nd6.c revision 278801
1/*- 2 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the project nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $ 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: stable/10/sys/netinet6/nd6.c 278801 2015-02-15 13:57:44Z rrs $"); 34 35#include "opt_inet.h" 36#include "opt_inet6.h" 37#include "opt_kdtrace.h" 38 39#include <sys/param.h> 40#include <sys/systm.h> 41#include <sys/callout.h> 42#include <sys/malloc.h> 43#include <sys/mbuf.h> 44#include <sys/socket.h> 45#include <sys/sockio.h> 46#include <sys/time.h> 47#include <sys/kernel.h> 48#include <sys/protosw.h> 49#include <sys/errno.h> 50#include <sys/syslog.h> 51#include <sys/lock.h> 52#include <sys/rwlock.h> 53#include <sys/queue.h> 54#include <sys/sdt.h> 55#include <sys/sysctl.h> 56 57#include <net/if.h> 58#include <net/if_arc.h> 59#include <net/if_dl.h> 60#include <net/if_types.h> 61#include <net/iso88025.h> 62#include <net/fddi.h> 63#include <net/route.h> 64#include <net/vnet.h> 65 66#include <netinet/in.h> 67#include <netinet/in_kdtrace.h> 68#include <net/if_llatbl.h> 69#define L3_ADDR_SIN6(le) ((struct sockaddr_in6 *) L3_ADDR(le)) 70#include <netinet/if_ether.h> 71#include <netinet6/in6_var.h> 72#include <netinet/ip6.h> 73#include <netinet6/ip6_var.h> 74#include <netinet6/scope6_var.h> 75#include <netinet6/nd6.h> 76#include <netinet6/in6_ifattach.h> 77#include <netinet/icmp6.h> 78#include <netinet6/send.h> 79 80#include <sys/limits.h> 81 82#include <security/mac/mac_framework.h> 83 84#define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ 85#define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ 86 87#define SIN6(s) ((const struct sockaddr_in6 *)(s)) 88 89/* timer values */ 90VNET_DEFINE(int, nd6_prune) = 1; /* walk list every 1 seconds */ 91VNET_DEFINE(int, nd6_delay) = 5; /* delay first probe time 5 second */ 92VNET_DEFINE(int, nd6_umaxtries) = 3; /* maximum unicast query */ 93VNET_DEFINE(int, nd6_mmaxtries) = 3; /* maximum multicast query */ 94VNET_DEFINE(int, nd6_useloopback) = 1; /* use loopback interface for 95 * local traffic */ 96VNET_DEFINE(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage 97 * collection timer */ 98 99/* preventing too many loops in ND option parsing */ 100static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */ 101 102VNET_DEFINE(int, nd6_maxnudhint) = 0; /* max # of subsequent upper 103 * layer hints */ 104static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved 105 * ND entries */ 106#define V_nd6_maxndopt VNET(nd6_maxndopt) 107#define V_nd6_maxqueuelen VNET(nd6_maxqueuelen) 108 109#ifdef ND6_DEBUG 110VNET_DEFINE(int, nd6_debug) = 1; 111#else 112VNET_DEFINE(int, nd6_debug) = 0; 113#endif 114 115/* for debugging? */ 116#if 0 117static int nd6_inuse, nd6_allocated; 118#endif 119 120VNET_DEFINE(struct nd_drhead, nd_defrouter); 121VNET_DEFINE(struct nd_prhead, nd_prefix); 122 123VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL; 124#define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval) 125 126int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int); 127 128static int nd6_is_new_addr_neighbor(struct sockaddr_in6 *, 129 struct ifnet *); 130static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *); 131static void nd6_slowtimo(void *); 132static int regen_tmpaddr(struct in6_ifaddr *); 133static struct llentry *nd6_free(struct llentry *, int); 134static void nd6_llinfo_timer(void *); 135static void clear_llinfo_pqueue(struct llentry *); 136 137static VNET_DEFINE(struct callout, nd6_slowtimo_ch); 138#define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch) 139 140VNET_DEFINE(struct callout, nd6_timer_ch); 141 142void 143nd6_init(void) 144{ 145 146 LIST_INIT(&V_nd_prefix); 147 148 /* initialization of the default router list */ 149 TAILQ_INIT(&V_nd_defrouter); 150 151 /* start timer */ 152 callout_init(&V_nd6_slowtimo_ch, 0); 153 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 154 nd6_slowtimo, curvnet); 155} 156 157#ifdef VIMAGE 158void 159nd6_destroy() 160{ 161 162 callout_drain(&V_nd6_slowtimo_ch); 163 callout_drain(&V_nd6_timer_ch); 164} 165#endif 166 167struct nd_ifinfo * 168nd6_ifattach(struct ifnet *ifp) 169{ 170 struct nd_ifinfo *nd; 171 172 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO); 173 nd->initialized = 1; 174 175 nd->chlim = IPV6_DEFHLIM; 176 nd->basereachable = REACHABLE_TIME; 177 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable); 178 nd->retrans = RETRANS_TIMER; 179 180 nd->flags = ND6_IFF_PERFORMNUD; 181 182 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. 183 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by 184 * default regardless of the V_ip6_auto_linklocal configuration to 185 * give a reasonable default behavior. 186 */ 187 if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) || 188 (ifp->if_flags & IFF_LOOPBACK)) 189 nd->flags |= ND6_IFF_AUTO_LINKLOCAL; 190 /* 191 * A loopback interface does not need to accept RTADV. 192 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by 193 * default regardless of the V_ip6_accept_rtadv configuration to 194 * prevent the interface from accepting RA messages arrived 195 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV. 196 */ 197 if (V_ip6_accept_rtadv && 198 !(ifp->if_flags & IFF_LOOPBACK) && 199 (ifp->if_type != IFT_BRIDGE)) 200 nd->flags |= ND6_IFF_ACCEPT_RTADV; 201 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK)) 202 nd->flags |= ND6_IFF_NO_RADR; 203 204 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */ 205 nd6_setmtu0(ifp, nd); 206 207 return nd; 208} 209 210void 211nd6_ifdetach(struct nd_ifinfo *nd) 212{ 213 214 free(nd, M_IP6NDP); 215} 216 217/* 218 * Reset ND level link MTU. This function is called when the physical MTU 219 * changes, which means we might have to adjust the ND level MTU. 220 */ 221void 222nd6_setmtu(struct ifnet *ifp) 223{ 224 225 nd6_setmtu0(ifp, ND_IFINFO(ifp)); 226} 227 228/* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */ 229void 230nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi) 231{ 232 u_int32_t omaxmtu; 233 234 omaxmtu = ndi->maxmtu; 235 236 switch (ifp->if_type) { 237 case IFT_ARCNET: 238 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */ 239 break; 240 case IFT_FDDI: 241 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */ 242 break; 243 case IFT_ISO88025: 244 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu); 245 break; 246 default: 247 ndi->maxmtu = ifp->if_mtu; 248 break; 249 } 250 251 /* 252 * Decreasing the interface MTU under IPV6 minimum MTU may cause 253 * undesirable situation. We thus notify the operator of the change 254 * explicitly. The check for omaxmtu is necessary to restrict the 255 * log to the case of changing the MTU, not initializing it. 256 */ 257 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { 258 log(LOG_NOTICE, "nd6_setmtu0: " 259 "new link MTU on %s (%lu) is too small for IPv6\n", 260 if_name(ifp), (unsigned long)ndi->maxmtu); 261 } 262 263 if (ndi->maxmtu > V_in6_maxmtu) 264 in6_setmaxmtu(); /* check all interfaces just in case */ 265 266} 267 268void 269nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts) 270{ 271 272 bzero(ndopts, sizeof(*ndopts)); 273 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; 274 ndopts->nd_opts_last 275 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); 276 277 if (icmp6len == 0) { 278 ndopts->nd_opts_done = 1; 279 ndopts->nd_opts_search = NULL; 280 } 281} 282 283/* 284 * Take one ND option. 285 */ 286struct nd_opt_hdr * 287nd6_option(union nd_opts *ndopts) 288{ 289 struct nd_opt_hdr *nd_opt; 290 int olen; 291 292 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 293 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 294 __func__)); 295 if (ndopts->nd_opts_search == NULL) 296 return NULL; 297 if (ndopts->nd_opts_done) 298 return NULL; 299 300 nd_opt = ndopts->nd_opts_search; 301 302 /* make sure nd_opt_len is inside the buffer */ 303 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) { 304 bzero(ndopts, sizeof(*ndopts)); 305 return NULL; 306 } 307 308 olen = nd_opt->nd_opt_len << 3; 309 if (olen == 0) { 310 /* 311 * Message validation requires that all included 312 * options have a length that is greater than zero. 313 */ 314 bzero(ndopts, sizeof(*ndopts)); 315 return NULL; 316 } 317 318 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen); 319 if (ndopts->nd_opts_search > ndopts->nd_opts_last) { 320 /* option overruns the end of buffer, invalid */ 321 bzero(ndopts, sizeof(*ndopts)); 322 return NULL; 323 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { 324 /* reached the end of options chain */ 325 ndopts->nd_opts_done = 1; 326 ndopts->nd_opts_search = NULL; 327 } 328 return nd_opt; 329} 330 331/* 332 * Parse multiple ND options. 333 * This function is much easier to use, for ND routines that do not need 334 * multiple options of the same type. 335 */ 336int 337nd6_options(union nd_opts *ndopts) 338{ 339 struct nd_opt_hdr *nd_opt; 340 int i = 0; 341 342 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 343 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 344 __func__)); 345 if (ndopts->nd_opts_search == NULL) 346 return 0; 347 348 while (1) { 349 nd_opt = nd6_option(ndopts); 350 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { 351 /* 352 * Message validation requires that all included 353 * options have a length that is greater than zero. 354 */ 355 ICMP6STAT_INC(icp6s_nd_badopt); 356 bzero(ndopts, sizeof(*ndopts)); 357 return -1; 358 } 359 360 if (nd_opt == NULL) 361 goto skip1; 362 363 switch (nd_opt->nd_opt_type) { 364 case ND_OPT_SOURCE_LINKADDR: 365 case ND_OPT_TARGET_LINKADDR: 366 case ND_OPT_MTU: 367 case ND_OPT_REDIRECTED_HEADER: 368 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { 369 nd6log((LOG_INFO, 370 "duplicated ND6 option found (type=%d)\n", 371 nd_opt->nd_opt_type)); 372 /* XXX bark? */ 373 } else { 374 ndopts->nd_opt_array[nd_opt->nd_opt_type] 375 = nd_opt; 376 } 377 break; 378 case ND_OPT_PREFIX_INFORMATION: 379 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { 380 ndopts->nd_opt_array[nd_opt->nd_opt_type] 381 = nd_opt; 382 } 383 ndopts->nd_opts_pi_end = 384 (struct nd_opt_prefix_info *)nd_opt; 385 break; 386 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */ 387 case ND_OPT_RDNSS: /* RFC 6106 */ 388 case ND_OPT_DNSSL: /* RFC 6106 */ 389 /* 390 * Silently ignore options we know and do not care about 391 * in the kernel. 392 */ 393 break; 394 default: 395 /* 396 * Unknown options must be silently ignored, 397 * to accomodate future extension to the protocol. 398 */ 399 nd6log((LOG_DEBUG, 400 "nd6_options: unsupported option %d - " 401 "option ignored\n", nd_opt->nd_opt_type)); 402 } 403 404skip1: 405 i++; 406 if (i > V_nd6_maxndopt) { 407 ICMP6STAT_INC(icp6s_nd_toomanyopt); 408 nd6log((LOG_INFO, "too many loop in nd opt\n")); 409 break; 410 } 411 412 if (ndopts->nd_opts_done) 413 break; 414 } 415 416 return 0; 417} 418 419/* 420 * ND6 timer routine to handle ND6 entries 421 */ 422void 423nd6_llinfo_settimer_locked(struct llentry *ln, long tick) 424{ 425 int canceled; 426 427 LLE_WLOCK_ASSERT(ln); 428 429 if (tick < 0) { 430 ln->la_expire = 0; 431 ln->ln_ntick = 0; 432 canceled = callout_stop(&ln->ln_timer_ch); 433 } else { 434 ln->la_expire = time_uptime + tick / hz; 435 LLE_ADDREF(ln); 436 if (tick > INT_MAX) { 437 ln->ln_ntick = tick - INT_MAX; 438 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX, 439 nd6_llinfo_timer, ln); 440 } else { 441 ln->ln_ntick = 0; 442 canceled = callout_reset(&ln->ln_timer_ch, tick, 443 nd6_llinfo_timer, ln); 444 } 445 } 446 if (canceled) 447 LLE_REMREF(ln); 448} 449 450void 451nd6_llinfo_settimer(struct llentry *ln, long tick) 452{ 453 454 LLE_WLOCK(ln); 455 nd6_llinfo_settimer_locked(ln, tick); 456 LLE_WUNLOCK(ln); 457} 458 459static void 460nd6_llinfo_timer(void *arg) 461{ 462 struct llentry *ln; 463 struct in6_addr *dst; 464 struct ifnet *ifp; 465 struct nd_ifinfo *ndi = NULL; 466 467 KASSERT(arg != NULL, ("%s: arg NULL", __func__)); 468 ln = (struct llentry *)arg; 469 LLE_WLOCK(ln); 470 if (callout_pending(&ln->la_timer)) { 471 /* 472 * Here we are a bit odd here in the treatment of 473 * active/pending. If the pending bit is set, it got 474 * rescheduled before I ran. The active 475 * bit we ignore, since if it was stopped 476 * in ll_tablefree() and was currently running 477 * it would have return 0 so the code would 478 * not have deleted it since the callout could 479 * not be stopped so we want to go through 480 * with the delete here now. If the callout 481 * was restarted, the pending bit will be back on and 482 * we just want to bail since the callout_reset would 483 * return 1 and our reference would have been removed 484 * by nd6_llinfo_settimer_locked above since canceled 485 * would have been 1. 486 */ 487 LLE_WUNLOCK(ln); 488 return; 489 } 490 ifp = ln->lle_tbl->llt_ifp; 491 CURVNET_SET(ifp->if_vnet); 492 493 if (ln->ln_ntick > 0) { 494 if (ln->ln_ntick > INT_MAX) { 495 ln->ln_ntick -= INT_MAX; 496 nd6_llinfo_settimer_locked(ln, INT_MAX); 497 } else { 498 ln->ln_ntick = 0; 499 nd6_llinfo_settimer_locked(ln, ln->ln_ntick); 500 } 501 goto done; 502 } 503 504 ndi = ND_IFINFO(ifp); 505 dst = &L3_ADDR_SIN6(ln)->sin6_addr; 506 if (ln->la_flags & LLE_STATIC) { 507 goto done; 508 } 509 510 if (ln->la_flags & LLE_DELETED) { 511 (void)nd6_free(ln, 0); 512 ln = NULL; 513 goto done; 514 } 515 516 switch (ln->ln_state) { 517 case ND6_LLINFO_INCOMPLETE: 518 if (ln->la_asked < V_nd6_mmaxtries) { 519 ln->la_asked++; 520 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 521 LLE_WUNLOCK(ln); 522 nd6_ns_output(ifp, NULL, dst, ln, 0); 523 LLE_WLOCK(ln); 524 } else { 525 struct mbuf *m = ln->la_hold; 526 if (m) { 527 struct mbuf *m0; 528 529 /* 530 * assuming every packet in la_hold has the 531 * same IP header. Send error after unlock. 532 */ 533 m0 = m->m_nextpkt; 534 m->m_nextpkt = NULL; 535 ln->la_hold = m0; 536 clear_llinfo_pqueue(ln); 537 } 538 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT); 539 (void)nd6_free(ln, 0); 540 ln = NULL; 541 if (m != NULL) 542 icmp6_error2(m, ICMP6_DST_UNREACH, 543 ICMP6_DST_UNREACH_ADDR, 0, ifp); 544 } 545 break; 546 case ND6_LLINFO_REACHABLE: 547 if (!ND6_LLINFO_PERMANENT(ln)) { 548 ln->ln_state = ND6_LLINFO_STALE; 549 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 550 } 551 break; 552 553 case ND6_LLINFO_STALE: 554 /* Garbage Collection(RFC 2461 5.3) */ 555 if (!ND6_LLINFO_PERMANENT(ln)) { 556 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 557 (void)nd6_free(ln, 1); 558 ln = NULL; 559 } 560 break; 561 562 case ND6_LLINFO_DELAY: 563 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 564 /* We need NUD */ 565 ln->la_asked = 1; 566 ln->ln_state = ND6_LLINFO_PROBE; 567 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 568 LLE_WUNLOCK(ln); 569 nd6_ns_output(ifp, dst, dst, ln, 0); 570 LLE_WLOCK(ln); 571 } else { 572 ln->ln_state = ND6_LLINFO_STALE; /* XXX */ 573 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 574 } 575 break; 576 case ND6_LLINFO_PROBE: 577 if (ln->la_asked < V_nd6_umaxtries) { 578 ln->la_asked++; 579 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 580 LLE_WUNLOCK(ln); 581 nd6_ns_output(ifp, dst, dst, ln, 0); 582 LLE_WLOCK(ln); 583 } else { 584 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 585 (void)nd6_free(ln, 0); 586 ln = NULL; 587 } 588 break; 589 default: 590 panic("%s: paths in a dark night can be confusing: %d", 591 __func__, ln->ln_state); 592 } 593done: 594 if (ln != NULL) 595 LLE_FREE_LOCKED(ln); 596 CURVNET_RESTORE(); 597} 598 599 600/* 601 * ND6 timer routine to expire default route list and prefix list 602 */ 603void 604nd6_timer(void *arg) 605{ 606 CURVNET_SET((struct vnet *) arg); 607 struct nd_defrouter *dr, *ndr; 608 struct nd_prefix *pr, *npr; 609 struct in6_ifaddr *ia6, *nia6; 610 611 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz, 612 nd6_timer, curvnet); 613 614 /* expire default router list */ 615 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 616 if (dr->expire && dr->expire < time_uptime) 617 defrtrlist_del(dr); 618 } 619 620 /* 621 * expire interface addresses. 622 * in the past the loop was inside prefix expiry processing. 623 * However, from a stricter speci-confrmance standpoint, we should 624 * rather separate address lifetimes and prefix lifetimes. 625 * 626 * XXXRW: in6_ifaddrhead locking. 627 */ 628 addrloop: 629 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) { 630 /* check address lifetime */ 631 if (IFA6_IS_INVALID(ia6)) { 632 int regen = 0; 633 634 /* 635 * If the expiring address is temporary, try 636 * regenerating a new one. This would be useful when 637 * we suspended a laptop PC, then turned it on after a 638 * period that could invalidate all temporary 639 * addresses. Although we may have to restart the 640 * loop (see below), it must be after purging the 641 * address. Otherwise, we'd see an infinite loop of 642 * regeneration. 643 */ 644 if (V_ip6_use_tempaddr && 645 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 646 if (regen_tmpaddr(ia6) == 0) 647 regen = 1; 648 } 649 650 in6_purgeaddr(&ia6->ia_ifa); 651 652 if (regen) 653 goto addrloop; /* XXX: see below */ 654 } else if (IFA6_IS_DEPRECATED(ia6)) { 655 int oldflags = ia6->ia6_flags; 656 657 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 658 659 /* 660 * If a temporary address has just become deprecated, 661 * regenerate a new one if possible. 662 */ 663 if (V_ip6_use_tempaddr && 664 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 665 (oldflags & IN6_IFF_DEPRECATED) == 0) { 666 667 if (regen_tmpaddr(ia6) == 0) { 668 /* 669 * A new temporary address is 670 * generated. 671 * XXX: this means the address chain 672 * has changed while we are still in 673 * the loop. Although the change 674 * would not cause disaster (because 675 * it's not a deletion, but an 676 * addition,) we'd rather restart the 677 * loop just for safety. Or does this 678 * significantly reduce performance?? 679 */ 680 goto addrloop; 681 } 682 } 683 } else { 684 /* 685 * A new RA might have made a deprecated address 686 * preferred. 687 */ 688 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 689 } 690 } 691 692 /* expire prefix list */ 693 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 694 /* 695 * check prefix lifetime. 696 * since pltime is just for autoconf, pltime processing for 697 * prefix is not necessary. 698 */ 699 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 700 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) { 701 702 /* 703 * address expiration and prefix expiration are 704 * separate. NEVER perform in6_purgeaddr here. 705 */ 706 prelist_remove(pr); 707 } 708 } 709 CURVNET_RESTORE(); 710} 711 712/* 713 * ia6 - deprecated/invalidated temporary address 714 */ 715static int 716regen_tmpaddr(struct in6_ifaddr *ia6) 717{ 718 struct ifaddr *ifa; 719 struct ifnet *ifp; 720 struct in6_ifaddr *public_ifa6 = NULL; 721 722 ifp = ia6->ia_ifa.ifa_ifp; 723 IF_ADDR_RLOCK(ifp); 724 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 725 struct in6_ifaddr *it6; 726 727 if (ifa->ifa_addr->sa_family != AF_INET6) 728 continue; 729 730 it6 = (struct in6_ifaddr *)ifa; 731 732 /* ignore no autoconf addresses. */ 733 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 734 continue; 735 736 /* ignore autoconf addresses with different prefixes. */ 737 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 738 continue; 739 740 /* 741 * Now we are looking at an autoconf address with the same 742 * prefix as ours. If the address is temporary and is still 743 * preferred, do not create another one. It would be rare, but 744 * could happen, for example, when we resume a laptop PC after 745 * a long period. 746 */ 747 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 748 !IFA6_IS_DEPRECATED(it6)) { 749 public_ifa6 = NULL; 750 break; 751 } 752 753 /* 754 * This is a public autoconf address that has the same prefix 755 * as ours. If it is preferred, keep it. We can't break the 756 * loop here, because there may be a still-preferred temporary 757 * address with the prefix. 758 */ 759 if (!IFA6_IS_DEPRECATED(it6)) 760 public_ifa6 = it6; 761 762 if (public_ifa6 != NULL) 763 ifa_ref(&public_ifa6->ia_ifa); 764 } 765 IF_ADDR_RUNLOCK(ifp); 766 767 if (public_ifa6 != NULL) { 768 int e; 769 770 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 771 ifa_free(&public_ifa6->ia_ifa); 772 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 773 " tmp addr,errno=%d\n", e); 774 return (-1); 775 } 776 ifa_free(&public_ifa6->ia_ifa); 777 return (0); 778 } 779 780 return (-1); 781} 782 783/* 784 * Nuke neighbor cache/prefix/default router management table, right before 785 * ifp goes away. 786 */ 787void 788nd6_purge(struct ifnet *ifp) 789{ 790 struct nd_defrouter *dr, *ndr; 791 struct nd_prefix *pr, *npr; 792 793 /* 794 * Nuke default router list entries toward ifp. 795 * We defer removal of default router list entries that is installed 796 * in the routing table, in order to keep additional side effects as 797 * small as possible. 798 */ 799 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 800 if (dr->installed) 801 continue; 802 803 if (dr->ifp == ifp) 804 defrtrlist_del(dr); 805 } 806 807 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 808 if (!dr->installed) 809 continue; 810 811 if (dr->ifp == ifp) 812 defrtrlist_del(dr); 813 } 814 815 /* Nuke prefix list entries toward ifp */ 816 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 817 if (pr->ndpr_ifp == ifp) { 818 /* 819 * Because if_detach() does *not* release prefixes 820 * while purging addresses the reference count will 821 * still be above zero. We therefore reset it to 822 * make sure that the prefix really gets purged. 823 */ 824 pr->ndpr_refcnt = 0; 825 826 /* 827 * Previously, pr->ndpr_addr is removed as well, 828 * but I strongly believe we don't have to do it. 829 * nd6_purge() is only called from in6_ifdetach(), 830 * which removes all the associated interface addresses 831 * by itself. 832 * (jinmei@kame.net 20010129) 833 */ 834 prelist_remove(pr); 835 } 836 } 837 838 /* cancel default outgoing interface setting */ 839 if (V_nd6_defifindex == ifp->if_index) 840 nd6_setdefaultiface(0); 841 842 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 843 /* Refresh default router list. */ 844 defrouter_select(); 845 } 846 847 /* XXXXX 848 * We do not nuke the neighbor cache entries here any more 849 * because the neighbor cache is kept in if_afdata[AF_INET6]. 850 * nd6_purge() is invoked by in6_ifdetach() which is called 851 * from if_detach() where everything gets purged. So let 852 * in6_domifdetach() do the actual L2 table purging work. 853 */ 854} 855 856/* 857 * the caller acquires and releases the lock on the lltbls 858 * Returns the llentry locked 859 */ 860struct llentry * 861nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp) 862{ 863 struct sockaddr_in6 sin6; 864 struct llentry *ln; 865 int llflags; 866 867 bzero(&sin6, sizeof(sin6)); 868 sin6.sin6_len = sizeof(struct sockaddr_in6); 869 sin6.sin6_family = AF_INET6; 870 sin6.sin6_addr = *addr6; 871 872 IF_AFDATA_LOCK_ASSERT(ifp); 873 874 llflags = 0; 875 if (flags & ND6_CREATE) 876 llflags |= LLE_CREATE; 877 if (flags & ND6_EXCLUSIVE) 878 llflags |= LLE_EXCLUSIVE; 879 880 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6); 881 if ((ln != NULL) && (llflags & LLE_CREATE)) 882 ln->ln_state = ND6_LLINFO_NOSTATE; 883 884 return (ln); 885} 886 887/* 888 * Test whether a given IPv6 address is a neighbor or not, ignoring 889 * the actual neighbor cache. The neighbor cache is ignored in order 890 * to not reenter the routing code from within itself. 891 */ 892static int 893nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 894{ 895 struct nd_prefix *pr; 896 struct ifaddr *dstaddr; 897 898 /* 899 * A link-local address is always a neighbor. 900 * XXX: a link does not necessarily specify a single interface. 901 */ 902 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 903 struct sockaddr_in6 sin6_copy; 904 u_int32_t zone; 905 906 /* 907 * We need sin6_copy since sa6_recoverscope() may modify the 908 * content (XXX). 909 */ 910 sin6_copy = *addr; 911 if (sa6_recoverscope(&sin6_copy)) 912 return (0); /* XXX: should be impossible */ 913 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 914 return (0); 915 if (sin6_copy.sin6_scope_id == zone) 916 return (1); 917 else 918 return (0); 919 } 920 921 /* 922 * If the address matches one of our addresses, 923 * it should be a neighbor. 924 * If the address matches one of our on-link prefixes, it should be a 925 * neighbor. 926 */ 927 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 928 if (pr->ndpr_ifp != ifp) 929 continue; 930 931 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) { 932 struct rtentry *rt; 933 934 /* Always use the default FIB here. */ 935 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 936 0, 0, RT_DEFAULT_FIB); 937 if (rt == NULL) 938 continue; 939 /* 940 * This is the case where multiple interfaces 941 * have the same prefix, but only one is installed 942 * into the routing table and that prefix entry 943 * is not the one being examined here. In the case 944 * where RADIX_MPATH is enabled, multiple route 945 * entries (of the same rt_key value) will be 946 * installed because the interface addresses all 947 * differ. 948 */ 949 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 950 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) { 951 RTFREE_LOCKED(rt); 952 continue; 953 } 954 RTFREE_LOCKED(rt); 955 } 956 957 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 958 &addr->sin6_addr, &pr->ndpr_mask)) 959 return (1); 960 } 961 962 /* 963 * If the address is assigned on the node of the other side of 964 * a p2p interface, the address should be a neighbor. 965 */ 966 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr); 967 if (dstaddr != NULL) { 968 if (dstaddr->ifa_ifp == ifp) { 969 ifa_free(dstaddr); 970 return (1); 971 } 972 ifa_free(dstaddr); 973 } 974 975 /* 976 * If the default router list is empty, all addresses are regarded 977 * as on-link, and thus, as a neighbor. 978 */ 979 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && 980 TAILQ_EMPTY(&V_nd_defrouter) && 981 V_nd6_defifindex == ifp->if_index) { 982 return (1); 983 } 984 985 return (0); 986} 987 988 989/* 990 * Detect if a given IPv6 address identifies a neighbor on a given link. 991 * XXX: should take care of the destination of a p2p link? 992 */ 993int 994nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 995{ 996 struct llentry *lle; 997 int rc = 0; 998 999 IF_AFDATA_UNLOCK_ASSERT(ifp); 1000 if (nd6_is_new_addr_neighbor(addr, ifp)) 1001 return (1); 1002 1003 /* 1004 * Even if the address matches none of our addresses, it might be 1005 * in the neighbor cache. 1006 */ 1007 IF_AFDATA_RLOCK(ifp); 1008 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) { 1009 LLE_RUNLOCK(lle); 1010 rc = 1; 1011 } 1012 IF_AFDATA_RUNLOCK(ifp); 1013 return (rc); 1014} 1015 1016/* 1017 * Free an nd6 llinfo entry. 1018 * Since the function would cause significant changes in the kernel, DO NOT 1019 * make it global, unless you have a strong reason for the change, and are sure 1020 * that the change is safe. 1021 */ 1022static struct llentry * 1023nd6_free(struct llentry *ln, int gc) 1024{ 1025 struct llentry *next; 1026 struct nd_defrouter *dr; 1027 struct ifnet *ifp; 1028 1029 LLE_WLOCK_ASSERT(ln); 1030 1031 /* 1032 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1033 * even though it is not harmful, it was not really necessary. 1034 */ 1035 1036 /* cancel timer */ 1037 nd6_llinfo_settimer_locked(ln, -1); 1038 1039 ifp = ln->lle_tbl->llt_ifp; 1040 1041 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1042 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1043 1044 if (dr != NULL && dr->expire && 1045 ln->ln_state == ND6_LLINFO_STALE && gc) { 1046 /* 1047 * If the reason for the deletion is just garbage 1048 * collection, and the neighbor is an active default 1049 * router, do not delete it. Instead, reset the GC 1050 * timer using the router's lifetime. 1051 * Simply deleting the entry would affect default 1052 * router selection, which is not necessarily a good 1053 * thing, especially when we're using router preference 1054 * values. 1055 * XXX: the check for ln_state would be redundant, 1056 * but we intentionally keep it just in case. 1057 */ 1058 if (dr->expire > time_uptime) 1059 nd6_llinfo_settimer_locked(ln, 1060 (dr->expire - time_uptime) * hz); 1061 else 1062 nd6_llinfo_settimer_locked(ln, 1063 (long)V_nd6_gctimer * hz); 1064 1065 next = LIST_NEXT(ln, lle_next); 1066 LLE_REMREF(ln); 1067 LLE_WUNLOCK(ln); 1068 return (next); 1069 } 1070 1071 if (dr) { 1072 /* 1073 * Unreachablity of a router might affect the default 1074 * router selection and on-link detection of advertised 1075 * prefixes. 1076 */ 1077 1078 /* 1079 * Temporarily fake the state to choose a new default 1080 * router and to perform on-link determination of 1081 * prefixes correctly. 1082 * Below the state will be set correctly, 1083 * or the entry itself will be deleted. 1084 */ 1085 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1086 } 1087 1088 if (ln->ln_router || dr) { 1089 1090 /* 1091 * We need to unlock to avoid a LOR with rt6_flush() with the 1092 * rnh and for the calls to pfxlist_onlink_check() and 1093 * defrouter_select() in the block further down for calls 1094 * into nd6_lookup(). We still hold a ref. 1095 */ 1096 LLE_WUNLOCK(ln); 1097 1098 /* 1099 * rt6_flush must be called whether or not the neighbor 1100 * is in the Default Router List. 1101 * See a corresponding comment in nd6_na_input(). 1102 */ 1103 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1104 } 1105 1106 if (dr) { 1107 /* 1108 * Since defrouter_select() does not affect the 1109 * on-link determination and MIP6 needs the check 1110 * before the default router selection, we perform 1111 * the check now. 1112 */ 1113 pfxlist_onlink_check(); 1114 1115 /* 1116 * Refresh default router list. 1117 */ 1118 defrouter_select(); 1119 } 1120 1121 if (ln->ln_router || dr) 1122 LLE_WLOCK(ln); 1123 } 1124 1125 /* 1126 * Before deleting the entry, remember the next entry as the 1127 * return value. We need this because pfxlist_onlink_check() above 1128 * might have freed other entries (particularly the old next entry) as 1129 * a side effect (XXX). 1130 */ 1131 next = LIST_NEXT(ln, lle_next); 1132 1133 /* 1134 * Save to unlock. We still hold an extra reference and will not 1135 * free(9) in llentry_free() if someone else holds one as well. 1136 */ 1137 LLE_WUNLOCK(ln); 1138 IF_AFDATA_LOCK(ifp); 1139 LLE_WLOCK(ln); 1140 1141 /* Guard against race with other llentry_free(). */ 1142 if (ln->la_flags & LLE_LINKED) { 1143 LLE_REMREF(ln); 1144 llentry_free(ln); 1145 } else 1146 LLE_FREE_LOCKED(ln); 1147 1148 IF_AFDATA_UNLOCK(ifp); 1149 1150 return (next); 1151} 1152 1153/* 1154 * Upper-layer reachability hint for Neighbor Unreachability Detection. 1155 * 1156 * XXX cost-effective methods? 1157 */ 1158void 1159nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force) 1160{ 1161 struct llentry *ln; 1162 struct ifnet *ifp; 1163 1164 if ((dst6 == NULL) || (rt == NULL)) 1165 return; 1166 1167 ifp = rt->rt_ifp; 1168 IF_AFDATA_RLOCK(ifp); 1169 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL); 1170 IF_AFDATA_RUNLOCK(ifp); 1171 if (ln == NULL) 1172 return; 1173 1174 if (ln->ln_state < ND6_LLINFO_REACHABLE) 1175 goto done; 1176 1177 /* 1178 * if we get upper-layer reachability confirmation many times, 1179 * it is possible we have false information. 1180 */ 1181 if (!force) { 1182 ln->ln_byhint++; 1183 if (ln->ln_byhint > V_nd6_maxnudhint) { 1184 goto done; 1185 } 1186 } 1187 1188 ln->ln_state = ND6_LLINFO_REACHABLE; 1189 if (!ND6_LLINFO_PERMANENT(ln)) { 1190 nd6_llinfo_settimer_locked(ln, 1191 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz); 1192 } 1193done: 1194 LLE_WUNLOCK(ln); 1195} 1196 1197 1198/* 1199 * Rejuvenate this function for routing operations related 1200 * processing. 1201 */ 1202void 1203nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info) 1204{ 1205 struct sockaddr_in6 *gateway; 1206 struct nd_defrouter *dr; 1207 struct ifnet *ifp; 1208 1209 RT_LOCK_ASSERT(rt); 1210 gateway = (struct sockaddr_in6 *)rt->rt_gateway; 1211 ifp = rt->rt_ifp; 1212 1213 switch (req) { 1214 case RTM_ADD: 1215 break; 1216 1217 case RTM_DELETE: 1218 if (!ifp) 1219 return; 1220 /* 1221 * Only indirect routes are interesting. 1222 */ 1223 if ((rt->rt_flags & RTF_GATEWAY) == 0) 1224 return; 1225 /* 1226 * check for default route 1227 */ 1228 if (IN6_ARE_ADDR_EQUAL(&in6addr_any, 1229 &SIN6(rt_key(rt))->sin6_addr)) { 1230 1231 dr = defrouter_lookup(&gateway->sin6_addr, ifp); 1232 if (dr != NULL) 1233 dr->installed = 0; 1234 } 1235 break; 1236 } 1237} 1238 1239 1240int 1241nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) 1242{ 1243 struct in6_drlist *drl = (struct in6_drlist *)data; 1244 struct in6_oprlist *oprl = (struct in6_oprlist *)data; 1245 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1246 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1247 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1248 struct nd_defrouter *dr; 1249 struct nd_prefix *pr; 1250 int i = 0, error = 0; 1251 1252 if (ifp->if_afdata[AF_INET6] == NULL) 1253 return (EPFNOSUPPORT); 1254 switch (cmd) { 1255 case SIOCGDRLST_IN6: 1256 /* 1257 * obsolete API, use sysctl under net.inet6.icmp6 1258 */ 1259 bzero(drl, sizeof(*drl)); 1260 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 1261 if (i >= DRLSTSIZ) 1262 break; 1263 drl->defrouter[i].rtaddr = dr->rtaddr; 1264 in6_clearscope(&drl->defrouter[i].rtaddr); 1265 1266 drl->defrouter[i].flags = dr->flags; 1267 drl->defrouter[i].rtlifetime = dr->rtlifetime; 1268 drl->defrouter[i].expire = dr->expire + 1269 (time_second - time_uptime); 1270 drl->defrouter[i].if_index = dr->ifp->if_index; 1271 i++; 1272 } 1273 break; 1274 case SIOCGPRLST_IN6: 1275 /* 1276 * obsolete API, use sysctl under net.inet6.icmp6 1277 * 1278 * XXX the structure in6_prlist was changed in backward- 1279 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6, 1280 * in6_prlist is used for nd6_sysctl() - fill_prlist(). 1281 */ 1282 /* 1283 * XXX meaning of fields, especialy "raflags", is very 1284 * differnet between RA prefix list and RR/static prefix list. 1285 * how about separating ioctls into two? 1286 */ 1287 bzero(oprl, sizeof(*oprl)); 1288 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 1289 struct nd_pfxrouter *pfr; 1290 int j; 1291 1292 if (i >= PRLSTSIZ) 1293 break; 1294 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr; 1295 oprl->prefix[i].raflags = pr->ndpr_raf; 1296 oprl->prefix[i].prefixlen = pr->ndpr_plen; 1297 oprl->prefix[i].vltime = pr->ndpr_vltime; 1298 oprl->prefix[i].pltime = pr->ndpr_pltime; 1299 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index; 1300 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 1301 oprl->prefix[i].expire = 0; 1302 else { 1303 time_t maxexpire; 1304 1305 /* XXX: we assume time_t is signed. */ 1306 maxexpire = (-1) & 1307 ~((time_t)1 << 1308 ((sizeof(maxexpire) * 8) - 1)); 1309 if (pr->ndpr_vltime < 1310 maxexpire - pr->ndpr_lastupdate) { 1311 oprl->prefix[i].expire = 1312 pr->ndpr_lastupdate + 1313 pr->ndpr_vltime + 1314 (time_second - time_uptime); 1315 } else 1316 oprl->prefix[i].expire = maxexpire; 1317 } 1318 1319 j = 0; 1320 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 1321 if (j < DRLSTSIZ) { 1322#define RTRADDR oprl->prefix[i].advrtr[j] 1323 RTRADDR = pfr->router->rtaddr; 1324 in6_clearscope(&RTRADDR); 1325#undef RTRADDR 1326 } 1327 j++; 1328 } 1329 oprl->prefix[i].advrtrs = j; 1330 oprl->prefix[i].origin = PR_ORIG_RA; 1331 1332 i++; 1333 } 1334 1335 break; 1336 case OSIOCGIFINFO_IN6: 1337#define ND ndi->ndi 1338 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1339 bzero(&ND, sizeof(ND)); 1340 ND.linkmtu = IN6_LINKMTU(ifp); 1341 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1342 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1343 ND.reachable = ND_IFINFO(ifp)->reachable; 1344 ND.retrans = ND_IFINFO(ifp)->retrans; 1345 ND.flags = ND_IFINFO(ifp)->flags; 1346 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1347 ND.chlim = ND_IFINFO(ifp)->chlim; 1348 break; 1349 case SIOCGIFINFO_IN6: 1350 ND = *ND_IFINFO(ifp); 1351 break; 1352 case SIOCSIFINFO_IN6: 1353 /* 1354 * used to change host variables from userland. 1355 * intented for a use on router to reflect RA configurations. 1356 */ 1357 /* 0 means 'unspecified' */ 1358 if (ND.linkmtu != 0) { 1359 if (ND.linkmtu < IPV6_MMTU || 1360 ND.linkmtu > IN6_LINKMTU(ifp)) { 1361 error = EINVAL; 1362 break; 1363 } 1364 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1365 } 1366 1367 if (ND.basereachable != 0) { 1368 int obasereachable = ND_IFINFO(ifp)->basereachable; 1369 1370 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1371 if (ND.basereachable != obasereachable) 1372 ND_IFINFO(ifp)->reachable = 1373 ND_COMPUTE_RTIME(ND.basereachable); 1374 } 1375 if (ND.retrans != 0) 1376 ND_IFINFO(ifp)->retrans = ND.retrans; 1377 if (ND.chlim != 0) 1378 ND_IFINFO(ifp)->chlim = ND.chlim; 1379 /* FALLTHROUGH */ 1380 case SIOCSIFINFO_FLAGS: 1381 { 1382 struct ifaddr *ifa; 1383 struct in6_ifaddr *ia; 1384 1385 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1386 !(ND.flags & ND6_IFF_IFDISABLED)) { 1387 /* ifdisabled 1->0 transision */ 1388 1389 /* 1390 * If the interface is marked as ND6_IFF_IFDISABLED and 1391 * has an link-local address with IN6_IFF_DUPLICATED, 1392 * do not clear ND6_IFF_IFDISABLED. 1393 * See RFC 4862, Section 5.4.5. 1394 */ 1395 int duplicated_linklocal = 0; 1396 1397 IF_ADDR_RLOCK(ifp); 1398 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1399 if (ifa->ifa_addr->sa_family != AF_INET6) 1400 continue; 1401 ia = (struct in6_ifaddr *)ifa; 1402 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) && 1403 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1404 duplicated_linklocal = 1; 1405 break; 1406 } 1407 } 1408 IF_ADDR_RUNLOCK(ifp); 1409 1410 if (duplicated_linklocal) { 1411 ND.flags |= ND6_IFF_IFDISABLED; 1412 log(LOG_ERR, "Cannot enable an interface" 1413 " with a link-local address marked" 1414 " duplicate.\n"); 1415 } else { 1416 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED; 1417 if (ifp->if_flags & IFF_UP) 1418 in6_if_up(ifp); 1419 } 1420 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1421 (ND.flags & ND6_IFF_IFDISABLED)) { 1422 /* ifdisabled 0->1 transision */ 1423 /* Mark all IPv6 address as tentative. */ 1424 1425 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; 1426 IF_ADDR_RLOCK(ifp); 1427 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1428 if (ifa->ifa_addr->sa_family != AF_INET6) 1429 continue; 1430 ia = (struct in6_ifaddr *)ifa; 1431 ia->ia6_flags |= IN6_IFF_TENTATIVE; 1432 } 1433 IF_ADDR_RUNLOCK(ifp); 1434 } 1435 1436 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) { 1437 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) { 1438 /* auto_linklocal 0->1 transision */ 1439 1440 /* If no link-local address on ifp, configure */ 1441 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL; 1442 in6_ifattach(ifp, NULL); 1443 } else if (!(ND.flags & ND6_IFF_IFDISABLED) && 1444 ifp->if_flags & IFF_UP) { 1445 /* 1446 * When the IF already has 1447 * ND6_IFF_AUTO_LINKLOCAL, no link-local 1448 * address is assigned, and IFF_UP, try to 1449 * assign one. 1450 */ 1451 int haslinklocal = 0; 1452 1453 IF_ADDR_RLOCK(ifp); 1454 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1455 if (ifa->ifa_addr->sa_family != AF_INET6) 1456 continue; 1457 ia = (struct in6_ifaddr *)ifa; 1458 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1459 haslinklocal = 1; 1460 break; 1461 } 1462 } 1463 IF_ADDR_RUNLOCK(ifp); 1464 if (!haslinklocal) 1465 in6_ifattach(ifp, NULL); 1466 } 1467 } 1468 } 1469 ND_IFINFO(ifp)->flags = ND.flags; 1470 break; 1471#undef ND 1472 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1473 /* sync kernel routing table with the default router list */ 1474 defrouter_reset(); 1475 defrouter_select(); 1476 break; 1477 case SIOCSPFXFLUSH_IN6: 1478 { 1479 /* flush all the prefix advertised by routers */ 1480 struct nd_prefix *pr, *next; 1481 1482 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) { 1483 struct in6_ifaddr *ia, *ia_next; 1484 1485 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1486 continue; /* XXX */ 1487 1488 /* do we really have to remove addresses as well? */ 1489 /* XXXRW: in6_ifaddrhead locking. */ 1490 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link, 1491 ia_next) { 1492 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1493 continue; 1494 1495 if (ia->ia6_ndpr == pr) 1496 in6_purgeaddr(&ia->ia_ifa); 1497 } 1498 prelist_remove(pr); 1499 } 1500 break; 1501 } 1502 case SIOCSRTRFLUSH_IN6: 1503 { 1504 /* flush all the default routers */ 1505 struct nd_defrouter *dr, *next; 1506 1507 defrouter_reset(); 1508 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) { 1509 defrtrlist_del(dr); 1510 } 1511 defrouter_select(); 1512 break; 1513 } 1514 case SIOCGNBRINFO_IN6: 1515 { 1516 struct llentry *ln; 1517 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1518 1519 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1520 return (error); 1521 1522 IF_AFDATA_RLOCK(ifp); 1523 ln = nd6_lookup(&nb_addr, 0, ifp); 1524 IF_AFDATA_RUNLOCK(ifp); 1525 1526 if (ln == NULL) { 1527 error = EINVAL; 1528 break; 1529 } 1530 nbi->state = ln->ln_state; 1531 nbi->asked = ln->la_asked; 1532 nbi->isrouter = ln->ln_router; 1533 if (ln->la_expire == 0) 1534 nbi->expire = 0; 1535 else 1536 nbi->expire = ln->la_expire + 1537 (time_second - time_uptime); 1538 LLE_RUNLOCK(ln); 1539 break; 1540 } 1541 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1542 ndif->ifindex = V_nd6_defifindex; 1543 break; 1544 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1545 return (nd6_setdefaultiface(ndif->ifindex)); 1546 } 1547 return (error); 1548} 1549 1550/* 1551 * Create neighbor cache entry and cache link-layer address, 1552 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1553 * 1554 * type - ICMP6 type 1555 * code - type dependent information 1556 * 1557 * XXXXX 1558 * The caller of this function already acquired the ndp 1559 * cache table lock because the cache entry is returned. 1560 */ 1561struct llentry * 1562nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, 1563 int lladdrlen, int type, int code) 1564{ 1565 struct llentry *ln = NULL; 1566 int is_newentry; 1567 int do_update; 1568 int olladdr; 1569 int llchange; 1570 int flags; 1571 int newstate = 0; 1572 uint16_t router = 0; 1573 struct sockaddr_in6 sin6; 1574 struct mbuf *chain = NULL; 1575 int static_route = 0; 1576 1577 IF_AFDATA_UNLOCK_ASSERT(ifp); 1578 1579 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__)); 1580 KASSERT(from != NULL, ("%s: from == NULL", __func__)); 1581 1582 /* nothing must be updated for unspecified address */ 1583 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1584 return NULL; 1585 1586 /* 1587 * Validation about ifp->if_addrlen and lladdrlen must be done in 1588 * the caller. 1589 * 1590 * XXX If the link does not have link-layer adderss, what should 1591 * we do? (ifp->if_addrlen == 0) 1592 * Spec says nothing in sections for RA, RS and NA. There's small 1593 * description on it in NS section (RFC 2461 7.2.3). 1594 */ 1595 flags = lladdr ? ND6_EXCLUSIVE : 0; 1596 IF_AFDATA_RLOCK(ifp); 1597 ln = nd6_lookup(from, flags, ifp); 1598 IF_AFDATA_RUNLOCK(ifp); 1599 if (ln == NULL) { 1600 flags |= ND6_EXCLUSIVE; 1601 IF_AFDATA_LOCK(ifp); 1602 ln = nd6_lookup(from, flags | ND6_CREATE, ifp); 1603 IF_AFDATA_UNLOCK(ifp); 1604 is_newentry = 1; 1605 } else { 1606 /* do nothing if static ndp is set */ 1607 if (ln->la_flags & LLE_STATIC) { 1608 static_route = 1; 1609 goto done; 1610 } 1611 is_newentry = 0; 1612 } 1613 if (ln == NULL) 1614 return (NULL); 1615 1616 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0; 1617 if (olladdr && lladdr) { 1618 llchange = bcmp(lladdr, &ln->ll_addr, 1619 ifp->if_addrlen); 1620 } else 1621 llchange = 0; 1622 1623 /* 1624 * newentry olladdr lladdr llchange (*=record) 1625 * 0 n n -- (1) 1626 * 0 y n -- (2) 1627 * 0 n y -- (3) * STALE 1628 * 0 y y n (4) * 1629 * 0 y y y (5) * STALE 1630 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1631 * 1 -- y -- (7) * STALE 1632 */ 1633 1634 if (lladdr) { /* (3-5) and (7) */ 1635 /* 1636 * Record source link-layer address 1637 * XXX is it dependent to ifp->if_type? 1638 */ 1639 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen); 1640 ln->la_flags |= LLE_VALID; 1641 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 1642 } 1643 1644 if (!is_newentry) { 1645 if ((!olladdr && lladdr != NULL) || /* (3) */ 1646 (olladdr && lladdr != NULL && llchange)) { /* (5) */ 1647 do_update = 1; 1648 newstate = ND6_LLINFO_STALE; 1649 } else /* (1-2,4) */ 1650 do_update = 0; 1651 } else { 1652 do_update = 1; 1653 if (lladdr == NULL) /* (6) */ 1654 newstate = ND6_LLINFO_NOSTATE; 1655 else /* (7) */ 1656 newstate = ND6_LLINFO_STALE; 1657 } 1658 1659 if (do_update) { 1660 /* 1661 * Update the state of the neighbor cache. 1662 */ 1663 ln->ln_state = newstate; 1664 1665 if (ln->ln_state == ND6_LLINFO_STALE) { 1666 /* 1667 * XXX: since nd6_output() below will cause 1668 * state tansition to DELAY and reset the timer, 1669 * we must set the timer now, although it is actually 1670 * meaningless. 1671 */ 1672 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 1673 1674 if (ln->la_hold) { 1675 struct mbuf *m_hold, *m_hold_next; 1676 1677 /* 1678 * reset the la_hold in advance, to explicitly 1679 * prevent a la_hold lookup in nd6_output() 1680 * (wouldn't happen, though...) 1681 */ 1682 for (m_hold = ln->la_hold, ln->la_hold = NULL; 1683 m_hold; m_hold = m_hold_next) { 1684 m_hold_next = m_hold->m_nextpkt; 1685 m_hold->m_nextpkt = NULL; 1686 1687 /* 1688 * we assume ifp is not a p2p here, so 1689 * just set the 2nd argument as the 1690 * 1st one. 1691 */ 1692 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain); 1693 } 1694 /* 1695 * If we have mbufs in the chain we need to do 1696 * deferred transmit. Copy the address from the 1697 * llentry before dropping the lock down below. 1698 */ 1699 if (chain != NULL) 1700 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6)); 1701 } 1702 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { 1703 /* probe right away */ 1704 nd6_llinfo_settimer_locked((void *)ln, 0); 1705 } 1706 } 1707 1708 /* 1709 * ICMP6 type dependent behavior. 1710 * 1711 * NS: clear IsRouter if new entry 1712 * RS: clear IsRouter 1713 * RA: set IsRouter if there's lladdr 1714 * redir: clear IsRouter if new entry 1715 * 1716 * RA case, (1): 1717 * The spec says that we must set IsRouter in the following cases: 1718 * - If lladdr exist, set IsRouter. This means (1-5). 1719 * - If it is old entry (!newentry), set IsRouter. This means (7). 1720 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1721 * A quetion arises for (1) case. (1) case has no lladdr in the 1722 * neighbor cache, this is similar to (6). 1723 * This case is rare but we figured that we MUST NOT set IsRouter. 1724 * 1725 * newentry olladdr lladdr llchange NS RS RA redir 1726 * D R 1727 * 0 n n -- (1) c ? s 1728 * 0 y n -- (2) c s s 1729 * 0 n y -- (3) c s s 1730 * 0 y y n (4) c s s 1731 * 0 y y y (5) c s s 1732 * 1 -- n -- (6) c c c s 1733 * 1 -- y -- (7) c c s c s 1734 * 1735 * (c=clear s=set) 1736 */ 1737 switch (type & 0xff) { 1738 case ND_NEIGHBOR_SOLICIT: 1739 /* 1740 * New entry must have is_router flag cleared. 1741 */ 1742 if (is_newentry) /* (6-7) */ 1743 ln->ln_router = 0; 1744 break; 1745 case ND_REDIRECT: 1746 /* 1747 * If the icmp is a redirect to a better router, always set the 1748 * is_router flag. Otherwise, if the entry is newly created, 1749 * clear the flag. [RFC 2461, sec 8.3] 1750 */ 1751 if (code == ND_REDIRECT_ROUTER) 1752 ln->ln_router = 1; 1753 else if (is_newentry) /* (6-7) */ 1754 ln->ln_router = 0; 1755 break; 1756 case ND_ROUTER_SOLICIT: 1757 /* 1758 * is_router flag must always be cleared. 1759 */ 1760 ln->ln_router = 0; 1761 break; 1762 case ND_ROUTER_ADVERT: 1763 /* 1764 * Mark an entry with lladdr as a router. 1765 */ 1766 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */ 1767 (is_newentry && lladdr)) { /* (7) */ 1768 ln->ln_router = 1; 1769 } 1770 break; 1771 } 1772 1773 if (ln != NULL) { 1774 static_route = (ln->la_flags & LLE_STATIC); 1775 router = ln->ln_router; 1776 1777 if (flags & ND6_EXCLUSIVE) 1778 LLE_WUNLOCK(ln); 1779 else 1780 LLE_RUNLOCK(ln); 1781 if (static_route) 1782 ln = NULL; 1783 } 1784 if (chain) 1785 nd6_output_flush(ifp, ifp, chain, &sin6, NULL); 1786 1787 /* 1788 * When the link-layer address of a router changes, select the 1789 * best router again. In particular, when the neighbor entry is newly 1790 * created, it might affect the selection policy. 1791 * Question: can we restrict the first condition to the "is_newentry" 1792 * case? 1793 * XXX: when we hear an RA from a new router with the link-layer 1794 * address option, defrouter_select() is called twice, since 1795 * defrtrlist_update called the function as well. However, I believe 1796 * we can compromise the overhead, since it only happens the first 1797 * time. 1798 * XXX: although defrouter_select() should not have a bad effect 1799 * for those are not autoconfigured hosts, we explicitly avoid such 1800 * cases for safety. 1801 */ 1802 if (do_update && router && 1803 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1804 /* 1805 * guaranteed recursion 1806 */ 1807 defrouter_select(); 1808 } 1809 1810 return (ln); 1811done: 1812 if (ln != NULL) { 1813 if (flags & ND6_EXCLUSIVE) 1814 LLE_WUNLOCK(ln); 1815 else 1816 LLE_RUNLOCK(ln); 1817 if (static_route) 1818 ln = NULL; 1819 } 1820 return (ln); 1821} 1822 1823static void 1824nd6_slowtimo(void *arg) 1825{ 1826 CURVNET_SET((struct vnet *) arg); 1827 struct nd_ifinfo *nd6if; 1828 struct ifnet *ifp; 1829 1830 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 1831 nd6_slowtimo, curvnet); 1832 IFNET_RLOCK_NOSLEEP(); 1833 TAILQ_FOREACH(ifp, &V_ifnet, if_list) { 1834 if (ifp->if_afdata[AF_INET6] == NULL) 1835 continue; 1836 nd6if = ND_IFINFO(ifp); 1837 if (nd6if->basereachable && /* already initialized */ 1838 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 1839 /* 1840 * Since reachable time rarely changes by router 1841 * advertisements, we SHOULD insure that a new random 1842 * value gets recomputed at least once every few hours. 1843 * (RFC 2461, 6.3.4) 1844 */ 1845 nd6if->recalctm = V_nd6_recalc_reachtm_interval; 1846 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 1847 } 1848 } 1849 IFNET_RUNLOCK_NOSLEEP(); 1850 CURVNET_RESTORE(); 1851} 1852 1853int 1854nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 1855 struct sockaddr_in6 *dst, struct rtentry *rt0) 1856{ 1857 1858 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL)); 1859} 1860 1861 1862/* 1863 * Note that I'm not enforcing any global serialization 1864 * lle state or asked changes here as the logic is too 1865 * complicated to avoid having to always acquire an exclusive 1866 * lock 1867 * KMM 1868 * 1869 */ 1870#define senderr(e) { error = (e); goto bad;} 1871 1872int 1873nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 1874 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle, 1875 struct mbuf **chain) 1876{ 1877 struct mbuf *m = m0; 1878 struct m_tag *mtag; 1879 struct llentry *ln = lle; 1880 struct ip6_hdr *ip6; 1881 int error = 0; 1882 int flags = 0; 1883 int ip6len; 1884 1885#ifdef INVARIANTS 1886 if (lle != NULL) { 1887 1888 LLE_WLOCK_ASSERT(lle); 1889 1890 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed")); 1891 } 1892#endif 1893 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) 1894 goto sendpkt; 1895 1896 if (nd6_need_cache(ifp) == 0) 1897 goto sendpkt; 1898 1899 /* 1900 * next hop determination. This routine is derived from ether_output. 1901 */ 1902 1903 /* 1904 * Address resolution or Neighbor Unreachability Detection 1905 * for the next hop. 1906 * At this point, the destination of the packet must be a unicast 1907 * or an anycast address(i.e. not a multicast). 1908 */ 1909 1910 flags = (lle != NULL) ? LLE_EXCLUSIVE : 0; 1911 if (ln == NULL) { 1912 retry: 1913 IF_AFDATA_RLOCK(ifp); 1914 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst); 1915 IF_AFDATA_RUNLOCK(ifp); 1916 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) { 1917 /* 1918 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 1919 * the condition below is not very efficient. But we believe 1920 * it is tolerable, because this should be a rare case. 1921 */ 1922 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0); 1923 IF_AFDATA_LOCK(ifp); 1924 ln = nd6_lookup(&dst->sin6_addr, flags, ifp); 1925 IF_AFDATA_UNLOCK(ifp); 1926 } 1927 } 1928 if (ln == NULL) { 1929 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && 1930 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 1931 char ip6buf[INET6_ADDRSTRLEN]; 1932 log(LOG_DEBUG, 1933 "nd6_output: can't allocate llinfo for %s " 1934 "(ln=%p)\n", 1935 ip6_sprintf(ip6buf, &dst->sin6_addr), ln); 1936 senderr(EIO); /* XXX: good error? */ 1937 } 1938 goto sendpkt; /* send anyway */ 1939 } 1940 1941 /* We don't have to do link-layer address resolution on a p2p link. */ 1942 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && 1943 ln->ln_state < ND6_LLINFO_REACHABLE) { 1944 if ((flags & LLE_EXCLUSIVE) == 0) { 1945 flags |= LLE_EXCLUSIVE; 1946 LLE_RUNLOCK(ln); 1947 goto retry; 1948 } 1949 ln->ln_state = ND6_LLINFO_STALE; 1950 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 1951 } 1952 1953 /* 1954 * The first time we send a packet to a neighbor whose entry is 1955 * STALE, we have to change the state to DELAY and a sets a timer to 1956 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 1957 * neighbor unreachability detection on expiration. 1958 * (RFC 2461 7.3.3) 1959 */ 1960 if (ln->ln_state == ND6_LLINFO_STALE) { 1961 if ((flags & LLE_EXCLUSIVE) == 0) { 1962 flags |= LLE_EXCLUSIVE; 1963 LLE_RUNLOCK(ln); 1964 goto retry; 1965 } 1966 ln->la_asked = 0; 1967 ln->ln_state = ND6_LLINFO_DELAY; 1968 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz); 1969 } 1970 1971 /* 1972 * If the neighbor cache entry has a state other than INCOMPLETE 1973 * (i.e. its link-layer address is already resolved), just 1974 * send the packet. 1975 */ 1976 if (ln->ln_state > ND6_LLINFO_INCOMPLETE) 1977 goto sendpkt; 1978 1979 /* 1980 * There is a neighbor cache entry, but no ethernet address 1981 * response yet. Append this latest packet to the end of the 1982 * packet queue in the mbuf, unless the number of the packet 1983 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 1984 * the oldest packet in the queue will be removed. 1985 */ 1986 if (ln->ln_state == ND6_LLINFO_NOSTATE) 1987 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1988 1989 if ((flags & LLE_EXCLUSIVE) == 0) { 1990 flags |= LLE_EXCLUSIVE; 1991 LLE_RUNLOCK(ln); 1992 goto retry; 1993 } 1994 1995 LLE_WLOCK_ASSERT(ln); 1996 1997 if (ln->la_hold) { 1998 struct mbuf *m_hold; 1999 int i; 2000 2001 i = 0; 2002 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) { 2003 i++; 2004 if (m_hold->m_nextpkt == NULL) { 2005 m_hold->m_nextpkt = m; 2006 break; 2007 } 2008 } 2009 while (i >= V_nd6_maxqueuelen) { 2010 m_hold = ln->la_hold; 2011 ln->la_hold = ln->la_hold->m_nextpkt; 2012 m_freem(m_hold); 2013 i--; 2014 } 2015 } else { 2016 ln->la_hold = m; 2017 } 2018 2019 /* 2020 * If there has been no NS for the neighbor after entering the 2021 * INCOMPLETE state, send the first solicitation. 2022 */ 2023 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) { 2024 ln->la_asked++; 2025 2026 nd6_llinfo_settimer_locked(ln, 2027 (long)ND_IFINFO(ifp)->retrans * hz / 1000); 2028 LLE_WUNLOCK(ln); 2029 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); 2030 if (lle != NULL && ln == lle) 2031 LLE_WLOCK(lle); 2032 2033 } else if (lle == NULL || ln != lle) { 2034 /* 2035 * We did the lookup (no lle arg) so we 2036 * need to do the unlock here. 2037 */ 2038 LLE_WUNLOCK(ln); 2039 } 2040 2041 return (0); 2042 2043 sendpkt: 2044 /* discard the packet if IPv6 operation is disabled on the interface */ 2045 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 2046 error = ENETDOWN; /* better error? */ 2047 goto bad; 2048 } 2049 /* 2050 * ln is valid and the caller did not pass in 2051 * an llentry 2052 */ 2053 if ((ln != NULL) && (lle == NULL)) { 2054 if (flags & LLE_EXCLUSIVE) 2055 LLE_WUNLOCK(ln); 2056 else 2057 LLE_RUNLOCK(ln); 2058 } 2059 2060#ifdef MAC 2061 mac_netinet6_nd6_send(ifp, m); 2062#endif 2063 2064 /* 2065 * If called from nd6_ns_output() (NS), nd6_na_output() (NA), 2066 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA 2067 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND 2068 * to be diverted to user space. When re-injected into the kernel, 2069 * send_output() will directly dispatch them to the outgoing interface. 2070 */ 2071 if (send_sendso_input_hook != NULL) { 2072 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); 2073 if (mtag != NULL) { 2074 ip6 = mtod(m, struct ip6_hdr *); 2075 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); 2076 /* Use the SEND socket */ 2077 error = send_sendso_input_hook(m, ifp, SND_OUT, 2078 ip6len); 2079 /* -1 == no app on SEND socket */ 2080 if (error == 0 || error != -1) 2081 return (error); 2082 } 2083 } 2084 2085 /* 2086 * We were passed in a pointer to an lle with the lock held 2087 * this means that we can't call if_output as we will 2088 * recurse on the lle lock - so what we do is we create 2089 * a list of mbufs to send and transmit them in the caller 2090 * after the lock is dropped 2091 */ 2092 if (lle != NULL) { 2093 if (*chain == NULL) 2094 *chain = m; 2095 else { 2096 struct mbuf *mb; 2097 2098 /* 2099 * append mbuf to end of deferred chain 2100 */ 2101 mb = *chain; 2102 while (mb->m_nextpkt != NULL) 2103 mb = mb->m_nextpkt; 2104 mb->m_nextpkt = m; 2105 } 2106 return (error); 2107 } 2108 m_clrprotoflags(m); /* Avoid confusing lower layers. */ 2109 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL, 2110 mtod(m, struct ip6_hdr *)); 2111 if ((ifp->if_flags & IFF_LOOPBACK) != 0) { 2112 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst, 2113 NULL)); 2114 } 2115 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL); 2116 return (error); 2117 2118 bad: 2119 /* 2120 * ln is valid and the caller did not pass in 2121 * an llentry 2122 */ 2123 if ((ln != NULL) && (lle == NULL)) { 2124 if (flags & LLE_EXCLUSIVE) 2125 LLE_WUNLOCK(ln); 2126 else 2127 LLE_RUNLOCK(ln); 2128 } 2129 if (m) 2130 m_freem(m); 2131 return (error); 2132} 2133#undef senderr 2134 2135 2136int 2137nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain, 2138 struct sockaddr_in6 *dst, struct route *ro) 2139{ 2140 struct mbuf *m, *m_head; 2141 struct ifnet *outifp; 2142 int error = 0; 2143 2144 m_head = chain; 2145 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2146 outifp = origifp; 2147 else 2148 outifp = ifp; 2149 2150 while (m_head) { 2151 m = m_head; 2152 m_head = m_head->m_nextpkt; 2153 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro); 2154 } 2155 2156 /* 2157 * XXX 2158 * note that intermediate errors are blindly ignored - but this is 2159 * the same convention as used with nd6_output when called by 2160 * nd6_cache_lladdr 2161 */ 2162 return (error); 2163} 2164 2165 2166int 2167nd6_need_cache(struct ifnet *ifp) 2168{ 2169 /* 2170 * XXX: we currently do not make neighbor cache on any interface 2171 * other than ARCnet, Ethernet, FDDI and GIF. 2172 * 2173 * RFC2893 says: 2174 * - unidirectional tunnels needs no ND 2175 */ 2176 switch (ifp->if_type) { 2177 case IFT_ARCNET: 2178 case IFT_ETHER: 2179 case IFT_FDDI: 2180 case IFT_IEEE1394: 2181#ifdef IFT_L2VLAN 2182 case IFT_L2VLAN: 2183#endif 2184#ifdef IFT_IEEE80211 2185 case IFT_IEEE80211: 2186#endif 2187 case IFT_INFINIBAND: 2188 case IFT_GIF: /* XXX need more cases? */ 2189 case IFT_PPP: 2190 case IFT_TUNNEL: 2191 case IFT_BRIDGE: 2192 case IFT_PROPVIRTUAL: 2193 return (1); 2194 default: 2195 return (0); 2196 } 2197} 2198 2199/* 2200 * the callers of this function need to be re-worked to drop 2201 * the lle lock, drop here for now 2202 */ 2203int 2204nd6_storelladdr(struct ifnet *ifp, struct mbuf *m, 2205 const struct sockaddr *dst, u_char *desten, struct llentry **lle) 2206{ 2207 struct llentry *ln; 2208 2209 *lle = NULL; 2210 IF_AFDATA_UNLOCK_ASSERT(ifp); 2211 if (m != NULL && m->m_flags & M_MCAST) { 2212 int i; 2213 2214 switch (ifp->if_type) { 2215 case IFT_ETHER: 2216 case IFT_FDDI: 2217#ifdef IFT_L2VLAN 2218 case IFT_L2VLAN: 2219#endif 2220#ifdef IFT_IEEE80211 2221 case IFT_IEEE80211: 2222#endif 2223 case IFT_BRIDGE: 2224 case IFT_ISO88025: 2225 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr, 2226 desten); 2227 return (0); 2228 case IFT_IEEE1394: 2229 /* 2230 * netbsd can use if_broadcastaddr, but we don't do so 2231 * to reduce # of ifdef. 2232 */ 2233 for (i = 0; i < ifp->if_addrlen; i++) 2234 desten[i] = ~0; 2235 return (0); 2236 case IFT_ARCNET: 2237 *desten = 0; 2238 return (0); 2239 default: 2240 m_freem(m); 2241 return (EAFNOSUPPORT); 2242 } 2243 } 2244 2245 2246 /* 2247 * the entry should have been created in nd6_store_lladdr 2248 */ 2249 IF_AFDATA_RLOCK(ifp); 2250 ln = lla_lookup(LLTABLE6(ifp), 0, dst); 2251 IF_AFDATA_RUNLOCK(ifp); 2252 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) { 2253 if (ln != NULL) 2254 LLE_RUNLOCK(ln); 2255 /* this could happen, if we could not allocate memory */ 2256 m_freem(m); 2257 return (1); 2258 } 2259 2260 bcopy(&ln->ll_addr, desten, ifp->if_addrlen); 2261 *lle = ln; 2262 LLE_RUNLOCK(ln); 2263 /* 2264 * A *small* use after free race exists here 2265 */ 2266 return (0); 2267} 2268 2269static void 2270clear_llinfo_pqueue(struct llentry *ln) 2271{ 2272 struct mbuf *m_hold, *m_hold_next; 2273 2274 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) { 2275 m_hold_next = m_hold->m_nextpkt; 2276 m_freem(m_hold); 2277 } 2278 2279 ln->la_hold = NULL; 2280 return; 2281} 2282 2283static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2284static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2285#ifdef SYSCTL_DECL 2286SYSCTL_DECL(_net_inet6_icmp6); 2287#endif 2288SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2289 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2290SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2291 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2292SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2293 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, ""); 2294SYSCTL_VNET_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer, 2295 CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), ""); 2296 2297static int 2298nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2299{ 2300 struct in6_defrouter d; 2301 struct nd_defrouter *dr; 2302 int error; 2303 2304 if (req->newptr) 2305 return (EPERM); 2306 2307 bzero(&d, sizeof(d)); 2308 d.rtaddr.sin6_family = AF_INET6; 2309 d.rtaddr.sin6_len = sizeof(d.rtaddr); 2310 2311 /* 2312 * XXX locking 2313 */ 2314 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 2315 d.rtaddr.sin6_addr = dr->rtaddr; 2316 error = sa6_recoverscope(&d.rtaddr); 2317 if (error != 0) 2318 return (error); 2319 d.flags = dr->flags; 2320 d.rtlifetime = dr->rtlifetime; 2321 d.expire = dr->expire + (time_second - time_uptime); 2322 d.if_index = dr->ifp->if_index; 2323 error = SYSCTL_OUT(req, &d, sizeof(d)); 2324 if (error != 0) 2325 return (error); 2326 } 2327 return (0); 2328} 2329 2330static int 2331nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2332{ 2333 struct in6_prefix p; 2334 struct sockaddr_in6 s6; 2335 struct nd_prefix *pr; 2336 struct nd_pfxrouter *pfr; 2337 time_t maxexpire; 2338 int error; 2339 char ip6buf[INET6_ADDRSTRLEN]; 2340 2341 if (req->newptr) 2342 return (EPERM); 2343 2344 bzero(&p, sizeof(p)); 2345 p.origin = PR_ORIG_RA; 2346 bzero(&s6, sizeof(s6)); 2347 s6.sin6_family = AF_INET6; 2348 s6.sin6_len = sizeof(s6); 2349 2350 /* 2351 * XXX locking 2352 */ 2353 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 2354 p.prefix = pr->ndpr_prefix; 2355 if (sa6_recoverscope(&p.prefix)) { 2356 log(LOG_ERR, "scope error in prefix list (%s)\n", 2357 ip6_sprintf(ip6buf, &p.prefix.sin6_addr)); 2358 /* XXX: press on... */ 2359 } 2360 p.raflags = pr->ndpr_raf; 2361 p.prefixlen = pr->ndpr_plen; 2362 p.vltime = pr->ndpr_vltime; 2363 p.pltime = pr->ndpr_pltime; 2364 p.if_index = pr->ndpr_ifp->if_index; 2365 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2366 p.expire = 0; 2367 else { 2368 /* XXX: we assume time_t is signed. */ 2369 maxexpire = (-1) & 2370 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1)); 2371 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate) 2372 p.expire = pr->ndpr_lastupdate + 2373 pr->ndpr_vltime + 2374 (time_second - time_uptime); 2375 else 2376 p.expire = maxexpire; 2377 } 2378 p.refcnt = pr->ndpr_refcnt; 2379 p.flags = pr->ndpr_stateflags; 2380 p.advrtrs = 0; 2381 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) 2382 p.advrtrs++; 2383 error = SYSCTL_OUT(req, &p, sizeof(p)); 2384 if (error != 0) 2385 return (error); 2386 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 2387 s6.sin6_addr = pfr->router->rtaddr; 2388 if (sa6_recoverscope(&s6)) 2389 log(LOG_ERR, 2390 "scope error in prefix list (%s)\n", 2391 ip6_sprintf(ip6buf, &pfr->router->rtaddr)); 2392 error = SYSCTL_OUT(req, &s6, sizeof(s6)); 2393 if (error != 0) 2394 return (error); 2395 } 2396 } 2397 return (0); 2398} 2399