mld6.c revision 233200
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: stable/9/sys/netinet6/mld6.c 233200 2012-03-19 20:49:16Z jhb $"); 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 104static struct mld_ifinfo * 105 mli_alloc_locked(struct ifnet *); 106static void mli_delete_locked(const struct ifnet *); 107static void mld_dispatch_packet(struct mbuf *); 108static void mld_dispatch_queue(struct ifqueue *, int); 109static void mld_final_leave(struct in6_multi *, struct mld_ifinfo *); 110static void mld_fasttimo_vnet(void); 111static int mld_handle_state_change(struct in6_multi *, 112 struct mld_ifinfo *); 113static int mld_initial_join(struct in6_multi *, struct mld_ifinfo *, 114 const int); 115#ifdef KTR 116static char * mld_rec_type_to_str(const int); 117#endif 118static void mld_set_version(struct mld_ifinfo *, const int); 119static void mld_slowtimo_vnet(void); 120static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *, 121 /*const*/ struct mld_hdr *); 122static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *, 123 /*const*/ struct mld_hdr *); 124static void mld_v1_process_group_timer(struct mld_ifinfo *, 125 struct in6_multi *); 126static void mld_v1_process_querier_timers(struct mld_ifinfo *); 127static int mld_v1_transmit_report(struct in6_multi *, const int); 128static void mld_v1_update_group(struct in6_multi *, const int); 129static void mld_v2_cancel_link_timers(struct mld_ifinfo *); 130static void mld_v2_dispatch_general_query(struct mld_ifinfo *); 131static struct mbuf * 132 mld_v2_encap_report(struct ifnet *, struct mbuf *); 133static int mld_v2_enqueue_filter_change(struct ifqueue *, 134 struct in6_multi *); 135static int mld_v2_enqueue_group_record(struct ifqueue *, 136 struct in6_multi *, const int, const int, const int, 137 const int); 138static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *, 139 struct mbuf *, const int, const int); 140static int mld_v2_merge_state_changes(struct in6_multi *, 141 struct ifqueue *); 142static void mld_v2_process_group_timers(struct mld_ifinfo *, 143 struct ifqueue *, struct ifqueue *, 144 struct in6_multi *, const int); 145static int mld_v2_process_group_query(struct in6_multi *, 146 struct mld_ifinfo *mli, int, struct mbuf *, const int); 147static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); 148static 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 */ 196static struct mtx mld_mtx; 197MALLOC_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 */ 207static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0}; 208static VNET_DEFINE(LIST_HEAD(, mld_ifinfo), mli_head); 209static VNET_DEFINE(int, interface_timers_running6); 210static VNET_DEFINE(int, state_change_timers_running6); 211static 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 219SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ 220 221SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, 222 "IPv6 Multicast Listener Discovery"); 223 224/* 225 * Virtualized sysctls. 226 */ 227SYSCTL_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 */ 235SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE, 236 sysctl_mld_ifinfo, "Per-interface MLDv2 state"); 237 238static int mld_v1enable = 1; 239SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW, 240 &mld_v1enable, 0, "Enable fallback to MLDv1"); 241TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable); 242 243static int mld_use_allow = 1; 244SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW, 245 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves"); 246TUNABLE_INT("net.inet6.mld.use_allow", &mld_use_allow); 247 248/* 249 * Packed Router Alert option structure declaration. 250 */ 251struct mld_raopt { 252 struct ip6_hbh hbh; 253 struct ip6_opt pad; 254 struct ip6_opt_router ra; 255} __packed; 256 257/* 258 * Router Alert hop-by-hop option header. 259 */ 260static struct mld_raopt mld_ra = { 261 .hbh = { 0, 0 }, 262 .pad = { .ip6o_type = IP6OPT_PADN, 0 }, 263 .ra = { 264 .ip6or_type = IP6OPT_ROUTER_ALERT, 265 .ip6or_len = IP6OPT_RTALERT_LEN - 2, 266 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), 267 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) 268 } 269}; 270static struct ip6_pktopts mld_po; 271 272static __inline void 273mld_save_context(struct mbuf *m, struct ifnet *ifp) 274{ 275 276#ifdef VIMAGE 277 m->m_pkthdr.header = ifp->if_vnet; 278#endif /* VIMAGE */ 279 m->m_pkthdr.flowid = ifp->if_index; 280} 281 282static __inline void 283mld_scrub_context(struct mbuf *m) 284{ 285 286 m->m_pkthdr.header = NULL; 287 m->m_pkthdr.flowid = 0; 288} 289 290/* 291 * Restore context from a queued output chain. 292 * Return saved ifindex. 293 * 294 * VIMAGE: The assertion is there to make sure that we 295 * actually called CURVNET_SET() with what's in the mbuf chain. 296 */ 297static __inline uint32_t 298mld_restore_context(struct mbuf *m) 299{ 300 301#if defined(VIMAGE) && defined(INVARIANTS) 302 KASSERT(curvnet == m->m_pkthdr.header, 303 ("%s: called when curvnet was not restored", __func__)); 304#endif 305 return (m->m_pkthdr.flowid); 306} 307 308/* 309 * Retrieve or set threshold between group-source queries in seconds. 310 * 311 * VIMAGE: Assume curvnet set by caller. 312 * SMPng: NOTE: Serialized by MLD lock. 313 */ 314static int 315sysctl_mld_gsr(SYSCTL_HANDLER_ARGS) 316{ 317 int error; 318 int i; 319 320 error = sysctl_wire_old_buffer(req, sizeof(int)); 321 if (error) 322 return (error); 323 324 MLD_LOCK(); 325 326 i = V_mld_gsrdelay.tv_sec; 327 328 error = sysctl_handle_int(oidp, &i, 0, req); 329 if (error || !req->newptr) 330 goto out_locked; 331 332 if (i < -1 || i >= 60) { 333 error = EINVAL; 334 goto out_locked; 335 } 336 337 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", 338 V_mld_gsrdelay.tv_sec, i); 339 V_mld_gsrdelay.tv_sec = i; 340 341out_locked: 342 MLD_UNLOCK(); 343 return (error); 344} 345 346/* 347 * Expose struct mld_ifinfo to userland, keyed by ifindex. 348 * For use by ifmcstat(8). 349 * 350 * SMPng: NOTE: Does an unlocked ifindex space read. 351 * VIMAGE: Assume curvnet set by caller. The node handler itself 352 * is not directly virtualized. 353 */ 354static int 355sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) 356{ 357 int *name; 358 int error; 359 u_int namelen; 360 struct ifnet *ifp; 361 struct mld_ifinfo *mli; 362 363 name = (int *)arg1; 364 namelen = arg2; 365 366 if (req->newptr != NULL) 367 return (EPERM); 368 369 if (namelen != 1) 370 return (EINVAL); 371 372 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo)); 373 if (error) 374 return (error); 375 376 IN6_MULTI_LOCK(); 377 MLD_LOCK(); 378 379 if (name[0] <= 0 || name[0] > V_if_index) { 380 error = ENOENT; 381 goto out_locked; 382 } 383 384 error = ENOENT; 385 386 ifp = ifnet_byindex(name[0]); 387 if (ifp == NULL) 388 goto out_locked; 389 390 LIST_FOREACH(mli, &V_mli_head, mli_link) { 391 if (ifp == mli->mli_ifp) { 392 error = SYSCTL_OUT(req, mli, 393 sizeof(struct mld_ifinfo)); 394 break; 395 } 396 } 397 398out_locked: 399 MLD_UNLOCK(); 400 IN6_MULTI_UNLOCK(); 401 return (error); 402} 403 404/* 405 * Dispatch an entire queue of pending packet chains. 406 * VIMAGE: Assumes the vnet pointer has been set. 407 */ 408static void 409mld_dispatch_queue(struct ifqueue *ifq, int limit) 410{ 411 struct mbuf *m; 412 413 for (;;) { 414 _IF_DEQUEUE(ifq, m); 415 if (m == NULL) 416 break; 417 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m); 418 mld_dispatch_packet(m); 419 if (--limit == 0) 420 break; 421 } 422} 423 424/* 425 * Filter outgoing MLD report state by group. 426 * 427 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1) 428 * and node-local addresses. However, kernel and socket consumers 429 * always embed the KAME scope ID in the address provided, so strip it 430 * when performing comparison. 431 * Note: This is not the same as the *multicast* scope. 432 * 433 * Return zero if the given group is one for which MLD reports 434 * should be suppressed, or non-zero if reports should be issued. 435 */ 436static __inline int 437mld_is_addr_reported(const struct in6_addr *addr) 438{ 439 440 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__)); 441 442 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) 443 return (0); 444 445 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) { 446 struct in6_addr tmp = *addr; 447 in6_clearscope(&tmp); 448 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) 449 return (0); 450 } 451 452 return (1); 453} 454 455/* 456 * Attach MLD when PF_INET6 is attached to an interface. 457 * 458 * SMPng: Normally called with IF_AFDATA_LOCK held. 459 */ 460struct mld_ifinfo * 461mld_domifattach(struct ifnet *ifp) 462{ 463 struct mld_ifinfo *mli; 464 465 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 466 __func__, ifp, ifp->if_xname); 467 468 MLD_LOCK(); 469 470 mli = mli_alloc_locked(ifp); 471 if (!(ifp->if_flags & IFF_MULTICAST)) 472 mli->mli_flags |= MLIF_SILENT; 473 if (mld_use_allow) 474 mli->mli_flags |= MLIF_USEALLOW; 475 476 MLD_UNLOCK(); 477 478 return (mli); 479} 480 481/* 482 * VIMAGE: assume curvnet set by caller. 483 */ 484static struct mld_ifinfo * 485mli_alloc_locked(/*const*/ struct ifnet *ifp) 486{ 487 struct mld_ifinfo *mli; 488 489 MLD_LOCK_ASSERT(); 490 491 mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO); 492 if (mli == NULL) 493 goto out; 494 495 mli->mli_ifp = ifp; 496 mli->mli_version = MLD_VERSION_2; 497 mli->mli_flags = 0; 498 mli->mli_rv = MLD_RV_INIT; 499 mli->mli_qi = MLD_QI_INIT; 500 mli->mli_qri = MLD_QRI_INIT; 501 mli->mli_uri = MLD_URI_INIT; 502 503 SLIST_INIT(&mli->mli_relinmhead); 504 505 /* 506 * Responses to general queries are subject to bounds. 507 */ 508 IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); 509 510 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); 511 512 CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)", 513 ifp, ifp->if_xname); 514 515out: 516 return (mli); 517} 518 519/* 520 * Hook for ifdetach. 521 * 522 * NOTE: Some finalization tasks need to run before the protocol domain 523 * is detached, but also before the link layer does its cleanup. 524 * Run before link-layer cleanup; cleanup groups, but do not free MLD state. 525 * 526 * SMPng: Caller must hold IN6_MULTI_LOCK(). 527 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. 528 * XXX This routine is also bitten by unlocked ifma_protospec access. 529 */ 530void 531mld_ifdetach(struct ifnet *ifp) 532{ 533 struct mld_ifinfo *mli; 534 struct ifmultiaddr *ifma; 535 struct in6_multi *inm, *tinm; 536 537 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, 538 ifp->if_xname); 539 540 IN6_MULTI_LOCK_ASSERT(); 541 MLD_LOCK(); 542 543 mli = MLD_IFINFO(ifp); 544 if (mli->mli_version == MLD_VERSION_2) { 545 IF_ADDR_RLOCK(ifp); 546 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 547 if (ifma->ifma_addr->sa_family != AF_INET6 || 548 ifma->ifma_protospec == NULL) 549 continue; 550 inm = (struct in6_multi *)ifma->ifma_protospec; 551 if (inm->in6m_state == MLD_LEAVING_MEMBER) { 552 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 553 inm, in6m_nrele); 554 } 555 in6m_clear_recorded(inm); 556 } 557 IF_ADDR_RUNLOCK(ifp); 558 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, 559 tinm) { 560 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 561 in6m_release_locked(inm); 562 } 563 } 564 565 MLD_UNLOCK(); 566} 567 568/* 569 * Hook for domifdetach. 570 * Runs after link-layer cleanup; free MLD state. 571 * 572 * SMPng: Normally called with IF_AFDATA_LOCK held. 573 */ 574void 575mld_domifdetach(struct ifnet *ifp) 576{ 577 578 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 579 __func__, ifp, ifp->if_xname); 580 581 MLD_LOCK(); 582 mli_delete_locked(ifp); 583 MLD_UNLOCK(); 584} 585 586static void 587mli_delete_locked(const struct ifnet *ifp) 588{ 589 struct mld_ifinfo *mli, *tmli; 590 591 CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)", 592 __func__, ifp, ifp->if_xname); 593 594 MLD_LOCK_ASSERT(); 595 596 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) { 597 if (mli->mli_ifp == ifp) { 598 /* 599 * Free deferred General Query responses. 600 */ 601 _IF_DRAIN(&mli->mli_gq); 602 603 LIST_REMOVE(mli, mli_link); 604 605 KASSERT(SLIST_EMPTY(&mli->mli_relinmhead), 606 ("%s: there are dangling in_multi references", 607 __func__)); 608 609 free(mli, M_MLD); 610 return; 611 } 612 } 613#ifdef INVARIANTS 614 panic("%s: mld_ifinfo not found for ifp %p\n", __func__, ifp); 615#endif 616} 617 618/* 619 * Process a received MLDv1 general or address-specific query. 620 * Assumes that the query header has been pulled up to sizeof(mld_hdr). 621 * 622 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 623 * mld_addr. This is OK as we own the mbuf chain. 624 */ 625static int 626mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 627 /*const*/ struct mld_hdr *mld) 628{ 629 struct ifmultiaddr *ifma; 630 struct mld_ifinfo *mli; 631 struct in6_multi *inm; 632 int is_general_query; 633 uint16_t timer; 634#ifdef KTR 635 char ip6tbuf[INET6_ADDRSTRLEN]; 636#endif 637 638 is_general_query = 0; 639 640 if (!mld_v1enable) { 641 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)", 642 ip6_sprintf(ip6tbuf, &mld->mld_addr), 643 ifp, ifp->if_xname); 644 return (0); 645 } 646 647 /* 648 * RFC3810 Section 6.2: MLD queries must originate from 649 * a router's link-local address. 650 */ 651 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 652 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 653 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 654 ifp, ifp->if_xname); 655 return (0); 656 } 657 658 /* 659 * Do address field validation upfront before we accept 660 * the query. 661 */ 662 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 663 /* 664 * MLDv1 General Query. 665 * If this was not sent to the all-nodes group, ignore it. 666 */ 667 struct in6_addr dst; 668 669 dst = ip6->ip6_dst; 670 in6_clearscope(&dst); 671 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) 672 return (EINVAL); 673 is_general_query = 1; 674 } else { 675 /* 676 * Embed scope ID of receiving interface in MLD query for 677 * lookup whilst we don't hold other locks. 678 */ 679 in6_setscope(&mld->mld_addr, ifp, NULL); 680 } 681 682 IN6_MULTI_LOCK(); 683 MLD_LOCK(); 684 685 /* 686 * Switch to MLDv1 host compatibility mode. 687 */ 688 mli = MLD_IFINFO(ifp); 689 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 690 mld_set_version(mli, MLD_VERSION_1); 691 692 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; 693 if (timer == 0) 694 timer = 1; 695 696 IF_ADDR_RLOCK(ifp); 697 if (is_general_query) { 698 /* 699 * For each reporting group joined on this 700 * interface, kick the report timer. 701 */ 702 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)", 703 ifp, ifp->if_xname); 704 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 705 if (ifma->ifma_addr->sa_family != AF_INET6 || 706 ifma->ifma_protospec == NULL) 707 continue; 708 inm = (struct in6_multi *)ifma->ifma_protospec; 709 mld_v1_update_group(inm, timer); 710 } 711 } else { 712 /* 713 * MLDv1 Group-Specific Query. 714 * If this is a group-specific MLDv1 query, we need only 715 * look up the single group to process it. 716 */ 717 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 718 if (inm != NULL) { 719 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)", 720 ip6_sprintf(ip6tbuf, &mld->mld_addr), 721 ifp, ifp->if_xname); 722 mld_v1_update_group(inm, timer); 723 } 724 /* XXX Clear embedded scope ID as userland won't expect it. */ 725 in6_clearscope(&mld->mld_addr); 726 } 727 728 IF_ADDR_RUNLOCK(ifp); 729 MLD_UNLOCK(); 730 IN6_MULTI_UNLOCK(); 731 732 return (0); 733} 734 735/* 736 * Update the report timer on a group in response to an MLDv1 query. 737 * 738 * If we are becoming the reporting member for this group, start the timer. 739 * If we already are the reporting member for this group, and timer is 740 * below the threshold, reset it. 741 * 742 * We may be updating the group for the first time since we switched 743 * to MLDv2. If we are, then we must clear any recorded source lists, 744 * and transition to REPORTING state; the group timer is overloaded 745 * for group and group-source query responses. 746 * 747 * Unlike MLDv2, the delay per group should be jittered 748 * to avoid bursts of MLDv1 reports. 749 */ 750static void 751mld_v1_update_group(struct in6_multi *inm, const int timer) 752{ 753#ifdef KTR 754 char ip6tbuf[INET6_ADDRSTRLEN]; 755#endif 756 757 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, 758 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 759 inm->in6m_ifp->if_xname, timer); 760 761 IN6_MULTI_LOCK_ASSERT(); 762 763 switch (inm->in6m_state) { 764 case MLD_NOT_MEMBER: 765 case MLD_SILENT_MEMBER: 766 break; 767 case MLD_REPORTING_MEMBER: 768 if (inm->in6m_timer != 0 && 769 inm->in6m_timer <= timer) { 770 CTR1(KTR_MLD, "%s: REPORTING and timer running, " 771 "skipping.", __func__); 772 break; 773 } 774 /* FALLTHROUGH */ 775 case MLD_SG_QUERY_PENDING_MEMBER: 776 case MLD_G_QUERY_PENDING_MEMBER: 777 case MLD_IDLE_MEMBER: 778 case MLD_LAZY_MEMBER: 779 case MLD_AWAKENING_MEMBER: 780 CTR1(KTR_MLD, "%s: ->REPORTING", __func__); 781 inm->in6m_state = MLD_REPORTING_MEMBER; 782 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 783 V_current_state_timers_running6 = 1; 784 break; 785 case MLD_SLEEPING_MEMBER: 786 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); 787 inm->in6m_state = MLD_AWAKENING_MEMBER; 788 break; 789 case MLD_LEAVING_MEMBER: 790 break; 791 } 792} 793 794/* 795 * Process a received MLDv2 general, group-specific or 796 * group-and-source-specific query. 797 * 798 * Assumes that the query header has been pulled up to sizeof(mldv2_query). 799 * 800 * Return 0 if successful, otherwise an appropriate error code is returned. 801 */ 802static int 803mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 804 struct mbuf *m, const int off, const int icmp6len) 805{ 806 struct mld_ifinfo *mli; 807 struct mldv2_query *mld; 808 struct in6_multi *inm; 809 uint32_t maxdelay, nsrc, qqi; 810 int is_general_query; 811 uint16_t timer; 812 uint8_t qrv; 813#ifdef KTR 814 char ip6tbuf[INET6_ADDRSTRLEN]; 815#endif 816 817 is_general_query = 0; 818 819 /* 820 * RFC3810 Section 6.2: MLD queries must originate from 821 * a router's link-local address. 822 */ 823 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 824 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 825 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 826 ifp, ifp->if_xname); 827 return (0); 828 } 829 830 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname); 831 832 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off); 833 834 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ 835 if (maxdelay >= 32678) { 836 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << 837 (MLD_MRC_EXP(maxdelay) + 3); 838 } 839 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; 840 if (timer == 0) 841 timer = 1; 842 843 qrv = MLD_QRV(mld->mld_misc); 844 if (qrv < 2) { 845 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, 846 qrv, MLD_RV_INIT); 847 qrv = MLD_RV_INIT; 848 } 849 850 qqi = mld->mld_qqi; 851 if (qqi >= 128) { 852 qqi = MLD_QQIC_MANT(mld->mld_qqi) << 853 (MLD_QQIC_EXP(mld->mld_qqi) + 3); 854 } 855 856 nsrc = ntohs(mld->mld_numsrc); 857 if (nsrc > MLD_MAX_GS_SOURCES) 858 return (EMSGSIZE); 859 if (icmp6len < sizeof(struct mldv2_query) + 860 (nsrc * sizeof(struct in6_addr))) 861 return (EMSGSIZE); 862 863 /* 864 * Do further input validation upfront to avoid resetting timers 865 * should we need to discard this query. 866 */ 867 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 868 /* 869 * General Queries SHOULD be directed to ff02::1. 870 * A general query with a source list has undefined 871 * behaviour; discard it. 872 */ 873 struct in6_addr dst; 874 875 dst = ip6->ip6_dst; 876 in6_clearscope(&dst); 877 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes) || 878 nsrc > 0) 879 return (EINVAL); 880 is_general_query = 1; 881 } else { 882 /* 883 * Embed scope ID of receiving interface in MLD query for 884 * lookup whilst we don't hold other locks (due to KAME 885 * locking lameness). We own this mbuf chain just now. 886 */ 887 in6_setscope(&mld->mld_addr, ifp, NULL); 888 } 889 890 IN6_MULTI_LOCK(); 891 MLD_LOCK(); 892 893 mli = MLD_IFINFO(ifp); 894 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 895 896 /* 897 * Discard the v2 query if we're in Compatibility Mode. 898 * The RFC is pretty clear that hosts need to stay in MLDv1 mode 899 * until the Old Version Querier Present timer expires. 900 */ 901 if (mli->mli_version != MLD_VERSION_2) 902 goto out_locked; 903 904 mld_set_version(mli, MLD_VERSION_2); 905 mli->mli_rv = qrv; 906 mli->mli_qi = qqi; 907 mli->mli_qri = maxdelay; 908 909 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, 910 maxdelay); 911 912 if (is_general_query) { 913 /* 914 * MLDv2 General Query. 915 * 916 * Schedule a current-state report on this ifp for 917 * all groups, possibly containing source lists. 918 * 919 * If there is a pending General Query response 920 * scheduled earlier than the selected delay, do 921 * not schedule any other reports. 922 * Otherwise, reset the interface timer. 923 */ 924 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", 925 ifp, ifp->if_xname); 926 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { 927 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); 928 V_interface_timers_running6 = 1; 929 } 930 } else { 931 /* 932 * MLDv2 Group-specific or Group-and-source-specific Query. 933 * 934 * Group-source-specific queries are throttled on 935 * a per-group basis to defeat denial-of-service attempts. 936 * Queries for groups we are not a member of on this 937 * link are simply ignored. 938 */ 939 IF_ADDR_RLOCK(ifp); 940 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 941 if (inm == NULL) { 942 IF_ADDR_RUNLOCK(ifp); 943 goto out_locked; 944 } 945 if (nsrc > 0) { 946 if (!ratecheck(&inm->in6m_lastgsrtv, 947 &V_mld_gsrdelay)) { 948 CTR1(KTR_MLD, "%s: GS query throttled.", 949 __func__); 950 IF_ADDR_RUNLOCK(ifp); 951 goto out_locked; 952 } 953 } 954 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", 955 ifp, ifp->if_xname); 956 /* 957 * If there is a pending General Query response 958 * scheduled sooner than the selected delay, no 959 * further report need be scheduled. 960 * Otherwise, prepare to respond to the 961 * group-specific or group-and-source query. 962 */ 963 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) 964 mld_v2_process_group_query(inm, mli, timer, m, off); 965 966 /* XXX Clear embedded scope ID as userland won't expect it. */ 967 in6_clearscope(&mld->mld_addr); 968 IF_ADDR_RUNLOCK(ifp); 969 } 970 971out_locked: 972 MLD_UNLOCK(); 973 IN6_MULTI_UNLOCK(); 974 975 return (0); 976} 977 978/* 979 * Process a recieved MLDv2 group-specific or group-and-source-specific 980 * query. 981 * Return <0 if any error occured. Currently this is ignored. 982 */ 983static int 984mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli, 985 int timer, struct mbuf *m0, const int off) 986{ 987 struct mldv2_query *mld; 988 int retval; 989 uint16_t nsrc; 990 991 IN6_MULTI_LOCK_ASSERT(); 992 MLD_LOCK_ASSERT(); 993 994 retval = 0; 995 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off); 996 997 switch (inm->in6m_state) { 998 case MLD_NOT_MEMBER: 999 case MLD_SILENT_MEMBER: 1000 case MLD_SLEEPING_MEMBER: 1001 case MLD_LAZY_MEMBER: 1002 case MLD_AWAKENING_MEMBER: 1003 case MLD_IDLE_MEMBER: 1004 case MLD_LEAVING_MEMBER: 1005 return (retval); 1006 break; 1007 case MLD_REPORTING_MEMBER: 1008 case MLD_G_QUERY_PENDING_MEMBER: 1009 case MLD_SG_QUERY_PENDING_MEMBER: 1010 break; 1011 } 1012 1013 nsrc = ntohs(mld->mld_numsrc); 1014 1015 /* 1016 * Deal with group-specific queries upfront. 1017 * If any group query is already pending, purge any recorded 1018 * source-list state if it exists, and schedule a query response 1019 * for this group-specific query. 1020 */ 1021 if (nsrc == 0) { 1022 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 1023 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { 1024 in6m_clear_recorded(inm); 1025 timer = min(inm->in6m_timer, timer); 1026 } 1027 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; 1028 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1029 V_current_state_timers_running6 = 1; 1030 return (retval); 1031 } 1032 1033 /* 1034 * Deal with the case where a group-and-source-specific query has 1035 * been received but a group-specific query is already pending. 1036 */ 1037 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { 1038 timer = min(inm->in6m_timer, timer); 1039 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1040 V_current_state_timers_running6 = 1; 1041 return (retval); 1042 } 1043 1044 /* 1045 * Finally, deal with the case where a group-and-source-specific 1046 * query has been received, where a response to a previous g-s-r 1047 * query exists, or none exists. 1048 * In this case, we need to parse the source-list which the Querier 1049 * has provided us with and check if we have any source list filter 1050 * entries at T1 for these sources. If we do not, there is no need 1051 * schedule a report and the query may be dropped. 1052 * If we do, we must record them and schedule a current-state 1053 * report for those sources. 1054 */ 1055 if (inm->in6m_nsrc > 0) { 1056 struct mbuf *m; 1057 uint8_t *sp; 1058 int i, nrecorded; 1059 int soff; 1060 1061 m = m0; 1062 soff = off + sizeof(struct mldv2_query); 1063 nrecorded = 0; 1064 for (i = 0; i < nsrc; i++) { 1065 sp = mtod(m, uint8_t *) + soff; 1066 retval = in6m_record_source(inm, 1067 (const struct in6_addr *)sp); 1068 if (retval < 0) 1069 break; 1070 nrecorded += retval; 1071 soff += sizeof(struct in6_addr); 1072 if (soff >= m->m_len) { 1073 soff = soff - m->m_len; 1074 m = m->m_next; 1075 if (m == NULL) 1076 break; 1077 } 1078 } 1079 if (nrecorded > 0) { 1080 CTR1(KTR_MLD, 1081 "%s: schedule response to SG query", __func__); 1082 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; 1083 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1084 V_current_state_timers_running6 = 1; 1085 } 1086 } 1087 1088 return (retval); 1089} 1090 1091/* 1092 * Process a received MLDv1 host membership report. 1093 * Assumes mld points to mld_hdr in pulled up mbuf chain. 1094 * 1095 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 1096 * mld_addr. This is OK as we own the mbuf chain. 1097 */ 1098static int 1099mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, 1100 /*const*/ struct mld_hdr *mld) 1101{ 1102 struct in6_addr src, dst; 1103 struct in6_ifaddr *ia; 1104 struct in6_multi *inm; 1105#ifdef KTR 1106 char ip6tbuf[INET6_ADDRSTRLEN]; 1107#endif 1108 1109 if (!mld_v1enable) { 1110 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", 1111 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1112 ifp, ifp->if_xname); 1113 return (0); 1114 } 1115 1116 if (ifp->if_flags & IFF_LOOPBACK) 1117 return (0); 1118 1119 /* 1120 * MLDv1 reports must originate from a host's link-local address, 1121 * or the unspecified address (when booting). 1122 */ 1123 src = ip6->ip6_src; 1124 in6_clearscope(&src); 1125 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { 1126 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 1127 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 1128 ifp, ifp->if_xname); 1129 return (EINVAL); 1130 } 1131 1132 /* 1133 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast 1134 * group, and must be directed to the group itself. 1135 */ 1136 dst = ip6->ip6_dst; 1137 in6_clearscope(&dst); 1138 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || 1139 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { 1140 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", 1141 ip6_sprintf(ip6tbuf, &ip6->ip6_dst), 1142 ifp, ifp->if_xname); 1143 return (EINVAL); 1144 } 1145 1146 /* 1147 * Make sure we don't hear our own membership report, as fast 1148 * leave requires knowing that we are the only member of a 1149 * group. Assume we used the link-local address if available, 1150 * otherwise look for ::. 1151 * 1152 * XXX Note that scope ID comparison is needed for the address 1153 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be 1154 * performed for the on-wire address. 1155 */ 1156 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1157 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || 1158 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) { 1159 if (ia != NULL) 1160 ifa_free(&ia->ia_ifa); 1161 return (0); 1162 } 1163 if (ia != NULL) 1164 ifa_free(&ia->ia_ifa); 1165 1166 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", 1167 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname); 1168 1169 /* 1170 * Embed scope ID of receiving interface in MLD query for lookup 1171 * whilst we don't hold other locks (due to KAME locking lameness). 1172 */ 1173 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) 1174 in6_setscope(&mld->mld_addr, ifp, NULL); 1175 1176 IN6_MULTI_LOCK(); 1177 MLD_LOCK(); 1178 IF_ADDR_RLOCK(ifp); 1179 1180 /* 1181 * MLDv1 report suppression. 1182 * If we are a member of this group, and our membership should be 1183 * reported, and our group timer is pending or about to be reset, 1184 * stop our group timer by transitioning to the 'lazy' state. 1185 */ 1186 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 1187 if (inm != NULL) { 1188 struct mld_ifinfo *mli; 1189 1190 mli = inm->in6m_mli; 1191 KASSERT(mli != NULL, 1192 ("%s: no mli for ifp %p", __func__, ifp)); 1193 1194 /* 1195 * If we are in MLDv2 host mode, do not allow the 1196 * other host's MLDv1 report to suppress our reports. 1197 */ 1198 if (mli->mli_version == MLD_VERSION_2) 1199 goto out_locked; 1200 1201 inm->in6m_timer = 0; 1202 1203 switch (inm->in6m_state) { 1204 case MLD_NOT_MEMBER: 1205 case MLD_SILENT_MEMBER: 1206 case MLD_SLEEPING_MEMBER: 1207 break; 1208 case MLD_REPORTING_MEMBER: 1209 case MLD_IDLE_MEMBER: 1210 case MLD_AWAKENING_MEMBER: 1211 CTR3(KTR_MLD, 1212 "report suppressed for %s on ifp %p(%s)", 1213 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1214 ifp, ifp->if_xname); 1215 case MLD_LAZY_MEMBER: 1216 inm->in6m_state = MLD_LAZY_MEMBER; 1217 break; 1218 case MLD_G_QUERY_PENDING_MEMBER: 1219 case MLD_SG_QUERY_PENDING_MEMBER: 1220 case MLD_LEAVING_MEMBER: 1221 break; 1222 } 1223 } 1224 1225out_locked: 1226 IF_ADDR_RUNLOCK(ifp); 1227 MLD_UNLOCK(); 1228 IN6_MULTI_UNLOCK(); 1229 1230 /* XXX Clear embedded scope ID as userland won't expect it. */ 1231 in6_clearscope(&mld->mld_addr); 1232 1233 return (0); 1234} 1235 1236/* 1237 * MLD input path. 1238 * 1239 * Assume query messages which fit in a single ICMPv6 message header 1240 * have been pulled up. 1241 * Assume that userland will want to see the message, even if it 1242 * otherwise fails kernel input validation; do not free it. 1243 * Pullup may however free the mbuf chain m if it fails. 1244 * 1245 * Return IPPROTO_DONE if we freed m. Otherwise, return 0. 1246 */ 1247int 1248mld_input(struct mbuf *m, int off, int icmp6len) 1249{ 1250 struct ifnet *ifp; 1251 struct ip6_hdr *ip6; 1252 struct mld_hdr *mld; 1253 int mldlen; 1254 1255 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); 1256 1257 ifp = m->m_pkthdr.rcvif; 1258 1259 ip6 = mtod(m, struct ip6_hdr *); 1260 1261 /* Pullup to appropriate size. */ 1262 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); 1263 if (mld->mld_type == MLD_LISTENER_QUERY && 1264 icmp6len >= sizeof(struct mldv2_query)) { 1265 mldlen = sizeof(struct mldv2_query); 1266 } else { 1267 mldlen = sizeof(struct mld_hdr); 1268 } 1269 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen); 1270 if (mld == NULL) { 1271 ICMP6STAT_INC(icp6s_badlen); 1272 return (IPPROTO_DONE); 1273 } 1274 1275 /* 1276 * Userland needs to see all of this traffic for implementing 1277 * the endpoint discovery portion of multicast routing. 1278 */ 1279 switch (mld->mld_type) { 1280 case MLD_LISTENER_QUERY: 1281 icmp6_ifstat_inc(ifp, ifs6_in_mldquery); 1282 if (icmp6len == sizeof(struct mld_hdr)) { 1283 if (mld_v1_input_query(ifp, ip6, mld) != 0) 1284 return (0); 1285 } else if (icmp6len >= sizeof(struct mldv2_query)) { 1286 if (mld_v2_input_query(ifp, ip6, m, off, 1287 icmp6len) != 0) 1288 return (0); 1289 } 1290 break; 1291 case MLD_LISTENER_REPORT: 1292 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1293 if (mld_v1_input_report(ifp, ip6, mld) != 0) 1294 return (0); 1295 break; 1296 case MLDV2_LISTENER_REPORT: 1297 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1298 break; 1299 case MLD_LISTENER_DONE: 1300 icmp6_ifstat_inc(ifp, ifs6_in_mlddone); 1301 break; 1302 default: 1303 break; 1304 } 1305 1306 return (0); 1307} 1308 1309/* 1310 * Fast timeout handler (global). 1311 * VIMAGE: Timeout handlers are expected to service all vimages. 1312 */ 1313void 1314mld_fasttimo(void) 1315{ 1316 VNET_ITERATOR_DECL(vnet_iter); 1317 1318 VNET_LIST_RLOCK_NOSLEEP(); 1319 VNET_FOREACH(vnet_iter) { 1320 CURVNET_SET(vnet_iter); 1321 mld_fasttimo_vnet(); 1322 CURVNET_RESTORE(); 1323 } 1324 VNET_LIST_RUNLOCK_NOSLEEP(); 1325} 1326 1327/* 1328 * Fast timeout handler (per-vnet). 1329 * 1330 * VIMAGE: Assume caller has set up our curvnet. 1331 */ 1332static void 1333mld_fasttimo_vnet(void) 1334{ 1335 struct ifqueue scq; /* State-change packets */ 1336 struct ifqueue qrq; /* Query response packets */ 1337 struct ifnet *ifp; 1338 struct mld_ifinfo *mli; 1339 struct ifmultiaddr *ifma; 1340 struct in6_multi *inm, *tinm; 1341 int uri_fasthz; 1342 1343 uri_fasthz = 0; 1344 1345 /* 1346 * Quick check to see if any work needs to be done, in order to 1347 * minimize the overhead of fasttimo processing. 1348 * SMPng: XXX Unlocked reads. 1349 */ 1350 if (!V_current_state_timers_running6 && 1351 !V_interface_timers_running6 && 1352 !V_state_change_timers_running6) 1353 return; 1354 1355 IN6_MULTI_LOCK(); 1356 MLD_LOCK(); 1357 1358 /* 1359 * MLDv2 General Query response timer processing. 1360 */ 1361 if (V_interface_timers_running6) { 1362 CTR1(KTR_MLD, "%s: interface timers running", __func__); 1363 1364 V_interface_timers_running6 = 0; 1365 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1366 if (mli->mli_v2_timer == 0) { 1367 /* Do nothing. */ 1368 } else if (--mli->mli_v2_timer == 0) { 1369 mld_v2_dispatch_general_query(mli); 1370 } else { 1371 V_interface_timers_running6 = 1; 1372 } 1373 } 1374 } 1375 1376 if (!V_current_state_timers_running6 && 1377 !V_state_change_timers_running6) 1378 goto out_locked; 1379 1380 V_current_state_timers_running6 = 0; 1381 V_state_change_timers_running6 = 0; 1382 1383 CTR1(KTR_MLD, "%s: state change timers running", __func__); 1384 1385 /* 1386 * MLD host report and state-change timer processing. 1387 * Note: Processing a v2 group timer may remove a node. 1388 */ 1389 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1390 ifp = mli->mli_ifp; 1391 1392 if (mli->mli_version == MLD_VERSION_2) { 1393 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * 1394 PR_FASTHZ); 1395 1396 memset(&qrq, 0, sizeof(struct ifqueue)); 1397 IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS); 1398 1399 memset(&scq, 0, sizeof(struct ifqueue)); 1400 IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS); 1401 } 1402 1403 IF_ADDR_RLOCK(ifp); 1404 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1405 if (ifma->ifma_addr->sa_family != AF_INET6 || 1406 ifma->ifma_protospec == NULL) 1407 continue; 1408 inm = (struct in6_multi *)ifma->ifma_protospec; 1409 switch (mli->mli_version) { 1410 case MLD_VERSION_1: 1411 mld_v1_process_group_timer(mli, inm); 1412 break; 1413 case MLD_VERSION_2: 1414 mld_v2_process_group_timers(mli, &qrq, 1415 &scq, inm, uri_fasthz); 1416 break; 1417 } 1418 } 1419 IF_ADDR_RUNLOCK(ifp); 1420 1421 switch (mli->mli_version) { 1422 case MLD_VERSION_1: 1423 /* 1424 * Transmit reports for this lifecycle. This 1425 * is done while not holding IF_ADDR_LOCK 1426 * since this can call 1427 * in6ifa_ifpforlinklocal() which locks 1428 * IF_ADDR_LOCK internally as well as 1429 * ip6_output() to transmit a packet. 1430 */ 1431 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1432 in6m_nrele, tinm) { 1433 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1434 in6m_nrele); 1435 (void)mld_v1_transmit_report(inm, 1436 MLD_LISTENER_REPORT); 1437 } 1438 break; 1439 case MLD_VERSION_2: 1440 mld_dispatch_queue(&qrq, 0); 1441 mld_dispatch_queue(&scq, 0); 1442 1443 /* 1444 * Free the in_multi reference(s) for 1445 * this lifecycle. 1446 */ 1447 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1448 in6m_nrele, tinm) { 1449 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1450 in6m_nrele); 1451 in6m_release_locked(inm); 1452 } 1453 break; 1454 } 1455 } 1456 1457out_locked: 1458 MLD_UNLOCK(); 1459 IN6_MULTI_UNLOCK(); 1460} 1461 1462/* 1463 * Update host report group timer. 1464 * Will update the global pending timer flags. 1465 */ 1466static void 1467mld_v1_process_group_timer(struct mld_ifinfo *mli, struct in6_multi *inm) 1468{ 1469 int report_timer_expired; 1470 1471 IN6_MULTI_LOCK_ASSERT(); 1472 MLD_LOCK_ASSERT(); 1473 1474 if (inm->in6m_timer == 0) { 1475 report_timer_expired = 0; 1476 } else if (--inm->in6m_timer == 0) { 1477 report_timer_expired = 1; 1478 } else { 1479 V_current_state_timers_running6 = 1; 1480 return; 1481 } 1482 1483 switch (inm->in6m_state) { 1484 case MLD_NOT_MEMBER: 1485 case MLD_SILENT_MEMBER: 1486 case MLD_IDLE_MEMBER: 1487 case MLD_LAZY_MEMBER: 1488 case MLD_SLEEPING_MEMBER: 1489 case MLD_AWAKENING_MEMBER: 1490 break; 1491 case MLD_REPORTING_MEMBER: 1492 if (report_timer_expired) { 1493 inm->in6m_state = MLD_IDLE_MEMBER; 1494 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1495 in6m_nrele); 1496 } 1497 break; 1498 case MLD_G_QUERY_PENDING_MEMBER: 1499 case MLD_SG_QUERY_PENDING_MEMBER: 1500 case MLD_LEAVING_MEMBER: 1501 break; 1502 } 1503} 1504 1505/* 1506 * Update a group's timers for MLDv2. 1507 * Will update the global pending timer flags. 1508 * Note: Unlocked read from mli. 1509 */ 1510static void 1511mld_v2_process_group_timers(struct mld_ifinfo *mli, 1512 struct ifqueue *qrq, struct ifqueue *scq, 1513 struct in6_multi *inm, const int uri_fasthz) 1514{ 1515 int query_response_timer_expired; 1516 int state_change_retransmit_timer_expired; 1517#ifdef KTR 1518 char ip6tbuf[INET6_ADDRSTRLEN]; 1519#endif 1520 1521 IN6_MULTI_LOCK_ASSERT(); 1522 MLD_LOCK_ASSERT(); 1523 1524 query_response_timer_expired = 0; 1525 state_change_retransmit_timer_expired = 0; 1526 1527 /* 1528 * During a transition from compatibility mode back to MLDv2, 1529 * a group record in REPORTING state may still have its group 1530 * timer active. This is a no-op in this function; it is easier 1531 * to deal with it here than to complicate the slow-timeout path. 1532 */ 1533 if (inm->in6m_timer == 0) { 1534 query_response_timer_expired = 0; 1535 } else if (--inm->in6m_timer == 0) { 1536 query_response_timer_expired = 1; 1537 } else { 1538 V_current_state_timers_running6 = 1; 1539 } 1540 1541 if (inm->in6m_sctimer == 0) { 1542 state_change_retransmit_timer_expired = 0; 1543 } else if (--inm->in6m_sctimer == 0) { 1544 state_change_retransmit_timer_expired = 1; 1545 } else { 1546 V_state_change_timers_running6 = 1; 1547 } 1548 1549 /* We are in fasttimo, so be quick about it. */ 1550 if (!state_change_retransmit_timer_expired && 1551 !query_response_timer_expired) 1552 return; 1553 1554 switch (inm->in6m_state) { 1555 case MLD_NOT_MEMBER: 1556 case MLD_SILENT_MEMBER: 1557 case MLD_SLEEPING_MEMBER: 1558 case MLD_LAZY_MEMBER: 1559 case MLD_AWAKENING_MEMBER: 1560 case MLD_IDLE_MEMBER: 1561 break; 1562 case MLD_G_QUERY_PENDING_MEMBER: 1563 case MLD_SG_QUERY_PENDING_MEMBER: 1564 /* 1565 * Respond to a previously pending Group-Specific 1566 * or Group-and-Source-Specific query by enqueueing 1567 * the appropriate Current-State report for 1568 * immediate transmission. 1569 */ 1570 if (query_response_timer_expired) { 1571 int retval; 1572 1573 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, 1574 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER), 1575 0); 1576 CTR2(KTR_MLD, "%s: enqueue record = %d", 1577 __func__, retval); 1578 inm->in6m_state = MLD_REPORTING_MEMBER; 1579 in6m_clear_recorded(inm); 1580 } 1581 /* FALLTHROUGH */ 1582 case MLD_REPORTING_MEMBER: 1583 case MLD_LEAVING_MEMBER: 1584 if (state_change_retransmit_timer_expired) { 1585 /* 1586 * State-change retransmission timer fired. 1587 * If there are any further pending retransmissions, 1588 * set the global pending state-change flag, and 1589 * reset the timer. 1590 */ 1591 if (--inm->in6m_scrv > 0) { 1592 inm->in6m_sctimer = uri_fasthz; 1593 V_state_change_timers_running6 = 1; 1594 } 1595 /* 1596 * Retransmit the previously computed state-change 1597 * report. If there are no further pending 1598 * retransmissions, the mbuf queue will be consumed. 1599 * Update T0 state to T1 as we have now sent 1600 * a state-change. 1601 */ 1602 (void)mld_v2_merge_state_changes(inm, scq); 1603 1604 in6m_commit(inm); 1605 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 1606 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1607 inm->in6m_ifp->if_xname); 1608 1609 /* 1610 * If we are leaving the group for good, make sure 1611 * we release MLD's reference to it. 1612 * This release must be deferred using a SLIST, 1613 * as we are called from a loop which traverses 1614 * the in_ifmultiaddr TAILQ. 1615 */ 1616 if (inm->in6m_state == MLD_LEAVING_MEMBER && 1617 inm->in6m_scrv == 0) { 1618 inm->in6m_state = MLD_NOT_MEMBER; 1619 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 1620 inm, in6m_nrele); 1621 } 1622 } 1623 break; 1624 } 1625} 1626 1627/* 1628 * Switch to a different version on the given interface, 1629 * as per Section 9.12. 1630 */ 1631static void 1632mld_set_version(struct mld_ifinfo *mli, const int version) 1633{ 1634 int old_version_timer; 1635 1636 MLD_LOCK_ASSERT(); 1637 1638 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, 1639 version, mli->mli_ifp, mli->mli_ifp->if_xname); 1640 1641 if (version == MLD_VERSION_1) { 1642 /* 1643 * Compute the "Older Version Querier Present" timer as per 1644 * Section 9.12. 1645 */ 1646 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; 1647 old_version_timer *= PR_SLOWHZ; 1648 mli->mli_v1_timer = old_version_timer; 1649 } 1650 1651 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { 1652 mli->mli_version = MLD_VERSION_1; 1653 mld_v2_cancel_link_timers(mli); 1654 } 1655} 1656 1657/* 1658 * Cancel pending MLDv2 timers for the given link and all groups 1659 * joined on it; state-change, general-query, and group-query timers. 1660 */ 1661static void 1662mld_v2_cancel_link_timers(struct mld_ifinfo *mli) 1663{ 1664 struct ifmultiaddr *ifma; 1665 struct ifnet *ifp; 1666 struct in6_multi *inm, *tinm; 1667 1668 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, 1669 mli->mli_ifp, mli->mli_ifp->if_xname); 1670 1671 IN6_MULTI_LOCK_ASSERT(); 1672 MLD_LOCK_ASSERT(); 1673 1674 /* 1675 * Fast-track this potentially expensive operation 1676 * by checking all the global 'timer pending' flags. 1677 */ 1678 if (!V_interface_timers_running6 && 1679 !V_state_change_timers_running6 && 1680 !V_current_state_timers_running6) 1681 return; 1682 1683 mli->mli_v2_timer = 0; 1684 1685 ifp = mli->mli_ifp; 1686 1687 IF_ADDR_RLOCK(ifp); 1688 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1689 if (ifma->ifma_addr->sa_family != AF_INET6) 1690 continue; 1691 inm = (struct in6_multi *)ifma->ifma_protospec; 1692 switch (inm->in6m_state) { 1693 case MLD_NOT_MEMBER: 1694 case MLD_SILENT_MEMBER: 1695 case MLD_IDLE_MEMBER: 1696 case MLD_LAZY_MEMBER: 1697 case MLD_SLEEPING_MEMBER: 1698 case MLD_AWAKENING_MEMBER: 1699 break; 1700 case MLD_LEAVING_MEMBER: 1701 /* 1702 * If we are leaving the group and switching 1703 * version, we need to release the final 1704 * reference held for issuing the INCLUDE {}. 1705 */ 1706 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1707 in6m_nrele); 1708 /* FALLTHROUGH */ 1709 case MLD_G_QUERY_PENDING_MEMBER: 1710 case MLD_SG_QUERY_PENDING_MEMBER: 1711 in6m_clear_recorded(inm); 1712 /* FALLTHROUGH */ 1713 case MLD_REPORTING_MEMBER: 1714 inm->in6m_sctimer = 0; 1715 inm->in6m_timer = 0; 1716 inm->in6m_state = MLD_REPORTING_MEMBER; 1717 /* 1718 * Free any pending MLDv2 state-change records. 1719 */ 1720 _IF_DRAIN(&inm->in6m_scq); 1721 break; 1722 } 1723 } 1724 IF_ADDR_RUNLOCK(ifp); 1725 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, tinm) { 1726 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 1727 in6m_release_locked(inm); 1728 } 1729} 1730 1731/* 1732 * Global slowtimo handler. 1733 * VIMAGE: Timeout handlers are expected to service all vimages. 1734 */ 1735void 1736mld_slowtimo(void) 1737{ 1738 VNET_ITERATOR_DECL(vnet_iter); 1739 1740 VNET_LIST_RLOCK_NOSLEEP(); 1741 VNET_FOREACH(vnet_iter) { 1742 CURVNET_SET(vnet_iter); 1743 mld_slowtimo_vnet(); 1744 CURVNET_RESTORE(); 1745 } 1746 VNET_LIST_RUNLOCK_NOSLEEP(); 1747} 1748 1749/* 1750 * Per-vnet slowtimo handler. 1751 */ 1752static void 1753mld_slowtimo_vnet(void) 1754{ 1755 struct mld_ifinfo *mli; 1756 1757 MLD_LOCK(); 1758 1759 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1760 mld_v1_process_querier_timers(mli); 1761 } 1762 1763 MLD_UNLOCK(); 1764} 1765 1766/* 1767 * Update the Older Version Querier Present timers for a link. 1768 * See Section 9.12 of RFC 3810. 1769 */ 1770static void 1771mld_v1_process_querier_timers(struct mld_ifinfo *mli) 1772{ 1773 1774 MLD_LOCK_ASSERT(); 1775 1776 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) { 1777 /* 1778 * MLDv1 Querier Present timer expired; revert to MLDv2. 1779 */ 1780 CTR5(KTR_MLD, 1781 "%s: transition from v%d -> v%d on %p(%s)", 1782 __func__, mli->mli_version, MLD_VERSION_2, 1783 mli->mli_ifp, mli->mli_ifp->if_xname); 1784 mli->mli_version = MLD_VERSION_2; 1785 } 1786} 1787 1788/* 1789 * Transmit an MLDv1 report immediately. 1790 */ 1791static int 1792mld_v1_transmit_report(struct in6_multi *in6m, const int type) 1793{ 1794 struct ifnet *ifp; 1795 struct in6_ifaddr *ia; 1796 struct ip6_hdr *ip6; 1797 struct mbuf *mh, *md; 1798 struct mld_hdr *mld; 1799 1800 IN6_MULTI_LOCK_ASSERT(); 1801 MLD_LOCK_ASSERT(); 1802 1803 ifp = in6m->in6m_ifp; 1804 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1805 /* ia may be NULL if link-local address is tentative. */ 1806 1807 MGETHDR(mh, M_DONTWAIT, MT_HEADER); 1808 if (mh == NULL) { 1809 if (ia != NULL) 1810 ifa_free(&ia->ia_ifa); 1811 return (ENOMEM); 1812 } 1813 MGET(md, M_DONTWAIT, MT_DATA); 1814 if (md == NULL) { 1815 m_free(mh); 1816 if (ia != NULL) 1817 ifa_free(&ia->ia_ifa); 1818 return (ENOMEM); 1819 } 1820 mh->m_next = md; 1821 1822 /* 1823 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so 1824 * that ether_output() does not need to allocate another mbuf 1825 * for the header in the most common case. 1826 */ 1827 MH_ALIGN(mh, sizeof(struct ip6_hdr)); 1828 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); 1829 mh->m_len = sizeof(struct ip6_hdr); 1830 1831 ip6 = mtod(mh, struct ip6_hdr *); 1832 ip6->ip6_flow = 0; 1833 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 1834 ip6->ip6_vfc |= IPV6_VERSION; 1835 ip6->ip6_nxt = IPPROTO_ICMPV6; 1836 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 1837 ip6->ip6_dst = in6m->in6m_addr; 1838 1839 md->m_len = sizeof(struct mld_hdr); 1840 mld = mtod(md, struct mld_hdr *); 1841 mld->mld_type = type; 1842 mld->mld_code = 0; 1843 mld->mld_cksum = 0; 1844 mld->mld_maxdelay = 0; 1845 mld->mld_reserved = 0; 1846 mld->mld_addr = in6m->in6m_addr; 1847 in6_clearscope(&mld->mld_addr); 1848 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 1849 sizeof(struct ip6_hdr), sizeof(struct mld_hdr)); 1850 1851 mld_save_context(mh, ifp); 1852 mh->m_flags |= M_MLDV1; 1853 1854 mld_dispatch_packet(mh); 1855 1856 if (ia != NULL) 1857 ifa_free(&ia->ia_ifa); 1858 return (0); 1859} 1860 1861/* 1862 * Process a state change from the upper layer for the given IPv6 group. 1863 * 1864 * Each socket holds a reference on the in_multi in its own ip_moptions. 1865 * The socket layer will have made the necessary updates to.the group 1866 * state, it is now up to MLD to issue a state change report if there 1867 * has been any change between T0 (when the last state-change was issued) 1868 * and T1 (now). 1869 * 1870 * We use the MLDv2 state machine at group level. The MLd module 1871 * however makes the decision as to which MLD protocol version to speak. 1872 * A state change *from* INCLUDE {} always means an initial join. 1873 * A state change *to* INCLUDE {} always means a final leave. 1874 * 1875 * If delay is non-zero, and the state change is an initial multicast 1876 * join, the state change report will be delayed by 'delay' ticks 1877 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise 1878 * the initial MLDv2 state change report will be delayed by whichever 1879 * is sooner, a pending state-change timer or delay itself. 1880 * 1881 * VIMAGE: curvnet should have been set by caller, as this routine 1882 * is called from the socket option handlers. 1883 */ 1884int 1885mld_change_state(struct in6_multi *inm, const int delay) 1886{ 1887 struct mld_ifinfo *mli; 1888 struct ifnet *ifp; 1889 int error; 1890 1891 IN6_MULTI_LOCK_ASSERT(); 1892 1893 error = 0; 1894 1895 /* 1896 * Try to detect if the upper layer just asked us to change state 1897 * for an interface which has now gone away. 1898 */ 1899 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 1900 ifp = inm->in6m_ifma->ifma_ifp; 1901 if (ifp != NULL) { 1902 /* 1903 * Sanity check that netinet6's notion of ifp is the 1904 * same as net's. 1905 */ 1906 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 1907 } 1908 1909 MLD_LOCK(); 1910 1911 mli = MLD_IFINFO(ifp); 1912 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 1913 1914 /* 1915 * If we detect a state transition to or from MCAST_UNDEFINED 1916 * for this group, then we are starting or finishing an MLD 1917 * life cycle for this group. 1918 */ 1919 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { 1920 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, 1921 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); 1922 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { 1923 CTR1(KTR_MLD, "%s: initial join", __func__); 1924 error = mld_initial_join(inm, mli, delay); 1925 goto out_locked; 1926 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { 1927 CTR1(KTR_MLD, "%s: final leave", __func__); 1928 mld_final_leave(inm, mli); 1929 goto out_locked; 1930 } 1931 } else { 1932 CTR1(KTR_MLD, "%s: filter set change", __func__); 1933 } 1934 1935 error = mld_handle_state_change(inm, mli); 1936 1937out_locked: 1938 MLD_UNLOCK(); 1939 return (error); 1940} 1941 1942/* 1943 * Perform the initial join for an MLD group. 1944 * 1945 * When joining a group: 1946 * If the group should have its MLD traffic suppressed, do nothing. 1947 * MLDv1 starts sending MLDv1 host membership reports. 1948 * MLDv2 will schedule an MLDv2 state-change report containing the 1949 * initial state of the membership. 1950 * 1951 * If the delay argument is non-zero, then we must delay sending the 1952 * initial state change for delay ticks (in units of PR_FASTHZ). 1953 */ 1954static int 1955mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli, 1956 const int delay) 1957{ 1958 struct ifnet *ifp; 1959 struct ifqueue *ifq; 1960 int error, retval, syncstates; 1961 int odelay; 1962#ifdef KTR 1963 char ip6tbuf[INET6_ADDRSTRLEN]; 1964#endif 1965 1966 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", 1967 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1968 inm->in6m_ifp, inm->in6m_ifp->if_xname); 1969 1970 error = 0; 1971 syncstates = 1; 1972 1973 ifp = inm->in6m_ifp; 1974 1975 IN6_MULTI_LOCK_ASSERT(); 1976 MLD_LOCK_ASSERT(); 1977 1978 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); 1979 1980 /* 1981 * Groups joined on loopback or marked as 'not reported', 1982 * enter the MLD_SILENT_MEMBER state and 1983 * are never reported in any protocol exchanges. 1984 * All other groups enter the appropriate state machine 1985 * for the version in use on this link. 1986 * A link marked as MLIF_SILENT causes MLD to be completely 1987 * disabled for the link. 1988 */ 1989 if ((ifp->if_flags & IFF_LOOPBACK) || 1990 (mli->mli_flags & MLIF_SILENT) || 1991 !mld_is_addr_reported(&inm->in6m_addr)) { 1992 CTR1(KTR_MLD, 1993"%s: not kicking state machine for silent group", __func__); 1994 inm->in6m_state = MLD_SILENT_MEMBER; 1995 inm->in6m_timer = 0; 1996 } else { 1997 /* 1998 * Deal with overlapping in_multi lifecycle. 1999 * If this group was LEAVING, then make sure 2000 * we drop the reference we picked up to keep the 2001 * group around for the final INCLUDE {} enqueue. 2002 */ 2003 if (mli->mli_version == MLD_VERSION_2 && 2004 inm->in6m_state == MLD_LEAVING_MEMBER) 2005 in6m_release_locked(inm); 2006 2007 inm->in6m_state = MLD_REPORTING_MEMBER; 2008 2009 switch (mli->mli_version) { 2010 case MLD_VERSION_1: 2011 /* 2012 * If a delay was provided, only use it if 2013 * it is greater than the delay normally 2014 * used for an MLDv1 state change report, 2015 * and delay sending the initial MLDv1 report 2016 * by not transitioning to the IDLE state. 2017 */ 2018 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); 2019 if (delay) { 2020 inm->in6m_timer = max(delay, odelay); 2021 V_current_state_timers_running6 = 1; 2022 } else { 2023 inm->in6m_state = MLD_IDLE_MEMBER; 2024 error = mld_v1_transmit_report(inm, 2025 MLD_LISTENER_REPORT); 2026 if (error == 0) { 2027 inm->in6m_timer = odelay; 2028 V_current_state_timers_running6 = 1; 2029 } 2030 } 2031 break; 2032 2033 case MLD_VERSION_2: 2034 /* 2035 * Defer update of T0 to T1, until the first copy 2036 * of the state change has been transmitted. 2037 */ 2038 syncstates = 0; 2039 2040 /* 2041 * Immediately enqueue a State-Change Report for 2042 * this interface, freeing any previous reports. 2043 * Don't kick the timers if there is nothing to do, 2044 * or if an error occurred. 2045 */ 2046 ifq = &inm->in6m_scq; 2047 _IF_DRAIN(ifq); 2048 retval = mld_v2_enqueue_group_record(ifq, inm, 1, 2049 0, 0, (mli->mli_flags & MLIF_USEALLOW)); 2050 CTR2(KTR_MLD, "%s: enqueue record = %d", 2051 __func__, retval); 2052 if (retval <= 0) { 2053 error = retval * -1; 2054 break; 2055 } 2056 2057 /* 2058 * Schedule transmission of pending state-change 2059 * report up to RV times for this link. The timer 2060 * will fire at the next mld_fasttimo (~200ms), 2061 * giving us an opportunity to merge the reports. 2062 * 2063 * If a delay was provided to this function, only 2064 * use this delay if sooner than the existing one. 2065 */ 2066 KASSERT(mli->mli_rv > 1, 2067 ("%s: invalid robustness %d", __func__, 2068 mli->mli_rv)); 2069 inm->in6m_scrv = mli->mli_rv; 2070 if (delay) { 2071 if (inm->in6m_sctimer > 1) { 2072 inm->in6m_sctimer = 2073 min(inm->in6m_sctimer, delay); 2074 } else 2075 inm->in6m_sctimer = delay; 2076 } else 2077 inm->in6m_sctimer = 1; 2078 V_state_change_timers_running6 = 1; 2079 2080 error = 0; 2081 break; 2082 } 2083 } 2084 2085 /* 2086 * Only update the T0 state if state change is atomic, 2087 * i.e. we don't need to wait for a timer to fire before we 2088 * can consider the state change to have been communicated. 2089 */ 2090 if (syncstates) { 2091 in6m_commit(inm); 2092 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2093 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2094 inm->in6m_ifp->if_xname); 2095 } 2096 2097 return (error); 2098} 2099 2100/* 2101 * Issue an intermediate state change during the life-cycle. 2102 */ 2103static int 2104mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli) 2105{ 2106 struct ifnet *ifp; 2107 int retval; 2108#ifdef KTR 2109 char ip6tbuf[INET6_ADDRSTRLEN]; 2110#endif 2111 2112 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)", 2113 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2114 inm->in6m_ifp, inm->in6m_ifp->if_xname); 2115 2116 ifp = inm->in6m_ifp; 2117 2118 IN6_MULTI_LOCK_ASSERT(); 2119 MLD_LOCK_ASSERT(); 2120 2121 KASSERT(mli && mli->mli_ifp == ifp, 2122 ("%s: inconsistent ifp", __func__)); 2123 2124 if ((ifp->if_flags & IFF_LOOPBACK) || 2125 (mli->mli_flags & MLIF_SILENT) || 2126 !mld_is_addr_reported(&inm->in6m_addr) || 2127 (mli->mli_version != MLD_VERSION_2)) { 2128 if (!mld_is_addr_reported(&inm->in6m_addr)) { 2129 CTR1(KTR_MLD, 2130"%s: not kicking state machine for silent group", __func__); 2131 } 2132 CTR1(KTR_MLD, "%s: nothing to do", __func__); 2133 in6m_commit(inm); 2134 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2135 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2136 inm->in6m_ifp->if_xname); 2137 return (0); 2138 } 2139 2140 _IF_DRAIN(&inm->in6m_scq); 2141 2142 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0, 2143 (mli->mli_flags & MLIF_USEALLOW)); 2144 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); 2145 if (retval <= 0) 2146 return (-retval); 2147 2148 /* 2149 * If record(s) were enqueued, start the state-change 2150 * report timer for this group. 2151 */ 2152 inm->in6m_scrv = mli->mli_rv; 2153 inm->in6m_sctimer = 1; 2154 V_state_change_timers_running6 = 1; 2155 2156 return (0); 2157} 2158 2159/* 2160 * Perform the final leave for a multicast address. 2161 * 2162 * When leaving a group: 2163 * MLDv1 sends a DONE message, if and only if we are the reporter. 2164 * MLDv2 enqueues a state-change report containing a transition 2165 * to INCLUDE {} for immediate transmission. 2166 */ 2167static void 2168mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli) 2169{ 2170 int syncstates; 2171#ifdef KTR 2172 char ip6tbuf[INET6_ADDRSTRLEN]; 2173#endif 2174 2175 syncstates = 1; 2176 2177 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)", 2178 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2179 inm->in6m_ifp, inm->in6m_ifp->if_xname); 2180 2181 IN6_MULTI_LOCK_ASSERT(); 2182 MLD_LOCK_ASSERT(); 2183 2184 switch (inm->in6m_state) { 2185 case MLD_NOT_MEMBER: 2186 case MLD_SILENT_MEMBER: 2187 case MLD_LEAVING_MEMBER: 2188 /* Already leaving or left; do nothing. */ 2189 CTR1(KTR_MLD, 2190"%s: not kicking state machine for silent group", __func__); 2191 break; 2192 case MLD_REPORTING_MEMBER: 2193 case MLD_IDLE_MEMBER: 2194 case MLD_G_QUERY_PENDING_MEMBER: 2195 case MLD_SG_QUERY_PENDING_MEMBER: 2196 if (mli->mli_version == MLD_VERSION_1) { 2197#ifdef INVARIANTS 2198 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 2199 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) 2200 panic("%s: MLDv2 state reached, not MLDv2 mode", 2201 __func__); 2202#endif 2203 mld_v1_transmit_report(inm, MLD_LISTENER_DONE); 2204 inm->in6m_state = MLD_NOT_MEMBER; 2205 } else if (mli->mli_version == MLD_VERSION_2) { 2206 /* 2207 * Stop group timer and all pending reports. 2208 * Immediately enqueue a state-change report 2209 * TO_IN {} to be sent on the next fast timeout, 2210 * giving us an opportunity to merge reports. 2211 */ 2212 _IF_DRAIN(&inm->in6m_scq); 2213 inm->in6m_timer = 0; 2214 inm->in6m_scrv = mli->mli_rv; 2215 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " 2216 "pending retransmissions.", __func__, 2217 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2218 inm->in6m_ifp->if_xname, inm->in6m_scrv); 2219 if (inm->in6m_scrv == 0) { 2220 inm->in6m_state = MLD_NOT_MEMBER; 2221 inm->in6m_sctimer = 0; 2222 } else { 2223 int retval; 2224 2225 in6m_acquire_locked(inm); 2226 2227 retval = mld_v2_enqueue_group_record( 2228 &inm->in6m_scq, inm, 1, 0, 0, 2229 (mli->mli_flags & MLIF_USEALLOW)); 2230 KASSERT(retval != 0, 2231 ("%s: enqueue record = %d", __func__, 2232 retval)); 2233 2234 inm->in6m_state = MLD_LEAVING_MEMBER; 2235 inm->in6m_sctimer = 1; 2236 V_state_change_timers_running6 = 1; 2237 syncstates = 0; 2238 } 2239 break; 2240 } 2241 break; 2242 case MLD_LAZY_MEMBER: 2243 case MLD_SLEEPING_MEMBER: 2244 case MLD_AWAKENING_MEMBER: 2245 /* Our reports are suppressed; do nothing. */ 2246 break; 2247 } 2248 2249 if (syncstates) { 2250 in6m_commit(inm); 2251 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2252 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2253 inm->in6m_ifp->if_xname); 2254 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 2255 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", 2256 __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname); 2257 } 2258} 2259 2260/* 2261 * Enqueue an MLDv2 group record to the given output queue. 2262 * 2263 * If is_state_change is zero, a current-state record is appended. 2264 * If is_state_change is non-zero, a state-change report is appended. 2265 * 2266 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2267 * If is_group_query is zero, and if there is a packet with free space 2268 * at the tail of the queue, it will be appended to providing there 2269 * is enough free space. 2270 * Otherwise a new mbuf packet chain is allocated. 2271 * 2272 * If is_source_query is non-zero, each source is checked to see if 2273 * it was recorded for a Group-Source query, and will be omitted if 2274 * it is not both in-mode and recorded. 2275 * 2276 * If use_block_allow is non-zero, state change reports for initial join 2277 * and final leave, on an inclusive mode group with a source list, will be 2278 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively. 2279 * 2280 * The function will attempt to allocate leading space in the packet 2281 * for the IPv6+ICMP headers to be prepended without fragmenting the chain. 2282 * 2283 * If successful the size of all data appended to the queue is returned, 2284 * otherwise an error code less than zero is returned, or zero if 2285 * no record(s) were appended. 2286 */ 2287static int 2288mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm, 2289 const int is_state_change, const int is_group_query, 2290 const int is_source_query, const int use_block_allow) 2291{ 2292 struct mldv2_record mr; 2293 struct mldv2_record *pmr; 2294 struct ifnet *ifp; 2295 struct ip6_msource *ims, *nims; 2296 struct mbuf *m0, *m, *md; 2297 int error, is_filter_list_change; 2298 int minrec0len, m0srcs, msrcs, nbytes, off; 2299 int record_has_sources; 2300 int now; 2301 int type; 2302 uint8_t mode; 2303#ifdef KTR 2304 char ip6tbuf[INET6_ADDRSTRLEN]; 2305#endif 2306 2307 IN6_MULTI_LOCK_ASSERT(); 2308 2309 error = 0; 2310 ifp = inm->in6m_ifp; 2311 is_filter_list_change = 0; 2312 m = NULL; 2313 m0 = NULL; 2314 m0srcs = 0; 2315 msrcs = 0; 2316 nbytes = 0; 2317 nims = NULL; 2318 record_has_sources = 1; 2319 pmr = NULL; 2320 type = MLD_DO_NOTHING; 2321 mode = inm->in6m_st[1].iss_fmode; 2322 2323 /* 2324 * If we did not transition out of ASM mode during t0->t1, 2325 * and there are no source nodes to process, we can skip 2326 * the generation of source records. 2327 */ 2328 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && 2329 inm->in6m_nsrc == 0) 2330 record_has_sources = 0; 2331 2332 if (is_state_change) { 2333 /* 2334 * Queue a state change record. 2335 * If the mode did not change, and there are non-ASM 2336 * listeners or source filters present, 2337 * we potentially need to issue two records for the group. 2338 * If there are ASM listeners, and there was no filter 2339 * mode transition of any kind, do nothing. 2340 * 2341 * If we are transitioning to MCAST_UNDEFINED, we need 2342 * not send any sources. A transition to/from this state is 2343 * considered inclusive with some special treatment. 2344 * 2345 * If we are rewriting initial joins/leaves to use 2346 * ALLOW/BLOCK, and the group's membership is inclusive, 2347 * we need to send sources in all cases. 2348 */ 2349 if (mode != inm->in6m_st[0].iss_fmode) { 2350 if (mode == MCAST_EXCLUDE) { 2351 CTR1(KTR_MLD, "%s: change to EXCLUDE", 2352 __func__); 2353 type = MLD_CHANGE_TO_EXCLUDE_MODE; 2354 } else { 2355 CTR1(KTR_MLD, "%s: change to INCLUDE", 2356 __func__); 2357 if (use_block_allow) { 2358 /* 2359 * XXX 2360 * Here we're interested in state 2361 * edges either direction between 2362 * MCAST_UNDEFINED and MCAST_INCLUDE. 2363 * Perhaps we should just check 2364 * the group state, rather than 2365 * the filter mode. 2366 */ 2367 if (mode == MCAST_UNDEFINED) { 2368 type = MLD_BLOCK_OLD_SOURCES; 2369 } else { 2370 type = MLD_ALLOW_NEW_SOURCES; 2371 } 2372 } else { 2373 type = MLD_CHANGE_TO_INCLUDE_MODE; 2374 if (mode == MCAST_UNDEFINED) 2375 record_has_sources = 0; 2376 } 2377 } 2378 } else { 2379 if (record_has_sources) { 2380 is_filter_list_change = 1; 2381 } else { 2382 type = MLD_DO_NOTHING; 2383 } 2384 } 2385 } else { 2386 /* 2387 * Queue a current state record. 2388 */ 2389 if (mode == MCAST_EXCLUDE) { 2390 type = MLD_MODE_IS_EXCLUDE; 2391 } else if (mode == MCAST_INCLUDE) { 2392 type = MLD_MODE_IS_INCLUDE; 2393 KASSERT(inm->in6m_st[1].iss_asm == 0, 2394 ("%s: inm %p is INCLUDE but ASM count is %d", 2395 __func__, inm, inm->in6m_st[1].iss_asm)); 2396 } 2397 } 2398 2399 /* 2400 * Generate the filter list changes using a separate function. 2401 */ 2402 if (is_filter_list_change) 2403 return (mld_v2_enqueue_filter_change(ifq, inm)); 2404 2405 if (type == MLD_DO_NOTHING) { 2406 CTR3(KTR_MLD, "%s: nothing to do for %s/%s", 2407 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2408 inm->in6m_ifp->if_xname); 2409 return (0); 2410 } 2411 2412 /* 2413 * If any sources are present, we must be able to fit at least 2414 * one in the trailing space of the tail packet's mbuf, 2415 * ideally more. 2416 */ 2417 minrec0len = sizeof(struct mldv2_record); 2418 if (record_has_sources) 2419 minrec0len += sizeof(struct in6_addr); 2420 2421 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, 2422 mld_rec_type_to_str(type), 2423 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2424 inm->in6m_ifp->if_xname); 2425 2426 /* 2427 * Check if we have a packet in the tail of the queue for this 2428 * group into which the first group record for this group will fit. 2429 * Otherwise allocate a new packet. 2430 * Always allocate leading space for IP6+RA+ICMPV6+REPORT. 2431 * Note: Group records for G/GSR query responses MUST be sent 2432 * in their own packet. 2433 */ 2434 m0 = ifq->ifq_tail; 2435 if (!is_group_query && 2436 m0 != NULL && 2437 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && 2438 (m0->m_pkthdr.len + minrec0len) < 2439 (ifp->if_mtu - MLD_MTUSPACE)) { 2440 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2441 sizeof(struct mldv2_record)) / 2442 sizeof(struct in6_addr); 2443 m = m0; 2444 CTR1(KTR_MLD, "%s: use existing packet", __func__); 2445 } else { 2446 if (_IF_QFULL(ifq)) { 2447 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2448 return (-ENOMEM); 2449 } 2450 m = NULL; 2451 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2452 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2453 if (!is_state_change && !is_group_query) 2454 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2455 if (m == NULL) 2456 m = m_gethdr(M_DONTWAIT, MT_DATA); 2457 if (m == NULL) 2458 return (-ENOMEM); 2459 2460 mld_save_context(m, ifp); 2461 2462 CTR1(KTR_MLD, "%s: allocated first packet", __func__); 2463 } 2464 2465 /* 2466 * Append group record. 2467 * If we have sources, we don't know how many yet. 2468 */ 2469 mr.mr_type = type; 2470 mr.mr_datalen = 0; 2471 mr.mr_numsrc = 0; 2472 mr.mr_addr = inm->in6m_addr; 2473 in6_clearscope(&mr.mr_addr); 2474 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2475 if (m != m0) 2476 m_freem(m); 2477 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2478 return (-ENOMEM); 2479 } 2480 nbytes += sizeof(struct mldv2_record); 2481 2482 /* 2483 * Append as many sources as will fit in the first packet. 2484 * If we are appending to a new packet, the chain allocation 2485 * may potentially use clusters; use m_getptr() in this case. 2486 * If we are appending to an existing packet, we need to obtain 2487 * a pointer to the group record after m_append(), in case a new 2488 * mbuf was allocated. 2489 * 2490 * Only append sources which are in-mode at t1. If we are 2491 * transitioning to MCAST_UNDEFINED state on the group, and 2492 * use_block_allow is zero, do not include source entries. 2493 * Otherwise, we need to include this source in the report. 2494 * 2495 * Only report recorded sources in our filter set when responding 2496 * to a group-source query. 2497 */ 2498 if (record_has_sources) { 2499 if (m == m0) { 2500 md = m_last(m); 2501 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2502 md->m_len - nbytes); 2503 } else { 2504 md = m_getptr(m, 0, &off); 2505 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2506 off); 2507 } 2508 msrcs = 0; 2509 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, 2510 nims) { 2511 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2512 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2513 now = im6s_get_mode(inm, ims, 1); 2514 CTR2(KTR_MLD, "%s: node is %d", __func__, now); 2515 if ((now != mode) || 2516 (now == mode && 2517 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2518 CTR1(KTR_MLD, "%s: skip node", __func__); 2519 continue; 2520 } 2521 if (is_source_query && ims->im6s_stp == 0) { 2522 CTR1(KTR_MLD, "%s: skip unrecorded node", 2523 __func__); 2524 continue; 2525 } 2526 CTR1(KTR_MLD, "%s: append node", __func__); 2527 if (!m_append(m, sizeof(struct in6_addr), 2528 (void *)&ims->im6s_addr)) { 2529 if (m != m0) 2530 m_freem(m); 2531 CTR1(KTR_MLD, "%s: m_append() failed.", 2532 __func__); 2533 return (-ENOMEM); 2534 } 2535 nbytes += sizeof(struct in6_addr); 2536 ++msrcs; 2537 if (msrcs == m0srcs) 2538 break; 2539 } 2540 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, 2541 msrcs); 2542 pmr->mr_numsrc = htons(msrcs); 2543 nbytes += (msrcs * sizeof(struct in6_addr)); 2544 } 2545 2546 if (is_source_query && msrcs == 0) { 2547 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); 2548 if (m != m0) 2549 m_freem(m); 2550 return (0); 2551 } 2552 2553 /* 2554 * We are good to go with first packet. 2555 */ 2556 if (m != m0) { 2557 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); 2558 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2559 _IF_ENQUEUE(ifq, m); 2560 } else 2561 m->m_pkthdr.PH_vt.vt_nrecs++; 2562 2563 /* 2564 * No further work needed if no source list in packet(s). 2565 */ 2566 if (!record_has_sources) 2567 return (nbytes); 2568 2569 /* 2570 * Whilst sources remain to be announced, we need to allocate 2571 * a new packet and fill out as many sources as will fit. 2572 * Always try for a cluster first. 2573 */ 2574 while (nims != NULL) { 2575 if (_IF_QFULL(ifq)) { 2576 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2577 return (-ENOMEM); 2578 } 2579 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2580 if (m == NULL) 2581 m = m_gethdr(M_DONTWAIT, MT_DATA); 2582 if (m == NULL) 2583 return (-ENOMEM); 2584 mld_save_context(m, ifp); 2585 md = m_getptr(m, 0, &off); 2586 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); 2587 CTR1(KTR_MLD, "%s: allocated next packet", __func__); 2588 2589 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2590 if (m != m0) 2591 m_freem(m); 2592 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2593 return (-ENOMEM); 2594 } 2595 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2596 nbytes += sizeof(struct mldv2_record); 2597 2598 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2599 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2600 2601 msrcs = 0; 2602 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2603 CTR2(KTR_MLD, "%s: visit node %s", 2604 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2605 now = im6s_get_mode(inm, ims, 1); 2606 if ((now != mode) || 2607 (now == mode && 2608 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2609 CTR1(KTR_MLD, "%s: skip node", __func__); 2610 continue; 2611 } 2612 if (is_source_query && ims->im6s_stp == 0) { 2613 CTR1(KTR_MLD, "%s: skip unrecorded node", 2614 __func__); 2615 continue; 2616 } 2617 CTR1(KTR_MLD, "%s: append node", __func__); 2618 if (!m_append(m, sizeof(struct in6_addr), 2619 (void *)&ims->im6s_addr)) { 2620 if (m != m0) 2621 m_freem(m); 2622 CTR1(KTR_MLD, "%s: m_append() failed.", 2623 __func__); 2624 return (-ENOMEM); 2625 } 2626 ++msrcs; 2627 if (msrcs == m0srcs) 2628 break; 2629 } 2630 pmr->mr_numsrc = htons(msrcs); 2631 nbytes += (msrcs * sizeof(struct in6_addr)); 2632 2633 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); 2634 _IF_ENQUEUE(ifq, m); 2635 } 2636 2637 return (nbytes); 2638} 2639 2640/* 2641 * Type used to mark record pass completion. 2642 * We exploit the fact we can cast to this easily from the 2643 * current filter modes on each ip_msource node. 2644 */ 2645typedef enum { 2646 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2647 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2648 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2649 REC_FULL = REC_ALLOW | REC_BLOCK 2650} rectype_t; 2651 2652/* 2653 * Enqueue an MLDv2 filter list change to the given output queue. 2654 * 2655 * Source list filter state is held in an RB-tree. When the filter list 2656 * for a group is changed without changing its mode, we need to compute 2657 * the deltas between T0 and T1 for each source in the filter set, 2658 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2659 * 2660 * As we may potentially queue two record types, and the entire R-B tree 2661 * needs to be walked at once, we break this out into its own function 2662 * so we can generate a tightly packed queue of packets. 2663 * 2664 * XXX This could be written to only use one tree walk, although that makes 2665 * serializing into the mbuf chains a bit harder. For now we do two walks 2666 * which makes things easier on us, and it may or may not be harder on 2667 * the L2 cache. 2668 * 2669 * If successful the size of all data appended to the queue is returned, 2670 * otherwise an error code less than zero is returned, or zero if 2671 * no record(s) were appended. 2672 */ 2673static int 2674mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm) 2675{ 2676 static const int MINRECLEN = 2677 sizeof(struct mldv2_record) + sizeof(struct in6_addr); 2678 struct ifnet *ifp; 2679 struct mldv2_record mr; 2680 struct mldv2_record *pmr; 2681 struct ip6_msource *ims, *nims; 2682 struct mbuf *m, *m0, *md; 2683 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 2684 int nallow, nblock; 2685 uint8_t mode, now, then; 2686 rectype_t crt, drt, nrt; 2687#ifdef KTR 2688 char ip6tbuf[INET6_ADDRSTRLEN]; 2689#endif 2690 2691 IN6_MULTI_LOCK_ASSERT(); 2692 2693 if (inm->in6m_nsrc == 0 || 2694 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) 2695 return (0); 2696 2697 ifp = inm->in6m_ifp; /* interface */ 2698 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ 2699 crt = REC_NONE; /* current group record type */ 2700 drt = REC_NONE; /* mask of completed group record types */ 2701 nrt = REC_NONE; /* record type for current node */ 2702 m0srcs = 0; /* # source which will fit in current mbuf chain */ 2703 npbytes = 0; /* # of bytes appended this packet */ 2704 nbytes = 0; /* # of bytes appended to group's state-change queue */ 2705 rsrcs = 0; /* # sources encoded in current record */ 2706 schanged = 0; /* # nodes encoded in overall filter change */ 2707 nallow = 0; /* # of source entries in ALLOW_NEW */ 2708 nblock = 0; /* # of source entries in BLOCK_OLD */ 2709 nims = NULL; /* next tree node pointer */ 2710 2711 /* 2712 * For each possible filter record mode. 2713 * The first kind of source we encounter tells us which 2714 * is the first kind of record we start appending. 2715 * If a node transitioned to UNDEFINED at t1, its mode is treated 2716 * as the inverse of the group's filter mode. 2717 */ 2718 while (drt != REC_FULL) { 2719 do { 2720 m0 = ifq->ifq_tail; 2721 if (m0 != NULL && 2722 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 2723 MLD_V2_REPORT_MAXRECS) && 2724 (m0->m_pkthdr.len + MINRECLEN) < 2725 (ifp->if_mtu - MLD_MTUSPACE)) { 2726 m = m0; 2727 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2728 sizeof(struct mldv2_record)) / 2729 sizeof(struct in6_addr); 2730 CTR1(KTR_MLD, 2731 "%s: use previous packet", __func__); 2732 } else { 2733 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2734 if (m == NULL) 2735 m = m_gethdr(M_DONTWAIT, MT_DATA); 2736 if (m == NULL) { 2737 CTR1(KTR_MLD, 2738 "%s: m_get*() failed", __func__); 2739 return (-ENOMEM); 2740 } 2741 m->m_pkthdr.PH_vt.vt_nrecs = 0; 2742 mld_save_context(m, ifp); 2743 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2744 sizeof(struct mldv2_record)) / 2745 sizeof(struct in6_addr); 2746 npbytes = 0; 2747 CTR1(KTR_MLD, 2748 "%s: allocated new packet", __func__); 2749 } 2750 /* 2751 * Append the MLD group record header to the 2752 * current packet's data area. 2753 * Recalculate pointer to free space for next 2754 * group record, in case m_append() allocated 2755 * a new mbuf or cluster. 2756 */ 2757 memset(&mr, 0, sizeof(mr)); 2758 mr.mr_addr = inm->in6m_addr; 2759 in6_clearscope(&mr.mr_addr); 2760 if (!m_append(m, sizeof(mr), (void *)&mr)) { 2761 if (m != m0) 2762 m_freem(m); 2763 CTR1(KTR_MLD, 2764 "%s: m_append() failed", __func__); 2765 return (-ENOMEM); 2766 } 2767 npbytes += sizeof(struct mldv2_record); 2768 if (m != m0) { 2769 /* new packet; offset in chain */ 2770 md = m_getptr(m, npbytes - 2771 sizeof(struct mldv2_record), &off); 2772 pmr = (struct mldv2_record *)(mtod(md, 2773 uint8_t *) + off); 2774 } else { 2775 /* current packet; offset from last append */ 2776 md = m_last(m); 2777 pmr = (struct mldv2_record *)(mtod(md, 2778 uint8_t *) + md->m_len - 2779 sizeof(struct mldv2_record)); 2780 } 2781 /* 2782 * Begin walking the tree for this record type 2783 * pass, or continue from where we left off 2784 * previously if we had to allocate a new packet. 2785 * Only report deltas in-mode at t1. 2786 * We need not report included sources as allowed 2787 * if we are in inclusive mode on the group, 2788 * however the converse is not true. 2789 */ 2790 rsrcs = 0; 2791 if (nims == NULL) { 2792 nims = RB_MIN(ip6_msource_tree, 2793 &inm->in6m_srcs); 2794 } 2795 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2796 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2797 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2798 now = im6s_get_mode(inm, ims, 1); 2799 then = im6s_get_mode(inm, ims, 0); 2800 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", 2801 __func__, then, now); 2802 if (now == then) { 2803 CTR1(KTR_MLD, 2804 "%s: skip unchanged", __func__); 2805 continue; 2806 } 2807 if (mode == MCAST_EXCLUDE && 2808 now == MCAST_INCLUDE) { 2809 CTR1(KTR_MLD, 2810 "%s: skip IN src on EX group", 2811 __func__); 2812 continue; 2813 } 2814 nrt = (rectype_t)now; 2815 if (nrt == REC_NONE) 2816 nrt = (rectype_t)(~mode & REC_FULL); 2817 if (schanged++ == 0) { 2818 crt = nrt; 2819 } else if (crt != nrt) 2820 continue; 2821 if (!m_append(m, sizeof(struct in6_addr), 2822 (void *)&ims->im6s_addr)) { 2823 if (m != m0) 2824 m_freem(m); 2825 CTR1(KTR_MLD, 2826 "%s: m_append() failed", __func__); 2827 return (-ENOMEM); 2828 } 2829 nallow += !!(crt == REC_ALLOW); 2830 nblock += !!(crt == REC_BLOCK); 2831 if (++rsrcs == m0srcs) 2832 break; 2833 } 2834 /* 2835 * If we did not append any tree nodes on this 2836 * pass, back out of allocations. 2837 */ 2838 if (rsrcs == 0) { 2839 npbytes -= sizeof(struct mldv2_record); 2840 if (m != m0) { 2841 CTR1(KTR_MLD, 2842 "%s: m_free(m)", __func__); 2843 m_freem(m); 2844 } else { 2845 CTR1(KTR_MLD, 2846 "%s: m_adj(m, -mr)", __func__); 2847 m_adj(m, -((int)sizeof( 2848 struct mldv2_record))); 2849 } 2850 continue; 2851 } 2852 npbytes += (rsrcs * sizeof(struct in6_addr)); 2853 if (crt == REC_ALLOW) 2854 pmr->mr_type = MLD_ALLOW_NEW_SOURCES; 2855 else if (crt == REC_BLOCK) 2856 pmr->mr_type = MLD_BLOCK_OLD_SOURCES; 2857 pmr->mr_numsrc = htons(rsrcs); 2858 /* 2859 * Count the new group record, and enqueue this 2860 * packet if it wasn't already queued. 2861 */ 2862 m->m_pkthdr.PH_vt.vt_nrecs++; 2863 if (m != m0) 2864 _IF_ENQUEUE(ifq, m); 2865 nbytes += npbytes; 2866 } while (nims != NULL); 2867 drt |= crt; 2868 crt = (~crt & REC_FULL); 2869 } 2870 2871 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 2872 nallow, nblock); 2873 2874 return (nbytes); 2875} 2876 2877static int 2878mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq) 2879{ 2880 struct ifqueue *gq; 2881 struct mbuf *m; /* pending state-change */ 2882 struct mbuf *m0; /* copy of pending state-change */ 2883 struct mbuf *mt; /* last state-change in packet */ 2884 int docopy, domerge; 2885 u_int recslen; 2886 2887 docopy = 0; 2888 domerge = 0; 2889 recslen = 0; 2890 2891 IN6_MULTI_LOCK_ASSERT(); 2892 MLD_LOCK_ASSERT(); 2893 2894 /* 2895 * If there are further pending retransmissions, make a writable 2896 * copy of each queued state-change message before merging. 2897 */ 2898 if (inm->in6m_scrv > 0) 2899 docopy = 1; 2900 2901 gq = &inm->in6m_scq; 2902#ifdef KTR 2903 if (gq->ifq_head == NULL) { 2904 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", 2905 __func__, inm); 2906 } 2907#endif 2908 2909 m = gq->ifq_head; 2910 while (m != NULL) { 2911 /* 2912 * Only merge the report into the current packet if 2913 * there is sufficient space to do so; an MLDv2 report 2914 * packet may only contain 65,535 group records. 2915 * Always use a simple mbuf chain concatentation to do this, 2916 * as large state changes for single groups may have 2917 * allocated clusters. 2918 */ 2919 domerge = 0; 2920 mt = ifscq->ifq_tail; 2921 if (mt != NULL) { 2922 recslen = m_length(m, NULL); 2923 2924 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 2925 m->m_pkthdr.PH_vt.vt_nrecs <= 2926 MLD_V2_REPORT_MAXRECS) && 2927 (mt->m_pkthdr.len + recslen <= 2928 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) 2929 domerge = 1; 2930 } 2931 2932 if (!domerge && _IF_QFULL(gq)) { 2933 CTR2(KTR_MLD, 2934 "%s: outbound queue full, skipping whole packet %p", 2935 __func__, m); 2936 mt = m->m_nextpkt; 2937 if (!docopy) 2938 m_freem(m); 2939 m = mt; 2940 continue; 2941 } 2942 2943 if (!docopy) { 2944 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); 2945 _IF_DEQUEUE(gq, m0); 2946 m = m0->m_nextpkt; 2947 } else { 2948 CTR2(KTR_MLD, "%s: copying %p", __func__, m); 2949 m0 = m_dup(m, M_NOWAIT); 2950 if (m0 == NULL) 2951 return (ENOMEM); 2952 m0->m_nextpkt = NULL; 2953 m = m->m_nextpkt; 2954 } 2955 2956 if (!domerge) { 2957 CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)", 2958 __func__, m0, ifscq); 2959 _IF_ENQUEUE(ifscq, m0); 2960 } else { 2961 struct mbuf *mtl; /* last mbuf of packet mt */ 2962 2963 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", 2964 __func__, m0, mt); 2965 2966 mtl = m_last(mt); 2967 m0->m_flags &= ~M_PKTHDR; 2968 mt->m_pkthdr.len += recslen; 2969 mt->m_pkthdr.PH_vt.vt_nrecs += 2970 m0->m_pkthdr.PH_vt.vt_nrecs; 2971 2972 mtl->m_next = m0; 2973 } 2974 } 2975 2976 return (0); 2977} 2978 2979/* 2980 * Respond to a pending MLDv2 General Query. 2981 */ 2982static void 2983mld_v2_dispatch_general_query(struct mld_ifinfo *mli) 2984{ 2985 struct ifmultiaddr *ifma; 2986 struct ifnet *ifp; 2987 struct in6_multi *inm; 2988 int retval; 2989 2990 IN6_MULTI_LOCK_ASSERT(); 2991 MLD_LOCK_ASSERT(); 2992 2993 KASSERT(mli->mli_version == MLD_VERSION_2, 2994 ("%s: called when version %d", __func__, mli->mli_version)); 2995 2996 ifp = mli->mli_ifp; 2997 2998 IF_ADDR_RLOCK(ifp); 2999 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3000 if (ifma->ifma_addr->sa_family != AF_INET6 || 3001 ifma->ifma_protospec == NULL) 3002 continue; 3003 3004 inm = (struct in6_multi *)ifma->ifma_protospec; 3005 KASSERT(ifp == inm->in6m_ifp, 3006 ("%s: inconsistent ifp", __func__)); 3007 3008 switch (inm->in6m_state) { 3009 case MLD_NOT_MEMBER: 3010 case MLD_SILENT_MEMBER: 3011 break; 3012 case MLD_REPORTING_MEMBER: 3013 case MLD_IDLE_MEMBER: 3014 case MLD_LAZY_MEMBER: 3015 case MLD_SLEEPING_MEMBER: 3016 case MLD_AWAKENING_MEMBER: 3017 inm->in6m_state = MLD_REPORTING_MEMBER; 3018 retval = mld_v2_enqueue_group_record(&mli->mli_gq, 3019 inm, 0, 0, 0, 0); 3020 CTR2(KTR_MLD, "%s: enqueue record = %d", 3021 __func__, retval); 3022 break; 3023 case MLD_G_QUERY_PENDING_MEMBER: 3024 case MLD_SG_QUERY_PENDING_MEMBER: 3025 case MLD_LEAVING_MEMBER: 3026 break; 3027 } 3028 } 3029 IF_ADDR_RUNLOCK(ifp); 3030 3031 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); 3032 3033 /* 3034 * Slew transmission of bursts over 500ms intervals. 3035 */ 3036 if (mli->mli_gq.ifq_head != NULL) { 3037 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( 3038 MLD_RESPONSE_BURST_INTERVAL); 3039 V_interface_timers_running6 = 1; 3040 } 3041} 3042 3043/* 3044 * Transmit the next pending message in the output queue. 3045 * 3046 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3047 * MRT: Nothing needs to be done, as MLD traffic is always local to 3048 * a link and uses a link-scope multicast address. 3049 */ 3050static void 3051mld_dispatch_packet(struct mbuf *m) 3052{ 3053 struct ip6_moptions im6o; 3054 struct ifnet *ifp; 3055 struct ifnet *oifp; 3056 struct mbuf *m0; 3057 struct mbuf *md; 3058 struct ip6_hdr *ip6; 3059 struct mld_hdr *mld; 3060 int error; 3061 int off; 3062 int type; 3063 uint32_t ifindex; 3064 3065 CTR2(KTR_MLD, "%s: transmit %p", __func__, m); 3066 3067 /* 3068 * Set VNET image pointer from enqueued mbuf chain 3069 * before doing anything else. Whilst we use interface 3070 * indexes to guard against interface detach, they are 3071 * unique to each VIMAGE and must be retrieved. 3072 */ 3073 ifindex = mld_restore_context(m); 3074 3075 /* 3076 * Check if the ifnet still exists. This limits the scope of 3077 * any race in the absence of a global ifp lock for low cost 3078 * (an array lookup). 3079 */ 3080 ifp = ifnet_byindex(ifindex); 3081 if (ifp == NULL) { 3082 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", 3083 __func__, m, ifindex); 3084 m_freem(m); 3085 IP6STAT_INC(ip6s_noroute); 3086 goto out; 3087 } 3088 3089 im6o.im6o_multicast_hlim = 1; 3090 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); 3091 im6o.im6o_multicast_ifp = ifp; 3092 3093 if (m->m_flags & M_MLDV1) { 3094 m0 = m; 3095 } else { 3096 m0 = mld_v2_encap_report(ifp, m); 3097 if (m0 == NULL) { 3098 CTR2(KTR_MLD, "%s: dropped %p", __func__, m); 3099 m_freem(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 } 3148out: 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 */ 3161static struct mbuf * 3162mld_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 3223static char * 3224mld_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 3253static void 3254mld_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} 3266SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL); 3267 3268static void 3269mld_uninit(void *unused __unused) 3270{ 3271 3272 CTR1(KTR_MLD, "%s: tearing down", __func__); 3273 MLD_LOCK_DESTROY(); 3274} 3275SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL); 3276 3277static void 3278vnet_mld_init(const void *unused __unused) 3279{ 3280 3281 CTR1(KTR_MLD, "%s: initializing", __func__); 3282 3283 LIST_INIT(&V_mli_head); 3284} 3285VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init, 3286 NULL); 3287 3288static void 3289vnet_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} 3297VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit, 3298 NULL); 3299 3300static int 3301mld_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 3314static moduledata_t mld_mod = { 3315 "mld", 3316 mld_modevent, 3317 0 3318}; 3319DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3320