xref: /freebsd-12.1/sys/netinet6/nd6.c (revision db6726b5)
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 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.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/vnet.h>
65 
66 #include <netinet/in.h>
67 #include <netinet/in_kdtrace.h>
68 #include <net/if_llatbl.h>
69 #include <netinet/if_ether.h>
70 #include <netinet6/in6_var.h>
71 #include <netinet/ip6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet6/scope6_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/in6_ifattach.h>
76 #include <netinet/icmp6.h>
77 #include <netinet6/send.h>
78 
79 #include <sys/limits.h>
80 
81 #include <security/mac/mac_framework.h>
82 
83 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
84 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
85 
86 #define SIN6(s) ((const struct sockaddr_in6 *)(s))
87 
88 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
89 
90 /* timer values */
91 VNET_DEFINE(int, nd6_prune)	= 1;	/* walk list every 1 seconds */
92 VNET_DEFINE(int, nd6_delay)	= 5;	/* delay first probe time 5 second */
93 VNET_DEFINE(int, nd6_umaxtries)	= 3;	/* maximum unicast query */
94 VNET_DEFINE(int, nd6_mmaxtries)	= 3;	/* maximum multicast query */
95 VNET_DEFINE(int, nd6_useloopback) = 1;	/* use loopback interface for
96 					 * local traffic */
97 VNET_DEFINE(int, nd6_gctimer)	= (60 * 60 * 24); /* 1 day: garbage
98 					 * collection timer */
99 
100 /* preventing too many loops in ND option parsing */
101 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
102 
103 VNET_DEFINE(int, nd6_maxnudhint) = 0;	/* max # of subsequent upper
104 					 * layer hints */
105 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved
106 					 * ND entries */
107 #define	V_nd6_maxndopt			VNET(nd6_maxndopt)
108 #define	V_nd6_maxqueuelen		VNET(nd6_maxqueuelen)
109 
110 #ifdef ND6_DEBUG
111 VNET_DEFINE(int, nd6_debug) = 1;
112 #else
113 VNET_DEFINE(int, nd6_debug) = 0;
114 #endif
115 
116 static eventhandler_tag lle_event_eh, iflladdr_event_eh;
117 
118 VNET_DEFINE(struct nd_drhead, nd_defrouter);
119 VNET_DEFINE(struct nd_prhead, nd_prefix);
120 VNET_DEFINE(struct rwlock, nd6_lock);
121 VNET_DEFINE(uint64_t, nd6_list_genid);
122 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
123 
124 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
125 #define	V_nd6_recalc_reachtm_interval	VNET(nd6_recalc_reachtm_interval)
126 
127 int	(*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
128 
129 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
130 	struct ifnet *);
131 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
132 static void nd6_slowtimo(void *);
133 static int regen_tmpaddr(struct in6_ifaddr *);
134 static void nd6_free(struct llentry **, int);
135 static void nd6_free_redirect(const struct llentry *);
136 static void nd6_llinfo_timer(void *);
137 static void nd6_llinfo_settimer_locked(struct llentry *, long);
138 static void clear_llinfo_pqueue(struct llentry *);
139 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *);
140 static int nd6_resolve_slow(struct ifnet *, int, struct mbuf *,
141     const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
142 static int nd6_need_cache(struct ifnet *);
143 
144 
145 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
146 #define	V_nd6_slowtimo_ch		VNET(nd6_slowtimo_ch)
147 
148 VNET_DEFINE(struct callout, nd6_timer_ch);
149 #define	V_nd6_timer_ch			VNET(nd6_timer_ch)
150 
151 static void
nd6_lle_event(void * arg __unused,struct llentry * lle,int evt)152 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
153 {
154 	struct rt_addrinfo rtinfo;
155 	struct sockaddr_in6 dst;
156 	struct sockaddr_dl gw;
157 	struct ifnet *ifp;
158 	int type;
159 	int fibnum;
160 
161 	LLE_WLOCK_ASSERT(lle);
162 
163 	if (lltable_get_af(lle->lle_tbl) != AF_INET6)
164 		return;
165 
166 	switch (evt) {
167 	case LLENTRY_RESOLVED:
168 		type = RTM_ADD;
169 		KASSERT(lle->la_flags & LLE_VALID,
170 		    ("%s: %p resolved but not valid?", __func__, lle));
171 		break;
172 	case LLENTRY_EXPIRED:
173 		type = RTM_DELETE;
174 		break;
175 	default:
176 		return;
177 	}
178 
179 	ifp = lltable_get_ifp(lle->lle_tbl);
180 
181 	bzero(&dst, sizeof(dst));
182 	bzero(&gw, sizeof(gw));
183 	bzero(&rtinfo, sizeof(rtinfo));
184 	lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
185 	dst.sin6_scope_id = in6_getscopezone(ifp,
186 	    in6_addrscope(&dst.sin6_addr));
187 	gw.sdl_len = sizeof(struct sockaddr_dl);
188 	gw.sdl_family = AF_LINK;
189 	gw.sdl_alen = ifp->if_addrlen;
190 	gw.sdl_index = ifp->if_index;
191 	gw.sdl_type = ifp->if_type;
192 	if (evt == LLENTRY_RESOLVED)
193 		bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
194 	rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
195 	rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
196 	rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
197 	fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
198 	rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
199 	    type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
200 }
201 
202 /*
203  * A handler for interface link layer address change event.
204  */
205 static void
nd6_iflladdr(void * arg __unused,struct ifnet * ifp)206 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
207 {
208 
209 	lltable_update_ifaddr(LLTABLE6(ifp));
210 }
211 
212 void
nd6_init(void)213 nd6_init(void)
214 {
215 
216 	mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
217 	rw_init(&V_nd6_lock, "nd6 list");
218 
219 	LIST_INIT(&V_nd_prefix);
220 	TAILQ_INIT(&V_nd_defrouter);
221 
222 	/* Start timers. */
223 	callout_init(&V_nd6_slowtimo_ch, 0);
224 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
225 	    nd6_slowtimo, curvnet);
226 
227 	callout_init(&V_nd6_timer_ch, 0);
228 	callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
229 
230 	nd6_dad_init();
231 	if (IS_DEFAULT_VNET(curvnet)) {
232 		lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
233 		    NULL, EVENTHANDLER_PRI_ANY);
234 		iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
235 		    nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
236 	}
237 }
238 
239 #ifdef VIMAGE
240 void
nd6_destroy()241 nd6_destroy()
242 {
243 
244 	callout_drain(&V_nd6_slowtimo_ch);
245 	callout_drain(&V_nd6_timer_ch);
246 	if (IS_DEFAULT_VNET(curvnet)) {
247 		EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
248 		EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
249 	}
250 	rw_destroy(&V_nd6_lock);
251 	mtx_destroy(&V_nd6_onlink_mtx);
252 }
253 #endif
254 
255 struct nd_ifinfo *
nd6_ifattach(struct ifnet * ifp)256 nd6_ifattach(struct ifnet *ifp)
257 {
258 	struct nd_ifinfo *nd;
259 
260 	nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
261 	nd->initialized = 1;
262 
263 	nd->chlim = IPV6_DEFHLIM;
264 	nd->basereachable = REACHABLE_TIME;
265 	nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
266 	nd->retrans = RETRANS_TIMER;
267 
268 	nd->flags = ND6_IFF_PERFORMNUD;
269 
270 	/* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
271 	 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
272 	 * default regardless of the V_ip6_auto_linklocal configuration to
273 	 * give a reasonable default behavior.
274 	 */
275 	if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) ||
276 	    (ifp->if_flags & IFF_LOOPBACK))
277 		nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
278 	/*
279 	 * A loopback interface does not need to accept RTADV.
280 	 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
281 	 * default regardless of the V_ip6_accept_rtadv configuration to
282 	 * prevent the interface from accepting RA messages arrived
283 	 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
284 	 */
285 	if (V_ip6_accept_rtadv &&
286 	    !(ifp->if_flags & IFF_LOOPBACK) &&
287 	    (ifp->if_type != IFT_BRIDGE))
288 			nd->flags |= ND6_IFF_ACCEPT_RTADV;
289 	if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
290 		nd->flags |= ND6_IFF_NO_RADR;
291 
292 	/* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
293 	nd6_setmtu0(ifp, nd);
294 
295 	return nd;
296 }
297 
298 void
nd6_ifdetach(struct ifnet * ifp,struct nd_ifinfo * nd)299 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
300 {
301 	struct ifaddr *ifa, *next;
302 
303 	IF_ADDR_RLOCK(ifp);
304 	CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
305 		if (ifa->ifa_addr->sa_family != AF_INET6)
306 			continue;
307 
308 		/* stop DAD processing */
309 		nd6_dad_stop(ifa);
310 	}
311 	IF_ADDR_RUNLOCK(ifp);
312 
313 	free(nd, M_IP6NDP);
314 }
315 
316 /*
317  * Reset ND level link MTU. This function is called when the physical MTU
318  * changes, which means we might have to adjust the ND level MTU.
319  */
320 void
nd6_setmtu(struct ifnet * ifp)321 nd6_setmtu(struct ifnet *ifp)
322 {
323 	if (ifp->if_afdata[AF_INET6] == NULL)
324 		return;
325 
326 	nd6_setmtu0(ifp, ND_IFINFO(ifp));
327 }
328 
329 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
330 void
nd6_setmtu0(struct ifnet * ifp,struct nd_ifinfo * ndi)331 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
332 {
333 	u_int32_t omaxmtu;
334 
335 	omaxmtu = ndi->maxmtu;
336 	ndi->maxmtu = ifp->if_mtu;
337 
338 	/*
339 	 * Decreasing the interface MTU under IPV6 minimum MTU may cause
340 	 * undesirable situation.  We thus notify the operator of the change
341 	 * explicitly.  The check for omaxmtu is necessary to restrict the
342 	 * log to the case of changing the MTU, not initializing it.
343 	 */
344 	if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
345 		log(LOG_NOTICE, "nd6_setmtu0: "
346 		    "new link MTU on %s (%lu) is too small for IPv6\n",
347 		    if_name(ifp), (unsigned long)ndi->maxmtu);
348 	}
349 
350 	if (ndi->maxmtu > V_in6_maxmtu)
351 		in6_setmaxmtu(); /* check all interfaces just in case */
352 
353 }
354 
355 void
nd6_option_init(void * opt,int icmp6len,union nd_opts * ndopts)356 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
357 {
358 
359 	bzero(ndopts, sizeof(*ndopts));
360 	ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
361 	ndopts->nd_opts_last
362 		= (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
363 
364 	if (icmp6len == 0) {
365 		ndopts->nd_opts_done = 1;
366 		ndopts->nd_opts_search = NULL;
367 	}
368 }
369 
370 /*
371  * Take one ND option.
372  */
373 struct nd_opt_hdr *
nd6_option(union nd_opts * ndopts)374 nd6_option(union nd_opts *ndopts)
375 {
376 	struct nd_opt_hdr *nd_opt;
377 	int olen;
378 
379 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
380 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
381 	    __func__));
382 	if (ndopts->nd_opts_search == NULL)
383 		return NULL;
384 	if (ndopts->nd_opts_done)
385 		return NULL;
386 
387 	nd_opt = ndopts->nd_opts_search;
388 
389 	/* make sure nd_opt_len is inside the buffer */
390 	if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
391 		bzero(ndopts, sizeof(*ndopts));
392 		return NULL;
393 	}
394 
395 	olen = nd_opt->nd_opt_len << 3;
396 	if (olen == 0) {
397 		/*
398 		 * Message validation requires that all included
399 		 * options have a length that is greater than zero.
400 		 */
401 		bzero(ndopts, sizeof(*ndopts));
402 		return NULL;
403 	}
404 
405 	ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
406 	if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
407 		/* option overruns the end of buffer, invalid */
408 		bzero(ndopts, sizeof(*ndopts));
409 		return NULL;
410 	} else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
411 		/* reached the end of options chain */
412 		ndopts->nd_opts_done = 1;
413 		ndopts->nd_opts_search = NULL;
414 	}
415 	return nd_opt;
416 }
417 
418 /*
419  * Parse multiple ND options.
420  * This function is much easier to use, for ND routines that do not need
421  * multiple options of the same type.
422  */
423 int
nd6_options(union nd_opts * ndopts)424 nd6_options(union nd_opts *ndopts)
425 {
426 	struct nd_opt_hdr *nd_opt;
427 	int i = 0;
428 
429 	KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
430 	KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
431 	    __func__));
432 	if (ndopts->nd_opts_search == NULL)
433 		return 0;
434 
435 	while (1) {
436 		nd_opt = nd6_option(ndopts);
437 		if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
438 			/*
439 			 * Message validation requires that all included
440 			 * options have a length that is greater than zero.
441 			 */
442 			ICMP6STAT_INC(icp6s_nd_badopt);
443 			bzero(ndopts, sizeof(*ndopts));
444 			return -1;
445 		}
446 
447 		if (nd_opt == NULL)
448 			goto skip1;
449 
450 		switch (nd_opt->nd_opt_type) {
451 		case ND_OPT_SOURCE_LINKADDR:
452 		case ND_OPT_TARGET_LINKADDR:
453 		case ND_OPT_MTU:
454 		case ND_OPT_REDIRECTED_HEADER:
455 		case ND_OPT_NONCE:
456 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
457 				nd6log((LOG_INFO,
458 				    "duplicated ND6 option found (type=%d)\n",
459 				    nd_opt->nd_opt_type));
460 				/* XXX bark? */
461 			} else {
462 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
463 					= nd_opt;
464 			}
465 			break;
466 		case ND_OPT_PREFIX_INFORMATION:
467 			if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
468 				ndopts->nd_opt_array[nd_opt->nd_opt_type]
469 					= nd_opt;
470 			}
471 			ndopts->nd_opts_pi_end =
472 				(struct nd_opt_prefix_info *)nd_opt;
473 			break;
474 		/* What about ND_OPT_ROUTE_INFO? RFC 4191 */
475 		case ND_OPT_RDNSS:	/* RFC 6106 */
476 		case ND_OPT_DNSSL:	/* RFC 6106 */
477 			/*
478 			 * Silently ignore options we know and do not care about
479 			 * in the kernel.
480 			 */
481 			break;
482 		default:
483 			/*
484 			 * Unknown options must be silently ignored,
485 			 * to accommodate future extension to the protocol.
486 			 */
487 			nd6log((LOG_DEBUG,
488 			    "nd6_options: unsupported option %d - "
489 			    "option ignored\n", nd_opt->nd_opt_type));
490 		}
491 
492 skip1:
493 		i++;
494 		if (i > V_nd6_maxndopt) {
495 			ICMP6STAT_INC(icp6s_nd_toomanyopt);
496 			nd6log((LOG_INFO, "too many loop in nd opt\n"));
497 			break;
498 		}
499 
500 		if (ndopts->nd_opts_done)
501 			break;
502 	}
503 
504 	return 0;
505 }
506 
507 /*
508  * ND6 timer routine to handle ND6 entries
509  */
510 static void
nd6_llinfo_settimer_locked(struct llentry * ln,long tick)511 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
512 {
513 	int canceled;
514 
515 	LLE_WLOCK_ASSERT(ln);
516 
517 	if (tick < 0) {
518 		ln->la_expire = 0;
519 		ln->ln_ntick = 0;
520 		canceled = callout_stop(&ln->lle_timer);
521 	} else {
522 		ln->la_expire = time_uptime + tick / hz;
523 		LLE_ADDREF(ln);
524 		if (tick > INT_MAX) {
525 			ln->ln_ntick = tick - INT_MAX;
526 			canceled = callout_reset(&ln->lle_timer, INT_MAX,
527 			    nd6_llinfo_timer, ln);
528 		} else {
529 			ln->ln_ntick = 0;
530 			canceled = callout_reset(&ln->lle_timer, tick,
531 			    nd6_llinfo_timer, ln);
532 		}
533 	}
534 	if (canceled > 0)
535 		LLE_REMREF(ln);
536 }
537 
538 /*
539  * Gets source address of the first packet in hold queue
540  * and stores it in @src.
541  * Returns pointer to @src (if hold queue is not empty) or NULL.
542  *
543  * Set noinline to be dtrace-friendly
544  */
545 static __noinline struct in6_addr *
nd6_llinfo_get_holdsrc(struct llentry * ln,struct in6_addr * src)546 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
547 {
548 	struct ip6_hdr hdr;
549 	struct mbuf *m;
550 
551 	if (ln->la_hold == NULL)
552 		return (NULL);
553 
554 	/*
555 	 * assume every packet in la_hold has the same IP header
556 	 */
557 	m = ln->la_hold;
558 	if (sizeof(hdr) > m->m_len)
559 		return (NULL);
560 
561 	m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
562 	*src = hdr.ip6_src;
563 
564 	return (src);
565 }
566 
567 /*
568  * Checks if we need to switch from STALE state.
569  *
570  * RFC 4861 requires switching from STALE to DELAY state
571  * on first packet matching entry, waiting V_nd6_delay and
572  * transition to PROBE state (if upper layer confirmation was
573  * not received).
574  *
575  * This code performs a bit differently:
576  * On packet hit we don't change state (but desired state
577  * can be guessed by control plane). However, after V_nd6_delay
578  * seconds code will transition to PROBE state (so DELAY state
579  * is kinda skipped in most situations).
580  *
581  * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
582  * we perform the following upon entering STALE state:
583  *
584  * 1) Arm timer to run each V_nd6_delay seconds to make sure that
585  * if packet was transmitted at the start of given interval, we
586  * would be able to switch to PROBE state in V_nd6_delay seconds
587  * as user expects.
588  *
589  * 2) Reschedule timer until original V_nd6_gctimer expires keeping
590  * lle in STALE state (remaining timer value stored in lle_remtime).
591  *
592  * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
593  * seconds ago.
594  *
595  * Returns non-zero value if the entry is still STALE (storing
596  * the next timer interval in @pdelay).
597  *
598  * Returns zero value if original timer expired or we need to switch to
599  * PROBE (store that in @do_switch variable).
600  */
601 static int
nd6_is_stale(struct llentry * lle,long * pdelay,int * do_switch)602 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
603 {
604 	int nd_delay, nd_gctimer, r_skip_req;
605 	time_t lle_hittime;
606 	long delay;
607 
608 	*do_switch = 0;
609 	nd_gctimer = V_nd6_gctimer;
610 	nd_delay = V_nd6_delay;
611 
612 	LLE_REQ_LOCK(lle);
613 	r_skip_req = lle->r_skip_req;
614 	lle_hittime = lle->lle_hittime;
615 	LLE_REQ_UNLOCK(lle);
616 
617 	if (r_skip_req > 0) {
618 
619 		/*
620 		 * Nonzero r_skip_req value was set upon entering
621 		 * STALE state. Since value was not changed, no
622 		 * packets were passed using this lle. Ask for
623 		 * timer reschedule and keep STALE state.
624 		 */
625 		delay = (long)(MIN(nd_gctimer, nd_delay));
626 		delay *= hz;
627 		if (lle->lle_remtime > delay)
628 			lle->lle_remtime -= delay;
629 		else {
630 			delay = lle->lle_remtime;
631 			lle->lle_remtime = 0;
632 		}
633 
634 		if (delay == 0) {
635 
636 			/*
637 			 * The original ng6_gctime timeout ended,
638 			 * no more rescheduling.
639 			 */
640 			return (0);
641 		}
642 
643 		*pdelay = delay;
644 		return (1);
645 	}
646 
647 	/*
648 	 * Packet received. Verify timestamp
649 	 */
650 	delay = (long)(time_uptime - lle_hittime);
651 	if (delay < nd_delay) {
652 
653 		/*
654 		 * V_nd6_delay still not passed since the first
655 		 * hit in STALE state.
656 		 * Reshedule timer and return.
657 		 */
658 		*pdelay = (long)(nd_delay - delay) * hz;
659 		return (1);
660 	}
661 
662 	/* Request switching to probe */
663 	*do_switch = 1;
664 	return (0);
665 }
666 
667 
668 /*
669  * Switch @lle state to new state optionally arming timers.
670  *
671  * Set noinline to be dtrace-friendly
672  */
673 __noinline void
nd6_llinfo_setstate(struct llentry * lle,int newstate)674 nd6_llinfo_setstate(struct llentry *lle, int newstate)
675 {
676 	struct ifnet *ifp;
677 	int nd_gctimer, nd_delay;
678 	long delay, remtime;
679 
680 	delay = 0;
681 	remtime = 0;
682 
683 	switch (newstate) {
684 	case ND6_LLINFO_INCOMPLETE:
685 		ifp = lle->lle_tbl->llt_ifp;
686 		delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
687 		break;
688 	case ND6_LLINFO_REACHABLE:
689 		if (!ND6_LLINFO_PERMANENT(lle)) {
690 			ifp = lle->lle_tbl->llt_ifp;
691 			delay = (long)ND_IFINFO(ifp)->reachable * hz;
692 		}
693 		break;
694 	case ND6_LLINFO_STALE:
695 
696 		/*
697 		 * Notify fast path that we want to know if any packet
698 		 * is transmitted by setting r_skip_req.
699 		 */
700 		LLE_REQ_LOCK(lle);
701 		lle->r_skip_req = 1;
702 		LLE_REQ_UNLOCK(lle);
703 		nd_delay = V_nd6_delay;
704 		nd_gctimer = V_nd6_gctimer;
705 
706 		delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
707 		remtime = (long)nd_gctimer * hz - delay;
708 		break;
709 	case ND6_LLINFO_DELAY:
710 		lle->la_asked = 0;
711 		delay = (long)V_nd6_delay * hz;
712 		break;
713 	}
714 
715 	if (delay > 0)
716 		nd6_llinfo_settimer_locked(lle, delay);
717 
718 	lle->lle_remtime = remtime;
719 	lle->ln_state = newstate;
720 }
721 
722 /*
723  * Timer-dependent part of nd state machine.
724  *
725  * Set noinline to be dtrace-friendly
726  */
727 static __noinline void
nd6_llinfo_timer(void * arg)728 nd6_llinfo_timer(void *arg)
729 {
730 	struct llentry *ln;
731 	struct in6_addr *dst, *pdst, *psrc, src;
732 	struct ifnet *ifp;
733 	struct nd_ifinfo *ndi;
734 	int do_switch, send_ns;
735 	long delay;
736 
737 	KASSERT(arg != NULL, ("%s: arg NULL", __func__));
738 	ln = (struct llentry *)arg;
739 	ifp = lltable_get_ifp(ln->lle_tbl);
740 	CURVNET_SET(ifp->if_vnet);
741 
742 	ND6_RLOCK();
743 	LLE_WLOCK(ln);
744 	if (callout_pending(&ln->lle_timer)) {
745 		/*
746 		 * Here we are a bit odd here in the treatment of
747 		 * active/pending. If the pending bit is set, it got
748 		 * rescheduled before I ran. The active
749 		 * bit we ignore, since if it was stopped
750 		 * in ll_tablefree() and was currently running
751 		 * it would have return 0 so the code would
752 		 * not have deleted it since the callout could
753 		 * not be stopped so we want to go through
754 		 * with the delete here now. If the callout
755 		 * was restarted, the pending bit will be back on and
756 		 * we just want to bail since the callout_reset would
757 		 * return 1 and our reference would have been removed
758 		 * by nd6_llinfo_settimer_locked above since canceled
759 		 * would have been 1.
760 		 */
761 		LLE_WUNLOCK(ln);
762 		ND6_RUNLOCK();
763 		CURVNET_RESTORE();
764 		return;
765 	}
766 	ndi = ND_IFINFO(ifp);
767 	send_ns = 0;
768 	dst = &ln->r_l3addr.addr6;
769 	pdst = dst;
770 
771 	if (ln->ln_ntick > 0) {
772 		if (ln->ln_ntick > INT_MAX) {
773 			ln->ln_ntick -= INT_MAX;
774 			nd6_llinfo_settimer_locked(ln, INT_MAX);
775 		} else {
776 			ln->ln_ntick = 0;
777 			nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
778 		}
779 		goto done;
780 	}
781 
782 	if (ln->la_flags & LLE_STATIC) {
783 		goto done;
784 	}
785 
786 	if (ln->la_flags & LLE_DELETED) {
787 		nd6_free(&ln, 0);
788 		goto done;
789 	}
790 
791 	switch (ln->ln_state) {
792 	case ND6_LLINFO_INCOMPLETE:
793 		if (ln->la_asked < V_nd6_mmaxtries) {
794 			ln->la_asked++;
795 			send_ns = 1;
796 			/* Send NS to multicast address */
797 			pdst = NULL;
798 		} else {
799 			struct mbuf *m = ln->la_hold;
800 			if (m) {
801 				struct mbuf *m0;
802 
803 				/*
804 				 * assuming every packet in la_hold has the
805 				 * same IP header.  Send error after unlock.
806 				 */
807 				m0 = m->m_nextpkt;
808 				m->m_nextpkt = NULL;
809 				ln->la_hold = m0;
810 				clear_llinfo_pqueue(ln);
811 			}
812 			nd6_free(&ln, 0);
813 			if (m != NULL)
814 				icmp6_error2(m, ICMP6_DST_UNREACH,
815 				    ICMP6_DST_UNREACH_ADDR, 0, ifp);
816 		}
817 		break;
818 	case ND6_LLINFO_REACHABLE:
819 		if (!ND6_LLINFO_PERMANENT(ln))
820 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
821 		break;
822 
823 	case ND6_LLINFO_STALE:
824 		if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
825 
826 			/*
827 			 * No packet has used this entry and GC timeout
828 			 * has not been passed. Reshedule timer and
829 			 * return.
830 			 */
831 			nd6_llinfo_settimer_locked(ln, delay);
832 			break;
833 		}
834 
835 		if (do_switch == 0) {
836 
837 			/*
838 			 * GC timer has ended and entry hasn't been used.
839 			 * Run Garbage collector (RFC 4861, 5.3)
840 			 */
841 			if (!ND6_LLINFO_PERMANENT(ln))
842 				nd6_free(&ln, 1);
843 			break;
844 		}
845 
846 		/* Entry has been used AND delay timer has ended. */
847 
848 		/* FALLTHROUGH */
849 
850 	case ND6_LLINFO_DELAY:
851 		if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
852 			/* We need NUD */
853 			ln->la_asked = 1;
854 			nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
855 			send_ns = 1;
856 		} else
857 			nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
858 		break;
859 	case ND6_LLINFO_PROBE:
860 		if (ln->la_asked < V_nd6_umaxtries) {
861 			ln->la_asked++;
862 			send_ns = 1;
863 		} else {
864 			nd6_free(&ln, 0);
865 		}
866 		break;
867 	default:
868 		panic("%s: paths in a dark night can be confusing: %d",
869 		    __func__, ln->ln_state);
870 	}
871 done:
872 	if (ln != NULL)
873 		ND6_RUNLOCK();
874 	if (send_ns != 0) {
875 		nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
876 		psrc = nd6_llinfo_get_holdsrc(ln, &src);
877 		LLE_FREE_LOCKED(ln);
878 		ln = NULL;
879 		nd6_ns_output(ifp, psrc, pdst, dst, NULL);
880 	}
881 
882 	if (ln != NULL)
883 		LLE_FREE_LOCKED(ln);
884 	CURVNET_RESTORE();
885 }
886 
887 
888 /*
889  * ND6 timer routine to expire default route list and prefix list
890  */
891 void
nd6_timer(void * arg)892 nd6_timer(void *arg)
893 {
894 	CURVNET_SET((struct vnet *) arg);
895 	struct nd_drhead drq;
896 	struct nd_prhead prl;
897 	struct nd_defrouter *dr, *ndr;
898 	struct nd_prefix *pr, *npr;
899 	struct ifnet *ifp;
900 	struct in6_ifaddr *ia6, *nia6;
901 	uint64_t genid;
902 
903 	TAILQ_INIT(&drq);
904 	LIST_INIT(&prl);
905 
906 	ND6_WLOCK();
907 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr)
908 		if (dr->expire && dr->expire < time_uptime)
909 			defrouter_unlink(dr, &drq);
910 	ND6_WUNLOCK();
911 
912 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
913 		TAILQ_REMOVE(&drq, dr, dr_entry);
914 		defrouter_del(dr);
915 	}
916 
917 	/*
918 	 * expire interface addresses.
919 	 * in the past the loop was inside prefix expiry processing.
920 	 * However, from a stricter speci-confrmance standpoint, we should
921 	 * rather separate address lifetimes and prefix lifetimes.
922 	 *
923 	 * XXXRW: in6_ifaddrhead locking.
924 	 */
925   addrloop:
926 	CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
927 		/* check address lifetime */
928 		if (IFA6_IS_INVALID(ia6)) {
929 			int regen = 0;
930 
931 			/*
932 			 * If the expiring address is temporary, try
933 			 * regenerating a new one.  This would be useful when
934 			 * we suspended a laptop PC, then turned it on after a
935 			 * period that could invalidate all temporary
936 			 * addresses.  Although we may have to restart the
937 			 * loop (see below), it must be after purging the
938 			 * address.  Otherwise, we'd see an infinite loop of
939 			 * regeneration.
940 			 */
941 			if (V_ip6_use_tempaddr &&
942 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
943 				if (regen_tmpaddr(ia6) == 0)
944 					regen = 1;
945 			}
946 
947 			in6_purgeaddr(&ia6->ia_ifa);
948 
949 			if (regen)
950 				goto addrloop; /* XXX: see below */
951 		} else if (IFA6_IS_DEPRECATED(ia6)) {
952 			int oldflags = ia6->ia6_flags;
953 
954 			ia6->ia6_flags |= IN6_IFF_DEPRECATED;
955 
956 			/*
957 			 * If a temporary address has just become deprecated,
958 			 * regenerate a new one if possible.
959 			 */
960 			if (V_ip6_use_tempaddr &&
961 			    (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
962 			    (oldflags & IN6_IFF_DEPRECATED) == 0) {
963 
964 				if (regen_tmpaddr(ia6) == 0) {
965 					/*
966 					 * A new temporary address is
967 					 * generated.
968 					 * XXX: this means the address chain
969 					 * has changed while we are still in
970 					 * the loop.  Although the change
971 					 * would not cause disaster (because
972 					 * it's not a deletion, but an
973 					 * addition,) we'd rather restart the
974 					 * loop just for safety.  Or does this
975 					 * significantly reduce performance??
976 					 */
977 					goto addrloop;
978 				}
979 			}
980 		} else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
981 			/*
982 			 * Schedule DAD for a tentative address.  This happens
983 			 * if the interface was down or not running
984 			 * when the address was configured.
985 			 */
986 			int delay;
987 
988 			delay = arc4random() %
989 			    (MAX_RTR_SOLICITATION_DELAY * hz);
990 			nd6_dad_start((struct ifaddr *)ia6, delay);
991 		} else {
992 			/*
993 			 * Check status of the interface.  If it is down,
994 			 * mark the address as tentative for future DAD.
995 			 */
996 			ifp = ia6->ia_ifp;
997 			if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
998 			    ((ifp->if_flags & IFF_UP) == 0 ||
999 			    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1000 			    (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1001 				ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1002 				ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1003 			}
1004 
1005 			/*
1006 			 * A new RA might have made a deprecated address
1007 			 * preferred.
1008 			 */
1009 			ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1010 		}
1011 	}
1012 
1013 	ND6_WLOCK();
1014 restart:
1015 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1016 		/*
1017 		 * Expire prefixes. Since the pltime is only used for
1018 		 * autoconfigured addresses, pltime processing for prefixes is
1019 		 * not necessary.
1020 		 *
1021 		 * Only unlink after all derived addresses have expired. This
1022 		 * may not occur until two hours after the prefix has expired
1023 		 * per RFC 4862. If the prefix expires before its derived
1024 		 * addresses, mark it off-link. This will be done automatically
1025 		 * after unlinking if no address references remain.
1026 		 */
1027 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1028 		    time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1029 			continue;
1030 
1031 		if (pr->ndpr_addrcnt == 0) {
1032 			nd6_prefix_unlink(pr, &prl);
1033 			continue;
1034 		}
1035 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1036 			genid = V_nd6_list_genid;
1037 			nd6_prefix_ref(pr);
1038 			ND6_WUNLOCK();
1039 			ND6_ONLINK_LOCK();
1040 			(void)nd6_prefix_offlink(pr);
1041 			ND6_ONLINK_UNLOCK();
1042 			ND6_WLOCK();
1043 			nd6_prefix_rele(pr);
1044 			if (genid != V_nd6_list_genid)
1045 				goto restart;
1046 		}
1047 	}
1048 	ND6_WUNLOCK();
1049 
1050 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1051 		LIST_REMOVE(pr, ndpr_entry);
1052 		nd6_prefix_del(pr);
1053 	}
1054 
1055 	callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1056 	    nd6_timer, curvnet);
1057 
1058 	CURVNET_RESTORE();
1059 }
1060 
1061 /*
1062  * ia6 - deprecated/invalidated temporary address
1063  */
1064 static int
regen_tmpaddr(struct in6_ifaddr * ia6)1065 regen_tmpaddr(struct in6_ifaddr *ia6)
1066 {
1067 	struct ifaddr *ifa;
1068 	struct ifnet *ifp;
1069 	struct in6_ifaddr *public_ifa6 = NULL;
1070 
1071 	ifp = ia6->ia_ifa.ifa_ifp;
1072 	IF_ADDR_RLOCK(ifp);
1073 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1074 		struct in6_ifaddr *it6;
1075 
1076 		if (ifa->ifa_addr->sa_family != AF_INET6)
1077 			continue;
1078 
1079 		it6 = (struct in6_ifaddr *)ifa;
1080 
1081 		/* ignore no autoconf addresses. */
1082 		if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1083 			continue;
1084 
1085 		/* ignore autoconf addresses with different prefixes. */
1086 		if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1087 			continue;
1088 
1089 		/*
1090 		 * Now we are looking at an autoconf address with the same
1091 		 * prefix as ours.  If the address is temporary and is still
1092 		 * preferred, do not create another one.  It would be rare, but
1093 		 * could happen, for example, when we resume a laptop PC after
1094 		 * a long period.
1095 		 */
1096 		if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1097 		    !IFA6_IS_DEPRECATED(it6)) {
1098 			public_ifa6 = NULL;
1099 			break;
1100 		}
1101 
1102 		/*
1103 		 * This is a public autoconf address that has the same prefix
1104 		 * as ours.  If it is preferred, keep it.  We can't break the
1105 		 * loop here, because there may be a still-preferred temporary
1106 		 * address with the prefix.
1107 		 */
1108 		if (!IFA6_IS_DEPRECATED(it6))
1109 			public_ifa6 = it6;
1110 	}
1111 	if (public_ifa6 != NULL)
1112 		ifa_ref(&public_ifa6->ia_ifa);
1113 	IF_ADDR_RUNLOCK(ifp);
1114 
1115 	if (public_ifa6 != NULL) {
1116 		int e;
1117 
1118 		if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1119 			ifa_free(&public_ifa6->ia_ifa);
1120 			log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1121 			    " tmp addr,errno=%d\n", e);
1122 			return (-1);
1123 		}
1124 		ifa_free(&public_ifa6->ia_ifa);
1125 		return (0);
1126 	}
1127 
1128 	return (-1);
1129 }
1130 
1131 /*
1132  * Remove prefix and default router list entries corresponding to ifp. Neighbor
1133  * cache entries are freed in in6_domifdetach().
1134  */
1135 void
nd6_purge(struct ifnet * ifp)1136 nd6_purge(struct ifnet *ifp)
1137 {
1138 	struct nd_drhead drq;
1139 	struct nd_prhead prl;
1140 	struct nd_defrouter *dr, *ndr;
1141 	struct nd_prefix *pr, *npr;
1142 
1143 	TAILQ_INIT(&drq);
1144 	LIST_INIT(&prl);
1145 
1146 	/*
1147 	 * Nuke default router list entries toward ifp.
1148 	 * We defer removal of default router list entries that is installed
1149 	 * in the routing table, in order to keep additional side effects as
1150 	 * small as possible.
1151 	 */
1152 	ND6_WLOCK();
1153 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1154 		if (dr->installed)
1155 			continue;
1156 		if (dr->ifp == ifp)
1157 			defrouter_unlink(dr, &drq);
1158 	}
1159 	TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) {
1160 		if (!dr->installed)
1161 			continue;
1162 		if (dr->ifp == ifp)
1163 			defrouter_unlink(dr, &drq);
1164 	}
1165 
1166 	/*
1167 	 * Remove prefixes on ifp. We should have already removed addresses on
1168 	 * this interface, so no addresses should be referencing these prefixes.
1169 	 */
1170 	LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1171 		if (pr->ndpr_ifp == ifp)
1172 			nd6_prefix_unlink(pr, &prl);
1173 	}
1174 	ND6_WUNLOCK();
1175 
1176 	/* Delete the unlinked router and prefix objects. */
1177 	while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1178 		TAILQ_REMOVE(&drq, dr, dr_entry);
1179 		defrouter_del(dr);
1180 	}
1181 	while ((pr = LIST_FIRST(&prl)) != NULL) {
1182 		LIST_REMOVE(pr, ndpr_entry);
1183 		nd6_prefix_del(pr);
1184 	}
1185 
1186 	/* cancel default outgoing interface setting */
1187 	if (V_nd6_defifindex == ifp->if_index)
1188 		nd6_setdefaultiface(0);
1189 
1190 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1191 		/* Refresh default router list. */
1192 		defrouter_select_fib(ifp->if_fib);
1193 	}
1194 }
1195 
1196 /*
1197  * the caller acquires and releases the lock on the lltbls
1198  * Returns the llentry locked
1199  */
1200 struct llentry *
nd6_lookup(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1201 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1202 {
1203 	struct sockaddr_in6 sin6;
1204 	struct llentry *ln;
1205 
1206 	bzero(&sin6, sizeof(sin6));
1207 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1208 	sin6.sin6_family = AF_INET6;
1209 	sin6.sin6_addr = *addr6;
1210 
1211 	IF_AFDATA_LOCK_ASSERT(ifp);
1212 
1213 	ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1214 
1215 	return (ln);
1216 }
1217 
1218 struct llentry *
nd6_alloc(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1219 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1220 {
1221 	struct sockaddr_in6 sin6;
1222 	struct llentry *ln;
1223 
1224 	bzero(&sin6, sizeof(sin6));
1225 	sin6.sin6_len = sizeof(struct sockaddr_in6);
1226 	sin6.sin6_family = AF_INET6;
1227 	sin6.sin6_addr = *addr6;
1228 
1229 	ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1230 	if (ln != NULL)
1231 		ln->ln_state = ND6_LLINFO_NOSTATE;
1232 
1233 	return (ln);
1234 }
1235 
1236 /*
1237  * Test whether a given IPv6 address is a neighbor or not, ignoring
1238  * the actual neighbor cache.  The neighbor cache is ignored in order
1239  * to not reenter the routing code from within itself.
1240  */
1241 static int
nd6_is_new_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1242 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1243 {
1244 	struct nd_prefix *pr;
1245 	struct ifaddr *ifa;
1246 	struct rt_addrinfo info;
1247 	struct sockaddr_in6 rt_key;
1248 	const struct sockaddr *dst6;
1249 	uint64_t genid;
1250 	int error, fibnum;
1251 
1252 	/*
1253 	 * A link-local address is always a neighbor.
1254 	 * XXX: a link does not necessarily specify a single interface.
1255 	 */
1256 	if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1257 		struct sockaddr_in6 sin6_copy;
1258 		u_int32_t zone;
1259 
1260 		/*
1261 		 * We need sin6_copy since sa6_recoverscope() may modify the
1262 		 * content (XXX).
1263 		 */
1264 		sin6_copy = *addr;
1265 		if (sa6_recoverscope(&sin6_copy))
1266 			return (0); /* XXX: should be impossible */
1267 		if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1268 			return (0);
1269 		if (sin6_copy.sin6_scope_id == zone)
1270 			return (1);
1271 		else
1272 			return (0);
1273 	}
1274 
1275 	bzero(&rt_key, sizeof(rt_key));
1276 	bzero(&info, sizeof(info));
1277 	info.rti_info[RTAX_DST] = (struct sockaddr *)&rt_key;
1278 
1279 	/*
1280 	 * If the address matches one of our addresses,
1281 	 * it should be a neighbor.
1282 	 * If the address matches one of our on-link prefixes, it should be a
1283 	 * neighbor.
1284 	 */
1285 	ND6_RLOCK();
1286 restart:
1287 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1288 		if (pr->ndpr_ifp != ifp)
1289 			continue;
1290 
1291 		if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0) {
1292 			dst6 = (const struct sockaddr *)&pr->ndpr_prefix;
1293 
1294 			/*
1295 			 * We only need to check all FIBs if add_addr_allfibs
1296 			 * is unset. If set, checking any FIB will suffice.
1297 			 */
1298 			fibnum = V_rt_add_addr_allfibs ? rt_numfibs - 1 : 0;
1299 			for (; fibnum < rt_numfibs; fibnum++) {
1300 				genid = V_nd6_list_genid;
1301 				ND6_RUNLOCK();
1302 
1303 				/*
1304 				 * Restore length field before
1305 				 * retrying lookup
1306 				 */
1307 				rt_key.sin6_len = sizeof(rt_key);
1308 				error = rib_lookup_info(fibnum, dst6, 0, 0,
1309 						        &info);
1310 
1311 				ND6_RLOCK();
1312 				if (genid != V_nd6_list_genid)
1313 					goto restart;
1314 				if (error == 0)
1315 					break;
1316 			}
1317 			if (error != 0)
1318 				continue;
1319 
1320 			/*
1321 			 * This is the case where multiple interfaces
1322 			 * have the same prefix, but only one is installed
1323 			 * into the routing table and that prefix entry
1324 			 * is not the one being examined here. In the case
1325 			 * where RADIX_MPATH is enabled, multiple route
1326 			 * entries (of the same rt_key value) will be
1327 			 * installed because the interface addresses all
1328 			 * differ.
1329 			 */
1330 			if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1331 			    &rt_key.sin6_addr))
1332 				continue;
1333 		}
1334 
1335 		if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1336 		    &addr->sin6_addr, &pr->ndpr_mask)) {
1337 			ND6_RUNLOCK();
1338 			return (1);
1339 		}
1340 	}
1341 	ND6_RUNLOCK();
1342 
1343 	/*
1344 	 * If the address is assigned on the node of the other side of
1345 	 * a p2p interface, the address should be a neighbor.
1346 	 */
1347 	if (ifp->if_flags & IFF_POINTOPOINT) {
1348 		IF_ADDR_RLOCK(ifp);
1349 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1350 			if (ifa->ifa_addr->sa_family != addr->sin6_family)
1351 				continue;
1352 			if (ifa->ifa_dstaddr != NULL &&
1353 			    sa_equal(addr, ifa->ifa_dstaddr)) {
1354 				IF_ADDR_RUNLOCK(ifp);
1355 				return 1;
1356 			}
1357 		}
1358 		IF_ADDR_RUNLOCK(ifp);
1359 	}
1360 
1361 	/*
1362 	 * If the default router list is empty, all addresses are regarded
1363 	 * as on-link, and thus, as a neighbor.
1364 	 */
1365 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1366 	    TAILQ_EMPTY(&V_nd_defrouter) &&
1367 	    V_nd6_defifindex == ifp->if_index) {
1368 		return (1);
1369 	}
1370 
1371 	return (0);
1372 }
1373 
1374 
1375 /*
1376  * Detect if a given IPv6 address identifies a neighbor on a given link.
1377  * XXX: should take care of the destination of a p2p link?
1378  */
1379 int
nd6_is_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1380 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1381 {
1382 	struct llentry *lle;
1383 	int rc = 0;
1384 
1385 	IF_AFDATA_UNLOCK_ASSERT(ifp);
1386 	if (nd6_is_new_addr_neighbor(addr, ifp))
1387 		return (1);
1388 
1389 	/*
1390 	 * Even if the address matches none of our addresses, it might be
1391 	 * in the neighbor cache.
1392 	 */
1393 	IF_AFDATA_RLOCK(ifp);
1394 	if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) {
1395 		LLE_RUNLOCK(lle);
1396 		rc = 1;
1397 	}
1398 	IF_AFDATA_RUNLOCK(ifp);
1399 	return (rc);
1400 }
1401 
1402 /*
1403  * Free an nd6 llinfo entry.
1404  * Since the function would cause significant changes in the kernel, DO NOT
1405  * make it global, unless you have a strong reason for the change, and are sure
1406  * that the change is safe.
1407  *
1408  * Set noinline to be dtrace-friendly
1409  */
1410 static __noinline void
nd6_free(struct llentry ** lnp,int gc)1411 nd6_free(struct llentry **lnp, int gc)
1412 {
1413 	struct ifnet *ifp;
1414 	struct llentry *ln;
1415 	struct nd_defrouter *dr;
1416 
1417 	ln = *lnp;
1418 	*lnp = NULL;
1419 
1420 	LLE_WLOCK_ASSERT(ln);
1421 	ND6_RLOCK_ASSERT();
1422 
1423 	ifp = lltable_get_ifp(ln->lle_tbl);
1424 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1425 		dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1426 	else
1427 		dr = NULL;
1428 	ND6_RUNLOCK();
1429 
1430 	if ((ln->la_flags & LLE_DELETED) == 0)
1431 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1432 
1433 	/*
1434 	 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1435 	 * even though it is not harmful, it was not really necessary.
1436 	 */
1437 
1438 	/* cancel timer */
1439 	nd6_llinfo_settimer_locked(ln, -1);
1440 
1441 	if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1442 		if (dr != NULL && dr->expire &&
1443 		    ln->ln_state == ND6_LLINFO_STALE && gc) {
1444 			/*
1445 			 * If the reason for the deletion is just garbage
1446 			 * collection, and the neighbor is an active default
1447 			 * router, do not delete it.  Instead, reset the GC
1448 			 * timer using the router's lifetime.
1449 			 * Simply deleting the entry would affect default
1450 			 * router selection, which is not necessarily a good
1451 			 * thing, especially when we're using router preference
1452 			 * values.
1453 			 * XXX: the check for ln_state would be redundant,
1454 			 *      but we intentionally keep it just in case.
1455 			 */
1456 			if (dr->expire > time_uptime)
1457 				nd6_llinfo_settimer_locked(ln,
1458 				    (dr->expire - time_uptime) * hz);
1459 			else
1460 				nd6_llinfo_settimer_locked(ln,
1461 				    (long)V_nd6_gctimer * hz);
1462 
1463 			LLE_REMREF(ln);
1464 			LLE_WUNLOCK(ln);
1465 			defrouter_rele(dr);
1466 			return;
1467 		}
1468 
1469 		if (dr) {
1470 			/*
1471 			 * Unreachablity of a router might affect the default
1472 			 * router selection and on-link detection of advertised
1473 			 * prefixes.
1474 			 */
1475 
1476 			/*
1477 			 * Temporarily fake the state to choose a new default
1478 			 * router and to perform on-link determination of
1479 			 * prefixes correctly.
1480 			 * Below the state will be set correctly,
1481 			 * or the entry itself will be deleted.
1482 			 */
1483 			ln->ln_state = ND6_LLINFO_INCOMPLETE;
1484 		}
1485 
1486 		if (ln->ln_router || dr) {
1487 
1488 			/*
1489 			 * We need to unlock to avoid a LOR with rt6_flush() with the
1490 			 * rnh and for the calls to pfxlist_onlink_check() and
1491 			 * defrouter_select_fib() in the block further down for calls
1492 			 * into nd6_lookup().  We still hold a ref.
1493 			 */
1494 			LLE_WUNLOCK(ln);
1495 
1496 			/*
1497 			 * rt6_flush must be called whether or not the neighbor
1498 			 * is in the Default Router List.
1499 			 * See a corresponding comment in nd6_na_input().
1500 			 */
1501 			rt6_flush(&ln->r_l3addr.addr6, ifp);
1502 		}
1503 
1504 		if (dr) {
1505 			/*
1506 			 * Since defrouter_select_fib() does not affect the
1507 			 * on-link determination and MIP6 needs the check
1508 			 * before the default router selection, we perform
1509 			 * the check now.
1510 			 */
1511 			pfxlist_onlink_check();
1512 
1513 			/*
1514 			 * Refresh default router list.
1515 			 */
1516 			defrouter_select_fib(dr->ifp->if_fib);
1517 		}
1518 
1519 		/*
1520 		 * If this entry was added by an on-link redirect, remove the
1521 		 * corresponding host route.
1522 		 */
1523 		if (ln->la_flags & LLE_REDIRECT)
1524 			nd6_free_redirect(ln);
1525 
1526 		if (ln->ln_router || dr)
1527 			LLE_WLOCK(ln);
1528 	}
1529 
1530 	/*
1531 	 * Save to unlock. We still hold an extra reference and will not
1532 	 * free(9) in llentry_free() if someone else holds one as well.
1533 	 */
1534 	LLE_WUNLOCK(ln);
1535 	IF_AFDATA_LOCK(ifp);
1536 	LLE_WLOCK(ln);
1537 	/* Guard against race with other llentry_free(). */
1538 	if (ln->la_flags & LLE_LINKED) {
1539 		/* Remove callout reference */
1540 		LLE_REMREF(ln);
1541 		lltable_unlink_entry(ln->lle_tbl, ln);
1542 	}
1543 	IF_AFDATA_UNLOCK(ifp);
1544 
1545 	llentry_free(ln);
1546 	if (dr != NULL)
1547 		defrouter_rele(dr);
1548 }
1549 
1550 static int
nd6_isdynrte(const struct rtentry * rt,void * xap)1551 nd6_isdynrte(const struct rtentry *rt, void *xap)
1552 {
1553 
1554 	if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC))
1555 		return (1);
1556 
1557 	return (0);
1558 }
1559 /*
1560  * Remove the rtentry for the given llentry,
1561  * both of which were installed by a redirect.
1562  */
1563 static void
nd6_free_redirect(const struct llentry * ln)1564 nd6_free_redirect(const struct llentry *ln)
1565 {
1566 	int fibnum;
1567 	struct sockaddr_in6 sin6;
1568 	struct rt_addrinfo info;
1569 
1570 	lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1571 	memset(&info, 0, sizeof(info));
1572 	info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6;
1573 	info.rti_filter = nd6_isdynrte;
1574 
1575 	for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1576 		rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum);
1577 }
1578 
1579 /*
1580  * Rejuvenate this function for routing operations related
1581  * processing.
1582  */
1583 void
nd6_rtrequest(int req,struct rtentry * rt,struct rt_addrinfo * info)1584 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info)
1585 {
1586 	struct sockaddr_in6 *gateway;
1587 	struct nd_defrouter *dr;
1588 	struct ifnet *ifp;
1589 
1590 	gateway = (struct sockaddr_in6 *)rt->rt_gateway;
1591 	ifp = rt->rt_ifp;
1592 
1593 	switch (req) {
1594 	case RTM_ADD:
1595 		break;
1596 
1597 	case RTM_DELETE:
1598 		if (!ifp)
1599 			return;
1600 		/*
1601 		 * Only indirect routes are interesting.
1602 		 */
1603 		if ((rt->rt_flags & RTF_GATEWAY) == 0)
1604 			return;
1605 		/*
1606 		 * check for default route
1607 		 */
1608 		if (IN6_ARE_ADDR_EQUAL(&in6addr_any,
1609 		    &SIN6(rt_key(rt))->sin6_addr)) {
1610 			dr = defrouter_lookup(&gateway->sin6_addr, ifp);
1611 			if (dr != NULL) {
1612 				dr->installed = 0;
1613 				defrouter_rele(dr);
1614 			}
1615 		}
1616 		break;
1617 	}
1618 }
1619 
1620 
1621 int
nd6_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp)1622 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1623 {
1624 	struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1625 	struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1626 	struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1627 	int error = 0;
1628 
1629 	if (ifp->if_afdata[AF_INET6] == NULL)
1630 		return (EPFNOSUPPORT);
1631 	switch (cmd) {
1632 	case OSIOCGIFINFO_IN6:
1633 #define ND	ndi->ndi
1634 		/* XXX: old ndp(8) assumes a positive value for linkmtu. */
1635 		bzero(&ND, sizeof(ND));
1636 		ND.linkmtu = IN6_LINKMTU(ifp);
1637 		ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1638 		ND.basereachable = ND_IFINFO(ifp)->basereachable;
1639 		ND.reachable = ND_IFINFO(ifp)->reachable;
1640 		ND.retrans = ND_IFINFO(ifp)->retrans;
1641 		ND.flags = ND_IFINFO(ifp)->flags;
1642 		ND.recalctm = ND_IFINFO(ifp)->recalctm;
1643 		ND.chlim = ND_IFINFO(ifp)->chlim;
1644 		break;
1645 	case SIOCGIFINFO_IN6:
1646 		ND = *ND_IFINFO(ifp);
1647 		break;
1648 	case SIOCSIFINFO_IN6:
1649 		/*
1650 		 * used to change host variables from userland.
1651 		 * intended for a use on router to reflect RA configurations.
1652 		 */
1653 		/* 0 means 'unspecified' */
1654 		if (ND.linkmtu != 0) {
1655 			if (ND.linkmtu < IPV6_MMTU ||
1656 			    ND.linkmtu > IN6_LINKMTU(ifp)) {
1657 				error = EINVAL;
1658 				break;
1659 			}
1660 			ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1661 		}
1662 
1663 		if (ND.basereachable != 0) {
1664 			int obasereachable = ND_IFINFO(ifp)->basereachable;
1665 
1666 			ND_IFINFO(ifp)->basereachable = ND.basereachable;
1667 			if (ND.basereachable != obasereachable)
1668 				ND_IFINFO(ifp)->reachable =
1669 				    ND_COMPUTE_RTIME(ND.basereachable);
1670 		}
1671 		if (ND.retrans != 0)
1672 			ND_IFINFO(ifp)->retrans = ND.retrans;
1673 		if (ND.chlim != 0)
1674 			ND_IFINFO(ifp)->chlim = ND.chlim;
1675 		/* FALLTHROUGH */
1676 	case SIOCSIFINFO_FLAGS:
1677 	{
1678 		struct ifaddr *ifa;
1679 		struct in6_ifaddr *ia;
1680 
1681 		if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1682 		    !(ND.flags & ND6_IFF_IFDISABLED)) {
1683 			/* ifdisabled 1->0 transision */
1684 
1685 			/*
1686 			 * If the interface is marked as ND6_IFF_IFDISABLED and
1687 			 * has an link-local address with IN6_IFF_DUPLICATED,
1688 			 * do not clear ND6_IFF_IFDISABLED.
1689 			 * See RFC 4862, Section 5.4.5.
1690 			 */
1691 			IF_ADDR_RLOCK(ifp);
1692 			CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1693 				if (ifa->ifa_addr->sa_family != AF_INET6)
1694 					continue;
1695 				ia = (struct in6_ifaddr *)ifa;
1696 				if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1697 				    IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1698 					break;
1699 			}
1700 			IF_ADDR_RUNLOCK(ifp);
1701 
1702 			if (ifa != NULL) {
1703 				/* LLA is duplicated. */
1704 				ND.flags |= ND6_IFF_IFDISABLED;
1705 				log(LOG_ERR, "Cannot enable an interface"
1706 				    " with a link-local address marked"
1707 				    " duplicate.\n");
1708 			} else {
1709 				ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1710 				if (ifp->if_flags & IFF_UP)
1711 					in6_if_up(ifp);
1712 			}
1713 		} else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1714 			    (ND.flags & ND6_IFF_IFDISABLED)) {
1715 			/* ifdisabled 0->1 transision */
1716 			/* Mark all IPv6 address as tentative. */
1717 
1718 			ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1719 			if (V_ip6_dad_count > 0 &&
1720 			    (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1721 				IF_ADDR_RLOCK(ifp);
1722 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1723 				    ifa_link) {
1724 					if (ifa->ifa_addr->sa_family !=
1725 					    AF_INET6)
1726 						continue;
1727 					ia = (struct in6_ifaddr *)ifa;
1728 					ia->ia6_flags |= IN6_IFF_TENTATIVE;
1729 				}
1730 				IF_ADDR_RUNLOCK(ifp);
1731 			}
1732 		}
1733 
1734 		if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1735 			if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1736 				/* auto_linklocal 0->1 transision */
1737 
1738 				/* If no link-local address on ifp, configure */
1739 				ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1740 				in6_ifattach(ifp, NULL);
1741 			} else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1742 			    ifp->if_flags & IFF_UP) {
1743 				/*
1744 				 * When the IF already has
1745 				 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1746 				 * address is assigned, and IFF_UP, try to
1747 				 * assign one.
1748 				 */
1749 				IF_ADDR_RLOCK(ifp);
1750 				CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1751 				    ifa_link) {
1752 					if (ifa->ifa_addr->sa_family !=
1753 					    AF_INET6)
1754 						continue;
1755 					ia = (struct in6_ifaddr *)ifa;
1756 					if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1757 						break;
1758 				}
1759 				IF_ADDR_RUNLOCK(ifp);
1760 				if (ifa != NULL)
1761 					/* No LLA is configured. */
1762 					in6_ifattach(ifp, NULL);
1763 			}
1764 		}
1765 	}
1766 		ND_IFINFO(ifp)->flags = ND.flags;
1767 		break;
1768 #undef ND
1769 	case SIOCSNDFLUSH_IN6:	/* XXX: the ioctl name is confusing... */
1770 		/* sync kernel routing table with the default router list */
1771 		defrouter_reset();
1772 		defrouter_select();
1773 		break;
1774 	case SIOCSPFXFLUSH_IN6:
1775 	{
1776 		/* flush all the prefix advertised by routers */
1777 		struct in6_ifaddr *ia, *ia_next;
1778 		struct nd_prefix *pr, *next;
1779 		struct nd_prhead prl;
1780 
1781 		LIST_INIT(&prl);
1782 
1783 		ND6_WLOCK();
1784 		LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1785 			if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr))
1786 				continue; /* XXX */
1787 			nd6_prefix_unlink(pr, &prl);
1788 		}
1789 		ND6_WUNLOCK();
1790 
1791 		while ((pr = LIST_FIRST(&prl)) != NULL) {
1792 			LIST_REMOVE(pr, ndpr_entry);
1793 			/* XXXRW: in6_ifaddrhead locking. */
1794 			CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1795 			    ia_next) {
1796 				if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1797 					continue;
1798 
1799 				if (ia->ia6_ndpr == pr)
1800 					in6_purgeaddr(&ia->ia_ifa);
1801 			}
1802 			nd6_prefix_del(pr);
1803 		}
1804 		break;
1805 	}
1806 	case SIOCSRTRFLUSH_IN6:
1807 	{
1808 		/* flush all the default routers */
1809 		struct nd_drhead drq;
1810 		struct nd_defrouter *dr;
1811 
1812 		TAILQ_INIT(&drq);
1813 
1814 		defrouter_reset();
1815 
1816 		ND6_WLOCK();
1817 		while ((dr = TAILQ_FIRST(&V_nd_defrouter)) != NULL)
1818 			defrouter_unlink(dr, &drq);
1819 		ND6_WUNLOCK();
1820 		while ((dr = TAILQ_FIRST(&drq)) != NULL) {
1821 			TAILQ_REMOVE(&drq, dr, dr_entry);
1822 			defrouter_del(dr);
1823 		}
1824 
1825 		defrouter_select();
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 		IF_AFDATA_RLOCK(ifp);
1837 		ln = nd6_lookup(&nb_addr, 0, ifp);
1838 		IF_AFDATA_RUNLOCK(ifp);
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 sockaddr_in6 sin6;
1963 	struct mbuf *chain = NULL;
1964 	u_char linkhdr[LLE_MAX_LINKHDR];
1965 	size_t linkhdrsize;
1966 	int lladdr_off;
1967 
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 	IF_AFDATA_RLOCK(ifp);
1988 	ln = nd6_lookup(from, flags, ifp);
1989 	IF_AFDATA_RUNLOCK(ifp);
1990 	is_newentry = 0;
1991 	if (ln == NULL) {
1992 		flags |= LLE_EXCLUSIVE;
1993 		ln = nd6_alloc(from, 0, ifp);
1994 		if (ln == NULL)
1995 			return;
1996 
1997 		/*
1998 		 * Since we already know all the data for the new entry,
1999 		 * fill it before insertion.
2000 		 */
2001 		if (lladdr != NULL) {
2002 			linkhdrsize = sizeof(linkhdr);
2003 			if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2004 			    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2005 				return;
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_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 		linkhdrsize = sizeof(linkhdr);
2069 		if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2070 		    linkhdr, &linkhdrsize, &lladdr_off) != 0)
2071 			return;
2072 
2073 		if (lltable_try_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2074 		    lladdr_off) == 0) {
2075 			/* Entry was deleted */
2076 			return;
2077 		}
2078 
2079 		nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2080 
2081 		EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2082 
2083 		if (ln->la_hold != NULL)
2084 			nd6_grab_holdchain(ln, &chain, &sin6);
2085 	}
2086 
2087 	/* Calculates new router status */
2088 	router = nd6_is_router(type, code, is_newentry, olladdr,
2089 	    lladdr != NULL ? 1 : 0, ln->ln_router);
2090 
2091 	ln->ln_router = router;
2092 	/* Mark non-router redirects with special flag */
2093 	if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2094 		ln->la_flags |= LLE_REDIRECT;
2095 
2096 	if (flags & LLE_EXCLUSIVE)
2097 		LLE_WUNLOCK(ln);
2098 	else
2099 		LLE_RUNLOCK(ln);
2100 
2101 	if (chain != NULL)
2102 		nd6_flush_holdchain(ifp, chain, &sin6);
2103 
2104 	/*
2105 	 * When the link-layer address of a router changes, select the
2106 	 * best router again.  In particular, when the neighbor entry is newly
2107 	 * created, it might affect the selection policy.
2108 	 * Question: can we restrict the first condition to the "is_newentry"
2109 	 * case?
2110 	 * XXX: when we hear an RA from a new router with the link-layer
2111 	 * address option, defrouter_select_fib() is called twice, since
2112 	 * defrtrlist_update called the function as well.  However, I believe
2113 	 * we can compromise the overhead, since it only happens the first
2114 	 * time.
2115 	 * XXX: although defrouter_select_fib() should not have a bad effect
2116 	 * for those are not autoconfigured hosts, we explicitly avoid such
2117 	 * cases for safety.
2118 	 */
2119 	if ((do_update || is_newentry) && router &&
2120 	    ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2121 		/*
2122 		 * guaranteed recursion
2123 		 */
2124 		defrouter_select_fib(ifp->if_fib);
2125 	}
2126 }
2127 
2128 static void
nd6_slowtimo(void * arg)2129 nd6_slowtimo(void *arg)
2130 {
2131 	CURVNET_SET((struct vnet *) arg);
2132 	struct nd_ifinfo *nd6if;
2133 	struct ifnet *ifp;
2134 
2135 	callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2136 	    nd6_slowtimo, curvnet);
2137 	IFNET_RLOCK_NOSLEEP();
2138 	CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2139 		if (ifp->if_afdata[AF_INET6] == NULL)
2140 			continue;
2141 		nd6if = ND_IFINFO(ifp);
2142 		if (nd6if->basereachable && /* already initialized */
2143 		    (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2144 			/*
2145 			 * Since reachable time rarely changes by router
2146 			 * advertisements, we SHOULD insure that a new random
2147 			 * value gets recomputed at least once every few hours.
2148 			 * (RFC 2461, 6.3.4)
2149 			 */
2150 			nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2151 			nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2152 		}
2153 	}
2154 	IFNET_RUNLOCK_NOSLEEP();
2155 	CURVNET_RESTORE();
2156 }
2157 
2158 void
nd6_grab_holdchain(struct llentry * ln,struct mbuf ** chain,struct sockaddr_in6 * sin6)2159 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain,
2160     struct sockaddr_in6 *sin6)
2161 {
2162 
2163 	LLE_WLOCK_ASSERT(ln);
2164 
2165 	*chain = ln->la_hold;
2166 	ln->la_hold = NULL;
2167 	lltable_fill_sa_entry(ln, (struct sockaddr *)sin6);
2168 
2169 	if (ln->ln_state == ND6_LLINFO_STALE) {
2170 
2171 		/*
2172 		 * The first time we send a packet to a
2173 		 * neighbor whose entry is STALE, we have
2174 		 * to change the state to DELAY and a sets
2175 		 * a timer to expire in DELAY_FIRST_PROBE_TIME
2176 		 * seconds to ensure do neighbor unreachability
2177 		 * detection on expiration.
2178 		 * (RFC 2461 7.3.3)
2179 		 */
2180 		nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2181 	}
2182 }
2183 
2184 int
nd6_output_ifp(struct ifnet * ifp,struct ifnet * origifp,struct mbuf * m,struct sockaddr_in6 * dst,struct route * ro)2185 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2186     struct sockaddr_in6 *dst, struct route *ro)
2187 {
2188 	int error;
2189 	int ip6len;
2190 	struct ip6_hdr *ip6;
2191 	struct m_tag *mtag;
2192 
2193 #ifdef MAC
2194 	mac_netinet6_nd6_send(ifp, m);
2195 #endif
2196 
2197 	/*
2198 	 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2199 	 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2200 	 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2201 	 * to be diverted to user space.  When re-injected into the kernel,
2202 	 * send_output() will directly dispatch them to the outgoing interface.
2203 	 */
2204 	if (send_sendso_input_hook != NULL) {
2205 		mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2206 		if (mtag != NULL) {
2207 			ip6 = mtod(m, struct ip6_hdr *);
2208 			ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2209 			/* Use the SEND socket */
2210 			error = send_sendso_input_hook(m, ifp, SND_OUT,
2211 			    ip6len);
2212 			/* -1 == no app on SEND socket */
2213 			if (error == 0 || error != -1)
2214 			    return (error);
2215 		}
2216 	}
2217 
2218 	m_clrprotoflags(m);	/* Avoid confusing lower layers. */
2219 	IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2220 	    mtod(m, struct ip6_hdr *));
2221 
2222 	if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2223 		origifp = ifp;
2224 
2225 	error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2226 	return (error);
2227 }
2228 
2229 /*
2230  * Lookup link headerfor @sa_dst address. Stores found
2231  * data in @desten buffer. Copy of lle ln_flags can be also
2232  * saved in @pflags if @pflags is non-NULL.
2233  *
2234  * If destination LLE does not exists or lle state modification
2235  * is required, call "slow" version.
2236  *
2237  * Return values:
2238  * - 0 on success (address copied to buffer).
2239  * - EWOULDBLOCK (no local error, but address is still unresolved)
2240  * - other errors (alloc failure, etc)
2241  */
2242 int
nd6_resolve(struct ifnet * ifp,int is_gw,struct mbuf * m,const struct sockaddr * sa_dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)2243 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m,
2244     const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2245     struct llentry **plle)
2246 {
2247 	struct llentry *ln = NULL;
2248 	const struct sockaddr_in6 *dst6;
2249 
2250 	if (pflags != NULL)
2251 		*pflags = 0;
2252 
2253 	dst6 = (const struct sockaddr_in6 *)sa_dst;
2254 
2255 	/* discard the packet if IPv6 operation is disabled on the interface */
2256 	if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2257 		m_freem(m);
2258 		return (ENETDOWN); /* better error? */
2259 	}
2260 
2261 	if (m != NULL && m->m_flags & M_MCAST) {
2262 		switch (ifp->if_type) {
2263 		case IFT_ETHER:
2264 		case IFT_L2VLAN:
2265 		case IFT_BRIDGE:
2266 			ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2267 						 desten);
2268 			return (0);
2269 		default:
2270 			m_freem(m);
2271 			return (EAFNOSUPPORT);
2272 		}
2273 	}
2274 
2275 	IF_AFDATA_RLOCK(ifp);
2276 	ln = nd6_lookup(&dst6->sin6_addr, plle ? LLE_EXCLUSIVE : LLE_UNLOCKED,
2277 	    ifp);
2278 	if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2279 		/* Entry found, let's copy lle info */
2280 		bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2281 		if (pflags != NULL)
2282 			*pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2283 		/* Check if we have feedback request from nd6 timer */
2284 		if (ln->r_skip_req != 0) {
2285 			LLE_REQ_LOCK(ln);
2286 			ln->r_skip_req = 0; /* Notify that entry was used */
2287 			ln->lle_hittime = time_uptime;
2288 			LLE_REQ_UNLOCK(ln);
2289 		}
2290 		if (plle) {
2291 			LLE_ADDREF(ln);
2292 			*plle = ln;
2293 			LLE_WUNLOCK(ln);
2294 		}
2295 		IF_AFDATA_RUNLOCK(ifp);
2296 		return (0);
2297 	} else if (plle && ln)
2298 		LLE_WUNLOCK(ln);
2299 	IF_AFDATA_RUNLOCK(ifp);
2300 
2301 	return (nd6_resolve_slow(ifp, 0, m, dst6, desten, pflags, plle));
2302 }
2303 
2304 
2305 /*
2306  * Do L2 address resolution for @sa_dst address. Stores found
2307  * address in @desten buffer. Copy of lle ln_flags can be also
2308  * saved in @pflags if @pflags is non-NULL.
2309  *
2310  * Heavy version.
2311  * Function assume that destination LLE does not exist,
2312  * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2313  *
2314  * Set noinline to be dtrace-friendly
2315  */
2316 static __noinline int
nd6_resolve_slow(struct ifnet * ifp,int flags,struct mbuf * m,const struct sockaddr_in6 * dst,u_char * desten,uint32_t * pflags,struct llentry ** plle)2317 nd6_resolve_slow(struct ifnet *ifp, int flags, struct mbuf *m,
2318     const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2319     struct llentry **plle)
2320 {
2321 	struct llentry *lle = NULL, *lle_tmp;
2322 	struct in6_addr *psrc, src;
2323 	int send_ns, ll_len;
2324 	char *lladdr;
2325 
2326 	/*
2327 	 * Address resolution or Neighbor Unreachability Detection
2328 	 * for the next hop.
2329 	 * At this point, the destination of the packet must be a unicast
2330 	 * or an anycast address(i.e. not a multicast).
2331 	 */
2332 	if (lle == NULL) {
2333 		IF_AFDATA_RLOCK(ifp);
2334 		lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2335 		IF_AFDATA_RUNLOCK(ifp);
2336 		if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp))  {
2337 			/*
2338 			 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2339 			 * the condition below is not very efficient.  But we believe
2340 			 * it is tolerable, because this should be a rare case.
2341 			 */
2342 			lle = nd6_alloc(&dst->sin6_addr, 0, ifp);
2343 			if (lle == NULL) {
2344 				char ip6buf[INET6_ADDRSTRLEN];
2345 				log(LOG_DEBUG,
2346 				    "nd6_output: can't allocate llinfo for %s "
2347 				    "(ln=%p)\n",
2348 				    ip6_sprintf(ip6buf, &dst->sin6_addr), lle);
2349 				m_freem(m);
2350 				return (ENOBUFS);
2351 			}
2352 
2353 			IF_AFDATA_WLOCK(ifp);
2354 			LLE_WLOCK(lle);
2355 			/* Prefer any existing entry over newly-created one */
2356 			lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp);
2357 			if (lle_tmp == NULL)
2358 				lltable_link_entry(LLTABLE6(ifp), lle);
2359 			IF_AFDATA_WUNLOCK(ifp);
2360 			if (lle_tmp != NULL) {
2361 				lltable_free_entry(LLTABLE6(ifp), lle);
2362 				lle = lle_tmp;
2363 				lle_tmp = NULL;
2364 			}
2365 		}
2366 	}
2367 	if (lle == NULL) {
2368 		if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) {
2369 			m_freem(m);
2370 			return (ENOBUFS);
2371 		}
2372 
2373 		if (m != NULL)
2374 			m_freem(m);
2375 		return (ENOBUFS);
2376 	}
2377 
2378 	LLE_WLOCK_ASSERT(lle);
2379 
2380 	/*
2381 	 * The first time we send a packet to a neighbor whose entry is
2382 	 * STALE, we have to change the state to DELAY and a sets a timer to
2383 	 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2384 	 * neighbor unreachability detection on expiration.
2385 	 * (RFC 2461 7.3.3)
2386 	 */
2387 	if (lle->ln_state == ND6_LLINFO_STALE)
2388 		nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2389 
2390 	/*
2391 	 * If the neighbor cache entry has a state other than INCOMPLETE
2392 	 * (i.e. its link-layer address is already resolved), just
2393 	 * send the packet.
2394 	 */
2395 	if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2396 		if (flags & LLE_ADDRONLY) {
2397 			lladdr = lle->ll_addr;
2398 			ll_len = ifp->if_addrlen;
2399 		} else {
2400 			lladdr = lle->r_linkdata;
2401 			ll_len = lle->r_hdrlen;
2402 		}
2403 		bcopy(lladdr, desten, ll_len);
2404 		if (pflags != NULL)
2405 			*pflags = lle->la_flags;
2406 		if (plle) {
2407 			LLE_ADDREF(lle);
2408 			*plle = lle;
2409 		}
2410 		LLE_WUNLOCK(lle);
2411 		return (0);
2412 	}
2413 
2414 	/*
2415 	 * There is a neighbor cache entry, but no ethernet address
2416 	 * response yet.  Append this latest packet to the end of the
2417 	 * packet queue in the mbuf.  When it exceeds nd6_maxqueuelen,
2418 	 * the oldest packet in the queue will be removed.
2419 	 */
2420 
2421 	if (lle->la_hold != NULL) {
2422 		struct mbuf *m_hold;
2423 		int i;
2424 
2425 		i = 0;
2426 		for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){
2427 			i++;
2428 			if (m_hold->m_nextpkt == NULL) {
2429 				m_hold->m_nextpkt = m;
2430 				break;
2431 			}
2432 		}
2433 		while (i >= V_nd6_maxqueuelen) {
2434 			m_hold = lle->la_hold;
2435 			lle->la_hold = lle->la_hold->m_nextpkt;
2436 			m_freem(m_hold);
2437 			i--;
2438 		}
2439 	} else {
2440 		lle->la_hold = m;
2441 	}
2442 
2443 	/*
2444 	 * If there has been no NS for the neighbor after entering the
2445 	 * INCOMPLETE state, send the first solicitation.
2446 	 * Note that for newly-created lle la_asked will be 0,
2447 	 * so we will transition from ND6_LLINFO_NOSTATE to
2448 	 * ND6_LLINFO_INCOMPLETE state here.
2449 	 */
2450 	psrc = NULL;
2451 	send_ns = 0;
2452 	if (lle->la_asked == 0) {
2453 		lle->la_asked++;
2454 		send_ns = 1;
2455 		psrc = nd6_llinfo_get_holdsrc(lle, &src);
2456 
2457 		nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2458 	}
2459 	LLE_WUNLOCK(lle);
2460 	if (send_ns != 0)
2461 		nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2462 
2463 	return (EWOULDBLOCK);
2464 }
2465 
2466 /*
2467  * Do L2 address resolution for @sa_dst address. Stores found
2468  * address in @desten buffer. Copy of lle ln_flags can be also
2469  * saved in @pflags if @pflags is non-NULL.
2470  *
2471  * Return values:
2472  * - 0 on success (address copied to buffer).
2473  * - EWOULDBLOCK (no local error, but address is still unresolved)
2474  * - other errors (alloc failure, etc)
2475  */
2476 int
nd6_resolve_addr(struct ifnet * ifp,int flags,const struct sockaddr * dst,char * desten,uint32_t * pflags)2477 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2478     char *desten, uint32_t *pflags)
2479 {
2480 	int error;
2481 
2482 	flags |= LLE_ADDRONLY;
2483 	error = nd6_resolve_slow(ifp, flags, NULL,
2484 	    (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2485 	return (error);
2486 }
2487 
2488 int
nd6_flush_holdchain(struct ifnet * ifp,struct mbuf * chain,struct sockaddr_in6 * dst)2489 nd6_flush_holdchain(struct ifnet *ifp, struct mbuf *chain,
2490     struct sockaddr_in6 *dst)
2491 {
2492 	struct mbuf *m, *m_head;
2493 	int error = 0;
2494 
2495 	m_head = chain;
2496 
2497 	while (m_head) {
2498 		m = m_head;
2499 		m_head = m_head->m_nextpkt;
2500 		error = nd6_output_ifp(ifp, ifp, m, dst, NULL);
2501 	}
2502 
2503 	/*
2504 	 * XXX
2505 	 * note that intermediate errors are blindly ignored
2506 	 */
2507 	return (error);
2508 }
2509 
2510 static int
nd6_need_cache(struct ifnet * ifp)2511 nd6_need_cache(struct ifnet *ifp)
2512 {
2513 	/*
2514 	 * XXX: we currently do not make neighbor cache on any interface
2515 	 * other than Ethernet and GIF.
2516 	 *
2517 	 * RFC2893 says:
2518 	 * - unidirectional tunnels needs no ND
2519 	 */
2520 	switch (ifp->if_type) {
2521 	case IFT_ETHER:
2522 	case IFT_IEEE1394:
2523 	case IFT_L2VLAN:
2524 	case IFT_INFINIBAND:
2525 	case IFT_BRIDGE:
2526 	case IFT_PROPVIRTUAL:
2527 		return (1);
2528 	default:
2529 		return (0);
2530 	}
2531 }
2532 
2533 /*
2534  * Add pernament ND6 link-layer record for given
2535  * interface address.
2536  *
2537  * Very similar to IPv4 arp_ifinit(), but:
2538  * 1) IPv6 DAD is performed in different place
2539  * 2) It is called by IPv6 protocol stack in contrast to
2540  * arp_ifinit() which is typically called in SIOCSIFADDR
2541  * driver ioctl handler.
2542  *
2543  */
2544 int
nd6_add_ifa_lle(struct in6_ifaddr * ia)2545 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2546 {
2547 	struct ifnet *ifp;
2548 	struct llentry *ln, *ln_tmp;
2549 	struct sockaddr *dst;
2550 
2551 	ifp = ia->ia_ifa.ifa_ifp;
2552 	if (nd6_need_cache(ifp) == 0)
2553 		return (0);
2554 
2555 	ia->ia_ifa.ifa_rtrequest = nd6_rtrequest;
2556 	dst = (struct sockaddr *)&ia->ia_addr;
2557 	ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2558 	if (ln == NULL)
2559 		return (ENOBUFS);
2560 
2561 	IF_AFDATA_WLOCK(ifp);
2562 	LLE_WLOCK(ln);
2563 	/* Unlink any entry if exists */
2564 	ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst);
2565 	if (ln_tmp != NULL)
2566 		lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2567 	lltable_link_entry(LLTABLE6(ifp), ln);
2568 	IF_AFDATA_WUNLOCK(ifp);
2569 
2570 	if (ln_tmp != NULL)
2571 		EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2572 	EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2573 
2574 	LLE_WUNLOCK(ln);
2575 	if (ln_tmp != NULL)
2576 		llentry_free(ln_tmp);
2577 
2578 	return (0);
2579 }
2580 
2581 /*
2582  * Removes either all lle entries for given @ia, or lle
2583  * corresponding to @ia address.
2584  */
2585 void
nd6_rem_ifa_lle(struct in6_ifaddr * ia,int all)2586 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2587 {
2588 	struct sockaddr_in6 mask, addr;
2589 	struct sockaddr *saddr, *smask;
2590 	struct ifnet *ifp;
2591 
2592 	ifp = ia->ia_ifa.ifa_ifp;
2593 	memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2594 	memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2595 	saddr = (struct sockaddr *)&addr;
2596 	smask = (struct sockaddr *)&mask;
2597 
2598 	if (all != 0)
2599 		lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2600 	else
2601 		lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2602 }
2603 
2604 static void
clear_llinfo_pqueue(struct llentry * ln)2605 clear_llinfo_pqueue(struct llentry *ln)
2606 {
2607 	struct mbuf *m_hold, *m_hold_next;
2608 
2609 	for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) {
2610 		m_hold_next = m_hold->m_nextpkt;
2611 		m_freem(m_hold);
2612 	}
2613 
2614 	ln->la_hold = NULL;
2615 }
2616 
2617 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS);
2618 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS);
2619 
2620 SYSCTL_DECL(_net_inet6_icmp6);
2621 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist,
2622 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2623 	NULL, 0, nd6_sysctl_drlist, "S,in6_defrouter",
2624 	"NDP default router list");
2625 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2626 	CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2627 	NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2628 	"NDP prefix list");
2629 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2630 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2631 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2632 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2633 
2634 static int
nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)2635 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS)
2636 {
2637 	struct in6_defrouter d;
2638 	struct nd_defrouter *dr;
2639 	int error;
2640 
2641 	if (req->newptr != NULL)
2642 		return (EPERM);
2643 
2644 	error = sysctl_wire_old_buffer(req, 0);
2645 	if (error != 0)
2646 		return (error);
2647 
2648 	bzero(&d, sizeof(d));
2649 	d.rtaddr.sin6_family = AF_INET6;
2650 	d.rtaddr.sin6_len = sizeof(d.rtaddr);
2651 
2652 	ND6_RLOCK();
2653 	TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) {
2654 		d.rtaddr.sin6_addr = dr->rtaddr;
2655 		error = sa6_recoverscope(&d.rtaddr);
2656 		if (error != 0)
2657 			break;
2658 		d.flags = dr->raflags;
2659 		d.rtlifetime = dr->rtlifetime;
2660 		d.expire = dr->expire + (time_second - time_uptime);
2661 		d.if_index = dr->ifp->if_index;
2662 		error = SYSCTL_OUT(req, &d, sizeof(d));
2663 		if (error != 0)
2664 			break;
2665 	}
2666 	ND6_RUNLOCK();
2667 	return (error);
2668 }
2669 
2670 static int
nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)2671 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2672 {
2673 	struct in6_prefix p;
2674 	struct sockaddr_in6 s6;
2675 	struct nd_prefix *pr;
2676 	struct nd_pfxrouter *pfr;
2677 	time_t maxexpire;
2678 	int error;
2679 	char ip6buf[INET6_ADDRSTRLEN];
2680 
2681 	if (req->newptr)
2682 		return (EPERM);
2683 
2684 	error = sysctl_wire_old_buffer(req, 0);
2685 	if (error != 0)
2686 		return (error);
2687 
2688 	bzero(&p, sizeof(p));
2689 	p.origin = PR_ORIG_RA;
2690 	bzero(&s6, sizeof(s6));
2691 	s6.sin6_family = AF_INET6;
2692 	s6.sin6_len = sizeof(s6);
2693 
2694 	ND6_RLOCK();
2695 	LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2696 		p.prefix = pr->ndpr_prefix;
2697 		if (sa6_recoverscope(&p.prefix)) {
2698 			log(LOG_ERR, "scope error in prefix list (%s)\n",
2699 			    ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2700 			/* XXX: press on... */
2701 		}
2702 		p.raflags = pr->ndpr_raf;
2703 		p.prefixlen = pr->ndpr_plen;
2704 		p.vltime = pr->ndpr_vltime;
2705 		p.pltime = pr->ndpr_pltime;
2706 		p.if_index = pr->ndpr_ifp->if_index;
2707 		if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2708 			p.expire = 0;
2709 		else {
2710 			/* XXX: we assume time_t is signed. */
2711 			maxexpire = (-1) &
2712 			    ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2713 			if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2714 				p.expire = pr->ndpr_lastupdate +
2715 				    pr->ndpr_vltime +
2716 				    (time_second - time_uptime);
2717 			else
2718 				p.expire = maxexpire;
2719 		}
2720 		p.refcnt = pr->ndpr_addrcnt;
2721 		p.flags = pr->ndpr_stateflags;
2722 		p.advrtrs = 0;
2723 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2724 			p.advrtrs++;
2725 		error = SYSCTL_OUT(req, &p, sizeof(p));
2726 		if (error != 0)
2727 			break;
2728 		LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2729 			s6.sin6_addr = pfr->router->rtaddr;
2730 			if (sa6_recoverscope(&s6))
2731 				log(LOG_ERR,
2732 				    "scope error in prefix list (%s)\n",
2733 				    ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2734 			error = SYSCTL_OUT(req, &s6, sizeof(s6));
2735 			if (error != 0)
2736 				goto out;
2737 		}
2738 	}
2739 out:
2740 	ND6_RUNLOCK();
2741 	return (error);
2742 }
2743