xref: /freebsd/sys/netinet6/nd6.c (revision af23369a6deaaeb612ab266eb88b8bb8d560c322)
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_private.h>
65 #include <net/if_types.h>
66 #include <net/route.h>
67 #include <net/route/route_ctl.h>
68 #include <net/route/nhop.h>
69 #include <net/vnet.h>
70 
71 #include <netinet/in.h>
72 #include <netinet/in_kdtrace.h>
73 #include <net/if_llatbl.h>
74 #include <netinet/if_ether.h>
75 #include <netinet6/in6_fib.h>
76 #include <netinet6/in6_var.h>
77 #include <netinet/ip6.h>
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #include <netinet6/in6_ifattach.h>
82 #include <netinet/icmp6.h>
83 #include <netinet6/send.h>
84 
85 #include <sys/limits.h>
86 
87 #include <security/mac/mac_framework.h>
88 
89 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
90 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
91 
92 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
93 
94 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
95 
96 /* timer values */
97 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
98 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
99 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
100 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
101 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
102 					 * local traffic */
103 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
104 					 * collection timer */
105 
106 /* preventing too many loops in ND option parsing */
107 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
108 
109 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
110 					 * layer hints */
111 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 16; /* max pkts cached in unresolved
112 					 * ND entries */
113 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
114 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
115 
116 #ifdef ND6_DEBUG
117 VNET_DEFINE(int, nd6_debug) = 1;
118 #else
119 VNET_DEFINE(int, nd6_debug) = 0;
120 #endif
121 
122 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
123 
124 VNET_DEFINE(struct nd_prhead, nd_prefix);
125 VNET_DEFINE(struct rwlock, nd6_lock);
126 VNET_DEFINE(uint64_t, nd6_list_genid);
127 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
128 
129 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
130 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
131 
132 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
133 
134 static bool nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
135 	struct ifnet *);
136 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
137 static void nd6_slowtimo(void *);
138 static int regen_tmpaddr(struct in6_ifaddr *);
139 static void nd6_free(struct llentry **, int);
140 static void nd6_free_redirect(const struct llentry *);
141 static void nd6_llinfo_timer(void *);
142 static void nd6_llinfo_settimer_locked(struct llentry *, long);
143 static int nd6_resolve_slow(struct ifnet *, int, int, struct mbuf *,
144     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
145 static int nd6_need_cache(struct ifnet *);
146 
147 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
148 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
149 
150 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
151 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
152 
153 SYSCTL_DECL(_net_inet6_icmp6);
154 
155 static void
156 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
157 {
158 	struct rt_addrinfo rtinfo;
159 	struct sockaddr_in6 dst;
160 	struct sockaddr_dl gw;
161 	struct ifnet *ifp;
162 	int type;
163 	int fibnum;
164 
165 	LLE_WLOCK_ASSERT(lle);
166 
167 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
168 		return;
169 
170 	switch (evt) {
171 	case LLENTRY_RESOLVED:
172 		type = RTM_ADD;
173 		KASSERT(lle->la_flags & LLE_VALID,
174 		    ("%s: %p resolved but not valid?", __func__, lle));
175 		break;
176 	case LLENTRY_EXPIRED:
177 		type = RTM_DELETE;
178 		break;
179 	default:
180 		return;
181 	}
182 
183 	ifp = lltable_get_ifp(lle->lle_tbl);
184 
185 	bzero(&dst, sizeof(dst));
186 	bzero(&gw, sizeof(gw));
187 	bzero(&rtinfo, sizeof(rtinfo));
188 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
189 	dst.sin6_scope_id = in6_getscopezone(ifp,
190 	    in6_addrscope(&dst.sin6_addr));
191 	gw.sdl_len = sizeof(struct sockaddr_dl);
192 	gw.sdl_family = AF_LINK;
193 	gw.sdl_alen = ifp->if_addrlen;
194 	gw.sdl_index = ifp->if_index;
195 	gw.sdl_type = ifp->if_type;
196 	if (evt == LLENTRY_RESOLVED)
197 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
198 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
199 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
200 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
201 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
202 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
203 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
204 }
205 
206 /*
207  * A handler for interface link layer address change event.
208  */
209 static void
210 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
211 {
212 	if (ifp->if_afdata[AF_INET6] == NULL)
213 		return;
214 
215 	lltable_update_ifaddr(LLTABLE6(ifp));
216 }
217 
218 void
219 nd6_init(void)
220 {
221 
222 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
223 	rw_init(&V_nd6_lock, "nd6 list");
224 
225 	LIST_INIT(&V_nd_prefix);
226 	nd6_defrouter_init();
227 
228 	/* Start timers. */
229 	callout_init(&V_nd6_slowtimo_ch, 1);
230 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
231 	    nd6_slowtimo, curvnet);
232 
233 	callout_init(&V_nd6_timer_ch, 1);
234 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
235 
236 	nd6_dad_init();
237 	if (IS_DEFAULT_VNET(curvnet)) {
238 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
239 		    NULL, EVENTHANDLER_PRI_ANY);
240 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
241 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
242 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
243 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
244 	}
245 }
246 
247 #ifdef VIMAGE
248 void
249 nd6_destroy(void)
250 {
251 
252 	callout_drain(&V_nd6_slowtimo_ch);
253 	callout_drain(&V_nd6_timer_ch);
254 	if (IS_DEFAULT_VNET(curvnet)) {
255 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
256 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
257 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
258 	}
259 	rw_destroy(&V_nd6_lock);
260 	mtx_destroy(&V_nd6_onlink_mtx);
261 }
262 #endif
263 
264 struct nd_ifinfo *
265 nd6_ifattach(struct ifnet *ifp)
266 {
267 	struct nd_ifinfo *nd;
268 
269 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
270 	nd->initialized = 1;
271 
272 	nd->chlim = IPV6_DEFHLIM;
273 	nd->basereachable = REACHABLE_TIME;
274 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
275 	nd->retrans = RETRANS_TIMER;
276 
277 	nd->flags = ND6_IFF_PERFORMNUD;
278 
279 	/* Set IPv6 disabled on all interfaces but loopback by default. */
280 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
281 		nd->flags |= ND6_IFF_IFDISABLED;
282 
283 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
284 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
285 	 * default regardless of the V_ip6_auto_linklocal configuration to
286 	 * give a reasonable default behavior.
287 	 */
288 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE &&
289 	    ifp->if_type != IFT_WIREGUARD) || (ifp->if_flags & IFF_LOOPBACK))
290 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
291 	/*
292 	 * A loopback interface does not need to accept RTADV.
293 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
294 	 * default regardless of the V_ip6_accept_rtadv configuration to
295 	 * prevent the interface from accepting RA messages arrived
296 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
297 	 */
298 	if (V_ip6_accept_rtadv &&
299 	    !(ifp->if_flags & IFF_LOOPBACK) &&
300 	    (ifp->if_type != IFT_BRIDGE)) {
301 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
302 			/* If we globally accept rtadv, assume IPv6 on. */
303 			nd->flags &= ~ND6_IFF_IFDISABLED;
304 	}
305 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
306 		nd->flags |= ND6_IFF_NO_RADR;
307 
308 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
309 	nd6_setmtu0(ifp, nd);
310 
311 	return nd;
312 }
313 
314 void
315 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
316 {
317 	struct epoch_tracker et;
318 	struct ifaddr *ifa, *next;
319 
320 	NET_EPOCH_ENTER(et);
321 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
322 		if (ifa->ifa_addr->sa_family != AF_INET6)
323 			continue;
324 
325 		/* stop DAD processing */
326 		nd6_dad_stop(ifa);
327 	}
328 	NET_EPOCH_EXIT(et);
329 
330 	free(nd, M_IP6NDP);
331 }
332 
333 /*
334  * Reset ND level link MTU. This function is called when the physical MTU
335  * changes, which means we might have to adjust the ND level MTU.
336  */
337 void
338 nd6_setmtu(struct ifnet *ifp)
339 {
340 	if (ifp->if_afdata[AF_INET6] == NULL)
341 		return;
342 
343 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
344 }
345 
346 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
347 void
348 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
349 {
350 	u_int32_t omaxmtu;
351 
352 	omaxmtu = ndi->maxmtu;
353 	ndi->maxmtu = ifp->if_mtu;
354 
355 	/*
356 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
357 	 * undesirable situation.  We thus notify the operator of the change
358 	 * explicitly.  The check for omaxmtu is necessary to restrict the
359 	 * log to the case of changing the MTU, not initializing it.
360 	 */
361 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
362 		log(LOG_NOTICE, "nd6_setmtu0: "
363 		    "new link MTU on %s (%lu) is too small for IPv6\n",
364 		    if_name(ifp), (unsigned long)ndi->maxmtu);
365 	}
366 
367 	if (ndi->maxmtu > V_in6_maxmtu)
368 		in6_setmaxmtu(); /* check all interfaces just in case */
369 
370 }
371 
372 void
373 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
374 {
375 
376 	bzero(ndopts, sizeof(*ndopts));
377 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
378 	ndopts->nd_opts_last
379 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
380 
381 	if (icmp6len == 0) {
382 		ndopts->nd_opts_done = 1;
383 		ndopts->nd_opts_search = NULL;
384 	}
385 }
386 
387 /*
388  * Take one ND option.
389  */
390 struct nd_opt_hdr *
391 nd6_option(union nd_opts *ndopts)
392 {
393 	struct nd_opt_hdr *nd_opt;
394 	int olen;
395 
396 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
397 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
398 	    __func__));
399 	if (ndopts->nd_opts_search == NULL)
400 		return NULL;
401 	if (ndopts->nd_opts_done)
402 		return NULL;
403 
404 	nd_opt = ndopts->nd_opts_search;
405 
406 	/* make sure nd_opt_len is inside the buffer */
407 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
408 		bzero(ndopts, sizeof(*ndopts));
409 		return NULL;
410 	}
411 
412 	olen = nd_opt->nd_opt_len << 3;
413 	if (olen == 0) {
414 		/*
415 		 * Message validation requires that all included
416 		 * options have a length that is greater than zero.
417 		 */
418 		bzero(ndopts, sizeof(*ndopts));
419 		return NULL;
420 	}
421 
422 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
423 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
424 		/* option overruns the end of buffer, invalid */
425 		bzero(ndopts, sizeof(*ndopts));
426 		return NULL;
427 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
428 		/* reached the end of options chain */
429 		ndopts->nd_opts_done = 1;
430 		ndopts->nd_opts_search = NULL;
431 	}
432 	return nd_opt;
433 }
434 
435 /*
436  * Parse multiple ND options.
437  * This function is much easier to use, for ND routines that do not need
438  * multiple options of the same type.
439  */
440 int
441 nd6_options(union nd_opts *ndopts)
442 {
443 	struct nd_opt_hdr *nd_opt;
444 	int i = 0;
445 
446 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
447 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
448 	    __func__));
449 	if (ndopts->nd_opts_search == NULL)
450 		return 0;
451 
452 	while (1) {
453 		nd_opt = nd6_option(ndopts);
454 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
455 			/*
456 			 * Message validation requires that all included
457 			 * options have a length that is greater than zero.
458 			 */
459 			ICMP6STAT_INC(icp6s_nd_badopt);
460 			bzero(ndopts, sizeof(*ndopts));
461 			return -1;
462 		}
463 
464 		if (nd_opt == NULL)
465 			goto skip1;
466 
467 		switch (nd_opt->nd_opt_type) {
468 		case ND_OPT_SOURCE_LINKADDR:
469 		case ND_OPT_TARGET_LINKADDR:
470 		case ND_OPT_MTU:
471 		case ND_OPT_REDIRECTED_HEADER:
472 		case ND_OPT_NONCE:
473 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
474 				nd6log((LOG_INFO,
475 				    "duplicated ND6 option found (type=%d)\n",
476 				    nd_opt->nd_opt_type));
477 				/* XXX bark? */
478 			} else {
479 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
480 					= nd_opt;
481 			}
482 			break;
483 		case ND_OPT_PREFIX_INFORMATION:
484 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
485 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
486 					= nd_opt;
487 			}
488 			ndopts->nd_opts_pi_end =
489 				(struct nd_opt_prefix_info *)nd_opt;
490 			break;
491 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
492 		case ND_OPT_RDNSS:	/* RFC 6106 */
493 		case ND_OPT_DNSSL:	/* RFC 6106 */
494 			/*
495 			 * Silently ignore options we know and do not care about
496 			 * in the kernel.
497 			 */
498 			break;
499 		default:
500 			/*
501 			 * Unknown options must be silently ignored,
502 			 * to accommodate future extension to the protocol.
503 			 */
504 			nd6log((LOG_DEBUG,
505 			    "nd6_options: unsupported option %d - "
506 			    "option ignored\n", nd_opt->nd_opt_type));
507 		}
508 
509 skip1:
510 		i++;
511 		if (i > V_nd6_maxndopt) {
512 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
513 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
514 			break;
515 		}
516 
517 		if (ndopts->nd_opts_done)
518 			break;
519 	}
520 
521 	return 0;
522 }
523 
524 /*
525  * ND6 timer routine to handle ND6 entries
526  */
527 static void
528 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
529 {
530 	int canceled;
531 
532 	LLE_WLOCK_ASSERT(ln);
533 
534 	/* Do not schedule timers for child LLEs. */
535 	if (ln->la_flags & LLE_CHILD)
536 		return;
537 
538 	if (tick < 0) {
539 		ln->la_expire = 0;
540 		ln->ln_ntick = 0;
541 		canceled = callout_stop(&ln->lle_timer);
542 	} else {
543 		ln->la_expire = time_uptime + tick / hz;
544 		LLE_ADDREF(ln);
545 		if (tick > INT_MAX) {
546 			ln->ln_ntick = tick - INT_MAX;
547 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
548 			    nd6_llinfo_timer, ln);
549 		} else {
550 			ln->ln_ntick = 0;
551 			canceled = callout_reset(&ln->lle_timer, tick,
552 			    nd6_llinfo_timer, ln);
553 		}
554 	}
555 	if (canceled > 0)
556 		LLE_REMREF(ln);
557 }
558 
559 /*
560  * Gets source address of the first packet in hold queue
561  * and stores it in @src.
562  * Returns pointer to @src (if hold queue is not empty) or NULL.
563  *
564  * Set noinline to be dtrace-friendly
565  */
566 static __noinline struct in6_addr *
567 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
568 {
569 	struct ip6_hdr hdr;
570 	struct mbuf *m;
571 
572 	if (ln->la_hold == NULL)
573 		return (NULL);
574 
575 	/*
576 	 * assume every packet in la_hold has the same IP header
577 	 */
578 	m = ln->la_hold;
579 	if (sizeof(hdr) > m->m_len)
580 		return (NULL);
581 
582 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
583 	*src = hdr.ip6_src;
584 
585 	return (src);
586 }
587 
588 /*
589  * Checks if we need to switch from STALE state.
590  *
591  * RFC 4861 requires switching from STALE to DELAY state
592  * on first packet matching entry, waiting V_nd6_delay and
593  * transition to PROBE state (if upper layer confirmation was
594  * not received).
595  *
596  * This code performs a bit differently:
597  * On packet hit we don't change state (but desired state
598  * can be guessed by control plane). However, after V_nd6_delay
599  * seconds code will transition to PROBE state (so DELAY state
600  * is kinda skipped in most situations).
601  *
602  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
603  * we perform the following upon entering STALE state:
604  *
605  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
606  * if packet was transmitted at the start of given interval, we
607  * would be able to switch to PROBE state in V_nd6_delay seconds
608  * as user expects.
609  *
610  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
611  * lle in STALE state (remaining timer value stored in lle_remtime).
612  *
613  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
614  * seconds ago.
615  *
616  * Returns non-zero value if the entry is still STALE (storing
617  * the next timer interval in @pdelay).
618  *
619  * Returns zero value if original timer expired or we need to switch to
620  * PROBE (store that in @do_switch variable).
621  */
622 static int
623 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
624 {
625 	int nd_delay, nd_gctimer;
626 	time_t lle_hittime;
627 	long delay;
628 
629 	*do_switch = 0;
630 	nd_gctimer = V_nd6_gctimer;
631 	nd_delay = V_nd6_delay;
632 
633 	lle_hittime = llentry_get_hittime(lle);
634 
635 	if (lle_hittime == 0) {
636 		/*
637 		 * Datapath feedback has been requested upon entering
638 		 * STALE state. No packets has been passed using this lle.
639 		 * Ask for the timer reschedule and keep STALE state.
640 		 */
641 		delay = (long)(MIN(nd_gctimer, nd_delay));
642 		delay *= hz;
643 		if (lle->lle_remtime > delay)
644 			lle->lle_remtime -= delay;
645 		else {
646 			delay = lle->lle_remtime;
647 			lle->lle_remtime = 0;
648 		}
649 
650 		if (delay == 0) {
651 			/*
652 			 * The original ng6_gctime timeout ended,
653 			 * no more rescheduling.
654 			 */
655 			return (0);
656 		}
657 
658 		*pdelay = delay;
659 		return (1);
660 	}
661 
662 	/*
663 	 * Packet received. Verify timestamp
664 	 */
665 	delay = (long)(time_uptime - lle_hittime);
666 	if (delay < nd_delay) {
667 		/*
668 		 * V_nd6_delay still not passed since the first
669 		 * hit in STALE state.
670 		 * Reschedule timer and return.
671 		 */
672 		*pdelay = (long)(nd_delay - delay) * hz;
673 		return (1);
674 	}
675 
676 	/* Request switching to probe */
677 	*do_switch = 1;
678 	return (0);
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 		llentry_request_feedback(lle);
710 		nd_delay = V_nd6_delay;
711 		nd_gctimer = V_nd6_gctimer;
712 
713 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
714 		remtime = (long)nd_gctimer * hz - delay;
715 		break;
716 	case ND6_LLINFO_DELAY:
717 		lle->la_asked = 0;
718 		delay = (long)V_nd6_delay * hz;
719 		break;
720 	}
721 
722 	if (delay > 0)
723 		nd6_llinfo_settimer_locked(lle, delay);
724 
725 	lle->lle_remtime = remtime;
726 	lle->ln_state = newstate;
727 }
728 
729 /*
730  * Timer-dependent part of nd state machine.
731  *
732  * Set noinline to be dtrace-friendly
733  */
734 static __noinline void
735 nd6_llinfo_timer(void *arg)
736 {
737 	struct epoch_tracker et;
738 	struct llentry *ln;
739 	struct in6_addr *dst, *pdst, *psrc, src;
740 	struct ifnet *ifp;
741 	struct nd_ifinfo *ndi;
742 	int do_switch, send_ns;
743 	long delay;
744 
745 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
746 	ln = (struct llentry *)arg;
747 	ifp = lltable_get_ifp(ln->lle_tbl);
748 	CURVNET_SET(ifp->if_vnet);
749 
750 	ND6_RLOCK();
751 	LLE_WLOCK(ln);
752 	if (callout_pending(&ln->lle_timer)) {
753 		/*
754 		 * Here we are a bit odd here in the treatment of
755 		 * active/pending. If the pending bit is set, it got
756 		 * rescheduled before I ran. The active
757 		 * bit we ignore, since if it was stopped
758 		 * in ll_tablefree() and was currently running
759 		 * it would have return 0 so the code would
760 		 * not have deleted it since the callout could
761 		 * not be stopped so we want to go through
762 		 * with the delete here now. If the callout
763 		 * was restarted, the pending bit will be back on and
764 		 * we just want to bail since the callout_reset would
765 		 * return 1 and our reference would have been removed
766 		 * by nd6_llinfo_settimer_locked above since canceled
767 		 * would have been 1.
768 		 */
769 		LLE_WUNLOCK(ln);
770 		ND6_RUNLOCK();
771 		CURVNET_RESTORE();
772 		return;
773 	}
774 	NET_EPOCH_ENTER(et);
775 	ndi = ND_IFINFO(ifp);
776 	send_ns = 0;
777 	dst = &ln->r_l3addr.addr6;
778 	pdst = dst;
779 
780 	if (ln->ln_ntick > 0) {
781 		if (ln->ln_ntick > INT_MAX) {
782 			ln->ln_ntick -= INT_MAX;
783 			nd6_llinfo_settimer_locked(ln, INT_MAX);
784 		} else {
785 			ln->ln_ntick = 0;
786 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
787 		}
788 		goto done;
789 	}
790 
791 	if (ln->la_flags & LLE_STATIC) {
792 		goto done;
793 	}
794 
795 	if (ln->la_flags & LLE_DELETED) {
796 		nd6_free(&ln, 0);
797 		goto done;
798 	}
799 
800 	switch (ln->ln_state) {
801 	case ND6_LLINFO_INCOMPLETE:
802 		if (ln->la_asked < V_nd6_mmaxtries) {
803 			ln->la_asked++;
804 			send_ns = 1;
805 			/* Send NS to multicast address */
806 			pdst = NULL;
807 		} else {
808 			struct mbuf *m;
809 
810 			ICMP6STAT_ADD(icp6s_dropped, ln->la_numheld);
811 
812 			m = ln->la_hold;
813 			if (m != NULL) {
814 				/*
815 				 * assuming every packet in la_hold has the
816 				 * same IP header.  Send error after unlock.
817 				 */
818 				ln->la_hold = m->m_nextpkt;
819 				m->m_nextpkt = NULL;
820 				ln->la_numheld--;
821 			}
822 			nd6_free(&ln, 0);
823 			if (m != NULL) {
824 				struct mbuf *n = m;
825 
826 				/*
827 				 * if there are any ummapped mbufs, we
828 				 * must free them, rather than using
829 				 * them for an ICMP, as they cannot be
830 				 * checksummed.
831 				 */
832 				while ((n = n->m_next) != NULL) {
833 					if (n->m_flags & M_EXTPG)
834 						break;
835 				}
836 				if (n != NULL) {
837 					m_freem(m);
838 					m = NULL;
839 				} else {
840 					icmp6_error2(m, ICMP6_DST_UNREACH,
841 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
842 				}
843 			}
844 		}
845 		break;
846 	case ND6_LLINFO_REACHABLE:
847 		if (!ND6_LLINFO_PERMANENT(ln))
848 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
849 		break;
850 
851 	case ND6_LLINFO_STALE:
852 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
853 			/*
854 			 * No packet has used this entry and GC timeout
855 			 * has not been passed. Reschedule timer and
856 			 * return.
857 			 */
858 			nd6_llinfo_settimer_locked(ln, delay);
859 			break;
860 		}
861 
862 		if (do_switch == 0) {
863 			/*
864 			 * GC timer has ended and entry hasn't been used.
865 			 * Run Garbage collector (RFC 4861, 5.3)
866 			 */
867 			if (!ND6_LLINFO_PERMANENT(ln))
868 				nd6_free(&ln, 1);
869 			break;
870 		}
871 
872 		/* Entry has been used AND delay timer has ended. */
873 
874 		/* FALLTHROUGH */
875 
876 	case ND6_LLINFO_DELAY:
877 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
878 			/* We need NUD */
879 			ln->la_asked = 1;
880 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
881 			send_ns = 1;
882 		} else
883 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
884 		break;
885 	case ND6_LLINFO_PROBE:
886 		if (ln->la_asked < V_nd6_umaxtries) {
887 			ln->la_asked++;
888 			send_ns = 1;
889 		} else {
890 			nd6_free(&ln, 0);
891 		}
892 		break;
893 	default:
894 		panic("%s: paths in a dark night can be confusing: %d",
895 		    __func__, ln->ln_state);
896 	}
897 done:
898 	if (ln != NULL)
899 		ND6_RUNLOCK();
900 	if (send_ns != 0) {
901 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
902 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
903 		LLE_FREE_LOCKED(ln);
904 		ln = NULL;
905 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
906 	}
907 
908 	if (ln != NULL)
909 		LLE_FREE_LOCKED(ln);
910 	NET_EPOCH_EXIT(et);
911 	CURVNET_RESTORE();
912 }
913 
914 /*
915  * ND6 timer routine to expire default route list and prefix list
916  */
917 void
918 nd6_timer(void *arg)
919 {
920 	CURVNET_SET((struct vnet *) arg);
921 	struct epoch_tracker et;
922 	struct nd_prhead prl;
923 	struct nd_prefix *pr, *npr;
924 	struct ifnet *ifp;
925 	struct in6_ifaddr *ia6, *nia6;
926 	uint64_t genid;
927 
928 	LIST_INIT(&prl);
929 
930 	NET_EPOCH_ENTER(et);
931 	nd6_defrouter_timer();
932 
933 	/*
934 	 * expire interface addresses.
935 	 * in the past the loop was inside prefix expiry processing.
936 	 * However, from a stricter speci-confrmance standpoint, we should
937 	 * rather separate address lifetimes and prefix lifetimes.
938 	 *
939 	 * XXXRW: in6_ifaddrhead locking.
940 	 */
941   addrloop:
942 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
943 		/* check address lifetime */
944 		if (IFA6_IS_INVALID(ia6)) {
945 			int regen = 0;
946 
947 			/*
948 			 * If the expiring address is temporary, try
949 			 * regenerating a new one.  This would be useful when
950 			 * we suspended a laptop PC, then turned it on after a
951 			 * period that could invalidate all temporary
952 			 * addresses.  Although we may have to restart the
953 			 * loop (see below), it must be after purging the
954 			 * address.  Otherwise, we'd see an infinite loop of
955 			 * regeneration.
956 			 */
957 			if (V_ip6_use_tempaddr &&
958 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
959 				if (regen_tmpaddr(ia6) == 0)
960 					regen = 1;
961 			}
962 
963 			in6_purgeaddr(&ia6->ia_ifa);
964 
965 			if (regen)
966 				goto addrloop; /* XXX: see below */
967 		} else if (IFA6_IS_DEPRECATED(ia6)) {
968 			int oldflags = ia6->ia6_flags;
969 
970 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
971 
972 			/*
973 			 * If a temporary address has just become deprecated,
974 			 * regenerate a new one if possible.
975 			 */
976 			if (V_ip6_use_tempaddr &&
977 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
978 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
979 				if (regen_tmpaddr(ia6) == 0) {
980 					/*
981 					 * A new temporary address is
982 					 * generated.
983 					 * XXX: this means the address chain
984 					 * has changed while we are still in
985 					 * the loop.  Although the change
986 					 * would not cause disaster (because
987 					 * it's not a deletion, but an
988 					 * addition,) we'd rather restart the
989 					 * loop just for safety.  Or does this
990 					 * significantly reduce performance??
991 					 */
992 					goto addrloop;
993 				}
994 			}
995 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
996 			/*
997 			 * Schedule DAD for a tentative address.  This happens
998 			 * if the interface was down or not running
999 			 * when the address was configured.
1000 			 */
1001 			int delay;
1002 
1003 			delay = arc4random() %
1004 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1005 			nd6_dad_start((struct ifaddr *)ia6, delay);
1006 		} else {
1007 			/*
1008 			 * Check status of the interface.  If it is down,
1009 			 * mark the address as tentative for future DAD.
1010 			 */
1011 			ifp = ia6->ia_ifp;
1012 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1013 			    ((ifp->if_flags & IFF_UP) == 0 ||
1014 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1015 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1016 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1017 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1018 			}
1019 
1020 			/*
1021 			 * A new RA might have made a deprecated address
1022 			 * preferred.
1023 			 */
1024 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1025 		}
1026 	}
1027 	NET_EPOCH_EXIT(et);
1028 
1029 	ND6_WLOCK();
1030 restart:
1031 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1032 		/*
1033 		 * Expire prefixes. Since the pltime is only used for
1034 		 * autoconfigured addresses, pltime processing for prefixes is
1035 		 * not necessary.
1036 		 *
1037 		 * Only unlink after all derived addresses have expired. This
1038 		 * may not occur until two hours after the prefix has expired
1039 		 * per RFC 4862. If the prefix expires before its derived
1040 		 * addresses, mark it off-link. This will be done automatically
1041 		 * after unlinking if no address references remain.
1042 		 */
1043 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1044 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1045 			continue;
1046 
1047 		if (pr->ndpr_addrcnt == 0) {
1048 			nd6_prefix_unlink(pr, &prl);
1049 			continue;
1050 		}
1051 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1052 			genid = V_nd6_list_genid;
1053 			nd6_prefix_ref(pr);
1054 			ND6_WUNLOCK();
1055 			ND6_ONLINK_LOCK();
1056 			(void)nd6_prefix_offlink(pr);
1057 			ND6_ONLINK_UNLOCK();
1058 			ND6_WLOCK();
1059 			nd6_prefix_rele(pr);
1060 			if (genid != V_nd6_list_genid)
1061 				goto restart;
1062 		}
1063 	}
1064 	ND6_WUNLOCK();
1065 
1066 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1067 		LIST_REMOVE(pr, ndpr_entry);
1068 		nd6_prefix_del(pr);
1069 	}
1070 
1071 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1072 	    nd6_timer, curvnet);
1073 
1074 	CURVNET_RESTORE();
1075 }
1076 
1077 /*
1078  * ia6 - deprecated/invalidated temporary address
1079  */
1080 static int
1081 regen_tmpaddr(struct in6_ifaddr *ia6)
1082 {
1083 	struct ifaddr *ifa;
1084 	struct ifnet *ifp;
1085 	struct in6_ifaddr *public_ifa6 = NULL;
1086 
1087 	NET_EPOCH_ASSERT();
1088 
1089 	ifp = ia6->ia_ifa.ifa_ifp;
1090 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1091 		struct in6_ifaddr *it6;
1092 
1093 		if (ifa->ifa_addr->sa_family != AF_INET6)
1094 			continue;
1095 
1096 		it6 = (struct in6_ifaddr *)ifa;
1097 
1098 		/* ignore no autoconf addresses. */
1099 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1100 			continue;
1101 
1102 		/* ignore autoconf addresses with different prefixes. */
1103 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1104 			continue;
1105 
1106 		/*
1107 		 * Now we are looking at an autoconf address with the same
1108 		 * prefix as ours.  If the address is temporary and is still
1109 		 * preferred, do not create another one.  It would be rare, but
1110 		 * could happen, for example, when we resume a laptop PC after
1111 		 * a long period.
1112 		 */
1113 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1114 		    !IFA6_IS_DEPRECATED(it6)) {
1115 			public_ifa6 = NULL;
1116 			break;
1117 		}
1118 
1119 		/*
1120 		 * This is a public autoconf address that has the same prefix
1121 		 * as ours.  If it is preferred, keep it.  We can't break the
1122 		 * loop here, because there may be a still-preferred temporary
1123 		 * address with the prefix.
1124 		 */
1125 		if (!IFA6_IS_DEPRECATED(it6))
1126 			public_ifa6 = it6;
1127 	}
1128 	if (public_ifa6 != NULL)
1129 		ifa_ref(&public_ifa6->ia_ifa);
1130 
1131 	if (public_ifa6 != NULL) {
1132 		int e;
1133 
1134 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1135 			ifa_free(&public_ifa6->ia_ifa);
1136 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1137 			    " tmp addr,errno=%d\n", e);
1138 			return (-1);
1139 		}
1140 		ifa_free(&public_ifa6->ia_ifa);
1141 		return (0);
1142 	}
1143 
1144 	return (-1);
1145 }
1146 
1147 /*
1148  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1149  * cache entries are freed in in6_domifdetach().
1150  */
1151 void
1152 nd6_purge(struct ifnet *ifp)
1153 {
1154 	struct nd_prhead prl;
1155 	struct nd_prefix *pr, *npr;
1156 
1157 	LIST_INIT(&prl);
1158 
1159 	/* Purge default router list entries toward ifp. */
1160 	nd6_defrouter_purge(ifp);
1161 
1162 	ND6_WLOCK();
1163 	/*
1164 	 * Remove prefixes on ifp. We should have already removed addresses on
1165 	 * this interface, so no addresses should be referencing these prefixes.
1166 	 */
1167 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1168 		if (pr->ndpr_ifp == ifp)
1169 			nd6_prefix_unlink(pr, &prl);
1170 	}
1171 	ND6_WUNLOCK();
1172 
1173 	/* Delete the unlinked prefix objects. */
1174 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1175 		LIST_REMOVE(pr, ndpr_entry);
1176 		nd6_prefix_del(pr);
1177 	}
1178 
1179 	/* cancel default outgoing interface setting */
1180 	if (V_nd6_defifindex == ifp->if_index)
1181 		nd6_setdefaultiface(0);
1182 
1183 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1184 		/* Refresh default router list. */
1185 		defrouter_select_fib(ifp->if_fib);
1186 	}
1187 }
1188 
1189 /*
1190  * the caller acquires and releases the lock on the lltbls
1191  * Returns the llentry locked
1192  */
1193 struct llentry *
1194 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1195 {
1196 	struct sockaddr_in6 sin6;
1197 	struct llentry *ln;
1198 
1199 	bzero(&sin6, sizeof(sin6));
1200 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1201 	sin6.sin6_family = AF_INET6;
1202 	sin6.sin6_addr = *addr6;
1203 
1204 	IF_AFDATA_LOCK_ASSERT(ifp);
1205 
1206 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1207 
1208 	return (ln);
1209 }
1210 
1211 static struct llentry *
1212 nd6_alloc(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 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1223 	if (ln != NULL)
1224 		ln->ln_state = ND6_LLINFO_NOSTATE;
1225 
1226 	return (ln);
1227 }
1228 
1229 /*
1230  * Test whether a given IPv6 address can be a neighbor.
1231  */
1232 static bool
1233 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1234 {
1235 
1236 	/*
1237 	 * A link-local address is always a neighbor.
1238 	 * XXX: a link does not necessarily specify a single interface.
1239 	 */
1240 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1241 		struct sockaddr_in6 sin6_copy;
1242 		u_int32_t zone;
1243 
1244 		/*
1245 		 * We need sin6_copy since sa6_recoverscope() may modify the
1246 		 * content (XXX).
1247 		 */
1248 		sin6_copy = *addr;
1249 		if (sa6_recoverscope(&sin6_copy))
1250 			return (0); /* XXX: should be impossible */
1251 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1252 			return (0);
1253 		if (sin6_copy.sin6_scope_id == zone)
1254 			return (1);
1255 		else
1256 			return (0);
1257 	}
1258 	/* Checking global unicast */
1259 
1260 	/* If an address is directly reachable, it is a neigbor */
1261 	struct nhop_object *nh;
1262 	nh = fib6_lookup(ifp->if_fib, &addr->sin6_addr, 0, NHR_NONE, 0);
1263 	if (nh != NULL && nh->nh_aifp == ifp && (nh->nh_flags & NHF_GATEWAY) == 0)
1264 		return (true);
1265 
1266 	/*
1267 	 * Check prefixes with desired on-link state, as some may be not
1268 	 * installed in the routing table.
1269 	 */
1270 	bool matched = false;
1271 	struct nd_prefix *pr;
1272 	ND6_RLOCK();
1273 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1274 		if (pr->ndpr_ifp != ifp)
1275 			continue;
1276 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0)
1277 			continue;
1278 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1279 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1280 			matched = true;
1281 			break;
1282 		}
1283 	}
1284 	ND6_RUNLOCK();
1285 	if (matched)
1286 		return (true);
1287 
1288 	/*
1289 	 * If the address is assigned on the node of the other side of
1290 	 * a p2p interface, the address should be a neighbor.
1291 	 */
1292 	if (ifp->if_flags & IFF_POINTOPOINT) {
1293 		struct ifaddr *ifa;
1294 
1295 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1296 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1297 				continue;
1298 			if (ifa->ifa_dstaddr != NULL &&
1299 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1300 				return (true);
1301 			}
1302 		}
1303 	}
1304 
1305 	/*
1306 	 * If the default router list is empty, all addresses are regarded
1307 	 * as on-link, and thus, as a neighbor.
1308 	 */
1309 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1310 	    nd6_defrouter_list_empty() &&
1311 	    V_nd6_defifindex == ifp->if_index) {
1312 		return (1);
1313 	}
1314 
1315 	return (0);
1316 }
1317 
1318 /*
1319  * Detect if a given IPv6 address identifies a neighbor on a given link.
1320  * XXX: should take care of the destination of a p2p link?
1321  */
1322 int
1323 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1324 {
1325 	struct llentry *lle;
1326 	int rc = 0;
1327 
1328 	NET_EPOCH_ASSERT();
1329 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1330 	if (nd6_is_new_addr_neighbor(addr, ifp))
1331 		return (1);
1332 
1333 	/*
1334 	 * Even if the address matches none of our addresses, it might be
1335 	 * in the neighbor cache.
1336 	 */
1337 	if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1338 		LLE_RUNLOCK(lle);
1339 		rc = 1;
1340 	}
1341 	return (rc);
1342 }
1343 
1344 static __noinline void
1345 nd6_free_children(struct llentry *lle)
1346 {
1347 	struct llentry *child_lle;
1348 
1349 	NET_EPOCH_ASSERT();
1350 	LLE_WLOCK_ASSERT(lle);
1351 
1352 	while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1353 		LLE_WLOCK(child_lle);
1354 		lltable_unlink_child_entry(child_lle);
1355 		llentry_free(child_lle);
1356 	}
1357 }
1358 
1359 /*
1360  * Tries to update @lle address/prepend data with new @lladdr.
1361  *
1362  * Returns true on success.
1363  * In any case, @lle is returned wlocked.
1364  */
1365 static __noinline bool
1366 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1367 {
1368 	u_char buf[LLE_MAX_LINKHDR];
1369 	int fam, off;
1370 	size_t sz;
1371 
1372 	sz = sizeof(buf);
1373 	if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1374 		return (false);
1375 
1376 	/* Update data */
1377 	lltable_set_entry_addr(ifp, lle, buf, sz, off);
1378 
1379 	struct llentry *child_lle;
1380 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1381 		LLE_WLOCK(child_lle);
1382 		fam = child_lle->r_family;
1383 		sz = sizeof(buf);
1384 		if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1385 			/* success */
1386 			lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1387 			child_lle->ln_state = ND6_LLINFO_REACHABLE;
1388 		}
1389 		LLE_WUNLOCK(child_lle);
1390 	}
1391 
1392 	return (true);
1393 }
1394 
1395 bool
1396 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1397 {
1398 	NET_EPOCH_ASSERT();
1399 	LLE_WLOCK_ASSERT(lle);
1400 
1401 	if (!lltable_acquire_wlock(ifp, lle))
1402 		return (false);
1403 	bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1404 	IF_AFDATA_WUNLOCK(ifp);
1405 
1406 	return (ret);
1407 }
1408 
1409 /*
1410  * Free an nd6 llinfo entry.
1411  * Since the function would cause significant changes in the kernel, DO NOT
1412  * make it global, unless you have a strong reason for the change, and are sure
1413  * that the change is safe.
1414  *
1415  * Set noinline to be dtrace-friendly
1416  */
1417 static __noinline void
1418 nd6_free(struct llentry **lnp, int gc)
1419 {
1420 	struct ifnet *ifp;
1421 	struct llentry *ln;
1422 	struct nd_defrouter *dr;
1423 
1424 	ln = *lnp;
1425 	*lnp = NULL;
1426 
1427 	LLE_WLOCK_ASSERT(ln);
1428 	ND6_RLOCK_ASSERT();
1429 
1430 	KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1431 
1432 	ifp = lltable_get_ifp(ln->lle_tbl);
1433 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1434 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1435 	else
1436 		dr = NULL;
1437 	ND6_RUNLOCK();
1438 
1439 	if ((ln->la_flags & LLE_DELETED) == 0)
1440 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1441 
1442 	/*
1443 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1444 	 * even though it is not harmful, it was not really necessary.
1445 	 */
1446 
1447 	/* cancel timer */
1448 	nd6_llinfo_settimer_locked(ln, -1);
1449 
1450 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1451 		if (dr != NULL && dr->expire &&
1452 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1453 			/*
1454 			 * If the reason for the deletion is just garbage
1455 			 * collection, and the neighbor is an active default
1456 			 * router, do not delete it.  Instead, reset the GC
1457 			 * timer using the router's lifetime.
1458 			 * Simply deleting the entry would affect default
1459 			 * router selection, which is not necessarily a good
1460 			 * thing, especially when we're using router preference
1461 			 * values.
1462 			 * XXX: the check for ln_state would be redundant,
1463 			 *      but we intentionally keep it just in case.
1464 			 */
1465 			if (dr->expire > time_uptime)
1466 				nd6_llinfo_settimer_locked(ln,
1467 				    (dr->expire - time_uptime) * hz);
1468 			else
1469 				nd6_llinfo_settimer_locked(ln,
1470 				    (long)V_nd6_gctimer * hz);
1471 
1472 			LLE_REMREF(ln);
1473 			LLE_WUNLOCK(ln);
1474 			defrouter_rele(dr);
1475 			return;
1476 		}
1477 
1478 		if (dr) {
1479 			/*
1480 			 * Unreachability of a router might affect the default
1481 			 * router selection and on-link detection of advertised
1482 			 * prefixes.
1483 			 */
1484 
1485 			/*
1486 			 * Temporarily fake the state to choose a new default
1487 			 * router and to perform on-link determination of
1488 			 * prefixes correctly.
1489 			 * Below the state will be set correctly,
1490 			 * or the entry itself will be deleted.
1491 			 */
1492 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1493 		}
1494 
1495 		if (ln->ln_router || dr) {
1496 			/*
1497 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1498 			 * rnh and for the calls to pfxlist_onlink_check() and
1499 			 * defrouter_select_fib() in the block further down for calls
1500 			 * into nd6_lookup().  We still hold a ref.
1501 			 */
1502 			LLE_WUNLOCK(ln);
1503 
1504 			/*
1505 			 * rt6_flush must be called whether or not the neighbor
1506 			 * is in the Default Router List.
1507 			 * See a corresponding comment in nd6_na_input().
1508 			 */
1509 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1510 		}
1511 
1512 		if (dr) {
1513 			/*
1514 			 * Since defrouter_select_fib() does not affect the
1515 			 * on-link determination and MIP6 needs the check
1516 			 * before the default router selection, we perform
1517 			 * the check now.
1518 			 */
1519 			pfxlist_onlink_check();
1520 
1521 			/*
1522 			 * Refresh default router list.
1523 			 */
1524 			defrouter_select_fib(dr->ifp->if_fib);
1525 		}
1526 
1527 		/*
1528 		 * If this entry was added by an on-link redirect, remove the
1529 		 * corresponding host route.
1530 		 */
1531 		if (ln->la_flags & LLE_REDIRECT)
1532 			nd6_free_redirect(ln);
1533 
1534 		if (ln->ln_router || dr)
1535 			LLE_WLOCK(ln);
1536 	}
1537 
1538 	/*
1539 	 * Save to unlock. We still hold an extra reference and will not
1540 	 * free(9) in llentry_free() if someone else holds one as well.
1541 	 */
1542 	LLE_WUNLOCK(ln);
1543 	IF_AFDATA_LOCK(ifp);
1544 	LLE_WLOCK(ln);
1545 	/* Guard against race with other llentry_free(). */
1546 	if (ln->la_flags & LLE_LINKED) {
1547 		/* Remove callout reference */
1548 		LLE_REMREF(ln);
1549 		lltable_unlink_entry(ln->lle_tbl, ln);
1550 	}
1551 	IF_AFDATA_UNLOCK(ifp);
1552 
1553 	nd6_free_children(ln);
1554 
1555 	llentry_free(ln);
1556 	if (dr != NULL)
1557 		defrouter_rele(dr);
1558 }
1559 
1560 static int
1561 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1562 {
1563 
1564 	if (nh->nh_flags & NHF_REDIRECT)
1565 		return (1);
1566 
1567 	return (0);
1568 }
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 rib_cmd_info rc;
1580 	struct epoch_tracker et;
1581 
1582 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1583 
1584 	NET_EPOCH_ENTER(et);
1585 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1586 		rib_del_route_px(fibnum, (struct sockaddr *)&sin6, 128,
1587 		    nd6_isdynrte, NULL, 0, &rc);
1588 	NET_EPOCH_EXIT(et);
1589 }
1590 
1591 /*
1592  * Updates status of the default router route.
1593  */
1594 static void
1595 check_release_defrouter(const struct rib_cmd_info *rc, void *_cbdata)
1596 {
1597 	struct nd_defrouter *dr;
1598 	struct nhop_object *nh;
1599 
1600 	nh = rc->rc_nh_old;
1601 
1602 	if ((nh != NULL) && (nh->nh_flags & NHF_DEFAULT)) {
1603 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1604 		if (dr != NULL) {
1605 			dr->installed = 0;
1606 			defrouter_rele(dr);
1607 		}
1608 	}
1609 }
1610 
1611 void
1612 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1613 {
1614 
1615 #ifdef ROUTE_MPATH
1616 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1617 #else
1618 	check_release_defrouter(rc, NULL);
1619 #endif
1620 }
1621 
1622 int
1623 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1624 {
1625 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1626 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1627 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1628 	struct epoch_tracker et;
1629 	int error = 0;
1630 
1631 	if (ifp->if_afdata[AF_INET6] == NULL)
1632 		return (EPFNOSUPPORT);
1633 	switch (cmd) {
1634 	case OSIOCGIFINFO_IN6:
1635 #define ND	ndi->ndi
1636 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1637 		bzero(&ND, sizeof(ND));
1638 		ND.linkmtu = IN6_LINKMTU(ifp);
1639 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1640 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1641 		ND.reachable = ND_IFINFO(ifp)->reachable;
1642 		ND.retrans = ND_IFINFO(ifp)->retrans;
1643 		ND.flags = ND_IFINFO(ifp)->flags;
1644 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1645 		ND.chlim = ND_IFINFO(ifp)->chlim;
1646 		break;
1647 	case SIOCGIFINFO_IN6:
1648 		ND = *ND_IFINFO(ifp);
1649 		break;
1650 	case SIOCSIFINFO_IN6:
1651 		/*
1652 		 * used to change host variables from userland.
1653 		 * intended for a use on router to reflect RA configurations.
1654 		 */
1655 		/* 0 means 'unspecified' */
1656 		if (ND.linkmtu != 0) {
1657 			if (ND.linkmtu < IPV6_MMTU ||
1658 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1659 				error = EINVAL;
1660 				break;
1661 			}
1662 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1663 		}
1664 
1665 		if (ND.basereachable != 0) {
1666 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1667 
1668 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1669 			if (ND.basereachable != obasereachable)
1670 				ND_IFINFO(ifp)->reachable =
1671 				    ND_COMPUTE_RTIME(ND.basereachable);
1672 		}
1673 		if (ND.retrans != 0)
1674 			ND_IFINFO(ifp)->retrans = ND.retrans;
1675 		if (ND.chlim != 0)
1676 			ND_IFINFO(ifp)->chlim = ND.chlim;
1677 		/* FALLTHROUGH */
1678 	case SIOCSIFINFO_FLAGS:
1679 	{
1680 		struct ifaddr *ifa;
1681 		struct in6_ifaddr *ia;
1682 
1683 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1684 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1685 			/* ifdisabled 1->0 transision */
1686 
1687 			/*
1688 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1689 			 * has an link-local address with IN6_IFF_DUPLICATED,
1690 			 * do not clear ND6_IFF_IFDISABLED.
1691 			 * See RFC 4862, Section 5.4.5.
1692 			 */
1693 			NET_EPOCH_ENTER(et);
1694 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1695 				if (ifa->ifa_addr->sa_family != AF_INET6)
1696 					continue;
1697 				ia = (struct in6_ifaddr *)ifa;
1698 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1699 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1700 					break;
1701 			}
1702 			NET_EPOCH_EXIT(et);
1703 
1704 			if (ifa != NULL) {
1705 				/* LLA is duplicated. */
1706 				ND.flags |= ND6_IFF_IFDISABLED;
1707 				log(LOG_ERR, "Cannot enable an interface"
1708 				    " with a link-local address marked"
1709 				    " duplicate.\n");
1710 			} else {
1711 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1712 				if (ifp->if_flags & IFF_UP)
1713 					in6_if_up(ifp);
1714 			}
1715 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1716 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1717 			/* ifdisabled 0->1 transision */
1718 			/* Mark all IPv6 address as tentative. */
1719 
1720 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1721 			if (V_ip6_dad_count > 0 &&
1722 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1723 				NET_EPOCH_ENTER(et);
1724 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1725 				    ifa_link) {
1726 					if (ifa->ifa_addr->sa_family !=
1727 					    AF_INET6)
1728 						continue;
1729 					ia = (struct in6_ifaddr *)ifa;
1730 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1731 				}
1732 				NET_EPOCH_EXIT(et);
1733 			}
1734 		}
1735 
1736 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1737 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1738 				/* auto_linklocal 0->1 transision */
1739 
1740 				/* If no link-local address on ifp, configure */
1741 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1742 				in6_ifattach(ifp, NULL);
1743 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1744 			    ifp->if_flags & IFF_UP) {
1745 				/*
1746 				 * When the IF already has
1747 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1748 				 * address is assigned, and IFF_UP, try to
1749 				 * assign one.
1750 				 */
1751 				NET_EPOCH_ENTER(et);
1752 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1753 				    ifa_link) {
1754 					if (ifa->ifa_addr->sa_family !=
1755 					    AF_INET6)
1756 						continue;
1757 					ia = (struct in6_ifaddr *)ifa;
1758 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1759 						break;
1760 				}
1761 				NET_EPOCH_EXIT(et);
1762 				if (ifa != NULL)
1763 					/* No LLA is configured. */
1764 					in6_ifattach(ifp, NULL);
1765 			}
1766 		}
1767 		ND_IFINFO(ifp)->flags = ND.flags;
1768 		break;
1769 	}
1770 #undef ND
1771 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1772 		/* sync kernel routing table with the default router list */
1773 		defrouter_reset();
1774 		defrouter_select_fib(RT_ALL_FIBS);
1775 		break;
1776 	case SIOCSPFXFLUSH_IN6:
1777 	{
1778 		/* flush all the prefix advertised by routers */
1779 		struct in6_ifaddr *ia, *ia_next;
1780 		struct nd_prefix *pr, *next;
1781 		struct nd_prhead prl;
1782 
1783 		LIST_INIT(&prl);
1784 
1785 		ND6_WLOCK();
1786 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1787 			if (pr->ndpr_raf_ra_derived)
1788 				nd6_prefix_unlink(pr, &prl);
1789 		}
1790 		ND6_WUNLOCK();
1791 
1792 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1793 			LIST_REMOVE(pr, ndpr_entry);
1794 			/* XXXRW: in6_ifaddrhead locking. */
1795 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1796 			    ia_next) {
1797 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1798 					continue;
1799 
1800 				if (ia->ia6_ndpr == pr)
1801 					in6_purgeaddr(&ia->ia_ifa);
1802 			}
1803 			nd6_prefix_del(pr);
1804 		}
1805 		break;
1806 	}
1807 	case SIOCSRTRFLUSH_IN6:
1808 	{
1809 		/* flush all the default routers */
1810 
1811 		defrouter_reset();
1812 		nd6_defrouter_flush_all();
1813 		defrouter_select_fib(RT_ALL_FIBS);
1814 		break;
1815 	}
1816 	case SIOCGNBRINFO_IN6:
1817 	{
1818 		struct llentry *ln;
1819 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1820 
1821 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1822 			return (error);
1823 
1824 		NET_EPOCH_ENTER(et);
1825 		ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1826 		NET_EPOCH_EXIT(et);
1827 
1828 		if (ln == NULL) {
1829 			error = EINVAL;
1830 			break;
1831 		}
1832 		nbi->state = ln->ln_state;
1833 		nbi->asked = ln->la_asked;
1834 		nbi->isrouter = ln->ln_router;
1835 		if (ln->la_expire == 0)
1836 			nbi->expire = 0;
1837 		else
1838 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1839 			    (time_second - time_uptime);
1840 		LLE_RUNLOCK(ln);
1841 		break;
1842 	}
1843 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1844 		ndif->ifindex = V_nd6_defifindex;
1845 		break;
1846 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1847 		return (nd6_setdefaultiface(ndif->ifindex));
1848 	}
1849 	return (error);
1850 }
1851 
1852 /*
1853  * Calculates new isRouter value based on provided parameters and
1854  * returns it.
1855  */
1856 static int
1857 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1858     int ln_router)
1859 {
1860 
1861 	/*
1862 	 * ICMP6 type dependent behavior.
1863 	 *
1864 	 * NS: clear IsRouter if new entry
1865 	 * RS: clear IsRouter
1866 	 * RA: set IsRouter if there's lladdr
1867 	 * redir: clear IsRouter if new entry
1868 	 *
1869 	 * RA case, (1):
1870 	 * The spec says that we must set IsRouter in the following cases:
1871 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1872 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1873 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1874 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1875 	 * neighbor cache, this is similar to (6).
1876 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1877 	 *
1878 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1879 	 *							D R
1880 	 *	0	n	n	(1)	c   ?     s
1881 	 *	0	y	n	(2)	c   s     s
1882 	 *	0	n	y	(3)	c   s     s
1883 	 *	0	y	y	(4)	c   s     s
1884 	 *	0	y	y	(5)	c   s     s
1885 	 *	1	--	n	(6) c	c	c s
1886 	 *	1	--	y	(7) c	c   s	c s
1887 	 *
1888 	 *					(c=clear s=set)
1889 	 */
1890 	switch (type & 0xff) {
1891 	case ND_NEIGHBOR_SOLICIT:
1892 		/*
1893 		 * New entry must have is_router flag cleared.
1894 		 */
1895 		if (is_new)					/* (6-7) */
1896 			ln_router = 0;
1897 		break;
1898 	case ND_REDIRECT:
1899 		/*
1900 		 * If the icmp is a redirect to a better router, always set the
1901 		 * is_router flag.  Otherwise, if the entry is newly created,
1902 		 * clear the flag.  [RFC 2461, sec 8.3]
1903 		 */
1904 		if (code == ND_REDIRECT_ROUTER)
1905 			ln_router = 1;
1906 		else {
1907 			if (is_new)				/* (6-7) */
1908 				ln_router = 0;
1909 		}
1910 		break;
1911 	case ND_ROUTER_SOLICIT:
1912 		/*
1913 		 * is_router flag must always be cleared.
1914 		 */
1915 		ln_router = 0;
1916 		break;
1917 	case ND_ROUTER_ADVERT:
1918 		/*
1919 		 * Mark an entry with lladdr as a router.
1920 		 */
1921 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1922 		    (is_new && new_addr)) {			/* (7) */
1923 			ln_router = 1;
1924 		}
1925 		break;
1926 	}
1927 
1928 	return (ln_router);
1929 }
1930 
1931 /*
1932  * Create neighbor cache entry and cache link-layer address,
1933  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1934  *
1935  * type - ICMP6 type
1936  * code - type dependent information
1937  *
1938  */
1939 void
1940 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1941     int lladdrlen, int type, int code)
1942 {
1943 	struct llentry *ln = NULL, *ln_tmp;
1944 	int is_newentry;
1945 	int do_update;
1946 	int olladdr;
1947 	int llchange;
1948 	int flags;
1949 	uint16_t router = 0;
1950 	struct mbuf *chain = NULL;
1951 	u_char linkhdr[LLE_MAX_LINKHDR];
1952 	size_t linkhdrsize;
1953 	int lladdr_off;
1954 
1955 	NET_EPOCH_ASSERT();
1956 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1957 
1958 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1959 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1960 
1961 	/* nothing must be updated for unspecified address */
1962 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1963 		return;
1964 
1965 	/*
1966 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1967 	 * the caller.
1968 	 *
1969 	 * XXX If the link does not have link-layer adderss, what should
1970 	 * we do? (ifp->if_addrlen == 0)
1971 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1972 	 * description on it in NS section (RFC 2461 7.2.3).
1973 	 */
1974 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1975 	ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
1976 	is_newentry = 0;
1977 	if (ln == NULL) {
1978 		flags |= LLE_EXCLUSIVE;
1979 		ln = nd6_alloc(from, 0, ifp);
1980 		if (ln == NULL)
1981 			return;
1982 
1983 		/*
1984 		 * Since we already know all the data for the new entry,
1985 		 * fill it before insertion.
1986 		 */
1987 		if (lladdr != NULL) {
1988 			linkhdrsize = sizeof(linkhdr);
1989 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1990 			    linkhdr, &linkhdrsize, &lladdr_off) != 0) {
1991 				lltable_free_entry(LLTABLE6(ifp), ln);
1992 				return;
1993 			}
1994 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1995 			    lladdr_off);
1996 		}
1997 
1998 		IF_AFDATA_WLOCK(ifp);
1999 		LLE_WLOCK(ln);
2000 		/* Prefer any existing lle over newly-created one */
2001 		ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2002 		if (ln_tmp == NULL)
2003 			lltable_link_entry(LLTABLE6(ifp), ln);
2004 		IF_AFDATA_WUNLOCK(ifp);
2005 		if (ln_tmp == NULL) {
2006 			/* No existing lle, mark as new entry (6,7) */
2007 			is_newentry = 1;
2008 			if (lladdr != NULL) {	/* (7) */
2009 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2010 				EVENTHANDLER_INVOKE(lle_event, ln,
2011 				    LLENTRY_RESOLVED);
2012 			}
2013 		} else {
2014 			lltable_free_entry(LLTABLE6(ifp), ln);
2015 			ln = ln_tmp;
2016 			ln_tmp = NULL;
2017 		}
2018 	}
2019 	/* do nothing if static ndp is set */
2020 	if ((ln->la_flags & LLE_STATIC)) {
2021 		if (flags & LLE_EXCLUSIVE)
2022 			LLE_WUNLOCK(ln);
2023 		else
2024 			LLE_RUNLOCK(ln);
2025 		return;
2026 	}
2027 
2028 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2029 	if (olladdr && lladdr) {
2030 		llchange = bcmp(lladdr, ln->ll_addr,
2031 		    ifp->if_addrlen);
2032 	} else if (!olladdr && lladdr)
2033 		llchange = 1;
2034 	else
2035 		llchange = 0;
2036 
2037 	/*
2038 	 * newentry olladdr  lladdr  llchange	(*=record)
2039 	 *	0	n	n	--	(1)
2040 	 *	0	y	n	--	(2)
2041 	 *	0	n	y	y	(3) * STALE
2042 	 *	0	y	y	n	(4) *
2043 	 *	0	y	y	y	(5) * STALE
2044 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2045 	 *	1	--	y	--	(7) * STALE
2046 	 */
2047 
2048 	do_update = 0;
2049 	if (is_newentry == 0 && llchange != 0) {
2050 		do_update = 1;	/* (3,5) */
2051 
2052 		/*
2053 		 * Record source link-layer address
2054 		 * XXX is it dependent to ifp->if_type?
2055 		 */
2056 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2057 			/* Entry was deleted */
2058 			LLE_WUNLOCK(ln);
2059 			return;
2060 		}
2061 
2062 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2063 
2064 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2065 
2066 		if (ln->la_hold != NULL)
2067 			chain = nd6_grab_holdchain(ln);
2068 	}
2069 
2070 	/* Calculates new router status */
2071 	router = nd6_is_router(type, code, is_newentry, olladdr,
2072 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2073 
2074 	ln->ln_router = router;
2075 	/* Mark non-router redirects with special flag */
2076 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2077 		ln->la_flags |= LLE_REDIRECT;
2078 
2079 	if (flags & LLE_EXCLUSIVE)
2080 		LLE_WUNLOCK(ln);
2081 	else
2082 		LLE_RUNLOCK(ln);
2083 
2084 	if (chain != NULL)
2085 		nd6_flush_holdchain(ifp, ln, chain);
2086 	if (do_update)
2087 		nd6_flush_children_holdchain(ifp, ln);
2088 
2089 	/*
2090 	 * When the link-layer address of a router changes, select the
2091 	 * best router again.  In particular, when the neighbor entry is newly
2092 	 * created, it might affect the selection policy.
2093 	 * Question: can we restrict the first condition to the "is_newentry"
2094 	 * case?
2095 	 * XXX: when we hear an RA from a new router with the link-layer
2096 	 * address option, defrouter_select_fib() is called twice, since
2097 	 * defrtrlist_update called the function as well.  However, I believe
2098 	 * we can compromise the overhead, since it only happens the first
2099 	 * time.
2100 	 * XXX: although defrouter_select_fib() should not have a bad effect
2101 	 * for those are not autoconfigured hosts, we explicitly avoid such
2102 	 * cases for safety.
2103 	 */
2104 	if ((do_update || is_newentry) && router &&
2105 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2106 		/*
2107 		 * guaranteed recursion
2108 		 */
2109 		defrouter_select_fib(ifp->if_fib);
2110 	}
2111 }
2112 
2113 static void
2114 nd6_slowtimo(void *arg)
2115 {
2116 	struct epoch_tracker et;
2117 	CURVNET_SET((struct vnet *) arg);
2118 	struct nd_ifinfo *nd6if;
2119 	struct ifnet *ifp;
2120 
2121 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2122 	    nd6_slowtimo, curvnet);
2123 	NET_EPOCH_ENTER(et);
2124 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2125 		if (ifp->if_afdata[AF_INET6] == NULL)
2126 			continue;
2127 		nd6if = ND_IFINFO(ifp);
2128 		if (nd6if->basereachable && /* already initialized */
2129 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2130 			/*
2131 			 * Since reachable time rarely changes by router
2132 			 * advertisements, we SHOULD insure that a new random
2133 			 * value gets recomputed at least once every few hours.
2134 			 * (RFC 2461, 6.3.4)
2135 			 */
2136 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2137 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2138 		}
2139 	}
2140 	NET_EPOCH_EXIT(et);
2141 	CURVNET_RESTORE();
2142 }
2143 
2144 struct mbuf *
2145 nd6_grab_holdchain(struct llentry *ln)
2146 {
2147 	struct mbuf *chain;
2148 
2149 	LLE_WLOCK_ASSERT(ln);
2150 
2151 	chain = ln->la_hold;
2152 	ln->la_hold = NULL;
2153 	ln->la_numheld = 0;
2154 
2155 	if (ln->ln_state == ND6_LLINFO_STALE) {
2156 		/*
2157 		 * The first time we send a packet to a
2158 		 * neighbor whose entry is STALE, we have
2159 		 * to change the state to DELAY and a sets
2160 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2161 		 * seconds to ensure do neighbor unreachability
2162 		 * detection on expiration.
2163 		 * (RFC 2461 7.3.3)
2164 		 */
2165 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2166 	}
2167 
2168 	return (chain);
2169 }
2170 
2171 int
2172 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2173     struct sockaddr_in6 *dst, struct route *ro)
2174 {
2175 	int error;
2176 	int ip6len;
2177 	struct ip6_hdr *ip6;
2178 	struct m_tag *mtag;
2179 
2180 #ifdef MAC
2181 	mac_netinet6_nd6_send(ifp, m);
2182 #endif
2183 
2184 	/*
2185 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2186 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2187 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2188 	 * to be diverted to user space.  When re-injected into the kernel,
2189 	 * send_output() will directly dispatch them to the outgoing interface.
2190 	 */
2191 	if (send_sendso_input_hook != NULL) {
2192 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2193 		if (mtag != NULL) {
2194 			ip6 = mtod(m, struct ip6_hdr *);
2195 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2196 			/* Use the SEND socket */
2197 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2198 			    ip6len);
2199 			/* -1 == no app on SEND socket */
2200 			if (error == 0 || error != -1)
2201 			    return (error);
2202 		}
2203 	}
2204 
2205 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2206 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2207 	    mtod(m, struct ip6_hdr *));
2208 
2209 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2210 		origifp = ifp;
2211 
2212 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2213 	return (error);
2214 }
2215 
2216 /*
2217  * Lookup link headerfor @sa_dst address. Stores found
2218  * data in @desten buffer. Copy of lle ln_flags can be also
2219  * saved in @pflags if @pflags is non-NULL.
2220  *
2221  * If destination LLE does not exists or lle state modification
2222  * is required, call "slow" version.
2223  *
2224  * Return values:
2225  * - 0 on success (address copied to buffer).
2226  * - EWOULDBLOCK (no local error, but address is still unresolved)
2227  * - other errors (alloc failure, etc)
2228  */
2229 int
2230 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2231     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2232     struct llentry **plle)
2233 {
2234 	struct llentry *ln = NULL;
2235 	const struct sockaddr_in6 *dst6;
2236 
2237 	NET_EPOCH_ASSERT();
2238 
2239 	if (pflags != NULL)
2240 		*pflags = 0;
2241 
2242 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2243 
2244 	/* discard the packet if IPv6 operation is disabled on the interface */
2245 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2246 		m_freem(m);
2247 		return (ENETDOWN); /* better error? */
2248 	}
2249 
2250 	if (m != NULL && m->m_flags & M_MCAST) {
2251 		switch (ifp->if_type) {
2252 		case IFT_ETHER:
2253 		case IFT_L2VLAN:
2254 		case IFT_BRIDGE:
2255 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2256 						 desten);
2257 			return (0);
2258 		default:
2259 			m_freem(m);
2260 			return (EAFNOSUPPORT);
2261 		}
2262 	}
2263 
2264 	int family = gw_flags >> 16;
2265 	int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2266 	ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2267 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2268 		/* Entry found, let's copy lle info */
2269 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2270 		if (pflags != NULL)
2271 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2272 		llentry_provide_feedback(ln);
2273 		if (plle) {
2274 			LLE_ADDREF(ln);
2275 			*plle = ln;
2276 			LLE_WUNLOCK(ln);
2277 		}
2278 		return (0);
2279 	} else if (plle && ln)
2280 		LLE_WUNLOCK(ln);
2281 
2282 	return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2283 }
2284 
2285 /*
2286  * Finds or creates a new llentry for @addr and @family.
2287  * Returns wlocked llentry or NULL.
2288  *
2289  *
2290  * Child LLEs.
2291  *
2292  * Do not have their own state machine (gets marked as static)
2293  *  settimer bails out for child LLEs just in case.
2294  *
2295  * Locking order: parent lle gets locked first, chen goes the child.
2296  */
2297 static __noinline struct llentry *
2298 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2299 {
2300 	struct llentry *child_lle = NULL;
2301 	struct llentry *lle, *lle_tmp;
2302 
2303 	lle = nd6_alloc(addr, 0, ifp);
2304 	if (lle != NULL && family != AF_INET6) {
2305 		child_lle = nd6_alloc(addr, 0, ifp);
2306 		if (child_lle == NULL) {
2307 			lltable_free_entry(LLTABLE6(ifp), lle);
2308 			return (NULL);
2309 		}
2310 		child_lle->r_family = family;
2311 		child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2312 		child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2313 	}
2314 
2315 	if (lle == NULL) {
2316 		char ip6buf[INET6_ADDRSTRLEN];
2317 		log(LOG_DEBUG,
2318 		    "nd6_get_llentry: can't allocate llinfo for %s "
2319 		    "(ln=%p)\n",
2320 		    ip6_sprintf(ip6buf, addr), lle);
2321 		return (NULL);
2322 	}
2323 
2324 	IF_AFDATA_WLOCK(ifp);
2325 	LLE_WLOCK(lle);
2326 	/* Prefer any existing entry over newly-created one */
2327 	lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2328 	if (lle_tmp == NULL)
2329 		lltable_link_entry(LLTABLE6(ifp), lle);
2330 	else {
2331 		lltable_free_entry(LLTABLE6(ifp), lle);
2332 		lle = lle_tmp;
2333 	}
2334 	if (child_lle != NULL) {
2335 		/* Check if child lle for the same family exists */
2336 		lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2337 		LLE_WLOCK(child_lle);
2338 		if (lle_tmp == NULL) {
2339 			/* Attach */
2340 			lltable_link_child_entry(lle, child_lle);
2341 		} else {
2342 			/* child lle already exists, free newly-created one */
2343 			lltable_free_entry(LLTABLE6(ifp), child_lle);
2344 			child_lle = lle_tmp;
2345 		}
2346 		LLE_WUNLOCK(lle);
2347 		lle = child_lle;
2348 	}
2349 	IF_AFDATA_WUNLOCK(ifp);
2350 	return (lle);
2351 }
2352 
2353 /*
2354  * Do L2 address resolution for @sa_dst address. Stores found
2355  * address in @desten buffer. Copy of lle ln_flags can be also
2356  * saved in @pflags if @pflags is non-NULL.
2357  *
2358  * Heavy version.
2359  * Function assume that destination LLE does not exist,
2360  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2361  *
2362  * Set noinline to be dtrace-friendly
2363  */
2364 static __noinline int
2365 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2366     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2367     struct llentry **plle)
2368 {
2369 	struct llentry *lle = NULL;
2370 	struct in6_addr *psrc, src;
2371 	int send_ns, ll_len;
2372 	char *lladdr;
2373 
2374 	NET_EPOCH_ASSERT();
2375 
2376 	/*
2377 	 * Address resolution or Neighbor Unreachability Detection
2378 	 * for the next hop.
2379 	 * At this point, the destination of the packet must be a unicast
2380 	 * or an anycast address(i.e. not a multicast).
2381 	 */
2382 	lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2383 	if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2384 		/*
2385 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2386 		 * the condition below is not very efficient.  But we believe
2387 		 * it is tolerable, because this should be a rare case.
2388 		 */
2389 		lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2390 	}
2391 
2392 	if (lle == NULL) {
2393 		m_freem(m);
2394 		return (ENOBUFS);
2395 	}
2396 
2397 	LLE_WLOCK_ASSERT(lle);
2398 
2399 	/*
2400 	 * The first time we send a packet to a neighbor whose entry is
2401 	 * STALE, we have to change the state to DELAY and a sets a timer to
2402 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2403 	 * neighbor unreachability detection on expiration.
2404 	 * (RFC 2461 7.3.3)
2405 	 */
2406 	if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2407 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2408 
2409 	/*
2410 	 * If the neighbor cache entry has a state other than INCOMPLETE
2411 	 * (i.e. its link-layer address is already resolved), just
2412 	 * send the packet.
2413 	 */
2414 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2415 		if (flags & LLE_ADDRONLY) {
2416 			lladdr = lle->ll_addr;
2417 			ll_len = ifp->if_addrlen;
2418 		} else {
2419 			lladdr = lle->r_linkdata;
2420 			ll_len = lle->r_hdrlen;
2421 		}
2422 		bcopy(lladdr, desten, ll_len);
2423 		if (pflags != NULL)
2424 			*pflags = lle->la_flags;
2425 		if (plle) {
2426 			LLE_ADDREF(lle);
2427 			*plle = lle;
2428 		}
2429 		LLE_WUNLOCK(lle);
2430 		return (0);
2431 	}
2432 
2433 	/*
2434 	 * There is a neighbor cache entry, but no ethernet address
2435 	 * response yet.  Append this latest packet to the end of the
2436 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2437 	 * the oldest packet in the queue will be removed.
2438 	 */
2439 	if (m != NULL) {
2440 		size_t dropped;
2441 
2442 		dropped = lltable_append_entry_queue(lle, m, V_nd6_maxqueuelen);
2443 		ICMP6STAT_ADD(icp6s_dropped, dropped);
2444 	}
2445 
2446 	/*
2447 	 * If there has been no NS for the neighbor after entering the
2448 	 * INCOMPLETE state, send the first solicitation.
2449 	 * Note that for newly-created lle la_asked will be 0,
2450 	 * so we will transition from ND6_LLINFO_NOSTATE to
2451 	 * ND6_LLINFO_INCOMPLETE state here.
2452 	 */
2453 	psrc = NULL;
2454 	send_ns = 0;
2455 
2456 	/* If we have child lle, switch to the parent to send NS */
2457 	if (lle->la_flags & LLE_CHILD) {
2458 		struct llentry *lle_parent = lle->lle_parent;
2459 		LLE_WUNLOCK(lle);
2460 		lle = lle_parent;
2461 		LLE_WLOCK(lle);
2462 	}
2463 	if (lle->la_asked == 0) {
2464 		lle->la_asked++;
2465 		send_ns = 1;
2466 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2467 
2468 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2469 	}
2470 	LLE_WUNLOCK(lle);
2471 	if (send_ns != 0)
2472 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2473 
2474 	return (EWOULDBLOCK);
2475 }
2476 
2477 /*
2478  * Do L2 address resolution for @sa_dst address. Stores found
2479  * address in @desten buffer. Copy of lle ln_flags can be also
2480  * saved in @pflags if @pflags is non-NULL.
2481  *
2482  * Return values:
2483  * - 0 on success (address copied to buffer).
2484  * - EWOULDBLOCK (no local error, but address is still unresolved)
2485  * - other errors (alloc failure, etc)
2486  */
2487 int
2488 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2489     char *desten, uint32_t *pflags)
2490 {
2491 	int error;
2492 
2493 	flags |= LLE_ADDRONLY;
2494 	error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2495 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2496 	return (error);
2497 }
2498 
2499 int
2500 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2501 {
2502 	struct mbuf *m, *m_head;
2503 	struct sockaddr_in6 dst6;
2504 	int error = 0;
2505 
2506 	NET_EPOCH_ASSERT();
2507 
2508 	struct route_in6 ro = {
2509 		.ro_prepend = lle->r_linkdata,
2510 		.ro_plen = lle->r_hdrlen,
2511 	};
2512 
2513 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2514 	m_head = chain;
2515 
2516 	while (m_head) {
2517 		m = m_head;
2518 		m_head = m_head->m_nextpkt;
2519 		m->m_nextpkt = NULL;
2520 		error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2521 	}
2522 
2523 	/*
2524 	 * XXX
2525 	 * note that intermediate errors are blindly ignored
2526 	 */
2527 	return (error);
2528 }
2529 
2530 __noinline void
2531 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2532 {
2533 	struct llentry *child_lle;
2534 	struct mbuf *chain;
2535 
2536 	NET_EPOCH_ASSERT();
2537 
2538 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2539 		LLE_WLOCK(child_lle);
2540 		chain = nd6_grab_holdchain(child_lle);
2541 		LLE_WUNLOCK(child_lle);
2542 		nd6_flush_holdchain(ifp, child_lle, chain);
2543 	}
2544 }
2545 
2546 static int
2547 nd6_need_cache(struct ifnet *ifp)
2548 {
2549 	/*
2550 	 * XXX: we currently do not make neighbor cache on any interface
2551 	 * other than Ethernet and GIF.
2552 	 *
2553 	 * RFC2893 says:
2554 	 * - unidirectional tunnels needs no ND
2555 	 */
2556 	switch (ifp->if_type) {
2557 	case IFT_ETHER:
2558 	case IFT_IEEE1394:
2559 	case IFT_L2VLAN:
2560 	case IFT_INFINIBAND:
2561 	case IFT_BRIDGE:
2562 	case IFT_PROPVIRTUAL:
2563 		return (1);
2564 	default:
2565 		return (0);
2566 	}
2567 }
2568 
2569 /*
2570  * Add pernament ND6 link-layer record for given
2571  * interface address.
2572  *
2573  * Very similar to IPv4 arp_ifinit(), but:
2574  * 1) IPv6 DAD is performed in different place
2575  * 2) It is called by IPv6 protocol stack in contrast to
2576  * arp_ifinit() which is typically called in SIOCSIFADDR
2577  * driver ioctl handler.
2578  *
2579  */
2580 int
2581 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2582 {
2583 	struct ifnet *ifp;
2584 	struct llentry *ln, *ln_tmp;
2585 	struct sockaddr *dst;
2586 
2587 	ifp = ia->ia_ifa.ifa_ifp;
2588 	if (nd6_need_cache(ifp) == 0)
2589 		return (0);
2590 
2591 	dst = (struct sockaddr *)&ia->ia_addr;
2592 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2593 	if (ln == NULL)
2594 		return (ENOBUFS);
2595 
2596 	IF_AFDATA_WLOCK(ifp);
2597 	LLE_WLOCK(ln);
2598 	/* Unlink any entry if exists */
2599 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2600 	if (ln_tmp != NULL)
2601 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2602 	lltable_link_entry(LLTABLE6(ifp), ln);
2603 	IF_AFDATA_WUNLOCK(ifp);
2604 
2605 	if (ln_tmp != NULL)
2606 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2607 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2608 
2609 	LLE_WUNLOCK(ln);
2610 	if (ln_tmp != NULL)
2611 		llentry_free(ln_tmp);
2612 
2613 	return (0);
2614 }
2615 
2616 /*
2617  * Removes either all lle entries for given @ia, or lle
2618  * corresponding to @ia address.
2619  */
2620 void
2621 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2622 {
2623 	struct sockaddr_in6 mask, addr;
2624 	struct sockaddr *saddr, *smask;
2625 	struct ifnet *ifp;
2626 
2627 	ifp = ia->ia_ifa.ifa_ifp;
2628 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2629 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2630 	saddr = (struct sockaddr *)&addr;
2631 	smask = (struct sockaddr *)&mask;
2632 
2633 	if (all != 0)
2634 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2635 	else
2636 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2637 }
2638 
2639 static int
2640 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2641 {
2642 	struct in6_prefix p;
2643 	struct sockaddr_in6 s6;
2644 	struct nd_prefix *pr;
2645 	struct nd_pfxrouter *pfr;
2646 	time_t maxexpire;
2647 	int error;
2648 	char ip6buf[INET6_ADDRSTRLEN];
2649 
2650 	if (req->newptr)
2651 		return (EPERM);
2652 
2653 	error = sysctl_wire_old_buffer(req, 0);
2654 	if (error != 0)
2655 		return (error);
2656 
2657 	bzero(&p, sizeof(p));
2658 	p.origin = PR_ORIG_RA;
2659 	bzero(&s6, sizeof(s6));
2660 	s6.sin6_family = AF_INET6;
2661 	s6.sin6_len = sizeof(s6);
2662 
2663 	ND6_RLOCK();
2664 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2665 		if (!pr->ndpr_raf_ra_derived)
2666 			continue;
2667 		p.prefix = pr->ndpr_prefix;
2668 		if (sa6_recoverscope(&p.prefix)) {
2669 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2670 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2671 			/* XXX: press on... */
2672 		}
2673 		p.raflags = pr->ndpr_raf;
2674 		p.prefixlen = pr->ndpr_plen;
2675 		p.vltime = pr->ndpr_vltime;
2676 		p.pltime = pr->ndpr_pltime;
2677 		p.if_index = pr->ndpr_ifp->if_index;
2678 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2679 			p.expire = 0;
2680 		else {
2681 			/* XXX: we assume time_t is signed. */
2682 			maxexpire = (-1) &
2683 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2684 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2685 				p.expire = pr->ndpr_lastupdate +
2686 				    pr->ndpr_vltime +
2687 				    (time_second - time_uptime);
2688 			else
2689 				p.expire = maxexpire;
2690 		}
2691 		p.refcnt = pr->ndpr_addrcnt;
2692 		p.flags = pr->ndpr_stateflags;
2693 		p.advrtrs = 0;
2694 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2695 			p.advrtrs++;
2696 		error = SYSCTL_OUT(req, &p, sizeof(p));
2697 		if (error != 0)
2698 			break;
2699 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2700 			s6.sin6_addr = pfr->router->rtaddr;
2701 			if (sa6_recoverscope(&s6))
2702 				log(LOG_ERR,
2703 				    "scope error in prefix list (%s)\n",
2704 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2705 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2706 			if (error != 0)
2707 				goto out;
2708 		}
2709 	}
2710 out:
2711 	ND6_RUNLOCK();
2712 	return (error);
2713 }
2714 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2715 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2716 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2717 	"NDP prefix list");
2718 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2719 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2720 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2721 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2722