xref: /freebsd/sys/netinet6/nd6.c (revision c8d25804f56ae7c35eaa492b6110807a2675d41a)
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 		case ND_OPT_ROUTE_INFO:
517 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
518 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
519 					= nd_opt;
520 			}
521 			ndopts->nd_opts_rti_end =
522 				(struct nd_opt_route_info *)nd_opt;
523 			break;
524 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
525 		case ND_OPT_RDNSS:	/* RFC 6106 */
526 		case ND_OPT_DNSSL:	/* RFC 6106 */
527 			/*
528 			 * Silently ignore options we know and do not care about
529 			 * in the kernel.
530 			 */
531 			break;
532 		default:
533 			/*
534 			 * Unknown options must be silently ignored,
535 			 * to accommodate future extension to the protocol.
536 			 */
537 			nd6log((LOG_DEBUG,
538 			    "nd6_options: unsupported option %d - "
539 			    "option ignored\n", nd_opt->nd_opt_type));
540 		}
541 
542 skip1:
543 		i++;
544 		if (i > V_nd6_maxndopt) {
545 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
546 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
547 			break;
548 		}
549 
550 		if (ndopts->nd_opts_done)
551 			break;
552 	}
553 
554 	return 0;
555 }
556 
557 /*
558  * ND6 timer routine to handle ND6 entries
559  */
560 static void
nd6_llinfo_settimer_locked(struct llentry * ln,long tick)561 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
562 {
563 	int canceled;
564 
565 	LLE_WLOCK_ASSERT(ln);
566 
567 	/* Do not schedule timers for child LLEs. */
568 	if (ln->la_flags & LLE_CHILD)
569 		return;
570 
571 	if (tick < 0) {
572 		ln->la_expire = 0;
573 		ln->ln_ntick = 0;
574 		canceled = callout_stop(&ln->lle_timer);
575 	} else {
576 		ln->la_expire = time_uptime + tick / hz;
577 		LLE_ADDREF(ln);
578 		if (tick > INT_MAX) {
579 			ln->ln_ntick = tick - INT_MAX;
580 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
581 			    nd6_llinfo_timer, ln);
582 		} else {
583 			ln->ln_ntick = 0;
584 			canceled = callout_reset(&ln->lle_timer, tick,
585 			    nd6_llinfo_timer, ln);
586 		}
587 	}
588 	if (canceled > 0)
589 		LLE_REMREF(ln);
590 }
591 
592 /*
593  * Gets source address of the first packet in hold queue
594  * and stores it in @src.
595  * Returns pointer to @src (if hold queue is not empty) or NULL.
596  *
597  * Set noinline to be dtrace-friendly
598  */
599 static __noinline struct in6_addr *
nd6_llinfo_get_holdsrc(struct llentry * ln,struct in6_addr * src)600 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
601 {
602 	struct ip6_hdr hdr;
603 	struct mbuf *m;
604 
605 	if (ln->la_hold == NULL)
606 		return (NULL);
607 
608 	/*
609 	 * assume every packet in la_hold has the same IP header
610 	 */
611 	m = ln->la_hold;
612 	if (sizeof(hdr) > m->m_len)
613 		return (NULL);
614 
615 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
616 	*src = hdr.ip6_src;
617 
618 	return (src);
619 }
620 
621 /*
622  * Checks if we need to switch from STALE state.
623  *
624  * RFC 4861 requires switching from STALE to DELAY state
625  * on first packet matching entry, waiting V_nd6_delay and
626  * transition to PROBE state (if upper layer confirmation was
627  * not received).
628  *
629  * This code performs a bit differently:
630  * On packet hit we don't change state (but desired state
631  * can be guessed by control plane). However, after V_nd6_delay
632  * seconds code will transition to PROBE state (so DELAY state
633  * is kinda skipped in most situations).
634  *
635  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
636  * we perform the following upon entering STALE state:
637  *
638  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
639  * if packet was transmitted at the start of given interval, we
640  * would be able to switch to PROBE state in V_nd6_delay seconds
641  * as user expects.
642  *
643  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
644  * lle in STALE state (remaining timer value stored in lle_remtime).
645  *
646  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
647  * seconds ago.
648  *
649  * Returns non-zero value if the entry is still STALE (storing
650  * the next timer interval in @pdelay).
651  *
652  * Returns zero value if original timer expired or we need to switch to
653  * PROBE (store that in @do_switch variable).
654  */
655 static int
nd6_is_stale(struct llentry * lle,long * pdelay,int * do_switch)656 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
657 {
658 	int nd_delay, nd_gctimer;
659 	time_t lle_hittime;
660 	long delay;
661 
662 	*do_switch = 0;
663 	nd_gctimer = V_nd6_gctimer;
664 	nd_delay = V_nd6_delay;
665 
666 	lle_hittime = llentry_get_hittime(lle);
667 
668 	if (lle_hittime == 0) {
669 		/*
670 		 * Datapath feedback has been requested upon entering
671 		 * STALE state. No packets has been passed using this lle.
672 		 * Ask for the timer reschedule and keep STALE state.
673 		 */
674 		delay = (long)(MIN(nd_gctimer, nd_delay));
675 		delay *= hz;
676 		if (lle->lle_remtime > delay)
677 			lle->lle_remtime -= delay;
678 		else {
679 			delay = lle->lle_remtime;
680 			lle->lle_remtime = 0;
681 		}
682 
683 		if (delay == 0) {
684 			/*
685 			 * The original ng6_gctime timeout ended,
686 			 * no more rescheduling.
687 			 */
688 			return (0);
689 		}
690 
691 		*pdelay = delay;
692 		return (1);
693 	}
694 
695 	/*
696 	 * Packet received. Verify timestamp
697 	 */
698 	delay = (long)(time_uptime - lle_hittime);
699 	if (delay < nd_delay) {
700 		/*
701 		 * V_nd6_delay still not passed since the first
702 		 * hit in STALE state.
703 		 * Reschedule timer and return.
704 		 */
705 		*pdelay = (long)(nd_delay - delay) * hz;
706 		return (1);
707 	}
708 
709 	/* Request switching to probe */
710 	*do_switch = 1;
711 	return (0);
712 }
713 
714 /*
715  * Switch @lle state to new state optionally arming timers.
716  *
717  * Set noinline to be dtrace-friendly
718  */
719 __noinline void
nd6_llinfo_setstate(struct llentry * lle,int newstate)720 nd6_llinfo_setstate(struct llentry *lle, int newstate)
721 {
722 	struct ifnet *ifp;
723 	int nd_gctimer, nd_delay;
724 	long delay, remtime;
725 
726 	delay = 0;
727 	remtime = 0;
728 
729 	switch (newstate) {
730 	case ND6_LLINFO_INCOMPLETE:
731 		ifp = lle->lle_tbl->llt_ifp;
732 		delay = (long)ifp->if_inet6->nd_retrans * hz / 1000;
733 		break;
734 	case ND6_LLINFO_REACHABLE:
735 		if (!ND6_LLINFO_PERMANENT(lle)) {
736 			ifp = lle->lle_tbl->llt_ifp;
737 			delay = (long)ifp->if_inet6->nd_reachable * hz;
738 		}
739 		break;
740 	case ND6_LLINFO_STALE:
741 
742 		llentry_request_feedback(lle);
743 		nd_delay = V_nd6_delay;
744 		nd_gctimer = V_nd6_gctimer;
745 
746 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
747 		remtime = (long)nd_gctimer * hz - delay;
748 		break;
749 	case ND6_LLINFO_DELAY:
750 		lle->la_asked = 0;
751 		delay = (long)V_nd6_delay * hz;
752 		break;
753 	}
754 
755 	if (delay > 0)
756 		nd6_llinfo_settimer_locked(lle, delay);
757 
758 	lle->lle_remtime = remtime;
759 	lle->ln_state = newstate;
760 }
761 
762 /*
763  * Timer-dependent part of nd state machine.
764  *
765  * Set noinline to be dtrace-friendly
766  */
767 static __noinline void
nd6_llinfo_timer(void * arg)768 nd6_llinfo_timer(void *arg)
769 {
770 	struct epoch_tracker et;
771 	struct llentry *ln;
772 	struct in6_addr *dst, *pdst, *psrc, src;
773 	struct ifnet *ifp;
774 	struct in6_ifextra *ndi;
775 	int do_switch, send_ns;
776 	long delay;
777 
778 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
779 	ln = (struct llentry *)arg;
780 	ifp = lltable_get_ifp(ln->lle_tbl);
781 	CURVNET_SET(ifp->if_vnet);
782 
783 	ND6_RLOCK();
784 	LLE_WLOCK(ln);
785 	if (callout_pending(&ln->lle_timer)) {
786 		/*
787 		 * Here we are a bit odd here in the treatment of
788 		 * active/pending. If the pending bit is set, it got
789 		 * rescheduled before I ran. The active
790 		 * bit we ignore, since if it was stopped
791 		 * in ll_tablefree() and was currently running
792 		 * it would have return 0 so the code would
793 		 * not have deleted it since the callout could
794 		 * not be stopped so we want to go through
795 		 * with the delete here now. If the callout
796 		 * was restarted, the pending bit will be back on and
797 		 * we just want to bail since the callout_reset would
798 		 * return 1 and our reference would have been removed
799 		 * by nd6_llinfo_settimer_locked above since canceled
800 		 * would have been 1.
801 		 */
802 		LLE_WUNLOCK(ln);
803 		ND6_RUNLOCK();
804 		CURVNET_RESTORE();
805 		return;
806 	}
807 	NET_EPOCH_ENTER(et);
808 	ndi = ifp->if_inet6;
809 	send_ns = 0;
810 	dst = &ln->r_l3addr.addr6;
811 	pdst = dst;
812 
813 	if (ln->ln_ntick > 0) {
814 		if (ln->ln_ntick > INT_MAX) {
815 			ln->ln_ntick -= INT_MAX;
816 			nd6_llinfo_settimer_locked(ln, INT_MAX);
817 		} else {
818 			ln->ln_ntick = 0;
819 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
820 		}
821 		goto done;
822 	}
823 
824 	if (ln->la_flags & LLE_STATIC) {
825 		goto done;
826 	}
827 
828 	if (ln->la_flags & LLE_DELETED) {
829 		nd6_free(&ln, 0);
830 		goto done;
831 	}
832 
833 	switch (ln->ln_state) {
834 	case ND6_LLINFO_INCOMPLETE:
835 		if (ln->la_asked < V_nd6_mmaxtries) {
836 			ln->la_asked++;
837 			send_ns = 1;
838 			/* Send NS to multicast address */
839 			pdst = NULL;
840 		} else {
841 			struct mbuf *m;
842 
843 			ICMP6STAT_ADD(icp6s_dropped, ln->la_numheld);
844 
845 			m = ln->la_hold;
846 			if (m != NULL) {
847 				/*
848 				 * assuming every packet in la_hold has the
849 				 * same IP header.  Send error after unlock.
850 				 */
851 				ln->la_hold = m->m_nextpkt;
852 				m->m_nextpkt = NULL;
853 				ln->la_numheld--;
854 			}
855 			nd6_free(&ln, 0);
856 			if (m != NULL) {
857 				struct mbuf *n = m;
858 
859 				/*
860 				 * if there are any ummapped mbufs, we
861 				 * must free them, rather than using
862 				 * them for an ICMP, as they cannot be
863 				 * checksummed.
864 				 */
865 				while ((n = n->m_next) != NULL) {
866 					if (n->m_flags & M_EXTPG)
867 						break;
868 				}
869 				if (n != NULL) {
870 					m_freem(m);
871 					m = NULL;
872 				} else {
873 					icmp6_error2(m, ICMP6_DST_UNREACH,
874 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
875 				}
876 			}
877 		}
878 		break;
879 	case ND6_LLINFO_REACHABLE:
880 		if (!ND6_LLINFO_PERMANENT(ln))
881 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
882 		break;
883 
884 	case ND6_LLINFO_STALE:
885 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
886 			/*
887 			 * No packet has used this entry and GC timeout
888 			 * has not been passed. Reschedule timer and
889 			 * return.
890 			 */
891 			nd6_llinfo_settimer_locked(ln, delay);
892 			break;
893 		}
894 
895 		if (do_switch == 0) {
896 			/*
897 			 * GC timer has ended and entry hasn't been used.
898 			 * Run Garbage collector (RFC 4861, 5.3)
899 			 */
900 			if (!ND6_LLINFO_PERMANENT(ln))
901 				nd6_free(&ln, 1);
902 			break;
903 		}
904 
905 		/* Entry has been used AND delay timer has ended. */
906 
907 		/* FALLTHROUGH */
908 
909 	case ND6_LLINFO_DELAY:
910 		if ((ndi->nd_flags & ND6_IFF_PERFORMNUD) != 0) {
911 			/* We need NUD */
912 			ln->la_asked = 1;
913 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
914 			send_ns = 1;
915 		} else
916 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
917 		break;
918 	case ND6_LLINFO_PROBE:
919 		if (ln->la_asked < V_nd6_umaxtries) {
920 			ln->la_asked++;
921 			send_ns = 1;
922 		} else {
923 			nd6_free(&ln, 0);
924 		}
925 		break;
926 	default:
927 		panic("%s: paths in a dark night can be confusing: %d",
928 		    __func__, ln->ln_state);
929 	}
930 done:
931 	if (ln != NULL)
932 		ND6_RUNLOCK();
933 	if (send_ns != 0) {
934 		nd6_llinfo_settimer_locked(ln,
935 		    (long)ndi->nd_retrans * hz / 1000);
936 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
937 		LLE_FREE_LOCKED(ln);
938 		ln = NULL;
939 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
940 	}
941 
942 	if (ln != NULL)
943 		LLE_FREE_LOCKED(ln);
944 	NET_EPOCH_EXIT(et);
945 	CURVNET_RESTORE();
946 }
947 
948 /*
949  * ND6 timer routine to expire default route list and prefix list
950  */
951 void
nd6_timer(void * arg)952 nd6_timer(void *arg)
953 {
954 	CURVNET_SET((struct vnet *) arg);
955 	struct epoch_tracker et;
956 	struct nd_prhead prl;
957 	struct nd_prefix *pr, *npr;
958 	struct ifnet *ifp;
959 	struct in6_ifaddr *ia6, *nia6;
960 	uint64_t genid;
961 
962 	LIST_INIT(&prl);
963 
964 	NET_EPOCH_ENTER(et);
965 	nd6_defrouter_timer();
966 
967 	/*
968 	 * expire interface addresses.
969 	 * in the past the loop was inside prefix expiry processing.
970 	 * However, from a stricter speci-confrmance standpoint, we should
971 	 * rather separate address lifetimes and prefix lifetimes.
972 	 *
973 	 * XXXRW: in6_ifaddrhead locking.
974 	 */
975   addrloop:
976 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
977 		/* check address lifetime */
978 		if (IFA6_IS_INVALID(ia6)) {
979 			int regen = 0;
980 
981 			/*
982 			 * If the expiring address is temporary, try
983 			 * regenerating a new one.  This would be useful when
984 			 * we suspended a laptop PC, then turned it on after a
985 			 * period that could invalidate all temporary
986 			 * addresses.  Although we may have to restart the
987 			 * loop (see below), it must be after purging the
988 			 * address.  Otherwise, we'd see an infinite loop of
989 			 * regeneration.
990 			 */
991 			if (V_ip6_use_tempaddr &&
992 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
993 				if (regen_tmpaddr(ia6) == 0)
994 					regen = 1;
995 			}
996 
997 			in6_purgeaddr(&ia6->ia_ifa);
998 
999 			if (regen)
1000 				goto addrloop; /* XXX: see below */
1001 		} else if (IFA6_IS_DEPRECATED(ia6)) {
1002 			int oldflags = ia6->ia6_flags;
1003 
1004 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
1005 
1006 			/*
1007 			 * If a temporary address has just become deprecated,
1008 			 * regenerate a new one if possible.
1009 			 */
1010 			if (V_ip6_use_tempaddr &&
1011 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1012 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
1013 				if (regen_tmpaddr(ia6) == 0) {
1014 					/*
1015 					 * A new temporary address is
1016 					 * generated.
1017 					 * XXX: this means the address chain
1018 					 * has changed while we are still in
1019 					 * the loop.  Although the change
1020 					 * would not cause disaster (because
1021 					 * it's not a deletion, but an
1022 					 * addition,) we'd rather restart the
1023 					 * loop just for safety.  Or does this
1024 					 * significantly reduce performance??
1025 					 */
1026 					goto addrloop;
1027 				}
1028 			}
1029 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
1030 			/*
1031 			 * Schedule DAD for a tentative address.  This happens
1032 			 * if the interface was down or not running
1033 			 * when the address was configured.
1034 			 */
1035 			int delay;
1036 
1037 			delay = arc4random() %
1038 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1039 			nd6_dad_start((struct ifaddr *)ia6, delay);
1040 		} else {
1041 			/*
1042 			 * Check status of the interface.  If it is down,
1043 			 * mark the address as tentative for future DAD.
1044 			 */
1045 			ifp = ia6->ia_ifp;
1046 			if ((ifp->if_inet6->nd_flags & ND6_IFF_NO_DAD) == 0 &&
1047 			    ((ifp->if_flags & IFF_UP) == 0 ||
1048 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1049 			    (ifp->if_inet6->nd_flags & ND6_IFF_IFDISABLED))){
1050 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1051 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1052 			}
1053 
1054 			/*
1055 			 * A new RA might have made a deprecated address
1056 			 * preferred.
1057 			 */
1058 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1059 		}
1060 	}
1061 	NET_EPOCH_EXIT(et);
1062 
1063 	ND6_WLOCK();
1064 restart:
1065 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1066 		/*
1067 		 * Expire prefixes. Since the pltime is only used for
1068 		 * autoconfigured addresses, pltime processing for prefixes is
1069 		 * not necessary.
1070 		 *
1071 		 * Only unlink after all derived addresses have expired. This
1072 		 * may not occur until two hours after the prefix has expired
1073 		 * per RFC 4862. If the prefix expires before its derived
1074 		 * addresses, mark it off-link. This will be done automatically
1075 		 * after unlinking if no address references remain.
1076 		 */
1077 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1078 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1079 			continue;
1080 
1081 		if (pr->ndpr_addrcnt == 0) {
1082 			nd6_prefix_unlink(pr, &prl);
1083 			continue;
1084 		}
1085 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1086 			genid = V_nd6_list_genid;
1087 			nd6_prefix_ref(pr);
1088 			ND6_WUNLOCK();
1089 			ND6_ONLINK_LOCK();
1090 			(void)nd6_prefix_offlink(pr);
1091 			ND6_ONLINK_UNLOCK();
1092 			ND6_WLOCK();
1093 			nd6_prefix_rele(pr);
1094 			if (genid != V_nd6_list_genid)
1095 				goto restart;
1096 		}
1097 	}
1098 	ND6_WUNLOCK();
1099 
1100 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1101 		LIST_REMOVE(pr, ndpr_entry);
1102 		nd6_prefix_del(pr);
1103 	}
1104 
1105 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1106 	    nd6_timer, curvnet);
1107 
1108 	CURVNET_RESTORE();
1109 }
1110 
1111 /*
1112  * ia6 - deprecated/invalidated temporary address
1113  */
1114 static int
regen_tmpaddr(struct in6_ifaddr * ia6)1115 regen_tmpaddr(struct in6_ifaddr *ia6)
1116 {
1117 	struct ifaddr *ifa;
1118 	struct ifnet *ifp;
1119 	struct in6_ifaddr *public_ifa6 = NULL;
1120 
1121 	NET_EPOCH_ASSERT();
1122 
1123 	ifp = ia6->ia_ifa.ifa_ifp;
1124 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1125 		struct in6_ifaddr *it6;
1126 
1127 		if (ifa->ifa_addr->sa_family != AF_INET6)
1128 			continue;
1129 
1130 		it6 = (struct in6_ifaddr *)ifa;
1131 
1132 		/* ignore no autoconf addresses. */
1133 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1134 			continue;
1135 
1136 		/* ignore autoconf addresses with different prefixes. */
1137 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1138 			continue;
1139 
1140 		/*
1141 		 * Now we are looking at an autoconf address with the same
1142 		 * prefix as ours.  If the address is temporary and is still
1143 		 * preferred, do not create another one.  It would be rare, but
1144 		 * could happen, for example, when we resume a laptop PC after
1145 		 * a long period.
1146 		 */
1147 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1148 		    !IFA6_IS_DEPRECATED(it6)) {
1149 			public_ifa6 = NULL;
1150 			break;
1151 		}
1152 
1153 		/*
1154 		 * This is a public autoconf address that has the same prefix
1155 		 * as ours.  If it is preferred, keep it.  We can't break the
1156 		 * loop here, because there may be a still-preferred temporary
1157 		 * address with the prefix.
1158 		 */
1159 		if (!IFA6_IS_DEPRECATED(it6))
1160 			public_ifa6 = it6;
1161 	}
1162 	if (public_ifa6 != NULL)
1163 		ifa_ref(&public_ifa6->ia_ifa);
1164 
1165 	if (public_ifa6 != NULL) {
1166 		int e;
1167 
1168 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1169 			ifa_free(&public_ifa6->ia_ifa);
1170 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1171 			    " tmp addr,errno=%d\n", e);
1172 			return (-1);
1173 		}
1174 		ifa_free(&public_ifa6->ia_ifa);
1175 		return (0);
1176 	}
1177 
1178 	return (-1);
1179 }
1180 
1181 /*
1182  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1183  * cache entries are freed in in6_domifdetach().
1184  */
1185 void
nd6_purge(struct ifnet * ifp)1186 nd6_purge(struct ifnet *ifp)
1187 {
1188 	struct nd_prhead prl;
1189 	struct nd_prefix *pr, *npr;
1190 
1191 	LIST_INIT(&prl);
1192 
1193 	/* Purge default router list entries toward ifp. */
1194 	nd6_defrouter_purge(ifp);
1195 
1196 	ND6_WLOCK();
1197 	/*
1198 	 * Remove prefixes on ifp. We should have already removed addresses on
1199 	 * this interface, so no addresses should be referencing these prefixes.
1200 	 */
1201 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1202 		if (pr->ndpr_ifp == ifp)
1203 			nd6_prefix_unlink(pr, &prl);
1204 	}
1205 	ND6_WUNLOCK();
1206 
1207 	/* Delete the unlinked prefix objects. */
1208 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1209 		LIST_REMOVE(pr, ndpr_entry);
1210 		nd6_prefix_del(pr);
1211 	}
1212 
1213 	/* cancel default outgoing interface setting */
1214 	if (V_nd6_defifindex == ifp->if_index)
1215 		nd6_setdefaultiface(0);
1216 
1217 	if (ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV) {
1218 		/* Refresh default router list. */
1219 		defrouter_select_fib(ifp->if_fib);
1220 	}
1221 }
1222 
1223 /*
1224  * the caller acquires and releases the lock on the lltbls
1225  * Returns the llentry locked
1226  */
1227 struct llentry *
nd6_lookup(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1228 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1229 {
1230 	struct sockaddr_in6 sin6;
1231 	struct llentry *ln;
1232 
1233 	bzero(&sin6, sizeof(sin6));
1234 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1235 	sin6.sin6_family = AF_INET6;
1236 	sin6.sin6_addr = *addr6;
1237 
1238 	LLTABLE_RLOCK_ASSERT(LLTABLE6(ifp));
1239 
1240 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1241 
1242 	return (ln);
1243 }
1244 
1245 static struct llentry *
nd6_alloc(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1246 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1247 {
1248 	struct sockaddr_in6 sin6;
1249 	struct llentry *ln;
1250 
1251 	bzero(&sin6, sizeof(sin6));
1252 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1253 	sin6.sin6_family = AF_INET6;
1254 	sin6.sin6_addr = *addr6;
1255 
1256 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1257 	if (ln != NULL)
1258 		ln->ln_state = ND6_LLINFO_NOSTATE;
1259 
1260 	return (ln);
1261 }
1262 
1263 /*
1264  * Test whether a given IPv6 address can be a neighbor.
1265  */
1266 static bool
nd6_is_new_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1267 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1268 {
1269 
1270 	/*
1271 	 * A link-local address is always a neighbor.
1272 	 * XXX: a link does not necessarily specify a single interface.
1273 	 */
1274 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1275 		struct sockaddr_in6 sin6_copy;
1276 		u_int32_t zone;
1277 
1278 		/*
1279 		 * We need sin6_copy since sa6_recoverscope() may modify the
1280 		 * content (XXX).
1281 		 */
1282 		sin6_copy = *addr;
1283 		if (sa6_recoverscope(&sin6_copy))
1284 			return (0); /* XXX: should be impossible */
1285 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1286 			return (0);
1287 		if (sin6_copy.sin6_scope_id == zone)
1288 			return (1);
1289 		else
1290 			return (0);
1291 	}
1292 	/* Checking global unicast */
1293 
1294 	/* If an address is directly reachable, it is a neigbor */
1295 	struct nhop_object *nh;
1296 	nh = fib6_lookup(ifp->if_fib, &addr->sin6_addr, 0, NHR_NONE, 0);
1297 	if (nh != NULL && nh->nh_aifp == ifp && (nh->nh_flags & NHF_GATEWAY) == 0)
1298 		return (true);
1299 
1300 	/*
1301 	 * Check prefixes with desired on-link state, as some may be not
1302 	 * installed in the routing table.
1303 	 */
1304 	bool matched = false;
1305 	struct nd_prefix *pr;
1306 	ND6_RLOCK();
1307 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1308 		if (pr->ndpr_ifp != ifp)
1309 			continue;
1310 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0)
1311 			continue;
1312 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1313 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1314 			matched = true;
1315 			break;
1316 		}
1317 	}
1318 	ND6_RUNLOCK();
1319 	if (matched)
1320 		return (true);
1321 
1322 	/*
1323 	 * If the address is assigned on the node of the other side of
1324 	 * a p2p interface, the address should be a neighbor.
1325 	 */
1326 	if (ifp->if_flags & IFF_POINTOPOINT) {
1327 		struct ifaddr *ifa;
1328 
1329 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1330 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1331 				continue;
1332 			if (ifa->ifa_dstaddr != NULL &&
1333 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1334 				return (true);
1335 			}
1336 		}
1337 	}
1338 
1339 	/*
1340 	 * If the default router list is empty, all addresses are regarded
1341 	 * as on-link, and thus, as a neighbor.
1342 	 */
1343 	if (ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV &&
1344 	    nd6_defrouter_list_empty() &&
1345 	    V_nd6_defifindex == ifp->if_index) {
1346 		return (1);
1347 	}
1348 
1349 	return (0);
1350 }
1351 
1352 /*
1353  * Detect if a given IPv6 address identifies a neighbor on a given link.
1354  * XXX: should take care of the destination of a p2p link?
1355  */
1356 int
nd6_is_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1357 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1358 {
1359 	struct llentry *lle;
1360 	int rc = 0;
1361 
1362 	NET_EPOCH_ASSERT();
1363 
1364 	if (nd6_is_new_addr_neighbor(addr, ifp))
1365 		return (1);
1366 
1367 	/*
1368 	 * Even if the address matches none of our addresses, it might be
1369 	 * in the neighbor cache.
1370 	 */
1371 	if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1372 		LLE_RUNLOCK(lle);
1373 		rc = 1;
1374 	}
1375 	return (rc);
1376 }
1377 
1378 static __noinline void
nd6_free_children(struct llentry * lle)1379 nd6_free_children(struct llentry *lle)
1380 {
1381 	struct llentry *child_lle;
1382 
1383 	NET_EPOCH_ASSERT();
1384 	LLE_WLOCK_ASSERT(lle);
1385 
1386 	while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1387 		LLE_WLOCK(child_lle);
1388 		lltable_unlink_child_entry(child_lle);
1389 		llentry_free(child_lle);
1390 	}
1391 }
1392 
1393 /*
1394  * Tries to update @lle address/prepend data with new @lladdr.
1395  *
1396  * Returns true on success.
1397  * In any case, @lle is returned wlocked.
1398  */
1399 static __noinline bool
nd6_try_set_entry_addr_locked(struct ifnet * ifp,struct llentry * lle,char * lladdr)1400 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1401 {
1402 	u_char buf[LLE_MAX_LINKHDR];
1403 	int fam, off;
1404 	size_t sz;
1405 
1406 	sz = sizeof(buf);
1407 	if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1408 		return (false);
1409 
1410 	/* Update data */
1411 	lltable_set_entry_addr(ifp, lle, buf, sz, off);
1412 
1413 	struct llentry *child_lle;
1414 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1415 		LLE_WLOCK(child_lle);
1416 		fam = child_lle->r_family;
1417 		sz = sizeof(buf);
1418 		if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1419 			/* success */
1420 			lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1421 			child_lle->ln_state = ND6_LLINFO_REACHABLE;
1422 		}
1423 		LLE_WUNLOCK(child_lle);
1424 	}
1425 
1426 	return (true);
1427 }
1428 
1429 bool
nd6_try_set_entry_addr(struct ifnet * ifp,struct llentry * lle,char * lladdr)1430 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1431 {
1432 	NET_EPOCH_ASSERT();
1433 	LLE_WLOCK_ASSERT(lle);
1434 
1435 	if (!lltable_trylock(lle))
1436 		return (false);
1437 	bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1438 	LLTABLE_UNLOCK(lle->lle_tbl);
1439 
1440 	return (ret);
1441 }
1442 
1443 /*
1444  * Free an nd6 llinfo entry.
1445  * Since the function would cause significant changes in the kernel, DO NOT
1446  * make it global, unless you have a strong reason for the change, and are sure
1447  * that the change is safe.
1448  *
1449  * Set noinline to be dtrace-friendly
1450  */
1451 static __noinline void
nd6_free(struct llentry ** lnp,int gc)1452 nd6_free(struct llentry **lnp, int gc)
1453 {
1454 	struct ifnet *ifp;
1455 	struct llentry *ln;
1456 	struct nd_defrouter *dr;
1457 
1458 	ln = *lnp;
1459 	*lnp = NULL;
1460 
1461 	LLE_WLOCK_ASSERT(ln);
1462 	ND6_RLOCK_ASSERT();
1463 
1464 	KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1465 
1466 	ifp = lltable_get_ifp(ln->lle_tbl);
1467 	if ((ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV) != 0)
1468 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1469 	else
1470 		dr = NULL;
1471 	ND6_RUNLOCK();
1472 
1473 	if ((ln->la_flags & LLE_DELETED) == 0)
1474 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1475 
1476 	/*
1477 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1478 	 * even though it is not harmful, it was not really necessary.
1479 	 */
1480 
1481 	/* cancel timer */
1482 	nd6_llinfo_settimer_locked(ln, -1);
1483 
1484 	if (ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV) {
1485 		if (dr != NULL && dr->expire &&
1486 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1487 			/*
1488 			 * If the reason for the deletion is just garbage
1489 			 * collection, and the neighbor is an active default
1490 			 * router, do not delete it.  Instead, reset the GC
1491 			 * timer using the router's lifetime.
1492 			 * Simply deleting the entry would affect default
1493 			 * router selection, which is not necessarily a good
1494 			 * thing, especially when we're using router preference
1495 			 * values.
1496 			 * XXX: the check for ln_state would be redundant,
1497 			 *      but we intentionally keep it just in case.
1498 			 */
1499 			if (dr->expire > time_uptime)
1500 				nd6_llinfo_settimer_locked(ln,
1501 				    (dr->expire - time_uptime) * hz);
1502 			else
1503 				nd6_llinfo_settimer_locked(ln,
1504 				    (long)V_nd6_gctimer * hz);
1505 
1506 			LLE_REMREF(ln);
1507 			LLE_WUNLOCK(ln);
1508 			defrouter_rele(dr);
1509 			return;
1510 		}
1511 
1512 		if (dr) {
1513 			/*
1514 			 * Unreachability of a router might affect the default
1515 			 * router selection and on-link detection of advertised
1516 			 * prefixes.
1517 			 */
1518 
1519 			/*
1520 			 * Temporarily fake the state to choose a new default
1521 			 * router and to perform on-link determination of
1522 			 * prefixes correctly.
1523 			 * Below the state will be set correctly,
1524 			 * or the entry itself will be deleted.
1525 			 */
1526 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1527 		}
1528 
1529 		if (ln->ln_router || dr) {
1530 			/*
1531 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1532 			 * rnh and for the calls to pfxlist_onlink_check() and
1533 			 * defrouter_select_fib() in the block further down for calls
1534 			 * into nd6_lookup().  We still hold a ref.
1535 			 */
1536 			LLE_WUNLOCK(ln);
1537 
1538 			/*
1539 			 * rt6_flush must be called whether or not the neighbor
1540 			 * is in the Default Router List.
1541 			 * See a corresponding comment in nd6_na_input().
1542 			 */
1543 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1544 		}
1545 
1546 		if (dr) {
1547 			/*
1548 			 * Since defrouter_select_fib() does not affect the
1549 			 * on-link determination and MIP6 needs the check
1550 			 * before the default router selection, we perform
1551 			 * the check now.
1552 			 */
1553 			pfxlist_onlink_check();
1554 
1555 			/*
1556 			 * Refresh default router list.
1557 			 */
1558 			defrouter_select_fib(dr->ifp->if_fib);
1559 		}
1560 
1561 		/*
1562 		 * If this entry was added by an on-link redirect, remove the
1563 		 * corresponding host route.
1564 		 */
1565 		if (ln->la_flags & LLE_REDIRECT)
1566 			nd6_free_redirect(ln);
1567 
1568 		if (ln->ln_router || dr)
1569 			LLE_WLOCK(ln);
1570 	}
1571 
1572 	/*
1573 	 * Save to unlock. We still hold an extra reference and will not
1574 	 * free(9) in llentry_free() if someone else holds one as well.
1575 	 */
1576 	LLE_WUNLOCK(ln);
1577 	LLTABLE_LOCK(ln->lle_tbl);
1578 	LLE_WLOCK(ln);
1579 	/* Guard against race with other llentry_free(). */
1580 	if (ln->la_flags & LLE_LINKED) {
1581 		/* Remove callout reference */
1582 		LLE_REMREF(ln);
1583 		lltable_unlink_entry(ln->lle_tbl, ln);
1584 	}
1585 	LLTABLE_UNLOCK(ln->lle_tbl);
1586 
1587 	nd6_free_children(ln);
1588 
1589 	llentry_free(ln);
1590 	if (dr != NULL)
1591 		defrouter_rele(dr);
1592 }
1593 
1594 static int
nd6_isdynrte(const struct rtentry * rt,const struct nhop_object * nh,void * xap)1595 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1596 {
1597 
1598 	if (nh->nh_flags & NHF_REDIRECT)
1599 		return (1);
1600 
1601 	return (0);
1602 }
1603 
1604 /*
1605  * Remove the rtentry for the given llentry,
1606  * both of which were installed by a redirect.
1607  */
1608 static void
nd6_free_redirect(const struct llentry * ln)1609 nd6_free_redirect(const struct llentry *ln)
1610 {
1611 	int fibnum;
1612 	struct sockaddr_in6 sin6;
1613 	struct rib_cmd_info rc;
1614 	struct epoch_tracker et;
1615 
1616 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1617 
1618 	NET_EPOCH_ENTER(et);
1619 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1620 		rib_del_route_px(fibnum, (struct sockaddr *)&sin6, 128,
1621 		    nd6_isdynrte, NULL, 0, &rc);
1622 	NET_EPOCH_EXIT(et);
1623 }
1624 
1625 /*
1626  * Updates status of the default router route.
1627  */
1628 static void
check_release_defrouter(const struct rib_cmd_info * rc,void * _cbdata)1629 check_release_defrouter(const struct rib_cmd_info *rc, void *_cbdata)
1630 {
1631 	struct nd_defrouter *dr;
1632 	struct nhop_object *nh;
1633 
1634 	nh = rc->rc_nh_old;
1635 	if (rc->rc_cmd == RTM_DELETE && (nh->nh_flags & NHF_DEFAULT) != 0) {
1636 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1637 		if (dr != NULL) {
1638 			dr->installed = 0;
1639 			defrouter_rele(dr);
1640 		}
1641 	}
1642 }
1643 
1644 void
nd6_subscription_cb(struct rib_head * rnh,struct rib_cmd_info * rc,void * arg)1645 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1646 {
1647 
1648 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1649 	if (rc->rc_cmd == RTM_DELETE && !NH_IS_NHGRP(rc->rc_nh_old))
1650 		check_release_defrouter(rc, NULL);
1651 }
1652 
1653 int
nd6_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp)1654 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1655 {
1656 	struct epoch_tracker et;
1657 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1658 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1659 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1660 	struct in6_ifextra *ext = ifp->if_inet6;
1661 	int error = 0;
1662 
1663 	/* XXXGL: safety against IFT_PFSYNC & IFT_PFLOG */
1664 	if (ext == NULL)
1665 		return (EPFNOSUPPORT);
1666 #define ND	ndi->ndi
1667 	switch (cmd) {
1668 	case SIOCGIFINFO_IN6:
1669 		ND = (struct nd_ifinfo){
1670 			.linkmtu = ext->nd_linkmtu,
1671 			.maxmtu = ext->nd_maxmtu,
1672 			.basereachable = ext->nd_basereachable,
1673 			.reachable = ext->nd_reachable,
1674 			.retrans = ext->nd_retrans,
1675 			.flags = ext->nd_flags,
1676 			.recalctm = ext->nd_recalc_timer,
1677 			.chlim = ext->nd_curhoplimit,
1678 			.initialized = 1,
1679 		};
1680 		break;
1681 	case SIOCSIFINFO_IN6:
1682 		/*
1683 		 * used to change host variables from userland.
1684 		 * intended for a use on router to reflect RA configurations.
1685 		 */
1686 		/* 0 means 'unspecified' */
1687 		if (ND.linkmtu != 0) {
1688 			if (ND.linkmtu < IPV6_MMTU ||
1689 			    ND.linkmtu > in6_ifmtu(ifp)) {
1690 				error = EINVAL;
1691 				break;
1692 			}
1693 			ext->nd_linkmtu = ND.linkmtu;
1694 		}
1695 
1696 		if (ND.basereachable != 0) {
1697 			uint32_t obasereachable = ext->nd_basereachable;
1698 
1699 			ext->nd_basereachable = ND.basereachable;
1700 			if (ND.basereachable != obasereachable)
1701 				ext->nd_reachable =
1702 				    ND_COMPUTE_RTIME(ND.basereachable);
1703 		}
1704 		if (ND.retrans != 0)
1705 			ext->nd_retrans = ND.retrans;
1706 		if (ND.chlim != 0)
1707 			ext->nd_curhoplimit = ND.chlim;
1708 		/* FALLTHROUGH */
1709 	case SIOCSIFINFO_FLAGS:
1710 	{
1711 		struct ifaddr *ifa;
1712 		struct in6_ifaddr *ia;
1713 
1714 		if ((ext->nd_flags & ND6_IFF_IFDISABLED) &&
1715 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1716 			/* ifdisabled 1->0 transision */
1717 
1718 			/*
1719 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1720 			 * has an link-local address with IN6_IFF_DUPLICATED,
1721 			 * do not clear ND6_IFF_IFDISABLED.
1722 			 * See RFC 4862, Section 5.4.5.
1723 			 */
1724 			NET_EPOCH_ENTER(et);
1725 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1726 				if (ifa->ifa_addr->sa_family != AF_INET6)
1727 					continue;
1728 				ia = (struct in6_ifaddr *)ifa;
1729 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1730 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1731 					break;
1732 			}
1733 			NET_EPOCH_EXIT(et);
1734 
1735 			if (ifa != NULL) {
1736 				/* LLA is duplicated. */
1737 				ND.flags |= ND6_IFF_IFDISABLED;
1738 				log(LOG_ERR, "Cannot enable an interface"
1739 				    " with a link-local address marked"
1740 				    " duplicate.\n");
1741 			} else {
1742 				ext->nd_flags &= ~ND6_IFF_IFDISABLED;
1743 				if (ifp->if_flags & IFF_UP)
1744 					in6_if_up(ifp);
1745 			}
1746 		} else if (!(ext->nd_flags & ND6_IFF_IFDISABLED) &&
1747 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1748 			/* ifdisabled 0->1 transision */
1749 			/* Mark all IPv6 address as tentative. */
1750 
1751 			ext->nd_flags |= ND6_IFF_IFDISABLED;
1752 			if (V_ip6_dad_count > 0 &&
1753 			    (ext->nd_flags & ND6_IFF_NO_DAD) == 0) {
1754 				NET_EPOCH_ENTER(et);
1755 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1756 				    ifa_link) {
1757 					if (ifa->ifa_addr->sa_family !=
1758 					    AF_INET6)
1759 						continue;
1760 					ia = (struct in6_ifaddr *)ifa;
1761 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1762 				}
1763 				NET_EPOCH_EXIT(et);
1764 			}
1765 		}
1766 
1767 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1768 			if (!(ext->nd_flags & ND6_IFF_AUTO_LINKLOCAL)) {
1769 				/* auto_linklocal 0->1 transision */
1770 
1771 				/* If no link-local address on ifp, configure */
1772 				ext->nd_flags |= ND6_IFF_AUTO_LINKLOCAL;
1773 				in6_ifattach(ifp, NULL);
1774 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1775 			    ifp->if_flags & IFF_UP) {
1776 				/*
1777 				 * When the IF already has
1778 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1779 				 * address is assigned, and IFF_UP, try to
1780 				 * assign one.
1781 				 */
1782 				NET_EPOCH_ENTER(et);
1783 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1784 				    ifa_link) {
1785 					if (ifa->ifa_addr->sa_family !=
1786 					    AF_INET6)
1787 						continue;
1788 					ia = (struct in6_ifaddr *)ifa;
1789 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1790 						break;
1791 				}
1792 				NET_EPOCH_EXIT(et);
1793 				if (ifa != NULL)
1794 					/* No LLA is configured. */
1795 					in6_ifattach(ifp, NULL);
1796 			}
1797 		}
1798 		ext->nd_flags = ND.flags;
1799 		break;
1800 	}
1801 #undef ND
1802 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1803 		/* sync kernel routing table with the default router list */
1804 		defrouter_reset();
1805 		defrouter_select_fib(RT_ALL_FIBS);
1806 		break;
1807 	case SIOCSPFXFLUSH_IN6:
1808 	{
1809 		/* flush all the prefix advertised by routers */
1810 		struct in6_ifaddr *ia, *ia_next;
1811 		struct nd_prefix *pr, *next;
1812 		struct nd_prhead prl;
1813 
1814 		LIST_INIT(&prl);
1815 
1816 		ND6_WLOCK();
1817 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1818 			if (ND6_PREFIX_WITH_ROUTER(pr))
1819 				nd6_prefix_unlink(pr, &prl);
1820 		}
1821 		ND6_WUNLOCK();
1822 
1823 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1824 			LIST_REMOVE(pr, ndpr_entry);
1825 			/* XXXRW: in6_ifaddrhead locking. */
1826 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1827 			    ia_next) {
1828 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1829 					continue;
1830 
1831 				if (ia->ia6_ndpr == pr)
1832 					in6_purgeaddr(&ia->ia_ifa);
1833 			}
1834 			nd6_prefix_del(pr);
1835 		}
1836 		break;
1837 	}
1838 	case SIOCSRTRFLUSH_IN6:
1839 	{
1840 		/* flush all the default routers */
1841 
1842 		defrouter_reset();
1843 		nd6_defrouter_flush_all();
1844 		defrouter_select_fib(RT_ALL_FIBS);
1845 		break;
1846 	}
1847 	case SIOCGNBRINFO_IN6:
1848 	{
1849 		struct llentry *ln;
1850 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1851 
1852 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1853 			return (error);
1854 
1855 		NET_EPOCH_ENTER(et);
1856 		ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1857 		NET_EPOCH_EXIT(et);
1858 
1859 		if (ln == NULL) {
1860 			error = EINVAL;
1861 			break;
1862 		}
1863 		nbi->state = ln->ln_state;
1864 		nbi->asked = ln->la_asked;
1865 		nbi->isrouter = ln->ln_router;
1866 		if (ln->la_expire == 0)
1867 			nbi->expire = 0;
1868 		else
1869 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1870 			    (time_second - time_uptime);
1871 		LLE_RUNLOCK(ln);
1872 		break;
1873 	}
1874 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1875 		ndif->ifindex = V_nd6_defifindex;
1876 		break;
1877 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1878 		return (nd6_setdefaultiface(ndif->ifindex));
1879 	}
1880 	return (error);
1881 }
1882 
1883 /*
1884  * Calculates new isRouter value based on provided parameters and
1885  * returns it.
1886  */
1887 static int
nd6_is_router(int type,int code,int is_new,int old_addr,int new_addr,int ln_router)1888 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1889     int ln_router)
1890 {
1891 
1892 	/*
1893 	 * ICMP6 type dependent behavior.
1894 	 *
1895 	 * NS: clear IsRouter if new entry
1896 	 * RS: clear IsRouter
1897 	 * RA: set IsRouter if there's lladdr
1898 	 * redir: clear IsRouter if new entry
1899 	 *
1900 	 * RA case, (1):
1901 	 * The spec says that we must set IsRouter in the following cases:
1902 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1903 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1904 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1905 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1906 	 * neighbor cache, this is similar to (6).
1907 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1908 	 *
1909 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1910 	 *							D R
1911 	 *	0	n	n	(1)	c   ?     s
1912 	 *	0	y	n	(2)	c   s     s
1913 	 *	0	n	y	(3)	c   s     s
1914 	 *	0	y	y	(4)	c   s     s
1915 	 *	0	y	y	(5)	c   s     s
1916 	 *	1	--	n	(6) c	c	c s
1917 	 *	1	--	y	(7) c	c   s	c s
1918 	 *
1919 	 *					(c=clear s=set)
1920 	 */
1921 	switch (type & 0xff) {
1922 	case ND_NEIGHBOR_SOLICIT:
1923 		/*
1924 		 * New entry must have is_router flag cleared.
1925 		 */
1926 		if (is_new)					/* (6-7) */
1927 			ln_router = 0;
1928 		break;
1929 	case ND_REDIRECT:
1930 		/*
1931 		 * If the icmp is a redirect to a better router, always set the
1932 		 * is_router flag.  Otherwise, if the entry is newly created,
1933 		 * clear the flag.  [RFC 2461, sec 8.3]
1934 		 */
1935 		if (code == ND_REDIRECT_ROUTER)
1936 			ln_router = 1;
1937 		else {
1938 			if (is_new)				/* (6-7) */
1939 				ln_router = 0;
1940 		}
1941 		break;
1942 	case ND_ROUTER_SOLICIT:
1943 		/*
1944 		 * is_router flag must always be cleared.
1945 		 */
1946 		ln_router = 0;
1947 		break;
1948 	case ND_ROUTER_ADVERT:
1949 		/*
1950 		 * Mark an entry with lladdr as a router.
1951 		 */
1952 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1953 		    (is_new && new_addr)) {			/* (7) */
1954 			ln_router = 1;
1955 		}
1956 		break;
1957 	}
1958 
1959 	return (ln_router);
1960 }
1961 
1962 /*
1963  * Create neighbor cache entry and cache link-layer address,
1964  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1965  *
1966  * type - ICMP6 type
1967  * code - type dependent information
1968  *
1969  */
1970 void
nd6_cache_lladdr(struct ifnet * ifp,struct in6_addr * from,char * lladdr,int lladdrlen,int type,int code)1971 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1972     int lladdrlen, int type, int code)
1973 {
1974 	struct llentry *ln = NULL, *ln_tmp;
1975 	int is_newentry;
1976 	int do_update;
1977 	int olladdr;
1978 	int llchange;
1979 	int flags;
1980 	uint16_t router = 0;
1981 	struct mbuf *chain = NULL;
1982 	u_char linkhdr[LLE_MAX_LINKHDR];
1983 	size_t linkhdrsize;
1984 	int lladdr_off;
1985 
1986 	NET_EPOCH_ASSERT();
1987 
1988 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1989 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1990 
1991 	/* nothing must be updated for unspecified address */
1992 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1993 		return;
1994 
1995 	/*
1996 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1997 	 * the caller.
1998 	 *
1999 	 * XXX If the link does not have link-layer adderss, what should
2000 	 * we do? (ifp->if_addrlen == 0)
2001 	 * Spec says nothing in sections for RA, RS and NA.  There's small
2002 	 * description on it in NS section (RFC 2461 7.2.3).
2003 	 */
2004 	flags = lladdr ? LLE_EXCLUSIVE : 0;
2005 	ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
2006 	is_newentry = 0;
2007 	if (ln == NULL) {
2008 		flags |= LLE_EXCLUSIVE;
2009 		ln = nd6_alloc(from, 0, ifp);
2010 		if (ln == NULL)
2011 			return;
2012 
2013 		/*
2014 		 * Since we already know all the data for the new entry,
2015 		 * fill it before insertion.
2016 		 */
2017 		if (lladdr != NULL) {
2018 			linkhdrsize = sizeof(linkhdr);
2019 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2020 			    linkhdr, &linkhdrsize, &lladdr_off) != 0) {
2021 				lltable_free_entry(LLTABLE6(ifp), ln);
2022 				return;
2023 			}
2024 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2025 			    lladdr_off);
2026 		}
2027 
2028 		LLTABLE_LOCK(LLTABLE6(ifp));
2029 		LLE_WLOCK(ln);
2030 		/* Prefer any existing lle over newly-created one */
2031 		ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2032 		if (ln_tmp == NULL)
2033 			lltable_link_entry(LLTABLE6(ifp), ln);
2034 		LLTABLE_UNLOCK(LLTABLE6(ifp));
2035 		if (ln_tmp == NULL) {
2036 			/* No existing lle, mark as new entry (6,7) */
2037 			is_newentry = 1;
2038 			if (lladdr != NULL) {	/* (7) */
2039 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2040 				EVENTHANDLER_INVOKE(lle_event, ln,
2041 				    LLENTRY_RESOLVED);
2042 			}
2043 		} else {
2044 			lltable_free_entry(LLTABLE6(ifp), ln);
2045 			ln = ln_tmp;
2046 			ln_tmp = NULL;
2047 		}
2048 	}
2049 	/* do nothing if static ndp is set */
2050 	if ((ln->la_flags & LLE_STATIC)) {
2051 		if (flags & LLE_EXCLUSIVE)
2052 			LLE_WUNLOCK(ln);
2053 		else
2054 			LLE_RUNLOCK(ln);
2055 		return;
2056 	}
2057 
2058 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2059 	if (olladdr && lladdr) {
2060 		llchange = bcmp(lladdr, ln->ll_addr,
2061 		    ifp->if_addrlen);
2062 	} else if (!olladdr && lladdr)
2063 		llchange = 1;
2064 	else
2065 		llchange = 0;
2066 
2067 	/*
2068 	 * newentry olladdr  lladdr  llchange	(*=record)
2069 	 *	0	n	n	--	(1)
2070 	 *	0	y	n	--	(2)
2071 	 *	0	n	y	y	(3) * STALE
2072 	 *	0	y	y	n	(4) *
2073 	 *	0	y	y	y	(5) * STALE
2074 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2075 	 *	1	--	y	--	(7) * STALE
2076 	 */
2077 
2078 	do_update = 0;
2079 	if (is_newentry == 0 && llchange != 0) {
2080 		do_update = 1;	/* (3,5) */
2081 
2082 		/*
2083 		 * Record source link-layer address
2084 		 * XXX is it dependent to ifp->if_type?
2085 		 */
2086 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2087 			/* Entry was deleted */
2088 			LLE_WUNLOCK(ln);
2089 			return;
2090 		}
2091 
2092 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2093 
2094 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2095 
2096 		if (ln->la_hold != NULL)
2097 			chain = nd6_grab_holdchain(ln);
2098 	}
2099 
2100 	/* Calculates new router status */
2101 	router = nd6_is_router(type, code, is_newentry, olladdr,
2102 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2103 
2104 	ln->ln_router = router;
2105 	/* Mark non-router redirects with special flag */
2106 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2107 		ln->la_flags |= LLE_REDIRECT;
2108 
2109 	if (flags & LLE_EXCLUSIVE)
2110 		LLE_WUNLOCK(ln);
2111 	else
2112 		LLE_RUNLOCK(ln);
2113 
2114 	if (chain != NULL)
2115 		nd6_flush_holdchain(ifp, ln, chain);
2116 	if (do_update)
2117 		nd6_flush_children_holdchain(ifp, ln);
2118 
2119 	/*
2120 	 * When the link-layer address of a router changes, select the
2121 	 * best router again.  In particular, when the neighbor entry is newly
2122 	 * created, it might affect the selection policy.
2123 	 * Question: can we restrict the first condition to the "is_newentry"
2124 	 * case?
2125 	 * XXX: when we hear an RA from a new router with the link-layer
2126 	 * address option, defrouter_select_fib() is called twice, since
2127 	 * defrtrlist_update called the function as well.  However, I believe
2128 	 * we can compromise the overhead, since it only happens the first
2129 	 * time.
2130 	 * XXX: although defrouter_select_fib() should not have a bad effect
2131 	 * for those are not autoconfigured hosts, we explicitly avoid such
2132 	 * cases for safety.
2133 	 */
2134 	if ((do_update || is_newentry) && router &&
2135 	    ifp->if_inet6->nd_flags & ND6_IFF_ACCEPT_RTADV) {
2136 		/*
2137 		 * guaranteed recursion
2138 		 */
2139 		defrouter_select_fib(ifp->if_fib);
2140 	}
2141 }
2142 
2143 static void
nd6_slowtimo(void * arg)2144 nd6_slowtimo(void *arg)
2145 {
2146 	struct epoch_tracker et;
2147 	CURVNET_SET((struct vnet *) arg);
2148 	struct in6_ifextra *nd6if;
2149 	struct ifnet *ifp;
2150 
2151 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2152 	    nd6_slowtimo, curvnet);
2153 	NET_EPOCH_ENTER(et);
2154 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2155 		if ((nd6if = ifp->if_inet6) == NULL)
2156 			continue;
2157 		if (nd6if->nd_basereachable && /* already initialized */
2158 		    (nd6if->nd_recalc_timer -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2159 			/*
2160 			 * Since reachable time rarely changes by router
2161 			 * advertisements, we SHOULD insure that a new random
2162 			 * value gets recomputed at least once every few hours.
2163 			 * (RFC 2461, 6.3.4)
2164 			 */
2165 			nd6if->nd_recalc_timer = V_nd6_recalc_reachtm_interval;
2166 			nd6if->nd_reachable =
2167 			    ND_COMPUTE_RTIME(nd6if->nd_basereachable);
2168 		}
2169 	}
2170 	NET_EPOCH_EXIT(et);
2171 	CURVNET_RESTORE();
2172 }
2173 
2174 struct mbuf *
nd6_grab_holdchain(struct llentry * ln)2175 nd6_grab_holdchain(struct llentry *ln)
2176 {
2177 	struct mbuf *chain;
2178 
2179 	LLE_WLOCK_ASSERT(ln);
2180 
2181 	chain = ln->la_hold;
2182 	ln->la_hold = NULL;
2183 	ln->la_numheld = 0;
2184 
2185 	if (ln->ln_state == ND6_LLINFO_STALE) {
2186 		/*
2187 		 * The first time we send a packet to a
2188 		 * neighbor whose entry is STALE, we have
2189 		 * to change the state to DELAY and a sets
2190 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2191 		 * seconds to ensure do neighbor unreachability
2192 		 * detection on expiration.
2193 		 * (RFC 2461 7.3.3)
2194 		 */
2195 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2196 	}
2197 
2198 	return (chain);
2199 }
2200 
2201 int
nd6_output_ifp(struct ifnet * ifp,struct ifnet * origifp,struct mbuf * m,struct sockaddr_in6 * dst,struct route * ro)2202 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2203     struct sockaddr_in6 *dst, struct route *ro)
2204 {
2205 	int error;
2206 	int ip6len;
2207 	struct ip6_hdr *ip6;
2208 	struct m_tag *mtag;
2209 
2210 #ifdef MAC
2211 	mac_netinet6_nd6_send(ifp, m);
2212 #endif
2213 
2214 	/*
2215 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2216 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2217 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2218 	 * to be diverted to user space.  When re-injected into the kernel,
2219 	 * send_output() will directly dispatch them to the outgoing interface.
2220 	 */
2221 	if (send_sendso_input_hook != NULL) {
2222 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2223 		if (mtag != NULL) {
2224 			ip6 = mtod(m, struct ip6_hdr *);
2225 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2226 			/* Use the SEND socket */
2227 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2228 			    ip6len);
2229 			/* -1 == no app on SEND socket */
2230 			if (error == 0 || error != -1)
2231 			    return (error);
2232 		}
2233 	}
2234 
2235 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2236 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2237 	    mtod(m, struct ip6_hdr *));
2238 
2239 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2240 		origifp = ifp;
2241 
2242 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2243 	return (error);
2244 }
2245 
2246 /*
2247  * Lookup link headerfor @sa_dst address. Stores found
2248  * data in @desten buffer. Copy of lle ln_flags can be also
2249  * saved in @pflags if @pflags is non-NULL.
2250  *
2251  * If destination LLE does not exists or lle state modification
2252  * is required, call "slow" version.
2253  *
2254  * Return values:
2255  * - 0 on success (address copied to buffer).
2256  * - EWOULDBLOCK (no local error, but address is still unresolved)
2257  * - other errors (alloc failure, etc)
2258  */
2259 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)2260 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2261     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2262     struct llentry **plle)
2263 {
2264 	struct llentry *ln = NULL;
2265 	const struct sockaddr_in6 *dst6;
2266 
2267 	NET_EPOCH_ASSERT();
2268 
2269 	if (pflags != NULL)
2270 		*pflags = 0;
2271 
2272 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2273 
2274 	/* discard the packet if IPv6 operation is disabled on the interface */
2275 	if ((ifp->if_inet6->nd_flags & ND6_IFF_IFDISABLED)) {
2276 		m_freem(m);
2277 		return (ENETDOWN); /* better error? */
2278 	}
2279 
2280 	if (m != NULL && m->m_flags & M_MCAST) {
2281 		switch (ifp->if_type) {
2282 		case IFT_ETHER:
2283 		case IFT_L2VLAN:
2284 		case IFT_BRIDGE:
2285 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2286 						 desten);
2287 			return (0);
2288 		default:
2289 			m_freem(m);
2290 			return (EAFNOSUPPORT);
2291 		}
2292 	}
2293 
2294 	int family = gw_flags >> 16;
2295 	int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2296 	ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2297 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2298 		/* Entry found, let's copy lle info */
2299 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2300 		if (pflags != NULL)
2301 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2302 		llentry_provide_feedback(ln);
2303 		if (plle) {
2304 			LLE_ADDREF(ln);
2305 			*plle = ln;
2306 			LLE_WUNLOCK(ln);
2307 		}
2308 		return (0);
2309 	} else if (plle && ln)
2310 		LLE_WUNLOCK(ln);
2311 
2312 	return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2313 }
2314 
2315 /*
2316  * Finds or creates a new llentry for @addr and @family.
2317  * Returns wlocked llentry or NULL.
2318  *
2319  *
2320  * Child LLEs.
2321  *
2322  * Do not have their own state machine (gets marked as static)
2323  *  settimer bails out for child LLEs just in case.
2324  *
2325  * Locking order: parent lle gets locked first, chen goes the child.
2326  */
2327 static __noinline struct llentry *
nd6_get_llentry(struct ifnet * ifp,const struct in6_addr * addr,int family)2328 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2329 {
2330 	struct llentry *child_lle = NULL;
2331 	struct llentry *lle, *lle_tmp;
2332 
2333 	lle = nd6_alloc(addr, 0, ifp);
2334 	if (lle != NULL && family != AF_INET6) {
2335 		child_lle = nd6_alloc(addr, 0, ifp);
2336 		if (child_lle == NULL) {
2337 			lltable_free_entry(LLTABLE6(ifp), lle);
2338 			return (NULL);
2339 		}
2340 		child_lle->r_family = family;
2341 		child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2342 		child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2343 	}
2344 
2345 	if (lle == NULL) {
2346 		char ip6buf[INET6_ADDRSTRLEN];
2347 		log(LOG_DEBUG,
2348 		    "nd6_get_llentry: can't allocate llinfo for %s "
2349 		    "(ln=%p)\n",
2350 		    ip6_sprintf(ip6buf, addr), lle);
2351 		return (NULL);
2352 	}
2353 
2354 	LLTABLE_LOCK(LLTABLE6(ifp));
2355 	LLE_WLOCK(lle);
2356 	/* Prefer any existing entry over newly-created one */
2357 	lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2358 	if (lle_tmp == NULL)
2359 		lltable_link_entry(LLTABLE6(ifp), lle);
2360 	else {
2361 		lltable_free_entry(LLTABLE6(ifp), lle);
2362 		lle = lle_tmp;
2363 	}
2364 	if (child_lle != NULL) {
2365 		/* Check if child lle for the same family exists */
2366 		lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2367 		LLE_WLOCK(child_lle);
2368 		if (lle_tmp == NULL) {
2369 			/* Attach */
2370 			lltable_link_child_entry(lle, child_lle);
2371 		} else {
2372 			/* child lle already exists, free newly-created one */
2373 			lltable_free_entry(LLTABLE6(ifp), child_lle);
2374 			LLE_WLOCK(lle_tmp);
2375 			child_lle = lle_tmp;
2376 		}
2377 		LLE_WUNLOCK(lle);
2378 		lle = child_lle;
2379 	}
2380 	LLTABLE_UNLOCK(LLTABLE6(ifp));
2381 	return (lle);
2382 }
2383 
2384 /*
2385  * Do L2 address resolution for @sa_dst address. Stores found
2386  * address in @desten buffer. Copy of lle ln_flags can be also
2387  * saved in @pflags if @pflags is non-NULL.
2388  *
2389  * Heavy version.
2390  * Function assume that destination LLE does not exist,
2391  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2392  *
2393  * Set noinline to be dtrace-friendly
2394  */
2395 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)2396 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2397     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2398     struct llentry **plle)
2399 {
2400 	struct llentry *lle = NULL;
2401 	struct in6_addr *psrc, src;
2402 	int send_ns, ll_len;
2403 	char *lladdr;
2404 
2405 	NET_EPOCH_ASSERT();
2406 
2407 	/*
2408 	 * Address resolution or Neighbor Unreachability Detection
2409 	 * for the next hop.
2410 	 * At this point, the destination of the packet must be a unicast
2411 	 * or an anycast address(i.e. not a multicast).
2412 	 */
2413 	lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2414 	if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2415 		/*
2416 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2417 		 * the condition below is not very efficient.  But we believe
2418 		 * it is tolerable, because this should be a rare case.
2419 		 */
2420 		lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2421 	}
2422 
2423 	if (lle == NULL) {
2424 		m_freem(m);
2425 		return (ENOBUFS);
2426 	}
2427 
2428 	LLE_WLOCK_ASSERT(lle);
2429 
2430 	/*
2431 	 * The first time we send a packet to a neighbor whose entry is
2432 	 * STALE, we have to change the state to DELAY and a sets a timer to
2433 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2434 	 * neighbor unreachability detection on expiration.
2435 	 * (RFC 2461 7.3.3)
2436 	 */
2437 	if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2438 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2439 
2440 	/*
2441 	 * If the neighbor cache entry has a state other than INCOMPLETE
2442 	 * (i.e. its link-layer address is already resolved), just
2443 	 * send the packet.
2444 	 */
2445 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2446 		if (flags & LLE_ADDRONLY) {
2447 			lladdr = lle->ll_addr;
2448 			ll_len = ifp->if_addrlen;
2449 		} else {
2450 			lladdr = lle->r_linkdata;
2451 			ll_len = lle->r_hdrlen;
2452 		}
2453 		bcopy(lladdr, desten, ll_len);
2454 		if (pflags != NULL)
2455 			*pflags = lle->la_flags;
2456 		if (plle) {
2457 			LLE_ADDREF(lle);
2458 			*plle = lle;
2459 		}
2460 		LLE_WUNLOCK(lle);
2461 		return (0);
2462 	}
2463 
2464 	/*
2465 	 * There is a neighbor cache entry, but no ethernet address
2466 	 * response yet.  Append this latest packet to the end of the
2467 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2468 	 * the oldest packet in the queue will be removed.
2469 	 */
2470 	if (m != NULL) {
2471 		size_t dropped;
2472 
2473 		dropped = lltable_append_entry_queue(lle, m, V_nd6_maxqueuelen);
2474 		ICMP6STAT_ADD(icp6s_dropped, dropped);
2475 	}
2476 
2477 	/*
2478 	 * If there has been no NS for the neighbor after entering the
2479 	 * INCOMPLETE state, send the first solicitation.
2480 	 * Note that for newly-created lle la_asked will be 0,
2481 	 * so we will transition from ND6_LLINFO_NOSTATE to
2482 	 * ND6_LLINFO_INCOMPLETE state here.
2483 	 */
2484 	psrc = NULL;
2485 	send_ns = 0;
2486 
2487 	/* If we have child lle, switch to the parent to send NS */
2488 	if (lle->la_flags & LLE_CHILD) {
2489 		struct llentry *lle_parent = lle->lle_parent;
2490 		LLE_WUNLOCK(lle);
2491 		lle = lle_parent;
2492 		LLE_WLOCK(lle);
2493 	}
2494 	if (lle->la_asked == 0) {
2495 		lle->la_asked++;
2496 		send_ns = 1;
2497 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2498 
2499 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2500 	}
2501 	LLE_WUNLOCK(lle);
2502 	if (send_ns != 0)
2503 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2504 
2505 	return (EWOULDBLOCK);
2506 }
2507 
2508 /*
2509  * Do L2 address resolution for @sa_dst address. Stores found
2510  * address in @desten buffer. Copy of lle ln_flags can be also
2511  * saved in @pflags if @pflags is non-NULL.
2512  *
2513  * Return values:
2514  * - 0 on success (address copied to buffer).
2515  * - EWOULDBLOCK (no local error, but address is still unresolved)
2516  * - other errors (alloc failure, etc)
2517  */
2518 int
nd6_resolve_addr(struct ifnet * ifp,int flags,const struct sockaddr * dst,char * desten,uint32_t * pflags)2519 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2520     char *desten, uint32_t *pflags)
2521 {
2522 	int error;
2523 
2524 	flags |= LLE_ADDRONLY;
2525 	error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2526 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2527 	return (error);
2528 }
2529 
2530 int
nd6_flush_holdchain(struct ifnet * ifp,struct llentry * lle,struct mbuf * chain)2531 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2532 {
2533 	struct mbuf *m, *m_head;
2534 	struct sockaddr_in6 dst6;
2535 	int error = 0;
2536 
2537 	NET_EPOCH_ASSERT();
2538 
2539 	struct route_in6 ro = {
2540 		.ro_prepend = lle->r_linkdata,
2541 		.ro_plen = lle->r_hdrlen,
2542 	};
2543 
2544 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2545 	m_head = chain;
2546 
2547 	while (m_head) {
2548 		m = m_head;
2549 		m_head = m_head->m_nextpkt;
2550 		m->m_nextpkt = NULL;
2551 		error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2552 	}
2553 
2554 	/*
2555 	 * XXX
2556 	 * note that intermediate errors are blindly ignored
2557 	 */
2558 	return (error);
2559 }
2560 
2561 __noinline void
nd6_flush_children_holdchain(struct ifnet * ifp,struct llentry * lle)2562 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2563 {
2564 	struct llentry *child_lle;
2565 	struct mbuf *chain;
2566 
2567 	NET_EPOCH_ASSERT();
2568 
2569 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2570 		LLE_WLOCK(child_lle);
2571 		chain = nd6_grab_holdchain(child_lle);
2572 		LLE_WUNLOCK(child_lle);
2573 		nd6_flush_holdchain(ifp, child_lle, chain);
2574 	}
2575 }
2576 
2577 static int
nd6_need_cache(struct ifnet * ifp)2578 nd6_need_cache(struct ifnet *ifp)
2579 {
2580 	/*
2581 	 * XXX: we currently do not make neighbor cache on any interface
2582 	 * other than Ethernet and GIF.
2583 	 *
2584 	 * RFC2893 says:
2585 	 * - unidirectional tunnels needs no ND
2586 	 */
2587 	switch (ifp->if_type) {
2588 	case IFT_ETHER:
2589 	case IFT_IEEE1394:
2590 	case IFT_L2VLAN:
2591 	case IFT_INFINIBAND:
2592 	case IFT_BRIDGE:
2593 	case IFT_PROPVIRTUAL:
2594 		return (1);
2595 	default:
2596 		return (0);
2597 	}
2598 }
2599 
2600 /*
2601  * Add pernament ND6 link-layer record for given
2602  * interface address.
2603  *
2604  * Very similar to IPv4 arp_ifinit(), but:
2605  * 1) IPv6 DAD is performed in different place
2606  * 2) It is called by IPv6 protocol stack in contrast to
2607  * arp_ifinit() which is typically called in SIOCSIFADDR
2608  * driver ioctl handler.
2609  *
2610  */
2611 int
nd6_add_ifa_lle(struct in6_ifaddr * ia)2612 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2613 {
2614 	struct ifnet *ifp;
2615 	struct llentry *ln, *ln_tmp;
2616 	struct sockaddr *dst;
2617 
2618 	ifp = ia->ia_ifa.ifa_ifp;
2619 	if (nd6_need_cache(ifp) == 0)
2620 		return (0);
2621 
2622 	dst = (struct sockaddr *)&ia->ia_addr;
2623 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2624 	if (ln == NULL)
2625 		return (ENOBUFS);
2626 
2627 	LLTABLE_LOCK(LLTABLE6(ifp));
2628 	LLE_WLOCK(ln);
2629 	/* Unlink any entry if exists */
2630 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2631 	if (ln_tmp != NULL)
2632 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2633 	lltable_link_entry(LLTABLE6(ifp), ln);
2634 	LLTABLE_UNLOCK(LLTABLE6(ifp));
2635 
2636 	if (ln_tmp != NULL)
2637 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2638 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2639 
2640 	LLE_WUNLOCK(ln);
2641 	if (ln_tmp != NULL)
2642 		llentry_free(ln_tmp);
2643 
2644 	return (0);
2645 }
2646 
2647 /*
2648  * Removes either all lle entries for given @ia, or lle
2649  * corresponding to @ia address.
2650  */
2651 void
nd6_rem_ifa_lle(struct in6_ifaddr * ia,int all)2652 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2653 {
2654 	struct sockaddr_in6 mask, addr;
2655 	struct sockaddr *saddr, *smask;
2656 	struct ifnet *ifp;
2657 
2658 	ifp = ia->ia_ifa.ifa_ifp;
2659 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2660 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2661 	saddr = (struct sockaddr *)&addr;
2662 	smask = (struct sockaddr *)&mask;
2663 
2664 	if (all != 0)
2665 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2666 	else
2667 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2668 }
2669 
2670 static int
nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)2671 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2672 {
2673 	struct in6_prefix p;
2674 	struct sockaddr_in6 s6;
2675 	struct nd_prefix *pr;
2676 	struct nd_pfxrouter *pfr;
2677 	time_t maxexpire;
2678 	int error;
2679 	char ip6buf[INET6_ADDRSTRLEN];
2680 
2681 	if (req->newptr)
2682 		return (EPERM);
2683 
2684 	error = sysctl_wire_old_buffer(req, 0);
2685 	if (error != 0)
2686 		return (error);
2687 
2688 	bzero(&p, sizeof(p));
2689 	p.origin = PR_ORIG_RA;
2690 	bzero(&s6, sizeof(s6));
2691 	s6.sin6_family = AF_INET6;
2692 	s6.sin6_len = sizeof(s6);
2693 
2694 	ND6_RLOCK();
2695 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2696 		p.prefix = pr->ndpr_prefix;
2697 		if (sa6_recoverscope(&p.prefix)) {
2698 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2699 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2700 			/* XXX: press on... */
2701 		}
2702 		p.raflags = pr->ndpr_raf;
2703 		p.prefixlen = pr->ndpr_plen;
2704 		p.vltime = pr->ndpr_vltime;
2705 		p.pltime = pr->ndpr_pltime;
2706 		p.if_index = pr->ndpr_ifp->if_index;
2707 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2708 			p.expire = 0;
2709 		else {
2710 			/* XXX: we assume time_t is signed. */
2711 			maxexpire = (-1) &
2712 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2713 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2714 				p.expire = pr->ndpr_lastupdate +
2715 				    pr->ndpr_vltime +
2716 				    (time_second - time_uptime);
2717 			else
2718 				p.expire = maxexpire;
2719 		}
2720 		p.refcnt = pr->ndpr_addrcnt;
2721 		p.flags = pr->ndpr_stateflags;
2722 		p.advrtrs = 0;
2723 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2724 			p.advrtrs++;
2725 		error = SYSCTL_OUT(req, &p, sizeof(p));
2726 		if (error != 0)
2727 			break;
2728 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2729 			s6.sin6_addr = pfr->router->rtaddr;
2730 			if (sa6_recoverscope(&s6))
2731 				log(LOG_ERR,
2732 				    "scope error in prefix list (%s)\n",
2733 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2734 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2735 			if (error != 0)
2736 				goto out;
2737 		}
2738 	}
2739 out:
2740 	ND6_RUNLOCK();
2741 	return (error);
2742 }
2743 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2744 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2745 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2746 	"NDP prefix list");
2747 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2748 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2749 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2750 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2751