1/*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2007-2009 Bruce Simpson. 5 * Copyright (c) 1988 Stephen Deering. 6 * Copyright (c) 1992, 1993 7 * The Regents of the University of California. All rights reserved. 8 * 9 * This code is derived from software contributed to Berkeley by 10 * Stephen Deering of Stanford University. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)igmp.c 8.1 (Berkeley) 7/19/93 37 */ 38 39/* 40 * Internet Group Management Protocol (IGMP) routines. 41 * [RFC1112, RFC2236, RFC3376] 42 * 43 * Written by Steve Deering, Stanford, May 1988. 44 * Modified by Rosen Sharma, Stanford, Aug 1994. 45 * Modified by Bill Fenner, Xerox PARC, Feb 1995. 46 * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995. 47 * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson. 48 * 49 * MULTICAST Revision: 3.5.1.4 50 */ 51 52#include <sys/cdefs.h> 53__FBSDID("$FreeBSD$"); 54 55#include "opt_ddb.h" 56 57#include <sys/param.h> 58#include <sys/systm.h> 59#include <sys/module.h> 60#include <sys/malloc.h> 61#include <sys/mbuf.h> 62#include <sys/socket.h> 63#include <sys/protosw.h> 64#include <sys/kernel.h> 65#include <sys/lock.h> 66#include <sys/rmlock.h> 67#include <sys/sysctl.h> 68#include <sys/ktr.h> 69#include <sys/condvar.h> 70 71#ifdef DDB 72#include <ddb/ddb.h> 73#endif 74 75#include <net/if.h> 76#include <net/if_var.h> 77#include <net/netisr.h> 78#include <net/vnet.h> 79 80#include <netinet/in.h> 81#include <netinet/in_var.h> 82#include <netinet/in_systm.h> 83#include <netinet/ip.h> 84#include <netinet/ip_var.h> 85#include <netinet/ip_options.h> 86#include <netinet/igmp.h> 87#include <netinet/igmp_var.h> 88 89#include <machine/in_cksum.h> 90 91#include <security/mac/mac_framework.h> 92 93#ifndef KTR_IGMPV3 94#define KTR_IGMPV3 KTR_INET 95#endif 96 97static struct igmp_ifsoftc * 98 igi_alloc_locked(struct ifnet *); 99static void igi_delete_locked(const struct ifnet *); 100static void igmp_dispatch_queue(struct mbufq *, int, const int); 101static void igmp_fasttimo_vnet(void); 102static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *); 103static int igmp_handle_state_change(struct in_multi *, 104 struct igmp_ifsoftc *); 105static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *); 106static int igmp_input_v1_query(struct ifnet *, const struct ip *, 107 const struct igmp *); 108static int igmp_input_v2_query(struct ifnet *, const struct ip *, 109 const struct igmp *); 110static int igmp_input_v3_query(struct ifnet *, const struct ip *, 111 /*const*/ struct igmpv3 *); 112static int igmp_input_v3_group_query(struct in_multi *, 113 struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *); 114static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *, 115 /*const*/ struct igmp *); 116static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *, 117 /*const*/ struct igmp *); 118static void igmp_intr(struct mbuf *); 119static int igmp_isgroupreported(const struct in_addr); 120static struct mbuf * 121 igmp_ra_alloc(void); 122#ifdef KTR 123static char * igmp_rec_type_to_str(const int); 124#endif 125static void igmp_set_version(struct igmp_ifsoftc *, const int); 126static void igmp_slowtimo_vnet(void); 127static int igmp_v1v2_queue_report(struct in_multi *, const int); 128static void igmp_v1v2_process_group_timer(struct in_multi *, const int); 129static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *); 130static void igmp_v2_update_group(struct in_multi *, const int); 131static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *); 132static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *); 133static struct mbuf * 134 igmp_v3_encap_report(struct ifnet *, struct mbuf *); 135static int igmp_v3_enqueue_group_record(struct mbufq *, 136 struct in_multi *, const int, const int, const int); 137static int igmp_v3_enqueue_filter_change(struct mbufq *, 138 struct in_multi *); 139static void igmp_v3_process_group_timers(struct in_multi_head *, 140 struct mbufq *, struct mbufq *, struct in_multi *, 141 const int); 142static int igmp_v3_merge_state_changes(struct in_multi *, 143 struct mbufq *); 144static void igmp_v3_suppress_group_record(struct in_multi *); 145static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS); 146static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS); 147static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS); 148 149static const struct netisr_handler igmp_nh = { 150 .nh_name = "igmp", 151 .nh_handler = igmp_intr, 152 .nh_proto = NETISR_IGMP, 153 .nh_policy = NETISR_POLICY_SOURCE, 154}; 155 156/* 157 * System-wide globals. 158 * 159 * Unlocked access to these is OK, except for the global IGMP output 160 * queue. The IGMP subsystem lock ends up being system-wide for the moment, 161 * because all VIMAGEs have to share a global output queue, as netisrs 162 * themselves are not virtualized. 163 * 164 * Locking: 165 * * The permitted lock order is: IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 166 * Any may be taken independently; if any are held at the same 167 * time, the above lock order must be followed. 168 * * All output is delegated to the netisr. 169 * Now that Giant has been eliminated, the netisr may be inlined. 170 * * IN_MULTI_LIST_LOCK covers in_multi. 171 * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file, 172 * including the output queue. 173 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 174 * per-link state iterators. 175 * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface, 176 * therefore it is not refcounted. 177 * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi. 178 * 179 * Reference counting 180 * * IGMP acquires its own reference every time an in_multi is passed to 181 * it and the group is being joined for the first time. 182 * * IGMP releases its reference(s) on in_multi in a deferred way, 183 * because the operations which process the release run as part of 184 * a loop whose control variables are directly affected by the release 185 * (that, and not recursing on the IF_ADDR_LOCK). 186 * 187 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds 188 * to a vnet in ifp->if_vnet. 189 * 190 * SMPng: XXX We may potentially race operations on ifma_protospec. 191 * The problem is that we currently lack a clean way of taking the 192 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing, 193 * as anything which modifies ifma needs to be covered by that lock. 194 * So check for ifma_protospec being NULL before proceeding. 195 */ 196struct mtx igmp_mtx; 197 198struct mbuf *m_raopt; /* Router Alert option */ 199static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state"); 200 201/* 202 * VIMAGE-wide globals. 203 * 204 * The IGMPv3 timers themselves need to run per-image, however, 205 * protosw timers run globally (see tcp). 206 * An ifnet can only be in one vimage at a time, and the loopback 207 * ifnet, loif, is itself virtualized. 208 * It would otherwise be possible to seriously hose IGMP state, 209 * and create inconsistencies in upstream multicast routing, if you have 210 * multiple VIMAGEs running on the same link joining different multicast 211 * groups, UNLESS the "primary IP address" is different. This is because 212 * IGMP for IPv4 does not force link-local addresses to be used for each 213 * node, unlike MLD for IPv6. 214 * Obviously the IGMPv3 per-interface state has per-vimage granularity 215 * also as a result. 216 * 217 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection 218 * policy to control the address used by IGMP on the link. 219 */ 220VNET_DEFINE_STATIC(int, interface_timers_running); /* IGMPv3 general 221 * query response */ 222VNET_DEFINE_STATIC(int, state_change_timers_running); /* IGMPv3 state-change 223 * retransmit */ 224VNET_DEFINE_STATIC(int, current_state_timers_running); /* IGMPv1/v2 host 225 * report; IGMPv3 g/sg 226 * query response */ 227 228#define V_interface_timers_running VNET(interface_timers_running) 229#define V_state_change_timers_running VNET(state_change_timers_running) 230#define V_current_state_timers_running VNET(current_state_timers_running) 231 232VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) = 233 LIST_HEAD_INITIALIZER(igi_head); 234VNET_DEFINE_STATIC(struct igmpstat, igmpstat) = { 235 .igps_version = IGPS_VERSION_3, 236 .igps_len = sizeof(struct igmpstat), 237}; 238VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0}; 239 240#define V_igi_head VNET(igi_head) 241#define V_igmpstat VNET(igmpstat) 242#define V_igmp_gsrdelay VNET(igmp_gsrdelay) 243 244VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1; 245VNET_DEFINE_STATIC(int, igmp_sendra) = 1; 246VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1; 247VNET_DEFINE_STATIC(int, igmp_v1enable) = 1; 248VNET_DEFINE_STATIC(int, igmp_v2enable) = 1; 249VNET_DEFINE_STATIC(int, igmp_legacysupp); 250VNET_DEFINE_STATIC(int, igmp_default_version) = IGMP_VERSION_3; 251 252#define V_igmp_recvifkludge VNET(igmp_recvifkludge) 253#define V_igmp_sendra VNET(igmp_sendra) 254#define V_igmp_sendlocal VNET(igmp_sendlocal) 255#define V_igmp_v1enable VNET(igmp_v1enable) 256#define V_igmp_v2enable VNET(igmp_v2enable) 257#define V_igmp_legacysupp VNET(igmp_legacysupp) 258#define V_igmp_default_version VNET(igmp_default_version) 259 260/* 261 * Virtualized sysctls. 262 */ 263SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW, 264 &VNET_NAME(igmpstat), igmpstat, ""); 265SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW, 266 &VNET_NAME(igmp_recvifkludge), 0, 267 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address"); 268SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW, 269 &VNET_NAME(igmp_sendra), 0, 270 "Send IP Router Alert option in IGMPv2/v3 messages"); 271SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW, 272 &VNET_NAME(igmp_sendlocal), 0, 273 "Send IGMP membership reports for 224.0.0.0/24 groups"); 274SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW, 275 &VNET_NAME(igmp_v1enable), 0, 276 "Enable backwards compatibility with IGMPv1"); 277SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW, 278 &VNET_NAME(igmp_v2enable), 0, 279 "Enable backwards compatibility with IGMPv2"); 280SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW, 281 &VNET_NAME(igmp_legacysupp), 0, 282 "Allow v1/v2 reports to suppress v3 group responses"); 283SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version, 284 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 285 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I", 286 "Default version of IGMP to run on each interface"); 287SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay, 288 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 289 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I", 290 "Rate limit for IGMPv3 Group-and-Source queries in seconds"); 291 292/* 293 * Non-virtualized sysctls. 294 */ 295static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, 296 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo, 297 "Per-interface IGMPv3 state"); 298 299static __inline void 300igmp_save_context(struct mbuf *m, struct ifnet *ifp) 301{ 302 303#ifdef VIMAGE 304 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; 305#endif /* VIMAGE */ 306 m->m_pkthdr.rcvif = ifp; 307 m->m_pkthdr.flowid = ifp->if_index; 308} 309 310static __inline void 311igmp_scrub_context(struct mbuf *m) 312{ 313 314 m->m_pkthdr.PH_loc.ptr = NULL; 315 m->m_pkthdr.flowid = 0; 316} 317 318/* 319 * Restore context from a queued IGMP output chain. 320 * Return saved ifindex. 321 * 322 * VIMAGE: The assertion is there to make sure that we 323 * actually called CURVNET_SET() with what's in the mbuf chain. 324 */ 325static __inline uint32_t 326igmp_restore_context(struct mbuf *m) 327{ 328 329#ifdef notyet 330#if defined(VIMAGE) && defined(INVARIANTS) 331 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr), 332 ("%s: called when curvnet was not restored", __func__)); 333#endif 334#endif 335 return (m->m_pkthdr.flowid); 336} 337 338/* 339 * Retrieve or set default IGMP version. 340 * 341 * VIMAGE: Assume curvnet set by caller. 342 * SMPng: NOTE: Serialized by IGMP lock. 343 */ 344static int 345sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS) 346{ 347 int error; 348 int new; 349 350 error = sysctl_wire_old_buffer(req, sizeof(int)); 351 if (error) 352 return (error); 353 354 IGMP_LOCK(); 355 356 new = V_igmp_default_version; 357 358 error = sysctl_handle_int(oidp, &new, 0, req); 359 if (error || !req->newptr) 360 goto out_locked; 361 362 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) { 363 error = EINVAL; 364 goto out_locked; 365 } 366 367 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d", 368 V_igmp_default_version, new); 369 370 V_igmp_default_version = new; 371 372out_locked: 373 IGMP_UNLOCK(); 374 return (error); 375} 376 377/* 378 * Retrieve or set threshold between group-source queries in seconds. 379 * 380 * VIMAGE: Assume curvnet set by caller. 381 * SMPng: NOTE: Serialized by IGMP lock. 382 */ 383static int 384sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS) 385{ 386 int error; 387 int i; 388 389 error = sysctl_wire_old_buffer(req, sizeof(int)); 390 if (error) 391 return (error); 392 393 IGMP_LOCK(); 394 395 i = V_igmp_gsrdelay.tv_sec; 396 397 error = sysctl_handle_int(oidp, &i, 0, req); 398 if (error || !req->newptr) 399 goto out_locked; 400 401 if (i < -1 || i >= 60) { 402 error = EINVAL; 403 goto out_locked; 404 } 405 406 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d", 407 V_igmp_gsrdelay.tv_sec, i); 408 V_igmp_gsrdelay.tv_sec = i; 409 410out_locked: 411 IGMP_UNLOCK(); 412 return (error); 413} 414 415/* 416 * Expose struct igmp_ifsoftc to userland, keyed by ifindex. 417 * For use by ifmcstat(8). 418 * 419 * SMPng: NOTE: Does an unlocked ifindex space read. 420 * VIMAGE: Assume curvnet set by caller. The node handler itself 421 * is not directly virtualized. 422 */ 423static int 424sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS) 425{ 426 int *name; 427 int error; 428 u_int namelen; 429 struct ifnet *ifp; 430 struct igmp_ifsoftc *igi; 431 432 name = (int *)arg1; 433 namelen = arg2; 434 435 if (req->newptr != NULL) 436 return (EPERM); 437 438 if (namelen != 1) 439 return (EINVAL); 440 441 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo)); 442 if (error) 443 return (error); 444 445 IN_MULTI_LIST_LOCK(); 446 IGMP_LOCK(); 447 448 if (name[0] <= 0 || name[0] > V_if_index) { 449 error = ENOENT; 450 goto out_locked; 451 } 452 453 error = ENOENT; 454 455 ifp = ifnet_byindex(name[0]); 456 if (ifp == NULL) 457 goto out_locked; 458 459 LIST_FOREACH(igi, &V_igi_head, igi_link) { 460 if (ifp == igi->igi_ifp) { 461 struct igmp_ifinfo info; 462 463 info.igi_version = igi->igi_version; 464 info.igi_v1_timer = igi->igi_v1_timer; 465 info.igi_v2_timer = igi->igi_v2_timer; 466 info.igi_v3_timer = igi->igi_v3_timer; 467 info.igi_flags = igi->igi_flags; 468 info.igi_rv = igi->igi_rv; 469 info.igi_qi = igi->igi_qi; 470 info.igi_qri = igi->igi_qri; 471 info.igi_uri = igi->igi_uri; 472 error = SYSCTL_OUT(req, &info, sizeof(info)); 473 break; 474 } 475 } 476 477out_locked: 478 IGMP_UNLOCK(); 479 IN_MULTI_LIST_UNLOCK(); 480 return (error); 481} 482 483/* 484 * Dispatch an entire queue of pending packet chains 485 * using the netisr. 486 * VIMAGE: Assumes the vnet pointer has been set. 487 */ 488static void 489igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop) 490{ 491 struct mbuf *m; 492 493 while ((m = mbufq_dequeue(mq)) != NULL) { 494 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m); 495 if (loop) 496 m->m_flags |= M_IGMP_LOOP; 497 netisr_dispatch(NETISR_IGMP, m); 498 if (--limit == 0) 499 break; 500 } 501} 502 503/* 504 * Filter outgoing IGMP report state by group. 505 * 506 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1). 507 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are 508 * disabled for all groups in the 224.0.0.0/24 link-local scope. However, 509 * this may break certain IGMP snooping switches which rely on the old 510 * report behaviour. 511 * 512 * Return zero if the given group is one for which IGMP reports 513 * should be suppressed, or non-zero if reports should be issued. 514 */ 515static __inline int 516igmp_isgroupreported(const struct in_addr addr) 517{ 518 519 if (in_allhosts(addr) || 520 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr))))) 521 return (0); 522 523 return (1); 524} 525 526/* 527 * Construct a Router Alert option to use in outgoing packets. 528 */ 529static struct mbuf * 530igmp_ra_alloc(void) 531{ 532 struct mbuf *m; 533 struct ipoption *p; 534 535 m = m_get(M_WAITOK, MT_DATA); 536 p = mtod(m, struct ipoption *); 537 p->ipopt_dst.s_addr = INADDR_ANY; 538 p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */ 539 p->ipopt_list[1] = 0x04; /* 4 bytes long */ 540 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */ 541 p->ipopt_list[3] = 0x00; /* pad byte */ 542 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1]; 543 544 return (m); 545} 546 547/* 548 * Attach IGMP when PF_INET is attached to an interface. 549 */ 550struct igmp_ifsoftc * 551igmp_domifattach(struct ifnet *ifp) 552{ 553 struct igmp_ifsoftc *igi; 554 555 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 556 __func__, ifp, ifp->if_xname); 557 558 IGMP_LOCK(); 559 560 igi = igi_alloc_locked(ifp); 561 if (!(ifp->if_flags & IFF_MULTICAST)) 562 igi->igi_flags |= IGIF_SILENT; 563 564 IGMP_UNLOCK(); 565 566 return (igi); 567} 568 569/* 570 * VIMAGE: assume curvnet set by caller. 571 */ 572static struct igmp_ifsoftc * 573igi_alloc_locked(/*const*/ struct ifnet *ifp) 574{ 575 struct igmp_ifsoftc *igi; 576 577 IGMP_LOCK_ASSERT(); 578 579 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO); 580 if (igi == NULL) 581 goto out; 582 583 igi->igi_ifp = ifp; 584 igi->igi_version = V_igmp_default_version; 585 igi->igi_flags = 0; 586 igi->igi_rv = IGMP_RV_INIT; 587 igi->igi_qi = IGMP_QI_INIT; 588 igi->igi_qri = IGMP_QRI_INIT; 589 igi->igi_uri = IGMP_URI_INIT; 590 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS); 591 592 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link); 593 594 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)", 595 ifp, ifp->if_xname); 596 597out: 598 return (igi); 599} 600 601/* 602 * Hook for ifdetach. 603 * 604 * NOTE: Some finalization tasks need to run before the protocol domain 605 * is detached, but also before the link layer does its cleanup. 606 * 607 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK(). 608 * XXX This is also bitten by unlocked ifma_protospec access. 609 */ 610void 611igmp_ifdetach(struct ifnet *ifp) 612{ 613 struct igmp_ifsoftc *igi; 614 struct ifmultiaddr *ifma, *next; 615 struct in_multi *inm; 616 struct in_multi_head inm_free_tmp; 617 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, 618 ifp->if_xname); 619 620 SLIST_INIT(&inm_free_tmp); 621 IGMP_LOCK(); 622 623 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 624 if (igi->igi_version == IGMP_VERSION_3) { 625 IF_ADDR_WLOCK(ifp); 626 restart: 627 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { 628 if (ifma->ifma_addr->sa_family != AF_INET || 629 ifma->ifma_protospec == NULL) 630 continue; 631 inm = (struct in_multi *)ifma->ifma_protospec; 632 if (inm->inm_state == IGMP_LEAVING_MEMBER) 633 inm_rele_locked(&inm_free_tmp, inm); 634 inm_clear_recorded(inm); 635 if (__predict_false(ifma_restart)) { 636 ifma_restart = false; 637 goto restart; 638 } 639 } 640 IF_ADDR_WUNLOCK(ifp); 641 inm_release_list_deferred(&inm_free_tmp); 642 } 643 IGMP_UNLOCK(); 644 645} 646 647/* 648 * Hook for domifdetach. 649 */ 650void 651igmp_domifdetach(struct ifnet *ifp) 652{ 653 654 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 655 __func__, ifp, ifp->if_xname); 656 657 IGMP_LOCK(); 658 igi_delete_locked(ifp); 659 IGMP_UNLOCK(); 660} 661 662static void 663igi_delete_locked(const struct ifnet *ifp) 664{ 665 struct igmp_ifsoftc *igi, *tigi; 666 667 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)", 668 __func__, ifp, ifp->if_xname); 669 670 IGMP_LOCK_ASSERT(); 671 672 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) { 673 if (igi->igi_ifp == ifp) { 674 /* 675 * Free deferred General Query responses. 676 */ 677 mbufq_drain(&igi->igi_gq); 678 679 LIST_REMOVE(igi, igi_link); 680 free(igi, M_IGMP); 681 return; 682 } 683 } 684} 685 686/* 687 * Process a received IGMPv1 query. 688 * Return non-zero if the message should be dropped. 689 * 690 * VIMAGE: The curvnet pointer is derived from the input ifp. 691 */ 692static int 693igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip, 694 const struct igmp *igmp) 695{ 696 struct ifmultiaddr *ifma; 697 struct igmp_ifsoftc *igi; 698 struct in_multi *inm; 699 700 /* 701 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to 702 * 224.0.0.1. They are always treated as General Queries. 703 * igmp_group is always ignored. Do not drop it as a userland 704 * daemon may wish to see it. 705 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 706 */ 707 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) { 708 IGMPSTAT_INC(igps_rcv_badqueries); 709 return (0); 710 } 711 IGMPSTAT_INC(igps_rcv_gen_queries); 712 713 IN_MULTI_LIST_LOCK(); 714 IGMP_LOCK(); 715 716 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 717 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 718 719 if (igi->igi_flags & IGIF_LOOPBACK) { 720 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)", 721 ifp, ifp->if_xname); 722 goto out_locked; 723 } 724 725 /* 726 * Switch to IGMPv1 host compatibility mode. 727 */ 728 igmp_set_version(igi, IGMP_VERSION_1); 729 730 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname); 731 732 /* 733 * Start the timers in all of our group records 734 * for the interface on which the query arrived, 735 * except those which are already running. 736 */ 737 IF_ADDR_RLOCK(ifp); 738 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 739 if (ifma->ifma_addr->sa_family != AF_INET || 740 ifma->ifma_protospec == NULL) 741 continue; 742 inm = (struct in_multi *)ifma->ifma_protospec; 743 if (inm->inm_timer != 0) 744 continue; 745 switch (inm->inm_state) { 746 case IGMP_NOT_MEMBER: 747 case IGMP_SILENT_MEMBER: 748 break; 749 case IGMP_G_QUERY_PENDING_MEMBER: 750 case IGMP_SG_QUERY_PENDING_MEMBER: 751 case IGMP_REPORTING_MEMBER: 752 case IGMP_IDLE_MEMBER: 753 case IGMP_LAZY_MEMBER: 754 case IGMP_SLEEPING_MEMBER: 755 case IGMP_AWAKENING_MEMBER: 756 inm->inm_state = IGMP_REPORTING_MEMBER; 757 inm->inm_timer = IGMP_RANDOM_DELAY( 758 IGMP_V1V2_MAX_RI * PR_FASTHZ); 759 V_current_state_timers_running = 1; 760 break; 761 case IGMP_LEAVING_MEMBER: 762 break; 763 } 764 } 765 IF_ADDR_RUNLOCK(ifp); 766 767out_locked: 768 IGMP_UNLOCK(); 769 IN_MULTI_LIST_UNLOCK(); 770 771 return (0); 772} 773 774/* 775 * Process a received IGMPv2 general or group-specific query. 776 */ 777static int 778igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip, 779 const struct igmp *igmp) 780{ 781 struct ifmultiaddr *ifma; 782 struct igmp_ifsoftc *igi; 783 struct in_multi *inm; 784 int is_general_query; 785 uint16_t timer; 786 787 is_general_query = 0; 788 789 /* 790 * Validate address fields upfront. 791 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 792 */ 793 if (in_nullhost(igmp->igmp_group)) { 794 /* 795 * IGMPv2 General Query. 796 * If this was not sent to the all-hosts group, ignore it. 797 */ 798 if (!in_allhosts(ip->ip_dst)) 799 return (0); 800 IGMPSTAT_INC(igps_rcv_gen_queries); 801 is_general_query = 1; 802 } else { 803 /* IGMPv2 Group-Specific Query. */ 804 IGMPSTAT_INC(igps_rcv_group_queries); 805 } 806 807 IN_MULTI_LIST_LOCK(); 808 IGMP_LOCK(); 809 810 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 811 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 812 813 if (igi->igi_flags & IGIF_LOOPBACK) { 814 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)", 815 ifp, ifp->if_xname); 816 goto out_locked; 817 } 818 819 /* 820 * Ignore v2 query if in v1 Compatibility Mode. 821 */ 822 if (igi->igi_version == IGMP_VERSION_1) 823 goto out_locked; 824 825 igmp_set_version(igi, IGMP_VERSION_2); 826 827 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE; 828 if (timer == 0) 829 timer = 1; 830 831 if (is_general_query) { 832 /* 833 * For each reporting group joined on this 834 * interface, kick the report timer. 835 */ 836 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)", 837 ifp, ifp->if_xname); 838 IF_ADDR_RLOCK(ifp); 839 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 840 if (ifma->ifma_addr->sa_family != AF_INET || 841 ifma->ifma_protospec == NULL) 842 continue; 843 inm = (struct in_multi *)ifma->ifma_protospec; 844 igmp_v2_update_group(inm, timer); 845 } 846 IF_ADDR_RUNLOCK(ifp); 847 } else { 848 /* 849 * Group-specific IGMPv2 query, we need only 850 * look up the single group to process it. 851 */ 852 inm = inm_lookup(ifp, igmp->igmp_group); 853 if (inm != NULL) { 854 CTR3(KTR_IGMPV3, 855 "process v2 query 0x%08x on ifp %p(%s)", 856 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 857 igmp_v2_update_group(inm, timer); 858 } 859 } 860 861out_locked: 862 IGMP_UNLOCK(); 863 IN_MULTI_LIST_UNLOCK(); 864 865 return (0); 866} 867 868/* 869 * Update the report timer on a group in response to an IGMPv2 query. 870 * 871 * If we are becoming the reporting member for this group, start the timer. 872 * If we already are the reporting member for this group, and timer is 873 * below the threshold, reset it. 874 * 875 * We may be updating the group for the first time since we switched 876 * to IGMPv3. If we are, then we must clear any recorded source lists, 877 * and transition to REPORTING state; the group timer is overloaded 878 * for group and group-source query responses. 879 * 880 * Unlike IGMPv3, the delay per group should be jittered 881 * to avoid bursts of IGMPv2 reports. 882 */ 883static void 884igmp_v2_update_group(struct in_multi *inm, const int timer) 885{ 886 887 CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__, 888 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer); 889 890 IN_MULTI_LIST_LOCK_ASSERT(); 891 892 switch (inm->inm_state) { 893 case IGMP_NOT_MEMBER: 894 case IGMP_SILENT_MEMBER: 895 break; 896 case IGMP_REPORTING_MEMBER: 897 if (inm->inm_timer != 0 && 898 inm->inm_timer <= timer) { 899 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, " 900 "skipping.", __func__); 901 break; 902 } 903 /* FALLTHROUGH */ 904 case IGMP_SG_QUERY_PENDING_MEMBER: 905 case IGMP_G_QUERY_PENDING_MEMBER: 906 case IGMP_IDLE_MEMBER: 907 case IGMP_LAZY_MEMBER: 908 case IGMP_AWAKENING_MEMBER: 909 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__); 910 inm->inm_state = IGMP_REPORTING_MEMBER; 911 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 912 V_current_state_timers_running = 1; 913 break; 914 case IGMP_SLEEPING_MEMBER: 915 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__); 916 inm->inm_state = IGMP_AWAKENING_MEMBER; 917 break; 918 case IGMP_LEAVING_MEMBER: 919 break; 920 } 921} 922 923/* 924 * Process a received IGMPv3 general, group-specific or 925 * group-and-source-specific query. 926 * Assumes m has already been pulled up to the full IGMP message length. 927 * Return 0 if successful, otherwise an appropriate error code is returned. 928 */ 929static int 930igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip, 931 /*const*/ struct igmpv3 *igmpv3) 932{ 933 struct igmp_ifsoftc *igi; 934 struct in_multi *inm; 935 int is_general_query; 936 uint32_t maxresp, nsrc, qqi; 937 uint16_t timer; 938 uint8_t qrv; 939 940 is_general_query = 0; 941 942 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname); 943 944 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */ 945 if (maxresp >= 128) { 946 maxresp = IGMP_MANT(igmpv3->igmp_code) << 947 (IGMP_EXP(igmpv3->igmp_code) + 3); 948 } 949 950 /* 951 * Robustness must never be less than 2 for on-wire IGMPv3. 952 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make 953 * an exception for interfaces whose IGMPv3 state changes 954 * are redirected to loopback (e.g. MANET). 955 */ 956 qrv = IGMP_QRV(igmpv3->igmp_misc); 957 if (qrv < 2) { 958 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__, 959 qrv, IGMP_RV_INIT); 960 qrv = IGMP_RV_INIT; 961 } 962 963 qqi = igmpv3->igmp_qqi; 964 if (qqi >= 128) { 965 qqi = IGMP_MANT(igmpv3->igmp_qqi) << 966 (IGMP_EXP(igmpv3->igmp_qqi) + 3); 967 } 968 969 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE; 970 if (timer == 0) 971 timer = 1; 972 973 nsrc = ntohs(igmpv3->igmp_numsrc); 974 975 /* 976 * Validate address fields and versions upfront before 977 * accepting v3 query. 978 * XXX SMPng: Unlocked access to igmpstat counters here. 979 */ 980 if (in_nullhost(igmpv3->igmp_group)) { 981 /* 982 * IGMPv3 General Query. 983 * 984 * General Queries SHOULD be directed to 224.0.0.1. 985 * A general query with a source list has undefined 986 * behaviour; discard it. 987 */ 988 IGMPSTAT_INC(igps_rcv_gen_queries); 989 if (!in_allhosts(ip->ip_dst) || nsrc > 0) { 990 IGMPSTAT_INC(igps_rcv_badqueries); 991 return (0); 992 } 993 is_general_query = 1; 994 } else { 995 /* Group or group-source specific query. */ 996 if (nsrc == 0) 997 IGMPSTAT_INC(igps_rcv_group_queries); 998 else 999 IGMPSTAT_INC(igps_rcv_gsr_queries); 1000 } 1001 1002 IN_MULTI_LIST_LOCK(); 1003 IGMP_LOCK(); 1004 1005 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 1006 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 1007 1008 if (igi->igi_flags & IGIF_LOOPBACK) { 1009 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)", 1010 ifp, ifp->if_xname); 1011 goto out_locked; 1012 } 1013 1014 /* 1015 * Discard the v3 query if we're in Compatibility Mode. 1016 * The RFC is not obviously worded that hosts need to stay in 1017 * compatibility mode until the Old Version Querier Present 1018 * timer expires. 1019 */ 1020 if (igi->igi_version != IGMP_VERSION_3) { 1021 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)", 1022 igi->igi_version, ifp, ifp->if_xname); 1023 goto out_locked; 1024 } 1025 1026 igmp_set_version(igi, IGMP_VERSION_3); 1027 igi->igi_rv = qrv; 1028 igi->igi_qi = qqi; 1029 igi->igi_qri = maxresp; 1030 1031 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi, 1032 maxresp); 1033 1034 if (is_general_query) { 1035 /* 1036 * Schedule a current-state report on this ifp for 1037 * all groups, possibly containing source lists. 1038 * If there is a pending General Query response 1039 * scheduled earlier than the selected delay, do 1040 * not schedule any other reports. 1041 * Otherwise, reset the interface timer. 1042 */ 1043 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)", 1044 ifp, ifp->if_xname); 1045 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) { 1046 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer); 1047 V_interface_timers_running = 1; 1048 } 1049 } else { 1050 /* 1051 * Group-source-specific queries are throttled on 1052 * a per-group basis to defeat denial-of-service attempts. 1053 * Queries for groups we are not a member of on this 1054 * link are simply ignored. 1055 */ 1056 inm = inm_lookup(ifp, igmpv3->igmp_group); 1057 if (inm == NULL) 1058 goto out_locked; 1059 if (nsrc > 0) { 1060 if (!ratecheck(&inm->inm_lastgsrtv, 1061 &V_igmp_gsrdelay)) { 1062 CTR1(KTR_IGMPV3, "%s: GS query throttled.", 1063 __func__); 1064 IGMPSTAT_INC(igps_drop_gsr_queries); 1065 goto out_locked; 1066 } 1067 } 1068 CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)", 1069 ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname); 1070 /* 1071 * If there is a pending General Query response 1072 * scheduled sooner than the selected delay, no 1073 * further report need be scheduled. 1074 * Otherwise, prepare to respond to the 1075 * group-specific or group-and-source query. 1076 */ 1077 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) 1078 igmp_input_v3_group_query(inm, igi, timer, igmpv3); 1079 } 1080 1081out_locked: 1082 IGMP_UNLOCK(); 1083 IN_MULTI_LIST_UNLOCK(); 1084 1085 return (0); 1086} 1087 1088/* 1089 * Process a received IGMPv3 group-specific or group-and-source-specific 1090 * query. 1091 * Return <0 if any error occurred. Currently this is ignored. 1092 */ 1093static int 1094igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi, 1095 int timer, /*const*/ struct igmpv3 *igmpv3) 1096{ 1097 int retval; 1098 uint16_t nsrc; 1099 1100 IN_MULTI_LIST_LOCK_ASSERT(); 1101 IGMP_LOCK_ASSERT(); 1102 1103 retval = 0; 1104 1105 switch (inm->inm_state) { 1106 case IGMP_NOT_MEMBER: 1107 case IGMP_SILENT_MEMBER: 1108 case IGMP_SLEEPING_MEMBER: 1109 case IGMP_LAZY_MEMBER: 1110 case IGMP_AWAKENING_MEMBER: 1111 case IGMP_IDLE_MEMBER: 1112 case IGMP_LEAVING_MEMBER: 1113 return (retval); 1114 break; 1115 case IGMP_REPORTING_MEMBER: 1116 case IGMP_G_QUERY_PENDING_MEMBER: 1117 case IGMP_SG_QUERY_PENDING_MEMBER: 1118 break; 1119 } 1120 1121 nsrc = ntohs(igmpv3->igmp_numsrc); 1122 1123 /* 1124 * Deal with group-specific queries upfront. 1125 * If any group query is already pending, purge any recorded 1126 * source-list state if it exists, and schedule a query response 1127 * for this group-specific query. 1128 */ 1129 if (nsrc == 0) { 1130 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || 1131 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) { 1132 inm_clear_recorded(inm); 1133 timer = min(inm->inm_timer, timer); 1134 } 1135 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER; 1136 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1137 V_current_state_timers_running = 1; 1138 return (retval); 1139 } 1140 1141 /* 1142 * Deal with the case where a group-and-source-specific query has 1143 * been received but a group-specific query is already pending. 1144 */ 1145 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) { 1146 timer = min(inm->inm_timer, timer); 1147 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1148 V_current_state_timers_running = 1; 1149 return (retval); 1150 } 1151 1152 /* 1153 * Finally, deal with the case where a group-and-source-specific 1154 * query has been received, where a response to a previous g-s-r 1155 * query exists, or none exists. 1156 * In this case, we need to parse the source-list which the Querier 1157 * has provided us with and check if we have any source list filter 1158 * entries at T1 for these sources. If we do not, there is no need 1159 * schedule a report and the query may be dropped. 1160 * If we do, we must record them and schedule a current-state 1161 * report for those sources. 1162 * FIXME: Handling source lists larger than 1 mbuf requires that 1163 * we pass the mbuf chain pointer down to this function, and use 1164 * m_getptr() to walk the chain. 1165 */ 1166 if (inm->inm_nsrc > 0) { 1167 const struct in_addr *ap; 1168 int i, nrecorded; 1169 1170 ap = (const struct in_addr *)(igmpv3 + 1); 1171 nrecorded = 0; 1172 for (i = 0; i < nsrc; i++, ap++) { 1173 retval = inm_record_source(inm, ap->s_addr); 1174 if (retval < 0) 1175 break; 1176 nrecorded += retval; 1177 } 1178 if (nrecorded > 0) { 1179 CTR1(KTR_IGMPV3, 1180 "%s: schedule response to SG query", __func__); 1181 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER; 1182 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1183 V_current_state_timers_running = 1; 1184 } 1185 } 1186 1187 return (retval); 1188} 1189 1190/* 1191 * Process a received IGMPv1 host membership report. 1192 * 1193 * NOTE: 0.0.0.0 workaround breaks const correctness. 1194 */ 1195static int 1196igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1197 /*const*/ struct igmp *igmp) 1198{ 1199 struct rm_priotracker in_ifa_tracker; 1200 struct in_ifaddr *ia; 1201 struct in_multi *inm; 1202 1203 IGMPSTAT_INC(igps_rcv_reports); 1204 1205 if (ifp->if_flags & IFF_LOOPBACK) 1206 return (0); 1207 1208 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || 1209 !in_hosteq(igmp->igmp_group, ip->ip_dst)) { 1210 IGMPSTAT_INC(igps_rcv_badreports); 1211 return (EINVAL); 1212 } 1213 1214 /* 1215 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1216 * Booting clients may use the source address 0.0.0.0. Some 1217 * IGMP daemons may not know how to use IP_RECVIF to determine 1218 * the interface upon which this message was received. 1219 * Replace 0.0.0.0 with the subnet address if told to do so. 1220 */ 1221 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1222 NET_EPOCH_ENTER(); 1223 IFP_TO_IA(ifp, ia, &in_ifa_tracker); 1224 if (ia != NULL) 1225 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1226 NET_EPOCH_EXIT(); 1227 } 1228 1229 CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)", 1230 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 1231 1232 /* 1233 * IGMPv1 report suppression. 1234 * If we are a member of this group, and our membership should be 1235 * reported, stop our group timer and transition to the 'lazy' state. 1236 */ 1237 IN_MULTI_LIST_LOCK(); 1238 inm = inm_lookup(ifp, igmp->igmp_group); 1239 if (inm != NULL) { 1240 struct igmp_ifsoftc *igi; 1241 1242 igi = inm->inm_igi; 1243 if (igi == NULL) { 1244 KASSERT(igi != NULL, 1245 ("%s: no igi for ifp %p", __func__, ifp)); 1246 goto out_locked; 1247 } 1248 1249 IGMPSTAT_INC(igps_rcv_ourreports); 1250 1251 /* 1252 * If we are in IGMPv3 host mode, do not allow the 1253 * other host's IGMPv1 report to suppress our reports 1254 * unless explicitly configured to do so. 1255 */ 1256 if (igi->igi_version == IGMP_VERSION_3) { 1257 if (V_igmp_legacysupp) 1258 igmp_v3_suppress_group_record(inm); 1259 goto out_locked; 1260 } 1261 1262 inm->inm_timer = 0; 1263 1264 switch (inm->inm_state) { 1265 case IGMP_NOT_MEMBER: 1266 case IGMP_SILENT_MEMBER: 1267 break; 1268 case IGMP_IDLE_MEMBER: 1269 case IGMP_LAZY_MEMBER: 1270 case IGMP_AWAKENING_MEMBER: 1271 CTR3(KTR_IGMPV3, 1272 "report suppressed for 0x%08x on ifp %p(%s)", 1273 ntohl(igmp->igmp_group.s_addr), ifp, 1274 ifp->if_xname); 1275 case IGMP_SLEEPING_MEMBER: 1276 inm->inm_state = IGMP_SLEEPING_MEMBER; 1277 break; 1278 case IGMP_REPORTING_MEMBER: 1279 CTR3(KTR_IGMPV3, 1280 "report suppressed for 0x%08x on ifp %p(%s)", 1281 ntohl(igmp->igmp_group.s_addr), ifp, 1282 ifp->if_xname); 1283 if (igi->igi_version == IGMP_VERSION_1) 1284 inm->inm_state = IGMP_LAZY_MEMBER; 1285 else if (igi->igi_version == IGMP_VERSION_2) 1286 inm->inm_state = IGMP_SLEEPING_MEMBER; 1287 break; 1288 case IGMP_G_QUERY_PENDING_MEMBER: 1289 case IGMP_SG_QUERY_PENDING_MEMBER: 1290 case IGMP_LEAVING_MEMBER: 1291 break; 1292 } 1293 } 1294 1295out_locked: 1296 IN_MULTI_LIST_UNLOCK(); 1297 1298 return (0); 1299} 1300 1301/* 1302 * Process a received IGMPv2 host membership report. 1303 * 1304 * NOTE: 0.0.0.0 workaround breaks const correctness. 1305 */ 1306static int 1307igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1308 /*const*/ struct igmp *igmp) 1309{ 1310 struct rm_priotracker in_ifa_tracker; 1311 struct in_ifaddr *ia; 1312 struct in_multi *inm; 1313 1314 /* 1315 * Make sure we don't hear our own membership report. Fast 1316 * leave requires knowing that we are the only member of a 1317 * group. 1318 */ 1319 NET_EPOCH_ENTER(); 1320 IFP_TO_IA(ifp, ia, &in_ifa_tracker); 1321 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) { 1322 NET_EPOCH_EXIT(); 1323 return (0); 1324 } 1325 1326 IGMPSTAT_INC(igps_rcv_reports); 1327 1328 if (ifp->if_flags & IFF_LOOPBACK) { 1329 NET_EPOCH_EXIT(); 1330 return (0); 1331 } 1332 1333 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || 1334 !in_hosteq(igmp->igmp_group, ip->ip_dst)) { 1335 NET_EPOCH_EXIT(); 1336 IGMPSTAT_INC(igps_rcv_badreports); 1337 return (EINVAL); 1338 } 1339 1340 /* 1341 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1342 * Booting clients may use the source address 0.0.0.0. Some 1343 * IGMP daemons may not know how to use IP_RECVIF to determine 1344 * the interface upon which this message was received. 1345 * Replace 0.0.0.0 with the subnet address if told to do so. 1346 */ 1347 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1348 if (ia != NULL) 1349 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1350 } 1351 NET_EPOCH_EXIT(); 1352 1353 CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)", 1354 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 1355 1356 /* 1357 * IGMPv2 report suppression. 1358 * If we are a member of this group, and our membership should be 1359 * reported, and our group timer is pending or about to be reset, 1360 * stop our group timer by transitioning to the 'lazy' state. 1361 */ 1362 IN_MULTI_LIST_LOCK(); 1363 inm = inm_lookup(ifp, igmp->igmp_group); 1364 if (inm != NULL) { 1365 struct igmp_ifsoftc *igi; 1366 1367 igi = inm->inm_igi; 1368 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp)); 1369 1370 IGMPSTAT_INC(igps_rcv_ourreports); 1371 1372 /* 1373 * If we are in IGMPv3 host mode, do not allow the 1374 * other host's IGMPv1 report to suppress our reports 1375 * unless explicitly configured to do so. 1376 */ 1377 if (igi->igi_version == IGMP_VERSION_3) { 1378 if (V_igmp_legacysupp) 1379 igmp_v3_suppress_group_record(inm); 1380 goto out_locked; 1381 } 1382 1383 inm->inm_timer = 0; 1384 1385 switch (inm->inm_state) { 1386 case IGMP_NOT_MEMBER: 1387 case IGMP_SILENT_MEMBER: 1388 case IGMP_SLEEPING_MEMBER: 1389 break; 1390 case IGMP_REPORTING_MEMBER: 1391 case IGMP_IDLE_MEMBER: 1392 case IGMP_AWAKENING_MEMBER: 1393 CTR3(KTR_IGMPV3, 1394 "report suppressed for 0x%08x on ifp %p(%s)", 1395 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 1396 case IGMP_LAZY_MEMBER: 1397 inm->inm_state = IGMP_LAZY_MEMBER; 1398 break; 1399 case IGMP_G_QUERY_PENDING_MEMBER: 1400 case IGMP_SG_QUERY_PENDING_MEMBER: 1401 case IGMP_LEAVING_MEMBER: 1402 break; 1403 } 1404 } 1405 1406out_locked: 1407 IN_MULTI_LIST_UNLOCK(); 1408 1409 return (0); 1410} 1411 1412int 1413igmp_input(struct mbuf **mp, int *offp, int proto) 1414{ 1415 int iphlen; 1416 struct ifnet *ifp; 1417 struct igmp *igmp; 1418 struct ip *ip; 1419 struct mbuf *m; 1420 int igmplen; 1421 int minlen; 1422 int queryver; 1423 1424 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp); 1425 1426 m = *mp; 1427 ifp = m->m_pkthdr.rcvif; 1428 *mp = NULL; 1429 1430 IGMPSTAT_INC(igps_rcv_total); 1431 1432 ip = mtod(m, struct ip *); 1433 iphlen = *offp; 1434 igmplen = ntohs(ip->ip_len) - iphlen; 1435 1436 /* 1437 * Validate lengths. 1438 */ 1439 if (igmplen < IGMP_MINLEN) { 1440 IGMPSTAT_INC(igps_rcv_tooshort); 1441 m_freem(m); 1442 return (IPPROTO_DONE); 1443 } 1444 1445 /* 1446 * Always pullup to the minimum size for v1/v2 or v3 1447 * to amortize calls to m_pullup(). 1448 */ 1449 minlen = iphlen; 1450 if (igmplen >= IGMP_V3_QUERY_MINLEN) 1451 minlen += IGMP_V3_QUERY_MINLEN; 1452 else 1453 minlen += IGMP_MINLEN; 1454 if ((!M_WRITABLE(m) || m->m_len < minlen) && 1455 (m = m_pullup(m, minlen)) == NULL) { 1456 IGMPSTAT_INC(igps_rcv_tooshort); 1457 return (IPPROTO_DONE); 1458 } 1459 ip = mtod(m, struct ip *); 1460 1461 /* 1462 * Validate checksum. 1463 */ 1464 m->m_data += iphlen; 1465 m->m_len -= iphlen; 1466 igmp = mtod(m, struct igmp *); 1467 if (in_cksum(m, igmplen)) { 1468 IGMPSTAT_INC(igps_rcv_badsum); 1469 m_freem(m); 1470 return (IPPROTO_DONE); 1471 } 1472 m->m_data -= iphlen; 1473 m->m_len += iphlen; 1474 1475 /* 1476 * IGMP control traffic is link-scope, and must have a TTL of 1. 1477 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception; 1478 * probe packets may come from beyond the LAN. 1479 */ 1480 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) { 1481 IGMPSTAT_INC(igps_rcv_badttl); 1482 m_freem(m); 1483 return (IPPROTO_DONE); 1484 } 1485 1486 switch (igmp->igmp_type) { 1487 case IGMP_HOST_MEMBERSHIP_QUERY: 1488 if (igmplen == IGMP_MINLEN) { 1489 if (igmp->igmp_code == 0) 1490 queryver = IGMP_VERSION_1; 1491 else 1492 queryver = IGMP_VERSION_2; 1493 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) { 1494 queryver = IGMP_VERSION_3; 1495 } else { 1496 IGMPSTAT_INC(igps_rcv_tooshort); 1497 m_freem(m); 1498 return (IPPROTO_DONE); 1499 } 1500 1501 switch (queryver) { 1502 case IGMP_VERSION_1: 1503 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1504 if (!V_igmp_v1enable) 1505 break; 1506 if (igmp_input_v1_query(ifp, ip, igmp) != 0) { 1507 m_freem(m); 1508 return (IPPROTO_DONE); 1509 } 1510 break; 1511 1512 case IGMP_VERSION_2: 1513 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1514 if (!V_igmp_v2enable) 1515 break; 1516 if (igmp_input_v2_query(ifp, ip, igmp) != 0) { 1517 m_freem(m); 1518 return (IPPROTO_DONE); 1519 } 1520 break; 1521 1522 case IGMP_VERSION_3: { 1523 struct igmpv3 *igmpv3; 1524 uint16_t igmpv3len; 1525 uint16_t nsrc; 1526 1527 IGMPSTAT_INC(igps_rcv_v3_queries); 1528 igmpv3 = (struct igmpv3 *)igmp; 1529 /* 1530 * Validate length based on source count. 1531 */ 1532 nsrc = ntohs(igmpv3->igmp_numsrc); 1533 if (nsrc * sizeof(in_addr_t) > 1534 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) { 1535 IGMPSTAT_INC(igps_rcv_tooshort); 1536 m_freem(m); 1537 return (IPPROTO_DONE); 1538 } 1539 /* 1540 * m_pullup() may modify m, so pullup in 1541 * this scope. 1542 */ 1543 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN + 1544 sizeof(struct in_addr) * nsrc; 1545 if ((!M_WRITABLE(m) || 1546 m->m_len < igmpv3len) && 1547 (m = m_pullup(m, igmpv3len)) == NULL) { 1548 IGMPSTAT_INC(igps_rcv_tooshort); 1549 return (IPPROTO_DONE); 1550 } 1551 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *) 1552 + iphlen); 1553 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) { 1554 m_freem(m); 1555 return (IPPROTO_DONE); 1556 } 1557 } 1558 break; 1559 } 1560 break; 1561 1562 case IGMP_v1_HOST_MEMBERSHIP_REPORT: 1563 if (!V_igmp_v1enable) 1564 break; 1565 if (igmp_input_v1_report(ifp, ip, igmp) != 0) { 1566 m_freem(m); 1567 return (IPPROTO_DONE); 1568 } 1569 break; 1570 1571 case IGMP_v2_HOST_MEMBERSHIP_REPORT: 1572 if (!V_igmp_v2enable) 1573 break; 1574 if (!ip_checkrouteralert(m)) 1575 IGMPSTAT_INC(igps_rcv_nora); 1576 if (igmp_input_v2_report(ifp, ip, igmp) != 0) { 1577 m_freem(m); 1578 return (IPPROTO_DONE); 1579 } 1580 break; 1581 1582 case IGMP_v3_HOST_MEMBERSHIP_REPORT: 1583 /* 1584 * Hosts do not need to process IGMPv3 membership reports, 1585 * as report suppression is no longer required. 1586 */ 1587 if (!ip_checkrouteralert(m)) 1588 IGMPSTAT_INC(igps_rcv_nora); 1589 break; 1590 1591 default: 1592 break; 1593 } 1594 1595 /* 1596 * Pass all valid IGMP packets up to any process(es) listening on a 1597 * raw IGMP socket. 1598 */ 1599 *mp = m; 1600 return (rip_input(mp, offp, proto)); 1601} 1602 1603 1604/* 1605 * Fast timeout handler (global). 1606 * VIMAGE: Timeout handlers are expected to service all vimages. 1607 */ 1608void 1609igmp_fasttimo(void) 1610{ 1611 VNET_ITERATOR_DECL(vnet_iter); 1612 1613 VNET_LIST_RLOCK_NOSLEEP(); 1614 VNET_FOREACH(vnet_iter) { 1615 CURVNET_SET(vnet_iter); 1616 igmp_fasttimo_vnet(); 1617 CURVNET_RESTORE(); 1618 } 1619 VNET_LIST_RUNLOCK_NOSLEEP(); 1620} 1621 1622/* 1623 * Fast timeout handler (per-vnet). 1624 * Sends are shuffled off to a netisr to deal with Giant. 1625 * 1626 * VIMAGE: Assume caller has set up our curvnet. 1627 */ 1628static void 1629igmp_fasttimo_vnet(void) 1630{ 1631 struct mbufq scq; /* State-change packets */ 1632 struct mbufq qrq; /* Query response packets */ 1633 struct ifnet *ifp; 1634 struct igmp_ifsoftc *igi; 1635 struct ifmultiaddr *ifma, *next; 1636 struct in_multi *inm; 1637 struct in_multi_head inm_free_tmp; 1638 int loop, uri_fasthz; 1639 1640 loop = 0; 1641 uri_fasthz = 0; 1642 1643 /* 1644 * Quick check to see if any work needs to be done, in order to 1645 * minimize the overhead of fasttimo processing. 1646 * SMPng: XXX Unlocked reads. 1647 */ 1648 if (!V_current_state_timers_running && 1649 !V_interface_timers_running && 1650 !V_state_change_timers_running) 1651 return; 1652 1653 SLIST_INIT(&inm_free_tmp); 1654 IN_MULTI_LIST_LOCK(); 1655 IGMP_LOCK(); 1656 1657 /* 1658 * IGMPv3 General Query response timer processing. 1659 */ 1660 if (V_interface_timers_running) { 1661 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__); 1662 1663 V_interface_timers_running = 0; 1664 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1665 if (igi->igi_v3_timer == 0) { 1666 /* Do nothing. */ 1667 } else if (--igi->igi_v3_timer == 0) { 1668 igmp_v3_dispatch_general_query(igi); 1669 } else { 1670 V_interface_timers_running = 1; 1671 } 1672 } 1673 } 1674 1675 if (!V_current_state_timers_running && 1676 !V_state_change_timers_running) 1677 goto out_locked; 1678 1679 V_current_state_timers_running = 0; 1680 V_state_change_timers_running = 0; 1681 1682 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__); 1683 1684 /* 1685 * IGMPv1/v2/v3 host report and state-change timer processing. 1686 * Note: Processing a v3 group timer may remove a node. 1687 */ 1688 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1689 ifp = igi->igi_ifp; 1690 1691 if (igi->igi_version == IGMP_VERSION_3) { 1692 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; 1693 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri * 1694 PR_FASTHZ); 1695 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS); 1696 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS); 1697 } 1698 1699 IF_ADDR_WLOCK(ifp); 1700 restart: 1701 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { 1702 if (ifma->ifma_addr->sa_family != AF_INET || 1703 ifma->ifma_protospec == NULL) 1704 continue; 1705 inm = (struct in_multi *)ifma->ifma_protospec; 1706 switch (igi->igi_version) { 1707 case IGMP_VERSION_1: 1708 case IGMP_VERSION_2: 1709 igmp_v1v2_process_group_timer(inm, 1710 igi->igi_version); 1711 break; 1712 case IGMP_VERSION_3: 1713 igmp_v3_process_group_timers(&inm_free_tmp, &qrq, 1714 &scq, inm, uri_fasthz); 1715 break; 1716 } 1717 if (__predict_false(ifma_restart)) { 1718 ifma_restart = false; 1719 goto restart; 1720 } 1721 } 1722 IF_ADDR_WUNLOCK(ifp); 1723 1724 if (igi->igi_version == IGMP_VERSION_3) { 1725 igmp_dispatch_queue(&qrq, 0, loop); 1726 igmp_dispatch_queue(&scq, 0, loop); 1727 1728 /* 1729 * Free the in_multi reference(s) for this 1730 * IGMP lifecycle. 1731 */ 1732 inm_release_list_deferred(&inm_free_tmp); 1733 } 1734 } 1735 1736out_locked: 1737 IGMP_UNLOCK(); 1738 IN_MULTI_LIST_UNLOCK(); 1739} 1740 1741/* 1742 * Update host report group timer for IGMPv1/v2. 1743 * Will update the global pending timer flags. 1744 */ 1745static void 1746igmp_v1v2_process_group_timer(struct in_multi *inm, const int version) 1747{ 1748 int report_timer_expired; 1749 1750 IN_MULTI_LIST_LOCK_ASSERT(); 1751 IGMP_LOCK_ASSERT(); 1752 1753 if (inm->inm_timer == 0) { 1754 report_timer_expired = 0; 1755 } else if (--inm->inm_timer == 0) { 1756 report_timer_expired = 1; 1757 } else { 1758 V_current_state_timers_running = 1; 1759 return; 1760 } 1761 1762 switch (inm->inm_state) { 1763 case IGMP_NOT_MEMBER: 1764 case IGMP_SILENT_MEMBER: 1765 case IGMP_IDLE_MEMBER: 1766 case IGMP_LAZY_MEMBER: 1767 case IGMP_SLEEPING_MEMBER: 1768 case IGMP_AWAKENING_MEMBER: 1769 break; 1770 case IGMP_REPORTING_MEMBER: 1771 if (report_timer_expired) { 1772 inm->inm_state = IGMP_IDLE_MEMBER; 1773 (void)igmp_v1v2_queue_report(inm, 1774 (version == IGMP_VERSION_2) ? 1775 IGMP_v2_HOST_MEMBERSHIP_REPORT : 1776 IGMP_v1_HOST_MEMBERSHIP_REPORT); 1777 } 1778 break; 1779 case IGMP_G_QUERY_PENDING_MEMBER: 1780 case IGMP_SG_QUERY_PENDING_MEMBER: 1781 case IGMP_LEAVING_MEMBER: 1782 break; 1783 } 1784} 1785 1786/* 1787 * Update a group's timers for IGMPv3. 1788 * Will update the global pending timer flags. 1789 * Note: Unlocked read from igi. 1790 */ 1791static void 1792igmp_v3_process_group_timers(struct in_multi_head *inmh, 1793 struct mbufq *qrq, struct mbufq *scq, 1794 struct in_multi *inm, const int uri_fasthz) 1795{ 1796 int query_response_timer_expired; 1797 int state_change_retransmit_timer_expired; 1798 1799 IN_MULTI_LIST_LOCK_ASSERT(); 1800 IGMP_LOCK_ASSERT(); 1801 1802 query_response_timer_expired = 0; 1803 state_change_retransmit_timer_expired = 0; 1804 1805 /* 1806 * During a transition from v1/v2 compatibility mode back to v3, 1807 * a group record in REPORTING state may still have its group 1808 * timer active. This is a no-op in this function; it is easier 1809 * to deal with it here than to complicate the slow-timeout path. 1810 */ 1811 if (inm->inm_timer == 0) { 1812 query_response_timer_expired = 0; 1813 } else if (--inm->inm_timer == 0) { 1814 query_response_timer_expired = 1; 1815 } else { 1816 V_current_state_timers_running = 1; 1817 } 1818 1819 if (inm->inm_sctimer == 0) { 1820 state_change_retransmit_timer_expired = 0; 1821 } else if (--inm->inm_sctimer == 0) { 1822 state_change_retransmit_timer_expired = 1; 1823 } else { 1824 V_state_change_timers_running = 1; 1825 } 1826 1827 /* We are in fasttimo, so be quick about it. */ 1828 if (!state_change_retransmit_timer_expired && 1829 !query_response_timer_expired) 1830 return; 1831 1832 switch (inm->inm_state) { 1833 case IGMP_NOT_MEMBER: 1834 case IGMP_SILENT_MEMBER: 1835 case IGMP_SLEEPING_MEMBER: 1836 case IGMP_LAZY_MEMBER: 1837 case IGMP_AWAKENING_MEMBER: 1838 case IGMP_IDLE_MEMBER: 1839 break; 1840 case IGMP_G_QUERY_PENDING_MEMBER: 1841 case IGMP_SG_QUERY_PENDING_MEMBER: 1842 /* 1843 * Respond to a previously pending Group-Specific 1844 * or Group-and-Source-Specific query by enqueueing 1845 * the appropriate Current-State report for 1846 * immediate transmission. 1847 */ 1848 if (query_response_timer_expired) { 1849 int retval __unused; 1850 1851 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1, 1852 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)); 1853 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 1854 __func__, retval); 1855 inm->inm_state = IGMP_REPORTING_MEMBER; 1856 /* XXX Clear recorded sources for next time. */ 1857 inm_clear_recorded(inm); 1858 } 1859 /* FALLTHROUGH */ 1860 case IGMP_REPORTING_MEMBER: 1861 case IGMP_LEAVING_MEMBER: 1862 if (state_change_retransmit_timer_expired) { 1863 /* 1864 * State-change retransmission timer fired. 1865 * If there are any further pending retransmissions, 1866 * set the global pending state-change flag, and 1867 * reset the timer. 1868 */ 1869 if (--inm->inm_scrv > 0) { 1870 inm->inm_sctimer = uri_fasthz; 1871 V_state_change_timers_running = 1; 1872 } 1873 /* 1874 * Retransmit the previously computed state-change 1875 * report. If there are no further pending 1876 * retransmissions, the mbuf queue will be consumed. 1877 * Update T0 state to T1 as we have now sent 1878 * a state-change. 1879 */ 1880 (void)igmp_v3_merge_state_changes(inm, scq); 1881 1882 inm_commit(inm); 1883 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 1884 ntohl(inm->inm_addr.s_addr), 1885 inm->inm_ifp->if_xname); 1886 1887 /* 1888 * If we are leaving the group for good, make sure 1889 * we release IGMP's reference to it. 1890 * This release must be deferred using a SLIST, 1891 * as we are called from a loop which traverses 1892 * the in_ifmultiaddr TAILQ. 1893 */ 1894 if (inm->inm_state == IGMP_LEAVING_MEMBER && 1895 inm->inm_scrv == 0) { 1896 inm->inm_state = IGMP_NOT_MEMBER; 1897 inm_rele_locked(inmh, inm); 1898 } 1899 } 1900 break; 1901 } 1902} 1903 1904 1905/* 1906 * Suppress a group's pending response to a group or source/group query. 1907 * 1908 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency. 1909 * Do NOT update ST1/ST0 as this operation merely suppresses 1910 * the currently pending group record. 1911 * Do NOT suppress the response to a general query. It is possible but 1912 * it would require adding another state or flag. 1913 */ 1914static void 1915igmp_v3_suppress_group_record(struct in_multi *inm) 1916{ 1917 1918 IN_MULTI_LIST_LOCK_ASSERT(); 1919 1920 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3, 1921 ("%s: not IGMPv3 mode on link", __func__)); 1922 1923 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER || 1924 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER) 1925 return; 1926 1927 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) 1928 inm_clear_recorded(inm); 1929 1930 inm->inm_timer = 0; 1931 inm->inm_state = IGMP_REPORTING_MEMBER; 1932} 1933 1934/* 1935 * Switch to a different IGMP version on the given interface, 1936 * as per Section 7.2.1. 1937 */ 1938static void 1939igmp_set_version(struct igmp_ifsoftc *igi, const int version) 1940{ 1941 int old_version_timer; 1942 1943 IGMP_LOCK_ASSERT(); 1944 1945 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__, 1946 version, igi->igi_ifp, igi->igi_ifp->if_xname); 1947 1948 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) { 1949 /* 1950 * Compute the "Older Version Querier Present" timer as per 1951 * Section 8.12. 1952 */ 1953 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri; 1954 old_version_timer *= PR_SLOWHZ; 1955 1956 if (version == IGMP_VERSION_1) { 1957 igi->igi_v1_timer = old_version_timer; 1958 igi->igi_v2_timer = 0; 1959 } else if (version == IGMP_VERSION_2) { 1960 igi->igi_v1_timer = 0; 1961 igi->igi_v2_timer = old_version_timer; 1962 } 1963 } 1964 1965 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 1966 if (igi->igi_version != IGMP_VERSION_2) { 1967 igi->igi_version = IGMP_VERSION_2; 1968 igmp_v3_cancel_link_timers(igi); 1969 } 1970 } else if (igi->igi_v1_timer > 0) { 1971 if (igi->igi_version != IGMP_VERSION_1) { 1972 igi->igi_version = IGMP_VERSION_1; 1973 igmp_v3_cancel_link_timers(igi); 1974 } 1975 } 1976} 1977 1978/* 1979 * Cancel pending IGMPv3 timers for the given link and all groups 1980 * joined on it; state-change, general-query, and group-query timers. 1981 * 1982 * Only ever called on a transition from v3 to Compatibility mode. Kill 1983 * the timers stone dead (this may be expensive for large N groups), they 1984 * will be restarted if Compatibility Mode deems that they must be due to 1985 * query processing. 1986 */ 1987static void 1988igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi) 1989{ 1990 struct ifmultiaddr *ifma, *ifmatmp; 1991 struct ifnet *ifp; 1992 struct in_multi *inm; 1993 struct in_multi_head inm_free_tmp; 1994 1995 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__, 1996 igi->igi_ifp, igi->igi_ifp->if_xname); 1997 1998 IN_MULTI_LIST_LOCK_ASSERT(); 1999 IGMP_LOCK_ASSERT(); 2000 SLIST_INIT(&inm_free_tmp); 2001 2002 /* 2003 * Stop the v3 General Query Response on this link stone dead. 2004 * If fasttimo is woken up due to V_interface_timers_running, 2005 * the flag will be cleared if there are no pending link timers. 2006 */ 2007 igi->igi_v3_timer = 0; 2008 2009 /* 2010 * Now clear the current-state and state-change report timers 2011 * for all memberships scoped to this link. 2012 */ 2013 ifp = igi->igi_ifp; 2014 IF_ADDR_WLOCK(ifp); 2015 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, ifmatmp) { 2016 if (ifma->ifma_addr->sa_family != AF_INET || 2017 ifma->ifma_protospec == NULL) 2018 continue; 2019 inm = (struct in_multi *)ifma->ifma_protospec; 2020 switch (inm->inm_state) { 2021 case IGMP_NOT_MEMBER: 2022 case IGMP_SILENT_MEMBER: 2023 case IGMP_IDLE_MEMBER: 2024 case IGMP_LAZY_MEMBER: 2025 case IGMP_SLEEPING_MEMBER: 2026 case IGMP_AWAKENING_MEMBER: 2027 /* 2028 * These states are either not relevant in v3 mode, 2029 * or are unreported. Do nothing. 2030 */ 2031 break; 2032 case IGMP_LEAVING_MEMBER: 2033 /* 2034 * If we are leaving the group and switching to 2035 * compatibility mode, we need to release the final 2036 * reference held for issuing the INCLUDE {}, and 2037 * transition to REPORTING to ensure the host leave 2038 * message is sent upstream to the old querier -- 2039 * transition to NOT would lose the leave and race. 2040 */ 2041 inm_rele_locked(&inm_free_tmp, inm); 2042 /* FALLTHROUGH */ 2043 case IGMP_G_QUERY_PENDING_MEMBER: 2044 case IGMP_SG_QUERY_PENDING_MEMBER: 2045 inm_clear_recorded(inm); 2046 /* FALLTHROUGH */ 2047 case IGMP_REPORTING_MEMBER: 2048 inm->inm_state = IGMP_REPORTING_MEMBER; 2049 break; 2050 } 2051 /* 2052 * Always clear state-change and group report timers. 2053 * Free any pending IGMPv3 state-change records. 2054 */ 2055 inm->inm_sctimer = 0; 2056 inm->inm_timer = 0; 2057 mbufq_drain(&inm->inm_scq); 2058 } 2059 IF_ADDR_WUNLOCK(ifp); 2060 2061 inm_release_list_deferred(&inm_free_tmp); 2062} 2063 2064/* 2065 * Update the Older Version Querier Present timers for a link. 2066 * See Section 7.2.1 of RFC 3376. 2067 */ 2068static void 2069igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi) 2070{ 2071 2072 IGMP_LOCK_ASSERT(); 2073 2074 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) { 2075 /* 2076 * IGMPv1 and IGMPv2 Querier Present timers expired. 2077 * 2078 * Revert to IGMPv3. 2079 */ 2080 if (igi->igi_version != IGMP_VERSION_3) { 2081 CTR5(KTR_IGMPV3, 2082 "%s: transition from v%d -> v%d on %p(%s)", 2083 __func__, igi->igi_version, IGMP_VERSION_3, 2084 igi->igi_ifp, igi->igi_ifp->if_xname); 2085 igi->igi_version = IGMP_VERSION_3; 2086 } 2087 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 2088 /* 2089 * IGMPv1 Querier Present timer expired, 2090 * IGMPv2 Querier Present timer running. 2091 * If IGMPv2 was disabled since last timeout, 2092 * revert to IGMPv3. 2093 * If IGMPv2 is enabled, revert to IGMPv2. 2094 */ 2095 if (!V_igmp_v2enable) { 2096 CTR5(KTR_IGMPV3, 2097 "%s: transition from v%d -> v%d on %p(%s)", 2098 __func__, igi->igi_version, IGMP_VERSION_3, 2099 igi->igi_ifp, igi->igi_ifp->if_xname); 2100 igi->igi_v2_timer = 0; 2101 igi->igi_version = IGMP_VERSION_3; 2102 } else { 2103 --igi->igi_v2_timer; 2104 if (igi->igi_version != IGMP_VERSION_2) { 2105 CTR5(KTR_IGMPV3, 2106 "%s: transition from v%d -> v%d on %p(%s)", 2107 __func__, igi->igi_version, IGMP_VERSION_2, 2108 igi->igi_ifp, igi->igi_ifp->if_xname); 2109 igi->igi_version = IGMP_VERSION_2; 2110 igmp_v3_cancel_link_timers(igi); 2111 } 2112 } 2113 } else if (igi->igi_v1_timer > 0) { 2114 /* 2115 * IGMPv1 Querier Present timer running. 2116 * Stop IGMPv2 timer if running. 2117 * 2118 * If IGMPv1 was disabled since last timeout, 2119 * revert to IGMPv3. 2120 * If IGMPv1 is enabled, reset IGMPv2 timer if running. 2121 */ 2122 if (!V_igmp_v1enable) { 2123 CTR5(KTR_IGMPV3, 2124 "%s: transition from v%d -> v%d on %p(%s)", 2125 __func__, igi->igi_version, IGMP_VERSION_3, 2126 igi->igi_ifp, igi->igi_ifp->if_xname); 2127 igi->igi_v1_timer = 0; 2128 igi->igi_version = IGMP_VERSION_3; 2129 } else { 2130 --igi->igi_v1_timer; 2131 } 2132 if (igi->igi_v2_timer > 0) { 2133 CTR3(KTR_IGMPV3, 2134 "%s: cancel v2 timer on %p(%s)", 2135 __func__, igi->igi_ifp, igi->igi_ifp->if_xname); 2136 igi->igi_v2_timer = 0; 2137 } 2138 } 2139} 2140 2141/* 2142 * Global slowtimo handler. 2143 * VIMAGE: Timeout handlers are expected to service all vimages. 2144 */ 2145void 2146igmp_slowtimo(void) 2147{ 2148 VNET_ITERATOR_DECL(vnet_iter); 2149 2150 VNET_LIST_RLOCK_NOSLEEP(); 2151 VNET_FOREACH(vnet_iter) { 2152 CURVNET_SET(vnet_iter); 2153 igmp_slowtimo_vnet(); 2154 CURVNET_RESTORE(); 2155 } 2156 VNET_LIST_RUNLOCK_NOSLEEP(); 2157} 2158 2159/* 2160 * Per-vnet slowtimo handler. 2161 */ 2162static void 2163igmp_slowtimo_vnet(void) 2164{ 2165 struct igmp_ifsoftc *igi; 2166 2167 IGMP_LOCK(); 2168 2169 LIST_FOREACH(igi, &V_igi_head, igi_link) { 2170 igmp_v1v2_process_querier_timers(igi); 2171 } 2172 2173 IGMP_UNLOCK(); 2174} 2175 2176/* 2177 * Dispatch an IGMPv1/v2 host report or leave message. 2178 * These are always small enough to fit inside a single mbuf. 2179 */ 2180static int 2181igmp_v1v2_queue_report(struct in_multi *inm, const int type) 2182{ 2183 struct ifnet *ifp; 2184 struct igmp *igmp; 2185 struct ip *ip; 2186 struct mbuf *m; 2187 2188 IN_MULTI_LIST_LOCK_ASSERT(); 2189 IGMP_LOCK_ASSERT(); 2190 2191 ifp = inm->inm_ifp; 2192 2193 m = m_gethdr(M_NOWAIT, MT_DATA); 2194 if (m == NULL) 2195 return (ENOMEM); 2196 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp)); 2197 2198 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp); 2199 2200 m->m_data += sizeof(struct ip); 2201 m->m_len = sizeof(struct igmp); 2202 2203 igmp = mtod(m, struct igmp *); 2204 igmp->igmp_type = type; 2205 igmp->igmp_code = 0; 2206 igmp->igmp_group = inm->inm_addr; 2207 igmp->igmp_cksum = 0; 2208 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp)); 2209 2210 m->m_data -= sizeof(struct ip); 2211 m->m_len += sizeof(struct ip); 2212 2213 ip = mtod(m, struct ip *); 2214 ip->ip_tos = 0; 2215 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp)); 2216 ip->ip_off = 0; 2217 ip->ip_p = IPPROTO_IGMP; 2218 ip->ip_src.s_addr = INADDR_ANY; 2219 2220 if (type == IGMP_HOST_LEAVE_MESSAGE) 2221 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP); 2222 else 2223 ip->ip_dst = inm->inm_addr; 2224 2225 igmp_save_context(m, ifp); 2226 2227 m->m_flags |= M_IGMPV2; 2228 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK) 2229 m->m_flags |= M_IGMP_LOOP; 2230 2231 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m); 2232 netisr_dispatch(NETISR_IGMP, m); 2233 2234 return (0); 2235} 2236 2237/* 2238 * Process a state change from the upper layer for the given IPv4 group. 2239 * 2240 * Each socket holds a reference on the in_multi in its own ip_moptions. 2241 * The socket layer will have made the necessary updates to.the group 2242 * state, it is now up to IGMP to issue a state change report if there 2243 * has been any change between T0 (when the last state-change was issued) 2244 * and T1 (now). 2245 * 2246 * We use the IGMPv3 state machine at group level. The IGMP module 2247 * however makes the decision as to which IGMP protocol version to speak. 2248 * A state change *from* INCLUDE {} always means an initial join. 2249 * A state change *to* INCLUDE {} always means a final leave. 2250 * 2251 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can 2252 * save ourselves a bunch of work; any exclusive mode groups need not 2253 * compute source filter lists. 2254 * 2255 * VIMAGE: curvnet should have been set by caller, as this routine 2256 * is called from the socket option handlers. 2257 */ 2258int 2259igmp_change_state(struct in_multi *inm) 2260{ 2261 struct igmp_ifsoftc *igi; 2262 struct ifnet *ifp; 2263 int error; 2264 2265 error = 0; 2266 IN_MULTI_LOCK_ASSERT(); 2267 /* 2268 * Try to detect if the upper layer just asked us to change state 2269 * for an interface which has now gone away. 2270 */ 2271 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 2272 ifp = inm->inm_ifma->ifma_ifp; 2273 /* 2274 * Sanity check that netinet's notion of ifp is the 2275 * same as net's. 2276 */ 2277 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 2278 2279 IGMP_LOCK(); 2280 2281 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 2282 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 2283 2284 /* 2285 * If we detect a state transition to or from MCAST_UNDEFINED 2286 * for this group, then we are starting or finishing an IGMP 2287 * life cycle for this group. 2288 */ 2289 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) { 2290 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__, 2291 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode); 2292 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) { 2293 CTR1(KTR_IGMPV3, "%s: initial join", __func__); 2294 error = igmp_initial_join(inm, igi); 2295 goto out_locked; 2296 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) { 2297 CTR1(KTR_IGMPV3, "%s: final leave", __func__); 2298 igmp_final_leave(inm, igi); 2299 goto out_locked; 2300 } 2301 } else { 2302 CTR1(KTR_IGMPV3, "%s: filter set change", __func__); 2303 } 2304 2305 error = igmp_handle_state_change(inm, igi); 2306 2307out_locked: 2308 IGMP_UNLOCK(); 2309 return (error); 2310} 2311 2312/* 2313 * Perform the initial join for an IGMP group. 2314 * 2315 * When joining a group: 2316 * If the group should have its IGMP traffic suppressed, do nothing. 2317 * IGMPv1 starts sending IGMPv1 host membership reports. 2318 * IGMPv2 starts sending IGMPv2 host membership reports. 2319 * IGMPv3 will schedule an IGMPv3 state-change report containing the 2320 * initial state of the membership. 2321 */ 2322static int 2323igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi) 2324{ 2325 struct ifnet *ifp; 2326 struct mbufq *mq; 2327 int error, retval, syncstates; 2328 2329 CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__, 2330 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); 2331 2332 error = 0; 2333 syncstates = 1; 2334 2335 ifp = inm->inm_ifp; 2336 2337 IN_MULTI_LOCK_ASSERT(); 2338 IGMP_LOCK_ASSERT(); 2339 2340 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); 2341 2342 /* 2343 * Groups joined on loopback or marked as 'not reported', 2344 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and 2345 * are never reported in any IGMP protocol exchanges. 2346 * All other groups enter the appropriate IGMP state machine 2347 * for the version in use on this link. 2348 * A link marked as IGIF_SILENT causes IGMP to be completely 2349 * disabled for the link. 2350 */ 2351 if ((ifp->if_flags & IFF_LOOPBACK) || 2352 (igi->igi_flags & IGIF_SILENT) || 2353 !igmp_isgroupreported(inm->inm_addr)) { 2354 CTR1(KTR_IGMPV3, 2355"%s: not kicking state machine for silent group", __func__); 2356 inm->inm_state = IGMP_SILENT_MEMBER; 2357 inm->inm_timer = 0; 2358 } else { 2359 /* 2360 * Deal with overlapping in_multi lifecycle. 2361 * If this group was LEAVING, then make sure 2362 * we drop the reference we picked up to keep the 2363 * group around for the final INCLUDE {} enqueue. 2364 */ 2365 if (igi->igi_version == IGMP_VERSION_3 && 2366 inm->inm_state == IGMP_LEAVING_MEMBER) { 2367 MPASS(inm->inm_refcount > 1); 2368 inm_rele_locked(NULL, inm); 2369 } 2370 inm->inm_state = IGMP_REPORTING_MEMBER; 2371 2372 switch (igi->igi_version) { 2373 case IGMP_VERSION_1: 2374 case IGMP_VERSION_2: 2375 inm->inm_state = IGMP_IDLE_MEMBER; 2376 error = igmp_v1v2_queue_report(inm, 2377 (igi->igi_version == IGMP_VERSION_2) ? 2378 IGMP_v2_HOST_MEMBERSHIP_REPORT : 2379 IGMP_v1_HOST_MEMBERSHIP_REPORT); 2380 if (error == 0) { 2381 inm->inm_timer = IGMP_RANDOM_DELAY( 2382 IGMP_V1V2_MAX_RI * PR_FASTHZ); 2383 V_current_state_timers_running = 1; 2384 } 2385 break; 2386 2387 case IGMP_VERSION_3: 2388 /* 2389 * Defer update of T0 to T1, until the first copy 2390 * of the state change has been transmitted. 2391 */ 2392 syncstates = 0; 2393 2394 /* 2395 * Immediately enqueue a State-Change Report for 2396 * this interface, freeing any previous reports. 2397 * Don't kick the timers if there is nothing to do, 2398 * or if an error occurred. 2399 */ 2400 mq = &inm->inm_scq; 2401 mbufq_drain(mq); 2402 retval = igmp_v3_enqueue_group_record(mq, inm, 1, 2403 0, 0); 2404 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 2405 __func__, retval); 2406 if (retval <= 0) { 2407 error = retval * -1; 2408 break; 2409 } 2410 2411 /* 2412 * Schedule transmission of pending state-change 2413 * report up to RV times for this link. The timer 2414 * will fire at the next igmp_fasttimo (~200ms), 2415 * giving us an opportunity to merge the reports. 2416 */ 2417 if (igi->igi_flags & IGIF_LOOPBACK) { 2418 inm->inm_scrv = 1; 2419 } else { 2420 KASSERT(igi->igi_rv > 1, 2421 ("%s: invalid robustness %d", __func__, 2422 igi->igi_rv)); 2423 inm->inm_scrv = igi->igi_rv; 2424 } 2425 inm->inm_sctimer = 1; 2426 V_state_change_timers_running = 1; 2427 2428 error = 0; 2429 break; 2430 } 2431 } 2432 2433 /* 2434 * Only update the T0 state if state change is atomic, 2435 * i.e. we don't need to wait for a timer to fire before we 2436 * can consider the state change to have been communicated. 2437 */ 2438 if (syncstates) { 2439 inm_commit(inm); 2440 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 2441 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2442 } 2443 2444 return (error); 2445} 2446 2447/* 2448 * Issue an intermediate state change during the IGMP life-cycle. 2449 */ 2450static int 2451igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi) 2452{ 2453 struct ifnet *ifp; 2454 int retval; 2455 2456 CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__, 2457 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); 2458 2459 ifp = inm->inm_ifp; 2460 2461 IN_MULTI_LIST_LOCK_ASSERT(); 2462 IGMP_LOCK_ASSERT(); 2463 2464 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); 2465 2466 if ((ifp->if_flags & IFF_LOOPBACK) || 2467 (igi->igi_flags & IGIF_SILENT) || 2468 !igmp_isgroupreported(inm->inm_addr) || 2469 (igi->igi_version != IGMP_VERSION_3)) { 2470 if (!igmp_isgroupreported(inm->inm_addr)) { 2471 CTR1(KTR_IGMPV3, 2472"%s: not kicking state machine for silent group", __func__); 2473 } 2474 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__); 2475 inm_commit(inm); 2476 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 2477 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2478 return (0); 2479 } 2480 2481 mbufq_drain(&inm->inm_scq); 2482 2483 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0); 2484 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); 2485 if (retval <= 0) 2486 return (-retval); 2487 2488 /* 2489 * If record(s) were enqueued, start the state-change 2490 * report timer for this group. 2491 */ 2492 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv); 2493 inm->inm_sctimer = 1; 2494 V_state_change_timers_running = 1; 2495 2496 return (0); 2497} 2498 2499/* 2500 * Perform the final leave for an IGMP group. 2501 * 2502 * When leaving a group: 2503 * IGMPv1 does nothing. 2504 * IGMPv2 sends a host leave message, if and only if we are the reporter. 2505 * IGMPv3 enqueues a state-change report containing a transition 2506 * to INCLUDE {} for immediate transmission. 2507 */ 2508static void 2509igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi) 2510{ 2511 int syncstates; 2512 2513 syncstates = 1; 2514 2515 CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)", 2516 __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp, 2517 inm->inm_ifp->if_xname); 2518 2519 IN_MULTI_LIST_LOCK_ASSERT(); 2520 IGMP_LOCK_ASSERT(); 2521 2522 switch (inm->inm_state) { 2523 case IGMP_NOT_MEMBER: 2524 case IGMP_SILENT_MEMBER: 2525 case IGMP_LEAVING_MEMBER: 2526 /* Already leaving or left; do nothing. */ 2527 CTR1(KTR_IGMPV3, 2528"%s: not kicking state machine for silent group", __func__); 2529 break; 2530 case IGMP_REPORTING_MEMBER: 2531 case IGMP_IDLE_MEMBER: 2532 case IGMP_G_QUERY_PENDING_MEMBER: 2533 case IGMP_SG_QUERY_PENDING_MEMBER: 2534 if (igi->igi_version == IGMP_VERSION_2) { 2535#ifdef INVARIANTS 2536 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || 2537 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) 2538 panic("%s: IGMPv3 state reached, not IGMPv3 mode", 2539 __func__); 2540#endif 2541 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE); 2542 inm->inm_state = IGMP_NOT_MEMBER; 2543 } else if (igi->igi_version == IGMP_VERSION_3) { 2544 /* 2545 * Stop group timer and all pending reports. 2546 * Immediately enqueue a state-change report 2547 * TO_IN {} to be sent on the next fast timeout, 2548 * giving us an opportunity to merge reports. 2549 */ 2550 mbufq_drain(&inm->inm_scq); 2551 inm->inm_timer = 0; 2552 if (igi->igi_flags & IGIF_LOOPBACK) { 2553 inm->inm_scrv = 1; 2554 } else { 2555 inm->inm_scrv = igi->igi_rv; 2556 } 2557 CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d " 2558 "pending retransmissions.", __func__, 2559 ntohl(inm->inm_addr.s_addr), 2560 inm->inm_ifp->if_xname, inm->inm_scrv); 2561 if (inm->inm_scrv == 0) { 2562 inm->inm_state = IGMP_NOT_MEMBER; 2563 inm->inm_sctimer = 0; 2564 } else { 2565 int retval __unused; 2566 2567 inm_acquire_locked(inm); 2568 2569 retval = igmp_v3_enqueue_group_record( 2570 &inm->inm_scq, inm, 1, 0, 0); 2571 KASSERT(retval != 0, 2572 ("%s: enqueue record = %d", __func__, 2573 retval)); 2574 2575 inm->inm_state = IGMP_LEAVING_MEMBER; 2576 inm->inm_sctimer = 1; 2577 V_state_change_timers_running = 1; 2578 syncstates = 0; 2579 } 2580 break; 2581 } 2582 break; 2583 case IGMP_LAZY_MEMBER: 2584 case IGMP_SLEEPING_MEMBER: 2585 case IGMP_AWAKENING_MEMBER: 2586 /* Our reports are suppressed; do nothing. */ 2587 break; 2588 } 2589 2590 if (syncstates) { 2591 inm_commit(inm); 2592 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 2593 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2594 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 2595 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s", 2596 __func__, ntohl(inm->inm_addr.s_addr), 2597 inm->inm_ifp->if_xname); 2598 } 2599} 2600 2601/* 2602 * Enqueue an IGMPv3 group record to the given output queue. 2603 * 2604 * XXX This function could do with having the allocation code 2605 * split out, and the multiple-tree-walks coalesced into a single 2606 * routine as has been done in igmp_v3_enqueue_filter_change(). 2607 * 2608 * If is_state_change is zero, a current-state record is appended. 2609 * If is_state_change is non-zero, a state-change report is appended. 2610 * 2611 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2612 * If is_group_query is zero, and if there is a packet with free space 2613 * at the tail of the queue, it will be appended to providing there 2614 * is enough free space. 2615 * Otherwise a new mbuf packet chain is allocated. 2616 * 2617 * If is_source_query is non-zero, each source is checked to see if 2618 * it was recorded for a Group-Source query, and will be omitted if 2619 * it is not both in-mode and recorded. 2620 * 2621 * The function will attempt to allocate leading space in the packet 2622 * for the IP/IGMP header to be prepended without fragmenting the chain. 2623 * 2624 * If successful the size of all data appended to the queue is returned, 2625 * otherwise an error code less than zero is returned, or zero if 2626 * no record(s) were appended. 2627 */ 2628static int 2629igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm, 2630 const int is_state_change, const int is_group_query, 2631 const int is_source_query) 2632{ 2633 struct igmp_grouprec ig; 2634 struct igmp_grouprec *pig; 2635 struct ifnet *ifp; 2636 struct ip_msource *ims, *nims; 2637 struct mbuf *m0, *m, *md; 2638 int is_filter_list_change; 2639 int minrec0len, m0srcs, msrcs, nbytes, off; 2640 int record_has_sources; 2641 int now; 2642 int type; 2643 in_addr_t naddr; 2644 uint8_t mode; 2645 2646 IN_MULTI_LIST_LOCK_ASSERT(); 2647 2648 ifp = inm->inm_ifp; 2649 is_filter_list_change = 0; 2650 m = NULL; 2651 m0 = NULL; 2652 m0srcs = 0; 2653 msrcs = 0; 2654 nbytes = 0; 2655 nims = NULL; 2656 record_has_sources = 1; 2657 pig = NULL; 2658 type = IGMP_DO_NOTHING; 2659 mode = inm->inm_st[1].iss_fmode; 2660 2661 /* 2662 * If we did not transition out of ASM mode during t0->t1, 2663 * and there are no source nodes to process, we can skip 2664 * the generation of source records. 2665 */ 2666 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 && 2667 inm->inm_nsrc == 0) 2668 record_has_sources = 0; 2669 2670 if (is_state_change) { 2671 /* 2672 * Queue a state change record. 2673 * If the mode did not change, and there are non-ASM 2674 * listeners or source filters present, 2675 * we potentially need to issue two records for the group. 2676 * If we are transitioning to MCAST_UNDEFINED, we need 2677 * not send any sources. 2678 * If there are ASM listeners, and there was no filter 2679 * mode transition of any kind, do nothing. 2680 */ 2681 if (mode != inm->inm_st[0].iss_fmode) { 2682 if (mode == MCAST_EXCLUDE) { 2683 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE", 2684 __func__); 2685 type = IGMP_CHANGE_TO_EXCLUDE_MODE; 2686 } else { 2687 CTR1(KTR_IGMPV3, "%s: change to INCLUDE", 2688 __func__); 2689 type = IGMP_CHANGE_TO_INCLUDE_MODE; 2690 if (mode == MCAST_UNDEFINED) 2691 record_has_sources = 0; 2692 } 2693 } else { 2694 if (record_has_sources) { 2695 is_filter_list_change = 1; 2696 } else { 2697 type = IGMP_DO_NOTHING; 2698 } 2699 } 2700 } else { 2701 /* 2702 * Queue a current state record. 2703 */ 2704 if (mode == MCAST_EXCLUDE) { 2705 type = IGMP_MODE_IS_EXCLUDE; 2706 } else if (mode == MCAST_INCLUDE) { 2707 type = IGMP_MODE_IS_INCLUDE; 2708 KASSERT(inm->inm_st[1].iss_asm == 0, 2709 ("%s: inm %p is INCLUDE but ASM count is %d", 2710 __func__, inm, inm->inm_st[1].iss_asm)); 2711 } 2712 } 2713 2714 /* 2715 * Generate the filter list changes using a separate function. 2716 */ 2717 if (is_filter_list_change) 2718 return (igmp_v3_enqueue_filter_change(mq, inm)); 2719 2720 if (type == IGMP_DO_NOTHING) { 2721 CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__, 2722 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2723 return (0); 2724 } 2725 2726 /* 2727 * If any sources are present, we must be able to fit at least 2728 * one in the trailing space of the tail packet's mbuf, 2729 * ideally more. 2730 */ 2731 minrec0len = sizeof(struct igmp_grouprec); 2732 if (record_has_sources) 2733 minrec0len += sizeof(in_addr_t); 2734 2735 CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__, 2736 igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr), 2737 inm->inm_ifp->if_xname); 2738 2739 /* 2740 * Check if we have a packet in the tail of the queue for this 2741 * group into which the first group record for this group will fit. 2742 * Otherwise allocate a new packet. 2743 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT. 2744 * Note: Group records for G/GSR query responses MUST be sent 2745 * in their own packet. 2746 */ 2747 m0 = mbufq_last(mq); 2748 if (!is_group_query && 2749 m0 != NULL && 2750 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) && 2751 (m0->m_pkthdr.len + minrec0len) < 2752 (ifp->if_mtu - IGMP_LEADINGSPACE)) { 2753 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2754 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2755 m = m0; 2756 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__); 2757 } else { 2758 if (mbufq_full(mq)) { 2759 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); 2760 return (-ENOMEM); 2761 } 2762 m = NULL; 2763 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 2764 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2765 if (!is_state_change && !is_group_query) { 2766 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2767 if (m) 2768 m->m_data += IGMP_LEADINGSPACE; 2769 } 2770 if (m == NULL) { 2771 m = m_gethdr(M_NOWAIT, MT_DATA); 2772 if (m) 2773 M_ALIGN(m, IGMP_LEADINGSPACE); 2774 } 2775 if (m == NULL) 2776 return (-ENOMEM); 2777 2778 igmp_save_context(m, ifp); 2779 2780 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__); 2781 } 2782 2783 /* 2784 * Append group record. 2785 * If we have sources, we don't know how many yet. 2786 */ 2787 ig.ig_type = type; 2788 ig.ig_datalen = 0; 2789 ig.ig_numsrc = 0; 2790 ig.ig_group = inm->inm_addr; 2791 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { 2792 if (m != m0) 2793 m_freem(m); 2794 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); 2795 return (-ENOMEM); 2796 } 2797 nbytes += sizeof(struct igmp_grouprec); 2798 2799 /* 2800 * Append as many sources as will fit in the first packet. 2801 * If we are appending to a new packet, the chain allocation 2802 * may potentially use clusters; use m_getptr() in this case. 2803 * If we are appending to an existing packet, we need to obtain 2804 * a pointer to the group record after m_append(), in case a new 2805 * mbuf was allocated. 2806 * Only append sources which are in-mode at t1. If we are 2807 * transitioning to MCAST_UNDEFINED state on the group, do not 2808 * include source entries. 2809 * Only report recorded sources in our filter set when responding 2810 * to a group-source query. 2811 */ 2812 if (record_has_sources) { 2813 if (m == m0) { 2814 md = m_last(m); 2815 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + 2816 md->m_len - nbytes); 2817 } else { 2818 md = m_getptr(m, 0, &off); 2819 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + 2820 off); 2821 } 2822 msrcs = 0; 2823 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) { 2824 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2825 ims->ims_haddr); 2826 now = ims_get_mode(inm, ims, 1); 2827 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now); 2828 if ((now != mode) || 2829 (now == mode && mode == MCAST_UNDEFINED)) { 2830 CTR1(KTR_IGMPV3, "%s: skip node", __func__); 2831 continue; 2832 } 2833 if (is_source_query && ims->ims_stp == 0) { 2834 CTR1(KTR_IGMPV3, "%s: skip unrecorded node", 2835 __func__); 2836 continue; 2837 } 2838 CTR1(KTR_IGMPV3, "%s: append node", __func__); 2839 naddr = htonl(ims->ims_haddr); 2840 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { 2841 if (m != m0) 2842 m_freem(m); 2843 CTR1(KTR_IGMPV3, "%s: m_append() failed.", 2844 __func__); 2845 return (-ENOMEM); 2846 } 2847 nbytes += sizeof(in_addr_t); 2848 ++msrcs; 2849 if (msrcs == m0srcs) 2850 break; 2851 } 2852 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__, 2853 msrcs); 2854 pig->ig_numsrc = htons(msrcs); 2855 nbytes += (msrcs * sizeof(in_addr_t)); 2856 } 2857 2858 if (is_source_query && msrcs == 0) { 2859 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__); 2860 if (m != m0) 2861 m_freem(m); 2862 return (0); 2863 } 2864 2865 /* 2866 * We are good to go with first packet. 2867 */ 2868 if (m != m0) { 2869 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__); 2870 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2871 mbufq_enqueue(mq, m); 2872 } else 2873 m->m_pkthdr.PH_vt.vt_nrecs++; 2874 2875 /* 2876 * No further work needed if no source list in packet(s). 2877 */ 2878 if (!record_has_sources) 2879 return (nbytes); 2880 2881 /* 2882 * Whilst sources remain to be announced, we need to allocate 2883 * a new packet and fill out as many sources as will fit. 2884 * Always try for a cluster first. 2885 */ 2886 while (nims != NULL) { 2887 if (mbufq_full(mq)) { 2888 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); 2889 return (-ENOMEM); 2890 } 2891 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2892 if (m) 2893 m->m_data += IGMP_LEADINGSPACE; 2894 if (m == NULL) { 2895 m = m_gethdr(M_NOWAIT, MT_DATA); 2896 if (m) 2897 M_ALIGN(m, IGMP_LEADINGSPACE); 2898 } 2899 if (m == NULL) 2900 return (-ENOMEM); 2901 igmp_save_context(m, ifp); 2902 md = m_getptr(m, 0, &off); 2903 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); 2904 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__); 2905 2906 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { 2907 if (m != m0) 2908 m_freem(m); 2909 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); 2910 return (-ENOMEM); 2911 } 2912 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2913 nbytes += sizeof(struct igmp_grouprec); 2914 2915 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 2916 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2917 2918 msrcs = 0; 2919 RB_FOREACH_FROM(ims, ip_msource_tree, nims) { 2920 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2921 ims->ims_haddr); 2922 now = ims_get_mode(inm, ims, 1); 2923 if ((now != mode) || 2924 (now == mode && mode == MCAST_UNDEFINED)) { 2925 CTR1(KTR_IGMPV3, "%s: skip node", __func__); 2926 continue; 2927 } 2928 if (is_source_query && ims->ims_stp == 0) { 2929 CTR1(KTR_IGMPV3, "%s: skip unrecorded node", 2930 __func__); 2931 continue; 2932 } 2933 CTR1(KTR_IGMPV3, "%s: append node", __func__); 2934 naddr = htonl(ims->ims_haddr); 2935 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { 2936 if (m != m0) 2937 m_freem(m); 2938 CTR1(KTR_IGMPV3, "%s: m_append() failed.", 2939 __func__); 2940 return (-ENOMEM); 2941 } 2942 ++msrcs; 2943 if (msrcs == m0srcs) 2944 break; 2945 } 2946 pig->ig_numsrc = htons(msrcs); 2947 nbytes += (msrcs * sizeof(in_addr_t)); 2948 2949 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__); 2950 mbufq_enqueue(mq, m); 2951 } 2952 2953 return (nbytes); 2954} 2955 2956/* 2957 * Type used to mark record pass completion. 2958 * We exploit the fact we can cast to this easily from the 2959 * current filter modes on each ip_msource node. 2960 */ 2961typedef enum { 2962 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2963 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2964 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2965 REC_FULL = REC_ALLOW | REC_BLOCK 2966} rectype_t; 2967 2968/* 2969 * Enqueue an IGMPv3 filter list change to the given output queue. 2970 * 2971 * Source list filter state is held in an RB-tree. When the filter list 2972 * for a group is changed without changing its mode, we need to compute 2973 * the deltas between T0 and T1 for each source in the filter set, 2974 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2975 * 2976 * As we may potentially queue two record types, and the entire R-B tree 2977 * needs to be walked at once, we break this out into its own function 2978 * so we can generate a tightly packed queue of packets. 2979 * 2980 * XXX This could be written to only use one tree walk, although that makes 2981 * serializing into the mbuf chains a bit harder. For now we do two walks 2982 * which makes things easier on us, and it may or may not be harder on 2983 * the L2 cache. 2984 * 2985 * If successful the size of all data appended to the queue is returned, 2986 * otherwise an error code less than zero is returned, or zero if 2987 * no record(s) were appended. 2988 */ 2989static int 2990igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm) 2991{ 2992 static const int MINRECLEN = 2993 sizeof(struct igmp_grouprec) + sizeof(in_addr_t); 2994 struct ifnet *ifp; 2995 struct igmp_grouprec ig; 2996 struct igmp_grouprec *pig; 2997 struct ip_msource *ims, *nims; 2998 struct mbuf *m, *m0, *md; 2999 in_addr_t naddr; 3000 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 3001 int nallow, nblock; 3002 uint8_t mode, now, then; 3003 rectype_t crt, drt, nrt; 3004 3005 IN_MULTI_LIST_LOCK_ASSERT(); 3006 3007 if (inm->inm_nsrc == 0 || 3008 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0)) 3009 return (0); 3010 3011 ifp = inm->inm_ifp; /* interface */ 3012 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */ 3013 crt = REC_NONE; /* current group record type */ 3014 drt = REC_NONE; /* mask of completed group record types */ 3015 nrt = REC_NONE; /* record type for current node */ 3016 m0srcs = 0; /* # source which will fit in current mbuf chain */ 3017 nbytes = 0; /* # of bytes appended to group's state-change queue */ 3018 npbytes = 0; /* # of bytes appended this packet */ 3019 rsrcs = 0; /* # sources encoded in current record */ 3020 schanged = 0; /* # nodes encoded in overall filter change */ 3021 nallow = 0; /* # of source entries in ALLOW_NEW */ 3022 nblock = 0; /* # of source entries in BLOCK_OLD */ 3023 nims = NULL; /* next tree node pointer */ 3024 3025 /* 3026 * For each possible filter record mode. 3027 * The first kind of source we encounter tells us which 3028 * is the first kind of record we start appending. 3029 * If a node transitioned to UNDEFINED at t1, its mode is treated 3030 * as the inverse of the group's filter mode. 3031 */ 3032 while (drt != REC_FULL) { 3033 do { 3034 m0 = mbufq_last(mq); 3035 if (m0 != NULL && 3036 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 3037 IGMP_V3_REPORT_MAXRECS) && 3038 (m0->m_pkthdr.len + MINRECLEN) < 3039 (ifp->if_mtu - IGMP_LEADINGSPACE)) { 3040 m = m0; 3041 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 3042 sizeof(struct igmp_grouprec)) / 3043 sizeof(in_addr_t); 3044 CTR1(KTR_IGMPV3, 3045 "%s: use previous packet", __func__); 3046 } else { 3047 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 3048 if (m) 3049 m->m_data += IGMP_LEADINGSPACE; 3050 if (m == NULL) { 3051 m = m_gethdr(M_NOWAIT, MT_DATA); 3052 if (m) 3053 M_ALIGN(m, IGMP_LEADINGSPACE); 3054 } 3055 if (m == NULL) { 3056 CTR1(KTR_IGMPV3, 3057 "%s: m_get*() failed", __func__); 3058 return (-ENOMEM); 3059 } 3060 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3061 igmp_save_context(m, ifp); 3062 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 3063 sizeof(struct igmp_grouprec)) / 3064 sizeof(in_addr_t); 3065 npbytes = 0; 3066 CTR1(KTR_IGMPV3, 3067 "%s: allocated new packet", __func__); 3068 } 3069 /* 3070 * Append the IGMP group record header to the 3071 * current packet's data area. 3072 * Recalculate pointer to free space for next 3073 * group record, in case m_append() allocated 3074 * a new mbuf or cluster. 3075 */ 3076 memset(&ig, 0, sizeof(ig)); 3077 ig.ig_group = inm->inm_addr; 3078 if (!m_append(m, sizeof(ig), (void *)&ig)) { 3079 if (m != m0) 3080 m_freem(m); 3081 CTR1(KTR_IGMPV3, 3082 "%s: m_append() failed", __func__); 3083 return (-ENOMEM); 3084 } 3085 npbytes += sizeof(struct igmp_grouprec); 3086 if (m != m0) { 3087 /* new packet; offset in c hain */ 3088 md = m_getptr(m, npbytes - 3089 sizeof(struct igmp_grouprec), &off); 3090 pig = (struct igmp_grouprec *)(mtod(md, 3091 uint8_t *) + off); 3092 } else { 3093 /* current packet; offset from last append */ 3094 md = m_last(m); 3095 pig = (struct igmp_grouprec *)(mtod(md, 3096 uint8_t *) + md->m_len - 3097 sizeof(struct igmp_grouprec)); 3098 } 3099 /* 3100 * Begin walking the tree for this record type 3101 * pass, or continue from where we left off 3102 * previously if we had to allocate a new packet. 3103 * Only report deltas in-mode at t1. 3104 * We need not report included sources as allowed 3105 * if we are in inclusive mode on the group, 3106 * however the converse is not true. 3107 */ 3108 rsrcs = 0; 3109 if (nims == NULL) 3110 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs); 3111 RB_FOREACH_FROM(ims, ip_msource_tree, nims) { 3112 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", 3113 __func__, ims->ims_haddr); 3114 now = ims_get_mode(inm, ims, 1); 3115 then = ims_get_mode(inm, ims, 0); 3116 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d", 3117 __func__, then, now); 3118 if (now == then) { 3119 CTR1(KTR_IGMPV3, 3120 "%s: skip unchanged", __func__); 3121 continue; 3122 } 3123 if (mode == MCAST_EXCLUDE && 3124 now == MCAST_INCLUDE) { 3125 CTR1(KTR_IGMPV3, 3126 "%s: skip IN src on EX group", 3127 __func__); 3128 continue; 3129 } 3130 nrt = (rectype_t)now; 3131 if (nrt == REC_NONE) 3132 nrt = (rectype_t)(~mode & REC_FULL); 3133 if (schanged++ == 0) { 3134 crt = nrt; 3135 } else if (crt != nrt) 3136 continue; 3137 naddr = htonl(ims->ims_haddr); 3138 if (!m_append(m, sizeof(in_addr_t), 3139 (void *)&naddr)) { 3140 if (m != m0) 3141 m_freem(m); 3142 CTR1(KTR_IGMPV3, 3143 "%s: m_append() failed", __func__); 3144 return (-ENOMEM); 3145 } 3146 nallow += !!(crt == REC_ALLOW); 3147 nblock += !!(crt == REC_BLOCK); 3148 if (++rsrcs == m0srcs) 3149 break; 3150 } 3151 /* 3152 * If we did not append any tree nodes on this 3153 * pass, back out of allocations. 3154 */ 3155 if (rsrcs == 0) { 3156 npbytes -= sizeof(struct igmp_grouprec); 3157 if (m != m0) { 3158 CTR1(KTR_IGMPV3, 3159 "%s: m_free(m)", __func__); 3160 m_freem(m); 3161 } else { 3162 CTR1(KTR_IGMPV3, 3163 "%s: m_adj(m, -ig)", __func__); 3164 m_adj(m, -((int)sizeof( 3165 struct igmp_grouprec))); 3166 } 3167 continue; 3168 } 3169 npbytes += (rsrcs * sizeof(in_addr_t)); 3170 if (crt == REC_ALLOW) 3171 pig->ig_type = IGMP_ALLOW_NEW_SOURCES; 3172 else if (crt == REC_BLOCK) 3173 pig->ig_type = IGMP_BLOCK_OLD_SOURCES; 3174 pig->ig_numsrc = htons(rsrcs); 3175 /* 3176 * Count the new group record, and enqueue this 3177 * packet if it wasn't already queued. 3178 */ 3179 m->m_pkthdr.PH_vt.vt_nrecs++; 3180 if (m != m0) 3181 mbufq_enqueue(mq, m); 3182 nbytes += npbytes; 3183 } while (nims != NULL); 3184 drt |= crt; 3185 crt = (~crt & REC_FULL); 3186 } 3187 3188 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 3189 nallow, nblock); 3190 3191 return (nbytes); 3192} 3193 3194static int 3195igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq) 3196{ 3197 struct mbufq *gq; 3198 struct mbuf *m; /* pending state-change */ 3199 struct mbuf *m0; /* copy of pending state-change */ 3200 struct mbuf *mt; /* last state-change in packet */ 3201 int docopy, domerge; 3202 u_int recslen; 3203 3204 docopy = 0; 3205 domerge = 0; 3206 recslen = 0; 3207 3208 IN_MULTI_LIST_LOCK_ASSERT(); 3209 IGMP_LOCK_ASSERT(); 3210 3211 /* 3212 * If there are further pending retransmissions, make a writable 3213 * copy of each queued state-change message before merging. 3214 */ 3215 if (inm->inm_scrv > 0) 3216 docopy = 1; 3217 3218 gq = &inm->inm_scq; 3219#ifdef KTR 3220 if (mbufq_first(gq) == NULL) { 3221 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty", 3222 __func__, inm); 3223 } 3224#endif 3225 3226 m = mbufq_first(gq); 3227 while (m != NULL) { 3228 /* 3229 * Only merge the report into the current packet if 3230 * there is sufficient space to do so; an IGMPv3 report 3231 * packet may only contain 65,535 group records. 3232 * Always use a simple mbuf chain concatentation to do this, 3233 * as large state changes for single groups may have 3234 * allocated clusters. 3235 */ 3236 domerge = 0; 3237 mt = mbufq_last(scq); 3238 if (mt != NULL) { 3239 recslen = m_length(m, NULL); 3240 3241 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 3242 m->m_pkthdr.PH_vt.vt_nrecs <= 3243 IGMP_V3_REPORT_MAXRECS) && 3244 (mt->m_pkthdr.len + recslen <= 3245 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE))) 3246 domerge = 1; 3247 } 3248 3249 if (!domerge && mbufq_full(gq)) { 3250 CTR2(KTR_IGMPV3, 3251 "%s: outbound queue full, skipping whole packet %p", 3252 __func__, m); 3253 mt = m->m_nextpkt; 3254 if (!docopy) 3255 m_freem(m); 3256 m = mt; 3257 continue; 3258 } 3259 3260 if (!docopy) { 3261 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m); 3262 m0 = mbufq_dequeue(gq); 3263 m = m0->m_nextpkt; 3264 } else { 3265 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m); 3266 m0 = m_dup(m, M_NOWAIT); 3267 if (m0 == NULL) 3268 return (ENOMEM); 3269 m0->m_nextpkt = NULL; 3270 m = m->m_nextpkt; 3271 } 3272 3273 if (!domerge) { 3274 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)", 3275 __func__, m0, scq); 3276 mbufq_enqueue(scq, m0); 3277 } else { 3278 struct mbuf *mtl; /* last mbuf of packet mt */ 3279 3280 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)", 3281 __func__, m0, mt); 3282 3283 mtl = m_last(mt); 3284 m0->m_flags &= ~M_PKTHDR; 3285 mt->m_pkthdr.len += recslen; 3286 mt->m_pkthdr.PH_vt.vt_nrecs += 3287 m0->m_pkthdr.PH_vt.vt_nrecs; 3288 3289 mtl->m_next = m0; 3290 } 3291 } 3292 3293 return (0); 3294} 3295 3296/* 3297 * Respond to a pending IGMPv3 General Query. 3298 */ 3299static void 3300igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi) 3301{ 3302 struct ifmultiaddr *ifma; 3303 struct ifnet *ifp; 3304 struct in_multi *inm; 3305 int retval __unused, loop; 3306 3307 IN_MULTI_LIST_LOCK_ASSERT(); 3308 IGMP_LOCK_ASSERT(); 3309 3310 KASSERT(igi->igi_version == IGMP_VERSION_3, 3311 ("%s: called when version %d", __func__, igi->igi_version)); 3312 3313 /* 3314 * Check that there are some packets queued. If so, send them first. 3315 * For large number of groups the reply to general query can take 3316 * many packets, we should finish sending them before starting of 3317 * queuing the new reply. 3318 */ 3319 if (mbufq_len(&igi->igi_gq) != 0) 3320 goto send; 3321 3322 ifp = igi->igi_ifp; 3323 3324 IF_ADDR_RLOCK(ifp); 3325 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3326 if (ifma->ifma_addr->sa_family != AF_INET || 3327 ifma->ifma_protospec == NULL) 3328 continue; 3329 3330 inm = (struct in_multi *)ifma->ifma_protospec; 3331 KASSERT(ifp == inm->inm_ifp, 3332 ("%s: inconsistent ifp", __func__)); 3333 3334 switch (inm->inm_state) { 3335 case IGMP_NOT_MEMBER: 3336 case IGMP_SILENT_MEMBER: 3337 break; 3338 case IGMP_REPORTING_MEMBER: 3339 case IGMP_IDLE_MEMBER: 3340 case IGMP_LAZY_MEMBER: 3341 case IGMP_SLEEPING_MEMBER: 3342 case IGMP_AWAKENING_MEMBER: 3343 inm->inm_state = IGMP_REPORTING_MEMBER; 3344 retval = igmp_v3_enqueue_group_record(&igi->igi_gq, 3345 inm, 0, 0, 0); 3346 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 3347 __func__, retval); 3348 break; 3349 case IGMP_G_QUERY_PENDING_MEMBER: 3350 case IGMP_SG_QUERY_PENDING_MEMBER: 3351 case IGMP_LEAVING_MEMBER: 3352 break; 3353 } 3354 } 3355 IF_ADDR_RUNLOCK(ifp); 3356 3357send: 3358 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; 3359 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop); 3360 3361 /* 3362 * Slew transmission of bursts over 500ms intervals. 3363 */ 3364 if (mbufq_first(&igi->igi_gq) != NULL) { 3365 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY( 3366 IGMP_RESPONSE_BURST_INTERVAL); 3367 V_interface_timers_running = 1; 3368 } 3369} 3370 3371/* 3372 * Transmit the next pending IGMP message in the output queue. 3373 * 3374 * We get called from netisr_processqueue(). A mutex private to igmpoq 3375 * will be acquired and released around this routine. 3376 * 3377 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3378 * MRT: Nothing needs to be done, as IGMP traffic is always local to 3379 * a link and uses a link-scope multicast address. 3380 */ 3381static void 3382igmp_intr(struct mbuf *m) 3383{ 3384 struct ip_moptions imo; 3385 struct ifnet *ifp; 3386 struct mbuf *ipopts, *m0; 3387 int error; 3388 uint32_t ifindex; 3389 3390 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m); 3391 3392 /* 3393 * Set VNET image pointer from enqueued mbuf chain 3394 * before doing anything else. Whilst we use interface 3395 * indexes to guard against interface detach, they are 3396 * unique to each VIMAGE and must be retrieved. 3397 */ 3398 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr)); 3399 ifindex = igmp_restore_context(m); 3400 3401 /* 3402 * Check if the ifnet still exists. This limits the scope of 3403 * any race in the absence of a global ifp lock for low cost 3404 * (an array lookup). 3405 */ 3406 ifp = ifnet_byindex(ifindex); 3407 if (ifp == NULL) { 3408 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.", 3409 __func__, m, ifindex); 3410 m_freem(m); 3411 IPSTAT_INC(ips_noroute); 3412 goto out; 3413 } 3414 3415 ipopts = V_igmp_sendra ? m_raopt : NULL; 3416 3417 imo.imo_multicast_ttl = 1; 3418 imo.imo_multicast_vif = -1; 3419 imo.imo_multicast_loop = (V_ip_mrouter != NULL); 3420 3421 /* 3422 * If the user requested that IGMP traffic be explicitly 3423 * redirected to the loopback interface (e.g. they are running a 3424 * MANET interface and the routing protocol needs to see the 3425 * updates), handle this now. 3426 */ 3427 if (m->m_flags & M_IGMP_LOOP) 3428 imo.imo_multicast_ifp = V_loif; 3429 else 3430 imo.imo_multicast_ifp = ifp; 3431 3432 if (m->m_flags & M_IGMPV2) { 3433 m0 = m; 3434 } else { 3435 m0 = igmp_v3_encap_report(ifp, m); 3436 if (m0 == NULL) { 3437 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m); 3438 m_freem(m); 3439 IPSTAT_INC(ips_odropped); 3440 goto out; 3441 } 3442 } 3443 3444 igmp_scrub_context(m0); 3445 m_clrprotoflags(m); 3446 m0->m_pkthdr.rcvif = V_loif; 3447#ifdef MAC 3448 mac_netinet_igmp_send(ifp, m0); 3449#endif 3450 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL); 3451 if (error) { 3452 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error); 3453 goto out; 3454 } 3455 3456 IGMPSTAT_INC(igps_snd_reports); 3457 3458out: 3459 /* 3460 * We must restore the existing vnet pointer before 3461 * continuing as we are run from netisr context. 3462 */ 3463 CURVNET_RESTORE(); 3464} 3465 3466/* 3467 * Encapsulate an IGMPv3 report. 3468 * 3469 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf 3470 * chain has already had its IP/IGMPv3 header prepended. In this case 3471 * the function will not attempt to prepend; the lengths and checksums 3472 * will however be re-computed. 3473 * 3474 * Returns a pointer to the new mbuf chain head, or NULL if the 3475 * allocation failed. 3476 */ 3477static struct mbuf * 3478igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m) 3479{ 3480 struct rm_priotracker in_ifa_tracker; 3481 struct igmp_report *igmp; 3482 struct ip *ip; 3483 int hdrlen, igmpreclen; 3484 3485 KASSERT((m->m_flags & M_PKTHDR), 3486 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3487 3488 igmpreclen = m_length(m, NULL); 3489 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report); 3490 3491 if (m->m_flags & M_IGMPV3_HDR) { 3492 igmpreclen -= hdrlen; 3493 } else { 3494 M_PREPEND(m, hdrlen, M_NOWAIT); 3495 if (m == NULL) 3496 return (NULL); 3497 m->m_flags |= M_IGMPV3_HDR; 3498 } 3499 3500 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen); 3501 3502 m->m_data += sizeof(struct ip); 3503 m->m_len -= sizeof(struct ip); 3504 3505 igmp = mtod(m, struct igmp_report *); 3506 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT; 3507 igmp->ir_rsv1 = 0; 3508 igmp->ir_rsv2 = 0; 3509 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs); 3510 igmp->ir_cksum = 0; 3511 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen); 3512 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3513 3514 m->m_data -= sizeof(struct ip); 3515 m->m_len += sizeof(struct ip); 3516 3517 ip = mtod(m, struct ip *); 3518 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL; 3519 ip->ip_len = htons(hdrlen + igmpreclen); 3520 ip->ip_off = htons(IP_DF); 3521 ip->ip_p = IPPROTO_IGMP; 3522 ip->ip_sum = 0; 3523 3524 ip->ip_src.s_addr = INADDR_ANY; 3525 3526 if (m->m_flags & M_IGMP_LOOP) { 3527 struct in_ifaddr *ia; 3528 3529 NET_EPOCH_ENTER(); 3530 IFP_TO_IA(ifp, ia, &in_ifa_tracker); 3531 if (ia != NULL) 3532 ip->ip_src = ia->ia_addr.sin_addr; 3533 NET_EPOCH_EXIT(); 3534 } 3535 3536 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP); 3537 3538 return (m); 3539} 3540 3541#ifdef KTR 3542static char * 3543igmp_rec_type_to_str(const int type) 3544{ 3545 3546 switch (type) { 3547 case IGMP_CHANGE_TO_EXCLUDE_MODE: 3548 return "TO_EX"; 3549 break; 3550 case IGMP_CHANGE_TO_INCLUDE_MODE: 3551 return "TO_IN"; 3552 break; 3553 case IGMP_MODE_IS_EXCLUDE: 3554 return "MODE_EX"; 3555 break; 3556 case IGMP_MODE_IS_INCLUDE: 3557 return "MODE_IN"; 3558 break; 3559 case IGMP_ALLOW_NEW_SOURCES: 3560 return "ALLOW_NEW"; 3561 break; 3562 case IGMP_BLOCK_OLD_SOURCES: 3563 return "BLOCK_OLD"; 3564 break; 3565 default: 3566 break; 3567 } 3568 return "unknown"; 3569} 3570#endif 3571 3572#ifdef VIMAGE 3573static void 3574vnet_igmp_init(const void *unused __unused) 3575{ 3576 3577 netisr_register_vnet(&igmp_nh); 3578} 3579VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, 3580 vnet_igmp_init, NULL); 3581 3582static void 3583vnet_igmp_uninit(const void *unused __unused) 3584{ 3585 3586 /* This can happen when we shutdown the entire network stack. */ 3587 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3588 3589 netisr_unregister_vnet(&igmp_nh); 3590} 3591VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, 3592 vnet_igmp_uninit, NULL); 3593#endif 3594 3595#ifdef DDB 3596DB_SHOW_COMMAND(igi_list, db_show_igi_list) 3597{ 3598 struct igmp_ifsoftc *igi, *tigi; 3599 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head; 3600 3601 if (!have_addr) { 3602 db_printf("usage: show igi_list <addr>\n"); 3603 return; 3604 } 3605 igi_head = (struct _igi_list *)addr; 3606 3607 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) { 3608 db_printf("igmp_ifsoftc %p:\n", igi); 3609 db_printf(" ifp %p\n", igi->igi_ifp); 3610 db_printf(" version %u\n", igi->igi_version); 3611 db_printf(" v1_timer %u\n", igi->igi_v1_timer); 3612 db_printf(" v2_timer %u\n", igi->igi_v2_timer); 3613 db_printf(" v3_timer %u\n", igi->igi_v3_timer); 3614 db_printf(" flags %#x\n", igi->igi_flags); 3615 db_printf(" rv %u\n", igi->igi_rv); 3616 db_printf(" qi %u\n", igi->igi_qi); 3617 db_printf(" qri %u\n", igi->igi_qri); 3618 db_printf(" uri %u\n", igi->igi_uri); 3619 /* struct mbufq igi_gq; */ 3620 db_printf("\n"); 3621 } 3622} 3623#endif 3624 3625static int 3626igmp_modevent(module_t mod, int type, void *unused __unused) 3627{ 3628 3629 switch (type) { 3630 case MOD_LOAD: 3631 CTR1(KTR_IGMPV3, "%s: initializing", __func__); 3632 IGMP_LOCK_INIT(); 3633 m_raopt = igmp_ra_alloc(); 3634 netisr_register(&igmp_nh); 3635 break; 3636 case MOD_UNLOAD: 3637 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3638 netisr_unregister(&igmp_nh); 3639 m_free(m_raopt); 3640 m_raopt = NULL; 3641 IGMP_LOCK_DESTROY(); 3642 break; 3643 default: 3644 return (EOPNOTSUPP); 3645 } 3646 return (0); 3647} 3648 3649static moduledata_t igmp_mod = { 3650 "igmp", 3651 igmp_modevent, 3652 0 3653}; 3654DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE); 3655