1 /*- 2 * Copyright (c) 2009 Bruce Simpson. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 3. The name of the author may not be used to endorse or promote 13 * products derived from this software without specific prior written 14 * permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $ 29 */ 30 31 /*- 32 * Copyright (c) 1988 Stephen Deering. 33 * Copyright (c) 1992, 1993 34 * The Regents of the University of California. All rights reserved. 35 * 36 * This code is derived from software contributed to Berkeley by 37 * Stephen Deering of Stanford University. 38 * 39 * Redistribution and use in source and binary forms, with or without 40 * modification, are permitted provided that the following conditions 41 * are met: 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 4. Neither the name of the University nor the names of its contributors 48 * may be used to endorse or promote products derived from this software 49 * without specific prior written permission. 50 * 51 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 52 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 53 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 54 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 55 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 56 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 57 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 58 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 59 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 60 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 61 * SUCH DAMAGE. 62 * 63 * @(#)igmp.c 8.1 (Berkeley) 7/19/93 64 */ 65 66 #include <sys/cdefs.h> 67 __FBSDID("$FreeBSD$"); 68 69 #include "opt_inet.h" 70 #include "opt_inet6.h" 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/mbuf.h> 75 #include <sys/socket.h> 76 #include <sys/protosw.h> 77 #include <sys/sysctl.h> 78 #include <sys/kernel.h> 79 #include <sys/callout.h> 80 #include <sys/malloc.h> 81 #include <sys/module.h> 82 #include <sys/ktr.h> 83 84 #include <net/if.h> 85 #include <net/route.h> 86 #include <net/vnet.h> 87 88 #include <netinet/in.h> 89 #include <netinet/in_var.h> 90 #include <netinet6/in6_var.h> 91 #include <netinet/ip6.h> 92 #include <netinet6/ip6_var.h> 93 #include <netinet6/scope6_var.h> 94 #include <netinet/icmp6.h> 95 #include <netinet6/mld6.h> 96 #include <netinet6/mld6_var.h> 97 98 #include <security/mac/mac_framework.h> 99 100 #ifndef KTR_MLD 101 #define KTR_MLD KTR_INET6 102 #endif 103 104 static struct mld_ifinfo * 105 mli_alloc_locked(struct ifnet *); 106 static void mli_delete_locked(const struct ifnet *); 107 static void mld_dispatch_packet(struct mbuf *); 108 static void mld_dispatch_queue(struct ifqueue *, int); 109 static void mld_final_leave(struct in6_multi *, struct mld_ifinfo *); 110 static void mld_fasttimo_vnet(void); 111 static int mld_handle_state_change(struct in6_multi *, 112 struct mld_ifinfo *); 113 static int mld_initial_join(struct in6_multi *, struct mld_ifinfo *, 114 const int); 115 #ifdef KTR 116 static char * mld_rec_type_to_str(const int); 117 #endif 118 static void mld_set_version(struct mld_ifinfo *, const int); 119 static void mld_slowtimo_vnet(void); 120 static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *, 121 /*const*/ struct mld_hdr *); 122 static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *, 123 /*const*/ struct mld_hdr *); 124 static void mld_v1_process_group_timer(struct mld_ifinfo *, 125 struct in6_multi *); 126 static void mld_v1_process_querier_timers(struct mld_ifinfo *); 127 static int mld_v1_transmit_report(struct in6_multi *, const int); 128 static void mld_v1_update_group(struct in6_multi *, const int); 129 static void mld_v2_cancel_link_timers(struct mld_ifinfo *); 130 static void mld_v2_dispatch_general_query(struct mld_ifinfo *); 131 static struct mbuf * 132 mld_v2_encap_report(struct ifnet *, struct mbuf *); 133 static int mld_v2_enqueue_filter_change(struct ifqueue *, 134 struct in6_multi *); 135 static int mld_v2_enqueue_group_record(struct ifqueue *, 136 struct in6_multi *, const int, const int, const int, 137 const int); 138 static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *, 139 struct mbuf *, const int, const int); 140 static int mld_v2_merge_state_changes(struct in6_multi *, 141 struct ifqueue *); 142 static void mld_v2_process_group_timers(struct mld_ifinfo *, 143 struct ifqueue *, struct ifqueue *, 144 struct in6_multi *, const int); 145 static int mld_v2_process_group_query(struct in6_multi *, 146 struct mld_ifinfo *mli, int, struct mbuf *, const int); 147 static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); 148 static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS); 149 150 /* 151 * Normative references: RFC 2710, RFC 3590, RFC 3810. 152 * 153 * Locking: 154 * * The MLD subsystem lock ends up being system-wide for the moment, 155 * but could be per-VIMAGE later on. 156 * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. 157 * Any may be taken independently; if any are held at the same 158 * time, the above lock order must be followed. 159 * * IN6_MULTI_LOCK covers in_multi. 160 * * MLD_LOCK covers per-link state and any global variables in this file. 161 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 162 * per-link state iterators. 163 * 164 * XXX LOR PREVENTION 165 * A special case for IPv6 is the in6_setscope() routine. ip6_output() 166 * will not accept an ifp; it wants an embedded scope ID, unlike 167 * ip_output(), which happily takes the ifp given to it. The embedded 168 * scope ID is only used by MLD to select the outgoing interface. 169 * 170 * During interface attach and detach, MLD will take MLD_LOCK *after* 171 * the IF_AFDATA_LOCK. 172 * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call 173 * it with MLD_LOCK held without triggering an LOR. A netisr with indirect 174 * dispatch could work around this, but we'd rather not do that, as it 175 * can introduce other races. 176 * 177 * As such, we exploit the fact that the scope ID is just the interface 178 * index, and embed it in the IPv6 destination address accordingly. 179 * This is potentially NOT VALID for MLDv1 reports, as they 180 * are always sent to the multicast group itself; as MLDv2 181 * reports are always sent to ff02::16, this is not an issue 182 * when MLDv2 is in use. 183 * 184 * This does not however eliminate the LOR when ip6_output() itself 185 * calls in6_setscope() internally whilst MLD_LOCK is held. This will 186 * trigger a LOR warning in WITNESS when the ifnet is detached. 187 * 188 * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given 189 * how it's used across the network stack. Here we're simply exploiting 190 * the fact that MLD runs at a similar layer in the stack to scope6.c. 191 * 192 * VIMAGE: 193 * * Each in6_multi corresponds to an ifp, and each ifp corresponds 194 * to a vnet in ifp->if_vnet. 195 */ 196 static struct mtx mld_mtx; 197 static MALLOC_DEFINE(M_MLD, "mld", "mld state"); 198 199 #define MLD_EMBEDSCOPE(pin6, zoneid) \ 200 if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \ 201 IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \ 202 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \ 203 204 /* 205 * VIMAGE-wide globals. 206 */ 207 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0}; 208 static VNET_DEFINE(LIST_HEAD(, mld_ifinfo), mli_head); 209 static VNET_DEFINE(int, interface_timers_running6); 210 static VNET_DEFINE(int, state_change_timers_running6); 211 static VNET_DEFINE(int, current_state_timers_running6); 212 213 #define V_mld_gsrdelay VNET(mld_gsrdelay) 214 #define V_mli_head VNET(mli_head) 215 #define V_interface_timers_running6 VNET(interface_timers_running6) 216 #define V_state_change_timers_running6 VNET(state_change_timers_running6) 217 #define V_current_state_timers_running6 VNET(current_state_timers_running6) 218 219 SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ 220 221 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, 222 "IPv6 Multicast Listener Discovery"); 223 224 /* 225 * Virtualized sysctls. 226 */ 227 SYSCTL_VNET_PROC(_net_inet6_mld, OID_AUTO, gsrdelay, 228 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 229 &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I", 230 "Rate limit for MLDv2 Group-and-Source queries in seconds"); 231 232 /* 233 * Non-virtualized sysctls. 234 */ 235 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, 236 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo, 237 "Per-interface MLDv2 state"); 238 239 static int mld_v1enable = 1; 240 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW, 241 &mld_v1enable, 0, "Enable fallback to MLDv1"); 242 TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable); 243 244 static int mld_use_allow = 1; 245 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW, 246 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves"); 247 TUNABLE_INT("net.inet6.mld.use_allow", &mld_use_allow); 248 249 /* 250 * Packed Router Alert option structure declaration. 251 */ 252 struct mld_raopt { 253 struct ip6_hbh hbh; 254 struct ip6_opt pad; 255 struct ip6_opt_router ra; 256 } __packed; 257 258 /* 259 * Router Alert hop-by-hop option header. 260 */ 261 static struct mld_raopt mld_ra = { 262 .hbh = { 0, 0 }, 263 .pad = { .ip6o_type = IP6OPT_PADN, 0 }, 264 .ra = { 265 .ip6or_type = IP6OPT_ROUTER_ALERT, 266 .ip6or_len = IP6OPT_RTALERT_LEN - 2, 267 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), 268 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) 269 } 270 }; 271 static struct ip6_pktopts mld_po; 272 273 static __inline void 274 mld_save_context(struct mbuf *m, struct ifnet *ifp) 275 { 276 277 #ifdef VIMAGE 278 m->m_pkthdr.header = ifp->if_vnet; 279 #endif /* VIMAGE */ 280 m->m_pkthdr.flowid = ifp->if_index; 281 } 282 283 static __inline void 284 mld_scrub_context(struct mbuf *m) 285 { 286 287 m->m_pkthdr.header = NULL; 288 m->m_pkthdr.flowid = 0; 289 } 290 291 /* 292 * Restore context from a queued output chain. 293 * Return saved ifindex. 294 * 295 * VIMAGE: The assertion is there to make sure that we 296 * actually called CURVNET_SET() with what's in the mbuf chain. 297 */ 298 static __inline uint32_t 299 mld_restore_context(struct mbuf *m) 300 { 301 302 #if defined(VIMAGE) && defined(INVARIANTS) 303 KASSERT(curvnet == m->m_pkthdr.header, 304 ("%s: called when curvnet was not restored", __func__)); 305 #endif 306 return (m->m_pkthdr.flowid); 307 } 308 309 /* 310 * Retrieve or set threshold between group-source queries in seconds. 311 * 312 * VIMAGE: Assume curvnet set by caller. 313 * SMPng: NOTE: Serialized by MLD lock. 314 */ 315 static int 316 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS) 317 { 318 int error; 319 int i; 320 321 error = sysctl_wire_old_buffer(req, sizeof(int)); 322 if (error) 323 return (error); 324 325 MLD_LOCK(); 326 327 i = V_mld_gsrdelay.tv_sec; 328 329 error = sysctl_handle_int(oidp, &i, 0, req); 330 if (error || !req->newptr) 331 goto out_locked; 332 333 if (i < -1 || i >= 60) { 334 error = EINVAL; 335 goto out_locked; 336 } 337 338 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", 339 V_mld_gsrdelay.tv_sec, i); 340 V_mld_gsrdelay.tv_sec = i; 341 342 out_locked: 343 MLD_UNLOCK(); 344 return (error); 345 } 346 347 /* 348 * Expose struct mld_ifinfo to userland, keyed by ifindex. 349 * For use by ifmcstat(8). 350 * 351 * SMPng: NOTE: Does an unlocked ifindex space read. 352 * VIMAGE: Assume curvnet set by caller. The node handler itself 353 * is not directly virtualized. 354 */ 355 static int 356 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) 357 { 358 int *name; 359 int error; 360 u_int namelen; 361 struct ifnet *ifp; 362 struct mld_ifinfo *mli; 363 364 name = (int *)arg1; 365 namelen = arg2; 366 367 if (req->newptr != NULL) 368 return (EPERM); 369 370 if (namelen != 1) 371 return (EINVAL); 372 373 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo)); 374 if (error) 375 return (error); 376 377 IN6_MULTI_LOCK(); 378 MLD_LOCK(); 379 380 if (name[0] <= 0 || name[0] > V_if_index) { 381 error = ENOENT; 382 goto out_locked; 383 } 384 385 error = ENOENT; 386 387 ifp = ifnet_byindex(name[0]); 388 if (ifp == NULL) 389 goto out_locked; 390 391 LIST_FOREACH(mli, &V_mli_head, mli_link) { 392 if (ifp == mli->mli_ifp) { 393 error = SYSCTL_OUT(req, mli, 394 sizeof(struct mld_ifinfo)); 395 break; 396 } 397 } 398 399 out_locked: 400 MLD_UNLOCK(); 401 IN6_MULTI_UNLOCK(); 402 return (error); 403 } 404 405 /* 406 * Dispatch an entire queue of pending packet chains. 407 * VIMAGE: Assumes the vnet pointer has been set. 408 */ 409 static void 410 mld_dispatch_queue(struct ifqueue *ifq, int limit) 411 { 412 struct mbuf *m; 413 414 for (;;) { 415 _IF_DEQUEUE(ifq, m); 416 if (m == NULL) 417 break; 418 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m); 419 mld_dispatch_packet(m); 420 if (--limit == 0) 421 break; 422 } 423 } 424 425 /* 426 * Filter outgoing MLD report state by group. 427 * 428 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1) 429 * and node-local addresses. However, kernel and socket consumers 430 * always embed the KAME scope ID in the address provided, so strip it 431 * when performing comparison. 432 * Note: This is not the same as the *multicast* scope. 433 * 434 * Return zero if the given group is one for which MLD reports 435 * should be suppressed, or non-zero if reports should be issued. 436 */ 437 static __inline int 438 mld_is_addr_reported(const struct in6_addr *addr) 439 { 440 441 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__)); 442 443 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) 444 return (0); 445 446 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) { 447 struct in6_addr tmp = *addr; 448 in6_clearscope(&tmp); 449 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) 450 return (0); 451 } 452 453 return (1); 454 } 455 456 /* 457 * Attach MLD when PF_INET6 is attached to an interface. 458 * 459 * SMPng: Normally called with IF_AFDATA_LOCK held. 460 */ 461 struct mld_ifinfo * 462 mld_domifattach(struct ifnet *ifp) 463 { 464 struct mld_ifinfo *mli; 465 466 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 467 __func__, ifp, ifp->if_xname); 468 469 MLD_LOCK(); 470 471 mli = mli_alloc_locked(ifp); 472 if (!(ifp->if_flags & IFF_MULTICAST)) 473 mli->mli_flags |= MLIF_SILENT; 474 if (mld_use_allow) 475 mli->mli_flags |= MLIF_USEALLOW; 476 477 MLD_UNLOCK(); 478 479 return (mli); 480 } 481 482 /* 483 * VIMAGE: assume curvnet set by caller. 484 */ 485 static struct mld_ifinfo * 486 mli_alloc_locked(/*const*/ struct ifnet *ifp) 487 { 488 struct mld_ifinfo *mli; 489 490 MLD_LOCK_ASSERT(); 491 492 mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO); 493 if (mli == NULL) 494 goto out; 495 496 mli->mli_ifp = ifp; 497 mli->mli_version = MLD_VERSION_2; 498 mli->mli_flags = 0; 499 mli->mli_rv = MLD_RV_INIT; 500 mli->mli_qi = MLD_QI_INIT; 501 mli->mli_qri = MLD_QRI_INIT; 502 mli->mli_uri = MLD_URI_INIT; 503 504 SLIST_INIT(&mli->mli_relinmhead); 505 506 /* 507 * Responses to general queries are subject to bounds. 508 */ 509 IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); 510 511 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); 512 513 CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)", 514 ifp, ifp->if_xname); 515 516 out: 517 return (mli); 518 } 519 520 /* 521 * Hook for ifdetach. 522 * 523 * NOTE: Some finalization tasks need to run before the protocol domain 524 * is detached, but also before the link layer does its cleanup. 525 * Run before link-layer cleanup; cleanup groups, but do not free MLD state. 526 * 527 * SMPng: Caller must hold IN6_MULTI_LOCK(). 528 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. 529 * XXX This routine is also bitten by unlocked ifma_protospec access. 530 */ 531 void 532 mld_ifdetach(struct ifnet *ifp) 533 { 534 struct mld_ifinfo *mli; 535 struct ifmultiaddr *ifma; 536 struct in6_multi *inm, *tinm; 537 538 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, 539 ifp->if_xname); 540 541 IN6_MULTI_LOCK_ASSERT(); 542 MLD_LOCK(); 543 544 mli = MLD_IFINFO(ifp); 545 if (mli->mli_version == MLD_VERSION_2) { 546 IF_ADDR_RLOCK(ifp); 547 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 548 if (ifma->ifma_addr->sa_family != AF_INET6 || 549 ifma->ifma_protospec == NULL) 550 continue; 551 inm = (struct in6_multi *)ifma->ifma_protospec; 552 if (inm->in6m_state == MLD_LEAVING_MEMBER) { 553 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 554 inm, in6m_nrele); 555 } 556 in6m_clear_recorded(inm); 557 } 558 IF_ADDR_RUNLOCK(ifp); 559 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, 560 tinm) { 561 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 562 in6m_release_locked(inm); 563 } 564 } 565 566 MLD_UNLOCK(); 567 } 568 569 /* 570 * Hook for domifdetach. 571 * Runs after link-layer cleanup; free MLD state. 572 * 573 * SMPng: Normally called with IF_AFDATA_LOCK held. 574 */ 575 void 576 mld_domifdetach(struct ifnet *ifp) 577 { 578 579 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 580 __func__, ifp, ifp->if_xname); 581 582 MLD_LOCK(); 583 mli_delete_locked(ifp); 584 MLD_UNLOCK(); 585 } 586 587 static void 588 mli_delete_locked(const struct ifnet *ifp) 589 { 590 struct mld_ifinfo *mli, *tmli; 591 592 CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)", 593 __func__, ifp, ifp->if_xname); 594 595 MLD_LOCK_ASSERT(); 596 597 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) { 598 if (mli->mli_ifp == ifp) { 599 /* 600 * Free deferred General Query responses. 601 */ 602 _IF_DRAIN(&mli->mli_gq); 603 604 LIST_REMOVE(mli, mli_link); 605 606 KASSERT(SLIST_EMPTY(&mli->mli_relinmhead), 607 ("%s: there are dangling in_multi references", 608 __func__)); 609 610 free(mli, M_MLD); 611 return; 612 } 613 } 614 #ifdef INVARIANTS 615 panic("%s: mld_ifinfo not found for ifp %p\n", __func__, ifp); 616 #endif 617 } 618 619 /* 620 * Process a received MLDv1 general or address-specific query. 621 * Assumes that the query header has been pulled up to sizeof(mld_hdr). 622 * 623 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 624 * mld_addr. This is OK as we own the mbuf chain. 625 */ 626 static int 627 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 628 /*const*/ struct mld_hdr *mld) 629 { 630 struct ifmultiaddr *ifma; 631 struct mld_ifinfo *mli; 632 struct in6_multi *inm; 633 int is_general_query; 634 uint16_t timer; 635 #ifdef KTR 636 char ip6tbuf[INET6_ADDRSTRLEN]; 637 #endif 638 639 is_general_query = 0; 640 641 if (!mld_v1enable) { 642 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)", 643 ip6_sprintf(ip6tbuf, &mld->mld_addr), 644 ifp, ifp->if_xname); 645 return (0); 646 } 647 648 /* 649 * RFC3810 Section 6.2: MLD queries must originate from 650 * a router's link-local address. 651 */ 652 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 653 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 654 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 655 ifp, ifp->if_xname); 656 return (0); 657 } 658 659 /* 660 * Do address field validation upfront before we accept 661 * the query. 662 */ 663 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 664 /* 665 * MLDv1 General Query. 666 * If this was not sent to the all-nodes group, ignore it. 667 */ 668 struct in6_addr dst; 669 670 dst = ip6->ip6_dst; 671 in6_clearscope(&dst); 672 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) 673 return (EINVAL); 674 is_general_query = 1; 675 } else { 676 /* 677 * Embed scope ID of receiving interface in MLD query for 678 * lookup whilst we don't hold other locks. 679 */ 680 in6_setscope(&mld->mld_addr, ifp, NULL); 681 } 682 683 IN6_MULTI_LOCK(); 684 MLD_LOCK(); 685 686 /* 687 * Switch to MLDv1 host compatibility mode. 688 */ 689 mli = MLD_IFINFO(ifp); 690 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 691 mld_set_version(mli, MLD_VERSION_1); 692 693 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; 694 if (timer == 0) 695 timer = 1; 696 697 IF_ADDR_RLOCK(ifp); 698 if (is_general_query) { 699 /* 700 * For each reporting group joined on this 701 * interface, kick the report timer. 702 */ 703 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)", 704 ifp, ifp->if_xname); 705 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 706 if (ifma->ifma_addr->sa_family != AF_INET6 || 707 ifma->ifma_protospec == NULL) 708 continue; 709 inm = (struct in6_multi *)ifma->ifma_protospec; 710 mld_v1_update_group(inm, timer); 711 } 712 } else { 713 /* 714 * MLDv1 Group-Specific Query. 715 * If this is a group-specific MLDv1 query, we need only 716 * look up the single group to process it. 717 */ 718 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 719 if (inm != NULL) { 720 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)", 721 ip6_sprintf(ip6tbuf, &mld->mld_addr), 722 ifp, ifp->if_xname); 723 mld_v1_update_group(inm, timer); 724 } 725 /* XXX Clear embedded scope ID as userland won't expect it. */ 726 in6_clearscope(&mld->mld_addr); 727 } 728 729 IF_ADDR_RUNLOCK(ifp); 730 MLD_UNLOCK(); 731 IN6_MULTI_UNLOCK(); 732 733 return (0); 734 } 735 736 /* 737 * Update the report timer on a group in response to an MLDv1 query. 738 * 739 * If we are becoming the reporting member for this group, start the timer. 740 * If we already are the reporting member for this group, and timer is 741 * below the threshold, reset it. 742 * 743 * We may be updating the group for the first time since we switched 744 * to MLDv2. If we are, then we must clear any recorded source lists, 745 * and transition to REPORTING state; the group timer is overloaded 746 * for group and group-source query responses. 747 * 748 * Unlike MLDv2, the delay per group should be jittered 749 * to avoid bursts of MLDv1 reports. 750 */ 751 static void 752 mld_v1_update_group(struct in6_multi *inm, const int timer) 753 { 754 #ifdef KTR 755 char ip6tbuf[INET6_ADDRSTRLEN]; 756 #endif 757 758 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, 759 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 760 inm->in6m_ifp->if_xname, timer); 761 762 IN6_MULTI_LOCK_ASSERT(); 763 764 switch (inm->in6m_state) { 765 case MLD_NOT_MEMBER: 766 case MLD_SILENT_MEMBER: 767 break; 768 case MLD_REPORTING_MEMBER: 769 if (inm->in6m_timer != 0 && 770 inm->in6m_timer <= timer) { 771 CTR1(KTR_MLD, "%s: REPORTING and timer running, " 772 "skipping.", __func__); 773 break; 774 } 775 /* FALLTHROUGH */ 776 case MLD_SG_QUERY_PENDING_MEMBER: 777 case MLD_G_QUERY_PENDING_MEMBER: 778 case MLD_IDLE_MEMBER: 779 case MLD_LAZY_MEMBER: 780 case MLD_AWAKENING_MEMBER: 781 CTR1(KTR_MLD, "%s: ->REPORTING", __func__); 782 inm->in6m_state = MLD_REPORTING_MEMBER; 783 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 784 V_current_state_timers_running6 = 1; 785 break; 786 case MLD_SLEEPING_MEMBER: 787 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); 788 inm->in6m_state = MLD_AWAKENING_MEMBER; 789 break; 790 case MLD_LEAVING_MEMBER: 791 break; 792 } 793 } 794 795 /* 796 * Process a received MLDv2 general, group-specific or 797 * group-and-source-specific query. 798 * 799 * Assumes that the query header has been pulled up to sizeof(mldv2_query). 800 * 801 * Return 0 if successful, otherwise an appropriate error code is returned. 802 */ 803 static int 804 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 805 struct mbuf *m, const int off, const int icmp6len) 806 { 807 struct mld_ifinfo *mli; 808 struct mldv2_query *mld; 809 struct in6_multi *inm; 810 uint32_t maxdelay, nsrc, qqi; 811 int is_general_query; 812 uint16_t timer; 813 uint8_t qrv; 814 #ifdef KTR 815 char ip6tbuf[INET6_ADDRSTRLEN]; 816 #endif 817 818 is_general_query = 0; 819 820 /* 821 * RFC3810 Section 6.2: MLD queries must originate from 822 * a router's link-local address. 823 */ 824 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 825 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 826 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 827 ifp, ifp->if_xname); 828 return (0); 829 } 830 831 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname); 832 833 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off); 834 835 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ 836 if (maxdelay >= 32678) { 837 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << 838 (MLD_MRC_EXP(maxdelay) + 3); 839 } 840 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; 841 if (timer == 0) 842 timer = 1; 843 844 qrv = MLD_QRV(mld->mld_misc); 845 if (qrv < 2) { 846 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, 847 qrv, MLD_RV_INIT); 848 qrv = MLD_RV_INIT; 849 } 850 851 qqi = mld->mld_qqi; 852 if (qqi >= 128) { 853 qqi = MLD_QQIC_MANT(mld->mld_qqi) << 854 (MLD_QQIC_EXP(mld->mld_qqi) + 3); 855 } 856 857 nsrc = ntohs(mld->mld_numsrc); 858 if (nsrc > MLD_MAX_GS_SOURCES) 859 return (EMSGSIZE); 860 if (icmp6len < sizeof(struct mldv2_query) + 861 (nsrc * sizeof(struct in6_addr))) 862 return (EMSGSIZE); 863 864 /* 865 * Do further input validation upfront to avoid resetting timers 866 * should we need to discard this query. 867 */ 868 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 869 /* 870 * General Queries SHOULD be directed to ff02::1. 871 * A general query with a source list has undefined 872 * behaviour; discard it. 873 */ 874 struct in6_addr dst; 875 876 dst = ip6->ip6_dst; 877 in6_clearscope(&dst); 878 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes) || 879 nsrc > 0) 880 return (EINVAL); 881 is_general_query = 1; 882 } else { 883 /* 884 * Embed scope ID of receiving interface in MLD query for 885 * lookup whilst we don't hold other locks (due to KAME 886 * locking lameness). We own this mbuf chain just now. 887 */ 888 in6_setscope(&mld->mld_addr, ifp, NULL); 889 } 890 891 IN6_MULTI_LOCK(); 892 MLD_LOCK(); 893 894 mli = MLD_IFINFO(ifp); 895 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 896 897 /* 898 * Discard the v2 query if we're in Compatibility Mode. 899 * The RFC is pretty clear that hosts need to stay in MLDv1 mode 900 * until the Old Version Querier Present timer expires. 901 */ 902 if (mli->mli_version != MLD_VERSION_2) 903 goto out_locked; 904 905 mld_set_version(mli, MLD_VERSION_2); 906 mli->mli_rv = qrv; 907 mli->mli_qi = qqi; 908 mli->mli_qri = maxdelay; 909 910 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, 911 maxdelay); 912 913 if (is_general_query) { 914 /* 915 * MLDv2 General Query. 916 * 917 * Schedule a current-state report on this ifp for 918 * all groups, possibly containing source lists. 919 * 920 * If there is a pending General Query response 921 * scheduled earlier than the selected delay, do 922 * not schedule any other reports. 923 * Otherwise, reset the interface timer. 924 */ 925 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", 926 ifp, ifp->if_xname); 927 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { 928 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); 929 V_interface_timers_running6 = 1; 930 } 931 } else { 932 /* 933 * MLDv2 Group-specific or Group-and-source-specific Query. 934 * 935 * Group-source-specific queries are throttled on 936 * a per-group basis to defeat denial-of-service attempts. 937 * Queries for groups we are not a member of on this 938 * link are simply ignored. 939 */ 940 IF_ADDR_RLOCK(ifp); 941 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 942 if (inm == NULL) { 943 IF_ADDR_RUNLOCK(ifp); 944 goto out_locked; 945 } 946 if (nsrc > 0) { 947 if (!ratecheck(&inm->in6m_lastgsrtv, 948 &V_mld_gsrdelay)) { 949 CTR1(KTR_MLD, "%s: GS query throttled.", 950 __func__); 951 IF_ADDR_RUNLOCK(ifp); 952 goto out_locked; 953 } 954 } 955 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", 956 ifp, ifp->if_xname); 957 /* 958 * If there is a pending General Query response 959 * scheduled sooner than the selected delay, no 960 * further report need be scheduled. 961 * Otherwise, prepare to respond to the 962 * group-specific or group-and-source query. 963 */ 964 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) 965 mld_v2_process_group_query(inm, mli, timer, m, off); 966 967 /* XXX Clear embedded scope ID as userland won't expect it. */ 968 in6_clearscope(&mld->mld_addr); 969 IF_ADDR_RUNLOCK(ifp); 970 } 971 972 out_locked: 973 MLD_UNLOCK(); 974 IN6_MULTI_UNLOCK(); 975 976 return (0); 977 } 978 979 /* 980 * Process a recieved MLDv2 group-specific or group-and-source-specific 981 * query. 982 * Return <0 if any error occured. Currently this is ignored. 983 */ 984 static int 985 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli, 986 int timer, struct mbuf *m0, const int off) 987 { 988 struct mldv2_query *mld; 989 int retval; 990 uint16_t nsrc; 991 992 IN6_MULTI_LOCK_ASSERT(); 993 MLD_LOCK_ASSERT(); 994 995 retval = 0; 996 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off); 997 998 switch (inm->in6m_state) { 999 case MLD_NOT_MEMBER: 1000 case MLD_SILENT_MEMBER: 1001 case MLD_SLEEPING_MEMBER: 1002 case MLD_LAZY_MEMBER: 1003 case MLD_AWAKENING_MEMBER: 1004 case MLD_IDLE_MEMBER: 1005 case MLD_LEAVING_MEMBER: 1006 return (retval); 1007 break; 1008 case MLD_REPORTING_MEMBER: 1009 case MLD_G_QUERY_PENDING_MEMBER: 1010 case MLD_SG_QUERY_PENDING_MEMBER: 1011 break; 1012 } 1013 1014 nsrc = ntohs(mld->mld_numsrc); 1015 1016 /* 1017 * Deal with group-specific queries upfront. 1018 * If any group query is already pending, purge any recorded 1019 * source-list state if it exists, and schedule a query response 1020 * for this group-specific query. 1021 */ 1022 if (nsrc == 0) { 1023 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 1024 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { 1025 in6m_clear_recorded(inm); 1026 timer = min(inm->in6m_timer, timer); 1027 } 1028 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; 1029 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1030 V_current_state_timers_running6 = 1; 1031 return (retval); 1032 } 1033 1034 /* 1035 * Deal with the case where a group-and-source-specific query has 1036 * been received but a group-specific query is already pending. 1037 */ 1038 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { 1039 timer = min(inm->in6m_timer, timer); 1040 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1041 V_current_state_timers_running6 = 1; 1042 return (retval); 1043 } 1044 1045 /* 1046 * Finally, deal with the case where a group-and-source-specific 1047 * query has been received, where a response to a previous g-s-r 1048 * query exists, or none exists. 1049 * In this case, we need to parse the source-list which the Querier 1050 * has provided us with and check if we have any source list filter 1051 * entries at T1 for these sources. If we do not, there is no need 1052 * schedule a report and the query may be dropped. 1053 * If we do, we must record them and schedule a current-state 1054 * report for those sources. 1055 */ 1056 if (inm->in6m_nsrc > 0) { 1057 struct mbuf *m; 1058 uint8_t *sp; 1059 int i, nrecorded; 1060 int soff; 1061 1062 m = m0; 1063 soff = off + sizeof(struct mldv2_query); 1064 nrecorded = 0; 1065 for (i = 0; i < nsrc; i++) { 1066 sp = mtod(m, uint8_t *) + soff; 1067 retval = in6m_record_source(inm, 1068 (const struct in6_addr *)sp); 1069 if (retval < 0) 1070 break; 1071 nrecorded += retval; 1072 soff += sizeof(struct in6_addr); 1073 if (soff >= m->m_len) { 1074 soff = soff - m->m_len; 1075 m = m->m_next; 1076 if (m == NULL) 1077 break; 1078 } 1079 } 1080 if (nrecorded > 0) { 1081 CTR1(KTR_MLD, 1082 "%s: schedule response to SG query", __func__); 1083 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; 1084 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1085 V_current_state_timers_running6 = 1; 1086 } 1087 } 1088 1089 return (retval); 1090 } 1091 1092 /* 1093 * Process a received MLDv1 host membership report. 1094 * Assumes mld points to mld_hdr in pulled up mbuf chain. 1095 * 1096 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 1097 * mld_addr. This is OK as we own the mbuf chain. 1098 */ 1099 static int 1100 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, 1101 /*const*/ struct mld_hdr *mld) 1102 { 1103 struct in6_addr src, dst; 1104 struct in6_ifaddr *ia; 1105 struct in6_multi *inm; 1106 #ifdef KTR 1107 char ip6tbuf[INET6_ADDRSTRLEN]; 1108 #endif 1109 1110 if (!mld_v1enable) { 1111 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", 1112 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1113 ifp, ifp->if_xname); 1114 return (0); 1115 } 1116 1117 if (ifp->if_flags & IFF_LOOPBACK) 1118 return (0); 1119 1120 /* 1121 * MLDv1 reports must originate from a host's link-local address, 1122 * or the unspecified address (when booting). 1123 */ 1124 src = ip6->ip6_src; 1125 in6_clearscope(&src); 1126 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { 1127 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 1128 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 1129 ifp, ifp->if_xname); 1130 return (EINVAL); 1131 } 1132 1133 /* 1134 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast 1135 * group, and must be directed to the group itself. 1136 */ 1137 dst = ip6->ip6_dst; 1138 in6_clearscope(&dst); 1139 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || 1140 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { 1141 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", 1142 ip6_sprintf(ip6tbuf, &ip6->ip6_dst), 1143 ifp, ifp->if_xname); 1144 return (EINVAL); 1145 } 1146 1147 /* 1148 * Make sure we don't hear our own membership report, as fast 1149 * leave requires knowing that we are the only member of a 1150 * group. Assume we used the link-local address if available, 1151 * otherwise look for ::. 1152 * 1153 * XXX Note that scope ID comparison is needed for the address 1154 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be 1155 * performed for the on-wire address. 1156 */ 1157 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1158 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || 1159 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) { 1160 if (ia != NULL) 1161 ifa_free(&ia->ia_ifa); 1162 return (0); 1163 } 1164 if (ia != NULL) 1165 ifa_free(&ia->ia_ifa); 1166 1167 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", 1168 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname); 1169 1170 /* 1171 * Embed scope ID of receiving interface in MLD query for lookup 1172 * whilst we don't hold other locks (due to KAME locking lameness). 1173 */ 1174 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) 1175 in6_setscope(&mld->mld_addr, ifp, NULL); 1176 1177 IN6_MULTI_LOCK(); 1178 MLD_LOCK(); 1179 IF_ADDR_RLOCK(ifp); 1180 1181 /* 1182 * MLDv1 report suppression. 1183 * If we are a member of this group, and our membership should be 1184 * reported, and our group timer is pending or about to be reset, 1185 * stop our group timer by transitioning to the 'lazy' state. 1186 */ 1187 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 1188 if (inm != NULL) { 1189 struct mld_ifinfo *mli; 1190 1191 mli = inm->in6m_mli; 1192 KASSERT(mli != NULL, 1193 ("%s: no mli for ifp %p", __func__, ifp)); 1194 1195 /* 1196 * If we are in MLDv2 host mode, do not allow the 1197 * other host's MLDv1 report to suppress our reports. 1198 */ 1199 if (mli->mli_version == MLD_VERSION_2) 1200 goto out_locked; 1201 1202 inm->in6m_timer = 0; 1203 1204 switch (inm->in6m_state) { 1205 case MLD_NOT_MEMBER: 1206 case MLD_SILENT_MEMBER: 1207 case MLD_SLEEPING_MEMBER: 1208 break; 1209 case MLD_REPORTING_MEMBER: 1210 case MLD_IDLE_MEMBER: 1211 case MLD_AWAKENING_MEMBER: 1212 CTR3(KTR_MLD, 1213 "report suppressed for %s on ifp %p(%s)", 1214 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1215 ifp, ifp->if_xname); 1216 case MLD_LAZY_MEMBER: 1217 inm->in6m_state = MLD_LAZY_MEMBER; 1218 break; 1219 case MLD_G_QUERY_PENDING_MEMBER: 1220 case MLD_SG_QUERY_PENDING_MEMBER: 1221 case MLD_LEAVING_MEMBER: 1222 break; 1223 } 1224 } 1225 1226 out_locked: 1227 IF_ADDR_RUNLOCK(ifp); 1228 MLD_UNLOCK(); 1229 IN6_MULTI_UNLOCK(); 1230 1231 /* XXX Clear embedded scope ID as userland won't expect it. */ 1232 in6_clearscope(&mld->mld_addr); 1233 1234 return (0); 1235 } 1236 1237 /* 1238 * MLD input path. 1239 * 1240 * Assume query messages which fit in a single ICMPv6 message header 1241 * have been pulled up. 1242 * Assume that userland will want to see the message, even if it 1243 * otherwise fails kernel input validation; do not free it. 1244 * Pullup may however free the mbuf chain m if it fails. 1245 * 1246 * Return IPPROTO_DONE if we freed m. Otherwise, return 0. 1247 */ 1248 int 1249 mld_input(struct mbuf *m, int off, int icmp6len) 1250 { 1251 struct ifnet *ifp; 1252 struct ip6_hdr *ip6; 1253 struct mld_hdr *mld; 1254 int mldlen; 1255 1256 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); 1257 1258 ifp = m->m_pkthdr.rcvif; 1259 1260 ip6 = mtod(m, struct ip6_hdr *); 1261 1262 /* Pullup to appropriate size. */ 1263 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); 1264 if (mld->mld_type == MLD_LISTENER_QUERY && 1265 icmp6len >= sizeof(struct mldv2_query)) { 1266 mldlen = sizeof(struct mldv2_query); 1267 } else { 1268 mldlen = sizeof(struct mld_hdr); 1269 } 1270 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen); 1271 if (mld == NULL) { 1272 ICMP6STAT_INC(icp6s_badlen); 1273 return (IPPROTO_DONE); 1274 } 1275 1276 /* 1277 * Userland needs to see all of this traffic for implementing 1278 * the endpoint discovery portion of multicast routing. 1279 */ 1280 switch (mld->mld_type) { 1281 case MLD_LISTENER_QUERY: 1282 icmp6_ifstat_inc(ifp, ifs6_in_mldquery); 1283 if (icmp6len == sizeof(struct mld_hdr)) { 1284 if (mld_v1_input_query(ifp, ip6, mld) != 0) 1285 return (0); 1286 } else if (icmp6len >= sizeof(struct mldv2_query)) { 1287 if (mld_v2_input_query(ifp, ip6, m, off, 1288 icmp6len) != 0) 1289 return (0); 1290 } 1291 break; 1292 case MLD_LISTENER_REPORT: 1293 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1294 if (mld_v1_input_report(ifp, ip6, mld) != 0) 1295 return (0); 1296 break; 1297 case MLDV2_LISTENER_REPORT: 1298 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1299 break; 1300 case MLD_LISTENER_DONE: 1301 icmp6_ifstat_inc(ifp, ifs6_in_mlddone); 1302 break; 1303 default: 1304 break; 1305 } 1306 1307 return (0); 1308 } 1309 1310 /* 1311 * Fast timeout handler (global). 1312 * VIMAGE: Timeout handlers are expected to service all vimages. 1313 */ 1314 void 1315 mld_fasttimo(void) 1316 { 1317 VNET_ITERATOR_DECL(vnet_iter); 1318 1319 VNET_LIST_RLOCK_NOSLEEP(); 1320 VNET_FOREACH(vnet_iter) { 1321 CURVNET_SET(vnet_iter); 1322 mld_fasttimo_vnet(); 1323 CURVNET_RESTORE(); 1324 } 1325 VNET_LIST_RUNLOCK_NOSLEEP(); 1326 } 1327 1328 /* 1329 * Fast timeout handler (per-vnet). 1330 * 1331 * VIMAGE: Assume caller has set up our curvnet. 1332 */ 1333 static void 1334 mld_fasttimo_vnet(void) 1335 { 1336 struct ifqueue scq; /* State-change packets */ 1337 struct ifqueue qrq; /* Query response packets */ 1338 struct ifnet *ifp; 1339 struct mld_ifinfo *mli; 1340 struct ifmultiaddr *ifma; 1341 struct in6_multi *inm, *tinm; 1342 int uri_fasthz; 1343 1344 uri_fasthz = 0; 1345 1346 /* 1347 * Quick check to see if any work needs to be done, in order to 1348 * minimize the overhead of fasttimo processing. 1349 * SMPng: XXX Unlocked reads. 1350 */ 1351 if (!V_current_state_timers_running6 && 1352 !V_interface_timers_running6 && 1353 !V_state_change_timers_running6) 1354 return; 1355 1356 IN6_MULTI_LOCK(); 1357 MLD_LOCK(); 1358 1359 /* 1360 * MLDv2 General Query response timer processing. 1361 */ 1362 if (V_interface_timers_running6) { 1363 CTR1(KTR_MLD, "%s: interface timers running", __func__); 1364 1365 V_interface_timers_running6 = 0; 1366 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1367 if (mli->mli_v2_timer == 0) { 1368 /* Do nothing. */ 1369 } else if (--mli->mli_v2_timer == 0) { 1370 mld_v2_dispatch_general_query(mli); 1371 } else { 1372 V_interface_timers_running6 = 1; 1373 } 1374 } 1375 } 1376 1377 if (!V_current_state_timers_running6 && 1378 !V_state_change_timers_running6) 1379 goto out_locked; 1380 1381 V_current_state_timers_running6 = 0; 1382 V_state_change_timers_running6 = 0; 1383 1384 CTR1(KTR_MLD, "%s: state change timers running", __func__); 1385 1386 /* 1387 * MLD host report and state-change timer processing. 1388 * Note: Processing a v2 group timer may remove a node. 1389 */ 1390 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1391 ifp = mli->mli_ifp; 1392 1393 if (mli->mli_version == MLD_VERSION_2) { 1394 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * 1395 PR_FASTHZ); 1396 1397 memset(&qrq, 0, sizeof(struct ifqueue)); 1398 IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS); 1399 1400 memset(&scq, 0, sizeof(struct ifqueue)); 1401 IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS); 1402 } 1403 1404 IF_ADDR_RLOCK(ifp); 1405 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1406 if (ifma->ifma_addr->sa_family != AF_INET6 || 1407 ifma->ifma_protospec == NULL) 1408 continue; 1409 inm = (struct in6_multi *)ifma->ifma_protospec; 1410 switch (mli->mli_version) { 1411 case MLD_VERSION_1: 1412 mld_v1_process_group_timer(mli, inm); 1413 break; 1414 case MLD_VERSION_2: 1415 mld_v2_process_group_timers(mli, &qrq, 1416 &scq, inm, uri_fasthz); 1417 break; 1418 } 1419 } 1420 IF_ADDR_RUNLOCK(ifp); 1421 1422 switch (mli->mli_version) { 1423 case MLD_VERSION_1: 1424 /* 1425 * Transmit reports for this lifecycle. This 1426 * is done while not holding IF_ADDR_LOCK 1427 * since this can call 1428 * in6ifa_ifpforlinklocal() which locks 1429 * IF_ADDR_LOCK internally as well as 1430 * ip6_output() to transmit a packet. 1431 */ 1432 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1433 in6m_nrele, tinm) { 1434 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1435 in6m_nrele); 1436 (void)mld_v1_transmit_report(inm, 1437 MLD_LISTENER_REPORT); 1438 } 1439 break; 1440 case MLD_VERSION_2: 1441 mld_dispatch_queue(&qrq, 0); 1442 mld_dispatch_queue(&scq, 0); 1443 1444 /* 1445 * Free the in_multi reference(s) for 1446 * this lifecycle. 1447 */ 1448 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1449 in6m_nrele, tinm) { 1450 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1451 in6m_nrele); 1452 in6m_release_locked(inm); 1453 } 1454 break; 1455 } 1456 } 1457 1458 out_locked: 1459 MLD_UNLOCK(); 1460 IN6_MULTI_UNLOCK(); 1461 } 1462 1463 /* 1464 * Update host report group timer. 1465 * Will update the global pending timer flags. 1466 */ 1467 static void 1468 mld_v1_process_group_timer(struct mld_ifinfo *mli, struct in6_multi *inm) 1469 { 1470 int report_timer_expired; 1471 1472 IN6_MULTI_LOCK_ASSERT(); 1473 MLD_LOCK_ASSERT(); 1474 1475 if (inm->in6m_timer == 0) { 1476 report_timer_expired = 0; 1477 } else if (--inm->in6m_timer == 0) { 1478 report_timer_expired = 1; 1479 } else { 1480 V_current_state_timers_running6 = 1; 1481 return; 1482 } 1483 1484 switch (inm->in6m_state) { 1485 case MLD_NOT_MEMBER: 1486 case MLD_SILENT_MEMBER: 1487 case MLD_IDLE_MEMBER: 1488 case MLD_LAZY_MEMBER: 1489 case MLD_SLEEPING_MEMBER: 1490 case MLD_AWAKENING_MEMBER: 1491 break; 1492 case MLD_REPORTING_MEMBER: 1493 if (report_timer_expired) { 1494 inm->in6m_state = MLD_IDLE_MEMBER; 1495 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1496 in6m_nrele); 1497 } 1498 break; 1499 case MLD_G_QUERY_PENDING_MEMBER: 1500 case MLD_SG_QUERY_PENDING_MEMBER: 1501 case MLD_LEAVING_MEMBER: 1502 break; 1503 } 1504 } 1505 1506 /* 1507 * Update a group's timers for MLDv2. 1508 * Will update the global pending timer flags. 1509 * Note: Unlocked read from mli. 1510 */ 1511 static void 1512 mld_v2_process_group_timers(struct mld_ifinfo *mli, 1513 struct ifqueue *qrq, struct ifqueue *scq, 1514 struct in6_multi *inm, const int uri_fasthz) 1515 { 1516 int query_response_timer_expired; 1517 int state_change_retransmit_timer_expired; 1518 #ifdef KTR 1519 char ip6tbuf[INET6_ADDRSTRLEN]; 1520 #endif 1521 1522 IN6_MULTI_LOCK_ASSERT(); 1523 MLD_LOCK_ASSERT(); 1524 1525 query_response_timer_expired = 0; 1526 state_change_retransmit_timer_expired = 0; 1527 1528 /* 1529 * During a transition from compatibility mode back to MLDv2, 1530 * a group record in REPORTING state may still have its group 1531 * timer active. This is a no-op in this function; it is easier 1532 * to deal with it here than to complicate the slow-timeout path. 1533 */ 1534 if (inm->in6m_timer == 0) { 1535 query_response_timer_expired = 0; 1536 } else if (--inm->in6m_timer == 0) { 1537 query_response_timer_expired = 1; 1538 } else { 1539 V_current_state_timers_running6 = 1; 1540 } 1541 1542 if (inm->in6m_sctimer == 0) { 1543 state_change_retransmit_timer_expired = 0; 1544 } else if (--inm->in6m_sctimer == 0) { 1545 state_change_retransmit_timer_expired = 1; 1546 } else { 1547 V_state_change_timers_running6 = 1; 1548 } 1549 1550 /* We are in fasttimo, so be quick about it. */ 1551 if (!state_change_retransmit_timer_expired && 1552 !query_response_timer_expired) 1553 return; 1554 1555 switch (inm->in6m_state) { 1556 case MLD_NOT_MEMBER: 1557 case MLD_SILENT_MEMBER: 1558 case MLD_SLEEPING_MEMBER: 1559 case MLD_LAZY_MEMBER: 1560 case MLD_AWAKENING_MEMBER: 1561 case MLD_IDLE_MEMBER: 1562 break; 1563 case MLD_G_QUERY_PENDING_MEMBER: 1564 case MLD_SG_QUERY_PENDING_MEMBER: 1565 /* 1566 * Respond to a previously pending Group-Specific 1567 * or Group-and-Source-Specific query by enqueueing 1568 * the appropriate Current-State report for 1569 * immediate transmission. 1570 */ 1571 if (query_response_timer_expired) { 1572 int retval; 1573 1574 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, 1575 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER), 1576 0); 1577 CTR2(KTR_MLD, "%s: enqueue record = %d", 1578 __func__, retval); 1579 inm->in6m_state = MLD_REPORTING_MEMBER; 1580 in6m_clear_recorded(inm); 1581 } 1582 /* FALLTHROUGH */ 1583 case MLD_REPORTING_MEMBER: 1584 case MLD_LEAVING_MEMBER: 1585 if (state_change_retransmit_timer_expired) { 1586 /* 1587 * State-change retransmission timer fired. 1588 * If there are any further pending retransmissions, 1589 * set the global pending state-change flag, and 1590 * reset the timer. 1591 */ 1592 if (--inm->in6m_scrv > 0) { 1593 inm->in6m_sctimer = uri_fasthz; 1594 V_state_change_timers_running6 = 1; 1595 } 1596 /* 1597 * Retransmit the previously computed state-change 1598 * report. If there are no further pending 1599 * retransmissions, the mbuf queue will be consumed. 1600 * Update T0 state to T1 as we have now sent 1601 * a state-change. 1602 */ 1603 (void)mld_v2_merge_state_changes(inm, scq); 1604 1605 in6m_commit(inm); 1606 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 1607 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1608 inm->in6m_ifp->if_xname); 1609 1610 /* 1611 * If we are leaving the group for good, make sure 1612 * we release MLD's reference to it. 1613 * This release must be deferred using a SLIST, 1614 * as we are called from a loop which traverses 1615 * the in_ifmultiaddr TAILQ. 1616 */ 1617 if (inm->in6m_state == MLD_LEAVING_MEMBER && 1618 inm->in6m_scrv == 0) { 1619 inm->in6m_state = MLD_NOT_MEMBER; 1620 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 1621 inm, in6m_nrele); 1622 } 1623 } 1624 break; 1625 } 1626 } 1627 1628 /* 1629 * Switch to a different version on the given interface, 1630 * as per Section 9.12. 1631 */ 1632 static void 1633 mld_set_version(struct mld_ifinfo *mli, const int version) 1634 { 1635 int old_version_timer; 1636 1637 MLD_LOCK_ASSERT(); 1638 1639 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, 1640 version, mli->mli_ifp, mli->mli_ifp->if_xname); 1641 1642 if (version == MLD_VERSION_1) { 1643 /* 1644 * Compute the "Older Version Querier Present" timer as per 1645 * Section 9.12. 1646 */ 1647 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; 1648 old_version_timer *= PR_SLOWHZ; 1649 mli->mli_v1_timer = old_version_timer; 1650 } 1651 1652 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { 1653 mli->mli_version = MLD_VERSION_1; 1654 mld_v2_cancel_link_timers(mli); 1655 } 1656 } 1657 1658 /* 1659 * Cancel pending MLDv2 timers for the given link and all groups 1660 * joined on it; state-change, general-query, and group-query timers. 1661 */ 1662 static void 1663 mld_v2_cancel_link_timers(struct mld_ifinfo *mli) 1664 { 1665 struct ifmultiaddr *ifma; 1666 struct ifnet *ifp; 1667 struct in6_multi *inm, *tinm; 1668 1669 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, 1670 mli->mli_ifp, mli->mli_ifp->if_xname); 1671 1672 IN6_MULTI_LOCK_ASSERT(); 1673 MLD_LOCK_ASSERT(); 1674 1675 /* 1676 * Fast-track this potentially expensive operation 1677 * by checking all the global 'timer pending' flags. 1678 */ 1679 if (!V_interface_timers_running6 && 1680 !V_state_change_timers_running6 && 1681 !V_current_state_timers_running6) 1682 return; 1683 1684 mli->mli_v2_timer = 0; 1685 1686 ifp = mli->mli_ifp; 1687 1688 IF_ADDR_RLOCK(ifp); 1689 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1690 if (ifma->ifma_addr->sa_family != AF_INET6) 1691 continue; 1692 inm = (struct in6_multi *)ifma->ifma_protospec; 1693 switch (inm->in6m_state) { 1694 case MLD_NOT_MEMBER: 1695 case MLD_SILENT_MEMBER: 1696 case MLD_IDLE_MEMBER: 1697 case MLD_LAZY_MEMBER: 1698 case MLD_SLEEPING_MEMBER: 1699 case MLD_AWAKENING_MEMBER: 1700 break; 1701 case MLD_LEAVING_MEMBER: 1702 /* 1703 * If we are leaving the group and switching 1704 * version, we need to release the final 1705 * reference held for issuing the INCLUDE {}. 1706 */ 1707 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1708 in6m_nrele); 1709 /* FALLTHROUGH */ 1710 case MLD_G_QUERY_PENDING_MEMBER: 1711 case MLD_SG_QUERY_PENDING_MEMBER: 1712 in6m_clear_recorded(inm); 1713 /* FALLTHROUGH */ 1714 case MLD_REPORTING_MEMBER: 1715 inm->in6m_sctimer = 0; 1716 inm->in6m_timer = 0; 1717 inm->in6m_state = MLD_REPORTING_MEMBER; 1718 /* 1719 * Free any pending MLDv2 state-change records. 1720 */ 1721 _IF_DRAIN(&inm->in6m_scq); 1722 break; 1723 } 1724 } 1725 IF_ADDR_RUNLOCK(ifp); 1726 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, tinm) { 1727 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 1728 in6m_release_locked(inm); 1729 } 1730 } 1731 1732 /* 1733 * Global slowtimo handler. 1734 * VIMAGE: Timeout handlers are expected to service all vimages. 1735 */ 1736 void 1737 mld_slowtimo(void) 1738 { 1739 VNET_ITERATOR_DECL(vnet_iter); 1740 1741 VNET_LIST_RLOCK_NOSLEEP(); 1742 VNET_FOREACH(vnet_iter) { 1743 CURVNET_SET(vnet_iter); 1744 mld_slowtimo_vnet(); 1745 CURVNET_RESTORE(); 1746 } 1747 VNET_LIST_RUNLOCK_NOSLEEP(); 1748 } 1749 1750 /* 1751 * Per-vnet slowtimo handler. 1752 */ 1753 static void 1754 mld_slowtimo_vnet(void) 1755 { 1756 struct mld_ifinfo *mli; 1757 1758 MLD_LOCK(); 1759 1760 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1761 mld_v1_process_querier_timers(mli); 1762 } 1763 1764 MLD_UNLOCK(); 1765 } 1766 1767 /* 1768 * Update the Older Version Querier Present timers for a link. 1769 * See Section 9.12 of RFC 3810. 1770 */ 1771 static void 1772 mld_v1_process_querier_timers(struct mld_ifinfo *mli) 1773 { 1774 1775 MLD_LOCK_ASSERT(); 1776 1777 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) { 1778 /* 1779 * MLDv1 Querier Present timer expired; revert to MLDv2. 1780 */ 1781 CTR5(KTR_MLD, 1782 "%s: transition from v%d -> v%d on %p(%s)", 1783 __func__, mli->mli_version, MLD_VERSION_2, 1784 mli->mli_ifp, mli->mli_ifp->if_xname); 1785 mli->mli_version = MLD_VERSION_2; 1786 } 1787 } 1788 1789 /* 1790 * Transmit an MLDv1 report immediately. 1791 */ 1792 static int 1793 mld_v1_transmit_report(struct in6_multi *in6m, const int type) 1794 { 1795 struct ifnet *ifp; 1796 struct in6_ifaddr *ia; 1797 struct ip6_hdr *ip6; 1798 struct mbuf *mh, *md; 1799 struct mld_hdr *mld; 1800 1801 IN6_MULTI_LOCK_ASSERT(); 1802 MLD_LOCK_ASSERT(); 1803 1804 ifp = in6m->in6m_ifp; 1805 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1806 /* ia may be NULL if link-local address is tentative. */ 1807 1808 MGETHDR(mh, M_DONTWAIT, MT_HEADER); 1809 if (mh == NULL) { 1810 if (ia != NULL) 1811 ifa_free(&ia->ia_ifa); 1812 return (ENOMEM); 1813 } 1814 MGET(md, M_DONTWAIT, MT_DATA); 1815 if (md == NULL) { 1816 m_free(mh); 1817 if (ia != NULL) 1818 ifa_free(&ia->ia_ifa); 1819 return (ENOMEM); 1820 } 1821 mh->m_next = md; 1822 1823 /* 1824 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so 1825 * that ether_output() does not need to allocate another mbuf 1826 * for the header in the most common case. 1827 */ 1828 MH_ALIGN(mh, sizeof(struct ip6_hdr)); 1829 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); 1830 mh->m_len = sizeof(struct ip6_hdr); 1831 1832 ip6 = mtod(mh, struct ip6_hdr *); 1833 ip6->ip6_flow = 0; 1834 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 1835 ip6->ip6_vfc |= IPV6_VERSION; 1836 ip6->ip6_nxt = IPPROTO_ICMPV6; 1837 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 1838 ip6->ip6_dst = in6m->in6m_addr; 1839 1840 md->m_len = sizeof(struct mld_hdr); 1841 mld = mtod(md, struct mld_hdr *); 1842 mld->mld_type = type; 1843 mld->mld_code = 0; 1844 mld->mld_cksum = 0; 1845 mld->mld_maxdelay = 0; 1846 mld->mld_reserved = 0; 1847 mld->mld_addr = in6m->in6m_addr; 1848 in6_clearscope(&mld->mld_addr); 1849 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 1850 sizeof(struct ip6_hdr), sizeof(struct mld_hdr)); 1851 1852 mld_save_context(mh, ifp); 1853 mh->m_flags |= M_MLDV1; 1854 1855 mld_dispatch_packet(mh); 1856 1857 if (ia != NULL) 1858 ifa_free(&ia->ia_ifa); 1859 return (0); 1860 } 1861 1862 /* 1863 * Process a state change from the upper layer for the given IPv6 group. 1864 * 1865 * Each socket holds a reference on the in_multi in its own ip_moptions. 1866 * The socket layer will have made the necessary updates to.the group 1867 * state, it is now up to MLD to issue a state change report if there 1868 * has been any change between T0 (when the last state-change was issued) 1869 * and T1 (now). 1870 * 1871 * We use the MLDv2 state machine at group level. The MLd module 1872 * however makes the decision as to which MLD protocol version to speak. 1873 * A state change *from* INCLUDE {} always means an initial join. 1874 * A state change *to* INCLUDE {} always means a final leave. 1875 * 1876 * If delay is non-zero, and the state change is an initial multicast 1877 * join, the state change report will be delayed by 'delay' ticks 1878 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise 1879 * the initial MLDv2 state change report will be delayed by whichever 1880 * is sooner, a pending state-change timer or delay itself. 1881 * 1882 * VIMAGE: curvnet should have been set by caller, as this routine 1883 * is called from the socket option handlers. 1884 */ 1885 int 1886 mld_change_state(struct in6_multi *inm, const int delay) 1887 { 1888 struct mld_ifinfo *mli; 1889 struct ifnet *ifp; 1890 int error; 1891 1892 IN6_MULTI_LOCK_ASSERT(); 1893 1894 error = 0; 1895 1896 /* 1897 * Try to detect if the upper layer just asked us to change state 1898 * for an interface which has now gone away. 1899 */ 1900 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 1901 ifp = inm->in6m_ifma->ifma_ifp; 1902 if (ifp != NULL) { 1903 /* 1904 * Sanity check that netinet6's notion of ifp is the 1905 * same as net's. 1906 */ 1907 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 1908 } 1909 1910 MLD_LOCK(); 1911 1912 mli = MLD_IFINFO(ifp); 1913 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 1914 1915 /* 1916 * If we detect a state transition to or from MCAST_UNDEFINED 1917 * for this group, then we are starting or finishing an MLD 1918 * life cycle for this group. 1919 */ 1920 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { 1921 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, 1922 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); 1923 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { 1924 CTR1(KTR_MLD, "%s: initial join", __func__); 1925 error = mld_initial_join(inm, mli, delay); 1926 goto out_locked; 1927 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { 1928 CTR1(KTR_MLD, "%s: final leave", __func__); 1929 mld_final_leave(inm, mli); 1930 goto out_locked; 1931 } 1932 } else { 1933 CTR1(KTR_MLD, "%s: filter set change", __func__); 1934 } 1935 1936 error = mld_handle_state_change(inm, mli); 1937 1938 out_locked: 1939 MLD_UNLOCK(); 1940 return (error); 1941 } 1942 1943 /* 1944 * Perform the initial join for an MLD group. 1945 * 1946 * When joining a group: 1947 * If the group should have its MLD traffic suppressed, do nothing. 1948 * MLDv1 starts sending MLDv1 host membership reports. 1949 * MLDv2 will schedule an MLDv2 state-change report containing the 1950 * initial state of the membership. 1951 * 1952 * If the delay argument is non-zero, then we must delay sending the 1953 * initial state change for delay ticks (in units of PR_FASTHZ). 1954 */ 1955 static int 1956 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli, 1957 const int delay) 1958 { 1959 struct ifnet *ifp; 1960 struct ifqueue *ifq; 1961 int error, retval, syncstates; 1962 int odelay; 1963 #ifdef KTR 1964 char ip6tbuf[INET6_ADDRSTRLEN]; 1965 #endif 1966 1967 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", 1968 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1969 inm->in6m_ifp, inm->in6m_ifp->if_xname); 1970 1971 error = 0; 1972 syncstates = 1; 1973 1974 ifp = inm->in6m_ifp; 1975 1976 IN6_MULTI_LOCK_ASSERT(); 1977 MLD_LOCK_ASSERT(); 1978 1979 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); 1980 1981 /* 1982 * Groups joined on loopback or marked as 'not reported', 1983 * enter the MLD_SILENT_MEMBER state and 1984 * are never reported in any protocol exchanges. 1985 * All other groups enter the appropriate state machine 1986 * for the version in use on this link. 1987 * A link marked as MLIF_SILENT causes MLD to be completely 1988 * disabled for the link. 1989 */ 1990 if ((ifp->if_flags & IFF_LOOPBACK) || 1991 (mli->mli_flags & MLIF_SILENT) || 1992 !mld_is_addr_reported(&inm->in6m_addr)) { 1993 CTR1(KTR_MLD, 1994 "%s: not kicking state machine for silent group", __func__); 1995 inm->in6m_state = MLD_SILENT_MEMBER; 1996 inm->in6m_timer = 0; 1997 } else { 1998 /* 1999 * Deal with overlapping in_multi lifecycle. 2000 * If this group was LEAVING, then make sure 2001 * we drop the reference we picked up to keep the 2002 * group around for the final INCLUDE {} enqueue. 2003 */ 2004 if (mli->mli_version == MLD_VERSION_2 && 2005 inm->in6m_state == MLD_LEAVING_MEMBER) 2006 in6m_release_locked(inm); 2007 2008 inm->in6m_state = MLD_REPORTING_MEMBER; 2009 2010 switch (mli->mli_version) { 2011 case MLD_VERSION_1: 2012 /* 2013 * If a delay was provided, only use it if 2014 * it is greater than the delay normally 2015 * used for an MLDv1 state change report, 2016 * and delay sending the initial MLDv1 report 2017 * by not transitioning to the IDLE state. 2018 */ 2019 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); 2020 if (delay) { 2021 inm->in6m_timer = max(delay, odelay); 2022 V_current_state_timers_running6 = 1; 2023 } else { 2024 inm->in6m_state = MLD_IDLE_MEMBER; 2025 error = mld_v1_transmit_report(inm, 2026 MLD_LISTENER_REPORT); 2027 if (error == 0) { 2028 inm->in6m_timer = odelay; 2029 V_current_state_timers_running6 = 1; 2030 } 2031 } 2032 break; 2033 2034 case MLD_VERSION_2: 2035 /* 2036 * Defer update of T0 to T1, until the first copy 2037 * of the state change has been transmitted. 2038 */ 2039 syncstates = 0; 2040 2041 /* 2042 * Immediately enqueue a State-Change Report for 2043 * this interface, freeing any previous reports. 2044 * Don't kick the timers if there is nothing to do, 2045 * or if an error occurred. 2046 */ 2047 ifq = &inm->in6m_scq; 2048 _IF_DRAIN(ifq); 2049 retval = mld_v2_enqueue_group_record(ifq, inm, 1, 2050 0, 0, (mli->mli_flags & MLIF_USEALLOW)); 2051 CTR2(KTR_MLD, "%s: enqueue record = %d", 2052 __func__, retval); 2053 if (retval <= 0) { 2054 error = retval * -1; 2055 break; 2056 } 2057 2058 /* 2059 * Schedule transmission of pending state-change 2060 * report up to RV times for this link. The timer 2061 * will fire at the next mld_fasttimo (~200ms), 2062 * giving us an opportunity to merge the reports. 2063 * 2064 * If a delay was provided to this function, only 2065 * use this delay if sooner than the existing one. 2066 */ 2067 KASSERT(mli->mli_rv > 1, 2068 ("%s: invalid robustness %d", __func__, 2069 mli->mli_rv)); 2070 inm->in6m_scrv = mli->mli_rv; 2071 if (delay) { 2072 if (inm->in6m_sctimer > 1) { 2073 inm->in6m_sctimer = 2074 min(inm->in6m_sctimer, delay); 2075 } else 2076 inm->in6m_sctimer = delay; 2077 } else 2078 inm->in6m_sctimer = 1; 2079 V_state_change_timers_running6 = 1; 2080 2081 error = 0; 2082 break; 2083 } 2084 } 2085 2086 /* 2087 * Only update the T0 state if state change is atomic, 2088 * i.e. we don't need to wait for a timer to fire before we 2089 * can consider the state change to have been communicated. 2090 */ 2091 if (syncstates) { 2092 in6m_commit(inm); 2093 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2094 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2095 inm->in6m_ifp->if_xname); 2096 } 2097 2098 return (error); 2099 } 2100 2101 /* 2102 * Issue an intermediate state change during the life-cycle. 2103 */ 2104 static int 2105 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli) 2106 { 2107 struct ifnet *ifp; 2108 int retval; 2109 #ifdef KTR 2110 char ip6tbuf[INET6_ADDRSTRLEN]; 2111 #endif 2112 2113 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)", 2114 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2115 inm->in6m_ifp, inm->in6m_ifp->if_xname); 2116 2117 ifp = inm->in6m_ifp; 2118 2119 IN6_MULTI_LOCK_ASSERT(); 2120 MLD_LOCK_ASSERT(); 2121 2122 KASSERT(mli && mli->mli_ifp == ifp, 2123 ("%s: inconsistent ifp", __func__)); 2124 2125 if ((ifp->if_flags & IFF_LOOPBACK) || 2126 (mli->mli_flags & MLIF_SILENT) || 2127 !mld_is_addr_reported(&inm->in6m_addr) || 2128 (mli->mli_version != MLD_VERSION_2)) { 2129 if (!mld_is_addr_reported(&inm->in6m_addr)) { 2130 CTR1(KTR_MLD, 2131 "%s: not kicking state machine for silent group", __func__); 2132 } 2133 CTR1(KTR_MLD, "%s: nothing to do", __func__); 2134 in6m_commit(inm); 2135 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2136 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2137 inm->in6m_ifp->if_xname); 2138 return (0); 2139 } 2140 2141 _IF_DRAIN(&inm->in6m_scq); 2142 2143 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0, 2144 (mli->mli_flags & MLIF_USEALLOW)); 2145 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); 2146 if (retval <= 0) 2147 return (-retval); 2148 2149 /* 2150 * If record(s) were enqueued, start the state-change 2151 * report timer for this group. 2152 */ 2153 inm->in6m_scrv = mli->mli_rv; 2154 inm->in6m_sctimer = 1; 2155 V_state_change_timers_running6 = 1; 2156 2157 return (0); 2158 } 2159 2160 /* 2161 * Perform the final leave for a multicast address. 2162 * 2163 * When leaving a group: 2164 * MLDv1 sends a DONE message, if and only if we are the reporter. 2165 * MLDv2 enqueues a state-change report containing a transition 2166 * to INCLUDE {} for immediate transmission. 2167 */ 2168 static void 2169 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli) 2170 { 2171 int syncstates; 2172 #ifdef KTR 2173 char ip6tbuf[INET6_ADDRSTRLEN]; 2174 #endif 2175 2176 syncstates = 1; 2177 2178 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)", 2179 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2180 inm->in6m_ifp, inm->in6m_ifp->if_xname); 2181 2182 IN6_MULTI_LOCK_ASSERT(); 2183 MLD_LOCK_ASSERT(); 2184 2185 switch (inm->in6m_state) { 2186 case MLD_NOT_MEMBER: 2187 case MLD_SILENT_MEMBER: 2188 case MLD_LEAVING_MEMBER: 2189 /* Already leaving or left; do nothing. */ 2190 CTR1(KTR_MLD, 2191 "%s: not kicking state machine for silent group", __func__); 2192 break; 2193 case MLD_REPORTING_MEMBER: 2194 case MLD_IDLE_MEMBER: 2195 case MLD_G_QUERY_PENDING_MEMBER: 2196 case MLD_SG_QUERY_PENDING_MEMBER: 2197 if (mli->mli_version == MLD_VERSION_1) { 2198 #ifdef INVARIANTS 2199 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 2200 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) 2201 panic("%s: MLDv2 state reached, not MLDv2 mode", 2202 __func__); 2203 #endif 2204 mld_v1_transmit_report(inm, MLD_LISTENER_DONE); 2205 inm->in6m_state = MLD_NOT_MEMBER; 2206 } else if (mli->mli_version == MLD_VERSION_2) { 2207 /* 2208 * Stop group timer and all pending reports. 2209 * Immediately enqueue a state-change report 2210 * TO_IN {} to be sent on the next fast timeout, 2211 * giving us an opportunity to merge reports. 2212 */ 2213 _IF_DRAIN(&inm->in6m_scq); 2214 inm->in6m_timer = 0; 2215 inm->in6m_scrv = mli->mli_rv; 2216 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " 2217 "pending retransmissions.", __func__, 2218 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2219 inm->in6m_ifp->if_xname, inm->in6m_scrv); 2220 if (inm->in6m_scrv == 0) { 2221 inm->in6m_state = MLD_NOT_MEMBER; 2222 inm->in6m_sctimer = 0; 2223 } else { 2224 int retval; 2225 2226 in6m_acquire_locked(inm); 2227 2228 retval = mld_v2_enqueue_group_record( 2229 &inm->in6m_scq, inm, 1, 0, 0, 2230 (mli->mli_flags & MLIF_USEALLOW)); 2231 KASSERT(retval != 0, 2232 ("%s: enqueue record = %d", __func__, 2233 retval)); 2234 2235 inm->in6m_state = MLD_LEAVING_MEMBER; 2236 inm->in6m_sctimer = 1; 2237 V_state_change_timers_running6 = 1; 2238 syncstates = 0; 2239 } 2240 break; 2241 } 2242 break; 2243 case MLD_LAZY_MEMBER: 2244 case MLD_SLEEPING_MEMBER: 2245 case MLD_AWAKENING_MEMBER: 2246 /* Our reports are suppressed; do nothing. */ 2247 break; 2248 } 2249 2250 if (syncstates) { 2251 in6m_commit(inm); 2252 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2253 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2254 inm->in6m_ifp->if_xname); 2255 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 2256 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", 2257 __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname); 2258 } 2259 } 2260 2261 /* 2262 * Enqueue an MLDv2 group record to the given output queue. 2263 * 2264 * If is_state_change is zero, a current-state record is appended. 2265 * If is_state_change is non-zero, a state-change report is appended. 2266 * 2267 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2268 * If is_group_query is zero, and if there is a packet with free space 2269 * at the tail of the queue, it will be appended to providing there 2270 * is enough free space. 2271 * Otherwise a new mbuf packet chain is allocated. 2272 * 2273 * If is_source_query is non-zero, each source is checked to see if 2274 * it was recorded for a Group-Source query, and will be omitted if 2275 * it is not both in-mode and recorded. 2276 * 2277 * If use_block_allow is non-zero, state change reports for initial join 2278 * and final leave, on an inclusive mode group with a source list, will be 2279 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively. 2280 * 2281 * The function will attempt to allocate leading space in the packet 2282 * for the IPv6+ICMP headers to be prepended without fragmenting the chain. 2283 * 2284 * If successful the size of all data appended to the queue is returned, 2285 * otherwise an error code less than zero is returned, or zero if 2286 * no record(s) were appended. 2287 */ 2288 static int 2289 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm, 2290 const int is_state_change, const int is_group_query, 2291 const int is_source_query, const int use_block_allow) 2292 { 2293 struct mldv2_record mr; 2294 struct mldv2_record *pmr; 2295 struct ifnet *ifp; 2296 struct ip6_msource *ims, *nims; 2297 struct mbuf *m0, *m, *md; 2298 int error, is_filter_list_change; 2299 int minrec0len, m0srcs, msrcs, nbytes, off; 2300 int record_has_sources; 2301 int now; 2302 int type; 2303 uint8_t mode; 2304 #ifdef KTR 2305 char ip6tbuf[INET6_ADDRSTRLEN]; 2306 #endif 2307 2308 IN6_MULTI_LOCK_ASSERT(); 2309 2310 error = 0; 2311 ifp = inm->in6m_ifp; 2312 is_filter_list_change = 0; 2313 m = NULL; 2314 m0 = NULL; 2315 m0srcs = 0; 2316 msrcs = 0; 2317 nbytes = 0; 2318 nims = NULL; 2319 record_has_sources = 1; 2320 pmr = NULL; 2321 type = MLD_DO_NOTHING; 2322 mode = inm->in6m_st[1].iss_fmode; 2323 2324 /* 2325 * If we did not transition out of ASM mode during t0->t1, 2326 * and there are no source nodes to process, we can skip 2327 * the generation of source records. 2328 */ 2329 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && 2330 inm->in6m_nsrc == 0) 2331 record_has_sources = 0; 2332 2333 if (is_state_change) { 2334 /* 2335 * Queue a state change record. 2336 * If the mode did not change, and there are non-ASM 2337 * listeners or source filters present, 2338 * we potentially need to issue two records for the group. 2339 * If there are ASM listeners, and there was no filter 2340 * mode transition of any kind, do nothing. 2341 * 2342 * If we are transitioning to MCAST_UNDEFINED, we need 2343 * not send any sources. A transition to/from this state is 2344 * considered inclusive with some special treatment. 2345 * 2346 * If we are rewriting initial joins/leaves to use 2347 * ALLOW/BLOCK, and the group's membership is inclusive, 2348 * we need to send sources in all cases. 2349 */ 2350 if (mode != inm->in6m_st[0].iss_fmode) { 2351 if (mode == MCAST_EXCLUDE) { 2352 CTR1(KTR_MLD, "%s: change to EXCLUDE", 2353 __func__); 2354 type = MLD_CHANGE_TO_EXCLUDE_MODE; 2355 } else { 2356 CTR1(KTR_MLD, "%s: change to INCLUDE", 2357 __func__); 2358 if (use_block_allow) { 2359 /* 2360 * XXX 2361 * Here we're interested in state 2362 * edges either direction between 2363 * MCAST_UNDEFINED and MCAST_INCLUDE. 2364 * Perhaps we should just check 2365 * the group state, rather than 2366 * the filter mode. 2367 */ 2368 if (mode == MCAST_UNDEFINED) { 2369 type = MLD_BLOCK_OLD_SOURCES; 2370 } else { 2371 type = MLD_ALLOW_NEW_SOURCES; 2372 } 2373 } else { 2374 type = MLD_CHANGE_TO_INCLUDE_MODE; 2375 if (mode == MCAST_UNDEFINED) 2376 record_has_sources = 0; 2377 } 2378 } 2379 } else { 2380 if (record_has_sources) { 2381 is_filter_list_change = 1; 2382 } else { 2383 type = MLD_DO_NOTHING; 2384 } 2385 } 2386 } else { 2387 /* 2388 * Queue a current state record. 2389 */ 2390 if (mode == MCAST_EXCLUDE) { 2391 type = MLD_MODE_IS_EXCLUDE; 2392 } else if (mode == MCAST_INCLUDE) { 2393 type = MLD_MODE_IS_INCLUDE; 2394 KASSERT(inm->in6m_st[1].iss_asm == 0, 2395 ("%s: inm %p is INCLUDE but ASM count is %d", 2396 __func__, inm, inm->in6m_st[1].iss_asm)); 2397 } 2398 } 2399 2400 /* 2401 * Generate the filter list changes using a separate function. 2402 */ 2403 if (is_filter_list_change) 2404 return (mld_v2_enqueue_filter_change(ifq, inm)); 2405 2406 if (type == MLD_DO_NOTHING) { 2407 CTR3(KTR_MLD, "%s: nothing to do for %s/%s", 2408 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2409 inm->in6m_ifp->if_xname); 2410 return (0); 2411 } 2412 2413 /* 2414 * If any sources are present, we must be able to fit at least 2415 * one in the trailing space of the tail packet's mbuf, 2416 * ideally more. 2417 */ 2418 minrec0len = sizeof(struct mldv2_record); 2419 if (record_has_sources) 2420 minrec0len += sizeof(struct in6_addr); 2421 2422 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, 2423 mld_rec_type_to_str(type), 2424 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2425 inm->in6m_ifp->if_xname); 2426 2427 /* 2428 * Check if we have a packet in the tail of the queue for this 2429 * group into which the first group record for this group will fit. 2430 * Otherwise allocate a new packet. 2431 * Always allocate leading space for IP6+RA+ICMPV6+REPORT. 2432 * Note: Group records for G/GSR query responses MUST be sent 2433 * in their own packet. 2434 */ 2435 m0 = ifq->ifq_tail; 2436 if (!is_group_query && 2437 m0 != NULL && 2438 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && 2439 (m0->m_pkthdr.len + minrec0len) < 2440 (ifp->if_mtu - MLD_MTUSPACE)) { 2441 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2442 sizeof(struct mldv2_record)) / 2443 sizeof(struct in6_addr); 2444 m = m0; 2445 CTR1(KTR_MLD, "%s: use existing packet", __func__); 2446 } else { 2447 if (_IF_QFULL(ifq)) { 2448 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2449 return (-ENOMEM); 2450 } 2451 m = NULL; 2452 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2453 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2454 if (!is_state_change && !is_group_query) 2455 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2456 if (m == NULL) 2457 m = m_gethdr(M_DONTWAIT, MT_DATA); 2458 if (m == NULL) 2459 return (-ENOMEM); 2460 2461 mld_save_context(m, ifp); 2462 2463 CTR1(KTR_MLD, "%s: allocated first packet", __func__); 2464 } 2465 2466 /* 2467 * Append group record. 2468 * If we have sources, we don't know how many yet. 2469 */ 2470 mr.mr_type = type; 2471 mr.mr_datalen = 0; 2472 mr.mr_numsrc = 0; 2473 mr.mr_addr = inm->in6m_addr; 2474 in6_clearscope(&mr.mr_addr); 2475 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2476 if (m != m0) 2477 m_freem(m); 2478 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2479 return (-ENOMEM); 2480 } 2481 nbytes += sizeof(struct mldv2_record); 2482 2483 /* 2484 * Append as many sources as will fit in the first packet. 2485 * If we are appending to a new packet, the chain allocation 2486 * may potentially use clusters; use m_getptr() in this case. 2487 * If we are appending to an existing packet, we need to obtain 2488 * a pointer to the group record after m_append(), in case a new 2489 * mbuf was allocated. 2490 * 2491 * Only append sources which are in-mode at t1. If we are 2492 * transitioning to MCAST_UNDEFINED state on the group, and 2493 * use_block_allow is zero, do not include source entries. 2494 * Otherwise, we need to include this source in the report. 2495 * 2496 * Only report recorded sources in our filter set when responding 2497 * to a group-source query. 2498 */ 2499 if (record_has_sources) { 2500 if (m == m0) { 2501 md = m_last(m); 2502 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2503 md->m_len - nbytes); 2504 } else { 2505 md = m_getptr(m, 0, &off); 2506 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2507 off); 2508 } 2509 msrcs = 0; 2510 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, 2511 nims) { 2512 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2513 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2514 now = im6s_get_mode(inm, ims, 1); 2515 CTR2(KTR_MLD, "%s: node is %d", __func__, now); 2516 if ((now != mode) || 2517 (now == mode && 2518 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2519 CTR1(KTR_MLD, "%s: skip node", __func__); 2520 continue; 2521 } 2522 if (is_source_query && ims->im6s_stp == 0) { 2523 CTR1(KTR_MLD, "%s: skip unrecorded node", 2524 __func__); 2525 continue; 2526 } 2527 CTR1(KTR_MLD, "%s: append node", __func__); 2528 if (!m_append(m, sizeof(struct in6_addr), 2529 (void *)&ims->im6s_addr)) { 2530 if (m != m0) 2531 m_freem(m); 2532 CTR1(KTR_MLD, "%s: m_append() failed.", 2533 __func__); 2534 return (-ENOMEM); 2535 } 2536 nbytes += sizeof(struct in6_addr); 2537 ++msrcs; 2538 if (msrcs == m0srcs) 2539 break; 2540 } 2541 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, 2542 msrcs); 2543 pmr->mr_numsrc = htons(msrcs); 2544 nbytes += (msrcs * sizeof(struct in6_addr)); 2545 } 2546 2547 if (is_source_query && msrcs == 0) { 2548 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); 2549 if (m != m0) 2550 m_freem(m); 2551 return (0); 2552 } 2553 2554 /* 2555 * We are good to go with first packet. 2556 */ 2557 if (m != m0) { 2558 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); 2559 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2560 _IF_ENQUEUE(ifq, m); 2561 } else 2562 m->m_pkthdr.PH_vt.vt_nrecs++; 2563 2564 /* 2565 * No further work needed if no source list in packet(s). 2566 */ 2567 if (!record_has_sources) 2568 return (nbytes); 2569 2570 /* 2571 * Whilst sources remain to be announced, we need to allocate 2572 * a new packet and fill out as many sources as will fit. 2573 * Always try for a cluster first. 2574 */ 2575 while (nims != NULL) { 2576 if (_IF_QFULL(ifq)) { 2577 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2578 return (-ENOMEM); 2579 } 2580 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2581 if (m == NULL) 2582 m = m_gethdr(M_DONTWAIT, MT_DATA); 2583 if (m == NULL) 2584 return (-ENOMEM); 2585 mld_save_context(m, ifp); 2586 md = m_getptr(m, 0, &off); 2587 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); 2588 CTR1(KTR_MLD, "%s: allocated next packet", __func__); 2589 2590 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2591 if (m != m0) 2592 m_freem(m); 2593 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2594 return (-ENOMEM); 2595 } 2596 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2597 nbytes += sizeof(struct mldv2_record); 2598 2599 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2600 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2601 2602 msrcs = 0; 2603 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2604 CTR2(KTR_MLD, "%s: visit node %s", 2605 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2606 now = im6s_get_mode(inm, ims, 1); 2607 if ((now != mode) || 2608 (now == mode && 2609 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2610 CTR1(KTR_MLD, "%s: skip node", __func__); 2611 continue; 2612 } 2613 if (is_source_query && ims->im6s_stp == 0) { 2614 CTR1(KTR_MLD, "%s: skip unrecorded node", 2615 __func__); 2616 continue; 2617 } 2618 CTR1(KTR_MLD, "%s: append node", __func__); 2619 if (!m_append(m, sizeof(struct in6_addr), 2620 (void *)&ims->im6s_addr)) { 2621 if (m != m0) 2622 m_freem(m); 2623 CTR1(KTR_MLD, "%s: m_append() failed.", 2624 __func__); 2625 return (-ENOMEM); 2626 } 2627 ++msrcs; 2628 if (msrcs == m0srcs) 2629 break; 2630 } 2631 pmr->mr_numsrc = htons(msrcs); 2632 nbytes += (msrcs * sizeof(struct in6_addr)); 2633 2634 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); 2635 _IF_ENQUEUE(ifq, m); 2636 } 2637 2638 return (nbytes); 2639 } 2640 2641 /* 2642 * Type used to mark record pass completion. 2643 * We exploit the fact we can cast to this easily from the 2644 * current filter modes on each ip_msource node. 2645 */ 2646 typedef enum { 2647 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2648 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2649 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2650 REC_FULL = REC_ALLOW | REC_BLOCK 2651 } rectype_t; 2652 2653 /* 2654 * Enqueue an MLDv2 filter list change to the given output queue. 2655 * 2656 * Source list filter state is held in an RB-tree. When the filter list 2657 * for a group is changed without changing its mode, we need to compute 2658 * the deltas between T0 and T1 for each source in the filter set, 2659 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2660 * 2661 * As we may potentially queue two record types, and the entire R-B tree 2662 * needs to be walked at once, we break this out into its own function 2663 * so we can generate a tightly packed queue of packets. 2664 * 2665 * XXX This could be written to only use one tree walk, although that makes 2666 * serializing into the mbuf chains a bit harder. For now we do two walks 2667 * which makes things easier on us, and it may or may not be harder on 2668 * the L2 cache. 2669 * 2670 * If successful the size of all data appended to the queue is returned, 2671 * otherwise an error code less than zero is returned, or zero if 2672 * no record(s) were appended. 2673 */ 2674 static int 2675 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm) 2676 { 2677 static const int MINRECLEN = 2678 sizeof(struct mldv2_record) + sizeof(struct in6_addr); 2679 struct ifnet *ifp; 2680 struct mldv2_record mr; 2681 struct mldv2_record *pmr; 2682 struct ip6_msource *ims, *nims; 2683 struct mbuf *m, *m0, *md; 2684 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 2685 int nallow, nblock; 2686 uint8_t mode, now, then; 2687 rectype_t crt, drt, nrt; 2688 #ifdef KTR 2689 char ip6tbuf[INET6_ADDRSTRLEN]; 2690 #endif 2691 2692 IN6_MULTI_LOCK_ASSERT(); 2693 2694 if (inm->in6m_nsrc == 0 || 2695 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) 2696 return (0); 2697 2698 ifp = inm->in6m_ifp; /* interface */ 2699 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ 2700 crt = REC_NONE; /* current group record type */ 2701 drt = REC_NONE; /* mask of completed group record types */ 2702 nrt = REC_NONE; /* record type for current node */ 2703 m0srcs = 0; /* # source which will fit in current mbuf chain */ 2704 npbytes = 0; /* # of bytes appended this packet */ 2705 nbytes = 0; /* # of bytes appended to group's state-change queue */ 2706 rsrcs = 0; /* # sources encoded in current record */ 2707 schanged = 0; /* # nodes encoded in overall filter change */ 2708 nallow = 0; /* # of source entries in ALLOW_NEW */ 2709 nblock = 0; /* # of source entries in BLOCK_OLD */ 2710 nims = NULL; /* next tree node pointer */ 2711 2712 /* 2713 * For each possible filter record mode. 2714 * The first kind of source we encounter tells us which 2715 * is the first kind of record we start appending. 2716 * If a node transitioned to UNDEFINED at t1, its mode is treated 2717 * as the inverse of the group's filter mode. 2718 */ 2719 while (drt != REC_FULL) { 2720 do { 2721 m0 = ifq->ifq_tail; 2722 if (m0 != NULL && 2723 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 2724 MLD_V2_REPORT_MAXRECS) && 2725 (m0->m_pkthdr.len + MINRECLEN) < 2726 (ifp->if_mtu - MLD_MTUSPACE)) { 2727 m = m0; 2728 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2729 sizeof(struct mldv2_record)) / 2730 sizeof(struct in6_addr); 2731 CTR1(KTR_MLD, 2732 "%s: use previous packet", __func__); 2733 } else { 2734 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2735 if (m == NULL) 2736 m = m_gethdr(M_DONTWAIT, MT_DATA); 2737 if (m == NULL) { 2738 CTR1(KTR_MLD, 2739 "%s: m_get*() failed", __func__); 2740 return (-ENOMEM); 2741 } 2742 m->m_pkthdr.PH_vt.vt_nrecs = 0; 2743 mld_save_context(m, ifp); 2744 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2745 sizeof(struct mldv2_record)) / 2746 sizeof(struct in6_addr); 2747 npbytes = 0; 2748 CTR1(KTR_MLD, 2749 "%s: allocated new packet", __func__); 2750 } 2751 /* 2752 * Append the MLD group record header to the 2753 * current packet's data area. 2754 * Recalculate pointer to free space for next 2755 * group record, in case m_append() allocated 2756 * a new mbuf or cluster. 2757 */ 2758 memset(&mr, 0, sizeof(mr)); 2759 mr.mr_addr = inm->in6m_addr; 2760 in6_clearscope(&mr.mr_addr); 2761 if (!m_append(m, sizeof(mr), (void *)&mr)) { 2762 if (m != m0) 2763 m_freem(m); 2764 CTR1(KTR_MLD, 2765 "%s: m_append() failed", __func__); 2766 return (-ENOMEM); 2767 } 2768 npbytes += sizeof(struct mldv2_record); 2769 if (m != m0) { 2770 /* new packet; offset in chain */ 2771 md = m_getptr(m, npbytes - 2772 sizeof(struct mldv2_record), &off); 2773 pmr = (struct mldv2_record *)(mtod(md, 2774 uint8_t *) + off); 2775 } else { 2776 /* current packet; offset from last append */ 2777 md = m_last(m); 2778 pmr = (struct mldv2_record *)(mtod(md, 2779 uint8_t *) + md->m_len - 2780 sizeof(struct mldv2_record)); 2781 } 2782 /* 2783 * Begin walking the tree for this record type 2784 * pass, or continue from where we left off 2785 * previously if we had to allocate a new packet. 2786 * Only report deltas in-mode at t1. 2787 * We need not report included sources as allowed 2788 * if we are in inclusive mode on the group, 2789 * however the converse is not true. 2790 */ 2791 rsrcs = 0; 2792 if (nims == NULL) { 2793 nims = RB_MIN(ip6_msource_tree, 2794 &inm->in6m_srcs); 2795 } 2796 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2797 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2798 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2799 now = im6s_get_mode(inm, ims, 1); 2800 then = im6s_get_mode(inm, ims, 0); 2801 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", 2802 __func__, then, now); 2803 if (now == then) { 2804 CTR1(KTR_MLD, 2805 "%s: skip unchanged", __func__); 2806 continue; 2807 } 2808 if (mode == MCAST_EXCLUDE && 2809 now == MCAST_INCLUDE) { 2810 CTR1(KTR_MLD, 2811 "%s: skip IN src on EX group", 2812 __func__); 2813 continue; 2814 } 2815 nrt = (rectype_t)now; 2816 if (nrt == REC_NONE) 2817 nrt = (rectype_t)(~mode & REC_FULL); 2818 if (schanged++ == 0) { 2819 crt = nrt; 2820 } else if (crt != nrt) 2821 continue; 2822 if (!m_append(m, sizeof(struct in6_addr), 2823 (void *)&ims->im6s_addr)) { 2824 if (m != m0) 2825 m_freem(m); 2826 CTR1(KTR_MLD, 2827 "%s: m_append() failed", __func__); 2828 return (-ENOMEM); 2829 } 2830 nallow += !!(crt == REC_ALLOW); 2831 nblock += !!(crt == REC_BLOCK); 2832 if (++rsrcs == m0srcs) 2833 break; 2834 } 2835 /* 2836 * If we did not append any tree nodes on this 2837 * pass, back out of allocations. 2838 */ 2839 if (rsrcs == 0) { 2840 npbytes -= sizeof(struct mldv2_record); 2841 if (m != m0) { 2842 CTR1(KTR_MLD, 2843 "%s: m_free(m)", __func__); 2844 m_freem(m); 2845 } else { 2846 CTR1(KTR_MLD, 2847 "%s: m_adj(m, -mr)", __func__); 2848 m_adj(m, -((int)sizeof( 2849 struct mldv2_record))); 2850 } 2851 continue; 2852 } 2853 npbytes += (rsrcs * sizeof(struct in6_addr)); 2854 if (crt == REC_ALLOW) 2855 pmr->mr_type = MLD_ALLOW_NEW_SOURCES; 2856 else if (crt == REC_BLOCK) 2857 pmr->mr_type = MLD_BLOCK_OLD_SOURCES; 2858 pmr->mr_numsrc = htons(rsrcs); 2859 /* 2860 * Count the new group record, and enqueue this 2861 * packet if it wasn't already queued. 2862 */ 2863 m->m_pkthdr.PH_vt.vt_nrecs++; 2864 if (m != m0) 2865 _IF_ENQUEUE(ifq, m); 2866 nbytes += npbytes; 2867 } while (nims != NULL); 2868 drt |= crt; 2869 crt = (~crt & REC_FULL); 2870 } 2871 2872 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 2873 nallow, nblock); 2874 2875 return (nbytes); 2876 } 2877 2878 static int 2879 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq) 2880 { 2881 struct ifqueue *gq; 2882 struct mbuf *m; /* pending state-change */ 2883 struct mbuf *m0; /* copy of pending state-change */ 2884 struct mbuf *mt; /* last state-change in packet */ 2885 int docopy, domerge; 2886 u_int recslen; 2887 2888 docopy = 0; 2889 domerge = 0; 2890 recslen = 0; 2891 2892 IN6_MULTI_LOCK_ASSERT(); 2893 MLD_LOCK_ASSERT(); 2894 2895 /* 2896 * If there are further pending retransmissions, make a writable 2897 * copy of each queued state-change message before merging. 2898 */ 2899 if (inm->in6m_scrv > 0) 2900 docopy = 1; 2901 2902 gq = &inm->in6m_scq; 2903 #ifdef KTR 2904 if (gq->ifq_head == NULL) { 2905 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", 2906 __func__, inm); 2907 } 2908 #endif 2909 2910 m = gq->ifq_head; 2911 while (m != NULL) { 2912 /* 2913 * Only merge the report into the current packet if 2914 * there is sufficient space to do so; an MLDv2 report 2915 * packet may only contain 65,535 group records. 2916 * Always use a simple mbuf chain concatentation to do this, 2917 * as large state changes for single groups may have 2918 * allocated clusters. 2919 */ 2920 domerge = 0; 2921 mt = ifscq->ifq_tail; 2922 if (mt != NULL) { 2923 recslen = m_length(m, NULL); 2924 2925 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 2926 m->m_pkthdr.PH_vt.vt_nrecs <= 2927 MLD_V2_REPORT_MAXRECS) && 2928 (mt->m_pkthdr.len + recslen <= 2929 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) 2930 domerge = 1; 2931 } 2932 2933 if (!domerge && _IF_QFULL(gq)) { 2934 CTR2(KTR_MLD, 2935 "%s: outbound queue full, skipping whole packet %p", 2936 __func__, m); 2937 mt = m->m_nextpkt; 2938 if (!docopy) 2939 m_freem(m); 2940 m = mt; 2941 continue; 2942 } 2943 2944 if (!docopy) { 2945 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); 2946 _IF_DEQUEUE(gq, m0); 2947 m = m0->m_nextpkt; 2948 } else { 2949 CTR2(KTR_MLD, "%s: copying %p", __func__, m); 2950 m0 = m_dup(m, M_NOWAIT); 2951 if (m0 == NULL) 2952 return (ENOMEM); 2953 m0->m_nextpkt = NULL; 2954 m = m->m_nextpkt; 2955 } 2956 2957 if (!domerge) { 2958 CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)", 2959 __func__, m0, ifscq); 2960 _IF_ENQUEUE(ifscq, m0); 2961 } else { 2962 struct mbuf *mtl; /* last mbuf of packet mt */ 2963 2964 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", 2965 __func__, m0, mt); 2966 2967 mtl = m_last(mt); 2968 m0->m_flags &= ~M_PKTHDR; 2969 mt->m_pkthdr.len += recslen; 2970 mt->m_pkthdr.PH_vt.vt_nrecs += 2971 m0->m_pkthdr.PH_vt.vt_nrecs; 2972 2973 mtl->m_next = m0; 2974 } 2975 } 2976 2977 return (0); 2978 } 2979 2980 /* 2981 * Respond to a pending MLDv2 General Query. 2982 */ 2983 static void 2984 mld_v2_dispatch_general_query(struct mld_ifinfo *mli) 2985 { 2986 struct ifmultiaddr *ifma; 2987 struct ifnet *ifp; 2988 struct in6_multi *inm; 2989 int retval; 2990 2991 IN6_MULTI_LOCK_ASSERT(); 2992 MLD_LOCK_ASSERT(); 2993 2994 KASSERT(mli->mli_version == MLD_VERSION_2, 2995 ("%s: called when version %d", __func__, mli->mli_version)); 2996 2997 ifp = mli->mli_ifp; 2998 2999 IF_ADDR_RLOCK(ifp); 3000 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3001 if (ifma->ifma_addr->sa_family != AF_INET6 || 3002 ifma->ifma_protospec == NULL) 3003 continue; 3004 3005 inm = (struct in6_multi *)ifma->ifma_protospec; 3006 KASSERT(ifp == inm->in6m_ifp, 3007 ("%s: inconsistent ifp", __func__)); 3008 3009 switch (inm->in6m_state) { 3010 case MLD_NOT_MEMBER: 3011 case MLD_SILENT_MEMBER: 3012 break; 3013 case MLD_REPORTING_MEMBER: 3014 case MLD_IDLE_MEMBER: 3015 case MLD_LAZY_MEMBER: 3016 case MLD_SLEEPING_MEMBER: 3017 case MLD_AWAKENING_MEMBER: 3018 inm->in6m_state = MLD_REPORTING_MEMBER; 3019 retval = mld_v2_enqueue_group_record(&mli->mli_gq, 3020 inm, 0, 0, 0, 0); 3021 CTR2(KTR_MLD, "%s: enqueue record = %d", 3022 __func__, retval); 3023 break; 3024 case MLD_G_QUERY_PENDING_MEMBER: 3025 case MLD_SG_QUERY_PENDING_MEMBER: 3026 case MLD_LEAVING_MEMBER: 3027 break; 3028 } 3029 } 3030 IF_ADDR_RUNLOCK(ifp); 3031 3032 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); 3033 3034 /* 3035 * Slew transmission of bursts over 500ms intervals. 3036 */ 3037 if (mli->mli_gq.ifq_head != NULL) { 3038 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( 3039 MLD_RESPONSE_BURST_INTERVAL); 3040 V_interface_timers_running6 = 1; 3041 } 3042 } 3043 3044 /* 3045 * Transmit the next pending message in the output queue. 3046 * 3047 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3048 * MRT: Nothing needs to be done, as MLD traffic is always local to 3049 * a link and uses a link-scope multicast address. 3050 */ 3051 static void 3052 mld_dispatch_packet(struct mbuf *m) 3053 { 3054 struct ip6_moptions im6o; 3055 struct ifnet *ifp; 3056 struct ifnet *oifp; 3057 struct mbuf *m0; 3058 struct mbuf *md; 3059 struct ip6_hdr *ip6; 3060 struct mld_hdr *mld; 3061 int error; 3062 int off; 3063 int type; 3064 uint32_t ifindex; 3065 3066 CTR2(KTR_MLD, "%s: transmit %p", __func__, m); 3067 3068 /* 3069 * Set VNET image pointer from enqueued mbuf chain 3070 * before doing anything else. Whilst we use interface 3071 * indexes to guard against interface detach, they are 3072 * unique to each VIMAGE and must be retrieved. 3073 */ 3074 ifindex = mld_restore_context(m); 3075 3076 /* 3077 * Check if the ifnet still exists. This limits the scope of 3078 * any race in the absence of a global ifp lock for low cost 3079 * (an array lookup). 3080 */ 3081 ifp = ifnet_byindex(ifindex); 3082 if (ifp == NULL) { 3083 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", 3084 __func__, m, ifindex); 3085 m_freem(m); 3086 IP6STAT_INC(ip6s_noroute); 3087 goto out; 3088 } 3089 3090 im6o.im6o_multicast_hlim = 1; 3091 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); 3092 im6o.im6o_multicast_ifp = ifp; 3093 3094 if (m->m_flags & M_MLDV1) { 3095 m0 = m; 3096 } else { 3097 m0 = mld_v2_encap_report(ifp, m); 3098 if (m0 == NULL) { 3099 CTR2(KTR_MLD, "%s: dropped %p", __func__, m); 3100 IP6STAT_INC(ip6s_odropped); 3101 goto out; 3102 } 3103 } 3104 3105 mld_scrub_context(m0); 3106 m->m_flags &= ~(M_PROTOFLAGS); 3107 m0->m_pkthdr.rcvif = V_loif; 3108 3109 ip6 = mtod(m0, struct ip6_hdr *); 3110 #if 0 3111 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */ 3112 #else 3113 /* 3114 * XXX XXX Break some KPI rules to prevent an LOR which would 3115 * occur if we called in6_setscope() at transmission. 3116 * See comments at top of file. 3117 */ 3118 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index); 3119 #endif 3120 3121 /* 3122 * Retrieve the ICMPv6 type before handoff to ip6_output(), 3123 * so we can bump the stats. 3124 */ 3125 md = m_getptr(m0, sizeof(struct ip6_hdr), &off); 3126 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off); 3127 type = mld->mld_type; 3128 3129 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o, 3130 &oifp, NULL); 3131 if (error) { 3132 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error); 3133 goto out; 3134 } 3135 ICMP6STAT_INC(icp6s_outhist[type]); 3136 if (oifp != NULL) { 3137 icmp6_ifstat_inc(oifp, ifs6_out_msg); 3138 switch (type) { 3139 case MLD_LISTENER_REPORT: 3140 case MLDV2_LISTENER_REPORT: 3141 icmp6_ifstat_inc(oifp, ifs6_out_mldreport); 3142 break; 3143 case MLD_LISTENER_DONE: 3144 icmp6_ifstat_inc(oifp, ifs6_out_mlddone); 3145 break; 3146 } 3147 } 3148 out: 3149 return; 3150 } 3151 3152 /* 3153 * Encapsulate an MLDv2 report. 3154 * 3155 * KAME IPv6 requires that hop-by-hop options be passed separately, 3156 * and that the IPv6 header be prepended in a separate mbuf. 3157 * 3158 * Returns a pointer to the new mbuf chain head, or NULL if the 3159 * allocation failed. 3160 */ 3161 static struct mbuf * 3162 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m) 3163 { 3164 struct mbuf *mh; 3165 struct mldv2_report *mld; 3166 struct ip6_hdr *ip6; 3167 struct in6_ifaddr *ia; 3168 int mldreclen; 3169 3170 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 3171 KASSERT((m->m_flags & M_PKTHDR), 3172 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3173 3174 /* 3175 * RFC3590: OK to send as :: or tentative during DAD. 3176 */ 3177 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 3178 if (ia == NULL) 3179 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__); 3180 3181 MGETHDR(mh, M_DONTWAIT, MT_HEADER); 3182 if (mh == NULL) { 3183 if (ia != NULL) 3184 ifa_free(&ia->ia_ifa); 3185 m_freem(m); 3186 return (NULL); 3187 } 3188 MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); 3189 3190 mldreclen = m_length(m, NULL); 3191 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); 3192 3193 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report); 3194 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + 3195 sizeof(struct mldv2_report) + mldreclen; 3196 3197 ip6 = mtod(mh, struct ip6_hdr *); 3198 ip6->ip6_flow = 0; 3199 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3200 ip6->ip6_vfc |= IPV6_VERSION; 3201 ip6->ip6_nxt = IPPROTO_ICMPV6; 3202 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 3203 if (ia != NULL) 3204 ifa_free(&ia->ia_ifa); 3205 ip6->ip6_dst = in6addr_linklocal_allv2routers; 3206 /* scope ID will be set in netisr */ 3207 3208 mld = (struct mldv2_report *)(ip6 + 1); 3209 mld->mld_type = MLDV2_LISTENER_REPORT; 3210 mld->mld_code = 0; 3211 mld->mld_cksum = 0; 3212 mld->mld_v2_reserved = 0; 3213 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs); 3214 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3215 3216 mh->m_next = m; 3217 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 3218 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen); 3219 return (mh); 3220 } 3221 3222 #ifdef KTR 3223 static char * 3224 mld_rec_type_to_str(const int type) 3225 { 3226 3227 switch (type) { 3228 case MLD_CHANGE_TO_EXCLUDE_MODE: 3229 return "TO_EX"; 3230 break; 3231 case MLD_CHANGE_TO_INCLUDE_MODE: 3232 return "TO_IN"; 3233 break; 3234 case MLD_MODE_IS_EXCLUDE: 3235 return "MODE_EX"; 3236 break; 3237 case MLD_MODE_IS_INCLUDE: 3238 return "MODE_IN"; 3239 break; 3240 case MLD_ALLOW_NEW_SOURCES: 3241 return "ALLOW_NEW"; 3242 break; 3243 case MLD_BLOCK_OLD_SOURCES: 3244 return "BLOCK_OLD"; 3245 break; 3246 default: 3247 break; 3248 } 3249 return "unknown"; 3250 } 3251 #endif 3252 3253 static void 3254 mld_init(void *unused __unused) 3255 { 3256 3257 CTR1(KTR_MLD, "%s: initializing", __func__); 3258 MLD_LOCK_INIT(); 3259 3260 ip6_initpktopts(&mld_po); 3261 mld_po.ip6po_hlim = 1; 3262 mld_po.ip6po_hbh = &mld_ra.hbh; 3263 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER; 3264 mld_po.ip6po_flags = IP6PO_DONTFRAG; 3265 } 3266 SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL); 3267 3268 static void 3269 mld_uninit(void *unused __unused) 3270 { 3271 3272 CTR1(KTR_MLD, "%s: tearing down", __func__); 3273 MLD_LOCK_DESTROY(); 3274 } 3275 SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL); 3276 3277 static void 3278 vnet_mld_init(const void *unused __unused) 3279 { 3280 3281 CTR1(KTR_MLD, "%s: initializing", __func__); 3282 3283 LIST_INIT(&V_mli_head); 3284 } 3285 VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init, 3286 NULL); 3287 3288 static void 3289 vnet_mld_uninit(const void *unused __unused) 3290 { 3291 3292 CTR1(KTR_MLD, "%s: tearing down", __func__); 3293 3294 KASSERT(LIST_EMPTY(&V_mli_head), 3295 ("%s: mli list not empty; ifnets not detached?", __func__)); 3296 } 3297 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit, 3298 NULL); 3299 3300 static int 3301 mld_modevent(module_t mod, int type, void *unused __unused) 3302 { 3303 3304 switch (type) { 3305 case MOD_LOAD: 3306 case MOD_UNLOAD: 3307 break; 3308 default: 3309 return (EOPNOTSUPP); 3310 } 3311 return (0); 3312 } 3313 3314 static moduledata_t mld_mod = { 3315 "mld", 3316 mld_modevent, 3317 0 3318 }; 3319 DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3320