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 > 0) 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 * Checks if we need to switch from STALE state. 546 * 547 * RFC 4861 requires switching from STALE to DELAY state 548 * on first packet matching entry, waiting V_nd6_delay and 549 * transition to PROBE state (if upper layer confirmation was 550 * not received). 551 * 552 * This code performs a bit differently: 553 * On packet hit we don't change state (but desired state 554 * can be guessed by control plane). However, after V_nd6_delay 555 * seconds code will transition to PROBE state (so DELAY state 556 * is kinda skipped in most situations). 557 * 558 * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so 559 * we perform the following upon entering STALE state: 560 * 561 * 1) Arm timer to run each V_nd6_delay seconds to make sure that 562 * if packet was transmitted at the start of given interval, we 563 * would be able to switch to PROBE state in V_nd6_delay seconds 564 * as user expects. 565 * 566 * 2) Reschedule timer until original V_nd6_gctimer expires keeping 567 * lle in STALE state (remaining timer value stored in lle_remtime). 568 * 569 * 3) Reschedule timer if packet was transmitted less that V_nd6_delay 570 * seconds ago. 571 * 572 * Returns non-zero value if the entry is still STALE (storing 573 * the next timer interval in @pdelay). 574 * 575 * Returns zero value if original timer expired or we need to switch to 576 * PROBE (store that in @do_switch variable). 577 */ 578 static int 579 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch) 580 { 581 int nd_delay, nd_gctimer, r_skip_req; 582 time_t lle_hittime; 583 long delay; 584 585 *do_switch = 0; 586 nd_gctimer = V_nd6_gctimer; 587 nd_delay = V_nd6_delay; 588 589 LLE_REQ_LOCK(lle); 590 r_skip_req = lle->r_skip_req; 591 lle_hittime = lle->lle_hittime; 592 LLE_REQ_UNLOCK(lle); 593 594 if (r_skip_req > 0) { 595 596 /* 597 * Nonzero r_skip_req value was set upon entering 598 * STALE state. Since value was not changed, no 599 * packets were passed using this lle. Ask for 600 * timer reschedule and keep STALE state. 601 */ 602 delay = (long)(MIN(nd_gctimer, nd_delay)); 603 delay *= hz; 604 if (lle->lle_remtime > delay) 605 lle->lle_remtime -= delay; 606 else { 607 delay = lle->lle_remtime; 608 lle->lle_remtime = 0; 609 } 610 611 if (delay == 0) { 612 613 /* 614 * The original ng6_gctime timeout ended, 615 * no more rescheduling. 616 */ 617 return (0); 618 } 619 620 *pdelay = delay; 621 return (1); 622 } 623 624 /* 625 * Packet received. Verify timestamp 626 */ 627 delay = (long)(time_uptime - lle_hittime); 628 if (delay < nd_delay) { 629 630 /* 631 * V_nd6_delay still not passed since the first 632 * hit in STALE state. 633 * Reshedule timer and return. 634 */ 635 *pdelay = (long)(nd_delay - delay) * hz; 636 return (1); 637 } 638 639 /* Request switching to probe */ 640 *do_switch = 1; 641 return (0); 642 } 643 644 645 /* 646 * Switch @lle state to new state optionally arming timers. 647 * 648 * Set noinline to be dtrace-friendly 649 */ 650 __noinline void 651 nd6_llinfo_setstate(struct llentry *lle, int newstate) 652 { 653 struct ifnet *ifp; 654 int nd_gctimer, nd_delay; 655 long delay, remtime; 656 657 delay = 0; 658 remtime = 0; 659 660 switch (newstate) { 661 case ND6_LLINFO_INCOMPLETE: 662 ifp = lle->lle_tbl->llt_ifp; 663 delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000; 664 break; 665 case ND6_LLINFO_REACHABLE: 666 if (!ND6_LLINFO_PERMANENT(lle)) { 667 ifp = lle->lle_tbl->llt_ifp; 668 delay = (long)ND_IFINFO(ifp)->reachable * hz; 669 } 670 break; 671 case ND6_LLINFO_STALE: 672 673 /* 674 * Notify fast path that we want to know if any packet 675 * is transmitted by setting r_skip_req. 676 */ 677 LLE_REQ_LOCK(lle); 678 lle->r_skip_req = 1; 679 LLE_REQ_UNLOCK(lle); 680 nd_delay = V_nd6_delay; 681 nd_gctimer = V_nd6_gctimer; 682 683 delay = (long)(MIN(nd_gctimer, nd_delay)) * hz; 684 remtime = (long)nd_gctimer * hz - delay; 685 break; 686 case ND6_LLINFO_DELAY: 687 lle->la_asked = 0; 688 delay = (long)V_nd6_delay * hz; 689 break; 690 } 691 692 if (delay > 0) 693 nd6_llinfo_settimer_locked(lle, delay); 694 695 lle->lle_remtime = remtime; 696 lle->ln_state = newstate; 697 } 698 699 /* 700 * Timer-dependent part of nd state machine. 701 * 702 * Set noinline to be dtrace-friendly 703 */ 704 static __noinline void 705 nd6_llinfo_timer(void *arg) 706 { 707 struct llentry *ln; 708 struct in6_addr *dst, *pdst, *psrc, src; 709 struct ifnet *ifp; 710 struct nd_ifinfo *ndi = NULL; 711 int do_switch, send_ns; 712 long delay; 713 714 KASSERT(arg != NULL, ("%s: arg NULL", __func__)); 715 ln = (struct llentry *)arg; 716 LLE_WLOCK(ln); 717 if (callout_pending(&ln->lle_timer)) { 718 /* 719 * Here we are a bit odd here in the treatment of 720 * active/pending. If the pending bit is set, it got 721 * rescheduled before I ran. The active 722 * bit we ignore, since if it was stopped 723 * in ll_tablefree() and was currently running 724 * it would have return 0 so the code would 725 * not have deleted it since the callout could 726 * not be stopped so we want to go through 727 * with the delete here now. If the callout 728 * was restarted, the pending bit will be back on and 729 * we just want to bail since the callout_reset would 730 * return 1 and our reference would have been removed 731 * by nd6_llinfo_settimer_locked above since canceled 732 * would have been 1. 733 */ 734 LLE_WUNLOCK(ln); 735 return; 736 } 737 ifp = ln->lle_tbl->llt_ifp; 738 CURVNET_SET(ifp->if_vnet); 739 ndi = ND_IFINFO(ifp); 740 send_ns = 0; 741 dst = &ln->r_l3addr.addr6; 742 pdst = dst; 743 744 if (ln->ln_ntick > 0) { 745 if (ln->ln_ntick > INT_MAX) { 746 ln->ln_ntick -= INT_MAX; 747 nd6_llinfo_settimer_locked(ln, INT_MAX); 748 } else { 749 ln->ln_ntick = 0; 750 nd6_llinfo_settimer_locked(ln, ln->ln_ntick); 751 } 752 goto done; 753 } 754 755 if (ln->la_flags & LLE_STATIC) { 756 goto done; 757 } 758 759 if (ln->la_flags & LLE_DELETED) { 760 nd6_free(ln, 0); 761 ln = NULL; 762 goto done; 763 } 764 765 switch (ln->ln_state) { 766 case ND6_LLINFO_INCOMPLETE: 767 if (ln->la_asked < V_nd6_mmaxtries) { 768 ln->la_asked++; 769 send_ns = 1; 770 /* Send NS to multicast address */ 771 pdst = NULL; 772 } else { 773 struct mbuf *m = ln->la_hold; 774 if (m) { 775 struct mbuf *m0; 776 777 /* 778 * assuming every packet in la_hold has the 779 * same IP header. Send error after unlock. 780 */ 781 m0 = m->m_nextpkt; 782 m->m_nextpkt = NULL; 783 ln->la_hold = m0; 784 clear_llinfo_pqueue(ln); 785 } 786 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT); 787 nd6_free(ln, 0); 788 ln = NULL; 789 if (m != NULL) 790 icmp6_error2(m, ICMP6_DST_UNREACH, 791 ICMP6_DST_UNREACH_ADDR, 0, ifp); 792 } 793 break; 794 case ND6_LLINFO_REACHABLE: 795 if (!ND6_LLINFO_PERMANENT(ln)) 796 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); 797 break; 798 799 case ND6_LLINFO_STALE: 800 if (nd6_is_stale(ln, &delay, &do_switch) != 0) { 801 802 /* 803 * No packet has used this entry and GC timeout 804 * has not been passed. Reshedule timer and 805 * return. 806 */ 807 nd6_llinfo_settimer_locked(ln, delay); 808 break; 809 } 810 811 if (do_switch == 0) { 812 813 /* 814 * GC timer has ended and entry hasn't been used. 815 * Run Garbage collector (RFC 4861, 5.3) 816 */ 817 if (!ND6_LLINFO_PERMANENT(ln)) { 818 EVENTHANDLER_INVOKE(lle_event, ln, 819 LLENTRY_EXPIRED); 820 nd6_free(ln, 1); 821 ln = NULL; 822 } 823 break; 824 } 825 826 /* Entry has been used AND delay timer has ended. */ 827 828 /* FALLTHROUGH */ 829 830 case ND6_LLINFO_DELAY: 831 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 832 /* We need NUD */ 833 ln->la_asked = 1; 834 nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE); 835 send_ns = 1; 836 } else 837 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */ 838 break; 839 case ND6_LLINFO_PROBE: 840 if (ln->la_asked < V_nd6_umaxtries) { 841 ln->la_asked++; 842 send_ns = 1; 843 } else { 844 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 845 nd6_free(ln, 0); 846 ln = NULL; 847 } 848 break; 849 default: 850 panic("%s: paths in a dark night can be confusing: %d", 851 __func__, ln->ln_state); 852 } 853 done: 854 if (send_ns != 0) { 855 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 856 psrc = nd6_llinfo_get_holdsrc(ln, &src); 857 LLE_FREE_LOCKED(ln); 858 ln = NULL; 859 nd6_ns_output(ifp, psrc, pdst, dst, NULL); 860 } 861 862 if (ln != NULL) 863 LLE_FREE_LOCKED(ln); 864 CURVNET_RESTORE(); 865 } 866 867 868 /* 869 * ND6 timer routine to expire default route list and prefix list 870 */ 871 void 872 nd6_timer(void *arg) 873 { 874 CURVNET_SET((struct vnet *) arg); 875 struct nd_defrouter *dr, *ndr; 876 struct nd_prefix *pr, *npr; 877 struct in6_ifaddr *ia6, *nia6; 878 879 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz, 880 nd6_timer, curvnet); 881 882 /* expire default router list */ 883 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 884 if (dr->expire && dr->expire < time_uptime) 885 defrtrlist_del(dr); 886 } 887 888 /* 889 * expire interface addresses. 890 * in the past the loop was inside prefix expiry processing. 891 * However, from a stricter speci-confrmance standpoint, we should 892 * rather separate address lifetimes and prefix lifetimes. 893 * 894 * XXXRW: in6_ifaddrhead locking. 895 */ 896 addrloop: 897 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) { 898 /* check address lifetime */ 899 if (IFA6_IS_INVALID(ia6)) { 900 int regen = 0; 901 902 /* 903 * If the expiring address is temporary, try 904 * regenerating a new one. This would be useful when 905 * we suspended a laptop PC, then turned it on after a 906 * period that could invalidate all temporary 907 * addresses. Although we may have to restart the 908 * loop (see below), it must be after purging the 909 * address. Otherwise, we'd see an infinite loop of 910 * regeneration. 911 */ 912 if (V_ip6_use_tempaddr && 913 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 914 if (regen_tmpaddr(ia6) == 0) 915 regen = 1; 916 } 917 918 in6_purgeaddr(&ia6->ia_ifa); 919 920 if (regen) 921 goto addrloop; /* XXX: see below */ 922 } else if (IFA6_IS_DEPRECATED(ia6)) { 923 int oldflags = ia6->ia6_flags; 924 925 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 926 927 /* 928 * If a temporary address has just become deprecated, 929 * regenerate a new one if possible. 930 */ 931 if (V_ip6_use_tempaddr && 932 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 933 (oldflags & IN6_IFF_DEPRECATED) == 0) { 934 935 if (regen_tmpaddr(ia6) == 0) { 936 /* 937 * A new temporary address is 938 * generated. 939 * XXX: this means the address chain 940 * has changed while we are still in 941 * the loop. Although the change 942 * would not cause disaster (because 943 * it's not a deletion, but an 944 * addition,) we'd rather restart the 945 * loop just for safety. Or does this 946 * significantly reduce performance?? 947 */ 948 goto addrloop; 949 } 950 } 951 } else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) { 952 /* 953 * Schedule DAD for a tentative address. This happens 954 * if the interface was down or not running 955 * when the address was configured. 956 */ 957 int delay; 958 959 delay = arc4random() % 960 (MAX_RTR_SOLICITATION_DELAY * hz); 961 nd6_dad_start((struct ifaddr *)ia6, delay); 962 } else { 963 /* 964 * Check status of the interface. If it is down, 965 * mark the address as tentative for future DAD. 966 */ 967 if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 || 968 (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING) 969 == 0 || 970 (ND_IFINFO(ia6->ia_ifp)->flags & 971 ND6_IFF_IFDISABLED) != 0) { 972 ia6->ia6_flags &= ~IN6_IFF_DUPLICATED; 973 ia6->ia6_flags |= IN6_IFF_TENTATIVE; 974 } 975 /* 976 * A new RA might have made a deprecated address 977 * preferred. 978 */ 979 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 980 } 981 } 982 983 /* expire prefix list */ 984 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 985 /* 986 * check prefix lifetime. 987 * since pltime is just for autoconf, pltime processing for 988 * prefix is not necessary. 989 */ 990 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 991 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) { 992 993 /* 994 * address expiration and prefix expiration are 995 * separate. NEVER perform in6_purgeaddr here. 996 */ 997 prelist_remove(pr); 998 } 999 } 1000 CURVNET_RESTORE(); 1001 } 1002 1003 /* 1004 * ia6 - deprecated/invalidated temporary address 1005 */ 1006 static int 1007 regen_tmpaddr(struct in6_ifaddr *ia6) 1008 { 1009 struct ifaddr *ifa; 1010 struct ifnet *ifp; 1011 struct in6_ifaddr *public_ifa6 = NULL; 1012 1013 ifp = ia6->ia_ifa.ifa_ifp; 1014 IF_ADDR_RLOCK(ifp); 1015 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1016 struct in6_ifaddr *it6; 1017 1018 if (ifa->ifa_addr->sa_family != AF_INET6) 1019 continue; 1020 1021 it6 = (struct in6_ifaddr *)ifa; 1022 1023 /* ignore no autoconf addresses. */ 1024 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1025 continue; 1026 1027 /* ignore autoconf addresses with different prefixes. */ 1028 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 1029 continue; 1030 1031 /* 1032 * Now we are looking at an autoconf address with the same 1033 * prefix as ours. If the address is temporary and is still 1034 * preferred, do not create another one. It would be rare, but 1035 * could happen, for example, when we resume a laptop PC after 1036 * a long period. 1037 */ 1038 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 1039 !IFA6_IS_DEPRECATED(it6)) { 1040 public_ifa6 = NULL; 1041 break; 1042 } 1043 1044 /* 1045 * This is a public autoconf address that has the same prefix 1046 * as ours. If it is preferred, keep it. We can't break the 1047 * loop here, because there may be a still-preferred temporary 1048 * address with the prefix. 1049 */ 1050 if (!IFA6_IS_DEPRECATED(it6)) 1051 public_ifa6 = it6; 1052 } 1053 if (public_ifa6 != NULL) 1054 ifa_ref(&public_ifa6->ia_ifa); 1055 IF_ADDR_RUNLOCK(ifp); 1056 1057 if (public_ifa6 != NULL) { 1058 int e; 1059 1060 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 1061 ifa_free(&public_ifa6->ia_ifa); 1062 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 1063 " tmp addr,errno=%d\n", e); 1064 return (-1); 1065 } 1066 ifa_free(&public_ifa6->ia_ifa); 1067 return (0); 1068 } 1069 1070 return (-1); 1071 } 1072 1073 /* 1074 * Nuke neighbor cache/prefix/default router management table, right before 1075 * ifp goes away. 1076 */ 1077 void 1078 nd6_purge(struct ifnet *ifp) 1079 { 1080 struct nd_defrouter *dr, *ndr; 1081 struct nd_prefix *pr, *npr; 1082 1083 /* 1084 * Nuke default router list entries toward ifp. 1085 * We defer removal of default router list entries that is installed 1086 * in the routing table, in order to keep additional side effects as 1087 * small as possible. 1088 */ 1089 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 1090 if (dr->installed) 1091 continue; 1092 1093 if (dr->ifp == ifp) 1094 defrtrlist_del(dr); 1095 } 1096 1097 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 1098 if (!dr->installed) 1099 continue; 1100 1101 if (dr->ifp == ifp) 1102 defrtrlist_del(dr); 1103 } 1104 1105 /* Nuke prefix list entries toward ifp */ 1106 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 1107 if (pr->ndpr_ifp == ifp) { 1108 /* 1109 * Because if_detach() does *not* release prefixes 1110 * while purging addresses the reference count will 1111 * still be above zero. We therefore reset it to 1112 * make sure that the prefix really gets purged. 1113 */ 1114 pr->ndpr_refcnt = 0; 1115 1116 /* 1117 * Previously, pr->ndpr_addr is removed as well, 1118 * but I strongly believe we don't have to do it. 1119 * nd6_purge() is only called from in6_ifdetach(), 1120 * which removes all the associated interface addresses 1121 * by itself. 1122 * (jinmei@kame.net 20010129) 1123 */ 1124 prelist_remove(pr); 1125 } 1126 } 1127 1128 /* cancel default outgoing interface setting */ 1129 if (V_nd6_defifindex == ifp->if_index) 1130 nd6_setdefaultiface(0); 1131 1132 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1133 /* Refresh default router list. */ 1134 defrouter_select(); 1135 } 1136 1137 /* XXXXX 1138 * We do not nuke the neighbor cache entries here any more 1139 * because the neighbor cache is kept in if_afdata[AF_INET6]. 1140 * nd6_purge() is invoked by in6_ifdetach() which is called 1141 * from if_detach() where everything gets purged. So let 1142 * in6_domifdetach() do the actual L2 table purging work. 1143 */ 1144 } 1145 1146 /* 1147 * the caller acquires and releases the lock on the lltbls 1148 * Returns the llentry locked 1149 */ 1150 struct llentry * 1151 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp) 1152 { 1153 struct sockaddr_in6 sin6; 1154 struct llentry *ln; 1155 1156 bzero(&sin6, sizeof(sin6)); 1157 sin6.sin6_len = sizeof(struct sockaddr_in6); 1158 sin6.sin6_family = AF_INET6; 1159 sin6.sin6_addr = *addr6; 1160 1161 IF_AFDATA_LOCK_ASSERT(ifp); 1162 1163 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6); 1164 1165 return (ln); 1166 } 1167 1168 struct llentry * 1169 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp) 1170 { 1171 struct sockaddr_in6 sin6; 1172 struct llentry *ln; 1173 1174 bzero(&sin6, sizeof(sin6)); 1175 sin6.sin6_len = sizeof(struct sockaddr_in6); 1176 sin6.sin6_family = AF_INET6; 1177 sin6.sin6_addr = *addr6; 1178 1179 ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6); 1180 if (ln != NULL) 1181 ln->ln_state = ND6_LLINFO_NOSTATE; 1182 1183 return (ln); 1184 } 1185 1186 /* 1187 * Test whether a given IPv6 address is a neighbor or not, ignoring 1188 * the actual neighbor cache. The neighbor cache is ignored in order 1189 * to not reenter the routing code from within itself. 1190 */ 1191 static int 1192 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp) 1193 { 1194 struct nd_prefix *pr; 1195 struct ifaddr *dstaddr; 1196 1197 /* 1198 * A link-local address is always a neighbor. 1199 * XXX: a link does not necessarily specify a single interface. 1200 */ 1201 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 1202 struct sockaddr_in6 sin6_copy; 1203 u_int32_t zone; 1204 1205 /* 1206 * We need sin6_copy since sa6_recoverscope() may modify the 1207 * content (XXX). 1208 */ 1209 sin6_copy = *addr; 1210 if (sa6_recoverscope(&sin6_copy)) 1211 return (0); /* XXX: should be impossible */ 1212 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 1213 return (0); 1214 if (sin6_copy.sin6_scope_id == zone) 1215 return (1); 1216 else 1217 return (0); 1218 } 1219 1220 /* 1221 * If the address matches one of our addresses, 1222 * it should be a neighbor. 1223 * If the address matches one of our on-link prefixes, it should be a 1224 * neighbor. 1225 */ 1226 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 1227 if (pr->ndpr_ifp != ifp) 1228 continue; 1229 1230 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) { 1231 struct rtentry *rt; 1232 1233 /* Always use the default FIB here. */ 1234 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 1235 0, 0, RT_DEFAULT_FIB); 1236 if (rt == NULL) 1237 continue; 1238 /* 1239 * This is the case where multiple interfaces 1240 * have the same prefix, but only one is installed 1241 * into the routing table and that prefix entry 1242 * is not the one being examined here. In the case 1243 * where RADIX_MPATH is enabled, multiple route 1244 * entries (of the same rt_key value) will be 1245 * installed because the interface addresses all 1246 * differ. 1247 */ 1248 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 1249 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) { 1250 RTFREE_LOCKED(rt); 1251 continue; 1252 } 1253 RTFREE_LOCKED(rt); 1254 } 1255 1256 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 1257 &addr->sin6_addr, &pr->ndpr_mask)) 1258 return (1); 1259 } 1260 1261 /* 1262 * If the address is assigned on the node of the other side of 1263 * a p2p interface, the address should be a neighbor. 1264 */ 1265 dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS); 1266 if (dstaddr != NULL) { 1267 if (dstaddr->ifa_ifp == ifp) { 1268 ifa_free(dstaddr); 1269 return (1); 1270 } 1271 ifa_free(dstaddr); 1272 } 1273 1274 /* 1275 * If the default router list is empty, all addresses are regarded 1276 * as on-link, and thus, as a neighbor. 1277 */ 1278 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && 1279 TAILQ_EMPTY(&V_nd_defrouter) && 1280 V_nd6_defifindex == ifp->if_index) { 1281 return (1); 1282 } 1283 1284 return (0); 1285 } 1286 1287 1288 /* 1289 * Detect if a given IPv6 address identifies a neighbor on a given link. 1290 * XXX: should take care of the destination of a p2p link? 1291 */ 1292 int 1293 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp) 1294 { 1295 struct llentry *lle; 1296 int rc = 0; 1297 1298 IF_AFDATA_UNLOCK_ASSERT(ifp); 1299 if (nd6_is_new_addr_neighbor(addr, ifp)) 1300 return (1); 1301 1302 /* 1303 * Even if the address matches none of our addresses, it might be 1304 * in the neighbor cache. 1305 */ 1306 IF_AFDATA_RLOCK(ifp); 1307 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) { 1308 LLE_RUNLOCK(lle); 1309 rc = 1; 1310 } 1311 IF_AFDATA_RUNLOCK(ifp); 1312 return (rc); 1313 } 1314 1315 /* 1316 * Free an nd6 llinfo entry. 1317 * Since the function would cause significant changes in the kernel, DO NOT 1318 * make it global, unless you have a strong reason for the change, and are sure 1319 * that the change is safe. 1320 * 1321 * Set noinline to be dtrace-friendly 1322 */ 1323 static __noinline void 1324 nd6_free(struct llentry *ln, int gc) 1325 { 1326 struct nd_defrouter *dr; 1327 struct ifnet *ifp; 1328 1329 LLE_WLOCK_ASSERT(ln); 1330 1331 /* 1332 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1333 * even though it is not harmful, it was not really necessary. 1334 */ 1335 1336 /* cancel timer */ 1337 nd6_llinfo_settimer_locked(ln, -1); 1338 1339 ifp = ln->lle_tbl->llt_ifp; 1340 1341 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1342 dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp); 1343 1344 if (dr != NULL && dr->expire && 1345 ln->ln_state == ND6_LLINFO_STALE && gc) { 1346 /* 1347 * If the reason for the deletion is just garbage 1348 * collection, and the neighbor is an active default 1349 * router, do not delete it. Instead, reset the GC 1350 * timer using the router's lifetime. 1351 * Simply deleting the entry would affect default 1352 * router selection, which is not necessarily a good 1353 * thing, especially when we're using router preference 1354 * values. 1355 * XXX: the check for ln_state would be redundant, 1356 * but we intentionally keep it just in case. 1357 */ 1358 if (dr->expire > time_uptime) 1359 nd6_llinfo_settimer_locked(ln, 1360 (dr->expire - time_uptime) * hz); 1361 else 1362 nd6_llinfo_settimer_locked(ln, 1363 (long)V_nd6_gctimer * hz); 1364 1365 LLE_REMREF(ln); 1366 LLE_WUNLOCK(ln); 1367 return; 1368 } 1369 1370 if (dr) { 1371 /* 1372 * Unreachablity of a router might affect the default 1373 * router selection and on-link detection of advertised 1374 * prefixes. 1375 */ 1376 1377 /* 1378 * Temporarily fake the state to choose a new default 1379 * router and to perform on-link determination of 1380 * prefixes correctly. 1381 * Below the state will be set correctly, 1382 * or the entry itself will be deleted. 1383 */ 1384 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1385 } 1386 1387 if (ln->ln_router || dr) { 1388 1389 /* 1390 * We need to unlock to avoid a LOR with rt6_flush() with the 1391 * rnh and for the calls to pfxlist_onlink_check() and 1392 * defrouter_select() in the block further down for calls 1393 * into nd6_lookup(). We still hold a ref. 1394 */ 1395 LLE_WUNLOCK(ln); 1396 1397 /* 1398 * rt6_flush must be called whether or not the neighbor 1399 * is in the Default Router List. 1400 * See a corresponding comment in nd6_na_input(). 1401 */ 1402 rt6_flush(&ln->r_l3addr.addr6, ifp); 1403 } 1404 1405 if (dr) { 1406 /* 1407 * Since defrouter_select() does not affect the 1408 * on-link determination and MIP6 needs the check 1409 * before the default router selection, we perform 1410 * the check now. 1411 */ 1412 pfxlist_onlink_check(); 1413 1414 /* 1415 * Refresh default router list. 1416 */ 1417 defrouter_select(); 1418 } 1419 1420 /* 1421 * If this entry was added by an on-link redirect, remove the 1422 * corresponding host route. 1423 */ 1424 if (ln->la_flags & LLE_REDIRECT) 1425 nd6_free_redirect(ln); 1426 1427 if (ln->ln_router || dr) 1428 LLE_WLOCK(ln); 1429 } 1430 1431 /* 1432 * Save to unlock. We still hold an extra reference and will not 1433 * free(9) in llentry_free() if someone else holds one as well. 1434 */ 1435 LLE_WUNLOCK(ln); 1436 IF_AFDATA_LOCK(ifp); 1437 LLE_WLOCK(ln); 1438 /* Guard against race with other llentry_free(). */ 1439 if (ln->la_flags & LLE_LINKED) { 1440 /* Remove callout reference */ 1441 LLE_REMREF(ln); 1442 lltable_unlink_entry(ln->lle_tbl, ln); 1443 } 1444 IF_AFDATA_UNLOCK(ifp); 1445 1446 llentry_free(ln); 1447 } 1448 1449 static int 1450 nd6_isdynrte(const struct rtentry *rt, void *xap) 1451 { 1452 1453 if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC)) 1454 return (1); 1455 1456 return (0); 1457 } 1458 /* 1459 * Remove the rtentry for the given llentry, 1460 * both of which were installed by a redirect. 1461 */ 1462 static void 1463 nd6_free_redirect(const struct llentry *ln) 1464 { 1465 int fibnum; 1466 struct sockaddr_in6 sin6; 1467 struct rt_addrinfo info; 1468 1469 lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6); 1470 memset(&info, 0, sizeof(info)); 1471 info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6; 1472 info.rti_filter = nd6_isdynrte; 1473 1474 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) 1475 rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum); 1476 } 1477 1478 /* 1479 * Rejuvenate this function for routing operations related 1480 * processing. 1481 */ 1482 void 1483 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info) 1484 { 1485 struct sockaddr_in6 *gateway; 1486 struct nd_defrouter *dr; 1487 struct ifnet *ifp; 1488 1489 gateway = (struct sockaddr_in6 *)rt->rt_gateway; 1490 ifp = rt->rt_ifp; 1491 1492 switch (req) { 1493 case RTM_ADD: 1494 break; 1495 1496 case RTM_DELETE: 1497 if (!ifp) 1498 return; 1499 /* 1500 * Only indirect routes are interesting. 1501 */ 1502 if ((rt->rt_flags & RTF_GATEWAY) == 0) 1503 return; 1504 /* 1505 * check for default route 1506 */ 1507 if (IN6_ARE_ADDR_EQUAL(&in6addr_any, 1508 &SIN6(rt_key(rt))->sin6_addr)) { 1509 1510 dr = defrouter_lookup(&gateway->sin6_addr, ifp); 1511 if (dr != NULL) 1512 dr->installed = 0; 1513 } 1514 break; 1515 } 1516 } 1517 1518 1519 int 1520 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) 1521 { 1522 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1523 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1524 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1525 int error = 0; 1526 1527 if (ifp->if_afdata[AF_INET6] == NULL) 1528 return (EPFNOSUPPORT); 1529 switch (cmd) { 1530 case OSIOCGIFINFO_IN6: 1531 #define ND ndi->ndi 1532 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1533 bzero(&ND, sizeof(ND)); 1534 ND.linkmtu = IN6_LINKMTU(ifp); 1535 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1536 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1537 ND.reachable = ND_IFINFO(ifp)->reachable; 1538 ND.retrans = ND_IFINFO(ifp)->retrans; 1539 ND.flags = ND_IFINFO(ifp)->flags; 1540 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1541 ND.chlim = ND_IFINFO(ifp)->chlim; 1542 break; 1543 case SIOCGIFINFO_IN6: 1544 ND = *ND_IFINFO(ifp); 1545 break; 1546 case SIOCSIFINFO_IN6: 1547 /* 1548 * used to change host variables from userland. 1549 * intented for a use on router to reflect RA configurations. 1550 */ 1551 /* 0 means 'unspecified' */ 1552 if (ND.linkmtu != 0) { 1553 if (ND.linkmtu < IPV6_MMTU || 1554 ND.linkmtu > IN6_LINKMTU(ifp)) { 1555 error = EINVAL; 1556 break; 1557 } 1558 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1559 } 1560 1561 if (ND.basereachable != 0) { 1562 int obasereachable = ND_IFINFO(ifp)->basereachable; 1563 1564 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1565 if (ND.basereachable != obasereachable) 1566 ND_IFINFO(ifp)->reachable = 1567 ND_COMPUTE_RTIME(ND.basereachable); 1568 } 1569 if (ND.retrans != 0) 1570 ND_IFINFO(ifp)->retrans = ND.retrans; 1571 if (ND.chlim != 0) 1572 ND_IFINFO(ifp)->chlim = ND.chlim; 1573 /* FALLTHROUGH */ 1574 case SIOCSIFINFO_FLAGS: 1575 { 1576 struct ifaddr *ifa; 1577 struct in6_ifaddr *ia; 1578 1579 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1580 !(ND.flags & ND6_IFF_IFDISABLED)) { 1581 /* ifdisabled 1->0 transision */ 1582 1583 /* 1584 * If the interface is marked as ND6_IFF_IFDISABLED and 1585 * has an link-local address with IN6_IFF_DUPLICATED, 1586 * do not clear ND6_IFF_IFDISABLED. 1587 * See RFC 4862, Section 5.4.5. 1588 */ 1589 IF_ADDR_RLOCK(ifp); 1590 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1591 if (ifa->ifa_addr->sa_family != AF_INET6) 1592 continue; 1593 ia = (struct in6_ifaddr *)ifa; 1594 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) && 1595 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) 1596 break; 1597 } 1598 IF_ADDR_RUNLOCK(ifp); 1599 1600 if (ifa != NULL) { 1601 /* LLA is duplicated. */ 1602 ND.flags |= ND6_IFF_IFDISABLED; 1603 log(LOG_ERR, "Cannot enable an interface" 1604 " with a link-local address marked" 1605 " duplicate.\n"); 1606 } else { 1607 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED; 1608 if (ifp->if_flags & IFF_UP) 1609 in6_if_up(ifp); 1610 } 1611 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1612 (ND.flags & ND6_IFF_IFDISABLED)) { 1613 /* ifdisabled 0->1 transision */ 1614 /* Mark all IPv6 address as tentative. */ 1615 1616 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; 1617 if (V_ip6_dad_count > 0 && 1618 (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) { 1619 IF_ADDR_RLOCK(ifp); 1620 TAILQ_FOREACH(ifa, &ifp->if_addrhead, 1621 ifa_link) { 1622 if (ifa->ifa_addr->sa_family != 1623 AF_INET6) 1624 continue; 1625 ia = (struct in6_ifaddr *)ifa; 1626 ia->ia6_flags |= IN6_IFF_TENTATIVE; 1627 } 1628 IF_ADDR_RUNLOCK(ifp); 1629 } 1630 } 1631 1632 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) { 1633 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) { 1634 /* auto_linklocal 0->1 transision */ 1635 1636 /* If no link-local address on ifp, configure */ 1637 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL; 1638 in6_ifattach(ifp, NULL); 1639 } else if (!(ND.flags & ND6_IFF_IFDISABLED) && 1640 ifp->if_flags & IFF_UP) { 1641 /* 1642 * When the IF already has 1643 * ND6_IFF_AUTO_LINKLOCAL, no link-local 1644 * address is assigned, and IFF_UP, try to 1645 * assign one. 1646 */ 1647 IF_ADDR_RLOCK(ifp); 1648 TAILQ_FOREACH(ifa, &ifp->if_addrhead, 1649 ifa_link) { 1650 if (ifa->ifa_addr->sa_family != 1651 AF_INET6) 1652 continue; 1653 ia = (struct in6_ifaddr *)ifa; 1654 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) 1655 break; 1656 } 1657 IF_ADDR_RUNLOCK(ifp); 1658 if (ifa != NULL) 1659 /* No LLA is configured. */ 1660 in6_ifattach(ifp, NULL); 1661 } 1662 } 1663 } 1664 ND_IFINFO(ifp)->flags = ND.flags; 1665 break; 1666 #undef ND 1667 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1668 /* sync kernel routing table with the default router list */ 1669 defrouter_reset(); 1670 defrouter_select(); 1671 break; 1672 case SIOCSPFXFLUSH_IN6: 1673 { 1674 /* flush all the prefix advertised by routers */ 1675 struct nd_prefix *pr, *next; 1676 1677 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) { 1678 struct in6_ifaddr *ia, *ia_next; 1679 1680 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1681 continue; /* XXX */ 1682 1683 /* do we really have to remove addresses as well? */ 1684 /* XXXRW: in6_ifaddrhead locking. */ 1685 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link, 1686 ia_next) { 1687 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1688 continue; 1689 1690 if (ia->ia6_ndpr == pr) 1691 in6_purgeaddr(&ia->ia_ifa); 1692 } 1693 prelist_remove(pr); 1694 } 1695 break; 1696 } 1697 case SIOCSRTRFLUSH_IN6: 1698 { 1699 /* flush all the default routers */ 1700 struct nd_defrouter *dr, *next; 1701 1702 defrouter_reset(); 1703 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) { 1704 defrtrlist_del(dr); 1705 } 1706 defrouter_select(); 1707 break; 1708 } 1709 case SIOCGNBRINFO_IN6: 1710 { 1711 struct llentry *ln; 1712 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1713 1714 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1715 return (error); 1716 1717 IF_AFDATA_RLOCK(ifp); 1718 ln = nd6_lookup(&nb_addr, 0, ifp); 1719 IF_AFDATA_RUNLOCK(ifp); 1720 1721 if (ln == NULL) { 1722 error = EINVAL; 1723 break; 1724 } 1725 nbi->state = ln->ln_state; 1726 nbi->asked = ln->la_asked; 1727 nbi->isrouter = ln->ln_router; 1728 if (ln->la_expire == 0) 1729 nbi->expire = 0; 1730 else 1731 nbi->expire = ln->la_expire + 1732 (time_second - time_uptime); 1733 LLE_RUNLOCK(ln); 1734 break; 1735 } 1736 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1737 ndif->ifindex = V_nd6_defifindex; 1738 break; 1739 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1740 return (nd6_setdefaultiface(ndif->ifindex)); 1741 } 1742 return (error); 1743 } 1744 1745 /* 1746 * Calculates new isRouter value based on provided parameters and 1747 * returns it. 1748 */ 1749 static int 1750 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr, 1751 int ln_router) 1752 { 1753 1754 /* 1755 * ICMP6 type dependent behavior. 1756 * 1757 * NS: clear IsRouter if new entry 1758 * RS: clear IsRouter 1759 * RA: set IsRouter if there's lladdr 1760 * redir: clear IsRouter if new entry 1761 * 1762 * RA case, (1): 1763 * The spec says that we must set IsRouter in the following cases: 1764 * - If lladdr exist, set IsRouter. This means (1-5). 1765 * - If it is old entry (!newentry), set IsRouter. This means (7). 1766 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1767 * A quetion arises for (1) case. (1) case has no lladdr in the 1768 * neighbor cache, this is similar to (6). 1769 * This case is rare but we figured that we MUST NOT set IsRouter. 1770 * 1771 * is_new old_addr new_addr NS RS RA redir 1772 * D R 1773 * 0 n n (1) c ? s 1774 * 0 y n (2) c s s 1775 * 0 n y (3) c s s 1776 * 0 y y (4) c s s 1777 * 0 y y (5) c s s 1778 * 1 -- n (6) c c c s 1779 * 1 -- y (7) c c s c s 1780 * 1781 * (c=clear s=set) 1782 */ 1783 switch (type & 0xff) { 1784 case ND_NEIGHBOR_SOLICIT: 1785 /* 1786 * New entry must have is_router flag cleared. 1787 */ 1788 if (is_new) /* (6-7) */ 1789 ln_router = 0; 1790 break; 1791 case ND_REDIRECT: 1792 /* 1793 * If the icmp is a redirect to a better router, always set the 1794 * is_router flag. Otherwise, if the entry is newly created, 1795 * clear the flag. [RFC 2461, sec 8.3] 1796 */ 1797 if (code == ND_REDIRECT_ROUTER) 1798 ln_router = 1; 1799 else { 1800 if (is_new) /* (6-7) */ 1801 ln_router = 0; 1802 } 1803 break; 1804 case ND_ROUTER_SOLICIT: 1805 /* 1806 * is_router flag must always be cleared. 1807 */ 1808 ln_router = 0; 1809 break; 1810 case ND_ROUTER_ADVERT: 1811 /* 1812 * Mark an entry with lladdr as a router. 1813 */ 1814 if ((!is_new && (old_addr || new_addr)) || /* (2-5) */ 1815 (is_new && new_addr)) { /* (7) */ 1816 ln_router = 1; 1817 } 1818 break; 1819 } 1820 1821 return (ln_router); 1822 } 1823 1824 /* 1825 * Create neighbor cache entry and cache link-layer address, 1826 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1827 * 1828 * type - ICMP6 type 1829 * code - type dependent information 1830 * 1831 */ 1832 void 1833 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, 1834 int lladdrlen, int type, int code) 1835 { 1836 struct llentry *ln = NULL, *ln_tmp; 1837 int is_newentry; 1838 int do_update; 1839 int olladdr; 1840 int llchange; 1841 int flags; 1842 uint16_t router = 0; 1843 struct sockaddr_in6 sin6; 1844 struct mbuf *chain = NULL; 1845 1846 IF_AFDATA_UNLOCK_ASSERT(ifp); 1847 1848 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__)); 1849 KASSERT(from != NULL, ("%s: from == NULL", __func__)); 1850 1851 /* nothing must be updated for unspecified address */ 1852 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1853 return; 1854 1855 /* 1856 * Validation about ifp->if_addrlen and lladdrlen must be done in 1857 * the caller. 1858 * 1859 * XXX If the link does not have link-layer adderss, what should 1860 * we do? (ifp->if_addrlen == 0) 1861 * Spec says nothing in sections for RA, RS and NA. There's small 1862 * description on it in NS section (RFC 2461 7.2.3). 1863 */ 1864 flags = lladdr ? LLE_EXCLUSIVE : 0; 1865 IF_AFDATA_RLOCK(ifp); 1866 ln = nd6_lookup(from, flags, ifp); 1867 IF_AFDATA_RUNLOCK(ifp); 1868 is_newentry = 0; 1869 if (ln == NULL) { 1870 flags |= LLE_EXCLUSIVE; 1871 ln = nd6_alloc(from, 0, ifp); 1872 if (ln == NULL) 1873 return; 1874 1875 /* 1876 * Since we already know all the data for the new entry, 1877 * fill it before insertion. 1878 */ 1879 if (lladdr != NULL) 1880 lltable_set_entry_addr(ifp, ln, lladdr); 1881 IF_AFDATA_WLOCK(ifp); 1882 LLE_WLOCK(ln); 1883 /* Prefer any existing lle over newly-created one */ 1884 ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp); 1885 if (ln_tmp == NULL) 1886 lltable_link_entry(LLTABLE6(ifp), ln); 1887 IF_AFDATA_WUNLOCK(ifp); 1888 if (ln_tmp == NULL) { 1889 /* No existing lle, mark as new entry (6,7) */ 1890 is_newentry = 1; 1891 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); 1892 if (lladdr != NULL) /* (7) */ 1893 EVENTHANDLER_INVOKE(lle_event, ln, 1894 LLENTRY_RESOLVED); 1895 } else { 1896 lltable_free_entry(LLTABLE6(ifp), ln); 1897 ln = ln_tmp; 1898 ln_tmp = NULL; 1899 } 1900 } 1901 /* do nothing if static ndp is set */ 1902 if ((ln->la_flags & LLE_STATIC)) { 1903 if (flags & LLE_EXCLUSIVE) 1904 LLE_WUNLOCK(ln); 1905 else 1906 LLE_RUNLOCK(ln); 1907 return; 1908 } 1909 1910 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0; 1911 if (olladdr && lladdr) { 1912 llchange = bcmp(lladdr, &ln->ll_addr, 1913 ifp->if_addrlen); 1914 } else if (!olladdr && lladdr) 1915 llchange = 1; 1916 else 1917 llchange = 0; 1918 1919 /* 1920 * newentry olladdr lladdr llchange (*=record) 1921 * 0 n n -- (1) 1922 * 0 y n -- (2) 1923 * 0 n y y (3) * STALE 1924 * 0 y y n (4) * 1925 * 0 y y y (5) * STALE 1926 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1927 * 1 -- y -- (7) * STALE 1928 */ 1929 1930 do_update = 0; 1931 if (is_newentry == 0 && llchange != 0) { 1932 do_update = 1; /* (3,5) */ 1933 1934 /* 1935 * Record source link-layer address 1936 * XXX is it dependent to ifp->if_type? 1937 */ 1938 if (lltable_try_set_entry_addr(ifp, ln, lladdr) == 0) { 1939 /* Entry was deleted */ 1940 return; 1941 } 1942 1943 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); 1944 1945 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 1946 1947 if (ln->la_hold != NULL) 1948 nd6_grab_holdchain(ln, &chain, &sin6); 1949 } 1950 1951 /* Calculates new router status */ 1952 router = nd6_is_router(type, code, is_newentry, olladdr, 1953 lladdr != NULL ? 1 : 0, ln->ln_router); 1954 1955 ln->ln_router = router; 1956 /* Mark non-router redirects with special flag */ 1957 if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER) 1958 ln->la_flags |= LLE_REDIRECT; 1959 1960 if (flags & LLE_EXCLUSIVE) 1961 LLE_WUNLOCK(ln); 1962 else 1963 LLE_RUNLOCK(ln); 1964 1965 if (chain != NULL) 1966 nd6_flush_holdchain(ifp, ifp, chain, &sin6); 1967 1968 /* 1969 * When the link-layer address of a router changes, select the 1970 * best router again. In particular, when the neighbor entry is newly 1971 * created, it might affect the selection policy. 1972 * Question: can we restrict the first condition to the "is_newentry" 1973 * case? 1974 * XXX: when we hear an RA from a new router with the link-layer 1975 * address option, defrouter_select() is called twice, since 1976 * defrtrlist_update called the function as well. However, I believe 1977 * we can compromise the overhead, since it only happens the first 1978 * time. 1979 * XXX: although defrouter_select() should not have a bad effect 1980 * for those are not autoconfigured hosts, we explicitly avoid such 1981 * cases for safety. 1982 */ 1983 if ((do_update || is_newentry) && router && 1984 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1985 /* 1986 * guaranteed recursion 1987 */ 1988 defrouter_select(); 1989 } 1990 } 1991 1992 static void 1993 nd6_slowtimo(void *arg) 1994 { 1995 CURVNET_SET((struct vnet *) arg); 1996 struct nd_ifinfo *nd6if; 1997 struct ifnet *ifp; 1998 1999 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 2000 nd6_slowtimo, curvnet); 2001 IFNET_RLOCK_NOSLEEP(); 2002 TAILQ_FOREACH(ifp, &V_ifnet, if_link) { 2003 if (ifp->if_afdata[AF_INET6] == NULL) 2004 continue; 2005 nd6if = ND_IFINFO(ifp); 2006 if (nd6if->basereachable && /* already initialized */ 2007 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 2008 /* 2009 * Since reachable time rarely changes by router 2010 * advertisements, we SHOULD insure that a new random 2011 * value gets recomputed at least once every few hours. 2012 * (RFC 2461, 6.3.4) 2013 */ 2014 nd6if->recalctm = V_nd6_recalc_reachtm_interval; 2015 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 2016 } 2017 } 2018 IFNET_RUNLOCK_NOSLEEP(); 2019 CURVNET_RESTORE(); 2020 } 2021 2022 void 2023 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain, 2024 struct sockaddr_in6 *sin6) 2025 { 2026 2027 LLE_WLOCK_ASSERT(ln); 2028 2029 *chain = ln->la_hold; 2030 ln->la_hold = NULL; 2031 lltable_fill_sa_entry(ln, (struct sockaddr *)sin6); 2032 2033 if (ln->ln_state == ND6_LLINFO_STALE) { 2034 2035 /* 2036 * The first time we send a packet to a 2037 * neighbor whose entry is STALE, we have 2038 * to change the state to DELAY and a sets 2039 * a timer to expire in DELAY_FIRST_PROBE_TIME 2040 * seconds to ensure do neighbor unreachability 2041 * detection on expiration. 2042 * (RFC 2461 7.3.3) 2043 */ 2044 nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY); 2045 } 2046 } 2047 2048 int 2049 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m, 2050 struct sockaddr_in6 *dst, struct route *ro) 2051 { 2052 int error; 2053 int ip6len; 2054 struct ip6_hdr *ip6; 2055 struct m_tag *mtag; 2056 2057 #ifdef MAC 2058 mac_netinet6_nd6_send(ifp, m); 2059 #endif 2060 2061 /* 2062 * If called from nd6_ns_output() (NS), nd6_na_output() (NA), 2063 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA 2064 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND 2065 * to be diverted to user space. When re-injected into the kernel, 2066 * send_output() will directly dispatch them to the outgoing interface. 2067 */ 2068 if (send_sendso_input_hook != NULL) { 2069 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); 2070 if (mtag != NULL) { 2071 ip6 = mtod(m, struct ip6_hdr *); 2072 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); 2073 /* Use the SEND socket */ 2074 error = send_sendso_input_hook(m, ifp, SND_OUT, 2075 ip6len); 2076 /* -1 == no app on SEND socket */ 2077 if (error == 0 || error != -1) 2078 return (error); 2079 } 2080 } 2081 2082 m_clrprotoflags(m); /* Avoid confusing lower layers. */ 2083 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL, 2084 mtod(m, struct ip6_hdr *)); 2085 2086 if ((ifp->if_flags & IFF_LOOPBACK) == 0) 2087 origifp = ifp; 2088 2089 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro); 2090 return (error); 2091 } 2092 2093 /* 2094 * Do L2 address resolution for @sa_dst address. Stores found 2095 * address in @desten buffer. Copy of lle ln_flags can be also 2096 * saved in @pflags if @pflags is non-NULL. 2097 * 2098 * If destination LLE does not exists or lle state modification 2099 * is required, call "slow" version. 2100 * 2101 * Return values: 2102 * - 0 on success (address copied to buffer). 2103 * - EWOULDBLOCK (no local error, but address is still unresolved) 2104 * - other errors (alloc failure, etc) 2105 */ 2106 int 2107 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m, 2108 const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags) 2109 { 2110 struct llentry *ln = NULL; 2111 const struct sockaddr_in6 *dst6; 2112 2113 if (pflags != NULL) 2114 *pflags = 0; 2115 2116 dst6 = (const struct sockaddr_in6 *)sa_dst; 2117 2118 /* discard the packet if IPv6 operation is disabled on the interface */ 2119 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 2120 m_freem(m); 2121 return (ENETDOWN); /* better error? */ 2122 } 2123 2124 if (m != NULL && m->m_flags & M_MCAST) { 2125 switch (ifp->if_type) { 2126 case IFT_ETHER: 2127 case IFT_FDDI: 2128 case IFT_L2VLAN: 2129 case IFT_IEEE80211: 2130 case IFT_BRIDGE: 2131 case IFT_ISO88025: 2132 ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr, 2133 desten); 2134 return (0); 2135 default: 2136 m_freem(m); 2137 return (EAFNOSUPPORT); 2138 } 2139 } 2140 2141 IF_AFDATA_RLOCK(ifp); 2142 ln = nd6_lookup(&dst6->sin6_addr, LLE_UNLOCKED, ifp); 2143 if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) { 2144 /* Entry found, let's copy lle info */ 2145 bcopy(&ln->ll_addr, desten, ifp->if_addrlen); 2146 if (pflags != NULL) 2147 *pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR); 2148 /* Check if we have feedback request from nd6 timer */ 2149 if (ln->r_skip_req != 0) { 2150 LLE_REQ_LOCK(ln); 2151 ln->r_skip_req = 0; /* Notify that entry was used */ 2152 ln->lle_hittime = time_uptime; 2153 LLE_REQ_UNLOCK(ln); 2154 } 2155 IF_AFDATA_RUNLOCK(ifp); 2156 return (0); 2157 } 2158 IF_AFDATA_RUNLOCK(ifp); 2159 2160 return (nd6_resolve_slow(ifp, m, dst6, desten, pflags)); 2161 } 2162 2163 2164 /* 2165 * Do L2 address resolution for @sa_dst address. Stores found 2166 * address in @desten buffer. Copy of lle ln_flags can be also 2167 * saved in @pflags if @pflags is non-NULL. 2168 * 2169 * Heavy version. 2170 * Function assume that destination LLE does not exist, 2171 * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired. 2172 * 2173 * Set noinline to be dtrace-friendly 2174 */ 2175 static __noinline int 2176 nd6_resolve_slow(struct ifnet *ifp, struct mbuf *m, 2177 const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags) 2178 { 2179 struct llentry *lle = NULL, *lle_tmp; 2180 struct in6_addr *psrc, src; 2181 int send_ns; 2182 2183 /* 2184 * Address resolution or Neighbor Unreachability Detection 2185 * for the next hop. 2186 * At this point, the destination of the packet must be a unicast 2187 * or an anycast address(i.e. not a multicast). 2188 */ 2189 if (lle == NULL) { 2190 IF_AFDATA_RLOCK(ifp); 2191 lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp); 2192 IF_AFDATA_RUNLOCK(ifp); 2193 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) { 2194 /* 2195 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 2196 * the condition below is not very efficient. But we believe 2197 * it is tolerable, because this should be a rare case. 2198 */ 2199 lle = nd6_alloc(&dst->sin6_addr, 0, ifp); 2200 if (lle == NULL) { 2201 char ip6buf[INET6_ADDRSTRLEN]; 2202 log(LOG_DEBUG, 2203 "nd6_output: can't allocate llinfo for %s " 2204 "(ln=%p)\n", 2205 ip6_sprintf(ip6buf, &dst->sin6_addr), lle); 2206 m_freem(m); 2207 return (ENOBUFS); 2208 } 2209 2210 IF_AFDATA_WLOCK(ifp); 2211 LLE_WLOCK(lle); 2212 /* Prefer any existing entry over newly-created one */ 2213 lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp); 2214 if (lle_tmp == NULL) 2215 lltable_link_entry(LLTABLE6(ifp), lle); 2216 IF_AFDATA_WUNLOCK(ifp); 2217 if (lle_tmp != NULL) { 2218 lltable_free_entry(LLTABLE6(ifp), lle); 2219 lle = lle_tmp; 2220 lle_tmp = NULL; 2221 } 2222 } 2223 } 2224 if (lle == NULL) { 2225 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 2226 m_freem(m); 2227 return (ENOBUFS); 2228 } 2229 2230 if (m != NULL) 2231 m_freem(m); 2232 return (ENOBUFS); 2233 } 2234 2235 LLE_WLOCK_ASSERT(lle); 2236 2237 /* 2238 * The first time we send a packet to a neighbor whose entry is 2239 * STALE, we have to change the state to DELAY and a sets a timer to 2240 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 2241 * neighbor unreachability detection on expiration. 2242 * (RFC 2461 7.3.3) 2243 */ 2244 if (lle->ln_state == ND6_LLINFO_STALE) 2245 nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY); 2246 2247 /* 2248 * If the neighbor cache entry has a state other than INCOMPLETE 2249 * (i.e. its link-layer address is already resolved), just 2250 * send the packet. 2251 */ 2252 if (lle->ln_state > ND6_LLINFO_INCOMPLETE) { 2253 bcopy(&lle->ll_addr, desten, ifp->if_addrlen); 2254 if (pflags != NULL) 2255 *pflags = lle->la_flags; 2256 LLE_WUNLOCK(lle); 2257 return (0); 2258 } 2259 2260 /* 2261 * There is a neighbor cache entry, but no ethernet address 2262 * response yet. Append this latest packet to the end of the 2263 * packet queue in the mbuf, unless the number of the packet 2264 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 2265 * the oldest packet in the queue will be removed. 2266 */ 2267 2268 if (lle->la_hold != NULL) { 2269 struct mbuf *m_hold; 2270 int i; 2271 2272 i = 0; 2273 for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){ 2274 i++; 2275 if (m_hold->m_nextpkt == NULL) { 2276 m_hold->m_nextpkt = m; 2277 break; 2278 } 2279 } 2280 while (i >= V_nd6_maxqueuelen) { 2281 m_hold = lle->la_hold; 2282 lle->la_hold = lle->la_hold->m_nextpkt; 2283 m_freem(m_hold); 2284 i--; 2285 } 2286 } else { 2287 lle->la_hold = m; 2288 } 2289 2290 /* 2291 * If there has been no NS for the neighbor after entering the 2292 * INCOMPLETE state, send the first solicitation. 2293 * Note that for newly-created lle la_asked will be 0, 2294 * so we will transition from ND6_LLINFO_NOSTATE to 2295 * ND6_LLINFO_INCOMPLETE state here. 2296 */ 2297 psrc = NULL; 2298 send_ns = 0; 2299 if (lle->la_asked == 0) { 2300 lle->la_asked++; 2301 send_ns = 1; 2302 psrc = nd6_llinfo_get_holdsrc(lle, &src); 2303 2304 nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE); 2305 } 2306 LLE_WUNLOCK(lle); 2307 if (send_ns != 0) 2308 nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL); 2309 2310 return (EWOULDBLOCK); 2311 } 2312 2313 2314 int 2315 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain, 2316 struct sockaddr_in6 *dst) 2317 { 2318 struct mbuf *m, *m_head; 2319 struct ifnet *outifp; 2320 int error = 0; 2321 2322 m_head = chain; 2323 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2324 outifp = origifp; 2325 else 2326 outifp = ifp; 2327 2328 while (m_head) { 2329 m = m_head; 2330 m_head = m_head->m_nextpkt; 2331 error = nd6_output_ifp(ifp, origifp, m, dst, NULL); 2332 } 2333 2334 /* 2335 * XXX 2336 * note that intermediate errors are blindly ignored 2337 */ 2338 return (error); 2339 } 2340 2341 static int 2342 nd6_need_cache(struct ifnet *ifp) 2343 { 2344 /* 2345 * XXX: we currently do not make neighbor cache on any interface 2346 * other than ARCnet, Ethernet, FDDI and GIF. 2347 * 2348 * RFC2893 says: 2349 * - unidirectional tunnels needs no ND 2350 */ 2351 switch (ifp->if_type) { 2352 case IFT_ARCNET: 2353 case IFT_ETHER: 2354 case IFT_FDDI: 2355 case IFT_IEEE1394: 2356 case IFT_L2VLAN: 2357 case IFT_IEEE80211: 2358 case IFT_INFINIBAND: 2359 case IFT_BRIDGE: 2360 case IFT_PROPVIRTUAL: 2361 return (1); 2362 default: 2363 return (0); 2364 } 2365 } 2366 2367 /* 2368 * Add pernament ND6 link-layer record for given 2369 * interface address. 2370 * 2371 * Very similar to IPv4 arp_ifinit(), but: 2372 * 1) IPv6 DAD is performed in different place 2373 * 2) It is called by IPv6 protocol stack in contrast to 2374 * arp_ifinit() which is typically called in SIOCSIFADDR 2375 * driver ioctl handler. 2376 * 2377 */ 2378 int 2379 nd6_add_ifa_lle(struct in6_ifaddr *ia) 2380 { 2381 struct ifnet *ifp; 2382 struct llentry *ln, *ln_tmp; 2383 struct sockaddr *dst; 2384 2385 ifp = ia->ia_ifa.ifa_ifp; 2386 if (nd6_need_cache(ifp) == 0) 2387 return (0); 2388 2389 ia->ia_ifa.ifa_rtrequest = nd6_rtrequest; 2390 dst = (struct sockaddr *)&ia->ia_addr; 2391 ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst); 2392 if (ln == NULL) 2393 return (ENOBUFS); 2394 2395 IF_AFDATA_WLOCK(ifp); 2396 LLE_WLOCK(ln); 2397 /* Unlink any entry if exists */ 2398 ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst); 2399 if (ln_tmp != NULL) 2400 lltable_unlink_entry(LLTABLE6(ifp), ln_tmp); 2401 lltable_link_entry(LLTABLE6(ifp), ln); 2402 IF_AFDATA_WUNLOCK(ifp); 2403 2404 if (ln_tmp != NULL) 2405 EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED); 2406 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 2407 2408 LLE_WUNLOCK(ln); 2409 if (ln_tmp != NULL) 2410 llentry_free(ln_tmp); 2411 2412 return (0); 2413 } 2414 2415 /* 2416 * Removes either all lle entries for given @ia, or lle 2417 * corresponding to @ia address. 2418 */ 2419 void 2420 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all) 2421 { 2422 struct sockaddr_in6 mask, addr; 2423 struct sockaddr *saddr, *smask; 2424 struct ifnet *ifp; 2425 2426 ifp = ia->ia_ifa.ifa_ifp; 2427 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr)); 2428 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask)); 2429 saddr = (struct sockaddr *)&addr; 2430 smask = (struct sockaddr *)&mask; 2431 2432 if (all != 0) 2433 lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC); 2434 else 2435 lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr); 2436 } 2437 2438 static void 2439 clear_llinfo_pqueue(struct llentry *ln) 2440 { 2441 struct mbuf *m_hold, *m_hold_next; 2442 2443 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) { 2444 m_hold_next = m_hold->m_nextpkt; 2445 m_freem(m_hold); 2446 } 2447 2448 ln->la_hold = NULL; 2449 return; 2450 } 2451 2452 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2453 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2454 #ifdef SYSCTL_DECL 2455 SYSCTL_DECL(_net_inet6_icmp6); 2456 #endif 2457 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2458 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2459 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2460 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2461 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2462 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, ""); 2463 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer, 2464 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), ""); 2465 2466 static int 2467 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2468 { 2469 struct in6_defrouter d; 2470 struct nd_defrouter *dr; 2471 int error; 2472 2473 if (req->newptr) 2474 return (EPERM); 2475 2476 bzero(&d, sizeof(d)); 2477 d.rtaddr.sin6_family = AF_INET6; 2478 d.rtaddr.sin6_len = sizeof(d.rtaddr); 2479 2480 /* 2481 * XXX locking 2482 */ 2483 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 2484 d.rtaddr.sin6_addr = dr->rtaddr; 2485 error = sa6_recoverscope(&d.rtaddr); 2486 if (error != 0) 2487 return (error); 2488 d.flags = dr->flags; 2489 d.rtlifetime = dr->rtlifetime; 2490 d.expire = dr->expire + (time_second - time_uptime); 2491 d.if_index = dr->ifp->if_index; 2492 error = SYSCTL_OUT(req, &d, sizeof(d)); 2493 if (error != 0) 2494 return (error); 2495 } 2496 return (0); 2497 } 2498 2499 static int 2500 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2501 { 2502 struct in6_prefix p; 2503 struct sockaddr_in6 s6; 2504 struct nd_prefix *pr; 2505 struct nd_pfxrouter *pfr; 2506 time_t maxexpire; 2507 int error; 2508 char ip6buf[INET6_ADDRSTRLEN]; 2509 2510 if (req->newptr) 2511 return (EPERM); 2512 2513 bzero(&p, sizeof(p)); 2514 p.origin = PR_ORIG_RA; 2515 bzero(&s6, sizeof(s6)); 2516 s6.sin6_family = AF_INET6; 2517 s6.sin6_len = sizeof(s6); 2518 2519 /* 2520 * XXX locking 2521 */ 2522 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 2523 p.prefix = pr->ndpr_prefix; 2524 if (sa6_recoverscope(&p.prefix)) { 2525 log(LOG_ERR, "scope error in prefix list (%s)\n", 2526 ip6_sprintf(ip6buf, &p.prefix.sin6_addr)); 2527 /* XXX: press on... */ 2528 } 2529 p.raflags = pr->ndpr_raf; 2530 p.prefixlen = pr->ndpr_plen; 2531 p.vltime = pr->ndpr_vltime; 2532 p.pltime = pr->ndpr_pltime; 2533 p.if_index = pr->ndpr_ifp->if_index; 2534 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2535 p.expire = 0; 2536 else { 2537 /* XXX: we assume time_t is signed. */ 2538 maxexpire = (-1) & 2539 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1)); 2540 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate) 2541 p.expire = pr->ndpr_lastupdate + 2542 pr->ndpr_vltime + 2543 (time_second - time_uptime); 2544 else 2545 p.expire = maxexpire; 2546 } 2547 p.refcnt = pr->ndpr_refcnt; 2548 p.flags = pr->ndpr_stateflags; 2549 p.advrtrs = 0; 2550 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) 2551 p.advrtrs++; 2552 error = SYSCTL_OUT(req, &p, sizeof(p)); 2553 if (error != 0) 2554 return (error); 2555 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 2556 s6.sin6_addr = pfr->router->rtaddr; 2557 if (sa6_recoverscope(&s6)) 2558 log(LOG_ERR, 2559 "scope error in prefix list (%s)\n", 2560 ip6_sprintf(ip6buf, &pfr->router->rtaddr)); 2561 error = SYSCTL_OUT(req, &s6, sizeof(s6)); 2562 if (error != 0) 2563 return (error); 2564 } 2565 } 2566 return (0); 2567 } 2568