1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the project nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $
32 */
33
34 #include "opt_inet.h"
35 #include "opt_inet6.h"
36 #include "opt_route.h"
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/eventhandler.h>
41 #include <sys/callout.h>
42 #include <sys/lock.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/mutex.h>
46 #include <sys/socket.h>
47 #include <sys/sockio.h>
48 #include <sys/time.h>
49 #include <sys/kernel.h>
50 #include <sys/protosw.h>
51 #include <sys/errno.h>
52 #include <sys/syslog.h>
53 #include <sys/rwlock.h>
54 #include <sys/queue.h>
55 #include <sys/sdt.h>
56 #include <sys/sysctl.h>
57
58 #include <net/if.h>
59 #include <net/if_var.h>
60 #include <net/if_dl.h>
61 #include <net/if_private.h>
62 #include <net/if_types.h>
63 #include <net/route.h>
64 #include <net/route/route_ctl.h>
65 #include <net/route/nhop.h>
66 #include <net/vnet.h>
67
68 #include <netinet/in.h>
69 #include <netinet/in_kdtrace.h>
70 #include <net/if_llatbl.h>
71 #include <netinet/if_ether.h>
72 #include <netinet6/in6_fib.h>
73 #include <netinet6/in6_var.h>
74 #include <netinet/ip6.h>
75 #include <netinet6/ip6_var.h>
76 #include <netinet6/scope6_var.h>
77 #include <netinet6/nd6.h>
78 #include <netinet6/in6_ifattach.h>
79 #include <netinet/icmp6.h>
80 #include <netinet6/send.h>
81
82 #include <sys/limits.h>
83
84 #include <security/mac/mac_framework.h>
85
86 #define ND6_PREFIX_WITH_ROUTER(pr) !LIST_EMPTY(&(pr)->ndpr_advrtrs)
87
88 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */
89 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */
90
91 MALLOC_DEFINE(M_IP6NDP, "ip6ndp", "IPv6 Neighbor Discovery");
92
93 VNET_DEFINE_STATIC(int, nd6_prune) = 1;
94 #define V_nd6_prune VNET(nd6_prune)
95 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_PRUNE, nd6_prune,
96 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_prune), 0,
97 "Frequency in seconds of checks for expired prefixes and routers");
98
99 VNET_DEFINE_STATIC(int, nd6_delay) = 5;
100 #define V_nd6_delay VNET(nd6_delay)
101 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DELAY, nd6_delay,
102 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_delay), 0,
103 "Delay in seconds before probing for reachability");
104
105 VNET_DEFINE_STATIC(int, nd6_umaxtries) = 3;
106 #define V_nd6_umaxtries VNET(nd6_umaxtries)
107 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_UMAXTRIES, nd6_umaxtries,
108 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_umaxtries), 0,
109 "Number of ICMPv6 NS messages sent during reachability detection");
110
111 VNET_DEFINE(int, nd6_mmaxtries) = 3;
112 #define V_nd6_mmaxtries VNET(nd6_mmaxtries)
113 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MMAXTRIES, nd6_mmaxtries,
114 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_mmaxtries), 0,
115 "Number of ICMPv6 NS messages sent during address resolution");
116
117 VNET_DEFINE_STATIC(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage
118 * collection timer */
119 #define V_nd6_gctimer VNET(nd6_gctimer)
120
121 /* preventing too many loops in ND option parsing */
122 VNET_DEFINE_STATIC(int, nd6_maxndopt) = 10; /* max # of ND options allowed */
123
124 VNET_DEFINE_STATIC(int, nd6_maxqueuelen) = 16; /* max pkts cached in unresolved
125 * ND entries */
126 #define V_nd6_maxndopt VNET(nd6_maxndopt)
127 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen)
128
129 #ifdef ND6_DEBUG
130 VNET_DEFINE(int, nd6_debug) = 1;
131 #else
132 VNET_DEFINE(int, nd6_debug) = 0;
133 #endif
134 #define V_nd6_debug VNET(nd6_debug)
135 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_DEBUG, nd6_debug,
136 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_debug), 0,
137 "Log NDP debug messages");
138
139 static eventhandler_tag lle_event_eh, iflladdr_event_eh, ifnet_link_event_eh;
140
141 VNET_DEFINE(struct nd_prhead, nd_prefix);
142 VNET_DEFINE(struct rwlock, nd6_lock);
143 VNET_DEFINE(uint64_t, nd6_list_genid);
144 VNET_DEFINE(struct mtx, nd6_onlink_mtx);
145
146 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL;
147 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval)
148
149 int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int);
150
151 static bool nd6_is_new_addr_neighbor(const struct sockaddr_in6 *,
152 struct ifnet *);
153 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *);
154 static void nd6_slowtimo(void *);
155 static int regen_tmpaddr(struct in6_ifaddr *);
156 static void nd6_free(struct llentry **, int);
157 static void nd6_free_redirect(const struct llentry *);
158 static void nd6_llinfo_timer(void *);
159 static void nd6_llinfo_settimer_locked(struct llentry *, long);
160 static int nd6_resolve_slow(struct ifnet *, int, int, struct mbuf *,
161 const struct sockaddr_in6 *, u_char *, uint32_t *, struct llentry **);
162 static int nd6_need_cache(struct ifnet *);
163
164 VNET_DEFINE_STATIC(struct callout, nd6_slowtimo_ch);
165 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch)
166
167 VNET_DEFINE_STATIC(struct callout, nd6_timer_ch);
168 #define V_nd6_timer_ch VNET(nd6_timer_ch)
169
170 static void
nd6_lle_event(void * arg __unused,struct llentry * lle,int evt)171 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt)
172 {
173 struct rt_addrinfo rtinfo;
174 struct sockaddr_in6 dst;
175 struct sockaddr_dl gw;
176 struct ifnet *ifp;
177 int type;
178 int fibnum;
179
180 LLE_WLOCK_ASSERT(lle);
181
182 if (lltable_get_af(lle->lle_tbl) != AF_INET6)
183 return;
184
185 switch (evt) {
186 case LLENTRY_RESOLVED:
187 type = RTM_ADD;
188 KASSERT(lle->la_flags & LLE_VALID,
189 ("%s: %p resolved but not valid?", __func__, lle));
190 break;
191 case LLENTRY_EXPIRED:
192 type = RTM_DELETE;
193 break;
194 default:
195 return;
196 }
197
198 ifp = lltable_get_ifp(lle->lle_tbl);
199
200 bzero(&dst, sizeof(dst));
201 bzero(&gw, sizeof(gw));
202 bzero(&rtinfo, sizeof(rtinfo));
203 lltable_fill_sa_entry(lle, (struct sockaddr *)&dst);
204 dst.sin6_scope_id = in6_getscopezone(ifp,
205 in6_addrscope(&dst.sin6_addr));
206 gw.sdl_len = sizeof(struct sockaddr_dl);
207 gw.sdl_family = AF_LINK;
208 gw.sdl_alen = ifp->if_addrlen;
209 gw.sdl_index = ifp->if_index;
210 gw.sdl_type = ifp->if_type;
211 if (evt == LLENTRY_RESOLVED)
212 bcopy(lle->ll_addr, gw.sdl_data, ifp->if_addrlen);
213 rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst;
214 rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw;
215 rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY;
216 fibnum = V_rt_add_addr_allfibs ? RT_ALL_FIBS : ifp->if_fib;
217 rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | (
218 type == RTM_ADD ? RTF_UP: 0), 0, fibnum);
219 }
220
221 /*
222 * A handler for interface link layer address change event.
223 */
224 static void
nd6_iflladdr(void * arg __unused,struct ifnet * ifp)225 nd6_iflladdr(void *arg __unused, struct ifnet *ifp)
226 {
227 /* XXXGL: ??? */
228 if (ifp->if_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 /* XXXGL: ??? */
362 if (ifp->if_inet6 == NULL)
363 return;
364
365 nd6_setmtu0(ifp, ND_IFINFO(ifp));
366 }
367
368 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */
369 void
nd6_setmtu0(struct ifnet * ifp,struct nd_ifinfo * ndi)370 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi)
371 {
372 u_int32_t omaxmtu;
373
374 omaxmtu = ndi->maxmtu;
375 ndi->maxmtu = ifp->if_mtu;
376
377 /*
378 * Decreasing the interface MTU under IPV6 minimum MTU may cause
379 * undesirable situation. We thus notify the operator of the change
380 * explicitly. The check for omaxmtu is necessary to restrict the
381 * log to the case of changing the MTU, not initializing it.
382 */
383 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) {
384 log(LOG_NOTICE, "nd6_setmtu0: "
385 "new link MTU on %s (%lu) is too small for IPv6\n",
386 if_name(ifp), (unsigned long)ndi->maxmtu);
387 }
388 }
389
390 void
nd6_option_init(void * opt,int icmp6len,union nd_opts * ndopts)391 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts)
392 {
393
394 bzero(ndopts, sizeof(*ndopts));
395 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt;
396 ndopts->nd_opts_last
397 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len);
398
399 if (icmp6len == 0) {
400 ndopts->nd_opts_done = 1;
401 ndopts->nd_opts_search = NULL;
402 }
403 }
404
405 /*
406 * Take one ND option.
407 */
408 struct nd_opt_hdr *
nd6_option(union nd_opts * ndopts)409 nd6_option(union nd_opts *ndopts)
410 {
411 struct nd_opt_hdr *nd_opt;
412 int olen;
413
414 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
415 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
416 __func__));
417 if (ndopts->nd_opts_search == NULL)
418 return NULL;
419 if (ndopts->nd_opts_done)
420 return NULL;
421
422 nd_opt = ndopts->nd_opts_search;
423
424 /* make sure nd_opt_len is inside the buffer */
425 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) {
426 bzero(ndopts, sizeof(*ndopts));
427 return NULL;
428 }
429
430 olen = nd_opt->nd_opt_len << 3;
431 if (olen == 0) {
432 /*
433 * Message validation requires that all included
434 * options have a length that is greater than zero.
435 */
436 bzero(ndopts, sizeof(*ndopts));
437 return NULL;
438 }
439
440 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen);
441 if (ndopts->nd_opts_search > ndopts->nd_opts_last) {
442 /* option overruns the end of buffer, invalid */
443 bzero(ndopts, sizeof(*ndopts));
444 return NULL;
445 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) {
446 /* reached the end of options chain */
447 ndopts->nd_opts_done = 1;
448 ndopts->nd_opts_search = NULL;
449 }
450 return nd_opt;
451 }
452
453 /*
454 * Parse multiple ND options.
455 * This function is much easier to use, for ND routines that do not need
456 * multiple options of the same type.
457 */
458 int
nd6_options(union nd_opts * ndopts)459 nd6_options(union nd_opts *ndopts)
460 {
461 struct nd_opt_hdr *nd_opt;
462 int i = 0;
463
464 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__));
465 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts",
466 __func__));
467 if (ndopts->nd_opts_search == NULL)
468 return 0;
469
470 while (1) {
471 nd_opt = nd6_option(ndopts);
472 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) {
473 /*
474 * Message validation requires that all included
475 * options have a length that is greater than zero.
476 */
477 ICMP6STAT_INC(icp6s_nd_badopt);
478 bzero(ndopts, sizeof(*ndopts));
479 return -1;
480 }
481
482 if (nd_opt == NULL)
483 goto skip1;
484
485 switch (nd_opt->nd_opt_type) {
486 case ND_OPT_SOURCE_LINKADDR:
487 case ND_OPT_TARGET_LINKADDR:
488 case ND_OPT_MTU:
489 case ND_OPT_REDIRECTED_HEADER:
490 case ND_OPT_NONCE:
491 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) {
492 nd6log((LOG_INFO,
493 "duplicated ND6 option found (type=%d)\n",
494 nd_opt->nd_opt_type));
495 /* XXX bark? */
496 } else {
497 ndopts->nd_opt_array[nd_opt->nd_opt_type]
498 = nd_opt;
499 }
500 break;
501 case ND_OPT_PREFIX_INFORMATION:
502 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) {
503 ndopts->nd_opt_array[nd_opt->nd_opt_type]
504 = nd_opt;
505 }
506 ndopts->nd_opts_pi_end =
507 (struct nd_opt_prefix_info *)nd_opt;
508 break;
509 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */
510 case ND_OPT_RDNSS: /* RFC 6106 */
511 case ND_OPT_DNSSL: /* RFC 6106 */
512 /*
513 * Silently ignore options we know and do not care about
514 * in the kernel.
515 */
516 break;
517 default:
518 /*
519 * Unknown options must be silently ignored,
520 * to accommodate future extension to the protocol.
521 */
522 nd6log((LOG_DEBUG,
523 "nd6_options: unsupported option %d - "
524 "option ignored\n", nd_opt->nd_opt_type));
525 }
526
527 skip1:
528 i++;
529 if (i > V_nd6_maxndopt) {
530 ICMP6STAT_INC(icp6s_nd_toomanyopt);
531 nd6log((LOG_INFO, "too many loop in nd opt\n"));
532 break;
533 }
534
535 if (ndopts->nd_opts_done)
536 break;
537 }
538
539 return 0;
540 }
541
542 /*
543 * ND6 timer routine to handle ND6 entries
544 */
545 static void
nd6_llinfo_settimer_locked(struct llentry * ln,long tick)546 nd6_llinfo_settimer_locked(struct llentry *ln, long tick)
547 {
548 int canceled;
549
550 LLE_WLOCK_ASSERT(ln);
551
552 /* Do not schedule timers for child LLEs. */
553 if (ln->la_flags & LLE_CHILD)
554 return;
555
556 if (tick < 0) {
557 ln->la_expire = 0;
558 ln->ln_ntick = 0;
559 canceled = callout_stop(&ln->lle_timer);
560 } else {
561 ln->la_expire = time_uptime + tick / hz;
562 LLE_ADDREF(ln);
563 if (tick > INT_MAX) {
564 ln->ln_ntick = tick - INT_MAX;
565 canceled = callout_reset(&ln->lle_timer, INT_MAX,
566 nd6_llinfo_timer, ln);
567 } else {
568 ln->ln_ntick = 0;
569 canceled = callout_reset(&ln->lle_timer, tick,
570 nd6_llinfo_timer, ln);
571 }
572 }
573 if (canceled > 0)
574 LLE_REMREF(ln);
575 }
576
577 /*
578 * Gets source address of the first packet in hold queue
579 * and stores it in @src.
580 * Returns pointer to @src (if hold queue is not empty) or NULL.
581 *
582 * Set noinline to be dtrace-friendly
583 */
584 static __noinline struct in6_addr *
nd6_llinfo_get_holdsrc(struct llentry * ln,struct in6_addr * src)585 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src)
586 {
587 struct ip6_hdr hdr;
588 struct mbuf *m;
589
590 if (ln->la_hold == NULL)
591 return (NULL);
592
593 /*
594 * assume every packet in la_hold has the same IP header
595 */
596 m = ln->la_hold;
597 if (sizeof(hdr) > m->m_len)
598 return (NULL);
599
600 m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr);
601 *src = hdr.ip6_src;
602
603 return (src);
604 }
605
606 /*
607 * Checks if we need to switch from STALE state.
608 *
609 * RFC 4861 requires switching from STALE to DELAY state
610 * on first packet matching entry, waiting V_nd6_delay and
611 * transition to PROBE state (if upper layer confirmation was
612 * not received).
613 *
614 * This code performs a bit differently:
615 * On packet hit we don't change state (but desired state
616 * can be guessed by control plane). However, after V_nd6_delay
617 * seconds code will transition to PROBE state (so DELAY state
618 * is kinda skipped in most situations).
619 *
620 * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so
621 * we perform the following upon entering STALE state:
622 *
623 * 1) Arm timer to run each V_nd6_delay seconds to make sure that
624 * if packet was transmitted at the start of given interval, we
625 * would be able to switch to PROBE state in V_nd6_delay seconds
626 * as user expects.
627 *
628 * 2) Reschedule timer until original V_nd6_gctimer expires keeping
629 * lle in STALE state (remaining timer value stored in lle_remtime).
630 *
631 * 3) Reschedule timer if packet was transmitted less that V_nd6_delay
632 * seconds ago.
633 *
634 * Returns non-zero value if the entry is still STALE (storing
635 * the next timer interval in @pdelay).
636 *
637 * Returns zero value if original timer expired or we need to switch to
638 * PROBE (store that in @do_switch variable).
639 */
640 static int
nd6_is_stale(struct llentry * lle,long * pdelay,int * do_switch)641 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch)
642 {
643 int nd_delay, nd_gctimer;
644 time_t lle_hittime;
645 long delay;
646
647 *do_switch = 0;
648 nd_gctimer = V_nd6_gctimer;
649 nd_delay = V_nd6_delay;
650
651 lle_hittime = llentry_get_hittime(lle);
652
653 if (lle_hittime == 0) {
654 /*
655 * Datapath feedback has been requested upon entering
656 * STALE state. No packets has been passed using this lle.
657 * Ask for the timer reschedule and keep STALE state.
658 */
659 delay = (long)(MIN(nd_gctimer, nd_delay));
660 delay *= hz;
661 if (lle->lle_remtime > delay)
662 lle->lle_remtime -= delay;
663 else {
664 delay = lle->lle_remtime;
665 lle->lle_remtime = 0;
666 }
667
668 if (delay == 0) {
669 /*
670 * The original ng6_gctime timeout ended,
671 * no more rescheduling.
672 */
673 return (0);
674 }
675
676 *pdelay = delay;
677 return (1);
678 }
679
680 /*
681 * Packet received. Verify timestamp
682 */
683 delay = (long)(time_uptime - lle_hittime);
684 if (delay < nd_delay) {
685 /*
686 * V_nd6_delay still not passed since the first
687 * hit in STALE state.
688 * Reschedule timer and return.
689 */
690 *pdelay = (long)(nd_delay - delay) * hz;
691 return (1);
692 }
693
694 /* Request switching to probe */
695 *do_switch = 1;
696 return (0);
697 }
698
699 /*
700 * Switch @lle state to new state optionally arming timers.
701 *
702 * Set noinline to be dtrace-friendly
703 */
704 __noinline void
nd6_llinfo_setstate(struct llentry * lle,int newstate)705 nd6_llinfo_setstate(struct llentry *lle, int newstate)
706 {
707 struct ifnet *ifp;
708 int nd_gctimer, nd_delay;
709 long delay, remtime;
710
711 delay = 0;
712 remtime = 0;
713
714 switch (newstate) {
715 case ND6_LLINFO_INCOMPLETE:
716 ifp = lle->lle_tbl->llt_ifp;
717 delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000;
718 break;
719 case ND6_LLINFO_REACHABLE:
720 if (!ND6_LLINFO_PERMANENT(lle)) {
721 ifp = lle->lle_tbl->llt_ifp;
722 delay = (long)ND_IFINFO(ifp)->reachable * hz;
723 }
724 break;
725 case ND6_LLINFO_STALE:
726
727 llentry_request_feedback(lle);
728 nd_delay = V_nd6_delay;
729 nd_gctimer = V_nd6_gctimer;
730
731 delay = (long)(MIN(nd_gctimer, nd_delay)) * hz;
732 remtime = (long)nd_gctimer * hz - delay;
733 break;
734 case ND6_LLINFO_DELAY:
735 lle->la_asked = 0;
736 delay = (long)V_nd6_delay * hz;
737 break;
738 }
739
740 if (delay > 0)
741 nd6_llinfo_settimer_locked(lle, delay);
742
743 lle->lle_remtime = remtime;
744 lle->ln_state = newstate;
745 }
746
747 /*
748 * Timer-dependent part of nd state machine.
749 *
750 * Set noinline to be dtrace-friendly
751 */
752 static __noinline void
nd6_llinfo_timer(void * arg)753 nd6_llinfo_timer(void *arg)
754 {
755 struct epoch_tracker et;
756 struct llentry *ln;
757 struct in6_addr *dst, *pdst, *psrc, src;
758 struct ifnet *ifp;
759 struct nd_ifinfo *ndi;
760 int do_switch, send_ns;
761 long delay;
762
763 KASSERT(arg != NULL, ("%s: arg NULL", __func__));
764 ln = (struct llentry *)arg;
765 ifp = lltable_get_ifp(ln->lle_tbl);
766 CURVNET_SET(ifp->if_vnet);
767
768 ND6_RLOCK();
769 LLE_WLOCK(ln);
770 if (callout_pending(&ln->lle_timer)) {
771 /*
772 * Here we are a bit odd here in the treatment of
773 * active/pending. If the pending bit is set, it got
774 * rescheduled before I ran. The active
775 * bit we ignore, since if it was stopped
776 * in ll_tablefree() and was currently running
777 * it would have return 0 so the code would
778 * not have deleted it since the callout could
779 * not be stopped so we want to go through
780 * with the delete here now. If the callout
781 * was restarted, the pending bit will be back on and
782 * we just want to bail since the callout_reset would
783 * return 1 and our reference would have been removed
784 * by nd6_llinfo_settimer_locked above since canceled
785 * would have been 1.
786 */
787 LLE_WUNLOCK(ln);
788 ND6_RUNLOCK();
789 CURVNET_RESTORE();
790 return;
791 }
792 NET_EPOCH_ENTER(et);
793 ndi = ND_IFINFO(ifp);
794 send_ns = 0;
795 dst = &ln->r_l3addr.addr6;
796 pdst = dst;
797
798 if (ln->ln_ntick > 0) {
799 if (ln->ln_ntick > INT_MAX) {
800 ln->ln_ntick -= INT_MAX;
801 nd6_llinfo_settimer_locked(ln, INT_MAX);
802 } else {
803 ln->ln_ntick = 0;
804 nd6_llinfo_settimer_locked(ln, ln->ln_ntick);
805 }
806 goto done;
807 }
808
809 if (ln->la_flags & LLE_STATIC) {
810 goto done;
811 }
812
813 if (ln->la_flags & LLE_DELETED) {
814 nd6_free(&ln, 0);
815 goto done;
816 }
817
818 switch (ln->ln_state) {
819 case ND6_LLINFO_INCOMPLETE:
820 if (ln->la_asked < V_nd6_mmaxtries) {
821 ln->la_asked++;
822 send_ns = 1;
823 /* Send NS to multicast address */
824 pdst = NULL;
825 } else {
826 struct mbuf *m;
827
828 ICMP6STAT_ADD(icp6s_dropped, ln->la_numheld);
829
830 m = ln->la_hold;
831 if (m != NULL) {
832 /*
833 * assuming every packet in la_hold has the
834 * same IP header. Send error after unlock.
835 */
836 ln->la_hold = m->m_nextpkt;
837 m->m_nextpkt = NULL;
838 ln->la_numheld--;
839 }
840 nd6_free(&ln, 0);
841 if (m != NULL) {
842 struct mbuf *n = m;
843
844 /*
845 * if there are any ummapped mbufs, we
846 * must free them, rather than using
847 * them for an ICMP, as they cannot be
848 * checksummed.
849 */
850 while ((n = n->m_next) != NULL) {
851 if (n->m_flags & M_EXTPG)
852 break;
853 }
854 if (n != NULL) {
855 m_freem(m);
856 m = NULL;
857 } else {
858 icmp6_error2(m, ICMP6_DST_UNREACH,
859 ICMP6_DST_UNREACH_ADDR, 0, ifp);
860 }
861 }
862 }
863 break;
864 case ND6_LLINFO_REACHABLE:
865 if (!ND6_LLINFO_PERMANENT(ln))
866 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
867 break;
868
869 case ND6_LLINFO_STALE:
870 if (nd6_is_stale(ln, &delay, &do_switch) != 0) {
871 /*
872 * No packet has used this entry and GC timeout
873 * has not been passed. Reschedule timer and
874 * return.
875 */
876 nd6_llinfo_settimer_locked(ln, delay);
877 break;
878 }
879
880 if (do_switch == 0) {
881 /*
882 * GC timer has ended and entry hasn't been used.
883 * Run Garbage collector (RFC 4861, 5.3)
884 */
885 if (!ND6_LLINFO_PERMANENT(ln))
886 nd6_free(&ln, 1);
887 break;
888 }
889
890 /* Entry has been used AND delay timer has ended. */
891
892 /* FALLTHROUGH */
893
894 case ND6_LLINFO_DELAY:
895 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) {
896 /* We need NUD */
897 ln->la_asked = 1;
898 nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE);
899 send_ns = 1;
900 } else
901 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */
902 break;
903 case ND6_LLINFO_PROBE:
904 if (ln->la_asked < V_nd6_umaxtries) {
905 ln->la_asked++;
906 send_ns = 1;
907 } else {
908 nd6_free(&ln, 0);
909 }
910 break;
911 default:
912 panic("%s: paths in a dark night can be confusing: %d",
913 __func__, ln->ln_state);
914 }
915 done:
916 if (ln != NULL)
917 ND6_RUNLOCK();
918 if (send_ns != 0) {
919 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000);
920 psrc = nd6_llinfo_get_holdsrc(ln, &src);
921 LLE_FREE_LOCKED(ln);
922 ln = NULL;
923 nd6_ns_output(ifp, psrc, pdst, dst, NULL);
924 }
925
926 if (ln != NULL)
927 LLE_FREE_LOCKED(ln);
928 NET_EPOCH_EXIT(et);
929 CURVNET_RESTORE();
930 }
931
932 /*
933 * ND6 timer routine to expire default route list and prefix list
934 */
935 void
nd6_timer(void * arg)936 nd6_timer(void *arg)
937 {
938 CURVNET_SET((struct vnet *) arg);
939 struct epoch_tracker et;
940 struct nd_prhead prl;
941 struct nd_prefix *pr, *npr;
942 struct ifnet *ifp;
943 struct in6_ifaddr *ia6, *nia6;
944 uint64_t genid;
945
946 LIST_INIT(&prl);
947
948 NET_EPOCH_ENTER(et);
949 nd6_defrouter_timer();
950
951 /*
952 * expire interface addresses.
953 * in the past the loop was inside prefix expiry processing.
954 * However, from a stricter speci-confrmance standpoint, we should
955 * rather separate address lifetimes and prefix lifetimes.
956 *
957 * XXXRW: in6_ifaddrhead locking.
958 */
959 addrloop:
960 CK_STAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) {
961 /* check address lifetime */
962 if (IFA6_IS_INVALID(ia6)) {
963 int regen = 0;
964
965 /*
966 * If the expiring address is temporary, try
967 * regenerating a new one. This would be useful when
968 * we suspended a laptop PC, then turned it on after a
969 * period that could invalidate all temporary
970 * addresses. Although we may have to restart the
971 * loop (see below), it must be after purging the
972 * address. Otherwise, we'd see an infinite loop of
973 * regeneration.
974 */
975 if (V_ip6_use_tempaddr &&
976 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) {
977 if (regen_tmpaddr(ia6) == 0)
978 regen = 1;
979 }
980
981 in6_purgeaddr(&ia6->ia_ifa);
982
983 if (regen)
984 goto addrloop; /* XXX: see below */
985 } else if (IFA6_IS_DEPRECATED(ia6)) {
986 int oldflags = ia6->ia6_flags;
987
988 ia6->ia6_flags |= IN6_IFF_DEPRECATED;
989
990 /*
991 * If a temporary address has just become deprecated,
992 * regenerate a new one if possible.
993 */
994 if (V_ip6_use_tempaddr &&
995 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
996 (oldflags & IN6_IFF_DEPRECATED) == 0) {
997 if (regen_tmpaddr(ia6) == 0) {
998 /*
999 * A new temporary address is
1000 * generated.
1001 * XXX: this means the address chain
1002 * has changed while we are still in
1003 * the loop. Although the change
1004 * would not cause disaster (because
1005 * it's not a deletion, but an
1006 * addition,) we'd rather restart the
1007 * loop just for safety. Or does this
1008 * significantly reduce performance??
1009 */
1010 goto addrloop;
1011 }
1012 }
1013 } else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) {
1014 /*
1015 * Schedule DAD for a tentative address. This happens
1016 * if the interface was down or not running
1017 * when the address was configured.
1018 */
1019 int delay;
1020
1021 delay = arc4random() %
1022 (MAX_RTR_SOLICITATION_DELAY * hz);
1023 nd6_dad_start((struct ifaddr *)ia6, delay);
1024 } else {
1025 /*
1026 * Check status of the interface. If it is down,
1027 * mark the address as tentative for future DAD.
1028 */
1029 ifp = ia6->ia_ifp;
1030 if ((ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0 &&
1031 ((ifp->if_flags & IFF_UP) == 0 ||
1032 (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
1033 (ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) != 0)){
1034 ia6->ia6_flags &= ~IN6_IFF_DUPLICATED;
1035 ia6->ia6_flags |= IN6_IFF_TENTATIVE;
1036 }
1037
1038 /*
1039 * A new RA might have made a deprecated address
1040 * preferred.
1041 */
1042 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED;
1043 }
1044 }
1045 NET_EPOCH_EXIT(et);
1046
1047 ND6_WLOCK();
1048 restart:
1049 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1050 /*
1051 * Expire prefixes. Since the pltime is only used for
1052 * autoconfigured addresses, pltime processing for prefixes is
1053 * not necessary.
1054 *
1055 * Only unlink after all derived addresses have expired. This
1056 * may not occur until two hours after the prefix has expired
1057 * per RFC 4862. If the prefix expires before its derived
1058 * addresses, mark it off-link. This will be done automatically
1059 * after unlinking if no address references remain.
1060 */
1061 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME ||
1062 time_uptime - pr->ndpr_lastupdate <= pr->ndpr_vltime)
1063 continue;
1064
1065 if (pr->ndpr_addrcnt == 0) {
1066 nd6_prefix_unlink(pr, &prl);
1067 continue;
1068 }
1069 if ((pr->ndpr_stateflags & NDPRF_ONLINK) != 0) {
1070 genid = V_nd6_list_genid;
1071 nd6_prefix_ref(pr);
1072 ND6_WUNLOCK();
1073 ND6_ONLINK_LOCK();
1074 (void)nd6_prefix_offlink(pr);
1075 ND6_ONLINK_UNLOCK();
1076 ND6_WLOCK();
1077 nd6_prefix_rele(pr);
1078 if (genid != V_nd6_list_genid)
1079 goto restart;
1080 }
1081 }
1082 ND6_WUNLOCK();
1083
1084 while ((pr = LIST_FIRST(&prl)) != NULL) {
1085 LIST_REMOVE(pr, ndpr_entry);
1086 nd6_prefix_del(pr);
1087 }
1088
1089 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz,
1090 nd6_timer, curvnet);
1091
1092 CURVNET_RESTORE();
1093 }
1094
1095 /*
1096 * ia6 - deprecated/invalidated temporary address
1097 */
1098 static int
regen_tmpaddr(struct in6_ifaddr * ia6)1099 regen_tmpaddr(struct in6_ifaddr *ia6)
1100 {
1101 struct ifaddr *ifa;
1102 struct ifnet *ifp;
1103 struct in6_ifaddr *public_ifa6 = NULL;
1104
1105 NET_EPOCH_ASSERT();
1106
1107 ifp = ia6->ia_ifa.ifa_ifp;
1108 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1109 struct in6_ifaddr *it6;
1110
1111 if (ifa->ifa_addr->sa_family != AF_INET6)
1112 continue;
1113
1114 it6 = (struct in6_ifaddr *)ifa;
1115
1116 /* ignore no autoconf addresses. */
1117 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1118 continue;
1119
1120 /* ignore autoconf addresses with different prefixes. */
1121 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr)
1122 continue;
1123
1124 /*
1125 * Now we are looking at an autoconf address with the same
1126 * prefix as ours. If the address is temporary and is still
1127 * preferred, do not create another one. It would be rare, but
1128 * could happen, for example, when we resume a laptop PC after
1129 * a long period.
1130 */
1131 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 &&
1132 !IFA6_IS_DEPRECATED(it6)) {
1133 public_ifa6 = NULL;
1134 break;
1135 }
1136
1137 /*
1138 * This is a public autoconf address that has the same prefix
1139 * as ours. If it is preferred, keep it. We can't break the
1140 * loop here, because there may be a still-preferred temporary
1141 * address with the prefix.
1142 */
1143 if (!IFA6_IS_DEPRECATED(it6))
1144 public_ifa6 = it6;
1145 }
1146 if (public_ifa6 != NULL)
1147 ifa_ref(&public_ifa6->ia_ifa);
1148
1149 if (public_ifa6 != NULL) {
1150 int e;
1151
1152 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) {
1153 ifa_free(&public_ifa6->ia_ifa);
1154 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new"
1155 " tmp addr,errno=%d\n", e);
1156 return (-1);
1157 }
1158 ifa_free(&public_ifa6->ia_ifa);
1159 return (0);
1160 }
1161
1162 return (-1);
1163 }
1164
1165 /*
1166 * Remove prefix and default router list entries corresponding to ifp. Neighbor
1167 * cache entries are freed in in6_domifdetach().
1168 */
1169 void
nd6_purge(struct ifnet * ifp)1170 nd6_purge(struct ifnet *ifp)
1171 {
1172 struct nd_prhead prl;
1173 struct nd_prefix *pr, *npr;
1174
1175 LIST_INIT(&prl);
1176
1177 /* Purge default router list entries toward ifp. */
1178 nd6_defrouter_purge(ifp);
1179
1180 ND6_WLOCK();
1181 /*
1182 * Remove prefixes on ifp. We should have already removed addresses on
1183 * this interface, so no addresses should be referencing these prefixes.
1184 */
1185 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) {
1186 if (pr->ndpr_ifp == ifp)
1187 nd6_prefix_unlink(pr, &prl);
1188 }
1189 ND6_WUNLOCK();
1190
1191 /* Delete the unlinked prefix objects. */
1192 while ((pr = LIST_FIRST(&prl)) != NULL) {
1193 LIST_REMOVE(pr, ndpr_entry);
1194 nd6_prefix_del(pr);
1195 }
1196
1197 /* cancel default outgoing interface setting */
1198 if (V_nd6_defifindex == ifp->if_index)
1199 nd6_setdefaultiface(0);
1200
1201 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1202 /* Refresh default router list. */
1203 defrouter_select_fib(ifp->if_fib);
1204 }
1205 }
1206
1207 /*
1208 * the caller acquires and releases the lock on the lltbls
1209 * Returns the llentry locked
1210 */
1211 struct llentry *
nd6_lookup(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1212 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1213 {
1214 struct sockaddr_in6 sin6;
1215 struct llentry *ln;
1216
1217 bzero(&sin6, sizeof(sin6));
1218 sin6.sin6_len = sizeof(struct sockaddr_in6);
1219 sin6.sin6_family = AF_INET6;
1220 sin6.sin6_addr = *addr6;
1221
1222 LLTABLE_RLOCK_ASSERT(LLTABLE6(ifp));
1223
1224 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6);
1225
1226 return (ln);
1227 }
1228
1229 static struct llentry *
nd6_alloc(const struct in6_addr * addr6,int flags,struct ifnet * ifp)1230 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp)
1231 {
1232 struct sockaddr_in6 sin6;
1233 struct llentry *ln;
1234
1235 bzero(&sin6, sizeof(sin6));
1236 sin6.sin6_len = sizeof(struct sockaddr_in6);
1237 sin6.sin6_family = AF_INET6;
1238 sin6.sin6_addr = *addr6;
1239
1240 ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6);
1241 if (ln != NULL)
1242 ln->ln_state = ND6_LLINFO_NOSTATE;
1243
1244 return (ln);
1245 }
1246
1247 /*
1248 * Test whether a given IPv6 address can be a neighbor.
1249 */
1250 static bool
nd6_is_new_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1251 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1252 {
1253
1254 /*
1255 * A link-local address is always a neighbor.
1256 * XXX: a link does not necessarily specify a single interface.
1257 */
1258 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) {
1259 struct sockaddr_in6 sin6_copy;
1260 u_int32_t zone;
1261
1262 /*
1263 * We need sin6_copy since sa6_recoverscope() may modify the
1264 * content (XXX).
1265 */
1266 sin6_copy = *addr;
1267 if (sa6_recoverscope(&sin6_copy))
1268 return (0); /* XXX: should be impossible */
1269 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone))
1270 return (0);
1271 if (sin6_copy.sin6_scope_id == zone)
1272 return (1);
1273 else
1274 return (0);
1275 }
1276 /* Checking global unicast */
1277
1278 /* If an address is directly reachable, it is a neigbor */
1279 struct nhop_object *nh;
1280 nh = fib6_lookup(ifp->if_fib, &addr->sin6_addr, 0, NHR_NONE, 0);
1281 if (nh != NULL && nh->nh_aifp == ifp && (nh->nh_flags & NHF_GATEWAY) == 0)
1282 return (true);
1283
1284 /*
1285 * Check prefixes with desired on-link state, as some may be not
1286 * installed in the routing table.
1287 */
1288 bool matched = false;
1289 struct nd_prefix *pr;
1290 ND6_RLOCK();
1291 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
1292 if (pr->ndpr_ifp != ifp)
1293 continue;
1294 if ((pr->ndpr_stateflags & NDPRF_ONLINK) == 0)
1295 continue;
1296 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr,
1297 &addr->sin6_addr, &pr->ndpr_mask)) {
1298 matched = true;
1299 break;
1300 }
1301 }
1302 ND6_RUNLOCK();
1303 if (matched)
1304 return (true);
1305
1306 /*
1307 * If the address is assigned on the node of the other side of
1308 * a p2p interface, the address should be a neighbor.
1309 */
1310 if (ifp->if_flags & IFF_POINTOPOINT) {
1311 struct ifaddr *ifa;
1312
1313 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1314 if (ifa->ifa_addr->sa_family != addr->sin6_family)
1315 continue;
1316 if (ifa->ifa_dstaddr != NULL &&
1317 sa_equal(addr, ifa->ifa_dstaddr)) {
1318 return (true);
1319 }
1320 }
1321 }
1322
1323 /*
1324 * If the default router list is empty, all addresses are regarded
1325 * as on-link, and thus, as a neighbor.
1326 */
1327 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV &&
1328 nd6_defrouter_list_empty() &&
1329 V_nd6_defifindex == ifp->if_index) {
1330 return (1);
1331 }
1332
1333 return (0);
1334 }
1335
1336 /*
1337 * Detect if a given IPv6 address identifies a neighbor on a given link.
1338 * XXX: should take care of the destination of a p2p link?
1339 */
1340 int
nd6_is_addr_neighbor(const struct sockaddr_in6 * addr,struct ifnet * ifp)1341 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp)
1342 {
1343 struct llentry *lle;
1344 int rc = 0;
1345
1346 NET_EPOCH_ASSERT();
1347
1348 if (nd6_is_new_addr_neighbor(addr, ifp))
1349 return (1);
1350
1351 /*
1352 * Even if the address matches none of our addresses, it might be
1353 * in the neighbor cache.
1354 */
1355 if ((lle = nd6_lookup(&addr->sin6_addr, LLE_SF(AF_INET6, 0), ifp)) != NULL) {
1356 LLE_RUNLOCK(lle);
1357 rc = 1;
1358 }
1359 return (rc);
1360 }
1361
1362 static __noinline void
nd6_free_children(struct llentry * lle)1363 nd6_free_children(struct llentry *lle)
1364 {
1365 struct llentry *child_lle;
1366
1367 NET_EPOCH_ASSERT();
1368 LLE_WLOCK_ASSERT(lle);
1369
1370 while ((child_lle = CK_SLIST_FIRST(&lle->lle_children)) != NULL) {
1371 LLE_WLOCK(child_lle);
1372 lltable_unlink_child_entry(child_lle);
1373 llentry_free(child_lle);
1374 }
1375 }
1376
1377 /*
1378 * Tries to update @lle address/prepend data with new @lladdr.
1379 *
1380 * Returns true on success.
1381 * In any case, @lle is returned wlocked.
1382 */
1383 static __noinline bool
nd6_try_set_entry_addr_locked(struct ifnet * ifp,struct llentry * lle,char * lladdr)1384 nd6_try_set_entry_addr_locked(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1385 {
1386 u_char buf[LLE_MAX_LINKHDR];
1387 int fam, off;
1388 size_t sz;
1389
1390 sz = sizeof(buf);
1391 if (lltable_calc_llheader(ifp, AF_INET6, lladdr, buf, &sz, &off) != 0)
1392 return (false);
1393
1394 /* Update data */
1395 lltable_set_entry_addr(ifp, lle, buf, sz, off);
1396
1397 struct llentry *child_lle;
1398 CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
1399 LLE_WLOCK(child_lle);
1400 fam = child_lle->r_family;
1401 sz = sizeof(buf);
1402 if (lltable_calc_llheader(ifp, fam, lladdr, buf, &sz, &off) == 0) {
1403 /* success */
1404 lltable_set_entry_addr(ifp, child_lle, buf, sz, off);
1405 child_lle->ln_state = ND6_LLINFO_REACHABLE;
1406 }
1407 LLE_WUNLOCK(child_lle);
1408 }
1409
1410 return (true);
1411 }
1412
1413 bool
nd6_try_set_entry_addr(struct ifnet * ifp,struct llentry * lle,char * lladdr)1414 nd6_try_set_entry_addr(struct ifnet *ifp, struct llentry *lle, char *lladdr)
1415 {
1416 NET_EPOCH_ASSERT();
1417 LLE_WLOCK_ASSERT(lle);
1418
1419 if (!lltable_trylock(lle))
1420 return (false);
1421 bool ret = nd6_try_set_entry_addr_locked(ifp, lle, lladdr);
1422 LLTABLE_UNLOCK(lle->lle_tbl);
1423
1424 return (ret);
1425 }
1426
1427 /*
1428 * Free an nd6 llinfo entry.
1429 * Since the function would cause significant changes in the kernel, DO NOT
1430 * make it global, unless you have a strong reason for the change, and are sure
1431 * that the change is safe.
1432 *
1433 * Set noinline to be dtrace-friendly
1434 */
1435 static __noinline void
nd6_free(struct llentry ** lnp,int gc)1436 nd6_free(struct llentry **lnp, int gc)
1437 {
1438 struct ifnet *ifp;
1439 struct llentry *ln;
1440 struct nd_defrouter *dr;
1441
1442 ln = *lnp;
1443 *lnp = NULL;
1444
1445 LLE_WLOCK_ASSERT(ln);
1446 ND6_RLOCK_ASSERT();
1447
1448 KASSERT((ln->la_flags & LLE_CHILD) == 0, ("child lle"));
1449
1450 ifp = lltable_get_ifp(ln->lle_tbl);
1451 if ((ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) != 0)
1452 dr = defrouter_lookup_locked(&ln->r_l3addr.addr6, ifp);
1453 else
1454 dr = NULL;
1455 ND6_RUNLOCK();
1456
1457 if ((ln->la_flags & LLE_DELETED) == 0)
1458 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED);
1459
1460 /*
1461 * we used to have pfctlinput(PRC_HOSTDEAD) here.
1462 * even though it is not harmful, it was not really necessary.
1463 */
1464
1465 /* cancel timer */
1466 nd6_llinfo_settimer_locked(ln, -1);
1467
1468 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
1469 if (dr != NULL && dr->expire &&
1470 ln->ln_state == ND6_LLINFO_STALE && gc) {
1471 /*
1472 * If the reason for the deletion is just garbage
1473 * collection, and the neighbor is an active default
1474 * router, do not delete it. Instead, reset the GC
1475 * timer using the router's lifetime.
1476 * Simply deleting the entry would affect default
1477 * router selection, which is not necessarily a good
1478 * thing, especially when we're using router preference
1479 * values.
1480 * XXX: the check for ln_state would be redundant,
1481 * but we intentionally keep it just in case.
1482 */
1483 if (dr->expire > time_uptime)
1484 nd6_llinfo_settimer_locked(ln,
1485 (dr->expire - time_uptime) * hz);
1486 else
1487 nd6_llinfo_settimer_locked(ln,
1488 (long)V_nd6_gctimer * hz);
1489
1490 LLE_REMREF(ln);
1491 LLE_WUNLOCK(ln);
1492 defrouter_rele(dr);
1493 return;
1494 }
1495
1496 if (dr) {
1497 /*
1498 * Unreachability of a router might affect the default
1499 * router selection and on-link detection of advertised
1500 * prefixes.
1501 */
1502
1503 /*
1504 * Temporarily fake the state to choose a new default
1505 * router and to perform on-link determination of
1506 * prefixes correctly.
1507 * Below the state will be set correctly,
1508 * or the entry itself will be deleted.
1509 */
1510 ln->ln_state = ND6_LLINFO_INCOMPLETE;
1511 }
1512
1513 if (ln->ln_router || dr) {
1514 /*
1515 * We need to unlock to avoid a LOR with rt6_flush() with the
1516 * rnh and for the calls to pfxlist_onlink_check() and
1517 * defrouter_select_fib() in the block further down for calls
1518 * into nd6_lookup(). We still hold a ref.
1519 */
1520 LLE_WUNLOCK(ln);
1521
1522 /*
1523 * rt6_flush must be called whether or not the neighbor
1524 * is in the Default Router List.
1525 * See a corresponding comment in nd6_na_input().
1526 */
1527 rt6_flush(&ln->r_l3addr.addr6, ifp);
1528 }
1529
1530 if (dr) {
1531 /*
1532 * Since defrouter_select_fib() does not affect the
1533 * on-link determination and MIP6 needs the check
1534 * before the default router selection, we perform
1535 * the check now.
1536 */
1537 pfxlist_onlink_check();
1538
1539 /*
1540 * Refresh default router list.
1541 */
1542 defrouter_select_fib(dr->ifp->if_fib);
1543 }
1544
1545 /*
1546 * If this entry was added by an on-link redirect, remove the
1547 * corresponding host route.
1548 */
1549 if (ln->la_flags & LLE_REDIRECT)
1550 nd6_free_redirect(ln);
1551
1552 if (ln->ln_router || dr)
1553 LLE_WLOCK(ln);
1554 }
1555
1556 /*
1557 * Save to unlock. We still hold an extra reference and will not
1558 * free(9) in llentry_free() if someone else holds one as well.
1559 */
1560 LLE_WUNLOCK(ln);
1561 LLTABLE_LOCK(ln->lle_tbl);
1562 LLE_WLOCK(ln);
1563 /* Guard against race with other llentry_free(). */
1564 if (ln->la_flags & LLE_LINKED) {
1565 /* Remove callout reference */
1566 LLE_REMREF(ln);
1567 lltable_unlink_entry(ln->lle_tbl, ln);
1568 }
1569 LLTABLE_UNLOCK(ln->lle_tbl);
1570
1571 nd6_free_children(ln);
1572
1573 llentry_free(ln);
1574 if (dr != NULL)
1575 defrouter_rele(dr);
1576 }
1577
1578 static int
nd6_isdynrte(const struct rtentry * rt,const struct nhop_object * nh,void * xap)1579 nd6_isdynrte(const struct rtentry *rt, const struct nhop_object *nh, void *xap)
1580 {
1581
1582 if (nh->nh_flags & NHF_REDIRECT)
1583 return (1);
1584
1585 return (0);
1586 }
1587
1588 /*
1589 * Remove the rtentry for the given llentry,
1590 * both of which were installed by a redirect.
1591 */
1592 static void
nd6_free_redirect(const struct llentry * ln)1593 nd6_free_redirect(const struct llentry *ln)
1594 {
1595 int fibnum;
1596 struct sockaddr_in6 sin6;
1597 struct rib_cmd_info rc;
1598 struct epoch_tracker et;
1599
1600 lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6);
1601
1602 NET_EPOCH_ENTER(et);
1603 for (fibnum = 0; fibnum < rt_numfibs; fibnum++)
1604 rib_del_route_px(fibnum, (struct sockaddr *)&sin6, 128,
1605 nd6_isdynrte, NULL, 0, &rc);
1606 NET_EPOCH_EXIT(et);
1607 }
1608
1609 /*
1610 * Updates status of the default router route.
1611 */
1612 static void
check_release_defrouter(const struct rib_cmd_info * rc,void * _cbdata)1613 check_release_defrouter(const struct rib_cmd_info *rc, void *_cbdata)
1614 {
1615 struct nd_defrouter *dr;
1616 struct nhop_object *nh;
1617
1618 nh = rc->rc_nh_old;
1619 if (rc->rc_cmd == RTM_DELETE && (nh->nh_flags & NHF_DEFAULT) != 0) {
1620 dr = defrouter_lookup(&nh->gw6_sa.sin6_addr, nh->nh_ifp);
1621 if (dr != NULL) {
1622 dr->installed = 0;
1623 defrouter_rele(dr);
1624 }
1625 }
1626 }
1627
1628 void
nd6_subscription_cb(struct rib_head * rnh,struct rib_cmd_info * rc,void * arg)1629 nd6_subscription_cb(struct rib_head *rnh, struct rib_cmd_info *rc, void *arg)
1630 {
1631 #ifdef ROUTE_MPATH
1632 rib_decompose_notification(rc, check_release_defrouter, NULL);
1633 if (rc->rc_cmd == RTM_DELETE && !NH_IS_NHGRP(rc->rc_nh_old))
1634 check_release_defrouter(rc, NULL);
1635 #else
1636 check_release_defrouter(rc, NULL);
1637 #endif
1638 }
1639
1640 int
nd6_ioctl(u_long cmd,caddr_t data,struct ifnet * ifp)1641 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp)
1642 {
1643 struct in6_ndireq *ndi = (struct in6_ndireq *)data;
1644 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data;
1645 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data;
1646 struct epoch_tracker et;
1647 int error = 0;
1648
1649 /* XXXGL: ??? */
1650 if (ifp->if_inet6 == NULL)
1651 return (EPFNOSUPPORT);
1652 switch (cmd) {
1653 case OSIOCGIFINFO_IN6:
1654 #define ND ndi->ndi
1655 /* XXX: old ndp(8) assumes a positive value for linkmtu. */
1656 bzero(&ND, sizeof(ND));
1657 ND.linkmtu = IN6_LINKMTU(ifp);
1658 ND.maxmtu = ND_IFINFO(ifp)->maxmtu;
1659 ND.basereachable = ND_IFINFO(ifp)->basereachable;
1660 ND.reachable = ND_IFINFO(ifp)->reachable;
1661 ND.retrans = ND_IFINFO(ifp)->retrans;
1662 ND.flags = ND_IFINFO(ifp)->flags;
1663 ND.recalctm = ND_IFINFO(ifp)->recalctm;
1664 ND.chlim = ND_IFINFO(ifp)->chlim;
1665 break;
1666 case SIOCGIFINFO_IN6:
1667 ND = *ND_IFINFO(ifp);
1668 break;
1669 case SIOCSIFINFO_IN6:
1670 /*
1671 * used to change host variables from userland.
1672 * intended for a use on router to reflect RA configurations.
1673 */
1674 /* 0 means 'unspecified' */
1675 if (ND.linkmtu != 0) {
1676 if (ND.linkmtu < IPV6_MMTU ||
1677 ND.linkmtu > IN6_LINKMTU(ifp)) {
1678 error = EINVAL;
1679 break;
1680 }
1681 ND_IFINFO(ifp)->linkmtu = ND.linkmtu;
1682 }
1683
1684 if (ND.basereachable != 0) {
1685 int obasereachable = ND_IFINFO(ifp)->basereachable;
1686
1687 ND_IFINFO(ifp)->basereachable = ND.basereachable;
1688 if (ND.basereachable != obasereachable)
1689 ND_IFINFO(ifp)->reachable =
1690 ND_COMPUTE_RTIME(ND.basereachable);
1691 }
1692 if (ND.retrans != 0)
1693 ND_IFINFO(ifp)->retrans = ND.retrans;
1694 if (ND.chlim != 0)
1695 ND_IFINFO(ifp)->chlim = ND.chlim;
1696 /* FALLTHROUGH */
1697 case SIOCSIFINFO_FLAGS:
1698 {
1699 struct ifaddr *ifa;
1700 struct in6_ifaddr *ia;
1701
1702 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1703 !(ND.flags & ND6_IFF_IFDISABLED)) {
1704 /* ifdisabled 1->0 transision */
1705
1706 /*
1707 * If the interface is marked as ND6_IFF_IFDISABLED and
1708 * has an link-local address with IN6_IFF_DUPLICATED,
1709 * do not clear ND6_IFF_IFDISABLED.
1710 * See RFC 4862, Section 5.4.5.
1711 */
1712 NET_EPOCH_ENTER(et);
1713 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1714 if (ifa->ifa_addr->sa_family != AF_INET6)
1715 continue;
1716 ia = (struct in6_ifaddr *)ifa;
1717 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) &&
1718 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1719 break;
1720 }
1721 NET_EPOCH_EXIT(et);
1722
1723 if (ifa != NULL) {
1724 /* LLA is duplicated. */
1725 ND.flags |= ND6_IFF_IFDISABLED;
1726 log(LOG_ERR, "Cannot enable an interface"
1727 " with a link-local address marked"
1728 " duplicate.\n");
1729 } else {
1730 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED;
1731 if (ifp->if_flags & IFF_UP)
1732 in6_if_up(ifp);
1733 }
1734 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) &&
1735 (ND.flags & ND6_IFF_IFDISABLED)) {
1736 /* ifdisabled 0->1 transision */
1737 /* Mark all IPv6 address as tentative. */
1738
1739 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED;
1740 if (V_ip6_dad_count > 0 &&
1741 (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) {
1742 NET_EPOCH_ENTER(et);
1743 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1744 ifa_link) {
1745 if (ifa->ifa_addr->sa_family !=
1746 AF_INET6)
1747 continue;
1748 ia = (struct in6_ifaddr *)ifa;
1749 ia->ia6_flags |= IN6_IFF_TENTATIVE;
1750 }
1751 NET_EPOCH_EXIT(et);
1752 }
1753 }
1754
1755 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) {
1756 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) {
1757 /* auto_linklocal 0->1 transision */
1758
1759 /* If no link-local address on ifp, configure */
1760 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL;
1761 in6_ifattach(ifp, NULL);
1762 } else if (!(ND.flags & ND6_IFF_IFDISABLED) &&
1763 ifp->if_flags & IFF_UP) {
1764 /*
1765 * When the IF already has
1766 * ND6_IFF_AUTO_LINKLOCAL, no link-local
1767 * address is assigned, and IFF_UP, try to
1768 * assign one.
1769 */
1770 NET_EPOCH_ENTER(et);
1771 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead,
1772 ifa_link) {
1773 if (ifa->ifa_addr->sa_family !=
1774 AF_INET6)
1775 continue;
1776 ia = (struct in6_ifaddr *)ifa;
1777 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia)))
1778 break;
1779 }
1780 NET_EPOCH_EXIT(et);
1781 if (ifa != NULL)
1782 /* No LLA is configured. */
1783 in6_ifattach(ifp, NULL);
1784 }
1785 }
1786 ND_IFINFO(ifp)->flags = ND.flags;
1787 break;
1788 }
1789 #undef ND
1790 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */
1791 /* sync kernel routing table with the default router list */
1792 defrouter_reset();
1793 defrouter_select_fib(RT_ALL_FIBS);
1794 break;
1795 case SIOCSPFXFLUSH_IN6:
1796 {
1797 /* flush all the prefix advertised by routers */
1798 struct in6_ifaddr *ia, *ia_next;
1799 struct nd_prefix *pr, *next;
1800 struct nd_prhead prl;
1801
1802 LIST_INIT(&prl);
1803
1804 ND6_WLOCK();
1805 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) {
1806 if (ND6_PREFIX_WITH_ROUTER(pr))
1807 nd6_prefix_unlink(pr, &prl);
1808 }
1809 ND6_WUNLOCK();
1810
1811 while ((pr = LIST_FIRST(&prl)) != NULL) {
1812 LIST_REMOVE(pr, ndpr_entry);
1813 /* XXXRW: in6_ifaddrhead locking. */
1814 CK_STAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link,
1815 ia_next) {
1816 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0)
1817 continue;
1818
1819 if (ia->ia6_ndpr == pr)
1820 in6_purgeaddr(&ia->ia_ifa);
1821 }
1822 nd6_prefix_del(pr);
1823 }
1824 break;
1825 }
1826 case SIOCSRTRFLUSH_IN6:
1827 {
1828 /* flush all the default routers */
1829
1830 defrouter_reset();
1831 nd6_defrouter_flush_all();
1832 defrouter_select_fib(RT_ALL_FIBS);
1833 break;
1834 }
1835 case SIOCGNBRINFO_IN6:
1836 {
1837 struct llentry *ln;
1838 struct in6_addr nb_addr = nbi->addr; /* make local for safety */
1839
1840 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0)
1841 return (error);
1842
1843 NET_EPOCH_ENTER(et);
1844 ln = nd6_lookup(&nb_addr, LLE_SF(AF_INET6, 0), ifp);
1845 NET_EPOCH_EXIT(et);
1846
1847 if (ln == NULL) {
1848 error = EINVAL;
1849 break;
1850 }
1851 nbi->state = ln->ln_state;
1852 nbi->asked = ln->la_asked;
1853 nbi->isrouter = ln->ln_router;
1854 if (ln->la_expire == 0)
1855 nbi->expire = 0;
1856 else
1857 nbi->expire = ln->la_expire + ln->lle_remtime / hz +
1858 (time_second - time_uptime);
1859 LLE_RUNLOCK(ln);
1860 break;
1861 }
1862 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1863 ndif->ifindex = V_nd6_defifindex;
1864 break;
1865 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */
1866 return (nd6_setdefaultiface(ndif->ifindex));
1867 }
1868 return (error);
1869 }
1870
1871 /*
1872 * Calculates new isRouter value based on provided parameters and
1873 * returns it.
1874 */
1875 static int
nd6_is_router(int type,int code,int is_new,int old_addr,int new_addr,int ln_router)1876 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr,
1877 int ln_router)
1878 {
1879
1880 /*
1881 * ICMP6 type dependent behavior.
1882 *
1883 * NS: clear IsRouter if new entry
1884 * RS: clear IsRouter
1885 * RA: set IsRouter if there's lladdr
1886 * redir: clear IsRouter if new entry
1887 *
1888 * RA case, (1):
1889 * The spec says that we must set IsRouter in the following cases:
1890 * - If lladdr exist, set IsRouter. This means (1-5).
1891 * - If it is old entry (!newentry), set IsRouter. This means (7).
1892 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter.
1893 * A quetion arises for (1) case. (1) case has no lladdr in the
1894 * neighbor cache, this is similar to (6).
1895 * This case is rare but we figured that we MUST NOT set IsRouter.
1896 *
1897 * is_new old_addr new_addr NS RS RA redir
1898 * D R
1899 * 0 n n (1) c ? s
1900 * 0 y n (2) c s s
1901 * 0 n y (3) c s s
1902 * 0 y y (4) c s s
1903 * 0 y y (5) c s s
1904 * 1 -- n (6) c c c s
1905 * 1 -- y (7) c c s c s
1906 *
1907 * (c=clear s=set)
1908 */
1909 switch (type & 0xff) {
1910 case ND_NEIGHBOR_SOLICIT:
1911 /*
1912 * New entry must have is_router flag cleared.
1913 */
1914 if (is_new) /* (6-7) */
1915 ln_router = 0;
1916 break;
1917 case ND_REDIRECT:
1918 /*
1919 * If the icmp is a redirect to a better router, always set the
1920 * is_router flag. Otherwise, if the entry is newly created,
1921 * clear the flag. [RFC 2461, sec 8.3]
1922 */
1923 if (code == ND_REDIRECT_ROUTER)
1924 ln_router = 1;
1925 else {
1926 if (is_new) /* (6-7) */
1927 ln_router = 0;
1928 }
1929 break;
1930 case ND_ROUTER_SOLICIT:
1931 /*
1932 * is_router flag must always be cleared.
1933 */
1934 ln_router = 0;
1935 break;
1936 case ND_ROUTER_ADVERT:
1937 /*
1938 * Mark an entry with lladdr as a router.
1939 */
1940 if ((!is_new && (old_addr || new_addr)) || /* (2-5) */
1941 (is_new && new_addr)) { /* (7) */
1942 ln_router = 1;
1943 }
1944 break;
1945 }
1946
1947 return (ln_router);
1948 }
1949
1950 /*
1951 * Create neighbor cache entry and cache link-layer address,
1952 * on reception of inbound ND6 packets. (RS/RA/NS/redirect)
1953 *
1954 * type - ICMP6 type
1955 * code - type dependent information
1956 *
1957 */
1958 void
nd6_cache_lladdr(struct ifnet * ifp,struct in6_addr * from,char * lladdr,int lladdrlen,int type,int code)1959 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr,
1960 int lladdrlen, int type, int code)
1961 {
1962 struct llentry *ln = NULL, *ln_tmp;
1963 int is_newentry;
1964 int do_update;
1965 int olladdr;
1966 int llchange;
1967 int flags;
1968 uint16_t router = 0;
1969 struct mbuf *chain = NULL;
1970 u_char linkhdr[LLE_MAX_LINKHDR];
1971 size_t linkhdrsize;
1972 int lladdr_off;
1973
1974 NET_EPOCH_ASSERT();
1975
1976 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__));
1977 KASSERT(from != NULL, ("%s: from == NULL", __func__));
1978
1979 /* nothing must be updated for unspecified address */
1980 if (IN6_IS_ADDR_UNSPECIFIED(from))
1981 return;
1982
1983 /*
1984 * Validation about ifp->if_addrlen and lladdrlen must be done in
1985 * the caller.
1986 *
1987 * XXX If the link does not have link-layer adderss, what should
1988 * we do? (ifp->if_addrlen == 0)
1989 * Spec says nothing in sections for RA, RS and NA. There's small
1990 * description on it in NS section (RFC 2461 7.2.3).
1991 */
1992 flags = lladdr ? LLE_EXCLUSIVE : 0;
1993 ln = nd6_lookup(from, LLE_SF(AF_INET6, flags), ifp);
1994 is_newentry = 0;
1995 if (ln == NULL) {
1996 flags |= LLE_EXCLUSIVE;
1997 ln = nd6_alloc(from, 0, ifp);
1998 if (ln == NULL)
1999 return;
2000
2001 /*
2002 * Since we already know all the data for the new entry,
2003 * fill it before insertion.
2004 */
2005 if (lladdr != NULL) {
2006 linkhdrsize = sizeof(linkhdr);
2007 if (lltable_calc_llheader(ifp, AF_INET6, lladdr,
2008 linkhdr, &linkhdrsize, &lladdr_off) != 0) {
2009 lltable_free_entry(LLTABLE6(ifp), ln);
2010 return;
2011 }
2012 lltable_set_entry_addr(ifp, ln, linkhdr, linkhdrsize,
2013 lladdr_off);
2014 }
2015
2016 LLTABLE_LOCK(LLTABLE6(ifp));
2017 LLE_WLOCK(ln);
2018 /* Prefer any existing lle over newly-created one */
2019 ln_tmp = nd6_lookup(from, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2020 if (ln_tmp == NULL)
2021 lltable_link_entry(LLTABLE6(ifp), ln);
2022 LLTABLE_UNLOCK(LLTABLE6(ifp));
2023 if (ln_tmp == NULL) {
2024 /* No existing lle, mark as new entry (6,7) */
2025 is_newentry = 1;
2026 if (lladdr != NULL) { /* (7) */
2027 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2028 EVENTHANDLER_INVOKE(lle_event, ln,
2029 LLENTRY_RESOLVED);
2030 }
2031 } else {
2032 lltable_free_entry(LLTABLE6(ifp), ln);
2033 ln = ln_tmp;
2034 ln_tmp = NULL;
2035 }
2036 }
2037 /* do nothing if static ndp is set */
2038 if ((ln->la_flags & LLE_STATIC)) {
2039 if (flags & LLE_EXCLUSIVE)
2040 LLE_WUNLOCK(ln);
2041 else
2042 LLE_RUNLOCK(ln);
2043 return;
2044 }
2045
2046 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0;
2047 if (olladdr && lladdr) {
2048 llchange = bcmp(lladdr, ln->ll_addr,
2049 ifp->if_addrlen);
2050 } else if (!olladdr && lladdr)
2051 llchange = 1;
2052 else
2053 llchange = 0;
2054
2055 /*
2056 * newentry olladdr lladdr llchange (*=record)
2057 * 0 n n -- (1)
2058 * 0 y n -- (2)
2059 * 0 n y y (3) * STALE
2060 * 0 y y n (4) *
2061 * 0 y y y (5) * STALE
2062 * 1 -- n -- (6) NOSTATE(= PASSIVE)
2063 * 1 -- y -- (7) * STALE
2064 */
2065
2066 do_update = 0;
2067 if (is_newentry == 0 && llchange != 0) {
2068 do_update = 1; /* (3,5) */
2069
2070 /*
2071 * Record source link-layer address
2072 * XXX is it dependent to ifp->if_type?
2073 */
2074 if (!nd6_try_set_entry_addr(ifp, ln, lladdr)) {
2075 /* Entry was deleted */
2076 LLE_WUNLOCK(ln);
2077 return;
2078 }
2079
2080 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE);
2081
2082 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2083
2084 if (ln->la_hold != NULL)
2085 chain = nd6_grab_holdchain(ln);
2086 }
2087
2088 /* Calculates new router status */
2089 router = nd6_is_router(type, code, is_newentry, olladdr,
2090 lladdr != NULL ? 1 : 0, ln->ln_router);
2091
2092 ln->ln_router = router;
2093 /* Mark non-router redirects with special flag */
2094 if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER)
2095 ln->la_flags |= LLE_REDIRECT;
2096
2097 if (flags & LLE_EXCLUSIVE)
2098 LLE_WUNLOCK(ln);
2099 else
2100 LLE_RUNLOCK(ln);
2101
2102 if (chain != NULL)
2103 nd6_flush_holdchain(ifp, ln, chain);
2104 if (do_update)
2105 nd6_flush_children_holdchain(ifp, ln);
2106
2107 /*
2108 * When the link-layer address of a router changes, select the
2109 * best router again. In particular, when the neighbor entry is newly
2110 * created, it might affect the selection policy.
2111 * Question: can we restrict the first condition to the "is_newentry"
2112 * case?
2113 * XXX: when we hear an RA from a new router with the link-layer
2114 * address option, defrouter_select_fib() is called twice, since
2115 * defrtrlist_update called the function as well. However, I believe
2116 * we can compromise the overhead, since it only happens the first
2117 * time.
2118 * XXX: although defrouter_select_fib() should not have a bad effect
2119 * for those are not autoconfigured hosts, we explicitly avoid such
2120 * cases for safety.
2121 */
2122 if ((do_update || is_newentry) && router &&
2123 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) {
2124 /*
2125 * guaranteed recursion
2126 */
2127 defrouter_select_fib(ifp->if_fib);
2128 }
2129 }
2130
2131 static void
nd6_slowtimo(void * arg)2132 nd6_slowtimo(void *arg)
2133 {
2134 struct epoch_tracker et;
2135 CURVNET_SET((struct vnet *) arg);
2136 struct nd_ifinfo *nd6if;
2137 struct ifnet *ifp;
2138
2139 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz,
2140 nd6_slowtimo, curvnet);
2141 NET_EPOCH_ENTER(et);
2142 CK_STAILQ_FOREACH(ifp, &V_ifnet, if_link) {
2143 if (ifp->if_inet6 == NULL)
2144 continue;
2145 nd6if = ND_IFINFO(ifp);
2146 if (nd6if->basereachable && /* already initialized */
2147 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) {
2148 /*
2149 * Since reachable time rarely changes by router
2150 * advertisements, we SHOULD insure that a new random
2151 * value gets recomputed at least once every few hours.
2152 * (RFC 2461, 6.3.4)
2153 */
2154 nd6if->recalctm = V_nd6_recalc_reachtm_interval;
2155 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable);
2156 }
2157 }
2158 NET_EPOCH_EXIT(et);
2159 CURVNET_RESTORE();
2160 }
2161
2162 struct mbuf *
nd6_grab_holdchain(struct llentry * ln)2163 nd6_grab_holdchain(struct llentry *ln)
2164 {
2165 struct mbuf *chain;
2166
2167 LLE_WLOCK_ASSERT(ln);
2168
2169 chain = ln->la_hold;
2170 ln->la_hold = NULL;
2171 ln->la_numheld = 0;
2172
2173 if (ln->ln_state == ND6_LLINFO_STALE) {
2174 /*
2175 * The first time we send a packet to a
2176 * neighbor whose entry is STALE, we have
2177 * to change the state to DELAY and a sets
2178 * a timer to expire in DELAY_FIRST_PROBE_TIME
2179 * seconds to ensure do neighbor unreachability
2180 * detection on expiration.
2181 * (RFC 2461 7.3.3)
2182 */
2183 nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY);
2184 }
2185
2186 return (chain);
2187 }
2188
2189 int
nd6_output_ifp(struct ifnet * ifp,struct ifnet * origifp,struct mbuf * m,struct sockaddr_in6 * dst,struct route * ro)2190 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m,
2191 struct sockaddr_in6 *dst, struct route *ro)
2192 {
2193 int error;
2194 int ip6len;
2195 struct ip6_hdr *ip6;
2196 struct m_tag *mtag;
2197
2198 #ifdef MAC
2199 mac_netinet6_nd6_send(ifp, m);
2200 #endif
2201
2202 /*
2203 * If called from nd6_ns_output() (NS), nd6_na_output() (NA),
2204 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA
2205 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND
2206 * to be diverted to user space. When re-injected into the kernel,
2207 * send_output() will directly dispatch them to the outgoing interface.
2208 */
2209 if (send_sendso_input_hook != NULL) {
2210 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL);
2211 if (mtag != NULL) {
2212 ip6 = mtod(m, struct ip6_hdr *);
2213 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen);
2214 /* Use the SEND socket */
2215 error = send_sendso_input_hook(m, ifp, SND_OUT,
2216 ip6len);
2217 /* -1 == no app on SEND socket */
2218 if (error == 0 || error != -1)
2219 return (error);
2220 }
2221 }
2222
2223 m_clrprotoflags(m); /* Avoid confusing lower layers. */
2224 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL,
2225 mtod(m, struct ip6_hdr *));
2226
2227 if ((ifp->if_flags & IFF_LOOPBACK) == 0)
2228 origifp = ifp;
2229
2230 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro);
2231 return (error);
2232 }
2233
2234 /*
2235 * Lookup link headerfor @sa_dst address. Stores found
2236 * data in @desten buffer. Copy of lle ln_flags can be also
2237 * saved in @pflags if @pflags is non-NULL.
2238 *
2239 * If destination LLE does not exists or lle state modification
2240 * is required, call "slow" version.
2241 *
2242 * Return values:
2243 * - 0 on success (address copied to buffer).
2244 * - EWOULDBLOCK (no local error, but address is still unresolved)
2245 * - other errors (alloc failure, etc)
2246 */
2247 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)2248 nd6_resolve(struct ifnet *ifp, int gw_flags, struct mbuf *m,
2249 const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags,
2250 struct llentry **plle)
2251 {
2252 struct llentry *ln = NULL;
2253 const struct sockaddr_in6 *dst6;
2254
2255 NET_EPOCH_ASSERT();
2256
2257 if (pflags != NULL)
2258 *pflags = 0;
2259
2260 dst6 = (const struct sockaddr_in6 *)sa_dst;
2261
2262 /* discard the packet if IPv6 operation is disabled on the interface */
2263 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) {
2264 m_freem(m);
2265 return (ENETDOWN); /* better error? */
2266 }
2267
2268 if (m != NULL && m->m_flags & M_MCAST) {
2269 switch (ifp->if_type) {
2270 case IFT_ETHER:
2271 case IFT_L2VLAN:
2272 case IFT_BRIDGE:
2273 ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr,
2274 desten);
2275 return (0);
2276 default:
2277 m_freem(m);
2278 return (EAFNOSUPPORT);
2279 }
2280 }
2281
2282 int family = gw_flags >> 16;
2283 int lookup_flags = plle ? LLE_EXCLUSIVE : LLE_UNLOCKED;
2284 ln = nd6_lookup(&dst6->sin6_addr, LLE_SF(family, lookup_flags), ifp);
2285 if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) {
2286 /* Entry found, let's copy lle info */
2287 bcopy(ln->r_linkdata, desten, ln->r_hdrlen);
2288 if (pflags != NULL)
2289 *pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR);
2290 llentry_provide_feedback(ln);
2291 if (plle) {
2292 LLE_ADDREF(ln);
2293 *plle = ln;
2294 LLE_WUNLOCK(ln);
2295 }
2296 return (0);
2297 } else if (plle && ln)
2298 LLE_WUNLOCK(ln);
2299
2300 return (nd6_resolve_slow(ifp, family, 0, m, dst6, desten, pflags, plle));
2301 }
2302
2303 /*
2304 * Finds or creates a new llentry for @addr and @family.
2305 * Returns wlocked llentry or NULL.
2306 *
2307 *
2308 * Child LLEs.
2309 *
2310 * Do not have their own state machine (gets marked as static)
2311 * settimer bails out for child LLEs just in case.
2312 *
2313 * Locking order: parent lle gets locked first, chen goes the child.
2314 */
2315 static __noinline struct llentry *
nd6_get_llentry(struct ifnet * ifp,const struct in6_addr * addr,int family)2316 nd6_get_llentry(struct ifnet *ifp, const struct in6_addr *addr, int family)
2317 {
2318 struct llentry *child_lle = NULL;
2319 struct llentry *lle, *lle_tmp;
2320
2321 lle = nd6_alloc(addr, 0, ifp);
2322 if (lle != NULL && family != AF_INET6) {
2323 child_lle = nd6_alloc(addr, 0, ifp);
2324 if (child_lle == NULL) {
2325 lltable_free_entry(LLTABLE6(ifp), lle);
2326 return (NULL);
2327 }
2328 child_lle->r_family = family;
2329 child_lle->la_flags |= LLE_CHILD | LLE_STATIC;
2330 child_lle->ln_state = ND6_LLINFO_INCOMPLETE;
2331 }
2332
2333 if (lle == NULL) {
2334 char ip6buf[INET6_ADDRSTRLEN];
2335 log(LOG_DEBUG,
2336 "nd6_get_llentry: can't allocate llinfo for %s "
2337 "(ln=%p)\n",
2338 ip6_sprintf(ip6buf, addr), lle);
2339 return (NULL);
2340 }
2341
2342 LLTABLE_LOCK(LLTABLE6(ifp));
2343 LLE_WLOCK(lle);
2344 /* Prefer any existing entry over newly-created one */
2345 lle_tmp = nd6_lookup(addr, LLE_SF(AF_INET6, LLE_EXCLUSIVE), ifp);
2346 if (lle_tmp == NULL)
2347 lltable_link_entry(LLTABLE6(ifp), lle);
2348 else {
2349 lltable_free_entry(LLTABLE6(ifp), lle);
2350 lle = lle_tmp;
2351 }
2352 if (child_lle != NULL) {
2353 /* Check if child lle for the same family exists */
2354 lle_tmp = llentry_lookup_family(lle, child_lle->r_family);
2355 LLE_WLOCK(child_lle);
2356 if (lle_tmp == NULL) {
2357 /* Attach */
2358 lltable_link_child_entry(lle, child_lle);
2359 } else {
2360 /* child lle already exists, free newly-created one */
2361 lltable_free_entry(LLTABLE6(ifp), child_lle);
2362 LLE_WLOCK(lle_tmp);
2363 child_lle = lle_tmp;
2364 }
2365 LLE_WUNLOCK(lle);
2366 lle = child_lle;
2367 }
2368 LLTABLE_UNLOCK(LLTABLE6(ifp));
2369 return (lle);
2370 }
2371
2372 /*
2373 * Do L2 address resolution for @sa_dst address. Stores found
2374 * address in @desten buffer. Copy of lle ln_flags can be also
2375 * saved in @pflags if @pflags is non-NULL.
2376 *
2377 * Heavy version.
2378 * Function assume that destination LLE does not exist,
2379 * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired.
2380 *
2381 * Set noinline to be dtrace-friendly
2382 */
2383 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)2384 nd6_resolve_slow(struct ifnet *ifp, int family, int flags, struct mbuf *m,
2385 const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags,
2386 struct llentry **plle)
2387 {
2388 struct llentry *lle = NULL;
2389 struct in6_addr *psrc, src;
2390 int send_ns, ll_len;
2391 char *lladdr;
2392
2393 NET_EPOCH_ASSERT();
2394
2395 /*
2396 * Address resolution or Neighbor Unreachability Detection
2397 * for the next hop.
2398 * At this point, the destination of the packet must be a unicast
2399 * or an anycast address(i.e. not a multicast).
2400 */
2401 lle = nd6_lookup(&dst->sin6_addr, LLE_SF(family, LLE_EXCLUSIVE), ifp);
2402 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) {
2403 /*
2404 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(),
2405 * the condition below is not very efficient. But we believe
2406 * it is tolerable, because this should be a rare case.
2407 */
2408 lle = nd6_get_llentry(ifp, &dst->sin6_addr, family);
2409 }
2410
2411 if (lle == NULL) {
2412 m_freem(m);
2413 return (ENOBUFS);
2414 }
2415
2416 LLE_WLOCK_ASSERT(lle);
2417
2418 /*
2419 * The first time we send a packet to a neighbor whose entry is
2420 * STALE, we have to change the state to DELAY and a sets a timer to
2421 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do
2422 * neighbor unreachability detection on expiration.
2423 * (RFC 2461 7.3.3)
2424 */
2425 if ((!(lle->la_flags & LLE_CHILD)) && (lle->ln_state == ND6_LLINFO_STALE))
2426 nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY);
2427
2428 /*
2429 * If the neighbor cache entry has a state other than INCOMPLETE
2430 * (i.e. its link-layer address is already resolved), just
2431 * send the packet.
2432 */
2433 if (lle->ln_state > ND6_LLINFO_INCOMPLETE) {
2434 if (flags & LLE_ADDRONLY) {
2435 lladdr = lle->ll_addr;
2436 ll_len = ifp->if_addrlen;
2437 } else {
2438 lladdr = lle->r_linkdata;
2439 ll_len = lle->r_hdrlen;
2440 }
2441 bcopy(lladdr, desten, ll_len);
2442 if (pflags != NULL)
2443 *pflags = lle->la_flags;
2444 if (plle) {
2445 LLE_ADDREF(lle);
2446 *plle = lle;
2447 }
2448 LLE_WUNLOCK(lle);
2449 return (0);
2450 }
2451
2452 /*
2453 * There is a neighbor cache entry, but no ethernet address
2454 * response yet. Append this latest packet to the end of the
2455 * packet queue in the mbuf. When it exceeds nd6_maxqueuelen,
2456 * the oldest packet in the queue will be removed.
2457 */
2458 if (m != NULL) {
2459 size_t dropped;
2460
2461 dropped = lltable_append_entry_queue(lle, m, V_nd6_maxqueuelen);
2462 ICMP6STAT_ADD(icp6s_dropped, dropped);
2463 }
2464
2465 /*
2466 * If there has been no NS for the neighbor after entering the
2467 * INCOMPLETE state, send the first solicitation.
2468 * Note that for newly-created lle la_asked will be 0,
2469 * so we will transition from ND6_LLINFO_NOSTATE to
2470 * ND6_LLINFO_INCOMPLETE state here.
2471 */
2472 psrc = NULL;
2473 send_ns = 0;
2474
2475 /* If we have child lle, switch to the parent to send NS */
2476 if (lle->la_flags & LLE_CHILD) {
2477 struct llentry *lle_parent = lle->lle_parent;
2478 LLE_WUNLOCK(lle);
2479 lle = lle_parent;
2480 LLE_WLOCK(lle);
2481 }
2482 if (lle->la_asked == 0) {
2483 lle->la_asked++;
2484 send_ns = 1;
2485 psrc = nd6_llinfo_get_holdsrc(lle, &src);
2486
2487 nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE);
2488 }
2489 LLE_WUNLOCK(lle);
2490 if (send_ns != 0)
2491 nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL);
2492
2493 return (EWOULDBLOCK);
2494 }
2495
2496 /*
2497 * Do L2 address resolution for @sa_dst address. Stores found
2498 * address in @desten buffer. Copy of lle ln_flags can be also
2499 * saved in @pflags if @pflags is non-NULL.
2500 *
2501 * Return values:
2502 * - 0 on success (address copied to buffer).
2503 * - EWOULDBLOCK (no local error, but address is still unresolved)
2504 * - other errors (alloc failure, etc)
2505 */
2506 int
nd6_resolve_addr(struct ifnet * ifp,int flags,const struct sockaddr * dst,char * desten,uint32_t * pflags)2507 nd6_resolve_addr(struct ifnet *ifp, int flags, const struct sockaddr *dst,
2508 char *desten, uint32_t *pflags)
2509 {
2510 int error;
2511
2512 flags |= LLE_ADDRONLY;
2513 error = nd6_resolve_slow(ifp, AF_INET6, flags, NULL,
2514 (const struct sockaddr_in6 *)dst, desten, pflags, NULL);
2515 return (error);
2516 }
2517
2518 int
nd6_flush_holdchain(struct ifnet * ifp,struct llentry * lle,struct mbuf * chain)2519 nd6_flush_holdchain(struct ifnet *ifp, struct llentry *lle, struct mbuf *chain)
2520 {
2521 struct mbuf *m, *m_head;
2522 struct sockaddr_in6 dst6;
2523 int error = 0;
2524
2525 NET_EPOCH_ASSERT();
2526
2527 struct route_in6 ro = {
2528 .ro_prepend = lle->r_linkdata,
2529 .ro_plen = lle->r_hdrlen,
2530 };
2531
2532 lltable_fill_sa_entry(lle, (struct sockaddr *)&dst6);
2533 m_head = chain;
2534
2535 while (m_head) {
2536 m = m_head;
2537 m_head = m_head->m_nextpkt;
2538 m->m_nextpkt = NULL;
2539 error = nd6_output_ifp(ifp, ifp, m, &dst6, (struct route *)&ro);
2540 }
2541
2542 /*
2543 * XXX
2544 * note that intermediate errors are blindly ignored
2545 */
2546 return (error);
2547 }
2548
2549 __noinline void
nd6_flush_children_holdchain(struct ifnet * ifp,struct llentry * lle)2550 nd6_flush_children_holdchain(struct ifnet *ifp, struct llentry *lle)
2551 {
2552 struct llentry *child_lle;
2553 struct mbuf *chain;
2554
2555 NET_EPOCH_ASSERT();
2556
2557 CK_SLIST_FOREACH(child_lle, &lle->lle_children, lle_child_next) {
2558 LLE_WLOCK(child_lle);
2559 chain = nd6_grab_holdchain(child_lle);
2560 LLE_WUNLOCK(child_lle);
2561 nd6_flush_holdchain(ifp, child_lle, chain);
2562 }
2563 }
2564
2565 static int
nd6_need_cache(struct ifnet * ifp)2566 nd6_need_cache(struct ifnet *ifp)
2567 {
2568 /*
2569 * XXX: we currently do not make neighbor cache on any interface
2570 * other than Ethernet and GIF.
2571 *
2572 * RFC2893 says:
2573 * - unidirectional tunnels needs no ND
2574 */
2575 switch (ifp->if_type) {
2576 case IFT_ETHER:
2577 case IFT_IEEE1394:
2578 case IFT_L2VLAN:
2579 case IFT_INFINIBAND:
2580 case IFT_BRIDGE:
2581 case IFT_PROPVIRTUAL:
2582 return (1);
2583 default:
2584 return (0);
2585 }
2586 }
2587
2588 /*
2589 * Add pernament ND6 link-layer record for given
2590 * interface address.
2591 *
2592 * Very similar to IPv4 arp_ifinit(), but:
2593 * 1) IPv6 DAD is performed in different place
2594 * 2) It is called by IPv6 protocol stack in contrast to
2595 * arp_ifinit() which is typically called in SIOCSIFADDR
2596 * driver ioctl handler.
2597 *
2598 */
2599 int
nd6_add_ifa_lle(struct in6_ifaddr * ia)2600 nd6_add_ifa_lle(struct in6_ifaddr *ia)
2601 {
2602 struct ifnet *ifp;
2603 struct llentry *ln, *ln_tmp;
2604 struct sockaddr *dst;
2605
2606 ifp = ia->ia_ifa.ifa_ifp;
2607 if (nd6_need_cache(ifp) == 0)
2608 return (0);
2609
2610 dst = (struct sockaddr *)&ia->ia_addr;
2611 ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst);
2612 if (ln == NULL)
2613 return (ENOBUFS);
2614
2615 LLTABLE_LOCK(LLTABLE6(ifp));
2616 LLE_WLOCK(ln);
2617 /* Unlink any entry if exists */
2618 ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_SF(AF_INET6, LLE_EXCLUSIVE), dst);
2619 if (ln_tmp != NULL)
2620 lltable_unlink_entry(LLTABLE6(ifp), ln_tmp);
2621 lltable_link_entry(LLTABLE6(ifp), ln);
2622 LLTABLE_UNLOCK(LLTABLE6(ifp));
2623
2624 if (ln_tmp != NULL)
2625 EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED);
2626 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED);
2627
2628 LLE_WUNLOCK(ln);
2629 if (ln_tmp != NULL)
2630 llentry_free(ln_tmp);
2631
2632 return (0);
2633 }
2634
2635 /*
2636 * Removes either all lle entries for given @ia, or lle
2637 * corresponding to @ia address.
2638 */
2639 void
nd6_rem_ifa_lle(struct in6_ifaddr * ia,int all)2640 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all)
2641 {
2642 struct sockaddr_in6 mask, addr;
2643 struct sockaddr *saddr, *smask;
2644 struct ifnet *ifp;
2645
2646 ifp = ia->ia_ifa.ifa_ifp;
2647 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr));
2648 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask));
2649 saddr = (struct sockaddr *)&addr;
2650 smask = (struct sockaddr *)&mask;
2651
2652 if (all != 0)
2653 lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC);
2654 else
2655 lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr);
2656 }
2657
2658 static int
nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)2659 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS)
2660 {
2661 struct in6_prefix p;
2662 struct sockaddr_in6 s6;
2663 struct nd_prefix *pr;
2664 struct nd_pfxrouter *pfr;
2665 time_t maxexpire;
2666 int error;
2667 char ip6buf[INET6_ADDRSTRLEN];
2668
2669 if (req->newptr)
2670 return (EPERM);
2671
2672 error = sysctl_wire_old_buffer(req, 0);
2673 if (error != 0)
2674 return (error);
2675
2676 bzero(&p, sizeof(p));
2677 p.origin = PR_ORIG_RA;
2678 bzero(&s6, sizeof(s6));
2679 s6.sin6_family = AF_INET6;
2680 s6.sin6_len = sizeof(s6);
2681
2682 ND6_RLOCK();
2683 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) {
2684 p.prefix = pr->ndpr_prefix;
2685 if (sa6_recoverscope(&p.prefix)) {
2686 log(LOG_ERR, "scope error in prefix list (%s)\n",
2687 ip6_sprintf(ip6buf, &p.prefix.sin6_addr));
2688 /* XXX: press on... */
2689 }
2690 p.raflags = pr->ndpr_raf;
2691 p.prefixlen = pr->ndpr_plen;
2692 p.vltime = pr->ndpr_vltime;
2693 p.pltime = pr->ndpr_pltime;
2694 p.if_index = pr->ndpr_ifp->if_index;
2695 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME)
2696 p.expire = 0;
2697 else {
2698 /* XXX: we assume time_t is signed. */
2699 maxexpire = (-1) &
2700 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1));
2701 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate)
2702 p.expire = pr->ndpr_lastupdate +
2703 pr->ndpr_vltime +
2704 (time_second - time_uptime);
2705 else
2706 p.expire = maxexpire;
2707 }
2708 p.refcnt = pr->ndpr_addrcnt;
2709 p.flags = pr->ndpr_stateflags;
2710 p.advrtrs = 0;
2711 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry)
2712 p.advrtrs++;
2713 error = SYSCTL_OUT(req, &p, sizeof(p));
2714 if (error != 0)
2715 break;
2716 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) {
2717 s6.sin6_addr = pfr->router->rtaddr;
2718 if (sa6_recoverscope(&s6))
2719 log(LOG_ERR,
2720 "scope error in prefix list (%s)\n",
2721 ip6_sprintf(ip6buf, &pfr->router->rtaddr));
2722 error = SYSCTL_OUT(req, &s6, sizeof(s6));
2723 if (error != 0)
2724 goto out;
2725 }
2726 }
2727 out:
2728 ND6_RUNLOCK();
2729 return (error);
2730 }
2731 SYSCTL_PROC(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist,
2732 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2733 NULL, 0, nd6_sysctl_prlist, "S,in6_prefix",
2734 "NDP prefix list");
2735 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen,
2736 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, "");
2737 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer,
2738 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), "");
2739