
1536 
1537 		if (ln->ln_router || dr)
1538 			LLE_WLOCK(ln);
1539 	}
1540 
1541 	/*
1542 	 * Save to unlock. We still hold an extra reference and will not
1543 	 * free(9) in llentry_free() if someone else holds one as well.
1544 	 */
1545 	LLE_WUNLOCK(ln);
1546 	IF_AFDATA_LOCK(ifp);
1547 	LLE_WLOCK(ln);
1548 	/* Guard against race with other llentry_free(). */
1549 	if (ln->la_flags & LLE_LINKED) {
1550 		/* Remove callout reference */
1551 		LLE_REMREF(ln);
1552 		lltable_unlink_entry(ln->lle_tbl, ln);
1553 	}
1554 	IF_AFDATA_UNLOCK(ifp);
1555 
1556 	llentry_free(ln);
1557 	if (dr != NULL)
1558 		defrouter_rele(dr);
1559 }
1560 
1561 static int
1562 nd6_isdynrte(const struct rtentry *rt, void *xap)
1563 {
1564 
1565 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1566 		return (1);
1567 
1568 	return (0);
1569 }
1570 /*
1571  * Remove the rtentry for the given llentry,
1572  * both of which were installed by a redirect.
1573  */
1574 static void
1575 nd6_free_redirect(const struct llentry *ln)
1576 {
1577 	int fibnum;
1578 	struct sockaddr_in6 sin6;
1579 	struct rt_addrinfo info;
1580 
1581 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1582 	memset(&info, 0, sizeof(info));
1583 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1584 	info.rti_filter = nd6_isdynrte;
1585 
1586 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1587 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1588 }
1589 
1590 /*
1591  * Rejuvenate this function for routing operations related
1592  * processing.
1593  */
1594 void
1595 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1596 {
1597 	struct sockaddr_in6 *gateway;
1598 	struct nd_defrouter *dr;
1599 	struct ifnet *ifp;
1600 
1601 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1602 	ifp = rt->rt_ifp;
1603 
1604 	switch (req) {
1605 	case RTM_ADD:
1606 		break;
1607 
1608 	case RTM_DELETE:
1609 		if (!ifp)
1610 			return;
1611 		/*
1612 		 * Only indirect routes are interesting.
1613 		 */
1614 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1615 			return;
1616 		/*
1617 		 * check for default route
1618 		 */
1619 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1620 		    &SIN6(rt_key(rt))->sin6_addr)) {
1621 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1622 			if (dr != NULL) {
1623 				dr->installed = 0;
1624 				defrouter_rele(dr);
1625 			}
1626 		}
1627 		break;
1628 	}
1629 }
1630 
1631 
1632 int
1633 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1634 {
1635 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1636 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1637 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1638 	int error = 0;
1639 
1640 	if (ifp->if_afdata[AF_INET6] == NULL)
1641 		return (EPFNOSUPPORT);
1642 	switch (cmd) {
1643 	case OSIOCGIFINFO_IN6:
1644 #define ND	ndi->ndi
1645 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1646 		bzero(&ND, sizeof(ND));
1647 		ND.linkmtu = IN6_LINKMTU(ifp);
1648 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1649 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1650 		ND.reachable = ND_IFINFO(ifp)->reachable;
1651 		ND.retrans = ND_IFINFO(ifp)->retrans;
1652 		ND.flags = ND_IFINFO(ifp)->flags;
1653 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1654 		ND.chlim = ND_IFINFO(ifp)->chlim;
1655 		break;
1656 	case SIOCGIFINFO_IN6:
1657 		ND = *ND_IFINFO(ifp);
1658 		break;
1659 	case SIOCSIFINFO_IN6:
1660 		/*
1661 		 * used to change host variables from userland.
1662 		 * intended for a use on router to reflect RA configurations.
1663 		 */
1664 		/* 0 means 'unspecified' */
1665 		if (ND.linkmtu != 0) {
1666 			if (ND.linkmtu < IPV6_MMTU ||
1667 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1668 				error = EINVAL;
1669 				break;
1670 			}
1671 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1672 		}
1673 
1674 		if (ND.basereachable != 0) {
1675 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1676 
1677 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1678 			if (ND.basereachable != obasereachable)
1679 				ND_IFINFO(ifp)->reachable =
1680 				    ND_COMPUTE_RTIME(ND.basereachable);
1681 		}
1682 		if (ND.retrans != 0)
1683 			ND_IFINFO(ifp)->retrans = ND.retrans;
1684 		if (ND.chlim != 0)
1685 			ND_IFINFO(ifp)->chlim = ND.chlim;
1686 		/* FALLTHROUGH */
1687 	case SIOCSIFINFO_FLAGS:
1688 	{
1689 		struct ifaddr *ifa;
1690 		struct in6_ifaddr *ia;
1691 
1692 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1693 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1694 			/* ifdisabled 1->0 transision */
1695 
1696 			/*
1697 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1698 			 * has an link-local address with IN6_IFF_DUPLICATED,
1699 			 * do not clear ND6_IFF_IFDISABLED.
1700 			 * See RFC 4862, Section 5.4.5.
1701 			 */
1702 			IF_ADDR_RLOCK(ifp);
1703 			TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1704 				if (ifa->ifa_addr->sa_family != AF_INET6)
1705 					continue;
1706 				ia = (struct in6_ifaddr *)ifa;
1707 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1708 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1709 					break;
1710 			}
1711 			IF_ADDR_RUNLOCK(ifp);
1712 
1713 			if (ifa != NULL) {
1714 				/* LLA is duplicated. */
1715 				ND.flags |= ND6_IFF_IFDISABLED;
1716 				log(LOG_ERR, "Cannot enable an interface"
1717 				    " with a link-local address marked"
1718 				    " duplicate.\n");
1719 			} else {
1720 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1721 				if (ifp->if_flags & IFF_UP)
1722 					in6_if_up(ifp);
1723 			}
1724 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1725 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1726 			/* ifdisabled 0->1 transision */
1727 			/* Mark all IPv6 address as tentative. */
1728 
1729 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1730 			if (V_ip6_dad_count > 0 &&
1731 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1732 				IF_ADDR_RLOCK(ifp);
1733 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1734 				    ifa_link) {
1735 					if (ifa->ifa_addr->sa_family !=
1736 					    AF_INET6)
1737 						continue;
1738 					ia = (struct in6_ifaddr *)ifa;
1739 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1740 				}
1741 				IF_ADDR_RUNLOCK(ifp);
1742 			}
1743 		}
1744 
1745 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1746 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1747 				/* auto_linklocal 0->1 transision */
1748 
1749 				/* If no link-local address on ifp, configure */
1750 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1751 				in6_ifattach(ifp, NULL);
1752 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1753 			    ifp->if_flags & IFF_UP) {
1754 				/*
1755 				 * When the IF already has
1756 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1757 				 * address is assigned, and IFF_UP, try to
1758 				 * assign one.
1759 				 */
1760 				IF_ADDR_RLOCK(ifp);
1761 				TAILQ_FOREACH(ifa, &ifp->if_addrhead,
1762 				    ifa_link) {
1763 					if (ifa->ifa_addr->sa_family !=
1764 					    AF_INET6)
1765 						continue;
1766 					ia = (struct in6_ifaddr *)ifa;
1767 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1768 						break;
1769 				}
1770 				IF_ADDR_RUNLOCK(ifp);
1771 				if (ifa != NULL)
1772 					/* No LLA is configured. */
1773 					in6_ifattach(ifp, NULL);
1774 			}
1775 		}
1776 	}
1777 		ND_IFINFO(ifp)->flags = ND.flags;
1778 		break;
1779 #undef ND
1780 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1781 		/* sync kernel routing table with the default router list */
1782 		defrouter_reset();
1783 		defrouter_select();
1784 		break;
1785 	case SIOCSPFXFLUSH_IN6:
1786 	{
1787 		/* flush all the prefix advertised by routers */
1788 		struct in6_ifaddr *ia, *ia_next;
1789 		struct nd_prefix *pr, *next;
1790 		struct nd_prhead prl;
1791 
1792 		LIST_INIT(&prl);
1793 
1794 		ND6_WLOCK();
1795 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1796 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1797 				continue; /* XXX */
1798 			nd6_prefix_unlink(pr, &prl);
1799 		}
1800 		ND6_WUNLOCK();
1801 
1802 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1803 			LIST_REMOVE(pr, ndpr_entry);
1804 			/* XXXRW: in6_ifaddrhead locking. */
1805 			TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1806 			    ia_next) {
1807 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1808 					continue;
1809 
1810 				if (ia->ia6_ndpr == pr)
1811 					in6_purgeaddr(&ia->ia_ifa);
1812 			}
1813 			nd6_prefix_del(pr);
1814 		}
1815 		break;
1816 	}
1817 	case SIOCSRTRFLUSH_IN6:
1818 	{
1819 		/* flush all the default routers */
1820 		struct nd_drhead drq;
1821 		struct nd_defrouter *dr;
1822 
1823 		TAILQ_INIT(&drq);
1824 
1825 		defrouter_reset();
1826 
1827 		ND6_WLOCK();
1828 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1829 			defrouter_unlink(dr, &drq);
1830 		ND6_WUNLOCK();
1831 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1832 			TAILQ_REMOVE(&drq, dr, dr_entry);
1833 			defrouter_del(dr);
1834 		}
1835 
1836 		defrouter_select();
1837 		break;
1838 	}
1839 	case SIOCGNBRINFO_IN6:
1840 	{
1841 		struct llentry *ln;
1842 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1843 
1844 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1845 			return (error);
1846 
1847 		IF_AFDATA_RLOCK(ifp);
1848 		ln = nd6_lookup(&nb_addr, 0, ifp);
1849 		IF_AFDATA_RUNLOCK(ifp);
1850 
1851 		if (ln == NULL) {
1852 			error = EINVAL;
1853 			break;
1854 		}
1855 		nbi->state = ln->ln_state;
1856 		nbi->asked = ln->la_asked;
1857 		nbi->isrouter = ln->ln_router;
1858 		if (ln->la_expire == 0)
1859 			nbi->expire = 0;
1860 		else
1861 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1862 			    (time_second - time_uptime);
1863 		LLE_RUNLOCK(ln);
1864 		break;
1865 	}
1866 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1867 		ndif->ifindex = V_nd6_defifindex;
1868 		break;
1869 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1870 		return (nd6_setdefaultiface(ndif->ifindex));
1871 	}
1872 	return (error);
1873 }
1874 
1875 /*
1876  * Calculates new isRouter value based on provided parameters and
1877  * returns it.
1878  */
1879 static int
1880 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1881     int ln_router)
1882 {
1883 
1884 	/*
1885 	 * ICMP6 type dependent behavior.
1886 	 *
1887 	 * NS: clear IsRouter if new entry
1888 	 * RS: clear IsRouter
1889 	 * RA: set IsRouter if there's lladdr
1890 	 * redir: clear IsRouter if new entry
1891 	 *
1892 	 * RA case, (1):
1893 	 * The spec says that we must set IsRouter in the following cases:
1894 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1895 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1896 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1897 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1898 	 * neighbor cache, this is similar to (6).
1899 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1900 	 *
1901 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1902 	 *							D R
1903 	 *	0	n	n	(1)	c   ?     s
1904 	 *	0	y	n	(2)	c   s     s
1905 	 *	0	n	y	(3)	c   s     s
1906 	 *	0	y	y	(4)	c   s     s
1907 	 *	0	y	y	(5)	c   s     s
1908 	 *	1	--	n	(6) c	c	c s
1909 	 *	1	--	y	(7) c	c   s	c s
1910 	 *
1911 	 *					(c=clear s=set)
1912 	 */
1913 	switch (type & 0xff) {
1914 	case ND_NEIGHBOR_SOLICIT:
1915 		/*
1916 		 * New entry must have is_router flag cleared.
1917 		 */
1918 		if (is_new)					/* (6-7) */
1919 			ln_router = 0;
1920 		break;
1921 	case ND_REDIRECT:
1922 		/*
1923 		 * If the icmp is a redirect to a better router, always set the
1924 		 * is_router flag.  Otherwise, if the entry is newly created,
1925 		 * clear the flag.  [RFC 2461, sec 8.3]
1926 		 */
1927 		if (code == ND_REDIRECT_ROUTER)
1928 			ln_router = 1;
1929 		else {
1930 			if (is_new)				/* (6-7) */
1931 				ln_router = 0;
1932 		}
1933 		break;
1934 	case ND_ROUTER_SOLICIT:
1935 		/*
1936 		 * is_router flag must always be cleared.
1937 		 */
1938 		ln_router = 0;
1939 		break;
1940 	case ND_ROUTER_ADVERT:
1941 		/*
1942 		 * Mark an entry with lladdr as a router.
1943 		 */
1944 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1945 		    (is_new && new_addr)) {			/* (7) */
1946 			ln_router = 1;
1947 		}
1948 		break;
1949 	}
1950 
1951 	return (ln_router);
1952 }
1953 
1954 /*
1955  * Create neighbor cache entry and cache link-layer address,
1956  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1957  *
1958  * type - ICMP6 type
1959  * code - type dependent information
1960  *
1961  */
1962 void
1963 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1964     int lladdrlen, int type, int code)
1965 {
1966 	struct llentry *ln = NULL, *ln_tmp;
1967 	int is_newentry;
1968 	int do_update;
1969 	int olladdr;
1970 	int llchange;
1971 	int flags;
1972 	uint16_t router = 0;
1973 	struct sockaddr_in6 sin6;
1974 	struct mbuf *chain = NULL;
1975 	u_char linkhdr[LLE_MAX_LINKHDR];
1976 	size_t linkhdrsize;
1977 	int lladdr_off;
1978 
1979 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1980 
1981 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1982 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1983 
1984 	/* nothing must be updated for unspecified address */
1985 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1986 		return;
1987 
1988 	/*
1989 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1990 	 * the caller.
1991 	 *
1992 	 * XXX If the link does not have link-layer adderss, what should
1993 	 * we do? (ifp->if_addrlen == 0)
1994 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1995 	 * description on it in NS section (RFC 2461 7.2.3).
1996 	 */
1997 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1998 	IF_AFDATA_RLOCK(ifp);
1999 	ln = nd6_lookup(from, flags, ifp);
2000 	IF_AFDATA_RUNLOCK(ifp);
2001 	is_newentry = 0;
2002 	if (ln == NULL) {
2003 		flags |= LLE_EXCLUSIVE;
2004 		ln = nd6_alloc(from, 0, ifp);
2005 		if (ln == NULL)
2006 			return;
2007 
2008 		/*
2009 		 * Since we already know all the data for the new entry,
2010 		 * fill it before insertion.
2011 		 */
2012 		if (lladdr != NULL) {
2013 			linkhdrsize = sizeof(linkhdr);
2014 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2015 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2016 				return;
2017 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2018 			    lladdr_off);
2019 		}
2020 
2021 		IF_AFDATA_WLOCK(ifp);
2022 		LLE_WLOCK(ln);
2023 		/* Prefer any existing lle over newly-created one */
2024 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
2025 		if (ln_tmp == NULL)
2026 			lltable_link_entry(LLTABLE6(ifp), ln);
2027 		IF_AFDATA_WUNLOCK(ifp);
2028 		if (ln_tmp == NULL) {
2029 			/* No existing lle, mark as new entry (6,7) */
2030 			is_newentry = 1;
2031 			if (lladdr != NULL) {	/* (7) */
2032 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2033 				EVENTHANDLER_INVOKE(lle_event, ln,
2034 				    LLENTRY_RESOLVED);
2035 			}
2036 		} else {
2037 			lltable_free_entry(LLTABLE6(ifp), ln);
2038 			ln = ln_tmp;
2039 			ln_tmp = NULL;
2040 		}
2041 	}
2042 	/* do nothing if static ndp is set */
2043 	if ((ln->la_flags & LLE_STATIC)) {
2044 		if (flags & LLE_EXCLUSIVE)
2045 			LLE_WUNLOCK(ln);
2046 		else
2047 			LLE_RUNLOCK(ln);
2048 		return;
2049 	}
2050 
2051 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2052 	if (olladdr && lladdr) {
2053 		llchange = bcmp(lladdr, ln->ll_addr,
2054 		    ifp->if_addrlen);
2055 	} else if (!olladdr && lladdr)
2056 		llchange = 1;
2057 	else
2058 		llchange = 0;
2059 
2060 	/*
2061 	 * newentry olladdr  lladdr  llchange	(*=record)
2062 	 *	0	n	n	--	(1)
2063 	 *	0	y	n	--	(2)
2064 	 *	0	n	y	y	(3) * STALE
2065 	 *	0	y	y	n	(4) *
2066 	 *	0	y	y	y	(5) * STALE
2067 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2068 	 *	1	--	y	--	(7) * STALE
2069 	 */
2070 
2071 	do_update = 0;
2072 	if (is_newentry == 0 && llchange != 0) {
2073 		do_update = 1;	/* (3,5) */
2074 
2075 		/*
2076 		 * Record source link-layer address
2077 		 * XXX is it dependent to ifp->if_type?
2078 		 */
2079 		linkhdrsize = sizeof(linkhdr);
2080 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2081 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2082 			return;
2083 
2084 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2085 		    lladdr_off) == 0) {
2086 			/* Entry was deleted */
2087 			return;
2088 		}
2089 
2090 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2091 
2092 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2093 
2094 		if (ln->la_hold != NULL)
2095 			nd6_grab_holdchain(ln, &chain, &sin6);
2096 	}
2097 
2098 	/* Calculates new router status */
2099 	router = nd6_is_router(type, code, is_newentry, olladdr,
2100 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2101 
2102 	ln->ln_router = router;
2103 	/* Mark non-router redirects with special flag */
2104 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2105 		ln->la_flags |= LLE_REDIRECT;
2106 
2107 	if (flags & LLE_EXCLUSIVE)
2108 		LLE_WUNLOCK(ln);
2109 	else
2110 		LLE_RUNLOCK(ln);
2111 
2112 	if (chain != NULL)
2113 		nd6_flush_holdchain(ifp, chain, &sin6);
2114 
2115 	/*
2116 	 * When the link-layer address of a router changes, select the
2117 	 * best router again.  In particular, when the neighbor entry is newly
2118 	 * created, it might affect the selection policy.
2119 	 * Question: can we restrict the first condition to the "is_newentry"
2120 	 * case?
2121 	 * XXX: when we hear an RA from a new router with the link-layer
2122 	 * address option, defrouter_select_fib() is called twice, since
2123 	 * defrtrlist_update called the function as well.  However, I believe
2124 	 * we can compromise the overhead, since it only happens the first
2125 	 * time.
2126 	 * XXX: although defrouter_select_fib() should not have a bad effect
2127 	 * for those are not autoconfigured hosts, we explicitly avoid such
2128 	 * cases for safety.
2129 	 */
2130 	if ((do_update || is_newentry) && router &&
2131 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2132 		/*
2133 		 * guaranteed recursion
2134 		 */
2135 		defrouter_select_fib(ifp->if_fib);
2136 	}
2137 }
2138 
2139 static void
2140 nd6_slowtimo(void *arg)
2141 {
2142 	CURVNET_SET((struct vnet *) arg);
2143 	struct nd_ifinfo *nd6if;
2144 	struct ifnet *ifp;
2145 
2146 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2147 	    nd6_slowtimo, curvnet);
2148 	IFNET_RLOCK_NOSLEEP();
2149 	TAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2150 		if (ifp->if_afdata[AF_INET6] == NULL)
2151 			continue;
2152 		nd6if = ND_IFINFO(ifp);
2153 		if (nd6if->basereachable && /* already initialized */
2154 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2155 			/*
2156 			 * Since reachable time rarely changes by router
2157 			 * advertisements, we SHOULD insure that a new random
2158 			 * value gets recomputed at least once every few hours.
2159 			 * (RFC 2461, 6.3.4)
2160 			 */
2161 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2162 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2163 		}
2164 	}
2165 	IFNET_RUNLOCK_NOSLEEP();
2166 	CURVNET_RESTORE();
2167 }
2168 
2169 void
2170 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2171     struct sockaddr_in6 *sin6)
2172 {
2173 
2174 	LLE_WLOCK_ASSERT(ln);
2175 
2176 	*chain = ln->la_hold;
2177 	ln->la_hold = NULL;
2178 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2179 
2180 	if (ln->ln_state == ND6_LLINFO_STALE) {
2181 
2182 		/*
2183 		 * The first time we send a packet to a
2184 		 * neighbor whose entry is STALE, we have
2185 		 * to change the state to DELAY and a sets
2186 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2187 		 * seconds to ensure do neighbor unreachability
2188 		 * detection on expiration.
2189 		 * (RFC 2461 7.3.3)
2190 		 */
2191 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2192 	}
2193 }
2194 
2195 int
2196 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2197     struct sockaddr_in6 *dst, struct route *ro)
2198 {
2199 	int error;
2200 	int ip6len;
2201 	struct ip6_hdr *ip6;
2202 	struct m_tag *mtag;
2203 
2204 #ifdef MAC
2205 	mac_netinet6_nd6_send(ifp, m);
2206 #endif
2207 
2208 	/*
2209 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2210 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2211 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2212 	 * to be diverted to user space.  When re-injected into the kernel,
2213 	 * send_output() will directly dispatch them to the outgoing interface.
2214 	 */
2215 	if (send_sendso_input_hook != NULL) {
2216 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2217 		if (mtag != NULL) {
2218 			ip6 = mtod(m, struct ip6_hdr *);
2219 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2220 			/* Use the SEND socket */
2221 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2222 			    ip6len);
2223 			/* -1 == no app on SEND socket */
2224 			if (error == 0 || error != -1)
2225 			    return (error);
2226 		}
2227 	}
2228 
2229 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2230 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2231 	    mtod(m, struct ip6_hdr *));
2232 
2233 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2234 		origifp = ifp;
2235 
2236 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2237 	return (error);
2238 }
2239 
2240 /*
2241  * Lookup link headerfor @sa_dst address. Stores found
2242  * data in @desten buffer. Copy of lle ln_flags can be also
2243  * saved in @pflags if @pflags is non-NULL.
2244  *
2245  * If destination LLE does not exists or lle state modification
2246  * is required, call "slow" version.
2247  *
2248  * Return values:
2249  * - 0 on success (address copied to buffer).
2250  * - EWOULDBLOCK (no local error, but address is still unresolved)
2251  * - other errors (alloc failure, etc)
2252  */
2253 int
2254 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2255     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2256     struct llentry **plle)
2257 {
2258 	struct llentry *ln = NULL;
2259 	const struct sockaddr_in6 *dst6;
2260 
2261 	if (pflags != NULL)
2262 		*pflags = 0;
2263 
2264 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2265 
2266 	/* discard the packet if IPv6 operation is disabled on the interface */
2267 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2268 		m_freem(m);
2269 		return (ENETDOWN); /* better error? */
2270 	}
2271 
2272 	if (m != NULL && m->m_flags & M_MCAST) {
2273 		switch (ifp->if_type) {
2274 		case IFT_ETHER:
2275 		case IFT_FDDI:
2276 		case IFT_L2VLAN:
2277 		case IFT_BRIDGE:
2278 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2279 						 desten);
2280 			return (0);
2281 		default:
2282 			m_freem(m);
2283 			return (EAFNOSUPPORT);
2284 		}
2285 	}
2286 
2287 	IF_AFDATA_RLOCK(ifp);
2288 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2289 	    ifp);
2290 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2291 		/* Entry found, let's copy lle info */
2292 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2293 		if (pflags != NULL)
2294 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2295 		/* Check if we have feedback request from nd6 timer */
2296 		if (ln->r_skip_req != 0) {
2297 			LLE_REQ_LOCK(ln);
2298 			ln->r_skip_req = 0; /* Notify that entry was used */
2299 			ln->lle_hittime = time_uptime;
2300 			LLE_REQ_UNLOCK(ln);
2301 		}
2302 		if (plle) {
2303 			LLE_ADDREF(ln);
2304 			*plle = ln;
2305 			LLE_WUNLOCK(ln);
2306 		}
2307 		IF_AFDATA_RUNLOCK(ifp);
2308 		return (0);
2309 	} else if (plle && ln)
2310 		LLE_WUNLOCK(ln);
2311 	IF_AFDATA_RUNLOCK(ifp);
2312 
2313 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2314 }
2315 
2316 
2317 /*
2318  * Do L2 address resolution for @sa_dst address. Stores found
2319  * address in @desten buffer. Copy of lle ln_flags can be also
2320  * saved in @pflags if @pflags is non-NULL.
2321  *
2322  * Heavy version.
2323  * Function assume that destination LLE does not exist,
2324  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2325  *
2326  * Set noinline to be dtrace-friendly
2327  */
2328 static __noinline int
2329 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2330     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2331     struct llentry **plle)
2332 {
2333 	struct llentry *lle = NULL, *lle_tmp;
2334 	struct in6_addr *psrc, src;
2335 	int send_ns, ll_len;
2336 	char *lladdr;
2337 
2338 	/*
2339 	 * Address resolution or Neighbor Unreachability Detection
2340 	 * for the next hop.
2341 	 * At this point, the destination of the packet must be a unicast
2342 	 * or an anycast address(i.e. not a multicast).
2343 	 */
2344 	if (lle == NULL) {
2345 		IF_AFDATA_RLOCK(ifp);
2346 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2347 		IF_AFDATA_RUNLOCK(ifp);
2348 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2349 			/*
2350 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2351 			 * the condition below is not very efficient.  But we believe
2352 			 * it is tolerable, because this should be a rare case.
2353 			 */
2354 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2355 			if (lle == NULL) {
2356 				char ip6buf[INET6_ADDRSTRLEN];
2357 				log(LOG_DEBUG,
2358 				    "nd6_output: can't allocate llinfo for %s "
2359 				    "(ln=%p)\n",
2360 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2361 				m_freem(m);
2362 				return (ENOBUFS);
2363 			}
2364 
2365 			IF_AFDATA_WLOCK(ifp);
2366 			LLE_WLOCK(lle);
2367 			/* Prefer any existing entry over newly-created one */
2368 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2369 			if (lle_tmp == NULL)
2370 				lltable_link_entry(LLTABLE6(ifp), lle);
2371 			IF_AFDATA_WUNLOCK(ifp);
2372 			if (lle_tmp != NULL) {
2373 				lltable_free_entry(LLTABLE6(ifp), lle);
2374 				lle = lle_tmp;
2375 				lle_tmp = NULL;
2376 			}
2377 		}
2378 	}
2379 	if (lle == NULL) {
2380 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2381 			m_freem(m);
2382 			return (ENOBUFS);
2383 		}
2384 
2385 		if (m != NULL)
2386 			m_freem(m);
2387 		return (ENOBUFS);
2388 	}
2389 
2390 	LLE_WLOCK_ASSERT(lle);
2391 
2392 	/*
2393 	 * The first time we send a packet to a neighbor whose entry is
2394 	 * STALE, we have to change the state to DELAY and a sets a timer to
2395 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2396 	 * neighbor unreachability detection on expiration.
2397 	 * (RFC 2461 7.3.3)
2398 	 */
2399 	if (lle->ln_state == ND6_LLINFO_STALE)
2400 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2401 
2402 	/*
2403 	 * If the neighbor cache entry has a state other than INCOMPLETE
2404 	 * (i.e. its link-layer address is already resolved), just
2405 	 * send the packet.
2406 	 */
2407 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2408 		if (flags & LLE_ADDRONLY) {
2409 			lladdr = lle->ll_addr;
2410 			ll_len = ifp->if_addrlen;
2411 		} else {
2412 			lladdr = lle->r_linkdata;
2413 			ll_len = lle->r_hdrlen;
2414 		}
2415 		bcopy(lladdr, desten, ll_len);
2416 		if (pflags != NULL)
2417 			*pflags = lle->la_flags;
2418 		if (plle) {
2419 			LLE_ADDREF(lle);
2420 			*plle = lle;
2421 		}
2422 		LLE_WUNLOCK(lle);
2423 		return (0);
2424 	}
2425 
2426 	/*
2427 	 * There is a neighbor cache entry, but no ethernet address
2428 	 * response yet.  Append this latest packet to the end of the
2429 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2430 	 * the oldest packet in the queue will be removed.
2431 	 */
2432 
2433 	if (lle->la_hold != NULL) {
2434 		struct mbuf *m_hold;
2435 		int i;
2436 
2437 		i = 0;
2438 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2439 			i++;
2440 			if (m_hold->m_nextpkt == NULL) {
2441 				m_hold->m_nextpkt = m;
2442 				break;
2443 			}
2444 		}
2445 		while (i >= V_nd6_maxqueuelen) {
2446 			m_hold = lle->la_hold;
2447 			lle->la_hold = lle->la_hold->m_nextpkt;
2448 			m_freem(m_hold);
2449 			i--;
2450 		}
2451 	} else {
2452 		lle->la_hold = m;
2453 	}
2454 
2455 	/*
2456 	 * If there has been no NS for the neighbor after entering the
2457 	 * INCOMPLETE state, send the first solicitation.
2458 	 * Note that for newly-created lle la_asked will be 0,
2459 	 * so we will transition from ND6_LLINFO_NOSTATE to
2460 	 * ND6_LLINFO_INCOMPLETE state here.
2461 	 */
2462 	psrc = NULL;
2463 	send_ns = 0;
2464 	if (lle->la_asked == 0) {
2465 		lle->la_asked++;
2466 		send_ns = 1;
2467 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2468 
2469 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2470 	}
2471 	LLE_WUNLOCK(lle);
2472 	if (send_ns != 0)
2473 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2474 
2475 	return (EWOULDBLOCK);
2476 }
2477 
2478 /*
2479  * Do L2 address resolution for @sa_dst address. Stores found
2480  * address in @desten buffer. Copy of lle ln_flags can be also
2481  * saved in @pflags if @pflags is non-NULL.
2482  *
2483  * Return values:
2484  * - 0 on success (address copied to buffer).
2485  * - EWOULDBLOCK (no local error, but address is still unresolved)
2486  * - other errors (alloc failure, etc)
2487  */
2488 int
2489 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2490     char *desten, uint32_t *pflags)
2491 {
2492 	int error;
2493 
2494 	flags |= LLE_ADDRONLY;
2495 	error = nd6_resolve_slow(ifp, flags, NULL,
2496 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2497 	return (error);
2498 }
2499 
2500 int
2501 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2502     struct sockaddr_in6 *dst)
2503 {
2504 	struct mbuf *m, *m_head;
2505 	int error = 0;
2506 
2507 	m_head = chain;
2508 
2509 	while (m_head) {
2510 		m = m_head;
2511 		m_head = m_head->m_nextpkt;
2512 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2513 	}
2514 
2515 	/*
2516 	 * XXX
2517 	 * note that intermediate errors are blindly ignored
2518 	 */
2519 	return (error);
2520 }
2521 
2522 static int
2523 nd6_need_cache(struct ifnet *ifp)
2524 {
2525 	/*
2526 	 * XXX: we currently do not make neighbor cache on any interface
2527 	 * other than ARCnet, Ethernet, FDDI and GIF.
2528 	 *
2529 	 * RFC2893 says:
2530 	 * - unidirectional tunnels needs no ND
2531 	 */
2532 	switch (ifp->if_type) {
2533 	case IFT_ARCNET:
2534 	case IFT_ETHER:
2535 	case IFT_FDDI:
2536 	case IFT_IEEE1394:
2537 	case IFT_L2VLAN:
2538 	case IFT_INFINIBAND:
2539 	case IFT_BRIDGE:
2540 	case IFT_PROPVIRTUAL:
2541 		return (1);
2542 	default:
2543 		return (0);
2544 	}
2545 }
2546 
2547 /*
2548  * Add pernament ND6 link-layer record for given
2549  * interface address.
2550  *
2551  * Very similar to IPv4 arp_ifinit(), but:
2552  * 1) IPv6 DAD is performed in different place
2553  * 2) It is called by IPv6 protocol stack in contrast to
2554  * arp_ifinit() which is typically called in SIOCSIFADDR
2555  * driver ioctl handler.
2556  *
2557  */
2558 int
2559 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2560 {
2561 	struct ifnet *ifp;
2562 	struct llentry *ln, *ln_tmp;
2563 	struct sockaddr *dst;
2564 
2565 	ifp = ia->ia_ifa.ifa_ifp;
2566 	if (nd6_need_cache(ifp) == 0)
2567 		return (0);
2568 
2569 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2570 	dst = (struct sockaddr *)&ia->ia_addr;
2571 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2572 	if (ln == NULL)
2573 		return (ENOBUFS);
2574 
2575 	IF_AFDATA_WLOCK(ifp);
2576 	LLE_WLOCK(ln);
2577 	/* Unlink any entry if exists */
2578 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2579 	if (ln_tmp != NULL)
2580 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2581 	lltable_link_entry(LLTABLE6(ifp), ln);
2582 	IF_AFDATA_WUNLOCK(ifp);
2583 
2584 	if (ln_tmp != NULL)
2585 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2586 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2587 
2588 	LLE_WUNLOCK(ln);
2589 	if (ln_tmp != NULL)
2590 		llentry_free(ln_tmp);
2591 
2592 	return (0);
2593 }
2594 
2595 /*
2596  * Removes either all lle entries for given @ia, or lle
2597  * corresponding to @ia address.
2598  */
2599 void
2600 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2601 {
2602 	struct sockaddr_in6 mask, addr;
2603 	struct sockaddr *saddr, *smask;
2604 	struct ifnet *ifp;
2605 
2606 	ifp = ia->ia_ifa.ifa_ifp;
2607 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2608 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2609 	saddr = (struct sockaddr *)&addr;
2610 	smask = (struct sockaddr *)&mask;
2611 
2612 	if (all != 0)
2613 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2614 	else
2615 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2616 }
2617 
2618 static void
2619 clear_llinfo_pqueue(struct llentry *ln)
2620 {
2621 	struct mbuf *m_hold, *m_hold_next;
2622 
2623 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2624 		m_hold_next = m_hold->m_nextpkt;
2625 		m_freem(m_hold);
2626 	}
2627 
2628 	ln->la_hold = NULL;
2629 }
2630 
2631 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2632 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2633 
2634 SYSCTL_DECL(_net_inet6_icmp6);
2635 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2636 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2637 	NULL, 0, nd6_sysctl_drlist, "S,in6_defrouter",
2638 	"NDP default router list");
2639 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2640 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2641 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2642 	"NDP prefix list");
2643 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2644 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2645 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2646 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2647 
2648 static int
2649 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2650 {
2651 	struct in6_defrouter d;
2652 	struct nd_defrouter *dr;
2653 	int error;
2654 
2655 	if (req->newptr != NULL)
2656 		return (EPERM);
2657 
2658 	error = sysctl_wire_old_buffer(req, 0);
2659 	if (error != 0)
2660 		return (error);
2661 
2662 	bzero(&d, sizeof(d));
2663 	d.rtaddr.sin6_family = AF_INET6;
2664 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2665 
2666 	ND6_RLOCK();
2667 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2668 		d.rtaddr.sin6_addr = dr->rtaddr;
2669 		error = sa6_recoverscope(&d.rtaddr);
2670 		if (error != 0)
2671 			break;
2672 		d.flags = dr->raflags;
2673 		d.rtlifetime = dr->rtlifetime;
2674 		d.expire = dr->expire + (time_second - time_uptime);
2675 		d.if_index = dr->ifp->if_index;
2676 		error = SYSCTL_OUT(req, &d, sizeof(d));
2677 		if (error != 0)
2678 			break;
2679 	}
2680 	ND6_RUNLOCK();
2681 	return (error);
2682 }
2683 
2684 static int
2685 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2686 {
2687 	struct in6_prefix p;
2688 	struct sockaddr_in6 s6;
2689 	struct nd_prefix *pr;
2690 	struct nd_pfxrouter *pfr;
2691 	time_t maxexpire;
2692 	int error;
2693 	char ip6buf[INET6_ADDRSTRLEN];
2694 
2695 	if (req->newptr)
2696 		return (EPERM);
2697 
2698 	error = sysctl_wire_old_buffer(req, 0);
2699 	if (error != 0)
2700 		return (error);
2701 
2702 	bzero(&p, sizeof(p));
2703 	p.origin = PR_ORIG_RA;
2704 	bzero(&s6, sizeof(s6));
2705 	s6.sin6_family = AF_INET6;
2706 	s6.sin6_len = sizeof(s6);
2707 
2708 	ND6_RLOCK();
2709 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2710 		p.prefix = pr->ndpr_prefix;
2711 		if (sa6_recoverscope(&p.prefix)) {
2712 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2713 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2714 			/* XXX: press on... */
2715 		}
2716 		p.raflags = pr->ndpr_raf;
2717 		p.prefixlen = pr->ndpr_plen;
2718 		p.vltime = pr->ndpr_vltime;
2719 		p.pltime = pr->ndpr_pltime;
2720 		p.if_index = pr->ndpr_ifp->if_index;
2721 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2722 			p.expire = 0;
2723 		else {
2724 			/* XXX: we assume time_t is signed. */
2725 			maxexpire = (-1) &
2726 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2727 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2728 				p.expire = pr->ndpr_lastupdate +
2729 				    pr->ndpr_vltime +
2730 				    (time_second - time_uptime);
2731 			else
2732 				p.expire = maxexpire;
2733 		}
2734 		p.refcnt = pr->ndpr_addrcnt;
2735 		p.flags = pr->ndpr_stateflags;
2736 		p.advrtrs = 0;
2737 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2738 			p.advrtrs++;
2739 		error = SYSCTL_OUT(req, &p, sizeof(p));
2740 		if (error != 0)
2741 			break;
2742 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2743 			s6.sin6_addr = pfr->router->rtaddr;
2744 			if (sa6_recoverscope(&s6))
2745 				log(LOG_ERR,
2746 				    "scope error in prefix list (%s)\n",
2747 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2748 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2749 			if (error != 0)
2750 				goto out;
2751 		}
2752 	}
2753 out:
2754 	ND6_RUNLOCK();
2755 	return (error);
2756 }
2757