xref: /freebsd/sys/netinet6/nd6.c (revision 254b23eb1f540844cf2a90f2781ae4231c5701ce)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32  */
33 
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/eventhandler.h>
40 #include <sys/callout.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/mutex.h>
45 #include <sys/socket.h>
46 #include <sys/sockio.h>
47 #include <sys/time.h>
48 #include <sys/kernel.h>
49 #include <sys/protosw.h>
50 #include <sys/errno.h>
51 #include <sys/syslog.h>
52 #include <sys/rwlock.h>
53 #include <sys/queue.h>
54 #include <sys/sdt.h>
55 #include <sys/sysctl.h>
56 
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/if_private.h>
61 #include <net/if_types.h>
62 #include <net/route.h>
63 #include <net/route/route_ctl.h>
64 #include <net/route/nhop.h>
65 #include <net/vnet.h>
66 
67 #include <netinet/in.h>
68 #include <netinet/in_kdtrace.h>
69 #include <net/if_llatbl.h>
70 #include <netinet/if_ether.h>
71 #include <netinet6/in6_fib.h>
72 #include <netinet6/in6_var.h>
73 #include <netinet/ip6.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/scope6_var.h>
76 #include <netinet6/nd6.h>
77 #include <netinet6/in6_ifattach.h>
78 #include <netinet/icmp6.h>
79 #include <netinet6/send.h>
80 
81 #include <sys/limits.h>
82 
83 #include <security/mac/mac_framework.h>
84 
85 #define	ND6_PREFIX_WITH_ROUTER(pr)	!LIST_EMPTY(&(pr)->ndpr_advrtrs)
86 
87 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
88 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
89 
90 VNET_DEFINE_STATIC(int, nd6_prune) = 1;
91 #define	V_nd6_prune	VNET(nd6_prune)
92 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_PRUNE, nd6_prune,
93     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_prune), 0,
94     "Frequency in seconds of checks for expired prefixes and routers");
95 
96 VNET_DEFINE_STATIC(int, nd6_delay) = 5;
97 #define	V_nd6_delay	VNET(nd6_delay)
98 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DELAY, nd6_delay,
99     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_delay), 0,
100     "Delay in seconds before probing for reachability");
101 
102 VNET_DEFINE_STATIC(int, nd6_umaxtries) = 3;
103 #define	V_nd6_umaxtries	VNET(nd6_umaxtries)
104 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_UMAXTRIES, nd6_umaxtries,
105     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_umaxtries), 0,
106     "Number of ICMPv6 NS messages sent during reachability detection");
107 
108 VNET_DEFINE(int, nd6_mmaxtries) = 3;
109 #define	V_nd6_mmaxtries	VNET(nd6_mmaxtries)
110 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MMAXTRIES, nd6_mmaxtries,
111     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_mmaxtries), 0,
112     "Number of ICMPv6 NS messages sent during address resolution");
113 
114 VNET_DEFINE_STATIC(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage
115 							* collection timer */
116 #define	V_nd6_gctimer	VNET(nd6_gctimer)
117 
118 /* preventing too many loops in ND option parsing */
119 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
120 
121 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 16; /* max pkts cached in unresolved
122 					 * ND entries */
123 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
124 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
125 
126 #ifdef ND6_DEBUG
127 VNET_DEFINE(int, nd6_debug) = 1;
128 #else
129 VNET_DEFINE(int, nd6_debug) = 0;
130 #endif
131 #define	V_nd6_debug	VNET(nd6_debug)
132 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DEBUG, nd6_debug,
133     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_debug), 0,
134     "Log NDP debug messages");
135 
136 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
137 
138 VNET_DEFINE(struct nd_prhead, nd_prefix);
139 VNET_DEFINE(struct rwlock, nd6_lock);
140 VNET_DEFINE(uint64_t, nd6_list_genid);
141 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
142 
143 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
144 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
145 
146 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
147 
148 static bool nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
149 	struct ifnet *);
150 static void nd6_slowtimo(void *);
151 static int regen_tmpaddr(struct in6_ifaddr *);
152 static void nd6_free(struct llentry **, int);
153 static void nd6_free_redirect(const struct llentry *);
154 static void nd6_llinfo_timer(void *);
155 static void nd6_llinfo_settimer_locked(struct llentry *, long);
156 static int nd6_resolve_slow(struct ifnet *, int, int, struct mbuf *,
157     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
158 static int nd6_need_cache(struct ifnet *);
159 
160 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
161 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
162 
163 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
164 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
165 
166 static void
nd6_lle_event(void * arg __unused,struct llentry * lle,int evt)167 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
168 {
169 	struct rt_addrinfo rtinfo;
170 	struct sockaddr_in6 dst;
171 	struct sockaddr_dl gw;
172 	struct ifnet *ifp;
173 	int type;
174 	int fibnum;
175 
176 	LLE_WLOCK_ASSERT(lle);
177 
178 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
179 		return;
180 
181 	switch (evt) {
182 	case LLENTRY_RESOLVED:
183 		type = RTM_ADD;
184 		KASSERT(lle->la_flags & LLE_VALID,
185 		    ("%s: %p resolved but not valid?", __func__, lle));
186 		break;
187 	case LLENTRY_EXPIRED:
188 		type = RTM_DELETE;
189 		break;
190 	default:
191 		return;
192 	}
193 
194 	ifp = lltable_get_ifp(lle->lle_tbl);
195 
196 	bzero(&dst, sizeof(dst));
197 	bzero(&gw, sizeof(gw));
198 	bzero(&rtinfo, sizeof(rtinfo));
199 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
200 	dst.sin6_scope_id = in6_getscopezone(ifp,
201 	    in6_addrscope(&dst.sin6_addr));
202 	gw.sdl_len = sizeof(struct sockaddr_dl);
203 	gw.sdl_family = AF_LINK;
204 	gw.sdl_alen = ifp->if_addrlen;
205 	gw.sdl_index = ifp->if_index;
206 	gw.sdl_type = ifp->if_type;
207 	if (evt == LLENTRY_RESOLVED)
208 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
209 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
210 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
211 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
212 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
213 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
214 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
215 }
216 
217 /*
218  * A handler for interface link layer address change event.
219  */
220 static void
nd6_iflladdr(void * arg __unused,struct ifnet * ifp)221 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
222 {
223 	struct ifaddr *ifa;
224 	struct epoch_tracker et;
225 
226 	/* XXXGL: ??? */
227 	if (ifp->if_inet6 == NULL)
228 		return;
229 
230 	lltable_update_ifaddr(LLTABLE6(ifp));
231 
232 	if ((ifp->if_flags & IFF_UP) == 0)
233 		return;
234 
235 	/*
236 	 * Sends gratuitous NAs for each ifaddr to notify other
237 	 * nodes about the address change.
238 	 */
239 	NET_EPOCH_ENTER(et);
240 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
241 		if (ifa->ifa_addr->sa_family == AF_INET6 &&
242 		    ! IN6_IS_ADDR_MULTICAST(IFA_IN6(ifa)))
243 			nd6_grand_start(ifa, ND6_QUEUE_FLAG_LLADDR);
244 	}
245 	NET_EPOCH_EXIT(et);
246 }
247 
248 void
nd6_init(void)249 nd6_init(void)
250 {
251 
252 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
253 	rw_init(&V_nd6_lock, "nd6 list");
254 
255 	LIST_INIT(&V_nd_prefix);
256 	nd6_defrouter_init();
257 
258 	/* Start timers. */
259 	callout_init(&V_nd6_slowtimo_ch, 1);
260 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
261 	    nd6_slowtimo, curvnet);
262 
263 	callout_init(&V_nd6_timer_ch, 1);
264 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
265 
266 	nd6_dad_init();
267 	if (IS_DEFAULT_VNET(curvnet)) {
268 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
269 		    NULL, EVENTHANDLER_PRI_ANY);
270 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
271 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
272 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
273 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
274 	}
275 }
276 
277 #ifdef VIMAGE
278 void
nd6_destroy(void)279 nd6_destroy(void)
280 {
281 
282 	callout_drain(&V_nd6_slowtimo_ch);
283 	callout_drain(&V_nd6_timer_ch);
284 	if (IS_DEFAULT_VNET(curvnet)) {
285 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
286 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
287 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
288 	}
289 	rw_destroy(&V_nd6_lock);
290 	mtx_destroy(&V_nd6_onlink_mtx);
291 }
292 #endif
293 
294 void
nd6_ifattach(struct ifnet * ifp)295 nd6_ifattach(struct ifnet *ifp)
296 {
297 	struct in6_ifextra *nd = ifp->if_inet6;
298 
299 	nd->nd_linkmtu = 0;
300 	nd->nd_maxmtu = ifp->if_mtu;
301 	nd->nd_basereachable = REACHABLE_TIME;
302 	nd->nd_reachable = ND_COMPUTE_RTIME(nd->nd_basereachable);
303 	nd->nd_retrans = RETRANS_TIMER;
304 	nd->nd_recalc_timer = 0;
305 	nd->nd_dad_failures = 0;
306 	nd->nd_curhoplimit = IPV6_DEFHLIM;
307 
308 	nd->nd_flags = ND6_IFF_PERFORMNUD;
309 
310 	/* Set IPv6 disabled on all interfaces but loopback by default. */
311 	if ((ifp->if_flags & IFF_LOOPBACK) == 0) {
312 		nd->nd_flags |= ND6_IFF_IFDISABLED;
313 		if (V_ip6_no_radr)
314 			nd->nd_flags |= ND6_IFF_NO_RADR;
315 		if (V_ip6_use_stableaddr)
316 			nd->nd_flags |= ND6_IFF_STABLEADDR;
317 	}
318 
319 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
320 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
321 	 * default regardless of the V_ip6_auto_linklocal configuration to
322 	 * give a reasonable default behavior.
323 	 */
324 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE &&
325 	    ifp->if_type != IFT_WIREGUARD) || (ifp->if_flags & IFF_LOOPBACK))
326 		nd->nd_flags |= ND6_IFF_AUTO_LINKLOCAL;
327 	/*
328 	 * A loopback interface does not need to accept RTADV.
329 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
330 	 * default regardless of the V_ip6_accept_rtadv configuration to
331 	 * prevent the interface from accepting RA messages arrived
332 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
333 	 */
334 	if (V_ip6_accept_rtadv &&
335 	    !(ifp->if_flags & IFF_LOOPBACK) &&
336 	    (ifp->if_type != IFT_BRIDGE)) {
337 			nd->nd_flags |= ND6_IFF_ACCEPT_RTADV;
338 			/* If we globally accept rtadv, assume IPv6 on. */
339 			nd->nd_flags &= ~ND6_IFF_IFDISABLED;
340 	}
341 
342 	/* nd6 queue initialization */
343 	TAILQ_INIT(&nd->nd_queue);
344 }
345 
346 void
nd6_ifdetach(struct ifnet * ifp)347 nd6_ifdetach(struct ifnet *ifp)
348 {
349 	struct epoch_tracker et;
350 	struct ifaddr *ifa, *next;
351 
352 	NET_EPOCH_ENTER(et);
353 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
354 		if (ifa->ifa_addr->sa_family != AF_INET6)
355 			continue;
356 
357 		/* make sure there are no queued ND6 */
358 		nd6_queue_stop(ifa);
359 
360 		/* stop DAD processing */
361 		nd6_dad_stop(ifa);
362 	}
363 	NET_EPOCH_EXIT(et);
364 }
365 
366 /*
367  * Reset ND level link MTU. This function is called when the physical MTU
368  * changes, which means we might have to adjust the ND level MTU.
369  * XXX todo: do not maintain copy of ifp->if_mtu in if_inet6->nd_maxmtu.
370  */
371 void
nd6_setmtu(struct ifnet * ifp)372 nd6_setmtu(struct ifnet *ifp)
373 {
374 	struct in6_ifextra *ndi = ifp->if_inet6;
375 	uint32_t omaxmtu;
376 
377 	/* XXXGL: safety against IFT_PFSYNC & IFT_PFLOG */
378 	if (ndi == NULL)
379 		return;
380 
381 	omaxmtu = ndi->nd_maxmtu;
382 	ndi->nd_maxmtu = ifp->if_mtu;
383 
384 	/*
385 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
386 	 * undesirable situation.  We thus notify the operator of the change
387 	 * explicitly.  The check for omaxmtu is necessary to restrict the
388 	 * log to the case of changing the MTU, not initializing it.
389 	 */
390 	if (omaxmtu >= IPV6_MMTU && ndi->nd_maxmtu < IPV6_MMTU) {
391 		log(LOG_NOTICE, "%s: "
392 		    "new link MTU on %s (%lu) is too small for IPv6\n",
393 		    __func__, if_name(ifp), (unsigned long)ndi->nd_maxmtu);
394 	}
395 }
396 
397 void
nd6_option_init(void * opt,int icmp6len,union nd_opts * ndopts)398 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
399 {
400 
401 	bzero(ndopts, sizeof(*ndopts));
402 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
403 	ndopts->nd_opts_last
404 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
405 
406 	if (icmp6len == 0) {
407 		ndopts->nd_opts_done = 1;
408 		ndopts->nd_opts_search = NULL;
409 	}
410 }
411 
412 /*
413  * Take one ND option.
414  */
415 struct nd_opt_hdr *
nd6_option(union nd_opts * ndopts)416 nd6_option(union nd_opts *ndopts)
417 {
418 	struct nd_opt_hdr *nd_opt;
419 	int olen;
420 
421 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
422 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
423 	    __func__));
424 	if (ndopts->nd_opts_search == NULL)
425 		return NULL;
426 	if (ndopts->nd_opts_done)
427 		return NULL;
428 
429 	nd_opt = ndopts->nd_opts_search;
430 
431 	/* make sure nd_opt_len is inside the buffer */
432 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
433 		bzero(ndopts, sizeof(*ndopts));
434 		return NULL;
435 	}
436 
437 	olen = nd_opt->nd_opt_len << 3;
438 	/*
439 	 * RFC 4861 section 6.1.2: All included options
440 	 * must have a length that is greater than zero.
441 	 */
442 	if (olen == 0) {
443 		bzero(ndopts, sizeof(*ndopts));
444 		return NULL;
445 	}
446 
447 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
448 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
449 		/* option overruns the end of buffer, invalid */
450 		bzero(ndopts, sizeof(*ndopts));
451 		return NULL;
452 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
453 		/* reached the end of options chain */
454 		ndopts->nd_opts_done = 1;
455 		ndopts->nd_opts_search = NULL;
456 	}
457 	return nd_opt;
458 }
459 
460 /*
461  * Parse multiple ND options.
462  * This function is much easier to use, for ND routines that do not need
463  * multiple options of the same type.
464  */
465 int
nd6_options(union nd_opts * ndopts)466 nd6_options(union nd_opts *ndopts)
467 {
468 	struct nd_opt_hdr *nd_opt;
469 	int i = 0;
470 
471 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
472 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
473 	    __func__));
474 	if (ndopts->nd_opts_search == NULL)
475 		return 0;
476 
477 	while (1) {
478 		nd_opt = nd6_option(ndopts);
479 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
480 			/*
481 			 * Message validation requires that all included
482 			 * options have a length that is greater than zero.
483 			 */
484 			ICMP6STAT_INC(icp6s_nd_badopt);
485 			bzero(ndopts, sizeof(*ndopts));
486 			return -1;
487 		}
488 
489 		if (nd_opt == NULL)
490 			goto skip1;
491 
492 		switch (nd_opt->nd_opt_type) {
493 		case ND_OPT_SOURCE_LINKADDR:
494 		case ND_OPT_TARGET_LINKADDR:
495 		case ND_OPT_MTU:
496 		case ND_OPT_REDIRECTED_HEADER:
497 		case ND_OPT_NONCE:
498 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
499 				nd6log((LOG_INFO,
500 				    "duplicated ND6 option found (type=%d)\n",
501 				    nd_opt->nd_opt_type));
502 				/* XXX bark? */
503 			} else {
504 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
505 					= nd_opt;
506 			}
507 			break;
508 		case ND_OPT_PREFIX_INFORMATION:
509 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
510 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
511 					= nd_opt;
512 			}
513 			ndopts->nd_opts_pi_end =
514 				(struct nd_opt_prefix_info *)nd_opt;
515 			break;
516 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
517 		case ND_OPT_RDNSS:	/* RFC 6106 */
518 		case ND_OPT_DNSSL:	/* RFC 6106 */
519 			/*
520 			 * Silently ignore options we know and do not care about
521 			 * in the kernel.
522 			 */
523 			break;
524 		default:
525 			/*
526 			 * Unknown options must be silently ignored,
527 			 * to accommodate future extension to the protocol.
528 			 */
529 			nd6log((LOG_DEBUG,
530 			    "nd6_options: unsupported option %d - "
531 			    "option ignored\n", nd_opt->nd_opt_type));
532 		}
533 
534 skip1:
535 		i++;
536 		if (i > V_nd6_maxndopt) {
537 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
538 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
539 			break;
540 		}
541 
542 		if (ndopts->nd_opts_done)
543 			break;
544 	}
545 
546 	return 0;
547 }
548 
549 /*
550  * ND6 timer routine to handle ND6 entries
551  */
552 static void
nd6_llinfo_settimer_locked(struct llentry * ln,long tick)553 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
554 {
555 	int canceled;
556 
557 	LLE_WLOCK_ASSERT(ln);
558 
559 	/* Do not schedule timers for child LLEs. */
560 	if (ln->la_flags & LLE_CHILD)
561 		return;
562 
563 	if (tick < 0) {
564 		ln->la_expire = 0;
565 		ln->ln_ntick = 0;
566 		canceled = callout_stop(&ln->lle_timer);
567 	} else {
568 		ln->la_expire = time_uptime + tick / hz;
569 		LLE_ADDREF(ln);
570 		if (tick > INT_MAX) {
571 			ln->ln_ntick = tick - INT_MAX;
572 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
573 			    nd6_llinfo_timer, ln);
574 		} else {
575 			ln->ln_ntick = 0;
576 			canceled = callout_reset(&ln->lle_timer, tick,
577 			    nd6_llinfo_timer, ln);
578 		}
579 	}
580 	if (canceled > 0)
581 		LLE_REMREF(ln);
582 }
583 
584 /*
585  * Gets source address of the first packet in hold queue
586  * and stores it in @src.
587  * Returns pointer to @src (if hold queue is not empty) or NULL.
588  *
589  * Set noinline to be dtrace-friendly
590  */
591 static __noinline struct in6_addr *
nd6_llinfo_get_holdsrc(struct llentry * ln,struct in6_addr * src)592 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
593 {
594 	struct ip6_hdr hdr;
595 	struct mbuf *m;
596 
597 	if (ln->la_hold == NULL)
598 		return (NULL);
599 
600 	/*
601 	 * assume every packet in la_hold has the same IP header
602 	 */
603 	m = ln->la_hold;
604 	if (sizeof(hdr) > m->m_len)
605 		return (NULL);
606 
607 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
608 	*src = hdr.ip6_src;
609 
610 	return (src);
611 }
612 
613 /*
614  * Checks if we need to switch from STALE state.
615  *
616  * RFC 4861 requires switching from STALE to DELAY state
617  * on first packet matching entry, waiting V_nd6_delay and
618  * transition to PROBE state (if upper layer confirmation was
619  * not received).
620  *
621  * This code performs a bit differently:
622  * On packet hit we don't change state (but desired state
623  * can be guessed by control plane). However, after V_nd6_delay
624  * seconds code will transition to PROBE state (so DELAY state
625  * is kinda skipped in most situations).
626  *
627  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
628  * we perform the following upon entering STALE state:
629  *
630  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
631  * if packet was transmitted at the start of given interval, we
632  * would be able to switch to PROBE state in V_nd6_delay seconds
633  * as user expects.
634  *
635  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
636  * lle in STALE state (remaining timer value stored in lle_remtime).
637  *
638  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
639  * seconds ago.
640  *
641  * Returns non-zero value if the entry is still STALE (storing
642  * the next timer interval in @pdelay).
643  *
644  * Returns zero value if original timer expired or we need to switch to
645  * PROBE (store that in @do_switch variable).
646  */
647 static int
nd6_is_stale(struct llentry * lle,long * pdelay,int * do_switch)648 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
649 {
650 	int nd_delay, nd_gctimer;
651 	time_t lle_hittime;
652 	long delay;
653 
654 	*do_switch = 0;
655 	nd_gctimer = V_nd6_gctimer;
656 	nd_delay = V_nd6_delay;
657 
658 	lle_hittime = llentry_get_hittime(lle);
659 
660 	if (lle_hittime == 0) {
661 		/*
662 		 * Datapath feedback has been requested upon entering
663 		 * STALE state. No packets has been passed using this lle.
664 		 * Ask for the timer reschedule and keep STALE state.
665 		 */
666 		delay = (long)(MIN(nd_gctimer, nd_delay));
667 		delay *= hz;
668 		if (lle->lle_remtime > delay)
669 			lle->lle_remtime -= delay;
670 		else {
671 			delay = lle->lle_remtime;
672 			lle->lle_remtime = 0;
673 		}
674 
675 		if (delay == 0) {
676 			/*
677 			 * The original ng6_gctime timeout ended,
678 			 * no more rescheduling.
679 			 */
680 			return (0);
681 		}
682 
683 		*pdelay = delay;
684 		return (1);
685 	}
686 
687 	/*
688 	 * Packet received. Verify timestamp
689 	 */
690 	delay = (long)(time_uptime - lle_hittime);
691 	if (delay < nd_delay) {
692 		/*
693 		 * V_nd6_delay still not passed since the first
694 		 * hit in STALE state.
695 		 * Reschedule timer and return.
696 		 */
697 		*pdelay = (long)(nd_delay - delay) * hz;
698 		return (1);
699 	}
700 
701 	/* Request switching to probe */
702 	*do_switch = 1;
703 	return (0);
704 }
705 
706 /*
707  * Switch @lle state to new state optionally arming timers.
708  *
709  * Set noinline to be dtrace-friendly
710  */
711 __noinline void
nd6_llinfo_setstate(struct llentry * lle,int newstate)712 nd6_llinfo_setstate(struct llentry *lle, int newstate)
713 {
714 	struct ifnet *ifp;
715 	int nd_gctimer, nd_delay;
716 	long delay, remtime;
717 
718 	delay = 0;
719 	remtime = 0;
720 
721 	switch (newstate) {
722 	case ND6_LLINFO_INCOMPLETE:
723 		ifp = lle->lle_tbl->llt_ifp;
724 		delay = (long)ifp->if_inet6->nd_retrans * hz / 1000;
725 		break;
726 	case ND6_LLINFO_REACHABLE:
727 		if (!ND6_LLINFO_PERMANENT(lle)) {
728 			ifp = lle->lle_tbl->llt_ifp;
729 			delay = (long)ifp->if_inet6->nd_reachable * hz;
730 		}
731 		break;
732 	case ND6_LLINFO_STALE:
733 
734 		llentry_request_feedback(lle);
735 		nd_delay = V_nd6_delay;
736 		nd_gctimer = V_nd6_gctimer;
737 
738 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
739 		remtime = (long)nd_gctimer * hz - delay;
740 		break;
741 	case ND6_LLINFO_DELAY:
742 		lle->la_asked = 0;
743 		delay = (long)V_nd6_delay * hz;
744 		break;
745 	}
746 
747 	if (delay > 0)
748 		nd6_llinfo_settimer_locked(lle, delay);
749 
750 	lle->lle_remtime = remtime;
751 	lle->ln_state = newstate;
752 }
753 
754 /*
755  * Timer-dependent part of nd state machine.
756  *
757  * Set noinline to be dtrace-friendly
758  */
759 static __noinline void
nd6_llinfo_timer(void * arg)760 nd6_llinfo_timer(void *arg)
761 {
762 	struct epoch_tracker et;
763 	struct llentry *ln;
764 	struct in6_addr *dst, *pdst, *psrc, src;
765 	struct ifnet *ifp;
766 	struct in6_ifextra *ndi;
767 	int do_switch, send_ns;
768 	long delay;
769 
770 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
771 	ln = (struct llentry *)arg;
772 	ifp = lltable_get_ifp(ln->lle_tbl);
773 	CURVNET_SET(ifp->if_vnet);
774 
775 	ND6_RLOCK();
776 	LLE_WLOCK(ln);
777 	if (callout_pending(&ln->lle_timer)) {
778 		/*
779 		 * Here we are a bit odd here in the treatment of
780 		 * active/pending. If the pending bit is set, it got
781 		 * rescheduled before I ran. The active
782 		 * bit we ignore, since if it was stopped
783 		 * in ll_tablefree() and was currently running
784 		 * it would have return 0 so the code would
785 		 * not have deleted it since the callout could
786 		 * not be stopped so we want to go through
787 		 * with the delete here now. If the callout
788 		 * was restarted, the pending bit will be back on and
789 		 * we just want to bail since the callout_reset would
790 		 * return 1 and our reference would have been removed
791 		 * by nd6_llinfo_settimer_locked above since canceled
792 		 * would have been 1.
793 		 */
794 		LLE_WUNLOCK(ln);
795 		ND6_RUNLOCK();
796 		CURVNET_RESTORE();
797 		return;
798 	}
799 	NET_EPOCH_ENTER(et);
800 	ndi = ifp->if_inet6;
801 	send_ns = 0;
802 	dst = &ln->r_l3addr.addr6;
803 	pdst = dst;
804 
805 	if (ln->ln_ntick > 0) {
806 		if (ln->ln_ntick > INT_MAX) {
807 			ln->ln_ntick -= INT_MAX;
808 			nd6_llinfo_settimer_locked(ln, INT_MAX);
809 		} else {
810 			ln->ln_ntick = 0;
811 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
812 		}
813 		goto done;
814 	}
815 
816 	if (ln->la_flags & LLE_STATIC) {
817 		goto done;
818 	}
819 
820 	if (ln->la_flags & LLE_DELETED) {
821 		nd6_free(&ln, 0);
822 		goto done;
823 	}
824 
825 	switch (ln->ln_state) {
826 	case ND6_LLINFO_INCOMPLETE:
827 		if (ln->la_asked < V_nd6_mmaxtries) {
828 			ln->la_asked++;
829 			send_ns = 1;
830 			/* Send NS to multicast address */
831 			pdst = NULL;
832 		} else {
833 			struct mbuf *m;
834 
835 			ICMP6STAT_ADD(icp6s_dropped, ln->la_numheld);
836 
837 			m = ln->la_hold;
838 			if (m != NULL) {
839 				/*
840 				 * assuming every packet in la_hold has the
841 				 * same IP header.  Send error after unlock.
842 				 */
843 				ln->la_hold = m->m_nextpkt;
844 				m->m_nextpkt = NULL;
845 				ln->la_numheld--;
846 			}
847 			nd6_free(&ln, 0);
848 			if (m != NULL) {
849 				struct mbuf *n = m;
850 
851 				/*
852 				 * if there are any ummapped mbufs, we
853 				 * must free them, rather than using
854 				 * them for an ICMP, as they cannot be
855 				 * checksummed.
856 				 */
857 				while ((n = n->m_next) != NULL) {
858 					if (n->m_flags & M_EXTPG)
859 						break;
860 				}
861 				if (n != NULL) {
862 					m_freem(m);
863 					m = NULL;
864 				} else {
865 					icmp6_error2(m, ICMP6_DST_UNREACH,
866 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
867 				}
868 			}
869 		}
870 		break;
871 	case ND6_LLINFO_REACHABLE:
872 		if (!ND6_LLINFO_PERMANENT(ln))
873 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
874 		break;
875 
876 	case ND6_LLINFO_STALE:
877 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
878 			/*
879 			 * No packet has used this entry and GC timeout
880 			 * has not been passed. Reschedule timer and
881 			 * return.
882 			 */
883 			nd6_llinfo_settimer_locked(ln, delay);
884 			break;
885 		}
886 
887 		if (do_switch == 0) {
888 			/*
889 			 * GC timer has ended and entry hasn't been used.
890 			 * Run Garbage collector (RFC 4861, 5.3)
891 			 */
892 			if (!ND6_LLINFO_PERMANENT(ln))
893 				nd6_free(&ln, 1);
894 			break;
895 		}
896 
897 		/* Entry has been used AND delay timer has ended. */
898 
899 		/* FALLTHROUGH */
900 
901 	case ND6_LLINFO_DELAY:
902 		if ((ndi->nd_flags & ND6_IFF_PERFORMNUD) != 0) {
903 			/* We need NUD */
904 			ln->la_asked = 1;
905 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
906 			send_ns = 1;
907 		} else
908 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
909 		break;
910 	case ND6_LLINFO_PROBE:
911 		if (ln->la_asked < V_nd6_umaxtries) {
912 			ln->la_asked++;
913 			send_ns = 1;
914 		} else {
915 			nd6_free(&ln, 0);
916 		}
917 		break;
918 	default:
919 		panic("%s: paths in a dark night can be confusing: %d",
920 		    __func__, ln->ln_state);
921 	}
922 done:
923 	if (ln != NULL)
924 		ND6_RUNLOCK();
925 	if (send_ns != 0) {
926 		nd6_llinfo_settimer_locked(ln,
927 		    (long)ndi->nd_retrans * hz / 1000);
928 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
929 		LLE_FREE_LOCKED(ln);
930 		ln = NULL;
931 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
932 	}
933 
934 	if (ln != NULL)
935 		LLE_FREE_LOCKED(ln);
936 	NET_EPOCH_EXIT(et);
937 	CURVNET_RESTORE();
938 }
939 
940 /*
941  * ND6 timer routine to expire default route list and prefix list
942  */
943 void
nd6_timer(void * arg)944 nd6_timer(void *arg)
945 {
946 	CURVNET_SET((struct vnet *) arg);
947 	struct epoch_tracker et;
948 	struct nd_prhead prl;
949 	struct nd_prefix *pr, *npr;
950 	struct ifnet *ifp;
951 	struct in6_ifaddr *ia6, *nia6;
952 	uint64_t genid;
953 
954 	LIST_INIT(&prl);
955 
956 	NET_EPOCH_ENTER(et);
957 	nd6_defrouter_timer();
958 
959 	/*
960 	 * expire interface addresses.
961 	 * in the past the loop was inside prefix expiry processing.
962 	 * However, from a stricter speci-confrmance standpoint, we should
963 	 * rather separate address lifetimes and prefix lifetimes.
964 	 *
965 	 * XXXRW: in6_ifaddrhead locking.
966 	 */
967   addrloop:
968 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
969 		/* check address lifetime */
970 		if (IFA6_IS_INVALID(ia6)) {
971 			int regen = 0;
972 
973 			/*
974 			 * If the expiring address is temporary, try
975 			 * regenerating a new one.  This would be useful when
976 			 * we suspended a laptop PC, then turned it on after a
977 			 * period that could invalidate all temporary
978 			 * addresses.  Although we may have to restart the
979 			 * loop (see below), it must be after purging the
980 			 * address.  Otherwise, we'd see an infinite loop of
981 			 * regeneration.
982 			 */
983 			if (V_ip6_use_tempaddr &&
984 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
985 				if (regen_tmpaddr(ia6) == 0)
986 					regen = 1;
987 			}
988 
989 			in6_purgeaddr(&ia6->ia_ifa);
990 
991 			if (regen)
992 				goto addrloop; /* XXX: see below */
993 		} else if (IFA6_IS_DEPRECATED(ia6)) {
994 			int oldflags = ia6->ia6_flags;
995 
996 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
997 
998 			/*
999 			 * If a temporary address has just become deprecated,
1000 			 * regenerate a new one if possible.
1001 			 */
1002 			if (V_ip6_use_tempaddr &&
1003 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1004 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
1005 				if (regen_tmpaddr(ia6) == 0) {
1006 					/*
1007 					 * A new temporary address is
1008 					 * generated.
1009 					 * XXX: this means the address chain
1010 					 * has changed while we are still in
1011 					 * the loop.  Although the change
1012 					 * would not cause disaster (because
1013 					 * it's not a deletion, but an
1014 					 * addition,) we'd rather restart the
1015 					 * loop just for safety.  Or does this
1016 					 * significantly reduce performance??
1017 					 */
1018 					goto addrloop;
1019 				}
1020 			}
1021 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
1022 			/*
1023 			 * Schedule DAD for a tentative address.  This happens
1024 			 * if the interface was down or not running
1025 			 * when the address was configured.
1026 			 */
1027 			int delay;
1028 
1029 			delay = arc4random() %
1030 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1031 			nd6_dad_start((struct ifaddr *)ia6, delay);
1032 		} else {
1033 			/*
1034 			 * Check status of the interface.  If it is down,
1035 			 * mark the address as tentative for future DAD.
1036 			 */
1037 			ifp = ia6->ia_ifp;
1038 			if ((ifp->if_inet6->nd_flags & ND6_IFF_NO_DAD) == 0 &&
1039 			    ((ifp->if_flags & IFF_UP) == 0 ||
1040 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1041 			    (ifp->if_inet6->nd_flags & ND6_IFF_IFDISABLED))){
1042 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1043 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1044 			}
1045 
1046 			/*
1047 			 * A new RA might have made a deprecated address
1048 			 * preferred.
1049 			 */
1050 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1051 		}
1052 	}
1053 	NET_EPOCH_EXIT(et);
1054 
1055 	ND6_WLOCK();
1056 restart:
1057 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1058 		/*
1059 		 * Expire prefixes. Since the pltime is only used for
1060 		 * autoconfigured addresses, pltime processing for prefixes is
1061 		 * not necessary.
1062 		 *
1063 		 * Only unlink after all derived addresses have expired. This
1064 		 * may not occur until two hours after the prefix has expired
1065 		 * per RFC 4862. If the prefix expires before its derived
1066 		 * addresses, mark it off-link. This will be done automatically
1067 		 * after unlinking if no address references remain.
1068 		 */
1069 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1070 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1071 			continue;
1072 
1073 		if (pr->ndpr_addrcnt == 0) {
1074 			nd6_prefix_unlink(pr, &prl);
1075 			continue;
1076 		}
1077 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1078 			genid = V_nd6_list_genid;
1079 			nd6_prefix_ref(pr);
1080 			ND6_WUNLOCK();
1081 			ND6_ONLINK_LOCK();
1082 			(void)nd6_prefix_offlink(pr);
1083 			ND6_ONLINK_UNLOCK();
1084 			ND6_WLOCK();
1085 			nd6_prefix_rele(pr);
1086 			if (genid != V_nd6_list_genid)
1087 				goto restart;
1088 		}
1089 	}
1090 	ND6_WUNLOCK();
1091 
1092 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1093 		LIST_REMOVE(pr, ndpr_entry);
1094 		nd6_prefix_del(pr);
1095 	}
1096 
1097 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1098 	    nd6_timer, curvnet);
1099 
1100 	CURVNET_RESTORE();
1101 }
1102 
1103 /*
1104  * ia6 - deprecated/invalidated temporary address
1105  */
1106 static int
regen_tmpaddr(struct in6_ifaddr * ia6)1107 regen_tmpaddr(struct in6_ifaddr *ia6)
1108 {
1109 	struct ifaddr *ifa;
1110 	struct ifnet *ifp;
1111 	struct in6_ifaddr *public_ifa6 = NULL;
1112 
1113 	NET_EPOCH_ASSERT();
1114 
1115 	ifp = ia6->ia_ifa.ifa_ifp;
1116 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1117 		struct in6_ifaddr *it6;
1118 
1119 		if (ifa->ifa_addr->sa_family != AF_INET6)
1120 			continue;
1121 
1122 		it6 = (struct in6_ifaddr *)ifa;
1123 
1124 		/* ignore no autoconf addresses. */
1125 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1126 			continue;
1127 
1128 		/* ignore autoconf addresses with different prefixes. */
1129 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1130 			continue;
1131 
1132 		/*
1133 		 * Now we are looking at an autoconf address with the same
1134 		 * prefix as ours.  If the address is temporary and is still
1135 		 * preferred, do not create another one.  It would be rare, but
1136 		 * could happen, for example, when we resume a laptop PC after
1137 		 * a long period.
1138 		 */
1139 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1140 		    !IFA6_IS_DEPRECATED(it6)) {
1141 			public_ifa6 = NULL;
1142 			break;
1143 		}
1144 
1145 		/*
1146 		 * This is a public autoconf address that has the same prefix
1147 		 * as ours.  If it is preferred, keep it.  We can't break the
1148 		 * loop here, because there may be a still-preferred temporary
1149 		 * address with the prefix.
1150 		 */
1151 		if (!IFA6_IS_DEPRECATED(it6))
1152 			public_ifa6 = it6;
1153 	}
1154 	if (public_ifa6 != NULL)
1155 		ifa_ref(&public_ifa6->ia_ifa);
1156 
1157 	if (public_ifa6 != NULL) {
1158 		int e;
1159 
1160 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1161 			ifa_free(&public_ifa6->ia_ifa);
1162 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1163 			    " tmp addr,errno=%d\n", e);
1164 			return (-1);
1165 		}
1166 		ifa_free(&public_ifa6->ia_ifa);
1167 		return (0);
1168 	}
1169 
1170 	return (-1);
1171 }
1172 
1173 /*
1174  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1175  * cache entries are freed in in6_domifdetach().
1176  */
1177 void
nd6_purge(struct ifnet * ifp)1178 nd6_purge(struct ifnet *ifp)
1179 {
1180 	struct nd_prhead prl;
1181 	struct nd_prefix *pr, *npr;
1182 
1183 	LIST_INIT(&prl);
1184 
1185 	/* Purge default router list entries toward ifp. */
1186 	nd6_defrouter_purge(ifp);
1187 
1188 	ND6_WLOCK();
1189 	/*
1190 	 * Remove prefixes on ifp. We should have already removed addresses on
1191 	 * this interface, so no addresses should be referencing these prefixes.
1192 	 */
1193 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1194 		if (pr->ndpr_ifp == ifp)
1195 			nd6_prefix_unlink(pr, &prl);
1196 	}
1197 	ND6_WUNLOCK();
1198 
1199 	/* Delete the unlinked prefix objects. */
1200 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1201 		LIST_REMOVE(pr, ndpr_entry);
1202 		nd6_prefix_del(pr);
1203 	}
1204 
1205 	/* cancel default outgoing interface setting */
1206 	if (V_nd6_defifindex == ifp->if_index)
1207 		nd6_setdefaultiface(0);
1208 
1209 	if (ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV) {
1210 		/* Refresh default router list. */
1211 		defrouter_select_fib(ifp->if_fib);
1212 	}
1213 }
1214 
1215 /*
1216  * the caller acquires and releases the lock on the lltbls
1217  * Returns the llentry locked
1218  */
1219 struct llentry *
nd6_lookup(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1220 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1221 {
1222 	struct sockaddr_in6 sin6;
1223 	struct llentry *ln;
1224 
1225 	bzero(&sin6, sizeof(sin6));
1226 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1227 	sin6.sin6_family = AF_INET6;
1228 	sin6.sin6_addr = *addr6;
1229 
1230 	LLTABLE_RLOCK_ASSERT(LLTABLE6(ifp));
1231 
1232 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1233 
1234 	return (ln);
1235 }
1236 
1237 static struct llentry *
nd6_alloc(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1238 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1239 {
1240 	struct sockaddr_in6 sin6;
1241 	struct llentry *ln;
1242 
1243 	bzero(&sin6, sizeof(sin6));
1244 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1245 	sin6.sin6_family = AF_INET6;
1246 	sin6.sin6_addr = *addr6;
1247 
1248 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1249 	if (ln != NULL)
1250 		ln->ln_state = ND6_LLINFO_NOSTATE;
1251 
1252 	return (ln);
1253 }
1254 
1255 /*
1256  * Test whether a given IPv6 address can be a neighbor.
1257  */
1258 static bool
nd6_is_new_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1259 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1260 {
1261 
1262 	/*
1263 	 * A link-local address is always a neighbor.
1264 	 * XXX: a link does not necessarily specify a single interface.
1265 	 */
1266 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1267 		struct sockaddr_in6 sin6_copy;
1268 		u_int32_t zone;
1269 
1270 		/*
1271 		 * We need sin6_copy since sa6_recoverscope() may modify the
1272 		 * content (XXX).
1273 		 */
1274 		sin6_copy = *addr;
1275 		if (sa6_recoverscope(&sin6_copy))
1276 			return (0); /* XXX: should be impossible */
1277 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1278 			return (0);
1279 		if (sin6_copy.sin6_scope_id == zone)
1280 			return (1);
1281 		else
1282 			return (0);
1283 	}
1284 	/* Checking global unicast */
1285 
1286 	/* If an address is directly reachable, it is a neigbor */
1287 	struct nhop_object *nh;
1288 	nh = fib6_lookup(ifp->if_fib, &addr->sin6_addr, 0, NHR_NONE, 0);
1289 	if (nh != NULL && nh->nh_aifp == ifp && (nh->nh_flags & NHF_GATEWAY) == 0)
1290 		return (true);
1291 
1292 	/*
1293 	 * Check prefixes with desired on-link state, as some may be not
1294 	 * installed in the routing table.
1295 	 */
1296 	bool matched = false;
1297 	struct nd_prefix *pr;
1298 	ND6_RLOCK();
1299 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1300 		if (pr->ndpr_ifp != ifp)
1301 			continue;
1302 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0)
1303 			continue;
1304 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1305 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1306 			matched = true;
1307 			break;
1308 		}
1309 	}
1310 	ND6_RUNLOCK();
1311 	if (matched)
1312 		return (true);
1313 
1314 	/*
1315 	 * If the address is assigned on the node of the other side of
1316 	 * a p2p interface, the address should be a neighbor.
1317 	 */
1318 	if (ifp->if_flags & IFF_POINTOPOINT) {
1319 		struct ifaddr *ifa;
1320 
1321 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1322 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1323 				continue;
1324 			if (ifa->ifa_dstaddr != NULL &&
1325 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1326 				return (true);
1327 			}
1328 		}
1329 	}
1330 
1331 	/*
1332 	 * If the default router list is empty, all addresses are regarded
1333 	 * as on-link, and thus, as a neighbor.
1334 	 */
1335 	if (ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV &&
1336 	    nd6_defrouter_list_empty() &&
1337 	    V_nd6_defifindex == ifp->if_index) {
1338 		return (1);
1339 	}
1340 
1341 	return (0);
1342 }
1343 
1344 /*
1345  * Detect if a given IPv6 address identifies a neighbor on a given link.
1346  * XXX: should take care of the destination of a p2p link?
1347  */
1348 int
nd6_is_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1349 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1350 {
1351 	struct llentry *lle;
1352 	int rc = 0;
1353 
1354 	NET_EPOCH_ASSERT();
1355 
1356 	if (nd6_is_new_addr_neighbor(addr, ifp))
1357 		return (1);
1358 
1359 	/*
1360 	 * Even if the address matches none of our addresses, it might be
1361 	 * in the neighbor cache.
1362 	 */
1363 	if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1364 		LLE_RUNLOCK(lle);
1365 		rc = 1;
1366 	}
1367 	return (rc);
1368 }
1369 
1370 static __noinline void
nd6_free_children(struct llentry * lle)1371 nd6_free_children(struct llentry *lle)
1372 {
1373 	struct llentry *child_lle;
1374 
1375 	NET_EPOCH_ASSERT();
1376 	LLE_WLOCK_ASSERT(lle);
1377 
1378 	while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1379 		LLE_WLOCK(child_lle);
1380 		lltable_unlink_child_entry(child_lle);
1381 		llentry_free(child_lle);
1382 	}
1383 }
1384 
1385 /*
1386  * Tries to update @lle address/prepend data with new @lladdr.
1387  *
1388  * Returns true on success.
1389  * In any case, @lle is returned wlocked.
1390  */
1391 static __noinline bool
nd6_try_set_entry_addr_locked(struct ifnet * ifp,struct llentry * lle,char * lladdr)1392 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1393 {
1394 	u_char buf[LLE_MAX_LINKHDR];
1395 	int fam, off;
1396 	size_t sz;
1397 
1398 	sz = sizeof(buf);
1399 	if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1400 		return (false);
1401 
1402 	/* Update data */
1403 	lltable_set_entry_addr(ifp, lle, buf, sz, off);
1404 
1405 	struct llentry *child_lle;
1406 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1407 		LLE_WLOCK(child_lle);
1408 		fam = child_lle->r_family;
1409 		sz = sizeof(buf);
1410 		if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1411 			/* success */
1412 			lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1413 			child_lle->ln_state = ND6_LLINFO_REACHABLE;
1414 		}
1415 		LLE_WUNLOCK(child_lle);
1416 	}
1417 
1418 	return (true);
1419 }
1420 
1421 bool
nd6_try_set_entry_addr(struct ifnet * ifp,struct llentry * lle,char * lladdr)1422 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1423 {
1424 	NET_EPOCH_ASSERT();
1425 	LLE_WLOCK_ASSERT(lle);
1426 
1427 	if (!lltable_trylock(lle))
1428 		return (false);
1429 	bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1430 	LLTABLE_UNLOCK(lle->lle_tbl);
1431 
1432 	return (ret);
1433 }
1434 
1435 /*
1436  * Free an nd6 llinfo entry.
1437  * Since the function would cause significant changes in the kernel, DO NOT
1438  * make it global, unless you have a strong reason for the change, and are sure
1439  * that the change is safe.
1440  *
1441  * Set noinline to be dtrace-friendly
1442  */
1443 static __noinline void
nd6_free(struct llentry ** lnp,int gc)1444 nd6_free(struct llentry **lnp, int gc)
1445 {
1446 	struct ifnet *ifp;
1447 	struct llentry *ln;
1448 	struct nd_defrouter *dr;
1449 
1450 	ln = *lnp;
1451 	*lnp = NULL;
1452 
1453 	LLE_WLOCK_ASSERT(ln);
1454 	ND6_RLOCK_ASSERT();
1455 
1456 	KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1457 
1458 	ifp = lltable_get_ifp(ln->lle_tbl);
1459 	if ((ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV) != 0)
1460 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1461 	else
1462 		dr = NULL;
1463 	ND6_RUNLOCK();
1464 
1465 	if ((ln->la_flags & LLE_DELETED) == 0)
1466 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1467 
1468 	/*
1469 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1470 	 * even though it is not harmful, it was not really necessary.
1471 	 */
1472 
1473 	/* cancel timer */
1474 	nd6_llinfo_settimer_locked(ln, -1);
1475 
1476 	if (ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV) {
1477 		if (dr != NULL && dr->expire &&
1478 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1479 			/*
1480 			 * If the reason for the deletion is just garbage
1481 			 * collection, and the neighbor is an active default
1482 			 * router, do not delete it.  Instead, reset the GC
1483 			 * timer using the router's lifetime.
1484 			 * Simply deleting the entry would affect default
1485 			 * router selection, which is not necessarily a good
1486 			 * thing, especially when we're using router preference
1487 			 * values.
1488 			 * XXX: the check for ln_state would be redundant,
1489 			 *      but we intentionally keep it just in case.
1490 			 */
1491 			if (dr->expire > time_uptime)
1492 				nd6_llinfo_settimer_locked(ln,
1493 				    (dr->expire - time_uptime) * hz);
1494 			else
1495 				nd6_llinfo_settimer_locked(ln,
1496 				    (long)V_nd6_gctimer * hz);
1497 
1498 			LLE_REMREF(ln);
1499 			LLE_WUNLOCK(ln);
1500 			defrouter_rele(dr);
1501 			return;
1502 		}
1503 
1504 		if (dr) {
1505 			/*
1506 			 * Unreachability of a router might affect the default
1507 			 * router selection and on-link detection of advertised
1508 			 * prefixes.
1509 			 */
1510 
1511 			/*
1512 			 * Temporarily fake the state to choose a new default
1513 			 * router and to perform on-link determination of
1514 			 * prefixes correctly.
1515 			 * Below the state will be set correctly,
1516 			 * or the entry itself will be deleted.
1517 			 */
1518 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1519 		}
1520 
1521 		if (ln->ln_router || dr) {
1522 			/*
1523 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1524 			 * rnh and for the calls to pfxlist_onlink_check() and
1525 			 * defrouter_select_fib() in the block further down for calls
1526 			 * into nd6_lookup().  We still hold a ref.
1527 			 */
1528 			LLE_WUNLOCK(ln);
1529 
1530 			/*
1531 			 * rt6_flush must be called whether or not the neighbor
1532 			 * is in the Default Router List.
1533 			 * See a corresponding comment in nd6_na_input().
1534 			 */
1535 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1536 		}
1537 
1538 		if (dr) {
1539 			/*
1540 			 * Since defrouter_select_fib() does not affect the
1541 			 * on-link determination and MIP6 needs the check
1542 			 * before the default router selection, we perform
1543 			 * the check now.
1544 			 */
1545 			pfxlist_onlink_check();
1546 
1547 			/*
1548 			 * Refresh default router list.
1549 			 */
1550 			defrouter_select_fib(dr->ifp->if_fib);
1551 		}
1552 
1553 		/*
1554 		 * If this entry was added by an on-link redirect, remove the
1555 		 * corresponding host route.
1556 		 */
1557 		if (ln->la_flags & LLE_REDIRECT)
1558 			nd6_free_redirect(ln);
1559 
1560 		if (ln->ln_router || dr)
1561 			LLE_WLOCK(ln);
1562 	}
1563 
1564 	/*
1565 	 * Save to unlock. We still hold an extra reference and will not
1566 	 * free(9) in llentry_free() if someone else holds one as well.
1567 	 */
1568 	LLE_WUNLOCK(ln);
1569 	LLTABLE_LOCK(ln->lle_tbl);
1570 	LLE_WLOCK(ln);
1571 	/* Guard against race with other llentry_free(). */
1572 	if (ln->la_flags & LLE_LINKED) {
1573 		/* Remove callout reference */
1574 		LLE_REMREF(ln);
1575 		lltable_unlink_entry(ln->lle_tbl, ln);
1576 	}
1577 	LLTABLE_UNLOCK(ln->lle_tbl);
1578 
1579 	nd6_free_children(ln);
1580 
1581 	llentry_free(ln);
1582 	if (dr != NULL)
1583 		defrouter_rele(dr);
1584 }
1585 
1586 static int
nd6_isdynrte(const struct rtentry * rt,const struct nhop_object * nh,void * xap)1587 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1588 {
1589 
1590 	if (nh->nh_flags & NHF_REDIRECT)
1591 		return (1);
1592 
1593 	return (0);
1594 }
1595 
1596 /*
1597  * Remove the rtentry for the given llentry,
1598  * both of which were installed by a redirect.
1599  */
1600 static void
nd6_free_redirect(const struct llentry * ln)1601 nd6_free_redirect(const struct llentry *ln)
1602 {
1603 	int fibnum;
1604 	struct sockaddr_in6 sin6;
1605 	struct rib_cmd_info rc;
1606 	struct epoch_tracker et;
1607 
1608 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1609 
1610 	NET_EPOCH_ENTER(et);
1611 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1612 		rib_del_route_px(fibnum, (struct sockaddr *)&sin6, 128,
1613 		    nd6_isdynrte, NULL, 0, &rc);
1614 	NET_EPOCH_EXIT(et);
1615 }
1616 
1617 /*
1618  * Updates status of the default router route.
1619  */
1620 static void
check_release_defrouter(const struct rib_cmd_info * rc,void * _cbdata)1621 check_release_defrouter(const struct rib_cmd_info *rc, void *_cbdata)
1622 {
1623 	struct nd_defrouter *dr;
1624 	struct nhop_object *nh;
1625 
1626 	nh = rc->rc_nh_old;
1627 	if (rc->rc_cmd == RTM_DELETE && (nh->nh_flags & NHF_DEFAULT) != 0) {
1628 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1629 		if (dr != NULL) {
1630 			dr->installed = 0;
1631 			defrouter_rele(dr);
1632 		}
1633 	}
1634 }
1635 
1636 void
nd6_subscription_cb(struct rib_head * rnh,struct rib_cmd_info * rc,void * arg)1637 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1638 {
1639 
1640 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1641 	if (rc->rc_cmd == RTM_DELETE && !NH_IS_NHGRP(rc->rc_nh_old))
1642 		check_release_defrouter(rc, NULL);
1643 }
1644 
1645 int
nd6_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp)1646 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1647 {
1648 	struct epoch_tracker et;
1649 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1650 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1651 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1652 	struct in6_ifextra *ext = ifp->if_inet6;
1653 	int error = 0;
1654 
1655 	/* XXXGL: safety against IFT_PFSYNC & IFT_PFLOG */
1656 	if (ext == NULL)
1657 		return (EPFNOSUPPORT);
1658 #define ND	ndi->ndi
1659 	switch (cmd) {
1660 	case SIOCGIFINFO_IN6:
1661 		ND = (struct nd_ifinfo){
1662 			.linkmtu = ext->nd_linkmtu,
1663 			.maxmtu = ext->nd_maxmtu,
1664 			.basereachable = ext->nd_basereachable,
1665 			.reachable = ext->nd_reachable,
1666 			.retrans = ext->nd_retrans,
1667 			.flags = ext->nd_flags,
1668 			.recalctm = ext->nd_recalc_timer,
1669 			.chlim = ext->nd_curhoplimit,
1670 			.initialized = 1,
1671 		};
1672 		break;
1673 	case SIOCSIFINFO_IN6:
1674 		/*
1675 		 * used to change host variables from userland.
1676 		 * intended for a use on router to reflect RA configurations.
1677 		 */
1678 		/* 0 means 'unspecified' */
1679 		if (ND.linkmtu != 0) {
1680 			if (ND.linkmtu < IPV6_MMTU ||
1681 			    ND.linkmtu > in6_ifmtu(ifp)) {
1682 				error = EINVAL;
1683 				break;
1684 			}
1685 			ext->nd_linkmtu = ND.linkmtu;
1686 		}
1687 
1688 		if (ND.basereachable != 0) {
1689 			uint32_t obasereachable = ext->nd_basereachable;
1690 
1691 			ext->nd_basereachable = ND.basereachable;
1692 			if (ND.basereachable != obasereachable)
1693 				ext->nd_reachable =
1694 				    ND_COMPUTE_RTIME(ND.basereachable);
1695 		}
1696 		if (ND.retrans != 0)
1697 			ext->nd_retrans = ND.retrans;
1698 		if (ND.chlim != 0)
1699 			ext->nd_curhoplimit = ND.chlim;
1700 		/* FALLTHROUGH */
1701 	case SIOCSIFINFO_FLAGS:
1702 	{
1703 		struct ifaddr *ifa;
1704 		struct in6_ifaddr *ia;
1705 
1706 		if ((ext->nd_flags & ND6_IFF_IFDISABLED) &&
1707 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1708 			/* ifdisabled 1->0 transision */
1709 
1710 			/*
1711 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1712 			 * has an link-local address with IN6_IFF_DUPLICATED,
1713 			 * do not clear ND6_IFF_IFDISABLED.
1714 			 * See RFC 4862, Section 5.4.5.
1715 			 */
1716 			NET_EPOCH_ENTER(et);
1717 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1718 				if (ifa->ifa_addr->sa_family != AF_INET6)
1719 					continue;
1720 				ia = (struct in6_ifaddr *)ifa;
1721 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1722 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1723 					break;
1724 			}
1725 			NET_EPOCH_EXIT(et);
1726 
1727 			if (ifa != NULL) {
1728 				/* LLA is duplicated. */
1729 				ND.flags |= ND6_IFF_IFDISABLED;
1730 				log(LOG_ERR, "Cannot enable an interface"
1731 				    " with a link-local address marked"
1732 				    " duplicate.\n");
1733 			} else {
1734 				ext->nd_flags &= ~ND6_IFF_IFDISABLED;
1735 				if (ifp->if_flags & IFF_UP)
1736 					in6_if_up(ifp);
1737 			}
1738 		} else if (!(ext->nd_flags & ND6_IFF_IFDISABLED) &&
1739 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1740 			/* ifdisabled 0->1 transision */
1741 			/* Mark all IPv6 address as tentative. */
1742 
1743 			ext->nd_flags |= ND6_IFF_IFDISABLED;
1744 			if (V_ip6_dad_count > 0 &&
1745 			    (ext->nd_flags & ND6_IFF_NO_DAD) == 0) {
1746 				NET_EPOCH_ENTER(et);
1747 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1748 				    ifa_link) {
1749 					if (ifa->ifa_addr->sa_family !=
1750 					    AF_INET6)
1751 						continue;
1752 					ia = (struct in6_ifaddr *)ifa;
1753 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1754 				}
1755 				NET_EPOCH_EXIT(et);
1756 			}
1757 		}
1758 
1759 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1760 			if (!(ext->nd_flags & ND6_IFF_AUTO_LINKLOCAL)) {
1761 				/* auto_linklocal 0->1 transision */
1762 
1763 				/* If no link-local address on ifp, configure */
1764 				ext->nd_flags |= ND6_IFF_AUTO_LINKLOCAL;
1765 				in6_ifattach(ifp, NULL);
1766 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1767 			    ifp->if_flags & IFF_UP) {
1768 				/*
1769 				 * When the IF already has
1770 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1771 				 * address is assigned, and IFF_UP, try to
1772 				 * assign one.
1773 				 */
1774 				NET_EPOCH_ENTER(et);
1775 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1776 				    ifa_link) {
1777 					if (ifa->ifa_addr->sa_family !=
1778 					    AF_INET6)
1779 						continue;
1780 					ia = (struct in6_ifaddr *)ifa;
1781 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1782 						break;
1783 				}
1784 				NET_EPOCH_EXIT(et);
1785 				if (ifa != NULL)
1786 					/* No LLA is configured. */
1787 					in6_ifattach(ifp, NULL);
1788 			}
1789 		}
1790 		ext->nd_flags = ND.flags;
1791 		break;
1792 	}
1793 #undef ND
1794 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1795 		/* sync kernel routing table with the default router list */
1796 		defrouter_reset();
1797 		defrouter_select_fib(RT_ALL_FIBS);
1798 		break;
1799 	case SIOCSPFXFLUSH_IN6:
1800 	{
1801 		/* flush all the prefix advertised by routers */
1802 		struct in6_ifaddr *ia, *ia_next;
1803 		struct nd_prefix *pr, *next;
1804 		struct nd_prhead prl;
1805 
1806 		LIST_INIT(&prl);
1807 
1808 		ND6_WLOCK();
1809 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1810 			if (ND6_PREFIX_WITH_ROUTER(pr))
1811 				nd6_prefix_unlink(pr, &prl);
1812 		}
1813 		ND6_WUNLOCK();
1814 
1815 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1816 			LIST_REMOVE(pr, ndpr_entry);
1817 			/* XXXRW: in6_ifaddrhead locking. */
1818 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1819 			    ia_next) {
1820 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1821 					continue;
1822 
1823 				if (ia->ia6_ndpr == pr)
1824 					in6_purgeaddr(&ia->ia_ifa);
1825 			}
1826 			nd6_prefix_del(pr);
1827 		}
1828 		break;
1829 	}
1830 	case SIOCSRTRFLUSH_IN6:
1831 	{
1832 		/* flush all the default routers */
1833 
1834 		defrouter_reset();
1835 		nd6_defrouter_flush_all();
1836 		defrouter_select_fib(RT_ALL_FIBS);
1837 		break;
1838 	}
1839 	case SIOCGNBRINFO_IN6:
1840 	{
1841 		struct llentry *ln;
1842 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1843 
1844 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1845 			return (error);
1846 
1847 		NET_EPOCH_ENTER(et);
1848 		ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1849 		NET_EPOCH_EXIT(et);
1850 
1851 		if (ln == NULL) {
1852 			error = EINVAL;
1853 			break;
1854 		}
1855 		nbi->state = ln->ln_state;
1856 		nbi->asked = ln->la_asked;
1857 		nbi->isrouter = ln->ln_router;
1858 		if (ln->la_expire == 0)
1859 			nbi->expire = 0;
1860 		else
1861 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1862 			    (time_second - time_uptime);
1863 		LLE_RUNLOCK(ln);
1864 		break;
1865 	}
1866 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1867 		ndif->ifindex = V_nd6_defifindex;
1868 		break;
1869 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1870 		return (nd6_setdefaultiface(ndif->ifindex));
1871 	}
1872 	return (error);
1873 }
1874 
1875 /*
1876  * Calculates new isRouter value based on provided parameters and
1877  * returns it.
1878  */
1879 static int
nd6_is_router(int type,int code,int is_new,int old_addr,int new_addr,int ln_router)1880 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1881     int ln_router)
1882 {
1883 
1884 	/*
1885 	 * ICMP6 type dependent behavior.
1886 	 *
1887 	 * NS: clear IsRouter if new entry
1888 	 * RS: clear IsRouter
1889 	 * RA: set IsRouter if there's lladdr
1890 	 * redir: clear IsRouter if new entry
1891 	 *
1892 	 * RA case, (1):
1893 	 * The spec says that we must set IsRouter in the following cases:
1894 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1895 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1896 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1897 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1898 	 * neighbor cache, this is similar to (6).
1899 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1900 	 *
1901 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1902 	 *							D R
1903 	 *	0	n	n	(1)	c   ?     s
1904 	 *	0	y	n	(2)	c   s     s
1905 	 *	0	n	y	(3)	c   s     s
1906 	 *	0	y	y	(4)	c   s     s
1907 	 *	0	y	y	(5)	c   s     s
1908 	 *	1	--	n	(6) c	c	c s
1909 	 *	1	--	y	(7) c	c   s	c s
1910 	 *
1911 	 *					(c=clear s=set)
1912 	 */
1913 	switch (type & 0xff) {
1914 	case ND_NEIGHBOR_SOLICIT:
1915 		/*
1916 		 * New entry must have is_router flag cleared.
1917 		 */
1918 		if (is_new)					/* (6-7) */
1919 			ln_router = 0;
1920 		break;
1921 	case ND_REDIRECT:
1922 		/*
1923 		 * If the icmp is a redirect to a better router, always set the
1924 		 * is_router flag.  Otherwise, if the entry is newly created,
1925 		 * clear the flag.  [RFC 2461, sec 8.3]
1926 		 */
1927 		if (code == ND_REDIRECT_ROUTER)
1928 			ln_router = 1;
1929 		else {
1930 			if (is_new)				/* (6-7) */
1931 				ln_router = 0;
1932 		}
1933 		break;
1934 	case ND_ROUTER_SOLICIT:
1935 		/*
1936 		 * is_router flag must always be cleared.
1937 		 */
1938 		ln_router = 0;
1939 		break;
1940 	case ND_ROUTER_ADVERT:
1941 		/*
1942 		 * Mark an entry with lladdr as a router.
1943 		 */
1944 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1945 		    (is_new && new_addr)) {			/* (7) */
1946 			ln_router = 1;
1947 		}
1948 		break;
1949 	}
1950 
1951 	return (ln_router);
1952 }
1953 
1954 /*
1955  * Create neighbor cache entry and cache link-layer address,
1956  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1957  *
1958  * type - ICMP6 type
1959  * code - type dependent information
1960  *
1961  */
1962 void
nd6_cache_lladdr(struct ifnet * ifp,struct in6_addr * from,char * lladdr,int lladdrlen,int type,int code)1963 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1964     int lladdrlen, int type, int code)
1965 {
1966 	struct llentry *ln = NULL, *ln_tmp;
1967 	int is_newentry;
1968 	int do_update;
1969 	int olladdr;
1970 	int llchange;
1971 	int flags;
1972 	uint16_t router = 0;
1973 	struct mbuf *chain = NULL;
1974 	u_char linkhdr[LLE_MAX_LINKHDR];
1975 	size_t linkhdrsize;
1976 	int lladdr_off;
1977 
1978 	NET_EPOCH_ASSERT();
1979 
1980 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1981 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1982 
1983 	/* nothing must be updated for unspecified address */
1984 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1985 		return;
1986 
1987 	/*
1988 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1989 	 * the caller.
1990 	 *
1991 	 * XXX If the link does not have link-layer adderss, what should
1992 	 * we do? (ifp->if_addrlen == 0)
1993 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1994 	 * description on it in NS section (RFC 2461 7.2.3).
1995 	 */
1996 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1997 	ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
1998 	is_newentry = 0;
1999 	if (ln == NULL) {
2000 		flags |= LLE_EXCLUSIVE;
2001 		ln = nd6_alloc(from, 0, ifp);
2002 		if (ln == NULL)
2003 			return;
2004 
2005 		/*
2006 		 * Since we already know all the data for the new entry,
2007 		 * fill it before insertion.
2008 		 */
2009 		if (lladdr != NULL) {
2010 			linkhdrsize = sizeof(linkhdr);
2011 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2012 			    linkhdr, &linkhdrsize, &lladdr_off) != 0) {
2013 				lltable_free_entry(LLTABLE6(ifp), ln);
2014 				return;
2015 			}
2016 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2017 			    lladdr_off);
2018 		}
2019 
2020 		LLTABLE_LOCK(LLTABLE6(ifp));
2021 		LLE_WLOCK(ln);
2022 		/* Prefer any existing lle over newly-created one */
2023 		ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2024 		if (ln_tmp == NULL)
2025 			lltable_link_entry(LLTABLE6(ifp), ln);
2026 		LLTABLE_UNLOCK(LLTABLE6(ifp));
2027 		if (ln_tmp == NULL) {
2028 			/* No existing lle, mark as new entry (6,7) */
2029 			is_newentry = 1;
2030 			if (lladdr != NULL) {	/* (7) */
2031 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2032 				EVENTHANDLER_INVOKE(lle_event, ln,
2033 				    LLENTRY_RESOLVED);
2034 			}
2035 		} else {
2036 			lltable_free_entry(LLTABLE6(ifp), ln);
2037 			ln = ln_tmp;
2038 			ln_tmp = NULL;
2039 		}
2040 	}
2041 	/* do nothing if static ndp is set */
2042 	if ((ln->la_flags & LLE_STATIC)) {
2043 		if (flags & LLE_EXCLUSIVE)
2044 			LLE_WUNLOCK(ln);
2045 		else
2046 			LLE_RUNLOCK(ln);
2047 		return;
2048 	}
2049 
2050 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2051 	if (olladdr && lladdr) {
2052 		llchange = bcmp(lladdr, ln->ll_addr,
2053 		    ifp->if_addrlen);
2054 	} else if (!olladdr && lladdr)
2055 		llchange = 1;
2056 	else
2057 		llchange = 0;
2058 
2059 	/*
2060 	 * newentry olladdr  lladdr  llchange	(*=record)
2061 	 *	0	n	n	--	(1)
2062 	 *	0	y	n	--	(2)
2063 	 *	0	n	y	y	(3) * STALE
2064 	 *	0	y	y	n	(4) *
2065 	 *	0	y	y	y	(5) * STALE
2066 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2067 	 *	1	--	y	--	(7) * STALE
2068 	 */
2069 
2070 	do_update = 0;
2071 	if (is_newentry == 0 && llchange != 0) {
2072 		do_update = 1;	/* (3,5) */
2073 
2074 		/*
2075 		 * Record source link-layer address
2076 		 * XXX is it dependent to ifp->if_type?
2077 		 */
2078 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2079 			/* Entry was deleted */
2080 			LLE_WUNLOCK(ln);
2081 			return;
2082 		}
2083 
2084 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2085 
2086 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2087 
2088 		if (ln->la_hold != NULL)
2089 			chain = nd6_grab_holdchain(ln);
2090 	}
2091 
2092 	/* Calculates new router status */
2093 	router = nd6_is_router(type, code, is_newentry, olladdr,
2094 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2095 
2096 	ln->ln_router = router;
2097 	/* Mark non-router redirects with special flag */
2098 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2099 		ln->la_flags |= LLE_REDIRECT;
2100 
2101 	if (flags & LLE_EXCLUSIVE)
2102 		LLE_WUNLOCK(ln);
2103 	else
2104 		LLE_RUNLOCK(ln);
2105 
2106 	if (chain != NULL)
2107 		nd6_flush_holdchain(ifp, ln, chain);
2108 	if (do_update)
2109 		nd6_flush_children_holdchain(ifp, ln);
2110 
2111 	/*
2112 	 * When the link-layer address of a router changes, select the
2113 	 * best router again.  In particular, when the neighbor entry is newly
2114 	 * created, it might affect the selection policy.
2115 	 * Question: can we restrict the first condition to the "is_newentry"
2116 	 * case?
2117 	 * XXX: when we hear an RA from a new router with the link-layer
2118 	 * address option, defrouter_select_fib() is called twice, since
2119 	 * defrtrlist_update called the function as well.  However, I believe
2120 	 * we can compromise the overhead, since it only happens the first
2121 	 * time.
2122 	 * XXX: although defrouter_select_fib() should not have a bad effect
2123 	 * for those are not autoconfigured hosts, we explicitly avoid such
2124 	 * cases for safety.
2125 	 */
2126 	if ((do_update || is_newentry) && router &&
2127 	    ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV) {
2128 		/*
2129 		 * guaranteed recursion
2130 		 */
2131 		defrouter_select_fib(ifp->if_fib);
2132 	}
2133 }
2134 
2135 static void
nd6_slowtimo(void * arg)2136 nd6_slowtimo(void *arg)
2137 {
2138 	struct epoch_tracker et;
2139 	CURVNET_SET((struct vnet *) arg);
2140 	struct in6_ifextra *nd6if;
2141 	struct ifnet *ifp;
2142 
2143 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2144 	    nd6_slowtimo, curvnet);
2145 	NET_EPOCH_ENTER(et);
2146 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2147 		if ((nd6if = ifp->if_inet6) == NULL)
2148 			continue;
2149 		if (nd6if->nd_basereachable && /* already initialized */
2150 		    (nd6if->nd_recalc_timer -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2151 			/*
2152 			 * Since reachable time rarely changes by router
2153 			 * advertisements, we SHOULD insure that a new random
2154 			 * value gets recomputed at least once every few hours.
2155 			 * (RFC 2461, 6.3.4)
2156 			 */
2157 			nd6if->nd_recalc_timer = V_nd6_recalc_reachtm_interval;
2158 			nd6if->nd_reachable =
2159 			    ND_COMPUTE_RTIME(nd6if->nd_basereachable);
2160 		}
2161 	}
2162 	NET_EPOCH_EXIT(et);
2163 	CURVNET_RESTORE();
2164 }
2165 
2166 struct mbuf *
nd6_grab_holdchain(struct llentry * ln)2167 nd6_grab_holdchain(struct llentry *ln)
2168 {
2169 	struct mbuf *chain;
2170 
2171 	LLE_WLOCK_ASSERT(ln);
2172 
2173 	chain = ln->la_hold;
2174 	ln->la_hold = NULL;
2175 	ln->la_numheld = 0;
2176 
2177 	if (ln->ln_state == ND6_LLINFO_STALE) {
2178 		/*
2179 		 * The first time we send a packet to a
2180 		 * neighbor whose entry is STALE, we have
2181 		 * to change the state to DELAY and a sets
2182 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2183 		 * seconds to ensure do neighbor unreachability
2184 		 * detection on expiration.
2185 		 * (RFC 2461 7.3.3)
2186 		 */
2187 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2188 	}
2189 
2190 	return (chain);
2191 }
2192 
2193 int
nd6_output_ifp(struct ifnet * ifp,struct ifnet * origifp,struct mbuf * m,struct sockaddr_in6 * dst,struct route * ro)2194 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2195     struct sockaddr_in6 *dst, struct route *ro)
2196 {
2197 	int error;
2198 	int ip6len;
2199 	struct ip6_hdr *ip6;
2200 	struct m_tag *mtag;
2201 
2202 #ifdef MAC
2203 	mac_netinet6_nd6_send(ifp, m);
2204 #endif
2205 
2206 	/*
2207 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2208 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2209 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2210 	 * to be diverted to user space.  When re-injected into the kernel,
2211 	 * send_output() will directly dispatch them to the outgoing interface.
2212 	 */
2213 	if (send_sendso_input_hook != NULL) {
2214 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2215 		if (mtag != NULL) {
2216 			ip6 = mtod(m, struct ip6_hdr *);
2217 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2218 			/* Use the SEND socket */
2219 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2220 			    ip6len);
2221 			/* -1 == no app on SEND socket */
2222 			if (error == 0 || error != -1)
2223 			    return (error);
2224 		}
2225 	}
2226 
2227 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2228 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2229 	    mtod(m, struct ip6_hdr *));
2230 
2231 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2232 		origifp = ifp;
2233 
2234 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2235 	return (error);
2236 }
2237 
2238 /*
2239  * Lookup link headerfor @sa_dst address. Stores found
2240  * data in @desten buffer. Copy of lle ln_flags can be also
2241  * saved in @pflags if @pflags is non-NULL.
2242  *
2243  * If destination LLE does not exists or lle state modification
2244  * is required, call "slow" version.
2245  *
2246  * Return values:
2247  * - 0 on success (address copied to buffer).
2248  * - EWOULDBLOCK (no local error, but address is still unresolved)
2249  * - other errors (alloc failure, etc)
2250  */
2251 int
nd6_resolve(struct ifnet * ifp,int gw_flags,struct mbuf * m,const struct sockaddr * sa_dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)2252 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2253     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2254     struct llentry **plle)
2255 {
2256 	struct llentry *ln = NULL;
2257 	const struct sockaddr_in6 *dst6;
2258 
2259 	NET_EPOCH_ASSERT();
2260 
2261 	if (pflags != NULL)
2262 		*pflags = 0;
2263 
2264 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2265 
2266 	/* discard the packet if IPv6 operation is disabled on the interface */
2267 	if ((ifp->if_inet6->nd_flags & ND6_IFF_IFDISABLED)) {
2268 		m_freem(m);
2269 		return (ENETDOWN); /* better error? */
2270 	}
2271 
2272 	if (m != NULL && m->m_flags & M_MCAST) {
2273 		switch (ifp->if_type) {
2274 		case IFT_ETHER:
2275 		case IFT_L2VLAN:
2276 		case IFT_BRIDGE:
2277 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2278 						 desten);
2279 			return (0);
2280 		default:
2281 			m_freem(m);
2282 			return (EAFNOSUPPORT);
2283 		}
2284 	}
2285 
2286 	int family = gw_flags >> 16;
2287 	int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2288 	ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2289 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2290 		/* Entry found, let's copy lle info */
2291 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2292 		if (pflags != NULL)
2293 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2294 		llentry_provide_feedback(ln);
2295 		if (plle) {
2296 			LLE_ADDREF(ln);
2297 			*plle = ln;
2298 			LLE_WUNLOCK(ln);
2299 		}
2300 		return (0);
2301 	} else if (plle && ln)
2302 		LLE_WUNLOCK(ln);
2303 
2304 	return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2305 }
2306 
2307 /*
2308  * Finds or creates a new llentry for @addr and @family.
2309  * Returns wlocked llentry or NULL.
2310  *
2311  *
2312  * Child LLEs.
2313  *
2314  * Do not have their own state machine (gets marked as static)
2315  *  settimer bails out for child LLEs just in case.
2316  *
2317  * Locking order: parent lle gets locked first, chen goes the child.
2318  */
2319 static __noinline struct llentry *
nd6_get_llentry(struct ifnet * ifp,const struct in6_addr * addr,int family)2320 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2321 {
2322 	struct llentry *child_lle = NULL;
2323 	struct llentry *lle, *lle_tmp;
2324 
2325 	lle = nd6_alloc(addr, 0, ifp);
2326 	if (lle != NULL && family != AF_INET6) {
2327 		child_lle = nd6_alloc(addr, 0, ifp);
2328 		if (child_lle == NULL) {
2329 			lltable_free_entry(LLTABLE6(ifp), lle);
2330 			return (NULL);
2331 		}
2332 		child_lle->r_family = family;
2333 		child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2334 		child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2335 	}
2336 
2337 	if (lle == NULL) {
2338 		char ip6buf[INET6_ADDRSTRLEN];
2339 		log(LOG_DEBUG,
2340 		    "nd6_get_llentry: can't allocate llinfo for %s "
2341 		    "(ln=%p)\n",
2342 		    ip6_sprintf(ip6buf, addr), lle);
2343 		return (NULL);
2344 	}
2345 
2346 	LLTABLE_LOCK(LLTABLE6(ifp));
2347 	LLE_WLOCK(lle);
2348 	/* Prefer any existing entry over newly-created one */
2349 	lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2350 	if (lle_tmp == NULL)
2351 		lltable_link_entry(LLTABLE6(ifp), lle);
2352 	else {
2353 		lltable_free_entry(LLTABLE6(ifp), lle);
2354 		lle = lle_tmp;
2355 	}
2356 	if (child_lle != NULL) {
2357 		/* Check if child lle for the same family exists */
2358 		lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2359 		LLE_WLOCK(child_lle);
2360 		if (lle_tmp == NULL) {
2361 			/* Attach */
2362 			lltable_link_child_entry(lle, child_lle);
2363 		} else {
2364 			/* child lle already exists, free newly-created one */
2365 			lltable_free_entry(LLTABLE6(ifp), child_lle);
2366 			LLE_WLOCK(lle_tmp);
2367 			child_lle = lle_tmp;
2368 		}
2369 		LLE_WUNLOCK(lle);
2370 		lle = child_lle;
2371 	}
2372 	LLTABLE_UNLOCK(LLTABLE6(ifp));
2373 	return (lle);
2374 }
2375 
2376 /*
2377  * Do L2 address resolution for @sa_dst address. Stores found
2378  * address in @desten buffer. Copy of lle ln_flags can be also
2379  * saved in @pflags if @pflags is non-NULL.
2380  *
2381  * Heavy version.
2382  * Function assume that destination LLE does not exist,
2383  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2384  *
2385  * Set noinline to be dtrace-friendly
2386  */
2387 static __noinline int
nd6_resolve_slow(struct ifnet * ifp,int family,int flags,struct mbuf * m,const struct sockaddr_in6 * dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)2388 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2389     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2390     struct llentry **plle)
2391 {
2392 	struct llentry *lle = NULL;
2393 	struct in6_addr *psrc, src;
2394 	int send_ns, ll_len;
2395 	char *lladdr;
2396 
2397 	NET_EPOCH_ASSERT();
2398 
2399 	/*
2400 	 * Address resolution or Neighbor Unreachability Detection
2401 	 * for the next hop.
2402 	 * At this point, the destination of the packet must be a unicast
2403 	 * or an anycast address(i.e. not a multicast).
2404 	 */
2405 	lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2406 	if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2407 		/*
2408 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2409 		 * the condition below is not very efficient.  But we believe
2410 		 * it is tolerable, because this should be a rare case.
2411 		 */
2412 		lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2413 	}
2414 
2415 	if (lle == NULL) {
2416 		m_freem(m);
2417 		return (ENOBUFS);
2418 	}
2419 
2420 	LLE_WLOCK_ASSERT(lle);
2421 
2422 	/*
2423 	 * The first time we send a packet to a neighbor whose entry is
2424 	 * STALE, we have to change the state to DELAY and a sets a timer to
2425 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2426 	 * neighbor unreachability detection on expiration.
2427 	 * (RFC 2461 7.3.3)
2428 	 */
2429 	if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2430 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2431 
2432 	/*
2433 	 * If the neighbor cache entry has a state other than INCOMPLETE
2434 	 * (i.e. its link-layer address is already resolved), just
2435 	 * send the packet.
2436 	 */
2437 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2438 		if (flags & LLE_ADDRONLY) {
2439 			lladdr = lle->ll_addr;
2440 			ll_len = ifp->if_addrlen;
2441 		} else {
2442 			lladdr = lle->r_linkdata;
2443 			ll_len = lle->r_hdrlen;
2444 		}
2445 		bcopy(lladdr, desten, ll_len);
2446 		if (pflags != NULL)
2447 			*pflags = lle->la_flags;
2448 		if (plle) {
2449 			LLE_ADDREF(lle);
2450 			*plle = lle;
2451 		}
2452 		LLE_WUNLOCK(lle);
2453 		return (0);
2454 	}
2455 
2456 	/*
2457 	 * There is a neighbor cache entry, but no ethernet address
2458 	 * response yet.  Append this latest packet to the end of the
2459 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2460 	 * the oldest packet in the queue will be removed.
2461 	 */
2462 	if (m != NULL) {
2463 		size_t dropped;
2464 
2465 		dropped = lltable_append_entry_queue(lle, m, V_nd6_maxqueuelen);
2466 		ICMP6STAT_ADD(icp6s_dropped, dropped);
2467 	}
2468 
2469 	/*
2470 	 * If there has been no NS for the neighbor after entering the
2471 	 * INCOMPLETE state, send the first solicitation.
2472 	 * Note that for newly-created lle la_asked will be 0,
2473 	 * so we will transition from ND6_LLINFO_NOSTATE to
2474 	 * ND6_LLINFO_INCOMPLETE state here.
2475 	 */
2476 	psrc = NULL;
2477 	send_ns = 0;
2478 
2479 	/* If we have child lle, switch to the parent to send NS */
2480 	if (lle->la_flags & LLE_CHILD) {
2481 		struct llentry *lle_parent = lle->lle_parent;
2482 		LLE_WUNLOCK(lle);
2483 		lle = lle_parent;
2484 		LLE_WLOCK(lle);
2485 	}
2486 	if (lle->la_asked == 0) {
2487 		lle->la_asked++;
2488 		send_ns = 1;
2489 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2490 
2491 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2492 	}
2493 	LLE_WUNLOCK(lle);
2494 	if (send_ns != 0)
2495 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2496 
2497 	return (EWOULDBLOCK);
2498 }
2499 
2500 /*
2501  * Do L2 address resolution for @sa_dst address. Stores found
2502  * address in @desten buffer. Copy of lle ln_flags can be also
2503  * saved in @pflags if @pflags is non-NULL.
2504  *
2505  * Return values:
2506  * - 0 on success (address copied to buffer).
2507  * - EWOULDBLOCK (no local error, but address is still unresolved)
2508  * - other errors (alloc failure, etc)
2509  */
2510 int
nd6_resolve_addr(struct ifnet * ifp,int flags,const struct sockaddr * dst,char * desten,uint32_t * pflags)2511 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2512     char *desten, uint32_t *pflags)
2513 {
2514 	int error;
2515 
2516 	flags |= LLE_ADDRONLY;
2517 	error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2518 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2519 	return (error);
2520 }
2521 
2522 int
nd6_flush_holdchain(struct ifnet * ifp,struct llentry * lle,struct mbuf * chain)2523 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2524 {
2525 	struct mbuf *m, *m_head;
2526 	struct sockaddr_in6 dst6;
2527 	int error = 0;
2528 
2529 	NET_EPOCH_ASSERT();
2530 
2531 	struct route_in6 ro = {
2532 		.ro_prepend = lle->r_linkdata,
2533 		.ro_plen = lle->r_hdrlen,
2534 	};
2535 
2536 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2537 	m_head = chain;
2538 
2539 	while (m_head) {
2540 		m = m_head;
2541 		m_head = m_head->m_nextpkt;
2542 		m->m_nextpkt = NULL;
2543 		error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2544 	}
2545 
2546 	/*
2547 	 * XXX
2548 	 * note that intermediate errors are blindly ignored
2549 	 */
2550 	return (error);
2551 }
2552 
2553 __noinline void
nd6_flush_children_holdchain(struct ifnet * ifp,struct llentry * lle)2554 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2555 {
2556 	struct llentry *child_lle;
2557 	struct mbuf *chain;
2558 
2559 	NET_EPOCH_ASSERT();
2560 
2561 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2562 		LLE_WLOCK(child_lle);
2563 		chain = nd6_grab_holdchain(child_lle);
2564 		LLE_WUNLOCK(child_lle);
2565 		nd6_flush_holdchain(ifp, child_lle, chain);
2566 	}
2567 }
2568 
2569 static int
nd6_need_cache(struct ifnet * ifp)2570 nd6_need_cache(struct ifnet *ifp)
2571 {
2572 	/*
2573 	 * XXX: we currently do not make neighbor cache on any interface
2574 	 * other than Ethernet and GIF.
2575 	 *
2576 	 * RFC2893 says:
2577 	 * - unidirectional tunnels needs no ND
2578 	 */
2579 	switch (ifp->if_type) {
2580 	case IFT_ETHER:
2581 	case IFT_IEEE1394:
2582 	case IFT_L2VLAN:
2583 	case IFT_INFINIBAND:
2584 	case IFT_BRIDGE:
2585 	case IFT_PROPVIRTUAL:
2586 		return (1);
2587 	default:
2588 		return (0);
2589 	}
2590 }
2591 
2592 /*
2593  * Add pernament ND6 link-layer record for given
2594  * interface address.
2595  *
2596  * Very similar to IPv4 arp_ifinit(), but:
2597  * 1) IPv6 DAD is performed in different place
2598  * 2) It is called by IPv6 protocol stack in contrast to
2599  * arp_ifinit() which is typically called in SIOCSIFADDR
2600  * driver ioctl handler.
2601  *
2602  */
2603 int
nd6_add_ifa_lle(struct in6_ifaddr * ia)2604 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2605 {
2606 	struct ifnet *ifp;
2607 	struct llentry *ln, *ln_tmp;
2608 	struct sockaddr *dst;
2609 
2610 	ifp = ia->ia_ifa.ifa_ifp;
2611 	if (nd6_need_cache(ifp) == 0)
2612 		return (0);
2613 
2614 	dst = (struct sockaddr *)&ia->ia_addr;
2615 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2616 	if (ln == NULL)
2617 		return (ENOBUFS);
2618 
2619 	LLTABLE_LOCK(LLTABLE6(ifp));
2620 	LLE_WLOCK(ln);
2621 	/* Unlink any entry if exists */
2622 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2623 	if (ln_tmp != NULL)
2624 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2625 	lltable_link_entry(LLTABLE6(ifp), ln);
2626 	LLTABLE_UNLOCK(LLTABLE6(ifp));
2627 
2628 	if (ln_tmp != NULL)
2629 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2630 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2631 
2632 	LLE_WUNLOCK(ln);
2633 	if (ln_tmp != NULL)
2634 		llentry_free(ln_tmp);
2635 
2636 	return (0);
2637 }
2638 
2639 /*
2640  * Removes either all lle entries for given @ia, or lle
2641  * corresponding to @ia address.
2642  */
2643 void
nd6_rem_ifa_lle(struct in6_ifaddr * ia,int all)2644 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2645 {
2646 	struct sockaddr_in6 mask, addr;
2647 	struct sockaddr *saddr, *smask;
2648 	struct ifnet *ifp;
2649 
2650 	ifp = ia->ia_ifa.ifa_ifp;
2651 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2652 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2653 	saddr = (struct sockaddr *)&addr;
2654 	smask = (struct sockaddr *)&mask;
2655 
2656 	if (all != 0)
2657 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2658 	else
2659 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2660 }
2661 
2662 static int
nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)2663 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2664 {
2665 	struct in6_prefix p;
2666 	struct sockaddr_in6 s6;
2667 	struct nd_prefix *pr;
2668 	struct nd_pfxrouter *pfr;
2669 	time_t maxexpire;
2670 	int error;
2671 	char ip6buf[INET6_ADDRSTRLEN];
2672 
2673 	if (req->newptr)
2674 		return (EPERM);
2675 
2676 	error = sysctl_wire_old_buffer(req, 0);
2677 	if (error != 0)
2678 		return (error);
2679 
2680 	bzero(&p, sizeof(p));
2681 	p.origin = PR_ORIG_RA;
2682 	bzero(&s6, sizeof(s6));
2683 	s6.sin6_family = AF_INET6;
2684 	s6.sin6_len = sizeof(s6);
2685 
2686 	ND6_RLOCK();
2687 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2688 		p.prefix = pr->ndpr_prefix;
2689 		if (sa6_recoverscope(&p.prefix)) {
2690 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2691 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2692 			/* XXX: press on... */
2693 		}
2694 		p.raflags = pr->ndpr_raf;
2695 		p.prefixlen = pr->ndpr_plen;
2696 		p.vltime = pr->ndpr_vltime;
2697 		p.pltime = pr->ndpr_pltime;
2698 		p.if_index = pr->ndpr_ifp->if_index;
2699 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2700 			p.expire = 0;
2701 		else {
2702 			/* XXX: we assume time_t is signed. */
2703 			maxexpire = (-1) &
2704 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2705 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2706 				p.expire = pr->ndpr_lastupdate +
2707 				    pr->ndpr_vltime +
2708 				    (time_second - time_uptime);
2709 			else
2710 				p.expire = maxexpire;
2711 		}
2712 		p.refcnt = pr->ndpr_addrcnt;
2713 		p.flags = pr->ndpr_stateflags;
2714 		p.advrtrs = 0;
2715 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2716 			p.advrtrs++;
2717 		error = SYSCTL_OUT(req, &p, sizeof(p));
2718 		if (error != 0)
2719 			break;
2720 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2721 			s6.sin6_addr = pfr->router->rtaddr;
2722 			if (sa6_recoverscope(&s6))
2723 				log(LOG_ERR,
2724 				    "scope error in prefix list (%s)\n",
2725 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2726 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2727 			if (error != 0)
2728 				goto out;
2729 		}
2730 	}
2731 out:
2732 	ND6_RUNLOCK();
2733 	return (error);
2734 }
2735 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2736 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2737 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2738 	"NDP prefix list");
2739 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2740 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2741 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2742 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2743