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 if (ifp->if_afdata[AF_INET6] == NULL) 289 return; 290 291 nd6_setmtu0(ifp, ND_IFINFO(ifp)); 292 } 293 294 /* XXX todo: do not maintain copy of ifp->if_mtu in ndi->maxmtu */ 295 void 296 nd6_setmtu0(struct ifnet *ifp, struct nd_ifinfo *ndi) 297 { 298 u_int32_t omaxmtu; 299 300 omaxmtu = ndi->maxmtu; 301 302 switch (ifp->if_type) { 303 case IFT_ARCNET: 304 ndi->maxmtu = MIN(ARC_PHDS_MAXMTU, ifp->if_mtu); /* RFC2497 */ 305 break; 306 case IFT_FDDI: 307 ndi->maxmtu = MIN(FDDIIPMTU, ifp->if_mtu); /* RFC2467 */ 308 break; 309 case IFT_ISO88025: 310 ndi->maxmtu = MIN(ISO88025_MAX_MTU, ifp->if_mtu); 311 break; 312 default: 313 ndi->maxmtu = ifp->if_mtu; 314 break; 315 } 316 317 /* 318 * Decreasing the interface MTU under IPV6 minimum MTU may cause 319 * undesirable situation. We thus notify the operator of the change 320 * explicitly. The check for omaxmtu is necessary to restrict the 321 * log to the case of changing the MTU, not initializing it. 322 */ 323 if (omaxmtu >= IPV6_MMTU && ndi->maxmtu < IPV6_MMTU) { 324 log(LOG_NOTICE, "nd6_setmtu0: " 325 "new link MTU on %s (%lu) is too small for IPv6\n", 326 if_name(ifp), (unsigned long)ndi->maxmtu); 327 } 328 329 if (ndi->maxmtu > V_in6_maxmtu) 330 in6_setmaxmtu(); /* check all interfaces just in case */ 331 332 } 333 334 void 335 nd6_option_init(void *opt, int icmp6len, union nd_opts *ndopts) 336 { 337 338 bzero(ndopts, sizeof(*ndopts)); 339 ndopts->nd_opts_search = (struct nd_opt_hdr *)opt; 340 ndopts->nd_opts_last 341 = (struct nd_opt_hdr *)(((u_char *)opt) + icmp6len); 342 343 if (icmp6len == 0) { 344 ndopts->nd_opts_done = 1; 345 ndopts->nd_opts_search = NULL; 346 } 347 } 348 349 /* 350 * Take one ND option. 351 */ 352 struct nd_opt_hdr * 353 nd6_option(union nd_opts *ndopts) 354 { 355 struct nd_opt_hdr *nd_opt; 356 int olen; 357 358 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 359 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 360 __func__)); 361 if (ndopts->nd_opts_search == NULL) 362 return NULL; 363 if (ndopts->nd_opts_done) 364 return NULL; 365 366 nd_opt = ndopts->nd_opts_search; 367 368 /* make sure nd_opt_len is inside the buffer */ 369 if ((caddr_t)&nd_opt->nd_opt_len >= (caddr_t)ndopts->nd_opts_last) { 370 bzero(ndopts, sizeof(*ndopts)); 371 return NULL; 372 } 373 374 olen = nd_opt->nd_opt_len << 3; 375 if (olen == 0) { 376 /* 377 * Message validation requires that all included 378 * options have a length that is greater than zero. 379 */ 380 bzero(ndopts, sizeof(*ndopts)); 381 return NULL; 382 } 383 384 ndopts->nd_opts_search = (struct nd_opt_hdr *)((caddr_t)nd_opt + olen); 385 if (ndopts->nd_opts_search > ndopts->nd_opts_last) { 386 /* option overruns the end of buffer, invalid */ 387 bzero(ndopts, sizeof(*ndopts)); 388 return NULL; 389 } else if (ndopts->nd_opts_search == ndopts->nd_opts_last) { 390 /* reached the end of options chain */ 391 ndopts->nd_opts_done = 1; 392 ndopts->nd_opts_search = NULL; 393 } 394 return nd_opt; 395 } 396 397 /* 398 * Parse multiple ND options. 399 * This function is much easier to use, for ND routines that do not need 400 * multiple options of the same type. 401 */ 402 int 403 nd6_options(union nd_opts *ndopts) 404 { 405 struct nd_opt_hdr *nd_opt; 406 int i = 0; 407 408 KASSERT(ndopts != NULL, ("%s: ndopts == NULL", __func__)); 409 KASSERT(ndopts->nd_opts_last != NULL, ("%s: uninitialized ndopts", 410 __func__)); 411 if (ndopts->nd_opts_search == NULL) 412 return 0; 413 414 while (1) { 415 nd_opt = nd6_option(ndopts); 416 if (nd_opt == NULL && ndopts->nd_opts_last == NULL) { 417 /* 418 * Message validation requires that all included 419 * options have a length that is greater than zero. 420 */ 421 ICMP6STAT_INC(icp6s_nd_badopt); 422 bzero(ndopts, sizeof(*ndopts)); 423 return -1; 424 } 425 426 if (nd_opt == NULL) 427 goto skip1; 428 429 switch (nd_opt->nd_opt_type) { 430 case ND_OPT_SOURCE_LINKADDR: 431 case ND_OPT_TARGET_LINKADDR: 432 case ND_OPT_MTU: 433 case ND_OPT_REDIRECTED_HEADER: 434 case ND_OPT_NONCE: 435 if (ndopts->nd_opt_array[nd_opt->nd_opt_type]) { 436 nd6log((LOG_INFO, 437 "duplicated ND6 option found (type=%d)\n", 438 nd_opt->nd_opt_type)); 439 /* XXX bark? */ 440 } else { 441 ndopts->nd_opt_array[nd_opt->nd_opt_type] 442 = nd_opt; 443 } 444 break; 445 case ND_OPT_PREFIX_INFORMATION: 446 if (ndopts->nd_opt_array[nd_opt->nd_opt_type] == 0) { 447 ndopts->nd_opt_array[nd_opt->nd_opt_type] 448 = nd_opt; 449 } 450 ndopts->nd_opts_pi_end = 451 (struct nd_opt_prefix_info *)nd_opt; 452 break; 453 /* What about ND_OPT_ROUTE_INFO? RFC 4191 */ 454 case ND_OPT_RDNSS: /* RFC 6106 */ 455 case ND_OPT_DNSSL: /* RFC 6106 */ 456 /* 457 * Silently ignore options we know and do not care about 458 * in the kernel. 459 */ 460 break; 461 default: 462 /* 463 * Unknown options must be silently ignored, 464 * to accomodate future extension to the protocol. 465 */ 466 nd6log((LOG_DEBUG, 467 "nd6_options: unsupported option %d - " 468 "option ignored\n", nd_opt->nd_opt_type)); 469 } 470 471 skip1: 472 i++; 473 if (i > V_nd6_maxndopt) { 474 ICMP6STAT_INC(icp6s_nd_toomanyopt); 475 nd6log((LOG_INFO, "too many loop in nd opt\n")); 476 break; 477 } 478 479 if (ndopts->nd_opts_done) 480 break; 481 } 482 483 return 0; 484 } 485 486 /* 487 * ND6 timer routine to handle ND6 entries 488 */ 489 static void 490 nd6_llinfo_settimer_locked(struct llentry *ln, long tick) 491 { 492 int canceled; 493 494 LLE_WLOCK_ASSERT(ln); 495 496 if (tick < 0) { 497 ln->la_expire = 0; 498 ln->ln_ntick = 0; 499 canceled = callout_stop(&ln->lle_timer); 500 } else { 501 ln->la_expire = time_uptime + tick / hz; 502 LLE_ADDREF(ln); 503 if (tick > INT_MAX) { 504 ln->ln_ntick = tick - INT_MAX; 505 canceled = callout_reset(&ln->lle_timer, INT_MAX, 506 nd6_llinfo_timer, ln); 507 } else { 508 ln->ln_ntick = 0; 509 canceled = callout_reset(&ln->lle_timer, tick, 510 nd6_llinfo_timer, ln); 511 } 512 } 513 if (canceled > 0) 514 LLE_REMREF(ln); 515 } 516 517 /* 518 * Gets source address of the first packet in hold queue 519 * and stores it in @src. 520 * Returns pointer to @src (if hold queue is not empty) or NULL. 521 * 522 * Set noinline to be dtrace-friendly 523 */ 524 static __noinline struct in6_addr * 525 nd6_llinfo_get_holdsrc(struct llentry *ln, struct in6_addr *src) 526 { 527 struct ip6_hdr hdr; 528 struct mbuf *m; 529 530 if (ln->la_hold == NULL) 531 return (NULL); 532 533 /* 534 * assume every packet in la_hold has the same IP header 535 */ 536 m = ln->la_hold; 537 if (sizeof(hdr) > m->m_len) 538 return (NULL); 539 540 m_copydata(m, 0, sizeof(hdr), (caddr_t)&hdr); 541 *src = hdr.ip6_src; 542 543 return (src); 544 } 545 546 /* 547 * Checks if we need to switch from STALE state. 548 * 549 * RFC 4861 requires switching from STALE to DELAY state 550 * on first packet matching entry, waiting V_nd6_delay and 551 * transition to PROBE state (if upper layer confirmation was 552 * not received). 553 * 554 * This code performs a bit differently: 555 * On packet hit we don't change state (but desired state 556 * can be guessed by control plane). However, after V_nd6_delay 557 * seconds code will transition to PROBE state (so DELAY state 558 * is kinda skipped in most situations). 559 * 560 * Typically, V_nd6_gctimer is bigger than V_nd6_delay, so 561 * we perform the following upon entering STALE state: 562 * 563 * 1) Arm timer to run each V_nd6_delay seconds to make sure that 564 * if packet was transmitted at the start of given interval, we 565 * would be able to switch to PROBE state in V_nd6_delay seconds 566 * as user expects. 567 * 568 * 2) Reschedule timer until original V_nd6_gctimer expires keeping 569 * lle in STALE state (remaining timer value stored in lle_remtime). 570 * 571 * 3) Reschedule timer if packet was transmitted less that V_nd6_delay 572 * seconds ago. 573 * 574 * Returns non-zero value if the entry is still STALE (storing 575 * the next timer interval in @pdelay). 576 * 577 * Returns zero value if original timer expired or we need to switch to 578 * PROBE (store that in @do_switch variable). 579 */ 580 static int 581 nd6_is_stale(struct llentry *lle, long *pdelay, int *do_switch) 582 { 583 int nd_delay, nd_gctimer, r_skip_req; 584 time_t lle_hittime; 585 long delay; 586 587 *do_switch = 0; 588 nd_gctimer = V_nd6_gctimer; 589 nd_delay = V_nd6_delay; 590 591 LLE_REQ_LOCK(lle); 592 r_skip_req = lle->r_skip_req; 593 lle_hittime = lle->lle_hittime; 594 LLE_REQ_UNLOCK(lle); 595 596 if (r_skip_req > 0) { 597 598 /* 599 * Nonzero r_skip_req value was set upon entering 600 * STALE state. Since value was not changed, no 601 * packets were passed using this lle. Ask for 602 * timer reschedule and keep STALE state. 603 */ 604 delay = (long)(MIN(nd_gctimer, nd_delay)); 605 delay *= hz; 606 if (lle->lle_remtime > delay) 607 lle->lle_remtime -= delay; 608 else { 609 delay = lle->lle_remtime; 610 lle->lle_remtime = 0; 611 } 612 613 if (delay == 0) { 614 615 /* 616 * The original ng6_gctime timeout ended, 617 * no more rescheduling. 618 */ 619 return (0); 620 } 621 622 *pdelay = delay; 623 return (1); 624 } 625 626 /* 627 * Packet received. Verify timestamp 628 */ 629 delay = (long)(time_uptime - lle_hittime); 630 if (delay < nd_delay) { 631 632 /* 633 * V_nd6_delay still not passed since the first 634 * hit in STALE state. 635 * Reshedule timer and return. 636 */ 637 *pdelay = (long)(nd_delay - delay) * hz; 638 return (1); 639 } 640 641 /* Request switching to probe */ 642 *do_switch = 1; 643 return (0); 644 } 645 646 647 /* 648 * Switch @lle state to new state optionally arming timers. 649 * 650 * Set noinline to be dtrace-friendly 651 */ 652 __noinline void 653 nd6_llinfo_setstate(struct llentry *lle, int newstate) 654 { 655 struct ifnet *ifp; 656 int nd_gctimer, nd_delay; 657 long delay, remtime; 658 659 delay = 0; 660 remtime = 0; 661 662 switch (newstate) { 663 case ND6_LLINFO_INCOMPLETE: 664 ifp = lle->lle_tbl->llt_ifp; 665 delay = (long)ND_IFINFO(ifp)->retrans * hz / 1000; 666 break; 667 case ND6_LLINFO_REACHABLE: 668 if (!ND6_LLINFO_PERMANENT(lle)) { 669 ifp = lle->lle_tbl->llt_ifp; 670 delay = (long)ND_IFINFO(ifp)->reachable * hz; 671 } 672 break; 673 case ND6_LLINFO_STALE: 674 675 /* 676 * Notify fast path that we want to know if any packet 677 * is transmitted by setting r_skip_req. 678 */ 679 LLE_REQ_LOCK(lle); 680 lle->r_skip_req = 1; 681 LLE_REQ_UNLOCK(lle); 682 nd_delay = V_nd6_delay; 683 nd_gctimer = V_nd6_gctimer; 684 685 delay = (long)(MIN(nd_gctimer, nd_delay)) * hz; 686 remtime = (long)nd_gctimer * hz - delay; 687 break; 688 case ND6_LLINFO_DELAY: 689 lle->la_asked = 0; 690 delay = (long)V_nd6_delay * hz; 691 break; 692 } 693 694 if (delay > 0) 695 nd6_llinfo_settimer_locked(lle, delay); 696 697 lle->lle_remtime = remtime; 698 lle->ln_state = newstate; 699 } 700 701 /* 702 * Timer-dependent part of nd state machine. 703 * 704 * Set noinline to be dtrace-friendly 705 */ 706 static __noinline void 707 nd6_llinfo_timer(void *arg) 708 { 709 struct llentry *ln; 710 struct in6_addr *dst, *pdst, *psrc, src; 711 struct ifnet *ifp; 712 struct nd_ifinfo *ndi = NULL; 713 int do_switch, send_ns; 714 long delay; 715 716 KASSERT(arg != NULL, ("%s: arg NULL", __func__)); 717 ln = (struct llentry *)arg; 718 LLE_WLOCK(ln); 719 if (callout_pending(&ln->lle_timer)) { 720 /* 721 * Here we are a bit odd here in the treatment of 722 * active/pending. If the pending bit is set, it got 723 * rescheduled before I ran. The active 724 * bit we ignore, since if it was stopped 725 * in ll_tablefree() and was currently running 726 * it would have return 0 so the code would 727 * not have deleted it since the callout could 728 * not be stopped so we want to go through 729 * with the delete here now. If the callout 730 * was restarted, the pending bit will be back on and 731 * we just want to bail since the callout_reset would 732 * return 1 and our reference would have been removed 733 * by nd6_llinfo_settimer_locked above since canceled 734 * would have been 1. 735 */ 736 LLE_WUNLOCK(ln); 737 return; 738 } 739 ifp = ln->lle_tbl->llt_ifp; 740 CURVNET_SET(ifp->if_vnet); 741 ndi = ND_IFINFO(ifp); 742 send_ns = 0; 743 dst = &ln->r_l3addr.addr6; 744 pdst = dst; 745 746 if (ln->ln_ntick > 0) { 747 if (ln->ln_ntick > INT_MAX) { 748 ln->ln_ntick -= INT_MAX; 749 nd6_llinfo_settimer_locked(ln, INT_MAX); 750 } else { 751 ln->ln_ntick = 0; 752 nd6_llinfo_settimer_locked(ln, ln->ln_ntick); 753 } 754 goto done; 755 } 756 757 if (ln->la_flags & LLE_STATIC) { 758 goto done; 759 } 760 761 if (ln->la_flags & LLE_DELETED) { 762 nd6_free(ln, 0); 763 ln = NULL; 764 goto done; 765 } 766 767 switch (ln->ln_state) { 768 case ND6_LLINFO_INCOMPLETE: 769 if (ln->la_asked < V_nd6_mmaxtries) { 770 ln->la_asked++; 771 send_ns = 1; 772 /* Send NS to multicast address */ 773 pdst = NULL; 774 } else { 775 struct mbuf *m = ln->la_hold; 776 if (m) { 777 struct mbuf *m0; 778 779 /* 780 * assuming every packet in la_hold has the 781 * same IP header. Send error after unlock. 782 */ 783 m0 = m->m_nextpkt; 784 m->m_nextpkt = NULL; 785 ln->la_hold = m0; 786 clear_llinfo_pqueue(ln); 787 } 788 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_TIMEDOUT); 789 nd6_free(ln, 0); 790 ln = NULL; 791 if (m != NULL) 792 icmp6_error2(m, ICMP6_DST_UNREACH, 793 ICMP6_DST_UNREACH_ADDR, 0, ifp); 794 } 795 break; 796 case ND6_LLINFO_REACHABLE: 797 if (!ND6_LLINFO_PERMANENT(ln)) 798 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); 799 break; 800 801 case ND6_LLINFO_STALE: 802 if (nd6_is_stale(ln, &delay, &do_switch) != 0) { 803 804 /* 805 * No packet has used this entry and GC timeout 806 * has not been passed. Reshedule timer and 807 * return. 808 */ 809 nd6_llinfo_settimer_locked(ln, delay); 810 break; 811 } 812 813 if (do_switch == 0) { 814 815 /* 816 * GC timer has ended and entry hasn't been used. 817 * Run Garbage collector (RFC 4861, 5.3) 818 */ 819 if (!ND6_LLINFO_PERMANENT(ln)) { 820 EVENTHANDLER_INVOKE(lle_event, ln, 821 LLENTRY_EXPIRED); 822 nd6_free(ln, 1); 823 ln = NULL; 824 } 825 break; 826 } 827 828 /* Entry has been used AND delay timer has ended. */ 829 830 /* FALLTHROUGH */ 831 832 case ND6_LLINFO_DELAY: 833 if (ndi && (ndi->flags & ND6_IFF_PERFORMNUD) != 0) { 834 /* We need NUD */ 835 ln->la_asked = 1; 836 nd6_llinfo_setstate(ln, ND6_LLINFO_PROBE); 837 send_ns = 1; 838 } else 839 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); /* XXX */ 840 break; 841 case ND6_LLINFO_PROBE: 842 if (ln->la_asked < V_nd6_umaxtries) { 843 ln->la_asked++; 844 send_ns = 1; 845 } else { 846 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_EXPIRED); 847 nd6_free(ln, 0); 848 ln = NULL; 849 } 850 break; 851 default: 852 panic("%s: paths in a dark night can be confusing: %d", 853 __func__, ln->ln_state); 854 } 855 done: 856 if (send_ns != 0) { 857 nd6_llinfo_settimer_locked(ln, (long)ndi->retrans * hz / 1000); 858 psrc = nd6_llinfo_get_holdsrc(ln, &src); 859 LLE_FREE_LOCKED(ln); 860 ln = NULL; 861 nd6_ns_output(ifp, psrc, pdst, dst, NULL); 862 } 863 864 if (ln != NULL) 865 LLE_FREE_LOCKED(ln); 866 CURVNET_RESTORE(); 867 } 868 869 870 /* 871 * ND6 timer routine to expire default route list and prefix list 872 */ 873 void 874 nd6_timer(void *arg) 875 { 876 CURVNET_SET((struct vnet *) arg); 877 struct nd_defrouter *dr, *ndr; 878 struct nd_prefix *pr, *npr; 879 struct in6_ifaddr *ia6, *nia6; 880 881 callout_reset(&V_nd6_timer_ch, V_nd6_prune * hz, 882 nd6_timer, curvnet); 883 884 /* expire default router list */ 885 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 886 if (dr->expire && dr->expire < time_uptime) 887 defrtrlist_del(dr); 888 } 889 890 /* 891 * expire interface addresses. 892 * in the past the loop was inside prefix expiry processing. 893 * However, from a stricter speci-confrmance standpoint, we should 894 * rather separate address lifetimes and prefix lifetimes. 895 * 896 * XXXRW: in6_ifaddrhead locking. 897 */ 898 addrloop: 899 TAILQ_FOREACH_SAFE(ia6, &V_in6_ifaddrhead, ia_link, nia6) { 900 /* check address lifetime */ 901 if (IFA6_IS_INVALID(ia6)) { 902 int regen = 0; 903 904 /* 905 * If the expiring address is temporary, try 906 * regenerating a new one. This would be useful when 907 * we suspended a laptop PC, then turned it on after a 908 * period that could invalidate all temporary 909 * addresses. Although we may have to restart the 910 * loop (see below), it must be after purging the 911 * address. Otherwise, we'd see an infinite loop of 912 * regeneration. 913 */ 914 if (V_ip6_use_tempaddr && 915 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0) { 916 if (regen_tmpaddr(ia6) == 0) 917 regen = 1; 918 } 919 920 in6_purgeaddr(&ia6->ia_ifa); 921 922 if (regen) 923 goto addrloop; /* XXX: see below */ 924 } else if (IFA6_IS_DEPRECATED(ia6)) { 925 int oldflags = ia6->ia6_flags; 926 927 ia6->ia6_flags |= IN6_IFF_DEPRECATED; 928 929 /* 930 * If a temporary address has just become deprecated, 931 * regenerate a new one if possible. 932 */ 933 if (V_ip6_use_tempaddr && 934 (ia6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 935 (oldflags & IN6_IFF_DEPRECATED) == 0) { 936 937 if (regen_tmpaddr(ia6) == 0) { 938 /* 939 * A new temporary address is 940 * generated. 941 * XXX: this means the address chain 942 * has changed while we are still in 943 * the loop. Although the change 944 * would not cause disaster (because 945 * it's not a deletion, but an 946 * addition,) we'd rather restart the 947 * loop just for safety. Or does this 948 * significantly reduce performance?? 949 */ 950 goto addrloop; 951 } 952 } 953 } else if ((ia6->ia6_flags & IN6_IFF_TENTATIVE) != 0) { 954 /* 955 * Schedule DAD for a tentative address. This happens 956 * if the interface was down or not running 957 * when the address was configured. 958 */ 959 int delay; 960 961 delay = arc4random() % 962 (MAX_RTR_SOLICITATION_DELAY * hz); 963 nd6_dad_start((struct ifaddr *)ia6, delay); 964 } else { 965 /* 966 * Check status of the interface. If it is down, 967 * mark the address as tentative for future DAD. 968 */ 969 if ((ia6->ia_ifp->if_flags & IFF_UP) == 0 || 970 (ia6->ia_ifp->if_drv_flags & IFF_DRV_RUNNING) 971 == 0 || 972 (ND_IFINFO(ia6->ia_ifp)->flags & 973 ND6_IFF_IFDISABLED) != 0) { 974 ia6->ia6_flags &= ~IN6_IFF_DUPLICATED; 975 ia6->ia6_flags |= IN6_IFF_TENTATIVE; 976 } 977 /* 978 * A new RA might have made a deprecated address 979 * preferred. 980 */ 981 ia6->ia6_flags &= ~IN6_IFF_DEPRECATED; 982 } 983 } 984 985 /* expire prefix list */ 986 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 987 /* 988 * check prefix lifetime. 989 * since pltime is just for autoconf, pltime processing for 990 * prefix is not necessary. 991 */ 992 if (pr->ndpr_vltime != ND6_INFINITE_LIFETIME && 993 time_uptime - pr->ndpr_lastupdate > pr->ndpr_vltime) { 994 995 /* 996 * address expiration and prefix expiration are 997 * separate. NEVER perform in6_purgeaddr here. 998 */ 999 prelist_remove(pr); 1000 } 1001 } 1002 CURVNET_RESTORE(); 1003 } 1004 1005 /* 1006 * ia6 - deprecated/invalidated temporary address 1007 */ 1008 static int 1009 regen_tmpaddr(struct in6_ifaddr *ia6) 1010 { 1011 struct ifaddr *ifa; 1012 struct ifnet *ifp; 1013 struct in6_ifaddr *public_ifa6 = NULL; 1014 1015 ifp = ia6->ia_ifa.ifa_ifp; 1016 IF_ADDR_RLOCK(ifp); 1017 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1018 struct in6_ifaddr *it6; 1019 1020 if (ifa->ifa_addr->sa_family != AF_INET6) 1021 continue; 1022 1023 it6 = (struct in6_ifaddr *)ifa; 1024 1025 /* ignore no autoconf addresses. */ 1026 if ((it6->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1027 continue; 1028 1029 /* ignore autoconf addresses with different prefixes. */ 1030 if (it6->ia6_ndpr == NULL || it6->ia6_ndpr != ia6->ia6_ndpr) 1031 continue; 1032 1033 /* 1034 * Now we are looking at an autoconf address with the same 1035 * prefix as ours. If the address is temporary and is still 1036 * preferred, do not create another one. It would be rare, but 1037 * could happen, for example, when we resume a laptop PC after 1038 * a long period. 1039 */ 1040 if ((it6->ia6_flags & IN6_IFF_TEMPORARY) != 0 && 1041 !IFA6_IS_DEPRECATED(it6)) { 1042 public_ifa6 = NULL; 1043 break; 1044 } 1045 1046 /* 1047 * This is a public autoconf address that has the same prefix 1048 * as ours. If it is preferred, keep it. We can't break the 1049 * loop here, because there may be a still-preferred temporary 1050 * address with the prefix. 1051 */ 1052 if (!IFA6_IS_DEPRECATED(it6)) 1053 public_ifa6 = it6; 1054 } 1055 if (public_ifa6 != NULL) 1056 ifa_ref(&public_ifa6->ia_ifa); 1057 IF_ADDR_RUNLOCK(ifp); 1058 1059 if (public_ifa6 != NULL) { 1060 int e; 1061 1062 if ((e = in6_tmpifadd(public_ifa6, 0, 0)) != 0) { 1063 ifa_free(&public_ifa6->ia_ifa); 1064 log(LOG_NOTICE, "regen_tmpaddr: failed to create a new" 1065 " tmp addr,errno=%d\n", e); 1066 return (-1); 1067 } 1068 ifa_free(&public_ifa6->ia_ifa); 1069 return (0); 1070 } 1071 1072 return (-1); 1073 } 1074 1075 /* 1076 * Nuke neighbor cache/prefix/default router management table, right before 1077 * ifp goes away. 1078 */ 1079 void 1080 nd6_purge(struct ifnet *ifp) 1081 { 1082 struct nd_defrouter *dr, *ndr; 1083 struct nd_prefix *pr, *npr; 1084 1085 /* 1086 * Nuke default router list entries toward ifp. 1087 * We defer removal of default router list entries that is installed 1088 * in the routing table, in order to keep additional side effects as 1089 * small as possible. 1090 */ 1091 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 1092 if (dr->installed) 1093 continue; 1094 1095 if (dr->ifp == ifp) 1096 defrtrlist_del(dr); 1097 } 1098 1099 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, ndr) { 1100 if (!dr->installed) 1101 continue; 1102 1103 if (dr->ifp == ifp) 1104 defrtrlist_del(dr); 1105 } 1106 1107 /* Nuke prefix list entries toward ifp */ 1108 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, npr) { 1109 if (pr->ndpr_ifp == ifp) { 1110 /* 1111 * Because if_detach() does *not* release prefixes 1112 * while purging addresses the reference count will 1113 * still be above zero. We therefore reset it to 1114 * make sure that the prefix really gets purged. 1115 */ 1116 pr->ndpr_refcnt = 0; 1117 1118 /* 1119 * Previously, pr->ndpr_addr is removed as well, 1120 * but I strongly believe we don't have to do it. 1121 * nd6_purge() is only called from in6_ifdetach(), 1122 * which removes all the associated interface addresses 1123 * by itself. 1124 * (jinmei@kame.net 20010129) 1125 */ 1126 prelist_remove(pr); 1127 } 1128 } 1129 1130 /* cancel default outgoing interface setting */ 1131 if (V_nd6_defifindex == ifp->if_index) 1132 nd6_setdefaultiface(0); 1133 1134 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1135 /* Refresh default router list. */ 1136 defrouter_select(); 1137 } 1138 1139 /* XXXXX 1140 * We do not nuke the neighbor cache entries here any more 1141 * because the neighbor cache is kept in if_afdata[AF_INET6]. 1142 * nd6_purge() is invoked by in6_ifdetach() which is called 1143 * from if_detach() where everything gets purged. So let 1144 * in6_domifdetach() do the actual L2 table purging work. 1145 */ 1146 } 1147 1148 /* 1149 * the caller acquires and releases the lock on the lltbls 1150 * Returns the llentry locked 1151 */ 1152 struct llentry * 1153 nd6_lookup(const struct in6_addr *addr6, int flags, struct ifnet *ifp) 1154 { 1155 struct sockaddr_in6 sin6; 1156 struct llentry *ln; 1157 1158 bzero(&sin6, sizeof(sin6)); 1159 sin6.sin6_len = sizeof(struct sockaddr_in6); 1160 sin6.sin6_family = AF_INET6; 1161 sin6.sin6_addr = *addr6; 1162 1163 IF_AFDATA_LOCK_ASSERT(ifp); 1164 1165 ln = lla_lookup(LLTABLE6(ifp), flags, (struct sockaddr *)&sin6); 1166 1167 return (ln); 1168 } 1169 1170 struct llentry * 1171 nd6_alloc(const struct in6_addr *addr6, int flags, struct ifnet *ifp) 1172 { 1173 struct sockaddr_in6 sin6; 1174 struct llentry *ln; 1175 1176 bzero(&sin6, sizeof(sin6)); 1177 sin6.sin6_len = sizeof(struct sockaddr_in6); 1178 sin6.sin6_family = AF_INET6; 1179 sin6.sin6_addr = *addr6; 1180 1181 ln = lltable_alloc_entry(LLTABLE6(ifp), 0, (struct sockaddr *)&sin6); 1182 if (ln != NULL) 1183 ln->ln_state = ND6_LLINFO_NOSTATE; 1184 1185 return (ln); 1186 } 1187 1188 /* 1189 * Test whether a given IPv6 address is a neighbor or not, ignoring 1190 * the actual neighbor cache. The neighbor cache is ignored in order 1191 * to not reenter the routing code from within itself. 1192 */ 1193 static int 1194 nd6_is_new_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp) 1195 { 1196 struct nd_prefix *pr; 1197 struct ifaddr *dstaddr; 1198 1199 /* 1200 * A link-local address is always a neighbor. 1201 * XXX: a link does not necessarily specify a single interface. 1202 */ 1203 if (IN6_IS_ADDR_LINKLOCAL(&addr->sin6_addr)) { 1204 struct sockaddr_in6 sin6_copy; 1205 u_int32_t zone; 1206 1207 /* 1208 * We need sin6_copy since sa6_recoverscope() may modify the 1209 * content (XXX). 1210 */ 1211 sin6_copy = *addr; 1212 if (sa6_recoverscope(&sin6_copy)) 1213 return (0); /* XXX: should be impossible */ 1214 if (in6_setscope(&sin6_copy.sin6_addr, ifp, &zone)) 1215 return (0); 1216 if (sin6_copy.sin6_scope_id == zone) 1217 return (1); 1218 else 1219 return (0); 1220 } 1221 1222 /* 1223 * If the address matches one of our addresses, 1224 * it should be a neighbor. 1225 * If the address matches one of our on-link prefixes, it should be a 1226 * neighbor. 1227 */ 1228 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 1229 if (pr->ndpr_ifp != ifp) 1230 continue; 1231 1232 if (!(pr->ndpr_stateflags & NDPRF_ONLINK)) { 1233 struct rtentry *rt; 1234 1235 /* Always use the default FIB here. */ 1236 rt = in6_rtalloc1((struct sockaddr *)&pr->ndpr_prefix, 1237 0, 0, RT_DEFAULT_FIB); 1238 if (rt == NULL) 1239 continue; 1240 /* 1241 * This is the case where multiple interfaces 1242 * have the same prefix, but only one is installed 1243 * into the routing table and that prefix entry 1244 * is not the one being examined here. In the case 1245 * where RADIX_MPATH is enabled, multiple route 1246 * entries (of the same rt_key value) will be 1247 * installed because the interface addresses all 1248 * differ. 1249 */ 1250 if (!IN6_ARE_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 1251 &((struct sockaddr_in6 *)rt_key(rt))->sin6_addr)) { 1252 RTFREE_LOCKED(rt); 1253 continue; 1254 } 1255 RTFREE_LOCKED(rt); 1256 } 1257 1258 if (IN6_ARE_MASKED_ADDR_EQUAL(&pr->ndpr_prefix.sin6_addr, 1259 &addr->sin6_addr, &pr->ndpr_mask)) 1260 return (1); 1261 } 1262 1263 /* 1264 * If the address is assigned on the node of the other side of 1265 * a p2p interface, the address should be a neighbor. 1266 */ 1267 dstaddr = ifa_ifwithdstaddr((const struct sockaddr *)addr, RT_ALL_FIBS); 1268 if (dstaddr != NULL) { 1269 if (dstaddr->ifa_ifp == ifp) { 1270 ifa_free(dstaddr); 1271 return (1); 1272 } 1273 ifa_free(dstaddr); 1274 } 1275 1276 /* 1277 * If the default router list is empty, all addresses are regarded 1278 * as on-link, and thus, as a neighbor. 1279 */ 1280 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV && 1281 TAILQ_EMPTY(&V_nd_defrouter) && 1282 V_nd6_defifindex == ifp->if_index) { 1283 return (1); 1284 } 1285 1286 return (0); 1287 } 1288 1289 1290 /* 1291 * Detect if a given IPv6 address identifies a neighbor on a given link. 1292 * XXX: should take care of the destination of a p2p link? 1293 */ 1294 int 1295 nd6_is_addr_neighbor(const struct sockaddr_in6 *addr, struct ifnet *ifp) 1296 { 1297 struct llentry *lle; 1298 int rc = 0; 1299 1300 IF_AFDATA_UNLOCK_ASSERT(ifp); 1301 if (nd6_is_new_addr_neighbor(addr, ifp)) 1302 return (1); 1303 1304 /* 1305 * Even if the address matches none of our addresses, it might be 1306 * in the neighbor cache. 1307 */ 1308 IF_AFDATA_RLOCK(ifp); 1309 if ((lle = nd6_lookup(&addr->sin6_addr, 0, ifp)) != NULL) { 1310 LLE_RUNLOCK(lle); 1311 rc = 1; 1312 } 1313 IF_AFDATA_RUNLOCK(ifp); 1314 return (rc); 1315 } 1316 1317 /* 1318 * Free an nd6 llinfo entry. 1319 * Since the function would cause significant changes in the kernel, DO NOT 1320 * make it global, unless you have a strong reason for the change, and are sure 1321 * that the change is safe. 1322 * 1323 * Set noinline to be dtrace-friendly 1324 */ 1325 static __noinline void 1326 nd6_free(struct llentry *ln, int gc) 1327 { 1328 struct nd_defrouter *dr; 1329 struct ifnet *ifp; 1330 1331 LLE_WLOCK_ASSERT(ln); 1332 1333 /* 1334 * we used to have pfctlinput(PRC_HOSTDEAD) here. 1335 * even though it is not harmful, it was not really necessary. 1336 */ 1337 1338 /* cancel timer */ 1339 nd6_llinfo_settimer_locked(ln, -1); 1340 1341 ifp = ln->lle_tbl->llt_ifp; 1342 1343 if (ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1344 dr = defrouter_lookup(&ln->r_l3addr.addr6, ifp); 1345 1346 if (dr != NULL && dr->expire && 1347 ln->ln_state == ND6_LLINFO_STALE && gc) { 1348 /* 1349 * If the reason for the deletion is just garbage 1350 * collection, and the neighbor is an active default 1351 * router, do not delete it. Instead, reset the GC 1352 * timer using the router's lifetime. 1353 * Simply deleting the entry would affect default 1354 * router selection, which is not necessarily a good 1355 * thing, especially when we're using router preference 1356 * values. 1357 * XXX: the check for ln_state would be redundant, 1358 * but we intentionally keep it just in case. 1359 */ 1360 if (dr->expire > time_uptime) 1361 nd6_llinfo_settimer_locked(ln, 1362 (dr->expire - time_uptime) * hz); 1363 else 1364 nd6_llinfo_settimer_locked(ln, 1365 (long)V_nd6_gctimer * hz); 1366 1367 LLE_REMREF(ln); 1368 LLE_WUNLOCK(ln); 1369 return; 1370 } 1371 1372 if (dr) { 1373 /* 1374 * Unreachablity of a router might affect the default 1375 * router selection and on-link detection of advertised 1376 * prefixes. 1377 */ 1378 1379 /* 1380 * Temporarily fake the state to choose a new default 1381 * router and to perform on-link determination of 1382 * prefixes correctly. 1383 * Below the state will be set correctly, 1384 * or the entry itself will be deleted. 1385 */ 1386 ln->ln_state = ND6_LLINFO_INCOMPLETE; 1387 } 1388 1389 if (ln->ln_router || dr) { 1390 1391 /* 1392 * We need to unlock to avoid a LOR with rt6_flush() with the 1393 * rnh and for the calls to pfxlist_onlink_check() and 1394 * defrouter_select() in the block further down for calls 1395 * into nd6_lookup(). We still hold a ref. 1396 */ 1397 LLE_WUNLOCK(ln); 1398 1399 /* 1400 * rt6_flush must be called whether or not the neighbor 1401 * is in the Default Router List. 1402 * See a corresponding comment in nd6_na_input(). 1403 */ 1404 rt6_flush(&ln->r_l3addr.addr6, ifp); 1405 } 1406 1407 if (dr) { 1408 /* 1409 * Since defrouter_select() does not affect the 1410 * on-link determination and MIP6 needs the check 1411 * before the default router selection, we perform 1412 * the check now. 1413 */ 1414 pfxlist_onlink_check(); 1415 1416 /* 1417 * Refresh default router list. 1418 */ 1419 defrouter_select(); 1420 } 1421 1422 /* 1423 * If this entry was added by an on-link redirect, remove the 1424 * corresponding host route. 1425 */ 1426 if (ln->la_flags & LLE_REDIRECT) 1427 nd6_free_redirect(ln); 1428 1429 if (ln->ln_router || dr) 1430 LLE_WLOCK(ln); 1431 } 1432 1433 /* 1434 * Save to unlock. We still hold an extra reference and will not 1435 * free(9) in llentry_free() if someone else holds one as well. 1436 */ 1437 LLE_WUNLOCK(ln); 1438 IF_AFDATA_LOCK(ifp); 1439 LLE_WLOCK(ln); 1440 /* Guard against race with other llentry_free(). */ 1441 if (ln->la_flags & LLE_LINKED) { 1442 /* Remove callout reference */ 1443 LLE_REMREF(ln); 1444 lltable_unlink_entry(ln->lle_tbl, ln); 1445 } 1446 IF_AFDATA_UNLOCK(ifp); 1447 1448 llentry_free(ln); 1449 } 1450 1451 static int 1452 nd6_isdynrte(const struct rtentry *rt, void *xap) 1453 { 1454 1455 if (rt->rt_flags == (RTF_UP | RTF_HOST | RTF_DYNAMIC)) 1456 return (1); 1457 1458 return (0); 1459 } 1460 /* 1461 * Remove the rtentry for the given llentry, 1462 * both of which were installed by a redirect. 1463 */ 1464 static void 1465 nd6_free_redirect(const struct llentry *ln) 1466 { 1467 int fibnum; 1468 struct sockaddr_in6 sin6; 1469 struct rt_addrinfo info; 1470 1471 lltable_fill_sa_entry(ln, (struct sockaddr *)&sin6); 1472 memset(&info, 0, sizeof(info)); 1473 info.rti_info[RTAX_DST] = (struct sockaddr *)&sin6; 1474 info.rti_filter = nd6_isdynrte; 1475 1476 for (fibnum = 0; fibnum < rt_numfibs; fibnum++) 1477 rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum); 1478 } 1479 1480 /* 1481 * Rejuvenate this function for routing operations related 1482 * processing. 1483 */ 1484 void 1485 nd6_rtrequest(int req, struct rtentry *rt, struct rt_addrinfo *info) 1486 { 1487 struct sockaddr_in6 *gateway; 1488 struct nd_defrouter *dr; 1489 struct ifnet *ifp; 1490 1491 gateway = (struct sockaddr_in6 *)rt->rt_gateway; 1492 ifp = rt->rt_ifp; 1493 1494 switch (req) { 1495 case RTM_ADD: 1496 break; 1497 1498 case RTM_DELETE: 1499 if (!ifp) 1500 return; 1501 /* 1502 * Only indirect routes are interesting. 1503 */ 1504 if ((rt->rt_flags & RTF_GATEWAY) == 0) 1505 return; 1506 /* 1507 * check for default route 1508 */ 1509 if (IN6_ARE_ADDR_EQUAL(&in6addr_any, 1510 &SIN6(rt_key(rt))->sin6_addr)) { 1511 1512 dr = defrouter_lookup(&gateway->sin6_addr, ifp); 1513 if (dr != NULL) 1514 dr->installed = 0; 1515 } 1516 break; 1517 } 1518 } 1519 1520 1521 int 1522 nd6_ioctl(u_long cmd, caddr_t data, struct ifnet *ifp) 1523 { 1524 struct in6_ndireq *ndi = (struct in6_ndireq *)data; 1525 struct in6_nbrinfo *nbi = (struct in6_nbrinfo *)data; 1526 struct in6_ndifreq *ndif = (struct in6_ndifreq *)data; 1527 int error = 0; 1528 1529 if (ifp->if_afdata[AF_INET6] == NULL) 1530 return (EPFNOSUPPORT); 1531 switch (cmd) { 1532 case OSIOCGIFINFO_IN6: 1533 #define ND ndi->ndi 1534 /* XXX: old ndp(8) assumes a positive value for linkmtu. */ 1535 bzero(&ND, sizeof(ND)); 1536 ND.linkmtu = IN6_LINKMTU(ifp); 1537 ND.maxmtu = ND_IFINFO(ifp)->maxmtu; 1538 ND.basereachable = ND_IFINFO(ifp)->basereachable; 1539 ND.reachable = ND_IFINFO(ifp)->reachable; 1540 ND.retrans = ND_IFINFO(ifp)->retrans; 1541 ND.flags = ND_IFINFO(ifp)->flags; 1542 ND.recalctm = ND_IFINFO(ifp)->recalctm; 1543 ND.chlim = ND_IFINFO(ifp)->chlim; 1544 break; 1545 case SIOCGIFINFO_IN6: 1546 ND = *ND_IFINFO(ifp); 1547 break; 1548 case SIOCSIFINFO_IN6: 1549 /* 1550 * used to change host variables from userland. 1551 * intented for a use on router to reflect RA configurations. 1552 */ 1553 /* 0 means 'unspecified' */ 1554 if (ND.linkmtu != 0) { 1555 if (ND.linkmtu < IPV6_MMTU || 1556 ND.linkmtu > IN6_LINKMTU(ifp)) { 1557 error = EINVAL; 1558 break; 1559 } 1560 ND_IFINFO(ifp)->linkmtu = ND.linkmtu; 1561 } 1562 1563 if (ND.basereachable != 0) { 1564 int obasereachable = ND_IFINFO(ifp)->basereachable; 1565 1566 ND_IFINFO(ifp)->basereachable = ND.basereachable; 1567 if (ND.basereachable != obasereachable) 1568 ND_IFINFO(ifp)->reachable = 1569 ND_COMPUTE_RTIME(ND.basereachable); 1570 } 1571 if (ND.retrans != 0) 1572 ND_IFINFO(ifp)->retrans = ND.retrans; 1573 if (ND.chlim != 0) 1574 ND_IFINFO(ifp)->chlim = ND.chlim; 1575 /* FALLTHROUGH */ 1576 case SIOCSIFINFO_FLAGS: 1577 { 1578 struct ifaddr *ifa; 1579 struct in6_ifaddr *ia; 1580 1581 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1582 !(ND.flags & ND6_IFF_IFDISABLED)) { 1583 /* ifdisabled 1->0 transision */ 1584 1585 /* 1586 * If the interface is marked as ND6_IFF_IFDISABLED and 1587 * has an link-local address with IN6_IFF_DUPLICATED, 1588 * do not clear ND6_IFF_IFDISABLED. 1589 * See RFC 4862, Section 5.4.5. 1590 */ 1591 IF_ADDR_RLOCK(ifp); 1592 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1593 if (ifa->ifa_addr->sa_family != AF_INET6) 1594 continue; 1595 ia = (struct in6_ifaddr *)ifa; 1596 if ((ia->ia6_flags & IN6_IFF_DUPLICATED) && 1597 IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) 1598 break; 1599 } 1600 IF_ADDR_RUNLOCK(ifp); 1601 1602 if (ifa != NULL) { 1603 /* LLA is duplicated. */ 1604 ND.flags |= ND6_IFF_IFDISABLED; 1605 log(LOG_ERR, "Cannot enable an interface" 1606 " with a link-local address marked" 1607 " duplicate.\n"); 1608 } else { 1609 ND_IFINFO(ifp)->flags &= ~ND6_IFF_IFDISABLED; 1610 if (ifp->if_flags & IFF_UP) 1611 in6_if_up(ifp); 1612 } 1613 } else if (!(ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED) && 1614 (ND.flags & ND6_IFF_IFDISABLED)) { 1615 /* ifdisabled 0->1 transision */ 1616 /* Mark all IPv6 address as tentative. */ 1617 1618 ND_IFINFO(ifp)->flags |= ND6_IFF_IFDISABLED; 1619 if (V_ip6_dad_count > 0 && 1620 (ND_IFINFO(ifp)->flags & ND6_IFF_NO_DAD) == 0) { 1621 IF_ADDR_RLOCK(ifp); 1622 TAILQ_FOREACH(ifa, &ifp->if_addrhead, 1623 ifa_link) { 1624 if (ifa->ifa_addr->sa_family != 1625 AF_INET6) 1626 continue; 1627 ia = (struct in6_ifaddr *)ifa; 1628 ia->ia6_flags |= IN6_IFF_TENTATIVE; 1629 } 1630 IF_ADDR_RUNLOCK(ifp); 1631 } 1632 } 1633 1634 if (ND.flags & ND6_IFF_AUTO_LINKLOCAL) { 1635 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_AUTO_LINKLOCAL)) { 1636 /* auto_linklocal 0->1 transision */ 1637 1638 /* If no link-local address on ifp, configure */ 1639 ND_IFINFO(ifp)->flags |= ND6_IFF_AUTO_LINKLOCAL; 1640 in6_ifattach(ifp, NULL); 1641 } else if (!(ND.flags & ND6_IFF_IFDISABLED) && 1642 ifp->if_flags & IFF_UP) { 1643 /* 1644 * When the IF already has 1645 * ND6_IFF_AUTO_LINKLOCAL, no link-local 1646 * address is assigned, and IFF_UP, try to 1647 * assign one. 1648 */ 1649 IF_ADDR_RLOCK(ifp); 1650 TAILQ_FOREACH(ifa, &ifp->if_addrhead, 1651 ifa_link) { 1652 if (ifa->ifa_addr->sa_family != 1653 AF_INET6) 1654 continue; 1655 ia = (struct in6_ifaddr *)ifa; 1656 if (IN6_IS_ADDR_LINKLOCAL(IA6_IN6(ia))) 1657 break; 1658 } 1659 IF_ADDR_RUNLOCK(ifp); 1660 if (ifa != NULL) 1661 /* No LLA is configured. */ 1662 in6_ifattach(ifp, NULL); 1663 } 1664 } 1665 } 1666 ND_IFINFO(ifp)->flags = ND.flags; 1667 break; 1668 #undef ND 1669 case SIOCSNDFLUSH_IN6: /* XXX: the ioctl name is confusing... */ 1670 /* sync kernel routing table with the default router list */ 1671 defrouter_reset(); 1672 defrouter_select(); 1673 break; 1674 case SIOCSPFXFLUSH_IN6: 1675 { 1676 /* flush all the prefix advertised by routers */ 1677 struct nd_prefix *pr, *next; 1678 1679 LIST_FOREACH_SAFE(pr, &V_nd_prefix, ndpr_entry, next) { 1680 struct in6_ifaddr *ia, *ia_next; 1681 1682 if (IN6_IS_ADDR_LINKLOCAL(&pr->ndpr_prefix.sin6_addr)) 1683 continue; /* XXX */ 1684 1685 /* do we really have to remove addresses as well? */ 1686 /* XXXRW: in6_ifaddrhead locking. */ 1687 TAILQ_FOREACH_SAFE(ia, &V_in6_ifaddrhead, ia_link, 1688 ia_next) { 1689 if ((ia->ia6_flags & IN6_IFF_AUTOCONF) == 0) 1690 continue; 1691 1692 if (ia->ia6_ndpr == pr) 1693 in6_purgeaddr(&ia->ia_ifa); 1694 } 1695 prelist_remove(pr); 1696 } 1697 break; 1698 } 1699 case SIOCSRTRFLUSH_IN6: 1700 { 1701 /* flush all the default routers */ 1702 struct nd_defrouter *dr, *next; 1703 1704 defrouter_reset(); 1705 TAILQ_FOREACH_SAFE(dr, &V_nd_defrouter, dr_entry, next) { 1706 defrtrlist_del(dr); 1707 } 1708 defrouter_select(); 1709 break; 1710 } 1711 case SIOCGNBRINFO_IN6: 1712 { 1713 struct llentry *ln; 1714 struct in6_addr nb_addr = nbi->addr; /* make local for safety */ 1715 1716 if ((error = in6_setscope(&nb_addr, ifp, NULL)) != 0) 1717 return (error); 1718 1719 IF_AFDATA_RLOCK(ifp); 1720 ln = nd6_lookup(&nb_addr, 0, ifp); 1721 IF_AFDATA_RUNLOCK(ifp); 1722 1723 if (ln == NULL) { 1724 error = EINVAL; 1725 break; 1726 } 1727 nbi->state = ln->ln_state; 1728 nbi->asked = ln->la_asked; 1729 nbi->isrouter = ln->ln_router; 1730 if (ln->la_expire == 0) 1731 nbi->expire = 0; 1732 else 1733 nbi->expire = ln->la_expire + ln->lle_remtime / hz + 1734 (time_second - time_uptime); 1735 LLE_RUNLOCK(ln); 1736 break; 1737 } 1738 case SIOCGDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1739 ndif->ifindex = V_nd6_defifindex; 1740 break; 1741 case SIOCSDEFIFACE_IN6: /* XXX: should be implemented as a sysctl? */ 1742 return (nd6_setdefaultiface(ndif->ifindex)); 1743 } 1744 return (error); 1745 } 1746 1747 /* 1748 * Calculates new isRouter value based on provided parameters and 1749 * returns it. 1750 */ 1751 static int 1752 nd6_is_router(int type, int code, int is_new, int old_addr, int new_addr, 1753 int ln_router) 1754 { 1755 1756 /* 1757 * ICMP6 type dependent behavior. 1758 * 1759 * NS: clear IsRouter if new entry 1760 * RS: clear IsRouter 1761 * RA: set IsRouter if there's lladdr 1762 * redir: clear IsRouter if new entry 1763 * 1764 * RA case, (1): 1765 * The spec says that we must set IsRouter in the following cases: 1766 * - If lladdr exist, set IsRouter. This means (1-5). 1767 * - If it is old entry (!newentry), set IsRouter. This means (7). 1768 * So, based on the spec, in (1-5) and (7) cases we must set IsRouter. 1769 * A quetion arises for (1) case. (1) case has no lladdr in the 1770 * neighbor cache, this is similar to (6). 1771 * This case is rare but we figured that we MUST NOT set IsRouter. 1772 * 1773 * is_new old_addr new_addr NS RS RA redir 1774 * D R 1775 * 0 n n (1) c ? s 1776 * 0 y n (2) c s s 1777 * 0 n y (3) c s s 1778 * 0 y y (4) c s s 1779 * 0 y y (5) c s s 1780 * 1 -- n (6) c c c s 1781 * 1 -- y (7) c c s c s 1782 * 1783 * (c=clear s=set) 1784 */ 1785 switch (type & 0xff) { 1786 case ND_NEIGHBOR_SOLICIT: 1787 /* 1788 * New entry must have is_router flag cleared. 1789 */ 1790 if (is_new) /* (6-7) */ 1791 ln_router = 0; 1792 break; 1793 case ND_REDIRECT: 1794 /* 1795 * If the icmp is a redirect to a better router, always set the 1796 * is_router flag. Otherwise, if the entry is newly created, 1797 * clear the flag. [RFC 2461, sec 8.3] 1798 */ 1799 if (code == ND_REDIRECT_ROUTER) 1800 ln_router = 1; 1801 else { 1802 if (is_new) /* (6-7) */ 1803 ln_router = 0; 1804 } 1805 break; 1806 case ND_ROUTER_SOLICIT: 1807 /* 1808 * is_router flag must always be cleared. 1809 */ 1810 ln_router = 0; 1811 break; 1812 case ND_ROUTER_ADVERT: 1813 /* 1814 * Mark an entry with lladdr as a router. 1815 */ 1816 if ((!is_new && (old_addr || new_addr)) || /* (2-5) */ 1817 (is_new && new_addr)) { /* (7) */ 1818 ln_router = 1; 1819 } 1820 break; 1821 } 1822 1823 return (ln_router); 1824 } 1825 1826 /* 1827 * Create neighbor cache entry and cache link-layer address, 1828 * on reception of inbound ND6 packets. (RS/RA/NS/redirect) 1829 * 1830 * type - ICMP6 type 1831 * code - type dependent information 1832 * 1833 */ 1834 void 1835 nd6_cache_lladdr(struct ifnet *ifp, struct in6_addr *from, char *lladdr, 1836 int lladdrlen, int type, int code) 1837 { 1838 struct llentry *ln = NULL, *ln_tmp; 1839 int is_newentry; 1840 int do_update; 1841 int olladdr; 1842 int llchange; 1843 int flags; 1844 uint16_t router = 0; 1845 struct sockaddr_in6 sin6; 1846 struct mbuf *chain = NULL; 1847 1848 IF_AFDATA_UNLOCK_ASSERT(ifp); 1849 1850 KASSERT(ifp != NULL, ("%s: ifp == NULL", __func__)); 1851 KASSERT(from != NULL, ("%s: from == NULL", __func__)); 1852 1853 /* nothing must be updated for unspecified address */ 1854 if (IN6_IS_ADDR_UNSPECIFIED(from)) 1855 return; 1856 1857 /* 1858 * Validation about ifp->if_addrlen and lladdrlen must be done in 1859 * the caller. 1860 * 1861 * XXX If the link does not have link-layer adderss, what should 1862 * we do? (ifp->if_addrlen == 0) 1863 * Spec says nothing in sections for RA, RS and NA. There's small 1864 * description on it in NS section (RFC 2461 7.2.3). 1865 */ 1866 flags = lladdr ? LLE_EXCLUSIVE : 0; 1867 IF_AFDATA_RLOCK(ifp); 1868 ln = nd6_lookup(from, flags, ifp); 1869 IF_AFDATA_RUNLOCK(ifp); 1870 is_newentry = 0; 1871 if (ln == NULL) { 1872 flags |= LLE_EXCLUSIVE; 1873 ln = nd6_alloc(from, 0, ifp); 1874 if (ln == NULL) 1875 return; 1876 1877 /* 1878 * Since we already know all the data for the new entry, 1879 * fill it before insertion. 1880 */ 1881 if (lladdr != NULL) 1882 lltable_set_entry_addr(ifp, ln, lladdr); 1883 IF_AFDATA_WLOCK(ifp); 1884 LLE_WLOCK(ln); 1885 /* Prefer any existing lle over newly-created one */ 1886 ln_tmp = nd6_lookup(from, LLE_EXCLUSIVE, ifp); 1887 if (ln_tmp == NULL) 1888 lltable_link_entry(LLTABLE6(ifp), ln); 1889 IF_AFDATA_WUNLOCK(ifp); 1890 if (ln_tmp == NULL) { 1891 /* No existing lle, mark as new entry (6,7) */ 1892 is_newentry = 1; 1893 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); 1894 if (lladdr != NULL) /* (7) */ 1895 EVENTHANDLER_INVOKE(lle_event, ln, 1896 LLENTRY_RESOLVED); 1897 } else { 1898 lltable_free_entry(LLTABLE6(ifp), ln); 1899 ln = ln_tmp; 1900 ln_tmp = NULL; 1901 } 1902 } 1903 /* do nothing if static ndp is set */ 1904 if ((ln->la_flags & LLE_STATIC)) { 1905 if (flags & LLE_EXCLUSIVE) 1906 LLE_WUNLOCK(ln); 1907 else 1908 LLE_RUNLOCK(ln); 1909 return; 1910 } 1911 1912 olladdr = (ln->la_flags & LLE_VALID) ? 1 : 0; 1913 if (olladdr && lladdr) { 1914 llchange = bcmp(lladdr, &ln->ll_addr, 1915 ifp->if_addrlen); 1916 } else if (!olladdr && lladdr) 1917 llchange = 1; 1918 else 1919 llchange = 0; 1920 1921 /* 1922 * newentry olladdr lladdr llchange (*=record) 1923 * 0 n n -- (1) 1924 * 0 y n -- (2) 1925 * 0 n y y (3) * STALE 1926 * 0 y y n (4) * 1927 * 0 y y y (5) * STALE 1928 * 1 -- n -- (6) NOSTATE(= PASSIVE) 1929 * 1 -- y -- (7) * STALE 1930 */ 1931 1932 do_update = 0; 1933 if (is_newentry == 0 && llchange != 0) { 1934 do_update = 1; /* (3,5) */ 1935 1936 /* 1937 * Record source link-layer address 1938 * XXX is it dependent to ifp->if_type? 1939 */ 1940 if (lltable_try_set_entry_addr(ifp, ln, lladdr) == 0) { 1941 /* Entry was deleted */ 1942 return; 1943 } 1944 1945 nd6_llinfo_setstate(ln, ND6_LLINFO_STALE); 1946 1947 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 1948 1949 if (ln->la_hold != NULL) 1950 nd6_grab_holdchain(ln, &chain, &sin6); 1951 } 1952 1953 /* Calculates new router status */ 1954 router = nd6_is_router(type, code, is_newentry, olladdr, 1955 lladdr != NULL ? 1 : 0, ln->ln_router); 1956 1957 ln->ln_router = router; 1958 /* Mark non-router redirects with special flag */ 1959 if ((type & 0xFF) == ND_REDIRECT && code != ND_REDIRECT_ROUTER) 1960 ln->la_flags |= LLE_REDIRECT; 1961 1962 if (flags & LLE_EXCLUSIVE) 1963 LLE_WUNLOCK(ln); 1964 else 1965 LLE_RUNLOCK(ln); 1966 1967 if (chain != NULL) 1968 nd6_flush_holdchain(ifp, ifp, chain, &sin6); 1969 1970 /* 1971 * When the link-layer address of a router changes, select the 1972 * best router again. In particular, when the neighbor entry is newly 1973 * created, it might affect the selection policy. 1974 * Question: can we restrict the first condition to the "is_newentry" 1975 * case? 1976 * XXX: when we hear an RA from a new router with the link-layer 1977 * address option, defrouter_select() is called twice, since 1978 * defrtrlist_update called the function as well. However, I believe 1979 * we can compromise the overhead, since it only happens the first 1980 * time. 1981 * XXX: although defrouter_select() should not have a bad effect 1982 * for those are not autoconfigured hosts, we explicitly avoid such 1983 * cases for safety. 1984 */ 1985 if ((do_update || is_newentry) && router && 1986 ND_IFINFO(ifp)->flags & ND6_IFF_ACCEPT_RTADV) { 1987 /* 1988 * guaranteed recursion 1989 */ 1990 defrouter_select(); 1991 } 1992 } 1993 1994 static void 1995 nd6_slowtimo(void *arg) 1996 { 1997 CURVNET_SET((struct vnet *) arg); 1998 struct nd_ifinfo *nd6if; 1999 struct ifnet *ifp; 2000 2001 callout_reset(&V_nd6_slowtimo_ch, ND6_SLOWTIMER_INTERVAL * hz, 2002 nd6_slowtimo, curvnet); 2003 IFNET_RLOCK_NOSLEEP(); 2004 TAILQ_FOREACH(ifp, &V_ifnet, if_link) { 2005 if (ifp->if_afdata[AF_INET6] == NULL) 2006 continue; 2007 nd6if = ND_IFINFO(ifp); 2008 if (nd6if->basereachable && /* already initialized */ 2009 (nd6if->recalctm -= ND6_SLOWTIMER_INTERVAL) <= 0) { 2010 /* 2011 * Since reachable time rarely changes by router 2012 * advertisements, we SHOULD insure that a new random 2013 * value gets recomputed at least once every few hours. 2014 * (RFC 2461, 6.3.4) 2015 */ 2016 nd6if->recalctm = V_nd6_recalc_reachtm_interval; 2017 nd6if->reachable = ND_COMPUTE_RTIME(nd6if->basereachable); 2018 } 2019 } 2020 IFNET_RUNLOCK_NOSLEEP(); 2021 CURVNET_RESTORE(); 2022 } 2023 2024 void 2025 nd6_grab_holdchain(struct llentry *ln, struct mbuf **chain, 2026 struct sockaddr_in6 *sin6) 2027 { 2028 2029 LLE_WLOCK_ASSERT(ln); 2030 2031 *chain = ln->la_hold; 2032 ln->la_hold = NULL; 2033 lltable_fill_sa_entry(ln, (struct sockaddr *)sin6); 2034 2035 if (ln->ln_state == ND6_LLINFO_STALE) { 2036 2037 /* 2038 * The first time we send a packet to a 2039 * neighbor whose entry is STALE, we have 2040 * to change the state to DELAY and a sets 2041 * a timer to expire in DELAY_FIRST_PROBE_TIME 2042 * seconds to ensure do neighbor unreachability 2043 * detection on expiration. 2044 * (RFC 2461 7.3.3) 2045 */ 2046 nd6_llinfo_setstate(ln, ND6_LLINFO_DELAY); 2047 } 2048 } 2049 2050 int 2051 nd6_output_ifp(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *m, 2052 struct sockaddr_in6 *dst, struct route *ro) 2053 { 2054 int error; 2055 int ip6len; 2056 struct ip6_hdr *ip6; 2057 struct m_tag *mtag; 2058 2059 #ifdef MAC 2060 mac_netinet6_nd6_send(ifp, m); 2061 #endif 2062 2063 /* 2064 * If called from nd6_ns_output() (NS), nd6_na_output() (NA), 2065 * icmp6_redirect_output() (REDIRECT) or from rip6_output() (RS, RA 2066 * as handled by rtsol and rtadvd), mbufs will be tagged for SeND 2067 * to be diverted to user space. When re-injected into the kernel, 2068 * send_output() will directly dispatch them to the outgoing interface. 2069 */ 2070 if (send_sendso_input_hook != NULL) { 2071 mtag = m_tag_find(m, PACKET_TAG_ND_OUTGOING, NULL); 2072 if (mtag != NULL) { 2073 ip6 = mtod(m, struct ip6_hdr *); 2074 ip6len = sizeof(struct ip6_hdr) + ntohs(ip6->ip6_plen); 2075 /* Use the SEND socket */ 2076 error = send_sendso_input_hook(m, ifp, SND_OUT, 2077 ip6len); 2078 /* -1 == no app on SEND socket */ 2079 if (error == 0 || error != -1) 2080 return (error); 2081 } 2082 } 2083 2084 m_clrprotoflags(m); /* Avoid confusing lower layers. */ 2085 IP_PROBE(send, NULL, NULL, mtod(m, struct ip6_hdr *), ifp, NULL, 2086 mtod(m, struct ip6_hdr *)); 2087 2088 if ((ifp->if_flags & IFF_LOOPBACK) == 0) 2089 origifp = ifp; 2090 2091 error = (*ifp->if_output)(origifp, m, (struct sockaddr *)dst, ro); 2092 return (error); 2093 } 2094 2095 /* 2096 * Do L2 address resolution for @sa_dst address. Stores found 2097 * address in @desten buffer. Copy of lle ln_flags can be also 2098 * saved in @pflags if @pflags is non-NULL. 2099 * 2100 * If destination LLE does not exists or lle state modification 2101 * is required, call "slow" version. 2102 * 2103 * Return values: 2104 * - 0 on success (address copied to buffer). 2105 * - EWOULDBLOCK (no local error, but address is still unresolved) 2106 * - other errors (alloc failure, etc) 2107 */ 2108 int 2109 nd6_resolve(struct ifnet *ifp, int is_gw, struct mbuf *m, 2110 const struct sockaddr *sa_dst, u_char *desten, uint32_t *pflags) 2111 { 2112 struct llentry *ln = NULL; 2113 const struct sockaddr_in6 *dst6; 2114 2115 if (pflags != NULL) 2116 *pflags = 0; 2117 2118 dst6 = (const struct sockaddr_in6 *)sa_dst; 2119 2120 /* discard the packet if IPv6 operation is disabled on the interface */ 2121 if ((ND_IFINFO(ifp)->flags & ND6_IFF_IFDISABLED)) { 2122 m_freem(m); 2123 return (ENETDOWN); /* better error? */ 2124 } 2125 2126 if (m != NULL && m->m_flags & M_MCAST) { 2127 switch (ifp->if_type) { 2128 case IFT_ETHER: 2129 case IFT_FDDI: 2130 case IFT_L2VLAN: 2131 case IFT_IEEE80211: 2132 case IFT_BRIDGE: 2133 case IFT_ISO88025: 2134 ETHER_MAP_IPV6_MULTICAST(&dst6->sin6_addr, 2135 desten); 2136 return (0); 2137 default: 2138 m_freem(m); 2139 return (EAFNOSUPPORT); 2140 } 2141 } 2142 2143 IF_AFDATA_RLOCK(ifp); 2144 ln = nd6_lookup(&dst6->sin6_addr, LLE_UNLOCKED, ifp); 2145 if (ln != NULL && (ln->r_flags & RLLE_VALID) != 0) { 2146 /* Entry found, let's copy lle info */ 2147 bcopy(&ln->ll_addr, desten, ifp->if_addrlen); 2148 if (pflags != NULL) 2149 *pflags = LLE_VALID | (ln->r_flags & RLLE_IFADDR); 2150 /* Check if we have feedback request from nd6 timer */ 2151 if (ln->r_skip_req != 0) { 2152 LLE_REQ_LOCK(ln); 2153 ln->r_skip_req = 0; /* Notify that entry was used */ 2154 ln->lle_hittime = time_uptime; 2155 LLE_REQ_UNLOCK(ln); 2156 } 2157 IF_AFDATA_RUNLOCK(ifp); 2158 return (0); 2159 } 2160 IF_AFDATA_RUNLOCK(ifp); 2161 2162 return (nd6_resolve_slow(ifp, m, dst6, desten, pflags)); 2163 } 2164 2165 2166 /* 2167 * Do L2 address resolution for @sa_dst address. Stores found 2168 * address in @desten buffer. Copy of lle ln_flags can be also 2169 * saved in @pflags if @pflags is non-NULL. 2170 * 2171 * Heavy version. 2172 * Function assume that destination LLE does not exist, 2173 * is invalid or stale, so LLE_EXCLUSIVE lock needs to be acquired. 2174 * 2175 * Set noinline to be dtrace-friendly 2176 */ 2177 static __noinline int 2178 nd6_resolve_slow(struct ifnet *ifp, struct mbuf *m, 2179 const struct sockaddr_in6 *dst, u_char *desten, uint32_t *pflags) 2180 { 2181 struct llentry *lle = NULL, *lle_tmp; 2182 struct in6_addr *psrc, src; 2183 int send_ns; 2184 2185 /* 2186 * Address resolution or Neighbor Unreachability Detection 2187 * for the next hop. 2188 * At this point, the destination of the packet must be a unicast 2189 * or an anycast address(i.e. not a multicast). 2190 */ 2191 if (lle == NULL) { 2192 IF_AFDATA_RLOCK(ifp); 2193 lle = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp); 2194 IF_AFDATA_RUNLOCK(ifp); 2195 if ((lle == NULL) && nd6_is_addr_neighbor(dst, ifp)) { 2196 /* 2197 * Since nd6_is_addr_neighbor() internally calls nd6_lookup(), 2198 * the condition below is not very efficient. But we believe 2199 * it is tolerable, because this should be a rare case. 2200 */ 2201 lle = nd6_alloc(&dst->sin6_addr, 0, ifp); 2202 if (lle == NULL) { 2203 char ip6buf[INET6_ADDRSTRLEN]; 2204 log(LOG_DEBUG, 2205 "nd6_output: can't allocate llinfo for %s " 2206 "(ln=%p)\n", 2207 ip6_sprintf(ip6buf, &dst->sin6_addr), lle); 2208 m_freem(m); 2209 return (ENOBUFS); 2210 } 2211 2212 IF_AFDATA_WLOCK(ifp); 2213 LLE_WLOCK(lle); 2214 /* Prefer any existing entry over newly-created one */ 2215 lle_tmp = nd6_lookup(&dst->sin6_addr, LLE_EXCLUSIVE, ifp); 2216 if (lle_tmp == NULL) 2217 lltable_link_entry(LLTABLE6(ifp), lle); 2218 IF_AFDATA_WUNLOCK(ifp); 2219 if (lle_tmp != NULL) { 2220 lltable_free_entry(LLTABLE6(ifp), lle); 2221 lle = lle_tmp; 2222 lle_tmp = NULL; 2223 } 2224 } 2225 } 2226 if (lle == NULL) { 2227 if (!(ND_IFINFO(ifp)->flags & ND6_IFF_PERFORMNUD)) { 2228 m_freem(m); 2229 return (ENOBUFS); 2230 } 2231 2232 if (m != NULL) 2233 m_freem(m); 2234 return (ENOBUFS); 2235 } 2236 2237 LLE_WLOCK_ASSERT(lle); 2238 2239 /* 2240 * The first time we send a packet to a neighbor whose entry is 2241 * STALE, we have to change the state to DELAY and a sets a timer to 2242 * expire in DELAY_FIRST_PROBE_TIME seconds to ensure do 2243 * neighbor unreachability detection on expiration. 2244 * (RFC 2461 7.3.3) 2245 */ 2246 if (lle->ln_state == ND6_LLINFO_STALE) 2247 nd6_llinfo_setstate(lle, ND6_LLINFO_DELAY); 2248 2249 /* 2250 * If the neighbor cache entry has a state other than INCOMPLETE 2251 * (i.e. its link-layer address is already resolved), just 2252 * send the packet. 2253 */ 2254 if (lle->ln_state > ND6_LLINFO_INCOMPLETE) { 2255 bcopy(&lle->ll_addr, desten, ifp->if_addrlen); 2256 if (pflags != NULL) 2257 *pflags = lle->la_flags; 2258 LLE_WUNLOCK(lle); 2259 return (0); 2260 } 2261 2262 /* 2263 * There is a neighbor cache entry, but no ethernet address 2264 * response yet. Append this latest packet to the end of the 2265 * packet queue in the mbuf, unless the number of the packet 2266 * does not exceed nd6_maxqueuelen. When it exceeds nd6_maxqueuelen, 2267 * the oldest packet in the queue will be removed. 2268 */ 2269 2270 if (lle->la_hold != NULL) { 2271 struct mbuf *m_hold; 2272 int i; 2273 2274 i = 0; 2275 for (m_hold = lle->la_hold; m_hold; m_hold = m_hold->m_nextpkt){ 2276 i++; 2277 if (m_hold->m_nextpkt == NULL) { 2278 m_hold->m_nextpkt = m; 2279 break; 2280 } 2281 } 2282 while (i >= V_nd6_maxqueuelen) { 2283 m_hold = lle->la_hold; 2284 lle->la_hold = lle->la_hold->m_nextpkt; 2285 m_freem(m_hold); 2286 i--; 2287 } 2288 } else { 2289 lle->la_hold = m; 2290 } 2291 2292 /* 2293 * If there has been no NS for the neighbor after entering the 2294 * INCOMPLETE state, send the first solicitation. 2295 * Note that for newly-created lle la_asked will be 0, 2296 * so we will transition from ND6_LLINFO_NOSTATE to 2297 * ND6_LLINFO_INCOMPLETE state here. 2298 */ 2299 psrc = NULL; 2300 send_ns = 0; 2301 if (lle->la_asked == 0) { 2302 lle->la_asked++; 2303 send_ns = 1; 2304 psrc = nd6_llinfo_get_holdsrc(lle, &src); 2305 2306 nd6_llinfo_setstate(lle, ND6_LLINFO_INCOMPLETE); 2307 } 2308 LLE_WUNLOCK(lle); 2309 if (send_ns != 0) 2310 nd6_ns_output(ifp, psrc, NULL, &dst->sin6_addr, NULL); 2311 2312 return (EWOULDBLOCK); 2313 } 2314 2315 2316 int 2317 nd6_flush_holdchain(struct ifnet *ifp, struct ifnet *origifp, struct mbuf *chain, 2318 struct sockaddr_in6 *dst) 2319 { 2320 struct mbuf *m, *m_head; 2321 struct ifnet *outifp; 2322 int error = 0; 2323 2324 m_head = chain; 2325 if ((ifp->if_flags & IFF_LOOPBACK) != 0) 2326 outifp = origifp; 2327 else 2328 outifp = ifp; 2329 2330 while (m_head) { 2331 m = m_head; 2332 m_head = m_head->m_nextpkt; 2333 error = nd6_output_ifp(ifp, origifp, m, dst, NULL); 2334 } 2335 2336 /* 2337 * XXX 2338 * note that intermediate errors are blindly ignored 2339 */ 2340 return (error); 2341 } 2342 2343 static int 2344 nd6_need_cache(struct ifnet *ifp) 2345 { 2346 /* 2347 * XXX: we currently do not make neighbor cache on any interface 2348 * other than ARCnet, Ethernet, FDDI and GIF. 2349 * 2350 * RFC2893 says: 2351 * - unidirectional tunnels needs no ND 2352 */ 2353 switch (ifp->if_type) { 2354 case IFT_ARCNET: 2355 case IFT_ETHER: 2356 case IFT_FDDI: 2357 case IFT_IEEE1394: 2358 case IFT_L2VLAN: 2359 case IFT_IEEE80211: 2360 case IFT_INFINIBAND: 2361 case IFT_BRIDGE: 2362 case IFT_PROPVIRTUAL: 2363 return (1); 2364 default: 2365 return (0); 2366 } 2367 } 2368 2369 /* 2370 * Add pernament ND6 link-layer record for given 2371 * interface address. 2372 * 2373 * Very similar to IPv4 arp_ifinit(), but: 2374 * 1) IPv6 DAD is performed in different place 2375 * 2) It is called by IPv6 protocol stack in contrast to 2376 * arp_ifinit() which is typically called in SIOCSIFADDR 2377 * driver ioctl handler. 2378 * 2379 */ 2380 int 2381 nd6_add_ifa_lle(struct in6_ifaddr *ia) 2382 { 2383 struct ifnet *ifp; 2384 struct llentry *ln, *ln_tmp; 2385 struct sockaddr *dst; 2386 2387 ifp = ia->ia_ifa.ifa_ifp; 2388 if (nd6_need_cache(ifp) == 0) 2389 return (0); 2390 2391 ia->ia_ifa.ifa_rtrequest = nd6_rtrequest; 2392 dst = (struct sockaddr *)&ia->ia_addr; 2393 ln = lltable_alloc_entry(LLTABLE6(ifp), LLE_IFADDR, dst); 2394 if (ln == NULL) 2395 return (ENOBUFS); 2396 2397 IF_AFDATA_WLOCK(ifp); 2398 LLE_WLOCK(ln); 2399 /* Unlink any entry if exists */ 2400 ln_tmp = lla_lookup(LLTABLE6(ifp), LLE_EXCLUSIVE, dst); 2401 if (ln_tmp != NULL) 2402 lltable_unlink_entry(LLTABLE6(ifp), ln_tmp); 2403 lltable_link_entry(LLTABLE6(ifp), ln); 2404 IF_AFDATA_WUNLOCK(ifp); 2405 2406 if (ln_tmp != NULL) 2407 EVENTHANDLER_INVOKE(lle_event, ln_tmp, LLENTRY_EXPIRED); 2408 EVENTHANDLER_INVOKE(lle_event, ln, LLENTRY_RESOLVED); 2409 2410 LLE_WUNLOCK(ln); 2411 if (ln_tmp != NULL) 2412 llentry_free(ln_tmp); 2413 2414 return (0); 2415 } 2416 2417 /* 2418 * Removes either all lle entries for given @ia, or lle 2419 * corresponding to @ia address. 2420 */ 2421 void 2422 nd6_rem_ifa_lle(struct in6_ifaddr *ia, int all) 2423 { 2424 struct sockaddr_in6 mask, addr; 2425 struct sockaddr *saddr, *smask; 2426 struct ifnet *ifp; 2427 2428 ifp = ia->ia_ifa.ifa_ifp; 2429 memcpy(&addr, &ia->ia_addr, sizeof(ia->ia_addr)); 2430 memcpy(&mask, &ia->ia_prefixmask, sizeof(ia->ia_prefixmask)); 2431 saddr = (struct sockaddr *)&addr; 2432 smask = (struct sockaddr *)&mask; 2433 2434 if (all != 0) 2435 lltable_prefix_free(AF_INET6, saddr, smask, LLE_STATIC); 2436 else 2437 lltable_delete_addr(LLTABLE6(ifp), LLE_IFADDR, saddr); 2438 } 2439 2440 static void 2441 clear_llinfo_pqueue(struct llentry *ln) 2442 { 2443 struct mbuf *m_hold, *m_hold_next; 2444 2445 for (m_hold = ln->la_hold; m_hold; m_hold = m_hold_next) { 2446 m_hold_next = m_hold->m_nextpkt; 2447 m_freem(m_hold); 2448 } 2449 2450 ln->la_hold = NULL; 2451 return; 2452 } 2453 2454 static int nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS); 2455 static int nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS); 2456 #ifdef SYSCTL_DECL 2457 SYSCTL_DECL(_net_inet6_icmp6); 2458 #endif 2459 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_DRLIST, nd6_drlist, 2460 CTLFLAG_RD, nd6_sysctl_drlist, ""); 2461 SYSCTL_NODE(_net_inet6_icmp6, ICMPV6CTL_ND6_PRLIST, nd6_prlist, 2462 CTLFLAG_RD, nd6_sysctl_prlist, ""); 2463 SYSCTL_INT(_net_inet6_icmp6, ICMPV6CTL_ND6_MAXQLEN, nd6_maxqueuelen, 2464 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_maxqueuelen), 1, ""); 2465 SYSCTL_INT(_net_inet6_icmp6, OID_AUTO, nd6_gctimer, 2466 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nd6_gctimer), (60 * 60 * 24), ""); 2467 2468 static int 2469 nd6_sysctl_drlist(SYSCTL_HANDLER_ARGS) 2470 { 2471 struct in6_defrouter d; 2472 struct nd_defrouter *dr; 2473 int error; 2474 2475 if (req->newptr) 2476 return (EPERM); 2477 2478 bzero(&d, sizeof(d)); 2479 d.rtaddr.sin6_family = AF_INET6; 2480 d.rtaddr.sin6_len = sizeof(d.rtaddr); 2481 2482 /* 2483 * XXX locking 2484 */ 2485 TAILQ_FOREACH(dr, &V_nd_defrouter, dr_entry) { 2486 d.rtaddr.sin6_addr = dr->rtaddr; 2487 error = sa6_recoverscope(&d.rtaddr); 2488 if (error != 0) 2489 return (error); 2490 d.flags = dr->flags; 2491 d.rtlifetime = dr->rtlifetime; 2492 d.expire = dr->expire + (time_second - time_uptime); 2493 d.if_index = dr->ifp->if_index; 2494 error = SYSCTL_OUT(req, &d, sizeof(d)); 2495 if (error != 0) 2496 return (error); 2497 } 2498 return (0); 2499 } 2500 2501 static int 2502 nd6_sysctl_prlist(SYSCTL_HANDLER_ARGS) 2503 { 2504 struct in6_prefix p; 2505 struct sockaddr_in6 s6; 2506 struct nd_prefix *pr; 2507 struct nd_pfxrouter *pfr; 2508 time_t maxexpire; 2509 int error; 2510 char ip6buf[INET6_ADDRSTRLEN]; 2511 2512 if (req->newptr) 2513 return (EPERM); 2514 2515 bzero(&p, sizeof(p)); 2516 p.origin = PR_ORIG_RA; 2517 bzero(&s6, sizeof(s6)); 2518 s6.sin6_family = AF_INET6; 2519 s6.sin6_len = sizeof(s6); 2520 2521 /* 2522 * XXX locking 2523 */ 2524 LIST_FOREACH(pr, &V_nd_prefix, ndpr_entry) { 2525 p.prefix = pr->ndpr_prefix; 2526 if (sa6_recoverscope(&p.prefix)) { 2527 log(LOG_ERR, "scope error in prefix list (%s)\n", 2528 ip6_sprintf(ip6buf, &p.prefix.sin6_addr)); 2529 /* XXX: press on... */ 2530 } 2531 p.raflags = pr->ndpr_raf; 2532 p.prefixlen = pr->ndpr_plen; 2533 p.vltime = pr->ndpr_vltime; 2534 p.pltime = pr->ndpr_pltime; 2535 p.if_index = pr->ndpr_ifp->if_index; 2536 if (pr->ndpr_vltime == ND6_INFINITE_LIFETIME) 2537 p.expire = 0; 2538 else { 2539 /* XXX: we assume time_t is signed. */ 2540 maxexpire = (-1) & 2541 ~((time_t)1 << ((sizeof(maxexpire) * 8) - 1)); 2542 if (pr->ndpr_vltime < maxexpire - pr->ndpr_lastupdate) 2543 p.expire = pr->ndpr_lastupdate + 2544 pr->ndpr_vltime + 2545 (time_second - time_uptime); 2546 else 2547 p.expire = maxexpire; 2548 } 2549 p.refcnt = pr->ndpr_refcnt; 2550 p.flags = pr->ndpr_stateflags; 2551 p.advrtrs = 0; 2552 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) 2553 p.advrtrs++; 2554 error = SYSCTL_OUT(req, &p, sizeof(p)); 2555 if (error != 0) 2556 return (error); 2557 LIST_FOREACH(pfr, &pr->ndpr_advrtrs, pfr_entry) { 2558 s6.sin6_addr = pfr->router->rtaddr; 2559 if (sa6_recoverscope(&s6)) 2560 log(LOG_ERR, 2561 "scope error in prefix list (%s)\n", 2562 ip6_sprintf(ip6buf, &pfr->router->rtaddr)); 2563 error = SYSCTL_OUT(req, &s6, sizeof(s6)); 2564 if (error != 0) 2565 return (error); 2566 } 2567 } 2568 return (0); 2569 } 2570