xref: /netbsd-src/sys/netinet6/nd6.c (revision 5a9ce3a7bf83e158cd6c1eb8f71cec8041d3630c)
1 /*	$NetBSD: nd6.c,v 1.282 2024/04/11 07:34:37 knakahara 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 1.282 2024/04/11 07:34:37 knakahara 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 #include <sys/compat_stub.h>
61 
62 #include <net/if.h>
63 #include <net/if_dl.h>
64 #include <net/if_llatbl.h>
65 #include <net/if_types.h>
66 #include <net/nd.h>
67 #include <net/route.h>
68 #include <net/if_ether.h>
69 #include <net/if_arc.h>
70 
71 #include <netinet/in.h>
72 #include <netinet6/in6_var.h>
73 #include <netinet/ip6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/scope6_var.h>
76 #include <netinet6/nd6.h>
77 #include <netinet6/in6_ifattach.h>
78 #include <netinet/icmp6.h>
79 #include <netinet6/icmp6_private.h>
80 
81 #include <compat/netinet6/in6_var.h>
82 #include <compat/netinet6/nd6.h>
83 
84 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
85 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
86 
87 /* timer values */
88 int	nd6_prune	= 1;	/* walk list every 1 seconds */
89 int	nd6_useloopback = 1;	/* use loopback interface for local traffic */
90 
91 /* preventing too many loops in ND option parsing */
92 int nd6_maxndopt = 10;	/* max # of ND options allowed */
93 
94 #ifdef ND6_DEBUG
95 int nd6_debug = 1;
96 #else
97 int nd6_debug = 0;
98 #endif
99 
100 krwlock_t nd6_lock __cacheline_aligned;
101 
102 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL;
103 
104 static void nd6_slowtimo(void *);
105 static void nd6_free(struct llentry *, int);
106 static bool nd6_nud_enabled(struct ifnet *);
107 static unsigned int nd6_llinfo_reachable(struct ifnet *);
108 static unsigned int nd6_llinfo_retrans(struct ifnet *);
109 static union l3addr *nd6_llinfo_holdsrc(struct llentry *, union l3addr *);
110 static void nd6_llinfo_output(struct ifnet *, const union l3addr *,
111     const union l3addr *, const uint8_t *, const union l3addr *);
112 static void nd6_llinfo_missed(struct ifnet *, const union l3addr *,
113     int16_t, struct mbuf *);
114 static void nd6_timer(void *);
115 static void nd6_timer_work(struct work *, void *);
116 static struct nd_opt_hdr *nd6_option(union nd_opts *);
117 
118 static callout_t nd6_slowtimo_ch;
119 static callout_t nd6_timer_ch;
120 static struct workqueue	*nd6_timer_wq;
121 static struct work	nd6_timer_wk;
122 
123 struct nd_domain nd6_nd_domain = {
124 	.nd_family = AF_INET6,
125 	.nd_delay = 5,		/* delay first probe time 5 second */
126 	.nd_mmaxtries = 3,	/* maximum unicast query */
127 	.nd_umaxtries = 3,	/* maximum multicast query */
128 	.nd_retransmultiple = BACKOFF_MULTIPLE,
129 	.nd_maxretrans = MAX_RETRANS_TIMER,
130 	.nd_maxnudhint = 0,	/* max # of subsequent upper layer hints */
131 	.nd_maxqueuelen = 1,	/* max # of packets in unresolved ND entries */
132 	.nd_nud_enabled = nd6_nud_enabled,
133 	.nd_reachable = nd6_llinfo_reachable,
134 	.nd_retrans = nd6_llinfo_retrans,
135 	.nd_holdsrc = nd6_llinfo_holdsrc,
136 	.nd_output = nd6_llinfo_output,
137 	.nd_missed = nd6_llinfo_missed,
138 	.nd_free = nd6_free,
139 };
140 
141 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery");
142 
143 void
nd6_init(void)144 nd6_init(void)
145 {
146 	int error;
147 
148 	nd_attach_domain(&nd6_nd_domain);
149 	nd6_nbr_init();
150 
151 	rw_init(&nd6_lock);
152 
153 	callout_init(&nd6_slowtimo_ch, CALLOUT_MPSAFE);
154 	callout_init(&nd6_timer_ch, CALLOUT_MPSAFE);
155 
156 	error = workqueue_create(&nd6_timer_wq, "nd6_timer",
157 	    nd6_timer_work, NULL, PRI_SOFTNET, IPL_SOFTNET, WQ_MPSAFE);
158 	if (error)
159 		panic("%s: workqueue_create failed (%d)\n", __func__, error);
160 
161 	/* start timer */
162 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
163 	    nd6_slowtimo, NULL);
164 	callout_reset(&nd6_timer_ch, hz, nd6_timer, NULL);
165 }
166 
167 struct nd_kifinfo *
nd6_ifattach(struct ifnet * ifp)168 nd6_ifattach(struct ifnet *ifp)
169 {
170 	struct nd_kifinfo *nd;
171 
172 	nd = kmem_zalloc(sizeof(*nd), KM_SLEEP);
173 
174 	nd->chlim = IPV6_DEFHLIM;
175 	nd->basereachable = REACHABLE_TIME;
176 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
177 	nd->retrans = RETRANS_TIMER;
178 
179 	nd->flags = ND6_IFF_PERFORMNUD;
180 
181 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
182 	 * A bridge interface should not have ND6_IFF_AUTO_LINKLOCAL
183 	 * because one of its members should. */
184 	if ((ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
185 	    (ifp->if_flags & IFF_LOOPBACK))
186 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
187 
188 	return nd;
189 }
190 
191 void
nd6_ifdetach(struct ifnet * ifp,struct in6_ifextra * ext)192 nd6_ifdetach(struct ifnet *ifp, struct in6_ifextra *ext)
193 {
194 
195 	/* Ensure all IPv6 addresses are purged before calling nd6_purge */
196 	if_purgeaddrs(ifp, AF_INET6, in6_purgeaddr);
197 	nd6_purge(ifp, ext);
198 	kmem_free(ext->nd_ifinfo, sizeof(struct nd_kifinfo));
199 }
200 
201 void
nd6_option_init(void * opt,int icmp6len,union nd_opts * ndopts)202 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
203 {
204 
205 	memset(ndopts, 0, sizeof(*ndopts));
206 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
207 	ndopts->nd_opts_last
208 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
209 
210 	if (icmp6len == 0) {
211 		ndopts->nd_opts_done = 1;
212 		ndopts->nd_opts_search = NULL;
213 	}
214 }
215 
216 /*
217  * Take one ND option.
218  */
219 static struct nd_opt_hdr *
nd6_option(union nd_opts * ndopts)220 nd6_option(union nd_opts *ndopts)
221 {
222 	struct nd_opt_hdr *nd_opt;
223 	int olen;
224 
225 	KASSERT(ndopts != NULL);
226 	KASSERT(ndopts->nd_opts_last != NULL);
227 
228 	if (ndopts->nd_opts_search == NULL)
229 		return NULL;
230 	if (ndopts->nd_opts_done)
231 		return NULL;
232 
233 	nd_opt = ndopts->nd_opts_search;
234 
235 	/* make sure nd_opt_len is inside the buffer */
236 	if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) {
237 		memset(ndopts, 0, sizeof(*ndopts));
238 		return NULL;
239 	}
240 
241 	olen = nd_opt->nd_opt_len << 3;
242 	if (olen == 0) {
243 		/*
244 		 * Message validation requires that all included
245 		 * options have a length that is greater than zero.
246 		 */
247 		memset(ndopts, 0, sizeof(*ndopts));
248 		return NULL;
249 	}
250 
251 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen);
252 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
253 		/* option overruns the end of buffer, invalid */
254 		memset(ndopts, 0, sizeof(*ndopts));
255 		return NULL;
256 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
257 		/* reached the end of options chain */
258 		ndopts->nd_opts_done = 1;
259 		ndopts->nd_opts_search = NULL;
260 	}
261 	return nd_opt;
262 }
263 
264 /*
265  * Parse multiple ND options.
266  * This function is much easier to use, for ND routines that do not need
267  * multiple options of the same type.
268  */
269 int
nd6_options(union nd_opts * ndopts)270 nd6_options(union nd_opts *ndopts)
271 {
272 	struct nd_opt_hdr *nd_opt;
273 	int i = 0;
274 
275 	KASSERT(ndopts != NULL);
276 	KASSERT(ndopts->nd_opts_last != NULL);
277 
278 	if (ndopts->nd_opts_search == NULL)
279 		return 0;
280 
281 	while (1) {
282 		nd_opt = nd6_option(ndopts);
283 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
284 			/*
285 			 * Message validation requires that all included
286 			 * options have a length that is greater than zero.
287 			 */
288 			ICMP6_STATINC(ICMP6_STAT_ND_BADOPT);
289 			memset(ndopts, 0, sizeof(*ndopts));
290 			return -1;
291 		}
292 
293 		if (nd_opt == NULL)
294 			goto skip1;
295 
296 		switch (nd_opt->nd_opt_type) {
297 		case ND_OPT_SOURCE_LINKADDR:
298 		case ND_OPT_TARGET_LINKADDR:
299 		case ND_OPT_MTU:
300 		case ND_OPT_REDIRECTED_HEADER:
301 		case ND_OPT_NONCE:
302 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
303 				nd6log(LOG_INFO,
304 				    "duplicated ND6 option found (type=%d)\n",
305 				    nd_opt->nd_opt_type);
306 				/* XXX bark? */
307 			} else {
308 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
309 					= nd_opt;
310 			}
311 			break;
312 		case ND_OPT_PREFIX_INFORMATION:
313 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
314 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
315 					= nd_opt;
316 			}
317 			ndopts->nd_opts_pi_end =
318 				(struct nd_opt_prefix_info *)nd_opt;
319 			break;
320 		default:
321 			/*
322 			 * Unknown options must be silently ignored,
323 			 * to accommodate future extension to the protocol.
324 			 */
325 			nd6log(LOG_DEBUG,
326 			    "nd6_options: unsupported option %d - "
327 			    "option ignored\n", nd_opt->nd_opt_type);
328 		}
329 
330 skip1:
331 		i++;
332 		if (i > nd6_maxndopt) {
333 			ICMP6_STATINC(ICMP6_STAT_ND_TOOMANYOPT);
334 			nd6log(LOG_INFO, "too many loop in nd opt\n");
335 			break;
336 		}
337 
338 		if (ndopts->nd_opts_done)
339 			break;
340 	}
341 
342 	return 0;
343 }
344 
345 /*
346  * Gets source address of the first packet in hold queue
347  * and stores it in @src.
348  * Returns pointer to @src (if hold queue is not empty) or NULL.
349  */
350 static struct in6_addr *
nd6_llinfo_get_holdsrc(struct llentry * ln,struct in6_addr * src)351 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
352 {
353 	struct ip6_hdr *hip6;
354 
355 	if (ln == NULL || ln->ln_hold == NULL)
356 		return NULL;
357 
358 	/*
359 	 * assuming every packet in ln_hold has the same IP header
360 	 */
361 	hip6 = mtod(ln->ln_hold, struct ip6_hdr *);
362 	/* XXX pullup? */
363 	if (sizeof(*hip6) < ln->ln_hold->m_len)
364 		*src = hip6->ip6_src;
365 	else
366 		src = NULL;
367 
368 	return src;
369 }
370 
371 static union l3addr *
nd6_llinfo_holdsrc(struct llentry * ln,union l3addr * src)372 nd6_llinfo_holdsrc(struct llentry *ln, union l3addr *src)
373 {
374 
375 	if (nd6_llinfo_get_holdsrc(ln, &src->addr6) == NULL)
376 		return NULL;
377 	return src;
378 }
379 
380 static void
nd6_llinfo_output(struct ifnet * ifp,const union l3addr * daddr,const union l3addr * taddr,__unused const uint8_t * tlladdr,const union l3addr * hsrc)381 nd6_llinfo_output(struct ifnet *ifp, const union l3addr *daddr,
382     const union l3addr *taddr, __unused const uint8_t *tlladdr,
383     const union l3addr *hsrc)
384 {
385 
386 	nd6_ns_output(ifp,
387 	    daddr != NULL ? &daddr->addr6 : NULL,
388 	    taddr != NULL ? &taddr->addr6 : NULL,
389 	    hsrc != NULL ? &hsrc->addr6 : NULL, NULL);
390 }
391 
392 static bool
nd6_nud_enabled(struct ifnet * ifp)393 nd6_nud_enabled(struct ifnet *ifp)
394 {
395 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
396 
397 	return ndi->flags & ND6_IFF_PERFORMNUD;
398 }
399 
400 static unsigned int
nd6_llinfo_reachable(struct ifnet * ifp)401 nd6_llinfo_reachable(struct ifnet *ifp)
402 {
403 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
404 
405 	return ndi->reachable;
406 }
407 
408 static unsigned int
nd6_llinfo_retrans(struct ifnet * ifp)409 nd6_llinfo_retrans(struct ifnet *ifp)
410 {
411 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
412 
413 	return ndi->retrans;
414 }
415 
416 static void
nd6_llinfo_missed(struct ifnet * ifp,const union l3addr * taddr,int16_t type,struct mbuf * m)417 nd6_llinfo_missed(struct ifnet *ifp, const union l3addr *taddr,
418     int16_t type, struct mbuf *m)
419 {
420 	struct in6_addr mdaddr6 = zeroin6_addr;
421 	struct sockaddr_in6 dsin6, tsin6;
422 	struct sockaddr *sa;
423 
424 	if (m != NULL) {
425 		if (type == ND_LLINFO_PROBE) {
426 			struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
427 
428 			/* XXX pullup? */
429 			if (sizeof(*ip6) < m->m_len)
430 				mdaddr6 = ip6->ip6_src;
431 			m_freem(m);
432 		} else
433 			icmp6_error2(m, ICMP6_DST_UNREACH,
434 			    ICMP6_DST_UNREACH_ADDR, 0, ifp, &mdaddr6);
435 	}
436 	if (!IN6_IS_ADDR_UNSPECIFIED(&mdaddr6)) {
437 		sockaddr_in6_init(&dsin6, &mdaddr6, 0, 0, 0);
438 		sa = sin6tosa(&dsin6);
439 	} else
440 		sa = NULL;
441 
442 	sockaddr_in6_init(&tsin6, &taddr->addr6, 0, 0, 0);
443 	rt_clonedmsg(RTM_MISS, sa, sin6tosa(&tsin6), NULL, ifp);
444 }
445 
446 /*
447  * ND6 timer routine to expire default route list and prefix list
448  */
449 static void
nd6_timer_work(struct work * wk,void * arg)450 nd6_timer_work(struct work *wk, void *arg)
451 {
452 	struct in6_ifaddr *ia6, *nia6;
453 	int s, bound;
454 	struct psref psref;
455 
456 	callout_reset(&nd6_timer_ch, nd6_prune * hz,
457 	    nd6_timer, NULL);
458 
459 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
460 
461 	/* expire interface addresses */
462 	bound = curlwp_bind();
463 	s = pserialize_read_enter();
464 	for (ia6 = IN6_ADDRLIST_READER_FIRST(); ia6; ia6 = nia6) {
465 		nia6 = IN6_ADDRLIST_READER_NEXT(ia6);
466 
467 		ia6_acquire(ia6, &psref);
468 		pserialize_read_exit(s);
469 
470 		/* check address lifetime */
471 		if (IFA6_IS_INVALID(ia6)) {
472 			struct ifnet *ifp;
473 
474 			ifp = ia6->ia_ifa.ifa_ifp;
475 			IFNET_LOCK(ifp);
476 			/*
477 			 * Need to take the lock first to prevent if_detach
478 			 * from running in6_purgeaddr concurrently.
479 			 */
480 			if (!if_is_deactivated(ifp)) {
481 				ia6_release(ia6, &psref);
482 				in6_purgeaddr(&ia6->ia_ifa);
483 			} else {
484 				/*
485 				 * ifp is being destroyed, ia6 will be destroyed
486 				 * by if_detach.
487 				 */
488 				ia6_release(ia6, &psref);
489 			}
490 			ia6 = NULL;
491 			IFNET_UNLOCK(ifp);
492 		} else if (IFA6_IS_DEPRECATED(ia6)) {
493 			int oldflags = ia6->ia6_flags;
494 
495 			if ((oldflags & IN6_IFF_DEPRECATED) == 0) {
496 				ia6->ia6_flags |= IN6_IFF_DEPRECATED;
497 				rt_addrmsg(RTM_NEWADDR, (struct ifaddr *)ia6);
498 			}
499 		} else {
500 			/*
501 			 * A new RA might have made a deprecated address
502 			 * preferred.
503 			 */
504 			if (ia6->ia6_flags & IN6_IFF_DEPRECATED) {
505 				ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
506 				rt_addrmsg(RTM_NEWADDR, (struct ifaddr *)ia6);
507 			}
508 		}
509 		s = pserialize_read_enter();
510 		ia6_release(ia6, &psref);
511 	}
512 	pserialize_read_exit(s);
513 	curlwp_bindx(bound);
514 
515 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
516 }
517 
518 static void
nd6_timer(void * ignored_arg)519 nd6_timer(void *ignored_arg)
520 {
521 
522 	workqueue_enqueue(nd6_timer_wq, &nd6_timer_wk, NULL);
523 }
524 
525 /*
526  * Nuke neighbor cache/prefix/default router management table, right before
527  * ifp goes away.
528  */
529 void
nd6_purge(struct ifnet * ifp,struct in6_ifextra * ext)530 nd6_purge(struct ifnet *ifp, struct in6_ifextra *ext)
531 {
532 
533 	/*
534 	 * During detach, the ND info might be already removed, but
535 	 * then is explitly passed as argument.
536 	 * Otherwise get it from ifp->if_afdata.
537 	 */
538 	if (ext == NULL)
539 		ext = ifp->if_afdata[AF_INET6];
540 	if (ext == NULL)
541 		return;
542 
543 	/*
544 	 * We may not need to nuke the neighbor cache entries here
545 	 * because the neighbor cache is kept in if_afdata[AF_INET6].
546 	 * nd6_purge() is invoked by in6_ifdetach() which is called
547 	 * from if_detach() where everything gets purged. However
548 	 * in6_ifdetach is directly called from vlan(4), so we still
549 	 * need to purge entries here.
550 	 */
551 	if (ext->lltable != NULL)
552 		lltable_purge_entries(ext->lltable);
553 }
554 
555 struct llentry *
nd6_lookup(const struct in6_addr * addr6,const struct ifnet * ifp,bool wlock)556 nd6_lookup(const struct in6_addr *addr6, const struct ifnet *ifp, bool wlock)
557 {
558 	struct sockaddr_in6 sin6;
559 	struct llentry *ln;
560 
561 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
562 
563 	IF_AFDATA_RLOCK(ifp);
564 	ln = lla_lookup(LLTABLE6(ifp), wlock ? LLE_EXCLUSIVE : 0,
565 	    sin6tosa(&sin6));
566 	IF_AFDATA_RUNLOCK(ifp);
567 
568 	return ln;
569 }
570 
571 struct llentry *
nd6_create(const struct in6_addr * addr6,const struct ifnet * ifp)572 nd6_create(const struct in6_addr *addr6, const struct ifnet *ifp)
573 {
574 	struct sockaddr_in6 sin6;
575 	struct llentry *ln;
576 	struct rtentry *rt;
577 
578 	sockaddr_in6_init(&sin6, addr6, 0, 0, 0);
579 	rt = rtalloc1(sin6tosa(&sin6), 0);
580 
581 	IF_AFDATA_WLOCK(ifp);
582 	ln = lla_create(LLTABLE6(ifp), LLE_EXCLUSIVE, sin6tosa(&sin6), rt);
583 	IF_AFDATA_WUNLOCK(ifp);
584 
585 	if (rt != NULL)
586 		rt_unref(rt);
587 	if (ln != NULL)
588 		ln->ln_state = ND_LLINFO_NOSTATE;
589 
590 	return ln;
591 }
592 
593 /*
594  * Test whether a given IPv6 address is a neighbor or not, ignoring
595  * the actual neighbor cache.  The neighbor cache is ignored in order
596  * to not reenter the routing code from within itself.
597  */
598 static int
nd6_is_new_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)599 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
600 {
601 	struct ifaddr *dstaddr;
602 	int s;
603 
604 	/*
605 	 * A link-local address is always a neighbor.
606 	 * XXX: a link does not necessarily specify a single interface.
607 	 */
608 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
609 		struct sockaddr_in6 sin6_copy;
610 		u_int32_t zone;
611 
612 		/*
613 		 * We need sin6_copy since sa6_recoverscope() may modify the
614 		 * content (XXX).
615 		 */
616 		sin6_copy = *addr;
617 		if (sa6_recoverscope(&sin6_copy))
618 			return 0; /* XXX: should be impossible */
619 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
620 			return 0;
621 		if (sin6_copy.sin6_scope_id == zone)
622 			return 1;
623 		else
624 			return 0;
625 	}
626 
627 	/*
628 	 * If the address is assigned on the node of the other side of
629 	 * a p2p interface, the address should be a neighbor.
630 	 */
631 	s = pserialize_read_enter();
632 	dstaddr = ifa_ifwithdstaddr(sin6tocsa(addr));
633 	if (dstaddr != NULL) {
634 		if (dstaddr->ifa_ifp == ifp) {
635 			pserialize_read_exit(s);
636 			return 1;
637 		}
638 	}
639 	pserialize_read_exit(s);
640 
641 	return 0;
642 }
643 
644 /*
645  * Detect if a given IPv6 address identifies a neighbor on a given link.
646  * XXX: should take care of the destination of a p2p link?
647  */
648 int
nd6_is_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)649 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
650 {
651 	struct llentry *ln;
652 	struct rtentry *rt;
653 
654 	/*
655 	 * A link-local address is always a neighbor.
656 	 * XXX: a link does not necessarily specify a single interface.
657 	 */
658 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
659 		struct sockaddr_in6 sin6_copy;
660 		u_int32_t zone;
661 
662 		/*
663 		 * We need sin6_copy since sa6_recoverscope() may modify the
664 		 * content (XXX).
665 		 */
666 		sin6_copy = *addr;
667 		if (sa6_recoverscope(&sin6_copy))
668 			return 0; /* XXX: should be impossible */
669 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
670 			return 0;
671 		if (sin6_copy.sin6_scope_id == zone)
672 			return 1;
673 		else
674 			return 0;
675 	}
676 
677 	if (nd6_is_new_addr_neighbor(addr, ifp))
678 		return 1;
679 
680 	/*
681 	 * Even if the address matches none of our addresses, it might be
682 	 * in the neighbor cache or a connected route.
683 	 */
684 	ln = nd6_lookup(&addr->sin6_addr, ifp, false);
685 	if (ln != NULL) {
686 		LLE_RUNLOCK(ln);
687 		return 1;
688 	}
689 
690 	rt = rtalloc1(sin6tocsa(addr), 0);
691 	if (rt == NULL)
692 		return 0;
693 
694 	if ((rt->rt_flags & RTF_CONNECTED) && (rt->rt_ifp == ifp
695 #if NBRIDGE > 0
696 	    || rt->rt_ifp->if_bridge == ifp->if_bridge
697 #endif
698 #if NCARP > 0
699 	    || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) ||
700 	    (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)||
701 	    (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP &&
702 	    rt->rt_ifp->if_carpdev == ifp->if_carpdev)
703 #endif
704 	    )) {
705 		rt_unref(rt);
706 		return 1;
707 	}
708 	rt_unref(rt);
709 
710 	return 0;
711 }
712 
713 /*
714  * Free an nd6 llinfo entry.
715  * Since the function would cause significant changes in the kernel, DO NOT
716  * make it global, unless you have a strong reason for the change, and are sure
717  * that the change is safe.
718  */
719 static void
nd6_free(struct llentry * ln,int gc)720 nd6_free(struct llentry *ln, int gc)
721 {
722 	struct ifnet *ifp;
723 
724 	KASSERT(ln != NULL);
725 	LLE_WLOCK_ASSERT(ln);
726 
727 	/*
728 	 * If the reason for the deletion is just garbage collection,
729 	 * and the neighbor is an active router, do not delete it.
730 	 * Instead, reset the GC timer using the router's lifetime.
731 	 * XXX: the check for ln_state should be redundant,
732 	 *      but we intentionally keep it just in case.
733 	 */
734 	if (!ip6_forwarding && ln->ln_router &&
735 	    ln->ln_state == ND_LLINFO_STALE && gc)
736 	{
737 		nd_set_timer(ln, ND_TIMER_EXPIRE);
738 		LLE_WUNLOCK(ln);
739 		return;
740 	}
741 
742 	ifp = ln->lle_tbl->llt_ifp;
743 
744 	if (ln->la_flags & LLE_VALID || gc) {
745 		struct sockaddr_in6 sin6;
746 		const char *lladdr;
747 
748 		sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
749 		lladdr = ln->la_flags & LLE_VALID ?
750 		    (const char *)&ln->ll_addr : NULL;
751 		rt_clonedmsg(RTM_DELETE, NULL, sin6tosa(&sin6), lladdr, ifp);
752 	}
753 
754 	/*
755 	 * Save to unlock. We still hold an extra reference and will not
756 	 * free(9) in llentry_free() if someone else holds one as well.
757 	 */
758 	LLE_WUNLOCK(ln);
759 	IF_AFDATA_LOCK(ifp);
760 	LLE_WLOCK(ln);
761 
762 	lltable_free_entry(LLTABLE6(ifp), ln);
763 
764 	IF_AFDATA_UNLOCK(ifp);
765 }
766 
767 /*
768  * Upper-layer reachability hint for Neighbor Unreachability Detection.
769  *
770  * XXX cost-effective methods?
771  */
772 void
nd6_nud_hint(struct rtentry * rt)773 nd6_nud_hint(struct rtentry *rt)
774 {
775 	struct llentry *ln;
776 	struct ifnet *ifp;
777 
778 	if (rt == NULL)
779 		return;
780 
781 	ifp = rt->rt_ifp;
782 	ln = nd6_lookup(&(satocsin6(rt_getkey(rt)))->sin6_addr, ifp, true);
783 	nd_nud_hint(ln);
784 }
785 
786 struct gc_args {
787 	int gc_entries;
788 	const struct in6_addr *skip_in6;
789 };
790 
791 static int
nd6_purge_entry(struct lltable * llt,struct llentry * ln,void * farg)792 nd6_purge_entry(struct lltable *llt, struct llentry *ln, void *farg)
793 {
794 	struct gc_args *args = farg;
795 	int *n = &args->gc_entries;
796 	const struct in6_addr *skip_in6 = args->skip_in6;
797 
798 	if (*n <= 0)
799 		return 0;
800 
801 	if (ND_IS_LLINFO_PERMANENT(ln))
802 		return 0;
803 
804 	if (IN6_ARE_ADDR_EQUAL(&ln->r_l3addr.addr6, skip_in6))
805 		return 0;
806 
807 	LLE_WLOCK(ln);
808 	if (ln->ln_state > ND_LLINFO_INCOMPLETE)
809 		ln->ln_state = ND_LLINFO_STALE;
810 	else
811 		ln->ln_state = ND_LLINFO_PURGE;
812 	nd_set_timer(ln, ND_TIMER_IMMEDIATE);
813 	LLE_WUNLOCK(ln);
814 
815 	(*n)--;
816 	return 0;
817 }
818 
819 static void
nd6_gc_neighbors(struct lltable * llt,const struct in6_addr * in6)820 nd6_gc_neighbors(struct lltable *llt, const struct in6_addr *in6)
821 {
822 
823 	if (ip6_neighborgcthresh >= 0 &&
824 	    lltable_get_entry_count(llt) >= ip6_neighborgcthresh) {
825 		struct gc_args gc_args = {10, in6};
826 		/*
827 		 * XXX entries that are "less recently used" should be
828 		 * freed first.
829 		 */
830 		lltable_foreach_lle(llt, nd6_purge_entry, &gc_args);
831 	}
832 }
833 
834 void
nd6_rtrequest(int req,struct rtentry * rt,const struct rt_addrinfo * info)835 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info)
836 {
837 	struct sockaddr *gate = rt->rt_gateway;
838 	struct ifnet *ifp = rt->rt_ifp;
839 	uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen;
840 	struct ifaddr *ifa;
841 
842 	RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
843 
844 	if (req == RTM_LLINFO_UPD) {
845 		int rc;
846 		struct in6_addr *in6;
847 		struct in6_addr in6_all;
848 		int anycast;
849 
850 		if ((ifa = info->rti_ifa) == NULL)
851 			return;
852 
853 		in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
854 		anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST;
855 
856 		in6_all = in6addr_linklocal_allnodes;
857 		if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) {
858 			log(LOG_ERR, "%s: failed to set scope %s "
859 			    "(errno=%d)\n", __func__, if_name(ifp), rc);
860 			return;
861 		}
862 
863 		/* XXX don't set Override for proxy addresses */
864 		nd6_na_output(ifa->ifa_ifp, &in6_all, in6,
865 		    (anycast ? 0 : ND_NA_FLAG_OVERRIDE)
866 #if 0
867 		    | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0)
868 #endif
869 		    , 1, NULL);
870 		return;
871 	}
872 
873 	if ((rt->rt_flags & RTF_GATEWAY) != 0) {
874 		if (req != RTM_ADD)
875 			return;
876 		/*
877 		 * linklayers with particular MTU limitation.
878 		 */
879 		switch(ifp->if_type) {
880 #if NARCNET > 0
881 		case IFT_ARCNET:
882 			if (rt->rt_rmx.rmx_mtu > ARC_PHDS_MAXMTU) /* RFC2497 */
883 				rt->rt_rmx.rmx_mtu = ARC_PHDS_MAXMTU;
884 			break;
885 #endif
886 		}
887 		return;
888 	}
889 
890 	if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) {
891 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
892 		/*
893 		 * This is probably an interface direct route for a link
894 		 * which does not need neighbor caches (e.g. fe80::%lo0/64).
895 		 * We do not need special treatment below for such a route.
896 		 * Moreover, the RTF_LLINFO flag which would be set below
897 		 * would annoy the ndp(8) command.
898 		 */
899 		return;
900 	}
901 
902 	switch (req) {
903 	case RTM_ADD: {
904 		struct psref psref;
905 
906 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
907 		/*
908 		 * There is no backward compatibility :)
909 		 *
910 		 * if ((rt->rt_flags & RTF_HOST) == 0 &&
911 		 *     SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff)
912 		 *	   rt->rt_flags |= RTF_CLONING;
913 		 */
914 		/* XXX should move to route.c? */
915 		if (rt->rt_flags & (RTF_CONNECTED | RTF_LOCAL)) {
916 			union {
917 				struct sockaddr sa;
918 				struct sockaddr_dl sdl;
919 				struct sockaddr_storage ss;
920 			} u;
921 			/*
922 			 * Case 1: This route should come from a route to
923 			 * interface (RTF_CLONING case) or the route should be
924 			 * treated as on-link but is currently not
925 			 * (RTF_LLINFO && ln == NULL case).
926 			 */
927 			if (sockaddr_dl_init(&u.sdl, sizeof(u.ss),
928 			    ifp->if_index, ifp->if_type,
929 			    NULL, namelen, NULL, addrlen) == NULL) {
930 				printf("%s.%d: sockaddr_dl_init(, %zu, ) "
931 				    "failed on %s\n", __func__, __LINE__,
932 				    sizeof(u.ss), if_name(ifp));
933 			}
934 			rt_setgate(rt, &u.sa);
935 			gate = rt->rt_gateway;
936 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
937 			if (gate == NULL) {
938 				log(LOG_ERR,
939 				    "%s: rt_setgate failed on %s\n", __func__,
940 				    if_name(ifp));
941 				break;
942 			}
943 
944 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
945 			if ((rt->rt_flags & RTF_CONNECTED) != 0)
946 				break;
947 		}
948 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
949 		/*
950 		 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here.
951 		 * We don't do that here since llinfo is not ready yet.
952 		 *
953 		 * There are also couple of other things to be discussed:
954 		 * - unsolicited NA code needs improvement beforehand
955 		 * - RFC2461 says we MAY send multicast unsolicited NA
956 		 *   (7.2.6 paragraph 4), however, it also says that we
957 		 *   SHOULD provide a mechanism to prevent multicast NA storm.
958 		 *   we don't have anything like it right now.
959 		 *   note that the mechanism needs a mutual agreement
960 		 *   between proxies, which means that we need to implement
961 		 *   a new protocol, or a new kludge.
962 		 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA.
963 		 *   we need to check ip6forwarding before sending it.
964 		 *   (or should we allow proxy ND configuration only for
965 		 *   routers?  there's no mention about proxy ND from hosts)
966 		 */
967 #if 0
968 		/* XXX it does not work */
969 		if (rt->rt_flags & RTF_ANNOUNCE)
970 			nd6_na_output(ifp,
971 			      &satocsin6(rt_getkey(rt))->sin6_addr,
972 			      &satocsin6(rt_getkey(rt))->sin6_addr,
973 			      ip6_forwarding ? ND_NA_FLAG_ROUTER : 0,
974 			      1, NULL);
975 #endif
976 
977 		if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) {
978 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
979 			/*
980 			 * Address resolution isn't necessary for a point to
981 			 * point link, so we can skip this test for a p2p link.
982 			 */
983 			if (gate->sa_family != AF_LINK ||
984 			    gate->sa_len <
985 			    sockaddr_dl_measure(namelen, addrlen)) {
986 				log(LOG_DEBUG,
987 				    "nd6_rtrequest: bad gateway value: %s\n",
988 				    if_name(ifp));
989 				break;
990 			}
991 			satosdl(gate)->sdl_type = ifp->if_type;
992 			satosdl(gate)->sdl_index = ifp->if_index;
993 			RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
994 		}
995 		RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt));
996 
997 		/*
998 		 * When called from rt_ifa_addlocal, we cannot depend on that
999 		 * the address (rt_getkey(rt)) exits in the address list of the
1000 		 * interface. So check RTF_LOCAL instead.
1001 		 */
1002 		if (rt->rt_flags & RTF_LOCAL) {
1003 			if (nd6_useloopback)
1004 				rt->rt_ifp = lo0ifp;	/* XXX */
1005 			break;
1006 		}
1007 
1008 		/*
1009 		 * check if rt_getkey(rt) is an address assigned
1010 		 * to the interface.
1011 		 */
1012 		ifa = (struct ifaddr *)in6ifa_ifpwithaddr_psref(ifp,
1013 		    &satocsin6(rt_getkey(rt))->sin6_addr, &psref);
1014 		if (ifa != NULL) {
1015 			if (nd6_useloopback) {
1016 				rt->rt_ifp = lo0ifp;	/* XXX */
1017 				/*
1018 				 * Make sure rt_ifa be equal to the ifaddr
1019 				 * corresponding to the address.
1020 				 * We need this because when we refer
1021 				 * rt_ifa->ia6_flags in ip6_input, we assume
1022 				 * that the rt_ifa points to the address instead
1023 				 * of the loopback address.
1024 				 */
1025 				if (!ISSET(info->rti_flags, RTF_DONTCHANGEIFA)
1026 				    && ifa != rt->rt_ifa)
1027 					rt_replace_ifa(rt, ifa);
1028 			}
1029 		} else if (rt->rt_flags & RTF_ANNOUNCE) {
1030 			/* join solicited node multicast for proxy ND */
1031 			if (ifp->if_flags & IFF_MULTICAST) {
1032 				struct in6_addr llsol;
1033 				int error;
1034 
1035 				llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1036 				llsol.s6_addr32[0] = htonl(0xff020000);
1037 				llsol.s6_addr32[1] = 0;
1038 				llsol.s6_addr32[2] = htonl(1);
1039 				llsol.s6_addr8[12] = 0xff;
1040 				if (in6_setscope(&llsol, ifp, NULL))
1041 					goto out;
1042 				if (!in6_addmulti(&llsol, ifp, &error, 0)) {
1043 					char ip6buf[INET6_ADDRSTRLEN];
1044 					nd6log(LOG_ERR, "%s: failed to join "
1045 					    "%s (errno=%d)\n", if_name(ifp),
1046 					    IN6_PRINT(ip6buf, &llsol), error);
1047 				}
1048 			}
1049 		}
1050 	out:
1051 		ifa_release(ifa, &psref);
1052 		/*
1053 		 * If we have too many cache entries, initiate immediate
1054 		 * purging for some entries.
1055 		 */
1056 		if (rt->rt_ifp != NULL)
1057 			nd6_gc_neighbors(LLTABLE6(rt->rt_ifp), NULL);
1058 		break;
1059 	    }
1060 
1061 	case RTM_DELETE:
1062 		/* leave from solicited node multicast for proxy ND */
1063 		if ((rt->rt_flags & RTF_ANNOUNCE) != 0 &&
1064 		    (ifp->if_flags & IFF_MULTICAST) != 0) {
1065 			struct in6_addr llsol;
1066 
1067 			llsol = satocsin6(rt_getkey(rt))->sin6_addr;
1068 			llsol.s6_addr32[0] = htonl(0xff020000);
1069 			llsol.s6_addr32[1] = 0;
1070 			llsol.s6_addr32[2] = htonl(1);
1071 			llsol.s6_addr8[12] = 0xff;
1072 			if (in6_setscope(&llsol, ifp, NULL) == 0)
1073 				in6_lookup_and_delete_multi(&llsol, ifp);
1074 		}
1075 		break;
1076 	}
1077 }
1078 
1079 static void
nd6_setifflags(struct ifnet * ifp,uint32_t flags)1080 nd6_setifflags(struct ifnet *ifp, uint32_t flags)
1081 {
1082 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
1083 	struct ifaddr *ifa;
1084 	struct in6_ifaddr *ia;
1085 	int s;
1086 
1087 	if (ndi->flags & ND6_IFF_IFDISABLED && !(flags & ND6_IFF_IFDISABLED)) {
1088 		/*
1089 		 * If the interface is marked as ND6_IFF_IFDISABLED and
1090 		 * has a link-local address with IN6_IFF_DUPLICATED,
1091 		 * do not clear ND6_IFF_IFDISABLED.
1092 		 * See RFC 4862, section 5.4.5.
1093 		 */
1094 		bool duplicated_linklocal = false;
1095 
1096 		s = pserialize_read_enter();
1097 		IFADDR_READER_FOREACH(ifa, ifp) {
1098 			if (ifa->ifa_addr->sa_family != AF_INET6)
1099 				continue;
1100 			ia = (struct in6_ifaddr *)ifa;
1101 			if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1102 			    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1103 			{
1104 				duplicated_linklocal = true;
1105 				break;
1106 			}
1107 		}
1108 		pserialize_read_exit(s);
1109 
1110 		if (duplicated_linklocal) {
1111 			flags |= ND6_IFF_IFDISABLED;
1112 			log(LOG_ERR, "%s: Cannot enable an interface"
1113 			    " with a link-local address marked"
1114 			    " duplicate.\n", if_name(ifp));
1115 		} else {
1116 			ndi->flags &= ~ND6_IFF_IFDISABLED;
1117 			if (ifp->if_flags & IFF_UP)
1118 				in6_if_up(ifp);
1119 		}
1120 	} else if (!(ndi->flags & ND6_IFF_IFDISABLED) &&
1121 	    (flags & ND6_IFF_IFDISABLED))
1122 	{
1123 		struct psref psref;
1124 		int bound = curlwp_bind();
1125 
1126 		/* Mark all IPv6 addresses as tentative. */
1127 
1128 		ndi->flags |= ND6_IFF_IFDISABLED;
1129 		s = pserialize_read_enter();
1130 		IFADDR_READER_FOREACH(ifa, ifp) {
1131 			if (ifa->ifa_addr->sa_family != AF_INET6)
1132 				continue;
1133 			ifa_acquire(ifa, &psref);
1134 			pserialize_read_exit(s);
1135 
1136 			nd6_dad_stop(ifa);
1137 
1138 			ia = (struct in6_ifaddr *)ifa;
1139 			ia->ia6_flags |= IN6_IFF_TENTATIVE;
1140 
1141 			s = pserialize_read_enter();
1142 			ifa_release(ifa, &psref);
1143 		}
1144 		pserialize_read_exit(s);
1145 		curlwp_bindx(bound);
1146 	}
1147 
1148 	if (flags & ND6_IFF_AUTO_LINKLOCAL) {
1149 		if (!(ndi->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1150 			/* auto_linklocal 0->1 transition */
1151 
1152 			ndi->flags |= ND6_IFF_AUTO_LINKLOCAL;
1153 			in6_ifattach(ifp, NULL);
1154 		} else if (!(flags & ND6_IFF_IFDISABLED) &&
1155 		    ifp->if_flags & IFF_UP)
1156 		{
1157 			/*
1158 			 * When the IF already has
1159 			 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1160 			 * address is assigned, and IFF_UP, try to
1161 			 * assign one.
1162 			 */
1163 			bool haslinklocal = 0;
1164 
1165 			s = pserialize_read_enter();
1166 			IFADDR_READER_FOREACH(ifa, ifp) {
1167 				if (ifa->ifa_addr->sa_family !=AF_INET6)
1168 					continue;
1169 				ia = (struct in6_ifaddr *)ifa;
1170 				if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))){
1171 					haslinklocal = true;
1172 					break;
1173 				}
1174 			}
1175 			pserialize_read_exit(s);
1176 			if (!haslinklocal)
1177 				in6_ifattach(ifp, NULL);
1178 		}
1179 	}
1180 
1181 	ndi->flags = flags;
1182 }
1183 
1184 int
nd6_ioctl(u_long cmd,void * data,struct ifnet * ifp)1185 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp)
1186 {
1187 #ifdef OSIOCGIFINFO_IN6_90
1188 	struct in6_ndireq90 *ondi = (struct in6_ndireq90 *)data;
1189 	struct in6_ndifreq90 *ndif = (struct in6_ndifreq90 *)data;
1190 #define OND	ondi->ndi
1191 #endif
1192 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1193 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1194 	struct nd_kifinfo *ifndi = ND_IFINFO(ifp);
1195 	int error = 0;
1196 #define ND     ndi->ndi
1197 
1198 	switch (cmd) {
1199 #ifdef OSIOCSRTRFLUSH_IN6
1200 	case OSIOCGDRLST_IN6:		/* FALLTHROUGH */
1201 	case OSIOCGPRLST_IN6:		/* FALLTHROUGH */
1202 	case OSIOCSNDFLUSH_IN6:		/* FALLTHROUGH */
1203 	case OSIOCSPFXFLUSH_IN6:	/* FALLTHROUGH */
1204 	case OSIOCSRTRFLUSH_IN6:	/* FALLTHROUGH */
1205 		break;
1206 	case OSIOCGDEFIFACE_IN6:
1207 		ndif->ifindex = 0;
1208 		break;
1209 	case OSIOCSDEFIFACE_IN6:
1210 		error = ENOTSUP;
1211 		break;
1212 #endif
1213 #ifdef OSIOCGIFINFO_IN6
1214 	case OSIOCGIFINFO_IN6:		/* FALLTHROUGH */
1215 #endif
1216 #ifdef OSIOCGIFINFO_IN6_90
1217 	case OSIOCGIFINFO_IN6_90:
1218 		memset(&OND, 0, sizeof(OND));
1219 		OND.initialized = 1;
1220 		OND.chlim = ifndi->chlim;
1221 		OND.basereachable = ifndi->basereachable;
1222 		OND.retrans = ifndi->retrans;
1223 		OND.flags = ifndi->flags;
1224 		break;
1225 	case OSIOCSIFINFO_IN6_90:
1226 		/* Allow userland to set Neighbor Unreachability Detection
1227 		 * timers. */
1228 		if (OND.chlim != 0)
1229 			ifndi->chlim = OND.chlim;
1230 		if (OND.basereachable != 0 &&
1231 		    OND.basereachable != ifndi->basereachable)
1232 		{
1233 			ifndi->basereachable = OND.basereachable;
1234 			ifndi->reachable = ND_COMPUTE_RTIME(OND.basereachable);
1235 		}
1236 		if (OND.retrans != 0)
1237 			ifndi->retrans = OND.retrans;
1238 		/* Retain the old behaviour .... */
1239 		/* FALLTHROUGH */
1240 	case OSIOCSIFINFO_FLAGS_90:
1241 		nd6_setifflags(ifp, OND.flags);
1242 		break;
1243 #undef OND
1244 #endif
1245 	case SIOCGIFINFO_IN6:
1246 		ND.chlim = ifndi->chlim;
1247 		ND.basereachable = ifndi->basereachable;
1248 		ND.retrans = ifndi->retrans;
1249 		ND.flags = ifndi->flags;
1250 		break;
1251 	case SIOCSIFINFO_IN6:
1252 		/* Allow userland to set Neighbor Unreachability Detection
1253 		 * timers. */
1254 		if (ND.chlim != 0)
1255 			ifndi->chlim = ND.chlim;
1256 		if (ND.basereachable != 0 &&
1257 		    ND.basereachable != ifndi->basereachable)
1258 		{
1259 			ifndi->basereachable = ND.basereachable;
1260 			ifndi->reachable = ND_COMPUTE_RTIME(ND.basereachable);
1261 		}
1262 		if (ND.retrans != 0)
1263 			ifndi->retrans = ND.retrans;
1264 		break;
1265 	case SIOCSIFINFO_FLAGS:
1266 		nd6_setifflags(ifp, ND.flags);
1267 		break;
1268 #undef ND
1269 	case SIOCGNBRINFO_IN6:
1270 	{
1271 		struct llentry *ln;
1272 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1273 
1274 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1275 			return error;
1276 
1277 		ln = nd6_lookup(&nb_addr, ifp, false);
1278 		if (ln == NULL) {
1279 			error = EINVAL;
1280 			break;
1281 		}
1282 		nbi->state = ln->ln_state;
1283 		nbi->asked = ln->ln_asked;
1284 		nbi->isrouter = ln->ln_router;
1285 		nbi->expire = ln->ln_expire ?
1286 		    time_mono_to_wall(ln->ln_expire) : 0;
1287 		LLE_RUNLOCK(ln);
1288 
1289 		break;
1290 	}
1291 	}
1292 	return error;
1293 }
1294 
1295 void
nd6_llinfo_release_pkts(struct llentry * ln,struct ifnet * ifp)1296 nd6_llinfo_release_pkts(struct llentry *ln, struct ifnet *ifp)
1297 {
1298 	struct mbuf *m_hold, *m_hold_next;
1299 	struct sockaddr_in6 sin6;
1300 
1301 	LLE_WLOCK_ASSERT(ln);
1302 
1303 	sockaddr_in6_init(&sin6, &ln->r_l3addr.addr6, 0, 0, 0);
1304 
1305 	m_hold = ln->la_hold, ln->la_hold = NULL, ln->la_numheld = 0;
1306 
1307 	LLE_ADDREF(ln);
1308 	LLE_WUNLOCK(ln);
1309 	for (; m_hold != NULL; m_hold = m_hold_next) {
1310 		m_hold_next = m_hold->m_nextpkt;
1311 		m_hold->m_nextpkt = NULL;
1312 
1313 		/*
1314 		 * we assume ifp is not a p2p here, so
1315 		 * just set the 2nd argument as the
1316 		 * 1st one.
1317 		 */
1318 		ip6_if_output(ifp, ifp, m_hold, &sin6, NULL);
1319 	}
1320 	LLE_WLOCK(ln);
1321 	LLE_REMREF(ln);
1322 }
1323 
1324 /*
1325  * Create neighbor cache entry and cache link-layer address,
1326  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1327  */
1328 void
nd6_cache_lladdr(struct ifnet * ifp,struct in6_addr * from,char * lladdr,int lladdrlen,int type,int code)1329 nd6_cache_lladdr(
1330     struct ifnet *ifp,
1331     struct in6_addr *from,
1332     char *lladdr,
1333     int lladdrlen,
1334     int type,	/* ICMP6 type */
1335     int code	/* type dependent information */
1336 )
1337 {
1338 	struct llentry *ln = NULL;
1339 	int is_newentry;
1340 	int do_update;
1341 	int olladdr;
1342 	int llchange;
1343 	int newstate = 0;
1344 
1345 	KASSERT(ifp != NULL);
1346 	KASSERT(from != NULL);
1347 
1348 	/* nothing must be updated for unspecified address */
1349 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1350 		return;
1351 
1352 	/*
1353 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1354 	 * the caller.
1355 	 *
1356 	 * XXX If the link does not have link-layer adderss, what should
1357 	 * we do? (ifp->if_addrlen == 0)
1358 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1359 	 * description on it in NS section (RFC 2461 7.2.3).
1360 	 */
1361 
1362 	ln = nd6_lookup(from, ifp, true);
1363 	if (ln == NULL) {
1364 #if 0
1365 		/* nothing must be done if there's no lladdr */
1366 		if (!lladdr || !lladdrlen)
1367 			return NULL;
1368 #endif
1369 
1370 		ln = nd6_create(from, ifp);
1371 		is_newentry = 1;
1372 	} else {
1373 		/* do nothing if static ndp is set */
1374 		if (ln->la_flags & LLE_STATIC) {
1375 			LLE_WUNLOCK(ln);
1376 			return;
1377 		}
1378 		is_newentry = 0;
1379 	}
1380 
1381 	if (ln == NULL)
1382 		return;
1383 
1384 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
1385 	if (olladdr && lladdr) {
1386 		llchange = memcmp(lladdr, &ln->ll_addr, ifp->if_addrlen);
1387 	} else
1388 		llchange = 0;
1389 
1390 	/*
1391 	 * newentry olladdr  lladdr  llchange	(*=record)
1392 	 *	0	n	n	--	(1)
1393 	 *	0	y	n	--	(2)
1394 	 *	0	n	y	--	(3) * STALE
1395 	 *	0	y	y	n	(4) *
1396 	 *	0	y	y	y	(5) * STALE
1397 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
1398 	 *	1	--	y	--	(7) * STALE
1399 	 */
1400 
1401 	if (lladdr) {		/* (3-5) and (7) */
1402 		/*
1403 		 * Record source link-layer address
1404 		 * XXX is it dependent to ifp->if_type?
1405 		 */
1406 		memcpy(&ln->ll_addr, lladdr, ifp->if_addrlen);
1407 		ln->la_flags |= LLE_VALID;
1408 	}
1409 
1410 	if (!is_newentry) {
1411 		if ((!olladdr && lladdr) ||		/* (3) */
1412 		    (olladdr && lladdr && llchange)) {	/* (5) */
1413 			do_update = 1;
1414 			newstate = ND_LLINFO_STALE;
1415 		} else					/* (1-2,4) */
1416 			do_update = 0;
1417 	} else {
1418 		do_update = 1;
1419 		if (lladdr == NULL)			/* (6) */
1420 			newstate = ND_LLINFO_NOSTATE;
1421 		else					/* (7) */
1422 			newstate = ND_LLINFO_STALE;
1423 	}
1424 
1425 	if (do_update) {
1426 		/*
1427 		 * Update the state of the neighbor cache.
1428 		 */
1429 		ln->ln_state = newstate;
1430 
1431 		if (ln->ln_state == ND_LLINFO_STALE) {
1432 			/*
1433 			 * XXX: since nd6_output() below will cause
1434 			 * state tansition to DELAY and reset the timer,
1435 			 * we must set the timer now, although it is actually
1436 			 * meaningless.
1437 			 */
1438 			nd_set_timer(ln, ND_TIMER_GC);
1439 
1440 			nd6_llinfo_release_pkts(ln, ifp);
1441 		} else if (ln->ln_state == ND_LLINFO_INCOMPLETE) {
1442 			/* probe right away */
1443 			nd_set_timer(ln, ND_TIMER_IMMEDIATE);
1444 		}
1445 	}
1446 
1447 	/*
1448 	 * ICMP6 type dependent behavior.
1449 	 *
1450 	 * NS: clear IsRouter if new entry
1451 	 * RS: clear IsRouter
1452 	 * RA: set IsRouter if there's lladdr
1453 	 * redir: clear IsRouter if new entry
1454 	 *
1455 	 * RA case, (1):
1456 	 * The spec says that we must set IsRouter in the following cases:
1457 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1458 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1459 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1460 	 * A question arises for (1) case.  (1) case has no lladdr in the
1461 	 * neighbor cache, this is similar to (6).
1462 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1463 	 *
1464 	 * newentry olladdr  lladdr  llchange	    NS  RS  RA	redir
1465 	 *							D R
1466 	 *	0	n	n	--	(1)	c   ?     s
1467 	 *	0	y	n	--	(2)	c   s     s
1468 	 *	0	n	y	--	(3)	c   s     s
1469 	 *	0	y	y	n	(4)	c   s     s
1470 	 *	0	y	y	y	(5)	c   s     s
1471 	 *	1	--	n	--	(6) c	c 	c s
1472 	 *	1	--	y	--	(7) c	c   s	c s
1473 	 *
1474 	 *					(c=clear s=set)
1475 	 */
1476 	switch (type & 0xff) {
1477 	case ND_NEIGHBOR_SOLICIT:
1478 		/*
1479 		 * New entry must have is_router flag cleared.
1480 		 */
1481 		if (is_newentry)	/* (6-7) */
1482 			ln->ln_router = 0;
1483 		break;
1484 	case ND_REDIRECT:
1485 		/*
1486 		 * If the icmp is a redirect to a better router, always set the
1487 		 * is_router flag.  Otherwise, if the entry is newly created,
1488 		 * clear the flag.  [RFC 2461, sec 8.3]
1489 		 */
1490 		if (code == ND_REDIRECT_ROUTER)
1491 			ln->ln_router = 1;
1492 		else if (is_newentry) /* (6-7) */
1493 			ln->ln_router = 0;
1494 		break;
1495 	case ND_ROUTER_SOLICIT:
1496 		/*
1497 		 * is_router flag must always be cleared.
1498 		 */
1499 		ln->ln_router = 0;
1500 		break;
1501 	case ND_ROUTER_ADVERT:
1502 		/*
1503 		 * Mark an entry with lladdr as a router.
1504 		 */
1505 		if ((!is_newentry && (olladdr || lladdr)) ||	/* (2-5) */
1506 		    (is_newentry && lladdr)) {			/* (7) */
1507 			ln->ln_router = 1;
1508 		}
1509 		break;
1510 	}
1511 
1512 	if (do_update && lladdr != NULL) {
1513 		struct sockaddr_in6 sin6;
1514 
1515 		sockaddr_in6_init(&sin6, from, 0, 0, 0);
1516 		rt_clonedmsg(is_newentry ? RTM_ADD : RTM_CHANGE,
1517 		    NULL, sin6tosa(&sin6), lladdr, ifp);
1518 	}
1519 
1520 	if (ln != NULL)
1521 		LLE_WUNLOCK(ln);
1522 
1523 	/*
1524 	 * If we have too many cache entries, initiate immediate
1525 	 * purging for some entries.
1526 	 */
1527 	if (is_newentry)
1528 		nd6_gc_neighbors(LLTABLE6(ifp), &ln->r_l3addr.addr6);
1529 }
1530 
1531 static void
nd6_slowtimo(void * ignored_arg)1532 nd6_slowtimo(void *ignored_arg)
1533 {
1534 	struct nd_kifinfo *ndi;
1535 	struct ifnet *ifp;
1536 	struct psref psref;
1537 	int s;
1538 
1539 	SOFTNET_KERNEL_LOCK_UNLESS_NET_MPSAFE();
1540 	callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
1541 	    nd6_slowtimo, NULL);
1542 
1543 	s = pserialize_read_enter();
1544 	IFNET_READER_FOREACH(ifp) {
1545 		ndi = ND_IFINFO(ifp);
1546 		if (ndi->basereachable && /* already initialized */
1547 		    (ndi->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
1548 			if_acquire(ifp, &psref);
1549 			pserialize_read_exit(s);
1550 			/*
1551 			 * Since reachable time rarely changes by router
1552 			 * advertisements, we SHOULD insure that a new random
1553 			 * value gets recomputed at least once every few hours.
1554 			 * (RFC 2461, 6.3.4)
1555 			 */
1556 			ndi->recalctm = nd6_recalc_reachtm_interval;
1557 			ndi->reachable = ND_COMPUTE_RTIME(ndi->basereachable);
1558 			s = pserialize_read_enter();
1559 			if_release(ifp, &psref);
1560 		}
1561 	}
1562 	pserialize_read_exit(s);
1563 
1564 	SOFTNET_KERNEL_UNLOCK_UNLESS_NET_MPSAFE();
1565 }
1566 
1567 /*
1568  * Return 0 if a neighbor cache is found. Return EWOULDBLOCK if a cache is not
1569  * found and trying to resolve a neighbor; in this case the mbuf is queued in
1570  * the list. Otherwise return errno after freeing the mbuf.
1571  */
1572 int
nd6_resolve(struct ifnet * ifp,const struct rtentry * rt,struct mbuf * m,const struct sockaddr * _dst,uint8_t * lldst,size_t dstsize)1573 nd6_resolve(struct ifnet *ifp, const struct rtentry *rt, struct mbuf *m,
1574     const struct sockaddr *_dst, uint8_t *lldst, size_t dstsize)
1575 {
1576 	struct llentry *ln = NULL;
1577 	bool created = false;
1578 	const struct sockaddr_in6 *dst = satocsin6(_dst);
1579 	int error;
1580 	struct nd_kifinfo *ndi = ND_IFINFO(ifp);
1581 
1582 	/* discard the packet if IPv6 operation is disabled on the interface */
1583 	if (ndi->flags & ND6_IFF_IFDISABLED) {
1584 		m_freem(m);
1585 		return ENETDOWN; /* better error? */
1586 	}
1587 
1588 	/*
1589 	 * Address resolution or Neighbor Unreachability Detection
1590 	 * for the next hop.
1591 	 * At this point, the destination of the packet must be a unicast
1592 	 * or an anycast address(i.e. not a multicast).
1593 	 */
1594 
1595 	/* Look up the neighbor cache for the nexthop */
1596 	ln = nd6_lookup(&dst->sin6_addr, ifp, false);
1597 
1598 	if (ln != NULL && (ln->la_flags & LLE_VALID) != 0 &&
1599 	    ln->ln_state == ND_LLINFO_REACHABLE) {
1600 		/* Fast path */
1601 		memcpy(lldst, &ln->ll_addr, MIN(dstsize, ifp->if_addrlen));
1602 		LLE_RUNLOCK(ln);
1603 		return 0;
1604 	}
1605 	if (ln != NULL)
1606 		LLE_RUNLOCK(ln);
1607 
1608 	/* Slow path */
1609 	ln = nd6_lookup(&dst->sin6_addr, ifp, true);
1610 	if (ln == NULL && nd6_is_addr_neighbor(dst, ifp))  {
1611 		/*
1612 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
1613 		 * the condition below is not very efficient.  But we believe
1614 		 * it is tolerable, because this should be a rare case.
1615 		 */
1616 		ln = nd6_create(&dst->sin6_addr, ifp);
1617 		if (ln == NULL) {
1618 			char ip6buf[INET6_ADDRSTRLEN];
1619 			log(LOG_DEBUG,
1620 			    "%s: can't allocate llinfo for %s "
1621 			    "(ln=%p, rt=%p)\n", __func__,
1622 			    IN6_PRINT(ip6buf, &dst->sin6_addr), ln, rt);
1623 			m_freem(m);
1624 			return ENOBUFS;
1625 		}
1626 		created = true;
1627 	}
1628 
1629 	if (ln == NULL) {
1630 		m_freem(m);
1631 		return ENETDOWN; /* better error? */
1632 	}
1633 
1634 	error = nd_resolve(ln, rt, m, lldst, dstsize);
1635 
1636 	if (created)
1637 		nd6_gc_neighbors(LLTABLE6(ifp), &dst->sin6_addr);
1638 
1639 	return error;
1640 }
1641 
1642 int
nd6_need_cache(struct ifnet * ifp)1643 nd6_need_cache(struct ifnet *ifp)
1644 {
1645 	/*
1646 	 * XXX: we currently do not make neighbor cache on any interface
1647 	 * other than ARCnet, Ethernet, and GIF.
1648 	 *
1649 	 * RFC2893 says:
1650 	 * - unidirectional tunnels needs no ND
1651 	 */
1652 	switch (ifp->if_type) {
1653 	case IFT_ARCNET:
1654 	case IFT_ETHER:
1655 	case IFT_IEEE1394:
1656 	case IFT_CARP:
1657 	case IFT_GIF:		/* XXX need more cases? */
1658 	case IFT_IPSEC:
1659 	case IFT_PPP:
1660 	case IFT_TUNNEL:
1661 		return 1;
1662 	default:
1663 		return 0;
1664 	}
1665 }
1666 
1667 int
nd6_sysctl(int name,void * oldp,size_t * oldlenp,void * newp,size_t newlen)1668 nd6_sysctl(
1669     int name,
1670     void *oldp,	/* syscall arg, need copyout */
1671     size_t *oldlenp,
1672     void *newp,	/* syscall arg, need copyin */
1673     size_t newlen
1674 )
1675 {
1676 	int error;
1677 
1678 	if (newp)
1679 		return EPERM;
1680 
1681 	switch (name) {
1682 
1683 /* call the nd6 compat_90 hook to validate the nd6-related names */
1684 	case OICMPV6CTL_ND6_DRLIST: /* FALLTHROUGH */
1685 	case OICMPV6CTL_ND6_PRLIST:
1686 		MODULE_HOOK_CALL(net_inet6_nd_90_hook, (name), ENOPROTOOPT,
1687 		    error);
1688 		if (error == 0)
1689 			*oldlenp = 0;
1690 		return error;
1691 
1692 	case ICMPV6CTL_ND6_MAXQLEN:
1693 		return 0;
1694 	default:
1695 		return ENOPROTOOPT;
1696 	}
1697 }
1698