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 #include "opt_mac.h" 38 39 #include <sys/param.h> 40 #include <sys/systm.h> 41 #include <sys/callout.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/socket.h> 45 #include <sys/sockio.h> 46 #include <sys/time.h> 47 #include <sys/kernel.h> 48 #include <sys/protosw.h> 49 #include <sys/errno.h> 50 #include <sys/syslog.h> 51 #include <sys/queue.h> 52 #include <sys/sysctl.h> 53 54 #include <net/if.h> 55 #include <net/if_arc.h> 56 #include <net/if_dl.h> 57 #include <net/if_types.h> 58 #include <net/iso88025.h> 59 #include <net/fddi.h> 60 #include <net/route.h> 61 62 #include <netinet/in.h> 63 #include <netinet/if_ether.h> 64 #include <netinet6/in6_var.h> 65 #include <netinet/ip6.h> 66 #include <netinet6/ip6_var.h> 67 #include <netinet6/scope6_var.h> 68 #include <netinet6/nd6.h> 69 #include <netinet/icmp6.h> 70 71 #include <sys/limits.h> 72 #include <sys/vimage.h> 73 74 #include <security/mac/mac_framework.h> 75 76 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ 77 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ 78 79 #define SIN6(s) ((struct sockaddr_in6 *)s) 80 #define SDL(s) ((struct sockaddr_dl *)s) 81 82 /* timer values */ 83 int nd6_prune = 1; /* walk list every 1 seconds */ 84 int nd6_delay = 5; /* delay first probe time 5 second */ 85 int nd6_umaxtries = 3; /* maximum unicast query */ 86 int nd6_mmaxtries = 3; /* maximum multicast query */ 87 int nd6_useloopback = 1; /* use loopback interface for local traffic */ 88 int nd6_gctimer = (60 * 60 * 24); /* 1 day: garbage collection timer */ 89 90 /* preventing too many loops in ND option parsing */ 91 int nd6_maxndopt = 10; /* max # of ND options allowed */ 92 93 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */ 94 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */ 95 96 #ifdef ND6_DEBUG 97 int nd6_debug = 1; 98 #else 99 int nd6_debug = 0; 100 #endif 101 102 /* for debugging? */ 103 static int nd6_inuse, nd6_allocated; 104 105 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6}; 106 struct nd_drhead nd_defrouter; 107 struct nd_prhead nd_prefix = { 0 }; 108 109 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL; 110 static struct sockaddr_in6 all1_sa; 111 112 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *, 113 struct ifnet *)); 114 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *); 115 static void nd6_slowtimo(void *); 116 static int regen_tmpaddr(struct in6_ifaddr *); 117 static struct llinfo_nd6 *nd6_free(struct rtentry *, int); 118 static void nd6_llinfo_timer(void *); 119 static void clear_llinfo_pqueue(struct llinfo_nd6 *); 120 121 struct callout nd6_slowtimo_ch; 122 struct callout nd6_timer_ch; 123 extern struct callout in6_tmpaddrtimer_ch; 124 125 void 126 nd6_init(void) 127 { 128 static int nd6_init_done = 0; 129 int i; 130 131 if (nd6_init_done) { 132 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n"); 133 return; 134 } 135 136 all1_sa.sin6_family = AF_INET6; 137 all1_sa.sin6_len = sizeof(struct sockaddr_in6); 138 for (i = 0; i < sizeof(all1_sa.sin6_addr); i++) 139 all1_sa.sin6_addr.s6_addr[i] = 0xff; 140 141 /* initialization of the default router list */ 142 TAILQ_INIT(&V_nd_defrouter); 143 /* start timer */ 144 callout_init(&V_nd6_slowtimo_ch, 0); 145 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 146 nd6_slowtimo, NULL); 147 148 nd6_init_done = 1; 149 150 } 151 152 struct nd_ifinfo * 153 nd6_ifattach(struct ifnet *ifp) 154 { 155 struct nd_ifinfo *nd; 156 157 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK); 158 bzero(nd, sizeof(*nd)); 159 160 nd->initialized = 1; 161 162 nd->chlim = IPV6_DEFHLIM; 163 nd->basereachable = REACHABLE_TIME; 164 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable); 165 nd->retrans = RETRANS_TIMER; 166 /* 167 * Note that the default value of ip6_accept_rtadv is 0, which means 168 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV 169 * here. 170 */ 171 nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV); 172 173 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */ 174 nd6_setmtu0(ifp, nd); 175 176 return nd; 177 } 178 179 void 180 nd6_ifdetach(struct nd_ifinfo *nd) 181 { 182 183 free(nd, M_IP6NDP); 184 } 185 186 /* 187 * Reset ND level link MTU. This function is called when the physical MTU 188 * changes, which means we might have to adjust the ND level MTU. 189 */ 190 void 191 nd6_setmtu(struct ifnet *ifp) 192 { 193 194 nd6_setmtu0(ifp, ND_IFINFO(ifp)); 195 } 196 197 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */ 198 void 199 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi) 200 { 201 u_int32_t omaxmtu; 202 203 omaxmtu = ndi->maxmtu; 204 205 switch (ifp->if_type) { 206 case IFT_ARCNET: 207 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */ 208 break; 209 case IFT_FDDI: 210 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */ 211 break; 212 case IFT_ISO88025: 213 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu); 214 break; 215 default: 216 ndi->maxmtu = ifp->if_mtu; 217 break; 218 } 219 220 /* 221 * Decreasing the interface MTU under IPV6 minimum MTU may cause 222 * undesirable situation. We thus notify the operator of the change 223 * explicitly. The check for omaxmtu is necessary to restrict the 224 * log to the case of changing the MTU, not initializing it. 225 */ 226 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { 227 log(LOG_NOTICE, "nd6_setmtu0: " 228 "new link MTU on %s (%lu) is too small for IPv6\n", 229 if_name(ifp), (unsigned long)ndi->maxmtu); 230 } 231 232 if (ndi->maxmtu > V_in6_maxmtu) 233 in6_setmaxmtu(); /* check all interfaces just in case */ 234 235 #undef MIN 236 } 237 238 void 239 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts) 240 { 241 242 bzero(ndopts, sizeof(*ndopts)); 243 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; 244 ndopts->nd_opts_last 245 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); 246 247 if (icmp6len == 0) { 248 ndopts->nd_opts_done = 1; 249 ndopts->nd_opts_search = NULL; 250 } 251 } 252 253 /* 254 * Take one ND option. 255 */ 256 struct nd_opt_hdr * 257 nd6_option(union nd_opts *ndopts) 258 { 259 struct nd_opt_hdr *nd_opt; 260 int olen; 261 262 if (ndopts == NULL) 263 panic("ndopts == NULL in nd6_option"); 264 if (ndopts->nd_opts_last == NULL) 265 panic("uninitialized ndopts in nd6_option"); 266 if (ndopts->nd_opts_search == NULL) 267 return NULL; 268 if (ndopts->nd_opts_done) 269 return NULL; 270 271 nd_opt = ndopts->nd_opts_search; 272 273 /* make sure nd_opt_len is inside the buffer */ 274 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) { 275 bzero(ndopts, sizeof(*ndopts)); 276 return NULL; 277 } 278 279 olen = nd_opt->nd_opt_len << 3; 280 if (olen == 0) { 281 /* 282 * Message validation requires that all included 283 * options have a length that is greater than zero. 284 */ 285 bzero(ndopts, sizeof(*ndopts)); 286 return NULL; 287 } 288 289 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen); 290 if (ndopts->nd_opts_search > ndopts->nd_opts_last) { 291 /* option overruns the end of buffer, invalid */ 292 bzero(ndopts, sizeof(*ndopts)); 293 return NULL; 294 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { 295 /* reached the end of options chain */ 296 ndopts->nd_opts_done = 1; 297 ndopts->nd_opts_search = NULL; 298 } 299 return nd_opt; 300 } 301 302 /* 303 * Parse multiple ND options. 304 * This function is much easier to use, for ND routines that do not need 305 * multiple options of the same type. 306 */ 307 int 308 nd6_options(union nd_opts *ndopts) 309 { 310 struct nd_opt_hdr *nd_opt; 311 int i = 0; 312 313 if (ndopts == NULL) 314 panic("ndopts == NULL in nd6_options"); 315 if (ndopts->nd_opts_last == NULL) 316 panic("uninitialized ndopts in nd6_options"); 317 if (ndopts->nd_opts_search == NULL) 318 return 0; 319 320 while (1) { 321 nd_opt = nd6_option(ndopts); 322 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { 323 /* 324 * Message validation requires that all included 325 * options have a length that is greater than zero. 326 */ 327 V_icmp6stat.icp6s_nd_badopt++; 328 bzero(ndopts, sizeof(*ndopts)); 329 return -1; 330 } 331 332 if (nd_opt == NULL) 333 goto skip1; 334 335 switch (nd_opt->nd_opt_type) { 336 case ND_OPT_SOURCE_LINKADDR: 337 case ND_OPT_TARGET_LINKADDR: 338 case ND_OPT_MTU: 339 case ND_OPT_REDIRECTED_HEADER: 340 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { 341 nd6log((LOG_INFO, 342 "duplicated ND6 option found (type=%d)\n", 343 nd_opt->nd_opt_type)); 344 /* XXX bark? */ 345 } else { 346 ndopts->nd_opt_array[nd_opt->nd_opt_type] 347 = nd_opt; 348 } 349 break; 350 case ND_OPT_PREFIX_INFORMATION: 351 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { 352 ndopts->nd_opt_array[nd_opt->nd_opt_type] 353 = nd_opt; 354 } 355 ndopts->nd_opts_pi_end = 356 (struct nd_opt_prefix_info *)nd_opt; 357 break; 358 default: 359 /* 360 * Unknown options must be silently ignored, 361 * to accomodate future extension to the protocol. 362 */ 363 nd6log((LOG_DEBUG, 364 "nd6_options: unsupported option %d - " 365 "option ignored\n", nd_opt->nd_opt_type)); 366 } 367 368 skip1: 369 i++; 370 if (i > V_nd6_maxndopt) { 371 V_icmp6stat.icp6s_nd_toomanyopt++; 372 nd6log((LOG_INFO, "too many loop in nd opt\n")); 373 break; 374 } 375 376 if (ndopts->nd_opts_done) 377 break; 378 } 379 380 return 0; 381 } 382 383 /* 384 * ND6 timer routine to handle ND6 entries 385 */ 386 void 387 nd6_llinfo_settimer(struct llinfo_nd6 *ln, long tick) 388 { 389 if (tick < 0) { 390 ln->ln_expire = 0; 391 ln->ln_ntick = 0; 392 callout_stop(&ln->ln_timer_ch); 393 } else { 394 ln->ln_expire = time_second + tick / hz; 395 if (tick > INT_MAX) { 396 ln->ln_ntick = tick - INT_MAX; 397 callout_reset(&ln->ln_timer_ch, INT_MAX, 398 nd6_llinfo_timer, ln); 399 } else { 400 ln->ln_ntick = 0; 401 callout_reset(&ln->ln_timer_ch, tick, 402 nd6_llinfo_timer, ln); 403 } 404 } 405 } 406 407 static void 408 nd6_llinfo_timer(void *arg) 409 { 410 struct llinfo_nd6 *ln; 411 struct rtentry *rt; 412 struct in6_addr *dst; 413 struct ifnet *ifp; 414 struct nd_ifinfo *ndi = NULL; 415 416 ln = (struct llinfo_nd6 *)arg; 417 418 if (ln->ln_ntick > 0) { 419 if (ln->ln_ntick > INT_MAX) { 420 ln->ln_ntick -= INT_MAX; 421 nd6_llinfo_settimer(ln, INT_MAX); 422 } else { 423 ln->ln_ntick = 0; 424 nd6_llinfo_settimer(ln, ln->ln_ntick); 425 } 426 return; 427 } 428 429 if ((rt = ln->ln_rt) == NULL) 430 panic("ln->ln_rt == NULL"); 431 if ((ifp = rt->rt_ifp) == NULL) 432 panic("ln->ln_rt->rt_ifp == NULL"); 433 ndi = ND_IFINFO(ifp); 434 435 /* sanity check */ 436 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln) 437 panic("rt_llinfo(%p) is not equal to ln(%p)", 438 rt->rt_llinfo, ln); 439 if (rt_key(rt) == NULL) 440 panic("rt key is NULL in nd6_timer(ln=%p)", ln); 441 442 dst = &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr; 443 444 switch (ln->ln_state) { 445 case ND6_LLINFO_INCOMPLETE: 446 if (ln->ln_asked < V_nd6_mmaxtries) { 447 ln->ln_asked++; 448 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000); 449 nd6_ns_output(ifp, NULL, dst, ln, 0); 450 } else { 451 struct mbuf *m = ln->ln_hold; 452 if (m) { 453 struct mbuf *m0; 454 455 /* 456 * assuming every packet in ln_hold has the 457 * same IP header 458 */ 459 m0 = m->m_nextpkt; 460 m->m_nextpkt = NULL; 461 icmp6_error2(m, ICMP6_DST_UNREACH, 462 ICMP6_DST_UNREACH_ADDR, 0, rt->rt_ifp); 463 464 ln->ln_hold = m0; 465 clear_llinfo_pqueue(ln); 466 } 467 if (rt && rt->rt_llinfo) 468 (void)nd6_free(rt, 0); 469 ln = NULL; 470 } 471 break; 472 case ND6_LLINFO_REACHABLE: 473 if (!ND6_LLINFO_PERMANENT(ln)) { 474 ln->ln_state = ND6_LLINFO_STALE; 475 nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz); 476 } 477 break; 478 479 case ND6_LLINFO_STALE: 480 /* Garbage Collection(RFC 2461 5.3) */ 481 if (!ND6_LLINFO_PERMANENT(ln)) { 482 if (rt && rt->rt_llinfo) 483 (void)nd6_free(rt, 1); 484 ln = NULL; 485 } 486 break; 487 488 case ND6_LLINFO_DELAY: 489 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 490 /* We need NUD */ 491 ln->ln_asked = 1; 492 ln->ln_state = ND6_LLINFO_PROBE; 493 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000); 494 nd6_ns_output(ifp, dst, dst, ln, 0); 495 } else { 496 ln->ln_state = ND6_LLINFO_STALE; /* XXX */ 497 nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz); 498 } 499 break; 500 case ND6_LLINFO_PROBE: 501 if (ln->ln_asked < V_nd6_umaxtries) { 502 ln->ln_asked++; 503 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000); 504 nd6_ns_output(ifp, dst, dst, ln, 0); 505 } else if (rt->rt_ifa != NULL && 506 rt->rt_ifa->ifa_addr->sa_family == AF_INET6 && 507 (((struct in6_ifaddr *)rt->rt_ifa)->ia_flags & IFA_ROUTE)) { 508 /* 509 * This is an unreachable neighbor whose address is 510 * specified as the destination of a p2p interface 511 * (see in6_ifinit()). We should not free the entry 512 * since this is sort of a "static" entry generated 513 * via interface address configuration. 514 */ 515 ln->ln_asked = 0; 516 ln->ln_expire = 0; /* make it permanent */ 517 ln->ln_state = ND6_LLINFO_STALE; 518 } else { 519 if (rt && rt->rt_llinfo) 520 (void)nd6_free(rt, 0); 521 ln = NULL; 522 } 523 break; 524 } 525 } 526 527 528 /* 529 * ND6 timer routine to expire default route list and prefix list 530 */ 531 void 532 nd6_timer(void *ignored_arg) 533 { 534 int s; 535 struct nd_defrouter *dr; 536 struct nd_prefix *pr; 537 struct in6_ifaddr *ia6, *nia6; 538 struct in6_addrlifetime *lt6; 539 540 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz, 541 nd6_timer, NULL); 542 543 /* expire default router list */ 544 s = splnet(); 545 dr = TAILQ_FIRST(&V_nd_defrouter); 546 while (dr) { 547 if (dr->expire && dr->expire < time_second) { 548 struct nd_defrouter *t; 549 t = TAILQ_NEXT(dr, dr_entry); 550 defrtrlist_del(dr); 551 dr = t; 552 } else { 553 dr = TAILQ_NEXT(dr, dr_entry); 554 } 555 } 556 557 /* 558 * expire interface addresses. 559 * in the past the loop was inside prefix expiry processing. 560 * However, from a stricter speci-confrmance standpoint, we should 561 * rather separate address lifetimes and prefix lifetimes. 562 */ 563 addrloop: 564 for (ia6 = V_in6_ifaddr; ia6; ia6 = nia6) { 565 nia6 = ia6->ia_next; 566 /* check address lifetime */ 567 lt6 = &ia6->ia6_lifetime; 568 if (IFA6_IS_INVALID(ia6)) { 569 int regen = 0; 570 571 /* 572 * If the expiring address is temporary, try 573 * regenerating a new one. This would be useful when 574 * we suspended a laptop PC, then turned it on after a 575 * period that could invalidate all temporary 576 * addresses. Although we may have to restart the 577 * loop (see below), it must be after purging the 578 * address. Otherwise, we'd see an infinite loop of 579 * regeneration. 580 */ 581 if (V_ip6_use_tempaddr && 582 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 583 if (regen_tmpaddr(ia6) == 0) 584 regen = 1; 585 } 586 587 in6_purgeaddr(&ia6->ia_ifa); 588 589 if (regen) 590 goto addrloop; /* XXX: see below */ 591 } else if (IFA6_IS_DEPRECATED(ia6)) { 592 int oldflags = ia6->ia6_flags; 593 594 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 595 596 /* 597 * If a temporary address has just become deprecated, 598 * regenerate a new one if possible. 599 */ 600 if (V_ip6_use_tempaddr && 601 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 602 (oldflags & IN6_IFF_DEPRECATED) == 0) { 603 604 if (regen_tmpaddr(ia6) == 0) { 605 /* 606 * A new temporary address is 607 * generated. 608 * XXX: this means the address chain 609 * has changed while we are still in 610 * the loop. Although the change 611 * would not cause disaster (because 612 * it's not a deletion, but an 613 * addition,) we'd rather restart the 614 * loop just for safety. Or does this 615 * significantly reduce performance?? 616 */ 617 goto addrloop; 618 } 619 } 620 } else { 621 /* 622 * A new RA might have made a deprecated address 623 * preferred. 624 */ 625 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 626 } 627 } 628 629 /* expire prefix list */ 630 pr = V_nd_prefix.lh_first; 631 while (pr) { 632 /* 633 * check prefix lifetime. 634 * since pltime is just for autoconf, pltime processing for 635 * prefix is not necessary. 636 */ 637 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 638 time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) { 639 struct nd_prefix *t; 640 t = pr->ndpr_next; 641 642 /* 643 * address expiration and prefix expiration are 644 * separate. NEVER perform in6_purgeaddr here. 645 */ 646 647 prelist_remove(pr); 648 pr = t; 649 } else 650 pr = pr->ndpr_next; 651 } 652 splx(s); 653 } 654 655 /* 656 * ia6 - deprecated/invalidated temporary address 657 */ 658 static int 659 regen_tmpaddr(struct in6_ifaddr *ia6) 660 { 661 struct ifaddr *ifa; 662 struct ifnet *ifp; 663 struct in6_ifaddr *public_ifa6 = NULL; 664 665 ifp = ia6->ia_ifa.ifa_ifp; 666 for (ifa = ifp->if_addrlist.tqh_first; ifa; 667 ifa = ifa->ifa_list.tqe_next) { 668 struct in6_ifaddr *it6; 669 670 if (ifa->ifa_addr->sa_family != AF_INET6) 671 continue; 672 673 it6 = (struct in6_ifaddr *)ifa; 674 675 /* ignore no autoconf addresses. */ 676 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 677 continue; 678 679 /* ignore autoconf addresses with different prefixes. */ 680 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 681 continue; 682 683 /* 684 * Now we are looking at an autoconf address with the same 685 * prefix as ours. If the address is temporary and is still 686 * preferred, do not create another one. It would be rare, but 687 * could happen, for example, when we resume a laptop PC after 688 * a long period. 689 */ 690 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 691 !IFA6_IS_DEPRECATED(it6)) { 692 public_ifa6 = NULL; 693 break; 694 } 695 696 /* 697 * This is a public autoconf address that has the same prefix 698 * as ours. If it is preferred, keep it. We can't break the 699 * loop here, because there may be a still-preferred temporary 700 * address with the prefix. 701 */ 702 if (!IFA6_IS_DEPRECATED(it6)) 703 public_ifa6 = it6; 704 } 705 706 if (public_ifa6 != NULL) { 707 int e; 708 709 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 710 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 711 " tmp addr,errno=%d\n", e); 712 return (-1); 713 } 714 return (0); 715 } 716 717 return (-1); 718 } 719 720 /* 721 * Nuke neighbor cache/prefix/default router management table, right before 722 * ifp goes away. 723 */ 724 void 725 nd6_purge(struct ifnet *ifp) 726 { 727 struct llinfo_nd6 *ln, *nln; 728 struct nd_defrouter *dr, *ndr; 729 struct nd_prefix *pr, *npr; 730 731 /* 732 * Nuke default router list entries toward ifp. 733 * We defer removal of default router list entries that is installed 734 * in the routing table, in order to keep additional side effects as 735 * small as possible. 736 */ 737 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) { 738 ndr = TAILQ_NEXT(dr, dr_entry); 739 if (dr->installed) 740 continue; 741 742 if (dr->ifp == ifp) 743 defrtrlist_del(dr); 744 } 745 746 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = ndr) { 747 ndr = TAILQ_NEXT(dr, dr_entry); 748 if (!dr->installed) 749 continue; 750 751 if (dr->ifp == ifp) 752 defrtrlist_del(dr); 753 } 754 755 /* Nuke prefix list entries toward ifp */ 756 for (pr = V_nd_prefix.lh_first; pr; pr = npr) { 757 npr = pr->ndpr_next; 758 if (pr->ndpr_ifp == ifp) { 759 /* 760 * Because if_detach() does *not* release prefixes 761 * while purging addresses the reference count will 762 * still be above zero. We therefore reset it to 763 * make sure that the prefix really gets purged. 764 */ 765 pr->ndpr_refcnt = 0; 766 767 /* 768 * Previously, pr->ndpr_addr is removed as well, 769 * but I strongly believe we don't have to do it. 770 * nd6_purge() is only called from in6_ifdetach(), 771 * which removes all the associated interface addresses 772 * by itself. 773 * (jinmei@kame.net 20010129) 774 */ 775 prelist_remove(pr); 776 } 777 } 778 779 /* cancel default outgoing interface setting */ 780 if (V_nd6_defifindex == ifp->if_index) 781 nd6_setdefaultiface(0); 782 783 if (!V_ip6_forwarding && V_ip6_accept_rtadv) { /* XXX: too restrictive? */ 784 /* refresh default router list */ 785 defrouter_select(); 786 } 787 788 /* 789 * Nuke neighbor cache entries for the ifp. 790 * Note that rt->rt_ifp may not be the same as ifp, 791 * due to KAME goto ours hack. See RTM_RESOLVE case in 792 * nd6_rtrequest(), and ip6_input(). 793 */ 794 ln = V_llinfo_nd6.ln_next; 795 while (ln && ln != &V_llinfo_nd6) { 796 struct rtentry *rt; 797 struct sockaddr_dl *sdl; 798 799 nln = ln->ln_next; 800 rt = ln->ln_rt; 801 if (rt && rt->rt_gateway && 802 rt->rt_gateway->sa_family == AF_LINK) { 803 sdl = (struct sockaddr_dl *)rt->rt_gateway; 804 if (sdl->sdl_index == ifp->if_index) 805 nln = nd6_free(rt, 0); 806 } 807 ln = nln; 808 } 809 } 810 811 struct rtentry * 812 nd6_lookup(struct in6_addr *addr6, int create, struct ifnet *ifp) 813 { 814 struct rtentry *rt; 815 struct sockaddr_in6 sin6; 816 char ip6buf[INET6_ADDRSTRLEN]; 817 818 bzero(&sin6, sizeof(sin6)); 819 sin6.sin6_len = sizeof(struct sockaddr_in6); 820 sin6.sin6_family = AF_INET6; 821 sin6.sin6_addr = *addr6; 822 rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL); 823 if (rt) { 824 if ((rt->rt_flags & RTF_LLINFO) == 0 && create) { 825 /* 826 * This is the case for the default route. 827 * If we want to create a neighbor cache for the 828 * address, we should free the route for the 829 * destination and allocate an interface route. 830 */ 831 RTFREE_LOCKED(rt); 832 rt = NULL; 833 } 834 } 835 if (rt == NULL) { 836 if (create && ifp) { 837 int e; 838 839 /* 840 * If no route is available and create is set, 841 * we allocate a host route for the destination 842 * and treat it like an interface route. 843 * This hack is necessary for a neighbor which can't 844 * be covered by our own prefix. 845 */ 846 struct ifaddr *ifa = 847 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp); 848 if (ifa == NULL) 849 return (NULL); 850 851 /* 852 * Create a new route. RTF_LLINFO is necessary 853 * to create a Neighbor Cache entry for the 854 * destination in nd6_rtrequest which will be 855 * called in rtrequest via ifa->ifa_rtrequest. 856 */ 857 if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6, 858 ifa->ifa_addr, (struct sockaddr *)&all1_sa, 859 (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) & 860 ~RTF_CLONING, &rt)) != 0) { 861 log(LOG_ERR, 862 "nd6_lookup: failed to add route for a " 863 "neighbor(%s), errno=%d\n", 864 ip6_sprintf(ip6buf, addr6), e); 865 } 866 if (rt == NULL) 867 return (NULL); 868 RT_LOCK(rt); 869 if (rt->rt_llinfo) { 870 struct llinfo_nd6 *ln = 871 (struct llinfo_nd6 *)rt->rt_llinfo; 872 ln->ln_state = ND6_LLINFO_NOSTATE; 873 } 874 } else 875 return (NULL); 876 } 877 RT_LOCK_ASSERT(rt); 878 RT_REMREF(rt); 879 /* 880 * Validation for the entry. 881 * Note that the check for rt_llinfo is necessary because a cloned 882 * route from a parent route that has the L flag (e.g. the default 883 * route to a p2p interface) may have the flag, too, while the 884 * destination is not actually a neighbor. 885 * XXX: we can't use rt->rt_ifp to check for the interface, since 886 * it might be the loopback interface if the entry is for our 887 * own address on a non-loopback interface. Instead, we should 888 * use rt->rt_ifa->ifa_ifp, which would specify the REAL 889 * interface. 890 * Note also that ifa_ifp and ifp may differ when we connect two 891 * interfaces to a same link, install a link prefix to an interface, 892 * and try to install a neighbor cache on an interface that does not 893 * have a route to the prefix. 894 */ 895 if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 || 896 rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL || 897 (ifp && rt->rt_ifa->ifa_ifp != ifp)) { 898 if (create) { 899 nd6log((LOG_DEBUG, 900 "nd6_lookup: failed to lookup %s (if = %s)\n", 901 ip6_sprintf(ip6buf, addr6), 902 ifp ? if_name(ifp) : "unspec")); 903 } 904 RT_UNLOCK(rt); 905 return (NULL); 906 } 907 RT_UNLOCK(rt); /* XXX not ready to return rt locked */ 908 return (rt); 909 } 910 911 /* 912 * Test whether a given IPv6 address is a neighbor or not, ignoring 913 * the actual neighbor cache. The neighbor cache is ignored in order 914 * to not reenter the routing code from within itself. 915 */ 916 static int 917 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 918 { 919 struct nd_prefix *pr; 920 struct ifaddr *dstaddr; 921 922 /* 923 * A link-local address is always a neighbor. 924 * XXX: a link does not necessarily specify a single interface. 925 */ 926 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 927 struct sockaddr_in6 sin6_copy; 928 u_int32_t zone; 929 930 /* 931 * We need sin6_copy since sa6_recoverscope() may modify the 932 * content (XXX). 933 */ 934 sin6_copy = *addr; 935 if (sa6_recoverscope(&sin6_copy)) 936 return (0); /* XXX: should be impossible */ 937 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 938 return (0); 939 if (sin6_copy.sin6_scope_id == zone) 940 return (1); 941 else 942 return (0); 943 } 944 945 /* 946 * If the address matches one of our addresses, 947 * it should be a neighbor. 948 * If the address matches one of our on-link prefixes, it should be a 949 * neighbor. 950 */ 951 for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) { 952 if (pr->ndpr_ifp != ifp) 953 continue; 954 955 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) 956 continue; 957 958 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 959 &addr->sin6_addr, &pr->ndpr_mask)) 960 return (1); 961 } 962 963 /* 964 * If the address is assigned on the node of the other side of 965 * a p2p interface, the address should be a neighbor. 966 */ 967 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr); 968 if ((dstaddr != NULL) && (dstaddr->ifa_ifp == ifp)) 969 return (1); 970 971 /* 972 * If the default router list is empty, all addresses are regarded 973 * as on-link, and thus, as a neighbor. 974 * XXX: we restrict the condition to hosts, because routers usually do 975 * not have the "default router list". 976 */ 977 if (!V_ip6_forwarding && TAILQ_FIRST(&V_nd_defrouter) == NULL && 978 V_nd6_defifindex == ifp->if_index) { 979 return (1); 980 } 981 982 return (0); 983 } 984 985 986 /* 987 * Detect if a given IPv6 address identifies a neighbor on a given link. 988 * XXX: should take care of the destination of a p2p link? 989 */ 990 int 991 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 992 { 993 994 if (nd6_is_new_addr_neighbor(addr, ifp)) 995 return (1); 996 997 /* 998 * Even if the address matches none of our addresses, it might be 999 * in the neighbor cache. 1000 */ 1001 if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL) 1002 return (1); 1003 1004 return (0); 1005 } 1006 1007 /* 1008 * Free an nd6 llinfo entry. 1009 * Since the function would cause significant changes in the kernel, DO NOT 1010 * make it global, unless you have a strong reason for the change, and are sure 1011 * that the change is safe. 1012 */ 1013 static struct llinfo_nd6 * 1014 nd6_free(struct rtentry *rt, int gc) 1015 { 1016 struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next; 1017 struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr; 1018 struct nd_defrouter *dr; 1019 1020 /* 1021 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1022 * even though it is not harmful, it was not really necessary. 1023 */ 1024 1025 /* cancel timer */ 1026 nd6_llinfo_settimer(ln, -1); 1027 1028 if (!V_ip6_forwarding) { 1029 int s; 1030 s = splnet(); 1031 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr, 1032 rt->rt_ifp); 1033 1034 if (dr != NULL && dr->expire && 1035 ln->ln_state == ND6_LLINFO_STALE && gc) { 1036 /* 1037 * If the reason for the deletion is just garbage 1038 * collection, and the neighbor is an active default 1039 * router, do not delete it. Instead, reset the GC 1040 * timer using the router's lifetime. 1041 * Simply deleting the entry would affect default 1042 * router selection, which is not necessarily a good 1043 * thing, especially when we're using router preference 1044 * values. 1045 * XXX: the check for ln_state would be redundant, 1046 * but we intentionally keep it just in case. 1047 */ 1048 if (dr->expire > time_second) 1049 nd6_llinfo_settimer(ln, 1050 (dr->expire - time_second) * hz); 1051 else 1052 nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz); 1053 splx(s); 1054 return (ln->ln_next); 1055 } 1056 1057 if (ln->ln_router || dr) { 1058 /* 1059 * rt6_flush must be called whether or not the neighbor 1060 * is in the Default Router List. 1061 * See a corresponding comment in nd6_na_input(). 1062 */ 1063 rt6_flush(&in6, rt->rt_ifp); 1064 } 1065 1066 if (dr) { 1067 /* 1068 * Unreachablity of a router might affect the default 1069 * router selection and on-link detection of advertised 1070 * prefixes. 1071 */ 1072 1073 /* 1074 * Temporarily fake the state to choose a new default 1075 * router and to perform on-link determination of 1076 * prefixes correctly. 1077 * Below the state will be set correctly, 1078 * or the entry itself will be deleted. 1079 */ 1080 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1081 1082 /* 1083 * Since defrouter_select() does not affect the 1084 * on-link determination and MIP6 needs the check 1085 * before the default router selection, we perform 1086 * the check now. 1087 */ 1088 pfxlist_onlink_check(); 1089 1090 /* 1091 * refresh default router list 1092 */ 1093 defrouter_select(); 1094 } 1095 splx(s); 1096 } 1097 1098 /* 1099 * Before deleting the entry, remember the next entry as the 1100 * return value. We need this because pfxlist_onlink_check() above 1101 * might have freed other entries (particularly the old next entry) as 1102 * a side effect (XXX). 1103 */ 1104 next = ln->ln_next; 1105 1106 /* 1107 * Detach the route from the routing tree and the list of neighbor 1108 * caches, and disable the route entry not to be used in already 1109 * cached routes. 1110 */ 1111 rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0, 1112 rt_mask(rt), 0, (struct rtentry **)0); 1113 1114 return (next); 1115 } 1116 1117 /* 1118 * Upper-layer reachability hint for Neighbor Unreachability Detection. 1119 * 1120 * XXX cost-effective methods? 1121 */ 1122 void 1123 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force) 1124 { 1125 struct llinfo_nd6 *ln; 1126 1127 /* 1128 * If the caller specified "rt", use that. Otherwise, resolve the 1129 * routing table by supplied "dst6". 1130 */ 1131 if (rt == NULL) { 1132 if (dst6 == NULL) 1133 return; 1134 if ((rt = nd6_lookup(dst6, 0, NULL)) == NULL) 1135 return; 1136 } 1137 1138 if ((rt->rt_flags & RTF_GATEWAY) != 0 || 1139 (rt->rt_flags & RTF_LLINFO) == 0 || 1140 rt->rt_llinfo == NULL || rt->rt_gateway == NULL || 1141 rt->rt_gateway->sa_family != AF_LINK) { 1142 /* This is not a host route. */ 1143 return; 1144 } 1145 1146 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1147 if (ln->ln_state < ND6_LLINFO_REACHABLE) 1148 return; 1149 1150 /* 1151 * if we get upper-layer reachability confirmation many times, 1152 * it is possible we have false information. 1153 */ 1154 if (!force) { 1155 ln->ln_byhint++; 1156 if (ln->ln_byhint > V_nd6_maxnudhint) 1157 return; 1158 } 1159 1160 ln->ln_state = ND6_LLINFO_REACHABLE; 1161 if (!ND6_LLINFO_PERMANENT(ln)) { 1162 nd6_llinfo_settimer(ln, 1163 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz); 1164 } 1165 } 1166 1167 /* 1168 * info - XXX unused 1169 */ 1170 void 1171 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info) 1172 { 1173 struct sockaddr *gate = rt->rt_gateway; 1174 struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1175 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK}; 1176 struct ifnet *ifp = rt->rt_ifp; 1177 struct ifaddr *ifa; 1178 1179 RT_LOCK_ASSERT(rt); 1180 1181 if ((rt->rt_flags & RTF_GATEWAY) != 0) 1182 return; 1183 1184 if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) { 1185 /* 1186 * This is probably an interface direct route for a link 1187 * which does not need neighbor caches (e.g. fe80::%lo0/64). 1188 * We do not need special treatment below for such a route. 1189 * Moreover, the RTF_LLINFO flag which would be set below 1190 * would annoy the ndp(8) command. 1191 */ 1192 return; 1193 } 1194 1195 if (req == RTM_RESOLVE && 1196 (nd6_need_cache(ifp) == 0 || /* stf case */ 1197 !nd6_is_new_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), 1198 ifp))) { 1199 /* 1200 * FreeBSD and BSD/OS often make a cloned host route based 1201 * on a less-specific route (e.g. the default route). 1202 * If the less specific route does not have a "gateway" 1203 * (this is the case when the route just goes to a p2p or an 1204 * stf interface), we'll mistakenly make a neighbor cache for 1205 * the host route, and will see strange neighbor solicitation 1206 * for the corresponding destination. In order to avoid the 1207 * confusion, we check if the destination of the route is 1208 * a neighbor in terms of neighbor discovery, and stop the 1209 * process if not. Additionally, we remove the LLINFO flag 1210 * so that ndp(8) will not try to get the neighbor information 1211 * of the destination. 1212 */ 1213 rt->rt_flags &= ~RTF_LLINFO; 1214 return; 1215 } 1216 1217 switch (req) { 1218 case RTM_ADD: 1219 /* 1220 * There is no backward compatibility :) 1221 * 1222 * if ((rt->rt_flags & RTF_HOST) == 0 && 1223 * SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff) 1224 * rt->rt_flags |= RTF_CLONING; 1225 */ 1226 if ((rt->rt_flags & RTF_CLONING) || 1227 ((rt->rt_flags & RTF_LLINFO) && ln == NULL)) { 1228 /* 1229 * Case 1: This route should come from a route to 1230 * interface (RTF_CLONING case) or the route should be 1231 * treated as on-link but is currently not 1232 * (RTF_LLINFO && ln == NULL case). 1233 */ 1234 rt_setgate(rt, rt_key(rt), 1235 (struct sockaddr *)&null_sdl); 1236 gate = rt->rt_gateway; 1237 SDL(gate)->sdl_type = ifp->if_type; 1238 SDL(gate)->sdl_index = ifp->if_index; 1239 if (ln) 1240 nd6_llinfo_settimer(ln, 0); 1241 if ((rt->rt_flags & RTF_CLONING) != 0) 1242 break; 1243 } 1244 /* 1245 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here. 1246 * We don't do that here since llinfo is not ready yet. 1247 * 1248 * There are also couple of other things to be discussed: 1249 * - unsolicited NA code needs improvement beforehand 1250 * - RFC2461 says we MAY send multicast unsolicited NA 1251 * (7.2.6 paragraph 4), however, it also says that we 1252 * SHOULD provide a mechanism to prevent multicast NA storm. 1253 * we don't have anything like it right now. 1254 * note that the mechanism needs a mutual agreement 1255 * between proxies, which means that we need to implement 1256 * a new protocol, or a new kludge. 1257 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA. 1258 * we need to check ip6forwarding before sending it. 1259 * (or should we allow proxy ND configuration only for 1260 * routers? there's no mention about proxy ND from hosts) 1261 */ 1262 /* FALLTHROUGH */ 1263 case RTM_RESOLVE: 1264 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) { 1265 /* 1266 * Address resolution isn't necessary for a point to 1267 * point link, so we can skip this test for a p2p link. 1268 */ 1269 if (gate->sa_family != AF_LINK || 1270 gate->sa_len < sizeof(null_sdl)) { 1271 log(LOG_DEBUG, 1272 "nd6_rtrequest: bad gateway value: %s\n", 1273 if_name(ifp)); 1274 break; 1275 } 1276 SDL(gate)->sdl_type = ifp->if_type; 1277 SDL(gate)->sdl_index = ifp->if_index; 1278 } 1279 if (ln != NULL) 1280 break; /* This happens on a route change */ 1281 /* 1282 * Case 2: This route may come from cloning, or a manual route 1283 * add with a LL address. 1284 */ 1285 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln)); 1286 rt->rt_llinfo = (caddr_t)ln; 1287 if (ln == NULL) { 1288 log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n"); 1289 break; 1290 } 1291 V_nd6_inuse++; 1292 V_nd6_allocated++; 1293 bzero(ln, sizeof(*ln)); 1294 RT_ADDREF(rt); 1295 ln->ln_rt = rt; 1296 callout_init(&ln->ln_timer_ch, 0); 1297 1298 /* this is required for "ndp" command. - shin */ 1299 if (req == RTM_ADD) { 1300 /* 1301 * gate should have some valid AF_LINK entry, 1302 * and ln->ln_expire should have some lifetime 1303 * which is specified by ndp command. 1304 */ 1305 ln->ln_state = ND6_LLINFO_REACHABLE; 1306 ln->ln_byhint = 0; 1307 } else { 1308 /* 1309 * When req == RTM_RESOLVE, rt is created and 1310 * initialized in rtrequest(), so rt_expire is 0. 1311 */ 1312 ln->ln_state = ND6_LLINFO_NOSTATE; 1313 nd6_llinfo_settimer(ln, 0); 1314 } 1315 rt->rt_flags |= RTF_LLINFO; 1316 ln->ln_next = V_llinfo_nd6.ln_next; 1317 V_llinfo_nd6.ln_next = ln; 1318 ln->ln_prev = &V_llinfo_nd6; 1319 ln->ln_next->ln_prev = ln; 1320 1321 /* 1322 * check if rt_key(rt) is one of my address assigned 1323 * to the interface. 1324 */ 1325 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp, 1326 &SIN6(rt_key(rt))->sin6_addr); 1327 if (ifa) { 1328 caddr_t macp = nd6_ifptomac(ifp); 1329 nd6_llinfo_settimer(ln, -1); 1330 ln->ln_state = ND6_LLINFO_REACHABLE; 1331 ln->ln_byhint = 0; 1332 if (macp) { 1333 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen); 1334 SDL(gate)->sdl_alen = ifp->if_addrlen; 1335 } 1336 if (V_nd6_useloopback) { 1337 rt->rt_ifp = &V_loif[0]; /* XXX */ 1338 /* 1339 * Make sure rt_ifa be equal to the ifaddr 1340 * corresponding to the address. 1341 * We need this because when we refer 1342 * rt_ifa->ia6_flags in ip6_input, we assume 1343 * that the rt_ifa points to the address instead 1344 * of the loopback address. 1345 */ 1346 if (ifa != rt->rt_ifa) { 1347 IFAFREE(rt->rt_ifa); 1348 IFAREF(ifa); 1349 rt->rt_ifa = ifa; 1350 } 1351 } 1352 } else if (rt->rt_flags & RTF_ANNOUNCE) { 1353 nd6_llinfo_settimer(ln, -1); 1354 ln->ln_state = ND6_LLINFO_REACHABLE; 1355 ln->ln_byhint = 0; 1356 1357 /* join solicited node multicast for proxy ND */ 1358 if (ifp->if_flags & IFF_MULTICAST) { 1359 struct in6_addr llsol; 1360 int error; 1361 1362 llsol = SIN6(rt_key(rt))->sin6_addr; 1363 llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL; 1364 llsol.s6_addr32[1] = 0; 1365 llsol.s6_addr32[2] = htonl(1); 1366 llsol.s6_addr8[12] = 0xff; 1367 if (in6_setscope(&llsol, ifp, NULL)) 1368 break; 1369 if (in6_addmulti(&llsol, ifp, 1370 &error, 0) == NULL) { 1371 char ip6buf[INET6_ADDRSTRLEN]; 1372 nd6log((LOG_ERR, "%s: failed to join " 1373 "%s (errno=%d)\n", if_name(ifp), 1374 ip6_sprintf(ip6buf, &llsol), 1375 error)); 1376 } 1377 } 1378 } 1379 break; 1380 1381 case RTM_DELETE: 1382 if (ln == NULL) 1383 break; 1384 /* leave from solicited node multicast for proxy ND */ 1385 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 && 1386 (ifp->if_flags & IFF_MULTICAST) != 0) { 1387 struct in6_addr llsol; 1388 struct in6_multi *in6m; 1389 1390 llsol = SIN6(rt_key(rt))->sin6_addr; 1391 llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL; 1392 llsol.s6_addr32[1] = 0; 1393 llsol.s6_addr32[2] = htonl(1); 1394 llsol.s6_addr8[12] = 0xff; 1395 if (in6_setscope(&llsol, ifp, NULL) == 0) { 1396 IN6_LOOKUP_MULTI(llsol, ifp, in6m); 1397 if (in6m) 1398 in6_delmulti(in6m); 1399 } else 1400 ; /* XXX: should not happen. bark here? */ 1401 } 1402 V_nd6_inuse--; 1403 ln->ln_next->ln_prev = ln->ln_prev; 1404 ln->ln_prev->ln_next = ln->ln_next; 1405 ln->ln_prev = NULL; 1406 nd6_llinfo_settimer(ln, -1); 1407 RT_REMREF(rt); 1408 rt->rt_llinfo = 0; 1409 rt->rt_flags &= ~RTF_LLINFO; 1410 clear_llinfo_pqueue(ln); 1411 Free((caddr_t)ln); 1412 } 1413 } 1414 1415 int 1416 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) 1417 { 1418 struct in6_drlist *drl = (struct in6_drlist *)data; 1419 struct in6_oprlist *oprl = (struct in6_oprlist *)data; 1420 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1421 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1422 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1423 struct nd_defrouter *dr; 1424 struct nd_prefix *pr; 1425 struct rtentry *rt; 1426 int i = 0, error = 0; 1427 int s; 1428 1429 switch (cmd) { 1430 case SIOCGDRLST_IN6: 1431 /* 1432 * obsolete API, use sysctl under net.inet6.icmp6 1433 */ 1434 bzero(drl, sizeof(*drl)); 1435 s = splnet(); 1436 dr = TAILQ_FIRST(&V_nd_defrouter); 1437 while (dr && i < DRLSTSIZ) { 1438 drl->defrouter[i].rtaddr = dr->rtaddr; 1439 in6_clearscope(&drl->defrouter[i].rtaddr); 1440 1441 drl->defrouter[i].flags = dr->flags; 1442 drl->defrouter[i].rtlifetime = dr->rtlifetime; 1443 drl->defrouter[i].expire = dr->expire; 1444 drl->defrouter[i].if_index = dr->ifp->if_index; 1445 i++; 1446 dr = TAILQ_NEXT(dr, dr_entry); 1447 } 1448 splx(s); 1449 break; 1450 case SIOCGPRLST_IN6: 1451 /* 1452 * obsolete API, use sysctl under net.inet6.icmp6 1453 * 1454 * XXX the structure in6_prlist was changed in backward- 1455 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6, 1456 * in6_prlist is used for nd6_sysctl() - fill_prlist(). 1457 */ 1458 /* 1459 * XXX meaning of fields, especialy "raflags", is very 1460 * differnet between RA prefix list and RR/static prefix list. 1461 * how about separating ioctls into two? 1462 */ 1463 bzero(oprl, sizeof(*oprl)); 1464 s = splnet(); 1465 pr = V_nd_prefix.lh_first; 1466 while (pr && i < PRLSTSIZ) { 1467 struct nd_pfxrouter *pfr; 1468 int j; 1469 1470 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr; 1471 oprl->prefix[i].raflags = pr->ndpr_raf; 1472 oprl->prefix[i].prefixlen = pr->ndpr_plen; 1473 oprl->prefix[i].vltime = pr->ndpr_vltime; 1474 oprl->prefix[i].pltime = pr->ndpr_pltime; 1475 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index; 1476 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 1477 oprl->prefix[i].expire = 0; 1478 else { 1479 time_t maxexpire; 1480 1481 /* XXX: we assume time_t is signed. */ 1482 maxexpire = (-1) & 1483 ~((time_t)1 << 1484 ((sizeof(maxexpire) * 8) - 1)); 1485 if (pr->ndpr_vltime < 1486 maxexpire - pr->ndpr_lastupdate) { 1487 oprl->prefix[i].expire = 1488 pr->ndpr_lastupdate + 1489 pr->ndpr_vltime; 1490 } else 1491 oprl->prefix[i].expire = maxexpire; 1492 } 1493 1494 pfr = pr->ndpr_advrtrs.lh_first; 1495 j = 0; 1496 while (pfr) { 1497 if (j < DRLSTSIZ) { 1498 #define RTRADDR oprl->prefix[i].advrtr[j] 1499 RTRADDR = pfr->router->rtaddr; 1500 in6_clearscope(&RTRADDR); 1501 #undef RTRADDR 1502 } 1503 j++; 1504 pfr = pfr->pfr_next; 1505 } 1506 oprl->prefix[i].advrtrs = j; 1507 oprl->prefix[i].origin = PR_ORIG_RA; 1508 1509 i++; 1510 pr = pr->ndpr_next; 1511 } 1512 splx(s); 1513 1514 break; 1515 case OSIOCGIFINFO_IN6: 1516 #define ND ndi->ndi 1517 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1518 bzero(&ND, sizeof(ND)); 1519 ND.linkmtu = IN6_LINKMTU(ifp); 1520 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1521 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1522 ND.reachable = ND_IFINFO(ifp)->reachable; 1523 ND.retrans = ND_IFINFO(ifp)->retrans; 1524 ND.flags = ND_IFINFO(ifp)->flags; 1525 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1526 ND.chlim = ND_IFINFO(ifp)->chlim; 1527 break; 1528 case SIOCGIFINFO_IN6: 1529 ND = *ND_IFINFO(ifp); 1530 break; 1531 case SIOCSIFINFO_IN6: 1532 /* 1533 * used to change host variables from userland. 1534 * intented for a use on router to reflect RA configurations. 1535 */ 1536 /* 0 means 'unspecified' */ 1537 if (ND.linkmtu != 0) { 1538 if (ND.linkmtu < IPV6_MMTU || 1539 ND.linkmtu > IN6_LINKMTU(ifp)) { 1540 error = EINVAL; 1541 break; 1542 } 1543 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1544 } 1545 1546 if (ND.basereachable != 0) { 1547 int obasereachable = ND_IFINFO(ifp)->basereachable; 1548 1549 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1550 if (ND.basereachable != obasereachable) 1551 ND_IFINFO(ifp)->reachable = 1552 ND_COMPUTE_RTIME(ND.basereachable); 1553 } 1554 if (ND.retrans != 0) 1555 ND_IFINFO(ifp)->retrans = ND.retrans; 1556 if (ND.chlim != 0) 1557 ND_IFINFO(ifp)->chlim = ND.chlim; 1558 /* FALLTHROUGH */ 1559 case SIOCSIFINFO_FLAGS: 1560 ND_IFINFO(ifp)->flags = ND.flags; 1561 break; 1562 #undef ND 1563 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1564 /* sync kernel routing table with the default router list */ 1565 defrouter_reset(); 1566 defrouter_select(); 1567 break; 1568 case SIOCSPFXFLUSH_IN6: 1569 { 1570 /* flush all the prefix advertised by routers */ 1571 struct nd_prefix *pr, *next; 1572 1573 s = splnet(); 1574 for (pr = V_nd_prefix.lh_first; pr; pr = next) { 1575 struct in6_ifaddr *ia, *ia_next; 1576 1577 next = pr->ndpr_next; 1578 1579 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1580 continue; /* XXX */ 1581 1582 /* do we really have to remove addresses as well? */ 1583 for (ia = V_in6_ifaddr; ia; ia = ia_next) { 1584 /* ia might be removed. keep the next ptr. */ 1585 ia_next = ia->ia_next; 1586 1587 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1588 continue; 1589 1590 if (ia->ia6_ndpr == pr) 1591 in6_purgeaddr(&ia->ia_ifa); 1592 } 1593 prelist_remove(pr); 1594 } 1595 splx(s); 1596 break; 1597 } 1598 case SIOCSRTRFLUSH_IN6: 1599 { 1600 /* flush all the default routers */ 1601 struct nd_defrouter *dr, *next; 1602 1603 s = splnet(); 1604 defrouter_reset(); 1605 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; dr = next) { 1606 next = TAILQ_NEXT(dr, dr_entry); 1607 defrtrlist_del(dr); 1608 } 1609 defrouter_select(); 1610 splx(s); 1611 break; 1612 } 1613 case SIOCGNBRINFO_IN6: 1614 { 1615 struct llinfo_nd6 *ln; 1616 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1617 1618 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1619 return (error); 1620 1621 s = splnet(); 1622 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) { 1623 error = EINVAL; 1624 splx(s); 1625 break; 1626 } 1627 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1628 nbi->state = ln->ln_state; 1629 nbi->asked = ln->ln_asked; 1630 nbi->isrouter = ln->ln_router; 1631 nbi->expire = ln->ln_expire; 1632 splx(s); 1633 1634 break; 1635 } 1636 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1637 ndif->ifindex = V_nd6_defifindex; 1638 break; 1639 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1640 return (nd6_setdefaultiface(ndif->ifindex)); 1641 } 1642 return (error); 1643 } 1644 1645 /* 1646 * Create neighbor cache entry and cache link-layer address, 1647 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1648 * 1649 * type - ICMP6 type 1650 * code - type dependent information 1651 */ 1652 struct rtentry * 1653 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, 1654 int lladdrlen, int type, int code) 1655 { 1656 struct rtentry *rt = NULL; 1657 struct llinfo_nd6 *ln = NULL; 1658 int is_newentry; 1659 struct sockaddr_dl *sdl = NULL; 1660 int do_update; 1661 int olladdr; 1662 int llchange; 1663 int newstate = 0; 1664 1665 if (ifp == NULL) 1666 panic("ifp == NULL in nd6_cache_lladdr"); 1667 if (from == NULL) 1668 panic("from == NULL in nd6_cache_lladdr"); 1669 1670 /* nothing must be updated for unspecified address */ 1671 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1672 return NULL; 1673 1674 /* 1675 * Validation about ifp->if_addrlen and lladdrlen must be done in 1676 * the caller. 1677 * 1678 * XXX If the link does not have link-layer adderss, what should 1679 * we do? (ifp->if_addrlen == 0) 1680 * Spec says nothing in sections for RA, RS and NA. There's small 1681 * description on it in NS section (RFC 2461 7.2.3). 1682 */ 1683 1684 rt = nd6_lookup(from, 0, ifp); 1685 if (rt == NULL) { 1686 rt = nd6_lookup(from, 1, ifp); 1687 is_newentry = 1; 1688 } else { 1689 /* do nothing if static ndp is set */ 1690 if (rt->rt_flags & RTF_STATIC) 1691 return NULL; 1692 is_newentry = 0; 1693 } 1694 1695 if (rt == NULL) 1696 return NULL; 1697 if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) { 1698 fail: 1699 (void)nd6_free(rt, 0); 1700 return NULL; 1701 } 1702 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1703 if (ln == NULL) 1704 goto fail; 1705 if (rt->rt_gateway == NULL) 1706 goto fail; 1707 if (rt->rt_gateway->sa_family != AF_LINK) 1708 goto fail; 1709 sdl = SDL(rt->rt_gateway); 1710 1711 olladdr = (sdl->sdl_alen) ? 1 : 0; 1712 if (olladdr && lladdr) { 1713 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen)) 1714 llchange = 1; 1715 else 1716 llchange = 0; 1717 } else 1718 llchange = 0; 1719 1720 /* 1721 * newentry olladdr lladdr llchange (*=record) 1722 * 0 n n -- (1) 1723 * 0 y n -- (2) 1724 * 0 n y -- (3) * STALE 1725 * 0 y y n (4) * 1726 * 0 y y y (5) * STALE 1727 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1728 * 1 -- y -- (7) * STALE 1729 */ 1730 1731 if (lladdr) { /* (3-5) and (7) */ 1732 /* 1733 * Record source link-layer address 1734 * XXX is it dependent to ifp->if_type? 1735 */ 1736 sdl->sdl_alen = ifp->if_addrlen; 1737 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen); 1738 } 1739 1740 if (!is_newentry) { 1741 if ((!olladdr && lladdr != NULL) || /* (3) */ 1742 (olladdr && lladdr != NULL && llchange)) { /* (5) */ 1743 do_update = 1; 1744 newstate = ND6_LLINFO_STALE; 1745 } else /* (1-2,4) */ 1746 do_update = 0; 1747 } else { 1748 do_update = 1; 1749 if (lladdr == NULL) /* (6) */ 1750 newstate = ND6_LLINFO_NOSTATE; 1751 else /* (7) */ 1752 newstate = ND6_LLINFO_STALE; 1753 } 1754 1755 if (do_update) { 1756 /* 1757 * Update the state of the neighbor cache. 1758 */ 1759 ln->ln_state = newstate; 1760 1761 if (ln->ln_state == ND6_LLINFO_STALE) { 1762 /* 1763 * XXX: since nd6_output() below will cause 1764 * state tansition to DELAY and reset the timer, 1765 * we must set the timer now, although it is actually 1766 * meaningless. 1767 */ 1768 nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz); 1769 1770 if (ln->ln_hold) { 1771 struct mbuf *m_hold, *m_hold_next; 1772 1773 /* 1774 * reset the ln_hold in advance, to explicitly 1775 * prevent a ln_hold lookup in nd6_output() 1776 * (wouldn't happen, though...) 1777 */ 1778 for (m_hold = ln->ln_hold, ln->ln_hold = NULL; 1779 m_hold; m_hold = m_hold_next) { 1780 m_hold_next = m_hold->m_nextpkt; 1781 m_hold->m_nextpkt = NULL; 1782 1783 /* 1784 * we assume ifp is not a p2p here, so 1785 * just set the 2nd argument as the 1786 * 1st one. 1787 */ 1788 nd6_output(ifp, ifp, m_hold, 1789 (struct sockaddr_in6 *)rt_key(rt), 1790 rt); 1791 } 1792 } 1793 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { 1794 /* probe right away */ 1795 nd6_llinfo_settimer((void *)ln, 0); 1796 } 1797 } 1798 1799 /* 1800 * ICMP6 type dependent behavior. 1801 * 1802 * NS: clear IsRouter if new entry 1803 * RS: clear IsRouter 1804 * RA: set IsRouter if there's lladdr 1805 * redir: clear IsRouter if new entry 1806 * 1807 * RA case, (1): 1808 * The spec says that we must set IsRouter in the following cases: 1809 * - If lladdr exist, set IsRouter. This means (1-5). 1810 * - If it is old entry (!newentry), set IsRouter. This means (7). 1811 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1812 * A quetion arises for (1) case. (1) case has no lladdr in the 1813 * neighbor cache, this is similar to (6). 1814 * This case is rare but we figured that we MUST NOT set IsRouter. 1815 * 1816 * newentry olladdr lladdr llchange NS RS RA redir 1817 * D R 1818 * 0 n n -- (1) c ? s 1819 * 0 y n -- (2) c s s 1820 * 0 n y -- (3) c s s 1821 * 0 y y n (4) c s s 1822 * 0 y y y (5) c s s 1823 * 1 -- n -- (6) c c c s 1824 * 1 -- y -- (7) c c s c s 1825 * 1826 * (c=clear s=set) 1827 */ 1828 switch (type & 0xff) { 1829 case ND_NEIGHBOR_SOLICIT: 1830 /* 1831 * New entry must have is_router flag cleared. 1832 */ 1833 if (is_newentry) /* (6-7) */ 1834 ln->ln_router = 0; 1835 break; 1836 case ND_REDIRECT: 1837 /* 1838 * If the icmp is a redirect to a better router, always set the 1839 * is_router flag. Otherwise, if the entry is newly created, 1840 * clear the flag. [RFC 2461, sec 8.3] 1841 */ 1842 if (code == ND_REDIRECT_ROUTER) 1843 ln->ln_router = 1; 1844 else if (is_newentry) /* (6-7) */ 1845 ln->ln_router = 0; 1846 break; 1847 case ND_ROUTER_SOLICIT: 1848 /* 1849 * is_router flag must always be cleared. 1850 */ 1851 ln->ln_router = 0; 1852 break; 1853 case ND_ROUTER_ADVERT: 1854 /* 1855 * Mark an entry with lladdr as a router. 1856 */ 1857 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */ 1858 (is_newentry && lladdr)) { /* (7) */ 1859 ln->ln_router = 1; 1860 } 1861 break; 1862 } 1863 1864 /* 1865 * When the link-layer address of a router changes, select the 1866 * best router again. In particular, when the neighbor entry is newly 1867 * created, it might affect the selection policy. 1868 * Question: can we restrict the first condition to the "is_newentry" 1869 * case? 1870 * XXX: when we hear an RA from a new router with the link-layer 1871 * address option, defrouter_select() is called twice, since 1872 * defrtrlist_update called the function as well. However, I believe 1873 * we can compromise the overhead, since it only happens the first 1874 * time. 1875 * XXX: although defrouter_select() should not have a bad effect 1876 * for those are not autoconfigured hosts, we explicitly avoid such 1877 * cases for safety. 1878 */ 1879 if (do_update && ln->ln_router && !V_ip6_forwarding && V_ip6_accept_rtadv) 1880 defrouter_select(); 1881 1882 return rt; 1883 } 1884 1885 static void 1886 nd6_slowtimo(void *ignored_arg) 1887 { 1888 struct nd_ifinfo *nd6if; 1889 struct ifnet *ifp; 1890 1891 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 1892 nd6_slowtimo, NULL); 1893 IFNET_RLOCK(); 1894 for (ifp = TAILQ_FIRST(&V_ifnet); ifp; 1895 ifp = TAILQ_NEXT(ifp, if_list)) { 1896 nd6if = ND_IFINFO(ifp); 1897 if (nd6if->basereachable && /* already initialized */ 1898 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 1899 /* 1900 * Since reachable time rarely changes by router 1901 * advertisements, we SHOULD insure that a new random 1902 * value gets recomputed at least once every few hours. 1903 * (RFC 2461, 6.3.4) 1904 */ 1905 nd6if->recalctm = V_nd6_recalc_reachtm_interval; 1906 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 1907 } 1908 } 1909 IFNET_RUNLOCK(); 1910 } 1911 1912 #define senderr(e) { error = (e); goto bad;} 1913 int 1914 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 1915 struct sockaddr_in6 *dst, struct rtentry *rt0) 1916 { 1917 struct mbuf *m = m0; 1918 struct rtentry *rt = rt0; 1919 struct sockaddr_in6 *gw6 = NULL; 1920 struct llinfo_nd6 *ln = NULL; 1921 int error = 0; 1922 1923 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) 1924 goto sendpkt; 1925 1926 if (nd6_need_cache(ifp) == 0) 1927 goto sendpkt; 1928 1929 /* 1930 * next hop determination. This routine is derived from ether_output. 1931 */ 1932 /* NB: the locking here is tortuous... */ 1933 if (rt != NULL) 1934 RT_LOCK(rt); 1935 again: 1936 if (rt != NULL) { 1937 if ((rt->rt_flags & RTF_UP) == 0) { 1938 RT_UNLOCK(rt); 1939 rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL); 1940 if (rt != NULL) { 1941 RT_REMREF(rt); 1942 if (rt->rt_ifp != ifp) 1943 /* 1944 * XXX maybe we should update ifp too, 1945 * but the original code didn't and I 1946 * don't know what is correct here. 1947 */ 1948 goto again; 1949 } else 1950 senderr(EHOSTUNREACH); 1951 } 1952 1953 if (rt->rt_flags & RTF_GATEWAY) { 1954 gw6 = (struct sockaddr_in6 *)rt->rt_gateway; 1955 1956 /* 1957 * We skip link-layer address resolution and NUD 1958 * if the gateway is not a neighbor from ND point 1959 * of view, regardless of the value of nd_ifinfo.flags. 1960 * The second condition is a bit tricky; we skip 1961 * if the gateway is our own address, which is 1962 * sometimes used to install a route to a p2p link. 1963 */ 1964 if (!nd6_is_addr_neighbor(gw6, ifp) || 1965 in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) { 1966 RT_UNLOCK(rt); 1967 /* 1968 * We allow this kind of tricky route only 1969 * when the outgoing interface is p2p. 1970 * XXX: we may need a more generic rule here. 1971 */ 1972 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 1973 senderr(EHOSTUNREACH); 1974 1975 goto sendpkt; 1976 } 1977 1978 if (rt->rt_gwroute == NULL) 1979 goto lookup; 1980 rt = rt->rt_gwroute; 1981 RT_LOCK(rt); /* NB: gwroute */ 1982 if ((rt->rt_flags & RTF_UP) == 0) { 1983 RTFREE_LOCKED(rt); /* unlock gwroute */ 1984 rt = rt0; 1985 rt0->rt_gwroute = NULL; 1986 lookup: 1987 RT_UNLOCK(rt0); 1988 rt = rtalloc1(rt->rt_gateway, 1, 0UL); 1989 if (rt == rt0) { 1990 RT_REMREF(rt0); 1991 RT_UNLOCK(rt0); 1992 senderr(EHOSTUNREACH); 1993 } 1994 RT_LOCK(rt0); 1995 if (rt0->rt_gwroute != NULL) 1996 RTFREE(rt0->rt_gwroute); 1997 rt0->rt_gwroute = rt; 1998 if (rt == NULL) { 1999 RT_UNLOCK(rt0); 2000 senderr(EHOSTUNREACH); 2001 } 2002 } 2003 RT_UNLOCK(rt0); 2004 } 2005 RT_UNLOCK(rt); 2006 } 2007 2008 /* 2009 * Address resolution or Neighbor Unreachability Detection 2010 * for the next hop. 2011 * At this point, the destination of the packet must be a unicast 2012 * or an anycast address(i.e. not a multicast). 2013 */ 2014 2015 /* Look up the neighbor cache for the nexthop */ 2016 if (rt && (rt->rt_flags & RTF_LLINFO) != 0) 2017 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 2018 else { 2019 /* 2020 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 2021 * the condition below is not very efficient. But we believe 2022 * it is tolerable, because this should be a rare case. 2023 */ 2024 if (nd6_is_addr_neighbor(dst, ifp) && 2025 (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL) 2026 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 2027 } 2028 if (ln == NULL || rt == NULL) { 2029 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && 2030 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 2031 char ip6buf[INET6_ADDRSTRLEN]; 2032 log(LOG_DEBUG, 2033 "nd6_output: can't allocate llinfo for %s " 2034 "(ln=%p, rt=%p)\n", 2035 ip6_sprintf(ip6buf, &dst->sin6_addr), ln, rt); 2036 senderr(EIO); /* XXX: good error? */ 2037 } 2038 2039 goto sendpkt; /* send anyway */ 2040 } 2041 2042 /* We don't have to do link-layer address resolution on a p2p link. */ 2043 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && 2044 ln->ln_state < ND6_LLINFO_REACHABLE) { 2045 ln->ln_state = ND6_LLINFO_STALE; 2046 nd6_llinfo_settimer(ln, (long)V_nd6_gctimer * hz); 2047 } 2048 2049 /* 2050 * The first time we send a packet to a neighbor whose entry is 2051 * STALE, we have to change the state to DELAY and a sets a timer to 2052 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 2053 * neighbor unreachability detection on expiration. 2054 * (RFC 2461 7.3.3) 2055 */ 2056 if (ln->ln_state == ND6_LLINFO_STALE) { 2057 ln->ln_asked = 0; 2058 ln->ln_state = ND6_LLINFO_DELAY; 2059 nd6_llinfo_settimer(ln, (long)V_nd6_delay * hz); 2060 } 2061 2062 /* 2063 * If the neighbor cache entry has a state other than INCOMPLETE 2064 * (i.e. its link-layer address is already resolved), just 2065 * send the packet. 2066 */ 2067 if (ln->ln_state > ND6_LLINFO_INCOMPLETE) 2068 goto sendpkt; 2069 2070 /* 2071 * There is a neighbor cache entry, but no ethernet address 2072 * response yet. Append this latest packet to the end of the 2073 * packet queue in the mbuf, unless the number of the packet 2074 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 2075 * the oldest packet in the queue will be removed. 2076 */ 2077 if (ln->ln_state == ND6_LLINFO_NOSTATE) 2078 ln->ln_state = ND6_LLINFO_INCOMPLETE; 2079 if (ln->ln_hold) { 2080 struct mbuf *m_hold; 2081 int i; 2082 2083 i = 0; 2084 for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) { 2085 i++; 2086 if (m_hold->m_nextpkt == NULL) { 2087 m_hold->m_nextpkt = m; 2088 break; 2089 } 2090 } 2091 while (i >= V_nd6_maxqueuelen) { 2092 m_hold = ln->ln_hold; 2093 ln->ln_hold = ln->ln_hold->m_nextpkt; 2094 m_freem(m_hold); 2095 i--; 2096 } 2097 } else { 2098 ln->ln_hold = m; 2099 } 2100 2101 /* 2102 * If there has been no NS for the neighbor after entering the 2103 * INCOMPLETE state, send the first solicitation. 2104 */ 2105 if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) { 2106 ln->ln_asked++; 2107 nd6_llinfo_settimer(ln, 2108 (long)ND_IFINFO(ifp)->retrans * hz / 1000); 2109 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); 2110 } 2111 return (0); 2112 2113 sendpkt: 2114 /* discard the packet if IPv6 operation is disabled on the interface */ 2115 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 2116 error = ENETDOWN; /* better error? */ 2117 goto bad; 2118 } 2119 2120 #ifdef MAC 2121 mac_netinet6_nd6_send(ifp, m); 2122 #endif 2123 if ((ifp->if_flags & IFF_LOOPBACK) != 0) { 2124 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst, 2125 rt)); 2126 } 2127 return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt)); 2128 2129 bad: 2130 if (m) 2131 m_freem(m); 2132 return (error); 2133 } 2134 #undef senderr 2135 2136 int 2137 nd6_need_cache(struct ifnet *ifp) 2138 { 2139 /* 2140 * XXX: we currently do not make neighbor cache on any interface 2141 * other than ARCnet, Ethernet, FDDI and GIF. 2142 * 2143 * RFC2893 says: 2144 * - unidirectional tunnels needs no ND 2145 */ 2146 switch (ifp->if_type) { 2147 case IFT_ARCNET: 2148 case IFT_ETHER: 2149 case IFT_FDDI: 2150 case IFT_IEEE1394: 2151 #ifdef IFT_L2VLAN 2152 case IFT_L2VLAN: 2153 #endif 2154 #ifdef IFT_IEEE80211 2155 case IFT_IEEE80211: 2156 #endif 2157 #ifdef IFT_CARP 2158 case IFT_CARP: 2159 #endif 2160 case IFT_GIF: /* XXX need more cases? */ 2161 case IFT_PPP: 2162 case IFT_TUNNEL: 2163 case IFT_BRIDGE: 2164 case IFT_PROPVIRTUAL: 2165 return (1); 2166 default: 2167 return (0); 2168 } 2169 } 2170 2171 int 2172 nd6_storelladdr(struct ifnet *ifp, struct rtentry *rt0, struct mbuf *m, 2173 struct sockaddr *dst, u_char *desten) 2174 { 2175 struct sockaddr_dl *sdl; 2176 struct rtentry *rt; 2177 int error; 2178 2179 if (m->m_flags & M_MCAST) { 2180 int i; 2181 2182 switch (ifp->if_type) { 2183 case IFT_ETHER: 2184 case IFT_FDDI: 2185 #ifdef IFT_L2VLAN 2186 case IFT_L2VLAN: 2187 #endif 2188 #ifdef IFT_IEEE80211 2189 case IFT_IEEE80211: 2190 #endif 2191 case IFT_BRIDGE: 2192 case IFT_ISO88025: 2193 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr, 2194 desten); 2195 return (0); 2196 case IFT_IEEE1394: 2197 /* 2198 * netbsd can use if_broadcastaddr, but we don't do so 2199 * to reduce # of ifdef. 2200 */ 2201 for (i = 0; i < ifp->if_addrlen; i++) 2202 desten[i] = ~0; 2203 return (0); 2204 case IFT_ARCNET: 2205 *desten = 0; 2206 return (0); 2207 default: 2208 m_freem(m); 2209 return (EAFNOSUPPORT); 2210 } 2211 } 2212 2213 if (rt0 == NULL) { 2214 /* this could happen, if we could not allocate memory */ 2215 m_freem(m); 2216 return (ENOMEM); 2217 } 2218 2219 error = rt_check(&rt, &rt0, dst); 2220 if (error) { 2221 m_freem(m); 2222 return (error); 2223 } 2224 RT_UNLOCK(rt); 2225 2226 if (rt->rt_gateway->sa_family != AF_LINK) { 2227 printf("nd6_storelladdr: something odd happens\n"); 2228 m_freem(m); 2229 return (EINVAL); 2230 } 2231 sdl = SDL(rt->rt_gateway); 2232 if (sdl->sdl_alen == 0) { 2233 /* this should be impossible, but we bark here for debugging */ 2234 printf("nd6_storelladdr: sdl_alen == 0\n"); 2235 m_freem(m); 2236 return (EINVAL); 2237 } 2238 2239 bcopy(LLADDR(sdl), desten, sdl->sdl_alen); 2240 return (0); 2241 } 2242 2243 static void 2244 clear_llinfo_pqueue(struct llinfo_nd6 *ln) 2245 { 2246 struct mbuf *m_hold, *m_hold_next; 2247 2248 for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) { 2249 m_hold_next = m_hold->m_nextpkt; 2250 m_hold->m_nextpkt = NULL; 2251 m_freem(m_hold); 2252 } 2253 2254 ln->ln_hold = NULL; 2255 return; 2256 } 2257 2258 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2259 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2260 #ifdef SYSCTL_DECL 2261 SYSCTL_DECL(_net_inet6_icmp6); 2262 #endif 2263 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2264 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2265 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2266 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2267 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2268 CTLFLAG_RW, &nd6_maxqueuelen, 1, ""); 2269 2270 static int 2271 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2272 { 2273 int error; 2274 char buf[1024] __aligned(4); 2275 struct in6_defrouter *d, *de; 2276 struct nd_defrouter *dr; 2277 2278 if (req->newptr) 2279 return EPERM; 2280 error = 0; 2281 2282 for (dr = TAILQ_FIRST(&V_nd_defrouter); dr; 2283 dr = TAILQ_NEXT(dr, dr_entry)) { 2284 d = (struct in6_defrouter *)buf; 2285 de = (struct in6_defrouter *)(buf + sizeof(buf)); 2286 2287 if (d + 1 <= de) { 2288 bzero(d, sizeof(*d)); 2289 d->rtaddr.sin6_family = AF_INET6; 2290 d->rtaddr.sin6_len = sizeof(d->rtaddr); 2291 d->rtaddr.sin6_addr = dr->rtaddr; 2292 error = sa6_recoverscope(&d->rtaddr); 2293 if (error != 0) 2294 return (error); 2295 d->flags = dr->flags; 2296 d->rtlifetime = dr->rtlifetime; 2297 d->expire = dr->expire; 2298 d->if_index = dr->ifp->if_index; 2299 } else 2300 panic("buffer too short"); 2301 2302 error = SYSCTL_OUT(req, buf, sizeof(*d)); 2303 if (error) 2304 break; 2305 } 2306 2307 return (error); 2308 } 2309 2310 static int 2311 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2312 { 2313 int error; 2314 char buf[1024] __aligned(4); 2315 struct in6_prefix *p, *pe; 2316 struct nd_prefix *pr; 2317 char ip6buf[INET6_ADDRSTRLEN]; 2318 2319 if (req->newptr) 2320 return EPERM; 2321 error = 0; 2322 2323 for (pr = V_nd_prefix.lh_first; pr; pr = pr->ndpr_next) { 2324 u_short advrtrs; 2325 size_t advance; 2326 struct sockaddr_in6 *sin6, *s6; 2327 struct nd_pfxrouter *pfr; 2328 2329 p = (struct in6_prefix *)buf; 2330 pe = (struct in6_prefix *)(buf + sizeof(buf)); 2331 2332 if (p + 1 <= pe) { 2333 bzero(p, sizeof(*p)); 2334 sin6 = (struct sockaddr_in6 *)(p + 1); 2335 2336 p->prefix = pr->ndpr_prefix; 2337 if (sa6_recoverscope(&p->prefix)) { 2338 log(LOG_ERR, 2339 "scope error in prefix list (%s)\n", 2340 ip6_sprintf(ip6buf, &p->prefix.sin6_addr)); 2341 /* XXX: press on... */ 2342 } 2343 p->raflags = pr->ndpr_raf; 2344 p->prefixlen = pr->ndpr_plen; 2345 p->vltime = pr->ndpr_vltime; 2346 p->pltime = pr->ndpr_pltime; 2347 p->if_index = pr->ndpr_ifp->if_index; 2348 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2349 p->expire = 0; 2350 else { 2351 time_t maxexpire; 2352 2353 /* XXX: we assume time_t is signed. */ 2354 maxexpire = (-1) & 2355 ~((time_t)1 << 2356 ((sizeof(maxexpire) * 8) - 1)); 2357 if (pr->ndpr_vltime < 2358 maxexpire - pr->ndpr_lastupdate) { 2359 p->expire = pr->ndpr_lastupdate + 2360 pr->ndpr_vltime; 2361 } else 2362 p->expire = maxexpire; 2363 } 2364 p->refcnt = pr->ndpr_refcnt; 2365 p->flags = pr->ndpr_stateflags; 2366 p->origin = PR_ORIG_RA; 2367 advrtrs = 0; 2368 for (pfr = pr->ndpr_advrtrs.lh_first; pfr; 2369 pfr = pfr->pfr_next) { 2370 if ((void *)&sin6[advrtrs + 1] > (void *)pe) { 2371 advrtrs++; 2372 continue; 2373 } 2374 s6 = &sin6[advrtrs]; 2375 bzero(s6, sizeof(*s6)); 2376 s6->sin6_family = AF_INET6; 2377 s6->sin6_len = sizeof(*sin6); 2378 s6->sin6_addr = pfr->router->rtaddr; 2379 if (sa6_recoverscope(s6)) { 2380 log(LOG_ERR, 2381 "scope error in " 2382 "prefix list (%s)\n", 2383 ip6_sprintf(ip6buf, 2384 &pfr->router->rtaddr)); 2385 } 2386 advrtrs++; 2387 } 2388 p->advrtrs = advrtrs; 2389 } else 2390 panic("buffer too short"); 2391 2392 advance = sizeof(*p) + sizeof(*sin6) * advrtrs; 2393 error = SYSCTL_OUT(req, buf, advance); 2394 if (error) 2395 break; 2396 } 2397 2398 return (error); 2399 } 2400