1 /* $FreeBSD$ */ 2 /* $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $ */ 3 4 /*- 5 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. Neither the name of the project nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 */ 32 33 #include "opt_inet.h" 34 #include "opt_inet6.h" 35 #include "opt_mac.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/callout.h> 40 #include <sys/malloc.h> 41 #include <sys/mbuf.h> 42 #include <sys/socket.h> 43 #include <sys/sockio.h> 44 #include <sys/time.h> 45 #include <sys/kernel.h> 46 #include <sys/protosw.h> 47 #include <sys/errno.h> 48 #include <sys/syslog.h> 49 #include <sys/queue.h> 50 #include <sys/sysctl.h> 51 52 #include <net/if.h> 53 #include <net/if_arc.h> 54 #include <net/if_dl.h> 55 #include <net/if_types.h> 56 #include <net/iso88025.h> 57 #include <net/fddi.h> 58 #include <net/route.h> 59 60 #include <netinet/in.h> 61 #include <netinet/if_ether.h> 62 #include <netinet6/in6_var.h> 63 #include <netinet/ip6.h> 64 #include <netinet6/ip6_var.h> 65 #include <netinet6/scope6_var.h> 66 #include <netinet6/nd6.h> 67 #include <netinet/icmp6.h> 68 69 #include <sys/limits.h> 70 71 #include <security/mac/mac_framework.h> 72 73 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ 74 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ 75 76 #define SIN6(s) ((struct sockaddr_in6 *)s) 77 #define SDL(s) ((struct sockaddr_dl *)s) 78 79 /* timer values */ 80 int nd6_prune = 1; /* walk list every 1 seconds */ 81 int nd6_delay = 5; /* delay first probe time 5 second */ 82 int nd6_umaxtries = 3; /* maximum unicast query */ 83 int nd6_mmaxtries = 3; /* maximum multicast query */ 84 int nd6_useloopback = 1; /* use loopback interface for local traffic */ 85 int nd6_gctimer = (60 * 60 * 24); /* 1 day: garbage collection timer */ 86 87 /* preventing too many loops in ND option parsing */ 88 int nd6_maxndopt = 10; /* max # of ND options allowed */ 89 90 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */ 91 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */ 92 93 #ifdef ND6_DEBUG 94 int nd6_debug = 1; 95 #else 96 int nd6_debug = 0; 97 #endif 98 99 /* for debugging? */ 100 static int nd6_inuse, nd6_allocated; 101 102 struct llinfo_nd6 llinfo_nd6 = {&llinfo_nd6, &llinfo_nd6}; 103 struct nd_drhead nd_defrouter; 104 struct nd_prhead nd_prefix = { 0 }; 105 106 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL; 107 static struct sockaddr_in6 all1_sa; 108 109 static int nd6_is_new_addr_neighbor __P((struct sockaddr_in6 *, 110 struct ifnet *)); 111 static void nd6_setmtu0 __P((struct ifnet *, struct nd_ifinfo *)); 112 static void nd6_slowtimo __P((void *)); 113 static int regen_tmpaddr __P((struct in6_ifaddr *)); 114 static struct llinfo_nd6 *nd6_free __P((struct rtentry *, int)); 115 static void nd6_llinfo_timer __P((void *)); 116 static void clear_llinfo_pqueue __P((struct llinfo_nd6 *)); 117 118 struct callout nd6_slowtimo_ch; 119 struct callout nd6_timer_ch; 120 extern struct callout in6_tmpaddrtimer_ch; 121 122 void 123 nd6_init() 124 { 125 static int nd6_init_done = 0; 126 int i; 127 128 if (nd6_init_done) { 129 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n"); 130 return; 131 } 132 133 all1_sa.sin6_family = AF_INET6; 134 all1_sa.sin6_len = sizeof(struct sockaddr_in6); 135 for (i = 0; i < sizeof(all1_sa.sin6_addr); i++) 136 all1_sa.sin6_addr.s6_addr[i] = 0xff; 137 138 /* initialization of the default router list */ 139 TAILQ_INIT(&nd_defrouter); 140 141 nd6_init_done = 1; 142 143 /* start timer */ 144 callout_init(&nd6_slowtimo_ch, 0); 145 callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 146 nd6_slowtimo, NULL); 147 } 148 149 struct nd_ifinfo * 150 nd6_ifattach(ifp) 151 struct ifnet *ifp; 152 { 153 struct nd_ifinfo *nd; 154 155 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK); 156 bzero(nd, sizeof(*nd)); 157 158 nd->initialized = 1; 159 160 nd->chlim = IPV6_DEFHLIM; 161 nd->basereachable = REACHABLE_TIME; 162 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable); 163 nd->retrans = RETRANS_TIMER; 164 /* 165 * Note that the default value of ip6_accept_rtadv is 0, which means 166 * we won't accept RAs by default even if we set ND6_IFF_ACCEPT_RTADV 167 * here. 168 */ 169 nd->flags = (ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV); 170 171 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */ 172 nd6_setmtu0(ifp, nd); 173 174 return nd; 175 } 176 177 void 178 nd6_ifdetach(nd) 179 struct nd_ifinfo *nd; 180 { 181 182 free(nd, M_IP6NDP); 183 } 184 185 /* 186 * Reset ND level link MTU. This function is called when the physical MTU 187 * changes, which means we might have to adjust the ND level MTU. 188 */ 189 void 190 nd6_setmtu(ifp) 191 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(ifp, ndi) 200 struct ifnet *ifp; 201 struct nd_ifinfo *ndi; 202 { 203 u_int32_t omaxmtu; 204 205 omaxmtu = ndi->maxmtu; 206 207 switch (ifp->if_type) { 208 case IFT_ARCNET: 209 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */ 210 break; 211 case IFT_FDDI: 212 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */ 213 break; 214 case IFT_ISO88025: 215 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu); 216 break; 217 default: 218 ndi->maxmtu = ifp->if_mtu; 219 break; 220 } 221 222 /* 223 * Decreasing the interface MTU under IPV6 minimum MTU may cause 224 * undesirable situation. We thus notify the operator of the change 225 * explicitly. The check for omaxmtu is necessary to restrict the 226 * log to the case of changing the MTU, not initializing it. 227 */ 228 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { 229 log(LOG_NOTICE, "nd6_setmtu0: " 230 "new link MTU on %s (%lu) is too small for IPv6\n", 231 if_name(ifp), (unsigned long)ndi->maxmtu); 232 } 233 234 if (ndi->maxmtu > in6_maxmtu) 235 in6_setmaxmtu(); /* check all interfaces just in case */ 236 237 #undef MIN 238 } 239 240 void 241 nd6_option_init(opt, icmp6len, ndopts) 242 void *opt; 243 int icmp6len; 244 union nd_opts *ndopts; 245 { 246 247 bzero(ndopts, sizeof(*ndopts)); 248 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; 249 ndopts->nd_opts_last 250 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); 251 252 if (icmp6len == 0) { 253 ndopts->nd_opts_done = 1; 254 ndopts->nd_opts_search = NULL; 255 } 256 } 257 258 /* 259 * Take one ND option. 260 */ 261 struct nd_opt_hdr * 262 nd6_option(ndopts) 263 union nd_opts *ndopts; 264 { 265 struct nd_opt_hdr *nd_opt; 266 int olen; 267 268 if (ndopts == NULL) 269 panic("ndopts == NULL in nd6_option"); 270 if (ndopts->nd_opts_last == NULL) 271 panic("uninitialized ndopts in nd6_option"); 272 if (ndopts->nd_opts_search == NULL) 273 return NULL; 274 if (ndopts->nd_opts_done) 275 return NULL; 276 277 nd_opt = ndopts->nd_opts_search; 278 279 /* make sure nd_opt_len is inside the buffer */ 280 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) { 281 bzero(ndopts, sizeof(*ndopts)); 282 return NULL; 283 } 284 285 olen = nd_opt->nd_opt_len << 3; 286 if (olen == 0) { 287 /* 288 * Message validation requires that all included 289 * options have a length that is greater than zero. 290 */ 291 bzero(ndopts, sizeof(*ndopts)); 292 return NULL; 293 } 294 295 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen); 296 if (ndopts->nd_opts_search > ndopts->nd_opts_last) { 297 /* option overruns the end of buffer, invalid */ 298 bzero(ndopts, sizeof(*ndopts)); 299 return NULL; 300 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { 301 /* reached the end of options chain */ 302 ndopts->nd_opts_done = 1; 303 ndopts->nd_opts_search = NULL; 304 } 305 return nd_opt; 306 } 307 308 /* 309 * Parse multiple ND options. 310 * This function is much easier to use, for ND routines that do not need 311 * multiple options of the same type. 312 */ 313 int 314 nd6_options(ndopts) 315 union nd_opts *ndopts; 316 { 317 struct nd_opt_hdr *nd_opt; 318 int i = 0; 319 320 if (ndopts == NULL) 321 panic("ndopts == NULL in nd6_options"); 322 if (ndopts->nd_opts_last == NULL) 323 panic("uninitialized ndopts in nd6_options"); 324 if (ndopts->nd_opts_search == NULL) 325 return 0; 326 327 while (1) { 328 nd_opt = nd6_option(ndopts); 329 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { 330 /* 331 * Message validation requires that all included 332 * options have a length that is greater than zero. 333 */ 334 icmp6stat.icp6s_nd_badopt++; 335 bzero(ndopts, sizeof(*ndopts)); 336 return -1; 337 } 338 339 if (nd_opt == NULL) 340 goto skip1; 341 342 switch (nd_opt->nd_opt_type) { 343 case ND_OPT_SOURCE_LINKADDR: 344 case ND_OPT_TARGET_LINKADDR: 345 case ND_OPT_MTU: 346 case ND_OPT_REDIRECTED_HEADER: 347 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { 348 nd6log((LOG_INFO, 349 "duplicated ND6 option found (type=%d)\n", 350 nd_opt->nd_opt_type)); 351 /* XXX bark? */ 352 } else { 353 ndopts->nd_opt_array[nd_opt->nd_opt_type] 354 = nd_opt; 355 } 356 break; 357 case ND_OPT_PREFIX_INFORMATION: 358 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { 359 ndopts->nd_opt_array[nd_opt->nd_opt_type] 360 = nd_opt; 361 } 362 ndopts->nd_opts_pi_end = 363 (struct nd_opt_prefix_info *)nd_opt; 364 break; 365 default: 366 /* 367 * Unknown options must be silently ignored, 368 * to accomodate future extension to the protocol. 369 */ 370 nd6log((LOG_DEBUG, 371 "nd6_options: unsupported option %d - " 372 "option ignored\n", nd_opt->nd_opt_type)); 373 } 374 375 skip1: 376 i++; 377 if (i > nd6_maxndopt) { 378 icmp6stat.icp6s_nd_toomanyopt++; 379 nd6log((LOG_INFO, "too many loop in nd opt\n")); 380 break; 381 } 382 383 if (ndopts->nd_opts_done) 384 break; 385 } 386 387 return 0; 388 } 389 390 /* 391 * ND6 timer routine to handle ND6 entries 392 */ 393 void 394 nd6_llinfo_settimer(ln, tick) 395 struct llinfo_nd6 *ln; 396 long tick; 397 { 398 if (tick < 0) { 399 ln->ln_expire = 0; 400 ln->ln_ntick = 0; 401 callout_stop(&ln->ln_timer_ch); 402 } else { 403 ln->ln_expire = time_second + tick / hz; 404 if (tick > INT_MAX) { 405 ln->ln_ntick = tick - INT_MAX; 406 callout_reset(&ln->ln_timer_ch, INT_MAX, 407 nd6_llinfo_timer, ln); 408 } else { 409 ln->ln_ntick = 0; 410 callout_reset(&ln->ln_timer_ch, tick, 411 nd6_llinfo_timer, ln); 412 } 413 } 414 } 415 416 static void 417 nd6_llinfo_timer(arg) 418 void *arg; 419 { 420 struct llinfo_nd6 *ln; 421 struct rtentry *rt; 422 struct in6_addr *dst; 423 struct ifnet *ifp; 424 struct nd_ifinfo *ndi = NULL; 425 426 ln = (struct llinfo_nd6 *)arg; 427 428 if (ln->ln_ntick > 0) { 429 if (ln->ln_ntick > INT_MAX) { 430 ln->ln_ntick -= INT_MAX; 431 nd6_llinfo_settimer(ln, INT_MAX); 432 } else { 433 ln->ln_ntick = 0; 434 nd6_llinfo_settimer(ln, ln->ln_ntick); 435 } 436 return; 437 } 438 439 if ((rt = ln->ln_rt) == NULL) 440 panic("ln->ln_rt == NULL"); 441 if ((ifp = rt->rt_ifp) == NULL) 442 panic("ln->ln_rt->rt_ifp == NULL"); 443 ndi = ND_IFINFO(ifp); 444 445 /* sanity check */ 446 if (rt->rt_llinfo && (struct llinfo_nd6 *)rt->rt_llinfo != ln) 447 panic("rt_llinfo(%p) is not equal to ln(%p)", 448 rt->rt_llinfo, ln); 449 if (rt_key(rt) == NULL) 450 panic("rt key is NULL in nd6_timer(ln=%p)", ln); 451 452 dst = &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr; 453 454 switch (ln->ln_state) { 455 case ND6_LLINFO_INCOMPLETE: 456 if (ln->ln_asked < nd6_mmaxtries) { 457 ln->ln_asked++; 458 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000); 459 nd6_ns_output(ifp, NULL, dst, ln, 0); 460 } else { 461 struct mbuf *m = ln->ln_hold; 462 if (m) { 463 struct mbuf *m0; 464 465 /* 466 * assuming every packet in ln_hold has the 467 * same IP header 468 */ 469 m0 = m->m_nextpkt; 470 m->m_nextpkt = NULL; 471 icmp6_error2(m, ICMP6_DST_UNREACH, 472 ICMP6_DST_UNREACH_ADDR, 0, rt->rt_ifp); 473 474 ln->ln_hold = m0; 475 clear_llinfo_pqueue(ln); 476 } 477 if (rt) 478 (void)nd6_free(rt, 0); 479 ln = NULL; 480 } 481 break; 482 case ND6_LLINFO_REACHABLE: 483 if (!ND6_LLINFO_PERMANENT(ln)) { 484 ln->ln_state = ND6_LLINFO_STALE; 485 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 486 } 487 break; 488 489 case ND6_LLINFO_STALE: 490 /* Garbage Collection(RFC 2461 5.3) */ 491 if (!ND6_LLINFO_PERMANENT(ln)) { 492 (void)nd6_free(rt, 1); 493 ln = NULL; 494 } 495 break; 496 497 case ND6_LLINFO_DELAY: 498 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 499 /* We need NUD */ 500 ln->ln_asked = 1; 501 ln->ln_state = ND6_LLINFO_PROBE; 502 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000); 503 nd6_ns_output(ifp, dst, dst, ln, 0); 504 } else { 505 ln->ln_state = ND6_LLINFO_STALE; /* XXX */ 506 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 507 } 508 break; 509 case ND6_LLINFO_PROBE: 510 if (ln->ln_asked < nd6_umaxtries) { 511 ln->ln_asked++; 512 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000); 513 nd6_ns_output(ifp, dst, dst, ln, 0); 514 } else if (rt->rt_ifa != NULL && 515 rt->rt_ifa->ifa_addr->sa_family == AF_INET6 && 516 (((struct in6_ifaddr *)rt->rt_ifa)->ia_flags & IFA_ROUTE)) { 517 /* 518 * This is an unreachable neighbor whose address is 519 * specified as the destination of a p2p interface 520 * (see in6_ifinit()). We should not free the entry 521 * since this is sort of a "static" entry generated 522 * via interface address configuration. 523 */ 524 ln->ln_asked = 0; 525 ln->ln_expire = 0; /* make it permanent */ 526 ln->ln_state = ND6_LLINFO_STALE; 527 } else { 528 (void)nd6_free(rt, 0); 529 ln = NULL; 530 } 531 break; 532 } 533 } 534 535 536 /* 537 * ND6 timer routine to expire default route list and prefix list 538 */ 539 void 540 nd6_timer(ignored_arg) 541 void *ignored_arg; 542 { 543 int s; 544 struct nd_defrouter *dr; 545 struct nd_prefix *pr; 546 struct in6_ifaddr *ia6, *nia6; 547 struct in6_addrlifetime *lt6; 548 549 callout_reset(&nd6_timer_ch, nd6_prune * hz, 550 nd6_timer, NULL); 551 552 /* expire default router list */ 553 s = splnet(); 554 dr = TAILQ_FIRST(&nd_defrouter); 555 while (dr) { 556 if (dr->expire && dr->expire < time_second) { 557 struct nd_defrouter *t; 558 t = TAILQ_NEXT(dr, dr_entry); 559 defrtrlist_del(dr); 560 dr = t; 561 } else { 562 dr = TAILQ_NEXT(dr, dr_entry); 563 } 564 } 565 566 /* 567 * expire interface addresses. 568 * in the past the loop was inside prefix expiry processing. 569 * However, from a stricter speci-confrmance standpoint, we should 570 * rather separate address lifetimes and prefix lifetimes. 571 */ 572 addrloop: 573 for (ia6 = in6_ifaddr; ia6; ia6 = nia6) { 574 nia6 = ia6->ia_next; 575 /* check address lifetime */ 576 lt6 = &ia6->ia6_lifetime; 577 if (IFA6_IS_INVALID(ia6)) { 578 int regen = 0; 579 580 /* 581 * If the expiring address is temporary, try 582 * regenerating a new one. This would be useful when 583 * we suspended a laptop PC, then turned it on after a 584 * period that could invalidate all temporary 585 * addresses. Although we may have to restart the 586 * loop (see below), it must be after purging the 587 * address. Otherwise, we'd see an infinite loop of 588 * regeneration. 589 */ 590 if (ip6_use_tempaddr && 591 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 592 if (regen_tmpaddr(ia6) == 0) 593 regen = 1; 594 } 595 596 in6_purgeaddr(&ia6->ia_ifa); 597 598 if (regen) 599 goto addrloop; /* XXX: see below */ 600 } else if (IFA6_IS_DEPRECATED(ia6)) { 601 int oldflags = ia6->ia6_flags; 602 603 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 604 605 /* 606 * If a temporary address has just become deprecated, 607 * regenerate a new one if possible. 608 */ 609 if (ip6_use_tempaddr && 610 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 611 (oldflags & IN6_IFF_DEPRECATED) == 0) { 612 613 if (regen_tmpaddr(ia6) == 0) { 614 /* 615 * A new temporary address is 616 * generated. 617 * XXX: this means the address chain 618 * has changed while we are still in 619 * the loop. Although the change 620 * would not cause disaster (because 621 * it's not a deletion, but an 622 * addition,) we'd rather restart the 623 * loop just for safety. Or does this 624 * significantly reduce performance?? 625 */ 626 goto addrloop; 627 } 628 } 629 } else { 630 /* 631 * A new RA might have made a deprecated address 632 * preferred. 633 */ 634 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 635 } 636 } 637 638 /* expire prefix list */ 639 pr = nd_prefix.lh_first; 640 while (pr) { 641 /* 642 * check prefix lifetime. 643 * since pltime is just for autoconf, pltime processing for 644 * prefix is not necessary. 645 */ 646 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 647 time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) { 648 struct nd_prefix *t; 649 t = pr->ndpr_next; 650 651 /* 652 * address expiration and prefix expiration are 653 * separate. NEVER perform in6_purgeaddr here. 654 */ 655 656 prelist_remove(pr); 657 pr = t; 658 } else 659 pr = pr->ndpr_next; 660 } 661 splx(s); 662 } 663 664 static int 665 regen_tmpaddr(ia6) 666 struct in6_ifaddr *ia6; /* deprecated/invalidated temporary address */ 667 { 668 struct ifaddr *ifa; 669 struct ifnet *ifp; 670 struct in6_ifaddr *public_ifa6 = NULL; 671 672 ifp = ia6->ia_ifa.ifa_ifp; 673 for (ifa = ifp->if_addrlist.tqh_first; ifa; 674 ifa = ifa->ifa_list.tqe_next) { 675 struct in6_ifaddr *it6; 676 677 if (ifa->ifa_addr->sa_family != AF_INET6) 678 continue; 679 680 it6 = (struct in6_ifaddr *)ifa; 681 682 /* ignore no autoconf addresses. */ 683 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 684 continue; 685 686 /* ignore autoconf addresses with different prefixes. */ 687 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 688 continue; 689 690 /* 691 * Now we are looking at an autoconf address with the same 692 * prefix as ours. If the address is temporary and is still 693 * preferred, do not create another one. It would be rare, but 694 * could happen, for example, when we resume a laptop PC after 695 * a long period. 696 */ 697 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 698 !IFA6_IS_DEPRECATED(it6)) { 699 public_ifa6 = NULL; 700 break; 701 } 702 703 /* 704 * This is a public autoconf address that has the same prefix 705 * as ours. If it is preferred, keep it. We can't break the 706 * loop here, because there may be a still-preferred temporary 707 * address with the prefix. 708 */ 709 if (!IFA6_IS_DEPRECATED(it6)) 710 public_ifa6 = it6; 711 } 712 713 if (public_ifa6 != NULL) { 714 int e; 715 716 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 717 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 718 " tmp addr,errno=%d\n", e); 719 return (-1); 720 } 721 return (0); 722 } 723 724 return (-1); 725 } 726 727 /* 728 * Nuke neighbor cache/prefix/default router management table, right before 729 * ifp goes away. 730 */ 731 void 732 nd6_purge(ifp) 733 struct ifnet *ifp; 734 { 735 struct llinfo_nd6 *ln, *nln; 736 struct nd_defrouter *dr, *ndr; 737 struct nd_prefix *pr, *npr; 738 739 /* 740 * Nuke default router list entries toward ifp. 741 * We defer removal of default router list entries that is installed 742 * in the routing table, in order to keep additional side effects as 743 * small as possible. 744 */ 745 for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) { 746 ndr = TAILQ_NEXT(dr, dr_entry); 747 if (dr->installed) 748 continue; 749 750 if (dr->ifp == ifp) 751 defrtrlist_del(dr); 752 } 753 754 for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = ndr) { 755 ndr = TAILQ_NEXT(dr, dr_entry); 756 if (!dr->installed) 757 continue; 758 759 if (dr->ifp == ifp) 760 defrtrlist_del(dr); 761 } 762 763 /* Nuke prefix list entries toward ifp */ 764 for (pr = nd_prefix.lh_first; pr; pr = npr) { 765 npr = pr->ndpr_next; 766 if (pr->ndpr_ifp == ifp) { 767 /* 768 * Because if_detach() does *not* release prefixes 769 * while purging addresses the reference count will 770 * still be above zero. We therefore reset it to 771 * make sure that the prefix really gets purged. 772 */ 773 pr->ndpr_refcnt = 0; 774 775 /* 776 * Previously, pr->ndpr_addr is removed as well, 777 * but I strongly believe we don't have to do it. 778 * nd6_purge() is only called from in6_ifdetach(), 779 * which removes all the associated interface addresses 780 * by itself. 781 * (jinmei@kame.net 20010129) 782 */ 783 prelist_remove(pr); 784 } 785 } 786 787 /* cancel default outgoing interface setting */ 788 if (nd6_defifindex == ifp->if_index) 789 nd6_setdefaultiface(0); 790 791 if (!ip6_forwarding && ip6_accept_rtadv) { /* XXX: too restrictive? */ 792 /* refresh default router list */ 793 defrouter_select(); 794 } 795 796 /* 797 * Nuke neighbor cache entries for the ifp. 798 * Note that rt->rt_ifp may not be the same as ifp, 799 * due to KAME goto ours hack. See RTM_RESOLVE case in 800 * nd6_rtrequest(), and ip6_input(). 801 */ 802 ln = llinfo_nd6.ln_next; 803 while (ln && ln != &llinfo_nd6) { 804 struct rtentry *rt; 805 struct sockaddr_dl *sdl; 806 807 nln = ln->ln_next; 808 rt = ln->ln_rt; 809 if (rt && rt->rt_gateway && 810 rt->rt_gateway->sa_family == AF_LINK) { 811 sdl = (struct sockaddr_dl *)rt->rt_gateway; 812 if (sdl->sdl_index == ifp->if_index) 813 nln = nd6_free(rt, 0); 814 } 815 ln = nln; 816 } 817 } 818 819 struct rtentry * 820 nd6_lookup(addr6, create, ifp) 821 struct in6_addr *addr6; 822 int create; 823 struct ifnet *ifp; 824 { 825 struct rtentry *rt; 826 struct sockaddr_in6 sin6; 827 828 bzero(&sin6, sizeof(sin6)); 829 sin6.sin6_len = sizeof(struct sockaddr_in6); 830 sin6.sin6_family = AF_INET6; 831 sin6.sin6_addr = *addr6; 832 rt = rtalloc1((struct sockaddr *)&sin6, create, 0UL); 833 if (rt) { 834 if ((rt->rt_flags & RTF_LLINFO) == 0 && create) { 835 /* 836 * This is the case for the default route. 837 * If we want to create a neighbor cache for the 838 * address, we should free the route for the 839 * destination and allocate an interface route. 840 */ 841 RTFREE_LOCKED(rt); 842 rt = NULL; 843 } 844 } 845 if (rt == NULL) { 846 if (create && ifp) { 847 int e; 848 849 /* 850 * If no route is available and create is set, 851 * we allocate a host route for the destination 852 * and treat it like an interface route. 853 * This hack is necessary for a neighbor which can't 854 * be covered by our own prefix. 855 */ 856 struct ifaddr *ifa = 857 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp); 858 if (ifa == NULL) 859 return (NULL); 860 861 /* 862 * Create a new route. RTF_LLINFO is necessary 863 * to create a Neighbor Cache entry for the 864 * destination in nd6_rtrequest which will be 865 * called in rtrequest via ifa->ifa_rtrequest. 866 */ 867 if ((e = rtrequest(RTM_ADD, (struct sockaddr *)&sin6, 868 ifa->ifa_addr, (struct sockaddr *)&all1_sa, 869 (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) & 870 ~RTF_CLONING, &rt)) != 0) { 871 log(LOG_ERR, 872 "nd6_lookup: failed to add route for a " 873 "neighbor(%s), errno=%d\n", 874 ip6_sprintf(addr6), e); 875 } 876 if (rt == NULL) 877 return (NULL); 878 RT_LOCK(rt); 879 if (rt->rt_llinfo) { 880 struct llinfo_nd6 *ln = 881 (struct llinfo_nd6 *)rt->rt_llinfo; 882 ln->ln_state = ND6_LLINFO_NOSTATE; 883 } 884 } else 885 return (NULL); 886 } 887 RT_LOCK_ASSERT(rt); 888 RT_REMREF(rt); 889 /* 890 * Validation for the entry. 891 * Note that the check for rt_llinfo is necessary because a cloned 892 * route from a parent route that has the L flag (e.g. the default 893 * route to a p2p interface) may have the flag, too, while the 894 * destination is not actually a neighbor. 895 * XXX: we can't use rt->rt_ifp to check for the interface, since 896 * it might be the loopback interface if the entry is for our 897 * own address on a non-loopback interface. Instead, we should 898 * use rt->rt_ifa->ifa_ifp, which would specify the REAL 899 * interface. 900 * Note also that ifa_ifp and ifp may differ when we connect two 901 * interfaces to a same link, install a link prefix to an interface, 902 * and try to install a neighbor cache on an interface that does not 903 * have a route to the prefix. 904 */ 905 if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 || 906 rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL || 907 (ifp && rt->rt_ifa->ifa_ifp != ifp)) { 908 if (create) { 909 nd6log((LOG_DEBUG, 910 "nd6_lookup: failed to lookup %s (if = %s)\n", 911 ip6_sprintf(addr6), 912 ifp ? if_name(ifp) : "unspec")); 913 } 914 RT_UNLOCK(rt); 915 return (NULL); 916 } 917 RT_UNLOCK(rt); /* XXX not ready to return rt locked */ 918 return (rt); 919 } 920 921 /* 922 * Test whether a given IPv6 address is a neighbor or not, ignoring 923 * the actual neighbor cache. The neighbor cache is ignored in order 924 * to not reenter the routing code from within itself. 925 */ 926 static int 927 nd6_is_new_addr_neighbor(addr, ifp) 928 struct sockaddr_in6 *addr; 929 struct ifnet *ifp; 930 { 931 struct nd_prefix *pr; 932 struct ifaddr *dstaddr; 933 934 /* 935 * A link-local address is always a neighbor. 936 * XXX: a link does not necessarily specify a single interface. 937 */ 938 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 939 struct sockaddr_in6 sin6_copy; 940 u_int32_t zone; 941 942 /* 943 * We need sin6_copy since sa6_recoverscope() may modify the 944 * content (XXX). 945 */ 946 sin6_copy = *addr; 947 if (sa6_recoverscope(&sin6_copy)) 948 return (0); /* XXX: should be impossible */ 949 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 950 return (0); 951 if (sin6_copy.sin6_scope_id == zone) 952 return (1); 953 else 954 return (0); 955 } 956 957 /* 958 * If the address matches one of our addresses, 959 * it should be a neighbor. 960 * If the address matches one of our on-link prefixes, it should be a 961 * neighbor. 962 */ 963 for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) { 964 if (pr->ndpr_ifp != ifp) 965 continue; 966 967 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) 968 continue; 969 970 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 971 &addr->sin6_addr, &pr->ndpr_mask)) 972 return (1); 973 } 974 975 /* 976 * If the address is assigned on the node of the other side of 977 * a p2p interface, the address should be a neighbor. 978 */ 979 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr); 980 if ((dstaddr != NULL) && (dstaddr->ifa_ifp == ifp)) 981 return (1); 982 983 /* 984 * If the default router list is empty, all addresses are regarded 985 * as on-link, and thus, as a neighbor. 986 * XXX: we restrict the condition to hosts, because routers usually do 987 * not have the "default router list". 988 */ 989 if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL && 990 nd6_defifindex == ifp->if_index) { 991 return (1); 992 } 993 994 return (0); 995 } 996 997 998 /* 999 * Detect if a given IPv6 address identifies a neighbor on a given link. 1000 * XXX: should take care of the destination of a p2p link? 1001 */ 1002 int 1003 nd6_is_addr_neighbor(addr, ifp) 1004 struct sockaddr_in6 *addr; 1005 struct ifnet *ifp; 1006 { 1007 1008 if (nd6_is_new_addr_neighbor(addr, ifp)) 1009 return (1); 1010 1011 /* 1012 * Even if the address matches none of our addresses, it might be 1013 * in the neighbor cache. 1014 */ 1015 if (nd6_lookup(&addr->sin6_addr, 0, ifp) != NULL) 1016 return (1); 1017 1018 return (0); 1019 } 1020 1021 /* 1022 * Free an nd6 llinfo entry. 1023 * Since the function would cause significant changes in the kernel, DO NOT 1024 * make it global, unless you have a strong reason for the change, and are sure 1025 * that the change is safe. 1026 */ 1027 static struct llinfo_nd6 * 1028 nd6_free(rt, gc) 1029 struct rtentry *rt; 1030 int gc; 1031 { 1032 struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next; 1033 struct in6_addr in6 = ((struct sockaddr_in6 *)rt_key(rt))->sin6_addr; 1034 struct nd_defrouter *dr; 1035 1036 /* 1037 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1038 * even though it is not harmful, it was not really necessary. 1039 */ 1040 1041 /* cancel timer */ 1042 nd6_llinfo_settimer(ln, -1); 1043 1044 if (!ip6_forwarding) { 1045 int s; 1046 s = splnet(); 1047 dr = defrouter_lookup(&((struct sockaddr_in6 *)rt_key(rt))->sin6_addr, 1048 rt->rt_ifp); 1049 1050 if (dr != NULL && dr->expire && 1051 ln->ln_state == ND6_LLINFO_STALE && gc) { 1052 /* 1053 * If the reason for the deletion is just garbage 1054 * collection, and the neighbor is an active default 1055 * router, do not delete it. Instead, reset the GC 1056 * timer using the router's lifetime. 1057 * Simply deleting the entry would affect default 1058 * router selection, which is not necessarily a good 1059 * thing, especially when we're using router preference 1060 * values. 1061 * XXX: the check for ln_state would be redundant, 1062 * but we intentionally keep it just in case. 1063 */ 1064 if (dr->expire > time_second) 1065 nd6_llinfo_settimer(ln, 1066 (dr->expire - time_second) * hz); 1067 else 1068 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 1069 splx(s); 1070 return (ln->ln_next); 1071 } 1072 1073 if (ln->ln_router || dr) { 1074 /* 1075 * rt6_flush must be called whether or not the neighbor 1076 * is in the Default Router List. 1077 * See a corresponding comment in nd6_na_input(). 1078 */ 1079 rt6_flush(&in6, rt->rt_ifp); 1080 } 1081 1082 if (dr) { 1083 /* 1084 * Unreachablity of a router might affect the default 1085 * router selection and on-link detection of advertised 1086 * prefixes. 1087 */ 1088 1089 /* 1090 * Temporarily fake the state to choose a new default 1091 * router and to perform on-link determination of 1092 * prefixes correctly. 1093 * Below the state will be set correctly, 1094 * or the entry itself will be deleted. 1095 */ 1096 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1097 1098 /* 1099 * Since defrouter_select() does not affect the 1100 * on-link determination and MIP6 needs the check 1101 * before the default router selection, we perform 1102 * the check now. 1103 */ 1104 pfxlist_onlink_check(); 1105 1106 /* 1107 * refresh default router list 1108 */ 1109 defrouter_select(); 1110 } 1111 splx(s); 1112 } 1113 1114 /* 1115 * Before deleting the entry, remember the next entry as the 1116 * return value. We need this because pfxlist_onlink_check() above 1117 * might have freed other entries (particularly the old next entry) as 1118 * a side effect (XXX). 1119 */ 1120 next = ln->ln_next; 1121 1122 /* 1123 * Detach the route from the routing tree and the list of neighbor 1124 * caches, and disable the route entry not to be used in already 1125 * cached routes. 1126 */ 1127 rtrequest(RTM_DELETE, rt_key(rt), (struct sockaddr *)0, 1128 rt_mask(rt), 0, (struct rtentry **)0); 1129 1130 return (next); 1131 } 1132 1133 /* 1134 * Upper-layer reachability hint for Neighbor Unreachability Detection. 1135 * 1136 * XXX cost-effective methods? 1137 */ 1138 void 1139 nd6_nud_hint(rt, dst6, force) 1140 struct rtentry *rt; 1141 struct in6_addr *dst6; 1142 int force; 1143 { 1144 struct llinfo_nd6 *ln; 1145 1146 /* 1147 * If the caller specified "rt", use that. Otherwise, resolve the 1148 * routing table by supplied "dst6". 1149 */ 1150 if (rt == NULL) { 1151 if (dst6 == NULL) 1152 return; 1153 if ((rt = nd6_lookup(dst6, 0, NULL)) == NULL) 1154 return; 1155 } 1156 1157 if ((rt->rt_flags & RTF_GATEWAY) != 0 || 1158 (rt->rt_flags & RTF_LLINFO) == 0 || 1159 rt->rt_llinfo == NULL || rt->rt_gateway == NULL || 1160 rt->rt_gateway->sa_family != AF_LINK) { 1161 /* This is not a host route. */ 1162 return; 1163 } 1164 1165 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1166 if (ln->ln_state < ND6_LLINFO_REACHABLE) 1167 return; 1168 1169 /* 1170 * if we get upper-layer reachability confirmation many times, 1171 * it is possible we have false information. 1172 */ 1173 if (!force) { 1174 ln->ln_byhint++; 1175 if (ln->ln_byhint > nd6_maxnudhint) 1176 return; 1177 } 1178 1179 ln->ln_state = ND6_LLINFO_REACHABLE; 1180 if (!ND6_LLINFO_PERMANENT(ln)) { 1181 nd6_llinfo_settimer(ln, 1182 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz); 1183 } 1184 } 1185 1186 void 1187 nd6_rtrequest(req, rt, info) 1188 int req; 1189 struct rtentry *rt; 1190 struct rt_addrinfo *info; /* xxx unused */ 1191 { 1192 struct sockaddr *gate = rt->rt_gateway; 1193 struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1194 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK}; 1195 struct ifnet *ifp = rt->rt_ifp; 1196 struct ifaddr *ifa; 1197 1198 RT_LOCK_ASSERT(rt); 1199 1200 if ((rt->rt_flags & RTF_GATEWAY) != 0) 1201 return; 1202 1203 if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) { 1204 /* 1205 * This is probably an interface direct route for a link 1206 * which does not need neighbor caches (e.g. fe80::%lo0/64). 1207 * We do not need special treatment below for such a route. 1208 * Moreover, the RTF_LLINFO flag which would be set below 1209 * would annoy the ndp(8) command. 1210 */ 1211 return; 1212 } 1213 1214 if (req == RTM_RESOLVE && 1215 (nd6_need_cache(ifp) == 0 || /* stf case */ 1216 !nd6_is_new_addr_neighbor((struct sockaddr_in6 *)rt_key(rt), 1217 ifp))) { 1218 /* 1219 * FreeBSD and BSD/OS often make a cloned host route based 1220 * on a less-specific route (e.g. the default route). 1221 * If the less specific route does not have a "gateway" 1222 * (this is the case when the route just goes to a p2p or an 1223 * stf interface), we'll mistakenly make a neighbor cache for 1224 * the host route, and will see strange neighbor solicitation 1225 * for the corresponding destination. In order to avoid the 1226 * confusion, we check if the destination of the route is 1227 * a neighbor in terms of neighbor discovery, and stop the 1228 * process if not. Additionally, we remove the LLINFO flag 1229 * so that ndp(8) will not try to get the neighbor information 1230 * of the destination. 1231 */ 1232 rt->rt_flags &= ~RTF_LLINFO; 1233 return; 1234 } 1235 1236 switch (req) { 1237 case RTM_ADD: 1238 /* 1239 * There is no backward compatibility :) 1240 * 1241 * if ((rt->rt_flags & RTF_HOST) == 0 && 1242 * SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff) 1243 * rt->rt_flags |= RTF_CLONING; 1244 */ 1245 if ((rt->rt_flags & RTF_CLONING) || 1246 ((rt->rt_flags & RTF_LLINFO) && ln == NULL)) { 1247 /* 1248 * Case 1: This route should come from a route to 1249 * interface (RTF_CLONING case) or the route should be 1250 * treated as on-link but is currently not 1251 * (RTF_LLINFO && ln == NULL case). 1252 */ 1253 rt_setgate(rt, rt_key(rt), 1254 (struct sockaddr *)&null_sdl); 1255 gate = rt->rt_gateway; 1256 SDL(gate)->sdl_type = ifp->if_type; 1257 SDL(gate)->sdl_index = ifp->if_index; 1258 if (ln) 1259 nd6_llinfo_settimer(ln, 0); 1260 if ((rt->rt_flags & RTF_CLONING) != 0) 1261 break; 1262 } 1263 /* 1264 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here. 1265 * We don't do that here since llinfo is not ready yet. 1266 * 1267 * There are also couple of other things to be discussed: 1268 * - unsolicited NA code needs improvement beforehand 1269 * - RFC2461 says we MAY send multicast unsolicited NA 1270 * (7.2.6 paragraph 4), however, it also says that we 1271 * SHOULD provide a mechanism to prevent multicast NA storm. 1272 * we don't have anything like it right now. 1273 * note that the mechanism needs a mutual agreement 1274 * between proxies, which means that we need to implement 1275 * a new protocol, or a new kludge. 1276 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA. 1277 * we need to check ip6forwarding before sending it. 1278 * (or should we allow proxy ND configuration only for 1279 * routers? there's no mention about proxy ND from hosts) 1280 */ 1281 /* FALLTHROUGH */ 1282 case RTM_RESOLVE: 1283 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) { 1284 /* 1285 * Address resolution isn't necessary for a point to 1286 * point link, so we can skip this test for a p2p link. 1287 */ 1288 if (gate->sa_family != AF_LINK || 1289 gate->sa_len < sizeof(null_sdl)) { 1290 log(LOG_DEBUG, 1291 "nd6_rtrequest: bad gateway value: %s\n", 1292 if_name(ifp)); 1293 break; 1294 } 1295 SDL(gate)->sdl_type = ifp->if_type; 1296 SDL(gate)->sdl_index = ifp->if_index; 1297 } 1298 if (ln != NULL) 1299 break; /* This happens on a route change */ 1300 /* 1301 * Case 2: This route may come from cloning, or a manual route 1302 * add with a LL address. 1303 */ 1304 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln)); 1305 rt->rt_llinfo = (caddr_t)ln; 1306 if (ln == NULL) { 1307 log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n"); 1308 break; 1309 } 1310 nd6_inuse++; 1311 nd6_allocated++; 1312 bzero(ln, sizeof(*ln)); 1313 ln->ln_rt = rt; 1314 callout_init(&ln->ln_timer_ch, 0); 1315 1316 /* this is required for "ndp" command. - shin */ 1317 if (req == RTM_ADD && (rt->rt_flags & RTF_STATIC)) { 1318 /* 1319 * gate should have some valid AF_LINK entry, 1320 * and ln->ln_expire should have some lifetime 1321 * which is specified by ndp command. 1322 */ 1323 ln->ln_state = ND6_LLINFO_REACHABLE; 1324 ln->ln_byhint = 0; 1325 } else { 1326 /* 1327 * When req == RTM_RESOLVE, rt is created and 1328 * initialized in rtrequest(), so rt_expire is 0. 1329 */ 1330 ln->ln_state = ND6_LLINFO_NOSTATE; 1331 nd6_llinfo_settimer(ln, 0); 1332 } 1333 rt->rt_flags |= RTF_LLINFO; 1334 ln->ln_next = llinfo_nd6.ln_next; 1335 llinfo_nd6.ln_next = ln; 1336 ln->ln_prev = &llinfo_nd6; 1337 ln->ln_next->ln_prev = ln; 1338 1339 /* 1340 * check if rt_key(rt) is one of my address assigned 1341 * to the interface. 1342 */ 1343 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(rt->rt_ifp, 1344 &SIN6(rt_key(rt))->sin6_addr); 1345 if (ifa) { 1346 caddr_t macp = nd6_ifptomac(ifp); 1347 nd6_llinfo_settimer(ln, -1); 1348 ln->ln_state = ND6_LLINFO_REACHABLE; 1349 ln->ln_byhint = 0; 1350 if (macp) { 1351 bcopy(macp, LLADDR(SDL(gate)), ifp->if_addrlen); 1352 SDL(gate)->sdl_alen = ifp->if_addrlen; 1353 } 1354 if (nd6_useloopback) { 1355 rt->rt_ifp = &loif[0]; /* XXX */ 1356 /* 1357 * Make sure rt_ifa be equal to the ifaddr 1358 * corresponding to the address. 1359 * We need this because when we refer 1360 * rt_ifa->ia6_flags in ip6_input, we assume 1361 * that the rt_ifa points to the address instead 1362 * of the loopback address. 1363 */ 1364 if (ifa != rt->rt_ifa) { 1365 IFAFREE(rt->rt_ifa); 1366 IFAREF(ifa); 1367 rt->rt_ifa = ifa; 1368 } 1369 } 1370 } else if (rt->rt_flags & RTF_ANNOUNCE) { 1371 nd6_llinfo_settimer(ln, -1); 1372 ln->ln_state = ND6_LLINFO_REACHABLE; 1373 ln->ln_byhint = 0; 1374 1375 /* join solicited node multicast for proxy ND */ 1376 if (ifp->if_flags & IFF_MULTICAST) { 1377 struct in6_addr llsol; 1378 int error; 1379 1380 llsol = SIN6(rt_key(rt))->sin6_addr; 1381 llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL; 1382 llsol.s6_addr32[1] = 0; 1383 llsol.s6_addr32[2] = htonl(1); 1384 llsol.s6_addr8[12] = 0xff; 1385 if (in6_setscope(&llsol, ifp, NULL)) 1386 break; 1387 if (in6_addmulti(&llsol, ifp, 1388 &error, 0) == NULL) { 1389 nd6log((LOG_ERR, "%s: failed to join " 1390 "%s (errno=%d)\n", if_name(ifp), 1391 ip6_sprintf(&llsol), error)); 1392 } 1393 } 1394 } 1395 break; 1396 1397 case RTM_DELETE: 1398 if (ln == NULL) 1399 break; 1400 /* leave from solicited node multicast for proxy ND */ 1401 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 && 1402 (ifp->if_flags & IFF_MULTICAST) != 0) { 1403 struct in6_addr llsol; 1404 struct in6_multi *in6m; 1405 1406 llsol = SIN6(rt_key(rt))->sin6_addr; 1407 llsol.s6_addr32[0] = IPV6_ADDR_INT32_MLL; 1408 llsol.s6_addr32[1] = 0; 1409 llsol.s6_addr32[2] = htonl(1); 1410 llsol.s6_addr8[12] = 0xff; 1411 if (in6_setscope(&llsol, ifp, NULL) == 0) { 1412 IN6_LOOKUP_MULTI(llsol, ifp, in6m); 1413 if (in6m) 1414 in6_delmulti(in6m); 1415 } else 1416 ; /* XXX: should not happen. bark here? */ 1417 } 1418 nd6_inuse--; 1419 ln->ln_next->ln_prev = ln->ln_prev; 1420 ln->ln_prev->ln_next = ln->ln_next; 1421 ln->ln_prev = NULL; 1422 nd6_llinfo_settimer(ln, -1); 1423 rt->rt_llinfo = 0; 1424 rt->rt_flags &= ~RTF_LLINFO; 1425 clear_llinfo_pqueue(ln); 1426 Free((caddr_t)ln); 1427 } 1428 } 1429 1430 int 1431 nd6_ioctl(cmd, data, ifp) 1432 u_long cmd; 1433 caddr_t data; 1434 struct ifnet *ifp; 1435 { 1436 struct in6_drlist *drl = (struct in6_drlist *)data; 1437 struct in6_oprlist *oprl = (struct in6_oprlist *)data; 1438 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1439 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1440 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1441 struct nd_defrouter *dr; 1442 struct nd_prefix *pr; 1443 struct rtentry *rt; 1444 int i = 0, error = 0; 1445 int s; 1446 1447 switch (cmd) { 1448 case SIOCGDRLST_IN6: 1449 /* 1450 * obsolete API, use sysctl under net.inet6.icmp6 1451 */ 1452 bzero(drl, sizeof(*drl)); 1453 s = splnet(); 1454 dr = TAILQ_FIRST(&nd_defrouter); 1455 while (dr && i < DRLSTSIZ) { 1456 drl->defrouter[i].rtaddr = dr->rtaddr; 1457 in6_clearscope(&drl->defrouter[i].rtaddr); 1458 1459 drl->defrouter[i].flags = dr->flags; 1460 drl->defrouter[i].rtlifetime = dr->rtlifetime; 1461 drl->defrouter[i].expire = dr->expire; 1462 drl->defrouter[i].if_index = dr->ifp->if_index; 1463 i++; 1464 dr = TAILQ_NEXT(dr, dr_entry); 1465 } 1466 splx(s); 1467 break; 1468 case SIOCGPRLST_IN6: 1469 /* 1470 * obsolete API, use sysctl under net.inet6.icmp6 1471 * 1472 * XXX the structure in6_prlist was changed in backward- 1473 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6, 1474 * in6_prlist is used for nd6_sysctl() - fill_prlist(). 1475 */ 1476 /* 1477 * XXX meaning of fields, especialy "raflags", is very 1478 * differnet between RA prefix list and RR/static prefix list. 1479 * how about separating ioctls into two? 1480 */ 1481 bzero(oprl, sizeof(*oprl)); 1482 s = splnet(); 1483 pr = nd_prefix.lh_first; 1484 while (pr && i < PRLSTSIZ) { 1485 struct nd_pfxrouter *pfr; 1486 int j; 1487 1488 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr; 1489 oprl->prefix[i].raflags = pr->ndpr_raf; 1490 oprl->prefix[i].prefixlen = pr->ndpr_plen; 1491 oprl->prefix[i].vltime = pr->ndpr_vltime; 1492 oprl->prefix[i].pltime = pr->ndpr_pltime; 1493 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index; 1494 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 1495 oprl->prefix[i].expire = 0; 1496 else { 1497 time_t maxexpire; 1498 1499 /* XXX: we assume time_t is signed. */ 1500 maxexpire = (-1) & 1501 ~((time_t)1 << 1502 ((sizeof(maxexpire) * 8) - 1)); 1503 if (pr->ndpr_vltime < 1504 maxexpire - pr->ndpr_lastupdate) { 1505 oprl->prefix[i].expire = 1506 pr->ndpr_lastupdate + 1507 pr->ndpr_vltime; 1508 } else 1509 oprl->prefix[i].expire = maxexpire; 1510 } 1511 1512 pfr = pr->ndpr_advrtrs.lh_first; 1513 j = 0; 1514 while (pfr) { 1515 if (j < DRLSTSIZ) { 1516 #define RTRADDR oprl->prefix[i].advrtr[j] 1517 RTRADDR = pfr->router->rtaddr; 1518 in6_clearscope(&RTRADDR); 1519 #undef RTRADDR 1520 } 1521 j++; 1522 pfr = pfr->pfr_next; 1523 } 1524 oprl->prefix[i].advrtrs = j; 1525 oprl->prefix[i].origin = PR_ORIG_RA; 1526 1527 i++; 1528 pr = pr->ndpr_next; 1529 } 1530 splx(s); 1531 1532 break; 1533 case OSIOCGIFINFO_IN6: 1534 #define ND ndi->ndi 1535 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1536 bzero(&ND, sizeof(ND)); 1537 ND.linkmtu = IN6_LINKMTU(ifp); 1538 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1539 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1540 ND.reachable = ND_IFINFO(ifp)->reachable; 1541 ND.retrans = ND_IFINFO(ifp)->retrans; 1542 ND.flags = ND_IFINFO(ifp)->flags; 1543 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1544 ND.chlim = ND_IFINFO(ifp)->chlim; 1545 break; 1546 case SIOCGIFINFO_IN6: 1547 ND = *ND_IFINFO(ifp); 1548 break; 1549 case SIOCSIFINFO_IN6: 1550 /* 1551 * used to change host variables from userland. 1552 * intented for a use on router to reflect RA configurations. 1553 */ 1554 /* 0 means 'unspecified' */ 1555 if (ND.linkmtu != 0) { 1556 if (ND.linkmtu < IPV6_MMTU || 1557 ND.linkmtu > IN6_LINKMTU(ifp)) { 1558 error = EINVAL; 1559 break; 1560 } 1561 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1562 } 1563 1564 if (ND.basereachable != 0) { 1565 int obasereachable = ND_IFINFO(ifp)->basereachable; 1566 1567 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1568 if (ND.basereachable != obasereachable) 1569 ND_IFINFO(ifp)->reachable = 1570 ND_COMPUTE_RTIME(ND.basereachable); 1571 } 1572 if (ND.retrans != 0) 1573 ND_IFINFO(ifp)->retrans = ND.retrans; 1574 if (ND.chlim != 0) 1575 ND_IFINFO(ifp)->chlim = ND.chlim; 1576 /* FALLTHROUGH */ 1577 case SIOCSIFINFO_FLAGS: 1578 ND_IFINFO(ifp)->flags = ND.flags; 1579 break; 1580 #undef ND 1581 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1582 /* sync kernel routing table with the default router list */ 1583 defrouter_reset(); 1584 defrouter_select(); 1585 break; 1586 case SIOCSPFXFLUSH_IN6: 1587 { 1588 /* flush all the prefix advertised by routers */ 1589 struct nd_prefix *pr, *next; 1590 1591 s = splnet(); 1592 for (pr = nd_prefix.lh_first; pr; pr = next) { 1593 struct in6_ifaddr *ia, *ia_next; 1594 1595 next = pr->ndpr_next; 1596 1597 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1598 continue; /* XXX */ 1599 1600 /* do we really have to remove addresses as well? */ 1601 for (ia = in6_ifaddr; ia; ia = ia_next) { 1602 /* ia might be removed. keep the next ptr. */ 1603 ia_next = ia->ia_next; 1604 1605 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1606 continue; 1607 1608 if (ia->ia6_ndpr == pr) 1609 in6_purgeaddr(&ia->ia_ifa); 1610 } 1611 prelist_remove(pr); 1612 } 1613 splx(s); 1614 break; 1615 } 1616 case SIOCSRTRFLUSH_IN6: 1617 { 1618 /* flush all the default routers */ 1619 struct nd_defrouter *dr, *next; 1620 1621 s = splnet(); 1622 defrouter_reset(); 1623 for (dr = TAILQ_FIRST(&nd_defrouter); dr; dr = next) { 1624 next = TAILQ_NEXT(dr, dr_entry); 1625 defrtrlist_del(dr); 1626 } 1627 defrouter_select(); 1628 splx(s); 1629 break; 1630 } 1631 case SIOCGNBRINFO_IN6: 1632 { 1633 struct llinfo_nd6 *ln; 1634 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1635 1636 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1637 return (error); 1638 1639 s = splnet(); 1640 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL) { 1641 error = EINVAL; 1642 splx(s); 1643 break; 1644 } 1645 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1646 nbi->state = ln->ln_state; 1647 nbi->asked = ln->ln_asked; 1648 nbi->isrouter = ln->ln_router; 1649 nbi->expire = ln->ln_expire; 1650 splx(s); 1651 1652 break; 1653 } 1654 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1655 ndif->ifindex = nd6_defifindex; 1656 break; 1657 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1658 return (nd6_setdefaultiface(ndif->ifindex)); 1659 } 1660 return (error); 1661 } 1662 1663 /* 1664 * Create neighbor cache entry and cache link-layer address, 1665 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1666 */ 1667 struct rtentry * 1668 nd6_cache_lladdr(ifp, from, lladdr, lladdrlen, type, code) 1669 struct ifnet *ifp; 1670 struct in6_addr *from; 1671 char *lladdr; 1672 int lladdrlen; 1673 int type; /* ICMP6 type */ 1674 int code; /* type dependent information */ 1675 { 1676 struct rtentry *rt = NULL; 1677 struct llinfo_nd6 *ln = NULL; 1678 int is_newentry; 1679 struct sockaddr_dl *sdl = NULL; 1680 int do_update; 1681 int olladdr; 1682 int llchange; 1683 int newstate = 0; 1684 1685 if (ifp == NULL) 1686 panic("ifp == NULL in nd6_cache_lladdr"); 1687 if (from == NULL) 1688 panic("from == NULL in nd6_cache_lladdr"); 1689 1690 /* nothing must be updated for unspecified address */ 1691 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1692 return NULL; 1693 1694 /* 1695 * Validation about ifp->if_addrlen and lladdrlen must be done in 1696 * the caller. 1697 * 1698 * XXX If the link does not have link-layer adderss, what should 1699 * we do? (ifp->if_addrlen == 0) 1700 * Spec says nothing in sections for RA, RS and NA. There's small 1701 * description on it in NS section (RFC 2461 7.2.3). 1702 */ 1703 1704 rt = nd6_lookup(from, 0, ifp); 1705 if (rt == NULL) { 1706 rt = nd6_lookup(from, 1, ifp); 1707 is_newentry = 1; 1708 } else { 1709 /* do nothing if static ndp is set */ 1710 if (rt->rt_flags & RTF_STATIC) 1711 return NULL; 1712 is_newentry = 0; 1713 } 1714 1715 if (rt == NULL) 1716 return NULL; 1717 if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) { 1718 fail: 1719 (void)nd6_free(rt, 0); 1720 return NULL; 1721 } 1722 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1723 if (ln == NULL) 1724 goto fail; 1725 if (rt->rt_gateway == NULL) 1726 goto fail; 1727 if (rt->rt_gateway->sa_family != AF_LINK) 1728 goto fail; 1729 sdl = SDL(rt->rt_gateway); 1730 1731 olladdr = (sdl->sdl_alen) ? 1 : 0; 1732 if (olladdr && lladdr) { 1733 if (bcmp(lladdr, LLADDR(sdl), ifp->if_addrlen)) 1734 llchange = 1; 1735 else 1736 llchange = 0; 1737 } else 1738 llchange = 0; 1739 1740 /* 1741 * newentry olladdr lladdr llchange (*=record) 1742 * 0 n n -- (1) 1743 * 0 y n -- (2) 1744 * 0 n y -- (3) * STALE 1745 * 0 y y n (4) * 1746 * 0 y y y (5) * STALE 1747 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1748 * 1 -- y -- (7) * STALE 1749 */ 1750 1751 if (lladdr) { /* (3-5) and (7) */ 1752 /* 1753 * Record source link-layer address 1754 * XXX is it dependent to ifp->if_type? 1755 */ 1756 sdl->sdl_alen = ifp->if_addrlen; 1757 bcopy(lladdr, LLADDR(sdl), ifp->if_addrlen); 1758 } 1759 1760 if (!is_newentry) { 1761 if ((!olladdr && lladdr != NULL) || /* (3) */ 1762 (olladdr && lladdr != NULL && llchange)) { /* (5) */ 1763 do_update = 1; 1764 newstate = ND6_LLINFO_STALE; 1765 } else /* (1-2,4) */ 1766 do_update = 0; 1767 } else { 1768 do_update = 1; 1769 if (lladdr == NULL) /* (6) */ 1770 newstate = ND6_LLINFO_NOSTATE; 1771 else /* (7) */ 1772 newstate = ND6_LLINFO_STALE; 1773 } 1774 1775 if (do_update) { 1776 /* 1777 * Update the state of the neighbor cache. 1778 */ 1779 ln->ln_state = newstate; 1780 1781 if (ln->ln_state == ND6_LLINFO_STALE) { 1782 /* 1783 * XXX: since nd6_output() below will cause 1784 * state tansition to DELAY and reset the timer, 1785 * we must set the timer now, although it is actually 1786 * meaningless. 1787 */ 1788 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 1789 1790 if (ln->ln_hold) { 1791 struct mbuf *m_hold, *m_hold_next; 1792 for (m_hold = ln->ln_hold; m_hold; 1793 m_hold = m_hold_next) { 1794 struct mbuf *mpkt = NULL; 1795 1796 m_hold_next = m_hold->m_nextpkt; 1797 mpkt = m_copym(m_hold, 0, M_COPYALL, M_DONTWAIT); 1798 if (mpkt == NULL) { 1799 m_freem(m_hold); 1800 break; 1801 } 1802 mpkt->m_nextpkt = NULL; 1803 1804 /* 1805 * we assume ifp is not a p2p here, so 1806 * just set the 2nd argument as the 1807 * 1st one. 1808 */ 1809 nd6_output(ifp, ifp, mpkt, 1810 (struct sockaddr_in6 *)rt_key(rt), 1811 rt); 1812 } 1813 ln->ln_hold = NULL; 1814 } 1815 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { 1816 /* probe right away */ 1817 nd6_llinfo_settimer((void *)ln, 0); 1818 } 1819 } 1820 1821 /* 1822 * ICMP6 type dependent behavior. 1823 * 1824 * NS: clear IsRouter if new entry 1825 * RS: clear IsRouter 1826 * RA: set IsRouter if there's lladdr 1827 * redir: clear IsRouter if new entry 1828 * 1829 * RA case, (1): 1830 * The spec says that we must set IsRouter in the following cases: 1831 * - If lladdr exist, set IsRouter. This means (1-5). 1832 * - If it is old entry (!newentry), set IsRouter. This means (7). 1833 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1834 * A quetion arises for (1) case. (1) case has no lladdr in the 1835 * neighbor cache, this is similar to (6). 1836 * This case is rare but we figured that we MUST NOT set IsRouter. 1837 * 1838 * newentry olladdr lladdr llchange NS RS RA redir 1839 * D R 1840 * 0 n n -- (1) c ? s 1841 * 0 y n -- (2) c s s 1842 * 0 n y -- (3) c s s 1843 * 0 y y n (4) c s s 1844 * 0 y y y (5) c s s 1845 * 1 -- n -- (6) c c c s 1846 * 1 -- y -- (7) c c s c s 1847 * 1848 * (c=clear s=set) 1849 */ 1850 switch (type & 0xff) { 1851 case ND_NEIGHBOR_SOLICIT: 1852 /* 1853 * New entry must have is_router flag cleared. 1854 */ 1855 if (is_newentry) /* (6-7) */ 1856 ln->ln_router = 0; 1857 break; 1858 case ND_REDIRECT: 1859 /* 1860 * If the icmp is a redirect to a better router, always set the 1861 * is_router flag. Otherwise, if the entry is newly created, 1862 * clear the flag. [RFC 2461, sec 8.3] 1863 */ 1864 if (code == ND_REDIRECT_ROUTER) 1865 ln->ln_router = 1; 1866 else if (is_newentry) /* (6-7) */ 1867 ln->ln_router = 0; 1868 break; 1869 case ND_ROUTER_SOLICIT: 1870 /* 1871 * is_router flag must always be cleared. 1872 */ 1873 ln->ln_router = 0; 1874 break; 1875 case ND_ROUTER_ADVERT: 1876 /* 1877 * Mark an entry with lladdr as a router. 1878 */ 1879 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */ 1880 (is_newentry && lladdr)) { /* (7) */ 1881 ln->ln_router = 1; 1882 } 1883 break; 1884 } 1885 1886 /* 1887 * When the link-layer address of a router changes, select the 1888 * best router again. In particular, when the neighbor entry is newly 1889 * created, it might affect the selection policy. 1890 * Question: can we restrict the first condition to the "is_newentry" 1891 * case? 1892 * XXX: when we hear an RA from a new router with the link-layer 1893 * address option, defrouter_select() is called twice, since 1894 * defrtrlist_update called the function as well. However, I believe 1895 * we can compromise the overhead, since it only happens the first 1896 * time. 1897 * XXX: although defrouter_select() should not have a bad effect 1898 * for those are not autoconfigured hosts, we explicitly avoid such 1899 * cases for safety. 1900 */ 1901 if (do_update && ln->ln_router && !ip6_forwarding && ip6_accept_rtadv) 1902 defrouter_select(); 1903 1904 return rt; 1905 } 1906 1907 static void 1908 nd6_slowtimo(ignored_arg) 1909 void *ignored_arg; 1910 { 1911 struct nd_ifinfo *nd6if; 1912 struct ifnet *ifp; 1913 1914 callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 1915 nd6_slowtimo, NULL); 1916 IFNET_RLOCK(); 1917 for (ifp = TAILQ_FIRST(&ifnet); ifp; ifp = TAILQ_NEXT(ifp, if_list)) { 1918 nd6if = ND_IFINFO(ifp); 1919 if (nd6if->basereachable && /* already initialized */ 1920 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 1921 /* 1922 * Since reachable time rarely changes by router 1923 * advertisements, we SHOULD insure that a new random 1924 * value gets recomputed at least once every few hours. 1925 * (RFC 2461, 6.3.4) 1926 */ 1927 nd6if->recalctm = nd6_recalc_reachtm_interval; 1928 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 1929 } 1930 } 1931 IFNET_RUNLOCK(); 1932 } 1933 1934 #define senderr(e) { error = (e); goto bad;} 1935 int 1936 nd6_output(ifp, origifp, m0, dst, rt0) 1937 struct ifnet *ifp; 1938 struct ifnet *origifp; 1939 struct mbuf *m0; 1940 struct sockaddr_in6 *dst; 1941 struct rtentry *rt0; 1942 { 1943 struct mbuf *m = m0; 1944 struct rtentry *rt = rt0; 1945 struct sockaddr_in6 *gw6 = NULL; 1946 struct llinfo_nd6 *ln = NULL; 1947 int error = 0; 1948 1949 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) 1950 goto sendpkt; 1951 1952 if (nd6_need_cache(ifp) == 0) 1953 goto sendpkt; 1954 1955 /* 1956 * next hop determination. This routine is derived from ether_output. 1957 */ 1958 again: 1959 if (rt) { 1960 if ((rt->rt_flags & RTF_UP) == 0) { 1961 rt0 = rt = rtalloc1((struct sockaddr *)dst, 1, 0UL); 1962 if (rt != NULL) { 1963 RT_REMREF(rt); 1964 RT_UNLOCK(rt); 1965 if (rt->rt_ifp != ifp) 1966 /* 1967 * XXX maybe we should update ifp too, 1968 * but the original code didn't and I 1969 * don't know what is correct here. 1970 */ 1971 goto again; 1972 } else 1973 senderr(EHOSTUNREACH); 1974 } 1975 1976 if (rt->rt_flags & RTF_GATEWAY) { 1977 gw6 = (struct sockaddr_in6 *)rt->rt_gateway; 1978 1979 /* 1980 * We skip link-layer address resolution and NUD 1981 * if the gateway is not a neighbor from ND point 1982 * of view, regardless of the value of nd_ifinfo.flags. 1983 * The second condition is a bit tricky; we skip 1984 * if the gateway is our own address, which is 1985 * sometimes used to install a route to a p2p link. 1986 */ 1987 if (!nd6_is_addr_neighbor(gw6, ifp) || 1988 in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) { 1989 /* 1990 * We allow this kind of tricky route only 1991 * when the outgoing interface is p2p. 1992 * XXX: we may need a more generic rule here. 1993 */ 1994 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 1995 senderr(EHOSTUNREACH); 1996 1997 goto sendpkt; 1998 } 1999 2000 if (rt->rt_gwroute == 0) 2001 goto lookup; 2002 if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) { 2003 RT_LOCK(rt); 2004 rtfree(rt); rt = rt0; 2005 lookup: 2006 rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1, 0UL); 2007 if ((rt = rt->rt_gwroute) == 0) 2008 senderr(EHOSTUNREACH); 2009 RT_UNLOCK(rt); 2010 } 2011 } 2012 } 2013 2014 /* 2015 * Address resolution or Neighbor Unreachability Detection 2016 * for the next hop. 2017 * At this point, the destination of the packet must be a unicast 2018 * or an anycast address(i.e. not a multicast). 2019 */ 2020 2021 /* Look up the neighbor cache for the nexthop */ 2022 if (rt && (rt->rt_flags & RTF_LLINFO) != 0) 2023 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 2024 else { 2025 /* 2026 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 2027 * the condition below is not very efficient. But we believe 2028 * it is tolerable, because this should be a rare case. 2029 */ 2030 if (nd6_is_addr_neighbor(dst, ifp) && 2031 (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL) 2032 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 2033 } 2034 if (ln == NULL || rt == NULL) { 2035 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && 2036 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 2037 log(LOG_DEBUG, 2038 "nd6_output: can't allocate llinfo for %s " 2039 "(ln=%p, rt=%p)\n", 2040 ip6_sprintf(&dst->sin6_addr), ln, rt); 2041 senderr(EIO); /* XXX: good error? */ 2042 } 2043 2044 goto sendpkt; /* send anyway */ 2045 } 2046 2047 /* We don't have to do link-layer address resolution on a p2p link. */ 2048 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && 2049 ln->ln_state < ND6_LLINFO_REACHABLE) { 2050 ln->ln_state = ND6_LLINFO_STALE; 2051 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 2052 } 2053 2054 /* 2055 * The first time we send a packet to a neighbor whose entry is 2056 * STALE, we have to change the state to DELAY and a sets a timer to 2057 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 2058 * neighbor unreachability detection on expiration. 2059 * (RFC 2461 7.3.3) 2060 */ 2061 if (ln->ln_state == ND6_LLINFO_STALE) { 2062 ln->ln_asked = 0; 2063 ln->ln_state = ND6_LLINFO_DELAY; 2064 nd6_llinfo_settimer(ln, (long)nd6_delay * hz); 2065 } 2066 2067 /* 2068 * If the neighbor cache entry has a state other than INCOMPLETE 2069 * (i.e. its link-layer address is already resolved), just 2070 * send the packet. 2071 */ 2072 if (ln->ln_state > ND6_LLINFO_INCOMPLETE) 2073 goto sendpkt; 2074 2075 /* 2076 * There is a neighbor cache entry, but no ethernet address 2077 * response yet. Append this latest packet to the end of the 2078 * packet queue in the mbuf, unless the number of the packet 2079 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 2080 * the oldest packet in the queue will be removed. 2081 */ 2082 if (ln->ln_state == ND6_LLINFO_NOSTATE) 2083 ln->ln_state = ND6_LLINFO_INCOMPLETE; 2084 if (ln->ln_hold) { 2085 struct mbuf *m_hold; 2086 int i; 2087 2088 i = 0; 2089 for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) { 2090 i++; 2091 if (m_hold->m_nextpkt == NULL) { 2092 m_hold->m_nextpkt = m; 2093 break; 2094 } 2095 } 2096 while (i >= nd6_maxqueuelen) { 2097 m_hold = ln->ln_hold; 2098 ln->ln_hold = ln->ln_hold->m_nextpkt; 2099 m_freem(m_hold); 2100 i--; 2101 } 2102 } else { 2103 ln->ln_hold = m; 2104 } 2105 2106 /* 2107 * If there has been no NS for the neighbor after entering the 2108 * INCOMPLETE state, send the first solicitation. 2109 */ 2110 if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) { 2111 ln->ln_asked++; 2112 nd6_llinfo_settimer(ln, 2113 (long)ND_IFINFO(ifp)->retrans * hz / 1000); 2114 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); 2115 } 2116 return (0); 2117 2118 sendpkt: 2119 /* discard the packet if IPv6 operation is disabled on the interface */ 2120 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 2121 error = ENETDOWN; /* better error? */ 2122 goto bad; 2123 } 2124 2125 #ifdef IPSEC 2126 /* clean ipsec history once it goes out of the node */ 2127 ipsec_delaux(m); 2128 #endif 2129 2130 #ifdef MAC 2131 mac_create_mbuf_linklayer(ifp, m); 2132 #endif 2133 if ((ifp->if_flags & IFF_LOOPBACK) != 0) { 2134 return ((*ifp->if_output)(origifp, m, (struct sockaddr *)dst, 2135 rt)); 2136 } 2137 return ((*ifp->if_output)(ifp, m, (struct sockaddr *)dst, rt)); 2138 2139 bad: 2140 if (m) 2141 m_freem(m); 2142 return (error); 2143 } 2144 #undef senderr 2145 2146 int 2147 nd6_need_cache(ifp) 2148 struct ifnet *ifp; 2149 { 2150 /* 2151 * XXX: we currently do not make neighbor cache on any interface 2152 * other than ARCnet, Ethernet, FDDI and GIF. 2153 * 2154 * RFC2893 says: 2155 * - unidirectional tunnels needs no ND 2156 */ 2157 switch (ifp->if_type) { 2158 case IFT_ARCNET: 2159 case IFT_ETHER: 2160 case IFT_FDDI: 2161 case IFT_IEEE1394: 2162 #ifdef IFT_L2VLAN 2163 case IFT_L2VLAN: 2164 #endif 2165 #ifdef IFT_IEEE80211 2166 case IFT_IEEE80211: 2167 #endif 2168 #ifdef IFT_CARP 2169 case IFT_CARP: 2170 #endif 2171 case IFT_GIF: /* XXX need more cases? */ 2172 case IFT_PPP: 2173 case IFT_TUNNEL: 2174 case IFT_BRIDGE: 2175 return (1); 2176 default: 2177 return (0); 2178 } 2179 } 2180 2181 int 2182 nd6_storelladdr(ifp, rt0, m, dst, desten) 2183 struct ifnet *ifp; 2184 struct rtentry *rt0; 2185 struct mbuf *m; 2186 struct sockaddr *dst; 2187 u_char *desten; 2188 { 2189 struct sockaddr_dl *sdl; 2190 struct rtentry *rt; 2191 int error; 2192 2193 if (m->m_flags & M_MCAST) { 2194 int i; 2195 2196 switch (ifp->if_type) { 2197 case IFT_ETHER: 2198 case IFT_FDDI: 2199 #ifdef IFT_L2VLAN 2200 case IFT_L2VLAN: 2201 #endif 2202 #ifdef IFT_IEEE80211 2203 case IFT_IEEE80211: 2204 #endif 2205 case IFT_BRIDGE: 2206 case IFT_ISO88025: 2207 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr, 2208 desten); 2209 return (0); 2210 case IFT_IEEE1394: 2211 /* 2212 * netbsd can use if_broadcastaddr, but we don't do so 2213 * to reduce # of ifdef. 2214 */ 2215 for (i = 0; i < ifp->if_addrlen; i++) 2216 desten[i] = ~0; 2217 return (0); 2218 case IFT_ARCNET: 2219 *desten = 0; 2220 return (0); 2221 default: 2222 m_freem(m); 2223 return (EAFNOSUPPORT); 2224 } 2225 } 2226 2227 if (rt0 == NULL) { 2228 /* this could happen, if we could not allocate memory */ 2229 m_freem(m); 2230 return (ENOMEM); 2231 } 2232 2233 error = rt_check(&rt, &rt0, dst); 2234 if (error) { 2235 m_freem(m); 2236 return (error); 2237 } 2238 RT_UNLOCK(rt); 2239 2240 if (rt->rt_gateway->sa_family != AF_LINK) { 2241 printf("nd6_storelladdr: something odd happens\n"); 2242 m_freem(m); 2243 return (EINVAL); 2244 } 2245 sdl = SDL(rt->rt_gateway); 2246 if (sdl->sdl_alen == 0) { 2247 /* this should be impossible, but we bark here for debugging */ 2248 printf("nd6_storelladdr: sdl_alen == 0\n"); 2249 m_freem(m); 2250 return (EINVAL); 2251 } 2252 2253 bcopy(LLADDR(sdl), desten, sdl->sdl_alen); 2254 return (0); 2255 } 2256 2257 static void 2258 clear_llinfo_pqueue(ln) 2259 struct llinfo_nd6 *ln; 2260 { 2261 struct mbuf *m_hold, *m_hold_next; 2262 2263 for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) { 2264 m_hold_next = m_hold->m_nextpkt; 2265 m_hold->m_nextpkt = NULL; 2266 m_freem(m_hold); 2267 } 2268 2269 ln->ln_hold = NULL; 2270 return; 2271 } 2272 2273 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2274 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2275 #ifdef SYSCTL_DECL 2276 SYSCTL_DECL(_net_inet6_icmp6); 2277 #endif 2278 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2279 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2280 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2281 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2282 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2283 CTLFLAG_RW, &nd6_maxqueuelen, 1, ""); 2284 2285 static int 2286 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2287 { 2288 int error; 2289 char buf[1024]; 2290 struct in6_defrouter *d, *de; 2291 struct nd_defrouter *dr; 2292 2293 if (req->newptr) 2294 return EPERM; 2295 error = 0; 2296 2297 for (dr = TAILQ_FIRST(&nd_defrouter); dr; 2298 dr = TAILQ_NEXT(dr, dr_entry)) { 2299 d = (struct in6_defrouter *)buf; 2300 de = (struct in6_defrouter *)(buf + sizeof(buf)); 2301 2302 if (d + 1 <= de) { 2303 bzero(d, sizeof(*d)); 2304 d->rtaddr.sin6_family = AF_INET6; 2305 d->rtaddr.sin6_len = sizeof(d->rtaddr); 2306 d->rtaddr.sin6_addr = dr->rtaddr; 2307 sa6_recoverscope(&d->rtaddr); 2308 d->flags = dr->flags; 2309 d->rtlifetime = dr->rtlifetime; 2310 d->expire = dr->expire; 2311 d->if_index = dr->ifp->if_index; 2312 } else 2313 panic("buffer too short"); 2314 2315 error = SYSCTL_OUT(req, buf, sizeof(*d)); 2316 if (error) 2317 break; 2318 } 2319 2320 return (error); 2321 } 2322 2323 static int 2324 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2325 { 2326 int error; 2327 char buf[1024]; 2328 struct in6_prefix *p, *pe; 2329 struct nd_prefix *pr; 2330 2331 if (req->newptr) 2332 return EPERM; 2333 error = 0; 2334 2335 for (pr = nd_prefix.lh_first; pr; pr = pr->ndpr_next) { 2336 u_short advrtrs; 2337 size_t advance; 2338 struct sockaddr_in6 *sin6, *s6; 2339 struct nd_pfxrouter *pfr; 2340 2341 p = (struct in6_prefix *)buf; 2342 pe = (struct in6_prefix *)(buf + sizeof(buf)); 2343 2344 if (p + 1 <= pe) { 2345 bzero(p, sizeof(*p)); 2346 sin6 = (struct sockaddr_in6 *)(p + 1); 2347 2348 p->prefix = pr->ndpr_prefix; 2349 if (sa6_recoverscope(&p->prefix)) { 2350 log(LOG_ERR, 2351 "scope error in prefix list (%s)\n", 2352 ip6_sprintf(&p->prefix.sin6_addr)); 2353 /* XXX: press on... */ 2354 } 2355 p->raflags = pr->ndpr_raf; 2356 p->prefixlen = pr->ndpr_plen; 2357 p->vltime = pr->ndpr_vltime; 2358 p->pltime = pr->ndpr_pltime; 2359 p->if_index = pr->ndpr_ifp->if_index; 2360 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2361 p->expire = 0; 2362 else { 2363 time_t maxexpire; 2364 2365 /* XXX: we assume time_t is signed. */ 2366 maxexpire = (-1) & 2367 ~((time_t)1 << 2368 ((sizeof(maxexpire) * 8) - 1)); 2369 if (pr->ndpr_vltime < 2370 maxexpire - pr->ndpr_lastupdate) { 2371 p->expire = pr->ndpr_lastupdate + 2372 pr->ndpr_vltime; 2373 } else 2374 p->expire = maxexpire; 2375 } 2376 p->refcnt = pr->ndpr_refcnt; 2377 p->flags = pr->ndpr_stateflags; 2378 p->origin = PR_ORIG_RA; 2379 advrtrs = 0; 2380 for (pfr = pr->ndpr_advrtrs.lh_first; pfr; 2381 pfr = pfr->pfr_next) { 2382 if ((void *)&sin6[advrtrs + 1] > (void *)pe) { 2383 advrtrs++; 2384 continue; 2385 } 2386 s6 = &sin6[advrtrs]; 2387 bzero(s6, sizeof(*s6)); 2388 s6->sin6_family = AF_INET6; 2389 s6->sin6_len = sizeof(*sin6); 2390 s6->sin6_addr = pfr->router->rtaddr; 2391 if (sa6_recoverscope(s6)) { 2392 log(LOG_ERR, 2393 "scope error in " 2394 "prefix list (%s)\n", 2395 ip6_sprintf(&pfr->router->rtaddr)); 2396 } 2397 advrtrs++; 2398 } 2399 p->advrtrs = advrtrs; 2400 } else 2401 panic("buffer too short"); 2402 2403 advance = sizeof(*p) + sizeof(*sin6) * advrtrs; 2404 error = SYSCTL_OUT(req, buf, advance); 2405 if (error) 2406 break; 2407 } 2408 2409 return (error); 2410 } 2411