xref: /freebsd/sys/netinet6/nd6.c (revision 0c891822310cebd11c22bb59391f85bc53c94fa3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the project nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	$KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32  */
33 
34 #include <sys/cdefs.h>
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_route.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/eventhandler.h>
42 #include <sys/callout.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/mutex.h>
47 #include <sys/socket.h>
48 #include <sys/sockio.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/protosw.h>
52 #include <sys/errno.h>
53 #include <sys/syslog.h>
54 #include <sys/rwlock.h>
55 #include <sys/queue.h>
56 #include <sys/sdt.h>
57 #include <sys/sysctl.h>
58 
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_private.h>
63 #include <net/if_types.h>
64 #include <net/route.h>
65 #include <net/route/route_ctl.h>
66 #include <net/route/nhop.h>
67 #include <net/vnet.h>
68 
69 #include <netinet/in.h>
70 #include <netinet/in_kdtrace.h>
71 #include <net/if_llatbl.h>
72 #include <netinet/if_ether.h>
73 #include <netinet6/in6_fib.h>
74 #include <netinet6/in6_var.h>
75 #include <netinet/ip6.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
79 #include <netinet6/in6_ifattach.h>
80 #include <netinet/icmp6.h>
81 #include <netinet6/send.h>
82 
83 #include <sys/limits.h>
84 
85 #include <security/mac/mac_framework.h>
86 
87 #define	ND6_PREFIX_WITH_ROUTER(pr)	!LIST_EMPTY(&(pr)->ndpr_advrtrs)
88 
89 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
90 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
91 
92 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
93 
94 VNET_DEFINE_STATIC(int, nd6_prune) = 1;
95 #define	V_nd6_prune	VNET(nd6_prune)
96 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_PRUNE, nd6_prune,
97     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_prune), 0,
98     "Frequency in seconds of checks for expired prefixes and routers");
99 
100 VNET_DEFINE_STATIC(int, nd6_delay) = 5;
101 #define	V_nd6_delay	VNET(nd6_delay)
102 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DELAY, nd6_delay,
103     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_delay), 0,
104     "Delay in seconds before probing for reachability");
105 
106 VNET_DEFINE_STATIC(int, nd6_umaxtries) = 3;
107 #define	V_nd6_umaxtries	VNET(nd6_umaxtries)
108 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_UMAXTRIES, nd6_umaxtries,
109     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_umaxtries), 0,
110     "Number of ICMPv6 NS messages sent during reachability detection");
111 
112 VNET_DEFINE(int, nd6_mmaxtries) = 3;
113 #define	V_nd6_mmaxtries	VNET(nd6_mmaxtries)
114 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MMAXTRIES, nd6_mmaxtries,
115     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_mmaxtries), 0,
116     "Number of ICMPv6 NS messages sent during address resolution");
117 
118 VNET_DEFINE_STATIC(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage
119 							* collection timer */
120 #define	V_nd6_gctimer	VNET(nd6_gctimer)
121 
122 /* preventing too many loops in ND option parsing */
123 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
124 
125 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 16; /* max pkts cached in unresolved
126 					 * ND entries */
127 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
128 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
129 
130 #ifdef ND6_DEBUG
131 VNET_DEFINE(int, nd6_debug) = 1;
132 #else
133 VNET_DEFINE(int, nd6_debug) = 0;
134 #endif
135 #define	V_nd6_debug	VNET(nd6_debug)
136 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DEBUG, nd6_debug,
137     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_debug), 0,
138     "Log NDP debug messages");
139 
140 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
141 
142 VNET_DEFINE(struct nd_prhead, nd_prefix);
143 VNET_DEFINE(struct rwlock, nd6_lock);
144 VNET_DEFINE(uint64_t, nd6_list_genid);
145 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
146 
147 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
148 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
149 
150 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
151 
152 static bool nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
153 	struct ifnet *);
154 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
155 static void nd6_slowtimo(void *);
156 static int regen_tmpaddr(struct in6_ifaddr *);
157 static void nd6_free(struct llentry **, int);
158 static void nd6_free_redirect(const struct llentry *);
159 static void nd6_llinfo_timer(void *);
160 static void nd6_llinfo_settimer_locked(struct llentry *, long);
161 static int nd6_resolve_slow(struct ifnet *, int, int, struct mbuf *,
162     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
163 static int nd6_need_cache(struct ifnet *);
164 
165 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
166 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
167 
168 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
169 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
170 
171 static void
nd6_lle_event(void * arg __unused,struct llentry * lle,int evt)172 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
173 {
174 	struct rt_addrinfo rtinfo;
175 	struct sockaddr_in6 dst;
176 	struct sockaddr_dl gw;
177 	struct ifnet *ifp;
178 	int type;
179 	int fibnum;
180 
181 	LLE_WLOCK_ASSERT(lle);
182 
183 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
184 		return;
185 
186 	switch (evt) {
187 	case LLENTRY_RESOLVED:
188 		type = RTM_ADD;
189 		KASSERT(lle->la_flags & LLE_VALID,
190 		    ("%s: %p resolved but not valid?", __func__, lle));
191 		break;
192 	case LLENTRY_EXPIRED:
193 		type = RTM_DELETE;
194 		break;
195 	default:
196 		return;
197 	}
198 
199 	ifp = lltable_get_ifp(lle->lle_tbl);
200 
201 	bzero(&dst, sizeof(dst));
202 	bzero(&gw, sizeof(gw));
203 	bzero(&rtinfo, sizeof(rtinfo));
204 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
205 	dst.sin6_scope_id = in6_getscopezone(ifp,
206 	    in6_addrscope(&dst.sin6_addr));
207 	gw.sdl_len = sizeof(struct sockaddr_dl);
208 	gw.sdl_family = AF_LINK;
209 	gw.sdl_alen = ifp->if_addrlen;
210 	gw.sdl_index = ifp->if_index;
211 	gw.sdl_type = ifp->if_type;
212 	if (evt == LLENTRY_RESOLVED)
213 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
214 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
215 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
216 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
217 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
218 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
219 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
220 }
221 
222 /*
223  * A handler for interface link layer address change event.
224  */
225 static void
nd6_iflladdr(void * arg __unused,struct ifnet * ifp)226 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
227 {
228 	if (ifp->if_afdata[AF_INET6] == NULL)
229 		return;
230 
231 	lltable_update_ifaddr(LLTABLE6(ifp));
232 }
233 
234 void
nd6_init(void)235 nd6_init(void)
236 {
237 
238 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
239 	rw_init(&V_nd6_lock, "nd6 list");
240 
241 	LIST_INIT(&V_nd_prefix);
242 	nd6_defrouter_init();
243 
244 	/* Start timers. */
245 	callout_init(&V_nd6_slowtimo_ch, 1);
246 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
247 	    nd6_slowtimo, curvnet);
248 
249 	callout_init(&V_nd6_timer_ch, 1);
250 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
251 
252 	nd6_dad_init();
253 	if (IS_DEFAULT_VNET(curvnet)) {
254 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
255 		    NULL, EVENTHANDLER_PRI_ANY);
256 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
257 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
258 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
259 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
260 	}
261 }
262 
263 #ifdef VIMAGE
264 void
nd6_destroy(void)265 nd6_destroy(void)
266 {
267 
268 	callout_drain(&V_nd6_slowtimo_ch);
269 	callout_drain(&V_nd6_timer_ch);
270 	if (IS_DEFAULT_VNET(curvnet)) {
271 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
272 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
273 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
274 	}
275 	rw_destroy(&V_nd6_lock);
276 	mtx_destroy(&V_nd6_onlink_mtx);
277 }
278 #endif
279 
280 struct nd_ifinfo *
nd6_ifattach(struct ifnet * ifp)281 nd6_ifattach(struct ifnet *ifp)
282 {
283 	struct nd_ifinfo *nd;
284 
285 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
286 	nd->initialized = 1;
287 
288 	nd->chlim = IPV6_DEFHLIM;
289 	nd->basereachable = REACHABLE_TIME;
290 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
291 	nd->retrans = RETRANS_TIMER;
292 
293 	nd->flags = ND6_IFF_PERFORMNUD;
294 
295 	/* Set IPv6 disabled on all interfaces but loopback by default. */
296 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
297 		nd->flags |= ND6_IFF_IFDISABLED;
298 
299 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
300 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
301 	 * default regardless of the V_ip6_auto_linklocal configuration to
302 	 * give a reasonable default behavior.
303 	 */
304 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE &&
305 	    ifp->if_type != IFT_WIREGUARD) || (ifp->if_flags & IFF_LOOPBACK))
306 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
307 	/*
308 	 * A loopback interface does not need to accept RTADV.
309 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
310 	 * default regardless of the V_ip6_accept_rtadv configuration to
311 	 * prevent the interface from accepting RA messages arrived
312 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
313 	 */
314 	if (V_ip6_accept_rtadv &&
315 	    !(ifp->if_flags & IFF_LOOPBACK) &&
316 	    (ifp->if_type != IFT_BRIDGE)) {
317 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
318 			/* If we globally accept rtadv, assume IPv6 on. */
319 			nd->flags &= ~ND6_IFF_IFDISABLED;
320 	}
321 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
322 		nd->flags |= ND6_IFF_NO_RADR;
323 
324 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
325 	nd6_setmtu0(ifp, nd);
326 
327 	return nd;
328 }
329 
330 void
nd6_ifdetach(struct ifnet * ifp,struct nd_ifinfo * nd)331 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
332 {
333 	struct epoch_tracker et;
334 	struct ifaddr *ifa, *next;
335 
336 	NET_EPOCH_ENTER(et);
337 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
338 		if (ifa->ifa_addr->sa_family != AF_INET6)
339 			continue;
340 
341 		/* stop DAD processing */
342 		nd6_dad_stop(ifa);
343 	}
344 	NET_EPOCH_EXIT(et);
345 
346 	free(nd, M_IP6NDP);
347 }
348 
349 /*
350  * Reset ND level link MTU. This function is called when the physical MTU
351  * changes, which means we might have to adjust the ND level MTU.
352  */
353 void
nd6_setmtu(struct ifnet * ifp)354 nd6_setmtu(struct ifnet *ifp)
355 {
356 	if (ifp->if_afdata[AF_INET6] == NULL)
357 		return;
358 
359 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
360 }
361 
362 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
363 void
nd6_setmtu0(struct ifnet * ifp,struct nd_ifinfo * ndi)364 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
365 {
366 	u_int32_t omaxmtu;
367 
368 	omaxmtu = ndi->maxmtu;
369 	ndi->maxmtu = ifp->if_mtu;
370 
371 	/*
372 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
373 	 * undesirable situation.  We thus notify the operator of the change
374 	 * explicitly.  The check for omaxmtu is necessary to restrict the
375 	 * log to the case of changing the MTU, not initializing it.
376 	 */
377 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
378 		log(LOG_NOTICE, "nd6_setmtu0: "
379 		    "new link MTU on %s (%lu) is too small for IPv6\n",
380 		    if_name(ifp), (unsigned long)ndi->maxmtu);
381 	}
382 }
383 
384 void
nd6_option_init(void * opt,int icmp6len,union nd_opts * ndopts)385 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
386 {
387 
388 	bzero(ndopts, sizeof(*ndopts));
389 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
390 	ndopts->nd_opts_last
391 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
392 
393 	if (icmp6len == 0) {
394 		ndopts->nd_opts_done = 1;
395 		ndopts->nd_opts_search = NULL;
396 	}
397 }
398 
399 /*
400  * Take one ND option.
401  */
402 struct nd_opt_hdr *
nd6_option(union nd_opts * ndopts)403 nd6_option(union nd_opts *ndopts)
404 {
405 	struct nd_opt_hdr *nd_opt;
406 	int olen;
407 
408 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
409 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
410 	    __func__));
411 	if (ndopts->nd_opts_search == NULL)
412 		return NULL;
413 	if (ndopts->nd_opts_done)
414 		return NULL;
415 
416 	nd_opt = ndopts->nd_opts_search;
417 
418 	/* make sure nd_opt_len is inside the buffer */
419 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
420 		bzero(ndopts, sizeof(*ndopts));
421 		return NULL;
422 	}
423 
424 	olen = nd_opt->nd_opt_len << 3;
425 	if (olen == 0) {
426 		/*
427 		 * Message validation requires that all included
428 		 * options have a length that is greater than zero.
429 		 */
430 		bzero(ndopts, sizeof(*ndopts));
431 		return NULL;
432 	}
433 
434 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
435 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
436 		/* option overruns the end of buffer, invalid */
437 		bzero(ndopts, sizeof(*ndopts));
438 		return NULL;
439 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
440 		/* reached the end of options chain */
441 		ndopts->nd_opts_done = 1;
442 		ndopts->nd_opts_search = NULL;
443 	}
444 	return nd_opt;
445 }
446 
447 /*
448  * Parse multiple ND options.
449  * This function is much easier to use, for ND routines that do not need
450  * multiple options of the same type.
451  */
452 int
nd6_options(union nd_opts * ndopts)453 nd6_options(union nd_opts *ndopts)
454 {
455 	struct nd_opt_hdr *nd_opt;
456 	int i = 0;
457 
458 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
459 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
460 	    __func__));
461 	if (ndopts->nd_opts_search == NULL)
462 		return 0;
463 
464 	while (1) {
465 		nd_opt = nd6_option(ndopts);
466 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
467 			/*
468 			 * Message validation requires that all included
469 			 * options have a length that is greater than zero.
470 			 */
471 			ICMP6STAT_INC(icp6s_nd_badopt);
472 			bzero(ndopts, sizeof(*ndopts));
473 			return -1;
474 		}
475 
476 		if (nd_opt == NULL)
477 			goto skip1;
478 
479 		switch (nd_opt->nd_opt_type) {
480 		case ND_OPT_SOURCE_LINKADDR:
481 		case ND_OPT_TARGET_LINKADDR:
482 		case ND_OPT_MTU:
483 		case ND_OPT_REDIRECTED_HEADER:
484 		case ND_OPT_NONCE:
485 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
486 				nd6log((LOG_INFO,
487 				    "duplicated ND6 option found (type=%d)\n",
488 				    nd_opt->nd_opt_type));
489 				/* XXX bark? */
490 			} else {
491 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
492 					= nd_opt;
493 			}
494 			break;
495 		case ND_OPT_PREFIX_INFORMATION:
496 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
497 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
498 					= nd_opt;
499 			}
500 			ndopts->nd_opts_pi_end =
501 				(struct nd_opt_prefix_info *)nd_opt;
502 			break;
503 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
504 		case ND_OPT_RDNSS:	/* RFC 6106 */
505 		case ND_OPT_DNSSL:	/* RFC 6106 */
506 			/*
507 			 * Silently ignore options we know and do not care about
508 			 * in the kernel.
509 			 */
510 			break;
511 		default:
512 			/*
513 			 * Unknown options must be silently ignored,
514 			 * to accommodate future extension to the protocol.
515 			 */
516 			nd6log((LOG_DEBUG,
517 			    "nd6_options: unsupported option %d - "
518 			    "option ignored\n", nd_opt->nd_opt_type));
519 		}
520 
521 skip1:
522 		i++;
523 		if (i > V_nd6_maxndopt) {
524 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
525 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
526 			break;
527 		}
528 
529 		if (ndopts->nd_opts_done)
530 			break;
531 	}
532 
533 	return 0;
534 }
535 
536 /*
537  * ND6 timer routine to handle ND6 entries
538  */
539 static void
nd6_llinfo_settimer_locked(struct llentry * ln,long tick)540 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
541 {
542 	int canceled;
543 
544 	LLE_WLOCK_ASSERT(ln);
545 
546 	/* Do not schedule timers for child LLEs. */
547 	if (ln->la_flags & LLE_CHILD)
548 		return;
549 
550 	if (tick < 0) {
551 		ln->la_expire = 0;
552 		ln->ln_ntick = 0;
553 		canceled = callout_stop(&ln->lle_timer);
554 	} else {
555 		ln->la_expire = time_uptime + tick / hz;
556 		LLE_ADDREF(ln);
557 		if (tick > INT_MAX) {
558 			ln->ln_ntick = tick - INT_MAX;
559 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
560 			    nd6_llinfo_timer, ln);
561 		} else {
562 			ln->ln_ntick = 0;
563 			canceled = callout_reset(&ln->lle_timer, tick,
564 			    nd6_llinfo_timer, ln);
565 		}
566 	}
567 	if (canceled > 0)
568 		LLE_REMREF(ln);
569 }
570 
571 /*
572  * Gets source address of the first packet in hold queue
573  * and stores it in @src.
574  * Returns pointer to @src (if hold queue is not empty) or NULL.
575  *
576  * Set noinline to be dtrace-friendly
577  */
578 static __noinline struct in6_addr *
nd6_llinfo_get_holdsrc(struct llentry * ln,struct in6_addr * src)579 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
580 {
581 	struct ip6_hdr hdr;
582 	struct mbuf *m;
583 
584 	if (ln->la_hold == NULL)
585 		return (NULL);
586 
587 	/*
588 	 * assume every packet in la_hold has the same IP header
589 	 */
590 	m = ln->la_hold;
591 	if (sizeof(hdr) > m->m_len)
592 		return (NULL);
593 
594 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
595 	*src = hdr.ip6_src;
596 
597 	return (src);
598 }
599 
600 /*
601  * Checks if we need to switch from STALE state.
602  *
603  * RFC 4861 requires switching from STALE to DELAY state
604  * on first packet matching entry, waiting V_nd6_delay and
605  * transition to PROBE state (if upper layer confirmation was
606  * not received).
607  *
608  * This code performs a bit differently:
609  * On packet hit we don't change state (but desired state
610  * can be guessed by control plane). However, after V_nd6_delay
611  * seconds code will transition to PROBE state (so DELAY state
612  * is kinda skipped in most situations).
613  *
614  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
615  * we perform the following upon entering STALE state:
616  *
617  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
618  * if packet was transmitted at the start of given interval, we
619  * would be able to switch to PROBE state in V_nd6_delay seconds
620  * as user expects.
621  *
622  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
623  * lle in STALE state (remaining timer value stored in lle_remtime).
624  *
625  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
626  * seconds ago.
627  *
628  * Returns non-zero value if the entry is still STALE (storing
629  * the next timer interval in @pdelay).
630  *
631  * Returns zero value if original timer expired or we need to switch to
632  * PROBE (store that in @do_switch variable).
633  */
634 static int
nd6_is_stale(struct llentry * lle,long * pdelay,int * do_switch)635 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
636 {
637 	int nd_delay, nd_gctimer;
638 	time_t lle_hittime;
639 	long delay;
640 
641 	*do_switch = 0;
642 	nd_gctimer = V_nd6_gctimer;
643 	nd_delay = V_nd6_delay;
644 
645 	lle_hittime = llentry_get_hittime(lle);
646 
647 	if (lle_hittime == 0) {
648 		/*
649 		 * Datapath feedback has been requested upon entering
650 		 * STALE state. No packets has been passed using this lle.
651 		 * Ask for the timer reschedule and keep STALE state.
652 		 */
653 		delay = (long)(MIN(nd_gctimer, nd_delay));
654 		delay *= hz;
655 		if (lle->lle_remtime > delay)
656 			lle->lle_remtime -= delay;
657 		else {
658 			delay = lle->lle_remtime;
659 			lle->lle_remtime = 0;
660 		}
661 
662 		if (delay == 0) {
663 			/*
664 			 * The original ng6_gctime timeout ended,
665 			 * no more rescheduling.
666 			 */
667 			return (0);
668 		}
669 
670 		*pdelay = delay;
671 		return (1);
672 	}
673 
674 	/*
675 	 * Packet received. Verify timestamp
676 	 */
677 	delay = (long)(time_uptime - lle_hittime);
678 	if (delay < nd_delay) {
679 		/*
680 		 * V_nd6_delay still not passed since the first
681 		 * hit in STALE state.
682 		 * Reschedule timer and return.
683 		 */
684 		*pdelay = (long)(nd_delay - delay) * hz;
685 		return (1);
686 	}
687 
688 	/* Request switching to probe */
689 	*do_switch = 1;
690 	return (0);
691 }
692 
693 /*
694  * Switch @lle state to new state optionally arming timers.
695  *
696  * Set noinline to be dtrace-friendly
697  */
698 __noinline void
nd6_llinfo_setstate(struct llentry * lle,int newstate)699 nd6_llinfo_setstate(struct llentry *lle, int newstate)
700 {
701 	struct ifnet *ifp;
702 	int nd_gctimer, nd_delay;
703 	long delay, remtime;
704 
705 	delay = 0;
706 	remtime = 0;
707 
708 	switch (newstate) {
709 	case ND6_LLINFO_INCOMPLETE:
710 		ifp = lle->lle_tbl->llt_ifp;
711 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
712 		break;
713 	case ND6_LLINFO_REACHABLE:
714 		if (!ND6_LLINFO_PERMANENT(lle)) {
715 			ifp = lle->lle_tbl->llt_ifp;
716 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
717 		}
718 		break;
719 	case ND6_LLINFO_STALE:
720 
721 		llentry_request_feedback(lle);
722 		nd_delay = V_nd6_delay;
723 		nd_gctimer = V_nd6_gctimer;
724 
725 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
726 		remtime = (long)nd_gctimer * hz - delay;
727 		break;
728 	case ND6_LLINFO_DELAY:
729 		lle->la_asked = 0;
730 		delay = (long)V_nd6_delay * hz;
731 		break;
732 	}
733 
734 	if (delay > 0)
735 		nd6_llinfo_settimer_locked(lle, delay);
736 
737 	lle->lle_remtime = remtime;
738 	lle->ln_state = newstate;
739 }
740 
741 /*
742  * Timer-dependent part of nd state machine.
743  *
744  * Set noinline to be dtrace-friendly
745  */
746 static __noinline void
nd6_llinfo_timer(void * arg)747 nd6_llinfo_timer(void *arg)
748 {
749 	struct epoch_tracker et;
750 	struct llentry *ln;
751 	struct in6_addr *dst, *pdst, *psrc, src;
752 	struct ifnet *ifp;
753 	struct nd_ifinfo *ndi;
754 	int do_switch, send_ns;
755 	long delay;
756 
757 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
758 	ln = (struct llentry *)arg;
759 	ifp = lltable_get_ifp(ln->lle_tbl);
760 	CURVNET_SET(ifp->if_vnet);
761 
762 	ND6_RLOCK();
763 	LLE_WLOCK(ln);
764 	if (callout_pending(&ln->lle_timer)) {
765 		/*
766 		 * Here we are a bit odd here in the treatment of
767 		 * active/pending. If the pending bit is set, it got
768 		 * rescheduled before I ran. The active
769 		 * bit we ignore, since if it was stopped
770 		 * in ll_tablefree() and was currently running
771 		 * it would have return 0 so the code would
772 		 * not have deleted it since the callout could
773 		 * not be stopped so we want to go through
774 		 * with the delete here now. If the callout
775 		 * was restarted, the pending bit will be back on and
776 		 * we just want to bail since the callout_reset would
777 		 * return 1 and our reference would have been removed
778 		 * by nd6_llinfo_settimer_locked above since canceled
779 		 * would have been 1.
780 		 */
781 		LLE_WUNLOCK(ln);
782 		ND6_RUNLOCK();
783 		CURVNET_RESTORE();
784 		return;
785 	}
786 	NET_EPOCH_ENTER(et);
787 	ndi = ND_IFINFO(ifp);
788 	send_ns = 0;
789 	dst = &ln->r_l3addr.addr6;
790 	pdst = dst;
791 
792 	if (ln->ln_ntick > 0) {
793 		if (ln->ln_ntick > INT_MAX) {
794 			ln->ln_ntick -= INT_MAX;
795 			nd6_llinfo_settimer_locked(ln, INT_MAX);
796 		} else {
797 			ln->ln_ntick = 0;
798 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
799 		}
800 		goto done;
801 	}
802 
803 	if (ln->la_flags & LLE_STATIC) {
804 		goto done;
805 	}
806 
807 	if (ln->la_flags & LLE_DELETED) {
808 		nd6_free(&ln, 0);
809 		goto done;
810 	}
811 
812 	switch (ln->ln_state) {
813 	case ND6_LLINFO_INCOMPLETE:
814 		if (ln->la_asked < V_nd6_mmaxtries) {
815 			ln->la_asked++;
816 			send_ns = 1;
817 			/* Send NS to multicast address */
818 			pdst = NULL;
819 		} else {
820 			struct mbuf *m;
821 
822 			ICMP6STAT_ADD(icp6s_dropped, ln->la_numheld);
823 
824 			m = ln->la_hold;
825 			if (m != NULL) {
826 				/*
827 				 * assuming every packet in la_hold has the
828 				 * same IP header.  Send error after unlock.
829 				 */
830 				ln->la_hold = m->m_nextpkt;
831 				m->m_nextpkt = NULL;
832 				ln->la_numheld--;
833 			}
834 			nd6_free(&ln, 0);
835 			if (m != NULL) {
836 				struct mbuf *n = m;
837 
838 				/*
839 				 * if there are any ummapped mbufs, we
840 				 * must free them, rather than using
841 				 * them for an ICMP, as they cannot be
842 				 * checksummed.
843 				 */
844 				while ((n = n->m_next) != NULL) {
845 					if (n->m_flags & M_EXTPG)
846 						break;
847 				}
848 				if (n != NULL) {
849 					m_freem(m);
850 					m = NULL;
851 				} else {
852 					icmp6_error2(m, ICMP6_DST_UNREACH,
853 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
854 				}
855 			}
856 		}
857 		break;
858 	case ND6_LLINFO_REACHABLE:
859 		if (!ND6_LLINFO_PERMANENT(ln))
860 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
861 		break;
862 
863 	case ND6_LLINFO_STALE:
864 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
865 			/*
866 			 * No packet has used this entry and GC timeout
867 			 * has not been passed. Reschedule timer and
868 			 * return.
869 			 */
870 			nd6_llinfo_settimer_locked(ln, delay);
871 			break;
872 		}
873 
874 		if (do_switch == 0) {
875 			/*
876 			 * GC timer has ended and entry hasn't been used.
877 			 * Run Garbage collector (RFC 4861, 5.3)
878 			 */
879 			if (!ND6_LLINFO_PERMANENT(ln))
880 				nd6_free(&ln, 1);
881 			break;
882 		}
883 
884 		/* Entry has been used AND delay timer has ended. */
885 
886 		/* FALLTHROUGH */
887 
888 	case ND6_LLINFO_DELAY:
889 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
890 			/* We need NUD */
891 			ln->la_asked = 1;
892 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
893 			send_ns = 1;
894 		} else
895 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
896 		break;
897 	case ND6_LLINFO_PROBE:
898 		if (ln->la_asked < V_nd6_umaxtries) {
899 			ln->la_asked++;
900 			send_ns = 1;
901 		} else {
902 			nd6_free(&ln, 0);
903 		}
904 		break;
905 	default:
906 		panic("%s: paths in a dark night can be confusing: %d",
907 		    __func__, ln->ln_state);
908 	}
909 done:
910 	if (ln != NULL)
911 		ND6_RUNLOCK();
912 	if (send_ns != 0) {
913 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
914 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
915 		LLE_FREE_LOCKED(ln);
916 		ln = NULL;
917 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
918 	}
919 
920 	if (ln != NULL)
921 		LLE_FREE_LOCKED(ln);
922 	NET_EPOCH_EXIT(et);
923 	CURVNET_RESTORE();
924 }
925 
926 /*
927  * ND6 timer routine to expire default route list and prefix list
928  */
929 void
nd6_timer(void * arg)930 nd6_timer(void *arg)
931 {
932 	CURVNET_SET((struct vnet *) arg);
933 	struct epoch_tracker et;
934 	struct nd_prhead prl;
935 	struct nd_prefix *pr, *npr;
936 	struct ifnet *ifp;
937 	struct in6_ifaddr *ia6, *nia6;
938 	uint64_t genid;
939 
940 	LIST_INIT(&prl);
941 
942 	NET_EPOCH_ENTER(et);
943 	nd6_defrouter_timer();
944 
945 	/*
946 	 * expire interface addresses.
947 	 * in the past the loop was inside prefix expiry processing.
948 	 * However, from a stricter speci-confrmance standpoint, we should
949 	 * rather separate address lifetimes and prefix lifetimes.
950 	 *
951 	 * XXXRW: in6_ifaddrhead locking.
952 	 */
953   addrloop:
954 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
955 		/* check address lifetime */
956 		if (IFA6_IS_INVALID(ia6)) {
957 			int regen = 0;
958 
959 			/*
960 			 * If the expiring address is temporary, try
961 			 * regenerating a new one.  This would be useful when
962 			 * we suspended a laptop PC, then turned it on after a
963 			 * period that could invalidate all temporary
964 			 * addresses.  Although we may have to restart the
965 			 * loop (see below), it must be after purging the
966 			 * address.  Otherwise, we'd see an infinite loop of
967 			 * regeneration.
968 			 */
969 			if (V_ip6_use_tempaddr &&
970 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
971 				if (regen_tmpaddr(ia6) == 0)
972 					regen = 1;
973 			}
974 
975 			in6_purgeaddr(&ia6->ia_ifa);
976 
977 			if (regen)
978 				goto addrloop; /* XXX: see below */
979 		} else if (IFA6_IS_DEPRECATED(ia6)) {
980 			int oldflags = ia6->ia6_flags;
981 
982 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
983 
984 			/*
985 			 * If a temporary address has just become deprecated,
986 			 * regenerate a new one if possible.
987 			 */
988 			if (V_ip6_use_tempaddr &&
989 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
990 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
991 				if (regen_tmpaddr(ia6) == 0) {
992 					/*
993 					 * A new temporary address is
994 					 * generated.
995 					 * XXX: this means the address chain
996 					 * has changed while we are still in
997 					 * the loop.  Although the change
998 					 * would not cause disaster (because
999 					 * it's not a deletion, but an
1000 					 * addition,) we'd rather restart the
1001 					 * loop just for safety.  Or does this
1002 					 * significantly reduce performance??
1003 					 */
1004 					goto addrloop;
1005 				}
1006 			}
1007 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
1008 			/*
1009 			 * Schedule DAD for a tentative address.  This happens
1010 			 * if the interface was down or not running
1011 			 * when the address was configured.
1012 			 */
1013 			int delay;
1014 
1015 			delay = arc4random() %
1016 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1017 			nd6_dad_start((struct ifaddr *)ia6, delay);
1018 		} else {
1019 			/*
1020 			 * Check status of the interface.  If it is down,
1021 			 * mark the address as tentative for future DAD.
1022 			 */
1023 			ifp = ia6->ia_ifp;
1024 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1025 			    ((ifp->if_flags & IFF_UP) == 0 ||
1026 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1027 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1028 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1029 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1030 			}
1031 
1032 			/*
1033 			 * A new RA might have made a deprecated address
1034 			 * preferred.
1035 			 */
1036 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1037 		}
1038 	}
1039 	NET_EPOCH_EXIT(et);
1040 
1041 	ND6_WLOCK();
1042 restart:
1043 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1044 		/*
1045 		 * Expire prefixes. Since the pltime is only used for
1046 		 * autoconfigured addresses, pltime processing for prefixes is
1047 		 * not necessary.
1048 		 *
1049 		 * Only unlink after all derived addresses have expired. This
1050 		 * may not occur until two hours after the prefix has expired
1051 		 * per RFC 4862. If the prefix expires before its derived
1052 		 * addresses, mark it off-link. This will be done automatically
1053 		 * after unlinking if no address references remain.
1054 		 */
1055 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1056 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1057 			continue;
1058 
1059 		if (pr->ndpr_addrcnt == 0) {
1060 			nd6_prefix_unlink(pr, &prl);
1061 			continue;
1062 		}
1063 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1064 			genid = V_nd6_list_genid;
1065 			nd6_prefix_ref(pr);
1066 			ND6_WUNLOCK();
1067 			ND6_ONLINK_LOCK();
1068 			(void)nd6_prefix_offlink(pr);
1069 			ND6_ONLINK_UNLOCK();
1070 			ND6_WLOCK();
1071 			nd6_prefix_rele(pr);
1072 			if (genid != V_nd6_list_genid)
1073 				goto restart;
1074 		}
1075 	}
1076 	ND6_WUNLOCK();
1077 
1078 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1079 		LIST_REMOVE(pr, ndpr_entry);
1080 		nd6_prefix_del(pr);
1081 	}
1082 
1083 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1084 	    nd6_timer, curvnet);
1085 
1086 	CURVNET_RESTORE();
1087 }
1088 
1089 /*
1090  * ia6 - deprecated/invalidated temporary address
1091  */
1092 static int
regen_tmpaddr(struct in6_ifaddr * ia6)1093 regen_tmpaddr(struct in6_ifaddr *ia6)
1094 {
1095 	struct ifaddr *ifa;
1096 	struct ifnet *ifp;
1097 	struct in6_ifaddr *public_ifa6 = NULL;
1098 
1099 	NET_EPOCH_ASSERT();
1100 
1101 	ifp = ia6->ia_ifa.ifa_ifp;
1102 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1103 		struct in6_ifaddr *it6;
1104 
1105 		if (ifa->ifa_addr->sa_family != AF_INET6)
1106 			continue;
1107 
1108 		it6 = (struct in6_ifaddr *)ifa;
1109 
1110 		/* ignore no autoconf addresses. */
1111 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1112 			continue;
1113 
1114 		/* ignore autoconf addresses with different prefixes. */
1115 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1116 			continue;
1117 
1118 		/*
1119 		 * Now we are looking at an autoconf address with the same
1120 		 * prefix as ours.  If the address is temporary and is still
1121 		 * preferred, do not create another one.  It would be rare, but
1122 		 * could happen, for example, when we resume a laptop PC after
1123 		 * a long period.
1124 		 */
1125 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1126 		    !IFA6_IS_DEPRECATED(it6)) {
1127 			public_ifa6 = NULL;
1128 			break;
1129 		}
1130 
1131 		/*
1132 		 * This is a public autoconf address that has the same prefix
1133 		 * as ours.  If it is preferred, keep it.  We can't break the
1134 		 * loop here, because there may be a still-preferred temporary
1135 		 * address with the prefix.
1136 		 */
1137 		if (!IFA6_IS_DEPRECATED(it6))
1138 			public_ifa6 = it6;
1139 	}
1140 	if (public_ifa6 != NULL)
1141 		ifa_ref(&public_ifa6->ia_ifa);
1142 
1143 	if (public_ifa6 != NULL) {
1144 		int e;
1145 
1146 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1147 			ifa_free(&public_ifa6->ia_ifa);
1148 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1149 			    " tmp addr,errno=%d\n", e);
1150 			return (-1);
1151 		}
1152 		ifa_free(&public_ifa6->ia_ifa);
1153 		return (0);
1154 	}
1155 
1156 	return (-1);
1157 }
1158 
1159 /*
1160  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1161  * cache entries are freed in in6_domifdetach().
1162  */
1163 void
nd6_purge(struct ifnet * ifp)1164 nd6_purge(struct ifnet *ifp)
1165 {
1166 	struct nd_prhead prl;
1167 	struct nd_prefix *pr, *npr;
1168 
1169 	LIST_INIT(&prl);
1170 
1171 	/* Purge default router list entries toward ifp. */
1172 	nd6_defrouter_purge(ifp);
1173 
1174 	ND6_WLOCK();
1175 	/*
1176 	 * Remove prefixes on ifp. We should have already removed addresses on
1177 	 * this interface, so no addresses should be referencing these prefixes.
1178 	 */
1179 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1180 		if (pr->ndpr_ifp == ifp)
1181 			nd6_prefix_unlink(pr, &prl);
1182 	}
1183 	ND6_WUNLOCK();
1184 
1185 	/* Delete the unlinked prefix objects. */
1186 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1187 		LIST_REMOVE(pr, ndpr_entry);
1188 		nd6_prefix_del(pr);
1189 	}
1190 
1191 	/* cancel default outgoing interface setting */
1192 	if (V_nd6_defifindex == ifp->if_index)
1193 		nd6_setdefaultiface(0);
1194 
1195 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1196 		/* Refresh default router list. */
1197 		defrouter_select_fib(ifp->if_fib);
1198 	}
1199 }
1200 
1201 /*
1202  * the caller acquires and releases the lock on the lltbls
1203  * Returns the llentry locked
1204  */
1205 struct llentry *
nd6_lookup(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1206 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1207 {
1208 	struct sockaddr_in6 sin6;
1209 	struct llentry *ln;
1210 
1211 	bzero(&sin6, sizeof(sin6));
1212 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1213 	sin6.sin6_family = AF_INET6;
1214 	sin6.sin6_addr = *addr6;
1215 
1216 	IF_AFDATA_LOCK_ASSERT(ifp);
1217 
1218 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1219 
1220 	return (ln);
1221 }
1222 
1223 static struct llentry *
nd6_alloc(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1224 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1225 {
1226 	struct sockaddr_in6 sin6;
1227 	struct llentry *ln;
1228 
1229 	bzero(&sin6, sizeof(sin6));
1230 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1231 	sin6.sin6_family = AF_INET6;
1232 	sin6.sin6_addr = *addr6;
1233 
1234 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1235 	if (ln != NULL)
1236 		ln->ln_state = ND6_LLINFO_NOSTATE;
1237 
1238 	return (ln);
1239 }
1240 
1241 /*
1242  * Test whether a given IPv6 address can be a neighbor.
1243  */
1244 static bool
nd6_is_new_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1245 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1246 {
1247 
1248 	/*
1249 	 * A link-local address is always a neighbor.
1250 	 * XXX: a link does not necessarily specify a single interface.
1251 	 */
1252 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1253 		struct sockaddr_in6 sin6_copy;
1254 		u_int32_t zone;
1255 
1256 		/*
1257 		 * We need sin6_copy since sa6_recoverscope() may modify the
1258 		 * content (XXX).
1259 		 */
1260 		sin6_copy = *addr;
1261 		if (sa6_recoverscope(&sin6_copy))
1262 			return (0); /* XXX: should be impossible */
1263 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1264 			return (0);
1265 		if (sin6_copy.sin6_scope_id == zone)
1266 			return (1);
1267 		else
1268 			return (0);
1269 	}
1270 	/* Checking global unicast */
1271 
1272 	/* If an address is directly reachable, it is a neigbor */
1273 	struct nhop_object *nh;
1274 	nh = fib6_lookup(ifp->if_fib, &addr->sin6_addr, 0, NHR_NONE, 0);
1275 	if (nh != NULL && nh->nh_aifp == ifp && (nh->nh_flags & NHF_GATEWAY) == 0)
1276 		return (true);
1277 
1278 	/*
1279 	 * Check prefixes with desired on-link state, as some may be not
1280 	 * installed in the routing table.
1281 	 */
1282 	bool matched = false;
1283 	struct nd_prefix *pr;
1284 	ND6_RLOCK();
1285 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1286 		if (pr->ndpr_ifp != ifp)
1287 			continue;
1288 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0)
1289 			continue;
1290 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1291 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1292 			matched = true;
1293 			break;
1294 		}
1295 	}
1296 	ND6_RUNLOCK();
1297 	if (matched)
1298 		return (true);
1299 
1300 	/*
1301 	 * If the address is assigned on the node of the other side of
1302 	 * a p2p interface, the address should be a neighbor.
1303 	 */
1304 	if (ifp->if_flags & IFF_POINTOPOINT) {
1305 		struct ifaddr *ifa;
1306 
1307 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1308 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1309 				continue;
1310 			if (ifa->ifa_dstaddr != NULL &&
1311 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1312 				return (true);
1313 			}
1314 		}
1315 	}
1316 
1317 	/*
1318 	 * If the default router list is empty, all addresses are regarded
1319 	 * as on-link, and thus, as a neighbor.
1320 	 */
1321 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1322 	    nd6_defrouter_list_empty() &&
1323 	    V_nd6_defifindex == ifp->if_index) {
1324 		return (1);
1325 	}
1326 
1327 	return (0);
1328 }
1329 
1330 /*
1331  * Detect if a given IPv6 address identifies a neighbor on a given link.
1332  * XXX: should take care of the destination of a p2p link?
1333  */
1334 int
nd6_is_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1335 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1336 {
1337 	struct llentry *lle;
1338 	int rc = 0;
1339 
1340 	NET_EPOCH_ASSERT();
1341 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1342 	if (nd6_is_new_addr_neighbor(addr, ifp))
1343 		return (1);
1344 
1345 	/*
1346 	 * Even if the address matches none of our addresses, it might be
1347 	 * in the neighbor cache.
1348 	 */
1349 	if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1350 		LLE_RUNLOCK(lle);
1351 		rc = 1;
1352 	}
1353 	return (rc);
1354 }
1355 
1356 static __noinline void
nd6_free_children(struct llentry * lle)1357 nd6_free_children(struct llentry *lle)
1358 {
1359 	struct llentry *child_lle;
1360 
1361 	NET_EPOCH_ASSERT();
1362 	LLE_WLOCK_ASSERT(lle);
1363 
1364 	while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1365 		LLE_WLOCK(child_lle);
1366 		lltable_unlink_child_entry(child_lle);
1367 		llentry_free(child_lle);
1368 	}
1369 }
1370 
1371 /*
1372  * Tries to update @lle address/prepend data with new @lladdr.
1373  *
1374  * Returns true on success.
1375  * In any case, @lle is returned wlocked.
1376  */
1377 static __noinline bool
nd6_try_set_entry_addr_locked(struct ifnet * ifp,struct llentry * lle,char * lladdr)1378 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1379 {
1380 	u_char buf[LLE_MAX_LINKHDR];
1381 	int fam, off;
1382 	size_t sz;
1383 
1384 	sz = sizeof(buf);
1385 	if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1386 		return (false);
1387 
1388 	/* Update data */
1389 	lltable_set_entry_addr(ifp, lle, buf, sz, off);
1390 
1391 	struct llentry *child_lle;
1392 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1393 		LLE_WLOCK(child_lle);
1394 		fam = child_lle->r_family;
1395 		sz = sizeof(buf);
1396 		if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1397 			/* success */
1398 			lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1399 			child_lle->ln_state = ND6_LLINFO_REACHABLE;
1400 		}
1401 		LLE_WUNLOCK(child_lle);
1402 	}
1403 
1404 	return (true);
1405 }
1406 
1407 bool
nd6_try_set_entry_addr(struct ifnet * ifp,struct llentry * lle,char * lladdr)1408 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1409 {
1410 	NET_EPOCH_ASSERT();
1411 	LLE_WLOCK_ASSERT(lle);
1412 
1413 	if (!lltable_acquire_wlock(ifp, lle))
1414 		return (false);
1415 	bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1416 	IF_AFDATA_WUNLOCK(ifp);
1417 
1418 	return (ret);
1419 }
1420 
1421 /*
1422  * Free an nd6 llinfo entry.
1423  * Since the function would cause significant changes in the kernel, DO NOT
1424  * make it global, unless you have a strong reason for the change, and are sure
1425  * that the change is safe.
1426  *
1427  * Set noinline to be dtrace-friendly
1428  */
1429 static __noinline void
nd6_free(struct llentry ** lnp,int gc)1430 nd6_free(struct llentry **lnp, int gc)
1431 {
1432 	struct ifnet *ifp;
1433 	struct llentry *ln;
1434 	struct nd_defrouter *dr;
1435 
1436 	ln = *lnp;
1437 	*lnp = NULL;
1438 
1439 	LLE_WLOCK_ASSERT(ln);
1440 	ND6_RLOCK_ASSERT();
1441 
1442 	KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1443 
1444 	ifp = lltable_get_ifp(ln->lle_tbl);
1445 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1446 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1447 	else
1448 		dr = NULL;
1449 	ND6_RUNLOCK();
1450 
1451 	if ((ln->la_flags & LLE_DELETED) == 0)
1452 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1453 
1454 	/*
1455 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1456 	 * even though it is not harmful, it was not really necessary.
1457 	 */
1458 
1459 	/* cancel timer */
1460 	nd6_llinfo_settimer_locked(ln, -1);
1461 
1462 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1463 		if (dr != NULL && dr->expire &&
1464 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1465 			/*
1466 			 * If the reason for the deletion is just garbage
1467 			 * collection, and the neighbor is an active default
1468 			 * router, do not delete it.  Instead, reset the GC
1469 			 * timer using the router's lifetime.
1470 			 * Simply deleting the entry would affect default
1471 			 * router selection, which is not necessarily a good
1472 			 * thing, especially when we're using router preference
1473 			 * values.
1474 			 * XXX: the check for ln_state would be redundant,
1475 			 *      but we intentionally keep it just in case.
1476 			 */
1477 			if (dr->expire > time_uptime)
1478 				nd6_llinfo_settimer_locked(ln,
1479 				    (dr->expire - time_uptime) * hz);
1480 			else
1481 				nd6_llinfo_settimer_locked(ln,
1482 				    (long)V_nd6_gctimer * hz);
1483 
1484 			LLE_REMREF(ln);
1485 			LLE_WUNLOCK(ln);
1486 			defrouter_rele(dr);
1487 			return;
1488 		}
1489 
1490 		if (dr) {
1491 			/*
1492 			 * Unreachability of a router might affect the default
1493 			 * router selection and on-link detection of advertised
1494 			 * prefixes.
1495 			 */
1496 
1497 			/*
1498 			 * Temporarily fake the state to choose a new default
1499 			 * router and to perform on-link determination of
1500 			 * prefixes correctly.
1501 			 * Below the state will be set correctly,
1502 			 * or the entry itself will be deleted.
1503 			 */
1504 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1505 		}
1506 
1507 		if (ln->ln_router || dr) {
1508 			/*
1509 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1510 			 * rnh and for the calls to pfxlist_onlink_check() and
1511 			 * defrouter_select_fib() in the block further down for calls
1512 			 * into nd6_lookup().  We still hold a ref.
1513 			 */
1514 			LLE_WUNLOCK(ln);
1515 
1516 			/*
1517 			 * rt6_flush must be called whether or not the neighbor
1518 			 * is in the Default Router List.
1519 			 * See a corresponding comment in nd6_na_input().
1520 			 */
1521 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1522 		}
1523 
1524 		if (dr) {
1525 			/*
1526 			 * Since defrouter_select_fib() does not affect the
1527 			 * on-link determination and MIP6 needs the check
1528 			 * before the default router selection, we perform
1529 			 * the check now.
1530 			 */
1531 			pfxlist_onlink_check();
1532 
1533 			/*
1534 			 * Refresh default router list.
1535 			 */
1536 			defrouter_select_fib(dr->ifp->if_fib);
1537 		}
1538 
1539 		/*
1540 		 * If this entry was added by an on-link redirect, remove the
1541 		 * corresponding host route.
1542 		 */
1543 		if (ln->la_flags & LLE_REDIRECT)
1544 			nd6_free_redirect(ln);
1545 
1546 		if (ln->ln_router || dr)
1547 			LLE_WLOCK(ln);
1548 	}
1549 
1550 	/*
1551 	 * Save to unlock. We still hold an extra reference and will not
1552 	 * free(9) in llentry_free() if someone else holds one as well.
1553 	 */
1554 	LLE_WUNLOCK(ln);
1555 	IF_AFDATA_LOCK(ifp);
1556 	LLE_WLOCK(ln);
1557 	/* Guard against race with other llentry_free(). */
1558 	if (ln->la_flags & LLE_LINKED) {
1559 		/* Remove callout reference */
1560 		LLE_REMREF(ln);
1561 		lltable_unlink_entry(ln->lle_tbl, ln);
1562 	}
1563 	IF_AFDATA_UNLOCK(ifp);
1564 
1565 	nd6_free_children(ln);
1566 
1567 	llentry_free(ln);
1568 	if (dr != NULL)
1569 		defrouter_rele(dr);
1570 }
1571 
1572 static int
nd6_isdynrte(const struct rtentry * rt,const struct nhop_object * nh,void * xap)1573 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1574 {
1575 
1576 	if (nh->nh_flags & NHF_REDIRECT)
1577 		return (1);
1578 
1579 	return (0);
1580 }
1581 
1582 /*
1583  * Remove the rtentry for the given llentry,
1584  * both of which were installed by a redirect.
1585  */
1586 static void
nd6_free_redirect(const struct llentry * ln)1587 nd6_free_redirect(const struct llentry *ln)
1588 {
1589 	int fibnum;
1590 	struct sockaddr_in6 sin6;
1591 	struct rib_cmd_info rc;
1592 	struct epoch_tracker et;
1593 
1594 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1595 
1596 	NET_EPOCH_ENTER(et);
1597 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1598 		rib_del_route_px(fibnum, (struct sockaddr *)&sin6, 128,
1599 		    nd6_isdynrte, NULL, 0, &rc);
1600 	NET_EPOCH_EXIT(et);
1601 }
1602 
1603 /*
1604  * Updates status of the default router route.
1605  */
1606 static void
check_release_defrouter(const struct rib_cmd_info * rc,void * _cbdata)1607 check_release_defrouter(const struct rib_cmd_info *rc, void *_cbdata)
1608 {
1609 	struct nd_defrouter *dr;
1610 	struct nhop_object *nh;
1611 
1612 	nh = rc->rc_nh_old;
1613 	if (rc->rc_cmd == RTM_DELETE && (nh->nh_flags & NHF_DEFAULT) != 0) {
1614 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1615 		if (dr != NULL) {
1616 			dr->installed = 0;
1617 			defrouter_rele(dr);
1618 		}
1619 	}
1620 }
1621 
1622 void
nd6_subscription_cb(struct rib_head * rnh,struct rib_cmd_info * rc,void * arg)1623 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1624 {
1625 #ifdef ROUTE_MPATH
1626 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1627 	if (rc->rc_cmd == RTM_DELETE && !NH_IS_NHGRP(rc->rc_nh_old))
1628 		check_release_defrouter(rc, NULL);
1629 #else
1630 	check_release_defrouter(rc, NULL);
1631 #endif
1632 }
1633 
1634 int
nd6_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp)1635 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1636 {
1637 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1638 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1639 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1640 	struct epoch_tracker et;
1641 	int error = 0;
1642 
1643 	if (ifp->if_afdata[AF_INET6] == NULL)
1644 		return (EPFNOSUPPORT);
1645 	switch (cmd) {
1646 	case OSIOCGIFINFO_IN6:
1647 #define ND	ndi->ndi
1648 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1649 		bzero(&ND, sizeof(ND));
1650 		ND.linkmtu = IN6_LINKMTU(ifp);
1651 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1652 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1653 		ND.reachable = ND_IFINFO(ifp)->reachable;
1654 		ND.retrans = ND_IFINFO(ifp)->retrans;
1655 		ND.flags = ND_IFINFO(ifp)->flags;
1656 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1657 		ND.chlim = ND_IFINFO(ifp)->chlim;
1658 		break;
1659 	case SIOCGIFINFO_IN6:
1660 		ND = *ND_IFINFO(ifp);
1661 		break;
1662 	case SIOCSIFINFO_IN6:
1663 		/*
1664 		 * used to change host variables from userland.
1665 		 * intended for a use on router to reflect RA configurations.
1666 		 */
1667 		/* 0 means 'unspecified' */
1668 		if (ND.linkmtu != 0) {
1669 			if (ND.linkmtu < IPV6_MMTU ||
1670 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1671 				error = EINVAL;
1672 				break;
1673 			}
1674 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1675 		}
1676 
1677 		if (ND.basereachable != 0) {
1678 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1679 
1680 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1681 			if (ND.basereachable != obasereachable)
1682 				ND_IFINFO(ifp)->reachable =
1683 				    ND_COMPUTE_RTIME(ND.basereachable);
1684 		}
1685 		if (ND.retrans != 0)
1686 			ND_IFINFO(ifp)->retrans = ND.retrans;
1687 		if (ND.chlim != 0)
1688 			ND_IFINFO(ifp)->chlim = ND.chlim;
1689 		/* FALLTHROUGH */
1690 	case SIOCSIFINFO_FLAGS:
1691 	{
1692 		struct ifaddr *ifa;
1693 		struct in6_ifaddr *ia;
1694 
1695 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1696 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1697 			/* ifdisabled 1->0 transision */
1698 
1699 			/*
1700 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1701 			 * has an link-local address with IN6_IFF_DUPLICATED,
1702 			 * do not clear ND6_IFF_IFDISABLED.
1703 			 * See RFC 4862, Section 5.4.5.
1704 			 */
1705 			NET_EPOCH_ENTER(et);
1706 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1707 				if (ifa->ifa_addr->sa_family != AF_INET6)
1708 					continue;
1709 				ia = (struct in6_ifaddr *)ifa;
1710 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1711 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1712 					break;
1713 			}
1714 			NET_EPOCH_EXIT(et);
1715 
1716 			if (ifa != NULL) {
1717 				/* LLA is duplicated. */
1718 				ND.flags |= ND6_IFF_IFDISABLED;
1719 				log(LOG_ERR, "Cannot enable an interface"
1720 				    " with a link-local address marked"
1721 				    " duplicate.\n");
1722 			} else {
1723 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1724 				if (ifp->if_flags & IFF_UP)
1725 					in6_if_up(ifp);
1726 			}
1727 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1728 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1729 			/* ifdisabled 0->1 transision */
1730 			/* Mark all IPv6 address as tentative. */
1731 
1732 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1733 			if (V_ip6_dad_count > 0 &&
1734 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1735 				NET_EPOCH_ENTER(et);
1736 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1737 				    ifa_link) {
1738 					if (ifa->ifa_addr->sa_family !=
1739 					    AF_INET6)
1740 						continue;
1741 					ia = (struct in6_ifaddr *)ifa;
1742 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1743 				}
1744 				NET_EPOCH_EXIT(et);
1745 			}
1746 		}
1747 
1748 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1749 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1750 				/* auto_linklocal 0->1 transision */
1751 
1752 				/* If no link-local address on ifp, configure */
1753 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1754 				in6_ifattach(ifp, NULL);
1755 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1756 			    ifp->if_flags & IFF_UP) {
1757 				/*
1758 				 * When the IF already has
1759 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1760 				 * address is assigned, and IFF_UP, try to
1761 				 * assign one.
1762 				 */
1763 				NET_EPOCH_ENTER(et);
1764 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1765 				    ifa_link) {
1766 					if (ifa->ifa_addr->sa_family !=
1767 					    AF_INET6)
1768 						continue;
1769 					ia = (struct in6_ifaddr *)ifa;
1770 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1771 						break;
1772 				}
1773 				NET_EPOCH_EXIT(et);
1774 				if (ifa != NULL)
1775 					/* No LLA is configured. */
1776 					in6_ifattach(ifp, NULL);
1777 			}
1778 		}
1779 		ND_IFINFO(ifp)->flags = ND.flags;
1780 		break;
1781 	}
1782 #undef ND
1783 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1784 		/* sync kernel routing table with the default router list */
1785 		defrouter_reset();
1786 		defrouter_select_fib(RT_ALL_FIBS);
1787 		break;
1788 	case SIOCSPFXFLUSH_IN6:
1789 	{
1790 		/* flush all the prefix advertised by routers */
1791 		struct in6_ifaddr *ia, *ia_next;
1792 		struct nd_prefix *pr, *next;
1793 		struct nd_prhead prl;
1794 
1795 		LIST_INIT(&prl);
1796 
1797 		ND6_WLOCK();
1798 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1799 			if (ND6_PREFIX_WITH_ROUTER(pr))
1800 				nd6_prefix_unlink(pr, &prl);
1801 		}
1802 		ND6_WUNLOCK();
1803 
1804 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1805 			LIST_REMOVE(pr, ndpr_entry);
1806 			/* XXXRW: in6_ifaddrhead locking. */
1807 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1808 			    ia_next) {
1809 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1810 					continue;
1811 
1812 				if (ia->ia6_ndpr == pr)
1813 					in6_purgeaddr(&ia->ia_ifa);
1814 			}
1815 			nd6_prefix_del(pr);
1816 		}
1817 		break;
1818 	}
1819 	case SIOCSRTRFLUSH_IN6:
1820 	{
1821 		/* flush all the default routers */
1822 
1823 		defrouter_reset();
1824 		nd6_defrouter_flush_all();
1825 		defrouter_select_fib(RT_ALL_FIBS);
1826 		break;
1827 	}
1828 	case SIOCGNBRINFO_IN6:
1829 	{
1830 		struct llentry *ln;
1831 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1832 
1833 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1834 			return (error);
1835 
1836 		NET_EPOCH_ENTER(et);
1837 		ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1838 		NET_EPOCH_EXIT(et);
1839 
1840 		if (ln == NULL) {
1841 			error = EINVAL;
1842 			break;
1843 		}
1844 		nbi->state = ln->ln_state;
1845 		nbi->asked = ln->la_asked;
1846 		nbi->isrouter = ln->ln_router;
1847 		if (ln->la_expire == 0)
1848 			nbi->expire = 0;
1849 		else
1850 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1851 			    (time_second - time_uptime);
1852 		LLE_RUNLOCK(ln);
1853 		break;
1854 	}
1855 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1856 		ndif->ifindex = V_nd6_defifindex;
1857 		break;
1858 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1859 		return (nd6_setdefaultiface(ndif->ifindex));
1860 	}
1861 	return (error);
1862 }
1863 
1864 /*
1865  * Calculates new isRouter value based on provided parameters and
1866  * returns it.
1867  */
1868 static int
nd6_is_router(int type,int code,int is_new,int old_addr,int new_addr,int ln_router)1869 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1870     int ln_router)
1871 {
1872 
1873 	/*
1874 	 * ICMP6 type dependent behavior.
1875 	 *
1876 	 * NS: clear IsRouter if new entry
1877 	 * RS: clear IsRouter
1878 	 * RA: set IsRouter if there's lladdr
1879 	 * redir: clear IsRouter if new entry
1880 	 *
1881 	 * RA case, (1):
1882 	 * The spec says that we must set IsRouter in the following cases:
1883 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1884 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1885 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1886 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1887 	 * neighbor cache, this is similar to (6).
1888 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1889 	 *
1890 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1891 	 *							D R
1892 	 *	0	n	n	(1)	c   ?     s
1893 	 *	0	y	n	(2)	c   s     s
1894 	 *	0	n	y	(3)	c   s     s
1895 	 *	0	y	y	(4)	c   s     s
1896 	 *	0	y	y	(5)	c   s     s
1897 	 *	1	--	n	(6) c	c	c s
1898 	 *	1	--	y	(7) c	c   s	c s
1899 	 *
1900 	 *					(c=clear s=set)
1901 	 */
1902 	switch (type & 0xff) {
1903 	case ND_NEIGHBOR_SOLICIT:
1904 		/*
1905 		 * New entry must have is_router flag cleared.
1906 		 */
1907 		if (is_new)					/* (6-7) */
1908 			ln_router = 0;
1909 		break;
1910 	case ND_REDIRECT:
1911 		/*
1912 		 * If the icmp is a redirect to a better router, always set the
1913 		 * is_router flag.  Otherwise, if the entry is newly created,
1914 		 * clear the flag.  [RFC 2461, sec 8.3]
1915 		 */
1916 		if (code == ND_REDIRECT_ROUTER)
1917 			ln_router = 1;
1918 		else {
1919 			if (is_new)				/* (6-7) */
1920 				ln_router = 0;
1921 		}
1922 		break;
1923 	case ND_ROUTER_SOLICIT:
1924 		/*
1925 		 * is_router flag must always be cleared.
1926 		 */
1927 		ln_router = 0;
1928 		break;
1929 	case ND_ROUTER_ADVERT:
1930 		/*
1931 		 * Mark an entry with lladdr as a router.
1932 		 */
1933 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1934 		    (is_new && new_addr)) {			/* (7) */
1935 			ln_router = 1;
1936 		}
1937 		break;
1938 	}
1939 
1940 	return (ln_router);
1941 }
1942 
1943 /*
1944  * Create neighbor cache entry and cache link-layer address,
1945  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1946  *
1947  * type - ICMP6 type
1948  * code - type dependent information
1949  *
1950  */
1951 void
nd6_cache_lladdr(struct ifnet * ifp,struct in6_addr * from,char * lladdr,int lladdrlen,int type,int code)1952 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1953     int lladdrlen, int type, int code)
1954 {
1955 	struct llentry *ln = NULL, *ln_tmp;
1956 	int is_newentry;
1957 	int do_update;
1958 	int olladdr;
1959 	int llchange;
1960 	int flags;
1961 	uint16_t router = 0;
1962 	struct mbuf *chain = NULL;
1963 	u_char linkhdr[LLE_MAX_LINKHDR];
1964 	size_t linkhdrsize;
1965 	int lladdr_off;
1966 
1967 	NET_EPOCH_ASSERT();
1968 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1969 
1970 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1971 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1972 
1973 	/* nothing must be updated for unspecified address */
1974 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1975 		return;
1976 
1977 	/*
1978 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1979 	 * the caller.
1980 	 *
1981 	 * XXX If the link does not have link-layer adderss, what should
1982 	 * we do? (ifp->if_addrlen == 0)
1983 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1984 	 * description on it in NS section (RFC 2461 7.2.3).
1985 	 */
1986 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1987 	ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
1988 	is_newentry = 0;
1989 	if (ln == NULL) {
1990 		flags |= LLE_EXCLUSIVE;
1991 		ln = nd6_alloc(from, 0, ifp);
1992 		if (ln == NULL)
1993 			return;
1994 
1995 		/*
1996 		 * Since we already know all the data for the new entry,
1997 		 * fill it before insertion.
1998 		 */
1999 		if (lladdr != NULL) {
2000 			linkhdrsize = sizeof(linkhdr);
2001 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2002 			    linkhdr, &linkhdrsize, &lladdr_off) != 0) {
2003 				lltable_free_entry(LLTABLE6(ifp), ln);
2004 				return;
2005 			}
2006 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2007 			    lladdr_off);
2008 		}
2009 
2010 		IF_AFDATA_WLOCK(ifp);
2011 		LLE_WLOCK(ln);
2012 		/* Prefer any existing lle over newly-created one */
2013 		ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2014 		if (ln_tmp == NULL)
2015 			lltable_link_entry(LLTABLE6(ifp), ln);
2016 		IF_AFDATA_WUNLOCK(ifp);
2017 		if (ln_tmp == NULL) {
2018 			/* No existing lle, mark as new entry (6,7) */
2019 			is_newentry = 1;
2020 			if (lladdr != NULL) {	/* (7) */
2021 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2022 				EVENTHANDLER_INVOKE(lle_event, ln,
2023 				    LLENTRY_RESOLVED);
2024 			}
2025 		} else {
2026 			lltable_free_entry(LLTABLE6(ifp), ln);
2027 			ln = ln_tmp;
2028 			ln_tmp = NULL;
2029 		}
2030 	}
2031 	/* do nothing if static ndp is set */
2032 	if ((ln->la_flags & LLE_STATIC)) {
2033 		if (flags & LLE_EXCLUSIVE)
2034 			LLE_WUNLOCK(ln);
2035 		else
2036 			LLE_RUNLOCK(ln);
2037 		return;
2038 	}
2039 
2040 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2041 	if (olladdr && lladdr) {
2042 		llchange = bcmp(lladdr, ln->ll_addr,
2043 		    ifp->if_addrlen);
2044 	} else if (!olladdr && lladdr)
2045 		llchange = 1;
2046 	else
2047 		llchange = 0;
2048 
2049 	/*
2050 	 * newentry olladdr  lladdr  llchange	(*=record)
2051 	 *	0	n	n	--	(1)
2052 	 *	0	y	n	--	(2)
2053 	 *	0	n	y	y	(3) * STALE
2054 	 *	0	y	y	n	(4) *
2055 	 *	0	y	y	y	(5) * STALE
2056 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2057 	 *	1	--	y	--	(7) * STALE
2058 	 */
2059 
2060 	do_update = 0;
2061 	if (is_newentry == 0 && llchange != 0) {
2062 		do_update = 1;	/* (3,5) */
2063 
2064 		/*
2065 		 * Record source link-layer address
2066 		 * XXX is it dependent to ifp->if_type?
2067 		 */
2068 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2069 			/* Entry was deleted */
2070 			LLE_WUNLOCK(ln);
2071 			return;
2072 		}
2073 
2074 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2075 
2076 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2077 
2078 		if (ln->la_hold != NULL)
2079 			chain = nd6_grab_holdchain(ln);
2080 	}
2081 
2082 	/* Calculates new router status */
2083 	router = nd6_is_router(type, code, is_newentry, olladdr,
2084 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2085 
2086 	ln->ln_router = router;
2087 	/* Mark non-router redirects with special flag */
2088 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2089 		ln->la_flags |= LLE_REDIRECT;
2090 
2091 	if (flags & LLE_EXCLUSIVE)
2092 		LLE_WUNLOCK(ln);
2093 	else
2094 		LLE_RUNLOCK(ln);
2095 
2096 	if (chain != NULL)
2097 		nd6_flush_holdchain(ifp, ln, chain);
2098 	if (do_update)
2099 		nd6_flush_children_holdchain(ifp, ln);
2100 
2101 	/*
2102 	 * When the link-layer address of a router changes, select the
2103 	 * best router again.  In particular, when the neighbor entry is newly
2104 	 * created, it might affect the selection policy.
2105 	 * Question: can we restrict the first condition to the "is_newentry"
2106 	 * case?
2107 	 * XXX: when we hear an RA from a new router with the link-layer
2108 	 * address option, defrouter_select_fib() is called twice, since
2109 	 * defrtrlist_update called the function as well.  However, I believe
2110 	 * we can compromise the overhead, since it only happens the first
2111 	 * time.
2112 	 * XXX: although defrouter_select_fib() should not have a bad effect
2113 	 * for those are not autoconfigured hosts, we explicitly avoid such
2114 	 * cases for safety.
2115 	 */
2116 	if ((do_update || is_newentry) && router &&
2117 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2118 		/*
2119 		 * guaranteed recursion
2120 		 */
2121 		defrouter_select_fib(ifp->if_fib);
2122 	}
2123 }
2124 
2125 static void
nd6_slowtimo(void * arg)2126 nd6_slowtimo(void *arg)
2127 {
2128 	struct epoch_tracker et;
2129 	CURVNET_SET((struct vnet *) arg);
2130 	struct nd_ifinfo *nd6if;
2131 	struct ifnet *ifp;
2132 
2133 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2134 	    nd6_slowtimo, curvnet);
2135 	NET_EPOCH_ENTER(et);
2136 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2137 		if (ifp->if_afdata[AF_INET6] == NULL)
2138 			continue;
2139 		nd6if = ND_IFINFO(ifp);
2140 		if (nd6if->basereachable && /* already initialized */
2141 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2142 			/*
2143 			 * Since reachable time rarely changes by router
2144 			 * advertisements, we SHOULD insure that a new random
2145 			 * value gets recomputed at least once every few hours.
2146 			 * (RFC 2461, 6.3.4)
2147 			 */
2148 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2149 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2150 		}
2151 	}
2152 	NET_EPOCH_EXIT(et);
2153 	CURVNET_RESTORE();
2154 }
2155 
2156 struct mbuf *
nd6_grab_holdchain(struct llentry * ln)2157 nd6_grab_holdchain(struct llentry *ln)
2158 {
2159 	struct mbuf *chain;
2160 
2161 	LLE_WLOCK_ASSERT(ln);
2162 
2163 	chain = ln->la_hold;
2164 	ln->la_hold = NULL;
2165 	ln->la_numheld = 0;
2166 
2167 	if (ln->ln_state == ND6_LLINFO_STALE) {
2168 		/*
2169 		 * The first time we send a packet to a
2170 		 * neighbor whose entry is STALE, we have
2171 		 * to change the state to DELAY and a sets
2172 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2173 		 * seconds to ensure do neighbor unreachability
2174 		 * detection on expiration.
2175 		 * (RFC 2461 7.3.3)
2176 		 */
2177 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2178 	}
2179 
2180 	return (chain);
2181 }
2182 
2183 int
nd6_output_ifp(struct ifnet * ifp,struct ifnet * origifp,struct mbuf * m,struct sockaddr_in6 * dst,struct route * ro)2184 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2185     struct sockaddr_in6 *dst, struct route *ro)
2186 {
2187 	int error;
2188 	int ip6len;
2189 	struct ip6_hdr *ip6;
2190 	struct m_tag *mtag;
2191 
2192 #ifdef MAC
2193 	mac_netinet6_nd6_send(ifp, m);
2194 #endif
2195 
2196 	/*
2197 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2198 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2199 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2200 	 * to be diverted to user space.  When re-injected into the kernel,
2201 	 * send_output() will directly dispatch them to the outgoing interface.
2202 	 */
2203 	if (send_sendso_input_hook != NULL) {
2204 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2205 		if (mtag != NULL) {
2206 			ip6 = mtod(m, struct ip6_hdr *);
2207 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2208 			/* Use the SEND socket */
2209 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2210 			    ip6len);
2211 			/* -1 == no app on SEND socket */
2212 			if (error == 0 || error != -1)
2213 			    return (error);
2214 		}
2215 	}
2216 
2217 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2218 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2219 	    mtod(m, struct ip6_hdr *));
2220 
2221 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2222 		origifp = ifp;
2223 
2224 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2225 	return (error);
2226 }
2227 
2228 /*
2229  * Lookup link headerfor @sa_dst address. Stores found
2230  * data in @desten buffer. Copy of lle ln_flags can be also
2231  * saved in @pflags if @pflags is non-NULL.
2232  *
2233  * If destination LLE does not exists or lle state modification
2234  * is required, call "slow" version.
2235  *
2236  * Return values:
2237  * - 0 on success (address copied to buffer).
2238  * - EWOULDBLOCK (no local error, but address is still unresolved)
2239  * - other errors (alloc failure, etc)
2240  */
2241 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)2242 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2243     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2244     struct llentry **plle)
2245 {
2246 	struct llentry *ln = NULL;
2247 	const struct sockaddr_in6 *dst6;
2248 
2249 	NET_EPOCH_ASSERT();
2250 
2251 	if (pflags != NULL)
2252 		*pflags = 0;
2253 
2254 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2255 
2256 	/* discard the packet if IPv6 operation is disabled on the interface */
2257 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2258 		m_freem(m);
2259 		return (ENETDOWN); /* better error? */
2260 	}
2261 
2262 	if (m != NULL && m->m_flags & M_MCAST) {
2263 		switch (ifp->if_type) {
2264 		case IFT_ETHER:
2265 		case IFT_L2VLAN:
2266 		case IFT_BRIDGE:
2267 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2268 						 desten);
2269 			return (0);
2270 		default:
2271 			m_freem(m);
2272 			return (EAFNOSUPPORT);
2273 		}
2274 	}
2275 
2276 	int family = gw_flags >> 16;
2277 	int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2278 	ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2279 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2280 		/* Entry found, let's copy lle info */
2281 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2282 		if (pflags != NULL)
2283 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2284 		llentry_provide_feedback(ln);
2285 		if (plle) {
2286 			LLE_ADDREF(ln);
2287 			*plle = ln;
2288 			LLE_WUNLOCK(ln);
2289 		}
2290 		return (0);
2291 	} else if (plle && ln)
2292 		LLE_WUNLOCK(ln);
2293 
2294 	return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2295 }
2296 
2297 /*
2298  * Finds or creates a new llentry for @addr and @family.
2299  * Returns wlocked llentry or NULL.
2300  *
2301  *
2302  * Child LLEs.
2303  *
2304  * Do not have their own state machine (gets marked as static)
2305  *  settimer bails out for child LLEs just in case.
2306  *
2307  * Locking order: parent lle gets locked first, chen goes the child.
2308  */
2309 static __noinline struct llentry *
nd6_get_llentry(struct ifnet * ifp,const struct in6_addr * addr,int family)2310 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2311 {
2312 	struct llentry *child_lle = NULL;
2313 	struct llentry *lle, *lle_tmp;
2314 
2315 	lle = nd6_alloc(addr, 0, ifp);
2316 	if (lle != NULL && family != AF_INET6) {
2317 		child_lle = nd6_alloc(addr, 0, ifp);
2318 		if (child_lle == NULL) {
2319 			lltable_free_entry(LLTABLE6(ifp), lle);
2320 			return (NULL);
2321 		}
2322 		child_lle->r_family = family;
2323 		child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2324 		child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2325 	}
2326 
2327 	if (lle == NULL) {
2328 		char ip6buf[INET6_ADDRSTRLEN];
2329 		log(LOG_DEBUG,
2330 		    "nd6_get_llentry: can't allocate llinfo for %s "
2331 		    "(ln=%p)\n",
2332 		    ip6_sprintf(ip6buf, addr), lle);
2333 		return (NULL);
2334 	}
2335 
2336 	IF_AFDATA_WLOCK(ifp);
2337 	LLE_WLOCK(lle);
2338 	/* Prefer any existing entry over newly-created one */
2339 	lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2340 	if (lle_tmp == NULL)
2341 		lltable_link_entry(LLTABLE6(ifp), lle);
2342 	else {
2343 		lltable_free_entry(LLTABLE6(ifp), lle);
2344 		lle = lle_tmp;
2345 	}
2346 	if (child_lle != NULL) {
2347 		/* Check if child lle for the same family exists */
2348 		lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2349 		LLE_WLOCK(child_lle);
2350 		if (lle_tmp == NULL) {
2351 			/* Attach */
2352 			lltable_link_child_entry(lle, child_lle);
2353 		} else {
2354 			/* child lle already exists, free newly-created one */
2355 			lltable_free_entry(LLTABLE6(ifp), child_lle);
2356 			LLE_WLOCK(lle_tmp);
2357 			child_lle = lle_tmp;
2358 		}
2359 		LLE_WUNLOCK(lle);
2360 		lle = child_lle;
2361 	}
2362 	IF_AFDATA_WUNLOCK(ifp);
2363 	return (lle);
2364 }
2365 
2366 /*
2367  * Do L2 address resolution for @sa_dst address. Stores found
2368  * address in @desten buffer. Copy of lle ln_flags can be also
2369  * saved in @pflags if @pflags is non-NULL.
2370  *
2371  * Heavy version.
2372  * Function assume that destination LLE does not exist,
2373  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2374  *
2375  * Set noinline to be dtrace-friendly
2376  */
2377 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)2378 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2379     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2380     struct llentry **plle)
2381 {
2382 	struct llentry *lle = NULL;
2383 	struct in6_addr *psrc, src;
2384 	int send_ns, ll_len;
2385 	char *lladdr;
2386 
2387 	NET_EPOCH_ASSERT();
2388 
2389 	/*
2390 	 * Address resolution or Neighbor Unreachability Detection
2391 	 * for the next hop.
2392 	 * At this point, the destination of the packet must be a unicast
2393 	 * or an anycast address(i.e. not a multicast).
2394 	 */
2395 	lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2396 	if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2397 		/*
2398 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2399 		 * the condition below is not very efficient.  But we believe
2400 		 * it is tolerable, because this should be a rare case.
2401 		 */
2402 		lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2403 	}
2404 
2405 	if (lle == NULL) {
2406 		m_freem(m);
2407 		return (ENOBUFS);
2408 	}
2409 
2410 	LLE_WLOCK_ASSERT(lle);
2411 
2412 	/*
2413 	 * The first time we send a packet to a neighbor whose entry is
2414 	 * STALE, we have to change the state to DELAY and a sets a timer to
2415 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2416 	 * neighbor unreachability detection on expiration.
2417 	 * (RFC 2461 7.3.3)
2418 	 */
2419 	if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2420 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2421 
2422 	/*
2423 	 * If the neighbor cache entry has a state other than INCOMPLETE
2424 	 * (i.e. its link-layer address is already resolved), just
2425 	 * send the packet.
2426 	 */
2427 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2428 		if (flags & LLE_ADDRONLY) {
2429 			lladdr = lle->ll_addr;
2430 			ll_len = ifp->if_addrlen;
2431 		} else {
2432 			lladdr = lle->r_linkdata;
2433 			ll_len = lle->r_hdrlen;
2434 		}
2435 		bcopy(lladdr, desten, ll_len);
2436 		if (pflags != NULL)
2437 			*pflags = lle->la_flags;
2438 		if (plle) {
2439 			LLE_ADDREF(lle);
2440 			*plle = lle;
2441 		}
2442 		LLE_WUNLOCK(lle);
2443 		return (0);
2444 	}
2445 
2446 	/*
2447 	 * There is a neighbor cache entry, but no ethernet address
2448 	 * response yet.  Append this latest packet to the end of the
2449 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2450 	 * the oldest packet in the queue will be removed.
2451 	 */
2452 	if (m != NULL) {
2453 		size_t dropped;
2454 
2455 		dropped = lltable_append_entry_queue(lle, m, V_nd6_maxqueuelen);
2456 		ICMP6STAT_ADD(icp6s_dropped, dropped);
2457 	}
2458 
2459 	/*
2460 	 * If there has been no NS for the neighbor after entering the
2461 	 * INCOMPLETE state, send the first solicitation.
2462 	 * Note that for newly-created lle la_asked will be 0,
2463 	 * so we will transition from ND6_LLINFO_NOSTATE to
2464 	 * ND6_LLINFO_INCOMPLETE state here.
2465 	 */
2466 	psrc = NULL;
2467 	send_ns = 0;
2468 
2469 	/* If we have child lle, switch to the parent to send NS */
2470 	if (lle->la_flags & LLE_CHILD) {
2471 		struct llentry *lle_parent = lle->lle_parent;
2472 		LLE_WUNLOCK(lle);
2473 		lle = lle_parent;
2474 		LLE_WLOCK(lle);
2475 	}
2476 	if (lle->la_asked == 0) {
2477 		lle->la_asked++;
2478 		send_ns = 1;
2479 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2480 
2481 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2482 	}
2483 	LLE_WUNLOCK(lle);
2484 	if (send_ns != 0)
2485 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2486 
2487 	return (EWOULDBLOCK);
2488 }
2489 
2490 /*
2491  * Do L2 address resolution for @sa_dst address. Stores found
2492  * address in @desten buffer. Copy of lle ln_flags can be also
2493  * saved in @pflags if @pflags is non-NULL.
2494  *
2495  * Return values:
2496  * - 0 on success (address copied to buffer).
2497  * - EWOULDBLOCK (no local error, but address is still unresolved)
2498  * - other errors (alloc failure, etc)
2499  */
2500 int
nd6_resolve_addr(struct ifnet * ifp,int flags,const struct sockaddr * dst,char * desten,uint32_t * pflags)2501 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2502     char *desten, uint32_t *pflags)
2503 {
2504 	int error;
2505 
2506 	flags |= LLE_ADDRONLY;
2507 	error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2508 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2509 	return (error);
2510 }
2511 
2512 int
nd6_flush_holdchain(struct ifnet * ifp,struct llentry * lle,struct mbuf * chain)2513 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2514 {
2515 	struct mbuf *m, *m_head;
2516 	struct sockaddr_in6 dst6;
2517 	int error = 0;
2518 
2519 	NET_EPOCH_ASSERT();
2520 
2521 	struct route_in6 ro = {
2522 		.ro_prepend = lle->r_linkdata,
2523 		.ro_plen = lle->r_hdrlen,
2524 	};
2525 
2526 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2527 	m_head = chain;
2528 
2529 	while (m_head) {
2530 		m = m_head;
2531 		m_head = m_head->m_nextpkt;
2532 		m->m_nextpkt = NULL;
2533 		error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2534 	}
2535 
2536 	/*
2537 	 * XXX
2538 	 * note that intermediate errors are blindly ignored
2539 	 */
2540 	return (error);
2541 }
2542 
2543 __noinline void
nd6_flush_children_holdchain(struct ifnet * ifp,struct llentry * lle)2544 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2545 {
2546 	struct llentry *child_lle;
2547 	struct mbuf *chain;
2548 
2549 	NET_EPOCH_ASSERT();
2550 
2551 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2552 		LLE_WLOCK(child_lle);
2553 		chain = nd6_grab_holdchain(child_lle);
2554 		LLE_WUNLOCK(child_lle);
2555 		nd6_flush_holdchain(ifp, child_lle, chain);
2556 	}
2557 }
2558 
2559 static int
nd6_need_cache(struct ifnet * ifp)2560 nd6_need_cache(struct ifnet *ifp)
2561 {
2562 	/*
2563 	 * XXX: we currently do not make neighbor cache on any interface
2564 	 * other than Ethernet and GIF.
2565 	 *
2566 	 * RFC2893 says:
2567 	 * - unidirectional tunnels needs no ND
2568 	 */
2569 	switch (ifp->if_type) {
2570 	case IFT_ETHER:
2571 	case IFT_IEEE1394:
2572 	case IFT_L2VLAN:
2573 	case IFT_INFINIBAND:
2574 	case IFT_BRIDGE:
2575 	case IFT_PROPVIRTUAL:
2576 		return (1);
2577 	default:
2578 		return (0);
2579 	}
2580 }
2581 
2582 /*
2583  * Add pernament ND6 link-layer record for given
2584  * interface address.
2585  *
2586  * Very similar to IPv4 arp_ifinit(), but:
2587  * 1) IPv6 DAD is performed in different place
2588  * 2) It is called by IPv6 protocol stack in contrast to
2589  * arp_ifinit() which is typically called in SIOCSIFADDR
2590  * driver ioctl handler.
2591  *
2592  */
2593 int
nd6_add_ifa_lle(struct in6_ifaddr * ia)2594 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2595 {
2596 	struct ifnet *ifp;
2597 	struct llentry *ln, *ln_tmp;
2598 	struct sockaddr *dst;
2599 
2600 	ifp = ia->ia_ifa.ifa_ifp;
2601 	if (nd6_need_cache(ifp) == 0)
2602 		return (0);
2603 
2604 	dst = (struct sockaddr *)&ia->ia_addr;
2605 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2606 	if (ln == NULL)
2607 		return (ENOBUFS);
2608 
2609 	IF_AFDATA_WLOCK(ifp);
2610 	LLE_WLOCK(ln);
2611 	/* Unlink any entry if exists */
2612 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2613 	if (ln_tmp != NULL)
2614 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2615 	lltable_link_entry(LLTABLE6(ifp), ln);
2616 	IF_AFDATA_WUNLOCK(ifp);
2617 
2618 	if (ln_tmp != NULL)
2619 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2620 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2621 
2622 	LLE_WUNLOCK(ln);
2623 	if (ln_tmp != NULL)
2624 		llentry_free(ln_tmp);
2625 
2626 	return (0);
2627 }
2628 
2629 /*
2630  * Removes either all lle entries for given @ia, or lle
2631  * corresponding to @ia address.
2632  */
2633 void
nd6_rem_ifa_lle(struct in6_ifaddr * ia,int all)2634 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2635 {
2636 	struct sockaddr_in6 mask, addr;
2637 	struct sockaddr *saddr, *smask;
2638 	struct ifnet *ifp;
2639 
2640 	ifp = ia->ia_ifa.ifa_ifp;
2641 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2642 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2643 	saddr = (struct sockaddr *)&addr;
2644 	smask = (struct sockaddr *)&mask;
2645 
2646 	if (all != 0)
2647 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2648 	else
2649 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2650 }
2651 
2652 static int
nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)2653 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2654 {
2655 	struct in6_prefix p;
2656 	struct sockaddr_in6 s6;
2657 	struct nd_prefix *pr;
2658 	struct nd_pfxrouter *pfr;
2659 	time_t maxexpire;
2660 	int error;
2661 	char ip6buf[INET6_ADDRSTRLEN];
2662 
2663 	if (req->newptr)
2664 		return (EPERM);
2665 
2666 	error = sysctl_wire_old_buffer(req, 0);
2667 	if (error != 0)
2668 		return (error);
2669 
2670 	bzero(&p, sizeof(p));
2671 	p.origin = PR_ORIG_RA;
2672 	bzero(&s6, sizeof(s6));
2673 	s6.sin6_family = AF_INET6;
2674 	s6.sin6_len = sizeof(s6);
2675 
2676 	ND6_RLOCK();
2677 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2678 		p.prefix = pr->ndpr_prefix;
2679 		if (sa6_recoverscope(&p.prefix)) {
2680 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2681 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2682 			/* XXX: press on... */
2683 		}
2684 		p.raflags = pr->ndpr_raf;
2685 		p.prefixlen = pr->ndpr_plen;
2686 		p.vltime = pr->ndpr_vltime;
2687 		p.pltime = pr->ndpr_pltime;
2688 		p.if_index = pr->ndpr_ifp->if_index;
2689 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2690 			p.expire = 0;
2691 		else {
2692 			/* XXX: we assume time_t is signed. */
2693 			maxexpire = (-1) &
2694 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2695 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2696 				p.expire = pr->ndpr_lastupdate +
2697 				    pr->ndpr_vltime +
2698 				    (time_second - time_uptime);
2699 			else
2700 				p.expire = maxexpire;
2701 		}
2702 		p.refcnt = pr->ndpr_addrcnt;
2703 		p.flags = pr->ndpr_stateflags;
2704 		p.advrtrs = 0;
2705 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2706 			p.advrtrs++;
2707 		error = SYSCTL_OUT(req, &p, sizeof(p));
2708 		if (error != 0)
2709 			break;
2710 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2711 			s6.sin6_addr = pfr->router->rtaddr;
2712 			if (sa6_recoverscope(&s6))
2713 				log(LOG_ERR,
2714 				    "scope error in prefix list (%s)\n",
2715 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2716 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2717 			if (error != 0)
2718 				goto out;
2719 		}
2720 	}
2721 out:
2722 	ND6_RUNLOCK();
2723 	return (error);
2724 }
2725 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2726 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2727 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2728 	"NDP prefix list");
2729 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2730 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2731 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2732 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2733