1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the project nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32 */
33
34 #include <sys/cdefs.h>
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_route.h"
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/eventhandler.h>
42 #include <sys/callout.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/mutex.h>
47 #include <sys/socket.h>
48 #include <sys/sockio.h>
49 #include <sys/time.h>
50 #include <sys/kernel.h>
51 #include <sys/protosw.h>
52 #include <sys/errno.h>
53 #include <sys/syslog.h>
54 #include <sys/rwlock.h>
55 #include <sys/queue.h>
56 #include <sys/sdt.h>
57 #include <sys/sysctl.h>
58
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/if_dl.h>
62 #include <net/if_private.h>
63 #include <net/if_types.h>
64 #include <net/route.h>
65 #include <net/route/route_ctl.h>
66 #include <net/route/nhop.h>
67 #include <net/vnet.h>
68
69 #include <netinet/in.h>
70 #include <netinet/in_kdtrace.h>
71 #include <net/if_llatbl.h>
72 #include <netinet/if_ether.h>
73 #include <netinet6/in6_fib.h>
74 #include <netinet6/in6_var.h>
75 #include <netinet/ip6.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/nd6.h>
79 #include <netinet6/in6_ifattach.h>
80 #include <netinet/icmp6.h>
81 #include <netinet6/send.h>
82
83 #include <sys/limits.h>
84
85 #include <security/mac/mac_framework.h>
86
87 #define ND6_PREFIX_WITH_ROUTER(pr) !LIST_EMPTY(&(pr)->ndpr_advrtrs)
88
89 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
90 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
91
92 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
93
94 VNET_DEFINE_STATIC(int, nd6_prune) = 1;
95 #define V_nd6_prune VNET(nd6_prune)
96 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_PRUNE, nd6_prune,
97 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_prune), 0,
98 "Frequency in seconds of checks for expired prefixes and routers");
99
100 VNET_DEFINE_STATIC(int, nd6_delay) = 5;
101 #define V_nd6_delay VNET(nd6_delay)
102 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DELAY, nd6_delay,
103 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_delay), 0,
104 "Delay in seconds before probing for reachability");
105
106 VNET_DEFINE_STATIC(int, nd6_umaxtries) = 3;
107 #define V_nd6_umaxtries VNET(nd6_umaxtries)
108 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_UMAXTRIES, nd6_umaxtries,
109 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_umaxtries), 0,
110 "Number of ICMPv6 NS messages sent during reachability detection");
111
112 VNET_DEFINE(int, nd6_mmaxtries) = 3;
113 #define V_nd6_mmaxtries VNET(nd6_mmaxtries)
114 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MMAXTRIES, nd6_mmaxtries,
115 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_mmaxtries), 0,
116 "Number of ICMPv6 NS messages sent during address resolution");
117
118 VNET_DEFINE_STATIC(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage
119 * collection timer */
120 #define V_nd6_gctimer VNET(nd6_gctimer)
121
122 /* preventing too many loops in ND option parsing */
123 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
124
125 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 16; /* max pkts cached in unresolved
126 * ND entries */
127 #define V_nd6_maxndopt VNET(nd6_maxndopt)
128 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen)
129
130 #ifdef ND6_DEBUG
131 VNET_DEFINE(int, nd6_debug) = 1;
132 #else
133 VNET_DEFINE(int, nd6_debug) = 0;
134 #endif
135 #define V_nd6_debug VNET(nd6_debug)
136 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DEBUG, nd6_debug,
137 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_debug), 0,
138 "Log NDP debug messages");
139
140 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
141
142 VNET_DEFINE(struct nd_prhead, nd_prefix);
143 VNET_DEFINE(struct rwlock, nd6_lock);
144 VNET_DEFINE(uint64_t, nd6_list_genid);
145 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
146
147 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
148 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval)
149
150 int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
151
152 static bool nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
153 struct ifnet *);
154 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
155 static void nd6_slowtimo(void *);
156 static int regen_tmpaddr(struct in6_ifaddr *);
157 static void nd6_free(struct llentry **, int);
158 static void nd6_free_redirect(const struct llentry *);
159 static void nd6_llinfo_timer(void *);
160 static void nd6_llinfo_settimer_locked(struct llentry *, long);
161 static int nd6_resolve_slow(struct ifnet *, int, int, struct mbuf *,
162 const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
163 static int nd6_need_cache(struct ifnet *);
164
165 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
166 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
167
168 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
169 #define V_nd6_timer_ch VNET(nd6_timer_ch)
170
171 static void
nd6_lle_event(void * arg __unused,struct llentry * lle,int evt)172 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
173 {
174 struct rt_addrinfo rtinfo;
175 struct sockaddr_in6 dst;
176 struct sockaddr_dl gw;
177 struct ifnet *ifp;
178 int type;
179 int fibnum;
180
181 LLE_WLOCK_ASSERT(lle);
182
183 if (lltable_get_af(lle->lle_tbl) != AF_INET6)
184 return;
185
186 switch (evt) {
187 case LLENTRY_RESOLVED:
188 type = RTM_ADD;
189 KASSERT(lle->la_flags & LLE_VALID,
190 ("%s: %p resolved but not valid?", __func__, lle));
191 break;
192 case LLENTRY_EXPIRED:
193 type = RTM_DELETE;
194 break;
195 default:
196 return;
197 }
198
199 ifp = lltable_get_ifp(lle->lle_tbl);
200
201 bzero(&dst, sizeof(dst));
202 bzero(&gw, sizeof(gw));
203 bzero(&rtinfo, sizeof(rtinfo));
204 lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
205 dst.sin6_scope_id = in6_getscopezone(ifp,
206 in6_addrscope(&dst.sin6_addr));
207 gw.sdl_len = sizeof(struct sockaddr_dl);
208 gw.sdl_family = AF_LINK;
209 gw.sdl_alen = ifp->if_addrlen;
210 gw.sdl_index = ifp->if_index;
211 gw.sdl_type = ifp->if_type;
212 if (evt == LLENTRY_RESOLVED)
213 bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
214 rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
215 rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
216 rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
217 fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
218 rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
219 type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
220 }
221
222 /*
223 * A handler for interface link layer address change event.
224 */
225 static void
nd6_iflladdr(void * arg __unused,struct ifnet * ifp)226 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
227 {
228 if (ifp->if_afdata[AF_INET6] == NULL)
229 return;
230
231 lltable_update_ifaddr(LLTABLE6(ifp));
232 }
233
234 void
nd6_init(void)235 nd6_init(void)
236 {
237
238 mtx_init(&V_nd6_onlink_mtx, "nd6 onlink", NULL, MTX_DEF);
239 rw_init(&V_nd6_lock, "nd6 list");
240
241 LIST_INIT(&V_nd_prefix);
242 nd6_defrouter_init();
243
244 /* Start timers. */
245 callout_init(&V_nd6_slowtimo_ch, 1);
246 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
247 nd6_slowtimo, curvnet);
248
249 callout_init(&V_nd6_timer_ch, 1);
250 callout_reset(&V_nd6_timer_ch, hz, nd6_timer, curvnet);
251
252 nd6_dad_init();
253 if (IS_DEFAULT_VNET(curvnet)) {
254 lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event,
255 NULL, EVENTHANDLER_PRI_ANY);
256 iflladdr_event_eh = EVENTHANDLER_REGISTER(iflladdr_event,
257 nd6_iflladdr, NULL, EVENTHANDLER_PRI_ANY);
258 ifnet_link_event_eh = EVENTHANDLER_REGISTER(ifnet_link_event,
259 nd6_ifnet_link_event, NULL, EVENTHANDLER_PRI_ANY);
260 }
261 }
262
263 #ifdef VIMAGE
264 void
nd6_destroy(void)265 nd6_destroy(void)
266 {
267
268 callout_drain(&V_nd6_slowtimo_ch);
269 callout_drain(&V_nd6_timer_ch);
270 if (IS_DEFAULT_VNET(curvnet)) {
271 EVENTHANDLER_DEREGISTER(ifnet_link_event, ifnet_link_event_eh);
272 EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh);
273 EVENTHANDLER_DEREGISTER(iflladdr_event, iflladdr_event_eh);
274 }
275 rw_destroy(&V_nd6_lock);
276 mtx_destroy(&V_nd6_onlink_mtx);
277 }
278 #endif
279
280 struct nd_ifinfo *
nd6_ifattach(struct ifnet * ifp)281 nd6_ifattach(struct ifnet *ifp)
282 {
283 struct nd_ifinfo *nd;
284
285 nd = malloc(sizeof(*nd), M_IP6NDP, M_WAITOK | M_ZERO);
286 nd->initialized = 1;
287
288 nd->chlim = IPV6_DEFHLIM;
289 nd->basereachable = REACHABLE_TIME;
290 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable);
291 nd->retrans = RETRANS_TIMER;
292
293 nd->flags = ND6_IFF_PERFORMNUD;
294
295 /* Set IPv6 disabled on all interfaces but loopback by default. */
296 if ((ifp->if_flags & IFF_LOOPBACK) == 0)
297 nd->flags |= ND6_IFF_IFDISABLED;
298
299 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL.
300 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by
301 * default regardless of the V_ip6_auto_linklocal configuration to
302 * give a reasonable default behavior.
303 */
304 if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE &&
305 ifp->if_type != IFT_WIREGUARD) || (ifp->if_flags & IFF_LOOPBACK))
306 nd->flags |= ND6_IFF_AUTO_LINKLOCAL;
307 /*
308 * A loopback interface does not need to accept RTADV.
309 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by
310 * default regardless of the V_ip6_accept_rtadv configuration to
311 * prevent the interface from accepting RA messages arrived
312 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV.
313 */
314 if (V_ip6_accept_rtadv &&
315 !(ifp->if_flags & IFF_LOOPBACK) &&
316 (ifp->if_type != IFT_BRIDGE)) {
317 nd->flags |= ND6_IFF_ACCEPT_RTADV;
318 /* If we globally accept rtadv, assume IPv6 on. */
319 nd->flags &= ~ND6_IFF_IFDISABLED;
320 }
321 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK))
322 nd->flags |= ND6_IFF_NO_RADR;
323
324 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */
325 nd6_setmtu0(ifp, nd);
326
327 /* Configure default value for stable addresses algorithm, skip loopback interface */
328 if (V_ip6_use_stableaddr && !(ifp->if_flags & IFF_LOOPBACK)) {
329 nd->flags |= ND6_IFF_STABLEADDR;
330 }
331
332 return nd;
333 }
334
335 void
nd6_ifdetach(struct ifnet * ifp,struct nd_ifinfo * nd)336 nd6_ifdetach(struct ifnet *ifp, struct nd_ifinfo *nd)
337 {
338 struct epoch_tracker et;
339 struct ifaddr *ifa, *next;
340
341 NET_EPOCH_ENTER(et);
342 CK_STAILQ_FOREACH_SAFE(ifa, &ifp->if_addrhead, ifa_link, next) {
343 if (ifa->ifa_addr->sa_family != AF_INET6)
344 continue;
345
346 /* stop DAD processing */
347 nd6_dad_stop(ifa);
348 }
349 NET_EPOCH_EXIT(et);
350
351 free(nd, M_IP6NDP);
352 }
353
354 /*
355 * Reset ND level link MTU. This function is called when the physical MTU
356 * changes, which means we might have to adjust the ND level MTU.
357 */
358 void
nd6_setmtu(struct ifnet * ifp)359 nd6_setmtu(struct ifnet *ifp)
360 {
361 if (ifp->if_afdata[AF_INET6] == NULL)
362 return;
363
364 nd6_setmtu0(ifp, ND_IFINFO(ifp));
365 }
366
367 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
368 void
nd6_setmtu0(struct ifnet * ifp,struct nd_ifinfo * ndi)369 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
370 {
371 u_int32_t omaxmtu;
372
373 omaxmtu = ndi->maxmtu;
374 ndi->maxmtu = ifp->if_mtu;
375
376 /*
377 * Decreasing the interface MTU under IPV6 minimum MTU may cause
378 * undesirable situation. We thus notify the operator of the change
379 * explicitly. The check for omaxmtu is necessary to restrict the
380 * log to the case of changing the MTU, not initializing it.
381 */
382 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
383 log(LOG_NOTICE, "nd6_setmtu0: "
384 "new link MTU on %s (%lu) is too small for IPv6\n",
385 if_name(ifp), (unsigned long)ndi->maxmtu);
386 }
387 }
388
389 void
nd6_option_init(void * opt,int icmp6len,union nd_opts * ndopts)390 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
391 {
392
393 bzero(ndopts, sizeof(*ndopts));
394 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
395 ndopts->nd_opts_last
396 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
397
398 if (icmp6len == 0) {
399 ndopts->nd_opts_done = 1;
400 ndopts->nd_opts_search = NULL;
401 }
402 }
403
404 /*
405 * Take one ND option.
406 */
407 struct nd_opt_hdr *
nd6_option(union nd_opts * ndopts)408 nd6_option(union nd_opts *ndopts)
409 {
410 struct nd_opt_hdr *nd_opt;
411 int olen;
412
413 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
414 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
415 __func__));
416 if (ndopts->nd_opts_search == NULL)
417 return NULL;
418 if (ndopts->nd_opts_done)
419 return NULL;
420
421 nd_opt = ndopts->nd_opts_search;
422
423 /* make sure nd_opt_len is inside the buffer */
424 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
425 bzero(ndopts, sizeof(*ndopts));
426 return NULL;
427 }
428
429 olen = nd_opt->nd_opt_len << 3;
430 if (olen == 0) {
431 /*
432 * Message validation requires that all included
433 * options have a length that is greater than zero.
434 */
435 bzero(ndopts, sizeof(*ndopts));
436 return NULL;
437 }
438
439 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
440 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
441 /* option overruns the end of buffer, invalid */
442 bzero(ndopts, sizeof(*ndopts));
443 return NULL;
444 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
445 /* reached the end of options chain */
446 ndopts->nd_opts_done = 1;
447 ndopts->nd_opts_search = NULL;
448 }
449 return nd_opt;
450 }
451
452 /*
453 * Parse multiple ND options.
454 * This function is much easier to use, for ND routines that do not need
455 * multiple options of the same type.
456 */
457 int
nd6_options(union nd_opts * ndopts)458 nd6_options(union nd_opts *ndopts)
459 {
460 struct nd_opt_hdr *nd_opt;
461 int i = 0;
462
463 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
464 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
465 __func__));
466 if (ndopts->nd_opts_search == NULL)
467 return 0;
468
469 while (1) {
470 nd_opt = nd6_option(ndopts);
471 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
472 /*
473 * Message validation requires that all included
474 * options have a length that is greater than zero.
475 */
476 ICMP6STAT_INC(icp6s_nd_badopt);
477 bzero(ndopts, sizeof(*ndopts));
478 return -1;
479 }
480
481 if (nd_opt == NULL)
482 goto skip1;
483
484 switch (nd_opt->nd_opt_type) {
485 case ND_OPT_SOURCE_LINKADDR:
486 case ND_OPT_TARGET_LINKADDR:
487 case ND_OPT_MTU:
488 case ND_OPT_REDIRECTED_HEADER:
489 case ND_OPT_NONCE:
490 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
491 nd6log((LOG_INFO,
492 "duplicated ND6 option found (type=%d)\n",
493 nd_opt->nd_opt_type));
494 /* XXX bark? */
495 } else {
496 ndopts->nd_opt_array[nd_opt->nd_opt_type]
497 = nd_opt;
498 }
499 break;
500 case ND_OPT_PREFIX_INFORMATION:
501 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
502 ndopts->nd_opt_array[nd_opt->nd_opt_type]
503 = nd_opt;
504 }
505 ndopts->nd_opts_pi_end =
506 (struct nd_opt_prefix_info *)nd_opt;
507 break;
508 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */
509 case ND_OPT_RDNSS: /* RFC 6106 */
510 case ND_OPT_DNSSL: /* RFC 6106 */
511 /*
512 * Silently ignore options we know and do not care about
513 * in the kernel.
514 */
515 break;
516 default:
517 /*
518 * Unknown options must be silently ignored,
519 * to accommodate future extension to the protocol.
520 */
521 nd6log((LOG_DEBUG,
522 "nd6_options: unsupported option %d - "
523 "option ignored\n", nd_opt->nd_opt_type));
524 }
525
526 skip1:
527 i++;
528 if (i > V_nd6_maxndopt) {
529 ICMP6STAT_INC(icp6s_nd_toomanyopt);
530 nd6log((LOG_INFO, "too many loop in nd opt\n"));
531 break;
532 }
533
534 if (ndopts->nd_opts_done)
535 break;
536 }
537
538 return 0;
539 }
540
541 /*
542 * ND6 timer routine to handle ND6 entries
543 */
544 static void
nd6_llinfo_settimer_locked(struct llentry * ln,long tick)545 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
546 {
547 int canceled;
548
549 LLE_WLOCK_ASSERT(ln);
550
551 /* Do not schedule timers for child LLEs. */
552 if (ln->la_flags & LLE_CHILD)
553 return;
554
555 if (tick < 0) {
556 ln->la_expire = 0;
557 ln->ln_ntick = 0;
558 canceled = callout_stop(&ln->lle_timer);
559 } else {
560 ln->la_expire = time_uptime + tick / hz;
561 LLE_ADDREF(ln);
562 if (tick > INT_MAX) {
563 ln->ln_ntick = tick - INT_MAX;
564 canceled = callout_reset(&ln->lle_timer, INT_MAX,
565 nd6_llinfo_timer, ln);
566 } else {
567 ln->ln_ntick = 0;
568 canceled = callout_reset(&ln->lle_timer, tick,
569 nd6_llinfo_timer, ln);
570 }
571 }
572 if (canceled > 0)
573 LLE_REMREF(ln);
574 }
575
576 /*
577 * Gets source address of the first packet in hold queue
578 * and stores it in @src.
579 * Returns pointer to @src (if hold queue is not empty) or NULL.
580 *
581 * Set noinline to be dtrace-friendly
582 */
583 static __noinline struct in6_addr *
nd6_llinfo_get_holdsrc(struct llentry * ln,struct in6_addr * src)584 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
585 {
586 struct ip6_hdr hdr;
587 struct mbuf *m;
588
589 if (ln->la_hold == NULL)
590 return (NULL);
591
592 /*
593 * assume every packet in la_hold has the same IP header
594 */
595 m = ln->la_hold;
596 if (sizeof(hdr) > m->m_len)
597 return (NULL);
598
599 m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
600 *src = hdr.ip6_src;
601
602 return (src);
603 }
604
605 /*
606 * Checks if we need to switch from STALE state.
607 *
608 * RFC 4861 requires switching from STALE to DELAY state
609 * on first packet matching entry, waiting V_nd6_delay and
610 * transition to PROBE state (if upper layer confirmation was
611 * not received).
612 *
613 * This code performs a bit differently:
614 * On packet hit we don't change state (but desired state
615 * can be guessed by control plane). However, after V_nd6_delay
616 * seconds code will transition to PROBE state (so DELAY state
617 * is kinda skipped in most situations).
618 *
619 * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
620 * we perform the following upon entering STALE state:
621 *
622 * 1) Arm timer to run each V_nd6_delay seconds to make sure that
623 * if packet was transmitted at the start of given interval, we
624 * would be able to switch to PROBE state in V_nd6_delay seconds
625 * as user expects.
626 *
627 * 2) Reschedule timer until original V_nd6_gctimer expires keeping
628 * lle in STALE state (remaining timer value stored in lle_remtime).
629 *
630 * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
631 * seconds ago.
632 *
633 * Returns non-zero value if the entry is still STALE (storing
634 * the next timer interval in @pdelay).
635 *
636 * Returns zero value if original timer expired or we need to switch to
637 * PROBE (store that in @do_switch variable).
638 */
639 static int
nd6_is_stale(struct llentry * lle,long * pdelay,int * do_switch)640 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
641 {
642 int nd_delay, nd_gctimer;
643 time_t lle_hittime;
644 long delay;
645
646 *do_switch = 0;
647 nd_gctimer = V_nd6_gctimer;
648 nd_delay = V_nd6_delay;
649
650 lle_hittime = llentry_get_hittime(lle);
651
652 if (lle_hittime == 0) {
653 /*
654 * Datapath feedback has been requested upon entering
655 * STALE state. No packets has been passed using this lle.
656 * Ask for the timer reschedule and keep STALE state.
657 */
658 delay = (long)(MIN(nd_gctimer, nd_delay));
659 delay *= hz;
660 if (lle->lle_remtime > delay)
661 lle->lle_remtime -= delay;
662 else {
663 delay = lle->lle_remtime;
664 lle->lle_remtime = 0;
665 }
666
667 if (delay == 0) {
668 /*
669 * The original ng6_gctime timeout ended,
670 * no more rescheduling.
671 */
672 return (0);
673 }
674
675 *pdelay = delay;
676 return (1);
677 }
678
679 /*
680 * Packet received. Verify timestamp
681 */
682 delay = (long)(time_uptime - lle_hittime);
683 if (delay < nd_delay) {
684 /*
685 * V_nd6_delay still not passed since the first
686 * hit in STALE state.
687 * Reschedule timer and return.
688 */
689 *pdelay = (long)(nd_delay - delay) * hz;
690 return (1);
691 }
692
693 /* Request switching to probe */
694 *do_switch = 1;
695 return (0);
696 }
697
698 /*
699 * Switch @lle state to new state optionally arming timers.
700 *
701 * Set noinline to be dtrace-friendly
702 */
703 __noinline void
nd6_llinfo_setstate(struct llentry * lle,int newstate)704 nd6_llinfo_setstate(struct llentry *lle, int newstate)
705 {
706 struct ifnet *ifp;
707 int nd_gctimer, nd_delay;
708 long delay, remtime;
709
710 delay = 0;
711 remtime = 0;
712
713 switch (newstate) {
714 case ND6_LLINFO_INCOMPLETE:
715 ifp = lle->lle_tbl->llt_ifp;
716 delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
717 break;
718 case ND6_LLINFO_REACHABLE:
719 if (!ND6_LLINFO_PERMANENT(lle)) {
720 ifp = lle->lle_tbl->llt_ifp;
721 delay = (long)ND_IFINFO(ifp)->reachable * hz;
722 }
723 break;
724 case ND6_LLINFO_STALE:
725
726 llentry_request_feedback(lle);
727 nd_delay = V_nd6_delay;
728 nd_gctimer = V_nd6_gctimer;
729
730 delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
731 remtime = (long)nd_gctimer * hz - delay;
732 break;
733 case ND6_LLINFO_DELAY:
734 lle->la_asked = 0;
735 delay = (long)V_nd6_delay * hz;
736 break;
737 }
738
739 if (delay > 0)
740 nd6_llinfo_settimer_locked(lle, delay);
741
742 lle->lle_remtime = remtime;
743 lle->ln_state = newstate;
744 }
745
746 /*
747 * Timer-dependent part of nd state machine.
748 *
749 * Set noinline to be dtrace-friendly
750 */
751 static __noinline void
nd6_llinfo_timer(void * arg)752 nd6_llinfo_timer(void *arg)
753 {
754 struct epoch_tracker et;
755 struct llentry *ln;
756 struct in6_addr *dst, *pdst, *psrc, src;
757 struct ifnet *ifp;
758 struct nd_ifinfo *ndi;
759 int do_switch, send_ns;
760 long delay;
761
762 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
763 ln = (struct llentry *)arg;
764 ifp = lltable_get_ifp(ln->lle_tbl);
765 CURVNET_SET(ifp->if_vnet);
766
767 ND6_RLOCK();
768 LLE_WLOCK(ln);
769 if (callout_pending(&ln->lle_timer)) {
770 /*
771 * Here we are a bit odd here in the treatment of
772 * active/pending. If the pending bit is set, it got
773 * rescheduled before I ran. The active
774 * bit we ignore, since if it was stopped
775 * in ll_tablefree() and was currently running
776 * it would have return 0 so the code would
777 * not have deleted it since the callout could
778 * not be stopped so we want to go through
779 * with the delete here now. If the callout
780 * was restarted, the pending bit will be back on and
781 * we just want to bail since the callout_reset would
782 * return 1 and our reference would have been removed
783 * by nd6_llinfo_settimer_locked above since canceled
784 * would have been 1.
785 */
786 LLE_WUNLOCK(ln);
787 ND6_RUNLOCK();
788 CURVNET_RESTORE();
789 return;
790 }
791 NET_EPOCH_ENTER(et);
792 ndi = ND_IFINFO(ifp);
793 send_ns = 0;
794 dst = &ln->r_l3addr.addr6;
795 pdst = dst;
796
797 if (ln->ln_ntick > 0) {
798 if (ln->ln_ntick > INT_MAX) {
799 ln->ln_ntick -= INT_MAX;
800 nd6_llinfo_settimer_locked(ln, INT_MAX);
801 } else {
802 ln->ln_ntick = 0;
803 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
804 }
805 goto done;
806 }
807
808 if (ln->la_flags & LLE_STATIC) {
809 goto done;
810 }
811
812 if (ln->la_flags & LLE_DELETED) {
813 nd6_free(&ln, 0);
814 goto done;
815 }
816
817 switch (ln->ln_state) {
818 case ND6_LLINFO_INCOMPLETE:
819 if (ln->la_asked < V_nd6_mmaxtries) {
820 ln->la_asked++;
821 send_ns = 1;
822 /* Send NS to multicast address */
823 pdst = NULL;
824 } else {
825 struct mbuf *m;
826
827 ICMP6STAT_ADD(icp6s_dropped, ln->la_numheld);
828
829 m = ln->la_hold;
830 if (m != NULL) {
831 /*
832 * assuming every packet in la_hold has the
833 * same IP header. Send error after unlock.
834 */
835 ln->la_hold = m->m_nextpkt;
836 m->m_nextpkt = NULL;
837 ln->la_numheld--;
838 }
839 nd6_free(&ln, 0);
840 if (m != NULL) {
841 struct mbuf *n = m;
842
843 /*
844 * if there are any ummapped mbufs, we
845 * must free them, rather than using
846 * them for an ICMP, as they cannot be
847 * checksummed.
848 */
849 while ((n = n->m_next) != NULL) {
850 if (n->m_flags & M_EXTPG)
851 break;
852 }
853 if (n != NULL) {
854 m_freem(m);
855 m = NULL;
856 } else {
857 icmp6_error2(m, ICMP6_DST_UNREACH,
858 ICMP6_DST_UNREACH_ADDR, 0, ifp);
859 }
860 }
861 }
862 break;
863 case ND6_LLINFO_REACHABLE:
864 if (!ND6_LLINFO_PERMANENT(ln))
865 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
866 break;
867
868 case ND6_LLINFO_STALE:
869 if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
870 /*
871 * No packet has used this entry and GC timeout
872 * has not been passed. Reschedule timer and
873 * return.
874 */
875 nd6_llinfo_settimer_locked(ln, delay);
876 break;
877 }
878
879 if (do_switch == 0) {
880 /*
881 * GC timer has ended and entry hasn't been used.
882 * Run Garbage collector (RFC 4861, 5.3)
883 */
884 if (!ND6_LLINFO_PERMANENT(ln))
885 nd6_free(&ln, 1);
886 break;
887 }
888
889 /* Entry has been used AND delay timer has ended. */
890
891 /* FALLTHROUGH */
892
893 case ND6_LLINFO_DELAY:
894 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
895 /* We need NUD */
896 ln->la_asked = 1;
897 nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
898 send_ns = 1;
899 } else
900 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
901 break;
902 case ND6_LLINFO_PROBE:
903 if (ln->la_asked < V_nd6_umaxtries) {
904 ln->la_asked++;
905 send_ns = 1;
906 } else {
907 nd6_free(&ln, 0);
908 }
909 break;
910 default:
911 panic("%s: paths in a dark night can be confusing: %d",
912 __func__, ln->ln_state);
913 }
914 done:
915 if (ln != NULL)
916 ND6_RUNLOCK();
917 if (send_ns != 0) {
918 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
919 psrc = nd6_llinfo_get_holdsrc(ln, &src);
920 LLE_FREE_LOCKED(ln);
921 ln = NULL;
922 nd6_ns_output(ifp, psrc, pdst, dst, NULL);
923 }
924
925 if (ln != NULL)
926 LLE_FREE_LOCKED(ln);
927 NET_EPOCH_EXIT(et);
928 CURVNET_RESTORE();
929 }
930
931 /*
932 * ND6 timer routine to expire default route list and prefix list
933 */
934 void
nd6_timer(void * arg)935 nd6_timer(void *arg)
936 {
937 CURVNET_SET((struct vnet *) arg);
938 struct epoch_tracker et;
939 struct nd_prhead prl;
940 struct nd_prefix *pr, *npr;
941 struct ifnet *ifp;
942 struct in6_ifaddr *ia6, *nia6;
943 uint64_t genid;
944
945 LIST_INIT(&prl);
946
947 NET_EPOCH_ENTER(et);
948 nd6_defrouter_timer();
949
950 /*
951 * expire interface addresses.
952 * in the past the loop was inside prefix expiry processing.
953 * However, from a stricter speci-confrmance standpoint, we should
954 * rather separate address lifetimes and prefix lifetimes.
955 *
956 * XXXRW: in6_ifaddrhead locking.
957 */
958 addrloop:
959 CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
960 /* check address lifetime */
961 if (IFA6_IS_INVALID(ia6)) {
962 int regen = 0;
963
964 /*
965 * If the expiring address is temporary, try
966 * regenerating a new one. This would be useful when
967 * we suspended a laptop PC, then turned it on after a
968 * period that could invalidate all temporary
969 * addresses. Although we may have to restart the
970 * loop (see below), it must be after purging the
971 * address. Otherwise, we'd see an infinite loop of
972 * regeneration.
973 */
974 if (V_ip6_use_tempaddr &&
975 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
976 if (regen_tmpaddr(ia6) == 0)
977 regen = 1;
978 }
979
980 in6_purgeaddr(&ia6->ia_ifa);
981
982 if (regen)
983 goto addrloop; /* XXX: see below */
984 } else if (IFA6_IS_DEPRECATED(ia6)) {
985 int oldflags = ia6->ia6_flags;
986
987 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
988
989 /*
990 * If a temporary address has just become deprecated,
991 * regenerate a new one if possible.
992 */
993 if (V_ip6_use_tempaddr &&
994 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
995 (oldflags & IN6_IFF_DEPRECATED) == 0) {
996 if (regen_tmpaddr(ia6) == 0) {
997 /*
998 * A new temporary address is
999 * generated.
1000 * XXX: this means the address chain
1001 * has changed while we are still in
1002 * the loop. Although the change
1003 * would not cause disaster (because
1004 * it's not a deletion, but an
1005 * addition,) we'd rather restart the
1006 * loop just for safety. Or does this
1007 * significantly reduce performance??
1008 */
1009 goto addrloop;
1010 }
1011 }
1012 } else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
1013 /*
1014 * Schedule DAD for a tentative address. This happens
1015 * if the interface was down or not running
1016 * when the address was configured.
1017 */
1018 int delay;
1019
1020 delay = arc4random() %
1021 (MAX_RTR_SOLICITATION_DELAY * hz);
1022 nd6_dad_start((struct ifaddr *)ia6, delay);
1023 } else {
1024 /*
1025 * Check status of the interface. If it is down,
1026 * mark the address as tentative for future DAD.
1027 */
1028 ifp = ia6->ia_ifp;
1029 if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1030 ((ifp->if_flags & IFF_UP) == 0 ||
1031 (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1032 (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1033 ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1034 ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1035 }
1036
1037 /*
1038 * A new RA might have made a deprecated address
1039 * preferred.
1040 */
1041 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1042 }
1043 }
1044 NET_EPOCH_EXIT(et);
1045
1046 ND6_WLOCK();
1047 restart:
1048 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1049 /*
1050 * Expire prefixes. Since the pltime is only used for
1051 * autoconfigured addresses, pltime processing for prefixes is
1052 * not necessary.
1053 *
1054 * Only unlink after all derived addresses have expired. This
1055 * may not occur until two hours after the prefix has expired
1056 * per RFC 4862. If the prefix expires before its derived
1057 * addresses, mark it off-link. This will be done automatically
1058 * after unlinking if no address references remain.
1059 */
1060 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1061 time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1062 continue;
1063
1064 if (pr->ndpr_addrcnt == 0) {
1065 nd6_prefix_unlink(pr, &prl);
1066 continue;
1067 }
1068 if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1069 genid = V_nd6_list_genid;
1070 nd6_prefix_ref(pr);
1071 ND6_WUNLOCK();
1072 ND6_ONLINK_LOCK();
1073 (void)nd6_prefix_offlink(pr);
1074 ND6_ONLINK_UNLOCK();
1075 ND6_WLOCK();
1076 nd6_prefix_rele(pr);
1077 if (genid != V_nd6_list_genid)
1078 goto restart;
1079 }
1080 }
1081 ND6_WUNLOCK();
1082
1083 while ((pr = LIST_FIRST(&prl)) != NULL) {
1084 LIST_REMOVE(pr, ndpr_entry);
1085 nd6_prefix_del(pr);
1086 }
1087
1088 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1089 nd6_timer, curvnet);
1090
1091 CURVNET_RESTORE();
1092 }
1093
1094 /*
1095 * ia6 - deprecated/invalidated temporary address
1096 */
1097 static int
regen_tmpaddr(struct in6_ifaddr * ia6)1098 regen_tmpaddr(struct in6_ifaddr *ia6)
1099 {
1100 struct ifaddr *ifa;
1101 struct ifnet *ifp;
1102 struct in6_ifaddr *public_ifa6 = NULL;
1103
1104 NET_EPOCH_ASSERT();
1105
1106 ifp = ia6->ia_ifa.ifa_ifp;
1107 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1108 struct in6_ifaddr *it6;
1109
1110 if (ifa->ifa_addr->sa_family != AF_INET6)
1111 continue;
1112
1113 it6 = (struct in6_ifaddr *)ifa;
1114
1115 /* ignore no autoconf addresses. */
1116 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1117 continue;
1118
1119 /* ignore autoconf addresses with different prefixes. */
1120 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1121 continue;
1122
1123 /*
1124 * Now we are looking at an autoconf address with the same
1125 * prefix as ours. If the address is temporary and is still
1126 * preferred, do not create another one. It would be rare, but
1127 * could happen, for example, when we resume a laptop PC after
1128 * a long period.
1129 */
1130 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1131 !IFA6_IS_DEPRECATED(it6)) {
1132 public_ifa6 = NULL;
1133 break;
1134 }
1135
1136 /*
1137 * This is a public autoconf address that has the same prefix
1138 * as ours. If it is preferred, keep it. We can't break the
1139 * loop here, because there may be a still-preferred temporary
1140 * address with the prefix.
1141 */
1142 if (!IFA6_IS_DEPRECATED(it6))
1143 public_ifa6 = it6;
1144 }
1145 if (public_ifa6 != NULL)
1146 ifa_ref(&public_ifa6->ia_ifa);
1147
1148 if (public_ifa6 != NULL) {
1149 int e;
1150
1151 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1152 ifa_free(&public_ifa6->ia_ifa);
1153 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1154 " tmp addr,errno=%d\n", e);
1155 return (-1);
1156 }
1157 ifa_free(&public_ifa6->ia_ifa);
1158 return (0);
1159 }
1160
1161 return (-1);
1162 }
1163
1164 /*
1165 * Remove prefix and default router list entries corresponding to ifp. Neighbor
1166 * cache entries are freed in in6_domifdetach().
1167 */
1168 void
nd6_purge(struct ifnet * ifp)1169 nd6_purge(struct ifnet *ifp)
1170 {
1171 struct nd_prhead prl;
1172 struct nd_prefix *pr, *npr;
1173
1174 LIST_INIT(&prl);
1175
1176 /* Purge default router list entries toward ifp. */
1177 nd6_defrouter_purge(ifp);
1178
1179 ND6_WLOCK();
1180 /*
1181 * Remove prefixes on ifp. We should have already removed addresses on
1182 * this interface, so no addresses should be referencing these prefixes.
1183 */
1184 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1185 if (pr->ndpr_ifp == ifp)
1186 nd6_prefix_unlink(pr, &prl);
1187 }
1188 ND6_WUNLOCK();
1189
1190 /* Delete the unlinked prefix objects. */
1191 while ((pr = LIST_FIRST(&prl)) != NULL) {
1192 LIST_REMOVE(pr, ndpr_entry);
1193 nd6_prefix_del(pr);
1194 }
1195
1196 /* cancel default outgoing interface setting */
1197 if (V_nd6_defifindex == ifp->if_index)
1198 nd6_setdefaultiface(0);
1199
1200 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1201 /* Refresh default router list. */
1202 defrouter_select_fib(ifp->if_fib);
1203 }
1204 }
1205
1206 /*
1207 * the caller acquires and releases the lock on the lltbls
1208 * Returns the llentry locked
1209 */
1210 struct llentry *
nd6_lookup(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1211 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1212 {
1213 struct sockaddr_in6 sin6;
1214 struct llentry *ln;
1215
1216 bzero(&sin6, sizeof(sin6));
1217 sin6.sin6_len = sizeof(struct sockaddr_in6);
1218 sin6.sin6_family = AF_INET6;
1219 sin6.sin6_addr = *addr6;
1220
1221 IF_AFDATA_LOCK_ASSERT(ifp);
1222
1223 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1224
1225 return (ln);
1226 }
1227
1228 static struct llentry *
nd6_alloc(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1229 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1230 {
1231 struct sockaddr_in6 sin6;
1232 struct llentry *ln;
1233
1234 bzero(&sin6, sizeof(sin6));
1235 sin6.sin6_len = sizeof(struct sockaddr_in6);
1236 sin6.sin6_family = AF_INET6;
1237 sin6.sin6_addr = *addr6;
1238
1239 ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1240 if (ln != NULL)
1241 ln->ln_state = ND6_LLINFO_NOSTATE;
1242
1243 return (ln);
1244 }
1245
1246 /*
1247 * Test whether a given IPv6 address can be a neighbor.
1248 */
1249 static bool
nd6_is_new_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1250 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1251 {
1252
1253 /*
1254 * A link-local address is always a neighbor.
1255 * XXX: a link does not necessarily specify a single interface.
1256 */
1257 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1258 struct sockaddr_in6 sin6_copy;
1259 u_int32_t zone;
1260
1261 /*
1262 * We need sin6_copy since sa6_recoverscope() may modify the
1263 * content (XXX).
1264 */
1265 sin6_copy = *addr;
1266 if (sa6_recoverscope(&sin6_copy))
1267 return (0); /* XXX: should be impossible */
1268 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1269 return (0);
1270 if (sin6_copy.sin6_scope_id == zone)
1271 return (1);
1272 else
1273 return (0);
1274 }
1275 /* Checking global unicast */
1276
1277 /* If an address is directly reachable, it is a neigbor */
1278 struct nhop_object *nh;
1279 nh = fib6_lookup(ifp->if_fib, &addr->sin6_addr, 0, NHR_NONE, 0);
1280 if (nh != NULL && nh->nh_aifp == ifp && (nh->nh_flags & NHF_GATEWAY) == 0)
1281 return (true);
1282
1283 /*
1284 * Check prefixes with desired on-link state, as some may be not
1285 * installed in the routing table.
1286 */
1287 bool matched = false;
1288 struct nd_prefix *pr;
1289 ND6_RLOCK();
1290 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1291 if (pr->ndpr_ifp != ifp)
1292 continue;
1293 if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0)
1294 continue;
1295 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1296 &addr->sin6_addr, &pr->ndpr_mask)) {
1297 matched = true;
1298 break;
1299 }
1300 }
1301 ND6_RUNLOCK();
1302 if (matched)
1303 return (true);
1304
1305 /*
1306 * If the address is assigned on the node of the other side of
1307 * a p2p interface, the address should be a neighbor.
1308 */
1309 if (ifp->if_flags & IFF_POINTOPOINT) {
1310 struct ifaddr *ifa;
1311
1312 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1313 if (ifa->ifa_addr->sa_family != addr->sin6_family)
1314 continue;
1315 if (ifa->ifa_dstaddr != NULL &&
1316 sa_equal(addr, ifa->ifa_dstaddr)) {
1317 return (true);
1318 }
1319 }
1320 }
1321
1322 /*
1323 * If the default router list is empty, all addresses are regarded
1324 * as on-link, and thus, as a neighbor.
1325 */
1326 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1327 nd6_defrouter_list_empty() &&
1328 V_nd6_defifindex == ifp->if_index) {
1329 return (1);
1330 }
1331
1332 return (0);
1333 }
1334
1335 /*
1336 * Detect if a given IPv6 address identifies a neighbor on a given link.
1337 * XXX: should take care of the destination of a p2p link?
1338 */
1339 int
nd6_is_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1340 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1341 {
1342 struct llentry *lle;
1343 int rc = 0;
1344
1345 NET_EPOCH_ASSERT();
1346 IF_AFDATA_UNLOCK_ASSERT(ifp);
1347 if (nd6_is_new_addr_neighbor(addr, ifp))
1348 return (1);
1349
1350 /*
1351 * Even if the address matches none of our addresses, it might be
1352 * in the neighbor cache.
1353 */
1354 if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1355 LLE_RUNLOCK(lle);
1356 rc = 1;
1357 }
1358 return (rc);
1359 }
1360
1361 static __noinline void
nd6_free_children(struct llentry * lle)1362 nd6_free_children(struct llentry *lle)
1363 {
1364 struct llentry *child_lle;
1365
1366 NET_EPOCH_ASSERT();
1367 LLE_WLOCK_ASSERT(lle);
1368
1369 while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1370 LLE_WLOCK(child_lle);
1371 lltable_unlink_child_entry(child_lle);
1372 llentry_free(child_lle);
1373 }
1374 }
1375
1376 /*
1377 * Tries to update @lle address/prepend data with new @lladdr.
1378 *
1379 * Returns true on success.
1380 * In any case, @lle is returned wlocked.
1381 */
1382 static __noinline bool
nd6_try_set_entry_addr_locked(struct ifnet * ifp,struct llentry * lle,char * lladdr)1383 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1384 {
1385 u_char buf[LLE_MAX_LINKHDR];
1386 int fam, off;
1387 size_t sz;
1388
1389 sz = sizeof(buf);
1390 if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1391 return (false);
1392
1393 /* Update data */
1394 lltable_set_entry_addr(ifp, lle, buf, sz, off);
1395
1396 struct llentry *child_lle;
1397 CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1398 LLE_WLOCK(child_lle);
1399 fam = child_lle->r_family;
1400 sz = sizeof(buf);
1401 if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1402 /* success */
1403 lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1404 child_lle->ln_state = ND6_LLINFO_REACHABLE;
1405 }
1406 LLE_WUNLOCK(child_lle);
1407 }
1408
1409 return (true);
1410 }
1411
1412 bool
nd6_try_set_entry_addr(struct ifnet * ifp,struct llentry * lle,char * lladdr)1413 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1414 {
1415 NET_EPOCH_ASSERT();
1416 LLE_WLOCK_ASSERT(lle);
1417
1418 if (!lltable_acquire_wlock(ifp, lle))
1419 return (false);
1420 bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1421 IF_AFDATA_WUNLOCK(ifp);
1422
1423 return (ret);
1424 }
1425
1426 /*
1427 * Free an nd6 llinfo entry.
1428 * Since the function would cause significant changes in the kernel, DO NOT
1429 * make it global, unless you have a strong reason for the change, and are sure
1430 * that the change is safe.
1431 *
1432 * Set noinline to be dtrace-friendly
1433 */
1434 static __noinline void
nd6_free(struct llentry ** lnp,int gc)1435 nd6_free(struct llentry **lnp, int gc)
1436 {
1437 struct ifnet *ifp;
1438 struct llentry *ln;
1439 struct nd_defrouter *dr;
1440
1441 ln = *lnp;
1442 *lnp = NULL;
1443
1444 LLE_WLOCK_ASSERT(ln);
1445 ND6_RLOCK_ASSERT();
1446
1447 KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1448
1449 ifp = lltable_get_ifp(ln->lle_tbl);
1450 if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1451 dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1452 else
1453 dr = NULL;
1454 ND6_RUNLOCK();
1455
1456 if ((ln->la_flags & LLE_DELETED) == 0)
1457 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1458
1459 /*
1460 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1461 * even though it is not harmful, it was not really necessary.
1462 */
1463
1464 /* cancel timer */
1465 nd6_llinfo_settimer_locked(ln, -1);
1466
1467 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1468 if (dr != NULL && dr->expire &&
1469 ln->ln_state == ND6_LLINFO_STALE && gc) {
1470 /*
1471 * If the reason for the deletion is just garbage
1472 * collection, and the neighbor is an active default
1473 * router, do not delete it. Instead, reset the GC
1474 * timer using the router's lifetime.
1475 * Simply deleting the entry would affect default
1476 * router selection, which is not necessarily a good
1477 * thing, especially when we're using router preference
1478 * values.
1479 * XXX: the check for ln_state would be redundant,
1480 * but we intentionally keep it just in case.
1481 */
1482 if (dr->expire > time_uptime)
1483 nd6_llinfo_settimer_locked(ln,
1484 (dr->expire - time_uptime) * hz);
1485 else
1486 nd6_llinfo_settimer_locked(ln,
1487 (long)V_nd6_gctimer * hz);
1488
1489 LLE_REMREF(ln);
1490 LLE_WUNLOCK(ln);
1491 defrouter_rele(dr);
1492 return;
1493 }
1494
1495 if (dr) {
1496 /*
1497 * Unreachability of a router might affect the default
1498 * router selection and on-link detection of advertised
1499 * prefixes.
1500 */
1501
1502 /*
1503 * Temporarily fake the state to choose a new default
1504 * router and to perform on-link determination of
1505 * prefixes correctly.
1506 * Below the state will be set correctly,
1507 * or the entry itself will be deleted.
1508 */
1509 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1510 }
1511
1512 if (ln->ln_router || dr) {
1513 /*
1514 * We need to unlock to avoid a LOR with rt6_flush() with the
1515 * rnh and for the calls to pfxlist_onlink_check() and
1516 * defrouter_select_fib() in the block further down for calls
1517 * into nd6_lookup(). We still hold a ref.
1518 */
1519 LLE_WUNLOCK(ln);
1520
1521 /*
1522 * rt6_flush must be called whether or not the neighbor
1523 * is in the Default Router List.
1524 * See a corresponding comment in nd6_na_input().
1525 */
1526 rt6_flush(&ln->r_l3addr.addr6, ifp);
1527 }
1528
1529 if (dr) {
1530 /*
1531 * Since defrouter_select_fib() does not affect the
1532 * on-link determination and MIP6 needs the check
1533 * before the default router selection, we perform
1534 * the check now.
1535 */
1536 pfxlist_onlink_check();
1537
1538 /*
1539 * Refresh default router list.
1540 */
1541 defrouter_select_fib(dr->ifp->if_fib);
1542 }
1543
1544 /*
1545 * If this entry was added by an on-link redirect, remove the
1546 * corresponding host route.
1547 */
1548 if (ln->la_flags & LLE_REDIRECT)
1549 nd6_free_redirect(ln);
1550
1551 if (ln->ln_router || dr)
1552 LLE_WLOCK(ln);
1553 }
1554
1555 /*
1556 * Save to unlock. We still hold an extra reference and will not
1557 * free(9) in llentry_free() if someone else holds one as well.
1558 */
1559 LLE_WUNLOCK(ln);
1560 IF_AFDATA_LOCK(ifp);
1561 LLE_WLOCK(ln);
1562 /* Guard against race with other llentry_free(). */
1563 if (ln->la_flags & LLE_LINKED) {
1564 /* Remove callout reference */
1565 LLE_REMREF(ln);
1566 lltable_unlink_entry(ln->lle_tbl, ln);
1567 }
1568 IF_AFDATA_UNLOCK(ifp);
1569
1570 nd6_free_children(ln);
1571
1572 llentry_free(ln);
1573 if (dr != NULL)
1574 defrouter_rele(dr);
1575 }
1576
1577 static int
nd6_isdynrte(const struct rtentry * rt,const struct nhop_object * nh,void * xap)1578 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1579 {
1580
1581 if (nh->nh_flags & NHF_REDIRECT)
1582 return (1);
1583
1584 return (0);
1585 }
1586
1587 /*
1588 * Remove the rtentry for the given llentry,
1589 * both of which were installed by a redirect.
1590 */
1591 static void
nd6_free_redirect(const struct llentry * ln)1592 nd6_free_redirect(const struct llentry *ln)
1593 {
1594 int fibnum;
1595 struct sockaddr_in6 sin6;
1596 struct rib_cmd_info rc;
1597 struct epoch_tracker et;
1598
1599 lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1600
1601 NET_EPOCH_ENTER(et);
1602 for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1603 rib_del_route_px(fibnum, (struct sockaddr *)&sin6, 128,
1604 nd6_isdynrte, NULL, 0, &rc);
1605 NET_EPOCH_EXIT(et);
1606 }
1607
1608 /*
1609 * Updates status of the default router route.
1610 */
1611 static void
check_release_defrouter(const struct rib_cmd_info * rc,void * _cbdata)1612 check_release_defrouter(const struct rib_cmd_info *rc, void *_cbdata)
1613 {
1614 struct nd_defrouter *dr;
1615 struct nhop_object *nh;
1616
1617 nh = rc->rc_nh_old;
1618 if (rc->rc_cmd == RTM_DELETE && (nh->nh_flags & NHF_DEFAULT) != 0) {
1619 dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1620 if (dr != NULL) {
1621 dr->installed = 0;
1622 defrouter_rele(dr);
1623 }
1624 }
1625 }
1626
1627 void
nd6_subscription_cb(struct rib_head * rnh,struct rib_cmd_info * rc,void * arg)1628 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1629 {
1630 #ifdef ROUTE_MPATH
1631 rib_decompose_notification(rc, check_release_defrouter, NULL);
1632 if (rc->rc_cmd == RTM_DELETE && !NH_IS_NHGRP(rc->rc_nh_old))
1633 check_release_defrouter(rc, NULL);
1634 #else
1635 check_release_defrouter(rc, NULL);
1636 #endif
1637 }
1638
1639 int
nd6_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp)1640 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1641 {
1642 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1643 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1644 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1645 struct epoch_tracker et;
1646 int error = 0;
1647
1648 if (ifp->if_afdata[AF_INET6] == NULL)
1649 return (EPFNOSUPPORT);
1650 switch (cmd) {
1651 case OSIOCGIFINFO_IN6:
1652 #define ND ndi->ndi
1653 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1654 bzero(&ND, sizeof(ND));
1655 ND.linkmtu = IN6_LINKMTU(ifp);
1656 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1657 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1658 ND.reachable = ND_IFINFO(ifp)->reachable;
1659 ND.retrans = ND_IFINFO(ifp)->retrans;
1660 ND.flags = ND_IFINFO(ifp)->flags;
1661 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1662 ND.chlim = ND_IFINFO(ifp)->chlim;
1663 break;
1664 case SIOCGIFINFO_IN6:
1665 ND = *ND_IFINFO(ifp);
1666 break;
1667 case SIOCSIFINFO_IN6:
1668 /*
1669 * used to change host variables from userland.
1670 * intended for a use on router to reflect RA configurations.
1671 */
1672 /* 0 means 'unspecified' */
1673 if (ND.linkmtu != 0) {
1674 if (ND.linkmtu < IPV6_MMTU ||
1675 ND.linkmtu > IN6_LINKMTU(ifp)) {
1676 error = EINVAL;
1677 break;
1678 }
1679 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1680 }
1681
1682 if (ND.basereachable != 0) {
1683 int obasereachable = ND_IFINFO(ifp)->basereachable;
1684
1685 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1686 if (ND.basereachable != obasereachable)
1687 ND_IFINFO(ifp)->reachable =
1688 ND_COMPUTE_RTIME(ND.basereachable);
1689 }
1690 if (ND.retrans != 0)
1691 ND_IFINFO(ifp)->retrans = ND.retrans;
1692 if (ND.chlim != 0)
1693 ND_IFINFO(ifp)->chlim = ND.chlim;
1694 /* FALLTHROUGH */
1695 case SIOCSIFINFO_FLAGS:
1696 {
1697 struct ifaddr *ifa;
1698 struct in6_ifaddr *ia;
1699
1700 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1701 !(ND.flags & ND6_IFF_IFDISABLED)) {
1702 /* ifdisabled 1->0 transision */
1703
1704 /*
1705 * If the interface is marked as ND6_IFF_IFDISABLED and
1706 * has an link-local address with IN6_IFF_DUPLICATED,
1707 * do not clear ND6_IFF_IFDISABLED.
1708 * See RFC 4862, Section 5.4.5.
1709 */
1710 NET_EPOCH_ENTER(et);
1711 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1712 if (ifa->ifa_addr->sa_family != AF_INET6)
1713 continue;
1714 ia = (struct in6_ifaddr *)ifa;
1715 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1716 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1717 break;
1718 }
1719 NET_EPOCH_EXIT(et);
1720
1721 if (ifa != NULL) {
1722 /* LLA is duplicated. */
1723 ND.flags |= ND6_IFF_IFDISABLED;
1724 log(LOG_ERR, "Cannot enable an interface"
1725 " with a link-local address marked"
1726 " duplicate.\n");
1727 } else {
1728 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1729 if (ifp->if_flags & IFF_UP)
1730 in6_if_up(ifp);
1731 }
1732 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1733 (ND.flags & ND6_IFF_IFDISABLED)) {
1734 /* ifdisabled 0->1 transision */
1735 /* Mark all IPv6 address as tentative. */
1736
1737 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1738 if (V_ip6_dad_count > 0 &&
1739 (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1740 NET_EPOCH_ENTER(et);
1741 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1742 ifa_link) {
1743 if (ifa->ifa_addr->sa_family !=
1744 AF_INET6)
1745 continue;
1746 ia = (struct in6_ifaddr *)ifa;
1747 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1748 }
1749 NET_EPOCH_EXIT(et);
1750 }
1751 }
1752
1753 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1754 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1755 /* auto_linklocal 0->1 transision */
1756
1757 /* If no link-local address on ifp, configure */
1758 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1759 in6_ifattach(ifp, NULL);
1760 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1761 ifp->if_flags & IFF_UP) {
1762 /*
1763 * When the IF already has
1764 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1765 * address is assigned, and IFF_UP, try to
1766 * assign one.
1767 */
1768 NET_EPOCH_ENTER(et);
1769 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1770 ifa_link) {
1771 if (ifa->ifa_addr->sa_family !=
1772 AF_INET6)
1773 continue;
1774 ia = (struct in6_ifaddr *)ifa;
1775 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1776 break;
1777 }
1778 NET_EPOCH_EXIT(et);
1779 if (ifa != NULL)
1780 /* No LLA is configured. */
1781 in6_ifattach(ifp, NULL);
1782 }
1783 }
1784 ND_IFINFO(ifp)->flags = ND.flags;
1785 break;
1786 }
1787 #undef ND
1788 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1789 /* sync kernel routing table with the default router list */
1790 defrouter_reset();
1791 defrouter_select_fib(RT_ALL_FIBS);
1792 break;
1793 case SIOCSPFXFLUSH_IN6:
1794 {
1795 /* flush all the prefix advertised by routers */
1796 struct in6_ifaddr *ia, *ia_next;
1797 struct nd_prefix *pr, *next;
1798 struct nd_prhead prl;
1799
1800 LIST_INIT(&prl);
1801
1802 ND6_WLOCK();
1803 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1804 if (ND6_PREFIX_WITH_ROUTER(pr))
1805 nd6_prefix_unlink(pr, &prl);
1806 }
1807 ND6_WUNLOCK();
1808
1809 while ((pr = LIST_FIRST(&prl)) != NULL) {
1810 LIST_REMOVE(pr, ndpr_entry);
1811 /* XXXRW: in6_ifaddrhead locking. */
1812 CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1813 ia_next) {
1814 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1815 continue;
1816
1817 if (ia->ia6_ndpr == pr)
1818 in6_purgeaddr(&ia->ia_ifa);
1819 }
1820 nd6_prefix_del(pr);
1821 }
1822 break;
1823 }
1824 case SIOCSRTRFLUSH_IN6:
1825 {
1826 /* flush all the default routers */
1827
1828 defrouter_reset();
1829 nd6_defrouter_flush_all();
1830 defrouter_select_fib(RT_ALL_FIBS);
1831 break;
1832 }
1833 case SIOCGNBRINFO_IN6:
1834 {
1835 struct llentry *ln;
1836 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1837
1838 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1839 return (error);
1840
1841 NET_EPOCH_ENTER(et);
1842 ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1843 NET_EPOCH_EXIT(et);
1844
1845 if (ln == NULL) {
1846 error = EINVAL;
1847 break;
1848 }
1849 nbi->state = ln->ln_state;
1850 nbi->asked = ln->la_asked;
1851 nbi->isrouter = ln->ln_router;
1852 if (ln->la_expire == 0)
1853 nbi->expire = 0;
1854 else
1855 nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1856 (time_second - time_uptime);
1857 LLE_RUNLOCK(ln);
1858 break;
1859 }
1860 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1861 ndif->ifindex = V_nd6_defifindex;
1862 break;
1863 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1864 return (nd6_setdefaultiface(ndif->ifindex));
1865 }
1866 return (error);
1867 }
1868
1869 /*
1870 * Calculates new isRouter value based on provided parameters and
1871 * returns it.
1872 */
1873 static int
nd6_is_router(int type,int code,int is_new,int old_addr,int new_addr,int ln_router)1874 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1875 int ln_router)
1876 {
1877
1878 /*
1879 * ICMP6 type dependent behavior.
1880 *
1881 * NS: clear IsRouter if new entry
1882 * RS: clear IsRouter
1883 * RA: set IsRouter if there's lladdr
1884 * redir: clear IsRouter if new entry
1885 *
1886 * RA case, (1):
1887 * The spec says that we must set IsRouter in the following cases:
1888 * - If lladdr exist, set IsRouter. This means (1-5).
1889 * - If it is old entry (!newentry), set IsRouter. This means (7).
1890 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1891 * A quetion arises for (1) case. (1) case has no lladdr in the
1892 * neighbor cache, this is similar to (6).
1893 * This case is rare but we figured that we MUST NOT set IsRouter.
1894 *
1895 * is_new old_addr new_addr NS RS RA redir
1896 * D R
1897 * 0 n n (1) c ? s
1898 * 0 y n (2) c s s
1899 * 0 n y (3) c s s
1900 * 0 y y (4) c s s
1901 * 0 y y (5) c s s
1902 * 1 -- n (6) c c c s
1903 * 1 -- y (7) c c s c s
1904 *
1905 * (c=clear s=set)
1906 */
1907 switch (type & 0xff) {
1908 case ND_NEIGHBOR_SOLICIT:
1909 /*
1910 * New entry must have is_router flag cleared.
1911 */
1912 if (is_new) /* (6-7) */
1913 ln_router = 0;
1914 break;
1915 case ND_REDIRECT:
1916 /*
1917 * If the icmp is a redirect to a better router, always set the
1918 * is_router flag. Otherwise, if the entry is newly created,
1919 * clear the flag. [RFC 2461, sec 8.3]
1920 */
1921 if (code == ND_REDIRECT_ROUTER)
1922 ln_router = 1;
1923 else {
1924 if (is_new) /* (6-7) */
1925 ln_router = 0;
1926 }
1927 break;
1928 case ND_ROUTER_SOLICIT:
1929 /*
1930 * is_router flag must always be cleared.
1931 */
1932 ln_router = 0;
1933 break;
1934 case ND_ROUTER_ADVERT:
1935 /*
1936 * Mark an entry with lladdr as a router.
1937 */
1938 if ((!is_new && (old_addr || new_addr)) || /* (2-5) */
1939 (is_new && new_addr)) { /* (7) */
1940 ln_router = 1;
1941 }
1942 break;
1943 }
1944
1945 return (ln_router);
1946 }
1947
1948 /*
1949 * Create neighbor cache entry and cache link-layer address,
1950 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1951 *
1952 * type - ICMP6 type
1953 * code - type dependent information
1954 *
1955 */
1956 void
nd6_cache_lladdr(struct ifnet * ifp,struct in6_addr * from,char * lladdr,int lladdrlen,int type,int code)1957 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1958 int lladdrlen, int type, int code)
1959 {
1960 struct llentry *ln = NULL, *ln_tmp;
1961 int is_newentry;
1962 int do_update;
1963 int olladdr;
1964 int llchange;
1965 int flags;
1966 uint16_t router = 0;
1967 struct mbuf *chain = NULL;
1968 u_char linkhdr[LLE_MAX_LINKHDR];
1969 size_t linkhdrsize;
1970 int lladdr_off;
1971
1972 NET_EPOCH_ASSERT();
1973 IF_AFDATA_UNLOCK_ASSERT(ifp);
1974
1975 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1976 KASSERT(from != NULL, ("%s: from == NULL", __func__));
1977
1978 /* nothing must be updated for unspecified address */
1979 if (IN6_IS_ADDR_UNSPECIFIED(from))
1980 return;
1981
1982 /*
1983 * Validation about ifp->if_addrlen and lladdrlen must be done in
1984 * the caller.
1985 *
1986 * XXX If the link does not have link-layer adderss, what should
1987 * we do? (ifp->if_addrlen == 0)
1988 * Spec says nothing in sections for RA, RS and NA. There's small
1989 * description on it in NS section (RFC 2461 7.2.3).
1990 */
1991 flags = lladdr ? LLE_EXCLUSIVE : 0;
1992 ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
1993 is_newentry = 0;
1994 if (ln == NULL) {
1995 flags |= LLE_EXCLUSIVE;
1996 ln = nd6_alloc(from, 0, ifp);
1997 if (ln == NULL)
1998 return;
1999
2000 /*
2001 * Since we already know all the data for the new entry,
2002 * fill it before insertion.
2003 */
2004 if (lladdr != NULL) {
2005 linkhdrsize = sizeof(linkhdr);
2006 if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2007 linkhdr, &linkhdrsize, &lladdr_off) != 0) {
2008 lltable_free_entry(LLTABLE6(ifp), ln);
2009 return;
2010 }
2011 lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2012 lladdr_off);
2013 }
2014
2015 IF_AFDATA_WLOCK(ifp);
2016 LLE_WLOCK(ln);
2017 /* Prefer any existing lle over newly-created one */
2018 ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2019 if (ln_tmp == NULL)
2020 lltable_link_entry(LLTABLE6(ifp), ln);
2021 IF_AFDATA_WUNLOCK(ifp);
2022 if (ln_tmp == NULL) {
2023 /* No existing lle, mark as new entry (6,7) */
2024 is_newentry = 1;
2025 if (lladdr != NULL) { /* (7) */
2026 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2027 EVENTHANDLER_INVOKE(lle_event, ln,
2028 LLENTRY_RESOLVED);
2029 }
2030 } else {
2031 lltable_free_entry(LLTABLE6(ifp), ln);
2032 ln = ln_tmp;
2033 ln_tmp = NULL;
2034 }
2035 }
2036 /* do nothing if static ndp is set */
2037 if ((ln->la_flags & LLE_STATIC)) {
2038 if (flags & LLE_EXCLUSIVE)
2039 LLE_WUNLOCK(ln);
2040 else
2041 LLE_RUNLOCK(ln);
2042 return;
2043 }
2044
2045 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2046 if (olladdr && lladdr) {
2047 llchange = bcmp(lladdr, ln->ll_addr,
2048 ifp->if_addrlen);
2049 } else if (!olladdr && lladdr)
2050 llchange = 1;
2051 else
2052 llchange = 0;
2053
2054 /*
2055 * newentry olladdr lladdr llchange (*=record)
2056 * 0 n n -- (1)
2057 * 0 y n -- (2)
2058 * 0 n y y (3) * STALE
2059 * 0 y y n (4) *
2060 * 0 y y y (5) * STALE
2061 * 1 -- n -- (6) NOSTATE(= PASSIVE)
2062 * 1 -- y -- (7) * STALE
2063 */
2064
2065 do_update = 0;
2066 if (is_newentry == 0 && llchange != 0) {
2067 do_update = 1; /* (3,5) */
2068
2069 /*
2070 * Record source link-layer address
2071 * XXX is it dependent to ifp->if_type?
2072 */
2073 if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2074 /* Entry was deleted */
2075 LLE_WUNLOCK(ln);
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 chain = nd6_grab_holdchain(ln);
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, ln, chain);
2103 if (do_update)
2104 nd6_flush_children_holdchain(ifp, ln);
2105
2106 /*
2107 * When the link-layer address of a router changes, select the
2108 * best router again. In particular, when the neighbor entry is newly
2109 * created, it might affect the selection policy.
2110 * Question: can we restrict the first condition to the "is_newentry"
2111 * case?
2112 * XXX: when we hear an RA from a new router with the link-layer
2113 * address option, defrouter_select_fib() is called twice, since
2114 * defrtrlist_update called the function as well. However, I believe
2115 * we can compromise the overhead, since it only happens the first
2116 * time.
2117 * XXX: although defrouter_select_fib() should not have a bad effect
2118 * for those are not autoconfigured hosts, we explicitly avoid such
2119 * cases for safety.
2120 */
2121 if ((do_update || is_newentry) && router &&
2122 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2123 /*
2124 * guaranteed recursion
2125 */
2126 defrouter_select_fib(ifp->if_fib);
2127 }
2128 }
2129
2130 static void
nd6_slowtimo(void * arg)2131 nd6_slowtimo(void *arg)
2132 {
2133 struct epoch_tracker et;
2134 CURVNET_SET((struct vnet *) arg);
2135 struct nd_ifinfo *nd6if;
2136 struct ifnet *ifp;
2137
2138 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2139 nd6_slowtimo, curvnet);
2140 NET_EPOCH_ENTER(et);
2141 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2142 if (ifp->if_afdata[AF_INET6] == NULL)
2143 continue;
2144 nd6if = ND_IFINFO(ifp);
2145 if (nd6if->basereachable && /* already initialized */
2146 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2147 /*
2148 * Since reachable time rarely changes by router
2149 * advertisements, we SHOULD insure that a new random
2150 * value gets recomputed at least once every few hours.
2151 * (RFC 2461, 6.3.4)
2152 */
2153 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2154 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2155 }
2156 }
2157 NET_EPOCH_EXIT(et);
2158 CURVNET_RESTORE();
2159 }
2160
2161 struct mbuf *
nd6_grab_holdchain(struct llentry * ln)2162 nd6_grab_holdchain(struct llentry *ln)
2163 {
2164 struct mbuf *chain;
2165
2166 LLE_WLOCK_ASSERT(ln);
2167
2168 chain = ln->la_hold;
2169 ln->la_hold = NULL;
2170 ln->la_numheld = 0;
2171
2172 if (ln->ln_state == ND6_LLINFO_STALE) {
2173 /*
2174 * The first time we send a packet to a
2175 * neighbor whose entry is STALE, we have
2176 * to change the state to DELAY and a sets
2177 * a timer to expire in DELAY_FIRST_PROBE_TIME
2178 * seconds to ensure do neighbor unreachability
2179 * detection on expiration.
2180 * (RFC 2461 7.3.3)
2181 */
2182 nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2183 }
2184
2185 return (chain);
2186 }
2187
2188 int
nd6_output_ifp(struct ifnet * ifp,struct ifnet * origifp,struct mbuf * m,struct sockaddr_in6 * dst,struct route * ro)2189 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2190 struct sockaddr_in6 *dst, struct route *ro)
2191 {
2192 int error;
2193 int ip6len;
2194 struct ip6_hdr *ip6;
2195 struct m_tag *mtag;
2196
2197 #ifdef MAC
2198 mac_netinet6_nd6_send(ifp, m);
2199 #endif
2200
2201 /*
2202 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2203 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2204 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2205 * to be diverted to user space. When re-injected into the kernel,
2206 * send_output() will directly dispatch them to the outgoing interface.
2207 */
2208 if (send_sendso_input_hook != NULL) {
2209 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2210 if (mtag != NULL) {
2211 ip6 = mtod(m, struct ip6_hdr *);
2212 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2213 /* Use the SEND socket */
2214 error = send_sendso_input_hook(m, ifp, SND_OUT,
2215 ip6len);
2216 /* -1 == no app on SEND socket */
2217 if (error == 0 || error != -1)
2218 return (error);
2219 }
2220 }
2221
2222 m_clrprotoflags(m); /* Avoid confusing lower layers. */
2223 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2224 mtod(m, struct ip6_hdr *));
2225
2226 if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2227 origifp = ifp;
2228
2229 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2230 return (error);
2231 }
2232
2233 /*
2234 * Lookup link headerfor @sa_dst address. Stores found
2235 * data in @desten buffer. Copy of lle ln_flags can be also
2236 * saved in @pflags if @pflags is non-NULL.
2237 *
2238 * If destination LLE does not exists or lle state modification
2239 * is required, call "slow" version.
2240 *
2241 * Return values:
2242 * - 0 on success (address copied to buffer).
2243 * - EWOULDBLOCK (no local error, but address is still unresolved)
2244 * - other errors (alloc failure, etc)
2245 */
2246 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)2247 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2248 const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2249 struct llentry **plle)
2250 {
2251 struct llentry *ln = NULL;
2252 const struct sockaddr_in6 *dst6;
2253
2254 NET_EPOCH_ASSERT();
2255
2256 if (pflags != NULL)
2257 *pflags = 0;
2258
2259 dst6 = (const struct sockaddr_in6 *)sa_dst;
2260
2261 /* discard the packet if IPv6 operation is disabled on the interface */
2262 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2263 m_freem(m);
2264 return (ENETDOWN); /* better error? */
2265 }
2266
2267 if (m != NULL && m->m_flags & M_MCAST) {
2268 switch (ifp->if_type) {
2269 case IFT_ETHER:
2270 case IFT_L2VLAN:
2271 case IFT_BRIDGE:
2272 ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2273 desten);
2274 return (0);
2275 default:
2276 m_freem(m);
2277 return (EAFNOSUPPORT);
2278 }
2279 }
2280
2281 int family = gw_flags >> 16;
2282 int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2283 ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2284 if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2285 /* Entry found, let's copy lle info */
2286 bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2287 if (pflags != NULL)
2288 *pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2289 llentry_provide_feedback(ln);
2290 if (plle) {
2291 LLE_ADDREF(ln);
2292 *plle = ln;
2293 LLE_WUNLOCK(ln);
2294 }
2295 return (0);
2296 } else if (plle && ln)
2297 LLE_WUNLOCK(ln);
2298
2299 return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2300 }
2301
2302 /*
2303 * Finds or creates a new llentry for @addr and @family.
2304 * Returns wlocked llentry or NULL.
2305 *
2306 *
2307 * Child LLEs.
2308 *
2309 * Do not have their own state machine (gets marked as static)
2310 * settimer bails out for child LLEs just in case.
2311 *
2312 * Locking order: parent lle gets locked first, chen goes the child.
2313 */
2314 static __noinline struct llentry *
nd6_get_llentry(struct ifnet * ifp,const struct in6_addr * addr,int family)2315 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2316 {
2317 struct llentry *child_lle = NULL;
2318 struct llentry *lle, *lle_tmp;
2319
2320 lle = nd6_alloc(addr, 0, ifp);
2321 if (lle != NULL && family != AF_INET6) {
2322 child_lle = nd6_alloc(addr, 0, ifp);
2323 if (child_lle == NULL) {
2324 lltable_free_entry(LLTABLE6(ifp), lle);
2325 return (NULL);
2326 }
2327 child_lle->r_family = family;
2328 child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2329 child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2330 }
2331
2332 if (lle == NULL) {
2333 char ip6buf[INET6_ADDRSTRLEN];
2334 log(LOG_DEBUG,
2335 "nd6_get_llentry: can't allocate llinfo for %s "
2336 "(ln=%p)\n",
2337 ip6_sprintf(ip6buf, addr), lle);
2338 return (NULL);
2339 }
2340
2341 IF_AFDATA_WLOCK(ifp);
2342 LLE_WLOCK(lle);
2343 /* Prefer any existing entry over newly-created one */
2344 lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2345 if (lle_tmp == NULL)
2346 lltable_link_entry(LLTABLE6(ifp), lle);
2347 else {
2348 lltable_free_entry(LLTABLE6(ifp), lle);
2349 lle = lle_tmp;
2350 }
2351 if (child_lle != NULL) {
2352 /* Check if child lle for the same family exists */
2353 lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2354 LLE_WLOCK(child_lle);
2355 if (lle_tmp == NULL) {
2356 /* Attach */
2357 lltable_link_child_entry(lle, child_lle);
2358 } else {
2359 /* child lle already exists, free newly-created one */
2360 lltable_free_entry(LLTABLE6(ifp), child_lle);
2361 LLE_WLOCK(lle_tmp);
2362 child_lle = lle_tmp;
2363 }
2364 LLE_WUNLOCK(lle);
2365 lle = child_lle;
2366 }
2367 IF_AFDATA_WUNLOCK(ifp);
2368 return (lle);
2369 }
2370
2371 /*
2372 * Do L2 address resolution for @sa_dst address. Stores found
2373 * address in @desten buffer. Copy of lle ln_flags can be also
2374 * saved in @pflags if @pflags is non-NULL.
2375 *
2376 * Heavy version.
2377 * Function assume that destination LLE does not exist,
2378 * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2379 *
2380 * Set noinline to be dtrace-friendly
2381 */
2382 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)2383 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2384 const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2385 struct llentry **plle)
2386 {
2387 struct llentry *lle = NULL;
2388 struct in6_addr *psrc, src;
2389 int send_ns, ll_len;
2390 char *lladdr;
2391
2392 NET_EPOCH_ASSERT();
2393
2394 /*
2395 * Address resolution or Neighbor Unreachability Detection
2396 * for the next hop.
2397 * At this point, the destination of the packet must be a unicast
2398 * or an anycast address(i.e. not a multicast).
2399 */
2400 lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2401 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
2402 /*
2403 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2404 * the condition below is not very efficient. But we believe
2405 * it is tolerable, because this should be a rare case.
2406 */
2407 lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2408 }
2409
2410 if (lle == NULL) {
2411 m_freem(m);
2412 return (ENOBUFS);
2413 }
2414
2415 LLE_WLOCK_ASSERT(lle);
2416
2417 /*
2418 * The first time we send a packet to a neighbor whose entry is
2419 * STALE, we have to change the state to DELAY and a sets a timer to
2420 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2421 * neighbor unreachability detection on expiration.
2422 * (RFC 2461 7.3.3)
2423 */
2424 if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2425 nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2426
2427 /*
2428 * If the neighbor cache entry has a state other than INCOMPLETE
2429 * (i.e. its link-layer address is already resolved), just
2430 * send the packet.
2431 */
2432 if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2433 if (flags & LLE_ADDRONLY) {
2434 lladdr = lle->ll_addr;
2435 ll_len = ifp->if_addrlen;
2436 } else {
2437 lladdr = lle->r_linkdata;
2438 ll_len = lle->r_hdrlen;
2439 }
2440 bcopy(lladdr, desten, ll_len);
2441 if (pflags != NULL)
2442 *pflags = lle->la_flags;
2443 if (plle) {
2444 LLE_ADDREF(lle);
2445 *plle = lle;
2446 }
2447 LLE_WUNLOCK(lle);
2448 return (0);
2449 }
2450
2451 /*
2452 * There is a neighbor cache entry, but no ethernet address
2453 * response yet. Append this latest packet to the end of the
2454 * packet queue in the mbuf. When it exceeds nd6_maxqueuelen,
2455 * the oldest packet in the queue will be removed.
2456 */
2457 if (m != NULL) {
2458 size_t dropped;
2459
2460 dropped = lltable_append_entry_queue(lle, m, V_nd6_maxqueuelen);
2461 ICMP6STAT_ADD(icp6s_dropped, dropped);
2462 }
2463
2464 /*
2465 * If there has been no NS for the neighbor after entering the
2466 * INCOMPLETE state, send the first solicitation.
2467 * Note that for newly-created lle la_asked will be 0,
2468 * so we will transition from ND6_LLINFO_NOSTATE to
2469 * ND6_LLINFO_INCOMPLETE state here.
2470 */
2471 psrc = NULL;
2472 send_ns = 0;
2473
2474 /* If we have child lle, switch to the parent to send NS */
2475 if (lle->la_flags & LLE_CHILD) {
2476 struct llentry *lle_parent = lle->lle_parent;
2477 LLE_WUNLOCK(lle);
2478 lle = lle_parent;
2479 LLE_WLOCK(lle);
2480 }
2481 if (lle->la_asked == 0) {
2482 lle->la_asked++;
2483 send_ns = 1;
2484 psrc = nd6_llinfo_get_holdsrc(lle, &src);
2485
2486 nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2487 }
2488 LLE_WUNLOCK(lle);
2489 if (send_ns != 0)
2490 nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2491
2492 return (EWOULDBLOCK);
2493 }
2494
2495 /*
2496 * Do L2 address resolution for @sa_dst address. Stores found
2497 * address in @desten buffer. Copy of lle ln_flags can be also
2498 * saved in @pflags if @pflags is non-NULL.
2499 *
2500 * Return values:
2501 * - 0 on success (address copied to buffer).
2502 * - EWOULDBLOCK (no local error, but address is still unresolved)
2503 * - other errors (alloc failure, etc)
2504 */
2505 int
nd6_resolve_addr(struct ifnet * ifp,int flags,const struct sockaddr * dst,char * desten,uint32_t * pflags)2506 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2507 char *desten, uint32_t *pflags)
2508 {
2509 int error;
2510
2511 flags |= LLE_ADDRONLY;
2512 error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2513 (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2514 return (error);
2515 }
2516
2517 int
nd6_flush_holdchain(struct ifnet * ifp,struct llentry * lle,struct mbuf * chain)2518 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2519 {
2520 struct mbuf *m, *m_head;
2521 struct sockaddr_in6 dst6;
2522 int error = 0;
2523
2524 NET_EPOCH_ASSERT();
2525
2526 struct route_in6 ro = {
2527 .ro_prepend = lle->r_linkdata,
2528 .ro_plen = lle->r_hdrlen,
2529 };
2530
2531 lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2532 m_head = chain;
2533
2534 while (m_head) {
2535 m = m_head;
2536 m_head = m_head->m_nextpkt;
2537 m->m_nextpkt = NULL;
2538 error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2539 }
2540
2541 /*
2542 * XXX
2543 * note that intermediate errors are blindly ignored
2544 */
2545 return (error);
2546 }
2547
2548 __noinline void
nd6_flush_children_holdchain(struct ifnet * ifp,struct llentry * lle)2549 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2550 {
2551 struct llentry *child_lle;
2552 struct mbuf *chain;
2553
2554 NET_EPOCH_ASSERT();
2555
2556 CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2557 LLE_WLOCK(child_lle);
2558 chain = nd6_grab_holdchain(child_lle);
2559 LLE_WUNLOCK(child_lle);
2560 nd6_flush_holdchain(ifp, child_lle, chain);
2561 }
2562 }
2563
2564 static int
nd6_need_cache(struct ifnet * ifp)2565 nd6_need_cache(struct ifnet *ifp)
2566 {
2567 /*
2568 * XXX: we currently do not make neighbor cache on any interface
2569 * other than Ethernet and GIF.
2570 *
2571 * RFC2893 says:
2572 * - unidirectional tunnels needs no ND
2573 */
2574 switch (ifp->if_type) {
2575 case IFT_ETHER:
2576 case IFT_IEEE1394:
2577 case IFT_L2VLAN:
2578 case IFT_INFINIBAND:
2579 case IFT_BRIDGE:
2580 case IFT_PROPVIRTUAL:
2581 return (1);
2582 default:
2583 return (0);
2584 }
2585 }
2586
2587 /*
2588 * Add pernament ND6 link-layer record for given
2589 * interface address.
2590 *
2591 * Very similar to IPv4 arp_ifinit(), but:
2592 * 1) IPv6 DAD is performed in different place
2593 * 2) It is called by IPv6 protocol stack in contrast to
2594 * arp_ifinit() which is typically called in SIOCSIFADDR
2595 * driver ioctl handler.
2596 *
2597 */
2598 int
nd6_add_ifa_lle(struct in6_ifaddr * ia)2599 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2600 {
2601 struct ifnet *ifp;
2602 struct llentry *ln, *ln_tmp;
2603 struct sockaddr *dst;
2604
2605 ifp = ia->ia_ifa.ifa_ifp;
2606 if (nd6_need_cache(ifp) == 0)
2607 return (0);
2608
2609 dst = (struct sockaddr *)&ia->ia_addr;
2610 ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2611 if (ln == NULL)
2612 return (ENOBUFS);
2613
2614 IF_AFDATA_WLOCK(ifp);
2615 LLE_WLOCK(ln);
2616 /* Unlink any entry if exists */
2617 ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2618 if (ln_tmp != NULL)
2619 lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2620 lltable_link_entry(LLTABLE6(ifp), ln);
2621 IF_AFDATA_WUNLOCK(ifp);
2622
2623 if (ln_tmp != NULL)
2624 EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2625 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2626
2627 LLE_WUNLOCK(ln);
2628 if (ln_tmp != NULL)
2629 llentry_free(ln_tmp);
2630
2631 return (0);
2632 }
2633
2634 /*
2635 * Removes either all lle entries for given @ia, or lle
2636 * corresponding to @ia address.
2637 */
2638 void
nd6_rem_ifa_lle(struct in6_ifaddr * ia,int all)2639 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2640 {
2641 struct sockaddr_in6 mask, addr;
2642 struct sockaddr *saddr, *smask;
2643 struct ifnet *ifp;
2644
2645 ifp = ia->ia_ifa.ifa_ifp;
2646 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2647 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2648 saddr = (struct sockaddr *)&addr;
2649 smask = (struct sockaddr *)&mask;
2650
2651 if (all != 0)
2652 lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2653 else
2654 lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2655 }
2656
2657 static int
nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)2658 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2659 {
2660 struct in6_prefix p;
2661 struct sockaddr_in6 s6;
2662 struct nd_prefix *pr;
2663 struct nd_pfxrouter *pfr;
2664 time_t maxexpire;
2665 int error;
2666 char ip6buf[INET6_ADDRSTRLEN];
2667
2668 if (req->newptr)
2669 return (EPERM);
2670
2671 error = sysctl_wire_old_buffer(req, 0);
2672 if (error != 0)
2673 return (error);
2674
2675 bzero(&p, sizeof(p));
2676 p.origin = PR_ORIG_RA;
2677 bzero(&s6, sizeof(s6));
2678 s6.sin6_family = AF_INET6;
2679 s6.sin6_len = sizeof(s6);
2680
2681 ND6_RLOCK();
2682 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2683 p.prefix = pr->ndpr_prefix;
2684 if (sa6_recoverscope(&p.prefix)) {
2685 log(LOG_ERR, "scope error in prefix list (%s)\n",
2686 ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2687 /* XXX: press on... */
2688 }
2689 p.raflags = pr->ndpr_raf;
2690 p.prefixlen = pr->ndpr_plen;
2691 p.vltime = pr->ndpr_vltime;
2692 p.pltime = pr->ndpr_pltime;
2693 p.if_index = pr->ndpr_ifp->if_index;
2694 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2695 p.expire = 0;
2696 else {
2697 /* XXX: we assume time_t is signed. */
2698 maxexpire = (-1) &
2699 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2700 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2701 p.expire = pr->ndpr_lastupdate +
2702 pr->ndpr_vltime +
2703 (time_second - time_uptime);
2704 else
2705 p.expire = maxexpire;
2706 }
2707 p.refcnt = pr->ndpr_addrcnt;
2708 p.flags = pr->ndpr_stateflags;
2709 p.advrtrs = 0;
2710 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2711 p.advrtrs++;
2712 error = SYSCTL_OUT(req, &p, sizeof(p));
2713 if (error != 0)
2714 break;
2715 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2716 s6.sin6_addr = pfr->router->rtaddr;
2717 if (sa6_recoverscope(&s6))
2718 log(LOG_ERR,
2719 "scope error in prefix list (%s)\n",
2720 ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2721 error = SYSCTL_OUT(req, &s6, sizeof(s6));
2722 if (error != 0)
2723 goto out;
2724 }
2725 }
2726 out:
2727 ND6_RUNLOCK();
2728 return (error);
2729 }
2730 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2731 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2732 NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2733 "NDP prefix list");
2734 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2735 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2736 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2737 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2738