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("$FreeBSD$"); 34 35 #include "opt_inet.h" 36 #include "opt_inet6.h" 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/callout.h> 41 #include <sys/malloc.h> 42 #include <sys/mbuf.h> 43 #include <sys/socket.h> 44 #include <sys/sockio.h> 45 #include <sys/time.h> 46 #include <sys/kernel.h> 47 #include <sys/protosw.h> 48 #include <sys/errno.h> 49 #include <sys/syslog.h> 50 #include <sys/lock.h> 51 #include <sys/rwlock.h> 52 #include <sys/queue.h> 53 #include <sys/sysctl.h> 54 55 #include <net/if.h> 56 #include <net/if_arc.h> 57 #include <net/if_dl.h> 58 #include <net/if_types.h> 59 #include <net/iso88025.h> 60 #include <net/fddi.h> 61 #include <net/route.h> 62 #include <net/vnet.h> 63 64 #include <netinet/in.h> 65 #include <net/if_llatbl.h> 66 #define L3_ADDR_SIN6(le) ((struct sockaddr_in6 *) L3_ADDR(le)) 67 #include <netinet/if_ether.h> 68 #include <netinet6/in6_var.h> 69 #include <netinet/ip6.h> 70 #include <netinet6/ip6_var.h> 71 #include <netinet6/scope6_var.h> 72 #include <netinet6/nd6.h> 73 #include <netinet6/in6_ifattach.h> 74 #include <netinet/icmp6.h> 75 #include <netinet6/send.h> 76 77 #include <sys/limits.h> 78 79 #include <security/mac/mac_framework.h> 80 81 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ 82 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ 83 84 #define SIN6(s) ((struct sockaddr_in6 *)s) 85 86 /* timer values */ 87 VNET_DEFINE(int, nd6_prune) = 1; /* walk list every 1 seconds */ 88 VNET_DEFINE(int, nd6_delay) = 5; /* delay first probe time 5 second */ 89 VNET_DEFINE(int, nd6_umaxtries) = 3; /* maximum unicast query */ 90 VNET_DEFINE(int, nd6_mmaxtries) = 3; /* maximum multicast query */ 91 VNET_DEFINE(int, nd6_useloopback) = 1; /* use loopback interface for 92 * local traffic */ 93 VNET_DEFINE(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage 94 * collection timer */ 95 96 /* preventing too many loops in ND option parsing */ 97 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */ 98 99 VNET_DEFINE(int, nd6_maxnudhint) = 0; /* max # of subsequent upper 100 * layer hints */ 101 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved 102 * ND entries */ 103 #define V_nd6_maxndopt VNET(nd6_maxndopt) 104 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen) 105 106 #ifdef ND6_DEBUG 107 VNET_DEFINE(int, nd6_debug) = 1; 108 #else 109 VNET_DEFINE(int, nd6_debug) = 0; 110 #endif 111 112 /* for debugging? */ 113 #if 0 114 static int nd6_inuse, nd6_allocated; 115 #endif 116 117 VNET_DEFINE(struct nd_drhead, nd_defrouter); 118 VNET_DEFINE(struct nd_prhead, nd_prefix); 119 120 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL; 121 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval) 122 123 static struct sockaddr_in6 all1_sa; 124 125 int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int); 126 127 static int nd6_is_new_addr_neighbor(struct sockaddr_in6 *, 128 struct ifnet *); 129 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *); 130 static void nd6_slowtimo(void *); 131 static int regen_tmpaddr(struct in6_ifaddr *); 132 static struct llentry *nd6_free(struct llentry *, int); 133 static void nd6_llinfo_timer(void *); 134 static void clear_llinfo_pqueue(struct llentry *); 135 136 static VNET_DEFINE(struct callout, nd6_slowtimo_ch); 137 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch) 138 139 VNET_DEFINE(struct callout, nd6_timer_ch); 140 141 void 142 nd6_init(void) 143 { 144 int i; 145 146 LIST_INIT(&V_nd_prefix); 147 148 all1_sa.sin6_family = AF_INET6; 149 all1_sa.sin6_len = sizeof(struct sockaddr_in6); 150 for (i = 0; i < sizeof(all1_sa.sin6_addr); i++) 151 all1_sa.sin6_addr.s6_addr[i] = 0xff; 152 153 /* initialization of the default router list */ 154 TAILQ_INIT(&V_nd_defrouter); 155 156 /* start timer */ 157 callout_init(&V_nd6_slowtimo_ch, 0); 158 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 159 nd6_slowtimo, curvnet); 160 } 161 162 #ifdef VIMAGE 163 void 164 nd6_destroy() 165 { 166 167 callout_drain(&V_nd6_slowtimo_ch); 168 callout_drain(&V_nd6_timer_ch); 169 } 170 #endif 171 172 struct nd_ifinfo * 173 nd6_ifattach(struct ifnet *ifp) 174 { 175 struct nd_ifinfo *nd; 176 177 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO); 178 nd->initialized = 1; 179 180 nd->chlim = IPV6_DEFHLIM; 181 nd->basereachable = REACHABLE_TIME; 182 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable); 183 nd->retrans = RETRANS_TIMER; 184 185 nd->flags = ND6_IFF_PERFORMNUD; 186 187 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. */ 188 if (V_ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK)) 189 nd->flags |= ND6_IFF_AUTO_LINKLOCAL; 190 191 /* A loopback interface does not need to accept RTADV. */ 192 if (V_ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK)) 193 nd->flags |= ND6_IFF_ACCEPT_RTADV; 194 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK)) 195 nd->flags |= ND6_IFF_NO_RADR; 196 197 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */ 198 nd6_setmtu0(ifp, nd); 199 200 return nd; 201 } 202 203 void 204 nd6_ifdetach(struct nd_ifinfo *nd) 205 { 206 207 free(nd, M_IP6NDP); 208 } 209 210 /* 211 * Reset ND level link MTU. This function is called when the physical MTU 212 * changes, which means we might have to adjust the ND level MTU. 213 */ 214 void 215 nd6_setmtu(struct ifnet *ifp) 216 { 217 218 nd6_setmtu0(ifp, ND_IFINFO(ifp)); 219 } 220 221 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */ 222 void 223 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi) 224 { 225 u_int32_t omaxmtu; 226 227 omaxmtu = ndi->maxmtu; 228 229 switch (ifp->if_type) { 230 case IFT_ARCNET: 231 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */ 232 break; 233 case IFT_FDDI: 234 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */ 235 break; 236 case IFT_ISO88025: 237 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu); 238 break; 239 default: 240 ndi->maxmtu = ifp->if_mtu; 241 break; 242 } 243 244 /* 245 * Decreasing the interface MTU under IPV6 minimum MTU may cause 246 * undesirable situation. We thus notify the operator of the change 247 * explicitly. The check for omaxmtu is necessary to restrict the 248 * log to the case of changing the MTU, not initializing it. 249 */ 250 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { 251 log(LOG_NOTICE, "nd6_setmtu0: " 252 "new link MTU on %s (%lu) is too small for IPv6\n", 253 if_name(ifp), (unsigned long)ndi->maxmtu); 254 } 255 256 if (ndi->maxmtu > V_in6_maxmtu) 257 in6_setmaxmtu(); /* check all interfaces just in case */ 258 259 } 260 261 void 262 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts) 263 { 264 265 bzero(ndopts, sizeof(*ndopts)); 266 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; 267 ndopts->nd_opts_last 268 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); 269 270 if (icmp6len == 0) { 271 ndopts->nd_opts_done = 1; 272 ndopts->nd_opts_search = NULL; 273 } 274 } 275 276 /* 277 * Take one ND option. 278 */ 279 struct nd_opt_hdr * 280 nd6_option(union nd_opts *ndopts) 281 { 282 struct nd_opt_hdr *nd_opt; 283 int olen; 284 285 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 286 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 287 __func__)); 288 if (ndopts->nd_opts_search == NULL) 289 return NULL; 290 if (ndopts->nd_opts_done) 291 return NULL; 292 293 nd_opt = ndopts->nd_opts_search; 294 295 /* make sure nd_opt_len is inside the buffer */ 296 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) { 297 bzero(ndopts, sizeof(*ndopts)); 298 return NULL; 299 } 300 301 olen = nd_opt->nd_opt_len << 3; 302 if (olen == 0) { 303 /* 304 * Message validation requires that all included 305 * options have a length that is greater than zero. 306 */ 307 bzero(ndopts, sizeof(*ndopts)); 308 return NULL; 309 } 310 311 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen); 312 if (ndopts->nd_opts_search > ndopts->nd_opts_last) { 313 /* option overruns the end of buffer, invalid */ 314 bzero(ndopts, sizeof(*ndopts)); 315 return NULL; 316 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { 317 /* reached the end of options chain */ 318 ndopts->nd_opts_done = 1; 319 ndopts->nd_opts_search = NULL; 320 } 321 return nd_opt; 322 } 323 324 /* 325 * Parse multiple ND options. 326 * This function is much easier to use, for ND routines that do not need 327 * multiple options of the same type. 328 */ 329 int 330 nd6_options(union nd_opts *ndopts) 331 { 332 struct nd_opt_hdr *nd_opt; 333 int i = 0; 334 335 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 336 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 337 __func__)); 338 if (ndopts->nd_opts_search == NULL) 339 return 0; 340 341 while (1) { 342 nd_opt = nd6_option(ndopts); 343 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { 344 /* 345 * Message validation requires that all included 346 * options have a length that is greater than zero. 347 */ 348 ICMP6STAT_INC(icp6s_nd_badopt); 349 bzero(ndopts, sizeof(*ndopts)); 350 return -1; 351 } 352 353 if (nd_opt == NULL) 354 goto skip1; 355 356 switch (nd_opt->nd_opt_type) { 357 case ND_OPT_SOURCE_LINKADDR: 358 case ND_OPT_TARGET_LINKADDR: 359 case ND_OPT_MTU: 360 case ND_OPT_REDIRECTED_HEADER: 361 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { 362 nd6log((LOG_INFO, 363 "duplicated ND6 option found (type=%d)\n", 364 nd_opt->nd_opt_type)); 365 /* XXX bark? */ 366 } else { 367 ndopts->nd_opt_array[nd_opt->nd_opt_type] 368 = nd_opt; 369 } 370 break; 371 case ND_OPT_PREFIX_INFORMATION: 372 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { 373 ndopts->nd_opt_array[nd_opt->nd_opt_type] 374 = nd_opt; 375 } 376 ndopts->nd_opts_pi_end = 377 (struct nd_opt_prefix_info *)nd_opt; 378 break; 379 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */ 380 case ND_OPT_RDNSS: /* RFC 6106 */ 381 case ND_OPT_DNSSL: /* RFC 6106 */ 382 /* 383 * Silently ignore options we know and do not care about 384 * in the kernel. 385 */ 386 break; 387 default: 388 /* 389 * Unknown options must be silently ignored, 390 * to accomodate future extension to the protocol. 391 */ 392 nd6log((LOG_DEBUG, 393 "nd6_options: unsupported option %d - " 394 "option ignored\n", nd_opt->nd_opt_type)); 395 } 396 397 skip1: 398 i++; 399 if (i > V_nd6_maxndopt) { 400 ICMP6STAT_INC(icp6s_nd_toomanyopt); 401 nd6log((LOG_INFO, "too many loop in nd opt\n")); 402 break; 403 } 404 405 if (ndopts->nd_opts_done) 406 break; 407 } 408 409 return 0; 410 } 411 412 /* 413 * ND6 timer routine to handle ND6 entries 414 */ 415 void 416 nd6_llinfo_settimer_locked(struct llentry *ln, long tick) 417 { 418 int canceled; 419 420 LLE_WLOCK_ASSERT(ln); 421 422 if (tick < 0) { 423 ln->la_expire = 0; 424 ln->ln_ntick = 0; 425 canceled = callout_stop(&ln->ln_timer_ch); 426 } else { 427 ln->la_expire = time_second + tick / hz; 428 LLE_ADDREF(ln); 429 if (tick > INT_MAX) { 430 ln->ln_ntick = tick - INT_MAX; 431 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX, 432 nd6_llinfo_timer, ln); 433 } else { 434 ln->ln_ntick = 0; 435 canceled = callout_reset(&ln->ln_timer_ch, tick, 436 nd6_llinfo_timer, ln); 437 } 438 } 439 if (canceled) 440 LLE_REMREF(ln); 441 } 442 443 void 444 nd6_llinfo_settimer(struct llentry *ln, long tick) 445 { 446 447 LLE_WLOCK(ln); 448 nd6_llinfo_settimer_locked(ln, tick); 449 LLE_WUNLOCK(ln); 450 } 451 452 static void 453 nd6_llinfo_timer(void *arg) 454 { 455 struct llentry *ln; 456 struct in6_addr *dst; 457 struct ifnet *ifp; 458 struct nd_ifinfo *ndi = NULL; 459 460 KASSERT(arg != NULL, ("%s: arg NULL", __func__)); 461 ln = (struct llentry *)arg; 462 LLE_WLOCK_ASSERT(ln); 463 ifp = ln->lle_tbl->llt_ifp; 464 465 CURVNET_SET(ifp->if_vnet); 466 467 if (ln->ln_ntick > 0) { 468 if (ln->ln_ntick > INT_MAX) { 469 ln->ln_ntick -= INT_MAX; 470 nd6_llinfo_settimer_locked(ln, INT_MAX); 471 } else { 472 ln->ln_ntick = 0; 473 nd6_llinfo_settimer_locked(ln, ln->ln_ntick); 474 } 475 goto done; 476 } 477 478 ndi = ND_IFINFO(ifp); 479 dst = &L3_ADDR_SIN6(ln)->sin6_addr; 480 if (ln->la_flags & LLE_STATIC) { 481 goto done; 482 } 483 484 if (ln->la_flags & LLE_DELETED) { 485 (void)nd6_free(ln, 0); 486 ln = NULL; 487 goto done; 488 } 489 490 switch (ln->ln_state) { 491 case ND6_LLINFO_INCOMPLETE: 492 if (ln->la_asked < V_nd6_mmaxtries) { 493 ln->la_asked++; 494 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 495 LLE_WUNLOCK(ln); 496 nd6_ns_output(ifp, NULL, dst, ln, 0); 497 LLE_WLOCK(ln); 498 } else { 499 struct mbuf *m = ln->la_hold; 500 if (m) { 501 struct mbuf *m0; 502 503 /* 504 * assuming every packet in la_hold has the 505 * same IP header. Send error after unlock. 506 */ 507 m0 = m->m_nextpkt; 508 m->m_nextpkt = NULL; 509 ln->la_hold = m0; 510 clear_llinfo_pqueue(ln); 511 } 512 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT); 513 (void)nd6_free(ln, 0); 514 ln = NULL; 515 if (m != NULL) 516 icmp6_error2(m, ICMP6_DST_UNREACH, 517 ICMP6_DST_UNREACH_ADDR, 0, ifp); 518 } 519 break; 520 case ND6_LLINFO_REACHABLE: 521 if (!ND6_LLINFO_PERMANENT(ln)) { 522 ln->ln_state = ND6_LLINFO_STALE; 523 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 524 } 525 break; 526 527 case ND6_LLINFO_STALE: 528 /* Garbage Collection(RFC 2461 5.3) */ 529 if (!ND6_LLINFO_PERMANENT(ln)) { 530 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 531 (void)nd6_free(ln, 1); 532 ln = NULL; 533 } 534 break; 535 536 case ND6_LLINFO_DELAY: 537 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 538 /* We need NUD */ 539 ln->la_asked = 1; 540 ln->ln_state = ND6_LLINFO_PROBE; 541 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 542 LLE_WUNLOCK(ln); 543 nd6_ns_output(ifp, dst, dst, ln, 0); 544 LLE_WLOCK(ln); 545 } else { 546 ln->ln_state = ND6_LLINFO_STALE; /* XXX */ 547 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 548 } 549 break; 550 case ND6_LLINFO_PROBE: 551 if (ln->la_asked < V_nd6_umaxtries) { 552 ln->la_asked++; 553 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 554 LLE_WUNLOCK(ln); 555 nd6_ns_output(ifp, dst, dst, ln, 0); 556 LLE_WLOCK(ln); 557 } else { 558 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 559 (void)nd6_free(ln, 0); 560 ln = NULL; 561 } 562 break; 563 default: 564 panic("%s: paths in a dark night can be confusing: %d", 565 __func__, ln->ln_state); 566 } 567 done: 568 if (ln != NULL) 569 LLE_FREE_LOCKED(ln); 570 CURVNET_RESTORE(); 571 } 572 573 574 /* 575 * ND6 timer routine to expire default route list and prefix list 576 */ 577 void 578 nd6_timer(void *arg) 579 { 580 CURVNET_SET((struct vnet *) arg); 581 struct nd_defrouter *dr, *ndr; 582 struct nd_prefix *pr, *npr; 583 struct in6_ifaddr *ia6, *nia6; 584 585 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz, 586 nd6_timer, curvnet); 587 588 /* expire default router list */ 589 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 590 if (dr->expire && dr->expire < time_second) 591 defrtrlist_del(dr); 592 } 593 594 /* 595 * expire interface addresses. 596 * in the past the loop was inside prefix expiry processing. 597 * However, from a stricter speci-confrmance standpoint, we should 598 * rather separate address lifetimes and prefix lifetimes. 599 * 600 * XXXRW: in6_ifaddrhead locking. 601 */ 602 addrloop: 603 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) { 604 /* check address lifetime */ 605 if (IFA6_IS_INVALID(ia6)) { 606 int regen = 0; 607 608 /* 609 * If the expiring address is temporary, try 610 * regenerating a new one. This would be useful when 611 * we suspended a laptop PC, then turned it on after a 612 * period that could invalidate all temporary 613 * addresses. Although we may have to restart the 614 * loop (see below), it must be after purging the 615 * address. Otherwise, we'd see an infinite loop of 616 * regeneration. 617 */ 618 if (V_ip6_use_tempaddr && 619 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 620 if (regen_tmpaddr(ia6) == 0) 621 regen = 1; 622 } 623 624 in6_purgeaddr(&ia6->ia_ifa); 625 626 if (regen) 627 goto addrloop; /* XXX: see below */ 628 } else if (IFA6_IS_DEPRECATED(ia6)) { 629 int oldflags = ia6->ia6_flags; 630 631 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 632 633 /* 634 * If a temporary address has just become deprecated, 635 * regenerate a new one if possible. 636 */ 637 if (V_ip6_use_tempaddr && 638 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 639 (oldflags & IN6_IFF_DEPRECATED) == 0) { 640 641 if (regen_tmpaddr(ia6) == 0) { 642 /* 643 * A new temporary address is 644 * generated. 645 * XXX: this means the address chain 646 * has changed while we are still in 647 * the loop. Although the change 648 * would not cause disaster (because 649 * it's not a deletion, but an 650 * addition,) we'd rather restart the 651 * loop just for safety. Or does this 652 * significantly reduce performance?? 653 */ 654 goto addrloop; 655 } 656 } 657 } else { 658 /* 659 * A new RA might have made a deprecated address 660 * preferred. 661 */ 662 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 663 } 664 } 665 666 /* expire prefix list */ 667 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 668 /* 669 * check prefix lifetime. 670 * since pltime is just for autoconf, pltime processing for 671 * prefix is not necessary. 672 */ 673 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 674 time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) { 675 676 /* 677 * address expiration and prefix expiration are 678 * separate. NEVER perform in6_purgeaddr here. 679 */ 680 prelist_remove(pr); 681 } 682 } 683 CURVNET_RESTORE(); 684 } 685 686 /* 687 * ia6 - deprecated/invalidated temporary address 688 */ 689 static int 690 regen_tmpaddr(struct in6_ifaddr *ia6) 691 { 692 struct ifaddr *ifa; 693 struct ifnet *ifp; 694 struct in6_ifaddr *public_ifa6 = NULL; 695 696 ifp = ia6->ia_ifa.ifa_ifp; 697 IF_ADDR_RLOCK(ifp); 698 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 699 struct in6_ifaddr *it6; 700 701 if (ifa->ifa_addr->sa_family != AF_INET6) 702 continue; 703 704 it6 = (struct in6_ifaddr *)ifa; 705 706 /* ignore no autoconf addresses. */ 707 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 708 continue; 709 710 /* ignore autoconf addresses with different prefixes. */ 711 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 712 continue; 713 714 /* 715 * Now we are looking at an autoconf address with the same 716 * prefix as ours. If the address is temporary and is still 717 * preferred, do not create another one. It would be rare, but 718 * could happen, for example, when we resume a laptop PC after 719 * a long period. 720 */ 721 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 722 !IFA6_IS_DEPRECATED(it6)) { 723 public_ifa6 = NULL; 724 break; 725 } 726 727 /* 728 * This is a public autoconf address that has the same prefix 729 * as ours. If it is preferred, keep it. We can't break the 730 * loop here, because there may be a still-preferred temporary 731 * address with the prefix. 732 */ 733 if (!IFA6_IS_DEPRECATED(it6)) 734 public_ifa6 = it6; 735 736 if (public_ifa6 != NULL) 737 ifa_ref(&public_ifa6->ia_ifa); 738 } 739 IF_ADDR_RUNLOCK(ifp); 740 741 if (public_ifa6 != NULL) { 742 int e; 743 744 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 745 ifa_free(&public_ifa6->ia_ifa); 746 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 747 " tmp addr,errno=%d\n", e); 748 return (-1); 749 } 750 ifa_free(&public_ifa6->ia_ifa); 751 return (0); 752 } 753 754 return (-1); 755 } 756 757 /* 758 * Nuke neighbor cache/prefix/default router management table, right before 759 * ifp goes away. 760 */ 761 void 762 nd6_purge(struct ifnet *ifp) 763 { 764 struct nd_defrouter *dr, *ndr; 765 struct nd_prefix *pr, *npr; 766 767 /* 768 * Nuke default router list entries toward ifp. 769 * We defer removal of default router list entries that is installed 770 * in the routing table, in order to keep additional side effects as 771 * small as possible. 772 */ 773 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 774 if (dr->installed) 775 continue; 776 777 if (dr->ifp == ifp) 778 defrtrlist_del(dr); 779 } 780 781 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 782 if (!dr->installed) 783 continue; 784 785 if (dr->ifp == ifp) 786 defrtrlist_del(dr); 787 } 788 789 /* Nuke prefix list entries toward ifp */ 790 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 791 if (pr->ndpr_ifp == ifp) { 792 /* 793 * Because if_detach() does *not* release prefixes 794 * while purging addresses the reference count will 795 * still be above zero. We therefore reset it to 796 * make sure that the prefix really gets purged. 797 */ 798 pr->ndpr_refcnt = 0; 799 800 /* 801 * Previously, pr->ndpr_addr is removed as well, 802 * but I strongly believe we don't have to do it. 803 * nd6_purge() is only called from in6_ifdetach(), 804 * which removes all the associated interface addresses 805 * by itself. 806 * (jinmei@kame.net 20010129) 807 */ 808 prelist_remove(pr); 809 } 810 } 811 812 /* cancel default outgoing interface setting */ 813 if (V_nd6_defifindex == ifp->if_index) 814 nd6_setdefaultiface(0); 815 816 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 817 /* Refresh default router list. */ 818 defrouter_select(); 819 } 820 821 /* XXXXX 822 * We do not nuke the neighbor cache entries here any more 823 * because the neighbor cache is kept in if_afdata[AF_INET6]. 824 * nd6_purge() is invoked by in6_ifdetach() which is called 825 * from if_detach() where everything gets purged. So let 826 * in6_domifdetach() do the actual L2 table purging work. 827 */ 828 } 829 830 /* 831 * the caller acquires and releases the lock on the lltbls 832 * Returns the llentry locked 833 */ 834 struct llentry * 835 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp) 836 { 837 struct sockaddr_in6 sin6; 838 struct llentry *ln; 839 int llflags; 840 841 bzero(&sin6, sizeof(sin6)); 842 sin6.sin6_len = sizeof(struct sockaddr_in6); 843 sin6.sin6_family = AF_INET6; 844 sin6.sin6_addr = *addr6; 845 846 IF_AFDATA_LOCK_ASSERT(ifp); 847 848 llflags = 0; 849 if (flags & ND6_CREATE) 850 llflags |= LLE_CREATE; 851 if (flags & ND6_EXCLUSIVE) 852 llflags |= LLE_EXCLUSIVE; 853 854 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6); 855 if ((ln != NULL) && (llflags & LLE_CREATE)) 856 ln->ln_state = ND6_LLINFO_NOSTATE; 857 858 return (ln); 859 } 860 861 /* 862 * Test whether a given IPv6 address is a neighbor or not, ignoring 863 * the actual neighbor cache. The neighbor cache is ignored in order 864 * to not reenter the routing code from within itself. 865 */ 866 static int 867 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 868 { 869 struct nd_prefix *pr; 870 struct ifaddr *dstaddr; 871 872 /* 873 * A link-local address is always a neighbor. 874 * XXX: a link does not necessarily specify a single interface. 875 */ 876 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 877 struct sockaddr_in6 sin6_copy; 878 u_int32_t zone; 879 880 /* 881 * We need sin6_copy since sa6_recoverscope() may modify the 882 * content (XXX). 883 */ 884 sin6_copy = *addr; 885 if (sa6_recoverscope(&sin6_copy)) 886 return (0); /* XXX: should be impossible */ 887 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 888 return (0); 889 if (sin6_copy.sin6_scope_id == zone) 890 return (1); 891 else 892 return (0); 893 } 894 895 /* 896 * If the address matches one of our addresses, 897 * it should be a neighbor. 898 * If the address matches one of our on-link prefixes, it should be a 899 * neighbor. 900 */ 901 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 902 if (pr->ndpr_ifp != ifp) 903 continue; 904 905 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) { 906 struct rtentry *rt; 907 908 /* Always use the default FIB here. */ 909 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 910 0, 0, RT_DEFAULT_FIB); 911 if (rt == NULL) 912 continue; 913 /* 914 * This is the case where multiple interfaces 915 * have the same prefix, but only one is installed 916 * into the routing table and that prefix entry 917 * is not the one being examined here. In the case 918 * where RADIX_MPATH is enabled, multiple route 919 * entries (of the same rt_key value) will be 920 * installed because the interface addresses all 921 * differ. 922 */ 923 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 924 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) { 925 RTFREE_LOCKED(rt); 926 continue; 927 } 928 RTFREE_LOCKED(rt); 929 } 930 931 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 932 &addr->sin6_addr, &pr->ndpr_mask)) 933 return (1); 934 } 935 936 /* 937 * If the address is assigned on the node of the other side of 938 * a p2p interface, the address should be a neighbor. 939 */ 940 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr); 941 if (dstaddr != NULL) { 942 if (dstaddr->ifa_ifp == ifp) { 943 ifa_free(dstaddr); 944 return (1); 945 } 946 ifa_free(dstaddr); 947 } 948 949 /* 950 * If the default router list is empty, all addresses are regarded 951 * as on-link, and thus, as a neighbor. 952 */ 953 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && 954 TAILQ_EMPTY(&V_nd_defrouter) && 955 V_nd6_defifindex == ifp->if_index) { 956 return (1); 957 } 958 959 return (0); 960 } 961 962 963 /* 964 * Detect if a given IPv6 address identifies a neighbor on a given link. 965 * XXX: should take care of the destination of a p2p link? 966 */ 967 int 968 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 969 { 970 struct llentry *lle; 971 int rc = 0; 972 973 IF_AFDATA_UNLOCK_ASSERT(ifp); 974 if (nd6_is_new_addr_neighbor(addr, ifp)) 975 return (1); 976 977 /* 978 * Even if the address matches none of our addresses, it might be 979 * in the neighbor cache. 980 */ 981 IF_AFDATA_RLOCK(ifp); 982 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) { 983 LLE_RUNLOCK(lle); 984 rc = 1; 985 } 986 IF_AFDATA_RUNLOCK(ifp); 987 return (rc); 988 } 989 990 /* 991 * Free an nd6 llinfo entry. 992 * Since the function would cause significant changes in the kernel, DO NOT 993 * make it global, unless you have a strong reason for the change, and are sure 994 * that the change is safe. 995 */ 996 static struct llentry * 997 nd6_free(struct llentry *ln, int gc) 998 { 999 struct llentry *next; 1000 struct nd_defrouter *dr; 1001 struct ifnet *ifp; 1002 1003 LLE_WLOCK_ASSERT(ln); 1004 1005 /* 1006 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1007 * even though it is not harmful, it was not really necessary. 1008 */ 1009 1010 /* cancel timer */ 1011 nd6_llinfo_settimer_locked(ln, -1); 1012 1013 ifp = ln->lle_tbl->llt_ifp; 1014 1015 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1016 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1017 1018 if (dr != NULL && dr->expire && 1019 ln->ln_state == ND6_LLINFO_STALE && gc) { 1020 /* 1021 * If the reason for the deletion is just garbage 1022 * collection, and the neighbor is an active default 1023 * router, do not delete it. Instead, reset the GC 1024 * timer using the router's lifetime. 1025 * Simply deleting the entry would affect default 1026 * router selection, which is not necessarily a good 1027 * thing, especially when we're using router preference 1028 * values. 1029 * XXX: the check for ln_state would be redundant, 1030 * but we intentionally keep it just in case. 1031 */ 1032 if (dr->expire > time_second) 1033 nd6_llinfo_settimer_locked(ln, 1034 (dr->expire - time_second) * hz); 1035 else 1036 nd6_llinfo_settimer_locked(ln, 1037 (long)V_nd6_gctimer * hz); 1038 1039 next = LIST_NEXT(ln, lle_next); 1040 LLE_REMREF(ln); 1041 LLE_WUNLOCK(ln); 1042 return (next); 1043 } 1044 1045 if (dr) { 1046 /* 1047 * Unreachablity of a router might affect the default 1048 * router selection and on-link detection of advertised 1049 * prefixes. 1050 */ 1051 1052 /* 1053 * Temporarily fake the state to choose a new default 1054 * router and to perform on-link determination of 1055 * prefixes correctly. 1056 * Below the state will be set correctly, 1057 * or the entry itself will be deleted. 1058 */ 1059 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1060 } 1061 1062 if (ln->ln_router || dr) { 1063 1064 /* 1065 * We need to unlock to avoid a LOR with rt6_flush() with the 1066 * rnh and for the calls to pfxlist_onlink_check() and 1067 * defrouter_select() in the block further down for calls 1068 * into nd6_lookup(). We still hold a ref. 1069 */ 1070 LLE_WUNLOCK(ln); 1071 1072 /* 1073 * rt6_flush must be called whether or not the neighbor 1074 * is in the Default Router List. 1075 * See a corresponding comment in nd6_na_input(). 1076 */ 1077 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1078 } 1079 1080 if (dr) { 1081 /* 1082 * Since defrouter_select() does not affect the 1083 * on-link determination and MIP6 needs the check 1084 * before the default router selection, we perform 1085 * the check now. 1086 */ 1087 pfxlist_onlink_check(); 1088 1089 /* 1090 * Refresh default router list. 1091 */ 1092 defrouter_select(); 1093 } 1094 1095 if (ln->ln_router || dr) 1096 LLE_WLOCK(ln); 1097 } 1098 1099 /* 1100 * Before deleting the entry, remember the next entry as the 1101 * return value. We need this because pfxlist_onlink_check() above 1102 * might have freed other entries (particularly the old next entry) as 1103 * a side effect (XXX). 1104 */ 1105 next = LIST_NEXT(ln, lle_next); 1106 1107 /* 1108 * Save to unlock. We still hold an extra reference and will not 1109 * free(9) in llentry_free() if someone else holds one as well. 1110 */ 1111 LLE_WUNLOCK(ln); 1112 IF_AFDATA_LOCK(ifp); 1113 LLE_WLOCK(ln); 1114 1115 /* Guard against race with other llentry_free(). */ 1116 if (ln->la_flags & LLE_LINKED) { 1117 LLE_REMREF(ln); 1118 llentry_free(ln); 1119 } else 1120 LLE_FREE_LOCKED(ln); 1121 1122 IF_AFDATA_UNLOCK(ifp); 1123 1124 return (next); 1125 } 1126 1127 /* 1128 * Upper-layer reachability hint for Neighbor Unreachability Detection. 1129 * 1130 * XXX cost-effective methods? 1131 */ 1132 void 1133 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force) 1134 { 1135 struct llentry *ln; 1136 struct ifnet *ifp; 1137 1138 if ((dst6 == NULL) || (rt == NULL)) 1139 return; 1140 1141 ifp = rt->rt_ifp; 1142 IF_AFDATA_LOCK(ifp); 1143 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL); 1144 IF_AFDATA_UNLOCK(ifp); 1145 if (ln == NULL) 1146 return; 1147 1148 if (ln->ln_state < ND6_LLINFO_REACHABLE) 1149 goto done; 1150 1151 /* 1152 * if we get upper-layer reachability confirmation many times, 1153 * it is possible we have false information. 1154 */ 1155 if (!force) { 1156 ln->ln_byhint++; 1157 if (ln->ln_byhint > V_nd6_maxnudhint) { 1158 goto done; 1159 } 1160 } 1161 1162 ln->ln_state = ND6_LLINFO_REACHABLE; 1163 if (!ND6_LLINFO_PERMANENT(ln)) { 1164 nd6_llinfo_settimer_locked(ln, 1165 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz); 1166 } 1167 done: 1168 LLE_WUNLOCK(ln); 1169 } 1170 1171 1172 /* 1173 * Rejuvenate this function for routing operations related 1174 * processing. 1175 */ 1176 void 1177 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info) 1178 { 1179 struct sockaddr_in6 *gateway; 1180 struct nd_defrouter *dr; 1181 struct ifnet *ifp; 1182 1183 RT_LOCK_ASSERT(rt); 1184 gateway = (struct sockaddr_in6 *)rt->rt_gateway; 1185 ifp = rt->rt_ifp; 1186 1187 switch (req) { 1188 case RTM_ADD: 1189 break; 1190 1191 case RTM_DELETE: 1192 if (!ifp) 1193 return; 1194 /* 1195 * Only indirect routes are interesting. 1196 */ 1197 if ((rt->rt_flags & RTF_GATEWAY) == 0) 1198 return; 1199 /* 1200 * check for default route 1201 */ 1202 if (IN6_ARE_ADDR_EQUAL(&in6addr_any, 1203 &SIN6(rt_key(rt))->sin6_addr)) { 1204 1205 dr = defrouter_lookup(&gateway->sin6_addr, ifp); 1206 if (dr != NULL) 1207 dr->installed = 0; 1208 } 1209 break; 1210 } 1211 } 1212 1213 1214 int 1215 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) 1216 { 1217 struct in6_drlist *drl = (struct in6_drlist *)data; 1218 struct in6_oprlist *oprl = (struct in6_oprlist *)data; 1219 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1220 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1221 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1222 struct nd_defrouter *dr; 1223 struct nd_prefix *pr; 1224 int i = 0, error = 0; 1225 1226 switch (cmd) { 1227 case SIOCGDRLST_IN6: 1228 /* 1229 * obsolete API, use sysctl under net.inet6.icmp6 1230 */ 1231 bzero(drl, sizeof(*drl)); 1232 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 1233 if (i >= DRLSTSIZ) 1234 break; 1235 drl->defrouter[i].rtaddr = dr->rtaddr; 1236 in6_clearscope(&drl->defrouter[i].rtaddr); 1237 1238 drl->defrouter[i].flags = dr->flags; 1239 drl->defrouter[i].rtlifetime = dr->rtlifetime; 1240 drl->defrouter[i].expire = dr->expire; 1241 drl->defrouter[i].if_index = dr->ifp->if_index; 1242 i++; 1243 } 1244 break; 1245 case SIOCGPRLST_IN6: 1246 /* 1247 * obsolete API, use sysctl under net.inet6.icmp6 1248 * 1249 * XXX the structure in6_prlist was changed in backward- 1250 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6, 1251 * in6_prlist is used for nd6_sysctl() - fill_prlist(). 1252 */ 1253 /* 1254 * XXX meaning of fields, especialy "raflags", is very 1255 * differnet between RA prefix list and RR/static prefix list. 1256 * how about separating ioctls into two? 1257 */ 1258 bzero(oprl, sizeof(*oprl)); 1259 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 1260 struct nd_pfxrouter *pfr; 1261 int j; 1262 1263 if (i >= PRLSTSIZ) 1264 break; 1265 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr; 1266 oprl->prefix[i].raflags = pr->ndpr_raf; 1267 oprl->prefix[i].prefixlen = pr->ndpr_plen; 1268 oprl->prefix[i].vltime = pr->ndpr_vltime; 1269 oprl->prefix[i].pltime = pr->ndpr_pltime; 1270 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index; 1271 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 1272 oprl->prefix[i].expire = 0; 1273 else { 1274 time_t maxexpire; 1275 1276 /* XXX: we assume time_t is signed. */ 1277 maxexpire = (-1) & 1278 ~((time_t)1 << 1279 ((sizeof(maxexpire) * 8) - 1)); 1280 if (pr->ndpr_vltime < 1281 maxexpire - pr->ndpr_lastupdate) { 1282 oprl->prefix[i].expire = 1283 pr->ndpr_lastupdate + 1284 pr->ndpr_vltime; 1285 } else 1286 oprl->prefix[i].expire = maxexpire; 1287 } 1288 1289 j = 0; 1290 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 1291 if (j < DRLSTSIZ) { 1292 #define RTRADDR oprl->prefix[i].advrtr[j] 1293 RTRADDR = pfr->router->rtaddr; 1294 in6_clearscope(&RTRADDR); 1295 #undef RTRADDR 1296 } 1297 j++; 1298 } 1299 oprl->prefix[i].advrtrs = j; 1300 oprl->prefix[i].origin = PR_ORIG_RA; 1301 1302 i++; 1303 } 1304 1305 break; 1306 case OSIOCGIFINFO_IN6: 1307 #define ND ndi->ndi 1308 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1309 bzero(&ND, sizeof(ND)); 1310 ND.linkmtu = IN6_LINKMTU(ifp); 1311 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1312 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1313 ND.reachable = ND_IFINFO(ifp)->reachable; 1314 ND.retrans = ND_IFINFO(ifp)->retrans; 1315 ND.flags = ND_IFINFO(ifp)->flags; 1316 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1317 ND.chlim = ND_IFINFO(ifp)->chlim; 1318 break; 1319 case SIOCGIFINFO_IN6: 1320 ND = *ND_IFINFO(ifp); 1321 break; 1322 case SIOCSIFINFO_IN6: 1323 /* 1324 * used to change host variables from userland. 1325 * intented for a use on router to reflect RA configurations. 1326 */ 1327 /* 0 means 'unspecified' */ 1328 if (ND.linkmtu != 0) { 1329 if (ND.linkmtu < IPV6_MMTU || 1330 ND.linkmtu > IN6_LINKMTU(ifp)) { 1331 error = EINVAL; 1332 break; 1333 } 1334 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1335 } 1336 1337 if (ND.basereachable != 0) { 1338 int obasereachable = ND_IFINFO(ifp)->basereachable; 1339 1340 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1341 if (ND.basereachable != obasereachable) 1342 ND_IFINFO(ifp)->reachable = 1343 ND_COMPUTE_RTIME(ND.basereachable); 1344 } 1345 if (ND.retrans != 0) 1346 ND_IFINFO(ifp)->retrans = ND.retrans; 1347 if (ND.chlim != 0) 1348 ND_IFINFO(ifp)->chlim = ND.chlim; 1349 /* FALLTHROUGH */ 1350 case SIOCSIFINFO_FLAGS: 1351 { 1352 struct ifaddr *ifa; 1353 struct in6_ifaddr *ia; 1354 1355 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1356 !(ND.flags & ND6_IFF_IFDISABLED)) { 1357 /* ifdisabled 1->0 transision */ 1358 1359 /* 1360 * If the interface is marked as ND6_IFF_IFDISABLED and 1361 * has an link-local address with IN6_IFF_DUPLICATED, 1362 * do not clear ND6_IFF_IFDISABLED. 1363 * See RFC 4862, Section 5.4.5. 1364 */ 1365 int duplicated_linklocal = 0; 1366 1367 IF_ADDR_RLOCK(ifp); 1368 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1369 if (ifa->ifa_addr->sa_family != AF_INET6) 1370 continue; 1371 ia = (struct in6_ifaddr *)ifa; 1372 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) && 1373 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1374 duplicated_linklocal = 1; 1375 break; 1376 } 1377 } 1378 IF_ADDR_RUNLOCK(ifp); 1379 1380 if (duplicated_linklocal) { 1381 ND.flags |= ND6_IFF_IFDISABLED; 1382 log(LOG_ERR, "Cannot enable an interface" 1383 " with a link-local address marked" 1384 " duplicate.\n"); 1385 } else { 1386 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED; 1387 if (ifp->if_flags & IFF_UP) 1388 in6_if_up(ifp); 1389 } 1390 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1391 (ND.flags & ND6_IFF_IFDISABLED)) { 1392 /* ifdisabled 0->1 transision */ 1393 /* Mark all IPv6 address as tentative. */ 1394 1395 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; 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 ia->ia6_flags |= IN6_IFF_TENTATIVE; 1402 } 1403 IF_ADDR_RUNLOCK(ifp); 1404 } 1405 1406 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) { 1407 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) { 1408 /* auto_linklocal 0->1 transision */ 1409 1410 /* If no link-local address on ifp, configure */ 1411 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL; 1412 in6_ifattach(ifp, NULL); 1413 } else if (!(ND.flags & ND6_IFF_IFDISABLED) && 1414 ifp->if_flags & IFF_UP) { 1415 /* 1416 * When the IF already has 1417 * ND6_IFF_AUTO_LINKLOCAL, no link-local 1418 * address is assigned, and IFF_UP, try to 1419 * assign one. 1420 */ 1421 int haslinklocal = 0; 1422 1423 IF_ADDR_RLOCK(ifp); 1424 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1425 if (ifa->ifa_addr->sa_family != AF_INET6) 1426 continue; 1427 ia = (struct in6_ifaddr *)ifa; 1428 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1429 haslinklocal = 1; 1430 break; 1431 } 1432 } 1433 IF_ADDR_RUNLOCK(ifp); 1434 if (!haslinklocal) 1435 in6_ifattach(ifp, NULL); 1436 } 1437 } 1438 } 1439 ND_IFINFO(ifp)->flags = ND.flags; 1440 break; 1441 #undef ND 1442 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1443 /* sync kernel routing table with the default router list */ 1444 defrouter_reset(); 1445 defrouter_select(); 1446 break; 1447 case SIOCSPFXFLUSH_IN6: 1448 { 1449 /* flush all the prefix advertised by routers */ 1450 struct nd_prefix *pr, *next; 1451 1452 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) { 1453 struct in6_ifaddr *ia, *ia_next; 1454 1455 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1456 continue; /* XXX */ 1457 1458 /* do we really have to remove addresses as well? */ 1459 /* XXXRW: in6_ifaddrhead locking. */ 1460 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link, 1461 ia_next) { 1462 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1463 continue; 1464 1465 if (ia->ia6_ndpr == pr) 1466 in6_purgeaddr(&ia->ia_ifa); 1467 } 1468 prelist_remove(pr); 1469 } 1470 break; 1471 } 1472 case SIOCSRTRFLUSH_IN6: 1473 { 1474 /* flush all the default routers */ 1475 struct nd_defrouter *dr, *next; 1476 1477 defrouter_reset(); 1478 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) { 1479 defrtrlist_del(dr); 1480 } 1481 defrouter_select(); 1482 break; 1483 } 1484 case SIOCGNBRINFO_IN6: 1485 { 1486 struct llentry *ln; 1487 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1488 1489 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1490 return (error); 1491 1492 IF_AFDATA_RLOCK(ifp); 1493 ln = nd6_lookup(&nb_addr, 0, ifp); 1494 IF_AFDATA_RUNLOCK(ifp); 1495 1496 if (ln == NULL) { 1497 error = EINVAL; 1498 break; 1499 } 1500 nbi->state = ln->ln_state; 1501 nbi->asked = ln->la_asked; 1502 nbi->isrouter = ln->ln_router; 1503 nbi->expire = ln->la_expire; 1504 LLE_RUNLOCK(ln); 1505 break; 1506 } 1507 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1508 ndif->ifindex = V_nd6_defifindex; 1509 break; 1510 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1511 return (nd6_setdefaultiface(ndif->ifindex)); 1512 } 1513 return (error); 1514 } 1515 1516 /* 1517 * Create neighbor cache entry and cache link-layer address, 1518 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1519 * 1520 * type - ICMP6 type 1521 * code - type dependent information 1522 * 1523 * XXXXX 1524 * The caller of this function already acquired the ndp 1525 * cache table lock because the cache entry is returned. 1526 */ 1527 struct llentry * 1528 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, 1529 int lladdrlen, int type, int code) 1530 { 1531 struct llentry *ln = NULL; 1532 int is_newentry; 1533 int do_update; 1534 int olladdr; 1535 int llchange; 1536 int flags; 1537 int newstate = 0; 1538 uint16_t router = 0; 1539 struct sockaddr_in6 sin6; 1540 struct mbuf *chain = NULL; 1541 int static_route = 0; 1542 1543 IF_AFDATA_UNLOCK_ASSERT(ifp); 1544 1545 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__)); 1546 KASSERT(from != NULL, ("%s: from == NULL", __func__)); 1547 1548 /* nothing must be updated for unspecified address */ 1549 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1550 return NULL; 1551 1552 /* 1553 * Validation about ifp->if_addrlen and lladdrlen must be done in 1554 * the caller. 1555 * 1556 * XXX If the link does not have link-layer adderss, what should 1557 * we do? (ifp->if_addrlen == 0) 1558 * Spec says nothing in sections for RA, RS and NA. There's small 1559 * description on it in NS section (RFC 2461 7.2.3). 1560 */ 1561 flags = lladdr ? ND6_EXCLUSIVE : 0; 1562 IF_AFDATA_LOCK(ifp); 1563 ln = nd6_lookup(from, flags, ifp); 1564 1565 if (ln == NULL) { 1566 flags |= ND6_EXCLUSIVE; 1567 ln = nd6_lookup(from, flags | ND6_CREATE, ifp); 1568 IF_AFDATA_UNLOCK(ifp); 1569 is_newentry = 1; 1570 } else { 1571 IF_AFDATA_UNLOCK(ifp); 1572 /* do nothing if static ndp is set */ 1573 if (ln->la_flags & LLE_STATIC) { 1574 static_route = 1; 1575 goto done; 1576 } 1577 is_newentry = 0; 1578 } 1579 if (ln == NULL) 1580 return (NULL); 1581 1582 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0; 1583 if (olladdr && lladdr) { 1584 llchange = bcmp(lladdr, &ln->ll_addr, 1585 ifp->if_addrlen); 1586 } else 1587 llchange = 0; 1588 1589 /* 1590 * newentry olladdr lladdr llchange (*=record) 1591 * 0 n n -- (1) 1592 * 0 y n -- (2) 1593 * 0 n y -- (3) * STALE 1594 * 0 y y n (4) * 1595 * 0 y y y (5) * STALE 1596 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1597 * 1 -- y -- (7) * STALE 1598 */ 1599 1600 if (lladdr) { /* (3-5) and (7) */ 1601 /* 1602 * Record source link-layer address 1603 * XXX is it dependent to ifp->if_type? 1604 */ 1605 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen); 1606 ln->la_flags |= LLE_VALID; 1607 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 1608 } 1609 1610 if (!is_newentry) { 1611 if ((!olladdr && lladdr != NULL) || /* (3) */ 1612 (olladdr && lladdr != NULL && llchange)) { /* (5) */ 1613 do_update = 1; 1614 newstate = ND6_LLINFO_STALE; 1615 } else /* (1-2,4) */ 1616 do_update = 0; 1617 } else { 1618 do_update = 1; 1619 if (lladdr == NULL) /* (6) */ 1620 newstate = ND6_LLINFO_NOSTATE; 1621 else /* (7) */ 1622 newstate = ND6_LLINFO_STALE; 1623 } 1624 1625 if (do_update) { 1626 /* 1627 * Update the state of the neighbor cache. 1628 */ 1629 ln->ln_state = newstate; 1630 1631 if (ln->ln_state == ND6_LLINFO_STALE) { 1632 /* 1633 * XXX: since nd6_output() below will cause 1634 * state tansition to DELAY and reset the timer, 1635 * we must set the timer now, although it is actually 1636 * meaningless. 1637 */ 1638 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 1639 1640 if (ln->la_hold) { 1641 struct mbuf *m_hold, *m_hold_next; 1642 1643 /* 1644 * reset the la_hold in advance, to explicitly 1645 * prevent a la_hold lookup in nd6_output() 1646 * (wouldn't happen, though...) 1647 */ 1648 for (m_hold = ln->la_hold, ln->la_hold = NULL; 1649 m_hold; m_hold = m_hold_next) { 1650 m_hold_next = m_hold->m_nextpkt; 1651 m_hold->m_nextpkt = NULL; 1652 1653 /* 1654 * we assume ifp is not a p2p here, so 1655 * just set the 2nd argument as the 1656 * 1st one. 1657 */ 1658 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain); 1659 } 1660 /* 1661 * If we have mbufs in the chain we need to do 1662 * deferred transmit. Copy the address from the 1663 * llentry before dropping the lock down below. 1664 */ 1665 if (chain != NULL) 1666 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6)); 1667 } 1668 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { 1669 /* probe right away */ 1670 nd6_llinfo_settimer_locked((void *)ln, 0); 1671 } 1672 } 1673 1674 /* 1675 * ICMP6 type dependent behavior. 1676 * 1677 * NS: clear IsRouter if new entry 1678 * RS: clear IsRouter 1679 * RA: set IsRouter if there's lladdr 1680 * redir: clear IsRouter if new entry 1681 * 1682 * RA case, (1): 1683 * The spec says that we must set IsRouter in the following cases: 1684 * - If lladdr exist, set IsRouter. This means (1-5). 1685 * - If it is old entry (!newentry), set IsRouter. This means (7). 1686 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1687 * A quetion arises for (1) case. (1) case has no lladdr in the 1688 * neighbor cache, this is similar to (6). 1689 * This case is rare but we figured that we MUST NOT set IsRouter. 1690 * 1691 * newentry olladdr lladdr llchange NS RS RA redir 1692 * D R 1693 * 0 n n -- (1) c ? s 1694 * 0 y n -- (2) c s s 1695 * 0 n y -- (3) c s s 1696 * 0 y y n (4) c s s 1697 * 0 y y y (5) c s s 1698 * 1 -- n -- (6) c c c s 1699 * 1 -- y -- (7) c c s c s 1700 * 1701 * (c=clear s=set) 1702 */ 1703 switch (type & 0xff) { 1704 case ND_NEIGHBOR_SOLICIT: 1705 /* 1706 * New entry must have is_router flag cleared. 1707 */ 1708 if (is_newentry) /* (6-7) */ 1709 ln->ln_router = 0; 1710 break; 1711 case ND_REDIRECT: 1712 /* 1713 * If the icmp is a redirect to a better router, always set the 1714 * is_router flag. Otherwise, if the entry is newly created, 1715 * clear the flag. [RFC 2461, sec 8.3] 1716 */ 1717 if (code == ND_REDIRECT_ROUTER) 1718 ln->ln_router = 1; 1719 else if (is_newentry) /* (6-7) */ 1720 ln->ln_router = 0; 1721 break; 1722 case ND_ROUTER_SOLICIT: 1723 /* 1724 * is_router flag must always be cleared. 1725 */ 1726 ln->ln_router = 0; 1727 break; 1728 case ND_ROUTER_ADVERT: 1729 /* 1730 * Mark an entry with lladdr as a router. 1731 */ 1732 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */ 1733 (is_newentry && lladdr)) { /* (7) */ 1734 ln->ln_router = 1; 1735 } 1736 break; 1737 } 1738 1739 if (ln != NULL) { 1740 static_route = (ln->la_flags & LLE_STATIC); 1741 router = ln->ln_router; 1742 1743 if (flags & ND6_EXCLUSIVE) 1744 LLE_WUNLOCK(ln); 1745 else 1746 LLE_RUNLOCK(ln); 1747 if (static_route) 1748 ln = NULL; 1749 } 1750 if (chain) 1751 nd6_output_flush(ifp, ifp, chain, &sin6, NULL); 1752 1753 /* 1754 * When the link-layer address of a router changes, select the 1755 * best router again. In particular, when the neighbor entry is newly 1756 * created, it might affect the selection policy. 1757 * Question: can we restrict the first condition to the "is_newentry" 1758 * case? 1759 * XXX: when we hear an RA from a new router with the link-layer 1760 * address option, defrouter_select() is called twice, since 1761 * defrtrlist_update called the function as well. However, I believe 1762 * we can compromise the overhead, since it only happens the first 1763 * time. 1764 * XXX: although defrouter_select() should not have a bad effect 1765 * for those are not autoconfigured hosts, we explicitly avoid such 1766 * cases for safety. 1767 */ 1768 if (do_update && router && 1769 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1770 /* 1771 * guaranteed recursion 1772 */ 1773 defrouter_select(); 1774 } 1775 1776 return (ln); 1777 done: 1778 if (ln != NULL) { 1779 if (flags & ND6_EXCLUSIVE) 1780 LLE_WUNLOCK(ln); 1781 else 1782 LLE_RUNLOCK(ln); 1783 if (static_route) 1784 ln = NULL; 1785 } 1786 return (ln); 1787 } 1788 1789 static void 1790 nd6_slowtimo(void *arg) 1791 { 1792 CURVNET_SET((struct vnet *) arg); 1793 struct nd_ifinfo *nd6if; 1794 struct ifnet *ifp; 1795 1796 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 1797 nd6_slowtimo, curvnet); 1798 IFNET_RLOCK_NOSLEEP(); 1799 TAILQ_FOREACH(ifp, &V_ifnet, if_list) { 1800 nd6if = ND_IFINFO(ifp); 1801 if (nd6if->basereachable && /* already initialized */ 1802 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 1803 /* 1804 * Since reachable time rarely changes by router 1805 * advertisements, we SHOULD insure that a new random 1806 * value gets recomputed at least once every few hours. 1807 * (RFC 2461, 6.3.4) 1808 */ 1809 nd6if->recalctm = V_nd6_recalc_reachtm_interval; 1810 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 1811 } 1812 } 1813 IFNET_RUNLOCK_NOSLEEP(); 1814 CURVNET_RESTORE(); 1815 } 1816 1817 int 1818 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 1819 struct sockaddr_in6 *dst, struct rtentry *rt0) 1820 { 1821 1822 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL)); 1823 } 1824 1825 1826 /* 1827 * Note that I'm not enforcing any global serialization 1828 * lle state or asked changes here as the logic is too 1829 * complicated to avoid having to always acquire an exclusive 1830 * lock 1831 * KMM 1832 * 1833 */ 1834 #define senderr(e) { error = (e); goto bad;} 1835 1836 int 1837 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 1838 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle, 1839 struct mbuf **chain) 1840 { 1841 struct mbuf *m = m0; 1842 struct m_tag *mtag; 1843 struct llentry *ln = lle; 1844 struct ip6_hdr *ip6; 1845 int error = 0; 1846 int flags = 0; 1847 int ip6len; 1848 1849 #ifdef INVARIANTS 1850 if (lle != NULL) { 1851 1852 LLE_WLOCK_ASSERT(lle); 1853 1854 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed")); 1855 } 1856 #endif 1857 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) 1858 goto sendpkt; 1859 1860 if (nd6_need_cache(ifp) == 0) 1861 goto sendpkt; 1862 1863 /* 1864 * next hop determination. This routine is derived from ether_output. 1865 */ 1866 1867 /* 1868 * Address resolution or Neighbor Unreachability Detection 1869 * for the next hop. 1870 * At this point, the destination of the packet must be a unicast 1871 * or an anycast address(i.e. not a multicast). 1872 */ 1873 1874 flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0; 1875 if (ln == NULL) { 1876 retry: 1877 IF_AFDATA_LOCK(ifp); 1878 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst); 1879 IF_AFDATA_UNLOCK(ifp); 1880 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) { 1881 /* 1882 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 1883 * the condition below is not very efficient. But we believe 1884 * it is tolerable, because this should be a rare case. 1885 */ 1886 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0); 1887 IF_AFDATA_LOCK(ifp); 1888 ln = nd6_lookup(&dst->sin6_addr, flags, ifp); 1889 IF_AFDATA_UNLOCK(ifp); 1890 } 1891 } 1892 if (ln == NULL) { 1893 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && 1894 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 1895 char ip6buf[INET6_ADDRSTRLEN]; 1896 log(LOG_DEBUG, 1897 "nd6_output: can't allocate llinfo for %s " 1898 "(ln=%p)\n", 1899 ip6_sprintf(ip6buf, &dst->sin6_addr), ln); 1900 senderr(EIO); /* XXX: good error? */ 1901 } 1902 goto sendpkt; /* send anyway */ 1903 } 1904 1905 /* We don't have to do link-layer address resolution on a p2p link. */ 1906 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && 1907 ln->ln_state < ND6_LLINFO_REACHABLE) { 1908 if ((flags & LLE_EXCLUSIVE) == 0) { 1909 flags |= LLE_EXCLUSIVE; 1910 goto retry; 1911 } 1912 ln->ln_state = ND6_LLINFO_STALE; 1913 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 1914 } 1915 1916 /* 1917 * The first time we send a packet to a neighbor whose entry is 1918 * STALE, we have to change the state to DELAY and a sets a timer to 1919 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 1920 * neighbor unreachability detection on expiration. 1921 * (RFC 2461 7.3.3) 1922 */ 1923 if (ln->ln_state == ND6_LLINFO_STALE) { 1924 if ((flags & LLE_EXCLUSIVE) == 0) { 1925 flags |= LLE_EXCLUSIVE; 1926 LLE_RUNLOCK(ln); 1927 goto retry; 1928 } 1929 ln->la_asked = 0; 1930 ln->ln_state = ND6_LLINFO_DELAY; 1931 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz); 1932 } 1933 1934 /* 1935 * If the neighbor cache entry has a state other than INCOMPLETE 1936 * (i.e. its link-layer address is already resolved), just 1937 * send the packet. 1938 */ 1939 if (ln->ln_state > ND6_LLINFO_INCOMPLETE) 1940 goto sendpkt; 1941 1942 /* 1943 * There is a neighbor cache entry, but no ethernet address 1944 * response yet. Append this latest packet to the end of the 1945 * packet queue in the mbuf, unless the number of the packet 1946 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 1947 * the oldest packet in the queue will be removed. 1948 */ 1949 if (ln->ln_state == ND6_LLINFO_NOSTATE) 1950 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1951 1952 if ((flags & LLE_EXCLUSIVE) == 0) { 1953 flags |= LLE_EXCLUSIVE; 1954 LLE_RUNLOCK(ln); 1955 goto retry; 1956 } 1957 1958 LLE_WLOCK_ASSERT(ln); 1959 1960 if (ln->la_hold) { 1961 struct mbuf *m_hold; 1962 int i; 1963 1964 i = 0; 1965 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) { 1966 i++; 1967 if (m_hold->m_nextpkt == NULL) { 1968 m_hold->m_nextpkt = m; 1969 break; 1970 } 1971 } 1972 while (i >= V_nd6_maxqueuelen) { 1973 m_hold = ln->la_hold; 1974 ln->la_hold = ln->la_hold->m_nextpkt; 1975 m_freem(m_hold); 1976 i--; 1977 } 1978 } else { 1979 ln->la_hold = m; 1980 } 1981 1982 /* 1983 * If there has been no NS for the neighbor after entering the 1984 * INCOMPLETE state, send the first solicitation. 1985 */ 1986 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) { 1987 ln->la_asked++; 1988 1989 nd6_llinfo_settimer_locked(ln, 1990 (long)ND_IFINFO(ifp)->retrans * hz / 1000); 1991 LLE_WUNLOCK(ln); 1992 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); 1993 if (lle != NULL && ln == lle) 1994 LLE_WLOCK(lle); 1995 1996 } else if (lle == NULL || ln != lle) { 1997 /* 1998 * We did the lookup (no lle arg) so we 1999 * need to do the unlock here. 2000 */ 2001 LLE_WUNLOCK(ln); 2002 } 2003 2004 return (0); 2005 2006 sendpkt: 2007 /* discard the packet if IPv6 operation is disabled on the interface */ 2008 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 2009 error = ENETDOWN; /* better error? */ 2010 goto bad; 2011 } 2012 /* 2013 * ln is valid and the caller did not pass in 2014 * an llentry 2015 */ 2016 if ((ln != NULL) && (lle == NULL)) { 2017 if (flags & LLE_EXCLUSIVE) 2018 LLE_WUNLOCK(ln); 2019 else 2020 LLE_RUNLOCK(ln); 2021 } 2022 2023 #ifdef MAC 2024 mac_netinet6_nd6_send(ifp, m); 2025 #endif 2026 2027 /* 2028 * If called from nd6_ns_output() (NS), nd6_na_output() (NA), 2029 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA 2030 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND 2031 * to be diverted to user space. When re-injected into the kernel, 2032 * send_output() will directly dispatch them to the outgoing interface. 2033 */ 2034 if (send_sendso_input_hook != NULL) { 2035 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); 2036 if (mtag != NULL) { 2037 ip6 = mtod(m, struct ip6_hdr *); 2038 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); 2039 /* Use the SEND socket */ 2040 error = send_sendso_input_hook(m, ifp, SND_OUT, 2041 ip6len); 2042 /* -1 == no app on SEND socket */ 2043 if (error == 0 || error != -1) 2044 return (error); 2045 } 2046 } 2047 2048 /* 2049 * We were passed in a pointer to an lle with the lock held 2050 * this means that we can't call if_output as we will 2051 * recurse on the lle lock - so what we do is we create 2052 * a list of mbufs to send and transmit them in the caller 2053 * after the lock is dropped 2054 */ 2055 if (lle != NULL) { 2056 if (*chain == NULL) 2057 *chain = m; 2058 else { 2059 struct mbuf *mb; 2060 2061 /* 2062 * append mbuf to end of deferred chain 2063 */ 2064 mb = *chain; 2065 while (mb->m_nextpkt != NULL) 2066 mb = mb->m_nextpkt; 2067 mb->m_nextpkt = m; 2068 } 2069 return (error); 2070 } 2071 /* Reset layer specific mbuf flags to avoid confusing lower layers. */ 2072 m->m_flags &= ~(M_PROTOFLAGS); 2073 if ((ifp->if_flags & IFF_LOOPBACK) != 0) { 2074 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst, 2075 NULL)); 2076 } 2077 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL); 2078 return (error); 2079 2080 bad: 2081 /* 2082 * ln is valid and the caller did not pass in 2083 * an llentry 2084 */ 2085 if ((ln != NULL) && (lle == NULL)) { 2086 if (flags & LLE_EXCLUSIVE) 2087 LLE_WUNLOCK(ln); 2088 else 2089 LLE_RUNLOCK(ln); 2090 } 2091 if (m) 2092 m_freem(m); 2093 return (error); 2094 } 2095 #undef senderr 2096 2097 2098 int 2099 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain, 2100 struct sockaddr_in6 *dst, struct route *ro) 2101 { 2102 struct mbuf *m, *m_head; 2103 struct ifnet *outifp; 2104 int error = 0; 2105 2106 m_head = chain; 2107 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2108 outifp = origifp; 2109 else 2110 outifp = ifp; 2111 2112 while (m_head) { 2113 m = m_head; 2114 m_head = m_head->m_nextpkt; 2115 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro); 2116 } 2117 2118 /* 2119 * XXX 2120 * note that intermediate errors are blindly ignored - but this is 2121 * the same convention as used with nd6_output when called by 2122 * nd6_cache_lladdr 2123 */ 2124 return (error); 2125 } 2126 2127 2128 int 2129 nd6_need_cache(struct ifnet *ifp) 2130 { 2131 /* 2132 * XXX: we currently do not make neighbor cache on any interface 2133 * other than ARCnet, Ethernet, FDDI and GIF. 2134 * 2135 * RFC2893 says: 2136 * - unidirectional tunnels needs no ND 2137 */ 2138 switch (ifp->if_type) { 2139 case IFT_ARCNET: 2140 case IFT_ETHER: 2141 case IFT_FDDI: 2142 case IFT_IEEE1394: 2143 #ifdef IFT_L2VLAN 2144 case IFT_L2VLAN: 2145 #endif 2146 #ifdef IFT_IEEE80211 2147 case IFT_IEEE80211: 2148 #endif 2149 case IFT_INFINIBAND: 2150 case IFT_GIF: /* XXX need more cases? */ 2151 case IFT_PPP: 2152 case IFT_TUNNEL: 2153 case IFT_BRIDGE: 2154 case IFT_PROPVIRTUAL: 2155 return (1); 2156 default: 2157 return (0); 2158 } 2159 } 2160 2161 /* 2162 * the callers of this function need to be re-worked to drop 2163 * the lle lock, drop here for now 2164 */ 2165 int 2166 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m, 2167 struct sockaddr *dst, u_char *desten, struct llentry **lle) 2168 { 2169 struct llentry *ln; 2170 2171 *lle = NULL; 2172 IF_AFDATA_UNLOCK_ASSERT(ifp); 2173 if (m != NULL && m->m_flags & M_MCAST) { 2174 int i; 2175 2176 switch (ifp->if_type) { 2177 case IFT_ETHER: 2178 case IFT_FDDI: 2179 #ifdef IFT_L2VLAN 2180 case IFT_L2VLAN: 2181 #endif 2182 #ifdef IFT_IEEE80211 2183 case IFT_IEEE80211: 2184 #endif 2185 case IFT_BRIDGE: 2186 case IFT_ISO88025: 2187 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr, 2188 desten); 2189 return (0); 2190 case IFT_IEEE1394: 2191 /* 2192 * netbsd can use if_broadcastaddr, but we don't do so 2193 * to reduce # of ifdef. 2194 */ 2195 for (i = 0; i < ifp->if_addrlen; i++) 2196 desten[i] = ~0; 2197 return (0); 2198 case IFT_ARCNET: 2199 *desten = 0; 2200 return (0); 2201 default: 2202 m_freem(m); 2203 return (EAFNOSUPPORT); 2204 } 2205 } 2206 2207 2208 /* 2209 * the entry should have been created in nd6_store_lladdr 2210 */ 2211 IF_AFDATA_RLOCK(ifp); 2212 ln = lla_lookup(LLTABLE6(ifp), 0, dst); 2213 IF_AFDATA_RUNLOCK(ifp); 2214 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) { 2215 if (ln != NULL) 2216 LLE_RUNLOCK(ln); 2217 /* this could happen, if we could not allocate memory */ 2218 m_freem(m); 2219 return (1); 2220 } 2221 2222 bcopy(&ln->ll_addr, desten, ifp->if_addrlen); 2223 *lle = ln; 2224 LLE_RUNLOCK(ln); 2225 /* 2226 * A *small* use after free race exists here 2227 */ 2228 return (0); 2229 } 2230 2231 static void 2232 clear_llinfo_pqueue(struct llentry *ln) 2233 { 2234 struct mbuf *m_hold, *m_hold_next; 2235 2236 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) { 2237 m_hold_next = m_hold->m_nextpkt; 2238 m_freem(m_hold); 2239 } 2240 2241 ln->la_hold = NULL; 2242 return; 2243 } 2244 2245 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2246 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2247 #ifdef SYSCTL_DECL 2248 SYSCTL_DECL(_net_inet6_icmp6); 2249 #endif 2250 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2251 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2252 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2253 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2254 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2255 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, ""); 2256 2257 static int 2258 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2259 { 2260 struct in6_defrouter d; 2261 struct nd_defrouter *dr; 2262 int error; 2263 2264 if (req->newptr) 2265 return (EPERM); 2266 2267 bzero(&d, sizeof(d)); 2268 d.rtaddr.sin6_family = AF_INET6; 2269 d.rtaddr.sin6_len = sizeof(d.rtaddr); 2270 2271 /* 2272 * XXX locking 2273 */ 2274 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 2275 d.rtaddr.sin6_addr = dr->rtaddr; 2276 error = sa6_recoverscope(&d.rtaddr); 2277 if (error != 0) 2278 return (error); 2279 d.flags = dr->flags; 2280 d.rtlifetime = dr->rtlifetime; 2281 d.expire = dr->expire; 2282 d.if_index = dr->ifp->if_index; 2283 error = SYSCTL_OUT(req, &d, sizeof(d)); 2284 if (error != 0) 2285 return (error); 2286 } 2287 return (0); 2288 } 2289 2290 static int 2291 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2292 { 2293 struct in6_prefix p; 2294 struct sockaddr_in6 s6; 2295 struct nd_prefix *pr; 2296 struct nd_pfxrouter *pfr; 2297 time_t maxexpire; 2298 int error; 2299 char ip6buf[INET6_ADDRSTRLEN]; 2300 2301 if (req->newptr) 2302 return (EPERM); 2303 2304 bzero(&p, sizeof(p)); 2305 p.origin = PR_ORIG_RA; 2306 bzero(&s6, sizeof(s6)); 2307 s6.sin6_family = AF_INET6; 2308 s6.sin6_len = sizeof(s6); 2309 2310 /* 2311 * XXX locking 2312 */ 2313 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 2314 p.prefix = pr->ndpr_prefix; 2315 if (sa6_recoverscope(&p.prefix)) { 2316 log(LOG_ERR, "scope error in prefix list (%s)\n", 2317 ip6_sprintf(ip6buf, &p.prefix.sin6_addr)); 2318 /* XXX: press on... */ 2319 } 2320 p.raflags = pr->ndpr_raf; 2321 p.prefixlen = pr->ndpr_plen; 2322 p.vltime = pr->ndpr_vltime; 2323 p.pltime = pr->ndpr_pltime; 2324 p.if_index = pr->ndpr_ifp->if_index; 2325 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2326 p.expire = 0; 2327 else { 2328 /* XXX: we assume time_t is signed. */ 2329 maxexpire = (-1) & 2330 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1)); 2331 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate) 2332 p.expire = pr->ndpr_lastupdate + 2333 pr->ndpr_vltime; 2334 else 2335 p.expire = maxexpire; 2336 } 2337 p.refcnt = pr->ndpr_refcnt; 2338 p.flags = pr->ndpr_stateflags; 2339 p.advrtrs = 0; 2340 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) 2341 p.advrtrs++; 2342 error = SYSCTL_OUT(req, &p, sizeof(p)); 2343 if (error != 0) 2344 return (error); 2345 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 2346 s6.sin6_addr = pfr->router->rtaddr; 2347 if (sa6_recoverscope(&s6)) 2348 log(LOG_ERR, 2349 "scope error in prefix list (%s)\n", 2350 ip6_sprintf(ip6buf, &pfr->router->rtaddr)); 2351 error = SYSCTL_OUT(req, &s6, sizeof(s6)); 2352 if (error != 0) 2353 return (error); 2354 } 2355 } 2356 return (0); 2357 } 2358