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 #include <netinet/if_ether.h> 70 #include <netinet6/in6_var.h> 71 #include <netinet/ip6.h> 72 #include <netinet6/ip6_var.h> 73 #include <netinet6/scope6_var.h> 74 #include <netinet6/nd6.h> 75 #include <netinet6/in6_ifattach.h> 76 #include <netinet/icmp6.h> 77 #include <netinet6/send.h> 78 79 #include <sys/limits.h> 80 81 #include <security/mac/mac_framework.h> 82 83 #define ND6_SLOWTIMER_INTERVAL (60 * 60) /* 1 hour */ 84 #define ND6_RECALC_REACHTM_INTERVAL (60 * 120) /* 2 hours */ 85 86 #define SIN6(s) ((const struct sockaddr_in6 *)(s)) 87 88 /* timer values */ 89 VNET_DEFINE(int, nd6_prune) = 1; /* walk list every 1 seconds */ 90 VNET_DEFINE(int, nd6_delay) = 5; /* delay first probe time 5 second */ 91 VNET_DEFINE(int, nd6_umaxtries) = 3; /* maximum unicast query */ 92 VNET_DEFINE(int, nd6_mmaxtries) = 3; /* maximum multicast query */ 93 VNET_DEFINE(int, nd6_useloopback) = 1; /* use loopback interface for 94 * local traffic */ 95 VNET_DEFINE(int, nd6_gctimer) = (60 * 60 * 24); /* 1 day: garbage 96 * collection timer */ 97 98 /* preventing too many loops in ND option parsing */ 99 static VNET_DEFINE(int, nd6_maxndopt) = 10; /* max # of ND options allowed */ 100 101 VNET_DEFINE(int, nd6_maxnudhint) = 0; /* max # of subsequent upper 102 * layer hints */ 103 static VNET_DEFINE(int, nd6_maxqueuelen) = 1; /* max pkts cached in unresolved 104 * ND entries */ 105 #define V_nd6_maxndopt VNET(nd6_maxndopt) 106 #define V_nd6_maxqueuelen VNET(nd6_maxqueuelen) 107 108 #ifdef ND6_DEBUG 109 VNET_DEFINE(int, nd6_debug) = 1; 110 #else 111 VNET_DEFINE(int, nd6_debug) = 0; 112 #endif 113 114 static eventhandler_tag lle_event_eh; 115 116 /* for debugging? */ 117 #if 0 118 static int nd6_inuse, nd6_allocated; 119 #endif 120 121 VNET_DEFINE(struct nd_drhead, nd_defrouter); 122 VNET_DEFINE(struct nd_prhead, nd_prefix); 123 124 VNET_DEFINE(int, nd6_recalc_reachtm_interval) = ND6_RECALC_REACHTM_INTERVAL; 125 #define V_nd6_recalc_reachtm_interval VNET(nd6_recalc_reachtm_interval) 126 127 int (*send_sendso_input_hook)(struct mbuf *, struct ifnet *, int, int); 128 129 static int nd6_is_new_addr_neighbor(const struct sockaddr_in6 *, 130 struct ifnet *); 131 static void nd6_setmtu0(struct ifnet *, struct nd_ifinfo *); 132 static void nd6_slowtimo(void *); 133 static int regen_tmpaddr(struct in6_ifaddr *); 134 static void nd6_free(struct llentry *, int); 135 static void nd6_free_redirect(const struct llentry *); 136 static void nd6_llinfo_timer(void *); 137 static void nd6_llinfo_settimer_locked(struct llentry *, long); 138 static void clear_llinfo_pqueue(struct llentry *); 139 static void nd6_rtrequest(int, struct rtentry *, struct rt_addrinfo *); 140 static int nd6_resolve_slow(struct ifnet *, struct mbuf *, 141 const struct sockaddr_in6 *, u_char *, uint32_t *); 142 static int nd6_need_cache(struct ifnet *); 143 144 145 static VNET_DEFINE(struct callout, nd6_slowtimo_ch); 146 #define V_nd6_slowtimo_ch VNET(nd6_slowtimo_ch) 147 148 VNET_DEFINE(struct callout, nd6_timer_ch); 149 150 static void 151 nd6_lle_event(void *arg __unused, struct llentry *lle, int evt) 152 { 153 struct rt_addrinfo rtinfo; 154 struct sockaddr_in6 dst; 155 struct sockaddr_dl gw; 156 struct ifnet *ifp; 157 int type; 158 159 LLE_WLOCK_ASSERT(lle); 160 161 if (lltable_get_af(lle->lle_tbl) != AF_INET6) 162 return; 163 164 switch (evt) { 165 case LLENTRY_RESOLVED: 166 type = RTM_ADD; 167 KASSERT(lle->la_flags & LLE_VALID, 168 ("%s: %p resolved but not valid?", __func__, lle)); 169 break; 170 case LLENTRY_EXPIRED: 171 type = RTM_DELETE; 172 break; 173 default: 174 return; 175 } 176 177 ifp = lltable_get_ifp(lle->lle_tbl); 178 179 bzero(&dst, sizeof(dst)); 180 bzero(&gw, sizeof(gw)); 181 bzero(&rtinfo, sizeof(rtinfo)); 182 lltable_fill_sa_entry(lle, (struct sockaddr *)&dst); 183 dst.sin6_scope_id = in6_getscopezone(ifp, 184 in6_addrscope(&dst.sin6_addr)); 185 gw.sdl_len = sizeof(struct sockaddr_dl); 186 gw.sdl_family = AF_LINK; 187 gw.sdl_alen = ifp->if_addrlen; 188 gw.sdl_index = ifp->if_index; 189 gw.sdl_type = ifp->if_type; 190 if (evt == LLENTRY_RESOLVED) 191 bcopy(&lle->ll_addr, gw.sdl_data, ifp->if_addrlen); 192 rtinfo.rti_info[RTAX_DST] = (struct sockaddr *)&dst; 193 rtinfo.rti_info[RTAX_GATEWAY] = (struct sockaddr *)&gw; 194 rtinfo.rti_addrs = RTA_DST | RTA_GATEWAY; 195 rt_missmsg_fib(type, &rtinfo, RTF_HOST | RTF_LLDATA | ( 196 type == RTM_ADD ? RTF_UP: 0), 0, RT_DEFAULT_FIB); 197 } 198 199 void 200 nd6_init(void) 201 { 202 203 LIST_INIT(&V_nd_prefix); 204 205 /* initialization of the default router list */ 206 TAILQ_INIT(&V_nd_defrouter); 207 208 /* start timer */ 209 callout_init(&V_nd6_slowtimo_ch, 0); 210 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 211 nd6_slowtimo, curvnet); 212 213 nd6_dad_init(); 214 if (IS_DEFAULT_VNET(curvnet)) 215 lle_event_eh = EVENTHANDLER_REGISTER(lle_event, nd6_lle_event, 216 NULL, EVENTHANDLER_PRI_ANY); 217 } 218 219 #ifdef VIMAGE 220 void 221 nd6_destroy() 222 { 223 224 callout_drain(&V_nd6_slowtimo_ch); 225 callout_drain(&V_nd6_timer_ch); 226 if (IS_DEFAULT_VNET(curvnet)) 227 EVENTHANDLER_DEREGISTER(lle_event, lle_event_eh); 228 } 229 #endif 230 231 struct nd_ifinfo * 232 nd6_ifattach(struct ifnet *ifp) 233 { 234 struct nd_ifinfo *nd; 235 236 nd = (struct nd_ifinfo *)malloc(sizeof(*nd), M_IP6NDP, M_WAITOK|M_ZERO); 237 nd->initialized = 1; 238 239 nd->chlim = IPV6_DEFHLIM; 240 nd->basereachable = REACHABLE_TIME; 241 nd->reachable = ND_COMPUTE_RTIME(nd->basereachable); 242 nd->retrans = RETRANS_TIMER; 243 244 nd->flags = ND6_IFF_PERFORMNUD; 245 246 /* A loopback interface always has ND6_IFF_AUTO_LINKLOCAL. 247 * XXXHRS: Clear ND6_IFF_AUTO_LINKLOCAL on an IFT_BRIDGE interface by 248 * default regardless of the V_ip6_auto_linklocal configuration to 249 * give a reasonable default behavior. 250 */ 251 if ((V_ip6_auto_linklocal && ifp->if_type != IFT_BRIDGE) || 252 (ifp->if_flags & IFF_LOOPBACK)) 253 nd->flags |= ND6_IFF_AUTO_LINKLOCAL; 254 /* 255 * A loopback interface does not need to accept RTADV. 256 * XXXHRS: Clear ND6_IFF_ACCEPT_RTADV on an IFT_BRIDGE interface by 257 * default regardless of the V_ip6_accept_rtadv configuration to 258 * prevent the interface from accepting RA messages arrived 259 * on one of the member interfaces with ND6_IFF_ACCEPT_RTADV. 260 */ 261 if (V_ip6_accept_rtadv && 262 !(ifp->if_flags & IFF_LOOPBACK) && 263 (ifp->if_type != IFT_BRIDGE)) 264 nd->flags |= ND6_IFF_ACCEPT_RTADV; 265 if (V_ip6_no_radr && !(ifp->if_flags & IFF_LOOPBACK)) 266 nd->flags |= ND6_IFF_NO_RADR; 267 268 /* XXX: we cannot call nd6_setmtu since ifp is not fully initialized */ 269 nd6_setmtu0(ifp, nd); 270 271 return nd; 272 } 273 274 void 275 nd6_ifdetach(struct nd_ifinfo *nd) 276 { 277 278 free(nd, M_IP6NDP); 279 } 280 281 /* 282 * Reset ND level link MTU. This function is called when the physical MTU 283 * changes, which means we might have to adjust the ND level MTU. 284 */ 285 void 286 nd6_setmtu(struct ifnet *ifp) 287 { 288 289 nd6_setmtu0(ifp, ND_IFINFO(ifp)); 290 } 291 292 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */ 293 void 294 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi) 295 { 296 u_int32_t omaxmtu; 297 298 omaxmtu = ndi->maxmtu; 299 300 switch (ifp->if_type) { 301 case IFT_ARCNET: 302 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */ 303 break; 304 case IFT_FDDI: 305 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */ 306 break; 307 case IFT_ISO88025: 308 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu); 309 break; 310 default: 311 ndi->maxmtu = ifp->if_mtu; 312 break; 313 } 314 315 /* 316 * Decreasing the interface MTU under IPV6 minimum MTU may cause 317 * undesirable situation. We thus notify the operator of the change 318 * explicitly. The check for omaxmtu is necessary to restrict the 319 * log to the case of changing the MTU, not initializing it. 320 */ 321 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { 322 log(LOG_NOTICE, "nd6_setmtu0: " 323 "new link MTU on %s (%lu) is too small for IPv6\n", 324 if_name(ifp), (unsigned long)ndi->maxmtu); 325 } 326 327 if (ndi->maxmtu > V_in6_maxmtu) 328 in6_setmaxmtu(); /* check all interfaces just in case */ 329 330 } 331 332 void 333 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts) 334 { 335 336 bzero(ndopts, sizeof(*ndopts)); 337 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; 338 ndopts->nd_opts_last 339 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); 340 341 if (icmp6len == 0) { 342 ndopts->nd_opts_done = 1; 343 ndopts->nd_opts_search = NULL; 344 } 345 } 346 347 /* 348 * Take one ND option. 349 */ 350 struct nd_opt_hdr * 351 nd6_option(union nd_opts *ndopts) 352 { 353 struct nd_opt_hdr *nd_opt; 354 int olen; 355 356 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 357 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 358 __func__)); 359 if (ndopts->nd_opts_search == NULL) 360 return NULL; 361 if (ndopts->nd_opts_done) 362 return NULL; 363 364 nd_opt = ndopts->nd_opts_search; 365 366 /* make sure nd_opt_len is inside the buffer */ 367 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) { 368 bzero(ndopts, sizeof(*ndopts)); 369 return NULL; 370 } 371 372 olen = nd_opt->nd_opt_len << 3; 373 if (olen == 0) { 374 /* 375 * Message validation requires that all included 376 * options have a length that is greater than zero. 377 */ 378 bzero(ndopts, sizeof(*ndopts)); 379 return NULL; 380 } 381 382 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen); 383 if (ndopts->nd_opts_search > ndopts->nd_opts_last) { 384 /* option overruns the end of buffer, invalid */ 385 bzero(ndopts, sizeof(*ndopts)); 386 return NULL; 387 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { 388 /* reached the end of options chain */ 389 ndopts->nd_opts_done = 1; 390 ndopts->nd_opts_search = NULL; 391 } 392 return nd_opt; 393 } 394 395 /* 396 * Parse multiple ND options. 397 * This function is much easier to use, for ND routines that do not need 398 * multiple options of the same type. 399 */ 400 int 401 nd6_options(union nd_opts *ndopts) 402 { 403 struct nd_opt_hdr *nd_opt; 404 int i = 0; 405 406 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 407 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 408 __func__)); 409 if (ndopts->nd_opts_search == NULL) 410 return 0; 411 412 while (1) { 413 nd_opt = nd6_option(ndopts); 414 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { 415 /* 416 * Message validation requires that all included 417 * options have a length that is greater than zero. 418 */ 419 ICMP6STAT_INC(icp6s_nd_badopt); 420 bzero(ndopts, sizeof(*ndopts)); 421 return -1; 422 } 423 424 if (nd_opt == NULL) 425 goto skip1; 426 427 switch (nd_opt->nd_opt_type) { 428 case ND_OPT_SOURCE_LINKADDR: 429 case ND_OPT_TARGET_LINKADDR: 430 case ND_OPT_MTU: 431 case ND_OPT_REDIRECTED_HEADER: 432 case ND_OPT_NONCE: 433 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { 434 nd6log((LOG_INFO, 435 "duplicated ND6 option found (type=%d)\n", 436 nd_opt->nd_opt_type)); 437 /* XXX bark? */ 438 } else { 439 ndopts->nd_opt_array[nd_opt->nd_opt_type] 440 = nd_opt; 441 } 442 break; 443 case ND_OPT_PREFIX_INFORMATION: 444 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { 445 ndopts->nd_opt_array[nd_opt->nd_opt_type] 446 = nd_opt; 447 } 448 ndopts->nd_opts_pi_end = 449 (struct nd_opt_prefix_info *)nd_opt; 450 break; 451 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */ 452 case ND_OPT_RDNSS: /* RFC 6106 */ 453 case ND_OPT_DNSSL: /* RFC 6106 */ 454 /* 455 * Silently ignore options we know and do not care about 456 * in the kernel. 457 */ 458 break; 459 default: 460 /* 461 * Unknown options must be silently ignored, 462 * to accomodate future extension to the protocol. 463 */ 464 nd6log((LOG_DEBUG, 465 "nd6_options: unsupported option %d - " 466 "option ignored\n", nd_opt->nd_opt_type)); 467 } 468 469 skip1: 470 i++; 471 if (i > V_nd6_maxndopt) { 472 ICMP6STAT_INC(icp6s_nd_toomanyopt); 473 nd6log((LOG_INFO, "too many loop in nd opt\n")); 474 break; 475 } 476 477 if (ndopts->nd_opts_done) 478 break; 479 } 480 481 return 0; 482 } 483 484 /* 485 * ND6 timer routine to handle ND6 entries 486 */ 487 static void 488 nd6_llinfo_settimer_locked(struct llentry *ln, long tick) 489 { 490 int canceled; 491 492 LLE_WLOCK_ASSERT(ln); 493 494 if (tick < 0) { 495 ln->la_expire = 0; 496 ln->ln_ntick = 0; 497 canceled = callout_stop(&ln->lle_timer); 498 } else { 499 ln->la_expire = time_uptime + tick / hz; 500 LLE_ADDREF(ln); 501 if (tick > INT_MAX) { 502 ln->ln_ntick = tick - INT_MAX; 503 canceled = callout_reset(&ln->lle_timer, INT_MAX, 504 nd6_llinfo_timer, ln); 505 } else { 506 ln->ln_ntick = 0; 507 canceled = callout_reset(&ln->lle_timer, tick, 508 nd6_llinfo_timer, ln); 509 } 510 } 511 if (canceled) 512 LLE_REMREF(ln); 513 } 514 515 /* 516 * Gets source address of the first packet in hold queue 517 * and stores it in @src. 518 * Returns pointer to @src (if hold queue is not empty) or NULL. 519 * 520 * Set noinline to be dtrace-friendly 521 */ 522 static __noinline struct in6_addr * 523 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src) 524 { 525 struct ip6_hdr hdr; 526 struct mbuf *m; 527 528 if (ln->la_hold == NULL) 529 return (NULL); 530 531 /* 532 * assume every packet in la_hold has the same IP header 533 */ 534 m = ln->la_hold; 535 if (sizeof(hdr) > m->m_len) 536 return (NULL); 537 538 m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr); 539 *src = hdr.ip6_src; 540 541 return (src); 542 } 543 544 /* 545 * Switch @lle state to new state optionally arming timers. 546 * 547 * Set noinline to be dtrace-friendly 548 */ 549 __noinline void 550 nd6_llinfo_setstate(struct llentry *lle, int newstate) 551 { 552 struct ifnet *ifp; 553 long delay; 554 555 delay = 0; 556 557 switch (newstate) { 558 case ND6_LLINFO_INCOMPLETE: 559 ifp = lle->lle_tbl->llt_ifp; 560 delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000; 561 break; 562 case ND6_LLINFO_REACHABLE: 563 if (!ND6_LLINFO_PERMANENT(lle)) { 564 ifp = lle->lle_tbl->llt_ifp; 565 delay = (long)ND_IFINFO(ifp)->reachable * hz; 566 } 567 break; 568 case ND6_LLINFO_STALE: 569 delay = (long)V_nd6_gctimer * hz; 570 break; 571 case ND6_LLINFO_DELAY: 572 lle->la_asked = 0; 573 delay = (long)V_nd6_delay * hz; 574 break; 575 } 576 577 if (delay > 0) 578 nd6_llinfo_settimer_locked(lle, delay); 579 580 lle->ln_state = newstate; 581 } 582 583 /* 584 * Timer-dependent part of nd state machine. 585 * 586 * Set noinline to be dtrace-friendly 587 */ 588 static __noinline void 589 nd6_llinfo_timer(void *arg) 590 { 591 struct llentry *ln; 592 struct in6_addr *dst, *pdst, *psrc, src; 593 struct ifnet *ifp; 594 struct nd_ifinfo *ndi = NULL; 595 int send_ns; 596 597 KASSERT(arg != NULL, ("%s: arg NULL", __func__)); 598 ln = (struct llentry *)arg; 599 LLE_WLOCK(ln); 600 if (callout_pending(&ln->lle_timer)) { 601 /* 602 * Here we are a bit odd here in the treatment of 603 * active/pending. If the pending bit is set, it got 604 * rescheduled before I ran. The active 605 * bit we ignore, since if it was stopped 606 * in ll_tablefree() and was currently running 607 * it would have return 0 so the code would 608 * not have deleted it since the callout could 609 * not be stopped so we want to go through 610 * with the delete here now. If the callout 611 * was restarted, the pending bit will be back on and 612 * we just want to bail since the callout_reset would 613 * return 1 and our reference would have been removed 614 * by nd6_llinfo_settimer_locked above since canceled 615 * would have been 1. 616 */ 617 LLE_WUNLOCK(ln); 618 return; 619 } 620 ifp = ln->lle_tbl->llt_ifp; 621 CURVNET_SET(ifp->if_vnet); 622 ndi = ND_IFINFO(ifp); 623 send_ns = 0; 624 dst = &ln->r_l3addr.addr6; 625 pdst = dst; 626 627 if (ln->ln_ntick > 0) { 628 if (ln->ln_ntick > INT_MAX) { 629 ln->ln_ntick -= INT_MAX; 630 nd6_llinfo_settimer_locked(ln, INT_MAX); 631 } else { 632 ln->ln_ntick = 0; 633 nd6_llinfo_settimer_locked(ln, ln->ln_ntick); 634 } 635 goto done; 636 } 637 638 if (ln->la_flags & LLE_STATIC) { 639 goto done; 640 } 641 642 if (ln->la_flags & LLE_DELETED) { 643 nd6_free(ln, 0); 644 ln = NULL; 645 goto done; 646 } 647 648 switch (ln->ln_state) { 649 case ND6_LLINFO_INCOMPLETE: 650 if (ln->la_asked < V_nd6_mmaxtries) { 651 ln->la_asked++; 652 send_ns = 1; 653 /* Send NS to multicast address */ 654 pdst = NULL; 655 } else { 656 struct mbuf *m = ln->la_hold; 657 if (m) { 658 struct mbuf *m0; 659 660 /* 661 * assuming every packet in la_hold has the 662 * same IP header. Send error after unlock. 663 */ 664 m0 = m->m_nextpkt; 665 m->m_nextpkt = NULL; 666 ln->la_hold = m0; 667 clear_llinfo_pqueue(ln); 668 } 669 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT); 670 nd6_free(ln, 0); 671 ln = NULL; 672 if (m != NULL) 673 icmp6_error2(m, ICMP6_DST_UNREACH, 674 ICMP6_DST_UNREACH_ADDR, 0, ifp); 675 } 676 break; 677 case ND6_LLINFO_REACHABLE: 678 if (!ND6_LLINFO_PERMANENT(ln)) 679 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); 680 break; 681 682 case ND6_LLINFO_STALE: 683 /* Garbage Collection(RFC 2461 5.3) */ 684 if (!ND6_LLINFO_PERMANENT(ln)) { 685 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 686 nd6_free(ln, 1); 687 ln = NULL; 688 } 689 break; 690 691 case ND6_LLINFO_DELAY: 692 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 693 /* We need NUD */ 694 ln->la_asked = 1; 695 nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE); 696 send_ns = 1; 697 } else 698 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */ 699 break; 700 case ND6_LLINFO_PROBE: 701 if (ln->la_asked < V_nd6_umaxtries) { 702 ln->la_asked++; 703 send_ns = 1; 704 } else { 705 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 706 nd6_free(ln, 0); 707 ln = NULL; 708 } 709 break; 710 default: 711 panic("%s: paths in a dark night can be confusing: %d", 712 __func__, ln->ln_state); 713 } 714 done: 715 if (send_ns != 0) { 716 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 717 psrc = nd6_llinfo_get_holdsrc(ln, &src); 718 LLE_FREE_LOCKED(ln); 719 ln = NULL; 720 nd6_ns_output(ifp, psrc, pdst, dst, NULL); 721 } 722 723 if (ln != NULL) 724 LLE_FREE_LOCKED(ln); 725 CURVNET_RESTORE(); 726 } 727 728 729 /* 730 * ND6 timer routine to expire default route list and prefix list 731 */ 732 void 733 nd6_timer(void *arg) 734 { 735 CURVNET_SET((struct vnet *) arg); 736 struct nd_defrouter *dr, *ndr; 737 struct nd_prefix *pr, *npr; 738 struct in6_ifaddr *ia6, *nia6; 739 740 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz, 741 nd6_timer, curvnet); 742 743 /* expire default router list */ 744 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 745 if (dr->expire && dr->expire < time_uptime) 746 defrtrlist_del(dr); 747 } 748 749 /* 750 * expire interface addresses. 751 * in the past the loop was inside prefix expiry processing. 752 * However, from a stricter speci-confrmance standpoint, we should 753 * rather separate address lifetimes and prefix lifetimes. 754 * 755 * XXXRW: in6_ifaddrhead locking. 756 */ 757 addrloop: 758 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) { 759 /* check address lifetime */ 760 if (IFA6_IS_INVALID(ia6)) { 761 int regen = 0; 762 763 /* 764 * If the expiring address is temporary, try 765 * regenerating a new one. This would be useful when 766 * we suspended a laptop PC, then turned it on after a 767 * period that could invalidate all temporary 768 * addresses. Although we may have to restart the 769 * loop (see below), it must be after purging the 770 * address. Otherwise, we'd see an infinite loop of 771 * regeneration. 772 */ 773 if (V_ip6_use_tempaddr && 774 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 775 if (regen_tmpaddr(ia6) == 0) 776 regen = 1; 777 } 778 779 in6_purgeaddr(&ia6->ia_ifa); 780 781 if (regen) 782 goto addrloop; /* XXX: see below */ 783 } else if (IFA6_IS_DEPRECATED(ia6)) { 784 int oldflags = ia6->ia6_flags; 785 786 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 787 788 /* 789 * If a temporary address has just become deprecated, 790 * regenerate a new one if possible. 791 */ 792 if (V_ip6_use_tempaddr && 793 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 794 (oldflags & IN6_IFF_DEPRECATED) == 0) { 795 796 if (regen_tmpaddr(ia6) == 0) { 797 /* 798 * A new temporary address is 799 * generated. 800 * XXX: this means the address chain 801 * has changed while we are still in 802 * the loop. Although the change 803 * would not cause disaster (because 804 * it's not a deletion, but an 805 * addition,) we'd rather restart the 806 * loop just for safety. Or does this 807 * significantly reduce performance?? 808 */ 809 goto addrloop; 810 } 811 } 812 } else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) { 813 /* 814 * Schedule DAD for a tentative address. This happens 815 * if the interface was down or not running 816 * when the address was configured. 817 */ 818 int delay; 819 820 delay = arc4random() % 821 (MAX_RTR_SOLICITATION_DELAY * hz); 822 nd6_dad_start((struct ifaddr *)ia6, delay); 823 } else { 824 /* 825 * Check status of the interface. If it is down, 826 * mark the address as tentative for future DAD. 827 */ 828 if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 || 829 (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING) 830 == 0 || 831 (ND_IFINFO(ia6->ia_ifp)->flags & 832 ND6_IFF_IFDISABLED) != 0) { 833 ia6->ia6_flags &= ~IN6_IFF_DUPLICATED; 834 ia6->ia6_flags |= IN6_IFF_TENTATIVE; 835 } 836 /* 837 * A new RA might have made a deprecated address 838 * preferred. 839 */ 840 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 841 } 842 } 843 844 /* expire prefix list */ 845 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 846 /* 847 * check prefix lifetime. 848 * since pltime is just for autoconf, pltime processing for 849 * prefix is not necessary. 850 */ 851 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 852 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) { 853 854 /* 855 * address expiration and prefix expiration are 856 * separate. NEVER perform in6_purgeaddr here. 857 */ 858 prelist_remove(pr); 859 } 860 } 861 CURVNET_RESTORE(); 862 } 863 864 /* 865 * ia6 - deprecated/invalidated temporary address 866 */ 867 static int 868 regen_tmpaddr(struct in6_ifaddr *ia6) 869 { 870 struct ifaddr *ifa; 871 struct ifnet *ifp; 872 struct in6_ifaddr *public_ifa6 = NULL; 873 874 ifp = ia6->ia_ifa.ifa_ifp; 875 IF_ADDR_RLOCK(ifp); 876 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 877 struct in6_ifaddr *it6; 878 879 if (ifa->ifa_addr->sa_family != AF_INET6) 880 continue; 881 882 it6 = (struct in6_ifaddr *)ifa; 883 884 /* ignore no autoconf addresses. */ 885 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 886 continue; 887 888 /* ignore autoconf addresses with different prefixes. */ 889 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 890 continue; 891 892 /* 893 * Now we are looking at an autoconf address with the same 894 * prefix as ours. If the address is temporary and is still 895 * preferred, do not create another one. It would be rare, but 896 * could happen, for example, when we resume a laptop PC after 897 * a long period. 898 */ 899 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 900 !IFA6_IS_DEPRECATED(it6)) { 901 public_ifa6 = NULL; 902 break; 903 } 904 905 /* 906 * This is a public autoconf address that has the same prefix 907 * as ours. If it is preferred, keep it. We can't break the 908 * loop here, because there may be a still-preferred temporary 909 * address with the prefix. 910 */ 911 if (!IFA6_IS_DEPRECATED(it6)) 912 public_ifa6 = it6; 913 } 914 if (public_ifa6 != NULL) 915 ifa_ref(&public_ifa6->ia_ifa); 916 IF_ADDR_RUNLOCK(ifp); 917 918 if (public_ifa6 != NULL) { 919 int e; 920 921 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 922 ifa_free(&public_ifa6->ia_ifa); 923 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 924 " tmp addr,errno=%d\n", e); 925 return (-1); 926 } 927 ifa_free(&public_ifa6->ia_ifa); 928 return (0); 929 } 930 931 return (-1); 932 } 933 934 /* 935 * Nuke neighbor cache/prefix/default router management table, right before 936 * ifp goes away. 937 */ 938 void 939 nd6_purge(struct ifnet *ifp) 940 { 941 struct nd_defrouter *dr, *ndr; 942 struct nd_prefix *pr, *npr; 943 944 /* 945 * Nuke default router list entries toward ifp. 946 * We defer removal of default router list entries that is installed 947 * in the routing table, in order to keep additional side effects as 948 * small as possible. 949 */ 950 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 951 if (dr->installed) 952 continue; 953 954 if (dr->ifp == ifp) 955 defrtrlist_del(dr); 956 } 957 958 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 959 if (!dr->installed) 960 continue; 961 962 if (dr->ifp == ifp) 963 defrtrlist_del(dr); 964 } 965 966 /* Nuke prefix list entries toward ifp */ 967 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 968 if (pr->ndpr_ifp == ifp) { 969 /* 970 * Because if_detach() does *not* release prefixes 971 * while purging addresses the reference count will 972 * still be above zero. We therefore reset it to 973 * make sure that the prefix really gets purged. 974 */ 975 pr->ndpr_refcnt = 0; 976 977 /* 978 * Previously, pr->ndpr_addr is removed as well, 979 * but I strongly believe we don't have to do it. 980 * nd6_purge() is only called from in6_ifdetach(), 981 * which removes all the associated interface addresses 982 * by itself. 983 * (jinmei@kame.net 20010129) 984 */ 985 prelist_remove(pr); 986 } 987 } 988 989 /* cancel default outgoing interface setting */ 990 if (V_nd6_defifindex == ifp->if_index) 991 nd6_setdefaultiface(0); 992 993 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 994 /* Refresh default router list. */ 995 defrouter_select(); 996 } 997 998 /* XXXXX 999 * We do not nuke the neighbor cache entries here any more 1000 * because the neighbor cache is kept in if_afdata[AF_INET6]. 1001 * nd6_purge() is invoked by in6_ifdetach() which is called 1002 * from if_detach() where everything gets purged. So let 1003 * in6_domifdetach() do the actual L2 table purging work. 1004 */ 1005 } 1006 1007 /* 1008 * the caller acquires and releases the lock on the lltbls 1009 * Returns the llentry locked 1010 */ 1011 struct llentry * 1012 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp) 1013 { 1014 struct sockaddr_in6 sin6; 1015 struct llentry *ln; 1016 1017 bzero(&sin6, sizeof(sin6)); 1018 sin6.sin6_len = sizeof(struct sockaddr_in6); 1019 sin6.sin6_family = AF_INET6; 1020 sin6.sin6_addr = *addr6; 1021 1022 IF_AFDATA_LOCK_ASSERT(ifp); 1023 1024 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6); 1025 1026 return (ln); 1027 } 1028 1029 struct llentry * 1030 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp) 1031 { 1032 struct sockaddr_in6 sin6; 1033 struct llentry *ln; 1034 1035 bzero(&sin6, sizeof(sin6)); 1036 sin6.sin6_len = sizeof(struct sockaddr_in6); 1037 sin6.sin6_family = AF_INET6; 1038 sin6.sin6_addr = *addr6; 1039 1040 ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6); 1041 if (ln != NULL) 1042 ln->ln_state = ND6_LLINFO_NOSTATE; 1043 1044 return (ln); 1045 } 1046 1047 /* 1048 * Test whether a given IPv6 address is a neighbor or not, ignoring 1049 * the actual neighbor cache. The neighbor cache is ignored in order 1050 * to not reenter the routing code from within itself. 1051 */ 1052 static int 1053 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp) 1054 { 1055 struct nd_prefix *pr; 1056 struct ifaddr *dstaddr; 1057 1058 /* 1059 * A link-local address is always a neighbor. 1060 * XXX: a link does not necessarily specify a single interface. 1061 */ 1062 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 1063 struct sockaddr_in6 sin6_copy; 1064 u_int32_t zone; 1065 1066 /* 1067 * We need sin6_copy since sa6_recoverscope() may modify the 1068 * content (XXX). 1069 */ 1070 sin6_copy = *addr; 1071 if (sa6_recoverscope(&sin6_copy)) 1072 return (0); /* XXX: should be impossible */ 1073 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 1074 return (0); 1075 if (sin6_copy.sin6_scope_id == zone) 1076 return (1); 1077 else 1078 return (0); 1079 } 1080 1081 /* 1082 * If the address matches one of our addresses, 1083 * it should be a neighbor. 1084 * If the address matches one of our on-link prefixes, it should be a 1085 * neighbor. 1086 */ 1087 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 1088 if (pr->ndpr_ifp != ifp) 1089 continue; 1090 1091 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) { 1092 struct rtentry *rt; 1093 1094 /* Always use the default FIB here. */ 1095 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 1096 0, 0, RT_DEFAULT_FIB); 1097 if (rt == NULL) 1098 continue; 1099 /* 1100 * This is the case where multiple interfaces 1101 * have the same prefix, but only one is installed 1102 * into the routing table and that prefix entry 1103 * is not the one being examined here. In the case 1104 * where RADIX_MPATH is enabled, multiple route 1105 * entries (of the same rt_key value) will be 1106 * installed because the interface addresses all 1107 * differ. 1108 */ 1109 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 1110 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) { 1111 RTFREE_LOCKED(rt); 1112 continue; 1113 } 1114 RTFREE_LOCKED(rt); 1115 } 1116 1117 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 1118 &addr->sin6_addr, &pr->ndpr_mask)) 1119 return (1); 1120 } 1121 1122 /* 1123 * If the address is assigned on the node of the other side of 1124 * a p2p interface, the address should be a neighbor. 1125 */ 1126 dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS); 1127 if (dstaddr != NULL) { 1128 if (dstaddr->ifa_ifp == ifp) { 1129 ifa_free(dstaddr); 1130 return (1); 1131 } 1132 ifa_free(dstaddr); 1133 } 1134 1135 /* 1136 * If the default router list is empty, all addresses are regarded 1137 * as on-link, and thus, as a neighbor. 1138 */ 1139 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && 1140 TAILQ_EMPTY(&V_nd_defrouter) && 1141 V_nd6_defifindex == ifp->if_index) { 1142 return (1); 1143 } 1144 1145 return (0); 1146 } 1147 1148 1149 /* 1150 * Detect if a given IPv6 address identifies a neighbor on a given link. 1151 * XXX: should take care of the destination of a p2p link? 1152 */ 1153 int 1154 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp) 1155 { 1156 struct llentry *lle; 1157 int rc = 0; 1158 1159 IF_AFDATA_UNLOCK_ASSERT(ifp); 1160 if (nd6_is_new_addr_neighbor(addr, ifp)) 1161 return (1); 1162 1163 /* 1164 * Even if the address matches none of our addresses, it might be 1165 * in the neighbor cache. 1166 */ 1167 IF_AFDATA_RLOCK(ifp); 1168 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) { 1169 LLE_RUNLOCK(lle); 1170 rc = 1; 1171 } 1172 IF_AFDATA_RUNLOCK(ifp); 1173 return (rc); 1174 } 1175 1176 /* 1177 * Free an nd6 llinfo entry. 1178 * Since the function would cause significant changes in the kernel, DO NOT 1179 * make it global, unless you have a strong reason for the change, and are sure 1180 * that the change is safe. 1181 * 1182 * Set noinline to be dtrace-friendly 1183 */ 1184 static __noinline void 1185 nd6_free(struct llentry *ln, int gc) 1186 { 1187 struct nd_defrouter *dr; 1188 struct ifnet *ifp; 1189 1190 LLE_WLOCK_ASSERT(ln); 1191 1192 /* 1193 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1194 * even though it is not harmful, it was not really necessary. 1195 */ 1196 1197 /* cancel timer */ 1198 nd6_llinfo_settimer_locked(ln, -1); 1199 1200 ifp = ln->lle_tbl->llt_ifp; 1201 1202 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1203 dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp); 1204 1205 if (dr != NULL && dr->expire && 1206 ln->ln_state == ND6_LLINFO_STALE && gc) { 1207 /* 1208 * If the reason for the deletion is just garbage 1209 * collection, and the neighbor is an active default 1210 * router, do not delete it. Instead, reset the GC 1211 * timer using the router's lifetime. 1212 * Simply deleting the entry would affect default 1213 * router selection, which is not necessarily a good 1214 * thing, especially when we're using router preference 1215 * values. 1216 * XXX: the check for ln_state would be redundant, 1217 * but we intentionally keep it just in case. 1218 */ 1219 if (dr->expire > time_uptime) 1220 nd6_llinfo_settimer_locked(ln, 1221 (dr->expire - time_uptime) * hz); 1222 else 1223 nd6_llinfo_settimer_locked(ln, 1224 (long)V_nd6_gctimer * hz); 1225 1226 LLE_REMREF(ln); 1227 LLE_WUNLOCK(ln); 1228 return; 1229 } 1230 1231 if (dr) { 1232 /* 1233 * Unreachablity of a router might affect the default 1234 * router selection and on-link detection of advertised 1235 * prefixes. 1236 */ 1237 1238 /* 1239 * Temporarily fake the state to choose a new default 1240 * router and to perform on-link determination of 1241 * prefixes correctly. 1242 * Below the state will be set correctly, 1243 * or the entry itself will be deleted. 1244 */ 1245 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1246 } 1247 1248 if (ln->ln_router || dr) { 1249 1250 /* 1251 * We need to unlock to avoid a LOR with rt6_flush() with the 1252 * rnh and for the calls to pfxlist_onlink_check() and 1253 * defrouter_select() in the block further down for calls 1254 * into nd6_lookup(). We still hold a ref. 1255 */ 1256 LLE_WUNLOCK(ln); 1257 1258 /* 1259 * rt6_flush must be called whether or not the neighbor 1260 * is in the Default Router List. 1261 * See a corresponding comment in nd6_na_input(). 1262 */ 1263 rt6_flush(&ln->r_l3addr.addr6, ifp); 1264 } 1265 1266 if (dr) { 1267 /* 1268 * Since defrouter_select() does not affect the 1269 * on-link determination and MIP6 needs the check 1270 * before the default router selection, we perform 1271 * the check now. 1272 */ 1273 pfxlist_onlink_check(); 1274 1275 /* 1276 * Refresh default router list. 1277 */ 1278 defrouter_select(); 1279 } 1280 1281 /* 1282 * If this entry was added by an on-link redirect, remove the 1283 * corresponding host route. 1284 */ 1285 if (ln->la_flags & LLE_REDIRECT) 1286 nd6_free_redirect(ln); 1287 1288 if (ln->ln_router || dr) 1289 LLE_WLOCK(ln); 1290 } 1291 1292 /* 1293 * Save to unlock. We still hold an extra reference and will not 1294 * free(9) in llentry_free() if someone else holds one as well. 1295 */ 1296 LLE_WUNLOCK(ln); 1297 IF_AFDATA_LOCK(ifp); 1298 LLE_WLOCK(ln); 1299 /* Guard against race with other llentry_free(). */ 1300 if (ln->la_flags & LLE_LINKED) { 1301 /* Remove callout reference */ 1302 LLE_REMREF(ln); 1303 lltable_unlink_entry(ln->lle_tbl, ln); 1304 } 1305 IF_AFDATA_UNLOCK(ifp); 1306 1307 llentry_free(ln); 1308 } 1309 1310 /* 1311 * Remove the rtentry for the given llentry, 1312 * both of which were installed by a redirect. 1313 */ 1314 static void 1315 nd6_free_redirect(const struct llentry *ln) 1316 { 1317 int fibnum; 1318 struct rtentry *rt; 1319 struct radix_node_head *rnh; 1320 struct sockaddr_in6 sin6; 1321 1322 lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6); 1323 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) { 1324 rnh = rt_tables_get_rnh(fibnum, AF_INET6); 1325 if (rnh == NULL) 1326 continue; 1327 1328 RADIX_NODE_HEAD_LOCK(rnh); 1329 rt = in6_rtalloc1((struct sockaddr *)&sin6, 0, 1330 RTF_RNH_LOCKED, fibnum); 1331 if (rt) { 1332 if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC)) 1333 rt_expunge(rnh, rt); 1334 RTFREE_LOCKED(rt); 1335 } 1336 RADIX_NODE_HEAD_UNLOCK(rnh); 1337 } 1338 } 1339 1340 /* 1341 * Rejuvenate this function for routing operations related 1342 * processing. 1343 */ 1344 void 1345 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info) 1346 { 1347 struct sockaddr_in6 *gateway; 1348 struct nd_defrouter *dr; 1349 struct ifnet *ifp; 1350 1351 gateway = (struct sockaddr_in6 *)rt->rt_gateway; 1352 ifp = rt->rt_ifp; 1353 1354 switch (req) { 1355 case RTM_ADD: 1356 break; 1357 1358 case RTM_DELETE: 1359 if (!ifp) 1360 return; 1361 /* 1362 * Only indirect routes are interesting. 1363 */ 1364 if ((rt->rt_flags & RTF_GATEWAY) == 0) 1365 return; 1366 /* 1367 * check for default route 1368 */ 1369 if (IN6_ARE_ADDR_EQUAL(&in6addr_any, 1370 &SIN6(rt_key(rt))->sin6_addr)) { 1371 1372 dr = defrouter_lookup(&gateway->sin6_addr, ifp); 1373 if (dr != NULL) 1374 dr->installed = 0; 1375 } 1376 break; 1377 } 1378 } 1379 1380 1381 int 1382 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) 1383 { 1384 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1385 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1386 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1387 int error = 0; 1388 1389 if (ifp->if_afdata[AF_INET6] == NULL) 1390 return (EPFNOSUPPORT); 1391 switch (cmd) { 1392 case OSIOCGIFINFO_IN6: 1393 #define ND ndi->ndi 1394 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1395 bzero(&ND, sizeof(ND)); 1396 ND.linkmtu = IN6_LINKMTU(ifp); 1397 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1398 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1399 ND.reachable = ND_IFINFO(ifp)->reachable; 1400 ND.retrans = ND_IFINFO(ifp)->retrans; 1401 ND.flags = ND_IFINFO(ifp)->flags; 1402 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1403 ND.chlim = ND_IFINFO(ifp)->chlim; 1404 break; 1405 case SIOCGIFINFO_IN6: 1406 ND = *ND_IFINFO(ifp); 1407 break; 1408 case SIOCSIFINFO_IN6: 1409 /* 1410 * used to change host variables from userland. 1411 * intented for a use on router to reflect RA configurations. 1412 */ 1413 /* 0 means 'unspecified' */ 1414 if (ND.linkmtu != 0) { 1415 if (ND.linkmtu < IPV6_MMTU || 1416 ND.linkmtu > IN6_LINKMTU(ifp)) { 1417 error = EINVAL; 1418 break; 1419 } 1420 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1421 } 1422 1423 if (ND.basereachable != 0) { 1424 int obasereachable = ND_IFINFO(ifp)->basereachable; 1425 1426 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1427 if (ND.basereachable != obasereachable) 1428 ND_IFINFO(ifp)->reachable = 1429 ND_COMPUTE_RTIME(ND.basereachable); 1430 } 1431 if (ND.retrans != 0) 1432 ND_IFINFO(ifp)->retrans = ND.retrans; 1433 if (ND.chlim != 0) 1434 ND_IFINFO(ifp)->chlim = ND.chlim; 1435 /* FALLTHROUGH */ 1436 case SIOCSIFINFO_FLAGS: 1437 { 1438 struct ifaddr *ifa; 1439 struct in6_ifaddr *ia; 1440 1441 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1442 !(ND.flags & ND6_IFF_IFDISABLED)) { 1443 /* ifdisabled 1->0 transision */ 1444 1445 /* 1446 * If the interface is marked as ND6_IFF_IFDISABLED and 1447 * has an link-local address with IN6_IFF_DUPLICATED, 1448 * do not clear ND6_IFF_IFDISABLED. 1449 * See RFC 4862, Section 5.4.5. 1450 */ 1451 IF_ADDR_RLOCK(ifp); 1452 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1453 if (ifa->ifa_addr->sa_family != AF_INET6) 1454 continue; 1455 ia = (struct in6_ifaddr *)ifa; 1456 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) && 1457 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) 1458 break; 1459 } 1460 IF_ADDR_RUNLOCK(ifp); 1461 1462 if (ifa != NULL) { 1463 /* LLA is duplicated. */ 1464 ND.flags |= ND6_IFF_IFDISABLED; 1465 log(LOG_ERR, "Cannot enable an interface" 1466 " with a link-local address marked" 1467 " duplicate.\n"); 1468 } else { 1469 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED; 1470 if (ifp->if_flags & IFF_UP) 1471 in6_if_up(ifp); 1472 } 1473 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1474 (ND.flags & ND6_IFF_IFDISABLED)) { 1475 /* ifdisabled 0->1 transision */ 1476 /* Mark all IPv6 address as tentative. */ 1477 1478 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; 1479 if (V_ip6_dad_count > 0 && 1480 (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) { 1481 IF_ADDR_RLOCK(ifp); 1482 TAILQ_FOREACH(ifa, &ifp->if_addrhead, 1483 ifa_link) { 1484 if (ifa->ifa_addr->sa_family != 1485 AF_INET6) 1486 continue; 1487 ia = (struct in6_ifaddr *)ifa; 1488 ia->ia6_flags |= IN6_IFF_TENTATIVE; 1489 } 1490 IF_ADDR_RUNLOCK(ifp); 1491 } 1492 } 1493 1494 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) { 1495 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) { 1496 /* auto_linklocal 0->1 transision */ 1497 1498 /* If no link-local address on ifp, configure */ 1499 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL; 1500 in6_ifattach(ifp, NULL); 1501 } else if (!(ND.flags & ND6_IFF_IFDISABLED) && 1502 ifp->if_flags & IFF_UP) { 1503 /* 1504 * When the IF already has 1505 * ND6_IFF_AUTO_LINKLOCAL, no link-local 1506 * address is assigned, and IFF_UP, try to 1507 * assign one. 1508 */ 1509 IF_ADDR_RLOCK(ifp); 1510 TAILQ_FOREACH(ifa, &ifp->if_addrhead, 1511 ifa_link) { 1512 if (ifa->ifa_addr->sa_family != 1513 AF_INET6) 1514 continue; 1515 ia = (struct in6_ifaddr *)ifa; 1516 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) 1517 break; 1518 } 1519 IF_ADDR_RUNLOCK(ifp); 1520 if (ifa != NULL) 1521 /* No LLA is configured. */ 1522 in6_ifattach(ifp, NULL); 1523 } 1524 } 1525 } 1526 ND_IFINFO(ifp)->flags = ND.flags; 1527 break; 1528 #undef ND 1529 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1530 /* sync kernel routing table with the default router list */ 1531 defrouter_reset(); 1532 defrouter_select(); 1533 break; 1534 case SIOCSPFXFLUSH_IN6: 1535 { 1536 /* flush all the prefix advertised by routers */ 1537 struct nd_prefix *pr, *next; 1538 1539 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) { 1540 struct in6_ifaddr *ia, *ia_next; 1541 1542 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1543 continue; /* XXX */ 1544 1545 /* do we really have to remove addresses as well? */ 1546 /* XXXRW: in6_ifaddrhead locking. */ 1547 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link, 1548 ia_next) { 1549 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1550 continue; 1551 1552 if (ia->ia6_ndpr == pr) 1553 in6_purgeaddr(&ia->ia_ifa); 1554 } 1555 prelist_remove(pr); 1556 } 1557 break; 1558 } 1559 case SIOCSRTRFLUSH_IN6: 1560 { 1561 /* flush all the default routers */ 1562 struct nd_defrouter *dr, *next; 1563 1564 defrouter_reset(); 1565 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) { 1566 defrtrlist_del(dr); 1567 } 1568 defrouter_select(); 1569 break; 1570 } 1571 case SIOCGNBRINFO_IN6: 1572 { 1573 struct llentry *ln; 1574 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1575 1576 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1577 return (error); 1578 1579 IF_AFDATA_RLOCK(ifp); 1580 ln = nd6_lookup(&nb_addr, 0, ifp); 1581 IF_AFDATA_RUNLOCK(ifp); 1582 1583 if (ln == NULL) { 1584 error = EINVAL; 1585 break; 1586 } 1587 nbi->state = ln->ln_state; 1588 nbi->asked = ln->la_asked; 1589 nbi->isrouter = ln->ln_router; 1590 if (ln->la_expire == 0) 1591 nbi->expire = 0; 1592 else 1593 nbi->expire = ln->la_expire + 1594 (time_second - time_uptime); 1595 LLE_RUNLOCK(ln); 1596 break; 1597 } 1598 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1599 ndif->ifindex = V_nd6_defifindex; 1600 break; 1601 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1602 return (nd6_setdefaultiface(ndif->ifindex)); 1603 } 1604 return (error); 1605 } 1606 1607 /* 1608 * Calculates new isRouter value based on provided parameters and 1609 * returns it. 1610 */ 1611 static int 1612 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr, 1613 int ln_router) 1614 { 1615 1616 /* 1617 * ICMP6 type dependent behavior. 1618 * 1619 * NS: clear IsRouter if new entry 1620 * RS: clear IsRouter 1621 * RA: set IsRouter if there's lladdr 1622 * redir: clear IsRouter if new entry 1623 * 1624 * RA case, (1): 1625 * The spec says that we must set IsRouter in the following cases: 1626 * - If lladdr exist, set IsRouter. This means (1-5). 1627 * - If it is old entry (!newentry), set IsRouter. This means (7). 1628 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1629 * A quetion arises for (1) case. (1) case has no lladdr in the 1630 * neighbor cache, this is similar to (6). 1631 * This case is rare but we figured that we MUST NOT set IsRouter. 1632 * 1633 * is_new old_addr new_addr NS RS RA redir 1634 * D R 1635 * 0 n n (1) c ? s 1636 * 0 y n (2) c s s 1637 * 0 n y (3) c s s 1638 * 0 y y (4) c s s 1639 * 0 y y (5) c s s 1640 * 1 -- n (6) c c c s 1641 * 1 -- y (7) c c s c s 1642 * 1643 * (c=clear s=set) 1644 */ 1645 switch (type & 0xff) { 1646 case ND_NEIGHBOR_SOLICIT: 1647 /* 1648 * New entry must have is_router flag cleared. 1649 */ 1650 if (is_new) /* (6-7) */ 1651 ln_router = 0; 1652 break; 1653 case ND_REDIRECT: 1654 /* 1655 * If the icmp is a redirect to a better router, always set the 1656 * is_router flag. Otherwise, if the entry is newly created, 1657 * clear the flag. [RFC 2461, sec 8.3] 1658 */ 1659 if (code == ND_REDIRECT_ROUTER) 1660 ln_router = 1; 1661 else { 1662 if (is_new) /* (6-7) */ 1663 ln_router = 0; 1664 } 1665 break; 1666 case ND_ROUTER_SOLICIT: 1667 /* 1668 * is_router flag must always be cleared. 1669 */ 1670 ln_router = 0; 1671 break; 1672 case ND_ROUTER_ADVERT: 1673 /* 1674 * Mark an entry with lladdr as a router. 1675 */ 1676 if ((!is_new && (old_addr || new_addr)) || /* (2-5) */ 1677 (is_new && new_addr)) { /* (7) */ 1678 ln_router = 1; 1679 } 1680 break; 1681 } 1682 1683 return (ln_router); 1684 } 1685 1686 /* 1687 * Create neighbor cache entry and cache link-layer address, 1688 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1689 * 1690 * type - ICMP6 type 1691 * code - type dependent information 1692 * 1693 */ 1694 void 1695 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, 1696 int lladdrlen, int type, int code) 1697 { 1698 struct llentry *ln = NULL, *ln_tmp; 1699 int is_newentry; 1700 int do_update; 1701 int olladdr; 1702 int llchange; 1703 int flags; 1704 uint16_t router = 0; 1705 struct sockaddr_in6 sin6; 1706 struct mbuf *chain = NULL; 1707 1708 IF_AFDATA_UNLOCK_ASSERT(ifp); 1709 1710 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__)); 1711 KASSERT(from != NULL, ("%s: from == NULL", __func__)); 1712 1713 /* nothing must be updated for unspecified address */ 1714 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1715 return; 1716 1717 /* 1718 * Validation about ifp->if_addrlen and lladdrlen must be done in 1719 * the caller. 1720 * 1721 * XXX If the link does not have link-layer adderss, what should 1722 * we do? (ifp->if_addrlen == 0) 1723 * Spec says nothing in sections for RA, RS and NA. There's small 1724 * description on it in NS section (RFC 2461 7.2.3). 1725 */ 1726 flags = lladdr ? LLE_EXCLUSIVE : 0; 1727 IF_AFDATA_RLOCK(ifp); 1728 ln = nd6_lookup(from, flags, ifp); 1729 IF_AFDATA_RUNLOCK(ifp); 1730 is_newentry = 0; 1731 if (ln == NULL) { 1732 flags |= LLE_EXCLUSIVE; 1733 ln = nd6_alloc(from, 0, ifp); 1734 if (ln == NULL) 1735 return; 1736 1737 /* 1738 * Since we already know all the data for the new entry, 1739 * fill it before insertion. 1740 */ 1741 if (lladdr != NULL) { 1742 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen); 1743 ln->la_flags |= LLE_VALID; 1744 } 1745 IF_AFDATA_WLOCK(ifp); 1746 LLE_WLOCK(ln); 1747 /* Prefer any existing lle over newly-created one */ 1748 ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp); 1749 if (ln_tmp == NULL) 1750 lltable_link_entry(LLTABLE6(ifp), ln); 1751 IF_AFDATA_WUNLOCK(ifp); 1752 if (ln_tmp == NULL) { 1753 /* No existing lle, mark as new entry (6,7) */ 1754 is_newentry = 1; 1755 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); 1756 if (lladdr != NULL) /* (7) */ 1757 EVENTHANDLER_INVOKE(lle_event, ln, 1758 LLENTRY_RESOLVED); 1759 } else { 1760 lltable_free_entry(LLTABLE6(ifp), ln); 1761 ln = ln_tmp; 1762 ln_tmp = NULL; 1763 } 1764 } 1765 /* do nothing if static ndp is set */ 1766 if ((ln->la_flags & LLE_STATIC)) { 1767 if (flags & LLE_EXCLUSIVE) 1768 LLE_WUNLOCK(ln); 1769 else 1770 LLE_RUNLOCK(ln); 1771 return; 1772 } 1773 1774 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0; 1775 if (olladdr && lladdr) { 1776 llchange = bcmp(lladdr, &ln->ll_addr, 1777 ifp->if_addrlen); 1778 } else if (!olladdr && lladdr) 1779 llchange = 1; 1780 else 1781 llchange = 0; 1782 1783 /* 1784 * newentry olladdr lladdr llchange (*=record) 1785 * 0 n n -- (1) 1786 * 0 y n -- (2) 1787 * 0 n y y (3) * STALE 1788 * 0 y y n (4) * 1789 * 0 y y y (5) * STALE 1790 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1791 * 1 -- y -- (7) * STALE 1792 */ 1793 1794 do_update = 0; 1795 if (is_newentry == 0 && llchange != 0) { 1796 do_update = 1; /* (3,5) */ 1797 1798 /* 1799 * Record source link-layer address 1800 * XXX is it dependent to ifp->if_type? 1801 */ 1802 bcopy(lladdr, &ln->ll_addr, ifp->if_addrlen); 1803 ln->la_flags |= LLE_VALID; 1804 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); 1805 1806 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 1807 1808 if (ln->la_hold != NULL) 1809 nd6_grab_holdchain(ln, &chain, &sin6); 1810 } 1811 1812 /* Calculates new router status */ 1813 router = nd6_is_router(type, code, is_newentry, olladdr, 1814 lladdr != NULL ? 1 : 0, ln->ln_router); 1815 1816 ln->ln_router = router; 1817 /* Mark non-router redirects with special flag */ 1818 if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER) 1819 ln->la_flags |= LLE_REDIRECT; 1820 1821 if (flags & LLE_EXCLUSIVE) 1822 LLE_WUNLOCK(ln); 1823 else 1824 LLE_RUNLOCK(ln); 1825 1826 if (chain != NULL) 1827 nd6_flush_holdchain(ifp, ifp, chain, &sin6); 1828 1829 /* 1830 * When the link-layer address of a router changes, select the 1831 * best router again. In particular, when the neighbor entry is newly 1832 * created, it might affect the selection policy. 1833 * Question: can we restrict the first condition to the "is_newentry" 1834 * case? 1835 * XXX: when we hear an RA from a new router with the link-layer 1836 * address option, defrouter_select() is called twice, since 1837 * defrtrlist_update called the function as well. However, I believe 1838 * we can compromise the overhead, since it only happens the first 1839 * time. 1840 * XXX: although defrouter_select() should not have a bad effect 1841 * for those are not autoconfigured hosts, we explicitly avoid such 1842 * cases for safety. 1843 */ 1844 if ((do_update || is_newentry) && router && 1845 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1846 /* 1847 * guaranteed recursion 1848 */ 1849 defrouter_select(); 1850 } 1851 } 1852 1853 static void 1854 nd6_slowtimo(void *arg) 1855 { 1856 CURVNET_SET((struct vnet *) arg); 1857 struct nd_ifinfo *nd6if; 1858 struct ifnet *ifp; 1859 1860 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 1861 nd6_slowtimo, curvnet); 1862 IFNET_RLOCK_NOSLEEP(); 1863 TAILQ_FOREACH(ifp, &V_ifnet, if_link) { 1864 if (ifp->if_afdata[AF_INET6] == NULL) 1865 continue; 1866 nd6if = ND_IFINFO(ifp); 1867 if (nd6if->basereachable && /* already initialized */ 1868 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 1869 /* 1870 * Since reachable time rarely changes by router 1871 * advertisements, we SHOULD insure that a new random 1872 * value gets recomputed at least once every few hours. 1873 * (RFC 2461, 6.3.4) 1874 */ 1875 nd6if->recalctm = V_nd6_recalc_reachtm_interval; 1876 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 1877 } 1878 } 1879 IFNET_RUNLOCK_NOSLEEP(); 1880 CURVNET_RESTORE(); 1881 } 1882 1883 void 1884 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain, 1885 struct sockaddr_in6 *sin6) 1886 { 1887 1888 LLE_WLOCK_ASSERT(ln); 1889 1890 *chain = ln->la_hold; 1891 ln->la_hold = NULL; 1892 lltable_fill_sa_entry(ln, (struct sockaddr *)sin6); 1893 1894 if (ln->ln_state == ND6_LLINFO_STALE) { 1895 1896 /* 1897 * The first time we send a packet to a 1898 * neighbor whose entry is STALE, we have 1899 * to change the state to DELAY and a sets 1900 * a timer to expire in DELAY_FIRST_PROBE_TIME 1901 * seconds to ensure do neighbor unreachability 1902 * detection on expiration. 1903 * (RFC 2461 7.3.3) 1904 */ 1905 nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY); 1906 } 1907 } 1908 1909 int 1910 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m, 1911 struct sockaddr_in6 *dst) 1912 { 1913 int error; 1914 int ip6len; 1915 struct ip6_hdr *ip6; 1916 struct m_tag *mtag; 1917 1918 #ifdef MAC 1919 mac_netinet6_nd6_send(ifp, m); 1920 #endif 1921 1922 /* 1923 * If called from nd6_ns_output() (NS), nd6_na_output() (NA), 1924 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA 1925 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND 1926 * to be diverted to user space. When re-injected into the kernel, 1927 * send_output() will directly dispatch them to the outgoing interface. 1928 */ 1929 if (send_sendso_input_hook != NULL) { 1930 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); 1931 if (mtag != NULL) { 1932 ip6 = mtod(m, struct ip6_hdr *); 1933 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); 1934 /* Use the SEND socket */ 1935 error = send_sendso_input_hook(m, ifp, SND_OUT, 1936 ip6len); 1937 /* -1 == no app on SEND socket */ 1938 if (error == 0 || error != -1) 1939 return (error); 1940 } 1941 } 1942 1943 m_clrprotoflags(m); /* Avoid confusing lower layers. */ 1944 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL, 1945 mtod(m, struct ip6_hdr *)); 1946 1947 if ((ifp->if_flags & IFF_LOOPBACK) == 0) 1948 origifp = ifp; 1949 1950 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, NULL); 1951 return (error); 1952 } 1953 1954 /* 1955 * Do L2 address resolution for @sa_dst address. Stores found 1956 * address in @desten buffer. Copy of lle ln_flags can be also 1957 * saved in @pflags if @pflags is non-NULL. 1958 * 1959 * If destination LLE does not exists or lle state modification 1960 * is required, call "slow" version. 1961 * 1962 * Return values: 1963 * - 0 on success (address copied to buffer). 1964 * - EWOULDBLOCK (no local error, but address is still unresolved) 1965 * - other errors (alloc failure, etc) 1966 */ 1967 int 1968 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m, 1969 const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags) 1970 { 1971 struct llentry *ln = NULL; 1972 const struct sockaddr_in6 *dst6; 1973 1974 if (pflags != NULL) 1975 *pflags = 0; 1976 1977 dst6 = (const struct sockaddr_in6 *)sa_dst; 1978 1979 /* discard the packet if IPv6 operation is disabled on the interface */ 1980 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 1981 m_freem(m); 1982 return (ENETDOWN); /* better error? */ 1983 } 1984 1985 if (m != NULL && m->m_flags & M_MCAST) { 1986 switch (ifp->if_type) { 1987 case IFT_ETHER: 1988 case IFT_FDDI: 1989 case IFT_L2VLAN: 1990 case IFT_IEEE80211: 1991 case IFT_BRIDGE: 1992 case IFT_ISO88025: 1993 ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr, 1994 desten); 1995 return (0); 1996 default: 1997 m_freem(m); 1998 return (EAFNOSUPPORT); 1999 } 2000 } 2001 2002 IF_AFDATA_RLOCK(ifp); 2003 ln = nd6_lookup(&dst6->sin6_addr, 0, ifp); 2004 IF_AFDATA_RUNLOCK(ifp); 2005 2006 /* 2007 * Perform fast path for the following cases: 2008 * 1) lle state is REACHABLE 2009 * 2) lle state is DELAY (NS message sent) 2010 * 2011 * Every other case involves lle modification, so we handle 2012 * them separately. 2013 */ 2014 if (ln == NULL || (ln->ln_state != ND6_LLINFO_REACHABLE && 2015 ln->ln_state != ND6_LLINFO_DELAY)) { 2016 /* Fall back to slow processing path */ 2017 if (ln != NULL) 2018 LLE_RUNLOCK(ln); 2019 return (nd6_resolve_slow(ifp, m, dst6, desten, pflags)); 2020 } 2021 2022 2023 bcopy(&ln->ll_addr, desten, ifp->if_addrlen); 2024 if (pflags != NULL) 2025 *pflags = ln->la_flags; 2026 LLE_RUNLOCK(ln); 2027 return (0); 2028 } 2029 2030 2031 /* 2032 * Do L2 address resolution for @sa_dst address. Stores found 2033 * address in @desten buffer. Copy of lle ln_flags can be also 2034 * saved in @pflags if @pflags is non-NULL. 2035 * 2036 * Heavy version. 2037 * Function assume that destination LLE does not exist, 2038 * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired. 2039 * 2040 * Set noinline to be dtrace-friendly 2041 */ 2042 static __noinline int 2043 nd6_resolve_slow(struct ifnet *ifp, struct mbuf *m, 2044 const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags) 2045 { 2046 struct llentry *lle = NULL, *lle_tmp; 2047 struct in6_addr *psrc, src; 2048 int send_ns; 2049 2050 /* 2051 * Address resolution or Neighbor Unreachability Detection 2052 * for the next hop. 2053 * At this point, the destination of the packet must be a unicast 2054 * or an anycast address(i.e. not a multicast). 2055 */ 2056 if (lle == NULL) { 2057 IF_AFDATA_RLOCK(ifp); 2058 lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp); 2059 IF_AFDATA_RUNLOCK(ifp); 2060 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) { 2061 /* 2062 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 2063 * the condition below is not very efficient. But we believe 2064 * it is tolerable, because this should be a rare case. 2065 */ 2066 lle = nd6_alloc(&dst->sin6_addr, 0, ifp); 2067 if (lle == NULL) { 2068 char ip6buf[INET6_ADDRSTRLEN]; 2069 log(LOG_DEBUG, 2070 "nd6_output: can't allocate llinfo for %s " 2071 "(ln=%p)\n", 2072 ip6_sprintf(ip6buf, &dst->sin6_addr), lle); 2073 m_freem(m); 2074 return (ENOBUFS); 2075 } 2076 2077 IF_AFDATA_WLOCK(ifp); 2078 LLE_WLOCK(lle); 2079 /* Prefer any existing entry over newly-created one */ 2080 lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp); 2081 if (lle_tmp == NULL) 2082 lltable_link_entry(LLTABLE6(ifp), lle); 2083 IF_AFDATA_WUNLOCK(ifp); 2084 if (lle_tmp != NULL) { 2085 lltable_free_entry(LLTABLE6(ifp), lle); 2086 lle = lle_tmp; 2087 lle_tmp = NULL; 2088 } 2089 } 2090 } 2091 if (lle == NULL) { 2092 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 2093 m_freem(m); 2094 return (ENOBUFS); 2095 } 2096 2097 if (m != NULL) 2098 m_freem(m); 2099 return (ENOBUFS); 2100 } 2101 2102 LLE_WLOCK_ASSERT(lle); 2103 2104 /* 2105 * The first time we send a packet to a neighbor whose entry is 2106 * STALE, we have to change the state to DELAY and a sets a timer to 2107 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 2108 * neighbor unreachability detection on expiration. 2109 * (RFC 2461 7.3.3) 2110 */ 2111 if (lle->ln_state == ND6_LLINFO_STALE) 2112 nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY); 2113 2114 /* 2115 * If the neighbor cache entry has a state other than INCOMPLETE 2116 * (i.e. its link-layer address is already resolved), just 2117 * send the packet. 2118 */ 2119 if (lle->ln_state > ND6_LLINFO_INCOMPLETE) { 2120 bcopy(&lle->ll_addr, desten, ifp->if_addrlen); 2121 if (pflags != NULL) 2122 *pflags = lle->la_flags; 2123 LLE_WUNLOCK(lle); 2124 return (0); 2125 } 2126 2127 /* 2128 * There is a neighbor cache entry, but no ethernet address 2129 * response yet. Append this latest packet to the end of the 2130 * packet queue in the mbuf, unless the number of the packet 2131 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 2132 * the oldest packet in the queue will be removed. 2133 */ 2134 2135 if (lle->la_hold != NULL) { 2136 struct mbuf *m_hold; 2137 int i; 2138 2139 i = 0; 2140 for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){ 2141 i++; 2142 if (m_hold->m_nextpkt == NULL) { 2143 m_hold->m_nextpkt = m; 2144 break; 2145 } 2146 } 2147 while (i >= V_nd6_maxqueuelen) { 2148 m_hold = lle->la_hold; 2149 lle->la_hold = lle->la_hold->m_nextpkt; 2150 m_freem(m_hold); 2151 i--; 2152 } 2153 } else { 2154 lle->la_hold = m; 2155 } 2156 2157 /* 2158 * If there has been no NS for the neighbor after entering the 2159 * INCOMPLETE state, send the first solicitation. 2160 * Note that for newly-created lle la_asked will be 0, 2161 * so we will transition from ND6_LLINFO_NOSTATE to 2162 * ND6_LLINFO_INCOMPLETE state here. 2163 */ 2164 psrc = NULL; 2165 send_ns = 0; 2166 if (lle->la_asked == 0) { 2167 lle->la_asked++; 2168 send_ns = 1; 2169 psrc = nd6_llinfo_get_holdsrc(lle, &src); 2170 2171 nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE); 2172 } 2173 LLE_WUNLOCK(lle); 2174 if (send_ns != 0) 2175 nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL); 2176 2177 return (EWOULDBLOCK); 2178 } 2179 2180 2181 int 2182 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain, 2183 struct sockaddr_in6 *dst) 2184 { 2185 struct mbuf *m, *m_head; 2186 struct ifnet *outifp; 2187 int error = 0; 2188 2189 m_head = chain; 2190 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2191 outifp = origifp; 2192 else 2193 outifp = ifp; 2194 2195 while (m_head) { 2196 m = m_head; 2197 m_head = m_head->m_nextpkt; 2198 error = nd6_output_ifp(ifp, origifp, m, dst); 2199 } 2200 2201 /* 2202 * XXX 2203 * note that intermediate errors are blindly ignored 2204 */ 2205 return (error); 2206 } 2207 2208 static int 2209 nd6_need_cache(struct ifnet *ifp) 2210 { 2211 /* 2212 * XXX: we currently do not make neighbor cache on any interface 2213 * other than ARCnet, Ethernet, FDDI and GIF. 2214 * 2215 * RFC2893 says: 2216 * - unidirectional tunnels needs no ND 2217 */ 2218 switch (ifp->if_type) { 2219 case IFT_ARCNET: 2220 case IFT_ETHER: 2221 case IFT_FDDI: 2222 case IFT_IEEE1394: 2223 case IFT_L2VLAN: 2224 case IFT_IEEE80211: 2225 case IFT_INFINIBAND: 2226 case IFT_BRIDGE: 2227 case IFT_PROPVIRTUAL: 2228 return (1); 2229 default: 2230 return (0); 2231 } 2232 } 2233 2234 /* 2235 * Add pernament ND6 link-layer record for given 2236 * interface address. 2237 * 2238 * Very similar to IPv4 arp_ifinit(), but: 2239 * 1) IPv6 DAD is performed in different place 2240 * 2) It is called by IPv6 protocol stack in contrast to 2241 * arp_ifinit() which is typically called in SIOCSIFADDR 2242 * driver ioctl handler. 2243 * 2244 */ 2245 int 2246 nd6_add_ifa_lle(struct in6_ifaddr *ia) 2247 { 2248 struct ifnet *ifp; 2249 struct llentry *ln, *ln_tmp; 2250 struct sockaddr *dst; 2251 2252 ifp = ia->ia_ifa.ifa_ifp; 2253 if (nd6_need_cache(ifp) == 0) 2254 return (0); 2255 2256 ia->ia_ifa.ifa_rtrequest = nd6_rtrequest; 2257 dst = (struct sockaddr *)&ia->ia_addr; 2258 ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst); 2259 if (ln == NULL) 2260 return (ENOBUFS); 2261 2262 IF_AFDATA_WLOCK(ifp); 2263 LLE_WLOCK(ln); 2264 /* Unlink any entry if exists */ 2265 ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst); 2266 if (ln_tmp != NULL) 2267 lltable_unlink_entry(LLTABLE6(ifp), ln_tmp); 2268 lltable_link_entry(LLTABLE6(ifp), ln); 2269 IF_AFDATA_WUNLOCK(ifp); 2270 2271 if (ln_tmp != NULL) 2272 EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED); 2273 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 2274 2275 LLE_WUNLOCK(ln); 2276 if (ln_tmp != NULL) 2277 llentry_free(ln_tmp); 2278 2279 return (0); 2280 } 2281 2282 /* 2283 * Removes either all lle entries for given @ia, or lle 2284 * corresponding to @ia address. 2285 */ 2286 void 2287 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all) 2288 { 2289 struct sockaddr_in6 mask, addr; 2290 struct sockaddr *saddr, *smask; 2291 struct ifnet *ifp; 2292 2293 ifp = ia->ia_ifa.ifa_ifp; 2294 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr)); 2295 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask)); 2296 saddr = (struct sockaddr *)&addr; 2297 smask = (struct sockaddr *)&mask; 2298 2299 if (all != 0) 2300 lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC); 2301 else 2302 lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr); 2303 } 2304 2305 static void 2306 clear_llinfo_pqueue(struct llentry *ln) 2307 { 2308 struct mbuf *m_hold, *m_hold_next; 2309 2310 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) { 2311 m_hold_next = m_hold->m_nextpkt; 2312 m_freem(m_hold); 2313 } 2314 2315 ln->la_hold = NULL; 2316 return; 2317 } 2318 2319 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2320 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2321 #ifdef SYSCTL_DECL 2322 SYSCTL_DECL(_net_inet6_icmp6); 2323 #endif 2324 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2325 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2326 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2327 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2328 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2329 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, ""); 2330 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer, 2331 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), ""); 2332 2333 static int 2334 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2335 { 2336 struct in6_defrouter d; 2337 struct nd_defrouter *dr; 2338 int error; 2339 2340 if (req->newptr) 2341 return (EPERM); 2342 2343 bzero(&d, sizeof(d)); 2344 d.rtaddr.sin6_family = AF_INET6; 2345 d.rtaddr.sin6_len = sizeof(d.rtaddr); 2346 2347 /* 2348 * XXX locking 2349 */ 2350 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 2351 d.rtaddr.sin6_addr = dr->rtaddr; 2352 error = sa6_recoverscope(&d.rtaddr); 2353 if (error != 0) 2354 return (error); 2355 d.flags = dr->flags; 2356 d.rtlifetime = dr->rtlifetime; 2357 d.expire = dr->expire + (time_second - time_uptime); 2358 d.if_index = dr->ifp->if_index; 2359 error = SYSCTL_OUT(req, &d, sizeof(d)); 2360 if (error != 0) 2361 return (error); 2362 } 2363 return (0); 2364 } 2365 2366 static int 2367 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2368 { 2369 struct in6_prefix p; 2370 struct sockaddr_in6 s6; 2371 struct nd_prefix *pr; 2372 struct nd_pfxrouter *pfr; 2373 time_t maxexpire; 2374 int error; 2375 char ip6buf[INET6_ADDRSTRLEN]; 2376 2377 if (req->newptr) 2378 return (EPERM); 2379 2380 bzero(&p, sizeof(p)); 2381 p.origin = PR_ORIG_RA; 2382 bzero(&s6, sizeof(s6)); 2383 s6.sin6_family = AF_INET6; 2384 s6.sin6_len = sizeof(s6); 2385 2386 /* 2387 * XXX locking 2388 */ 2389 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 2390 p.prefix = pr->ndpr_prefix; 2391 if (sa6_recoverscope(&p.prefix)) { 2392 log(LOG_ERR, "scope error in prefix list (%s)\n", 2393 ip6_sprintf(ip6buf, &p.prefix.sin6_addr)); 2394 /* XXX: press on... */ 2395 } 2396 p.raflags = pr->ndpr_raf; 2397 p.prefixlen = pr->ndpr_plen; 2398 p.vltime = pr->ndpr_vltime; 2399 p.pltime = pr->ndpr_pltime; 2400 p.if_index = pr->ndpr_ifp->if_index; 2401 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2402 p.expire = 0; 2403 else { 2404 /* XXX: we assume time_t is signed. */ 2405 maxexpire = (-1) & 2406 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1)); 2407 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate) 2408 p.expire = pr->ndpr_lastupdate + 2409 pr->ndpr_vltime + 2410 (time_second - time_uptime); 2411 else 2412 p.expire = maxexpire; 2413 } 2414 p.refcnt = pr->ndpr_refcnt; 2415 p.flags = pr->ndpr_stateflags; 2416 p.advrtrs = 0; 2417 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) 2418 p.advrtrs++; 2419 error = SYSCTL_OUT(req, &p, sizeof(p)); 2420 if (error != 0) 2421 return (error); 2422 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 2423 s6.sin6_addr = pfr->router->rtaddr; 2424 if (sa6_recoverscope(&s6)) 2425 log(LOG_ERR, 2426 "scope error in prefix list (%s)\n", 2427 ip6_sprintf(ip6buf, &pfr->router->rtaddr)); 2428 error = SYSCTL_OUT(req, &s6, sizeof(s6)); 2429 if (error != 0) 2430 return (error); 2431 } 2432 } 2433 return (0); 2434 } 2435