1 /* $NetBSD: nd6.c,v 1.147 2014/01/15 10:25:04 roy Exp $ */ 2 /* $KAME: nd6.c,v 1.279 2002/06/08 11:16:51 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 <sys/cdefs.h> 34 __KERNEL_RCSID(0, "$NetBSD: nd6.c,v 1.147 2014/01/15 10:25:04 roy Exp $"); 35 36 #include "opt_ipsec.h" 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/callout.h> 41 #include <sys/malloc.h> 42 #include <sys/mbuf.h> 43 #include <sys/socket.h> 44 #include <sys/socketvar.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/ioctl.h> 51 #include <sys/syslog.h> 52 #include <sys/queue.h> 53 #include <sys/cprng.h> 54 55 #include <net/if.h> 56 #include <net/if_dl.h> 57 #include <net/if_types.h> 58 #include <net/route.h> 59 #include <net/if_ether.h> 60 #include <net/if_fddi.h> 61 #include <net/if_arc.h> 62 63 #include <netinet/in.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 #include <netinet6/icmp6_private.h> 71 72 #include <net/net_osdep.h> 73 74 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ 75 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ 76 77 /* timer values */ 78 int nd6_prune = 1; /* walk list every 1 seconds */ 79 int nd6_delay = 5; /* delay first probe time 5 second */ 80 int nd6_umaxtries = 3; /* maximum unicast query */ 81 int nd6_mmaxtries = 3; /* maximum multicast query */ 82 int nd6_useloopback = 1; /* use loopback interface for local traffic */ 83 int nd6_gctimer = (60 * 60 * 24); /* 1 day: garbage collection timer */ 84 85 /* preventing too many loops in ND option parsing */ 86 int nd6_maxndopt = 10; /* max # of ND options allowed */ 87 88 int nd6_maxnudhint = 0; /* max # of subsequent upper layer hints */ 89 90 int nd6_maxqueuelen = 1; /* max # of packets cached in unresolved ND entries */ 91 92 #ifdef ND6_DEBUG 93 int nd6_debug = 1; 94 #else 95 int nd6_debug = 0; 96 #endif 97 98 /* for debugging? */ 99 static int nd6_inuse, nd6_allocated; 100 101 struct llinfo_nd6 llinfo_nd6 = { 102 .ln_prev = &llinfo_nd6, 103 .ln_next = &llinfo_nd6, 104 }; 105 struct nd_drhead nd_defrouter; 106 struct nd_prhead nd_prefix = { 0 }; 107 108 int nd6_recalc_reachtm_interval = ND6_RECALC_REACHTM_INTERVAL; 109 static const struct sockaddr_in6 all1_sa = { 110 .sin6_family = AF_INET6 111 , .sin6_len = sizeof(struct sockaddr_in6) 112 , .sin6_addr = {.s6_addr = {0xff, 0xff, 0xff, 0xff, 113 0xff, 0xff, 0xff, 0xff, 114 0xff, 0xff, 0xff, 0xff, 115 0xff, 0xff, 0xff, 0xff}} 116 }; 117 118 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *); 119 static void nd6_slowtimo(void *); 120 static int regen_tmpaddr(struct in6_ifaddr *); 121 static struct llinfo_nd6 *nd6_free(struct rtentry *, int); 122 static void nd6_llinfo_timer(void *); 123 static void clear_llinfo_pqueue(struct llinfo_nd6 *); 124 125 callout_t nd6_slowtimo_ch; 126 callout_t nd6_timer_ch; 127 extern callout_t in6_tmpaddrtimer_ch; 128 129 static int fill_drlist(void *, size_t *, size_t); 130 static int fill_prlist(void *, size_t *, size_t); 131 132 MALLOC_DEFINE(M_IP6NDP, "NDP", "IPv6 Neighbour Discovery"); 133 134 #define LN_DEQUEUE(ln) do { \ 135 (ln)->ln_next->ln_prev = (ln)->ln_prev; \ 136 (ln)->ln_prev->ln_next = (ln)->ln_next; \ 137 } while (/*CONSTCOND*/0) 138 #define LN_INSERTHEAD(ln) do { \ 139 (ln)->ln_next = llinfo_nd6.ln_next; \ 140 llinfo_nd6.ln_next = (ln); \ 141 (ln)->ln_prev = &llinfo_nd6; \ 142 (ln)->ln_next->ln_prev = (ln); \ 143 } while (/*CONSTCOND*/0) 144 void 145 nd6_init(void) 146 { 147 static int nd6_init_done = 0; 148 149 if (nd6_init_done) { 150 log(LOG_NOTICE, "nd6_init called more than once(ignored)\n"); 151 return; 152 } 153 154 /* initialization of the default router list */ 155 TAILQ_INIT(&nd_defrouter); 156 157 nd6_init_done = 1; 158 159 callout_init(&nd6_slowtimo_ch, CALLOUT_MPSAFE); 160 callout_init(&nd6_timer_ch, CALLOUT_MPSAFE); 161 162 /* start timer */ 163 callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 164 nd6_slowtimo, NULL); 165 } 166 167 struct nd_ifinfo * 168 nd6_ifattach(struct ifnet *ifp) 169 { 170 struct nd_ifinfo *nd; 171 172 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO); 173 174 nd->initialized = 1; 175 176 nd->chlim = IPV6_DEFHLIM; 177 nd->basereachable = REACHABLE_TIME; 178 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable); 179 nd->retrans = RETRANS_TIMER; 180 /* 181 * Note that the default value of ip6_accept_rtadv is 0. 182 * Because we do not set ND6_IFF_OVERRIDE_RTADV here, we won't 183 * accept RAs by default. 184 */ 185 nd->flags = ND6_IFF_PERFORMNUD | ND6_IFF_ACCEPT_RTADV; 186 187 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */ 188 nd6_setmtu0(ifp, nd); 189 190 return nd; 191 } 192 193 void 194 nd6_ifdetach(struct nd_ifinfo *nd) 195 { 196 197 free(nd, M_IP6NDP); 198 } 199 200 void 201 nd6_setmtu(struct ifnet *ifp) 202 { 203 nd6_setmtu0(ifp, ND_IFINFO(ifp)); 204 } 205 206 void 207 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi) 208 { 209 u_int32_t omaxmtu; 210 211 omaxmtu = ndi->maxmtu; 212 213 switch (ifp->if_type) { 214 case IFT_ARCNET: 215 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */ 216 break; 217 case IFT_FDDI: 218 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); 219 break; 220 default: 221 ndi->maxmtu = ifp->if_mtu; 222 break; 223 } 224 225 /* 226 * Decreasing the interface MTU under IPV6 minimum MTU may cause 227 * undesirable situation. We thus notify the operator of the change 228 * explicitly. The check for omaxmtu is necessary to restrict the 229 * log to the case of changing the MTU, not initializing it. 230 */ 231 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { 232 log(LOG_NOTICE, "nd6_setmtu0: new link MTU on %s (%lu) is too" 233 " small for IPv6 which needs %lu\n", 234 if_name(ifp), (unsigned long)ndi->maxmtu, (unsigned long) 235 IPV6_MMTU); 236 } 237 238 if (ndi->maxmtu > in6_maxmtu) 239 in6_setmaxmtu(); /* check all interfaces just in case */ 240 } 241 242 void 243 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts) 244 { 245 246 memset(ndopts, 0, sizeof(*ndopts)); 247 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; 248 ndopts->nd_opts_last 249 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); 250 251 if (icmp6len == 0) { 252 ndopts->nd_opts_done = 1; 253 ndopts->nd_opts_search = NULL; 254 } 255 } 256 257 /* 258 * Take one ND option. 259 */ 260 struct nd_opt_hdr * 261 nd6_option(union nd_opts *ndopts) 262 { 263 struct nd_opt_hdr *nd_opt; 264 int olen; 265 266 if (ndopts == NULL) 267 panic("ndopts == NULL in nd6_option"); 268 if (ndopts->nd_opts_last == NULL) 269 panic("uninitialized ndopts in nd6_option"); 270 if (ndopts->nd_opts_search == NULL) 271 return NULL; 272 if (ndopts->nd_opts_done) 273 return NULL; 274 275 nd_opt = ndopts->nd_opts_search; 276 277 /* make sure nd_opt_len is inside the buffer */ 278 if ((void *)&nd_opt->nd_opt_len >= (void *)ndopts->nd_opts_last) { 279 memset(ndopts, 0, sizeof(*ndopts)); 280 return NULL; 281 } 282 283 olen = nd_opt->nd_opt_len << 3; 284 if (olen == 0) { 285 /* 286 * Message validation requires that all included 287 * options have a length that is greater than zero. 288 */ 289 memset(ndopts, 0, sizeof(*ndopts)); 290 return NULL; 291 } 292 293 ndopts->nd_opts_search = (struct nd_opt_hdr *)((char *)nd_opt + olen); 294 if (ndopts->nd_opts_search > ndopts->nd_opts_last) { 295 /* option overruns the end of buffer, invalid */ 296 memset(ndopts, 0, sizeof(*ndopts)); 297 return NULL; 298 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { 299 /* reached the end of options chain */ 300 ndopts->nd_opts_done = 1; 301 ndopts->nd_opts_search = NULL; 302 } 303 return nd_opt; 304 } 305 306 /* 307 * Parse multiple ND options. 308 * This function is much easier to use, for ND routines that do not need 309 * multiple options of the same type. 310 */ 311 int 312 nd6_options(union nd_opts *ndopts) 313 { 314 struct nd_opt_hdr *nd_opt; 315 int i = 0; 316 317 if (ndopts == NULL) 318 panic("ndopts == NULL in nd6_options"); 319 if (ndopts->nd_opts_last == NULL) 320 panic("uninitialized ndopts in nd6_options"); 321 if (ndopts->nd_opts_search == NULL) 322 return 0; 323 324 while (1) { 325 nd_opt = nd6_option(ndopts); 326 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { 327 /* 328 * Message validation requires that all included 329 * options have a length that is greater than zero. 330 */ 331 ICMP6_STATINC(ICMP6_STAT_ND_BADOPT); 332 memset(ndopts, 0, sizeof(*ndopts)); 333 return -1; 334 } 335 336 if (nd_opt == NULL) 337 goto skip1; 338 339 switch (nd_opt->nd_opt_type) { 340 case ND_OPT_SOURCE_LINKADDR: 341 case ND_OPT_TARGET_LINKADDR: 342 case ND_OPT_MTU: 343 case ND_OPT_REDIRECTED_HEADER: 344 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { 345 nd6log((LOG_INFO, 346 "duplicated ND6 option found (type=%d)\n", 347 nd_opt->nd_opt_type)); 348 /* XXX bark? */ 349 } else { 350 ndopts->nd_opt_array[nd_opt->nd_opt_type] 351 = nd_opt; 352 } 353 break; 354 case ND_OPT_PREFIX_INFORMATION: 355 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { 356 ndopts->nd_opt_array[nd_opt->nd_opt_type] 357 = nd_opt; 358 } 359 ndopts->nd_opts_pi_end = 360 (struct nd_opt_prefix_info *)nd_opt; 361 break; 362 default: 363 /* 364 * Unknown options must be silently ignored, 365 * to accommodate future extension to the protocol. 366 */ 367 nd6log((LOG_DEBUG, 368 "nd6_options: unsupported option %d - " 369 "option ignored\n", nd_opt->nd_opt_type)); 370 } 371 372 skip1: 373 i++; 374 if (i > nd6_maxndopt) { 375 ICMP6_STATINC(ICMP6_STAT_ND_TOOMANYOPT); 376 nd6log((LOG_INFO, "too many loop in nd opt\n")); 377 break; 378 } 379 380 if (ndopts->nd_opts_done) 381 break; 382 } 383 384 return 0; 385 } 386 387 /* 388 * ND6 timer routine to handle ND6 entries 389 */ 390 void 391 nd6_llinfo_settimer(struct llinfo_nd6 *ln, long xtick) 392 { 393 int s; 394 395 s = splsoftnet(); 396 397 if (xtick < 0) { 398 ln->ln_expire = 0; 399 ln->ln_ntick = 0; 400 callout_stop(&ln->ln_timer_ch); 401 } else { 402 ln->ln_expire = time_second + xtick / hz; 403 if (xtick > INT_MAX) { 404 ln->ln_ntick = xtick - INT_MAX; 405 callout_reset(&ln->ln_timer_ch, INT_MAX, 406 nd6_llinfo_timer, ln); 407 } else { 408 ln->ln_ntick = 0; 409 callout_reset(&ln->ln_timer_ch, xtick, 410 nd6_llinfo_timer, ln); 411 } 412 } 413 414 splx(s); 415 } 416 417 static void 418 nd6_llinfo_timer(void *arg) 419 { 420 struct llinfo_nd6 *ln; 421 struct rtentry *rt; 422 const struct sockaddr_in6 *dst; 423 struct ifnet *ifp; 424 struct nd_ifinfo *ndi = NULL; 425 426 mutex_enter(softnet_lock); 427 KERNEL_LOCK(1, NULL); 428 429 ln = (struct llinfo_nd6 *)arg; 430 431 if (ln->ln_ntick > 0) { 432 nd6_llinfo_settimer(ln, ln->ln_ntick); 433 KERNEL_UNLOCK_ONE(NULL); 434 mutex_exit(softnet_lock); 435 return; 436 } 437 438 if ((rt = ln->ln_rt) == NULL) 439 panic("ln->ln_rt == NULL"); 440 if ((ifp = rt->rt_ifp) == NULL) 441 panic("ln->ln_rt->rt_ifp == NULL"); 442 ndi = ND_IFINFO(ifp); 443 dst = satocsin6(rt_getkey(rt)); 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 (!dst) 450 panic("dst=0 in nd6_timer(ln=%p)", ln); 451 452 switch (ln->ln_state) { 453 case ND6_LLINFO_INCOMPLETE: 454 if (ln->ln_asked < nd6_mmaxtries) { 455 ln->ln_asked++; 456 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000); 457 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); 458 } else { 459 struct mbuf *m = ln->ln_hold; 460 if (m) { 461 struct mbuf *m0; 462 463 /* 464 * assuming every packet in ln_hold has 465 * the same IP header 466 */ 467 m0 = m->m_nextpkt; 468 m->m_nextpkt = NULL; 469 icmp6_error2(m, ICMP6_DST_UNREACH, 470 ICMP6_DST_UNREACH_ADDR, 0, rt->rt_ifp); 471 472 ln->ln_hold = m0; 473 clear_llinfo_pqueue(ln); 474 } 475 (void)nd6_free(rt, 0); 476 ln = NULL; 477 } 478 break; 479 case ND6_LLINFO_REACHABLE: 480 if (!ND6_LLINFO_PERMANENT(ln)) { 481 ln->ln_state = ND6_LLINFO_STALE; 482 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 483 } 484 break; 485 486 case ND6_LLINFO_PURGE: 487 case ND6_LLINFO_STALE: 488 /* Garbage Collection(RFC 2461 5.3) */ 489 if (!ND6_LLINFO_PERMANENT(ln)) { 490 (void)nd6_free(rt, 1); 491 ln = NULL; 492 } 493 break; 494 495 case ND6_LLINFO_DELAY: 496 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 497 /* We need NUD */ 498 ln->ln_asked = 1; 499 ln->ln_state = ND6_LLINFO_PROBE; 500 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000); 501 nd6_ns_output(ifp, &dst->sin6_addr, 502 &dst->sin6_addr, ln, 0); 503 } else { 504 ln->ln_state = ND6_LLINFO_STALE; /* XXX */ 505 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 506 } 507 break; 508 case ND6_LLINFO_PROBE: 509 if (ln->ln_asked < nd6_umaxtries) { 510 ln->ln_asked++; 511 nd6_llinfo_settimer(ln, (long)ndi->retrans * hz / 1000); 512 nd6_ns_output(ifp, &dst->sin6_addr, 513 &dst->sin6_addr, ln, 0); 514 } else { 515 (void)nd6_free(rt, 0); 516 ln = NULL; 517 } 518 break; 519 } 520 521 KERNEL_UNLOCK_ONE(NULL); 522 mutex_exit(softnet_lock); 523 } 524 525 /* 526 * ND6 timer routine to expire default route list and prefix list 527 */ 528 void 529 nd6_timer(void *ignored_arg) 530 { 531 struct nd_defrouter *next_dr, *dr; 532 struct nd_prefix *next_pr, *pr; 533 struct in6_ifaddr *ia6, *nia6; 534 535 callout_reset(&nd6_timer_ch, nd6_prune * hz, 536 nd6_timer, NULL); 537 538 mutex_enter(softnet_lock); 539 KERNEL_LOCK(1, NULL); 540 541 /* expire default router list */ 542 543 TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, next_dr) { 544 if (dr->expire && dr->expire < time_second) { 545 defrtrlist_del(dr); 546 } 547 } 548 549 /* 550 * expire interface addresses. 551 * in the past the loop was inside prefix expiry processing. 552 * However, from a stricter speci-confrmance standpoint, we should 553 * rather separate address lifetimes and prefix lifetimes. 554 */ 555 addrloop: 556 for (ia6 = in6_ifaddr; ia6; ia6 = nia6) { 557 nia6 = ia6->ia_next; 558 /* check address lifetime */ 559 if (IFA6_IS_INVALID(ia6)) { 560 int regen = 0; 561 562 /* 563 * If the expiring address is temporary, try 564 * regenerating a new one. This would be useful when 565 * we suspended a laptop PC, then turned it on after a 566 * period that could invalidate all temporary 567 * addresses. Although we may have to restart the 568 * loop (see below), it must be after purging the 569 * address. Otherwise, we'd see an infinite loop of 570 * regeneration. 571 */ 572 if (ip6_use_tempaddr && 573 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 574 if (regen_tmpaddr(ia6) == 0) 575 regen = 1; 576 } 577 578 in6_purgeaddr(&ia6->ia_ifa); 579 580 if (regen) 581 goto addrloop; /* XXX: see below */ 582 } else if (IFA6_IS_DEPRECATED(ia6)) { 583 int oldflags = ia6->ia6_flags; 584 585 if ((oldflags & IN6_IFF_DEPRECATED) == 0) { 586 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 587 nd6_newaddrmsg((struct ifaddr *)ia6); 588 } 589 590 /* 591 * If a temporary address has just become deprecated, 592 * regenerate a new one if possible. 593 */ 594 if (ip6_use_tempaddr && 595 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 596 (oldflags & IN6_IFF_DEPRECATED) == 0) { 597 598 if (regen_tmpaddr(ia6) == 0) { 599 /* 600 * A new temporary address is 601 * generated. 602 * XXX: this means the address chain 603 * has changed while we are still in 604 * the loop. Although the change 605 * would not cause disaster (because 606 * it's not a deletion, but an 607 * addition,) we'd rather restart the 608 * loop just for safety. Or does this 609 * significantly reduce performance?? 610 */ 611 goto addrloop; 612 } 613 } 614 } else { 615 /* 616 * A new RA might have made a deprecated address 617 * preferred. 618 */ 619 if (ia6->ia6_flags & IN6_IFF_DEPRECATED) { 620 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 621 nd6_newaddrmsg((struct ifaddr *)ia6); 622 } 623 } 624 } 625 626 /* expire prefix list */ 627 LIST_FOREACH_SAFE(pr, &nd_prefix, ndpr_entry, next_pr) { 628 /* 629 * check prefix lifetime. 630 * since pltime is just for autoconf, pltime processing for 631 * prefix is not necessary. 632 */ 633 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 634 time_second - pr->ndpr_lastupdate > pr->ndpr_vltime) { 635 636 /* 637 * address expiration and prefix expiration are 638 * separate. NEVER perform in6_purgeaddr here. 639 */ 640 641 prelist_remove(pr); 642 } 643 } 644 645 KERNEL_UNLOCK_ONE(NULL); 646 mutex_exit(softnet_lock); 647 } 648 649 /* ia6: deprecated/invalidated temporary address */ 650 static int 651 regen_tmpaddr(struct in6_ifaddr *ia6) 652 { 653 struct ifaddr *ifa; 654 struct ifnet *ifp; 655 struct in6_ifaddr *public_ifa6 = NULL; 656 657 ifp = ia6->ia_ifa.ifa_ifp; 658 IFADDR_FOREACH(ifa, ifp) { 659 struct in6_ifaddr *it6; 660 661 if (ifa->ifa_addr->sa_family != AF_INET6) 662 continue; 663 664 it6 = (struct in6_ifaddr *)ifa; 665 666 /* ignore no autoconf addresses. */ 667 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 668 continue; 669 670 /* ignore autoconf addresses with different prefixes. */ 671 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 672 continue; 673 674 /* 675 * Now we are looking at an autoconf address with the same 676 * prefix as ours. If the address is temporary and is still 677 * preferred, do not create another one. It would be rare, but 678 * could happen, for example, when we resume a laptop PC after 679 * a long period. 680 */ 681 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 682 !IFA6_IS_DEPRECATED(it6)) { 683 public_ifa6 = NULL; 684 break; 685 } 686 687 /* 688 * This is a public autoconf address that has the same prefix 689 * as ours. If it is preferred, keep it. We can't break the 690 * loop here, because there may be a still-preferred temporary 691 * address with the prefix. 692 */ 693 if (!IFA6_IS_DEPRECATED(it6)) 694 public_ifa6 = it6; 695 } 696 697 if (public_ifa6 != NULL) { 698 int e; 699 700 /* 701 * Random factor is introduced in the preferred lifetime, so 702 * we do not need additional delay (3rd arg to in6_tmpifadd). 703 */ 704 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 705 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 706 " tmp addr, errno=%d\n", e); 707 return -1; 708 } 709 return 0; 710 } 711 712 return -1; 713 } 714 715 bool 716 nd6_accepts_rtadv(const struct nd_ifinfo *ndi) 717 { 718 switch (ndi->flags & (ND6_IFF_ACCEPT_RTADV|ND6_IFF_OVERRIDE_RTADV)) { 719 case ND6_IFF_OVERRIDE_RTADV|ND6_IFF_ACCEPT_RTADV: 720 return true; 721 case ND6_IFF_ACCEPT_RTADV: 722 return ip6_accept_rtadv != 0; 723 case ND6_IFF_OVERRIDE_RTADV: 724 case 0: 725 default: 726 return false; 727 } 728 } 729 730 /* 731 * Nuke neighbor cache/prefix/default router management table, right before 732 * ifp goes away. 733 */ 734 void 735 nd6_purge(struct ifnet *ifp) 736 { 737 struct llinfo_nd6 *ln, *nln; 738 struct nd_defrouter *dr, *ndr; 739 struct nd_prefix *pr, *npr; 740 741 /* 742 * Nuke default router list entries toward ifp. 743 * We defer removal of default router list entries that is installed 744 * in the routing table, in order to keep additional side effects as 745 * small as possible. 746 */ 747 TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, ndr) { 748 if (dr->installed) 749 continue; 750 751 if (dr->ifp == ifp) 752 defrtrlist_del(dr); 753 } 754 755 TAILQ_FOREACH_SAFE(dr, &nd_defrouter, dr_entry, ndr) { 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 LIST_FOREACH_SAFE(pr, &nd_prefix, ndpr_entry, npr) { 765 if (pr->ndpr_ifp == ifp) { 766 /* 767 * Because if_detach() does *not* release prefixes 768 * while purging addresses the reference count will 769 * still be above zero. We therefore reset it to 770 * make sure that the prefix really gets purged. 771 */ 772 pr->ndpr_refcnt = 0; 773 /* 774 * Previously, pr->ndpr_addr is removed as well, 775 * but I strongly believe we don't have to do it. 776 * nd6_purge() is only called from in6_ifdetach(), 777 * which removes all the associated interface addresses 778 * by itself. 779 * (jinmei@kame.net 20010129) 780 */ 781 prelist_remove(pr); 782 } 783 } 784 785 /* cancel default outgoing interface setting */ 786 if (nd6_defifindex == ifp->if_index) 787 nd6_setdefaultiface(0); 788 789 /* XXX: too restrictive? */ 790 if (!ip6_forwarding && ifp->if_afdata[AF_INET6]) { 791 struct nd_ifinfo *ndi = ND_IFINFO(ifp); 792 if (ndi && nd6_accepts_rtadv(ndi)) { 793 /* refresh default router list */ 794 defrouter_select(); 795 } 796 } 797 798 /* 799 * Nuke neighbor cache entries for the ifp. 800 * Note that rt->rt_ifp may not be the same as ifp, 801 * due to KAME goto ours hack. See RTM_RESOLVE case in 802 * nd6_rtrequest(), and ip6_input(). 803 */ 804 ln = llinfo_nd6.ln_next; 805 while (ln != NULL && ln != &llinfo_nd6) { 806 struct rtentry *rt; 807 const struct sockaddr_dl *sdl; 808 809 nln = ln->ln_next; 810 rt = ln->ln_rt; 811 if (rt && rt->rt_gateway && 812 rt->rt_gateway->sa_family == AF_LINK) { 813 sdl = satocsdl(rt->rt_gateway); 814 if (sdl->sdl_index == ifp->if_index) 815 nln = nd6_free(rt, 0); 816 } 817 ln = nln; 818 } 819 } 820 821 static struct rtentry * 822 nd6_lookup1(const struct in6_addr *addr6, int create, struct ifnet *ifp, 823 int cloning) 824 { 825 struct rtentry *rt; 826 struct sockaddr_in6 sin6; 827 828 sockaddr_in6_init(&sin6, addr6, 0, 0, 0); 829 rt = rtalloc1((struct sockaddr *)&sin6, create); 830 if (rt != NULL && (rt->rt_flags & RTF_LLINFO) == 0) { 831 /* 832 * This is the case for the default route. 833 * If we want to create a neighbor cache for the address, we 834 * should free the route for the destination and allocate an 835 * interface route. 836 */ 837 if (create) { 838 RTFREE(rt); 839 rt = NULL; 840 } 841 } 842 if (rt != NULL) 843 ; 844 else if (create && ifp) { 845 int e; 846 847 /* 848 * If no route is available and create is set, 849 * we allocate a host route for the destination 850 * and treat it like an interface route. 851 * This hack is necessary for a neighbor which can't 852 * be covered by our own prefix. 853 */ 854 struct ifaddr *ifa = 855 ifaof_ifpforaddr((struct sockaddr *)&sin6, ifp); 856 if (ifa == NULL) 857 return NULL; 858 859 /* 860 * Create a new route. RTF_LLINFO is necessary 861 * to create a Neighbor Cache entry for the 862 * destination in nd6_rtrequest which will be 863 * called in rtrequest via ifa->ifa_rtrequest. 864 */ 865 if ((e = rtrequest(RTM_ADD, (const struct sockaddr *)&sin6, 866 ifa->ifa_addr, (const struct sockaddr *)&all1_sa, 867 (ifa->ifa_flags | RTF_HOST | RTF_LLINFO) & 868 ~RTF_CLONING, &rt)) != 0) { 869 #if 0 870 log(LOG_ERR, 871 "nd6_lookup: failed to add route for a " 872 "neighbor(%s), errno=%d\n", 873 ip6_sprintf(addr6), e); 874 #endif 875 return NULL; 876 } 877 if (rt == NULL) 878 return NULL; 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 rt->rt_refcnt--; 887 888 /* 889 * Check for a cloning route to match the address. 890 * This should only be set from in6_is_addr_neighbor so we avoid 891 * a potentially expensive second call to rtalloc1. 892 */ 893 if (cloning && 894 rt->rt_flags & (RTF_CLONING | RTF_CLONED) && 895 (rt->rt_ifp == ifp 896 #if NBRIDGE > 0 897 || SAME_BRIDGE(rt->rt_ifp->if_bridgeport, ifp->if_bridgeport) 898 #endif 899 #if NCARP > 0 900 || (ifp->if_type == IFT_CARP && rt->rt_ifp == ifp->if_carpdev) || 901 (rt->rt_ifp->if_type == IFT_CARP && rt->rt_ifp->if_carpdev == ifp)|| 902 (ifp->if_type == IFT_CARP && rt->rt_ifp->if_type == IFT_CARP && 903 rt->rt_ifp->if_carpdev == ifp->if_carpdev) 904 #endif 905 )) 906 return rt; 907 908 /* 909 * Validation for the entry. 910 * Note that the check for rt_llinfo is necessary because a cloned 911 * route from a parent route that has the L flag (e.g. the default 912 * route to a p2p interface) may have the flag, too, while the 913 * destination is not actually a neighbor. 914 * XXX: we can't use rt->rt_ifp to check for the interface, since 915 * it might be the loopback interface if the entry is for our 916 * own address on a non-loopback interface. Instead, we should 917 * use rt->rt_ifa->ifa_ifp, which would specify the REAL 918 * interface. 919 * Note also that ifa_ifp and ifp may differ when we connect two 920 * interfaces to a same link, install a link prefix to an interface, 921 * and try to install a neighbor cache on an interface that does not 922 * have a route to the prefix. 923 */ 924 if ((rt->rt_flags & RTF_GATEWAY) || (rt->rt_flags & RTF_LLINFO) == 0 || 925 rt->rt_gateway->sa_family != AF_LINK || rt->rt_llinfo == NULL || 926 (ifp && rt->rt_ifa->ifa_ifp != ifp)) { 927 if (create) { 928 nd6log((LOG_DEBUG, 929 "nd6_lookup: failed to lookup %s (if = %s)\n", 930 ip6_sprintf(addr6), 931 ifp ? if_name(ifp) : "unspec")); 932 } 933 return NULL; 934 } 935 return rt; 936 } 937 938 struct rtentry * 939 nd6_lookup(const struct in6_addr *addr6, int create, struct ifnet *ifp) 940 { 941 942 return nd6_lookup1(addr6, create, ifp, 0); 943 } 944 945 /* 946 * Detect if a given IPv6 address identifies a neighbor on a given link. 947 * XXX: should take care of the destination of a p2p link? 948 */ 949 int 950 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp) 951 { 952 struct nd_prefix *pr; 953 954 /* 955 * A link-local address is always a neighbor. 956 * XXX: a link does not necessarily specify a single interface. 957 */ 958 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 959 struct sockaddr_in6 sin6_copy; 960 u_int32_t zone; 961 962 /* 963 * We need sin6_copy since sa6_recoverscope() may modify the 964 * content (XXX). 965 */ 966 sin6_copy = *addr; 967 if (sa6_recoverscope(&sin6_copy)) 968 return 0; /* XXX: should be impossible */ 969 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 970 return 0; 971 if (sin6_copy.sin6_scope_id == zone) 972 return 1; 973 else 974 return 0; 975 } 976 977 /* 978 * If the address matches one of our on-link prefixes, it should be a 979 * neighbor. 980 */ 981 LIST_FOREACH(pr, &nd_prefix, ndpr_entry) { 982 if (pr->ndpr_ifp != ifp) 983 continue; 984 985 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) 986 continue; 987 988 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 989 &addr->sin6_addr, &pr->ndpr_mask)) 990 return 1; 991 } 992 993 /* 994 * If the default router list is empty, all addresses are regarded 995 * as on-link, and thus, as a neighbor. 996 * XXX: we restrict the condition to hosts, because routers usually do 997 * not have the "default router list". 998 */ 999 if (!ip6_forwarding && TAILQ_FIRST(&nd_defrouter) == NULL && 1000 nd6_defifindex == ifp->if_index) { 1001 return 1; 1002 } 1003 1004 /* 1005 * Even if the address matches none of our addresses, it might match 1006 * a cloning route or be in the neighbor cache. 1007 */ 1008 if (nd6_lookup1(&addr->sin6_addr, 0, ifp, 1) != NULL) 1009 return 1; 1010 1011 return 0; 1012 } 1013 1014 /* 1015 * Free an nd6 llinfo entry. 1016 * Since the function would cause significant changes in the kernel, DO NOT 1017 * make it global, unless you have a strong reason for the change, and are sure 1018 * that the change is safe. 1019 */ 1020 static struct llinfo_nd6 * 1021 nd6_free(struct rtentry *rt, int gc) 1022 { 1023 struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo, *next; 1024 struct in6_addr in6 = satocsin6(rt_getkey(rt))->sin6_addr; 1025 struct nd_defrouter *dr; 1026 1027 /* 1028 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1029 * even though it is not harmful, it was not really necessary. 1030 */ 1031 1032 /* cancel timer */ 1033 nd6_llinfo_settimer(ln, -1); 1034 1035 if (!ip6_forwarding) { 1036 int s; 1037 s = splsoftnet(); 1038 dr = defrouter_lookup(&satocsin6(rt_getkey(rt))->sin6_addr, 1039 rt->rt_ifp); 1040 1041 if (dr != NULL && dr->expire && 1042 ln->ln_state == ND6_LLINFO_STALE && gc) { 1043 /* 1044 * If the reason for the deletion is just garbage 1045 * collection, and the neighbor is an active default 1046 * router, do not delete it. Instead, reset the GC 1047 * timer using the router's lifetime. 1048 * Simply deleting the entry would affect default 1049 * router selection, which is not necessarily a good 1050 * thing, especially when we're using router preference 1051 * values. 1052 * XXX: the check for ln_state would be redundant, 1053 * but we intentionally keep it just in case. 1054 */ 1055 if (dr->expire > time_second) 1056 nd6_llinfo_settimer(ln, 1057 (dr->expire - time_second) * hz); 1058 else 1059 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 1060 splx(s); 1061 return ln->ln_next; 1062 } 1063 1064 if (ln->ln_router || dr) { 1065 /* 1066 * rt6_flush must be called whether or not the neighbor 1067 * is in the Default Router List. 1068 * See a corresponding comment in nd6_na_input(). 1069 */ 1070 rt6_flush(&in6, rt->rt_ifp); 1071 } 1072 1073 if (dr) { 1074 /* 1075 * Unreachablity of a router might affect the default 1076 * router selection and on-link detection of advertised 1077 * prefixes. 1078 */ 1079 1080 /* 1081 * Temporarily fake the state to choose a new default 1082 * router and to perform on-link determination of 1083 * prefixes correctly. 1084 * Below the state will be set correctly, 1085 * or the entry itself will be deleted. 1086 */ 1087 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1088 1089 /* 1090 * Since defrouter_select() does not affect the 1091 * on-link determination and MIP6 needs the check 1092 * before the default router selection, we perform 1093 * the check now. 1094 */ 1095 pfxlist_onlink_check(); 1096 1097 /* 1098 * refresh default router list 1099 */ 1100 defrouter_select(); 1101 } 1102 splx(s); 1103 } 1104 1105 /* 1106 * Before deleting the entry, remember the next entry as the 1107 * return value. We need this because pfxlist_onlink_check() above 1108 * might have freed other entries (particularly the old next entry) as 1109 * a side effect (XXX). 1110 */ 1111 next = ln->ln_next; 1112 1113 /* 1114 * Detach the route from the routing tree and the list of neighbor 1115 * caches, and disable the route entry not to be used in already 1116 * cached routes. 1117 */ 1118 rtrequest(RTM_DELETE, rt_getkey(rt), NULL, rt_mask(rt), 0, NULL); 1119 1120 return next; 1121 } 1122 1123 /* 1124 * Upper-layer reachability hint for Neighbor Unreachability Detection. 1125 * 1126 * XXX cost-effective methods? 1127 */ 1128 void 1129 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force) 1130 { 1131 struct llinfo_nd6 *ln; 1132 1133 /* 1134 * If the caller specified "rt", use that. Otherwise, resolve the 1135 * routing table by supplied "dst6". 1136 */ 1137 if (rt == NULL) { 1138 if (dst6 == NULL) 1139 return; 1140 if ((rt = nd6_lookup(dst6, 0, NULL)) == NULL) 1141 return; 1142 } 1143 1144 if ((rt->rt_flags & RTF_GATEWAY) != 0 || 1145 (rt->rt_flags & RTF_LLINFO) == 0 || 1146 !rt->rt_llinfo || !rt->rt_gateway || 1147 rt->rt_gateway->sa_family != AF_LINK) { 1148 /* This is not a host route. */ 1149 return; 1150 } 1151 1152 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1153 if (ln->ln_state < ND6_LLINFO_REACHABLE) 1154 return; 1155 1156 /* 1157 * if we get upper-layer reachability confirmation many times, 1158 * it is possible we have false information. 1159 */ 1160 if (!force) { 1161 ln->ln_byhint++; 1162 if (ln->ln_byhint > nd6_maxnudhint) 1163 return; 1164 } 1165 1166 ln->ln_state = ND6_LLINFO_REACHABLE; 1167 if (!ND6_LLINFO_PERMANENT(ln)) { 1168 nd6_llinfo_settimer(ln, 1169 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz); 1170 } 1171 } 1172 1173 void 1174 nd6_rtrequest(int req, struct rtentry *rt, const struct rt_addrinfo *info) 1175 { 1176 struct sockaddr *gate = rt->rt_gateway; 1177 struct llinfo_nd6 *ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1178 struct ifnet *ifp = rt->rt_ifp; 1179 uint8_t namelen = strlen(ifp->if_xname), addrlen = ifp->if_addrlen; 1180 struct ifaddr *ifa; 1181 1182 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1183 1184 if (req == RTM_LLINFO_UPD) { 1185 int rc; 1186 struct in6_addr *in6; 1187 struct in6_addr in6_all; 1188 int anycast; 1189 1190 if ((ifa = info->rti_ifa) == NULL) 1191 return; 1192 1193 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1194 anycast = ifatoia6(ifa)->ia6_flags & IN6_IFF_ANYCAST; 1195 1196 in6_all = in6addr_linklocal_allnodes; 1197 if ((rc = in6_setscope(&in6_all, ifa->ifa_ifp, NULL)) != 0) { 1198 log(LOG_ERR, "%s: failed to set scope %s " 1199 "(errno=%d)\n", __func__, if_name(ifp), rc); 1200 return; 1201 } 1202 1203 /* XXX don't set Override for proxy addresses */ 1204 nd6_na_output(ifa->ifa_ifp, &in6_all, in6, 1205 (anycast ? 0 : ND_NA_FLAG_OVERRIDE) 1206 #if 0 1207 | (ip6_forwarding ? ND_NA_FLAG_ROUTER : 0) 1208 #endif 1209 , 1, NULL); 1210 return; 1211 } 1212 1213 if ((rt->rt_flags & RTF_GATEWAY) != 0) 1214 return; 1215 1216 if (nd6_need_cache(ifp) == 0 && (rt->rt_flags & RTF_HOST) == 0) { 1217 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1218 /* 1219 * This is probably an interface direct route for a link 1220 * which does not need neighbor caches (e.g. fe80::%lo0/64). 1221 * We do not need special treatment below for such a route. 1222 * Moreover, the RTF_LLINFO flag which would be set below 1223 * would annoy the ndp(8) command. 1224 */ 1225 return; 1226 } 1227 1228 if (req == RTM_RESOLVE && 1229 (nd6_need_cache(ifp) == 0 || /* stf case */ 1230 !nd6_is_addr_neighbor(satocsin6(rt_getkey(rt)), ifp))) { 1231 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1232 /* 1233 * FreeBSD and BSD/OS often make a cloned host route based 1234 * on a less-specific route (e.g. the default route). 1235 * If the less specific route does not have a "gateway" 1236 * (this is the case when the route just goes to a p2p or an 1237 * stf interface), we'll mistakenly make a neighbor cache for 1238 * the host route, and will see strange neighbor solicitation 1239 * for the corresponding destination. In order to avoid the 1240 * confusion, we check if the destination of the route is 1241 * a neighbor in terms of neighbor discovery, and stop the 1242 * process if not. Additionally, we remove the LLINFO flag 1243 * so that ndp(8) will not try to get the neighbor information 1244 * of the destination. 1245 */ 1246 rt->rt_flags &= ~RTF_LLINFO; 1247 return; 1248 } 1249 1250 switch (req) { 1251 case RTM_ADD: 1252 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1253 /* 1254 * There is no backward compatibility :) 1255 * 1256 * if ((rt->rt_flags & RTF_HOST) == 0 && 1257 * SIN(rt_mask(rt))->sin_addr.s_addr != 0xffffffff) 1258 * rt->rt_flags |= RTF_CLONING; 1259 */ 1260 if ((rt->rt_flags & RTF_CLONING) || 1261 ((rt->rt_flags & RTF_LLINFO) && ln == NULL)) { 1262 union { 1263 struct sockaddr sa; 1264 struct sockaddr_dl sdl; 1265 struct sockaddr_storage ss; 1266 } u; 1267 /* 1268 * Case 1: This route should come from a route to 1269 * interface (RTF_CLONING case) or the route should be 1270 * treated as on-link but is currently not 1271 * (RTF_LLINFO && ln == NULL case). 1272 */ 1273 if (sockaddr_dl_init(&u.sdl, sizeof(u.ss), 1274 ifp->if_index, ifp->if_type, 1275 NULL, namelen, NULL, addrlen) == NULL) { 1276 printf("%s.%d: sockaddr_dl_init(, %zu, ) " 1277 "failed on %s\n", __func__, __LINE__, 1278 sizeof(u.ss), if_name(ifp)); 1279 } 1280 rt_setgate(rt, &u.sa); 1281 gate = rt->rt_gateway; 1282 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1283 if (ln != NULL) 1284 nd6_llinfo_settimer(ln, 0); 1285 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1286 if ((rt->rt_flags & RTF_CLONING) != 0) 1287 break; 1288 } 1289 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1290 /* 1291 * In IPv4 code, we try to annonuce new RTF_ANNOUNCE entry here. 1292 * We don't do that here since llinfo is not ready yet. 1293 * 1294 * There are also couple of other things to be discussed: 1295 * - unsolicited NA code needs improvement beforehand 1296 * - RFC2461 says we MAY send multicast unsolicited NA 1297 * (7.2.6 paragraph 4), however, it also says that we 1298 * SHOULD provide a mechanism to prevent multicast NA storm. 1299 * we don't have anything like it right now. 1300 * note that the mechanism needs a mutual agreement 1301 * between proxies, which means that we need to implement 1302 * a new protocol, or a new kludge. 1303 * - from RFC2461 6.2.4, host MUST NOT send an unsolicited NA. 1304 * we need to check ip6forwarding before sending it. 1305 * (or should we allow proxy ND configuration only for 1306 * routers? there's no mention about proxy ND from hosts) 1307 */ 1308 #if 0 1309 /* XXX it does not work */ 1310 if (rt->rt_flags & RTF_ANNOUNCE) 1311 nd6_na_output(ifp, 1312 &satocsin6(rt_getkey(rt))->sin6_addr, 1313 &satocsin6(rt_getkey(rt))->sin6_addr, 1314 ip6_forwarding ? ND_NA_FLAG_ROUTER : 0, 1315 1, NULL); 1316 #endif 1317 /* FALLTHROUGH */ 1318 case RTM_RESOLVE: 1319 if ((ifp->if_flags & (IFF_POINTOPOINT | IFF_LOOPBACK)) == 0) { 1320 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1321 /* 1322 * Address resolution isn't necessary for a point to 1323 * point link, so we can skip this test for a p2p link. 1324 */ 1325 if (gate->sa_family != AF_LINK || 1326 gate->sa_len < 1327 sockaddr_dl_measure(namelen, addrlen)) { 1328 log(LOG_DEBUG, 1329 "nd6_rtrequest: bad gateway value: %s\n", 1330 if_name(ifp)); 1331 break; 1332 } 1333 satosdl(gate)->sdl_type = ifp->if_type; 1334 satosdl(gate)->sdl_index = ifp->if_index; 1335 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1336 } 1337 if (ln != NULL) 1338 break; /* This happens on a route change */ 1339 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1340 /* 1341 * Case 2: This route may come from cloning, or a manual route 1342 * add with a LL address. 1343 */ 1344 R_Malloc(ln, struct llinfo_nd6 *, sizeof(*ln)); 1345 rt->rt_llinfo = ln; 1346 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1347 if (ln == NULL) { 1348 log(LOG_DEBUG, "nd6_rtrequest: malloc failed\n"); 1349 break; 1350 } 1351 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1352 nd6_inuse++; 1353 nd6_allocated++; 1354 memset(ln, 0, sizeof(*ln)); 1355 ln->ln_rt = rt; 1356 callout_init(&ln->ln_timer_ch, CALLOUT_MPSAFE); 1357 /* this is required for "ndp" command. - shin */ 1358 if (req == RTM_ADD) { 1359 /* 1360 * gate should have some valid AF_LINK entry, 1361 * and ln->ln_expire should have some lifetime 1362 * which is specified by ndp command. 1363 */ 1364 ln->ln_state = ND6_LLINFO_REACHABLE; 1365 ln->ln_byhint = 0; 1366 } else { 1367 /* 1368 * When req == RTM_RESOLVE, rt is created and 1369 * initialized in rtrequest(), so rt_expire is 0. 1370 */ 1371 ln->ln_state = ND6_LLINFO_NOSTATE; 1372 nd6_llinfo_settimer(ln, 0); 1373 } 1374 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1375 rt->rt_flags |= RTF_LLINFO; 1376 ln->ln_next = llinfo_nd6.ln_next; 1377 llinfo_nd6.ln_next = ln; 1378 ln->ln_prev = &llinfo_nd6; 1379 ln->ln_next->ln_prev = ln; 1380 1381 /* 1382 * If we have too many cache entries, initiate immediate 1383 * purging for some "less recently used" entries. Note that 1384 * we cannot directly call nd6_free() here because it would 1385 * cause re-entering rtable related routines triggering an LOR 1386 * problem for FreeBSD. 1387 */ 1388 if (ip6_neighborgcthresh >= 0 && 1389 nd6_inuse >= ip6_neighborgcthresh) { 1390 int i; 1391 1392 for (i = 0; i < 10 && llinfo_nd6.ln_prev != ln; i++) { 1393 struct llinfo_nd6 *ln_end = llinfo_nd6.ln_prev; 1394 1395 /* Move this entry to the head */ 1396 LN_DEQUEUE(ln_end); 1397 LN_INSERTHEAD(ln_end); 1398 1399 if (ND6_LLINFO_PERMANENT(ln_end)) 1400 continue; 1401 1402 if (ln_end->ln_state > ND6_LLINFO_INCOMPLETE) 1403 ln_end->ln_state = ND6_LLINFO_STALE; 1404 else 1405 ln_end->ln_state = ND6_LLINFO_PURGE; 1406 nd6_llinfo_settimer(ln_end, 0); 1407 } 1408 } 1409 1410 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1411 /* 1412 * check if rt_getkey(rt) is an address assigned 1413 * to the interface. 1414 */ 1415 ifa = (struct ifaddr *)in6ifa_ifpwithaddr(ifp, 1416 &satocsin6(rt_getkey(rt))->sin6_addr); 1417 RT_DPRINTF("rt_getkey(rt) = %p\n", rt_getkey(rt)); 1418 if (ifa != NULL) { 1419 const void *mac; 1420 nd6_llinfo_settimer(ln, -1); 1421 ln->ln_state = ND6_LLINFO_REACHABLE; 1422 ln->ln_byhint = 0; 1423 if ((mac = nd6_ifptomac(ifp)) != NULL) { 1424 /* XXX check for error */ 1425 if (sockaddr_dl_setaddr(satosdl(gate), 1426 gate->sa_len, mac, 1427 ifp->if_addrlen) == NULL) { 1428 printf("%s.%d: " 1429 "sockaddr_dl_setaddr(, %d, ) " 1430 "failed on %s\n", __func__, 1431 __LINE__, gate->sa_len, 1432 if_name(ifp)); 1433 } 1434 } 1435 if (nd6_useloopback) { 1436 ifp = rt->rt_ifp = lo0ifp; /* XXX */ 1437 /* 1438 * Make sure rt_ifa be equal to the ifaddr 1439 * corresponding to the address. 1440 * We need this because when we refer 1441 * rt_ifa->ia6_flags in ip6_input, we assume 1442 * that the rt_ifa points to the address instead 1443 * of the loopback address. 1444 */ 1445 if (ifa != rt->rt_ifa) 1446 rt_replace_ifa(rt, ifa); 1447 rt->rt_flags &= ~RTF_CLONED; 1448 } 1449 } else if (rt->rt_flags & RTF_ANNOUNCE) { 1450 nd6_llinfo_settimer(ln, -1); 1451 ln->ln_state = ND6_LLINFO_REACHABLE; 1452 ln->ln_byhint = 0; 1453 1454 /* join solicited node multicast for proxy ND */ 1455 if (ifp->if_flags & IFF_MULTICAST) { 1456 struct in6_addr llsol; 1457 int error; 1458 1459 llsol = satocsin6(rt_getkey(rt))->sin6_addr; 1460 llsol.s6_addr32[0] = htonl(0xff020000); 1461 llsol.s6_addr32[1] = 0; 1462 llsol.s6_addr32[2] = htonl(1); 1463 llsol.s6_addr8[12] = 0xff; 1464 if (in6_setscope(&llsol, ifp, NULL)) 1465 break; 1466 if (!in6_addmulti(&llsol, ifp, &error, 0)) { 1467 nd6log((LOG_ERR, "%s: failed to join " 1468 "%s (errno=%d)\n", if_name(ifp), 1469 ip6_sprintf(&llsol), error)); 1470 } 1471 } 1472 } 1473 break; 1474 1475 case RTM_DELETE: 1476 if (ln == NULL) 1477 break; 1478 /* leave from solicited node multicast for proxy ND */ 1479 if ((rt->rt_flags & RTF_ANNOUNCE) != 0 && 1480 (ifp->if_flags & IFF_MULTICAST) != 0) { 1481 struct in6_addr llsol; 1482 struct in6_multi *in6m; 1483 1484 llsol = satocsin6(rt_getkey(rt))->sin6_addr; 1485 llsol.s6_addr32[0] = htonl(0xff020000); 1486 llsol.s6_addr32[1] = 0; 1487 llsol.s6_addr32[2] = htonl(1); 1488 llsol.s6_addr8[12] = 0xff; 1489 if (in6_setscope(&llsol, ifp, NULL) == 0) { 1490 IN6_LOOKUP_MULTI(llsol, ifp, in6m); 1491 if (in6m) 1492 in6_delmulti(in6m); 1493 } 1494 } 1495 nd6_inuse--; 1496 ln->ln_next->ln_prev = ln->ln_prev; 1497 ln->ln_prev->ln_next = ln->ln_next; 1498 ln->ln_prev = NULL; 1499 nd6_llinfo_settimer(ln, -1); 1500 rt->rt_llinfo = 0; 1501 rt->rt_flags &= ~RTF_LLINFO; 1502 clear_llinfo_pqueue(ln); 1503 Free(ln); 1504 } 1505 } 1506 1507 int 1508 nd6_ioctl(u_long cmd, void *data, struct ifnet *ifp) 1509 { 1510 struct in6_drlist *drl = (struct in6_drlist *)data; 1511 struct in6_oprlist *oprl = (struct in6_oprlist *)data; 1512 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1513 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1514 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1515 struct nd_defrouter *dr; 1516 struct nd_prefix *pr; 1517 struct rtentry *rt; 1518 int i = 0, error = 0; 1519 int s; 1520 1521 switch (cmd) { 1522 case SIOCGDRLST_IN6: 1523 /* 1524 * obsolete API, use sysctl under net.inet6.icmp6 1525 */ 1526 memset(drl, 0, sizeof(*drl)); 1527 s = splsoftnet(); 1528 TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) { 1529 if (i >= DRLSTSIZ) 1530 break; 1531 drl->defrouter[i].rtaddr = dr->rtaddr; 1532 in6_clearscope(&drl->defrouter[i].rtaddr); 1533 1534 drl->defrouter[i].flags = dr->flags; 1535 drl->defrouter[i].rtlifetime = dr->rtlifetime; 1536 drl->defrouter[i].expire = dr->expire; 1537 drl->defrouter[i].if_index = dr->ifp->if_index; 1538 i++; 1539 } 1540 splx(s); 1541 break; 1542 case SIOCGPRLST_IN6: 1543 /* 1544 * obsolete API, use sysctl under net.inet6.icmp6 1545 * 1546 * XXX the structure in6_prlist was changed in backward- 1547 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6, 1548 * in6_prlist is used for nd6_sysctl() - fill_prlist(). 1549 */ 1550 /* 1551 * XXX meaning of fields, especialy "raflags", is very 1552 * differnet between RA prefix list and RR/static prefix list. 1553 * how about separating ioctls into two? 1554 */ 1555 memset(oprl, 0, sizeof(*oprl)); 1556 s = splsoftnet(); 1557 LIST_FOREACH(pr, &nd_prefix, ndpr_entry) { 1558 struct nd_pfxrouter *pfr; 1559 int j; 1560 1561 if (i >= PRLSTSIZ) 1562 break; 1563 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr; 1564 oprl->prefix[i].raflags = pr->ndpr_raf; 1565 oprl->prefix[i].prefixlen = pr->ndpr_plen; 1566 oprl->prefix[i].vltime = pr->ndpr_vltime; 1567 oprl->prefix[i].pltime = pr->ndpr_pltime; 1568 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index; 1569 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 1570 oprl->prefix[i].expire = 0; 1571 else { 1572 time_t maxexpire; 1573 1574 /* XXX: we assume time_t is signed. */ 1575 maxexpire = (-1) & 1576 ~((time_t)1 << 1577 ((sizeof(maxexpire) * 8) - 1)); 1578 if (pr->ndpr_vltime < 1579 maxexpire - pr->ndpr_lastupdate) { 1580 oprl->prefix[i].expire = 1581 pr->ndpr_lastupdate + 1582 pr->ndpr_vltime; 1583 } else 1584 oprl->prefix[i].expire = maxexpire; 1585 } 1586 1587 j = 0; 1588 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 1589 if (j < DRLSTSIZ) { 1590 #define RTRADDR oprl->prefix[i].advrtr[j] 1591 RTRADDR = pfr->router->rtaddr; 1592 in6_clearscope(&RTRADDR); 1593 #undef RTRADDR 1594 } 1595 j++; 1596 } 1597 oprl->prefix[i].advrtrs = j; 1598 oprl->prefix[i].origin = PR_ORIG_RA; 1599 1600 i++; 1601 } 1602 splx(s); 1603 1604 break; 1605 case OSIOCGIFINFO_IN6: 1606 #define ND ndi->ndi 1607 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1608 memset(&ND, 0, sizeof(ND)); 1609 ND.linkmtu = IN6_LINKMTU(ifp); 1610 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1611 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1612 ND.reachable = ND_IFINFO(ifp)->reachable; 1613 ND.retrans = ND_IFINFO(ifp)->retrans; 1614 ND.flags = ND_IFINFO(ifp)->flags; 1615 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1616 ND.chlim = ND_IFINFO(ifp)->chlim; 1617 break; 1618 case SIOCGIFINFO_IN6: 1619 ND = *ND_IFINFO(ifp); 1620 break; 1621 case SIOCSIFINFO_IN6: 1622 /* 1623 * used to change host variables from userland. 1624 * intented for a use on router to reflect RA configurations. 1625 */ 1626 /* 0 means 'unspecified' */ 1627 if (ND.linkmtu != 0) { 1628 if (ND.linkmtu < IPV6_MMTU || 1629 ND.linkmtu > IN6_LINKMTU(ifp)) { 1630 error = EINVAL; 1631 break; 1632 } 1633 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1634 } 1635 1636 if (ND.basereachable != 0) { 1637 int obasereachable = ND_IFINFO(ifp)->basereachable; 1638 1639 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1640 if (ND.basereachable != obasereachable) 1641 ND_IFINFO(ifp)->reachable = 1642 ND_COMPUTE_RTIME(ND.basereachable); 1643 } 1644 if (ND.retrans != 0) 1645 ND_IFINFO(ifp)->retrans = ND.retrans; 1646 if (ND.chlim != 0) 1647 ND_IFINFO(ifp)->chlim = ND.chlim; 1648 /* FALLTHROUGH */ 1649 case SIOCSIFINFO_FLAGS: 1650 ND_IFINFO(ifp)->flags = ND.flags; 1651 break; 1652 #undef ND 1653 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1654 /* sync kernel routing table with the default router list */ 1655 defrouter_reset(); 1656 defrouter_select(); 1657 break; 1658 case SIOCSPFXFLUSH_IN6: 1659 { 1660 /* flush all the prefix advertised by routers */ 1661 struct nd_prefix *pfx, *next; 1662 1663 s = splsoftnet(); 1664 LIST_FOREACH_SAFE(pfx, &nd_prefix, ndpr_entry, next) { 1665 struct in6_ifaddr *ia, *ia_next; 1666 1667 if (IN6_IS_ADDR_LINKLOCAL(&pfx->ndpr_prefix.sin6_addr)) 1668 continue; /* XXX */ 1669 1670 /* do we really have to remove addresses as well? */ 1671 for (ia = in6_ifaddr; ia; ia = ia_next) { 1672 /* ia might be removed. keep the next ptr. */ 1673 ia_next = ia->ia_next; 1674 1675 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1676 continue; 1677 1678 if (ia->ia6_ndpr == pfx) 1679 in6_purgeaddr(&ia->ia_ifa); 1680 } 1681 prelist_remove(pfx); 1682 } 1683 splx(s); 1684 break; 1685 } 1686 case SIOCSRTRFLUSH_IN6: 1687 { 1688 /* flush all the default routers */ 1689 struct nd_defrouter *drtr, *next; 1690 1691 s = splsoftnet(); 1692 defrouter_reset(); 1693 TAILQ_FOREACH_SAFE(drtr, &nd_defrouter, dr_entry, next) { 1694 defrtrlist_del(drtr); 1695 } 1696 defrouter_select(); 1697 splx(s); 1698 break; 1699 } 1700 case SIOCGNBRINFO_IN6: 1701 { 1702 struct llinfo_nd6 *ln; 1703 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1704 1705 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1706 return error; 1707 1708 s = splsoftnet(); 1709 if ((rt = nd6_lookup(&nb_addr, 0, ifp)) == NULL || 1710 (ln = (struct llinfo_nd6 *)rt->rt_llinfo) == NULL) { 1711 error = EINVAL; 1712 splx(s); 1713 break; 1714 } 1715 nbi->state = ln->ln_state; 1716 nbi->asked = ln->ln_asked; 1717 nbi->isrouter = ln->ln_router; 1718 nbi->expire = ln->ln_expire; 1719 splx(s); 1720 1721 break; 1722 } 1723 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1724 ndif->ifindex = nd6_defifindex; 1725 break; 1726 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1727 return nd6_setdefaultiface(ndif->ifindex); 1728 } 1729 return error; 1730 } 1731 1732 void 1733 nd6_llinfo_release_pkts(struct llinfo_nd6 *ln, struct ifnet *ifp, 1734 struct rtentry *rt) 1735 { 1736 struct mbuf *m_hold, *m_hold_next; 1737 1738 for (m_hold = ln->ln_hold, ln->ln_hold = NULL; 1739 m_hold != NULL; 1740 m_hold = m_hold_next) { 1741 m_hold_next = m_hold->m_nextpkt; 1742 m_hold->m_nextpkt = NULL; 1743 1744 /* 1745 * we assume ifp is not a p2p here, so 1746 * just set the 2nd argument as the 1747 * 1st one. 1748 */ 1749 nd6_output(ifp, ifp, m_hold, satocsin6(rt_getkey(rt)), rt); 1750 } 1751 } 1752 1753 /* 1754 * Create neighbor cache entry and cache link-layer address, 1755 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1756 */ 1757 struct rtentry * 1758 nd6_cache_lladdr( 1759 struct ifnet *ifp, 1760 struct in6_addr *from, 1761 char *lladdr, 1762 int lladdrlen, 1763 int type, /* ICMP6 type */ 1764 int code /* type dependent information */ 1765 ) 1766 { 1767 struct nd_ifinfo *ndi = ND_IFINFO(ifp); 1768 struct rtentry *rt = NULL; 1769 struct llinfo_nd6 *ln = NULL; 1770 int is_newentry; 1771 struct sockaddr_dl *sdl = NULL; 1772 int do_update; 1773 int olladdr; 1774 int llchange; 1775 int newstate = 0; 1776 1777 if (ifp == NULL) 1778 panic("ifp == NULL in nd6_cache_lladdr"); 1779 if (from == NULL) 1780 panic("from == NULL in nd6_cache_lladdr"); 1781 1782 /* nothing must be updated for unspecified address */ 1783 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1784 return NULL; 1785 1786 /* 1787 * Validation about ifp->if_addrlen and lladdrlen must be done in 1788 * the caller. 1789 * 1790 * XXX If the link does not have link-layer adderss, what should 1791 * we do? (ifp->if_addrlen == 0) 1792 * Spec says nothing in sections for RA, RS and NA. There's small 1793 * description on it in NS section (RFC 2461 7.2.3). 1794 */ 1795 1796 rt = nd6_lookup(from, 0, ifp); 1797 if (rt == NULL) { 1798 #if 0 1799 /* nothing must be done if there's no lladdr */ 1800 if (!lladdr || !lladdrlen) 1801 return NULL; 1802 #endif 1803 1804 rt = nd6_lookup(from, 1, ifp); 1805 is_newentry = 1; 1806 } else { 1807 /* do nothing if static ndp is set */ 1808 if (rt->rt_flags & RTF_STATIC) 1809 return NULL; 1810 is_newentry = 0; 1811 } 1812 1813 if (rt == NULL) 1814 return NULL; 1815 if ((rt->rt_flags & (RTF_GATEWAY | RTF_LLINFO)) != RTF_LLINFO) { 1816 fail: 1817 (void)nd6_free(rt, 0); 1818 return NULL; 1819 } 1820 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 1821 if (ln == NULL) 1822 goto fail; 1823 if (rt->rt_gateway == NULL) 1824 goto fail; 1825 if (rt->rt_gateway->sa_family != AF_LINK) 1826 goto fail; 1827 sdl = satosdl(rt->rt_gateway); 1828 1829 olladdr = (sdl->sdl_alen) ? 1 : 0; 1830 if (olladdr && lladdr) { 1831 if (memcmp(lladdr, CLLADDR(sdl), ifp->if_addrlen)) 1832 llchange = 1; 1833 else 1834 llchange = 0; 1835 } else 1836 llchange = 0; 1837 1838 /* 1839 * newentry olladdr lladdr llchange (*=record) 1840 * 0 n n -- (1) 1841 * 0 y n -- (2) 1842 * 0 n y -- (3) * STALE 1843 * 0 y y n (4) * 1844 * 0 y y y (5) * STALE 1845 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1846 * 1 -- y -- (7) * STALE 1847 */ 1848 1849 if (lladdr) { /* (3-5) and (7) */ 1850 /* 1851 * Record source link-layer address 1852 * XXX is it dependent to ifp->if_type? 1853 */ 1854 /* XXX check for error */ 1855 if (sockaddr_dl_setaddr(sdl, sdl->sdl_len, lladdr, 1856 ifp->if_addrlen) == NULL) { 1857 printf("%s.%d: sockaddr_dl_setaddr(, %d, ) " 1858 "failed on %s\n", __func__, __LINE__, 1859 sdl->sdl_len, if_name(ifp)); 1860 } 1861 } 1862 1863 if (!is_newentry) { 1864 if ((!olladdr && lladdr) || /* (3) */ 1865 (olladdr && lladdr && llchange)) { /* (5) */ 1866 do_update = 1; 1867 newstate = ND6_LLINFO_STALE; 1868 } else /* (1-2,4) */ 1869 do_update = 0; 1870 } else { 1871 do_update = 1; 1872 if (lladdr == NULL) /* (6) */ 1873 newstate = ND6_LLINFO_NOSTATE; 1874 else /* (7) */ 1875 newstate = ND6_LLINFO_STALE; 1876 } 1877 1878 if (do_update) { 1879 /* 1880 * Update the state of the neighbor cache. 1881 */ 1882 ln->ln_state = newstate; 1883 1884 if (ln->ln_state == ND6_LLINFO_STALE) { 1885 /* 1886 * XXX: since nd6_output() below will cause 1887 * state tansition to DELAY and reset the timer, 1888 * we must set the timer now, although it is actually 1889 * meaningless. 1890 */ 1891 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 1892 1893 nd6_llinfo_release_pkts(ln, ifp, rt); 1894 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { 1895 /* probe right away */ 1896 nd6_llinfo_settimer((void *)ln, 0); 1897 } 1898 } 1899 1900 /* 1901 * ICMP6 type dependent behavior. 1902 * 1903 * NS: clear IsRouter if new entry 1904 * RS: clear IsRouter 1905 * RA: set IsRouter if there's lladdr 1906 * redir: clear IsRouter if new entry 1907 * 1908 * RA case, (1): 1909 * The spec says that we must set IsRouter in the following cases: 1910 * - If lladdr exist, set IsRouter. This means (1-5). 1911 * - If it is old entry (!newentry), set IsRouter. This means (7). 1912 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1913 * A quetion arises for (1) case. (1) case has no lladdr in the 1914 * neighbor cache, this is similar to (6). 1915 * This case is rare but we figured that we MUST NOT set IsRouter. 1916 * 1917 * newentry olladdr lladdr llchange NS RS RA redir 1918 * D R 1919 * 0 n n -- (1) c ? s 1920 * 0 y n -- (2) c s s 1921 * 0 n y -- (3) c s s 1922 * 0 y y n (4) c s s 1923 * 0 y y y (5) c s s 1924 * 1 -- n -- (6) c c c s 1925 * 1 -- y -- (7) c c s c s 1926 * 1927 * (c=clear s=set) 1928 */ 1929 switch (type & 0xff) { 1930 case ND_NEIGHBOR_SOLICIT: 1931 /* 1932 * New entry must have is_router flag cleared. 1933 */ 1934 if (is_newentry) /* (6-7) */ 1935 ln->ln_router = 0; 1936 break; 1937 case ND_REDIRECT: 1938 /* 1939 * If the icmp is a redirect to a better router, always set the 1940 * is_router flag. Otherwise, if the entry is newly created, 1941 * clear the flag. [RFC 2461, sec 8.3] 1942 */ 1943 if (code == ND_REDIRECT_ROUTER) 1944 ln->ln_router = 1; 1945 else if (is_newentry) /* (6-7) */ 1946 ln->ln_router = 0; 1947 break; 1948 case ND_ROUTER_SOLICIT: 1949 /* 1950 * is_router flag must always be cleared. 1951 */ 1952 ln->ln_router = 0; 1953 break; 1954 case ND_ROUTER_ADVERT: 1955 /* 1956 * Mark an entry with lladdr as a router. 1957 */ 1958 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */ 1959 (is_newentry && lladdr)) { /* (7) */ 1960 ln->ln_router = 1; 1961 } 1962 break; 1963 } 1964 1965 /* 1966 * When the link-layer address of a router changes, select the 1967 * best router again. In particular, when the neighbor entry is newly 1968 * created, it might affect the selection policy. 1969 * Question: can we restrict the first condition to the "is_newentry" 1970 * case? 1971 * XXX: when we hear an RA from a new router with the link-layer 1972 * address option, defrouter_select() is called twice, since 1973 * defrtrlist_update called the function as well. However, I believe 1974 * we can compromise the overhead, since it only happens the first 1975 * time. 1976 * XXX: although defrouter_select() should not have a bad effect 1977 * for those are not autoconfigured hosts, we explicitly avoid such 1978 * cases for safety. 1979 */ 1980 if (do_update && ln->ln_router && !ip6_forwarding && 1981 nd6_accepts_rtadv(ndi)) 1982 defrouter_select(); 1983 1984 return rt; 1985 } 1986 1987 static void 1988 nd6_slowtimo(void *ignored_arg) 1989 { 1990 struct nd_ifinfo *nd6if; 1991 struct ifnet *ifp; 1992 1993 mutex_enter(softnet_lock); 1994 KERNEL_LOCK(1, NULL); 1995 callout_reset(&nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 1996 nd6_slowtimo, NULL); 1997 TAILQ_FOREACH(ifp, &ifnet, if_list) { 1998 nd6if = ND_IFINFO(ifp); 1999 if (nd6if->basereachable && /* already initialized */ 2000 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 2001 /* 2002 * Since reachable time rarely changes by router 2003 * advertisements, we SHOULD insure that a new random 2004 * value gets recomputed at least once every few hours. 2005 * (RFC 2461, 6.3.4) 2006 */ 2007 nd6if->recalctm = nd6_recalc_reachtm_interval; 2008 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 2009 } 2010 } 2011 KERNEL_UNLOCK_ONE(NULL); 2012 mutex_exit(softnet_lock); 2013 } 2014 2015 #define senderr(e) { error = (e); goto bad;} 2016 int 2017 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m0, 2018 const struct sockaddr_in6 *dst, struct rtentry *rt0) 2019 { 2020 struct mbuf *m = m0; 2021 struct rtentry *rt = rt0; 2022 struct sockaddr_in6 *gw6 = NULL; 2023 struct llinfo_nd6 *ln = NULL; 2024 int error = 0; 2025 2026 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) 2027 goto sendpkt; 2028 2029 if (nd6_need_cache(ifp) == 0) 2030 goto sendpkt; 2031 2032 /* 2033 * next hop determination. This routine is derived from ether_output. 2034 */ 2035 if (rt) { 2036 if ((rt->rt_flags & RTF_UP) == 0) { 2037 if ((rt0 = rt = rtalloc1(sin6tocsa(dst), 1)) != NULL) { 2038 rt->rt_refcnt--; 2039 if (rt->rt_ifp != ifp) 2040 senderr(EHOSTUNREACH); 2041 } else 2042 senderr(EHOSTUNREACH); 2043 } 2044 2045 if (rt->rt_flags & RTF_GATEWAY) { 2046 gw6 = (struct sockaddr_in6 *)rt->rt_gateway; 2047 2048 /* 2049 * We skip link-layer address resolution and NUD 2050 * if the gateway is not a neighbor from ND point 2051 * of view, regardless of the value of nd_ifinfo.flags. 2052 * The second condition is a bit tricky; we skip 2053 * if the gateway is our own address, which is 2054 * sometimes used to install a route to a p2p link. 2055 */ 2056 if (!nd6_is_addr_neighbor(gw6, ifp) || 2057 in6ifa_ifpwithaddr(ifp, &gw6->sin6_addr)) { 2058 /* 2059 * We allow this kind of tricky route only 2060 * when the outgoing interface is p2p. 2061 * XXX: we may need a more generic rule here. 2062 */ 2063 if ((ifp->if_flags & IFF_POINTOPOINT) == 0) 2064 senderr(EHOSTUNREACH); 2065 2066 goto sendpkt; 2067 } 2068 2069 if (rt->rt_gwroute == NULL) 2070 goto lookup; 2071 if (((rt = rt->rt_gwroute)->rt_flags & RTF_UP) == 0) { 2072 rtfree(rt); rt = rt0; 2073 lookup: 2074 rt->rt_gwroute = rtalloc1(rt->rt_gateway, 1); 2075 if ((rt = rt->rt_gwroute) == NULL) 2076 senderr(EHOSTUNREACH); 2077 /* the "G" test below also prevents rt == rt0 */ 2078 if ((rt->rt_flags & RTF_GATEWAY) || 2079 (rt->rt_ifp != ifp)) { 2080 rt->rt_refcnt--; 2081 rt0->rt_gwroute = NULL; 2082 senderr(EHOSTUNREACH); 2083 } 2084 } 2085 } 2086 } 2087 2088 /* 2089 * Address resolution or Neighbor Unreachability Detection 2090 * for the next hop. 2091 * At this point, the destination of the packet must be a unicast 2092 * or an anycast address(i.e. not a multicast). 2093 */ 2094 2095 /* Look up the neighbor cache for the nexthop */ 2096 if (rt != NULL && (rt->rt_flags & RTF_LLINFO) != 0) 2097 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 2098 else { 2099 /* 2100 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 2101 * the condition below is not very efficient. But we believe 2102 * it is tolerable, because this should be a rare case. 2103 */ 2104 if (nd6_is_addr_neighbor(dst, ifp) && 2105 (rt = nd6_lookup(&dst->sin6_addr, 1, ifp)) != NULL) 2106 ln = (struct llinfo_nd6 *)rt->rt_llinfo; 2107 } 2108 if (ln == NULL || rt == NULL) { 2109 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && 2110 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 2111 log(LOG_DEBUG, 2112 "nd6_output: can't allocate llinfo for %s " 2113 "(ln=%p, rt=%p)\n", 2114 ip6_sprintf(&dst->sin6_addr), ln, rt); 2115 senderr(EIO); /* XXX: good error? */ 2116 } 2117 2118 goto sendpkt; /* send anyway */ 2119 } 2120 2121 /* 2122 * Move this entry to the head of the queue so that it is less likely 2123 * for this entry to be a target of forced garbage collection (see 2124 * nd6_rtrequest()). 2125 */ 2126 LN_DEQUEUE(ln); 2127 LN_INSERTHEAD(ln); 2128 2129 /* We don't have to do link-layer address resolution on a p2p link. */ 2130 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && 2131 ln->ln_state < ND6_LLINFO_REACHABLE) { 2132 ln->ln_state = ND6_LLINFO_STALE; 2133 nd6_llinfo_settimer(ln, (long)nd6_gctimer * hz); 2134 } 2135 2136 /* 2137 * The first time we send a packet to a neighbor whose entry is 2138 * STALE, we have to change the state to DELAY and a sets a timer to 2139 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 2140 * neighbor unreachability detection on expiration. 2141 * (RFC 2461 7.3.3) 2142 */ 2143 if (ln->ln_state == ND6_LLINFO_STALE) { 2144 ln->ln_asked = 0; 2145 ln->ln_state = ND6_LLINFO_DELAY; 2146 nd6_llinfo_settimer(ln, (long)nd6_delay * hz); 2147 } 2148 2149 /* 2150 * If the neighbor cache entry has a state other than INCOMPLETE 2151 * (i.e. its link-layer address is already resolved), just 2152 * send the packet. 2153 */ 2154 if (ln->ln_state > ND6_LLINFO_INCOMPLETE) 2155 goto sendpkt; 2156 2157 /* 2158 * There is a neighbor cache entry, but no ethernet address 2159 * response yet. Append this latest packet to the end of the 2160 * packet queue in the mbuf, unless the number of the packet 2161 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 2162 * the oldest packet in the queue will be removed. 2163 */ 2164 if (ln->ln_state == ND6_LLINFO_NOSTATE) 2165 ln->ln_state = ND6_LLINFO_INCOMPLETE; 2166 if (ln->ln_hold) { 2167 struct mbuf *m_hold; 2168 int i; 2169 2170 i = 0; 2171 for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold->m_nextpkt) { 2172 i++; 2173 if (m_hold->m_nextpkt == NULL) { 2174 m_hold->m_nextpkt = m; 2175 break; 2176 } 2177 } 2178 while (i >= nd6_maxqueuelen) { 2179 m_hold = ln->ln_hold; 2180 ln->ln_hold = ln->ln_hold->m_nextpkt; 2181 m_freem(m_hold); 2182 i--; 2183 } 2184 } else { 2185 ln->ln_hold = m; 2186 } 2187 2188 /* 2189 * If there has been no NS for the neighbor after entering the 2190 * INCOMPLETE state, send the first solicitation. 2191 */ 2192 if (!ND6_LLINFO_PERMANENT(ln) && ln->ln_asked == 0) { 2193 ln->ln_asked++; 2194 nd6_llinfo_settimer(ln, 2195 (long)ND_IFINFO(ifp)->retrans * hz / 1000); 2196 nd6_ns_output(ifp, NULL, &dst->sin6_addr, ln, 0); 2197 } 2198 return 0; 2199 2200 sendpkt: 2201 /* discard the packet if IPv6 operation is disabled on the interface */ 2202 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 2203 error = ENETDOWN; /* better error? */ 2204 goto bad; 2205 } 2206 2207 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2208 return (*ifp->if_output)(origifp, m, sin6tocsa(dst), rt); 2209 return (*ifp->if_output)(ifp, m, sin6tocsa(dst), rt); 2210 2211 bad: 2212 if (m != NULL) 2213 m_freem(m); 2214 return error; 2215 } 2216 #undef senderr 2217 2218 int 2219 nd6_need_cache(struct ifnet *ifp) 2220 { 2221 /* 2222 * XXX: we currently do not make neighbor cache on any interface 2223 * other than ARCnet, Ethernet, FDDI and GIF. 2224 * 2225 * RFC2893 says: 2226 * - unidirectional tunnels needs no ND 2227 */ 2228 switch (ifp->if_type) { 2229 case IFT_ARCNET: 2230 case IFT_ETHER: 2231 case IFT_FDDI: 2232 case IFT_IEEE1394: 2233 case IFT_CARP: 2234 case IFT_GIF: /* XXX need more cases? */ 2235 case IFT_PPP: 2236 case IFT_TUNNEL: 2237 return 1; 2238 default: 2239 return 0; 2240 } 2241 } 2242 2243 int 2244 nd6_storelladdr(const struct ifnet *ifp, const struct rtentry *rt, 2245 struct mbuf *m, const struct sockaddr *dst, uint8_t *lldst, 2246 size_t dstsize) 2247 { 2248 const struct sockaddr_dl *sdl; 2249 2250 if (m->m_flags & M_MCAST) { 2251 switch (ifp->if_type) { 2252 case IFT_ETHER: 2253 case IFT_FDDI: 2254 ETHER_MAP_IPV6_MULTICAST(&satocsin6(dst)->sin6_addr, 2255 lldst); 2256 return 1; 2257 case IFT_IEEE1394: 2258 memcpy(lldst, ifp->if_broadcastaddr, 2259 MIN(dstsize, ifp->if_addrlen)); 2260 return 1; 2261 case IFT_ARCNET: 2262 *lldst = 0; 2263 return 1; 2264 default: 2265 m_freem(m); 2266 return 0; 2267 } 2268 } 2269 2270 if (rt == NULL) { 2271 /* this could happen, if we could not allocate memory */ 2272 m_freem(m); 2273 return 0; 2274 } 2275 if (rt->rt_gateway->sa_family != AF_LINK) { 2276 printf("%s: something odd happens\n", __func__); 2277 m_freem(m); 2278 return 0; 2279 } 2280 sdl = satocsdl(rt->rt_gateway); 2281 if (sdl->sdl_alen == 0 || sdl->sdl_alen > dstsize) { 2282 /* this should be impossible, but we bark here for debugging */ 2283 printf("%s: sdl_alen == %" PRIu8 ", dst=%s, if=%s\n", __func__, 2284 sdl->sdl_alen, ip6_sprintf(&satocsin6(dst)->sin6_addr), 2285 if_name(ifp)); 2286 m_freem(m); 2287 return 0; 2288 } 2289 2290 memcpy(lldst, CLLADDR(sdl), MIN(dstsize, sdl->sdl_alen)); 2291 return 1; 2292 } 2293 2294 static void 2295 clear_llinfo_pqueue(struct llinfo_nd6 *ln) 2296 { 2297 struct mbuf *m_hold, *m_hold_next; 2298 2299 for (m_hold = ln->ln_hold; m_hold; m_hold = m_hold_next) { 2300 m_hold_next = m_hold->m_nextpkt; 2301 m_hold->m_nextpkt = NULL; 2302 m_freem(m_hold); 2303 } 2304 2305 ln->ln_hold = NULL; 2306 return; 2307 } 2308 2309 int 2310 nd6_sysctl( 2311 int name, 2312 void *oldp, /* syscall arg, need copyout */ 2313 size_t *oldlenp, 2314 void *newp, /* syscall arg, need copyin */ 2315 size_t newlen 2316 ) 2317 { 2318 void *p; 2319 size_t ol; 2320 int error; 2321 2322 error = 0; 2323 2324 if (newp) 2325 return EPERM; 2326 if (oldp && !oldlenp) 2327 return EINVAL; 2328 ol = oldlenp ? *oldlenp : 0; 2329 2330 if (oldp) { 2331 p = malloc(*oldlenp, M_TEMP, M_WAITOK); 2332 if (p == NULL) 2333 return ENOMEM; 2334 } else 2335 p = NULL; 2336 switch (name) { 2337 case ICMPV6CTL_ND6_DRLIST: 2338 error = fill_drlist(p, oldlenp, ol); 2339 if (!error && p != NULL && oldp != NULL) 2340 error = copyout(p, oldp, *oldlenp); 2341 break; 2342 2343 case ICMPV6CTL_ND6_PRLIST: 2344 error = fill_prlist(p, oldlenp, ol); 2345 if (!error && p != NULL && oldp != NULL) 2346 error = copyout(p, oldp, *oldlenp); 2347 break; 2348 2349 case ICMPV6CTL_ND6_MAXQLEN: 2350 break; 2351 2352 default: 2353 error = ENOPROTOOPT; 2354 break; 2355 } 2356 if (p) 2357 free(p, M_TEMP); 2358 2359 return error; 2360 } 2361 2362 static int 2363 fill_drlist(void *oldp, size_t *oldlenp, size_t ol) 2364 { 2365 int error = 0, s; 2366 struct in6_defrouter *d = NULL, *de = NULL; 2367 struct nd_defrouter *dr; 2368 size_t l; 2369 2370 s = splsoftnet(); 2371 2372 if (oldp) { 2373 d = (struct in6_defrouter *)oldp; 2374 de = (struct in6_defrouter *)((char *)oldp + *oldlenp); 2375 } 2376 l = 0; 2377 2378 TAILQ_FOREACH(dr, &nd_defrouter, dr_entry) { 2379 2380 if (oldp && d + 1 <= de) { 2381 memset(d, 0, sizeof(*d)); 2382 sockaddr_in6_init(&d->rtaddr, &dr->rtaddr, 0, 0, 0); 2383 if (sa6_recoverscope(&d->rtaddr)) { 2384 log(LOG_ERR, 2385 "scope error in router list (%s)\n", 2386 ip6_sprintf(&d->rtaddr.sin6_addr)); 2387 /* XXX: press on... */ 2388 } 2389 d->flags = dr->flags; 2390 d->rtlifetime = dr->rtlifetime; 2391 d->expire = dr->expire; 2392 d->if_index = dr->ifp->if_index; 2393 } 2394 2395 l += sizeof(*d); 2396 if (d) 2397 d++; 2398 } 2399 2400 if (oldp) { 2401 if (l > ol) 2402 error = ENOMEM; 2403 } 2404 if (oldlenp) 2405 *oldlenp = l; /* (void *)d - (void *)oldp */ 2406 2407 splx(s); 2408 2409 return error; 2410 } 2411 2412 static int 2413 fill_prlist(void *oldp, size_t *oldlenp, size_t ol) 2414 { 2415 int error = 0, s; 2416 struct nd_prefix *pr; 2417 uint8_t *p = NULL, *ps = NULL; 2418 uint8_t *pe = NULL; 2419 size_t l; 2420 2421 s = splsoftnet(); 2422 2423 if (oldp) { 2424 ps = p = (uint8_t*)oldp; 2425 pe = (uint8_t*)oldp + *oldlenp; 2426 } 2427 l = 0; 2428 2429 LIST_FOREACH(pr, &nd_prefix, ndpr_entry) { 2430 u_short advrtrs; 2431 struct sockaddr_in6 sin6; 2432 struct nd_pfxrouter *pfr; 2433 struct in6_prefix pfx; 2434 2435 if (oldp && p + sizeof(struct in6_prefix) <= pe) 2436 { 2437 memset(&pfx, 0, sizeof(pfx)); 2438 ps = p; 2439 pfx.prefix = pr->ndpr_prefix; 2440 2441 if (sa6_recoverscope(&pfx.prefix)) { 2442 log(LOG_ERR, 2443 "scope error in prefix list (%s)\n", 2444 ip6_sprintf(&pfx.prefix.sin6_addr)); 2445 /* XXX: press on... */ 2446 } 2447 pfx.raflags = pr->ndpr_raf; 2448 pfx.prefixlen = pr->ndpr_plen; 2449 pfx.vltime = pr->ndpr_vltime; 2450 pfx.pltime = pr->ndpr_pltime; 2451 pfx.if_index = pr->ndpr_ifp->if_index; 2452 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2453 pfx.expire = 0; 2454 else { 2455 time_t maxexpire; 2456 2457 /* XXX: we assume time_t is signed. */ 2458 maxexpire = (-1) & 2459 ~((time_t)1 << 2460 ((sizeof(maxexpire) * 8) - 1)); 2461 if (pr->ndpr_vltime < 2462 maxexpire - pr->ndpr_lastupdate) { 2463 pfx.expire = pr->ndpr_lastupdate + 2464 pr->ndpr_vltime; 2465 } else 2466 pfx.expire = maxexpire; 2467 } 2468 pfx.refcnt = pr->ndpr_refcnt; 2469 pfx.flags = pr->ndpr_stateflags; 2470 pfx.origin = PR_ORIG_RA; 2471 2472 p += sizeof(pfx); l += sizeof(pfx); 2473 2474 advrtrs = 0; 2475 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 2476 if (p + sizeof(sin6) > pe) { 2477 advrtrs++; 2478 continue; 2479 } 2480 2481 sockaddr_in6_init(&sin6, &pfr->router->rtaddr, 2482 0, 0, 0); 2483 if (sa6_recoverscope(&sin6)) { 2484 log(LOG_ERR, 2485 "scope error in " 2486 "prefix list (%s)\n", 2487 ip6_sprintf(&pfr->router->rtaddr)); 2488 } 2489 advrtrs++; 2490 memcpy(p, &sin6, sizeof(sin6)); 2491 p += sizeof(sin6); 2492 l += sizeof(sin6); 2493 } 2494 pfx.advrtrs = advrtrs; 2495 memcpy(ps, &pfx, sizeof(pfx)); 2496 } 2497 else { 2498 l += sizeof(pfx); 2499 advrtrs = 0; 2500 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 2501 advrtrs++; 2502 l += sizeof(sin6); 2503 } 2504 } 2505 } 2506 2507 if (oldp) { 2508 *oldlenp = l; /* (void *)d - (void *)oldp */ 2509 if (l > ol) 2510 error = ENOMEM; 2511 } else 2512 *oldlenp = l; 2513 2514 splx(s); 2515 2516 return error; 2517 } 2518