nd6.c revision 282445
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 282445 2015-05-05 03:17:32Z markj $"); 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 IF_ADDR_RUNLOCK(ifp); 765 766 if (public_ifa6 != NULL) { 767 int e; 768 769 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 770 ifa_free(&public_ifa6->ia_ifa); 771 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 772 " tmp addr,errno=%d\n", e); 773 return (-1); 774 } 775 ifa_free(&public_ifa6->ia_ifa); 776 return (0); 777 } 778 779 return (-1); 780} 781 782/* 783 * Nuke neighbor cache/prefix/default router management table, right before 784 * ifp goes away. 785 */ 786void 787nd6_purge(struct ifnet *ifp) 788{ 789 struct nd_defrouter *dr, *ndr; 790 struct nd_prefix *pr, *npr; 791 792 /* 793 * Nuke default router list entries toward ifp. 794 * We defer removal of default router list entries that is installed 795 * in the routing table, in order to keep additional side effects as 796 * small as possible. 797 */ 798 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 799 if (dr->installed) 800 continue; 801 802 if (dr->ifp == ifp) 803 defrtrlist_del(dr); 804 } 805 806 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 807 if (!dr->installed) 808 continue; 809 810 if (dr->ifp == ifp) 811 defrtrlist_del(dr); 812 } 813 814 /* Nuke prefix list entries toward ifp */ 815 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 816 if (pr->ndpr_ifp == ifp) { 817 /* 818 * Because if_detach() does *not* release prefixes 819 * while purging addresses the reference count will 820 * still be above zero. We therefore reset it to 821 * make sure that the prefix really gets purged. 822 */ 823 pr->ndpr_refcnt = 0; 824 825 /* 826 * Previously, pr->ndpr_addr is removed as well, 827 * but I strongly believe we don't have to do it. 828 * nd6_purge() is only called from in6_ifdetach(), 829 * which removes all the associated interface addresses 830 * by itself. 831 * (jinmei@kame.net 20010129) 832 */ 833 prelist_remove(pr); 834 } 835 } 836 837 /* cancel default outgoing interface setting */ 838 if (V_nd6_defifindex == ifp->if_index) 839 nd6_setdefaultiface(0); 840 841 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 842 /* Refresh default router list. */ 843 defrouter_select(); 844 } 845 846 /* XXXXX 847 * We do not nuke the neighbor cache entries here any more 848 * because the neighbor cache is kept in if_afdata[AF_INET6]. 849 * nd6_purge() is invoked by in6_ifdetach() which is called 850 * from if_detach() where everything gets purged. So let 851 * in6_domifdetach() do the actual L2 table purging work. 852 */ 853} 854 855/* 856 * the caller acquires and releases the lock on the lltbls 857 * Returns the llentry locked 858 */ 859struct llentry * 860nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp) 861{ 862 struct sockaddr_in6 sin6; 863 struct llentry *ln; 864 int llflags; 865 866 bzero(&sin6, sizeof(sin6)); 867 sin6.sin6_len = sizeof(struct sockaddr_in6); 868 sin6.sin6_family = AF_INET6; 869 sin6.sin6_addr = *addr6; 870 871 IF_AFDATA_LOCK_ASSERT(ifp); 872 873 llflags = 0; 874 if (flags & ND6_CREATE) 875 llflags |= LLE_CREATE; 876 if (flags & ND6_EXCLUSIVE) 877 llflags |= LLE_EXCLUSIVE; 878 879 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6); 880 if ((ln != NULL) && (llflags & LLE_CREATE)) 881 ln->ln_state = ND6_LLINFO_NOSTATE; 882 883 return (ln); 884} 885 886/* 887 * Test whether a given IPv6 address is a neighbor or not, ignoring 888 * the actual neighbor cache. The neighbor cache is ignored in order 889 * to not reenter the routing code from within itself. 890 */ 891static int 892nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 893{ 894 struct nd_prefix *pr; 895 struct ifaddr *dstaddr; 896 897 /* 898 * A link-local address is always a neighbor. 899 * XXX: a link does not necessarily specify a single interface. 900 */ 901 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 902 struct sockaddr_in6 sin6_copy; 903 u_int32_t zone; 904 905 /* 906 * We need sin6_copy since sa6_recoverscope() may modify the 907 * content (XXX). 908 */ 909 sin6_copy = *addr; 910 if (sa6_recoverscope(&sin6_copy)) 911 return (0); /* XXX: should be impossible */ 912 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 913 return (0); 914 if (sin6_copy.sin6_scope_id == zone) 915 return (1); 916 else 917 return (0); 918 } 919 920 /* 921 * If the address matches one of our addresses, 922 * it should be a neighbor. 923 * If the address matches one of our on-link prefixes, it should be a 924 * neighbor. 925 */ 926 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 927 if (pr->ndpr_ifp != ifp) 928 continue; 929 930 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) { 931 struct rtentry *rt; 932 933 /* Always use the default FIB here. */ 934 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 935 0, 0, RT_DEFAULT_FIB); 936 if (rt == NULL) 937 continue; 938 /* 939 * This is the case where multiple interfaces 940 * have the same prefix, but only one is installed 941 * into the routing table and that prefix entry 942 * is not the one being examined here. In the case 943 * where RADIX_MPATH is enabled, multiple route 944 * entries (of the same rt_key value) will be 945 * installed because the interface addresses all 946 * differ. 947 */ 948 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 949 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) { 950 RTFREE_LOCKED(rt); 951 continue; 952 } 953 RTFREE_LOCKED(rt); 954 } 955 956 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 957 &addr->sin6_addr, &pr->ndpr_mask)) 958 return (1); 959 } 960 961 /* 962 * If the address is assigned on the node of the other side of 963 * a p2p interface, the address should be a neighbor. 964 */ 965 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr); 966 if (dstaddr != NULL) { 967 if (dstaddr->ifa_ifp == ifp) { 968 ifa_free(dstaddr); 969 return (1); 970 } 971 ifa_free(dstaddr); 972 } 973 974 /* 975 * If the default router list is empty, all addresses are regarded 976 * as on-link, and thus, as a neighbor. 977 */ 978 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && 979 TAILQ_EMPTY(&V_nd_defrouter) && 980 V_nd6_defifindex == ifp->if_index) { 981 return (1); 982 } 983 984 return (0); 985} 986 987 988/* 989 * Detect if a given IPv6 address identifies a neighbor on a given link. 990 * XXX: should take care of the destination of a p2p link? 991 */ 992int 993nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 994{ 995 struct llentry *lle; 996 int rc = 0; 997 998 IF_AFDATA_UNLOCK_ASSERT(ifp); 999 if (nd6_is_new_addr_neighbor(addr, ifp)) 1000 return (1); 1001 1002 /* 1003 * Even if the address matches none of our addresses, it might be 1004 * in the neighbor cache. 1005 */ 1006 IF_AFDATA_RLOCK(ifp); 1007 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) { 1008 LLE_RUNLOCK(lle); 1009 rc = 1; 1010 } 1011 IF_AFDATA_RUNLOCK(ifp); 1012 return (rc); 1013} 1014 1015/* 1016 * Free an nd6 llinfo entry. 1017 * Since the function would cause significant changes in the kernel, DO NOT 1018 * make it global, unless you have a strong reason for the change, and are sure 1019 * that the change is safe. 1020 */ 1021static struct llentry * 1022nd6_free(struct llentry *ln, int gc) 1023{ 1024 struct llentry *next; 1025 struct nd_defrouter *dr; 1026 struct ifnet *ifp; 1027 1028 LLE_WLOCK_ASSERT(ln); 1029 1030 /* 1031 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1032 * even though it is not harmful, it was not really necessary. 1033 */ 1034 1035 /* cancel timer */ 1036 nd6_llinfo_settimer_locked(ln, -1); 1037 1038 ifp = ln->lle_tbl->llt_ifp; 1039 1040 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1041 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1042 1043 if (dr != NULL && dr->expire && 1044 ln->ln_state == ND6_LLINFO_STALE && gc) { 1045 /* 1046 * If the reason for the deletion is just garbage 1047 * collection, and the neighbor is an active default 1048 * router, do not delete it. Instead, reset the GC 1049 * timer using the router's lifetime. 1050 * Simply deleting the entry would affect default 1051 * router selection, which is not necessarily a good 1052 * thing, especially when we're using router preference 1053 * values. 1054 * XXX: the check for ln_state would be redundant, 1055 * but we intentionally keep it just in case. 1056 */ 1057 if (dr->expire > time_uptime) 1058 nd6_llinfo_settimer_locked(ln, 1059 (dr->expire - time_uptime) * hz); 1060 else 1061 nd6_llinfo_settimer_locked(ln, 1062 (long)V_nd6_gctimer * hz); 1063 1064 next = LIST_NEXT(ln, lle_next); 1065 LLE_REMREF(ln); 1066 LLE_WUNLOCK(ln); 1067 return (next); 1068 } 1069 1070 if (dr) { 1071 /* 1072 * Unreachablity of a router might affect the default 1073 * router selection and on-link detection of advertised 1074 * prefixes. 1075 */ 1076 1077 /* 1078 * Temporarily fake the state to choose a new default 1079 * router and to perform on-link determination of 1080 * prefixes correctly. 1081 * Below the state will be set correctly, 1082 * or the entry itself will be deleted. 1083 */ 1084 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1085 } 1086 1087 if (ln->ln_router || dr) { 1088 1089 /* 1090 * We need to unlock to avoid a LOR with rt6_flush() with the 1091 * rnh and for the calls to pfxlist_onlink_check() and 1092 * defrouter_select() in the block further down for calls 1093 * into nd6_lookup(). We still hold a ref. 1094 */ 1095 LLE_WUNLOCK(ln); 1096 1097 /* 1098 * rt6_flush must be called whether or not the neighbor 1099 * is in the Default Router List. 1100 * See a corresponding comment in nd6_na_input(). 1101 */ 1102 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1103 } 1104 1105 if (dr) { 1106 /* 1107 * Since defrouter_select() does not affect the 1108 * on-link determination and MIP6 needs the check 1109 * before the default router selection, we perform 1110 * the check now. 1111 */ 1112 pfxlist_onlink_check(); 1113 1114 /* 1115 * Refresh default router list. 1116 */ 1117 defrouter_select(); 1118 } 1119 1120 if (ln->ln_router || dr) 1121 LLE_WLOCK(ln); 1122 } 1123 1124 /* 1125 * Before deleting the entry, remember the next entry as the 1126 * return value. We need this because pfxlist_onlink_check() above 1127 * might have freed other entries (particularly the old next entry) as 1128 * a side effect (XXX). 1129 */ 1130 next = LIST_NEXT(ln, lle_next); 1131 1132 /* 1133 * Save to unlock. We still hold an extra reference and will not 1134 * free(9) in llentry_free() if someone else holds one as well. 1135 */ 1136 LLE_WUNLOCK(ln); 1137 IF_AFDATA_LOCK(ifp); 1138 LLE_WLOCK(ln); 1139 1140 /* Guard against race with other llentry_free(). */ 1141 if (ln->la_flags & LLE_LINKED) { 1142 LLE_REMREF(ln); 1143 llentry_free(ln); 1144 } else 1145 LLE_FREE_LOCKED(ln); 1146 1147 IF_AFDATA_UNLOCK(ifp); 1148 1149 return (next); 1150} 1151 1152/* 1153 * Upper-layer reachability hint for Neighbor Unreachability Detection. 1154 * 1155 * XXX cost-effective methods? 1156 */ 1157void 1158nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force) 1159{ 1160 struct llentry *ln; 1161 struct ifnet *ifp; 1162 1163 if ((dst6 == NULL) || (rt == NULL)) 1164 return; 1165 1166 ifp = rt->rt_ifp; 1167 IF_AFDATA_RLOCK(ifp); 1168 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL); 1169 IF_AFDATA_RUNLOCK(ifp); 1170 if (ln == NULL) 1171 return; 1172 1173 if (ln->ln_state < ND6_LLINFO_REACHABLE) 1174 goto done; 1175 1176 /* 1177 * if we get upper-layer reachability confirmation many times, 1178 * it is possible we have false information. 1179 */ 1180 if (!force) { 1181 ln->ln_byhint++; 1182 if (ln->ln_byhint > V_nd6_maxnudhint) { 1183 goto done; 1184 } 1185 } 1186 1187 ln->ln_state = ND6_LLINFO_REACHABLE; 1188 if (!ND6_LLINFO_PERMANENT(ln)) { 1189 nd6_llinfo_settimer_locked(ln, 1190 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz); 1191 } 1192done: 1193 LLE_WUNLOCK(ln); 1194} 1195 1196 1197/* 1198 * Rejuvenate this function for routing operations related 1199 * processing. 1200 */ 1201void 1202nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info) 1203{ 1204 struct sockaddr_in6 *gateway; 1205 struct nd_defrouter *dr; 1206 struct ifnet *ifp; 1207 1208 RT_LOCK_ASSERT(rt); 1209 gateway = (struct sockaddr_in6 *)rt->rt_gateway; 1210 ifp = rt->rt_ifp; 1211 1212 switch (req) { 1213 case RTM_ADD: 1214 break; 1215 1216 case RTM_DELETE: 1217 if (!ifp) 1218 return; 1219 /* 1220 * Only indirect routes are interesting. 1221 */ 1222 if ((rt->rt_flags & RTF_GATEWAY) == 0) 1223 return; 1224 /* 1225 * check for default route 1226 */ 1227 if (IN6_ARE_ADDR_EQUAL(&in6addr_any, 1228 &SIN6(rt_key(rt))->sin6_addr)) { 1229 1230 dr = defrouter_lookup(&gateway->sin6_addr, ifp); 1231 if (dr != NULL) 1232 dr->installed = 0; 1233 } 1234 break; 1235 } 1236} 1237 1238 1239int 1240nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) 1241{ 1242 struct in6_drlist *drl = (struct in6_drlist *)data; 1243 struct in6_oprlist *oprl = (struct in6_oprlist *)data; 1244 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1245 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1246 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1247 struct nd_defrouter *dr; 1248 struct nd_prefix *pr; 1249 int i = 0, error = 0; 1250 1251 if (ifp->if_afdata[AF_INET6] == NULL) 1252 return (EPFNOSUPPORT); 1253 switch (cmd) { 1254 case SIOCGDRLST_IN6: 1255 /* 1256 * obsolete API, use sysctl under net.inet6.icmp6 1257 */ 1258 bzero(drl, sizeof(*drl)); 1259 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 1260 if (i >= DRLSTSIZ) 1261 break; 1262 drl->defrouter[i].rtaddr = dr->rtaddr; 1263 in6_clearscope(&drl->defrouter[i].rtaddr); 1264 1265 drl->defrouter[i].flags = dr->flags; 1266 drl->defrouter[i].rtlifetime = dr->rtlifetime; 1267 drl->defrouter[i].expire = dr->expire + 1268 (time_second - time_uptime); 1269 drl->defrouter[i].if_index = dr->ifp->if_index; 1270 i++; 1271 } 1272 break; 1273 case SIOCGPRLST_IN6: 1274 /* 1275 * obsolete API, use sysctl under net.inet6.icmp6 1276 * 1277 * XXX the structure in6_prlist was changed in backward- 1278 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6, 1279 * in6_prlist is used for nd6_sysctl() - fill_prlist(). 1280 */ 1281 /* 1282 * XXX meaning of fields, especialy "raflags", is very 1283 * differnet between RA prefix list and RR/static prefix list. 1284 * how about separating ioctls into two? 1285 */ 1286 bzero(oprl, sizeof(*oprl)); 1287 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 1288 struct nd_pfxrouter *pfr; 1289 int j; 1290 1291 if (i >= PRLSTSIZ) 1292 break; 1293 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr; 1294 oprl->prefix[i].raflags = pr->ndpr_raf; 1295 oprl->prefix[i].prefixlen = pr->ndpr_plen; 1296 oprl->prefix[i].vltime = pr->ndpr_vltime; 1297 oprl->prefix[i].pltime = pr->ndpr_pltime; 1298 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index; 1299 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 1300 oprl->prefix[i].expire = 0; 1301 else { 1302 time_t maxexpire; 1303 1304 /* XXX: we assume time_t is signed. */ 1305 maxexpire = (-1) & 1306 ~((time_t)1 << 1307 ((sizeof(maxexpire) * 8) - 1)); 1308 if (pr->ndpr_vltime < 1309 maxexpire - pr->ndpr_lastupdate) { 1310 oprl->prefix[i].expire = 1311 pr->ndpr_lastupdate + 1312 pr->ndpr_vltime + 1313 (time_second - time_uptime); 1314 } else 1315 oprl->prefix[i].expire = maxexpire; 1316 } 1317 1318 j = 0; 1319 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 1320 if (j < DRLSTSIZ) { 1321#define RTRADDR oprl->prefix[i].advrtr[j] 1322 RTRADDR = pfr->router->rtaddr; 1323 in6_clearscope(&RTRADDR); 1324#undef RTRADDR 1325 } 1326 j++; 1327 } 1328 oprl->prefix[i].advrtrs = j; 1329 oprl->prefix[i].origin = PR_ORIG_RA; 1330 1331 i++; 1332 } 1333 1334 break; 1335 case OSIOCGIFINFO_IN6: 1336#define ND ndi->ndi 1337 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1338 bzero(&ND, sizeof(ND)); 1339 ND.linkmtu = IN6_LINKMTU(ifp); 1340 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1341 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1342 ND.reachable = ND_IFINFO(ifp)->reachable; 1343 ND.retrans = ND_IFINFO(ifp)->retrans; 1344 ND.flags = ND_IFINFO(ifp)->flags; 1345 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1346 ND.chlim = ND_IFINFO(ifp)->chlim; 1347 break; 1348 case SIOCGIFINFO_IN6: 1349 ND = *ND_IFINFO(ifp); 1350 break; 1351 case SIOCSIFINFO_IN6: 1352 /* 1353 * used to change host variables from userland. 1354 * intented for a use on router to reflect RA configurations. 1355 */ 1356 /* 0 means 'unspecified' */ 1357 if (ND.linkmtu != 0) { 1358 if (ND.linkmtu < IPV6_MMTU || 1359 ND.linkmtu > IN6_LINKMTU(ifp)) { 1360 error = EINVAL; 1361 break; 1362 } 1363 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1364 } 1365 1366 if (ND.basereachable != 0) { 1367 int obasereachable = ND_IFINFO(ifp)->basereachable; 1368 1369 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1370 if (ND.basereachable != obasereachable) 1371 ND_IFINFO(ifp)->reachable = 1372 ND_COMPUTE_RTIME(ND.basereachable); 1373 } 1374 if (ND.retrans != 0) 1375 ND_IFINFO(ifp)->retrans = ND.retrans; 1376 if (ND.chlim != 0) 1377 ND_IFINFO(ifp)->chlim = ND.chlim; 1378 /* FALLTHROUGH */ 1379 case SIOCSIFINFO_FLAGS: 1380 { 1381 struct ifaddr *ifa; 1382 struct in6_ifaddr *ia; 1383 1384 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1385 !(ND.flags & ND6_IFF_IFDISABLED)) { 1386 /* ifdisabled 1->0 transision */ 1387 1388 /* 1389 * If the interface is marked as ND6_IFF_IFDISABLED and 1390 * has an link-local address with IN6_IFF_DUPLICATED, 1391 * do not clear ND6_IFF_IFDISABLED. 1392 * See RFC 4862, Section 5.4.5. 1393 */ 1394 int duplicated_linklocal = 0; 1395 1396 IF_ADDR_RLOCK(ifp); 1397 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1398 if (ifa->ifa_addr->sa_family != AF_INET6) 1399 continue; 1400 ia = (struct in6_ifaddr *)ifa; 1401 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) && 1402 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1403 duplicated_linklocal = 1; 1404 break; 1405 } 1406 } 1407 IF_ADDR_RUNLOCK(ifp); 1408 1409 if (duplicated_linklocal) { 1410 ND.flags |= ND6_IFF_IFDISABLED; 1411 log(LOG_ERR, "Cannot enable an interface" 1412 " with a link-local address marked" 1413 " duplicate.\n"); 1414 } else { 1415 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED; 1416 if (ifp->if_flags & IFF_UP) 1417 in6_if_up(ifp); 1418 } 1419 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1420 (ND.flags & ND6_IFF_IFDISABLED)) { 1421 /* ifdisabled 0->1 transision */ 1422 /* Mark all IPv6 address as tentative. */ 1423 1424 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; 1425 IF_ADDR_RLOCK(ifp); 1426 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1427 if (ifa->ifa_addr->sa_family != AF_INET6) 1428 continue; 1429 ia = (struct in6_ifaddr *)ifa; 1430 ia->ia6_flags |= IN6_IFF_TENTATIVE; 1431 } 1432 IF_ADDR_RUNLOCK(ifp); 1433 } 1434 1435 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) { 1436 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) { 1437 /* auto_linklocal 0->1 transision */ 1438 1439 /* If no link-local address on ifp, configure */ 1440 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL; 1441 in6_ifattach(ifp, NULL); 1442 } else if (!(ND.flags & ND6_IFF_IFDISABLED) && 1443 ifp->if_flags & IFF_UP) { 1444 /* 1445 * When the IF already has 1446 * ND6_IFF_AUTO_LINKLOCAL, no link-local 1447 * address is assigned, and IFF_UP, try to 1448 * assign one. 1449 */ 1450 int haslinklocal = 0; 1451 1452 IF_ADDR_RLOCK(ifp); 1453 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1454 if (ifa->ifa_addr->sa_family != AF_INET6) 1455 continue; 1456 ia = (struct in6_ifaddr *)ifa; 1457 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1458 haslinklocal = 1; 1459 break; 1460 } 1461 } 1462 IF_ADDR_RUNLOCK(ifp); 1463 if (!haslinklocal) 1464 in6_ifattach(ifp, NULL); 1465 } 1466 } 1467 } 1468 ND_IFINFO(ifp)->flags = ND.flags; 1469 break; 1470#undef ND 1471 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1472 /* sync kernel routing table with the default router list */ 1473 defrouter_reset(); 1474 defrouter_select(); 1475 break; 1476 case SIOCSPFXFLUSH_IN6: 1477 { 1478 /* flush all the prefix advertised by routers */ 1479 struct nd_prefix *pr, *next; 1480 1481 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) { 1482 struct in6_ifaddr *ia, *ia_next; 1483 1484 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1485 continue; /* XXX */ 1486 1487 /* do we really have to remove addresses as well? */ 1488 /* XXXRW: in6_ifaddrhead locking. */ 1489 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link, 1490 ia_next) { 1491 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1492 continue; 1493 1494 if (ia->ia6_ndpr == pr) 1495 in6_purgeaddr(&ia->ia_ifa); 1496 } 1497 prelist_remove(pr); 1498 } 1499 break; 1500 } 1501 case SIOCSRTRFLUSH_IN6: 1502 { 1503 /* flush all the default routers */ 1504 struct nd_defrouter *dr, *next; 1505 1506 defrouter_reset(); 1507 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) { 1508 defrtrlist_del(dr); 1509 } 1510 defrouter_select(); 1511 break; 1512 } 1513 case SIOCGNBRINFO_IN6: 1514 { 1515 struct llentry *ln; 1516 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1517 1518 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1519 return (error); 1520 1521 IF_AFDATA_RLOCK(ifp); 1522 ln = nd6_lookup(&nb_addr, 0, ifp); 1523 IF_AFDATA_RUNLOCK(ifp); 1524 1525 if (ln == NULL) { 1526 error = EINVAL; 1527 break; 1528 } 1529 nbi->state = ln->ln_state; 1530 nbi->asked = ln->la_asked; 1531 nbi->isrouter = ln->ln_router; 1532 if (ln->la_expire == 0) 1533 nbi->expire = 0; 1534 else 1535 nbi->expire = ln->la_expire + 1536 (time_second - time_uptime); 1537 LLE_RUNLOCK(ln); 1538 break; 1539 } 1540 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1541 ndif->ifindex = V_nd6_defifindex; 1542 break; 1543 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1544 return (nd6_setdefaultiface(ndif->ifindex)); 1545 } 1546 return (error); 1547} 1548 1549/* 1550 * Create neighbor cache entry and cache link-layer address, 1551 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1552 * 1553 * type - ICMP6 type 1554 * code - type dependent information 1555 * 1556 * XXXXX 1557 * The caller of this function already acquired the ndp 1558 * cache table lock because the cache entry is returned. 1559 */ 1560struct llentry * 1561nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, 1562 int lladdrlen, int type, int code) 1563{ 1564 struct llentry *ln = NULL; 1565 int is_newentry; 1566 int do_update; 1567 int olladdr; 1568 int llchange; 1569 int flags; 1570 int newstate = 0; 1571 uint16_t router = 0; 1572 struct sockaddr_in6 sin6; 1573 struct mbuf *chain = NULL; 1574 int static_route = 0; 1575 1576 IF_AFDATA_UNLOCK_ASSERT(ifp); 1577 1578 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__)); 1579 KASSERT(from != NULL, ("%s: from == NULL", __func__)); 1580 1581 /* nothing must be updated for unspecified address */ 1582 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1583 return NULL; 1584 1585 /* 1586 * Validation about ifp->if_addrlen and lladdrlen must be done in 1587 * the caller. 1588 * 1589 * XXX If the link does not have link-layer adderss, what should 1590 * we do? (ifp->if_addrlen == 0) 1591 * Spec says nothing in sections for RA, RS and NA. There's small 1592 * description on it in NS section (RFC 2461 7.2.3). 1593 */ 1594 flags = lladdr ? ND6_EXCLUSIVE : 0; 1595 IF_AFDATA_RLOCK(ifp); 1596 ln = nd6_lookup(from, flags, ifp); 1597 IF_AFDATA_RUNLOCK(ifp); 1598 if (ln == NULL) { 1599 flags |= ND6_EXCLUSIVE; 1600 IF_AFDATA_LOCK(ifp); 1601 ln = nd6_lookup(from, flags | ND6_CREATE, ifp); 1602 IF_AFDATA_UNLOCK(ifp); 1603 is_newentry = 1; 1604 } else { 1605 /* do nothing if static ndp is set */ 1606 if (ln->la_flags & LLE_STATIC) { 1607 static_route = 1; 1608 goto done; 1609 } 1610 is_newentry = 0; 1611 } 1612 if (ln == NULL) 1613 return (NULL); 1614 1615 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0; 1616 if (olladdr && lladdr) { 1617 llchange = bcmp(lladdr, &ln->ll_addr, 1618 ifp->if_addrlen); 1619 } else 1620 llchange = 0; 1621 1622 /* 1623 * newentry olladdr lladdr llchange (*=record) 1624 * 0 n n -- (1) 1625 * 0 y n -- (2) 1626 * 0 n y -- (3) * STALE 1627 * 0 y y n (4) * 1628 * 0 y y y (5) * STALE 1629 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1630 * 1 -- y -- (7) * STALE 1631 */ 1632 1633 if (lladdr) { /* (3-5) and (7) */ 1634 /* 1635 * Record source link-layer address 1636 * XXX is it dependent to ifp->if_type? 1637 */ 1638 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen); 1639 ln->la_flags |= LLE_VALID; 1640 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 1641 } 1642 1643 if (!is_newentry) { 1644 if ((!olladdr && lladdr != NULL) || /* (3) */ 1645 (olladdr && lladdr != NULL && llchange)) { /* (5) */ 1646 do_update = 1; 1647 newstate = ND6_LLINFO_STALE; 1648 } else /* (1-2,4) */ 1649 do_update = 0; 1650 } else { 1651 do_update = 1; 1652 if (lladdr == NULL) /* (6) */ 1653 newstate = ND6_LLINFO_NOSTATE; 1654 else /* (7) */ 1655 newstate = ND6_LLINFO_STALE; 1656 } 1657 1658 if (do_update) { 1659 /* 1660 * Update the state of the neighbor cache. 1661 */ 1662 ln->ln_state = newstate; 1663 1664 if (ln->ln_state == ND6_LLINFO_STALE) { 1665 /* 1666 * XXX: since nd6_output() below will cause 1667 * state tansition to DELAY and reset the timer, 1668 * we must set the timer now, although it is actually 1669 * meaningless. 1670 */ 1671 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 1672 1673 if (ln->la_hold) { 1674 struct mbuf *m_hold, *m_hold_next; 1675 1676 /* 1677 * reset the la_hold in advance, to explicitly 1678 * prevent a la_hold lookup in nd6_output() 1679 * (wouldn't happen, though...) 1680 */ 1681 for (m_hold = ln->la_hold, ln->la_hold = NULL; 1682 m_hold; m_hold = m_hold_next) { 1683 m_hold_next = m_hold->m_nextpkt; 1684 m_hold->m_nextpkt = NULL; 1685 1686 /* 1687 * we assume ifp is not a p2p here, so 1688 * just set the 2nd argument as the 1689 * 1st one. 1690 */ 1691 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain); 1692 } 1693 /* 1694 * If we have mbufs in the chain we need to do 1695 * deferred transmit. Copy the address from the 1696 * llentry before dropping the lock down below. 1697 */ 1698 if (chain != NULL) 1699 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6)); 1700 } 1701 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { 1702 /* probe right away */ 1703 nd6_llinfo_settimer_locked((void *)ln, 0); 1704 } 1705 } 1706 1707 /* 1708 * ICMP6 type dependent behavior. 1709 * 1710 * NS: clear IsRouter if new entry 1711 * RS: clear IsRouter 1712 * RA: set IsRouter if there's lladdr 1713 * redir: clear IsRouter if new entry 1714 * 1715 * RA case, (1): 1716 * The spec says that we must set IsRouter in the following cases: 1717 * - If lladdr exist, set IsRouter. This means (1-5). 1718 * - If it is old entry (!newentry), set IsRouter. This means (7). 1719 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1720 * A quetion arises for (1) case. (1) case has no lladdr in the 1721 * neighbor cache, this is similar to (6). 1722 * This case is rare but we figured that we MUST NOT set IsRouter. 1723 * 1724 * newentry olladdr lladdr llchange NS RS RA redir 1725 * D R 1726 * 0 n n -- (1) c ? s 1727 * 0 y n -- (2) c s s 1728 * 0 n y -- (3) c s s 1729 * 0 y y n (4) c s s 1730 * 0 y y y (5) c s s 1731 * 1 -- n -- (6) c c c s 1732 * 1 -- y -- (7) c c s c s 1733 * 1734 * (c=clear s=set) 1735 */ 1736 switch (type & 0xff) { 1737 case ND_NEIGHBOR_SOLICIT: 1738 /* 1739 * New entry must have is_router flag cleared. 1740 */ 1741 if (is_newentry) /* (6-7) */ 1742 ln->ln_router = 0; 1743 break; 1744 case ND_REDIRECT: 1745 /* 1746 * If the icmp is a redirect to a better router, always set the 1747 * is_router flag. Otherwise, if the entry is newly created, 1748 * clear the flag. [RFC 2461, sec 8.3] 1749 */ 1750 if (code == ND_REDIRECT_ROUTER) 1751 ln->ln_router = 1; 1752 else if (is_newentry) /* (6-7) */ 1753 ln->ln_router = 0; 1754 break; 1755 case ND_ROUTER_SOLICIT: 1756 /* 1757 * is_router flag must always be cleared. 1758 */ 1759 ln->ln_router = 0; 1760 break; 1761 case ND_ROUTER_ADVERT: 1762 /* 1763 * Mark an entry with lladdr as a router. 1764 */ 1765 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */ 1766 (is_newentry && lladdr)) { /* (7) */ 1767 ln->ln_router = 1; 1768 } 1769 break; 1770 } 1771 1772 if (ln != NULL) { 1773 static_route = (ln->la_flags & LLE_STATIC); 1774 router = ln->ln_router; 1775 1776 if (flags & ND6_EXCLUSIVE) 1777 LLE_WUNLOCK(ln); 1778 else 1779 LLE_RUNLOCK(ln); 1780 if (static_route) 1781 ln = NULL; 1782 } 1783 if (chain) 1784 nd6_output_flush(ifp, ifp, chain, &sin6, NULL); 1785 1786 /* 1787 * When the link-layer address of a router changes, select the 1788 * best router again. In particular, when the neighbor entry is newly 1789 * created, it might affect the selection policy. 1790 * Question: can we restrict the first condition to the "is_newentry" 1791 * case? 1792 * XXX: when we hear an RA from a new router with the link-layer 1793 * address option, defrouter_select() is called twice, since 1794 * defrtrlist_update called the function as well. However, I believe 1795 * we can compromise the overhead, since it only happens the first 1796 * time. 1797 * XXX: although defrouter_select() should not have a bad effect 1798 * for those are not autoconfigured hosts, we explicitly avoid such 1799 * cases for safety. 1800 */ 1801 if (do_update && router && 1802 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1803 /* 1804 * guaranteed recursion 1805 */ 1806 defrouter_select(); 1807 } 1808 1809 return (ln); 1810done: 1811 if (ln != NULL) { 1812 if (flags & ND6_EXCLUSIVE) 1813 LLE_WUNLOCK(ln); 1814 else 1815 LLE_RUNLOCK(ln); 1816 if (static_route) 1817 ln = NULL; 1818 } 1819 return (ln); 1820} 1821 1822static void 1823nd6_slowtimo(void *arg) 1824{ 1825 CURVNET_SET((struct vnet *) arg); 1826 struct nd_ifinfo *nd6if; 1827 struct ifnet *ifp; 1828 1829 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 1830 nd6_slowtimo, curvnet); 1831 IFNET_RLOCK_NOSLEEP(); 1832 TAILQ_FOREACH(ifp, &V_ifnet, if_list) { 1833 if (ifp->if_afdata[AF_INET6] == NULL) 1834 continue; 1835 nd6if = ND_IFINFO(ifp); 1836 if (nd6if->basereachable && /* already initialized */ 1837 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 1838 /* 1839 * Since reachable time rarely changes by router 1840 * advertisements, we SHOULD insure that a new random 1841 * value gets recomputed at least once every few hours. 1842 * (RFC 2461, 6.3.4) 1843 */ 1844 nd6if->recalctm = V_nd6_recalc_reachtm_interval; 1845 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 1846 } 1847 } 1848 IFNET_RUNLOCK_NOSLEEP(); 1849 CURVNET_RESTORE(); 1850} 1851 1852int 1853nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 1854 struct sockaddr_in6 *dst, struct rtentry *rt0) 1855{ 1856 1857 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL)); 1858} 1859 1860 1861/* 1862 * Note that I'm not enforcing any global serialization 1863 * lle state or asked changes here as the logic is too 1864 * complicated to avoid having to always acquire an exclusive 1865 * lock 1866 * KMM 1867 * 1868 */ 1869#define senderr(e) { error = (e); goto bad;} 1870 1871int 1872nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 1873 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle, 1874 struct mbuf **chain) 1875{ 1876 struct mbuf *m = m0; 1877 struct m_tag *mtag; 1878 struct llentry *ln = lle; 1879 struct ip6_hdr *ip6; 1880 int error = 0; 1881 int flags = 0; 1882 int ip6len; 1883 1884#ifdef INVARIANTS 1885 if (lle != NULL) { 1886 1887 LLE_WLOCK_ASSERT(lle); 1888 1889 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed")); 1890 } 1891#endif 1892 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) 1893 goto sendpkt; 1894 1895 if (nd6_need_cache(ifp) == 0) 1896 goto sendpkt; 1897 1898 /* 1899 * next hop determination. This routine is derived from ether_output. 1900 */ 1901 1902 /* 1903 * Address resolution or Neighbor Unreachability Detection 1904 * for the next hop. 1905 * At this point, the destination of the packet must be a unicast 1906 * or an anycast address(i.e. not a multicast). 1907 */ 1908 1909 flags = (lle != NULL) ? LLE_EXCLUSIVE : 0; 1910 if (ln == NULL) { 1911 retry: 1912 IF_AFDATA_RLOCK(ifp); 1913 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst); 1914 IF_AFDATA_RUNLOCK(ifp); 1915 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) { 1916 /* 1917 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 1918 * the condition below is not very efficient. But we believe 1919 * it is tolerable, because this should be a rare case. 1920 */ 1921 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0); 1922 IF_AFDATA_LOCK(ifp); 1923 ln = nd6_lookup(&dst->sin6_addr, flags, ifp); 1924 IF_AFDATA_UNLOCK(ifp); 1925 } 1926 } 1927 if (ln == NULL) { 1928 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && 1929 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 1930 char ip6buf[INET6_ADDRSTRLEN]; 1931 log(LOG_DEBUG, 1932 "nd6_output: can't allocate llinfo for %s " 1933 "(ln=%p)\n", 1934 ip6_sprintf(ip6buf, &dst->sin6_addr), ln); 1935 senderr(EIO); /* XXX: good error? */ 1936 } 1937 goto sendpkt; /* send anyway */ 1938 } 1939 1940 /* We don't have to do link-layer address resolution on a p2p link. */ 1941 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && 1942 ln->ln_state < ND6_LLINFO_REACHABLE) { 1943 if ((flags & LLE_EXCLUSIVE) == 0) { 1944 flags |= LLE_EXCLUSIVE; 1945 LLE_RUNLOCK(ln); 1946 goto retry; 1947 } 1948 ln->ln_state = ND6_LLINFO_STALE; 1949 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 1950 } 1951 1952 /* 1953 * The first time we send a packet to a neighbor whose entry is 1954 * STALE, we have to change the state to DELAY and a sets a timer to 1955 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 1956 * neighbor unreachability detection on expiration. 1957 * (RFC 2461 7.3.3) 1958 */ 1959 if (ln->ln_state == ND6_LLINFO_STALE) { 1960 if ((flags & LLE_EXCLUSIVE) == 0) { 1961 flags |= LLE_EXCLUSIVE; 1962 LLE_RUNLOCK(ln); 1963 goto retry; 1964 } 1965 ln->la_asked = 0; 1966 ln->ln_state = ND6_LLINFO_DELAY; 1967 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz); 1968 } 1969 1970 /* 1971 * If the neighbor cache entry has a state other than INCOMPLETE 1972 * (i.e. its link-layer address is already resolved), just 1973 * send the packet. 1974 */ 1975 if (ln->ln_state > ND6_LLINFO_INCOMPLETE) 1976 goto sendpkt; 1977 1978 /* 1979 * There is a neighbor cache entry, but no ethernet address 1980 * response yet. Append this latest packet to the end of the 1981 * packet queue in the mbuf, unless the number of the packet 1982 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 1983 * the oldest packet in the queue will be removed. 1984 */ 1985 if (ln->ln_state == ND6_LLINFO_NOSTATE) 1986 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1987 1988 if ((flags & LLE_EXCLUSIVE) == 0) { 1989 flags |= LLE_EXCLUSIVE; 1990 LLE_RUNLOCK(ln); 1991 goto retry; 1992 } 1993 1994 LLE_WLOCK_ASSERT(ln); 1995 1996 if (ln->la_hold) { 1997 struct mbuf *m_hold; 1998 int i; 1999 2000 i = 0; 2001 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) { 2002 i++; 2003 if (m_hold->m_nextpkt == NULL) { 2004 m_hold->m_nextpkt = m; 2005 break; 2006 } 2007 } 2008 while (i >= V_nd6_maxqueuelen) { 2009 m_hold = ln->la_hold; 2010 ln->la_hold = ln->la_hold->m_nextpkt; 2011 m_freem(m_hold); 2012 i--; 2013 } 2014 } else { 2015 ln->la_hold = m; 2016 } 2017 2018 /* 2019 * If there has been no NS for the neighbor after entering the 2020 * INCOMPLETE state, send the first solicitation. 2021 */ 2022 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) { 2023 ln->la_asked++; 2024 2025 nd6_llinfo_settimer_locked(ln, 2026 (long)ND_IFINFO(ifp)->retrans * hz / 1000); 2027 LLE_WUNLOCK(ln); 2028 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); 2029 if (lle != NULL && ln == lle) 2030 LLE_WLOCK(lle); 2031 2032 } else if (lle == NULL || ln != lle) { 2033 /* 2034 * We did the lookup (no lle arg) so we 2035 * need to do the unlock here. 2036 */ 2037 LLE_WUNLOCK(ln); 2038 } 2039 2040 return (0); 2041 2042 sendpkt: 2043 /* discard the packet if IPv6 operation is disabled on the interface */ 2044 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 2045 error = ENETDOWN; /* better error? */ 2046 goto bad; 2047 } 2048 /* 2049 * ln is valid and the caller did not pass in 2050 * an llentry 2051 */ 2052 if ((ln != NULL) && (lle == NULL)) { 2053 if (flags & LLE_EXCLUSIVE) 2054 LLE_WUNLOCK(ln); 2055 else 2056 LLE_RUNLOCK(ln); 2057 } 2058 2059#ifdef MAC 2060 mac_netinet6_nd6_send(ifp, m); 2061#endif 2062 2063 /* 2064 * If called from nd6_ns_output() (NS), nd6_na_output() (NA), 2065 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA 2066 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND 2067 * to be diverted to user space. When re-injected into the kernel, 2068 * send_output() will directly dispatch them to the outgoing interface. 2069 */ 2070 if (send_sendso_input_hook != NULL) { 2071 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); 2072 if (mtag != NULL) { 2073 ip6 = mtod(m, struct ip6_hdr *); 2074 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); 2075 /* Use the SEND socket */ 2076 error = send_sendso_input_hook(m, ifp, SND_OUT, 2077 ip6len); 2078 /* -1 == no app on SEND socket */ 2079 if (error == 0 || error != -1) 2080 return (error); 2081 } 2082 } 2083 2084 /* 2085 * We were passed in a pointer to an lle with the lock held 2086 * this means that we can't call if_output as we will 2087 * recurse on the lle lock - so what we do is we create 2088 * a list of mbufs to send and transmit them in the caller 2089 * after the lock is dropped 2090 */ 2091 if (lle != NULL) { 2092 if (*chain == NULL) 2093 *chain = m; 2094 else { 2095 struct mbuf *mb; 2096 2097 /* 2098 * append mbuf to end of deferred chain 2099 */ 2100 mb = *chain; 2101 while (mb->m_nextpkt != NULL) 2102 mb = mb->m_nextpkt; 2103 mb->m_nextpkt = m; 2104 } 2105 return (error); 2106 } 2107 m_clrprotoflags(m); /* Avoid confusing lower layers. */ 2108 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL, 2109 mtod(m, struct ip6_hdr *)); 2110 if ((ifp->if_flags & IFF_LOOPBACK) != 0) { 2111 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst, 2112 NULL)); 2113 } 2114 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL); 2115 return (error); 2116 2117 bad: 2118 /* 2119 * ln is valid and the caller did not pass in 2120 * an llentry 2121 */ 2122 if ((ln != NULL) && (lle == NULL)) { 2123 if (flags & LLE_EXCLUSIVE) 2124 LLE_WUNLOCK(ln); 2125 else 2126 LLE_RUNLOCK(ln); 2127 } 2128 if (m) 2129 m_freem(m); 2130 return (error); 2131} 2132#undef senderr 2133 2134 2135int 2136nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain, 2137 struct sockaddr_in6 *dst, struct route *ro) 2138{ 2139 struct mbuf *m, *m_head; 2140 struct ifnet *outifp; 2141 int error = 0; 2142 2143 m_head = chain; 2144 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2145 outifp = origifp; 2146 else 2147 outifp = ifp; 2148 2149 while (m_head) { 2150 m = m_head; 2151 m_head = m_head->m_nextpkt; 2152 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro); 2153 } 2154 2155 /* 2156 * XXX 2157 * note that intermediate errors are blindly ignored - but this is 2158 * the same convention as used with nd6_output when called by 2159 * nd6_cache_lladdr 2160 */ 2161 return (error); 2162} 2163 2164 2165int 2166nd6_need_cache(struct ifnet *ifp) 2167{ 2168 /* 2169 * XXX: we currently do not make neighbor cache on any interface 2170 * other than ARCnet, Ethernet, FDDI and GIF. 2171 * 2172 * RFC2893 says: 2173 * - unidirectional tunnels needs no ND 2174 */ 2175 switch (ifp->if_type) { 2176 case IFT_ARCNET: 2177 case IFT_ETHER: 2178 case IFT_FDDI: 2179 case IFT_IEEE1394: 2180#ifdef IFT_L2VLAN 2181 case IFT_L2VLAN: 2182#endif 2183#ifdef IFT_IEEE80211 2184 case IFT_IEEE80211: 2185#endif 2186 case IFT_INFINIBAND: 2187 case IFT_BRIDGE: 2188 case IFT_PROPVIRTUAL: 2189 return (1); 2190 default: 2191 return (0); 2192 } 2193} 2194 2195/* 2196 * the callers of this function need to be re-worked to drop 2197 * the lle lock, drop here for now 2198 */ 2199int 2200nd6_storelladdr(struct ifnet *ifp, struct mbuf *m, 2201 const struct sockaddr *dst, u_char *desten, struct llentry **lle) 2202{ 2203 struct llentry *ln; 2204 2205 *lle = NULL; 2206 IF_AFDATA_UNLOCK_ASSERT(ifp); 2207 if (m != NULL && m->m_flags & M_MCAST) { 2208 int i; 2209 2210 switch (ifp->if_type) { 2211 case IFT_ETHER: 2212 case IFT_FDDI: 2213#ifdef IFT_L2VLAN 2214 case IFT_L2VLAN: 2215#endif 2216#ifdef IFT_IEEE80211 2217 case IFT_IEEE80211: 2218#endif 2219 case IFT_BRIDGE: 2220 case IFT_ISO88025: 2221 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr, 2222 desten); 2223 return (0); 2224 case IFT_IEEE1394: 2225 /* 2226 * netbsd can use if_broadcastaddr, but we don't do so 2227 * to reduce # of ifdef. 2228 */ 2229 for (i = 0; i < ifp->if_addrlen; i++) 2230 desten[i] = ~0; 2231 return (0); 2232 case IFT_ARCNET: 2233 *desten = 0; 2234 return (0); 2235 default: 2236 m_freem(m); 2237 return (EAFNOSUPPORT); 2238 } 2239 } 2240 2241 2242 /* 2243 * the entry should have been created in nd6_store_lladdr 2244 */ 2245 IF_AFDATA_RLOCK(ifp); 2246 ln = lla_lookup(LLTABLE6(ifp), 0, dst); 2247 IF_AFDATA_RUNLOCK(ifp); 2248 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) { 2249 if (ln != NULL) 2250 LLE_RUNLOCK(ln); 2251 /* this could happen, if we could not allocate memory */ 2252 m_freem(m); 2253 return (1); 2254 } 2255 2256 bcopy(&ln->ll_addr, desten, ifp->if_addrlen); 2257 *lle = ln; 2258 LLE_RUNLOCK(ln); 2259 /* 2260 * A *small* use after free race exists here 2261 */ 2262 return (0); 2263} 2264 2265static void 2266clear_llinfo_pqueue(struct llentry *ln) 2267{ 2268 struct mbuf *m_hold, *m_hold_next; 2269 2270 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) { 2271 m_hold_next = m_hold->m_nextpkt; 2272 m_freem(m_hold); 2273 } 2274 2275 ln->la_hold = NULL; 2276 return; 2277} 2278 2279static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2280static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2281#ifdef SYSCTL_DECL 2282SYSCTL_DECL(_net_inet6_icmp6); 2283#endif 2284SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2285 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2286SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2287 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2288SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2289 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, ""); 2290SYSCTL_VNET_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer, 2291 CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), ""); 2292 2293static int 2294nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2295{ 2296 struct in6_defrouter d; 2297 struct nd_defrouter *dr; 2298 int error; 2299 2300 if (req->newptr) 2301 return (EPERM); 2302 2303 bzero(&d, sizeof(d)); 2304 d.rtaddr.sin6_family = AF_INET6; 2305 d.rtaddr.sin6_len = sizeof(d.rtaddr); 2306 2307 /* 2308 * XXX locking 2309 */ 2310 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 2311 d.rtaddr.sin6_addr = dr->rtaddr; 2312 error = sa6_recoverscope(&d.rtaddr); 2313 if (error != 0) 2314 return (error); 2315 d.flags = dr->flags; 2316 d.rtlifetime = dr->rtlifetime; 2317 d.expire = dr->expire + (time_second - time_uptime); 2318 d.if_index = dr->ifp->if_index; 2319 error = SYSCTL_OUT(req, &d, sizeof(d)); 2320 if (error != 0) 2321 return (error); 2322 } 2323 return (0); 2324} 2325 2326static int 2327nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2328{ 2329 struct in6_prefix p; 2330 struct sockaddr_in6 s6; 2331 struct nd_prefix *pr; 2332 struct nd_pfxrouter *pfr; 2333 time_t maxexpire; 2334 int error; 2335 char ip6buf[INET6_ADDRSTRLEN]; 2336 2337 if (req->newptr) 2338 return (EPERM); 2339 2340 bzero(&p, sizeof(p)); 2341 p.origin = PR_ORIG_RA; 2342 bzero(&s6, sizeof(s6)); 2343 s6.sin6_family = AF_INET6; 2344 s6.sin6_len = sizeof(s6); 2345 2346 /* 2347 * XXX locking 2348 */ 2349 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 2350 p.prefix = pr->ndpr_prefix; 2351 if (sa6_recoverscope(&p.prefix)) { 2352 log(LOG_ERR, "scope error in prefix list (%s)\n", 2353 ip6_sprintf(ip6buf, &p.prefix.sin6_addr)); 2354 /* XXX: press on... */ 2355 } 2356 p.raflags = pr->ndpr_raf; 2357 p.prefixlen = pr->ndpr_plen; 2358 p.vltime = pr->ndpr_vltime; 2359 p.pltime = pr->ndpr_pltime; 2360 p.if_index = pr->ndpr_ifp->if_index; 2361 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2362 p.expire = 0; 2363 else { 2364 /* XXX: we assume time_t is signed. */ 2365 maxexpire = (-1) & 2366 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1)); 2367 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate) 2368 p.expire = pr->ndpr_lastupdate + 2369 pr->ndpr_vltime + 2370 (time_second - time_uptime); 2371 else 2372 p.expire = maxexpire; 2373 } 2374 p.refcnt = pr->ndpr_refcnt; 2375 p.flags = pr->ndpr_stateflags; 2376 p.advrtrs = 0; 2377 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) 2378 p.advrtrs++; 2379 error = SYSCTL_OUT(req, &p, sizeof(p)); 2380 if (error != 0) 2381 return (error); 2382 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 2383 s6.sin6_addr = pfr->router->rtaddr; 2384 if (sa6_recoverscope(&s6)) 2385 log(LOG_ERR, 2386 "scope error in prefix list (%s)\n", 2387 ip6_sprintf(ip6buf, &pfr->router->rtaddr)); 2388 error = SYSCTL_OUT(req, &s6, sizeof(s6)); 2389 if (error != 0) 2390 return (error); 2391 } 2392 } 2393 return (0); 2394} 2395