xref: /freebsd-13-stable/sys/netinet6/nd6.c (revision d0105578f05f507614185436cf5aba92a2b61782)
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_types.h>
63 #include <net/route.h>
64 #include <net/route/route_ctl.h>
65 #include <net/route/nhop.h>
66 #include <net/vnet.h>
67 
68 #include <netinet/in.h>
69 #include <netinet/in_kdtrace.h>
70 #include <net/if_llatbl.h>
71 #include <netinet/if_ether.h>
72 #include <netinet6/in6_fib.h>
73 #include <netinet6/in6_var.h>
74 #include <netinet/ip6.h>
75 #include <netinet6/ip6_var.h>
76 #include <netinet6/scope6_var.h>
77 #include <netinet6/nd6.h>
78 #include <netinet6/in6_ifattach.h>
79 #include <netinet/icmp6.h>
80 #include <netinet6/send.h>
81 
82 #include <sys/limits.h>
83 
84 #include <security/mac/mac_framework.h>
85 
86 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
87 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
88 
89 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
90 
91 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
92 
93 /* timer values */
94 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
95 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
96 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
97 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
98 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
99 					 * local traffic */
100 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
101 					 * collection timer */
102 
103 /* preventing too many loops in ND option parsing */
104 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
105 
106 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
107 					 * layer hints */
108 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 16; /* max pkts cached in unresolved
109 					 * ND entries */
110 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
111 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
112 
113 #ifdef ND6_DEBUG
114 VNET_DEFINE(int, nd6_debug) = 1;
115 #else
116 VNET_DEFINE(int, nd6_debug) = 0;
117 #endif
118 
119 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
120 
121 VNET_DEFINE(struct nd_prhead, nd_prefix);
122 VNET_DEFINE(struct rwlock, nd6_lock);
123 VNET_DEFINE(uint64_t, nd6_list_genid);
124 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
125 
126 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
127 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
128 
129 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
130 
131 static bool nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
132 	struct ifnet *);
133 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
134 static void nd6_slowtimo(void *);
135 static int regen_tmpaddr(struct in6_ifaddr *);
136 static void nd6_free(struct llentry **, int);
137 static void nd6_free_redirect(const struct llentry *);
138 static void nd6_llinfo_timer(void *);
139 static void nd6_llinfo_settimer_locked(struct llentry *, long);
140 static int nd6_resolve_slow(struct ifnet *, int, int, struct mbuf *,
141     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
142 static int nd6_need_cache(struct ifnet *);
143 
144 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
145 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
146 
147 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
148 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
149 
150 SYSCTL_DECL(_net_inet6_icmp6);
151 
152 static void
nd6_lle_event(void * arg __unused,struct llentry * lle,int evt)153 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
154 {
155 	struct rt_addrinfo rtinfo;
156 	struct sockaddr_in6 dst;
157 	struct sockaddr_dl gw;
158 	struct ifnet *ifp;
159 	int type;
160 	int fibnum;
161 
162 	LLE_WLOCK_ASSERT(lle);
163 
164 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
165 		return;
166 
167 	switch (evt) {
168 	case LLENTRY_RESOLVED:
169 		type = RTM_ADD;
170 		KASSERT(lle->la_flags & LLE_VALID,
171 		    ("%s: %p resolved but not valid?", __func__, lle));
172 		break;
173 	case LLENTRY_EXPIRED:
174 		type = RTM_DELETE;
175 		break;
176 	default:
177 		return;
178 	}
179 
180 	ifp = lltable_get_ifp(lle->lle_tbl);
181 
182 	bzero(&dst, sizeof(dst));
183 	bzero(&gw, sizeof(gw));
184 	bzero(&rtinfo, sizeof(rtinfo));
185 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
186 	dst.sin6_scope_id = in6_getscopezone(ifp,
187 	    in6_addrscope(&dst.sin6_addr));
188 	gw.sdl_len = sizeof(struct sockaddr_dl);
189 	gw.sdl_family = AF_LINK;
190 	gw.sdl_alen = ifp->if_addrlen;
191 	gw.sdl_index = ifp->if_index;
192 	gw.sdl_type = ifp->if_type;
193 	if (evt == LLENTRY_RESOLVED)
194 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
195 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
196 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
197 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
198 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
199 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
200 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
201 }
202 
203 /*
204  * A handler for interface link layer address change event.
205  */
206 static void
nd6_iflladdr(void * arg __unused,struct ifnet * ifp)207 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
208 {
209 	if (ifp->if_afdata[AF_INET6] == NULL)
210 		return;
211 
212 	lltable_update_ifaddr(LLTABLE6(ifp));
213 }
214 
215 void
nd6_init(void)216 nd6_init(void)
217 {
218 
219 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
220 	rw_init(&V_nd6_lock, "nd6 list");
221 
222 	LIST_INIT(&V_nd_prefix);
223 	nd6_defrouter_init();
224 
225 	/* Start timers. */
226 	callout_init(&V_nd6_slowtimo_ch, 1);
227 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
228 	    nd6_slowtimo, curvnet);
229 
230 	callout_init(&V_nd6_timer_ch, 1);
231 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
232 
233 	nd6_dad_init();
234 	if (IS_DEFAULT_VNET(curvnet)) {
235 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
236 		    NULL, EVENTHANDLER_PRI_ANY);
237 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
238 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
239 		ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
240 		    nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
241 	}
242 }
243 
244 #ifdef VIMAGE
245 void
nd6_destroy(void)246 nd6_destroy(void)
247 {
248 
249 	callout_drain(&V_nd6_slowtimo_ch);
250 	callout_drain(&V_nd6_timer_ch);
251 	if (IS_DEFAULT_VNET(curvnet)) {
252 		EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
253 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
254 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
255 	}
256 	rw_destroy(&V_nd6_lock);
257 	mtx_destroy(&V_nd6_onlink_mtx);
258 }
259 #endif
260 
261 struct nd_ifinfo *
nd6_ifattach(struct ifnet * ifp)262 nd6_ifattach(struct ifnet *ifp)
263 {
264 	struct nd_ifinfo *nd;
265 
266 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
267 	nd->initialized = 1;
268 
269 	nd->chlim = IPV6_DEFHLIM;
270 	nd->basereachable = REACHABLE_TIME;
271 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
272 	nd->retrans = RETRANS_TIMER;
273 
274 	nd->flags = ND6_IFF_PERFORMNUD;
275 
276 	/* Set IPv6 disabled on all interfaces but loopback by default. */
277 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
278 		nd->flags |= ND6_IFF_IFDISABLED;
279 
280 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
281 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
282 	 * default regardless of the V_ip6_auto_linklocal configuration to
283 	 * give a reasonable default behavior.
284 	 */
285 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE &&
286 	    ifp->if_type != IFT_WIREGUARD) || (ifp->if_flags & IFF_LOOPBACK))
287 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
288 	/*
289 	 * A loopback interface does not need to accept RTADV.
290 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
291 	 * default regardless of the V_ip6_accept_rtadv configuration to
292 	 * prevent the interface from accepting RA messages arrived
293 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
294 	 */
295 	if (V_ip6_accept_rtadv &&
296 	    !(ifp->if_flags & IFF_LOOPBACK) &&
297 	    (ifp->if_type != IFT_BRIDGE)) {
298 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
299 			/* If we globally accept rtadv, assume IPv6 on. */
300 			nd->flags &= ~ND6_IFF_IFDISABLED;
301 	}
302 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
303 		nd->flags |= ND6_IFF_NO_RADR;
304 
305 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
306 	nd6_setmtu0(ifp, nd);
307 
308 	return nd;
309 }
310 
311 void
nd6_ifdetach(struct ifnet * ifp,struct nd_ifinfo * nd)312 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
313 {
314 	struct epoch_tracker et;
315 	struct ifaddr *ifa, *next;
316 
317 	NET_EPOCH_ENTER(et);
318 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
319 		if (ifa->ifa_addr->sa_family != AF_INET6)
320 			continue;
321 
322 		/* stop DAD processing */
323 		nd6_dad_stop(ifa);
324 	}
325 	NET_EPOCH_EXIT(et);
326 
327 	free(nd, M_IP6NDP);
328 }
329 
330 /*
331  * Reset ND level link MTU. This function is called when the physical MTU
332  * changes, which means we might have to adjust the ND level MTU.
333  */
334 void
nd6_setmtu(struct ifnet * ifp)335 nd6_setmtu(struct ifnet *ifp)
336 {
337 	if (ifp->if_afdata[AF_INET6] == NULL)
338 		return;
339 
340 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
341 }
342 
343 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
344 void
nd6_setmtu0(struct ifnet * ifp,struct nd_ifinfo * ndi)345 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
346 {
347 	u_int32_t omaxmtu;
348 
349 	omaxmtu = ndi->maxmtu;
350 	ndi->maxmtu = ifp->if_mtu;
351 
352 	/*
353 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
354 	 * undesirable situation.  We thus notify the operator of the change
355 	 * explicitly.  The check for omaxmtu is necessary to restrict the
356 	 * log to the case of changing the MTU, not initializing it.
357 	 */
358 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
359 		log(LOG_NOTICE, "nd6_setmtu0: "
360 		    "new link MTU on %s (%lu) is too small for IPv6\n",
361 		    if_name(ifp), (unsigned long)ndi->maxmtu);
362 	}
363 
364 	if (ndi->maxmtu > V_in6_maxmtu)
365 		in6_setmaxmtu(); /* check all interfaces just in case */
366 
367 }
368 
369 void
nd6_option_init(void * opt,int icmp6len,union nd_opts * ndopts)370 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
371 {
372 
373 	bzero(ndopts, sizeof(*ndopts));
374 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
375 	ndopts->nd_opts_last
376 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
377 
378 	if (icmp6len == 0) {
379 		ndopts->nd_opts_done = 1;
380 		ndopts->nd_opts_search = NULL;
381 	}
382 }
383 
384 /*
385  * Take one ND option.
386  */
387 struct nd_opt_hdr *
nd6_option(union nd_opts * ndopts)388 nd6_option(union nd_opts *ndopts)
389 {
390 	struct nd_opt_hdr *nd_opt;
391 	int olen;
392 
393 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
394 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
395 	    __func__));
396 	if (ndopts->nd_opts_search == NULL)
397 		return NULL;
398 	if (ndopts->nd_opts_done)
399 		return NULL;
400 
401 	nd_opt = ndopts->nd_opts_search;
402 
403 	/* make sure nd_opt_len is inside the buffer */
404 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
405 		bzero(ndopts, sizeof(*ndopts));
406 		return NULL;
407 	}
408 
409 	olen = nd_opt->nd_opt_len << 3;
410 	if (olen == 0) {
411 		/*
412 		 * Message validation requires that all included
413 		 * options have a length that is greater than zero.
414 		 */
415 		bzero(ndopts, sizeof(*ndopts));
416 		return NULL;
417 	}
418 
419 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
420 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
421 		/* option overruns the end of buffer, invalid */
422 		bzero(ndopts, sizeof(*ndopts));
423 		return NULL;
424 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
425 		/* reached the end of options chain */
426 		ndopts->nd_opts_done = 1;
427 		ndopts->nd_opts_search = NULL;
428 	}
429 	return nd_opt;
430 }
431 
432 /*
433  * Parse multiple ND options.
434  * This function is much easier to use, for ND routines that do not need
435  * multiple options of the same type.
436  */
437 int
nd6_options(union nd_opts * ndopts)438 nd6_options(union nd_opts *ndopts)
439 {
440 	struct nd_opt_hdr *nd_opt;
441 	int i = 0;
442 
443 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
444 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
445 	    __func__));
446 	if (ndopts->nd_opts_search == NULL)
447 		return 0;
448 
449 	while (1) {
450 		nd_opt = nd6_option(ndopts);
451 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
452 			/*
453 			 * Message validation requires that all included
454 			 * options have a length that is greater than zero.
455 			 */
456 			ICMP6STAT_INC(icp6s_nd_badopt);
457 			bzero(ndopts, sizeof(*ndopts));
458 			return -1;
459 		}
460 
461 		if (nd_opt == NULL)
462 			goto skip1;
463 
464 		switch (nd_opt->nd_opt_type) {
465 		case ND_OPT_SOURCE_LINKADDR:
466 		case ND_OPT_TARGET_LINKADDR:
467 		case ND_OPT_MTU:
468 		case ND_OPT_REDIRECTED_HEADER:
469 		case ND_OPT_NONCE:
470 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
471 				nd6log((LOG_INFO,
472 				    "duplicated ND6 option found (type=%d)\n",
473 				    nd_opt->nd_opt_type));
474 				/* XXX bark? */
475 			} else {
476 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
477 					= nd_opt;
478 			}
479 			break;
480 		case ND_OPT_PREFIX_INFORMATION:
481 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
482 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
483 					= nd_opt;
484 			}
485 			ndopts->nd_opts_pi_end =
486 				(struct nd_opt_prefix_info *)nd_opt;
487 			break;
488 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
489 		case ND_OPT_RDNSS:	/* RFC 6106 */
490 		case ND_OPT_DNSSL:	/* RFC 6106 */
491 			/*
492 			 * Silently ignore options we know and do not care about
493 			 * in the kernel.
494 			 */
495 			break;
496 		default:
497 			/*
498 			 * Unknown options must be silently ignored,
499 			 * to accommodate future extension to the protocol.
500 			 */
501 			nd6log((LOG_DEBUG,
502 			    "nd6_options: unsupported option %d - "
503 			    "option ignored\n", nd_opt->nd_opt_type));
504 		}
505 
506 skip1:
507 		i++;
508 		if (i > V_nd6_maxndopt) {
509 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
510 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
511 			break;
512 		}
513 
514 		if (ndopts->nd_opts_done)
515 			break;
516 	}
517 
518 	return 0;
519 }
520 
521 /*
522  * ND6 timer routine to handle ND6 entries
523  */
524 static void
nd6_llinfo_settimer_locked(struct llentry * ln,long tick)525 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
526 {
527 	int canceled;
528 
529 	LLE_WLOCK_ASSERT(ln);
530 
531 	/* Do not schedule timers for child LLEs. */
532 	if (ln->la_flags & LLE_CHILD)
533 		return;
534 
535 	if (tick < 0) {
536 		ln->la_expire = 0;
537 		ln->ln_ntick = 0;
538 		canceled = callout_stop(&ln->lle_timer);
539 	} else {
540 		ln->la_expire = time_uptime + tick / hz;
541 		LLE_ADDREF(ln);
542 		if (tick > INT_MAX) {
543 			ln->ln_ntick = tick - INT_MAX;
544 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
545 			    nd6_llinfo_timer, ln);
546 		} else {
547 			ln->ln_ntick = 0;
548 			canceled = callout_reset(&ln->lle_timer, tick,
549 			    nd6_llinfo_timer, ln);
550 		}
551 	}
552 	if (canceled > 0)
553 		LLE_REMREF(ln);
554 }
555 
556 /*
557  * Gets source address of the first packet in hold queue
558  * and stores it in @src.
559  * Returns pointer to @src (if hold queue is not empty) or NULL.
560  *
561  * Set noinline to be dtrace-friendly
562  */
563 static __noinline struct in6_addr *
nd6_llinfo_get_holdsrc(struct llentry * ln,struct in6_addr * src)564 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
565 {
566 	struct ip6_hdr hdr;
567 	struct mbuf *m;
568 
569 	if (ln->la_hold == NULL)
570 		return (NULL);
571 
572 	/*
573 	 * assume every packet in la_hold has the same IP header
574 	 */
575 	m = ln->la_hold;
576 	if (sizeof(hdr) > m->m_len)
577 		return (NULL);
578 
579 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
580 	*src = hdr.ip6_src;
581 
582 	return (src);
583 }
584 
585 /*
586  * Checks if we need to switch from STALE state.
587  *
588  * RFC 4861 requires switching from STALE to DELAY state
589  * on first packet matching entry, waiting V_nd6_delay and
590  * transition to PROBE state (if upper layer confirmation was
591  * not received).
592  *
593  * This code performs a bit differently:
594  * On packet hit we don't change state (but desired state
595  * can be guessed by control plane). However, after V_nd6_delay
596  * seconds code will transition to PROBE state (so DELAY state
597  * is kinda skipped in most situations).
598  *
599  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
600  * we perform the following upon entering STALE state:
601  *
602  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
603  * if packet was transmitted at the start of given interval, we
604  * would be able to switch to PROBE state in V_nd6_delay seconds
605  * as user expects.
606  *
607  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
608  * lle in STALE state (remaining timer value stored in lle_remtime).
609  *
610  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
611  * seconds ago.
612  *
613  * Returns non-zero value if the entry is still STALE (storing
614  * the next timer interval in @pdelay).
615  *
616  * Returns zero value if original timer expired or we need to switch to
617  * PROBE (store that in @do_switch variable).
618  */
619 static int
nd6_is_stale(struct llentry * lle,long * pdelay,int * do_switch)620 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
621 {
622 	int nd_delay, nd_gctimer;
623 	time_t lle_hittime;
624 	long delay;
625 
626 	*do_switch = 0;
627 	nd_gctimer = V_nd6_gctimer;
628 	nd_delay = V_nd6_delay;
629 
630 	lle_hittime = llentry_get_hittime(lle);
631 
632 	if (lle_hittime == 0) {
633 		/*
634 		 * Datapath feedback has been requested upon entering
635 		 * STALE state. No packets has been passed using this lle.
636 		 * Ask for the timer reschedule and keep STALE state.
637 		 */
638 		delay = (long)(MIN(nd_gctimer, nd_delay));
639 		delay *= hz;
640 		if (lle->lle_remtime > delay)
641 			lle->lle_remtime -= delay;
642 		else {
643 			delay = lle->lle_remtime;
644 			lle->lle_remtime = 0;
645 		}
646 
647 		if (delay == 0) {
648 			/*
649 			 * The original ng6_gctime timeout ended,
650 			 * no more rescheduling.
651 			 */
652 			return (0);
653 		}
654 
655 		*pdelay = delay;
656 		return (1);
657 	}
658 
659 	/*
660 	 * Packet received. Verify timestamp
661 	 */
662 	delay = (long)(time_uptime - lle_hittime);
663 	if (delay < nd_delay) {
664 		/*
665 		 * V_nd6_delay still not passed since the first
666 		 * hit in STALE state.
667 		 * Reschedule timer and return.
668 		 */
669 		*pdelay = (long)(nd_delay - delay) * hz;
670 		return (1);
671 	}
672 
673 	/* Request switching to probe */
674 	*do_switch = 1;
675 	return (0);
676 }
677 
678 /*
679  * Switch @lle state to new state optionally arming timers.
680  *
681  * Set noinline to be dtrace-friendly
682  */
683 __noinline void
nd6_llinfo_setstate(struct llentry * lle,int newstate)684 nd6_llinfo_setstate(struct llentry *lle, int newstate)
685 {
686 	struct ifnet *ifp;
687 	int nd_gctimer, nd_delay;
688 	long delay, remtime;
689 
690 	delay = 0;
691 	remtime = 0;
692 
693 	switch (newstate) {
694 	case ND6_LLINFO_INCOMPLETE:
695 		ifp = lle->lle_tbl->llt_ifp;
696 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
697 		break;
698 	case ND6_LLINFO_REACHABLE:
699 		if (!ND6_LLINFO_PERMANENT(lle)) {
700 			ifp = lle->lle_tbl->llt_ifp;
701 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
702 		}
703 		break;
704 	case ND6_LLINFO_STALE:
705 
706 		llentry_request_feedback(lle);
707 		nd_delay = V_nd6_delay;
708 		nd_gctimer = V_nd6_gctimer;
709 
710 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
711 		remtime = (long)nd_gctimer * hz - delay;
712 		break;
713 	case ND6_LLINFO_DELAY:
714 		lle->la_asked = 0;
715 		delay = (long)V_nd6_delay * hz;
716 		break;
717 	}
718 
719 	if (delay > 0)
720 		nd6_llinfo_settimer_locked(lle, delay);
721 
722 	lle->lle_remtime = remtime;
723 	lle->ln_state = newstate;
724 }
725 
726 /*
727  * Timer-dependent part of nd state machine.
728  *
729  * Set noinline to be dtrace-friendly
730  */
731 static __noinline void
nd6_llinfo_timer(void * arg)732 nd6_llinfo_timer(void *arg)
733 {
734 	struct epoch_tracker et;
735 	struct llentry *ln;
736 	struct in6_addr *dst, *pdst, *psrc, src;
737 	struct ifnet *ifp;
738 	struct nd_ifinfo *ndi;
739 	int do_switch, send_ns;
740 	long delay;
741 
742 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
743 	ln = (struct llentry *)arg;
744 	ifp = lltable_get_ifp(ln->lle_tbl);
745 	CURVNET_SET(ifp->if_vnet);
746 
747 	ND6_RLOCK();
748 	LLE_WLOCK(ln);
749 	if (callout_pending(&ln->lle_timer)) {
750 		/*
751 		 * Here we are a bit odd here in the treatment of
752 		 * active/pending. If the pending bit is set, it got
753 		 * rescheduled before I ran. The active
754 		 * bit we ignore, since if it was stopped
755 		 * in ll_tablefree() and was currently running
756 		 * it would have return 0 so the code would
757 		 * not have deleted it since the callout could
758 		 * not be stopped so we want to go through
759 		 * with the delete here now. If the callout
760 		 * was restarted, the pending bit will be back on and
761 		 * we just want to bail since the callout_reset would
762 		 * return 1 and our reference would have been removed
763 		 * by nd6_llinfo_settimer_locked above since canceled
764 		 * would have been 1.
765 		 */
766 		LLE_WUNLOCK(ln);
767 		ND6_RUNLOCK();
768 		CURVNET_RESTORE();
769 		return;
770 	}
771 	NET_EPOCH_ENTER(et);
772 	ndi = ND_IFINFO(ifp);
773 	send_ns = 0;
774 	dst = &ln->r_l3addr.addr6;
775 	pdst = dst;
776 
777 	if (ln->ln_ntick > 0) {
778 		if (ln->ln_ntick > INT_MAX) {
779 			ln->ln_ntick -= INT_MAX;
780 			nd6_llinfo_settimer_locked(ln, INT_MAX);
781 		} else {
782 			ln->ln_ntick = 0;
783 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
784 		}
785 		goto done;
786 	}
787 
788 	if (ln->la_flags & LLE_STATIC) {
789 		goto done;
790 	}
791 
792 	if (ln->la_flags & LLE_DELETED) {
793 		nd6_free(&ln, 0);
794 		goto done;
795 	}
796 
797 	switch (ln->ln_state) {
798 	case ND6_LLINFO_INCOMPLETE:
799 		if (ln->la_asked < V_nd6_mmaxtries) {
800 			ln->la_asked++;
801 			send_ns = 1;
802 			/* Send NS to multicast address */
803 			pdst = NULL;
804 		} else {
805 			struct mbuf *m;
806 
807 			ICMP6STAT_ADD(icp6s_dropped, ln->la_numheld);
808 
809 			m = ln->la_hold;
810 			if (m != NULL) {
811 				/*
812 				 * assuming every packet in la_hold has the
813 				 * same IP header.  Send error after unlock.
814 				 */
815 				ln->la_hold = m->m_nextpkt;
816 				m->m_nextpkt = NULL;
817 				ln->la_numheld--;
818 			}
819 			nd6_free(&ln, 0);
820 			if (m != NULL) {
821 				struct mbuf *n = m;
822 
823 				/*
824 				 * if there are any ummapped mbufs, we
825 				 * must free them, rather than using
826 				 * them for an ICMP, as they cannot be
827 				 * checksummed.
828 				 */
829 				while ((n = n->m_next) != NULL) {
830 					if (n->m_flags & M_EXTPG)
831 						break;
832 				}
833 				if (n != NULL) {
834 					m_freem(m);
835 					m = NULL;
836 				} else {
837 					icmp6_error2(m, ICMP6_DST_UNREACH,
838 					    ICMP6_DST_UNREACH_ADDR, 0, ifp);
839 				}
840 			}
841 		}
842 		break;
843 	case ND6_LLINFO_REACHABLE:
844 		if (!ND6_LLINFO_PERMANENT(ln))
845 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
846 		break;
847 
848 	case ND6_LLINFO_STALE:
849 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
850 			/*
851 			 * No packet has used this entry and GC timeout
852 			 * has not been passed. Reschedule timer and
853 			 * return.
854 			 */
855 			nd6_llinfo_settimer_locked(ln, delay);
856 			break;
857 		}
858 
859 		if (do_switch == 0) {
860 			/*
861 			 * GC timer has ended and entry hasn't been used.
862 			 * Run Garbage collector (RFC 4861, 5.3)
863 			 */
864 			if (!ND6_LLINFO_PERMANENT(ln))
865 				nd6_free(&ln, 1);
866 			break;
867 		}
868 
869 		/* Entry has been used AND delay timer has ended. */
870 
871 		/* FALLTHROUGH */
872 
873 	case ND6_LLINFO_DELAY:
874 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
875 			/* We need NUD */
876 			ln->la_asked = 1;
877 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
878 			send_ns = 1;
879 		} else
880 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
881 		break;
882 	case ND6_LLINFO_PROBE:
883 		if (ln->la_asked < V_nd6_umaxtries) {
884 			ln->la_asked++;
885 			send_ns = 1;
886 		} else {
887 			nd6_free(&ln, 0);
888 		}
889 		break;
890 	default:
891 		panic("%s: paths in a dark night can be confusing: %d",
892 		    __func__, ln->ln_state);
893 	}
894 done:
895 	if (ln != NULL)
896 		ND6_RUNLOCK();
897 	if (send_ns != 0) {
898 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
899 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
900 		LLE_FREE_LOCKED(ln);
901 		ln = NULL;
902 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
903 	}
904 
905 	if (ln != NULL)
906 		LLE_FREE_LOCKED(ln);
907 	NET_EPOCH_EXIT(et);
908 	CURVNET_RESTORE();
909 }
910 
911 /*
912  * ND6 timer routine to expire default route list and prefix list
913  */
914 void
nd6_timer(void * arg)915 nd6_timer(void *arg)
916 {
917 	CURVNET_SET((struct vnet *) arg);
918 	struct epoch_tracker et;
919 	struct nd_prhead prl;
920 	struct nd_prefix *pr, *npr;
921 	struct ifnet *ifp;
922 	struct in6_ifaddr *ia6, *nia6;
923 	uint64_t genid;
924 
925 	LIST_INIT(&prl);
926 
927 	NET_EPOCH_ENTER(et);
928 	nd6_defrouter_timer();
929 
930 	/*
931 	 * expire interface addresses.
932 	 * in the past the loop was inside prefix expiry processing.
933 	 * However, from a stricter speci-confrmance standpoint, we should
934 	 * rather separate address lifetimes and prefix lifetimes.
935 	 *
936 	 * XXXRW: in6_ifaddrhead locking.
937 	 */
938   addrloop:
939 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
940 		/* check address lifetime */
941 		if (IFA6_IS_INVALID(ia6)) {
942 			int regen = 0;
943 
944 			/*
945 			 * If the expiring address is temporary, try
946 			 * regenerating a new one.  This would be useful when
947 			 * we suspended a laptop PC, then turned it on after a
948 			 * period that could invalidate all temporary
949 			 * addresses.  Although we may have to restart the
950 			 * loop (see below), it must be after purging the
951 			 * address.  Otherwise, we'd see an infinite loop of
952 			 * regeneration.
953 			 */
954 			if (V_ip6_use_tempaddr &&
955 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
956 				if (regen_tmpaddr(ia6) == 0)
957 					regen = 1;
958 			}
959 
960 			in6_purgeaddr(&ia6->ia_ifa);
961 
962 			if (regen)
963 				goto addrloop; /* XXX: see below */
964 		} else if (IFA6_IS_DEPRECATED(ia6)) {
965 			int oldflags = ia6->ia6_flags;
966 
967 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
968 
969 			/*
970 			 * If a temporary address has just become deprecated,
971 			 * regenerate a new one if possible.
972 			 */
973 			if (V_ip6_use_tempaddr &&
974 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
975 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
976 				if (regen_tmpaddr(ia6) == 0) {
977 					/*
978 					 * A new temporary address is
979 					 * generated.
980 					 * XXX: this means the address chain
981 					 * has changed while we are still in
982 					 * the loop.  Although the change
983 					 * would not cause disaster (because
984 					 * it's not a deletion, but an
985 					 * addition,) we'd rather restart the
986 					 * loop just for safety.  Or does this
987 					 * significantly reduce performance??
988 					 */
989 					goto addrloop;
990 				}
991 			}
992 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
993 			/*
994 			 * Schedule DAD for a tentative address.  This happens
995 			 * if the interface was down or not running
996 			 * when the address was configured.
997 			 */
998 			int delay;
999 
1000 			delay = arc4random() %
1001 			    (MAX_RTR_SOLICITATION_DELAY * hz);
1002 			nd6_dad_start((struct ifaddr *)ia6, delay);
1003 		} else {
1004 			/*
1005 			 * Check status of the interface.  If it is down,
1006 			 * mark the address as tentative for future DAD.
1007 			 */
1008 			ifp = ia6->ia_ifp;
1009 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1010 			    ((ifp->if_flags & IFF_UP) == 0 ||
1011 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1012 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1013 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1014 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1015 			}
1016 
1017 			/*
1018 			 * A new RA might have made a deprecated address
1019 			 * preferred.
1020 			 */
1021 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1022 		}
1023 	}
1024 	NET_EPOCH_EXIT(et);
1025 
1026 	ND6_WLOCK();
1027 restart:
1028 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1029 		/*
1030 		 * Expire prefixes. Since the pltime is only used for
1031 		 * autoconfigured addresses, pltime processing for prefixes is
1032 		 * not necessary.
1033 		 *
1034 		 * Only unlink after all derived addresses have expired. This
1035 		 * may not occur until two hours after the prefix has expired
1036 		 * per RFC 4862. If the prefix expires before its derived
1037 		 * addresses, mark it off-link. This will be done automatically
1038 		 * after unlinking if no address references remain.
1039 		 */
1040 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1041 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1042 			continue;
1043 
1044 		if (pr->ndpr_addrcnt == 0) {
1045 			nd6_prefix_unlink(pr, &prl);
1046 			continue;
1047 		}
1048 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1049 			genid = V_nd6_list_genid;
1050 			nd6_prefix_ref(pr);
1051 			ND6_WUNLOCK();
1052 			ND6_ONLINK_LOCK();
1053 			(void)nd6_prefix_offlink(pr);
1054 			ND6_ONLINK_UNLOCK();
1055 			ND6_WLOCK();
1056 			nd6_prefix_rele(pr);
1057 			if (genid != V_nd6_list_genid)
1058 				goto restart;
1059 		}
1060 	}
1061 	ND6_WUNLOCK();
1062 
1063 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1064 		LIST_REMOVE(pr, ndpr_entry);
1065 		nd6_prefix_del(pr);
1066 	}
1067 
1068 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1069 	    nd6_timer, curvnet);
1070 
1071 	CURVNET_RESTORE();
1072 }
1073 
1074 /*
1075  * ia6 - deprecated/invalidated temporary address
1076  */
1077 static int
regen_tmpaddr(struct in6_ifaddr * ia6)1078 regen_tmpaddr(struct in6_ifaddr *ia6)
1079 {
1080 	struct ifaddr *ifa;
1081 	struct ifnet *ifp;
1082 	struct in6_ifaddr *public_ifa6 = NULL;
1083 
1084 	NET_EPOCH_ASSERT();
1085 
1086 	ifp = ia6->ia_ifa.ifa_ifp;
1087 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1088 		struct in6_ifaddr *it6;
1089 
1090 		if (ifa->ifa_addr->sa_family != AF_INET6)
1091 			continue;
1092 
1093 		it6 = (struct in6_ifaddr *)ifa;
1094 
1095 		/* ignore no autoconf addresses. */
1096 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1097 			continue;
1098 
1099 		/* ignore autoconf addresses with different prefixes. */
1100 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1101 			continue;
1102 
1103 		/*
1104 		 * Now we are looking at an autoconf address with the same
1105 		 * prefix as ours.  If the address is temporary and is still
1106 		 * preferred, do not create another one.  It would be rare, but
1107 		 * could happen, for example, when we resume a laptop PC after
1108 		 * a long period.
1109 		 */
1110 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1111 		    !IFA6_IS_DEPRECATED(it6)) {
1112 			public_ifa6 = NULL;
1113 			break;
1114 		}
1115 
1116 		/*
1117 		 * This is a public autoconf address that has the same prefix
1118 		 * as ours.  If it is preferred, keep it.  We can't break the
1119 		 * loop here, because there may be a still-preferred temporary
1120 		 * address with the prefix.
1121 		 */
1122 		if (!IFA6_IS_DEPRECATED(it6))
1123 			public_ifa6 = it6;
1124 	}
1125 	if (public_ifa6 != NULL)
1126 		ifa_ref(&public_ifa6->ia_ifa);
1127 
1128 	if (public_ifa6 != NULL) {
1129 		int e;
1130 
1131 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1132 			ifa_free(&public_ifa6->ia_ifa);
1133 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1134 			    " tmp addr,errno=%d\n", e);
1135 			return (-1);
1136 		}
1137 		ifa_free(&public_ifa6->ia_ifa);
1138 		return (0);
1139 	}
1140 
1141 	return (-1);
1142 }
1143 
1144 /*
1145  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1146  * cache entries are freed in in6_domifdetach().
1147  */
1148 void
nd6_purge(struct ifnet * ifp)1149 nd6_purge(struct ifnet *ifp)
1150 {
1151 	struct nd_prhead prl;
1152 	struct nd_prefix *pr, *npr;
1153 
1154 	LIST_INIT(&prl);
1155 
1156 	/* Purge default router list entries toward ifp. */
1157 	nd6_defrouter_purge(ifp);
1158 
1159 	ND6_WLOCK();
1160 	/*
1161 	 * Remove prefixes on ifp. We should have already removed addresses on
1162 	 * this interface, so no addresses should be referencing these prefixes.
1163 	 */
1164 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1165 		if (pr->ndpr_ifp == ifp)
1166 			nd6_prefix_unlink(pr, &prl);
1167 	}
1168 	ND6_WUNLOCK();
1169 
1170 	/* Delete the unlinked prefix objects. */
1171 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1172 		LIST_REMOVE(pr, ndpr_entry);
1173 		nd6_prefix_del(pr);
1174 	}
1175 
1176 	/* cancel default outgoing interface setting */
1177 	if (V_nd6_defifindex == ifp->if_index)
1178 		nd6_setdefaultiface(0);
1179 
1180 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1181 		/* Refresh default router list. */
1182 		defrouter_select_fib(ifp->if_fib);
1183 	}
1184 }
1185 
1186 /*
1187  * the caller acquires and releases the lock on the lltbls
1188  * Returns the llentry locked
1189  */
1190 struct llentry *
nd6_lookup(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1191 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1192 {
1193 	struct sockaddr_in6 sin6;
1194 	struct llentry *ln;
1195 
1196 	bzero(&sin6, sizeof(sin6));
1197 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1198 	sin6.sin6_family = AF_INET6;
1199 	sin6.sin6_addr = *addr6;
1200 
1201 	IF_AFDATA_LOCK_ASSERT(ifp);
1202 
1203 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1204 
1205 	return (ln);
1206 }
1207 
1208 static struct llentry *
nd6_alloc(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1209 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1210 {
1211 	struct sockaddr_in6 sin6;
1212 	struct llentry *ln;
1213 
1214 	bzero(&sin6, sizeof(sin6));
1215 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1216 	sin6.sin6_family = AF_INET6;
1217 	sin6.sin6_addr = *addr6;
1218 
1219 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1220 	if (ln != NULL)
1221 		ln->ln_state = ND6_LLINFO_NOSTATE;
1222 
1223 	return (ln);
1224 }
1225 
1226 /*
1227  * Test whether a given IPv6 address can be a neighbor.
1228  */
1229 static bool
nd6_is_new_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1230 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1231 {
1232 
1233 	/*
1234 	 * A link-local address is always a neighbor.
1235 	 * XXX: a link does not necessarily specify a single interface.
1236 	 */
1237 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1238 		struct sockaddr_in6 sin6_copy;
1239 		u_int32_t zone;
1240 
1241 		/*
1242 		 * We need sin6_copy since sa6_recoverscope() may modify the
1243 		 * content (XXX).
1244 		 */
1245 		sin6_copy = *addr;
1246 		if (sa6_recoverscope(&sin6_copy))
1247 			return (0); /* XXX: should be impossible */
1248 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1249 			return (0);
1250 		if (sin6_copy.sin6_scope_id == zone)
1251 			return (1);
1252 		else
1253 			return (0);
1254 	}
1255 	/* Checking global unicast */
1256 
1257 	/* If an address is directly reachable, it is a neigbor */
1258 	struct nhop_object *nh;
1259 	nh = fib6_lookup(ifp->if_fib, &addr->sin6_addr, 0, NHR_NONE, 0);
1260 	if (nh != NULL && nh->nh_aifp == ifp && (nh->nh_flags & NHF_GATEWAY) == 0)
1261 		return (true);
1262 
1263 	/*
1264 	 * Check prefixes with desired on-link state, as some may be not
1265 	 * installed in the routing table.
1266 	 */
1267 	bool matched = false;
1268 	struct nd_prefix *pr;
1269 	ND6_RLOCK();
1270 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1271 		if (pr->ndpr_ifp != ifp)
1272 			continue;
1273 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0)
1274 			continue;
1275 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1276 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1277 			matched = true;
1278 			break;
1279 		}
1280 	}
1281 	ND6_RUNLOCK();
1282 	if (matched)
1283 		return (true);
1284 
1285 	/*
1286 	 * If the address is assigned on the node of the other side of
1287 	 * a p2p interface, the address should be a neighbor.
1288 	 */
1289 	if (ifp->if_flags & IFF_POINTOPOINT) {
1290 		struct ifaddr *ifa;
1291 
1292 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1293 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1294 				continue;
1295 			if (ifa->ifa_dstaddr != NULL &&
1296 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1297 				return (true);
1298 			}
1299 		}
1300 	}
1301 
1302 	/*
1303 	 * If the default router list is empty, all addresses are regarded
1304 	 * as on-link, and thus, as a neighbor.
1305 	 */
1306 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1307 	    nd6_defrouter_list_empty() &&
1308 	    V_nd6_defifindex == ifp->if_index) {
1309 		return (1);
1310 	}
1311 
1312 	return (0);
1313 }
1314 
1315 /*
1316  * Detect if a given IPv6 address identifies a neighbor on a given link.
1317  * XXX: should take care of the destination of a p2p link?
1318  */
1319 int
nd6_is_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1320 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1321 {
1322 	struct llentry *lle;
1323 	int rc = 0;
1324 
1325 	NET_EPOCH_ASSERT();
1326 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1327 	if (nd6_is_new_addr_neighbor(addr, ifp))
1328 		return (1);
1329 
1330 	/*
1331 	 * Even if the address matches none of our addresses, it might be
1332 	 * in the neighbor cache.
1333 	 */
1334 	if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1335 		LLE_RUNLOCK(lle);
1336 		rc = 1;
1337 	}
1338 	return (rc);
1339 }
1340 
1341 static __noinline void
nd6_free_children(struct llentry * lle)1342 nd6_free_children(struct llentry *lle)
1343 {
1344 	struct llentry *child_lle;
1345 
1346 	NET_EPOCH_ASSERT();
1347 	LLE_WLOCK_ASSERT(lle);
1348 
1349 	while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1350 		LLE_WLOCK(child_lle);
1351 		lltable_unlink_child_entry(child_lle);
1352 		llentry_free(child_lle);
1353 	}
1354 }
1355 
1356 /*
1357  * Tries to update @lle address/prepend data with new @lladdr.
1358  *
1359  * Returns true on success.
1360  * In any case, @lle is returned wlocked.
1361  */
1362 static __noinline bool
nd6_try_set_entry_addr_locked(struct ifnet * ifp,struct llentry * lle,char * lladdr)1363 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1364 {
1365 	u_char buf[LLE_MAX_LINKHDR];
1366 	int fam, off;
1367 	size_t sz;
1368 
1369 	sz = sizeof(buf);
1370 	if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1371 		return (false);
1372 
1373 	/* Update data */
1374 	lltable_set_entry_addr(ifp, lle, buf, sz, off);
1375 
1376 	struct llentry *child_lle;
1377 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1378 		LLE_WLOCK(child_lle);
1379 		fam = child_lle->r_family;
1380 		sz = sizeof(buf);
1381 		if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1382 			/* success */
1383 			lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1384 			child_lle->ln_state = ND6_LLINFO_REACHABLE;
1385 		}
1386 		LLE_WUNLOCK(child_lle);
1387 	}
1388 
1389 	return (true);
1390 }
1391 
1392 bool
nd6_try_set_entry_addr(struct ifnet * ifp,struct llentry * lle,char * lladdr)1393 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1394 {
1395 	NET_EPOCH_ASSERT();
1396 	LLE_WLOCK_ASSERT(lle);
1397 
1398 	if (!lltable_acquire_wlock(ifp, lle))
1399 		return (false);
1400 	bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1401 	IF_AFDATA_WUNLOCK(ifp);
1402 
1403 	return (ret);
1404 }
1405 
1406 /*
1407  * Free an nd6 llinfo entry.
1408  * Since the function would cause significant changes in the kernel, DO NOT
1409  * make it global, unless you have a strong reason for the change, and are sure
1410  * that the change is safe.
1411  *
1412  * Set noinline to be dtrace-friendly
1413  */
1414 static __noinline void
nd6_free(struct llentry ** lnp,int gc)1415 nd6_free(struct llentry **lnp, int gc)
1416 {
1417 	struct ifnet *ifp;
1418 	struct llentry *ln;
1419 	struct nd_defrouter *dr;
1420 
1421 	ln = *lnp;
1422 	*lnp = NULL;
1423 
1424 	LLE_WLOCK_ASSERT(ln);
1425 	ND6_RLOCK_ASSERT();
1426 
1427 	KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1428 
1429 	ifp = lltable_get_ifp(ln->lle_tbl);
1430 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1431 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1432 	else
1433 		dr = NULL;
1434 	ND6_RUNLOCK();
1435 
1436 	if ((ln->la_flags & LLE_DELETED) == 0)
1437 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1438 
1439 	/*
1440 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1441 	 * even though it is not harmful, it was not really necessary.
1442 	 */
1443 
1444 	/* cancel timer */
1445 	nd6_llinfo_settimer_locked(ln, -1);
1446 
1447 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1448 		if (dr != NULL && dr->expire &&
1449 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1450 			/*
1451 			 * If the reason for the deletion is just garbage
1452 			 * collection, and the neighbor is an active default
1453 			 * router, do not delete it.  Instead, reset the GC
1454 			 * timer using the router's lifetime.
1455 			 * Simply deleting the entry would affect default
1456 			 * router selection, which is not necessarily a good
1457 			 * thing, especially when we're using router preference
1458 			 * values.
1459 			 * XXX: the check for ln_state would be redundant,
1460 			 *      but we intentionally keep it just in case.
1461 			 */
1462 			if (dr->expire > time_uptime)
1463 				nd6_llinfo_settimer_locked(ln,
1464 				    (dr->expire - time_uptime) * hz);
1465 			else
1466 				nd6_llinfo_settimer_locked(ln,
1467 				    (long)V_nd6_gctimer * hz);
1468 
1469 			LLE_REMREF(ln);
1470 			LLE_WUNLOCK(ln);
1471 			defrouter_rele(dr);
1472 			return;
1473 		}
1474 
1475 		if (dr) {
1476 			/*
1477 			 * Unreachability of a router might affect the default
1478 			 * router selection and on-link detection of advertised
1479 			 * prefixes.
1480 			 */
1481 
1482 			/*
1483 			 * Temporarily fake the state to choose a new default
1484 			 * router and to perform on-link determination of
1485 			 * prefixes correctly.
1486 			 * Below the state will be set correctly,
1487 			 * or the entry itself will be deleted.
1488 			 */
1489 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1490 		}
1491 
1492 		if (ln->ln_router || dr) {
1493 			/*
1494 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1495 			 * rnh and for the calls to pfxlist_onlink_check() and
1496 			 * defrouter_select_fib() in the block further down for calls
1497 			 * into nd6_lookup().  We still hold a ref.
1498 			 */
1499 			LLE_WUNLOCK(ln);
1500 
1501 			/*
1502 			 * rt6_flush must be called whether or not the neighbor
1503 			 * is in the Default Router List.
1504 			 * See a corresponding comment in nd6_na_input().
1505 			 */
1506 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1507 		}
1508 
1509 		if (dr) {
1510 			/*
1511 			 * Since defrouter_select_fib() does not affect the
1512 			 * on-link determination and MIP6 needs the check
1513 			 * before the default router selection, we perform
1514 			 * the check now.
1515 			 */
1516 			pfxlist_onlink_check();
1517 
1518 			/*
1519 			 * Refresh default router list.
1520 			 */
1521 			defrouter_select_fib(dr->ifp->if_fib);
1522 		}
1523 
1524 		/*
1525 		 * If this entry was added by an on-link redirect, remove the
1526 		 * corresponding host route.
1527 		 */
1528 		if (ln->la_flags & LLE_REDIRECT)
1529 			nd6_free_redirect(ln);
1530 
1531 		if (ln->ln_router || dr)
1532 			LLE_WLOCK(ln);
1533 	}
1534 
1535 	/*
1536 	 * Save to unlock. We still hold an extra reference and will not
1537 	 * free(9) in llentry_free() if someone else holds one as well.
1538 	 */
1539 	LLE_WUNLOCK(ln);
1540 	IF_AFDATA_LOCK(ifp);
1541 	LLE_WLOCK(ln);
1542 	/* Guard against race with other llentry_free(). */
1543 	if (ln->la_flags & LLE_LINKED) {
1544 		/* Remove callout reference */
1545 		LLE_REMREF(ln);
1546 		lltable_unlink_entry(ln->lle_tbl, ln);
1547 	}
1548 	IF_AFDATA_UNLOCK(ifp);
1549 
1550 	nd6_free_children(ln);
1551 
1552 	llentry_free(ln);
1553 	if (dr != NULL)
1554 		defrouter_rele(dr);
1555 }
1556 
1557 static int
nd6_isdynrte(const struct rtentry * rt,const struct nhop_object * nh,void * xap)1558 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1559 {
1560 
1561 	if (nh->nh_flags & NHF_REDIRECT)
1562 		return (1);
1563 
1564 	return (0);
1565 }
1566 
1567 /*
1568  * Remove the rtentry for the given llentry,
1569  * both of which were installed by a redirect.
1570  */
1571 static void
nd6_free_redirect(const struct llentry * ln)1572 nd6_free_redirect(const struct llentry *ln)
1573 {
1574 	int fibnum;
1575 	struct sockaddr_in6 sin6;
1576 	struct rib_cmd_info rc;
1577 	struct epoch_tracker et;
1578 
1579 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1580 
1581 	NET_EPOCH_ENTER(et);
1582 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1583 		rib_del_route_px(fibnum, (struct sockaddr *)&sin6, 128,
1584 		    nd6_isdynrte, NULL, 0, &rc);
1585 	NET_EPOCH_EXIT(et);
1586 }
1587 
1588 /*
1589  * Updates status of the default router route.
1590  */
1591 static void
check_release_defrouter(const struct rib_cmd_info * rc,void * _cbdata)1592 check_release_defrouter(const struct rib_cmd_info *rc, void *_cbdata)
1593 {
1594 	struct nd_defrouter *dr;
1595 	struct nhop_object *nh;
1596 
1597 	nh = rc->rc_nh_old;
1598 	if (rc->rc_cmd == RTM_DELETE && (nh->nh_flags & NHF_DEFAULT) != 0) {
1599 		dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1600 		if (dr != NULL) {
1601 			dr->installed = 0;
1602 			defrouter_rele(dr);
1603 		}
1604 	}
1605 }
1606 
1607 void
nd6_subscription_cb(struct rib_head * rnh,struct rib_cmd_info * rc,void * arg)1608 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1609 {
1610 #ifdef ROUTE_MPATH
1611 	rib_decompose_notification(rc, check_release_defrouter, NULL);
1612 	if (rc->rc_cmd == RTM_DELETE && !NH_IS_NHGRP(rc->rc_nh_old))
1613 		check_release_defrouter(rc, NULL);
1614 #else
1615 	check_release_defrouter(rc, NULL);
1616 #endif
1617 }
1618 
1619 int
nd6_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp)1620 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1621 {
1622 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1623 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1624 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1625 	struct epoch_tracker et;
1626 	int error = 0;
1627 
1628 	if (ifp->if_afdata[AF_INET6] == NULL)
1629 		return (EPFNOSUPPORT);
1630 	switch (cmd) {
1631 	case OSIOCGIFINFO_IN6:
1632 #define ND	ndi->ndi
1633 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1634 		bzero(&ND, sizeof(ND));
1635 		ND.linkmtu = IN6_LINKMTU(ifp);
1636 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1637 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1638 		ND.reachable = ND_IFINFO(ifp)->reachable;
1639 		ND.retrans = ND_IFINFO(ifp)->retrans;
1640 		ND.flags = ND_IFINFO(ifp)->flags;
1641 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1642 		ND.chlim = ND_IFINFO(ifp)->chlim;
1643 		break;
1644 	case SIOCGIFINFO_IN6:
1645 		ND = *ND_IFINFO(ifp);
1646 		break;
1647 	case SIOCSIFINFO_IN6:
1648 		/*
1649 		 * used to change host variables from userland.
1650 		 * intended for a use on router to reflect RA configurations.
1651 		 */
1652 		/* 0 means 'unspecified' */
1653 		if (ND.linkmtu != 0) {
1654 			if (ND.linkmtu < IPV6_MMTU ||
1655 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1656 				error = EINVAL;
1657 				break;
1658 			}
1659 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1660 		}
1661 
1662 		if (ND.basereachable != 0) {
1663 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1664 
1665 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1666 			if (ND.basereachable != obasereachable)
1667 				ND_IFINFO(ifp)->reachable =
1668 				    ND_COMPUTE_RTIME(ND.basereachable);
1669 		}
1670 		if (ND.retrans != 0)
1671 			ND_IFINFO(ifp)->retrans = ND.retrans;
1672 		if (ND.chlim != 0)
1673 			ND_IFINFO(ifp)->chlim = ND.chlim;
1674 		/* FALLTHROUGH */
1675 	case SIOCSIFINFO_FLAGS:
1676 	{
1677 		struct ifaddr *ifa;
1678 		struct in6_ifaddr *ia;
1679 
1680 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1681 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1682 			/* ifdisabled 1->0 transision */
1683 
1684 			/*
1685 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1686 			 * has an link-local address with IN6_IFF_DUPLICATED,
1687 			 * do not clear ND6_IFF_IFDISABLED.
1688 			 * See RFC 4862, Section 5.4.5.
1689 			 */
1690 			NET_EPOCH_ENTER(et);
1691 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1692 				if (ifa->ifa_addr->sa_family != AF_INET6)
1693 					continue;
1694 				ia = (struct in6_ifaddr *)ifa;
1695 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1696 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1697 					break;
1698 			}
1699 			NET_EPOCH_EXIT(et);
1700 
1701 			if (ifa != NULL) {
1702 				/* LLA is duplicated. */
1703 				ND.flags |= ND6_IFF_IFDISABLED;
1704 				log(LOG_ERR, "Cannot enable an interface"
1705 				    " with a link-local address marked"
1706 				    " duplicate.\n");
1707 			} else {
1708 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1709 				if (ifp->if_flags & IFF_UP)
1710 					in6_if_up(ifp);
1711 			}
1712 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1713 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1714 			/* ifdisabled 0->1 transision */
1715 			/* Mark all IPv6 address as tentative. */
1716 
1717 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1718 			if (V_ip6_dad_count > 0 &&
1719 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1720 				NET_EPOCH_ENTER(et);
1721 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1722 				    ifa_link) {
1723 					if (ifa->ifa_addr->sa_family !=
1724 					    AF_INET6)
1725 						continue;
1726 					ia = (struct in6_ifaddr *)ifa;
1727 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1728 				}
1729 				NET_EPOCH_EXIT(et);
1730 			}
1731 		}
1732 
1733 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1734 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1735 				/* auto_linklocal 0->1 transision */
1736 
1737 				/* If no link-local address on ifp, configure */
1738 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1739 				in6_ifattach(ifp, NULL);
1740 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1741 			    ifp->if_flags & IFF_UP) {
1742 				/*
1743 				 * When the IF already has
1744 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1745 				 * address is assigned, and IFF_UP, try to
1746 				 * assign one.
1747 				 */
1748 				NET_EPOCH_ENTER(et);
1749 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1750 				    ifa_link) {
1751 					if (ifa->ifa_addr->sa_family !=
1752 					    AF_INET6)
1753 						continue;
1754 					ia = (struct in6_ifaddr *)ifa;
1755 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1756 						break;
1757 				}
1758 				NET_EPOCH_EXIT(et);
1759 				if (ifa != NULL)
1760 					/* No LLA is configured. */
1761 					in6_ifattach(ifp, NULL);
1762 			}
1763 		}
1764 		ND_IFINFO(ifp)->flags = ND.flags;
1765 		break;
1766 	}
1767 #undef ND
1768 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1769 		/* sync kernel routing table with the default router list */
1770 		defrouter_reset();
1771 		defrouter_select_fib(RT_ALL_FIBS);
1772 		break;
1773 	case SIOCSPFXFLUSH_IN6:
1774 	{
1775 		/* flush all the prefix advertised by routers */
1776 		struct in6_ifaddr *ia, *ia_next;
1777 		struct nd_prefix *pr, *next;
1778 		struct nd_prhead prl;
1779 
1780 		LIST_INIT(&prl);
1781 
1782 		ND6_WLOCK();
1783 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1784 			if (pr->ndpr_raf_ra_derived)
1785 				nd6_prefix_unlink(pr, &prl);
1786 		}
1787 		ND6_WUNLOCK();
1788 
1789 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1790 			LIST_REMOVE(pr, ndpr_entry);
1791 			/* XXXRW: in6_ifaddrhead locking. */
1792 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1793 			    ia_next) {
1794 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1795 					continue;
1796 
1797 				if (ia->ia6_ndpr == pr)
1798 					in6_purgeaddr(&ia->ia_ifa);
1799 			}
1800 			nd6_prefix_del(pr);
1801 		}
1802 		break;
1803 	}
1804 	case SIOCSRTRFLUSH_IN6:
1805 	{
1806 		/* flush all the default routers */
1807 
1808 		defrouter_reset();
1809 		nd6_defrouter_flush_all();
1810 		defrouter_select_fib(RT_ALL_FIBS);
1811 		break;
1812 	}
1813 	case SIOCGNBRINFO_IN6:
1814 	{
1815 		struct llentry *ln;
1816 		struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1817 
1818 		if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1819 			return (error);
1820 
1821 		NET_EPOCH_ENTER(et);
1822 		ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1823 		NET_EPOCH_EXIT(et);
1824 
1825 		if (ln == NULL) {
1826 			error = EINVAL;
1827 			break;
1828 		}
1829 		nbi->state = ln->ln_state;
1830 		nbi->asked = ln->la_asked;
1831 		nbi->isrouter = ln->ln_router;
1832 		if (ln->la_expire == 0)
1833 			nbi->expire = 0;
1834 		else
1835 			nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1836 			    (time_second - time_uptime);
1837 		LLE_RUNLOCK(ln);
1838 		break;
1839 	}
1840 	case SIOCGDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1841 		ndif->ifindex = V_nd6_defifindex;
1842 		break;
1843 	case SIOCSDEFIFACE_IN6:	/* XXX: should be implemented as a sysctl? */
1844 		return (nd6_setdefaultiface(ndif->ifindex));
1845 	}
1846 	return (error);
1847 }
1848 
1849 /*
1850  * Calculates new isRouter value based on provided parameters and
1851  * returns it.
1852  */
1853 static int
nd6_is_router(int type,int code,int is_new,int old_addr,int new_addr,int ln_router)1854 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1855     int ln_router)
1856 {
1857 
1858 	/*
1859 	 * ICMP6 type dependent behavior.
1860 	 *
1861 	 * NS: clear IsRouter if new entry
1862 	 * RS: clear IsRouter
1863 	 * RA: set IsRouter if there's lladdr
1864 	 * redir: clear IsRouter if new entry
1865 	 *
1866 	 * RA case, (1):
1867 	 * The spec says that we must set IsRouter in the following cases:
1868 	 * - If lladdr exist, set IsRouter.  This means (1-5).
1869 	 * - If it is old entry (!newentry), set IsRouter.  This means (7).
1870 	 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1871 	 * A quetion arises for (1) case.  (1) case has no lladdr in the
1872 	 * neighbor cache, this is similar to (6).
1873 	 * This case is rare but we figured that we MUST NOT set IsRouter.
1874 	 *
1875 	 *   is_new  old_addr new_addr 	    NS  RS  RA	redir
1876 	 *							D R
1877 	 *	0	n	n	(1)	c   ?     s
1878 	 *	0	y	n	(2)	c   s     s
1879 	 *	0	n	y	(3)	c   s     s
1880 	 *	0	y	y	(4)	c   s     s
1881 	 *	0	y	y	(5)	c   s     s
1882 	 *	1	--	n	(6) c	c	c s
1883 	 *	1	--	y	(7) c	c   s	c s
1884 	 *
1885 	 *					(c=clear s=set)
1886 	 */
1887 	switch (type & 0xff) {
1888 	case ND_NEIGHBOR_SOLICIT:
1889 		/*
1890 		 * New entry must have is_router flag cleared.
1891 		 */
1892 		if (is_new)					/* (6-7) */
1893 			ln_router = 0;
1894 		break;
1895 	case ND_REDIRECT:
1896 		/*
1897 		 * If the icmp is a redirect to a better router, always set the
1898 		 * is_router flag.  Otherwise, if the entry is newly created,
1899 		 * clear the flag.  [RFC 2461, sec 8.3]
1900 		 */
1901 		if (code == ND_REDIRECT_ROUTER)
1902 			ln_router = 1;
1903 		else {
1904 			if (is_new)				/* (6-7) */
1905 				ln_router = 0;
1906 		}
1907 		break;
1908 	case ND_ROUTER_SOLICIT:
1909 		/*
1910 		 * is_router flag must always be cleared.
1911 		 */
1912 		ln_router = 0;
1913 		break;
1914 	case ND_ROUTER_ADVERT:
1915 		/*
1916 		 * Mark an entry with lladdr as a router.
1917 		 */
1918 		if ((!is_new && (old_addr || new_addr)) ||	/* (2-5) */
1919 		    (is_new && new_addr)) {			/* (7) */
1920 			ln_router = 1;
1921 		}
1922 		break;
1923 	}
1924 
1925 	return (ln_router);
1926 }
1927 
1928 /*
1929  * Create neighbor cache entry and cache link-layer address,
1930  * on reception of inbound ND6 packets.  (RS/RA/NS/redirect)
1931  *
1932  * type - ICMP6 type
1933  * code - type dependent information
1934  *
1935  */
1936 void
nd6_cache_lladdr(struct ifnet * ifp,struct in6_addr * from,char * lladdr,int lladdrlen,int type,int code)1937 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1938     int lladdrlen, int type, int code)
1939 {
1940 	struct llentry *ln = NULL, *ln_tmp;
1941 	int is_newentry;
1942 	int do_update;
1943 	int olladdr;
1944 	int llchange;
1945 	int flags;
1946 	uint16_t router = 0;
1947 	struct mbuf *chain = NULL;
1948 	u_char linkhdr[LLE_MAX_LINKHDR];
1949 	size_t linkhdrsize;
1950 	int lladdr_off;
1951 
1952 	NET_EPOCH_ASSERT();
1953 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1954 
1955 	KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1956 	KASSERT(from != NULL, ("%s: from == NULL", __func__));
1957 
1958 	/* nothing must be updated for unspecified address */
1959 	if (IN6_IS_ADDR_UNSPECIFIED(from))
1960 		return;
1961 
1962 	/*
1963 	 * Validation about ifp->if_addrlen and lladdrlen must be done in
1964 	 * the caller.
1965 	 *
1966 	 * XXX If the link does not have link-layer adderss, what should
1967 	 * we do? (ifp->if_addrlen == 0)
1968 	 * Spec says nothing in sections for RA, RS and NA.  There's small
1969 	 * description on it in NS section (RFC 2461 7.2.3).
1970 	 */
1971 	flags = lladdr ? LLE_EXCLUSIVE : 0;
1972 	ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
1973 	is_newentry = 0;
1974 	if (ln == NULL) {
1975 		flags |= LLE_EXCLUSIVE;
1976 		ln = nd6_alloc(from, 0, ifp);
1977 		if (ln == NULL)
1978 			return;
1979 
1980 		/*
1981 		 * Since we already know all the data for the new entry,
1982 		 * fill it before insertion.
1983 		 */
1984 		if (lladdr != NULL) {
1985 			linkhdrsize = sizeof(linkhdr);
1986 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
1987 			    linkhdr, &linkhdrsize, &lladdr_off) != 0) {
1988 				lltable_free_entry(LLTABLE6(ifp), ln);
1989 				return;
1990 			}
1991 			lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
1992 			    lladdr_off);
1993 		}
1994 
1995 		IF_AFDATA_WLOCK(ifp);
1996 		LLE_WLOCK(ln);
1997 		/* Prefer any existing lle over newly-created one */
1998 		ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
1999 		if (ln_tmp == NULL)
2000 			lltable_link_entry(LLTABLE6(ifp), ln);
2001 		IF_AFDATA_WUNLOCK(ifp);
2002 		if (ln_tmp == NULL) {
2003 			/* No existing lle, mark as new entry (6,7) */
2004 			is_newentry = 1;
2005 			if (lladdr != NULL) {	/* (7) */
2006 				nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2007 				EVENTHANDLER_INVOKE(lle_event, ln,
2008 				    LLENTRY_RESOLVED);
2009 			}
2010 		} else {
2011 			lltable_free_entry(LLTABLE6(ifp), ln);
2012 			ln = ln_tmp;
2013 			ln_tmp = NULL;
2014 		}
2015 	}
2016 	/* do nothing if static ndp is set */
2017 	if ((ln->la_flags & LLE_STATIC)) {
2018 		if (flags & LLE_EXCLUSIVE)
2019 			LLE_WUNLOCK(ln);
2020 		else
2021 			LLE_RUNLOCK(ln);
2022 		return;
2023 	}
2024 
2025 	olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2026 	if (olladdr && lladdr) {
2027 		llchange = bcmp(lladdr, ln->ll_addr,
2028 		    ifp->if_addrlen);
2029 	} else if (!olladdr && lladdr)
2030 		llchange = 1;
2031 	else
2032 		llchange = 0;
2033 
2034 	/*
2035 	 * newentry olladdr  lladdr  llchange	(*=record)
2036 	 *	0	n	n	--	(1)
2037 	 *	0	y	n	--	(2)
2038 	 *	0	n	y	y	(3) * STALE
2039 	 *	0	y	y	n	(4) *
2040 	 *	0	y	y	y	(5) * STALE
2041 	 *	1	--	n	--	(6)   NOSTATE(= PASSIVE)
2042 	 *	1	--	y	--	(7) * STALE
2043 	 */
2044 
2045 	do_update = 0;
2046 	if (is_newentry == 0 && llchange != 0) {
2047 		do_update = 1;	/* (3,5) */
2048 
2049 		/*
2050 		 * Record source link-layer address
2051 		 * XXX is it dependent to ifp->if_type?
2052 		 */
2053 		if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2054 			/* Entry was deleted */
2055 			LLE_WUNLOCK(ln);
2056 			return;
2057 		}
2058 
2059 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2060 
2061 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2062 
2063 		if (ln->la_hold != NULL)
2064 			chain = nd6_grab_holdchain(ln);
2065 	}
2066 
2067 	/* Calculates new router status */
2068 	router = nd6_is_router(type, code, is_newentry, olladdr,
2069 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2070 
2071 	ln->ln_router = router;
2072 	/* Mark non-router redirects with special flag */
2073 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2074 		ln->la_flags |= LLE_REDIRECT;
2075 
2076 	if (flags & LLE_EXCLUSIVE)
2077 		LLE_WUNLOCK(ln);
2078 	else
2079 		LLE_RUNLOCK(ln);
2080 
2081 	if (chain != NULL)
2082 		nd6_flush_holdchain(ifp, ln, chain);
2083 	if (do_update)
2084 		nd6_flush_children_holdchain(ifp, ln);
2085 
2086 	/*
2087 	 * When the link-layer address of a router changes, select the
2088 	 * best router again.  In particular, when the neighbor entry is newly
2089 	 * created, it might affect the selection policy.
2090 	 * Question: can we restrict the first condition to the "is_newentry"
2091 	 * case?
2092 	 * XXX: when we hear an RA from a new router with the link-layer
2093 	 * address option, defrouter_select_fib() is called twice, since
2094 	 * defrtrlist_update called the function as well.  However, I believe
2095 	 * we can compromise the overhead, since it only happens the first
2096 	 * time.
2097 	 * XXX: although defrouter_select_fib() should not have a bad effect
2098 	 * for those are not autoconfigured hosts, we explicitly avoid such
2099 	 * cases for safety.
2100 	 */
2101 	if ((do_update || is_newentry) && router &&
2102 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2103 		/*
2104 		 * guaranteed recursion
2105 		 */
2106 		defrouter_select_fib(ifp->if_fib);
2107 	}
2108 }
2109 
2110 static void
nd6_slowtimo(void * arg)2111 nd6_slowtimo(void *arg)
2112 {
2113 	struct epoch_tracker et;
2114 	CURVNET_SET((struct vnet *) arg);
2115 	struct nd_ifinfo *nd6if;
2116 	struct ifnet *ifp;
2117 
2118 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2119 	    nd6_slowtimo, curvnet);
2120 	NET_EPOCH_ENTER(et);
2121 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2122 		if (ifp->if_afdata[AF_INET6] == NULL)
2123 			continue;
2124 		nd6if = ND_IFINFO(ifp);
2125 		if (nd6if->basereachable && /* already initialized */
2126 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2127 			/*
2128 			 * Since reachable time rarely changes by router
2129 			 * advertisements, we SHOULD insure that a new random
2130 			 * value gets recomputed at least once every few hours.
2131 			 * (RFC 2461, 6.3.4)
2132 			 */
2133 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2134 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2135 		}
2136 	}
2137 	NET_EPOCH_EXIT(et);
2138 	CURVNET_RESTORE();
2139 }
2140 
2141 struct mbuf *
nd6_grab_holdchain(struct llentry * ln)2142 nd6_grab_holdchain(struct llentry *ln)
2143 {
2144 	struct mbuf *chain;
2145 
2146 	LLE_WLOCK_ASSERT(ln);
2147 
2148 	chain = ln->la_hold;
2149 	ln->la_hold = NULL;
2150 	ln->la_numheld = 0;
2151 
2152 	if (ln->ln_state == ND6_LLINFO_STALE) {
2153 		/*
2154 		 * The first time we send a packet to a
2155 		 * neighbor whose entry is STALE, we have
2156 		 * to change the state to DELAY and a sets
2157 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2158 		 * seconds to ensure do neighbor unreachability
2159 		 * detection on expiration.
2160 		 * (RFC 2461 7.3.3)
2161 		 */
2162 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2163 	}
2164 
2165 	return (chain);
2166 }
2167 
2168 int
nd6_output_ifp(struct ifnet * ifp,struct ifnet * origifp,struct mbuf * m,struct sockaddr_in6 * dst,struct route * ro)2169 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2170     struct sockaddr_in6 *dst, struct route *ro)
2171 {
2172 	int error;
2173 	int ip6len;
2174 	struct ip6_hdr *ip6;
2175 	struct m_tag *mtag;
2176 
2177 #ifdef MAC
2178 	mac_netinet6_nd6_send(ifp, m);
2179 #endif
2180 
2181 	/*
2182 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2183 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2184 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2185 	 * to be diverted to user space.  When re-injected into the kernel,
2186 	 * send_output() will directly dispatch them to the outgoing interface.
2187 	 */
2188 	if (send_sendso_input_hook != NULL) {
2189 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2190 		if (mtag != NULL) {
2191 			ip6 = mtod(m, struct ip6_hdr *);
2192 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2193 			/* Use the SEND socket */
2194 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2195 			    ip6len);
2196 			/* -1 == no app on SEND socket */
2197 			if (error == 0 || error != -1)
2198 			    return (error);
2199 		}
2200 	}
2201 
2202 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2203 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2204 	    mtod(m, struct ip6_hdr *));
2205 
2206 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2207 		origifp = ifp;
2208 
2209 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2210 	return (error);
2211 }
2212 
2213 /*
2214  * Lookup link headerfor @sa_dst address. Stores found
2215  * data in @desten buffer. Copy of lle ln_flags can be also
2216  * saved in @pflags if @pflags is non-NULL.
2217  *
2218  * If destination LLE does not exists or lle state modification
2219  * is required, call "slow" version.
2220  *
2221  * Return values:
2222  * - 0 on success (address copied to buffer).
2223  * - EWOULDBLOCK (no local error, but address is still unresolved)
2224  * - other errors (alloc failure, etc)
2225  */
2226 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)2227 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2228     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2229     struct llentry **plle)
2230 {
2231 	struct llentry *ln = NULL;
2232 	const struct sockaddr_in6 *dst6;
2233 
2234 	NET_EPOCH_ASSERT();
2235 
2236 	if (pflags != NULL)
2237 		*pflags = 0;
2238 
2239 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2240 
2241 	/* discard the packet if IPv6 operation is disabled on the interface */
2242 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2243 		m_freem(m);
2244 		return (ENETDOWN); /* better error? */
2245 	}
2246 
2247 	if (m != NULL && m->m_flags & M_MCAST) {
2248 		switch (ifp->if_type) {
2249 		case IFT_ETHER:
2250 		case IFT_L2VLAN:
2251 		case IFT_BRIDGE:
2252 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2253 						 desten);
2254 			return (0);
2255 		default:
2256 			m_freem(m);
2257 			return (EAFNOSUPPORT);
2258 		}
2259 	}
2260 
2261 	int family = gw_flags >> 16;
2262 	int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2263 	ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2264 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2265 		/* Entry found, let's copy lle info */
2266 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2267 		if (pflags != NULL)
2268 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2269 		llentry_provide_feedback(ln);
2270 		if (plle) {
2271 			LLE_ADDREF(ln);
2272 			*plle = ln;
2273 			LLE_WUNLOCK(ln);
2274 		}
2275 		return (0);
2276 	} else if (plle && ln)
2277 		LLE_WUNLOCK(ln);
2278 
2279 	return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2280 }
2281 
2282 /*
2283  * Finds or creates a new llentry for @addr and @family.
2284  * Returns wlocked llentry or NULL.
2285  *
2286  *
2287  * Child LLEs.
2288  *
2289  * Do not have their own state machine (gets marked as static)
2290  *  settimer bails out for child LLEs just in case.
2291  *
2292  * Locking order: parent lle gets locked first, chen goes the child.
2293  */
2294 static __noinline struct llentry *
nd6_get_llentry(struct ifnet * ifp,const struct in6_addr * addr,int family)2295 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2296 {
2297 	struct llentry *child_lle = NULL;
2298 	struct llentry *lle, *lle_tmp;
2299 
2300 	lle = nd6_alloc(addr, 0, ifp);
2301 	if (lle != NULL && family != AF_INET6) {
2302 		child_lle = nd6_alloc(addr, 0, ifp);
2303 		if (child_lle == NULL) {
2304 			lltable_free_entry(LLTABLE6(ifp), lle);
2305 			return (NULL);
2306 		}
2307 		child_lle->r_family = family;
2308 		child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2309 		child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2310 	}
2311 
2312 	if (lle == NULL) {
2313 		char ip6buf[INET6_ADDRSTRLEN];
2314 		log(LOG_DEBUG,
2315 		    "nd6_get_llentry: can't allocate llinfo for %s "
2316 		    "(ln=%p)\n",
2317 		    ip6_sprintf(ip6buf, addr), lle);
2318 		return (NULL);
2319 	}
2320 
2321 	IF_AFDATA_WLOCK(ifp);
2322 	LLE_WLOCK(lle);
2323 	/* Prefer any existing entry over newly-created one */
2324 	lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2325 	if (lle_tmp == NULL)
2326 		lltable_link_entry(LLTABLE6(ifp), lle);
2327 	else {
2328 		lltable_free_entry(LLTABLE6(ifp), lle);
2329 		lle = lle_tmp;
2330 	}
2331 	if (child_lle != NULL) {
2332 		/* Check if child lle for the same family exists */
2333 		lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2334 		LLE_WLOCK(child_lle);
2335 		if (lle_tmp == NULL) {
2336 			/* Attach */
2337 			lltable_link_child_entry(lle, child_lle);
2338 		} else {
2339 			/* child lle already exists, free newly-created one */
2340 			lltable_free_entry(LLTABLE6(ifp), child_lle);
2341 			LLE_WLOCK(lle_tmp);
2342 			child_lle = lle_tmp;
2343 		}
2344 		LLE_WUNLOCK(lle);
2345 		lle = child_lle;
2346 	}
2347 	IF_AFDATA_WUNLOCK(ifp);
2348 	return (lle);
2349 }
2350 
2351 /*
2352  * Do L2 address resolution for @sa_dst address. Stores found
2353  * address in @desten buffer. Copy of lle ln_flags can be also
2354  * saved in @pflags if @pflags is non-NULL.
2355  *
2356  * Heavy version.
2357  * Function assume that destination LLE does not exist,
2358  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2359  *
2360  * Set noinline to be dtrace-friendly
2361  */
2362 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)2363 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2364     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2365     struct llentry **plle)
2366 {
2367 	struct llentry *lle = NULL;
2368 	struct in6_addr *psrc, src;
2369 	int send_ns, ll_len;
2370 	char *lladdr;
2371 
2372 	NET_EPOCH_ASSERT();
2373 
2374 	/*
2375 	 * Address resolution or Neighbor Unreachability Detection
2376 	 * for the next hop.
2377 	 * At this point, the destination of the packet must be a unicast
2378 	 * or an anycast address(i.e. not a multicast).
2379 	 */
2380 	lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2381 	if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2382 		/*
2383 		 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2384 		 * the condition below is not very efficient.  But we believe
2385 		 * it is tolerable, because this should be a rare case.
2386 		 */
2387 		lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2388 	}
2389 
2390 	if (lle == NULL) {
2391 		m_freem(m);
2392 		return (ENOBUFS);
2393 	}
2394 
2395 	LLE_WLOCK_ASSERT(lle);
2396 
2397 	/*
2398 	 * The first time we send a packet to a neighbor whose entry is
2399 	 * STALE, we have to change the state to DELAY and a sets a timer to
2400 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2401 	 * neighbor unreachability detection on expiration.
2402 	 * (RFC 2461 7.3.3)
2403 	 */
2404 	if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2405 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2406 
2407 	/*
2408 	 * If the neighbor cache entry has a state other than INCOMPLETE
2409 	 * (i.e. its link-layer address is already resolved), just
2410 	 * send the packet.
2411 	 */
2412 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2413 		if (flags & LLE_ADDRONLY) {
2414 			lladdr = lle->ll_addr;
2415 			ll_len = ifp->if_addrlen;
2416 		} else {
2417 			lladdr = lle->r_linkdata;
2418 			ll_len = lle->r_hdrlen;
2419 		}
2420 		bcopy(lladdr, desten, ll_len);
2421 		if (pflags != NULL)
2422 			*pflags = lle->la_flags;
2423 		if (plle) {
2424 			LLE_ADDREF(lle);
2425 			*plle = lle;
2426 		}
2427 		LLE_WUNLOCK(lle);
2428 		return (0);
2429 	}
2430 
2431 	/*
2432 	 * There is a neighbor cache entry, but no ethernet address
2433 	 * response yet.  Append this latest packet to the end of the
2434 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2435 	 * the oldest packet in the queue will be removed.
2436 	 */
2437 	if (m != NULL) {
2438 		size_t dropped;
2439 
2440 		dropped = lltable_append_entry_queue(lle, m, V_nd6_maxqueuelen);
2441 		ICMP6STAT_ADD(icp6s_dropped, dropped);
2442 	}
2443 
2444 	/*
2445 	 * If there has been no NS for the neighbor after entering the
2446 	 * INCOMPLETE state, send the first solicitation.
2447 	 * Note that for newly-created lle la_asked will be 0,
2448 	 * so we will transition from ND6_LLINFO_NOSTATE to
2449 	 * ND6_LLINFO_INCOMPLETE state here.
2450 	 */
2451 	psrc = NULL;
2452 	send_ns = 0;
2453 
2454 	/* If we have child lle, switch to the parent to send NS */
2455 	if (lle->la_flags & LLE_CHILD) {
2456 		struct llentry *lle_parent = lle->lle_parent;
2457 		LLE_WUNLOCK(lle);
2458 		lle = lle_parent;
2459 		LLE_WLOCK(lle);
2460 	}
2461 	if (lle->la_asked == 0) {
2462 		lle->la_asked++;
2463 		send_ns = 1;
2464 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2465 
2466 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2467 	}
2468 	LLE_WUNLOCK(lle);
2469 	if (send_ns != 0)
2470 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2471 
2472 	return (EWOULDBLOCK);
2473 }
2474 
2475 /*
2476  * Do L2 address resolution for @sa_dst address. Stores found
2477  * address in @desten buffer. Copy of lle ln_flags can be also
2478  * saved in @pflags if @pflags is non-NULL.
2479  *
2480  * Return values:
2481  * - 0 on success (address copied to buffer).
2482  * - EWOULDBLOCK (no local error, but address is still unresolved)
2483  * - other errors (alloc failure, etc)
2484  */
2485 int
nd6_resolve_addr(struct ifnet * ifp,int flags,const struct sockaddr * dst,char * desten,uint32_t * pflags)2486 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2487     char *desten, uint32_t *pflags)
2488 {
2489 	int error;
2490 
2491 	flags |= LLE_ADDRONLY;
2492 	error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2493 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2494 	return (error);
2495 }
2496 
2497 int
nd6_flush_holdchain(struct ifnet * ifp,struct llentry * lle,struct mbuf * chain)2498 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2499 {
2500 	struct mbuf *m, *m_head;
2501 	struct sockaddr_in6 dst6;
2502 	int error = 0;
2503 
2504 	NET_EPOCH_ASSERT();
2505 
2506 	struct route_in6 ro = {
2507 		.ro_prepend = lle->r_linkdata,
2508 		.ro_plen = lle->r_hdrlen,
2509 	};
2510 
2511 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2512 	m_head = chain;
2513 
2514 	while (m_head) {
2515 		m = m_head;
2516 		m_head = m_head->m_nextpkt;
2517 		m->m_nextpkt = NULL;
2518 		error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2519 	}
2520 
2521 	/*
2522 	 * XXX
2523 	 * note that intermediate errors are blindly ignored
2524 	 */
2525 	return (error);
2526 }
2527 
2528 __noinline void
nd6_flush_children_holdchain(struct ifnet * ifp,struct llentry * lle)2529 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2530 {
2531 	struct llentry *child_lle;
2532 	struct mbuf *chain;
2533 
2534 	NET_EPOCH_ASSERT();
2535 
2536 	CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2537 		LLE_WLOCK(child_lle);
2538 		chain = nd6_grab_holdchain(child_lle);
2539 		LLE_WUNLOCK(child_lle);
2540 		nd6_flush_holdchain(ifp, child_lle, chain);
2541 	}
2542 }
2543 
2544 static int
nd6_need_cache(struct ifnet * ifp)2545 nd6_need_cache(struct ifnet *ifp)
2546 {
2547 	/*
2548 	 * XXX: we currently do not make neighbor cache on any interface
2549 	 * other than Ethernet and GIF.
2550 	 *
2551 	 * RFC2893 says:
2552 	 * - unidirectional tunnels needs no ND
2553 	 */
2554 	switch (ifp->if_type) {
2555 	case IFT_ETHER:
2556 	case IFT_IEEE1394:
2557 	case IFT_L2VLAN:
2558 	case IFT_INFINIBAND:
2559 	case IFT_BRIDGE:
2560 	case IFT_PROPVIRTUAL:
2561 		return (1);
2562 	default:
2563 		return (0);
2564 	}
2565 }
2566 
2567 /*
2568  * Add pernament ND6 link-layer record for given
2569  * interface address.
2570  *
2571  * Very similar to IPv4 arp_ifinit(), but:
2572  * 1) IPv6 DAD is performed in different place
2573  * 2) It is called by IPv6 protocol stack in contrast to
2574  * arp_ifinit() which is typically called in SIOCSIFADDR
2575  * driver ioctl handler.
2576  *
2577  */
2578 int
nd6_add_ifa_lle(struct in6_ifaddr * ia)2579 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2580 {
2581 	struct ifnet *ifp;
2582 	struct llentry *ln, *ln_tmp;
2583 	struct sockaddr *dst;
2584 
2585 	ifp = ia->ia_ifa.ifa_ifp;
2586 	if (nd6_need_cache(ifp) == 0)
2587 		return (0);
2588 
2589 	dst = (struct sockaddr *)&ia->ia_addr;
2590 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2591 	if (ln == NULL)
2592 		return (ENOBUFS);
2593 
2594 	IF_AFDATA_WLOCK(ifp);
2595 	LLE_WLOCK(ln);
2596 	/* Unlink any entry if exists */
2597 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2598 	if (ln_tmp != NULL)
2599 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2600 	lltable_link_entry(LLTABLE6(ifp), ln);
2601 	IF_AFDATA_WUNLOCK(ifp);
2602 
2603 	if (ln_tmp != NULL)
2604 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2605 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2606 
2607 	LLE_WUNLOCK(ln);
2608 	if (ln_tmp != NULL)
2609 		llentry_free(ln_tmp);
2610 
2611 	return (0);
2612 }
2613 
2614 /*
2615  * Removes either all lle entries for given @ia, or lle
2616  * corresponding to @ia address.
2617  */
2618 void
nd6_rem_ifa_lle(struct in6_ifaddr * ia,int all)2619 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2620 {
2621 	struct sockaddr_in6 mask, addr;
2622 	struct sockaddr *saddr, *smask;
2623 	struct ifnet *ifp;
2624 
2625 	ifp = ia->ia_ifa.ifa_ifp;
2626 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2627 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2628 	saddr = (struct sockaddr *)&addr;
2629 	smask = (struct sockaddr *)&mask;
2630 
2631 	if (all != 0)
2632 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2633 	else
2634 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2635 }
2636 
2637 static int
nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)2638 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2639 {
2640 	struct in6_prefix p;
2641 	struct sockaddr_in6 s6;
2642 	struct nd_prefix *pr;
2643 	struct nd_pfxrouter *pfr;
2644 	time_t maxexpire;
2645 	int error;
2646 	char ip6buf[INET6_ADDRSTRLEN];
2647 
2648 	if (req->newptr)
2649 		return (EPERM);
2650 
2651 	error = sysctl_wire_old_buffer(req, 0);
2652 	if (error != 0)
2653 		return (error);
2654 
2655 	bzero(&p, sizeof(p));
2656 	p.origin = PR_ORIG_RA;
2657 	bzero(&s6, sizeof(s6));
2658 	s6.sin6_family = AF_INET6;
2659 	s6.sin6_len = sizeof(s6);
2660 
2661 	ND6_RLOCK();
2662 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2663 		if (!pr->ndpr_raf_ra_derived)
2664 			continue;
2665 		p.prefix = pr->ndpr_prefix;
2666 		if (sa6_recoverscope(&p.prefix)) {
2667 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2668 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2669 			/* XXX: press on... */
2670 		}
2671 		p.raflags = pr->ndpr_raf;
2672 		p.prefixlen = pr->ndpr_plen;
2673 		p.vltime = pr->ndpr_vltime;
2674 		p.pltime = pr->ndpr_pltime;
2675 		p.if_index = pr->ndpr_ifp->if_index;
2676 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2677 			p.expire = 0;
2678 		else {
2679 			/* XXX: we assume time_t is signed. */
2680 			maxexpire = (-1) &
2681 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2682 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2683 				p.expire = pr->ndpr_lastupdate +
2684 				    pr->ndpr_vltime +
2685 				    (time_second - time_uptime);
2686 			else
2687 				p.expire = maxexpire;
2688 		}
2689 		p.refcnt = pr->ndpr_addrcnt;
2690 		p.flags = pr->ndpr_stateflags;
2691 		p.advrtrs = 0;
2692 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2693 			p.advrtrs++;
2694 		error = SYSCTL_OUT(req, &p, sizeof(p));
2695 		if (error != 0)
2696 			break;
2697 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2698 			s6.sin6_addr = pfr->router->rtaddr;
2699 			if (sa6_recoverscope(&s6))
2700 				log(LOG_ERR,
2701 				    "scope error in prefix list (%s)\n",
2702 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2703 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2704 			if (error != 0)
2705 				goto out;
2706 		}
2707 	}
2708 out:
2709 	ND6_RUNLOCK();
2710 	return (error);
2711 }
2712 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2713 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2714 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2715 	"NDP prefix list");
2716 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2717 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2718 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2719 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2720