1 /*- 2 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project. 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the project nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * $KAME: nd6.c,v 1.144 2001/05/24 07:44:00 itojun Exp $ 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_inet.h" 36 #include "opt_inet6.h" 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/callout.h> 41 #include <sys/malloc.h> 42 #include <sys/mbuf.h> 43 #include <sys/socket.h> 44 #include <sys/sockio.h> 45 #include <sys/time.h> 46 #include <sys/kernel.h> 47 #include <sys/protosw.h> 48 #include <sys/errno.h> 49 #include <sys/syslog.h> 50 #include <sys/lock.h> 51 #include <sys/rwlock.h> 52 #include <sys/queue.h> 53 #include <sys/sdt.h> 54 #include <sys/sysctl.h> 55 56 #include <net/if.h> 57 #include <net/if_var.h> 58 #include <net/if_arc.h> 59 #include <net/if_dl.h> 60 #include <net/if_types.h> 61 #include <net/iso88025.h> 62 #include <net/fddi.h> 63 #include <net/route.h> 64 #include <net/vnet.h> 65 66 #include <netinet/in.h> 67 #include <netinet/in_kdtrace.h> 68 #include <net/if_llatbl.h> 69 #define L3_ADDR_SIN6(le) ((struct sockaddr_in6 *) L3_ADDR(le)) 70 #include <netinet/if_ether.h> 71 #include <netinet6/in6_var.h> 72 #include <netinet/ip6.h> 73 #include <netinet6/ip6_var.h> 74 #include <netinet6/scope6_var.h> 75 #include <netinet6/nd6.h> 76 #include <netinet6/in6_ifattach.h> 77 #include <netinet/icmp6.h> 78 #include <netinet6/send.h> 79 80 #include <sys/limits.h> 81 82 #include <security/mac/mac_framework.h> 83 84 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ 85 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ 86 87 #define SIN6(s) ((const struct sockaddr_in6 *)(s)) 88 89 /* timer values */ 90 VNET_DEFINE(int, nd6_prune) = 1; /* walk list every 1 seconds */ 91 VNET_DEFINE(int, nd6_delay) = 5; /* delay first probe time 5 second */ 92 VNET_DEFINE(int, nd6_umaxtries) = 3; /* maximum unicast query */ 93 VNET_DEFINE(int, nd6_mmaxtries) = 3; /* maximum multicast query */ 94 VNET_DEFINE(int, nd6_useloopback) = 1; /* use loopback interface for 95 * local traffic */ 96 VNET_DEFINE(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage 97 * collection timer */ 98 99 /* preventing too many loops in ND option parsing */ 100 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */ 101 102 VNET_DEFINE(int, nd6_maxnudhint) = 0; /* max # of subsequent upper 103 * layer hints */ 104 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved 105 * ND entries */ 106 #define V_nd6_maxndopt VNET(nd6_maxndopt) 107 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen) 108 109 #ifdef ND6_DEBUG 110 VNET_DEFINE(int, nd6_debug) = 1; 111 #else 112 VNET_DEFINE(int, nd6_debug) = 0; 113 #endif 114 115 /* for debugging? */ 116 #if 0 117 static int nd6_inuse, nd6_allocated; 118 #endif 119 120 VNET_DEFINE(struct nd_drhead, nd_defrouter); 121 VNET_DEFINE(struct nd_prhead, nd_prefix); 122 123 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL; 124 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval) 125 126 int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int); 127 128 static int nd6_is_new_addr_neighbor(struct sockaddr_in6 *, 129 struct ifnet *); 130 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *); 131 static void nd6_slowtimo(void *); 132 static int regen_tmpaddr(struct in6_ifaddr *); 133 static struct llentry *nd6_free(struct llentry *, int); 134 static void nd6_llinfo_timer(void *); 135 static void clear_llinfo_pqueue(struct llentry *); 136 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *); 137 138 static VNET_DEFINE(struct callout, nd6_slowtimo_ch); 139 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch) 140 141 VNET_DEFINE(struct callout, nd6_timer_ch); 142 143 void 144 nd6_init(void) 145 { 146 147 LIST_INIT(&V_nd_prefix); 148 149 /* initialization of the default router list */ 150 TAILQ_INIT(&V_nd_defrouter); 151 152 /* start timer */ 153 callout_init(&V_nd6_slowtimo_ch, 0); 154 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 155 nd6_slowtimo, curvnet); 156 } 157 158 #ifdef VIMAGE 159 void 160 nd6_destroy() 161 { 162 163 callout_drain(&V_nd6_slowtimo_ch); 164 callout_drain(&V_nd6_timer_ch); 165 } 166 #endif 167 168 struct nd_ifinfo * 169 nd6_ifattach(struct ifnet *ifp) 170 { 171 struct nd_ifinfo *nd; 172 173 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO); 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 nd->flags = ND6_IFF_PERFORMNUD; 182 183 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. 184 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by 185 * default regardless of the V_ip6_auto_linklocal configuration to 186 * give a reasonable default behavior. 187 */ 188 if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) || 189 (ifp->if_flags & IFF_LOOPBACK)) 190 nd->flags |= ND6_IFF_AUTO_LINKLOCAL; 191 /* 192 * A loopback interface does not need to accept RTADV. 193 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by 194 * default regardless of the V_ip6_accept_rtadv configuration to 195 * prevent the interface from accepting RA messages arrived 196 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV. 197 */ 198 if (V_ip6_accept_rtadv && 199 !(ifp->if_flags & IFF_LOOPBACK) && 200 (ifp->if_type != IFT_BRIDGE)) 201 nd->flags |= ND6_IFF_ACCEPT_RTADV; 202 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK)) 203 nd->flags |= ND6_IFF_NO_RADR; 204 205 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */ 206 nd6_setmtu0(ifp, nd); 207 208 return nd; 209 } 210 211 void 212 nd6_ifdetach(struct nd_ifinfo *nd) 213 { 214 215 free(nd, M_IP6NDP); 216 } 217 218 /* 219 * Reset ND level link MTU. This function is called when the physical MTU 220 * changes, which means we might have to adjust the ND level MTU. 221 */ 222 void 223 nd6_setmtu(struct ifnet *ifp) 224 { 225 226 nd6_setmtu0(ifp, ND_IFINFO(ifp)); 227 } 228 229 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */ 230 void 231 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi) 232 { 233 u_int32_t omaxmtu; 234 235 omaxmtu = ndi->maxmtu; 236 237 switch (ifp->if_type) { 238 case IFT_ARCNET: 239 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */ 240 break; 241 case IFT_FDDI: 242 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */ 243 break; 244 case IFT_ISO88025: 245 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu); 246 break; 247 default: 248 ndi->maxmtu = ifp->if_mtu; 249 break; 250 } 251 252 /* 253 * Decreasing the interface MTU under IPV6 minimum MTU may cause 254 * undesirable situation. We thus notify the operator of the change 255 * explicitly. The check for omaxmtu is necessary to restrict the 256 * log to the case of changing the MTU, not initializing it. 257 */ 258 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { 259 log(LOG_NOTICE, "nd6_setmtu0: " 260 "new link MTU on %s (%lu) is too small for IPv6\n", 261 if_name(ifp), (unsigned long)ndi->maxmtu); 262 } 263 264 if (ndi->maxmtu > V_in6_maxmtu) 265 in6_setmaxmtu(); /* check all interfaces just in case */ 266 267 } 268 269 void 270 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts) 271 { 272 273 bzero(ndopts, sizeof(*ndopts)); 274 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; 275 ndopts->nd_opts_last 276 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); 277 278 if (icmp6len == 0) { 279 ndopts->nd_opts_done = 1; 280 ndopts->nd_opts_search = NULL; 281 } 282 } 283 284 /* 285 * Take one ND option. 286 */ 287 struct nd_opt_hdr * 288 nd6_option(union nd_opts *ndopts) 289 { 290 struct nd_opt_hdr *nd_opt; 291 int olen; 292 293 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 294 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 295 __func__)); 296 if (ndopts->nd_opts_search == NULL) 297 return NULL; 298 if (ndopts->nd_opts_done) 299 return NULL; 300 301 nd_opt = ndopts->nd_opts_search; 302 303 /* make sure nd_opt_len is inside the buffer */ 304 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) { 305 bzero(ndopts, sizeof(*ndopts)); 306 return NULL; 307 } 308 309 olen = nd_opt->nd_opt_len << 3; 310 if (olen == 0) { 311 /* 312 * Message validation requires that all included 313 * options have a length that is greater than zero. 314 */ 315 bzero(ndopts, sizeof(*ndopts)); 316 return NULL; 317 } 318 319 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen); 320 if (ndopts->nd_opts_search > ndopts->nd_opts_last) { 321 /* option overruns the end of buffer, invalid */ 322 bzero(ndopts, sizeof(*ndopts)); 323 return NULL; 324 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { 325 /* reached the end of options chain */ 326 ndopts->nd_opts_done = 1; 327 ndopts->nd_opts_search = NULL; 328 } 329 return nd_opt; 330 } 331 332 /* 333 * Parse multiple ND options. 334 * This function is much easier to use, for ND routines that do not need 335 * multiple options of the same type. 336 */ 337 int 338 nd6_options(union nd_opts *ndopts) 339 { 340 struct nd_opt_hdr *nd_opt; 341 int i = 0; 342 343 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 344 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 345 __func__)); 346 if (ndopts->nd_opts_search == NULL) 347 return 0; 348 349 while (1) { 350 nd_opt = nd6_option(ndopts); 351 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { 352 /* 353 * Message validation requires that all included 354 * options have a length that is greater than zero. 355 */ 356 ICMP6STAT_INC(icp6s_nd_badopt); 357 bzero(ndopts, sizeof(*ndopts)); 358 return -1; 359 } 360 361 if (nd_opt == NULL) 362 goto skip1; 363 364 switch (nd_opt->nd_opt_type) { 365 case ND_OPT_SOURCE_LINKADDR: 366 case ND_OPT_TARGET_LINKADDR: 367 case ND_OPT_MTU: 368 case ND_OPT_REDIRECTED_HEADER: 369 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { 370 nd6log((LOG_INFO, 371 "duplicated ND6 option found (type=%d)\n", 372 nd_opt->nd_opt_type)); 373 /* XXX bark? */ 374 } else { 375 ndopts->nd_opt_array[nd_opt->nd_opt_type] 376 = nd_opt; 377 } 378 break; 379 case ND_OPT_PREFIX_INFORMATION: 380 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { 381 ndopts->nd_opt_array[nd_opt->nd_opt_type] 382 = nd_opt; 383 } 384 ndopts->nd_opts_pi_end = 385 (struct nd_opt_prefix_info *)nd_opt; 386 break; 387 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */ 388 case ND_OPT_RDNSS: /* RFC 6106 */ 389 case ND_OPT_DNSSL: /* RFC 6106 */ 390 /* 391 * Silently ignore options we know and do not care about 392 * in the kernel. 393 */ 394 break; 395 default: 396 /* 397 * Unknown options must be silently ignored, 398 * to accomodate future extension to the protocol. 399 */ 400 nd6log((LOG_DEBUG, 401 "nd6_options: unsupported option %d - " 402 "option ignored\n", nd_opt->nd_opt_type)); 403 } 404 405 skip1: 406 i++; 407 if (i > V_nd6_maxndopt) { 408 ICMP6STAT_INC(icp6s_nd_toomanyopt); 409 nd6log((LOG_INFO, "too many loop in nd opt\n")); 410 break; 411 } 412 413 if (ndopts->nd_opts_done) 414 break; 415 } 416 417 return 0; 418 } 419 420 /* 421 * ND6 timer routine to handle ND6 entries 422 */ 423 void 424 nd6_llinfo_settimer_locked(struct llentry *ln, long tick) 425 { 426 int canceled; 427 428 LLE_WLOCK_ASSERT(ln); 429 430 if (tick < 0) { 431 ln->la_expire = 0; 432 ln->ln_ntick = 0; 433 canceled = callout_stop(&ln->ln_timer_ch); 434 } else { 435 ln->la_expire = time_uptime + tick / hz; 436 LLE_ADDREF(ln); 437 if (tick > INT_MAX) { 438 ln->ln_ntick = tick - INT_MAX; 439 canceled = callout_reset(&ln->ln_timer_ch, INT_MAX, 440 nd6_llinfo_timer, ln); 441 } else { 442 ln->ln_ntick = 0; 443 canceled = callout_reset(&ln->ln_timer_ch, tick, 444 nd6_llinfo_timer, ln); 445 } 446 } 447 if (canceled) 448 LLE_REMREF(ln); 449 } 450 451 void 452 nd6_llinfo_settimer(struct llentry *ln, long tick) 453 { 454 455 LLE_WLOCK(ln); 456 nd6_llinfo_settimer_locked(ln, tick); 457 LLE_WUNLOCK(ln); 458 } 459 460 static void 461 nd6_llinfo_timer(void *arg) 462 { 463 struct llentry *ln; 464 struct in6_addr *dst; 465 struct ifnet *ifp; 466 struct nd_ifinfo *ndi = NULL; 467 468 KASSERT(arg != NULL, ("%s: arg NULL", __func__)); 469 ln = (struct llentry *)arg; 470 LLE_WLOCK_ASSERT(ln); 471 ifp = ln->lle_tbl->llt_ifp; 472 473 CURVNET_SET(ifp->if_vnet); 474 475 if (ln->ln_ntick > 0) { 476 if (ln->ln_ntick > INT_MAX) { 477 ln->ln_ntick -= INT_MAX; 478 nd6_llinfo_settimer_locked(ln, INT_MAX); 479 } else { 480 ln->ln_ntick = 0; 481 nd6_llinfo_settimer_locked(ln, ln->ln_ntick); 482 } 483 goto done; 484 } 485 486 ndi = ND_IFINFO(ifp); 487 dst = &L3_ADDR_SIN6(ln)->sin6_addr; 488 if (ln->la_flags & LLE_STATIC) { 489 goto done; 490 } 491 492 if (ln->la_flags & LLE_DELETED) { 493 (void)nd6_free(ln, 0); 494 ln = NULL; 495 goto done; 496 } 497 498 switch (ln->ln_state) { 499 case ND6_LLINFO_INCOMPLETE: 500 if (ln->la_asked < V_nd6_mmaxtries) { 501 ln->la_asked++; 502 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 503 LLE_WUNLOCK(ln); 504 nd6_ns_output(ifp, NULL, dst, ln, 0); 505 LLE_WLOCK(ln); 506 } else { 507 struct mbuf *m = ln->la_hold; 508 if (m) { 509 struct mbuf *m0; 510 511 /* 512 * assuming every packet in la_hold has the 513 * same IP header. Send error after unlock. 514 */ 515 m0 = m->m_nextpkt; 516 m->m_nextpkt = NULL; 517 ln->la_hold = m0; 518 clear_llinfo_pqueue(ln); 519 } 520 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT); 521 (void)nd6_free(ln, 0); 522 ln = NULL; 523 if (m != NULL) 524 icmp6_error2(m, ICMP6_DST_UNREACH, 525 ICMP6_DST_UNREACH_ADDR, 0, ifp); 526 } 527 break; 528 case ND6_LLINFO_REACHABLE: 529 if (!ND6_LLINFO_PERMANENT(ln)) { 530 ln->ln_state = ND6_LLINFO_STALE; 531 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 532 } 533 break; 534 535 case ND6_LLINFO_STALE: 536 /* Garbage Collection(RFC 2461 5.3) */ 537 if (!ND6_LLINFO_PERMANENT(ln)) { 538 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 539 (void)nd6_free(ln, 1); 540 ln = NULL; 541 } 542 break; 543 544 case ND6_LLINFO_DELAY: 545 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 546 /* We need NUD */ 547 ln->la_asked = 1; 548 ln->ln_state = ND6_LLINFO_PROBE; 549 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 550 LLE_WUNLOCK(ln); 551 nd6_ns_output(ifp, dst, dst, ln, 0); 552 LLE_WLOCK(ln); 553 } else { 554 ln->ln_state = ND6_LLINFO_STALE; /* XXX */ 555 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 556 } 557 break; 558 case ND6_LLINFO_PROBE: 559 if (ln->la_asked < V_nd6_umaxtries) { 560 ln->la_asked++; 561 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 562 LLE_WUNLOCK(ln); 563 nd6_ns_output(ifp, dst, dst, ln, 0); 564 LLE_WLOCK(ln); 565 } else { 566 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 567 (void)nd6_free(ln, 0); 568 ln = NULL; 569 } 570 break; 571 default: 572 panic("%s: paths in a dark night can be confusing: %d", 573 __func__, ln->ln_state); 574 } 575 done: 576 if (ln != NULL) 577 LLE_FREE_LOCKED(ln); 578 CURVNET_RESTORE(); 579 } 580 581 582 /* 583 * ND6 timer routine to expire default route list and prefix list 584 */ 585 void 586 nd6_timer(void *arg) 587 { 588 CURVNET_SET((struct vnet *) arg); 589 struct nd_defrouter *dr, *ndr; 590 struct nd_prefix *pr, *npr; 591 struct in6_ifaddr *ia6, *nia6; 592 593 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz, 594 nd6_timer, curvnet); 595 596 /* expire default router list */ 597 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 598 if (dr->expire && dr->expire < time_uptime) 599 defrtrlist_del(dr); 600 } 601 602 /* 603 * expire interface addresses. 604 * in the past the loop was inside prefix expiry processing. 605 * However, from a stricter speci-confrmance standpoint, we should 606 * rather separate address lifetimes and prefix lifetimes. 607 * 608 * XXXRW: in6_ifaddrhead locking. 609 */ 610 addrloop: 611 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) { 612 /* check address lifetime */ 613 if (IFA6_IS_INVALID(ia6)) { 614 int regen = 0; 615 616 /* 617 * If the expiring address is temporary, try 618 * regenerating a new one. This would be useful when 619 * we suspended a laptop PC, then turned it on after a 620 * period that could invalidate all temporary 621 * addresses. Although we may have to restart the 622 * loop (see below), it must be after purging the 623 * address. Otherwise, we'd see an infinite loop of 624 * regeneration. 625 */ 626 if (V_ip6_use_tempaddr && 627 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 628 if (regen_tmpaddr(ia6) == 0) 629 regen = 1; 630 } 631 632 in6_purgeaddr(&ia6->ia_ifa); 633 634 if (regen) 635 goto addrloop; /* XXX: see below */ 636 } else if (IFA6_IS_DEPRECATED(ia6)) { 637 int oldflags = ia6->ia6_flags; 638 639 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 640 641 /* 642 * If a temporary address has just become deprecated, 643 * regenerate a new one if possible. 644 */ 645 if (V_ip6_use_tempaddr && 646 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 647 (oldflags & IN6_IFF_DEPRECATED) == 0) { 648 649 if (regen_tmpaddr(ia6) == 0) { 650 /* 651 * A new temporary address is 652 * generated. 653 * XXX: this means the address chain 654 * has changed while we are still in 655 * the loop. Although the change 656 * would not cause disaster (because 657 * it's not a deletion, but an 658 * addition,) we'd rather restart the 659 * loop just for safety. Or does this 660 * significantly reduce performance?? 661 */ 662 goto addrloop; 663 } 664 } 665 } else { 666 /* 667 * A new RA might have made a deprecated address 668 * preferred. 669 */ 670 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 671 } 672 } 673 674 /* expire prefix list */ 675 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 676 /* 677 * check prefix lifetime. 678 * since pltime is just for autoconf, pltime processing for 679 * prefix is not necessary. 680 */ 681 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 682 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) { 683 684 /* 685 * address expiration and prefix expiration are 686 * separate. NEVER perform in6_purgeaddr here. 687 */ 688 prelist_remove(pr); 689 } 690 } 691 CURVNET_RESTORE(); 692 } 693 694 /* 695 * ia6 - deprecated/invalidated temporary address 696 */ 697 static int 698 regen_tmpaddr(struct in6_ifaddr *ia6) 699 { 700 struct ifaddr *ifa; 701 struct ifnet *ifp; 702 struct in6_ifaddr *public_ifa6 = NULL; 703 704 ifp = ia6->ia_ifa.ifa_ifp; 705 IF_ADDR_RLOCK(ifp); 706 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 707 struct in6_ifaddr *it6; 708 709 if (ifa->ifa_addr->sa_family != AF_INET6) 710 continue; 711 712 it6 = (struct in6_ifaddr *)ifa; 713 714 /* ignore no autoconf addresses. */ 715 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 716 continue; 717 718 /* ignore autoconf addresses with different prefixes. */ 719 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 720 continue; 721 722 /* 723 * Now we are looking at an autoconf address with the same 724 * prefix as ours. If the address is temporary and is still 725 * preferred, do not create another one. It would be rare, but 726 * could happen, for example, when we resume a laptop PC after 727 * a long period. 728 */ 729 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 730 !IFA6_IS_DEPRECATED(it6)) { 731 public_ifa6 = NULL; 732 break; 733 } 734 735 /* 736 * This is a public autoconf address that has the same prefix 737 * as ours. If it is preferred, keep it. We can't break the 738 * loop here, because there may be a still-preferred temporary 739 * address with the prefix. 740 */ 741 if (!IFA6_IS_DEPRECATED(it6)) 742 public_ifa6 = it6; 743 744 if (public_ifa6 != NULL) 745 ifa_ref(&public_ifa6->ia_ifa); 746 } 747 IF_ADDR_RUNLOCK(ifp); 748 749 if (public_ifa6 != NULL) { 750 int e; 751 752 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 753 ifa_free(&public_ifa6->ia_ifa); 754 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 755 " tmp addr,errno=%d\n", e); 756 return (-1); 757 } 758 ifa_free(&public_ifa6->ia_ifa); 759 return (0); 760 } 761 762 return (-1); 763 } 764 765 /* 766 * Nuke neighbor cache/prefix/default router management table, right before 767 * ifp goes away. 768 */ 769 void 770 nd6_purge(struct ifnet *ifp) 771 { 772 struct nd_defrouter *dr, *ndr; 773 struct nd_prefix *pr, *npr; 774 775 /* 776 * Nuke default router list entries toward ifp. 777 * We defer removal of default router list entries that is installed 778 * in the routing table, in order to keep additional side effects as 779 * small as possible. 780 */ 781 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 782 if (dr->installed) 783 continue; 784 785 if (dr->ifp == ifp) 786 defrtrlist_del(dr); 787 } 788 789 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 790 if (!dr->installed) 791 continue; 792 793 if (dr->ifp == ifp) 794 defrtrlist_del(dr); 795 } 796 797 /* Nuke prefix list entries toward ifp */ 798 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 799 if (pr->ndpr_ifp == ifp) { 800 /* 801 * Because if_detach() does *not* release prefixes 802 * while purging addresses the reference count will 803 * still be above zero. We therefore reset it to 804 * make sure that the prefix really gets purged. 805 */ 806 pr->ndpr_refcnt = 0; 807 808 /* 809 * Previously, pr->ndpr_addr is removed as well, 810 * but I strongly believe we don't have to do it. 811 * nd6_purge() is only called from in6_ifdetach(), 812 * which removes all the associated interface addresses 813 * by itself. 814 * (jinmei@kame.net 20010129) 815 */ 816 prelist_remove(pr); 817 } 818 } 819 820 /* cancel default outgoing interface setting */ 821 if (V_nd6_defifindex == ifp->if_index) 822 nd6_setdefaultiface(0); 823 824 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 825 /* Refresh default router list. */ 826 defrouter_select(); 827 } 828 829 /* XXXXX 830 * We do not nuke the neighbor cache entries here any more 831 * because the neighbor cache is kept in if_afdata[AF_INET6]. 832 * nd6_purge() is invoked by in6_ifdetach() which is called 833 * from if_detach() where everything gets purged. So let 834 * in6_domifdetach() do the actual L2 table purging work. 835 */ 836 } 837 838 /* 839 * the caller acquires and releases the lock on the lltbls 840 * Returns the llentry locked 841 */ 842 struct llentry * 843 nd6_lookup(struct in6_addr *addr6, int flags, struct ifnet *ifp) 844 { 845 struct sockaddr_in6 sin6; 846 struct llentry *ln; 847 int llflags; 848 849 bzero(&sin6, sizeof(sin6)); 850 sin6.sin6_len = sizeof(struct sockaddr_in6); 851 sin6.sin6_family = AF_INET6; 852 sin6.sin6_addr = *addr6; 853 854 IF_AFDATA_LOCK_ASSERT(ifp); 855 856 llflags = 0; 857 if (flags & ND6_CREATE) 858 llflags |= LLE_CREATE; 859 if (flags & ND6_EXCLUSIVE) 860 llflags |= LLE_EXCLUSIVE; 861 862 ln = lla_lookup(LLTABLE6(ifp), llflags, (struct sockaddr *)&sin6); 863 if ((ln != NULL) && (llflags & LLE_CREATE)) 864 ln->ln_state = ND6_LLINFO_NOSTATE; 865 866 return (ln); 867 } 868 869 /* 870 * Test whether a given IPv6 address is a neighbor or not, ignoring 871 * the actual neighbor cache. The neighbor cache is ignored in order 872 * to not reenter the routing code from within itself. 873 */ 874 static int 875 nd6_is_new_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 876 { 877 struct nd_prefix *pr; 878 struct ifaddr *dstaddr; 879 880 /* 881 * A link-local address is always a neighbor. 882 * XXX: a link does not necessarily specify a single interface. 883 */ 884 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 885 struct sockaddr_in6 sin6_copy; 886 u_int32_t zone; 887 888 /* 889 * We need sin6_copy since sa6_recoverscope() may modify the 890 * content (XXX). 891 */ 892 sin6_copy = *addr; 893 if (sa6_recoverscope(&sin6_copy)) 894 return (0); /* XXX: should be impossible */ 895 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 896 return (0); 897 if (sin6_copy.sin6_scope_id == zone) 898 return (1); 899 else 900 return (0); 901 } 902 903 /* 904 * If the address matches one of our addresses, 905 * it should be a neighbor. 906 * If the address matches one of our on-link prefixes, it should be a 907 * neighbor. 908 */ 909 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 910 if (pr->ndpr_ifp != ifp) 911 continue; 912 913 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) { 914 struct rtentry *rt; 915 916 /* Always use the default FIB here. */ 917 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 918 0, 0, RT_DEFAULT_FIB); 919 if (rt == NULL) 920 continue; 921 /* 922 * This is the case where multiple interfaces 923 * have the same prefix, but only one is installed 924 * into the routing table and that prefix entry 925 * is not the one being examined here. In the case 926 * where RADIX_MPATH is enabled, multiple route 927 * entries (of the same rt_key value) will be 928 * installed because the interface addresses all 929 * differ. 930 */ 931 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 932 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) { 933 RTFREE_LOCKED(rt); 934 continue; 935 } 936 RTFREE_LOCKED(rt); 937 } 938 939 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 940 &addr->sin6_addr, &pr->ndpr_mask)) 941 return (1); 942 } 943 944 /* 945 * If the address is assigned on the node of the other side of 946 * a p2p interface, the address should be a neighbor. 947 */ 948 dstaddr = ifa_ifwithdstaddr((struct sockaddr *)addr, RT_ALL_FIBS); 949 if (dstaddr != NULL) { 950 if (dstaddr->ifa_ifp == ifp) { 951 ifa_free(dstaddr); 952 return (1); 953 } 954 ifa_free(dstaddr); 955 } 956 957 /* 958 * If the default router list is empty, all addresses are regarded 959 * as on-link, and thus, as a neighbor. 960 */ 961 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && 962 TAILQ_EMPTY(&V_nd_defrouter) && 963 V_nd6_defifindex == ifp->if_index) { 964 return (1); 965 } 966 967 return (0); 968 } 969 970 971 /* 972 * Detect if a given IPv6 address identifies a neighbor on a given link. 973 * XXX: should take care of the destination of a p2p link? 974 */ 975 int 976 nd6_is_addr_neighbor(struct sockaddr_in6 *addr, struct ifnet *ifp) 977 { 978 struct llentry *lle; 979 int rc = 0; 980 981 IF_AFDATA_UNLOCK_ASSERT(ifp); 982 if (nd6_is_new_addr_neighbor(addr, ifp)) 983 return (1); 984 985 /* 986 * Even if the address matches none of our addresses, it might be 987 * in the neighbor cache. 988 */ 989 IF_AFDATA_RLOCK(ifp); 990 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) { 991 LLE_RUNLOCK(lle); 992 rc = 1; 993 } 994 IF_AFDATA_RUNLOCK(ifp); 995 return (rc); 996 } 997 998 /* 999 * Free an nd6 llinfo entry. 1000 * Since the function would cause significant changes in the kernel, DO NOT 1001 * make it global, unless you have a strong reason for the change, and are sure 1002 * that the change is safe. 1003 */ 1004 static struct llentry * 1005 nd6_free(struct llentry *ln, int gc) 1006 { 1007 struct llentry *next; 1008 struct nd_defrouter *dr; 1009 struct ifnet *ifp; 1010 1011 LLE_WLOCK_ASSERT(ln); 1012 1013 /* 1014 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1015 * even though it is not harmful, it was not really necessary. 1016 */ 1017 1018 /* cancel timer */ 1019 nd6_llinfo_settimer_locked(ln, -1); 1020 1021 ifp = ln->lle_tbl->llt_ifp; 1022 1023 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1024 dr = defrouter_lookup(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1025 1026 if (dr != NULL && dr->expire && 1027 ln->ln_state == ND6_LLINFO_STALE && gc) { 1028 /* 1029 * If the reason for the deletion is just garbage 1030 * collection, and the neighbor is an active default 1031 * router, do not delete it. Instead, reset the GC 1032 * timer using the router's lifetime. 1033 * Simply deleting the entry would affect default 1034 * router selection, which is not necessarily a good 1035 * thing, especially when we're using router preference 1036 * values. 1037 * XXX: the check for ln_state would be redundant, 1038 * but we intentionally keep it just in case. 1039 */ 1040 if (dr->expire > time_uptime) 1041 nd6_llinfo_settimer_locked(ln, 1042 (dr->expire - time_uptime) * hz); 1043 else 1044 nd6_llinfo_settimer_locked(ln, 1045 (long)V_nd6_gctimer * hz); 1046 1047 next = LIST_NEXT(ln, lle_next); 1048 LLE_REMREF(ln); 1049 LLE_WUNLOCK(ln); 1050 return (next); 1051 } 1052 1053 if (dr) { 1054 /* 1055 * Unreachablity of a router might affect the default 1056 * router selection and on-link detection of advertised 1057 * prefixes. 1058 */ 1059 1060 /* 1061 * Temporarily fake the state to choose a new default 1062 * router and to perform on-link determination of 1063 * prefixes correctly. 1064 * Below the state will be set correctly, 1065 * or the entry itself will be deleted. 1066 */ 1067 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1068 } 1069 1070 if (ln->ln_router || dr) { 1071 1072 /* 1073 * We need to unlock to avoid a LOR with rt6_flush() with the 1074 * rnh and for the calls to pfxlist_onlink_check() and 1075 * defrouter_select() in the block further down for calls 1076 * into nd6_lookup(). We still hold a ref. 1077 */ 1078 LLE_WUNLOCK(ln); 1079 1080 /* 1081 * rt6_flush must be called whether or not the neighbor 1082 * is in the Default Router List. 1083 * See a corresponding comment in nd6_na_input(). 1084 */ 1085 rt6_flush(&L3_ADDR_SIN6(ln)->sin6_addr, ifp); 1086 } 1087 1088 if (dr) { 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 1103 if (ln->ln_router || dr) 1104 LLE_WLOCK(ln); 1105 } 1106 1107 /* 1108 * Before deleting the entry, remember the next entry as the 1109 * return value. We need this because pfxlist_onlink_check() above 1110 * might have freed other entries (particularly the old next entry) as 1111 * a side effect (XXX). 1112 */ 1113 next = LIST_NEXT(ln, lle_next); 1114 1115 /* 1116 * Save to unlock. We still hold an extra reference and will not 1117 * free(9) in llentry_free() if someone else holds one as well. 1118 */ 1119 LLE_WUNLOCK(ln); 1120 IF_AFDATA_LOCK(ifp); 1121 LLE_WLOCK(ln); 1122 1123 /* Guard against race with other llentry_free(). */ 1124 if (ln->la_flags & LLE_LINKED) { 1125 LLE_REMREF(ln); 1126 llentry_free(ln); 1127 } else 1128 LLE_FREE_LOCKED(ln); 1129 1130 IF_AFDATA_UNLOCK(ifp); 1131 1132 return (next); 1133 } 1134 1135 /* 1136 * Upper-layer reachability hint for Neighbor Unreachability Detection. 1137 * 1138 * XXX cost-effective methods? 1139 */ 1140 void 1141 nd6_nud_hint(struct rtentry *rt, struct in6_addr *dst6, int force) 1142 { 1143 struct llentry *ln; 1144 struct ifnet *ifp; 1145 1146 if ((dst6 == NULL) || (rt == NULL)) 1147 return; 1148 1149 ifp = rt->rt_ifp; 1150 IF_AFDATA_RLOCK(ifp); 1151 ln = nd6_lookup(dst6, ND6_EXCLUSIVE, NULL); 1152 IF_AFDATA_RUNLOCK(ifp); 1153 if (ln == NULL) 1154 return; 1155 1156 if (ln->ln_state < ND6_LLINFO_REACHABLE) 1157 goto done; 1158 1159 /* 1160 * if we get upper-layer reachability confirmation many times, 1161 * it is possible we have false information. 1162 */ 1163 if (!force) { 1164 ln->ln_byhint++; 1165 if (ln->ln_byhint > V_nd6_maxnudhint) { 1166 goto done; 1167 } 1168 } 1169 1170 ln->ln_state = ND6_LLINFO_REACHABLE; 1171 if (!ND6_LLINFO_PERMANENT(ln)) { 1172 nd6_llinfo_settimer_locked(ln, 1173 (long)ND_IFINFO(rt->rt_ifp)->reachable * hz); 1174 } 1175 done: 1176 LLE_WUNLOCK(ln); 1177 } 1178 1179 1180 /* 1181 * Rejuvenate this function for routing operations related 1182 * processing. 1183 */ 1184 void 1185 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info) 1186 { 1187 struct sockaddr_in6 *gateway; 1188 struct nd_defrouter *dr; 1189 struct ifnet *ifp; 1190 1191 gateway = (struct sockaddr_in6 *)rt->rt_gateway; 1192 ifp = rt->rt_ifp; 1193 1194 switch (req) { 1195 case RTM_ADD: 1196 break; 1197 1198 case RTM_DELETE: 1199 if (!ifp) 1200 return; 1201 /* 1202 * Only indirect routes are interesting. 1203 */ 1204 if ((rt->rt_flags & RTF_GATEWAY) == 0) 1205 return; 1206 /* 1207 * check for default route 1208 */ 1209 if (IN6_ARE_ADDR_EQUAL(&in6addr_any, 1210 &SIN6(rt_key(rt))->sin6_addr)) { 1211 1212 dr = defrouter_lookup(&gateway->sin6_addr, ifp); 1213 if (dr != NULL) 1214 dr->installed = 0; 1215 } 1216 break; 1217 } 1218 } 1219 1220 1221 int 1222 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) 1223 { 1224 struct in6_drlist *drl = (struct in6_drlist *)data; 1225 struct in6_oprlist *oprl = (struct in6_oprlist *)data; 1226 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1227 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1228 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1229 struct nd_defrouter *dr; 1230 struct nd_prefix *pr; 1231 int i = 0, error = 0; 1232 1233 if (ifp->if_afdata[AF_INET6] == NULL) 1234 return (EPFNOSUPPORT); 1235 switch (cmd) { 1236 case SIOCGDRLST_IN6: 1237 /* 1238 * obsolete API, use sysctl under net.inet6.icmp6 1239 */ 1240 bzero(drl, sizeof(*drl)); 1241 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 1242 if (i >= DRLSTSIZ) 1243 break; 1244 drl->defrouter[i].rtaddr = dr->rtaddr; 1245 in6_clearscope(&drl->defrouter[i].rtaddr); 1246 1247 drl->defrouter[i].flags = dr->flags; 1248 drl->defrouter[i].rtlifetime = dr->rtlifetime; 1249 drl->defrouter[i].expire = dr->expire + 1250 (time_second - time_uptime); 1251 drl->defrouter[i].if_index = dr->ifp->if_index; 1252 i++; 1253 } 1254 break; 1255 case SIOCGPRLST_IN6: 1256 /* 1257 * obsolete API, use sysctl under net.inet6.icmp6 1258 * 1259 * XXX the structure in6_prlist was changed in backward- 1260 * incompatible manner. in6_oprlist is used for SIOCGPRLST_IN6, 1261 * in6_prlist is used for nd6_sysctl() - fill_prlist(). 1262 */ 1263 /* 1264 * XXX meaning of fields, especialy "raflags", is very 1265 * differnet between RA prefix list and RR/static prefix list. 1266 * how about separating ioctls into two? 1267 */ 1268 bzero(oprl, sizeof(*oprl)); 1269 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 1270 struct nd_pfxrouter *pfr; 1271 int j; 1272 1273 if (i >= PRLSTSIZ) 1274 break; 1275 oprl->prefix[i].prefix = pr->ndpr_prefix.sin6_addr; 1276 oprl->prefix[i].raflags = pr->ndpr_raf; 1277 oprl->prefix[i].prefixlen = pr->ndpr_plen; 1278 oprl->prefix[i].vltime = pr->ndpr_vltime; 1279 oprl->prefix[i].pltime = pr->ndpr_pltime; 1280 oprl->prefix[i].if_index = pr->ndpr_ifp->if_index; 1281 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 1282 oprl->prefix[i].expire = 0; 1283 else { 1284 time_t maxexpire; 1285 1286 /* XXX: we assume time_t is signed. */ 1287 maxexpire = (-1) & 1288 ~((time_t)1 << 1289 ((sizeof(maxexpire) * 8) - 1)); 1290 if (pr->ndpr_vltime < 1291 maxexpire - pr->ndpr_lastupdate) { 1292 oprl->prefix[i].expire = 1293 pr->ndpr_lastupdate + 1294 pr->ndpr_vltime + 1295 (time_second - time_uptime); 1296 } else 1297 oprl->prefix[i].expire = maxexpire; 1298 } 1299 1300 j = 0; 1301 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 1302 if (j < DRLSTSIZ) { 1303 #define RTRADDR oprl->prefix[i].advrtr[j] 1304 RTRADDR = pfr->router->rtaddr; 1305 in6_clearscope(&RTRADDR); 1306 #undef RTRADDR 1307 } 1308 j++; 1309 } 1310 oprl->prefix[i].advrtrs = j; 1311 oprl->prefix[i].origin = PR_ORIG_RA; 1312 1313 i++; 1314 } 1315 1316 break; 1317 case OSIOCGIFINFO_IN6: 1318 #define ND ndi->ndi 1319 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1320 bzero(&ND, sizeof(ND)); 1321 ND.linkmtu = IN6_LINKMTU(ifp); 1322 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1323 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1324 ND.reachable = ND_IFINFO(ifp)->reachable; 1325 ND.retrans = ND_IFINFO(ifp)->retrans; 1326 ND.flags = ND_IFINFO(ifp)->flags; 1327 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1328 ND.chlim = ND_IFINFO(ifp)->chlim; 1329 break; 1330 case SIOCGIFINFO_IN6: 1331 ND = *ND_IFINFO(ifp); 1332 break; 1333 case SIOCSIFINFO_IN6: 1334 /* 1335 * used to change host variables from userland. 1336 * intented for a use on router to reflect RA configurations. 1337 */ 1338 /* 0 means 'unspecified' */ 1339 if (ND.linkmtu != 0) { 1340 if (ND.linkmtu < IPV6_MMTU || 1341 ND.linkmtu > IN6_LINKMTU(ifp)) { 1342 error = EINVAL; 1343 break; 1344 } 1345 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1346 } 1347 1348 if (ND.basereachable != 0) { 1349 int obasereachable = ND_IFINFO(ifp)->basereachable; 1350 1351 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1352 if (ND.basereachable != obasereachable) 1353 ND_IFINFO(ifp)->reachable = 1354 ND_COMPUTE_RTIME(ND.basereachable); 1355 } 1356 if (ND.retrans != 0) 1357 ND_IFINFO(ifp)->retrans = ND.retrans; 1358 if (ND.chlim != 0) 1359 ND_IFINFO(ifp)->chlim = ND.chlim; 1360 /* FALLTHROUGH */ 1361 case SIOCSIFINFO_FLAGS: 1362 { 1363 struct ifaddr *ifa; 1364 struct in6_ifaddr *ia; 1365 1366 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1367 !(ND.flags & ND6_IFF_IFDISABLED)) { 1368 /* ifdisabled 1->0 transision */ 1369 1370 /* 1371 * If the interface is marked as ND6_IFF_IFDISABLED and 1372 * has an link-local address with IN6_IFF_DUPLICATED, 1373 * do not clear ND6_IFF_IFDISABLED. 1374 * See RFC 4862, Section 5.4.5. 1375 */ 1376 int duplicated_linklocal = 0; 1377 1378 IF_ADDR_RLOCK(ifp); 1379 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1380 if (ifa->ifa_addr->sa_family != AF_INET6) 1381 continue; 1382 ia = (struct in6_ifaddr *)ifa; 1383 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) && 1384 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1385 duplicated_linklocal = 1; 1386 break; 1387 } 1388 } 1389 IF_ADDR_RUNLOCK(ifp); 1390 1391 if (duplicated_linklocal) { 1392 ND.flags |= ND6_IFF_IFDISABLED; 1393 log(LOG_ERR, "Cannot enable an interface" 1394 " with a link-local address marked" 1395 " duplicate.\n"); 1396 } else { 1397 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED; 1398 if (ifp->if_flags & IFF_UP) 1399 in6_if_up(ifp); 1400 } 1401 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1402 (ND.flags & ND6_IFF_IFDISABLED)) { 1403 /* ifdisabled 0->1 transision */ 1404 /* Mark all IPv6 address as tentative. */ 1405 1406 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; 1407 IF_ADDR_RLOCK(ifp); 1408 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1409 if (ifa->ifa_addr->sa_family != AF_INET6) 1410 continue; 1411 ia = (struct in6_ifaddr *)ifa; 1412 ia->ia6_flags |= IN6_IFF_TENTATIVE; 1413 } 1414 IF_ADDR_RUNLOCK(ifp); 1415 } 1416 1417 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) { 1418 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) { 1419 /* auto_linklocal 0->1 transision */ 1420 1421 /* If no link-local address on ifp, configure */ 1422 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL; 1423 in6_ifattach(ifp, NULL); 1424 } else if (!(ND.flags & ND6_IFF_IFDISABLED) && 1425 ifp->if_flags & IFF_UP) { 1426 /* 1427 * When the IF already has 1428 * ND6_IFF_AUTO_LINKLOCAL, no link-local 1429 * address is assigned, and IFF_UP, try to 1430 * assign one. 1431 */ 1432 int haslinklocal = 0; 1433 1434 IF_ADDR_RLOCK(ifp); 1435 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1436 if (ifa->ifa_addr->sa_family != AF_INET6) 1437 continue; 1438 ia = (struct in6_ifaddr *)ifa; 1439 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) { 1440 haslinklocal = 1; 1441 break; 1442 } 1443 } 1444 IF_ADDR_RUNLOCK(ifp); 1445 if (!haslinklocal) 1446 in6_ifattach(ifp, NULL); 1447 } 1448 } 1449 } 1450 ND_IFINFO(ifp)->flags = ND.flags; 1451 break; 1452 #undef ND 1453 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1454 /* sync kernel routing table with the default router list */ 1455 defrouter_reset(); 1456 defrouter_select(); 1457 break; 1458 case SIOCSPFXFLUSH_IN6: 1459 { 1460 /* flush all the prefix advertised by routers */ 1461 struct nd_prefix *pr, *next; 1462 1463 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) { 1464 struct in6_ifaddr *ia, *ia_next; 1465 1466 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1467 continue; /* XXX */ 1468 1469 /* do we really have to remove addresses as well? */ 1470 /* XXXRW: in6_ifaddrhead locking. */ 1471 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link, 1472 ia_next) { 1473 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1474 continue; 1475 1476 if (ia->ia6_ndpr == pr) 1477 in6_purgeaddr(&ia->ia_ifa); 1478 } 1479 prelist_remove(pr); 1480 } 1481 break; 1482 } 1483 case SIOCSRTRFLUSH_IN6: 1484 { 1485 /* flush all the default routers */ 1486 struct nd_defrouter *dr, *next; 1487 1488 defrouter_reset(); 1489 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) { 1490 defrtrlist_del(dr); 1491 } 1492 defrouter_select(); 1493 break; 1494 } 1495 case SIOCGNBRINFO_IN6: 1496 { 1497 struct llentry *ln; 1498 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1499 1500 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1501 return (error); 1502 1503 IF_AFDATA_RLOCK(ifp); 1504 ln = nd6_lookup(&nb_addr, 0, ifp); 1505 IF_AFDATA_RUNLOCK(ifp); 1506 1507 if (ln == NULL) { 1508 error = EINVAL; 1509 break; 1510 } 1511 nbi->state = ln->ln_state; 1512 nbi->asked = ln->la_asked; 1513 nbi->isrouter = ln->ln_router; 1514 if (ln->la_expire == 0) 1515 nbi->expire = 0; 1516 else 1517 nbi->expire = ln->la_expire + 1518 (time_second - time_uptime); 1519 LLE_RUNLOCK(ln); 1520 break; 1521 } 1522 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1523 ndif->ifindex = V_nd6_defifindex; 1524 break; 1525 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1526 return (nd6_setdefaultiface(ndif->ifindex)); 1527 } 1528 return (error); 1529 } 1530 1531 /* 1532 * Create neighbor cache entry and cache link-layer address, 1533 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1534 * 1535 * type - ICMP6 type 1536 * code - type dependent information 1537 * 1538 * XXXXX 1539 * The caller of this function already acquired the ndp 1540 * cache table lock because the cache entry is returned. 1541 */ 1542 struct llentry * 1543 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, 1544 int lladdrlen, int type, int code) 1545 { 1546 struct llentry *ln = NULL; 1547 int is_newentry; 1548 int do_update; 1549 int olladdr; 1550 int llchange; 1551 int flags; 1552 int newstate = 0; 1553 uint16_t router = 0; 1554 struct sockaddr_in6 sin6; 1555 struct mbuf *chain = NULL; 1556 int static_route = 0; 1557 1558 IF_AFDATA_UNLOCK_ASSERT(ifp); 1559 1560 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__)); 1561 KASSERT(from != NULL, ("%s: from == NULL", __func__)); 1562 1563 /* nothing must be updated for unspecified address */ 1564 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1565 return NULL; 1566 1567 /* 1568 * Validation about ifp->if_addrlen and lladdrlen must be done in 1569 * the caller. 1570 * 1571 * XXX If the link does not have link-layer adderss, what should 1572 * we do? (ifp->if_addrlen == 0) 1573 * Spec says nothing in sections for RA, RS and NA. There's small 1574 * description on it in NS section (RFC 2461 7.2.3). 1575 */ 1576 flags = lladdr ? ND6_EXCLUSIVE : 0; 1577 IF_AFDATA_RLOCK(ifp); 1578 ln = nd6_lookup(from, flags, ifp); 1579 IF_AFDATA_RUNLOCK(ifp); 1580 if (ln == NULL) { 1581 flags |= ND6_EXCLUSIVE; 1582 IF_AFDATA_LOCK(ifp); 1583 ln = nd6_lookup(from, flags | ND6_CREATE, ifp); 1584 IF_AFDATA_UNLOCK(ifp); 1585 is_newentry = 1; 1586 } else { 1587 /* do nothing if static ndp is set */ 1588 if (ln->la_flags & LLE_STATIC) { 1589 static_route = 1; 1590 goto done; 1591 } 1592 is_newentry = 0; 1593 } 1594 if (ln == NULL) 1595 return (NULL); 1596 1597 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0; 1598 if (olladdr && lladdr) { 1599 llchange = bcmp(lladdr, &ln->ll_addr, 1600 ifp->if_addrlen); 1601 } else 1602 llchange = 0; 1603 1604 /* 1605 * newentry olladdr lladdr llchange (*=record) 1606 * 0 n n -- (1) 1607 * 0 y n -- (2) 1608 * 0 n y -- (3) * STALE 1609 * 0 y y n (4) * 1610 * 0 y y y (5) * STALE 1611 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1612 * 1 -- y -- (7) * STALE 1613 */ 1614 1615 if (lladdr) { /* (3-5) and (7) */ 1616 /* 1617 * Record source link-layer address 1618 * XXX is it dependent to ifp->if_type? 1619 */ 1620 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen); 1621 ln->la_flags |= LLE_VALID; 1622 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 1623 } 1624 1625 if (!is_newentry) { 1626 if ((!olladdr && lladdr != NULL) || /* (3) */ 1627 (olladdr && lladdr != NULL && llchange)) { /* (5) */ 1628 do_update = 1; 1629 newstate = ND6_LLINFO_STALE; 1630 } else /* (1-2,4) */ 1631 do_update = 0; 1632 } else { 1633 do_update = 1; 1634 if (lladdr == NULL) /* (6) */ 1635 newstate = ND6_LLINFO_NOSTATE; 1636 else /* (7) */ 1637 newstate = ND6_LLINFO_STALE; 1638 } 1639 1640 if (do_update) { 1641 /* 1642 * Update the state of the neighbor cache. 1643 */ 1644 ln->ln_state = newstate; 1645 1646 if (ln->ln_state == ND6_LLINFO_STALE) { 1647 /* 1648 * XXX: since nd6_output() below will cause 1649 * state tansition to DELAY and reset the timer, 1650 * we must set the timer now, although it is actually 1651 * meaningless. 1652 */ 1653 nd6_llinfo_settimer_locked(ln, (long)V_nd6_gctimer * hz); 1654 1655 if (ln->la_hold) { 1656 struct mbuf *m_hold, *m_hold_next; 1657 1658 /* 1659 * reset the la_hold in advance, to explicitly 1660 * prevent a la_hold lookup in nd6_output() 1661 * (wouldn't happen, though...) 1662 */ 1663 for (m_hold = ln->la_hold, ln->la_hold = NULL; 1664 m_hold; m_hold = m_hold_next) { 1665 m_hold_next = m_hold->m_nextpkt; 1666 m_hold->m_nextpkt = NULL; 1667 1668 /* 1669 * we assume ifp is not a p2p here, so 1670 * just set the 2nd argument as the 1671 * 1st one. 1672 */ 1673 nd6_output_lle(ifp, ifp, m_hold, L3_ADDR_SIN6(ln), NULL, ln, &chain); 1674 } 1675 /* 1676 * If we have mbufs in the chain we need to do 1677 * deferred transmit. Copy the address from the 1678 * llentry before dropping the lock down below. 1679 */ 1680 if (chain != NULL) 1681 memcpy(&sin6, L3_ADDR_SIN6(ln), sizeof(sin6)); 1682 } 1683 } else if (ln->ln_state == ND6_LLINFO_INCOMPLETE) { 1684 /* probe right away */ 1685 nd6_llinfo_settimer_locked((void *)ln, 0); 1686 } 1687 } 1688 1689 /* 1690 * ICMP6 type dependent behavior. 1691 * 1692 * NS: clear IsRouter if new entry 1693 * RS: clear IsRouter 1694 * RA: set IsRouter if there's lladdr 1695 * redir: clear IsRouter if new entry 1696 * 1697 * RA case, (1): 1698 * The spec says that we must set IsRouter in the following cases: 1699 * - If lladdr exist, set IsRouter. This means (1-5). 1700 * - If it is old entry (!newentry), set IsRouter. This means (7). 1701 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1702 * A quetion arises for (1) case. (1) case has no lladdr in the 1703 * neighbor cache, this is similar to (6). 1704 * This case is rare but we figured that we MUST NOT set IsRouter. 1705 * 1706 * newentry olladdr lladdr llchange NS RS RA redir 1707 * D R 1708 * 0 n n -- (1) c ? s 1709 * 0 y n -- (2) c s s 1710 * 0 n y -- (3) c s s 1711 * 0 y y n (4) c s s 1712 * 0 y y y (5) c s s 1713 * 1 -- n -- (6) c c c s 1714 * 1 -- y -- (7) c c s c s 1715 * 1716 * (c=clear s=set) 1717 */ 1718 switch (type & 0xff) { 1719 case ND_NEIGHBOR_SOLICIT: 1720 /* 1721 * New entry must have is_router flag cleared. 1722 */ 1723 if (is_newentry) /* (6-7) */ 1724 ln->ln_router = 0; 1725 break; 1726 case ND_REDIRECT: 1727 /* 1728 * If the icmp is a redirect to a better router, always set the 1729 * is_router flag. Otherwise, if the entry is newly created, 1730 * clear the flag. [RFC 2461, sec 8.3] 1731 */ 1732 if (code == ND_REDIRECT_ROUTER) 1733 ln->ln_router = 1; 1734 else if (is_newentry) /* (6-7) */ 1735 ln->ln_router = 0; 1736 break; 1737 case ND_ROUTER_SOLICIT: 1738 /* 1739 * is_router flag must always be cleared. 1740 */ 1741 ln->ln_router = 0; 1742 break; 1743 case ND_ROUTER_ADVERT: 1744 /* 1745 * Mark an entry with lladdr as a router. 1746 */ 1747 if ((!is_newentry && (olladdr || lladdr)) || /* (2-5) */ 1748 (is_newentry && lladdr)) { /* (7) */ 1749 ln->ln_router = 1; 1750 } 1751 break; 1752 } 1753 1754 if (ln != NULL) { 1755 static_route = (ln->la_flags & LLE_STATIC); 1756 router = ln->ln_router; 1757 1758 if (flags & ND6_EXCLUSIVE) 1759 LLE_WUNLOCK(ln); 1760 else 1761 LLE_RUNLOCK(ln); 1762 if (static_route) 1763 ln = NULL; 1764 } 1765 if (chain) 1766 nd6_output_flush(ifp, ifp, chain, &sin6); 1767 1768 /* 1769 * When the link-layer address of a router changes, select the 1770 * best router again. In particular, when the neighbor entry is newly 1771 * created, it might affect the selection policy. 1772 * Question: can we restrict the first condition to the "is_newentry" 1773 * case? 1774 * XXX: when we hear an RA from a new router with the link-layer 1775 * address option, defrouter_select() is called twice, since 1776 * defrtrlist_update called the function as well. However, I believe 1777 * we can compromise the overhead, since it only happens the first 1778 * time. 1779 * XXX: although defrouter_select() should not have a bad effect 1780 * for those are not autoconfigured hosts, we explicitly avoid such 1781 * cases for safety. 1782 */ 1783 if (do_update && router && 1784 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1785 /* 1786 * guaranteed recursion 1787 */ 1788 defrouter_select(); 1789 } 1790 1791 return (ln); 1792 done: 1793 if (ln != NULL) { 1794 if (flags & ND6_EXCLUSIVE) 1795 LLE_WUNLOCK(ln); 1796 else 1797 LLE_RUNLOCK(ln); 1798 if (static_route) 1799 ln = NULL; 1800 } 1801 return (ln); 1802 } 1803 1804 static void 1805 nd6_slowtimo(void *arg) 1806 { 1807 CURVNET_SET((struct vnet *) arg); 1808 struct nd_ifinfo *nd6if; 1809 struct ifnet *ifp; 1810 1811 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 1812 nd6_slowtimo, curvnet); 1813 IFNET_RLOCK_NOSLEEP(); 1814 TAILQ_FOREACH(ifp, &V_ifnet, if_link) { 1815 if (ifp->if_afdata[AF_INET6] == NULL) 1816 continue; 1817 nd6if = ND_IFINFO(ifp); 1818 if (nd6if->basereachable && /* already initialized */ 1819 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 1820 /* 1821 * Since reachable time rarely changes by router 1822 * advertisements, we SHOULD insure that a new random 1823 * value gets recomputed at least once every few hours. 1824 * (RFC 2461, 6.3.4) 1825 */ 1826 nd6if->recalctm = V_nd6_recalc_reachtm_interval; 1827 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 1828 } 1829 } 1830 IFNET_RUNLOCK_NOSLEEP(); 1831 CURVNET_RESTORE(); 1832 } 1833 1834 /* 1835 * IPv6 packet output - light version. 1836 * Checks if destination LLE exists and is in proper state 1837 * (e.g no modification required). If not true, fall back to 1838 * "heavy" version. 1839 */ 1840 int 1841 nd6_output(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m, 1842 struct sockaddr_in6 *dst, struct rtentry *rt0) 1843 { 1844 struct llentry *ln = NULL; 1845 int error = 0; 1846 1847 /* discard the packet if IPv6 operation is disabled on the interface */ 1848 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 1849 m_freem(m); 1850 return (ENETDOWN); /* better error? */ 1851 } 1852 1853 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) 1854 goto sendpkt; 1855 1856 if (nd6_need_cache(ifp) == 0) 1857 goto sendpkt; 1858 1859 IF_AFDATA_RLOCK(ifp); 1860 ln = nd6_lookup(&dst->sin6_addr, 0, ifp); 1861 IF_AFDATA_RUNLOCK(ifp); 1862 1863 /* 1864 * Perform fast path for the following cases: 1865 * 1) lle state is REACHABLE 1866 * 2) lle state is DELAY (NS message sentNS message sent) 1867 * 1868 * Every other case involves lle modification, so we handle 1869 * them separately. 1870 */ 1871 if (ln == NULL || (ln->ln_state != ND6_LLINFO_REACHABLE && 1872 ln->ln_state != ND6_LLINFO_DELAY)) { 1873 /* Fall back to slow processing path */ 1874 if (ln != NULL) 1875 LLE_RUNLOCK(ln); 1876 return (nd6_output_lle(ifp, origifp, m, dst, rt0, NULL, NULL)); 1877 } 1878 1879 sendpkt: 1880 if (ln != NULL) 1881 LLE_RUNLOCK(ln); 1882 1883 #ifdef MAC 1884 mac_netinet6_nd6_send(ifp, m); 1885 #endif 1886 1887 /* 1888 * If called from nd6_ns_output() (NS), nd6_na_output() (NA), 1889 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA 1890 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND 1891 * to be diverted to user space. When re-injected into the kernel, 1892 * send_output() will directly dispatch them to the outgoing interface. 1893 */ 1894 if (send_sendso_input_hook != NULL) { 1895 struct m_tag *mtag; 1896 struct ip6_hdr *ip6; 1897 int ip6len; 1898 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); 1899 if (mtag != NULL) { 1900 ip6 = mtod(m, struct ip6_hdr *); 1901 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); 1902 /* Use the SEND socket */ 1903 error = send_sendso_input_hook(m, ifp, SND_OUT, 1904 ip6len); 1905 /* -1 == no app on SEND socket */ 1906 if (error == 0 || error != -1) 1907 return (error); 1908 } 1909 } 1910 1911 m_clrprotoflags(m); /* Avoid confusing lower layers. */ 1912 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL, 1913 mtod(m, struct ip6_hdr *)); 1914 1915 if ((ifp->if_flags & IFF_LOOPBACK) == 0) 1916 origifp = ifp; 1917 1918 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL); 1919 return (error); 1920 } 1921 1922 1923 /* 1924 * Output IPv6 packet - heavy version. 1925 * Function assume that either 1926 * 1) destination LLE does not exist, is invalid or stale, so 1927 * ND6_EXCLUSIVE lock needs to be acquired 1928 * 2) destination lle is provided (with ND6_EXCLUSIVE lock), 1929 * in that case packets are queued in &chain. 1930 * 1931 */ 1932 int 1933 nd6_output_lle(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m, 1934 struct sockaddr_in6 *dst, struct rtentry *rt0, struct llentry *lle, 1935 struct mbuf **chain) 1936 { 1937 struct m_tag *mtag; 1938 struct ip6_hdr *ip6; 1939 int error = 0; 1940 int flags = 0; 1941 int has_lle = 0; 1942 int ip6len; 1943 1944 #ifdef INVARIANTS 1945 if (lle != NULL) { 1946 1947 LLE_WLOCK_ASSERT(lle); 1948 1949 KASSERT(chain != NULL, (" lle locked but no mbuf chain pointer passed")); 1950 } 1951 #endif 1952 KASSERT(m != NULL, ("NULL mbuf, nothing to send")); 1953 /* discard the packet if IPv6 operation is disabled on the interface */ 1954 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 1955 m_freem(m); 1956 return (ENETDOWN); /* better error? */ 1957 } 1958 1959 if (lle != NULL) 1960 has_lle = 1; 1961 1962 if (IN6_IS_ADDR_MULTICAST(&dst->sin6_addr)) 1963 goto sendpkt; 1964 1965 if (nd6_need_cache(ifp) == 0) 1966 goto sendpkt; 1967 1968 /* 1969 * Address resolution or Neighbor Unreachability Detection 1970 * for the next hop. 1971 * At this point, the destination of the packet must be a unicast 1972 * or an anycast address(i.e. not a multicast). 1973 */ 1974 if (lle == NULL) { 1975 IF_AFDATA_RLOCK(ifp); 1976 lle = nd6_lookup(&dst->sin6_addr, ND6_EXCLUSIVE, ifp); 1977 IF_AFDATA_RUNLOCK(ifp); 1978 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) { 1979 /* 1980 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 1981 * the condition below is not very efficient. But we believe 1982 * it is tolerable, because this should be a rare case. 1983 */ 1984 flags = ND6_CREATE | ND6_EXCLUSIVE; 1985 IF_AFDATA_LOCK(ifp); 1986 lle = nd6_lookup(&dst->sin6_addr, flags, ifp); 1987 IF_AFDATA_UNLOCK(ifp); 1988 } 1989 } 1990 if (lle == NULL) { 1991 if ((ifp->if_flags & IFF_POINTOPOINT) == 0 && 1992 !(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 1993 char ip6buf[INET6_ADDRSTRLEN]; 1994 log(LOG_DEBUG, 1995 "nd6_output: can't allocate llinfo for %s " 1996 "(ln=%p)\n", 1997 ip6_sprintf(ip6buf, &dst->sin6_addr), lle); 1998 m_freem(m); 1999 return (ENOBUFS); 2000 } 2001 goto sendpkt; /* send anyway */ 2002 } 2003 2004 LLE_WLOCK_ASSERT(lle); 2005 2006 /* We don't have to do link-layer address resolution on a p2p link. */ 2007 if ((ifp->if_flags & IFF_POINTOPOINT) != 0 && 2008 lle->ln_state < ND6_LLINFO_REACHABLE) { 2009 lle->ln_state = ND6_LLINFO_STALE; 2010 nd6_llinfo_settimer_locked(lle, (long)V_nd6_gctimer * hz); 2011 } 2012 2013 /* 2014 * The first time we send a packet to a neighbor whose entry is 2015 * STALE, we have to change the state to DELAY and a sets a timer to 2016 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 2017 * neighbor unreachability detection on expiration. 2018 * (RFC 2461 7.3.3) 2019 */ 2020 if (lle->ln_state == ND6_LLINFO_STALE) { 2021 lle->la_asked = 0; 2022 lle->ln_state = ND6_LLINFO_DELAY; 2023 nd6_llinfo_settimer_locked(lle, (long)V_nd6_delay * hz); 2024 } 2025 2026 /* 2027 * If the neighbor cache entry has a state other than INCOMPLETE 2028 * (i.e. its link-layer address is already resolved), just 2029 * send the packet. 2030 */ 2031 if (lle->ln_state > ND6_LLINFO_INCOMPLETE) 2032 goto sendpkt; 2033 2034 /* 2035 * There is a neighbor cache entry, but no ethernet address 2036 * response yet. Append this latest packet to the end of the 2037 * packet queue in the mbuf, unless the number of the packet 2038 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 2039 * the oldest packet in the queue will be removed. 2040 */ 2041 if (lle->ln_state == ND6_LLINFO_NOSTATE) 2042 lle->ln_state = ND6_LLINFO_INCOMPLETE; 2043 2044 if (lle->la_hold != NULL) { 2045 struct mbuf *m_hold; 2046 int i; 2047 2048 i = 0; 2049 for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){ 2050 i++; 2051 if (m_hold->m_nextpkt == NULL) { 2052 m_hold->m_nextpkt = m; 2053 break; 2054 } 2055 } 2056 while (i >= V_nd6_maxqueuelen) { 2057 m_hold = lle->la_hold; 2058 lle->la_hold = lle->la_hold->m_nextpkt; 2059 m_freem(m_hold); 2060 i--; 2061 } 2062 } else { 2063 lle->la_hold = m; 2064 } 2065 2066 /* 2067 * If there has been no NS for the neighbor after entering the 2068 * INCOMPLETE state, send the first solicitation. 2069 */ 2070 if (!ND6_LLINFO_PERMANENT(lle) && lle->la_asked == 0) { 2071 lle->la_asked++; 2072 2073 nd6_llinfo_settimer_locked(lle, 2074 (long)ND_IFINFO(ifp)->retrans * hz / 1000); 2075 LLE_WUNLOCK(lle); 2076 nd6_ns_output(ifp, NULL, &dst->sin6_addr, lle, 0); 2077 if (has_lle != 0) 2078 LLE_WLOCK(lle); 2079 } else if (has_lle == 0) { 2080 /* 2081 * We did the lookup (no lle arg) so we 2082 * need to do the unlock here. 2083 */ 2084 LLE_WUNLOCK(lle); 2085 } 2086 2087 return (0); 2088 2089 sendpkt: 2090 /* 2091 * ln is valid and the caller did not pass in 2092 * an llentry 2093 */ 2094 if (lle != NULL && has_lle == 0) 2095 LLE_WUNLOCK(lle); 2096 2097 #ifdef MAC 2098 mac_netinet6_nd6_send(ifp, m); 2099 #endif 2100 2101 /* 2102 * If called from nd6_ns_output() (NS), nd6_na_output() (NA), 2103 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA 2104 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND 2105 * to be diverted to user space. When re-injected into the kernel, 2106 * send_output() will directly dispatch them to the outgoing interface. 2107 */ 2108 if (send_sendso_input_hook != NULL) { 2109 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); 2110 if (mtag != NULL) { 2111 ip6 = mtod(m, struct ip6_hdr *); 2112 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); 2113 /* Use the SEND socket */ 2114 error = send_sendso_input_hook(m, ifp, SND_OUT, 2115 ip6len); 2116 /* -1 == no app on SEND socket */ 2117 if (error == 0 || error != -1) 2118 return (error); 2119 } 2120 } 2121 2122 /* 2123 * We were passed in a pointer to an lle with the lock held 2124 * this means that we can't call if_output as we will 2125 * recurse on the lle lock - so what we do is we create 2126 * a list of mbufs to send and transmit them in the caller 2127 * after the lock is dropped 2128 */ 2129 if (has_lle != 0) { 2130 if (*chain == NULL) 2131 *chain = m; 2132 else { 2133 struct mbuf *mb; 2134 2135 /* 2136 * append mbuf to end of deferred chain 2137 */ 2138 mb = *chain; 2139 while (mb->m_nextpkt != NULL) 2140 mb = mb->m_nextpkt; 2141 mb->m_nextpkt = m; 2142 } 2143 return (error); 2144 } 2145 m_clrprotoflags(m); /* Avoid confusing lower layers. */ 2146 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL, 2147 mtod(m, struct ip6_hdr *)); 2148 2149 if ((ifp->if_flags & IFF_LOOPBACK) == 0) 2150 origifp = ifp; 2151 2152 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL); 2153 return (error); 2154 } 2155 2156 2157 int 2158 nd6_output_flush(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain, 2159 struct sockaddr_in6 *dst) 2160 { 2161 struct mbuf *m, *m_head; 2162 struct ifnet *outifp; 2163 int error = 0; 2164 2165 m_head = chain; 2166 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2167 outifp = origifp; 2168 else 2169 outifp = ifp; 2170 2171 while (m_head) { 2172 m = m_head; 2173 m_head = m_head->m_nextpkt; 2174 error = (*ifp->if_output)(ifp, m, (struct sockaddr *)dst, NULL); 2175 } 2176 2177 /* 2178 * XXX 2179 * note that intermediate errors are blindly ignored - but this is 2180 * the same convention as used with nd6_output when called by 2181 * nd6_cache_lladdr 2182 */ 2183 return (error); 2184 } 2185 2186 2187 int 2188 nd6_need_cache(struct ifnet *ifp) 2189 { 2190 /* 2191 * XXX: we currently do not make neighbor cache on any interface 2192 * other than ARCnet, Ethernet, FDDI and GIF. 2193 * 2194 * RFC2893 says: 2195 * - unidirectional tunnels needs no ND 2196 */ 2197 switch (ifp->if_type) { 2198 case IFT_ARCNET: 2199 case IFT_ETHER: 2200 case IFT_FDDI: 2201 case IFT_IEEE1394: 2202 #ifdef IFT_L2VLAN 2203 case IFT_L2VLAN: 2204 #endif 2205 #ifdef IFT_IEEE80211 2206 case IFT_IEEE80211: 2207 #endif 2208 case IFT_INFINIBAND: 2209 case IFT_BRIDGE: 2210 case IFT_PROPVIRTUAL: 2211 return (1); 2212 default: 2213 return (0); 2214 } 2215 } 2216 2217 /* 2218 * Add pernament ND6 link-layer record for given 2219 * interface address. 2220 * 2221 * Very similar to IPv4 arp_ifinit(), but: 2222 * 1) IPv6 DAD is performed in different place 2223 * 2) It is called by IPv6 protocol stack in contrast to 2224 * arp_ifinit() which is typically called in SIOCSIFADDR 2225 * driver ioctl handler. 2226 * 2227 */ 2228 int 2229 nd6_add_ifa_lle(struct in6_ifaddr *ia) 2230 { 2231 struct ifnet *ifp; 2232 struct llentry *ln; 2233 2234 ifp = ia->ia_ifa.ifa_ifp; 2235 if (nd6_need_cache(ifp) == 0) 2236 return (0); 2237 IF_AFDATA_LOCK(ifp); 2238 ia->ia_ifa.ifa_rtrequest = nd6_rtrequest; 2239 ln = lla_lookup(LLTABLE6(ifp), (LLE_CREATE | LLE_IFADDR | 2240 LLE_EXCLUSIVE), (struct sockaddr *)&ia->ia_addr); 2241 IF_AFDATA_UNLOCK(ifp); 2242 if (ln != NULL) { 2243 ln->la_expire = 0; /* for IPv6 this means permanent */ 2244 ln->ln_state = ND6_LLINFO_REACHABLE; 2245 LLE_WUNLOCK(ln); 2246 in6_newaddrmsg(ia, RTM_ADD); 2247 return (0); 2248 } 2249 2250 return (ENOBUFS); 2251 } 2252 2253 /* 2254 * Removes ALL lle records for interface address prefix. 2255 * XXXME: That's probably not we really want to do, we need 2256 * to remove address record only and keep other records 2257 * until we determine if given prefix is really going 2258 * to be removed. 2259 */ 2260 void 2261 nd6_rem_ifa_lle(struct in6_ifaddr *ia) 2262 { 2263 struct sockaddr_in6 mask, addr; 2264 struct ifnet *ifp; 2265 2266 in6_newaddrmsg(ia, RTM_DELETE); 2267 2268 ifp = ia->ia_ifa.ifa_ifp; 2269 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr)); 2270 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask)); 2271 lltable_prefix_free(AF_INET6, (struct sockaddr *)&addr, 2272 (struct sockaddr *)&mask, LLE_STATIC); 2273 } 2274 2275 /* 2276 * the callers of this function need to be re-worked to drop 2277 * the lle lock, drop here for now 2278 */ 2279 int 2280 nd6_storelladdr(struct ifnet *ifp, struct mbuf *m, 2281 const struct sockaddr *dst, u_char *desten, struct llentry **lle) 2282 { 2283 struct llentry *ln; 2284 2285 *lle = NULL; 2286 IF_AFDATA_UNLOCK_ASSERT(ifp); 2287 if (m != NULL && m->m_flags & M_MCAST) { 2288 int i; 2289 2290 switch (ifp->if_type) { 2291 case IFT_ETHER: 2292 case IFT_FDDI: 2293 #ifdef IFT_L2VLAN 2294 case IFT_L2VLAN: 2295 #endif 2296 #ifdef IFT_IEEE80211 2297 case IFT_IEEE80211: 2298 #endif 2299 case IFT_BRIDGE: 2300 case IFT_ISO88025: 2301 ETHER_MAP_IPV6_MULTICAST(&SIN6(dst)->sin6_addr, 2302 desten); 2303 return (0); 2304 case IFT_IEEE1394: 2305 /* 2306 * netbsd can use if_broadcastaddr, but we don't do so 2307 * to reduce # of ifdef. 2308 */ 2309 for (i = 0; i < ifp->if_addrlen; i++) 2310 desten[i] = ~0; 2311 return (0); 2312 case IFT_ARCNET: 2313 *desten = 0; 2314 return (0); 2315 default: 2316 m_freem(m); 2317 return (EAFNOSUPPORT); 2318 } 2319 } 2320 2321 2322 /* 2323 * the entry should have been created in nd6_store_lladdr 2324 */ 2325 IF_AFDATA_RLOCK(ifp); 2326 ln = lla_lookup(LLTABLE6(ifp), 0, dst); 2327 IF_AFDATA_RUNLOCK(ifp); 2328 if ((ln == NULL) || !(ln->la_flags & LLE_VALID)) { 2329 if (ln != NULL) 2330 LLE_RUNLOCK(ln); 2331 /* this could happen, if we could not allocate memory */ 2332 m_freem(m); 2333 return (1); 2334 } 2335 2336 bcopy(&ln->ll_addr, desten, ifp->if_addrlen); 2337 *lle = ln; 2338 LLE_RUNLOCK(ln); 2339 /* 2340 * A *small* use after free race exists here 2341 */ 2342 return (0); 2343 } 2344 2345 static void 2346 clear_llinfo_pqueue(struct llentry *ln) 2347 { 2348 struct mbuf *m_hold, *m_hold_next; 2349 2350 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) { 2351 m_hold_next = m_hold->m_nextpkt; 2352 m_freem(m_hold); 2353 } 2354 2355 ln->la_hold = NULL; 2356 return; 2357 } 2358 2359 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2360 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2361 #ifdef SYSCTL_DECL 2362 SYSCTL_DECL(_net_inet6_icmp6); 2363 #endif 2364 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2365 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2366 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2367 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2368 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2369 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, ""); 2370 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer, 2371 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), ""); 2372 2373 static int 2374 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2375 { 2376 struct in6_defrouter d; 2377 struct nd_defrouter *dr; 2378 int error; 2379 2380 if (req->newptr) 2381 return (EPERM); 2382 2383 bzero(&d, sizeof(d)); 2384 d.rtaddr.sin6_family = AF_INET6; 2385 d.rtaddr.sin6_len = sizeof(d.rtaddr); 2386 2387 /* 2388 * XXX locking 2389 */ 2390 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 2391 d.rtaddr.sin6_addr = dr->rtaddr; 2392 error = sa6_recoverscope(&d.rtaddr); 2393 if (error != 0) 2394 return (error); 2395 d.flags = dr->flags; 2396 d.rtlifetime = dr->rtlifetime; 2397 d.expire = dr->expire + (time_second - time_uptime); 2398 d.if_index = dr->ifp->if_index; 2399 error = SYSCTL_OUT(req, &d, sizeof(d)); 2400 if (error != 0) 2401 return (error); 2402 } 2403 return (0); 2404 } 2405 2406 static int 2407 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2408 { 2409 struct in6_prefix p; 2410 struct sockaddr_in6 s6; 2411 struct nd_prefix *pr; 2412 struct nd_pfxrouter *pfr; 2413 time_t maxexpire; 2414 int error; 2415 char ip6buf[INET6_ADDRSTRLEN]; 2416 2417 if (req->newptr) 2418 return (EPERM); 2419 2420 bzero(&p, sizeof(p)); 2421 p.origin = PR_ORIG_RA; 2422 bzero(&s6, sizeof(s6)); 2423 s6.sin6_family = AF_INET6; 2424 s6.sin6_len = sizeof(s6); 2425 2426 /* 2427 * XXX locking 2428 */ 2429 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 2430 p.prefix = pr->ndpr_prefix; 2431 if (sa6_recoverscope(&p.prefix)) { 2432 log(LOG_ERR, "scope error in prefix list (%s)\n", 2433 ip6_sprintf(ip6buf, &p.prefix.sin6_addr)); 2434 /* XXX: press on... */ 2435 } 2436 p.raflags = pr->ndpr_raf; 2437 p.prefixlen = pr->ndpr_plen; 2438 p.vltime = pr->ndpr_vltime; 2439 p.pltime = pr->ndpr_pltime; 2440 p.if_index = pr->ndpr_ifp->if_index; 2441 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2442 p.expire = 0; 2443 else { 2444 /* XXX: we assume time_t is signed. */ 2445 maxexpire = (-1) & 2446 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1)); 2447 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate) 2448 p.expire = pr->ndpr_lastupdate + 2449 pr->ndpr_vltime + 2450 (time_second - time_uptime); 2451 else 2452 p.expire = maxexpire; 2453 } 2454 p.refcnt = pr->ndpr_refcnt; 2455 p.flags = pr->ndpr_stateflags; 2456 p.advrtrs = 0; 2457 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) 2458 p.advrtrs++; 2459 error = SYSCTL_OUT(req, &p, sizeof(p)); 2460 if (error != 0) 2461 return (error); 2462 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 2463 s6.sin6_addr = pfr->router->rtaddr; 2464 if (sa6_recoverscope(&s6)) 2465 log(LOG_ERR, 2466 "scope error in prefix list (%s)\n", 2467 ip6_sprintf(ip6buf, &pfr->router->rtaddr)); 2468 error = SYSCTL_OUT(req, &s6, sizeof(s6)); 2469 if (error != 0) 2470 return (error); 2471 } 2472 } 2473 return (0); 2474 } 2475