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 __P((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); 178 bzero(nd, sizeof(*nd)); 179 180 nd->initialized = 1; 181 182 nd->chlim = IPV6_DEFHLIM; 183 nd->basereachable = REACHABLE_TIME; 184 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable); 185 nd->retrans = RETRANS_TIMER; 186 187 nd->flags = ND6_IFF_PERFORMNUD; 188 189 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. */ 190 if (V_ip6_auto_linklocal || (ifp->if_flags & IFF_LOOPBACK)) 191 nd->flags |= ND6_IFF_AUTO_LINKLOCAL; 192 193 /* A loopback interface does not need to accept RTADV. */ 194 if (V_ip6_accept_rtadv && !(ifp->if_flags & IFF_LOOPBACK)) 195 nd->flags |= ND6_IFF_ACCEPT_RTADV; 196 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK)) 197 nd->flags |= ND6_IFF_NO_RADR; 198 199 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */ 200 nd6_setmtu0(ifp, nd); 201 202 return nd; 203 } 204 205 void 206 nd6_ifdetach(struct nd_ifinfo *nd) 207 { 208 209 free(nd, M_IP6NDP); 210 } 211 212 /* 213 * Reset ND level link MTU. This function is called when the physical MTU 214 * changes, which means we might have to adjust the ND level MTU. 215 */ 216 void 217 nd6_setmtu(struct ifnet *ifp) 218 { 219 220 nd6_setmtu0(ifp, ND_IFINFO(ifp)); 221 } 222 223 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */ 224 void 225 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi) 226 { 227 u_int32_t omaxmtu; 228 229 omaxmtu = ndi->maxmtu; 230 231 switch (ifp->if_type) { 232 case IFT_ARCNET: 233 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */ 234 break; 235 case IFT_FDDI: 236 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */ 237 break; 238 case IFT_ISO88025: 239 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu); 240 break; 241 default: 242 ndi->maxmtu = ifp->if_mtu; 243 break; 244 } 245 246 /* 247 * Decreasing the interface MTU under IPV6 minimum MTU may cause 248 * undesirable situation. We thus notify the operator of the change 249 * explicitly. The check for omaxmtu is necessary to restrict the 250 * log to the case of changing the MTU, not initializing it. 251 */ 252 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { 253 log(LOG_NOTICE, "nd6_setmtu0: " 254 "new link MTU on %s (%lu) is too small for IPv6\n", 255 if_name(ifp), (unsigned long)ndi->maxmtu); 256 } 257 258 if (ndi->maxmtu > V_in6_maxmtu) 259 in6_setmaxmtu(); /* check all interfaces just in case */ 260 261 } 262 263 void 264 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts) 265 { 266 267 bzero(ndopts, sizeof(*ndopts)); 268 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; 269 ndopts->nd_opts_last 270 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); 271 272 if (icmp6len == 0) { 273 ndopts->nd_opts_done = 1; 274 ndopts->nd_opts_search = NULL; 275 } 276 } 277 278 /* 279 * Take one ND option. 280 */ 281 struct nd_opt_hdr * 282 nd6_option(union nd_opts *ndopts) 283 { 284 struct nd_opt_hdr *nd_opt; 285 int olen; 286 287 if (ndopts == NULL) 288 panic("ndopts == NULL in nd6_option"); 289 if (ndopts->nd_opts_last == NULL) 290 panic("uninitialized ndopts in nd6_option"); 291 if (ndopts->nd_opts_search == NULL) 292 return NULL; 293 if (ndopts->nd_opts_done) 294 return NULL; 295 296 nd_opt = ndopts->nd_opts_search; 297 298 /* make sure nd_opt_len is inside the buffer */ 299 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) { 300 bzero(ndopts, sizeof(*ndopts)); 301 return NULL; 302 } 303 304 olen = nd_opt->nd_opt_len << 3; 305 if (olen == 0) { 306 /* 307 * Message validation requires that all included 308 * options have a length that is greater than zero. 309 */ 310 bzero(ndopts, sizeof(*ndopts)); 311 return NULL; 312 } 313 314 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen); 315 if (ndopts->nd_opts_search > ndopts->nd_opts_last) { 316 /* option overruns the end of buffer, invalid */ 317 bzero(ndopts, sizeof(*ndopts)); 318 return NULL; 319 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { 320 /* reached the end of options chain */ 321 ndopts->nd_opts_done = 1; 322 ndopts->nd_opts_search = NULL; 323 } 324 return nd_opt; 325 } 326 327 /* 328 * Parse multiple ND options. 329 * This function is much easier to use, for ND routines that do not need 330 * multiple options of the same type. 331 */ 332 int 333 nd6_options(union nd_opts *ndopts) 334 { 335 struct nd_opt_hdr *nd_opt; 336 int i = 0; 337 338 if (ndopts == NULL) 339 panic("ndopts == NULL in nd6_options"); 340 if (ndopts->nd_opts_last == NULL) 341 panic("uninitialized ndopts in nd6_options"); 342 if (ndopts->nd_opts_search == NULL) 343 return 0; 344 345 while (1) { 346 nd_opt = nd6_option(ndopts); 347 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { 348 /* 349 * Message validation requires that all included 350 * options have a length that is greater than zero. 351 */ 352 ICMP6STAT_INC(icp6s_nd_badopt); 353 bzero(ndopts, sizeof(*ndopts)); 354 return -1; 355 } 356 357 if (nd_opt == NULL) 358 goto skip1; 359 360 switch (nd_opt->nd_opt_type) { 361 case ND_OPT_SOURCE_LINKADDR: 362 case ND_OPT_TARGET_LINKADDR: 363 case ND_OPT_MTU: 364 case ND_OPT_REDIRECTED_HEADER: 365 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { 366 nd6log((LOG_INFO, 367 "duplicated ND6 option found (type=%d)\n", 368 nd_opt->nd_opt_type)); 369 /* XXX bark? */ 370 } else { 371 ndopts->nd_opt_array[nd_opt->nd_opt_type] 372 = nd_opt; 373 } 374 break; 375 case ND_OPT_PREFIX_INFORMATION: 376 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { 377 ndopts->nd_opt_array[nd_opt->nd_opt_type] 378 = nd_opt; 379 } 380 ndopts->nd_opts_pi_end = 381 (struct nd_opt_prefix_info *)nd_opt; 382 break; 383 default: 384 /* 385 * Unknown options must be silently ignored, 386 * to accomodate future extension to the protocol. 387 */ 388 nd6log((LOG_DEBUG, 389 "nd6_options: unsupported option %d - " 390 "option ignored\n", nd_opt->nd_opt_type)); 391 } 392 393 skip1: 394 i++; 395 if (i > V_nd6_maxndopt) { 396 ICMP6STAT_INC(icp6s_nd_toomanyopt); 397 nd6log((LOG_INFO, "too many loop in nd opt\n")); 398 break; 399 } 400 401 if (ndopts->nd_opts_done) 402 break; 403 } 404 405 return 0; 406 } 407 408 /* 409 * ND6 timer routine to handle ND6 entries 410 */ 411 void 412 nd6_llinfo_settimer_locked(struct llentry *ln, long tick) 413 { 414 int canceled; 415 416 LLE_WLOCK_ASSERT(ln); 417 418 if (tick < 0) { 419 ln->la_expire = 0; 420 ln->ln_ntick = 0; 421 canceled = callout_stop(&ln->ln_timer_ch); 422 } else { 423 ln->la_expire = time_second + tick / hz; 424 LLE_ADDREF(ln); 425 if (tick > INT_MAX) { 426 ln->ln_ntick = tick - INT_MAX; 427 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX, 428 nd6_llinfo_timer, ln); 429 } else { 430 ln->ln_ntick = 0; 431 canceled = callout_reset(&ln->ln_timer_ch, tick, 432 nd6_llinfo_timer, ln); 433 } 434 } 435 if (canceled) 436 LLE_REMREF(ln); 437 } 438 439 void 440 nd6_llinfo_settimer(struct llentry *ln, long tick) 441 { 442 443 LLE_WLOCK(ln); 444 nd6_llinfo_settimer_locked(ln, tick); 445 LLE_WUNLOCK(ln); 446 } 447 448 static void 449 nd6_llinfo_timer(void *arg) 450 { 451 struct llentry *ln; 452 struct in6_addr *dst; 453 struct ifnet *ifp; 454 struct nd_ifinfo *ndi = NULL; 455 456 KASSERT(arg != NULL, ("%s: arg NULL", __func__)); 457 ln = (struct llentry *)arg; 458 LLE_WLOCK_ASSERT(ln); 459 ifp = ln->lle_tbl->llt_ifp; 460 461 CURVNET_SET(ifp->if_vnet); 462 463 if (ln->ln_ntick > 0) { 464 if (ln->ln_ntick > INT_MAX) { 465 ln->ln_ntick -= INT_MAX; 466 nd6_llinfo_settimer_locked(ln, INT_MAX); 467 } else { 468 ln->ln_ntick = 0; 469 nd6_llinfo_settimer_locked(ln, ln->ln_ntick); 470 } 471 goto done; 472 } 473 474 ndi = ND_IFINFO(ifp); 475 dst = &L3_ADDR_SIN6(ln)->sin6_addr; 476 if (ln->la_flags & LLE_STATIC) { 477 goto done; 478 } 479 480 if (ln->la_flags & LLE_DELETED) { 481 (void)nd6_free(ln, 0); 482 ln = NULL; 483 goto done; 484 } 485 486 switch (ln->ln_state) { 487 case ND6_LLINFO_INCOMPLETE: 488 if (ln->la_asked < V_nd6_mmaxtries) { 489 ln->la_asked++; 490 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 491 LLE_WUNLOCK(ln); 492 nd6_ns_output(ifp, NULL, dst, ln, 0); 493 LLE_WLOCK(ln); 494 } else { 495 struct mbuf *m = ln->la_hold; 496 if (m) { 497 struct mbuf *m0; 498 499 /* 500 * assuming every packet in la_hold has the 501 * same IP header. Send error after unlock. 502 */ 503 m0 = m->m_nextpkt; 504 m->m_nextpkt = NULL; 505 ln->la_hold = m0; 506 clear_llinfo_pqueue(ln); 507 } 508 (void)nd6_free(ln, 0); 509 ln = NULL; 510 if (m != NULL) 511 icmp6_error2(m, ICMP6_DST_UNREACH, 512 ICMP6_DST_UNREACH_ADDR, 0, ifp); 513 } 514 break; 515 case ND6_LLINFO_REACHABLE: 516 if (!ND6_LLINFO_PERMANENT(ln)) { 517 ln->ln_state = ND6_LLINFO_STALE; 518 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 519 } 520 break; 521 522 case ND6_LLINFO_STALE: 523 /* Garbage Collection(RFC 2461 5.3) */ 524 if (!ND6_LLINFO_PERMANENT(ln)) { 525 (void)nd6_free(ln, 1); 526 ln = NULL; 527 } 528 break; 529 530 case ND6_LLINFO_DELAY: 531 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 532 /* We need NUD */ 533 ln->la_asked = 1; 534 ln->ln_state = ND6_LLINFO_PROBE; 535 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 536 LLE_WUNLOCK(ln); 537 nd6_ns_output(ifp, dst, dst, ln, 0); 538 LLE_WLOCK(ln); 539 } else { 540 ln->ln_state = ND6_LLINFO_STALE; /* XXX */ 541 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 542 } 543 break; 544 case ND6_LLINFO_PROBE: 545 if (ln->la_asked < V_nd6_umaxtries) { 546 ln->la_asked++; 547 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 548 LLE_WUNLOCK(ln); 549 nd6_ns_output(ifp, dst, dst, ln, 0); 550 LLE_WLOCK(ln); 551 } else { 552 (void)nd6_free(ln, 0); 553 ln = NULL; 554 } 555 break; 556 default: 557 panic("%s: paths in a dark night can be confusing: %d", 558 __func__, ln->ln_state); 559 } 560 done: 561 if (ln != NULL) 562 LLE_FREE_LOCKED(ln); 563 CURVNET_RESTORE(); 564 } 565 566 567 /* 568 * ND6 timer routine to expire default route list and prefix list 569 */ 570 void 571 nd6_timer(void *arg) 572 { 573 CURVNET_SET((struct vnet *) arg); 574 int s; 575 struct nd_defrouter *dr; 576 struct nd_prefix *pr; 577 struct in6_ifaddr *ia6, *nia6; 578 struct in6_addrlifetime *lt6; 579 580 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz, 581 nd6_timer, curvnet); 582 583 /* expire default router list */ 584 s = splnet(); 585 dr = TAILQ_FIRST(&V_nd_defrouter); 586 while (dr) { 587 if (dr->expire && dr->expire < time_second) { 588 struct nd_defrouter *t; 589 t = TAILQ_NEXT(dr, dr_entry); 590 defrtrlist_del(dr); 591 dr = t; 592 } else { 593 dr = TAILQ_NEXT(dr, dr_entry); 594 } 595 } 596 597 /* 598 * expire interface addresses. 599 * in the past the loop was inside prefix expiry processing. 600 * However, from a stricter speci-confrmance standpoint, we should 601 * rather separate address lifetimes and prefix lifetimes. 602 * 603 * XXXRW: in6_ifaddrhead locking. 604 */ 605 addrloop: 606 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) { 607 /* check address lifetime */ 608 lt6 = &ia6->ia6_lifetime; 609 if (IFA6_IS_INVALID(ia6)) { 610 int regen = 0; 611 612 /* 613 * If the expiring address is temporary, try 614 * regenerating a new one. This would be useful when 615 * we suspended a laptop PC, then turned it on after a 616 * period that could invalidate all temporary 617 * addresses. Although we may have to restart the 618 * loop (see below), it must be after purging the 619 * address. Otherwise, we'd see an infinite loop of 620 * regeneration. 621 */ 622 if (V_ip6_use_tempaddr && 623 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 624 if (regen_tmpaddr(ia6) == 0) 625 regen = 1; 626 } 627 628 in6_purgeaddr(&ia6->ia_ifa); 629 630 if (regen) 631 goto addrloop; /* XXX: see below */ 632 } else if (IFA6_IS_DEPRECATED(ia6)) { 633 int oldflags = ia6->ia6_flags; 634 635 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 636 637 /* 638 * If a temporary address has just become deprecated, 639 * regenerate a new one if possible. 640 */ 641 if (V_ip6_use_tempaddr && 642 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 643 (oldflags & IN6_IFF_DEPRECATED) == 0) { 644 645 if (regen_tmpaddr(ia6) == 0) { 646 /* 647 * A new temporary address is 648 * generated. 649 * XXX: this means the address chain 650 * has changed while we are still in 651 * the loop. Although the change 652 * would not cause disaster (because 653 * it's not a deletion, but an 654 * addition,) we'd rather restart the 655 * loop just for safety. Or does this 656 * significantly reduce performance?? 657 */ 658 goto addrloop; 659 } 660 } 661 } else { 662 /* 663 * A new RA might have made a deprecated address 664 * preferred. 665 */ 666 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 667 } 668 } 669 670 /* expire prefix list */ 671 pr = V_nd_prefix.lh_first; 672 while (pr) { 673 /* 674 * check prefix lifetime. 675 * since pltime is just for autoconf, pltime processing for 676 * prefix is not necessary. 677 */ 678 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 679 time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) { 680 struct nd_prefix *t; 681 t = pr->ndpr_next; 682 683 /* 684 * address expiration and prefix expiration are 685 * separate. NEVER perform in6_purgeaddr here. 686 */ 687 688 prelist_remove(pr); 689 pr = t; 690 } else 691 pr = pr->ndpr_next; 692 } 693 splx(s); 694 CURVNET_RESTORE(); 695 } 696 697 /* 698 * ia6 - deprecated/invalidated temporary address 699 */ 700 static int 701 regen_tmpaddr(struct in6_ifaddr *ia6) 702 { 703 struct ifaddr *ifa; 704 struct ifnet *ifp; 705 struct in6_ifaddr *public_ifa6 = NULL; 706 707 ifp = ia6->ia_ifa.ifa_ifp; 708 IF_ADDR_LOCK(ifp); 709 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 710 struct in6_ifaddr *it6; 711 712 if (ifa->ifa_addr->sa_family != AF_INET6) 713 continue; 714 715 it6 = (struct in6_ifaddr *)ifa; 716 717 /* ignore no autoconf addresses. */ 718 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 719 continue; 720 721 /* ignore autoconf addresses with different prefixes. */ 722 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 723 continue; 724 725 /* 726 * Now we are looking at an autoconf address with the same 727 * prefix as ours. If the address is temporary and is still 728 * preferred, do not create another one. It would be rare, but 729 * could happen, for example, when we resume a laptop PC after 730 * a long period. 731 */ 732 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 733 !IFA6_IS_DEPRECATED(it6)) { 734 public_ifa6 = NULL; 735 break; 736 } 737 738 /* 739 * This is a public autoconf address that has the same prefix 740 * as ours. If it is preferred, keep it. We can't break the 741 * loop here, because there may be a still-preferred temporary 742 * address with the prefix. 743 */ 744 if (!IFA6_IS_DEPRECATED(it6)) 745 public_ifa6 = it6; 746 747 if (public_ifa6 != NULL) 748 ifa_ref(&public_ifa6->ia_ifa); 749 } 750 IF_ADDR_UNLOCK(ifp); 751 752 if (public_ifa6 != NULL) { 753 int e; 754 755 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 756 ifa_free(&public_ifa6->ia_ifa); 757 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 758 " tmp addr,errno=%d\n", e); 759 return (-1); 760 } 761 ifa_free(&public_ifa6->ia_ifa); 762 return (0); 763 } 764 765 return (-1); 766 } 767 768 /* 769 * Nuke neighbor cache/prefix/default router management table, right before 770 * ifp goes away. 771 */ 772 void 773 nd6_purge(struct ifnet *ifp) 774 { 775 struct nd_defrouter *dr, *ndr; 776 struct nd_prefix *pr, *npr; 777 778 /* 779 * Nuke default router list entries toward ifp. 780 * We defer removal of default router list entries that is installed 781 * in the routing table, in order to keep additional side effects as 782 * small as possible. 783 */ 784 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) { 785 ndr = TAILQ_NEXT(dr, dr_entry); 786 if (dr->installed) 787 continue; 788 789 if (dr->ifp == ifp) 790 defrtrlist_del(dr); 791 } 792 793 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) { 794 ndr = TAILQ_NEXT(dr, dr_entry); 795 if (!dr->installed) 796 continue; 797 798 if (dr->ifp == ifp) 799 defrtrlist_del(dr); 800 } 801 802 /* Nuke prefix list entries toward ifp */ 803 for (pr = V_nd_prefix.lh_first; pr; pr = npr) { 804 npr = pr->ndpr_next; 805 if (pr->ndpr_ifp == ifp) { 806 /* 807 * Because if_detach() does *not* release prefixes 808 * while purging addresses the reference count will 809 * still be above zero. We therefore reset it to 810 * make sure that the prefix really gets purged. 811 */ 812 pr->ndpr_refcnt = 0; 813 814 /* 815 * Previously, pr->ndpr_addr is removed as well, 816 * but I strongly believe we don't have to do it. 817 * nd6_purge() is only called from in6_ifdetach(), 818 * which removes all the associated interface addresses 819 * by itself. 820 * (jinmei@kame.net 20010129) 821 */ 822 prelist_remove(pr); 823 } 824 } 825 826 /* cancel default outgoing interface setting */ 827 if (V_nd6_defifindex == ifp->if_index) 828 nd6_setdefaultiface(0); 829 830 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 831 /* Refresh default router list. */ 832 defrouter_select(); 833 } 834 835 /* XXXXX 836 * We do not nuke the neighbor cache entries here any more 837 * because the neighbor cache is kept in if_afdata[AF_INET6]. 838 * nd6_purge() is invoked by in6_ifdetach() which is called 839 * from if_detach() where everything gets purged. So let 840 * in6_domifdetach() do the actual L2 table purging work. 841 */ 842 } 843 844 /* 845 * the caller acquires and releases the lock on the lltbls 846 * Returns the llentry locked 847 */ 848 struct llentry * 849 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp) 850 { 851 struct sockaddr_in6 sin6; 852 struct llentry *ln; 853 int llflags; 854 855 bzero(&sin6, sizeof(sin6)); 856 sin6.sin6_len = sizeof(struct sockaddr_in6); 857 sin6.sin6_family = AF_INET6; 858 sin6.sin6_addr = *addr6; 859 860 IF_AFDATA_LOCK_ASSERT(ifp); 861 862 llflags = 0; 863 if (flags & ND6_CREATE) 864 llflags |= LLE_CREATE; 865 if (flags & ND6_EXCLUSIVE) 866 llflags |= LLE_EXCLUSIVE; 867 868 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6); 869 if ((ln != NULL) && (llflags & LLE_CREATE)) 870 ln->ln_state = ND6_LLINFO_NOSTATE; 871 872 return (ln); 873 } 874 875 /* 876 * Test whether a given IPv6 address is a neighbor or not, ignoring 877 * the actual neighbor cache. The neighbor cache is ignored in order 878 * to not reenter the routing code from within itself. 879 */ 880 static int 881 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 882 { 883 struct nd_prefix *pr; 884 struct ifaddr *dstaddr; 885 886 /* 887 * A link-local address is always a neighbor. 888 * XXX: a link does not necessarily specify a single interface. 889 */ 890 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 891 struct sockaddr_in6 sin6_copy; 892 u_int32_t zone; 893 894 /* 895 * We need sin6_copy since sa6_recoverscope() may modify the 896 * content (XXX). 897 */ 898 sin6_copy = *addr; 899 if (sa6_recoverscope(&sin6_copy)) 900 return (0); /* XXX: should be impossible */ 901 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 902 return (0); 903 if (sin6_copy.sin6_scope_id == zone) 904 return (1); 905 else 906 return (0); 907 } 908 909 /* 910 * If the address matches one of our addresses, 911 * it should be a neighbor. 912 * If the address matches one of our on-link prefixes, it should be a 913 * neighbor. 914 */ 915 for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) { 916 if (pr->ndpr_ifp != ifp) 917 continue; 918 919 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) { 920 struct rtentry *rt; 921 rt = rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 0, 0); 922 if (rt == NULL) 923 continue; 924 /* 925 * This is the case where multiple interfaces 926 * have the same prefix, but only one is installed 927 * into the routing table and that prefix entry 928 * is not the one being examined here. In the case 929 * where RADIX_MPATH is enabled, multiple route 930 * entries (of the same rt_key value) will be 931 * installed because the interface addresses all 932 * differ. 933 */ 934 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 935 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) { 936 RTFREE_LOCKED(rt); 937 continue; 938 } 939 RTFREE_LOCKED(rt); 940 } 941 942 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 943 &addr->sin6_addr, &pr->ndpr_mask)) 944 return (1); 945 } 946 947 /* 948 * If the address is assigned on the node of the other side of 949 * a p2p interface, the address should be a neighbor. 950 */ 951 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr); 952 if (dstaddr != NULL) { 953 if (dstaddr->ifa_ifp == ifp) { 954 ifa_free(dstaddr); 955 return (1); 956 } 957 ifa_free(dstaddr); 958 } 959 960 /* 961 * If the default router list is empty, all addresses are regarded 962 * as on-link, and thus, as a neighbor. 963 */ 964 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && 965 TAILQ_FIRST(&V_nd_defrouter) == NULL && 966 V_nd6_defifindex == ifp->if_index) { 967 return (1); 968 } 969 970 return (0); 971 } 972 973 974 /* 975 * Detect if a given IPv6 address identifies a neighbor on a given link. 976 * XXX: should take care of the destination of a p2p link? 977 */ 978 int 979 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 980 { 981 struct llentry *lle; 982 int rc = 0; 983 984 IF_AFDATA_UNLOCK_ASSERT(ifp); 985 if (nd6_is_new_addr_neighbor(addr, ifp)) 986 return (1); 987 988 /* 989 * Even if the address matches none of our addresses, it might be 990 * in the neighbor cache. 991 */ 992 IF_AFDATA_LOCK(ifp); 993 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) { 994 LLE_RUNLOCK(lle); 995 rc = 1; 996 } 997 IF_AFDATA_UNLOCK(ifp); 998 return (rc); 999 } 1000 1001 /* 1002 * Free an nd6 llinfo entry. 1003 * Since the function would cause significant changes in the kernel, DO NOT 1004 * make it global, unless you have a strong reason for the change, and are sure 1005 * that the change is safe. 1006 */ 1007 static struct llentry * 1008 nd6_free(struct llentry *ln, int gc) 1009 { 1010 struct llentry *next; 1011 struct nd_defrouter *dr; 1012 struct ifnet *ifp; 1013 1014 LLE_WLOCK_ASSERT(ln); 1015 1016 /* 1017 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1018 * even though it is not harmful, it was not really necessary. 1019 */ 1020 1021 /* cancel timer */ 1022 nd6_llinfo_settimer_locked(ln, -1); 1023 1024 ifp = ln->lle_tbl->llt_ifp; 1025 1026 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1027 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1028 1029 if (dr != NULL && dr->expire && 1030 ln->ln_state == ND6_LLINFO_STALE && gc) { 1031 /* 1032 * If the reason for the deletion is just garbage 1033 * collection, and the neighbor is an active default 1034 * router, do not delete it. Instead, reset the GC 1035 * timer using the router's lifetime. 1036 * Simply deleting the entry would affect default 1037 * router selection, which is not necessarily a good 1038 * thing, especially when we're using router preference 1039 * values. 1040 * XXX: the check for ln_state would be redundant, 1041 * but we intentionally keep it just in case. 1042 */ 1043 if (dr->expire > time_second) 1044 nd6_llinfo_settimer_locked(ln, 1045 (dr->expire - time_second) * hz); 1046 else 1047 nd6_llinfo_settimer_locked(ln, 1048 (long)V_nd6_gctimer * hz); 1049 1050 next = LIST_NEXT(ln, lle_next); 1051 LLE_REMREF(ln); 1052 LLE_WUNLOCK(ln); 1053 return (next); 1054 } 1055 1056 if (dr) { 1057 /* 1058 * Unreachablity of a router might affect the default 1059 * router selection and on-link detection of advertised 1060 * prefixes. 1061 */ 1062 1063 /* 1064 * Temporarily fake the state to choose a new default 1065 * router and to perform on-link determination of 1066 * prefixes correctly. 1067 * Below the state will be set correctly, 1068 * or the entry itself will be deleted. 1069 */ 1070 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1071 } 1072 1073 if (ln->ln_router || dr) { 1074 1075 /* 1076 * We need to unlock to avoid a LOR with rt6_flush() with the 1077 * rnh and for the calls to pfxlist_onlink_check() and 1078 * defrouter_select() in the block further down for calls 1079 * into nd6_lookup(). We still hold a ref. 1080 */ 1081 LLE_WUNLOCK(ln); 1082 1083 /* 1084 * rt6_flush must be called whether or not the neighbor 1085 * is in the Default Router List. 1086 * See a corresponding comment in nd6_na_input(). 1087 */ 1088 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1089 } 1090 1091 if (dr) { 1092 /* 1093 * Since defrouter_select() does not affect the 1094 * on-link determination and MIP6 needs the check 1095 * before the default router selection, we perform 1096 * the check now. 1097 */ 1098 pfxlist_onlink_check(); 1099 1100 /* 1101 * Refresh default router list. 1102 */ 1103 defrouter_select(); 1104 } 1105 1106 if (ln->ln_router || dr) 1107 LLE_WLOCK(ln); 1108 } 1109 1110 /* 1111 * Before deleting the entry, remember the next entry as the 1112 * return value. We need this because pfxlist_onlink_check() above 1113 * might have freed other entries (particularly the old next entry) as 1114 * a side effect (XXX). 1115 */ 1116 next = LIST_NEXT(ln, lle_next); 1117 1118 /* 1119 * Save to unlock. We still hold an extra reference and will not 1120 * free(9) in llentry_free() if someone else holds one as well. 1121 */ 1122 LLE_WUNLOCK(ln); 1123 IF_AFDATA_LOCK(ifp); 1124 LLE_WLOCK(ln); 1125 LLE_REMREF(ln); 1126 llentry_free(ln); 1127 IF_AFDATA_UNLOCK(ifp); 1128 1129 return (next); 1130 } 1131 1132 /* 1133 * Upper-layer reachability hint for Neighbor Unreachability Detection. 1134 * 1135 * XXX cost-effective methods? 1136 */ 1137 void 1138 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force) 1139 { 1140 struct llentry *ln; 1141 struct ifnet *ifp; 1142 1143 if ((dst6 == NULL) || (rt == NULL)) 1144 return; 1145 1146 ifp = rt->rt_ifp; 1147 IF_AFDATA_LOCK(ifp); 1148 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL); 1149 IF_AFDATA_UNLOCK(ifp); 1150 if (ln == NULL) 1151 return; 1152 1153 if (ln->ln_state < ND6_LLINFO_REACHABLE) 1154 goto done; 1155 1156 /* 1157 * if we get upper-layer reachability confirmation many times, 1158 * it is possible we have false information. 1159 */ 1160 if (!force) { 1161 ln->ln_byhint++; 1162 if (ln->ln_byhint > V_nd6_maxnudhint) { 1163 goto done; 1164 } 1165 } 1166 1167 ln->ln_state = ND6_LLINFO_REACHABLE; 1168 if (!ND6_LLINFO_PERMANENT(ln)) { 1169 nd6_llinfo_settimer_locked(ln, 1170 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz); 1171 } 1172 done: 1173 LLE_WUNLOCK(ln); 1174 } 1175 1176 1177 int 1178 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) 1179 { 1180 struct in6_drlist *drl = (struct in6_drlist *)data; 1181 struct in6_oprlist *oprl = (struct in6_oprlist *)data; 1182 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1183 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1184 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1185 struct nd_defrouter *dr; 1186 struct nd_prefix *pr; 1187 int i = 0, error = 0; 1188 int s; 1189 1190 switch (cmd) { 1191 case SIOCGDRLST_IN6: 1192 /* 1193 * obsolete API, use sysctl under net.inet6.icmp6 1194 */ 1195 bzero(drl, sizeof(*drl)); 1196 s = splnet(); 1197 dr = TAILQ_FIRST(&V_nd_defrouter); 1198 while (dr && i < DRLSTSIZ) { 1199 drl->defrouter[i].rtaddr = dr->rtaddr; 1200 in6_clearscope(&drl->defrouter[i].rtaddr); 1201 1202 drl->defrouter[i].flags = dr->flags; 1203 drl->defrouter[i].rtlifetime = dr->rtlifetime; 1204 drl->defrouter[i].expire = dr->expire; 1205 drl->defrouter[i].if_index = dr->ifp->if_index; 1206 i++; 1207 dr = TAILQ_NEXT(dr, dr_entry); 1208 } 1209 splx(s); 1210 break; 1211 case SIOCGPRLST_IN6: 1212 /* 1213 * obsolete API, use sysctl under net.inet6.icmp6 1214 * 1215 * XXX the structure in6_prlist was changed in backward- 1216 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6, 1217 * in6_prlist is used for nd6_sysctl() - fill_prlist(). 1218 */ 1219 /* 1220 * XXX meaning of fields, especialy "raflags", is very 1221 * differnet between RA prefix list and RR/static prefix list. 1222 * how about separating ioctls into two? 1223 */ 1224 bzero(oprl, sizeof(*oprl)); 1225 s = splnet(); 1226 pr = V_nd_prefix.lh_first; 1227 while (pr && i < PRLSTSIZ) { 1228 struct nd_pfxrouter *pfr; 1229 int j; 1230 1231 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr; 1232 oprl->prefix[i].raflags = pr->ndpr_raf; 1233 oprl->prefix[i].prefixlen = pr->ndpr_plen; 1234 oprl->prefix[i].vltime = pr->ndpr_vltime; 1235 oprl->prefix[i].pltime = pr->ndpr_pltime; 1236 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index; 1237 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 1238 oprl->prefix[i].expire = 0; 1239 else { 1240 time_t maxexpire; 1241 1242 /* XXX: we assume time_t is signed. */ 1243 maxexpire = (-1) & 1244 ~((time_t)1 << 1245 ((sizeof(maxexpire) * 8) - 1)); 1246 if (pr->ndpr_vltime < 1247 maxexpire - pr->ndpr_lastupdate) { 1248 oprl->prefix[i].expire = 1249 pr->ndpr_lastupdate + 1250 pr->ndpr_vltime; 1251 } else 1252 oprl->prefix[i].expire = maxexpire; 1253 } 1254 1255 pfr = pr->ndpr_advrtrs.lh_first; 1256 j = 0; 1257 while (pfr) { 1258 if (j < DRLSTSIZ) { 1259 #define RTRADDR oprl->prefix[i].advrtr[j] 1260 RTRADDR = pfr->router->rtaddr; 1261 in6_clearscope(&RTRADDR); 1262 #undef RTRADDR 1263 } 1264 j++; 1265 pfr = pfr->pfr_next; 1266 } 1267 oprl->prefix[i].advrtrs = j; 1268 oprl->prefix[i].origin = PR_ORIG_RA; 1269 1270 i++; 1271 pr = pr->ndpr_next; 1272 } 1273 splx(s); 1274 1275 break; 1276 case OSIOCGIFINFO_IN6: 1277 #define ND ndi->ndi 1278 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1279 bzero(&ND, sizeof(ND)); 1280 ND.linkmtu = IN6_LINKMTU(ifp); 1281 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1282 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1283 ND.reachable = ND_IFINFO(ifp)->reachable; 1284 ND.retrans = ND_IFINFO(ifp)->retrans; 1285 ND.flags = ND_IFINFO(ifp)->flags; 1286 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1287 ND.chlim = ND_IFINFO(ifp)->chlim; 1288 break; 1289 case SIOCGIFINFO_IN6: 1290 ND = *ND_IFINFO(ifp); 1291 break; 1292 case SIOCSIFINFO_IN6: 1293 /* 1294 * used to change host variables from userland. 1295 * intented for a use on router to reflect RA configurations. 1296 */ 1297 /* 0 means 'unspecified' */ 1298 if (ND.linkmtu != 0) { 1299 if (ND.linkmtu < IPV6_MMTU || 1300 ND.linkmtu > IN6_LINKMTU(ifp)) { 1301 error = EINVAL; 1302 break; 1303 } 1304 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1305 } 1306 1307 if (ND.basereachable != 0) { 1308 int obasereachable = ND_IFINFO(ifp)->basereachable; 1309 1310 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1311 if (ND.basereachable != obasereachable) 1312 ND_IFINFO(ifp)->reachable = 1313 ND_COMPUTE_RTIME(ND.basereachable); 1314 } 1315 if (ND.retrans != 0) 1316 ND_IFINFO(ifp)->retrans = ND.retrans; 1317 if (ND.chlim != 0) 1318 ND_IFINFO(ifp)->chlim = ND.chlim; 1319 /* FALLTHROUGH */ 1320 case SIOCSIFINFO_FLAGS: 1321 { 1322 struct ifaddr *ifa; 1323 struct in6_ifaddr *ia; 1324 1325 /* 1326 * Try to clear ifdisabled flag when enabling 1327 * accept_rtadv or auto_linklocal. 1328 */ 1329 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1330 !(ND.flags & ND6_IFF_IFDISABLED) && 1331 (ND.flags & (ND6_IFF_ACCEPT_RTADV | 1332 ND6_IFF_AUTO_LINKLOCAL))) 1333 ND.flags &= ~ND6_IFF_IFDISABLED; 1334 1335 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1336 !(ND.flags & ND6_IFF_IFDISABLED)) { 1337 /* ifdisabled 1->0 transision */ 1338 1339 /* 1340 * If the interface is marked as ND6_IFF_IFDISABLED and 1341 * has an link-local address with IN6_IFF_DUPLICATED, 1342 * do not clear ND6_IFF_IFDISABLED. 1343 * See RFC 4862, Section 5.4.5. 1344 */ 1345 int duplicated_linklocal = 0; 1346 1347 IF_ADDR_LOCK(ifp); 1348 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1349 if (ifa->ifa_addr->sa_family != AF_INET6) 1350 continue; 1351 ia = (struct in6_ifaddr *)ifa; 1352 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) && 1353 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1354 duplicated_linklocal = 1; 1355 break; 1356 } 1357 } 1358 IF_ADDR_UNLOCK(ifp); 1359 1360 if (duplicated_linklocal) { 1361 ND.flags |= ND6_IFF_IFDISABLED; 1362 log(LOG_ERR, "Cannot enable an interface" 1363 " with a link-local address marked" 1364 " duplicate.\n"); 1365 } else { 1366 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED; 1367 in6_if_up(ifp); 1368 } 1369 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1370 (ND.flags & ND6_IFF_IFDISABLED)) { 1371 /* ifdisabled 0->1 transision */ 1372 /* Mark all IPv6 address as tentative. */ 1373 1374 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; 1375 IF_ADDR_LOCK(ifp); 1376 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1377 if (ifa->ifa_addr->sa_family != AF_INET6) 1378 continue; 1379 ia = (struct in6_ifaddr *)ifa; 1380 ia->ia6_flags |= IN6_IFF_TENTATIVE; 1381 } 1382 IF_ADDR_UNLOCK(ifp); 1383 } 1384 1385 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL) && 1386 (ND.flags & ND6_IFF_AUTO_LINKLOCAL)) { 1387 /* auto_linklocal 0->1 transision */ 1388 1389 /* If no link-local address on ifp, configure */ 1390 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL; 1391 in6_ifattach(ifp, NULL); 1392 } else if ((ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL) && 1393 !(ND.flags & ND6_IFF_IFDISABLED)) { 1394 /* 1395 * When the IF already has 1396 * ND6_IFF_AUTO_LINKLOCAL and no link-local 1397 * address is assigned, try to assign one. 1398 */ 1399 int haslinklocal = 0; 1400 1401 IF_ADDR_LOCK(ifp); 1402 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1403 if (ifa->ifa_addr->sa_family != AF_INET6) 1404 continue; 1405 ia = (struct in6_ifaddr *)ifa; 1406 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1407 haslinklocal = 1; 1408 break; 1409 } 1410 } 1411 IF_ADDR_UNLOCK(ifp); 1412 if (!haslinklocal) 1413 in6_ifattach(ifp, NULL); 1414 } 1415 } 1416 ND_IFINFO(ifp)->flags = ND.flags; 1417 break; 1418 #undef ND 1419 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1420 /* sync kernel routing table with the default router list */ 1421 defrouter_reset(); 1422 defrouter_select(); 1423 break; 1424 case SIOCSPFXFLUSH_IN6: 1425 { 1426 /* flush all the prefix advertised by routers */ 1427 struct nd_prefix *pr, *next; 1428 1429 s = splnet(); 1430 for (pr = V_nd_prefix.lh_first; pr; pr = next) { 1431 struct in6_ifaddr *ia, *ia_next; 1432 1433 next = pr->ndpr_next; 1434 1435 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1436 continue; /* XXX */ 1437 1438 /* do we really have to remove addresses as well? */ 1439 /* XXXRW: in6_ifaddrhead locking. */ 1440 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link, 1441 ia_next) { 1442 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1443 continue; 1444 1445 if (ia->ia6_ndpr == pr) 1446 in6_purgeaddr(&ia->ia_ifa); 1447 } 1448 prelist_remove(pr); 1449 } 1450 splx(s); 1451 break; 1452 } 1453 case SIOCSRTRFLUSH_IN6: 1454 { 1455 /* flush all the default routers */ 1456 struct nd_defrouter *dr, *next; 1457 1458 s = splnet(); 1459 defrouter_reset(); 1460 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) { 1461 next = TAILQ_NEXT(dr, dr_entry); 1462 defrtrlist_del(dr); 1463 } 1464 defrouter_select(); 1465 splx(s); 1466 break; 1467 } 1468 case SIOCGNBRINFO_IN6: 1469 { 1470 struct llentry *ln; 1471 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1472 1473 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1474 return (error); 1475 1476 IF_AFDATA_LOCK(ifp); 1477 ln = nd6_lookup(&nb_addr, 0, ifp); 1478 IF_AFDATA_UNLOCK(ifp); 1479 1480 if (ln == NULL) { 1481 error = EINVAL; 1482 break; 1483 } 1484 nbi->state = ln->ln_state; 1485 nbi->asked = ln->la_asked; 1486 nbi->isrouter = ln->ln_router; 1487 nbi->expire = ln->la_expire; 1488 LLE_RUNLOCK(ln); 1489 break; 1490 } 1491 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1492 ndif->ifindex = V_nd6_defifindex; 1493 break; 1494 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1495 return (nd6_setdefaultiface(ndif->ifindex)); 1496 } 1497 return (error); 1498 } 1499 1500 /* 1501 * Create neighbor cache entry and cache link-layer address, 1502 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1503 * 1504 * type - ICMP6 type 1505 * code - type dependent information 1506 * 1507 * XXXXX 1508 * The caller of this function already acquired the ndp 1509 * cache table lock because the cache entry is returned. 1510 */ 1511 struct llentry * 1512 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, 1513 int lladdrlen, int type, int code) 1514 { 1515 struct llentry *ln = NULL; 1516 int is_newentry; 1517 int do_update; 1518 int olladdr; 1519 int llchange; 1520 int flags; 1521 int newstate = 0; 1522 uint16_t router = 0; 1523 struct sockaddr_in6 sin6; 1524 struct mbuf *chain = NULL; 1525 int static_route = 0; 1526 1527 IF_AFDATA_UNLOCK_ASSERT(ifp); 1528 1529 if (ifp == NULL) 1530 panic("ifp == NULL in nd6_cache_lladdr"); 1531 if (from == NULL) 1532 panic("from == NULL in nd6_cache_lladdr"); 1533 1534 /* nothing must be updated for unspecified address */ 1535 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1536 return NULL; 1537 1538 /* 1539 * Validation about ifp->if_addrlen and lladdrlen must be done in 1540 * the caller. 1541 * 1542 * XXX If the link does not have link-layer adderss, what should 1543 * we do? (ifp->if_addrlen == 0) 1544 * Spec says nothing in sections for RA, RS and NA. There's small 1545 * description on it in NS section (RFC 2461 7.2.3). 1546 */ 1547 flags = lladdr ? ND6_EXCLUSIVE : 0; 1548 IF_AFDATA_LOCK(ifp); 1549 ln = nd6_lookup(from, flags, ifp); 1550 1551 if (ln == NULL) { 1552 flags |= ND6_EXCLUSIVE; 1553 ln = nd6_lookup(from, flags | ND6_CREATE, ifp); 1554 IF_AFDATA_UNLOCK(ifp); 1555 is_newentry = 1; 1556 } else { 1557 IF_AFDATA_UNLOCK(ifp); 1558 /* do nothing if static ndp is set */ 1559 if (ln->la_flags & LLE_STATIC) { 1560 static_route = 1; 1561 goto done; 1562 } 1563 is_newentry = 0; 1564 } 1565 if (ln == NULL) 1566 return (NULL); 1567 1568 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0; 1569 if (olladdr && lladdr) { 1570 llchange = bcmp(lladdr, &ln->ll_addr, 1571 ifp->if_addrlen); 1572 } else 1573 llchange = 0; 1574 1575 /* 1576 * newentry olladdr lladdr llchange (*=record) 1577 * 0 n n -- (1) 1578 * 0 y n -- (2) 1579 * 0 n y -- (3) * STALE 1580 * 0 y y n (4) * 1581 * 0 y y y (5) * STALE 1582 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1583 * 1 -- y -- (7) * STALE 1584 */ 1585 1586 if (lladdr) { /* (3-5) and (7) */ 1587 /* 1588 * Record source link-layer address 1589 * XXX is it dependent to ifp->if_type? 1590 */ 1591 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen); 1592 ln->la_flags |= LLE_VALID; 1593 } 1594 1595 if (!is_newentry) { 1596 if ((!olladdr && lladdr != NULL) || /* (3) */ 1597 (olladdr && lladdr != NULL && llchange)) { /* (5) */ 1598 do_update = 1; 1599 newstate = ND6_LLINFO_STALE; 1600 } else /* (1-2,4) */ 1601 do_update = 0; 1602 } else { 1603 do_update = 1; 1604 if (lladdr == NULL) /* (6) */ 1605 newstate = ND6_LLINFO_NOSTATE; 1606 else /* (7) */ 1607 newstate = ND6_LLINFO_STALE; 1608 } 1609 1610 if (do_update) { 1611 /* 1612 * Update the state of the neighbor cache. 1613 */ 1614 ln->ln_state = newstate; 1615 1616 if (ln->ln_state == ND6_LLINFO_STALE) { 1617 /* 1618 * XXX: since nd6_output() below will cause 1619 * state tansition to DELAY and reset the timer, 1620 * we must set the timer now, although it is actually 1621 * meaningless. 1622 */ 1623 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 1624 1625 if (ln->la_hold) { 1626 struct mbuf *m_hold, *m_hold_next; 1627 1628 /* 1629 * reset the la_hold in advance, to explicitly 1630 * prevent a la_hold lookup in nd6_output() 1631 * (wouldn't happen, though...) 1632 */ 1633 for (m_hold = ln->la_hold, ln->la_hold = NULL; 1634 m_hold; m_hold = m_hold_next) { 1635 m_hold_next = m_hold->m_nextpkt; 1636 m_hold->m_nextpkt = NULL; 1637 1638 /* 1639 * we assume ifp is not a p2p here, so 1640 * just set the 2nd argument as the 1641 * 1st one. 1642 */ 1643 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain); 1644 } 1645 /* 1646 * If we have mbufs in the chain we need to do 1647 * deferred transmit. Copy the address from the 1648 * llentry before dropping the lock down below. 1649 */ 1650 if (chain != NULL) 1651 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6)); 1652 } 1653 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { 1654 /* probe right away */ 1655 nd6_llinfo_settimer_locked((void *)ln, 0); 1656 } 1657 } 1658 1659 /* 1660 * ICMP6 type dependent behavior. 1661 * 1662 * NS: clear IsRouter if new entry 1663 * RS: clear IsRouter 1664 * RA: set IsRouter if there's lladdr 1665 * redir: clear IsRouter if new entry 1666 * 1667 * RA case, (1): 1668 * The spec says that we must set IsRouter in the following cases: 1669 * - If lladdr exist, set IsRouter. This means (1-5). 1670 * - If it is old entry (!newentry), set IsRouter. This means (7). 1671 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1672 * A quetion arises for (1) case. (1) case has no lladdr in the 1673 * neighbor cache, this is similar to (6). 1674 * This case is rare but we figured that we MUST NOT set IsRouter. 1675 * 1676 * newentry olladdr lladdr llchange NS RS RA redir 1677 * D R 1678 * 0 n n -- (1) c ? s 1679 * 0 y n -- (2) c s s 1680 * 0 n y -- (3) c s s 1681 * 0 y y n (4) c s s 1682 * 0 y y y (5) c s s 1683 * 1 -- n -- (6) c c c s 1684 * 1 -- y -- (7) c c s c s 1685 * 1686 * (c=clear s=set) 1687 */ 1688 switch (type & 0xff) { 1689 case ND_NEIGHBOR_SOLICIT: 1690 /* 1691 * New entry must have is_router flag cleared. 1692 */ 1693 if (is_newentry) /* (6-7) */ 1694 ln->ln_router = 0; 1695 break; 1696 case ND_REDIRECT: 1697 /* 1698 * If the icmp is a redirect to a better router, always set the 1699 * is_router flag. Otherwise, if the entry is newly created, 1700 * clear the flag. [RFC 2461, sec 8.3] 1701 */ 1702 if (code == ND_REDIRECT_ROUTER) 1703 ln->ln_router = 1; 1704 else if (is_newentry) /* (6-7) */ 1705 ln->ln_router = 0; 1706 break; 1707 case ND_ROUTER_SOLICIT: 1708 /* 1709 * is_router flag must always be cleared. 1710 */ 1711 ln->ln_router = 0; 1712 break; 1713 case ND_ROUTER_ADVERT: 1714 /* 1715 * Mark an entry with lladdr as a router. 1716 */ 1717 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */ 1718 (is_newentry && lladdr)) { /* (7) */ 1719 ln->ln_router = 1; 1720 } 1721 break; 1722 } 1723 1724 if (ln != NULL) { 1725 static_route = (ln->la_flags & LLE_STATIC); 1726 router = ln->ln_router; 1727 1728 if (flags & ND6_EXCLUSIVE) 1729 LLE_WUNLOCK(ln); 1730 else 1731 LLE_RUNLOCK(ln); 1732 if (static_route) 1733 ln = NULL; 1734 } 1735 if (chain) 1736 nd6_output_flush(ifp, ifp, chain, &sin6, NULL); 1737 1738 /* 1739 * When the link-layer address of a router changes, select the 1740 * best router again. In particular, when the neighbor entry is newly 1741 * created, it might affect the selection policy. 1742 * Question: can we restrict the first condition to the "is_newentry" 1743 * case? 1744 * XXX: when we hear an RA from a new router with the link-layer 1745 * address option, defrouter_select() is called twice, since 1746 * defrtrlist_update called the function as well. However, I believe 1747 * we can compromise the overhead, since it only happens the first 1748 * time. 1749 * XXX: although defrouter_select() should not have a bad effect 1750 * for those are not autoconfigured hosts, we explicitly avoid such 1751 * cases for safety. 1752 */ 1753 if (do_update && router && 1754 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1755 /* 1756 * guaranteed recursion 1757 */ 1758 defrouter_select(); 1759 } 1760 1761 return (ln); 1762 done: 1763 if (ln != NULL) { 1764 if (flags & ND6_EXCLUSIVE) 1765 LLE_WUNLOCK(ln); 1766 else 1767 LLE_RUNLOCK(ln); 1768 if (static_route) 1769 ln = NULL; 1770 } 1771 return (ln); 1772 } 1773 1774 static void 1775 nd6_slowtimo(void *arg) 1776 { 1777 CURVNET_SET((struct vnet *) arg); 1778 struct nd_ifinfo *nd6if; 1779 struct ifnet *ifp; 1780 1781 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 1782 nd6_slowtimo, curvnet); 1783 IFNET_RLOCK_NOSLEEP(); 1784 for (ifp = TAILQ_FIRST(&V_ifnet); ifp; 1785 ifp = TAILQ_NEXT(ifp, if_list)) { 1786 nd6if = ND_IFINFO(ifp); 1787 if (nd6if->basereachable && /* already initialized */ 1788 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 1789 /* 1790 * Since reachable time rarely changes by router 1791 * advertisements, we SHOULD insure that a new random 1792 * value gets recomputed at least once every few hours. 1793 * (RFC 2461, 6.3.4) 1794 */ 1795 nd6if->recalctm = V_nd6_recalc_reachtm_interval; 1796 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 1797 } 1798 } 1799 IFNET_RUNLOCK_NOSLEEP(); 1800 CURVNET_RESTORE(); 1801 } 1802 1803 int 1804 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 1805 struct sockaddr_in6 *dst, struct rtentry *rt0) 1806 { 1807 1808 return (nd6_output_lle(ifp, origifp, m0, dst, rt0, NULL, NULL)); 1809 } 1810 1811 1812 /* 1813 * Note that I'm not enforcing any global serialization 1814 * lle state or asked changes here as the logic is too 1815 * complicated to avoid having to always acquire an exclusive 1816 * lock 1817 * KMM 1818 * 1819 */ 1820 #define senderr(e) { error = (e); goto bad;} 1821 1822 int 1823 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 1824 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle, 1825 struct mbuf **chain) 1826 { 1827 struct mbuf *m = m0; 1828 struct m_tag *mtag; 1829 struct llentry *ln = lle; 1830 struct ip6_hdr *ip6; 1831 int error = 0; 1832 int flags = 0; 1833 int ip6len; 1834 1835 #ifdef INVARIANTS 1836 if (lle != NULL) { 1837 1838 LLE_WLOCK_ASSERT(lle); 1839 1840 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed")); 1841 } 1842 #endif 1843 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) 1844 goto sendpkt; 1845 1846 if (nd6_need_cache(ifp) == 0) 1847 goto sendpkt; 1848 1849 /* 1850 * next hop determination. This routine is derived from ether_output. 1851 */ 1852 1853 /* 1854 * Address resolution or Neighbor Unreachability Detection 1855 * for the next hop. 1856 * At this point, the destination of the packet must be a unicast 1857 * or an anycast address(i.e. not a multicast). 1858 */ 1859 1860 flags = ((m != NULL) || (lle != NULL)) ? LLE_EXCLUSIVE : 0; 1861 if (ln == NULL) { 1862 retry: 1863 IF_AFDATA_LOCK(ifp); 1864 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)dst); 1865 IF_AFDATA_UNLOCK(ifp); 1866 if ((ln == NULL) && nd6_is_addr_neighbor(dst, ifp)) { 1867 /* 1868 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 1869 * the condition below is not very efficient. But we believe 1870 * it is tolerable, because this should be a rare case. 1871 */ 1872 flags = ND6_CREATE | (m ? ND6_EXCLUSIVE : 0); 1873 IF_AFDATA_LOCK(ifp); 1874 ln = nd6_lookup(&dst->sin6_addr, flags, ifp); 1875 IF_AFDATA_UNLOCK(ifp); 1876 } 1877 } 1878 if (ln == NULL) { 1879 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && 1880 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 1881 char ip6buf[INET6_ADDRSTRLEN]; 1882 log(LOG_DEBUG, 1883 "nd6_output: can't allocate llinfo for %s " 1884 "(ln=%p)\n", 1885 ip6_sprintf(ip6buf, &dst->sin6_addr), ln); 1886 senderr(EIO); /* XXX: good error? */ 1887 } 1888 goto sendpkt; /* send anyway */ 1889 } 1890 1891 /* We don't have to do link-layer address resolution on a p2p link. */ 1892 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && 1893 ln->ln_state < ND6_LLINFO_REACHABLE) { 1894 if ((flags & LLE_EXCLUSIVE) == 0) { 1895 flags |= LLE_EXCLUSIVE; 1896 goto retry; 1897 } 1898 ln->ln_state = ND6_LLINFO_STALE; 1899 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 1900 } 1901 1902 /* 1903 * The first time we send a packet to a neighbor whose entry is 1904 * STALE, we have to change the state to DELAY and a sets a timer to 1905 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 1906 * neighbor unreachability detection on expiration. 1907 * (RFC 2461 7.3.3) 1908 */ 1909 if (ln->ln_state == ND6_LLINFO_STALE) { 1910 if ((flags & LLE_EXCLUSIVE) == 0) { 1911 flags |= LLE_EXCLUSIVE; 1912 LLE_RUNLOCK(ln); 1913 goto retry; 1914 } 1915 ln->la_asked = 0; 1916 ln->ln_state = ND6_LLINFO_DELAY; 1917 nd6_llinfo_settimer_locked(ln, (long)V_nd6_delay * hz); 1918 } 1919 1920 /* 1921 * If the neighbor cache entry has a state other than INCOMPLETE 1922 * (i.e. its link-layer address is already resolved), just 1923 * send the packet. 1924 */ 1925 if (ln->ln_state > ND6_LLINFO_INCOMPLETE) 1926 goto sendpkt; 1927 1928 /* 1929 * There is a neighbor cache entry, but no ethernet address 1930 * response yet. Append this latest packet to the end of the 1931 * packet queue in the mbuf, unless the number of the packet 1932 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 1933 * the oldest packet in the queue will be removed. 1934 */ 1935 if (ln->ln_state == ND6_LLINFO_NOSTATE) 1936 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1937 1938 if ((flags & LLE_EXCLUSIVE) == 0) { 1939 flags |= LLE_EXCLUSIVE; 1940 LLE_RUNLOCK(ln); 1941 goto retry; 1942 } 1943 1944 LLE_WLOCK_ASSERT(ln); 1945 1946 if (ln->la_hold) { 1947 struct mbuf *m_hold; 1948 int i; 1949 1950 i = 0; 1951 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold->m_nextpkt) { 1952 i++; 1953 if (m_hold->m_nextpkt == NULL) { 1954 m_hold->m_nextpkt = m; 1955 break; 1956 } 1957 } 1958 while (i >= V_nd6_maxqueuelen) { 1959 m_hold = ln->la_hold; 1960 ln->la_hold = ln->la_hold->m_nextpkt; 1961 m_freem(m_hold); 1962 i--; 1963 } 1964 } else { 1965 ln->la_hold = m; 1966 } 1967 1968 /* 1969 * If there has been no NS for the neighbor after entering the 1970 * INCOMPLETE state, send the first solicitation. 1971 */ 1972 if (!ND6_LLINFO_PERMANENT(ln) && ln->la_asked == 0) { 1973 ln->la_asked++; 1974 1975 nd6_llinfo_settimer_locked(ln, 1976 (long)ND_IFINFO(ifp)->retrans * hz / 1000); 1977 LLE_WUNLOCK(ln); 1978 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); 1979 if (lle != NULL && ln == lle) 1980 LLE_WLOCK(lle); 1981 1982 } else if (lle == NULL || ln != lle) { 1983 /* 1984 * We did the lookup (no lle arg) so we 1985 * need to do the unlock here. 1986 */ 1987 LLE_WUNLOCK(ln); 1988 } 1989 1990 return (0); 1991 1992 sendpkt: 1993 /* discard the packet if IPv6 operation is disabled on the interface */ 1994 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 1995 error = ENETDOWN; /* better error? */ 1996 goto bad; 1997 } 1998 /* 1999 * ln is valid and the caller did not pass in 2000 * an llentry 2001 */ 2002 if ((ln != NULL) && (lle == NULL)) { 2003 if (flags & LLE_EXCLUSIVE) 2004 LLE_WUNLOCK(ln); 2005 else 2006 LLE_RUNLOCK(ln); 2007 } 2008 2009 #ifdef MAC 2010 mac_netinet6_nd6_send(ifp, m); 2011 #endif 2012 2013 /* 2014 * If called from nd6_ns_output() (NS), nd6_na_output() (NA), 2015 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA 2016 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND 2017 * to be diverted to user space. When re-injected into the kernel, 2018 * send_output() will directly dispatch them to the outgoing interface. 2019 */ 2020 if (send_sendso_input_hook != NULL) { 2021 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); 2022 if (mtag != NULL) { 2023 ip6 = mtod(m, struct ip6_hdr *); 2024 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); 2025 /* Use the SEND socket */ 2026 error = send_sendso_input_hook(m, ifp, SND_OUT, 2027 ip6len); 2028 /* -1 == no app on SEND socket */ 2029 if (error == 0 || error != -1) 2030 return (error); 2031 } 2032 } 2033 2034 /* 2035 * We were passed in a pointer to an lle with the lock held 2036 * this means that we can't call if_output as we will 2037 * recurse on the lle lock - so what we do is we create 2038 * a list of mbufs to send and transmit them in the caller 2039 * after the lock is dropped 2040 */ 2041 if (lle != NULL) { 2042 if (*chain == NULL) 2043 *chain = m; 2044 else { 2045 struct mbuf *mb; 2046 2047 /* 2048 * append mbuf to end of deferred chain 2049 */ 2050 mb = *chain; 2051 while (mb->m_nextpkt != NULL) 2052 mb = mb->m_nextpkt; 2053 mb->m_nextpkt = m; 2054 } 2055 return (error); 2056 } 2057 if ((ifp->if_flags & IFF_LOOPBACK) != 0) { 2058 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst, 2059 NULL)); 2060 } 2061 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL); 2062 return (error); 2063 2064 bad: 2065 /* 2066 * ln is valid and the caller did not pass in 2067 * an llentry 2068 */ 2069 if ((ln != NULL) && (lle == NULL)) { 2070 if (flags & LLE_EXCLUSIVE) 2071 LLE_WUNLOCK(ln); 2072 else 2073 LLE_RUNLOCK(ln); 2074 } 2075 if (m) 2076 m_freem(m); 2077 return (error); 2078 } 2079 #undef senderr 2080 2081 2082 int 2083 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain, 2084 struct sockaddr_in6 *dst, struct route *ro) 2085 { 2086 struct mbuf *m, *m_head; 2087 struct ifnet *outifp; 2088 int error = 0; 2089 2090 m_head = chain; 2091 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2092 outifp = origifp; 2093 else 2094 outifp = ifp; 2095 2096 while (m_head) { 2097 m = m_head; 2098 m_head = m_head->m_nextpkt; 2099 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, ro); 2100 } 2101 2102 /* 2103 * XXX 2104 * note that intermediate errors are blindly ignored - but this is 2105 * the same convention as used with nd6_output when called by 2106 * nd6_cache_lladdr 2107 */ 2108 return (error); 2109 } 2110 2111 2112 int 2113 nd6_need_cache(struct ifnet *ifp) 2114 { 2115 /* 2116 * XXX: we currently do not make neighbor cache on any interface 2117 * other than ARCnet, Ethernet, FDDI and GIF. 2118 * 2119 * RFC2893 says: 2120 * - unidirectional tunnels needs no ND 2121 */ 2122 switch (ifp->if_type) { 2123 case IFT_ARCNET: 2124 case IFT_ETHER: 2125 case IFT_FDDI: 2126 case IFT_IEEE1394: 2127 #ifdef IFT_L2VLAN 2128 case IFT_L2VLAN: 2129 #endif 2130 #ifdef IFT_IEEE80211 2131 case IFT_IEEE80211: 2132 #endif 2133 #ifdef IFT_CARP 2134 case IFT_CARP: 2135 #endif 2136 case IFT_INFINIBAND: 2137 case IFT_GIF: /* XXX need more cases? */ 2138 case IFT_PPP: 2139 case IFT_TUNNEL: 2140 case IFT_BRIDGE: 2141 case IFT_PROPVIRTUAL: 2142 return (1); 2143 default: 2144 return (0); 2145 } 2146 } 2147 2148 /* 2149 * the callers of this function need to be re-worked to drop 2150 * the lle lock, drop here for now 2151 */ 2152 int 2153 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m, 2154 struct sockaddr *dst, u_char *desten, struct llentry **lle) 2155 { 2156 struct llentry *ln; 2157 2158 *lle = NULL; 2159 IF_AFDATA_UNLOCK_ASSERT(ifp); 2160 if (m->m_flags & M_MCAST) { 2161 int i; 2162 2163 switch (ifp->if_type) { 2164 case IFT_ETHER: 2165 case IFT_FDDI: 2166 #ifdef IFT_L2VLAN 2167 case IFT_L2VLAN: 2168 #endif 2169 #ifdef IFT_IEEE80211 2170 case IFT_IEEE80211: 2171 #endif 2172 case IFT_BRIDGE: 2173 case IFT_ISO88025: 2174 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr, 2175 desten); 2176 return (0); 2177 case IFT_IEEE1394: 2178 /* 2179 * netbsd can use if_broadcastaddr, but we don't do so 2180 * to reduce # of ifdef. 2181 */ 2182 for (i = 0; i < ifp->if_addrlen; i++) 2183 desten[i] = ~0; 2184 return (0); 2185 case IFT_ARCNET: 2186 *desten = 0; 2187 return (0); 2188 default: 2189 m_freem(m); 2190 return (EAFNOSUPPORT); 2191 } 2192 } 2193 2194 2195 /* 2196 * the entry should have been created in nd6_store_lladdr 2197 */ 2198 IF_AFDATA_LOCK(ifp); 2199 ln = lla_lookup(LLTABLE6(ifp), 0, dst); 2200 IF_AFDATA_UNLOCK(ifp); 2201 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) { 2202 if (ln != NULL) 2203 LLE_RUNLOCK(ln); 2204 /* this could happen, if we could not allocate memory */ 2205 m_freem(m); 2206 return (1); 2207 } 2208 2209 bcopy(&ln->ll_addr, desten, ifp->if_addrlen); 2210 *lle = ln; 2211 LLE_RUNLOCK(ln); 2212 /* 2213 * A *small* use after free race exists here 2214 */ 2215 return (0); 2216 } 2217 2218 static void 2219 clear_llinfo_pqueue(struct llentry *ln) 2220 { 2221 struct mbuf *m_hold, *m_hold_next; 2222 2223 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) { 2224 m_hold_next = m_hold->m_nextpkt; 2225 m_hold->m_nextpkt = NULL; 2226 m_freem(m_hold); 2227 } 2228 2229 ln->la_hold = NULL; 2230 return; 2231 } 2232 2233 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2234 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2235 #ifdef SYSCTL_DECL 2236 SYSCTL_DECL(_net_inet6_icmp6); 2237 #endif 2238 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2239 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2240 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2241 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2242 SYSCTL_VNET_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2243 CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, ""); 2244 2245 static int 2246 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2247 { 2248 int error; 2249 char buf[1024] __aligned(4); 2250 struct in6_defrouter *d, *de; 2251 struct nd_defrouter *dr; 2252 2253 if (req->newptr) 2254 return EPERM; 2255 error = 0; 2256 2257 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; 2258 dr = TAILQ_NEXT(dr, dr_entry)) { 2259 d = (struct in6_defrouter *)buf; 2260 de = (struct in6_defrouter *)(buf + sizeof(buf)); 2261 2262 if (d + 1 <= de) { 2263 bzero(d, sizeof(*d)); 2264 d->rtaddr.sin6_family = AF_INET6; 2265 d->rtaddr.sin6_len = sizeof(d->rtaddr); 2266 d->rtaddr.sin6_addr = dr->rtaddr; 2267 error = sa6_recoverscope(&d->rtaddr); 2268 if (error != 0) 2269 return (error); 2270 d->flags = dr->flags; 2271 d->rtlifetime = dr->rtlifetime; 2272 d->expire = dr->expire; 2273 d->if_index = dr->ifp->if_index; 2274 } else 2275 panic("buffer too short"); 2276 2277 error = SYSCTL_OUT(req, buf, sizeof(*d)); 2278 if (error) 2279 break; 2280 } 2281 2282 return (error); 2283 } 2284 2285 static int 2286 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2287 { 2288 int error; 2289 char buf[1024] __aligned(4); 2290 struct in6_prefix *p, *pe; 2291 struct nd_prefix *pr; 2292 char ip6buf[INET6_ADDRSTRLEN]; 2293 2294 if (req->newptr) 2295 return EPERM; 2296 error = 0; 2297 2298 for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) { 2299 u_short advrtrs; 2300 size_t advance; 2301 struct sockaddr_in6 *sin6, *s6; 2302 struct nd_pfxrouter *pfr; 2303 2304 p = (struct in6_prefix *)buf; 2305 pe = (struct in6_prefix *)(buf + sizeof(buf)); 2306 2307 if (p + 1 <= pe) { 2308 bzero(p, sizeof(*p)); 2309 sin6 = (struct sockaddr_in6 *)(p + 1); 2310 2311 p->prefix = pr->ndpr_prefix; 2312 if (sa6_recoverscope(&p->prefix)) { 2313 log(LOG_ERR, 2314 "scope error in prefix list (%s)\n", 2315 ip6_sprintf(ip6buf, &p->prefix.sin6_addr)); 2316 /* XXX: press on... */ 2317 } 2318 p->raflags = pr->ndpr_raf; 2319 p->prefixlen = pr->ndpr_plen; 2320 p->vltime = pr->ndpr_vltime; 2321 p->pltime = pr->ndpr_pltime; 2322 p->if_index = pr->ndpr_ifp->if_index; 2323 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2324 p->expire = 0; 2325 else { 2326 time_t maxexpire; 2327 2328 /* XXX: we assume time_t is signed. */ 2329 maxexpire = (-1) & 2330 ~((time_t)1 << 2331 ((sizeof(maxexpire) * 8) - 1)); 2332 if (pr->ndpr_vltime < 2333 maxexpire - pr->ndpr_lastupdate) { 2334 p->expire = pr->ndpr_lastupdate + 2335 pr->ndpr_vltime; 2336 } else 2337 p->expire = maxexpire; 2338 } 2339 p->refcnt = pr->ndpr_refcnt; 2340 p->flags = pr->ndpr_stateflags; 2341 p->origin = PR_ORIG_RA; 2342 advrtrs = 0; 2343 for (pfr = pr->ndpr_advrtrs.lh_first; pfr; 2344 pfr = pfr->pfr_next) { 2345 if ((void *)&sin6[advrtrs + 1] > (void *)pe) { 2346 advrtrs++; 2347 continue; 2348 } 2349 s6 = &sin6[advrtrs]; 2350 bzero(s6, sizeof(*s6)); 2351 s6->sin6_family = AF_INET6; 2352 s6->sin6_len = sizeof(*sin6); 2353 s6->sin6_addr = pfr->router->rtaddr; 2354 if (sa6_recoverscope(s6)) { 2355 log(LOG_ERR, 2356 "scope error in " 2357 "prefix list (%s)\n", 2358 ip6_sprintf(ip6buf, 2359 &pfr->router->rtaddr)); 2360 } 2361 advrtrs++; 2362 } 2363 p->advrtrs = advrtrs; 2364 } else 2365 panic("buffer too short"); 2366 2367 advance = sizeof(*p) + sizeof(*sin6) * advrtrs; 2368 error = SYSCTL_OUT(req, buf, advance); 2369 if (error) 2370 break; 2371 } 2372 2373 return (error); 2374 } 2375