xref: /freebsd/sys/netinet6/nd6.c (revision e2eeea75eb8b6dd50c1298067a0655880d186734)
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 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
277 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
278 	 * default regardless of the V_ip6_auto_linklocal configuration to
279 	 * give a reasonable default behavior.
280 	 */
281 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
282 	    (ifp->if_flags & IFF_LOOPBACK))
283 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
284 	/*
285 	 * A loopback interface does not need to accept RTADV.
286 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
287 	 * default regardless of the V_ip6_accept_rtadv configuration to
288 	 * prevent the interface from accepting RA messages arrived
289 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
290 	 */
291 	if (V_ip6_accept_rtadv &&
292 	    !(ifp->if_flags & IFF_LOOPBACK) &&
293 	    (ifp->if_type != IFT_BRIDGE))
294 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
295 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
296 		nd->flags |= ND6_IFF_NO_RADR;
297 
298 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
299 	nd6_setmtu0(ifp, nd);
300 
301 	return nd;
302 }
303 
304 void
305 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
306 {
307 	struct epoch_tracker et;
308 	struct ifaddr *ifa, *next;
309 
310 	NET_EPOCH_ENTER(et);
311 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
312 		if (ifa->ifa_addr->sa_family != AF_INET6)
313 			continue;
314 
315 		/* stop DAD processing */
316 		nd6_dad_stop(ifa);
317 	}
318 	NET_EPOCH_EXIT(et);
319 
320 	free(nd, M_IP6NDP);
321 }
322 
323 /*
324  * Reset ND level link MTU. This function is called when the physical MTU
325  * changes, which means we might have to adjust the ND level MTU.
326  */
327 void
328 nd6_setmtu(struct ifnet *ifp)
329 {
330 	if (ifp->if_afdata[AF_INET6] == NULL)
331 		return;
332 
333 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
334 }
335 
336 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
337 void
338 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
339 {
340 	u_int32_t omaxmtu;
341 
342 	omaxmtu = ndi->maxmtu;
343 	ndi->maxmtu = ifp->if_mtu;
344 
345 	/*
346 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
347 	 * undesirable situation.  We thus notify the operator of the change
348 	 * explicitly.  The check for omaxmtu is necessary to restrict the
349 	 * log to the case of changing the MTU, not initializing it.
350 	 */
351 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
352 		log(LOG_NOTICE, "nd6_setmtu0: "
353 		    "new link MTU on %s (%lu) is too small for IPv6\n",
354 		    if_name(ifp), (unsigned long)ndi->maxmtu);
355 	}
356 
357 	if (ndi->maxmtu > V_in6_maxmtu)
358 		in6_setmaxmtu(); /* check all interfaces just in case */
359 
360 }
361 
362 void
363 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
364 {
365 
366 	bzero(ndopts, sizeof(*ndopts));
367 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
368 	ndopts->nd_opts_last
369 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
370 
371 	if (icmp6len == 0) {
372 		ndopts->nd_opts_done = 1;
373 		ndopts->nd_opts_search = NULL;
374 	}
375 }
376 
377 /*
378  * Take one ND option.
379  */
380 struct nd_opt_hdr *
381 nd6_option(union nd_opts *ndopts)
382 {
383 	struct nd_opt_hdr *nd_opt;
384 	int olen;
385 
386 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
387 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
388 	    __func__));
389 	if (ndopts->nd_opts_search == NULL)
390 		return NULL;
391 	if (ndopts->nd_opts_done)
392 		return NULL;
393 
394 	nd_opt = ndopts->nd_opts_search;
395 
396 	/* make sure nd_opt_len is inside the buffer */
397 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
398 		bzero(ndopts, sizeof(*ndopts));
399 		return NULL;
400 	}
401 
402 	olen = nd_opt->nd_opt_len << 3;
403 	if (olen == 0) {
404 		/*
405 		 * Message validation requires that all included
406 		 * options have a length that is greater than zero.
407 		 */
408 		bzero(ndopts, sizeof(*ndopts));
409 		return NULL;
410 	}
411 
412 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
413 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
414 		/* option overruns the end of buffer, invalid */
415 		bzero(ndopts, sizeof(*ndopts));
416 		return NULL;
417 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
418 		/* reached the end of options chain */
419 		ndopts->nd_opts_done = 1;
420 		ndopts->nd_opts_search = NULL;
421 	}
422 	return nd_opt;
423 }
424 
425 /*
426  * Parse multiple ND options.
427  * This function is much easier to use, for ND routines that do not need
428  * multiple options of the same type.
429  */
430 int
431 nd6_options(union nd_opts *ndopts)
432 {
433 	struct nd_opt_hdr *nd_opt;
434 	int i = 0;
435 
436 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
437 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
438 	    __func__));
439 	if (ndopts->nd_opts_search == NULL)
440 		return 0;
441 
442 	while (1) {
443 		nd_opt = nd6_option(ndopts);
444 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
445 			/*
446 			 * Message validation requires that all included
447 			 * options have a length that is greater than zero.
448 			 */
449 			ICMP6STAT_INC(icp6s_nd_badopt);
450 			bzero(ndopts, sizeof(*ndopts));
451 			return -1;
452 		}
453 
454 		if (nd_opt == NULL)
455 			goto skip1;
456 
457 		switch (nd_opt->nd_opt_type) {
458 		case ND_OPT_SOURCE_LINKADDR:
459 		case ND_OPT_TARGET_LINKADDR:
460 		case ND_OPT_MTU:
461 		case ND_OPT_REDIRECTED_HEADER:
462 		case ND_OPT_NONCE:
463 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
464 				nd6log((LOG_INFO,
465 				    "duplicated ND6 option found (type=%d)\n",
466 				    nd_opt->nd_opt_type));
467 				/* XXX bark? */
468 			} else {
469 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
470 					= nd_opt;
471 			}
472 			break;
473 		case ND_OPT_PREFIX_INFORMATION:
474 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
475 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
476 					= nd_opt;
477 			}
478 			ndopts->nd_opts_pi_end =
479 				(struct nd_opt_prefix_info *)nd_opt;
480 			break;
481 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
482 		case ND_OPT_RDNSS:	/* RFC 6106 */
483 		case ND_OPT_DNSSL:	/* RFC 6106 */
484 			/*
485 			 * Silently ignore options we know and do not care about
486 			 * in the kernel.
487 			 */
488 			break;
489 		default:
490 			/*
491 			 * Unknown options must be silently ignored,
492 			 * to accommodate future extension to the protocol.
493 			 */
494 			nd6log((LOG_DEBUG,
495 			    "nd6_options: unsupported option %d - "
496 			    "option ignored\n", nd_opt->nd_opt_type));
497 		}
498 
499 skip1:
500 		i++;
501 		if (i > V_nd6_maxndopt) {
502 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
503 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
504 			break;
505 		}
506 
507 		if (ndopts->nd_opts_done)
508 			break;
509 	}
510 
511 	return 0;
512 }
513 
514 /*
515  * ND6 timer routine to handle ND6 entries
516  */
517 static void
518 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
519 {
520 	int canceled;
521 
522 	LLE_WLOCK_ASSERT(ln);
523 
524 	if (tick < 0) {
525 		ln->la_expire = 0;
526 		ln->ln_ntick = 0;
527 		canceled = callout_stop(&ln->lle_timer);
528 	} else {
529 		ln->la_expire = time_uptime + tick / hz;
530 		LLE_ADDREF(ln);
531 		if (tick > INT_MAX) {
532 			ln->ln_ntick = tick - INT_MAX;
533 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
534 			    nd6_llinfo_timer, ln);
535 		} else {
536 			ln->ln_ntick = 0;
537 			canceled = callout_reset(&ln->lle_timer, tick,
538 			    nd6_llinfo_timer, ln);
539 		}
540 	}
541 	if (canceled > 0)
542 		LLE_REMREF(ln);
543 }
544 
545 /*
546  * Gets source address of the first packet in hold queue
547  * and stores it in @src.
548  * Returns pointer to @src (if hold queue is not empty) or NULL.
549  *
550  * Set noinline to be dtrace-friendly
551  */
552 static __noinline struct in6_addr *
553 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
554 {
555 	struct ip6_hdr hdr;
556 	struct mbuf *m;
557 
558 	if (ln->la_hold == NULL)
559 		return (NULL);
560 
561 	/*
562 	 * assume every packet in la_hold has the same IP header
563 	 */
564 	m = ln->la_hold;
565 	if (sizeof(hdr) > m->m_len)
566 		return (NULL);
567 
568 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
569 	*src = hdr.ip6_src;
570 
571 	return (src);
572 }
573 
574 /*
575  * Checks if we need to switch from STALE state.
576  *
577  * RFC 4861 requires switching from STALE to DELAY state
578  * on first packet matching entry, waiting V_nd6_delay and
579  * transition to PROBE state (if upper layer confirmation was
580  * not received).
581  *
582  * This code performs a bit differently:
583  * On packet hit we don't change state (but desired state
584  * can be guessed by control plane). However, after V_nd6_delay
585  * seconds code will transition to PROBE state (so DELAY state
586  * is kinda skipped in most situations).
587  *
588  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
589  * we perform the following upon entering STALE state:
590  *
591  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
592  * if packet was transmitted at the start of given interval, we
593  * would be able to switch to PROBE state in V_nd6_delay seconds
594  * as user expects.
595  *
596  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
597  * lle in STALE state (remaining timer value stored in lle_remtime).
598  *
599  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
600  * seconds ago.
601  *
602  * Returns non-zero value if the entry is still STALE (storing
603  * the next timer interval in @pdelay).
604  *
605  * Returns zero value if original timer expired or we need to switch to
606  * PROBE (store that in @do_switch variable).
607  */
608 static int
609 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
610 {
611 	int nd_delay, nd_gctimer, r_skip_req;
612 	time_t lle_hittime;
613 	long delay;
614 
615 	*do_switch = 0;
616 	nd_gctimer = V_nd6_gctimer;
617 	nd_delay = V_nd6_delay;
618 
619 	LLE_REQ_LOCK(lle);
620 	r_skip_req = lle->r_skip_req;
621 	lle_hittime = lle->lle_hittime;
622 	LLE_REQ_UNLOCK(lle);
623 
624 	if (r_skip_req > 0) {
625 		/*
626 		 * Nonzero r_skip_req value was set upon entering
627 		 * STALE state. Since value was not changed, no
628 		 * packets were passed using this lle. Ask for
629 		 * timer reschedule and keep STALE state.
630 		 */
631 		delay = (long)(MIN(nd_gctimer, nd_delay));
632 		delay *= hz;
633 		if (lle->lle_remtime > delay)
634 			lle->lle_remtime -= delay;
635 		else {
636 			delay = lle->lle_remtime;
637 			lle->lle_remtime = 0;
638 		}
639 
640 		if (delay == 0) {
641 			/*
642 			 * The original ng6_gctime timeout ended,
643 			 * no more rescheduling.
644 			 */
645 			return (0);
646 		}
647 
648 		*pdelay = delay;
649 		return (1);
650 	}
651 
652 	/*
653 	 * Packet received. Verify timestamp
654 	 */
655 	delay = (long)(time_uptime - lle_hittime);
656 	if (delay < nd_delay) {
657 		/*
658 		 * V_nd6_delay still not passed since the first
659 		 * hit in STALE state.
660 		 * Reshedule timer and return.
661 		 */
662 		*pdelay = (long)(nd_delay - delay) * hz;
663 		return (1);
664 	}
665 
666 	/* Request switching to probe */
667 	*do_switch = 1;
668 	return (0);
669 }
670 
671 /*
672  * Switch @lle state to new state optionally arming timers.
673  *
674  * Set noinline to be dtrace-friendly
675  */
676 __noinline void
677 nd6_llinfo_setstate(struct llentry *lle, int newstate)
678 {
679 	struct ifnet *ifp;
680 	int nd_gctimer, nd_delay;
681 	long delay, remtime;
682 
683 	delay = 0;
684 	remtime = 0;
685 
686 	switch (newstate) {
687 	case ND6_LLINFO_INCOMPLETE:
688 		ifp = lle->lle_tbl->llt_ifp;
689 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
690 		break;
691 	case ND6_LLINFO_REACHABLE:
692 		if (!ND6_LLINFO_PERMANENT(lle)) {
693 			ifp = lle->lle_tbl->llt_ifp;
694 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
695 		}
696 		break;
697 	case ND6_LLINFO_STALE:
698 
699 		/*
700 		 * Notify fast path that we want to know if any packet
701 		 * is transmitted by setting r_skip_req.
702 		 */
703 		LLE_REQ_LOCK(lle);
704 		lle->r_skip_req = 1;
705 		LLE_REQ_UNLOCK(lle);
706 		nd_delay = V_nd6_delay;
707 		nd_gctimer = V_nd6_gctimer;
708 
709 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
710 		remtime = (long)nd_gctimer * hz - delay;
711 		break;
712 	case ND6_LLINFO_DELAY:
713 		lle->la_asked = 0;
714 		delay = (long)V_nd6_delay * hz;
715 		break;
716 	}
717 
718 	if (delay > 0)
719 		nd6_llinfo_settimer_locked(lle, delay);
720 
721 	lle->lle_remtime = remtime;
722 	lle->ln_state = newstate;
723 }
724 
725 /*
726  * Timer-dependent part of nd state machine.
727  *
728  * Set noinline to be dtrace-friendly
729  */
730 static __noinline void
731 nd6_llinfo_timer(void *arg)
732 {
733 	struct epoch_tracker et;
734 	struct llentry *ln;
735 	struct in6_addr *dst, *pdst, *psrc, src;
736 	struct ifnet *ifp;
737 	struct nd_ifinfo *ndi;
738 	int do_switch, send_ns;
739 	long delay;
740 
741 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
742 	ln = (struct llentry *)arg;
743 	ifp = lltable_get_ifp(ln->lle_tbl);
744 	CURVNET_SET(ifp->if_vnet);
745 
746 	ND6_RLOCK();
747 	LLE_WLOCK(ln);
748 	if (callout_pending(&ln->lle_timer)) {
749 		/*
750 		 * Here we are a bit odd here in the treatment of
751 		 * active/pending. If the pending bit is set, it got
752 		 * rescheduled before I ran. The active
753 		 * bit we ignore, since if it was stopped
754 		 * in ll_tablefree() and was currently running
755 		 * it would have return 0 so the code would
756 		 * not have deleted it since the callout could
757 		 * not be stopped so we want to go through
758 		 * with the delete here now. If the callout
759 		 * was restarted, the pending bit will be back on and
760 		 * we just want to bail since the callout_reset would
761 		 * return 1 and our reference would have been removed
762 		 * by nd6_llinfo_settimer_locked above since canceled
763 		 * would have been 1.
764 		 */
765 		LLE_WUNLOCK(ln);
766 		ND6_RUNLOCK();
767 		CURVNET_RESTORE();
768 		return;
769 	}
770 	NET_EPOCH_ENTER(et);
771 	ndi = ND_IFINFO(ifp);
772 	send_ns = 0;
773 	dst = &ln->r_l3addr.addr6;
774 	pdst = dst;
775 
776 	if (ln->ln_ntick > 0) {
777 		if (ln->ln_ntick > INT_MAX) {
778 			ln->ln_ntick -= INT_MAX;
779 			nd6_llinfo_settimer_locked(ln, INT_MAX);
780 		} else {
781 			ln->ln_ntick = 0;
782 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
783 		}
784 		goto done;
785 	}
786 
787 	if (ln->la_flags & LLE_STATIC) {
788 		goto done;
789 	}
790 
791 	if (ln->la_flags & LLE_DELETED) {
792 		nd6_free(&ln, 0);
793 		goto done;
794 	}
795 
796 	switch (ln->ln_state) {
797 	case ND6_LLINFO_INCOMPLETE:
798 		if (ln->la_asked < V_nd6_mmaxtries) {
799 			ln->la_asked++;
800 			send_ns = 1;
801 			/* Send NS to multicast address */
802 			pdst = NULL;
803 		} else {
804 			struct mbuf *m = ln->la_hold;
805 			if (m) {
806 				struct mbuf *m0;
807 
808 				/*
809 				 * assuming every packet in la_hold has the
810 				 * same IP header.  Send error after unlock.
811 				 */
812 				m0 = m->m_nextpkt;
813 				m->m_nextpkt = NULL;
814 				ln->la_hold = m0;
815 				clear_llinfo_pqueue(ln);
816 			}
817 			nd6_free(&ln, 0);
818 			if (m != NULL) {
819 				struct mbuf *n = m;
820 
821 				/*
822 				 * if there are any ummapped mbufs, we
823 				 * must free them, rather than using
824 				 * them for an ICMP, as they cannot be
825 				 * checksummed.
826 				 */
827 				while ((n = n->m_next) != NULL) {
828 					if (n->m_flags & M_EXTPG)
829 						break;
830 				}
831 				if (n != NULL) {
832 					m_freem(m);
833 					m = NULL;
834 				} else {
835 					icmp6_error2(m, ICMP6_DST_UNREACH,
836 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
837 				}
838 			}
839 		}
840 		break;
841 	case ND6_LLINFO_REACHABLE:
842 		if (!ND6_LLINFO_PERMANENT(ln))
843 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
844 		break;
845 
846 	case ND6_LLINFO_STALE:
847 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
848 			/*
849 			 * No packet has used this entry and GC timeout
850 			 * has not been passed. Reshedule timer and
851 			 * return.
852 			 */
853 			nd6_llinfo_settimer_locked(ln, delay);
854 			break;
855 		}
856 
857 		if (do_switch == 0) {
858 			/*
859 			 * GC timer has ended and entry hasn't been used.
860 			 * Run Garbage collector (RFC 4861, 5.3)
861 			 */
862 			if (!ND6_LLINFO_PERMANENT(ln))
863 				nd6_free(&ln, 1);
864 			break;
865 		}
866 
867 		/* Entry has been used AND delay timer has ended. */
868 
869 		/* FALLTHROUGH */
870 
871 	case ND6_LLINFO_DELAY:
872 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
873 			/* We need NUD */
874 			ln->la_asked = 1;
875 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
876 			send_ns = 1;
877 		} else
878 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
879 		break;
880 	case ND6_LLINFO_PROBE:
881 		if (ln->la_asked < V_nd6_umaxtries) {
882 			ln->la_asked++;
883 			send_ns = 1;
884 		} else {
885 			nd6_free(&ln, 0);
886 		}
887 		break;
888 	default:
889 		panic("%s: paths in a dark night can be confusing: %d",
890 		    __func__, ln->ln_state);
891 	}
892 done:
893 	if (ln != NULL)
894 		ND6_RUNLOCK();
895 	if (send_ns != 0) {
896 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
897 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
898 		LLE_FREE_LOCKED(ln);
899 		ln = NULL;
900 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
901 	}
902 
903 	if (ln != NULL)
904 		LLE_FREE_LOCKED(ln);
905 	NET_EPOCH_EXIT(et);
906 	CURVNET_RESTORE();
907 }
908 
909 /*
910  * ND6 timer routine to expire default route list and prefix list
911  */
912 void
913 nd6_timer(void *arg)
914 {
915 	CURVNET_SET((struct vnet *) arg);
916 	struct epoch_tracker et;
917 	struct nd_prhead prl;
918 	struct nd_prefix *pr, *npr;
919 	struct ifnet *ifp;
920 	struct in6_ifaddr *ia6, *nia6;
921 	uint64_t genid;
922 
923 	LIST_INIT(&prl);
924 
925 	NET_EPOCH_ENTER(et);
926 	nd6_defrouter_timer();
927 
928 	/*
929 	 * expire interface addresses.
930 	 * in the past the loop was inside prefix expiry processing.
931 	 * However, from a stricter speci-confrmance standpoint, we should
932 	 * rather separate address lifetimes and prefix lifetimes.
933 	 *
934 	 * XXXRW: in6_ifaddrhead locking.
935 	 */
936   addrloop:
937 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
938 		/* check address lifetime */
939 		if (IFA6_IS_INVALID(ia6)) {
940 			int regen = 0;
941 
942 			/*
943 			 * If the expiring address is temporary, try
944 			 * regenerating a new one.  This would be useful when
945 			 * we suspended a laptop PC, then turned it on after a
946 			 * period that could invalidate all temporary
947 			 * addresses.  Although we may have to restart the
948 			 * loop (see below), it must be after purging the
949 			 * address.  Otherwise, we'd see an infinite loop of
950 			 * regeneration.
951 			 */
952 			if (V_ip6_use_tempaddr &&
953 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
954 				if (regen_tmpaddr(ia6) == 0)
955 					regen = 1;
956 			}
957 
958 			in6_purgeaddr(&ia6->ia_ifa);
959 
960 			if (regen)
961 				goto addrloop; /* XXX: see below */
962 		} else if (IFA6_IS_DEPRECATED(ia6)) {
963 			int oldflags = ia6->ia6_flags;
964 
965 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
966 
967 			/*
968 			 * If a temporary address has just become deprecated,
969 			 * regenerate a new one if possible.
970 			 */
971 			if (V_ip6_use_tempaddr &&
972 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
973 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
974 				if (regen_tmpaddr(ia6) == 0) {
975 					/*
976 					 * A new temporary address is
977 					 * generated.
978 					 * XXX: this means the address chain
979 					 * has changed while we are still in
980 					 * the loop.  Although the change
981 					 * would not cause disaster (because
982 					 * it's not a deletion, but an
983 					 * addition,) we'd rather restart the
984 					 * loop just for safety.  Or does this
985 					 * significantly reduce performance??
986 					 */
987 					goto addrloop;
988 				}
989 			}
990 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
991 			/*
992 			 * Schedule DAD for a tentative address.  This happens
993 			 * if the interface was down or not running
994 			 * when the address was configured.
995 			 */
996 			int delay;
997 
998 			delay = arc4random() %
999 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1000 			nd6_dad_start((struct ifaddr *)ia6, delay);
1001 		} else {
1002 			/*
1003 			 * Check status of the interface.  If it is down,
1004 			 * mark the address as tentative for future DAD.
1005 			 */
1006 			ifp = ia6->ia_ifp;
1007 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1008 			    ((ifp->if_flags & IFF_UP) == 0 ||
1009 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1010 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1011 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1012 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1013 			}
1014 
1015 			/*
1016 			 * A new RA might have made a deprecated address
1017 			 * preferred.
1018 			 */
1019 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1020 		}
1021 	}
1022 	NET_EPOCH_EXIT(et);
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 	NET_EPOCH_ASSERT();
1083 
1084 	ifp = ia6->ia_ifa.ifa_ifp;
1085 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1086 		struct in6_ifaddr *it6;
1087 
1088 		if (ifa->ifa_addr->sa_family != AF_INET6)
1089 			continue;
1090 
1091 		it6 = (struct in6_ifaddr *)ifa;
1092 
1093 		/* ignore no autoconf addresses. */
1094 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1095 			continue;
1096 
1097 		/* ignore autoconf addresses with different prefixes. */
1098 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1099 			continue;
1100 
1101 		/*
1102 		 * Now we are looking at an autoconf address with the same
1103 		 * prefix as ours.  If the address is temporary and is still
1104 		 * preferred, do not create another one.  It would be rare, but
1105 		 * could happen, for example, when we resume a laptop PC after
1106 		 * a long period.
1107 		 */
1108 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1109 		    !IFA6_IS_DEPRECATED(it6)) {
1110 			public_ifa6 = NULL;
1111 			break;
1112 		}
1113 
1114 		/*
1115 		 * This is a public autoconf address that has the same prefix
1116 		 * as ours.  If it is preferred, keep it.  We can't break the
1117 		 * loop here, because there may be a still-preferred temporary
1118 		 * address with the prefix.
1119 		 */
1120 		if (!IFA6_IS_DEPRECATED(it6))
1121 			public_ifa6 = it6;
1122 	}
1123 	if (public_ifa6 != NULL)
1124 		ifa_ref(&public_ifa6->ia_ifa);
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_prhead prl;
1150 	struct nd_prefix *pr, *npr;
1151 
1152 	LIST_INIT(&prl);
1153 
1154 	/* Purge default router list entries toward ifp. */
1155 	nd6_defrouter_purge(ifp);
1156 
1157 	ND6_WLOCK();
1158 	/*
1159 	 * Remove prefixes on ifp. We should have already removed addresses on
1160 	 * this interface, so no addresses should be referencing these prefixes.
1161 	 */
1162 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1163 		if (pr->ndpr_ifp == ifp)
1164 			nd6_prefix_unlink(pr, &prl);
1165 	}
1166 	ND6_WUNLOCK();
1167 
1168 	/* Delete the unlinked prefix objects. */
1169 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1170 		LIST_REMOVE(pr, ndpr_entry);
1171 		nd6_prefix_del(pr);
1172 	}
1173 
1174 	/* cancel default outgoing interface setting */
1175 	if (V_nd6_defifindex == ifp->if_index)
1176 		nd6_setdefaultiface(0);
1177 
1178 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1179 		/* Refresh default router list. */
1180 		defrouter_select_fib(ifp->if_fib);
1181 	}
1182 }
1183 
1184 /*
1185  * the caller acquires and releases the lock on the lltbls
1186  * Returns the llentry locked
1187  */
1188 struct llentry *
1189 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1190 {
1191 	struct sockaddr_in6 sin6;
1192 	struct llentry *ln;
1193 
1194 	bzero(&sin6, sizeof(sin6));
1195 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1196 	sin6.sin6_family = AF_INET6;
1197 	sin6.sin6_addr = *addr6;
1198 
1199 	IF_AFDATA_LOCK_ASSERT(ifp);
1200 
1201 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1202 
1203 	return (ln);
1204 }
1205 
1206 static struct llentry *
1207 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1208 {
1209 	struct sockaddr_in6 sin6;
1210 	struct llentry *ln;
1211 
1212 	bzero(&sin6, sizeof(sin6));
1213 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1214 	sin6.sin6_family = AF_INET6;
1215 	sin6.sin6_addr = *addr6;
1216 
1217 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1218 	if (ln != NULL)
1219 		ln->ln_state = ND6_LLINFO_NOSTATE;
1220 
1221 	return (ln);
1222 }
1223 
1224 /*
1225  * Test whether a given IPv6 address is a neighbor or not, ignoring
1226  * the actual neighbor cache.  The neighbor cache is ignored in order
1227  * to not reenter the routing code from within itself.
1228  */
1229 static int
1230 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1231 {
1232 	struct nd_prefix *pr;
1233 	struct ifaddr *ifa;
1234 	struct rt_addrinfo info;
1235 	struct sockaddr_in6 rt_key;
1236 	const struct sockaddr *dst6;
1237 	uint64_t genid;
1238 	int error, fibnum;
1239 
1240 	/*
1241 	 * A link-local address is always a neighbor.
1242 	 * XXX: a link does not necessarily specify a single interface.
1243 	 */
1244 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1245 		struct sockaddr_in6 sin6_copy;
1246 		u_int32_t zone;
1247 
1248 		/*
1249 		 * We need sin6_copy since sa6_recoverscope() may modify the
1250 		 * content (XXX).
1251 		 */
1252 		sin6_copy = *addr;
1253 		if (sa6_recoverscope(&sin6_copy))
1254 			return (0); /* XXX: should be impossible */
1255 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1256 			return (0);
1257 		if (sin6_copy.sin6_scope_id == zone)
1258 			return (1);
1259 		else
1260 			return (0);
1261 	}
1262 
1263 	bzero(&rt_key, sizeof(rt_key));
1264 	bzero(&info, sizeof(info));
1265 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1266 
1267 	/*
1268 	 * If the address matches one of our addresses,
1269 	 * it should be a neighbor.
1270 	 * If the address matches one of our on-link prefixes, it should be a
1271 	 * neighbor.
1272 	 */
1273 	ND6_RLOCK();
1274 restart:
1275 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1276 		if (pr->ndpr_ifp != ifp)
1277 			continue;
1278 
1279 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1280 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1281 
1282 			/*
1283 			 * We only need to check all FIBs if add_addr_allfibs
1284 			 * is unset. If set, checking any FIB will suffice.
1285 			 */
1286 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1287 			for (; fibnum < rt_numfibs; fibnum++) {
1288 				genid = V_nd6_list_genid;
1289 				ND6_RUNLOCK();
1290 
1291 				/*
1292 				 * Restore length field before
1293 				 * retrying lookup
1294 				 */
1295 				rt_key.sin6_len = sizeof(rt_key);
1296 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1297 						        &info);
1298 
1299 				ND6_RLOCK();
1300 				if (genid != V_nd6_list_genid)
1301 					goto restart;
1302 				if (error == 0)
1303 					break;
1304 			}
1305 			if (error != 0)
1306 				continue;
1307 
1308 			/*
1309 			 * This is the case where multiple interfaces
1310 			 * have the same prefix, but only one is installed
1311 			 * into the routing table and that prefix entry
1312 			 * is not the one being examined here. In the case
1313 			 * where RADIX_MPATH is enabled, multiple route
1314 			 * entries (of the same rt_key value) will be
1315 			 * installed because the interface addresses all
1316 			 * differ.
1317 			 */
1318 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1319 			    &rt_key.sin6_addr))
1320 				continue;
1321 		}
1322 
1323 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1324 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1325 			ND6_RUNLOCK();
1326 			return (1);
1327 		}
1328 	}
1329 	ND6_RUNLOCK();
1330 
1331 	/*
1332 	 * If the address is assigned on the node of the other side of
1333 	 * a p2p interface, the address should be a neighbor.
1334 	 */
1335 	if (ifp->if_flags & IFF_POINTOPOINT) {
1336 		struct epoch_tracker et;
1337 
1338 		NET_EPOCH_ENTER(et);
1339 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1340 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1341 				continue;
1342 			if (ifa->ifa_dstaddr != NULL &&
1343 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1344 				NET_EPOCH_EXIT(et);
1345 				return 1;
1346 			}
1347 		}
1348 		NET_EPOCH_EXIT(et);
1349 	}
1350 
1351 	/*
1352 	 * If the default router list is empty, all addresses are regarded
1353 	 * as on-link, and thus, as a neighbor.
1354 	 */
1355 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1356 	    nd6_defrouter_list_empty() &&
1357 	    V_nd6_defifindex == ifp->if_index) {
1358 		return (1);
1359 	}
1360 
1361 	return (0);
1362 }
1363 
1364 /*
1365  * Detect if a given IPv6 address identifies a neighbor on a given link.
1366  * XXX: should take care of the destination of a p2p link?
1367  */
1368 int
1369 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1370 {
1371 	struct llentry *lle;
1372 	int rc = 0;
1373 
1374 	NET_EPOCH_ASSERT();
1375 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1376 	if (nd6_is_new_addr_neighbor(addr, ifp))
1377 		return (1);
1378 
1379 	/*
1380 	 * Even if the address matches none of our addresses, it might be
1381 	 * in the neighbor cache.
1382 	 */
1383 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1384 		LLE_RUNLOCK(lle);
1385 		rc = 1;
1386 	}
1387 	return (rc);
1388 }
1389 
1390 /*
1391  * Free an nd6 llinfo entry.
1392  * Since the function would cause significant changes in the kernel, DO NOT
1393  * make it global, unless you have a strong reason for the change, and are sure
1394  * that the change is safe.
1395  *
1396  * Set noinline to be dtrace-friendly
1397  */
1398 static __noinline void
1399 nd6_free(struct llentry **lnp, int gc)
1400 {
1401 	struct ifnet *ifp;
1402 	struct llentry *ln;
1403 	struct nd_defrouter *dr;
1404 
1405 	ln = *lnp;
1406 	*lnp = NULL;
1407 
1408 	LLE_WLOCK_ASSERT(ln);
1409 	ND6_RLOCK_ASSERT();
1410 
1411 	ifp = lltable_get_ifp(ln->lle_tbl);
1412 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1413 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1414 	else
1415 		dr = NULL;
1416 	ND6_RUNLOCK();
1417 
1418 	if ((ln->la_flags & LLE_DELETED) == 0)
1419 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1420 
1421 	/*
1422 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1423 	 * even though it is not harmful, it was not really necessary.
1424 	 */
1425 
1426 	/* cancel timer */
1427 	nd6_llinfo_settimer_locked(ln, -1);
1428 
1429 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1430 		if (dr != NULL && dr->expire &&
1431 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1432 			/*
1433 			 * If the reason for the deletion is just garbage
1434 			 * collection, and the neighbor is an active default
1435 			 * router, do not delete it.  Instead, reset the GC
1436 			 * timer using the router's lifetime.
1437 			 * Simply deleting the entry would affect default
1438 			 * router selection, which is not necessarily a good
1439 			 * thing, especially when we're using router preference
1440 			 * values.
1441 			 * XXX: the check for ln_state would be redundant,
1442 			 *      but we intentionally keep it just in case.
1443 			 */
1444 			if (dr->expire > time_uptime)
1445 				nd6_llinfo_settimer_locked(ln,
1446 				    (dr->expire - time_uptime) * hz);
1447 			else
1448 				nd6_llinfo_settimer_locked(ln,
1449 				    (long)V_nd6_gctimer * hz);
1450 
1451 			LLE_REMREF(ln);
1452 			LLE_WUNLOCK(ln);
1453 			defrouter_rele(dr);
1454 			return;
1455 		}
1456 
1457 		if (dr) {
1458 			/*
1459 			 * Unreachablity of a router might affect the default
1460 			 * router selection and on-link detection of advertised
1461 			 * prefixes.
1462 			 */
1463 
1464 			/*
1465 			 * Temporarily fake the state to choose a new default
1466 			 * router and to perform on-link determination of
1467 			 * prefixes correctly.
1468 			 * Below the state will be set correctly,
1469 			 * or the entry itself will be deleted.
1470 			 */
1471 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1472 		}
1473 
1474 		if (ln->ln_router || dr) {
1475 			/*
1476 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1477 			 * rnh and for the calls to pfxlist_onlink_check() and
1478 			 * defrouter_select_fib() in the block further down for calls
1479 			 * into nd6_lookup().  We still hold a ref.
1480 			 */
1481 			LLE_WUNLOCK(ln);
1482 
1483 			/*
1484 			 * rt6_flush must be called whether or not the neighbor
1485 			 * is in the Default Router List.
1486 			 * See a corresponding comment in nd6_na_input().
1487 			 */
1488 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1489 		}
1490 
1491 		if (dr) {
1492 			/*
1493 			 * Since defrouter_select_fib() does not affect the
1494 			 * on-link determination and MIP6 needs the check
1495 			 * before the default router selection, we perform
1496 			 * the check now.
1497 			 */
1498 			pfxlist_onlink_check();
1499 
1500 			/*
1501 			 * Refresh default router list.
1502 			 */
1503 			defrouter_select_fib(dr->ifp->if_fib);
1504 		}
1505 
1506 		/*
1507 		 * If this entry was added by an on-link redirect, remove the
1508 		 * corresponding host route.
1509 		 */
1510 		if (ln->la_flags & LLE_REDIRECT)
1511 			nd6_free_redirect(ln);
1512 
1513 		if (ln->ln_router || dr)
1514 			LLE_WLOCK(ln);
1515 	}
1516 
1517 	/*
1518 	 * Save to unlock. We still hold an extra reference and will not
1519 	 * free(9) in llentry_free() if someone else holds one as well.
1520 	 */
1521 	LLE_WUNLOCK(ln);
1522 	IF_AFDATA_LOCK(ifp);
1523 	LLE_WLOCK(ln);
1524 	/* Guard against race with other llentry_free(). */
1525 	if (ln->la_flags & LLE_LINKED) {
1526 		/* Remove callout reference */
1527 		LLE_REMREF(ln);
1528 		lltable_unlink_entry(ln->lle_tbl, ln);
1529 	}
1530 	IF_AFDATA_UNLOCK(ifp);
1531 
1532 	llentry_free(ln);
1533 	if (dr != NULL)
1534 		defrouter_rele(dr);
1535 }
1536 
1537 static int
1538 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1539 {
1540 
1541 	if (nh->nh_flags & NHF_REDIRECT)
1542 		return (1);
1543 
1544 	return (0);
1545 }
1546 
1547 /*
1548  * Remove the rtentry for the given llentry,
1549  * both of which were installed by a redirect.
1550  */
1551 static void
1552 nd6_free_redirect(const struct llentry *ln)
1553 {
1554 	int fibnum;
1555 	struct sockaddr_in6 sin6;
1556 	struct rt_addrinfo info;
1557 	struct rib_cmd_info rc;
1558 	struct epoch_tracker et;
1559 
1560 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1561 	memset(&info, 0, sizeof(info));
1562 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1563 	info.rti_filter = nd6_isdynrte;
1564 
1565 	NET_EPOCH_ENTER(et);
1566 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1567 		rib_action(fibnum, RTM_DELETE, &info, &rc);
1568 	NET_EPOCH_EXIT(et);
1569 }
1570 
1571 /*
1572  * Updates status of the default router route.
1573  */
1574 static void
1575 check_release_defrouter(struct rib_cmd_info *rc, void *_cbdata)
1576 {
1577 	struct nd_defrouter *dr;
1578 	struct nhop_object *nh;
1579 
1580 	nh = rc->rc_nh_old;
1581 
1582 	if ((nh != NULL) && (nh->nh_flags & NHF_DEFAULT)) {
1583 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1584 		if (dr != NULL) {
1585 			dr->installed = 0;
1586 			defrouter_rele(dr);
1587 		}
1588 	}
1589 }
1590 
1591 void
1592 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1593 {
1594 
1595 #ifdef ROUTE_MPATH
1596 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1597 #else
1598 	check_release_defrouter(rc, NULL);
1599 #endif
1600 }
1601 
1602 int
1603 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1604 {
1605 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1606 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1607 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1608 	struct epoch_tracker et;
1609 	int error = 0;
1610 
1611 	if (ifp->if_afdata[AF_INET6] == NULL)
1612 		return (EPFNOSUPPORT);
1613 	switch (cmd) {
1614 	case OSIOCGIFINFO_IN6:
1615 #define ND	ndi->ndi
1616 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1617 		bzero(&ND, sizeof(ND));
1618 		ND.linkmtu = IN6_LINKMTU(ifp);
1619 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1620 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1621 		ND.reachable = ND_IFINFO(ifp)->reachable;
1622 		ND.retrans = ND_IFINFO(ifp)->retrans;
1623 		ND.flags = ND_IFINFO(ifp)->flags;
1624 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1625 		ND.chlim = ND_IFINFO(ifp)->chlim;
1626 		break;
1627 	case SIOCGIFINFO_IN6:
1628 		ND = *ND_IFINFO(ifp);
1629 		break;
1630 	case SIOCSIFINFO_IN6:
1631 		/*
1632 		 * used to change host variables from userland.
1633 		 * intended for a use on router to reflect RA configurations.
1634 		 */
1635 		/* 0 means 'unspecified' */
1636 		if (ND.linkmtu != 0) {
1637 			if (ND.linkmtu < IPV6_MMTU ||
1638 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1639 				error = EINVAL;
1640 				break;
1641 			}
1642 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1643 		}
1644 
1645 		if (ND.basereachable != 0) {
1646 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1647 
1648 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1649 			if (ND.basereachable != obasereachable)
1650 				ND_IFINFO(ifp)->reachable =
1651 				    ND_COMPUTE_RTIME(ND.basereachable);
1652 		}
1653 		if (ND.retrans != 0)
1654 			ND_IFINFO(ifp)->retrans = ND.retrans;
1655 		if (ND.chlim != 0)
1656 			ND_IFINFO(ifp)->chlim = ND.chlim;
1657 		/* FALLTHROUGH */
1658 	case SIOCSIFINFO_FLAGS:
1659 	{
1660 		struct ifaddr *ifa;
1661 		struct in6_ifaddr *ia;
1662 
1663 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1664 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1665 			/* ifdisabled 1->0 transision */
1666 
1667 			/*
1668 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1669 			 * has an link-local address with IN6_IFF_DUPLICATED,
1670 			 * do not clear ND6_IFF_IFDISABLED.
1671 			 * See RFC 4862, Section 5.4.5.
1672 			 */
1673 			NET_EPOCH_ENTER(et);
1674 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1675 				if (ifa->ifa_addr->sa_family != AF_INET6)
1676 					continue;
1677 				ia = (struct in6_ifaddr *)ifa;
1678 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1679 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1680 					break;
1681 			}
1682 			NET_EPOCH_EXIT(et);
1683 
1684 			if (ifa != NULL) {
1685 				/* LLA is duplicated. */
1686 				ND.flags |= ND6_IFF_IFDISABLED;
1687 				log(LOG_ERR, "Cannot enable an interface"
1688 				    " with a link-local address marked"
1689 				    " duplicate.\n");
1690 			} else {
1691 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1692 				if (ifp->if_flags & IFF_UP)
1693 					in6_if_up(ifp);
1694 			}
1695 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1696 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1697 			/* ifdisabled 0->1 transision */
1698 			/* Mark all IPv6 address as tentative. */
1699 
1700 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1701 			if (V_ip6_dad_count > 0 &&
1702 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1703 				NET_EPOCH_ENTER(et);
1704 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1705 				    ifa_link) {
1706 					if (ifa->ifa_addr->sa_family !=
1707 					    AF_INET6)
1708 						continue;
1709 					ia = (struct in6_ifaddr *)ifa;
1710 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1711 				}
1712 				NET_EPOCH_EXIT(et);
1713 			}
1714 		}
1715 
1716 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1717 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1718 				/* auto_linklocal 0->1 transision */
1719 
1720 				/* If no link-local address on ifp, configure */
1721 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1722 				in6_ifattach(ifp, NULL);
1723 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1724 			    ifp->if_flags & IFF_UP) {
1725 				/*
1726 				 * When the IF already has
1727 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1728 				 * address is assigned, and IFF_UP, try to
1729 				 * assign one.
1730 				 */
1731 				NET_EPOCH_ENTER(et);
1732 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1733 				    ifa_link) {
1734 					if (ifa->ifa_addr->sa_family !=
1735 					    AF_INET6)
1736 						continue;
1737 					ia = (struct in6_ifaddr *)ifa;
1738 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1739 						break;
1740 				}
1741 				NET_EPOCH_EXIT(et);
1742 				if (ifa != NULL)
1743 					/* No LLA is configured. */
1744 					in6_ifattach(ifp, NULL);
1745 			}
1746 		}
1747 		ND_IFINFO(ifp)->flags = ND.flags;
1748 		break;
1749 	}
1750 #undef ND
1751 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1752 		/* sync kernel routing table with the default router list */
1753 		defrouter_reset();
1754 		defrouter_select_fib(RT_ALL_FIBS);
1755 		break;
1756 	case SIOCSPFXFLUSH_IN6:
1757 	{
1758 		/* flush all the prefix advertised by routers */
1759 		struct in6_ifaddr *ia, *ia_next;
1760 		struct nd_prefix *pr, *next;
1761 		struct nd_prhead prl;
1762 
1763 		LIST_INIT(&prl);
1764 
1765 		ND6_WLOCK();
1766 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1767 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1768 				continue; /* XXX */
1769 			nd6_prefix_unlink(pr, &prl);
1770 		}
1771 		ND6_WUNLOCK();
1772 
1773 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1774 			LIST_REMOVE(pr, ndpr_entry);
1775 			/* XXXRW: in6_ifaddrhead locking. */
1776 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1777 			    ia_next) {
1778 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1779 					continue;
1780 
1781 				if (ia->ia6_ndpr == pr)
1782 					in6_purgeaddr(&ia->ia_ifa);
1783 			}
1784 			nd6_prefix_del(pr);
1785 		}
1786 		break;
1787 	}
1788 	case SIOCSRTRFLUSH_IN6:
1789 	{
1790 		/* flush all the default routers */
1791 
1792 		defrouter_reset();
1793 		nd6_defrouter_flush_all();
1794 		defrouter_select_fib(RT_ALL_FIBS);
1795 		break;
1796 	}
1797 	case SIOCGNBRINFO_IN6:
1798 	{
1799 		struct llentry *ln;
1800 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1801 
1802 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1803 			return (error);
1804 
1805 		NET_EPOCH_ENTER(et);
1806 		ln = nd6_lookup(&nb_addr, 0, ifp);
1807 		NET_EPOCH_EXIT(et);
1808 
1809 		if (ln == NULL) {
1810 			error = EINVAL;
1811 			break;
1812 		}
1813 		nbi->state = ln->ln_state;
1814 		nbi->asked = ln->la_asked;
1815 		nbi->isrouter = ln->ln_router;
1816 		if (ln->la_expire == 0)
1817 			nbi->expire = 0;
1818 		else
1819 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1820 			    (time_second - time_uptime);
1821 		LLE_RUNLOCK(ln);
1822 		break;
1823 	}
1824 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1825 		ndif->ifindex = V_nd6_defifindex;
1826 		break;
1827 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1828 		return (nd6_setdefaultiface(ndif->ifindex));
1829 	}
1830 	return (error);
1831 }
1832 
1833 /*
1834  * Calculates new isRouter value based on provided parameters and
1835  * returns it.
1836  */
1837 static int
1838 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1839     int ln_router)
1840 {
1841 
1842 	/*
1843 	 * ICMP6 type dependent behavior.
1844 	 *
1845 	 * NS: clear IsRouter if new entry
1846 	 * RS: clear IsRouter
1847 	 * RA: set IsRouter if there's lladdr
1848 	 * redir: clear IsRouter if new entry
1849 	 *
1850 	 * RA case, (1):
1851 	 * The spec says that we must set IsRouter in the following cases:
1852 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1853 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1854 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1855 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1856 	 * neighbor cache, this is similar to (6).
1857 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1858 	 *
1859 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1860 	 *							D R
1861 	 *	0	n	n	(1)	c   ?     s
1862 	 *	0	y	n	(2)	c   s     s
1863 	 *	0	n	y	(3)	c   s     s
1864 	 *	0	y	y	(4)	c   s     s
1865 	 *	0	y	y	(5)	c   s     s
1866 	 *	1	--	n	(6) c	c	c s
1867 	 *	1	--	y	(7) c	c   s	c s
1868 	 *
1869 	 *					(c=clear s=set)
1870 	 */
1871 	switch (type & 0xff) {
1872 	case ND_NEIGHBOR_SOLICIT:
1873 		/*
1874 		 * New entry must have is_router flag cleared.
1875 		 */
1876 		if (is_new)					/* (6-7) */
1877 			ln_router = 0;
1878 		break;
1879 	case ND_REDIRECT:
1880 		/*
1881 		 * If the icmp is a redirect to a better router, always set the
1882 		 * is_router flag.  Otherwise, if the entry is newly created,
1883 		 * clear the flag.  [RFC 2461, sec 8.3]
1884 		 */
1885 		if (code == ND_REDIRECT_ROUTER)
1886 			ln_router = 1;
1887 		else {
1888 			if (is_new)				/* (6-7) */
1889 				ln_router = 0;
1890 		}
1891 		break;
1892 	case ND_ROUTER_SOLICIT:
1893 		/*
1894 		 * is_router flag must always be cleared.
1895 		 */
1896 		ln_router = 0;
1897 		break;
1898 	case ND_ROUTER_ADVERT:
1899 		/*
1900 		 * Mark an entry with lladdr as a router.
1901 		 */
1902 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1903 		    (is_new && new_addr)) {			/* (7) */
1904 			ln_router = 1;
1905 		}
1906 		break;
1907 	}
1908 
1909 	return (ln_router);
1910 }
1911 
1912 /*
1913  * Create neighbor cache entry and cache link-layer address,
1914  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1915  *
1916  * type - ICMP6 type
1917  * code - type dependent information
1918  *
1919  */
1920 void
1921 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1922     int lladdrlen, int type, int code)
1923 {
1924 	struct llentry *ln = NULL, *ln_tmp;
1925 	int is_newentry;
1926 	int do_update;
1927 	int olladdr;
1928 	int llchange;
1929 	int flags;
1930 	uint16_t router = 0;
1931 	struct sockaddr_in6 sin6;
1932 	struct mbuf *chain = NULL;
1933 	u_char linkhdr[LLE_MAX_LINKHDR];
1934 	size_t linkhdrsize;
1935 	int lladdr_off;
1936 
1937 	NET_EPOCH_ASSERT();
1938 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1939 
1940 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1941 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1942 
1943 	/* nothing must be updated for unspecified address */
1944 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1945 		return;
1946 
1947 	/*
1948 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1949 	 * the caller.
1950 	 *
1951 	 * XXX If the link does not have link-layer adderss, what should
1952 	 * we do? (ifp->if_addrlen == 0)
1953 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1954 	 * description on it in NS section (RFC 2461 7.2.3).
1955 	 */
1956 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1957 	ln = nd6_lookup(from, flags, ifp);
1958 	is_newentry = 0;
1959 	if (ln == NULL) {
1960 		flags |= LLE_EXCLUSIVE;
1961 		ln = nd6_alloc(from, 0, ifp);
1962 		if (ln == NULL)
1963 			return;
1964 
1965 		/*
1966 		 * Since we already know all the data for the new entry,
1967 		 * fill it before insertion.
1968 		 */
1969 		if (lladdr != NULL) {
1970 			linkhdrsize = sizeof(linkhdr);
1971 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1972 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
1973 				return;
1974 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1975 			    lladdr_off);
1976 		}
1977 
1978 		IF_AFDATA_WLOCK(ifp);
1979 		LLE_WLOCK(ln);
1980 		/* Prefer any existing lle over newly-created one */
1981 		ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp);
1982 		if (ln_tmp == NULL)
1983 			lltable_link_entry(LLTABLE6(ifp), ln);
1984 		IF_AFDATA_WUNLOCK(ifp);
1985 		if (ln_tmp == NULL) {
1986 			/* No existing lle, mark as new entry (6,7) */
1987 			is_newentry = 1;
1988 			if (lladdr != NULL) {	/* (7) */
1989 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
1990 				EVENTHANDLER_INVOKE(lle_event, ln,
1991 				    LLENTRY_RESOLVED);
1992 			}
1993 		} else {
1994 			lltable_free_entry(LLTABLE6(ifp), ln);
1995 			ln = ln_tmp;
1996 			ln_tmp = NULL;
1997 		}
1998 	}
1999 	/* do nothing if static ndp is set */
2000 	if ((ln->la_flags & LLE_STATIC)) {
2001 		if (flags & LLE_EXCLUSIVE)
2002 			LLE_WUNLOCK(ln);
2003 		else
2004 			LLE_RUNLOCK(ln);
2005 		return;
2006 	}
2007 
2008 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2009 	if (olladdr && lladdr) {
2010 		llchange = bcmp(lladdr, ln->ll_addr,
2011 		    ifp->if_addrlen);
2012 	} else if (!olladdr && lladdr)
2013 		llchange = 1;
2014 	else
2015 		llchange = 0;
2016 
2017 	/*
2018 	 * newentry olladdr  lladdr  llchange	(*=record)
2019 	 *	0	n	n	--	(1)
2020 	 *	0	y	n	--	(2)
2021 	 *	0	n	y	y	(3) * STALE
2022 	 *	0	y	y	n	(4) *
2023 	 *	0	y	y	y	(5) * STALE
2024 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2025 	 *	1	--	y	--	(7) * STALE
2026 	 */
2027 
2028 	do_update = 0;
2029 	if (is_newentry == 0 && llchange != 0) {
2030 		do_update = 1;	/* (3,5) */
2031 
2032 		/*
2033 		 * Record source link-layer address
2034 		 * XXX is it dependent to ifp->if_type?
2035 		 */
2036 		linkhdrsize = sizeof(linkhdr);
2037 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2038 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2039 			return;
2040 
2041 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2042 		    lladdr_off) == 0) {
2043 			/* Entry was deleted */
2044 			return;
2045 		}
2046 
2047 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2048 
2049 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2050 
2051 		if (ln->la_hold != NULL)
2052 			nd6_grab_holdchain(ln, &chain, &sin6);
2053 	}
2054 
2055 	/* Calculates new router status */
2056 	router = nd6_is_router(type, code, is_newentry, olladdr,
2057 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2058 
2059 	ln->ln_router = router;
2060 	/* Mark non-router redirects with special flag */
2061 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2062 		ln->la_flags |= LLE_REDIRECT;
2063 
2064 	if (flags & LLE_EXCLUSIVE)
2065 		LLE_WUNLOCK(ln);
2066 	else
2067 		LLE_RUNLOCK(ln);
2068 
2069 	if (chain != NULL)
2070 		nd6_flush_holdchain(ifp, chain, &sin6);
2071 
2072 	/*
2073 	 * When the link-layer address of a router changes, select the
2074 	 * best router again.  In particular, when the neighbor entry is newly
2075 	 * created, it might affect the selection policy.
2076 	 * Question: can we restrict the first condition to the "is_newentry"
2077 	 * case?
2078 	 * XXX: when we hear an RA from a new router with the link-layer
2079 	 * address option, defrouter_select_fib() is called twice, since
2080 	 * defrtrlist_update called the function as well.  However, I believe
2081 	 * we can compromise the overhead, since it only happens the first
2082 	 * time.
2083 	 * XXX: although defrouter_select_fib() should not have a bad effect
2084 	 * for those are not autoconfigured hosts, we explicitly avoid such
2085 	 * cases for safety.
2086 	 */
2087 	if ((do_update || is_newentry) && router &&
2088 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2089 		/*
2090 		 * guaranteed recursion
2091 		 */
2092 		defrouter_select_fib(ifp->if_fib);
2093 	}
2094 }
2095 
2096 static void
2097 nd6_slowtimo(void *arg)
2098 {
2099 	struct epoch_tracker et;
2100 	CURVNET_SET((struct vnet *) arg);
2101 	struct nd_ifinfo *nd6if;
2102 	struct ifnet *ifp;
2103 
2104 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2105 	    nd6_slowtimo, curvnet);
2106 	NET_EPOCH_ENTER(et);
2107 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2108 		if (ifp->if_afdata[AF_INET6] == NULL)
2109 			continue;
2110 		nd6if = ND_IFINFO(ifp);
2111 		if (nd6if->basereachable && /* already initialized */
2112 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2113 			/*
2114 			 * Since reachable time rarely changes by router
2115 			 * advertisements, we SHOULD insure that a new random
2116 			 * value gets recomputed at least once every few hours.
2117 			 * (RFC 2461, 6.3.4)
2118 			 */
2119 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2120 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2121 		}
2122 	}
2123 	NET_EPOCH_EXIT(et);
2124 	CURVNET_RESTORE();
2125 }
2126 
2127 void
2128 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2129     struct sockaddr_in6 *sin6)
2130 {
2131 
2132 	LLE_WLOCK_ASSERT(ln);
2133 
2134 	*chain = ln->la_hold;
2135 	ln->la_hold = NULL;
2136 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2137 
2138 	if (ln->ln_state == ND6_LLINFO_STALE) {
2139 		/*
2140 		 * The first time we send a packet to a
2141 		 * neighbor whose entry is STALE, we have
2142 		 * to change the state to DELAY and a sets
2143 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2144 		 * seconds to ensure do neighbor unreachability
2145 		 * detection on expiration.
2146 		 * (RFC 2461 7.3.3)
2147 		 */
2148 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2149 	}
2150 }
2151 
2152 int
2153 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2154     struct sockaddr_in6 *dst, struct route *ro)
2155 {
2156 	int error;
2157 	int ip6len;
2158 	struct ip6_hdr *ip6;
2159 	struct m_tag *mtag;
2160 
2161 #ifdef MAC
2162 	mac_netinet6_nd6_send(ifp, m);
2163 #endif
2164 
2165 	/*
2166 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2167 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2168 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2169 	 * to be diverted to user space.  When re-injected into the kernel,
2170 	 * send_output() will directly dispatch them to the outgoing interface.
2171 	 */
2172 	if (send_sendso_input_hook != NULL) {
2173 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2174 		if (mtag != NULL) {
2175 			ip6 = mtod(m, struct ip6_hdr *);
2176 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2177 			/* Use the SEND socket */
2178 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2179 			    ip6len);
2180 			/* -1 == no app on SEND socket */
2181 			if (error == 0 || error != -1)
2182 			    return (error);
2183 		}
2184 	}
2185 
2186 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2187 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2188 	    mtod(m, struct ip6_hdr *));
2189 
2190 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2191 		origifp = ifp;
2192 
2193 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2194 	return (error);
2195 }
2196 
2197 /*
2198  * Lookup link headerfor @sa_dst address. Stores found
2199  * data in @desten buffer. Copy of lle ln_flags can be also
2200  * saved in @pflags if @pflags is non-NULL.
2201  *
2202  * If destination LLE does not exists or lle state modification
2203  * is required, call "slow" version.
2204  *
2205  * Return values:
2206  * - 0 on success (address copied to buffer).
2207  * - EWOULDBLOCK (no local error, but address is still unresolved)
2208  * - other errors (alloc failure, etc)
2209  */
2210 int
2211 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2212     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2213     struct llentry **plle)
2214 {
2215 	struct llentry *ln = NULL;
2216 	const struct sockaddr_in6 *dst6;
2217 
2218 	NET_EPOCH_ASSERT();
2219 
2220 	if (pflags != NULL)
2221 		*pflags = 0;
2222 
2223 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2224 
2225 	/* discard the packet if IPv6 operation is disabled on the interface */
2226 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2227 		m_freem(m);
2228 		return (ENETDOWN); /* better error? */
2229 	}
2230 
2231 	if (m != NULL && m->m_flags & M_MCAST) {
2232 		switch (ifp->if_type) {
2233 		case IFT_ETHER:
2234 		case IFT_L2VLAN:
2235 		case IFT_BRIDGE:
2236 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2237 						 desten);
2238 			return (0);
2239 		default:
2240 			m_freem(m);
2241 			return (EAFNOSUPPORT);
2242 		}
2243 	}
2244 
2245 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2246 	    ifp);
2247 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2248 		/* Entry found, let's copy lle info */
2249 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2250 		if (pflags != NULL)
2251 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2252 		/* Check if we have feedback request from nd6 timer */
2253 		if (ln->r_skip_req != 0) {
2254 			LLE_REQ_LOCK(ln);
2255 			ln->r_skip_req = 0; /* Notify that entry was used */
2256 			ln->lle_hittime = time_uptime;
2257 			LLE_REQ_UNLOCK(ln);
2258 		}
2259 		if (plle) {
2260 			LLE_ADDREF(ln);
2261 			*plle = ln;
2262 			LLE_WUNLOCK(ln);
2263 		}
2264 		return (0);
2265 	} else if (plle && ln)
2266 		LLE_WUNLOCK(ln);
2267 
2268 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2269 }
2270 
2271 /*
2272  * Do L2 address resolution for @sa_dst address. Stores found
2273  * address in @desten buffer. Copy of lle ln_flags can be also
2274  * saved in @pflags if @pflags is non-NULL.
2275  *
2276  * Heavy version.
2277  * Function assume that destination LLE does not exist,
2278  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2279  *
2280  * Set noinline to be dtrace-friendly
2281  */
2282 static __noinline int
2283 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2284     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2285     struct llentry **plle)
2286 {
2287 	struct llentry *lle = NULL, *lle_tmp;
2288 	struct in6_addr *psrc, src;
2289 	int send_ns, ll_len;
2290 	char *lladdr;
2291 
2292 	NET_EPOCH_ASSERT();
2293 
2294 	/*
2295 	 * Address resolution or Neighbor Unreachability Detection
2296 	 * for the next hop.
2297 	 * At this point, the destination of the packet must be a unicast
2298 	 * or an anycast address(i.e. not a multicast).
2299 	 */
2300 	if (lle == NULL) {
2301 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2302 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2303 			/*
2304 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2305 			 * the condition below is not very efficient.  But we believe
2306 			 * it is tolerable, because this should be a rare case.
2307 			 */
2308 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2309 			if (lle == NULL) {
2310 				char ip6buf[INET6_ADDRSTRLEN];
2311 				log(LOG_DEBUG,
2312 				    "nd6_output: can't allocate llinfo for %s "
2313 				    "(ln=%p)\n",
2314 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2315 				m_freem(m);
2316 				return (ENOBUFS);
2317 			}
2318 
2319 			IF_AFDATA_WLOCK(ifp);
2320 			LLE_WLOCK(lle);
2321 			/* Prefer any existing entry over newly-created one */
2322 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2323 			if (lle_tmp == NULL)
2324 				lltable_link_entry(LLTABLE6(ifp), lle);
2325 			IF_AFDATA_WUNLOCK(ifp);
2326 			if (lle_tmp != NULL) {
2327 				lltable_free_entry(LLTABLE6(ifp), lle);
2328 				lle = lle_tmp;
2329 				lle_tmp = NULL;
2330 			}
2331 		}
2332 	}
2333 	if (lle == NULL) {
2334 		m_freem(m);
2335 		return (ENOBUFS);
2336 	}
2337 
2338 	LLE_WLOCK_ASSERT(lle);
2339 
2340 	/*
2341 	 * The first time we send a packet to a neighbor whose entry is
2342 	 * STALE, we have to change the state to DELAY and a sets a timer to
2343 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2344 	 * neighbor unreachability detection on expiration.
2345 	 * (RFC 2461 7.3.3)
2346 	 */
2347 	if (lle->ln_state == ND6_LLINFO_STALE)
2348 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2349 
2350 	/*
2351 	 * If the neighbor cache entry has a state other than INCOMPLETE
2352 	 * (i.e. its link-layer address is already resolved), just
2353 	 * send the packet.
2354 	 */
2355 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2356 		if (flags & LLE_ADDRONLY) {
2357 			lladdr = lle->ll_addr;
2358 			ll_len = ifp->if_addrlen;
2359 		} else {
2360 			lladdr = lle->r_linkdata;
2361 			ll_len = lle->r_hdrlen;
2362 		}
2363 		bcopy(lladdr, desten, ll_len);
2364 		if (pflags != NULL)
2365 			*pflags = lle->la_flags;
2366 		if (plle) {
2367 			LLE_ADDREF(lle);
2368 			*plle = lle;
2369 		}
2370 		LLE_WUNLOCK(lle);
2371 		return (0);
2372 	}
2373 
2374 	/*
2375 	 * There is a neighbor cache entry, but no ethernet address
2376 	 * response yet.  Append this latest packet to the end of the
2377 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2378 	 * the oldest packet in the queue will be removed.
2379 	 */
2380 
2381 	if (lle->la_hold != NULL) {
2382 		struct mbuf *m_hold;
2383 		int i;
2384 
2385 		i = 0;
2386 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2387 			i++;
2388 			if (m_hold->m_nextpkt == NULL) {
2389 				m_hold->m_nextpkt = m;
2390 				break;
2391 			}
2392 		}
2393 		while (i >= V_nd6_maxqueuelen) {
2394 			m_hold = lle->la_hold;
2395 			lle->la_hold = lle->la_hold->m_nextpkt;
2396 			m_freem(m_hold);
2397 			i--;
2398 		}
2399 	} else {
2400 		lle->la_hold = m;
2401 	}
2402 
2403 	/*
2404 	 * If there has been no NS for the neighbor after entering the
2405 	 * INCOMPLETE state, send the first solicitation.
2406 	 * Note that for newly-created lle la_asked will be 0,
2407 	 * so we will transition from ND6_LLINFO_NOSTATE to
2408 	 * ND6_LLINFO_INCOMPLETE state here.
2409 	 */
2410 	psrc = NULL;
2411 	send_ns = 0;
2412 	if (lle->la_asked == 0) {
2413 		lle->la_asked++;
2414 		send_ns = 1;
2415 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2416 
2417 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2418 	}
2419 	LLE_WUNLOCK(lle);
2420 	if (send_ns != 0)
2421 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2422 
2423 	return (EWOULDBLOCK);
2424 }
2425 
2426 /*
2427  * Do L2 address resolution for @sa_dst address. Stores found
2428  * address in @desten buffer. Copy of lle ln_flags can be also
2429  * saved in @pflags if @pflags is non-NULL.
2430  *
2431  * Return values:
2432  * - 0 on success (address copied to buffer).
2433  * - EWOULDBLOCK (no local error, but address is still unresolved)
2434  * - other errors (alloc failure, etc)
2435  */
2436 int
2437 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2438     char *desten, uint32_t *pflags)
2439 {
2440 	int error;
2441 
2442 	flags |= LLE_ADDRONLY;
2443 	error = nd6_resolve_slow(ifp, flags, NULL,
2444 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2445 	return (error);
2446 }
2447 
2448 int
2449 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2450     struct sockaddr_in6 *dst)
2451 {
2452 	struct mbuf *m, *m_head;
2453 	int error = 0;
2454 
2455 	m_head = chain;
2456 
2457 	while (m_head) {
2458 		m = m_head;
2459 		m_head = m_head->m_nextpkt;
2460 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2461 	}
2462 
2463 	/*
2464 	 * XXX
2465 	 * note that intermediate errors are blindly ignored
2466 	 */
2467 	return (error);
2468 }
2469 
2470 static int
2471 nd6_need_cache(struct ifnet *ifp)
2472 {
2473 	/*
2474 	 * XXX: we currently do not make neighbor cache on any interface
2475 	 * other than Ethernet and GIF.
2476 	 *
2477 	 * RFC2893 says:
2478 	 * - unidirectional tunnels needs no ND
2479 	 */
2480 	switch (ifp->if_type) {
2481 	case IFT_ETHER:
2482 	case IFT_IEEE1394:
2483 	case IFT_L2VLAN:
2484 	case IFT_INFINIBAND:
2485 	case IFT_BRIDGE:
2486 	case IFT_PROPVIRTUAL:
2487 		return (1);
2488 	default:
2489 		return (0);
2490 	}
2491 }
2492 
2493 /*
2494  * Add pernament ND6 link-layer record for given
2495  * interface address.
2496  *
2497  * Very similar to IPv4 arp_ifinit(), but:
2498  * 1) IPv6 DAD is performed in different place
2499  * 2) It is called by IPv6 protocol stack in contrast to
2500  * arp_ifinit() which is typically called in SIOCSIFADDR
2501  * driver ioctl handler.
2502  *
2503  */
2504 int
2505 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2506 {
2507 	struct ifnet *ifp;
2508 	struct llentry *ln, *ln_tmp;
2509 	struct sockaddr *dst;
2510 
2511 	ifp = ia->ia_ifa.ifa_ifp;
2512 	if (nd6_need_cache(ifp) == 0)
2513 		return (0);
2514 
2515 	dst = (struct sockaddr *)&ia->ia_addr;
2516 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2517 	if (ln == NULL)
2518 		return (ENOBUFS);
2519 
2520 	IF_AFDATA_WLOCK(ifp);
2521 	LLE_WLOCK(ln);
2522 	/* Unlink any entry if exists */
2523 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2524 	if (ln_tmp != NULL)
2525 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2526 	lltable_link_entry(LLTABLE6(ifp), ln);
2527 	IF_AFDATA_WUNLOCK(ifp);
2528 
2529 	if (ln_tmp != NULL)
2530 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2531 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2532 
2533 	LLE_WUNLOCK(ln);
2534 	if (ln_tmp != NULL)
2535 		llentry_free(ln_tmp);
2536 
2537 	return (0);
2538 }
2539 
2540 /*
2541  * Removes either all lle entries for given @ia, or lle
2542  * corresponding to @ia address.
2543  */
2544 void
2545 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2546 {
2547 	struct sockaddr_in6 mask, addr;
2548 	struct sockaddr *saddr, *smask;
2549 	struct ifnet *ifp;
2550 
2551 	ifp = ia->ia_ifa.ifa_ifp;
2552 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2553 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2554 	saddr = (struct sockaddr *)&addr;
2555 	smask = (struct sockaddr *)&mask;
2556 
2557 	if (all != 0)
2558 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2559 	else
2560 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2561 }
2562 
2563 static void
2564 clear_llinfo_pqueue(struct llentry *ln)
2565 {
2566 	struct mbuf *m_hold, *m_hold_next;
2567 
2568 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2569 		m_hold_next = m_hold->m_nextpkt;
2570 		m_freem(m_hold);
2571 	}
2572 
2573 	ln->la_hold = NULL;
2574 }
2575 
2576 static int
2577 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2578 {
2579 	struct in6_prefix p;
2580 	struct sockaddr_in6 s6;
2581 	struct nd_prefix *pr;
2582 	struct nd_pfxrouter *pfr;
2583 	time_t maxexpire;
2584 	int error;
2585 	char ip6buf[INET6_ADDRSTRLEN];
2586 
2587 	if (req->newptr)
2588 		return (EPERM);
2589 
2590 	error = sysctl_wire_old_buffer(req, 0);
2591 	if (error != 0)
2592 		return (error);
2593 
2594 	bzero(&p, sizeof(p));
2595 	p.origin = PR_ORIG_RA;
2596 	bzero(&s6, sizeof(s6));
2597 	s6.sin6_family = AF_INET6;
2598 	s6.sin6_len = sizeof(s6);
2599 
2600 	ND6_RLOCK();
2601 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2602 		p.prefix = pr->ndpr_prefix;
2603 		if (sa6_recoverscope(&p.prefix)) {
2604 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2605 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2606 			/* XXX: press on... */
2607 		}
2608 		p.raflags = pr->ndpr_raf;
2609 		p.prefixlen = pr->ndpr_plen;
2610 		p.vltime = pr->ndpr_vltime;
2611 		p.pltime = pr->ndpr_pltime;
2612 		p.if_index = pr->ndpr_ifp->if_index;
2613 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2614 			p.expire = 0;
2615 		else {
2616 			/* XXX: we assume time_t is signed. */
2617 			maxexpire = (-1) &
2618 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2619 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2620 				p.expire = pr->ndpr_lastupdate +
2621 				    pr->ndpr_vltime +
2622 				    (time_second - time_uptime);
2623 			else
2624 				p.expire = maxexpire;
2625 		}
2626 		p.refcnt = pr->ndpr_addrcnt;
2627 		p.flags = pr->ndpr_stateflags;
2628 		p.advrtrs = 0;
2629 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2630 			p.advrtrs++;
2631 		error = SYSCTL_OUT(req, &p, sizeof(p));
2632 		if (error != 0)
2633 			break;
2634 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2635 			s6.sin6_addr = pfr->router->rtaddr;
2636 			if (sa6_recoverscope(&s6))
2637 				log(LOG_ERR,
2638 				    "scope error in prefix list (%s)\n",
2639 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2640 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2641 			if (error != 0)
2642 				goto out;
2643 		}
2644 	}
2645 out:
2646 	ND6_RUNLOCK();
2647 	return (error);
2648 }
2649 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2650 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2651 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2652 	"NDP prefix list");
2653 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2654 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2655 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2656 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2657