xref: /netbsd-src/sys/netinet6/nd6.c (revision 4d342c046e3288fb5a1edcd33cfec48c41c80664)
1 /*	$NetBSD: nd6.c,v 1.274 2020/09/15 10:05:36 roy Exp $	*/
2 /*	$KAME: nd6.c,v 1.279 2002/06/08 11:16:51 itojun Exp $	*/
3 
4 /*
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 #include <sys/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: nd6.c,v 1.274 2020/09/15 10:05:36 roy Exp $");
35 
36 #ifdef _KERNEL_OPT
37 #include "opt_compat_netbsd.h"
38 #include "opt_net_mpsafe.h"
39 #endif
40 
41 #include "bridge.h"
42 #include "carp.h"
43 
44 #include <sys/param.h>
45 #include <sys/systm.h>
46 #include <sys/callout.h>
47 #include <sys/kmem.h>
48 #include <sys/mbuf.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sockio.h>
52 #include <sys/time.h>
53 #include <sys/kernel.h>
54 #include <sys/errno.h>
55 #include <sys/ioctl.h>
56 #include <sys/syslog.h>
57 #include <sys/queue.h>
58 #include <sys/cprng.h>
59 #include <sys/workqueue.h>
60 
61 #include <net/if.h>
62 #include <net/if_dl.h>
63 #include <net/if_llatbl.h>
64 #include <net/if_types.h>
65 #include <net/nd.h>
66 #include <net/route.h>
67 #include <net/if_ether.h>
68 #include <net/if_arc.h>
69 
70 #include <netinet/in.h>
71 #include <netinet6/in6_var.h>
72 #include <netinet/ip6.h>
73 #include <netinet6/ip6_var.h>
74 #include <netinet6/scope6_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet6/in6_ifattach.h>
77 #include <netinet/icmp6.h>
78 #include <netinet6/icmp6_private.h>
79 
80 #ifdef COMPAT_90
81 #include <compat/netinet6/in6_var.h>
82 #include <compat/netinet6/nd6.h>
83 #endif
84 
85 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
86 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
87 
88 /* timer values */
89 int	nd6_prune	= 1;	/* walk list every 1 seconds */
90 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
91 
92 /* preventing too many loops in ND option parsing */
93 int nd6_maxndopt = 10;	/* max # of ND options allowed */
94 
95 #ifdef ND6_DEBUG
96 int nd6_debug = 1;
97 #else
98 int nd6_debug = 0;
99 #endif
100 
101 krwlock_t nd6_lock __cacheline_aligned;
102 
103 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
104 
105 static void nd6_slowtimo(void *);
106 static void nd6_free(struct llentry *, int);
107 static bool nd6_nud_enabled(struct ifnet *);
108 static unsigned int nd6_llinfo_reachable(struct ifnet *);
109 static unsigned int nd6_llinfo_retrans(struct ifnet *);
110 static union l3addr *nd6_llinfo_holdsrc(struct llentry *, union l3addr *);
111 static void nd6_llinfo_output(struct ifnet *, const union l3addr *,
112     const union l3addr *, const uint8_t *, const union l3addr *);
113 static void nd6_llinfo_missed(struct ifnet *, const union l3addr *,
114     int16_t, struct mbuf *);
115 static void nd6_timer(void *);
116 static void nd6_timer_work(struct work *, void *);
117 static struct nd_opt_hdr *nd6_option(union nd_opts *);
118 
119 static callout_t nd6_slowtimo_ch;
120 static callout_t nd6_timer_ch;
121 static struct workqueue	*nd6_timer_wq;
122 static struct work	nd6_timer_wk;
123 
124 struct nd_domain nd6_nd_domain = {
125 	.nd_family = AF_INET6,
126 	.nd_delay = 5,		/* delay first probe time 5 second */
127 	.nd_mmaxtries = 3,	/* maximum unicast query */
128 	.nd_umaxtries = 3,	/* maximum multicast query */
129 	.nd_retransmultiple = BACKOFF_MULTIPLE,
130 	.nd_maxretrans = MAX_RETRANS_TIMER,
131 	.nd_maxnudhint = 0,	/* max # of subsequent upper layer hints */
132 	.nd_maxqueuelen = 1,	/* max # of packets in unresolved ND entries */
133 	.nd_nud_enabled = nd6_nud_enabled,
134 	.nd_reachable = nd6_llinfo_reachable,
135 	.nd_retrans = nd6_llinfo_retrans,
136 	.nd_holdsrc = nd6_llinfo_holdsrc,
137 	.nd_output = nd6_llinfo_output,
138 	.nd_missed = nd6_llinfo_missed,
139 	.nd_free = nd6_free,
140 };
141 
142 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
143 
144 void
145 nd6_init(void)
146 {
147 	int error;
148 
149 	nd_attach_domain(&nd6_nd_domain);
150 	nd6_nbr_init();
151 
152 	rw_init(&nd6_lock);
153 
154 	callout_init(&nd6_slowtimo_ch, CALLOUT_MPSAFE);
155 	callout_init(&nd6_timer_ch, CALLOUT_MPSAFE);
156 
157 	error = workqueue_create(&nd6_timer_wq, "nd6_timer",
158 	    nd6_timer_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
159 	if (error)
160 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
161 
162 	/* start timer */
163 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
164 	    nd6_slowtimo, NULL);
165 	callout_reset(&nd6_timer_ch, hz, nd6_timer, NULL);
166 }
167 
168 struct nd_kifinfo *
169 nd6_ifattach(struct ifnet *ifp)
170 {
171 	struct nd_kifinfo *nd;
172 
173 	nd = kmem_zalloc(sizeof(*nd), KM_SLEEP);
174 
175 	nd->chlim = IPV6_DEFHLIM;
176 	nd->basereachable = REACHABLE_TIME;
177 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
178 	nd->retrans = RETRANS_TIMER;
179 
180 	nd->flags = ND6_IFF_PERFORMNUD;
181 
182 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
183 	 * A bridge interface should not have ND6_IFF_AUTO_LINKLOCAL
184 	 * because one of its members should. */
185 	if ((ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
186 	    (ifp->if_flags & IFF_LOOPBACK))
187 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
188 
189 	return nd;
190 }
191 
192 void
193 nd6_ifdetach(struct ifnet *ifp, struct in6_ifextra *ext)
194 {
195 
196 	/* Ensure all IPv6 addresses are purged before calling nd6_purge */
197 	if_purgeaddrs(ifp, AF_INET6, in6_purgeaddr);
198 	nd6_purge(ifp, ext);
199 	kmem_free(ext->nd_ifinfo, sizeof(struct nd_kifinfo));
200 }
201 
202 void
203 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
204 {
205 
206 	memset(ndopts, 0, sizeof(*ndopts));
207 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
208 	ndopts->nd_opts_last
209 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
210 
211 	if (icmp6len == 0) {
212 		ndopts->nd_opts_done = 1;
213 		ndopts->nd_opts_search = NULL;
214 	}
215 }
216 
217 /*
218  * Take one ND option.
219  */
220 static struct nd_opt_hdr *
221 nd6_option(union nd_opts *ndopts)
222 {
223 	struct nd_opt_hdr *nd_opt;
224 	int olen;
225 
226 	KASSERT(ndopts != NULL);
227 	KASSERT(ndopts->nd_opts_last != NULL);
228 
229 	if (ndopts->nd_opts_search == NULL)
230 		return NULL;
231 	if (ndopts->nd_opts_done)
232 		return NULL;
233 
234 	nd_opt = ndopts->nd_opts_search;
235 
236 	/* make sure nd_opt_len is inside the buffer */
237 	if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) {
238 		memset(ndopts, 0, sizeof(*ndopts));
239 		return NULL;
240 	}
241 
242 	olen = nd_opt->nd_opt_len << 3;
243 	if (olen == 0) {
244 		/*
245 		 * Message validation requires that all included
246 		 * options have a length that is greater than zero.
247 		 */
248 		memset(ndopts, 0, sizeof(*ndopts));
249 		return NULL;
250 	}
251 
252 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen);
253 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
254 		/* option overruns the end of buffer, invalid */
255 		memset(ndopts, 0, sizeof(*ndopts));
256 		return NULL;
257 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
258 		/* reached the end of options chain */
259 		ndopts->nd_opts_done = 1;
260 		ndopts->nd_opts_search = NULL;
261 	}
262 	return nd_opt;
263 }
264 
265 /*
266  * Parse multiple ND options.
267  * This function is much easier to use, for ND routines that do not need
268  * multiple options of the same type.
269  */
270 int
271 nd6_options(union nd_opts *ndopts)
272 {
273 	struct nd_opt_hdr *nd_opt;
274 	int i = 0;
275 
276 	KASSERT(ndopts != NULL);
277 	KASSERT(ndopts->nd_opts_last != NULL);
278 
279 	if (ndopts->nd_opts_search == NULL)
280 		return 0;
281 
282 	while (1) {
283 		nd_opt = nd6_option(ndopts);
284 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
285 			/*
286 			 * Message validation requires that all included
287 			 * options have a length that is greater than zero.
288 			 */
289 			ICMP6_STATINC(ICMP6_STAT_ND_BADOPT);
290 			memset(ndopts, 0, sizeof(*ndopts));
291 			return -1;
292 		}
293 
294 		if (nd_opt == NULL)
295 			goto skip1;
296 
297 		switch (nd_opt->nd_opt_type) {
298 		case ND_OPT_SOURCE_LINKADDR:
299 		case ND_OPT_TARGET_LINKADDR:
300 		case ND_OPT_MTU:
301 		case ND_OPT_REDIRECTED_HEADER:
302 		case ND_OPT_NONCE:
303 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
304 				nd6log(LOG_INFO,
305 				    "duplicated ND6 option found (type=%d)\n",
306 				    nd_opt->nd_opt_type);
307 				/* XXX bark? */
308 			} else {
309 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
310 					= nd_opt;
311 			}
312 			break;
313 		case ND_OPT_PREFIX_INFORMATION:
314 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
315 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
316 					= nd_opt;
317 			}
318 			ndopts->nd_opts_pi_end =
319 				(struct nd_opt_prefix_info *)nd_opt;
320 			break;
321 		default:
322 			/*
323 			 * Unknown options must be silently ignored,
324 			 * to accommodate future extension to the protocol.
325 			 */
326 			nd6log(LOG_DEBUG,
327 			    "nd6_options: unsupported option %d - "
328 			    "option ignored\n", nd_opt->nd_opt_type);
329 		}
330 
331 skip1:
332 		i++;
333 		if (i > nd6_maxndopt) {
334 			ICMP6_STATINC(ICMP6_STAT_ND_TOOMANYOPT);
335 			nd6log(LOG_INFO, "too many loop in nd opt\n");
336 			break;
337 		}
338 
339 		if (ndopts->nd_opts_done)
340 			break;
341 	}
342 
343 	return 0;
344 }
345 
346 /*
347  * Gets source address of the first packet in hold queue
348  * and stores it in @src.
349  * Returns pointer to @src (if hold queue is not empty) or NULL.
350  */
351 static struct in6_addr *
352 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
353 {
354 	struct ip6_hdr *hip6;
355 
356 	if (ln == NULL || ln->ln_hold == NULL)
357 		return NULL;
358 
359 	/*
360 	 * assuming every packet in ln_hold has the same IP header
361 	 */
362 	hip6 = mtod(ln->ln_hold, struct ip6_hdr *);
363 	/* XXX pullup? */
364 	if (sizeof(*hip6) < ln->ln_hold->m_len)
365 		*src = hip6->ip6_src;
366 	else
367 		src = NULL;
368 
369 	return src;
370 }
371 
372 static union l3addr *
373 nd6_llinfo_holdsrc(struct llentry *ln, union l3addr *src)
374 {
375 
376 	if (nd6_llinfo_get_holdsrc(ln, &src->addr6) == NULL)
377 		return NULL;
378 	return src;
379 }
380 
381 static void
382 nd6_llinfo_output(struct ifnet *ifp, const union l3addr *daddr,
383     const union l3addr *taddr, __unused const uint8_t *tlladdr,
384     const union l3addr *hsrc)
385 {
386 
387 	nd6_ns_output(ifp, &daddr->addr6, &taddr->addr6,
388 	    &hsrc->addr6, NULL);
389 }
390 
391 static bool
392 nd6_nud_enabled(struct ifnet *ifp)
393 {
394 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
395 
396 	return ndi->flags & ND6_IFF_PERFORMNUD;
397 }
398 
399 static unsigned int
400 nd6_llinfo_reachable(struct ifnet *ifp)
401 {
402 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
403 
404 	return ndi->reachable;
405 }
406 
407 static unsigned int
408 nd6_llinfo_retrans(struct ifnet *ifp)
409 {
410 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
411 
412 	return ndi->retrans;
413 }
414 
415 static void
416 nd6_llinfo_missed(struct ifnet *ifp, const union l3addr *taddr,
417     int16_t type, struct mbuf *m)
418 {
419 	struct in6_addr mdaddr6 = zeroin6_addr;
420 	struct sockaddr_in6 dsin6, tsin6;
421 	struct sockaddr *sa;
422 
423 	if (m != NULL) {
424 		if (type == ND_LLINFO_PROBE) {
425 			struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
426 
427 			/* XXX pullup? */
428 			if (sizeof(*ip6) < m->m_len)
429 				mdaddr6 = ip6->ip6_src;
430 			m_freem(m);
431 		} else
432 			icmp6_error2(m, ICMP6_DST_UNREACH,
433 			    ICMP6_DST_UNREACH_ADDR, 0, ifp, &mdaddr6);
434 	}
435 	if (!IN6_IS_ADDR_UNSPECIFIED(&mdaddr6)) {
436 		sockaddr_in6_init(&dsin6, &mdaddr6, 0, 0, 0);
437 		sa = sin6tosa(&dsin6);
438 	} else
439 		sa = NULL;
440 
441 	sockaddr_in6_init(&tsin6, &taddr->addr6, 0, 0, 0);
442 	rt_clonedmsg(RTM_MISS, sa, sin6tosa(&tsin6), NULL, ifp);
443 }
444 
445 /*
446  * ND6 timer routine to expire default route list and prefix list
447  */
448 static void
449 nd6_timer_work(struct work *wk, void *arg)
450 {
451 	struct in6_ifaddr *ia6, *nia6;
452 	int s, bound;
453 	struct psref psref;
454 
455 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
456 	    nd6_timer, NULL);
457 
458 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
459 
460 	/* expire interface addresses */
461 	bound = curlwp_bind();
462 	s = pserialize_read_enter();
463 	for (ia6 = IN6_ADDRLIST_READER_FIRST(); ia6; ia6 = nia6) {
464 		nia6 = IN6_ADDRLIST_READER_NEXT(ia6);
465 
466 		ia6_acquire(ia6, &psref);
467 		pserialize_read_exit(s);
468 
469 		/* check address lifetime */
470 		if (IFA6_IS_INVALID(ia6)) {
471 			struct ifnet *ifp;
472 
473 			ifp = ia6->ia_ifa.ifa_ifp;
474 			IFNET_LOCK(ifp);
475 			/*
476 			 * Need to take the lock first to prevent if_detach
477 			 * from running in6_purgeaddr concurrently.
478 			 */
479 			if (!if_is_deactivated(ifp)) {
480 				ia6_release(ia6, &psref);
481 				in6_purgeaddr(&ia6->ia_ifa);
482 			} else {
483 				/*
484 				 * ifp is being destroyed, ia6 will be destroyed
485 				 * by if_detach.
486 				 */
487 				ia6_release(ia6, &psref);
488 			}
489 			ia6 = NULL;
490 			IFNET_UNLOCK(ifp);
491 		} else if (IFA6_IS_DEPRECATED(ia6)) {
492 			int oldflags = ia6->ia6_flags;
493 
494 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
495 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
496 				rt_addrmsg(RTM_NEWADDR, (struct ifaddr *)ia6);
497 			}
498 		} else {
499 			/*
500 			 * A new RA might have made a deprecated address
501 			 * preferred.
502 			 */
503 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
504 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
505 				rt_addrmsg(RTM_NEWADDR, (struct ifaddr *)ia6);
506 			}
507 		}
508 		s = pserialize_read_enter();
509 		ia6_release(ia6, &psref);
510 	}
511 	pserialize_read_exit(s);
512 	curlwp_bindx(bound);
513 
514 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
515 }
516 
517 static void
518 nd6_timer(void *ignored_arg)
519 {
520 
521 	workqueue_enqueue(nd6_timer_wq, &nd6_timer_wk, NULL);
522 }
523 
524 /*
525  * Nuke neighbor cache/prefix/default router management table, right before
526  * ifp goes away.
527  */
528 void
529 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
530 {
531 
532 	/*
533 	 * During detach, the ND info might be already removed, but
534 	 * then is explitly passed as argument.
535 	 * Otherwise get it from ifp->if_afdata.
536 	 */
537 	if (ext == NULL)
538 		ext = ifp->if_afdata[AF_INET6];
539 	if (ext == NULL)
540 		return;
541 
542 	/*
543 	 * We may not need to nuke the neighbor cache entries here
544 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
545 	 * nd6_purge() is invoked by in6_ifdetach() which is called
546 	 * from if_detach() where everything gets purged. However
547 	 * in6_ifdetach is directly called from vlan(4), so we still
548 	 * need to purge entries here.
549 	 */
550 	if (ext->lltable != NULL)
551 		lltable_purge_entries(ext->lltable);
552 }
553 
554 struct llentry *
555 nd6_lookup(const struct in6_addr *addr6, const struct ifnet *ifp, bool wlock)
556 {
557 	struct sockaddr_in6 sin6;
558 	struct llentry *ln;
559 
560 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
561 
562 	IF_AFDATA_RLOCK(ifp);
563 	ln = lla_lookup(LLTABLE6(ifp), wlock ? LLE_EXCLUSIVE : 0,
564 	    sin6tosa(&sin6));
565 	IF_AFDATA_RUNLOCK(ifp);
566 
567 	return ln;
568 }
569 
570 struct llentry *
571 nd6_create(const struct in6_addr *addr6, const struct ifnet *ifp)
572 {
573 	struct sockaddr_in6 sin6;
574 	struct llentry *ln;
575 	struct rtentry *rt;
576 
577 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
578 	rt = rtalloc1(sin6tosa(&sin6), 0);
579 
580 	IF_AFDATA_WLOCK(ifp);
581 	ln = lla_create(LLTABLE6(ifp), LLE_EXCLUSIVE, sin6tosa(&sin6), rt);
582 	IF_AFDATA_WUNLOCK(ifp);
583 
584 	if (rt != NULL)
585 		rt_unref(rt);
586 	if (ln != NULL)
587 		ln->ln_state = ND_LLINFO_NOSTATE;
588 
589 	return ln;
590 }
591 
592 /*
593  * Test whether a given IPv6 address is a neighbor or not, ignoring
594  * the actual neighbor cache.  The neighbor cache is ignored in order
595  * to not reenter the routing code from within itself.
596  */
597 static int
598 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
599 {
600 	struct ifaddr *dstaddr;
601 	int s;
602 
603 	/*
604 	 * A link-local address is always a neighbor.
605 	 * XXX: a link does not necessarily specify a single interface.
606 	 */
607 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
608 		struct sockaddr_in6 sin6_copy;
609 		u_int32_t zone;
610 
611 		/*
612 		 * We need sin6_copy since sa6_recoverscope() may modify the
613 		 * content (XXX).
614 		 */
615 		sin6_copy = *addr;
616 		if (sa6_recoverscope(&sin6_copy))
617 			return 0; /* XXX: should be impossible */
618 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
619 			return 0;
620 		if (sin6_copy.sin6_scope_id == zone)
621 			return 1;
622 		else
623 			return 0;
624 	}
625 
626 	/*
627 	 * If the address is assigned on the node of the other side of
628 	 * a p2p interface, the address should be a neighbor.
629 	 */
630 	s = pserialize_read_enter();
631 	dstaddr = ifa_ifwithdstaddr(sin6tocsa(addr));
632 	if (dstaddr != NULL) {
633 		if (dstaddr->ifa_ifp == ifp) {
634 			pserialize_read_exit(s);
635 			return 1;
636 		}
637 	}
638 	pserialize_read_exit(s);
639 
640 	return 0;
641 }
642 
643 /*
644  * Detect if a given IPv6 address identifies a neighbor on a given link.
645  * XXX: should take care of the destination of a p2p link?
646  */
647 int
648 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
649 {
650 	struct llentry *ln;
651 	struct rtentry *rt;
652 
653 	/*
654 	 * A link-local address is always a neighbor.
655 	 * XXX: a link does not necessarily specify a single interface.
656 	 */
657 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
658 		struct sockaddr_in6 sin6_copy;
659 		u_int32_t zone;
660 
661 		/*
662 		 * We need sin6_copy since sa6_recoverscope() may modify the
663 		 * content (XXX).
664 		 */
665 		sin6_copy = *addr;
666 		if (sa6_recoverscope(&sin6_copy))
667 			return 0; /* XXX: should be impossible */
668 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
669 			return 0;
670 		if (sin6_copy.sin6_scope_id == zone)
671 			return 1;
672 		else
673 			return 0;
674 	}
675 
676 	if (nd6_is_new_addr_neighbor(addr, ifp))
677 		return 1;
678 
679 	/*
680 	 * Even if the address matches none of our addresses, it might be
681 	 * in the neighbor cache or a connected route.
682 	 */
683 	ln = nd6_lookup(&addr->sin6_addr, ifp, false);
684 	if (ln != NULL) {
685 		LLE_RUNLOCK(ln);
686 		return 1;
687 	}
688 
689 	rt = rtalloc1(sin6tocsa(addr), 0);
690 	if (rt == NULL)
691 		return 0;
692 
693 	if ((rt->rt_flags & RTF_CONNECTED) && (rt->rt_ifp == ifp
694 #if NBRIDGE > 0
695 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
696 #endif
697 #if NCARP > 0
698 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
699 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
700 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
701 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
702 #endif
703 	    )) {
704 		rt_unref(rt);
705 		return 1;
706 	}
707 	rt_unref(rt);
708 
709 	return 0;
710 }
711 
712 /*
713  * Free an nd6 llinfo entry.
714  * Since the function would cause significant changes in the kernel, DO NOT
715  * make it global, unless you have a strong reason for the change, and are sure
716  * that the change is safe.
717  */
718 static void
719 nd6_free(struct llentry *ln, int gc)
720 {
721 	struct ifnet *ifp;
722 
723 	KASSERT(ln != NULL);
724 	LLE_WLOCK_ASSERT(ln);
725 
726 	/*
727 	 * If the reason for the deletion is just garbage collection,
728 	 * and the neighbor is an active router, do not delete it.
729 	 * Instead, reset the GC timer using the router's lifetime.
730 	 * XXX: the check for ln_state should be redundant,
731 	 *      but we intentionally keep it just in case.
732 	 */
733 	if (!ip6_forwarding && ln->ln_router &&
734 	    ln->ln_state == ND_LLINFO_STALE && gc)
735 	{
736 		nd_set_timer(ln, ND_TIMER_EXPIRE);
737 		LLE_WUNLOCK(ln);
738 		return;
739 	}
740 
741 	ifp = ln->lle_tbl->llt_ifp;
742 
743 	if (ln->la_flags & LLE_VALID || gc) {
744 		struct sockaddr_in6 sin6;
745 		const char *lladdr;
746 
747 		sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
748 		lladdr = ln->la_flags & LLE_VALID ?
749 		    (const char *)&ln->ll_addr : NULL;
750 		rt_clonedmsg(RTM_DELETE, NULL, sin6tosa(&sin6), lladdr, ifp);
751 	}
752 
753 	/*
754 	 * Save to unlock. We still hold an extra reference and will not
755 	 * free(9) in llentry_free() if someone else holds one as well.
756 	 */
757 	LLE_WUNLOCK(ln);
758 	IF_AFDATA_LOCK(ifp);
759 	LLE_WLOCK(ln);
760 
761 	lltable_free_entry(LLTABLE6(ifp), ln);
762 
763 	IF_AFDATA_UNLOCK(ifp);
764 }
765 
766 /*
767  * Upper-layer reachability hint for Neighbor Unreachability Detection.
768  *
769  * XXX cost-effective methods?
770  */
771 void
772 nd6_nud_hint(struct rtentry *rt)
773 {
774 	struct llentry *ln;
775 	struct ifnet *ifp;
776 
777 	if (rt == NULL)
778 		return;
779 
780 	ifp = rt->rt_ifp;
781 	ln = nd6_lookup(&(satocsin6(rt_getkey(rt)))->sin6_addr, ifp, true);
782 	nd_nud_hint(ln);
783 }
784 
785 struct gc_args {
786 	int gc_entries;
787 	const struct in6_addr *skip_in6;
788 };
789 
790 static int
791 nd6_purge_entry(struct lltable *llt, struct llentry *ln, void *farg)
792 {
793 	struct gc_args *args = farg;
794 	int *n = &args->gc_entries;
795 	const struct in6_addr *skip_in6 = args->skip_in6;
796 
797 	if (*n <= 0)
798 		return 0;
799 
800 	if (ND_IS_LLINFO_PERMANENT(ln))
801 		return 0;
802 
803 	if (IN6_ARE_ADDR_EQUAL(&ln->r_l3addr.addr6, skip_in6))
804 		return 0;
805 
806 	LLE_WLOCK(ln);
807 	if (ln->ln_state > ND_LLINFO_INCOMPLETE)
808 		ln->ln_state = ND_LLINFO_STALE;
809 	else
810 		ln->ln_state = ND_LLINFO_PURGE;
811 	nd_set_timer(ln, ND_TIMER_IMMEDIATE);
812 	LLE_WUNLOCK(ln);
813 
814 	(*n)--;
815 	return 0;
816 }
817 
818 static void
819 nd6_gc_neighbors(struct lltable *llt, const struct in6_addr *in6)
820 {
821 
822 	if (ip6_neighborgcthresh >= 0 &&
823 	    lltable_get_entry_count(llt) >= ip6_neighborgcthresh) {
824 		struct gc_args gc_args = {10, in6};
825 		/*
826 		 * XXX entries that are "less recently used" should be
827 		 * freed first.
828 		 */
829 		lltable_foreach_lle(llt, nd6_purge_entry, &gc_args);
830 	}
831 }
832 
833 void
834 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
835 {
836 	struct sockaddr *gate = rt->rt_gateway;
837 	struct ifnet *ifp = rt->rt_ifp;
838 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
839 	struct ifaddr *ifa;
840 
841 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
842 
843 	if (req == RTM_LLINFO_UPD) {
844 		int rc;
845 		struct in6_addr *in6;
846 		struct in6_addr in6_all;
847 		int anycast;
848 
849 		if ((ifa = info->rti_ifa) == NULL)
850 			return;
851 
852 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
853 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
854 
855 		in6_all = in6addr_linklocal_allnodes;
856 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
857 			log(LOG_ERR, "%s: failed to set scope %s "
858 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
859 			return;
860 		}
861 
862 		/* XXX don't set Override for proxy addresses */
863 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
864 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
865 #if 0
866 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
867 #endif
868 		    , 1, NULL);
869 		return;
870 	}
871 
872 	if ((rt->rt_flags & RTF_GATEWAY) != 0) {
873 		if (req != RTM_ADD)
874 			return;
875 		/*
876 		 * linklayers with particular MTU limitation.
877 		 */
878 		switch(ifp->if_type) {
879 #if NARCNET > 0
880 		case IFT_ARCNET:
881 			if (rt->rt_rmx.rmx_mtu > ARC_PHDS_MAXMTU) /* RFC2497 */
882 				rt->rt_rmx.rmx_mtu = ARC_PHDS_MAXMTU;
883 			break;
884 #endif
885 		}
886 		return;
887 	}
888 
889 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
890 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
891 		/*
892 		 * This is probably an interface direct route for a link
893 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
894 		 * We do not need special treatment below for such a route.
895 		 * Moreover, the RTF_LLINFO flag which would be set below
896 		 * would annoy the ndp(8) command.
897 		 */
898 		return;
899 	}
900 
901 	switch (req) {
902 	case RTM_ADD: {
903 		struct psref psref;
904 
905 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
906 		/*
907 		 * There is no backward compatibility :)
908 		 *
909 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
910 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
911 		 *	   rt->rt_flags |= RTF_CLONING;
912 		 */
913 		/* XXX should move to route.c? */
914 		if (rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL)) {
915 			union {
916 				struct sockaddr sa;
917 				struct sockaddr_dl sdl;
918 				struct sockaddr_storage ss;
919 			} u;
920 			/*
921 			 * Case 1: This route should come from a route to
922 			 * interface (RTF_CLONING case) or the route should be
923 			 * treated as on-link but is currently not
924 			 * (RTF_LLINFO && ln == NULL case).
925 			 */
926 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
927 			    ifp->if_index, ifp->if_type,
928 			    NULL, namelen, NULL, addrlen) == NULL) {
929 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
930 				    "failed on %s\n", __func__, __LINE__,
931 				    sizeof(u.ss), if_name(ifp));
932 			}
933 			rt_setgate(rt, &u.sa);
934 			gate = rt->rt_gateway;
935 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
936 			if (gate == NULL) {
937 				log(LOG_ERR,
938 				    "%s: rt_setgate failed on %s\n", __func__,
939 				    if_name(ifp));
940 				break;
941 			}
942 
943 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
944 			if ((rt->rt_flags & RTF_CONNECTED) != 0)
945 				break;
946 		}
947 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
948 		/*
949 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
950 		 * We don't do that here since llinfo is not ready yet.
951 		 *
952 		 * There are also couple of other things to be discussed:
953 		 * - unsolicited NA code needs improvement beforehand
954 		 * - RFC2461 says we MAY send multicast unsolicited NA
955 		 *   (7.2.6 paragraph 4), however, it also says that we
956 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
957 		 *   we don't have anything like it right now.
958 		 *   note that the mechanism needs a mutual agreement
959 		 *   between proxies, which means that we need to implement
960 		 *   a new protocol, or a new kludge.
961 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
962 		 *   we need to check ip6forwarding before sending it.
963 		 *   (or should we allow proxy ND configuration only for
964 		 *   routers?  there's no mention about proxy ND from hosts)
965 		 */
966 #if 0
967 		/* XXX it does not work */
968 		if (rt->rt_flags & RTF_ANNOUNCE)
969 			nd6_na_output(ifp,
970 			      &satocsin6(rt_getkey(rt))->sin6_addr,
971 			      &satocsin6(rt_getkey(rt))->sin6_addr,
972 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
973 			      1, NULL);
974 #endif
975 
976 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
977 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
978 			/*
979 			 * Address resolution isn't necessary for a point to
980 			 * point link, so we can skip this test for a p2p link.
981 			 */
982 			if (gate->sa_family != AF_LINK ||
983 			    gate->sa_len <
984 			    sockaddr_dl_measure(namelen, addrlen)) {
985 				log(LOG_DEBUG,
986 				    "nd6_rtrequest: bad gateway value: %s\n",
987 				    if_name(ifp));
988 				break;
989 			}
990 			satosdl(gate)->sdl_type = ifp->if_type;
991 			satosdl(gate)->sdl_index = ifp->if_index;
992 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
993 		}
994 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
995 
996 		/*
997 		 * When called from rt_ifa_addlocal, we cannot depend on that
998 		 * the address (rt_getkey(rt)) exits in the address list of the
999 		 * interface. So check RTF_LOCAL instead.
1000 		 */
1001 		if (rt->rt_flags & RTF_LOCAL) {
1002 			if (nd6_useloopback)
1003 				rt->rt_ifp = lo0ifp;	/* XXX */
1004 			break;
1005 		}
1006 
1007 		/*
1008 		 * check if rt_getkey(rt) is an address assigned
1009 		 * to the interface.
1010 		 */
1011 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr_psref(ifp,
1012 		    &satocsin6(rt_getkey(rt))->sin6_addr, &psref);
1013 		if (ifa != NULL) {
1014 			if (nd6_useloopback) {
1015 				rt->rt_ifp = lo0ifp;	/* XXX */
1016 				/*
1017 				 * Make sure rt_ifa be equal to the ifaddr
1018 				 * corresponding to the address.
1019 				 * We need this because when we refer
1020 				 * rt_ifa->ia6_flags in ip6_input, we assume
1021 				 * that the rt_ifa points to the address instead
1022 				 * of the loopback address.
1023 				 */
1024 				if (!ISSET(info->rti_flags, RTF_DONTCHANGEIFA)
1025 				    && ifa != rt->rt_ifa)
1026 					rt_replace_ifa(rt, ifa);
1027 			}
1028 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1029 			/* join solicited node multicast for proxy ND */
1030 			if (ifp->if_flags & IFF_MULTICAST) {
1031 				struct in6_addr llsol;
1032 				int error;
1033 
1034 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1035 				llsol.s6_addr32[0] = htonl(0xff020000);
1036 				llsol.s6_addr32[1] = 0;
1037 				llsol.s6_addr32[2] = htonl(1);
1038 				llsol.s6_addr8[12] = 0xff;
1039 				if (in6_setscope(&llsol, ifp, NULL))
1040 					goto out;
1041 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1042 					char ip6buf[INET6_ADDRSTRLEN];
1043 					nd6log(LOG_ERR, "%s: failed to join "
1044 					    "%s (errno=%d)\n", if_name(ifp),
1045 					    IN6_PRINT(ip6buf, &llsol), error);
1046 				}
1047 			}
1048 		}
1049 	out:
1050 		ifa_release(ifa, &psref);
1051 		/*
1052 		 * If we have too many cache entries, initiate immediate
1053 		 * purging for some entries.
1054 		 */
1055 		if (rt->rt_ifp != NULL)
1056 			nd6_gc_neighbors(LLTABLE6(rt->rt_ifp), NULL);
1057 		break;
1058 	    }
1059 
1060 	case RTM_DELETE:
1061 		/* leave from solicited node multicast for proxy ND */
1062 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1063 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1064 			struct in6_addr llsol;
1065 
1066 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1067 			llsol.s6_addr32[0] = htonl(0xff020000);
1068 			llsol.s6_addr32[1] = 0;
1069 			llsol.s6_addr32[2] = htonl(1);
1070 			llsol.s6_addr8[12] = 0xff;
1071 			if (in6_setscope(&llsol, ifp, NULL) == 0)
1072 				in6_lookup_and_delete_multi(&llsol, ifp);
1073 		}
1074 		break;
1075 	}
1076 }
1077 
1078 static void
1079 nd6_setifflags(struct ifnet *ifp, uint32_t flags)
1080 {
1081 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
1082 	struct ifaddr *ifa;
1083 	struct in6_ifaddr *ia;
1084 	int s;
1085 
1086 	if (ndi->flags & ND6_IFF_IFDISABLED && !(flags & ND6_IFF_IFDISABLED)) {
1087 		/*
1088 		 * If the interface is marked as ND6_IFF_IFDISABLED and
1089 		 * has a link-local address with IN6_IFF_DUPLICATED,
1090 		 * do not clear ND6_IFF_IFDISABLED.
1091 		 * See RFC 4862, section 5.4.5.
1092 		 */
1093 		bool duplicated_linklocal = false;
1094 
1095 		s = pserialize_read_enter();
1096 		IFADDR_READER_FOREACH(ifa, ifp) {
1097 			if (ifa->ifa_addr->sa_family != AF_INET6)
1098 				continue;
1099 			ia = (struct in6_ifaddr *)ifa;
1100 			if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1101 			    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1102 			{
1103 				duplicated_linklocal = true;
1104 				break;
1105 			}
1106 		}
1107 		pserialize_read_exit(s);
1108 
1109 		if (duplicated_linklocal) {
1110 			flags |= ND6_IFF_IFDISABLED;
1111 			log(LOG_ERR, "%s: Cannot enable an interface"
1112 			    " with a link-local address marked"
1113 			    " duplicate.\n", if_name(ifp));
1114 		} else {
1115 			ndi->flags &= ~ND6_IFF_IFDISABLED;
1116 			if (ifp->if_flags & IFF_UP)
1117 				in6_if_up(ifp);
1118 		}
1119 	} else if (!(ndi->flags & ND6_IFF_IFDISABLED) &&
1120 	    (flags & ND6_IFF_IFDISABLED))
1121 	{
1122 		struct psref psref;
1123 		int bound = curlwp_bind();
1124 
1125 		/* Mark all IPv6 addresses as tentative. */
1126 
1127 		ndi->flags |= ND6_IFF_IFDISABLED;
1128 		s = pserialize_read_enter();
1129 		IFADDR_READER_FOREACH(ifa, ifp) {
1130 			if (ifa->ifa_addr->sa_family != AF_INET6)
1131 				continue;
1132 			ifa_acquire(ifa, &psref);
1133 			pserialize_read_exit(s);
1134 
1135 			nd6_dad_stop(ifa);
1136 
1137 			ia = (struct in6_ifaddr *)ifa;
1138 			ia->ia6_flags |= IN6_IFF_TENTATIVE;
1139 
1140 			s = pserialize_read_enter();
1141 			ifa_release(ifa, &psref);
1142 		}
1143 		pserialize_read_exit(s);
1144 		curlwp_bindx(bound);
1145 	}
1146 
1147 	if (flags & ND6_IFF_AUTO_LINKLOCAL) {
1148 		if (!(ndi->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1149 			/* auto_linklocal 0->1 transition */
1150 
1151 			ndi->flags |= ND6_IFF_AUTO_LINKLOCAL;
1152 			in6_ifattach(ifp, NULL);
1153 		} else if (!(flags & ND6_IFF_IFDISABLED) &&
1154 		    ifp->if_flags & IFF_UP)
1155 		{
1156 			/*
1157 			 * When the IF already has
1158 			 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1159 			 * address is assigned, and IFF_UP, try to
1160 			 * assign one.
1161 			 */
1162 			bool haslinklocal = 0;
1163 
1164 			s = pserialize_read_enter();
1165 			IFADDR_READER_FOREACH(ifa, ifp) {
1166 				if (ifa->ifa_addr->sa_family !=AF_INET6)
1167 					continue;
1168 				ia = (struct in6_ifaddr *)ifa;
1169 				if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1170 					haslinklocal = true;
1171 					break;
1172 				}
1173 			}
1174 			pserialize_read_exit(s);
1175 			if (!haslinklocal)
1176 				in6_ifattach(ifp, NULL);
1177 		}
1178 	}
1179 
1180 	ndi->flags = flags;
1181 }
1182 
1183 int
1184 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1185 {
1186 #ifdef OSIOCGIFINFO_IN6_90
1187 	struct in6_ndireq90 *ondi = (struct in6_ndireq90 *)data;
1188 	struct in6_ndifreq90 *ndif = (struct in6_ndifreq90 *)data;
1189 #define OND	ondi->ndi
1190 #endif
1191 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1192 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1193 	struct nd_kifinfo *ifndi = ND_IFINFO(ifp);
1194 	int error = 0;
1195 #define ND     ndi->ndi
1196 
1197 	switch (cmd) {
1198 #ifdef OSIOCSRTRFLUSH_IN6
1199 	case OSIOCGDRLST_IN6:		/* FALLTHROUGH */
1200 	case OSIOCGPRLST_IN6:		/* FALLTHROUGH */
1201 	case OSIOCSNDFLUSH_IN6:		/* FALLTHROUGH */
1202 	case OSIOCSPFXFLUSH_IN6:	/* FALLTHROUGH */
1203 	case OSIOCSRTRFLUSH_IN6:	/* FALLTHROUGH */
1204 		break;
1205 	case OSIOCGDEFIFACE_IN6:
1206 		ndif->ifindex = 0;
1207 		break;
1208 	case OSIOCSDEFIFACE_IN6:
1209 		error = ENOTSUP;
1210 		break;
1211 #endif
1212 #ifdef OSIOCGIFINFO_IN6
1213 	case OSIOCGIFINFO_IN6:		/* FALLTHROUGH */
1214 #endif
1215 #ifdef OSIOCGIFINFO_IN6_90
1216 	case OSIOCGIFINFO_IN6_90:
1217 		memset(&OND, 0, sizeof(OND));
1218 		OND.initialized = 1;
1219 		OND.chlim = ifndi->chlim;
1220 		OND.basereachable = ifndi->basereachable;
1221 		OND.retrans = ifndi->retrans;
1222 		OND.flags = ifndi->flags;
1223 		break;
1224 	case OSIOCSIFINFO_IN6_90:
1225 		/* Allow userland to set Neighour Unreachability Detection
1226 		 * timers. */
1227 		if (OND.chlim != 0)
1228 			ifndi->chlim = OND.chlim;
1229 		if (OND.basereachable != 0 &&
1230 		    OND.basereachable != ifndi->basereachable)
1231 		{
1232 			ifndi->basereachable = OND.basereachable;
1233 			ifndi->reachable = ND_COMPUTE_RTIME(OND.basereachable);
1234 		}
1235 		if (OND.retrans != 0)
1236 			ifndi->retrans = OND.retrans;
1237 		/* Retain the old behaviour .... */
1238 		/* FALLTHROUGH */
1239 	case OSIOCSIFINFO_FLAGS_90:
1240 		nd6_setifflags(ifp, OND.flags);
1241 		break;
1242 #undef OND
1243 #endif
1244 	case SIOCGIFINFO_IN6:
1245 		ND.chlim = ifndi->chlim;
1246 		ND.basereachable = ifndi->basereachable;
1247 		ND.retrans = ifndi->retrans;
1248 		ND.flags = ifndi->flags;
1249 		break;
1250 	case SIOCSIFINFO_IN6:
1251 		/* Allow userland to set Neighour Unreachability Detection
1252 		 * timers. */
1253 		if (ND.chlim != 0)
1254 			ifndi->chlim = ND.chlim;
1255 		if (ND.basereachable != 0 &&
1256 		    ND.basereachable != ifndi->basereachable)
1257 		{
1258 			ifndi->basereachable = ND.basereachable;
1259 			ifndi->reachable = ND_COMPUTE_RTIME(ND.basereachable);
1260 		}
1261 		if (ND.retrans != 0)
1262 			ifndi->retrans = ND.retrans;
1263 		break;
1264 	case SIOCSIFINFO_FLAGS:
1265 		nd6_setifflags(ifp, ND.flags);
1266 		break;
1267 #undef ND
1268 	case SIOCGNBRINFO_IN6:
1269 	{
1270 		struct llentry *ln;
1271 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1272 
1273 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1274 			return error;
1275 
1276 		ln = nd6_lookup(&nb_addr, ifp, false);
1277 		if (ln == NULL) {
1278 			error = EINVAL;
1279 			break;
1280 		}
1281 		nbi->state = ln->ln_state;
1282 		nbi->asked = ln->ln_asked;
1283 		nbi->isrouter = ln->ln_router;
1284 		nbi->expire = ln->ln_expire ?
1285 		    time_mono_to_wall(ln->ln_expire) : 0;
1286 		LLE_RUNLOCK(ln);
1287 
1288 		break;
1289 	}
1290 	}
1291 	return error;
1292 }
1293 
1294 void
1295 nd6_llinfo_release_pkts(struct llentry *ln, struct ifnet *ifp)
1296 {
1297 	struct mbuf *m_hold, *m_hold_next;
1298 	struct sockaddr_in6 sin6;
1299 
1300 	LLE_WLOCK_ASSERT(ln);
1301 
1302 	sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
1303 
1304 	m_hold = ln->la_hold, ln->la_hold = NULL, ln->la_numheld = 0;
1305 
1306 	LLE_WUNLOCK(ln);
1307 	for (; m_hold != NULL; m_hold = m_hold_next) {
1308 		m_hold_next = m_hold->m_nextpkt;
1309 		m_hold->m_nextpkt = NULL;
1310 
1311 		/*
1312 		 * we assume ifp is not a p2p here, so
1313 		 * just set the 2nd argument as the
1314 		 * 1st one.
1315 		 */
1316 		ip6_if_output(ifp, ifp, m_hold, &sin6, NULL);
1317 	}
1318 	LLE_WLOCK(ln);
1319 }
1320 
1321 /*
1322  * Create neighbor cache entry and cache link-layer address,
1323  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1324  */
1325 void
1326 nd6_cache_lladdr(
1327     struct ifnet *ifp,
1328     struct in6_addr *from,
1329     char *lladdr,
1330     int lladdrlen,
1331     int type,	/* ICMP6 type */
1332     int code	/* type dependent information */
1333 )
1334 {
1335 	struct llentry *ln = NULL;
1336 	int is_newentry;
1337 	int do_update;
1338 	int olladdr;
1339 	int llchange;
1340 	int newstate = 0;
1341 
1342 	KASSERT(ifp != NULL);
1343 	KASSERT(from != NULL);
1344 
1345 	/* nothing must be updated for unspecified address */
1346 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1347 		return;
1348 
1349 	/*
1350 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1351 	 * the caller.
1352 	 *
1353 	 * XXX If the link does not have link-layer adderss, what should
1354 	 * we do? (ifp->if_addrlen == 0)
1355 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1356 	 * description on it in NS section (RFC 2461 7.2.3).
1357 	 */
1358 
1359 	ln = nd6_lookup(from, ifp, true);
1360 	if (ln == NULL) {
1361 #if 0
1362 		/* nothing must be done if there's no lladdr */
1363 		if (!lladdr || !lladdrlen)
1364 			return NULL;
1365 #endif
1366 
1367 		ln = nd6_create(from, ifp);
1368 		is_newentry = 1;
1369 	} else {
1370 		/* do nothing if static ndp is set */
1371 		if (ln->la_flags & LLE_STATIC) {
1372 			LLE_WUNLOCK(ln);
1373 			return;
1374 		}
1375 		is_newentry = 0;
1376 	}
1377 
1378 	if (ln == NULL)
1379 		return;
1380 
1381 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1382 	if (olladdr && lladdr) {
1383 		llchange = memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen);
1384 	} else
1385 		llchange = 0;
1386 
1387 	/*
1388 	 * newentry olladdr  lladdr  llchange	(*=record)
1389 	 *	0	n	n	--	(1)
1390 	 *	0	y	n	--	(2)
1391 	 *	0	n	y	--	(3) * STALE
1392 	 *	0	y	y	n	(4) *
1393 	 *	0	y	y	y	(5) * STALE
1394 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1395 	 *	1	--	y	--	(7) * STALE
1396 	 */
1397 
1398 	if (lladdr) {		/* (3-5) and (7) */
1399 		/*
1400 		 * Record source link-layer address
1401 		 * XXX is it dependent to ifp->if_type?
1402 		 */
1403 		memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
1404 		ln->la_flags |= LLE_VALID;
1405 	}
1406 
1407 	if (!is_newentry) {
1408 		if ((!olladdr && lladdr) ||		/* (3) */
1409 		    (olladdr && lladdr && llchange)) {	/* (5) */
1410 			do_update = 1;
1411 			newstate = ND_LLINFO_STALE;
1412 		} else					/* (1-2,4) */
1413 			do_update = 0;
1414 	} else {
1415 		do_update = 1;
1416 		if (lladdr == NULL)			/* (6) */
1417 			newstate = ND_LLINFO_NOSTATE;
1418 		else					/* (7) */
1419 			newstate = ND_LLINFO_STALE;
1420 	}
1421 
1422 	if (do_update) {
1423 		/*
1424 		 * Update the state of the neighbor cache.
1425 		 */
1426 		ln->ln_state = newstate;
1427 
1428 		if (ln->ln_state == ND_LLINFO_STALE) {
1429 			/*
1430 			 * XXX: since nd6_output() below will cause
1431 			 * state tansition to DELAY and reset the timer,
1432 			 * we must set the timer now, although it is actually
1433 			 * meaningless.
1434 			 */
1435 			nd_set_timer(ln, ND_TIMER_GC);
1436 
1437 			nd6_llinfo_release_pkts(ln, ifp);
1438 		} else if (ln->ln_state == ND_LLINFO_INCOMPLETE) {
1439 			/* probe right away */
1440 			nd_set_timer(ln, ND_TIMER_IMMEDIATE);
1441 		}
1442 	}
1443 
1444 	/*
1445 	 * ICMP6 type dependent behavior.
1446 	 *
1447 	 * NS: clear IsRouter if new entry
1448 	 * RS: clear IsRouter
1449 	 * RA: set IsRouter if there's lladdr
1450 	 * redir: clear IsRouter if new entry
1451 	 *
1452 	 * RA case, (1):
1453 	 * The spec says that we must set IsRouter in the following cases:
1454 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1455 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1456 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1457 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1458 	 * neighbor cache, this is similar to (6).
1459 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1460 	 *
1461 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1462 	 *							D R
1463 	 *	0	n	n	--	(1)	c   ?     s
1464 	 *	0	y	n	--	(2)	c   s     s
1465 	 *	0	n	y	--	(3)	c   s     s
1466 	 *	0	y	y	n	(4)	c   s     s
1467 	 *	0	y	y	y	(5)	c   s     s
1468 	 *	1	--	n	--	(6) c	c 	c s
1469 	 *	1	--	y	--	(7) c	c   s	c s
1470 	 *
1471 	 *					(c=clear s=set)
1472 	 */
1473 	switch (type & 0xff) {
1474 	case ND_NEIGHBOR_SOLICIT:
1475 		/*
1476 		 * New entry must have is_router flag cleared.
1477 		 */
1478 		if (is_newentry)	/* (6-7) */
1479 			ln->ln_router = 0;
1480 		break;
1481 	case ND_REDIRECT:
1482 		/*
1483 		 * If the icmp is a redirect to a better router, always set the
1484 		 * is_router flag.  Otherwise, if the entry is newly created,
1485 		 * clear the flag.  [RFC 2461, sec 8.3]
1486 		 */
1487 		if (code == ND_REDIRECT_ROUTER)
1488 			ln->ln_router = 1;
1489 		else if (is_newentry) /* (6-7) */
1490 			ln->ln_router = 0;
1491 		break;
1492 	case ND_ROUTER_SOLICIT:
1493 		/*
1494 		 * is_router flag must always be cleared.
1495 		 */
1496 		ln->ln_router = 0;
1497 		break;
1498 	case ND_ROUTER_ADVERT:
1499 		/*
1500 		 * Mark an entry with lladdr as a router.
1501 		 */
1502 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1503 		    (is_newentry && lladdr)) {			/* (7) */
1504 			ln->ln_router = 1;
1505 		}
1506 		break;
1507 	}
1508 
1509 	if (do_update && lladdr != NULL) {
1510 		struct sockaddr_in6 sin6;
1511 
1512 		sockaddr_in6_init(&sin6, from, 0, 0, 0);
1513 		rt_clonedmsg(is_newentry ? RTM_ADD : RTM_CHANGE,
1514 		    NULL, sin6tosa(&sin6), lladdr, ifp);
1515 	}
1516 
1517 	if (ln != NULL)
1518 		LLE_WUNLOCK(ln);
1519 
1520 	/*
1521 	 * If we have too many cache entries, initiate immediate
1522 	 * purging for some entries.
1523 	 */
1524 	if (is_newentry)
1525 		nd6_gc_neighbors(LLTABLE6(ifp), &ln->r_l3addr.addr6);
1526 }
1527 
1528 static void
1529 nd6_slowtimo(void *ignored_arg)
1530 {
1531 	struct nd_kifinfo *ndi;
1532 	struct ifnet *ifp;
1533 	int s;
1534 
1535 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
1536 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1537 	    nd6_slowtimo, NULL);
1538 
1539 	s = pserialize_read_enter();
1540 	IFNET_READER_FOREACH(ifp) {
1541 		ndi = ND_IFINFO(ifp);
1542 		if (ndi->basereachable && /* already initialized */
1543 		    (ndi->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1544 			/*
1545 			 * Since reachable time rarely changes by router
1546 			 * advertisements, we SHOULD insure that a new random
1547 			 * value gets recomputed at least once every few hours.
1548 			 * (RFC 2461, 6.3.4)
1549 			 */
1550 			ndi->recalctm = nd6_recalc_reachtm_interval;
1551 			ndi->reachable = ND_COMPUTE_RTIME(ndi->basereachable);
1552 		}
1553 	}
1554 	pserialize_read_exit(s);
1555 
1556 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
1557 }
1558 
1559 /*
1560  * Return 0 if a neighbor cache is found. Return EWOULDBLOCK if a cache is not
1561  * found and trying to resolve a neighbor; in this case the mbuf is queued in
1562  * the list. Otherwise return errno after freeing the mbuf.
1563  */
1564 int
1565 nd6_resolve(struct ifnet *ifp, const struct rtentry *rt, struct mbuf *m,
1566     const struct sockaddr *_dst, uint8_t *lldst, size_t dstsize)
1567 {
1568 	struct llentry *ln = NULL;
1569 	bool created = false;
1570 	const struct sockaddr_in6 *dst = satocsin6(_dst);
1571 	int error;
1572 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
1573 
1574 	/* discard the packet if IPv6 operation is disabled on the interface */
1575 	if (ndi->flags & ND6_IFF_IFDISABLED) {
1576 		m_freem(m);
1577 		return ENETDOWN; /* better error? */
1578 	}
1579 
1580 	/*
1581 	 * Address resolution or Neighbor Unreachability Detection
1582 	 * for the next hop.
1583 	 * At this point, the destination of the packet must be a unicast
1584 	 * or an anycast address(i.e. not a multicast).
1585 	 */
1586 
1587 	/* Look up the neighbor cache for the nexthop */
1588 	ln = nd6_lookup(&dst->sin6_addr, ifp, false);
1589 
1590 	if (ln != NULL && (ln->la_flags & LLE_VALID) != 0 &&
1591 	    ln->ln_state == ND_LLINFO_REACHABLE) {
1592 		/* Fast path */
1593 		memcpy(lldst, &ln->ll_addr, MIN(dstsize, ifp->if_addrlen));
1594 		LLE_RUNLOCK(ln);
1595 		return 0;
1596 	}
1597 	if (ln != NULL)
1598 		LLE_RUNLOCK(ln);
1599 
1600 	/* Slow path */
1601 	ln = nd6_lookup(&dst->sin6_addr, ifp, true);
1602 	if (ln == NULL && nd6_is_addr_neighbor(dst, ifp))  {
1603 		/*
1604 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1605 		 * the condition below is not very efficient.  But we believe
1606 		 * it is tolerable, because this should be a rare case.
1607 		 */
1608 		ln = nd6_create(&dst->sin6_addr, ifp);
1609 		if (ln == NULL) {
1610 			char ip6buf[INET6_ADDRSTRLEN];
1611 			log(LOG_DEBUG,
1612 			    "%s: can't allocate llinfo for %s "
1613 			    "(ln=%p, rt=%p)\n", __func__,
1614 			    IN6_PRINT(ip6buf, &dst->sin6_addr), ln, rt);
1615 			m_freem(m);
1616 			return ENOBUFS;
1617 		}
1618 		created = true;
1619 	}
1620 
1621 	if (ln == NULL) {
1622 		m_freem(m);
1623 		return ENETDOWN; /* better error? */
1624 	}
1625 
1626 	error = nd_resolve(ln, rt, m, lldst, dstsize);
1627 
1628 	if (created)
1629 		nd6_gc_neighbors(LLTABLE6(ifp), &dst->sin6_addr);
1630 
1631 	return error;
1632 }
1633 
1634 int
1635 nd6_need_cache(struct ifnet *ifp)
1636 {
1637 	/*
1638 	 * XXX: we currently do not make neighbor cache on any interface
1639 	 * other than ARCnet, Ethernet, and GIF.
1640 	 *
1641 	 * RFC2893 says:
1642 	 * - unidirectional tunnels needs no ND
1643 	 */
1644 	switch (ifp->if_type) {
1645 	case IFT_ARCNET:
1646 	case IFT_ETHER:
1647 	case IFT_IEEE1394:
1648 	case IFT_CARP:
1649 	case IFT_GIF:		/* XXX need more cases? */
1650 	case IFT_PPP:
1651 	case IFT_TUNNEL:
1652 		return 1;
1653 	default:
1654 		return 0;
1655 	}
1656 }
1657 
1658 int
1659 nd6_sysctl(
1660     int name,
1661     void *oldp,	/* syscall arg, need copyout */
1662     size_t *oldlenp,
1663     void *newp,	/* syscall arg, need copyin */
1664     size_t newlen
1665 )
1666 {
1667 
1668 	if (newp)
1669 		return EPERM;
1670 
1671 	switch (name) {
1672 #ifdef COMPAT_90
1673 	case OICMPV6CTL_ND6_DRLIST: /* FALLTHROUGH */
1674 	case OICMPV6CTL_ND6_PRLIST:
1675 		*oldlenp = 0;
1676 		return 0;
1677 #endif
1678 	case ICMPV6CTL_ND6_MAXQLEN:
1679 		return 0;
1680 	default:
1681 		return ENOPROTOOPT;
1682 	}
1683 }
1684