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 #include "opt_ddb.h" 54 55 #include <sys/param.h> 56 #include <sys/systm.h> 57 #include <sys/module.h> 58 #include <sys/malloc.h> 59 #include <sys/mbuf.h> 60 #include <sys/socket.h> 61 #include <sys/kernel.h> 62 #include <sys/lock.h> 63 #include <sys/sysctl.h> 64 #include <sys/ktr.h> 65 #include <sys/condvar.h> 66 67 #ifdef DDB 68 #include <ddb/ddb.h> 69 #endif 70 71 #include <net/if.h> 72 #include <net/if_var.h> 73 #include <net/if_private.h> 74 #include <net/netisr.h> 75 #include <net/vnet.h> 76 77 #include <netinet/in.h> 78 #include <netinet/in_var.h> 79 #include <netinet/in_systm.h> 80 #include <netinet/ip.h> 81 #include <netinet/ip_var.h> 82 #include <netinet/ip_options.h> 83 #include <netinet/igmp.h> 84 #include <netinet/igmp_var.h> 85 86 #include <machine/in_cksum.h> 87 88 #include <security/mac/mac_framework.h> 89 90 #ifndef KTR_IGMPV3 91 #define KTR_IGMPV3 KTR_INET 92 #endif 93 94 #define IGMP_SLOWHZ 2 /* 2 slow timeouts per second */ 95 #define IGMP_FASTHZ 5 /* 5 fast timeouts per second */ 96 #define IGMP_RESPONSE_BURST_INTERVAL (IGMP_FASTHZ / 2) 97 98 static struct igmp_ifsoftc * 99 igi_alloc_locked(struct ifnet *); 100 static void igi_delete_locked(const struct ifnet *); 101 static void igmp_dispatch_queue(struct mbufq *, int, const int); 102 static void igmp_fasttimo_vnet(void); 103 static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *); 104 static int igmp_handle_state_change(struct in_multi *, 105 struct igmp_ifsoftc *); 106 static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *); 107 static int igmp_input_v1_query(struct ifnet *, const struct ip *, 108 const struct igmp *); 109 static int igmp_input_v2_query(struct ifnet *, const struct ip *, 110 const struct igmp *); 111 static int igmp_input_v3_query(struct ifnet *, const struct ip *, 112 /*const*/ struct igmpv3 *); 113 static int igmp_input_v3_group_query(struct in_multi *, 114 struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *); 115 static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *, 116 /*const*/ struct igmp *); 117 static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *, 118 /*const*/ struct igmp *); 119 static void igmp_intr(struct mbuf *); 120 static int igmp_isgroupreported(const struct in_addr); 121 static struct mbuf * 122 igmp_ra_alloc(void); 123 #ifdef KTR 124 static char * igmp_rec_type_to_str(const int); 125 #endif 126 static void igmp_set_version(struct igmp_ifsoftc *, const int); 127 static void igmp_slowtimo_vnet(void); 128 static int igmp_v1v2_queue_report(struct in_multi *, const int); 129 static void igmp_v1v2_process_group_timer(struct in_multi *, const int); 130 static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *); 131 static void igmp_v2_update_group(struct in_multi *, const int); 132 static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *); 133 static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *); 134 static struct mbuf * 135 igmp_v3_encap_report(struct ifnet *, struct mbuf *); 136 static int igmp_v3_enqueue_group_record(struct mbufq *, 137 struct in_multi *, const int, const int, const int); 138 static int igmp_v3_enqueue_filter_change(struct mbufq *, 139 struct in_multi *); 140 static void igmp_v3_process_group_timers(struct in_multi_head *, 141 struct mbufq *, struct mbufq *, struct in_multi *, 142 const int); 143 static int igmp_v3_merge_state_changes(struct in_multi *, 144 struct mbufq *); 145 static void igmp_v3_suppress_group_record(struct in_multi *); 146 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS); 147 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS); 148 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS); 149 static int sysctl_igmp_stat(SYSCTL_HANDLER_ARGS); 150 151 static const struct netisr_handler igmp_nh = { 152 .nh_name = "igmp", 153 .nh_handler = igmp_intr, 154 .nh_proto = NETISR_IGMP, 155 .nh_policy = NETISR_POLICY_SOURCE, 156 }; 157 158 /* 159 * System-wide globals. 160 * 161 * Unlocked access to these is OK, except for the global IGMP output 162 * queue. The IGMP subsystem lock ends up being system-wide for the moment, 163 * because all VIMAGEs have to share a global output queue, as netisrs 164 * themselves are not virtualized. 165 * 166 * Locking: 167 * * The permitted lock order is: IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 168 * Any may be taken independently; if any are held at the same 169 * time, the above lock order must be followed. 170 * * All output is delegated to the netisr. 171 * * IN_MULTI_LIST_LOCK covers in_multi. 172 * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file, 173 * including the output queue. 174 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 175 * per-link state iterators. 176 * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface, 177 * therefore it is not refcounted. 178 * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi. 179 * 180 * Reference counting 181 * * IGMP acquires its own reference every time an in_multi is passed to 182 * it and the group is being joined for the first time. 183 * * IGMP releases its reference(s) on in_multi in a deferred way, 184 * because the operations which process the release run as part of 185 * a loop whose control variables are directly affected by the release 186 * (that, and not recursing on the IF_ADDR_LOCK). 187 * 188 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds 189 * to a vnet in ifp->if_vnet. 190 * 191 * SMPng: XXX We may potentially race operations on ifma_protospec. 192 * The problem is that we currently lack a clean way of taking the 193 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing, 194 * as anything which modifies ifma needs to be covered by that lock. 195 * So check for ifma_protospec being NULL before proceeding. 196 */ 197 struct mtx igmp_mtx; 198 199 struct mbuf *m_raopt; /* Router Alert option */ 200 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state"); 201 202 /* 203 * VIMAGE-wide globals. 204 * 205 * The IGMPv3 timers themselves need to run per-image, however, for 206 * historical reasons, timers run globally. This needs to be improved. 207 * An ifnet can only be in one vimage at a time, and the loopback 208 * ifnet, loif, is itself virtualized. 209 * It would otherwise be possible to seriously hose IGMP state, 210 * and create inconsistencies in upstream multicast routing, if you have 211 * multiple VIMAGEs running on the same link joining different multicast 212 * groups, UNLESS the "primary IP address" is different. This is because 213 * IGMP for IPv4 does not force link-local addresses to be used for each 214 * node, unlike MLD for IPv6. 215 * Obviously the IGMPv3 per-interface state has per-vimage granularity 216 * also as a result. 217 * 218 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection 219 * policy to control the address used by IGMP on the link. 220 */ 221 VNET_DEFINE_STATIC(int, interface_timers_running); /* IGMPv3 general 222 * query response */ 223 VNET_DEFINE_STATIC(int, state_change_timers_running); /* IGMPv3 state-change 224 * retransmit */ 225 VNET_DEFINE_STATIC(int, current_state_timers_running); /* IGMPv1/v2 host 226 * report; IGMPv3 g/sg 227 * query response */ 228 229 #define V_interface_timers_running VNET(interface_timers_running) 230 #define V_state_change_timers_running VNET(state_change_timers_running) 231 #define V_current_state_timers_running VNET(current_state_timers_running) 232 233 VNET_PCPUSTAT_DEFINE(struct igmpstat, igmpstat); 234 VNET_PCPUSTAT_SYSINIT(igmpstat); 235 VNET_PCPUSTAT_SYSUNINIT(igmpstat); 236 237 VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) = 238 LIST_HEAD_INITIALIZER(igi_head); 239 VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0}; 240 241 #define V_igi_head VNET(igi_head) 242 #define V_igmp_gsrdelay VNET(igmp_gsrdelay) 243 244 VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1; 245 VNET_DEFINE_STATIC(int, igmp_sendra) = 1; 246 VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1; 247 VNET_DEFINE_STATIC(int, igmp_v1enable) = 1; 248 VNET_DEFINE_STATIC(int, igmp_v2enable) = 1; 249 VNET_DEFINE_STATIC(int, igmp_legacysupp); 250 VNET_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 */ 263 SYSCTL_PROC(_net_inet_igmp, IGMPCTL_STATS, stats, 264 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_MPSAFE, 265 &VNET_NAME(igmpstat), 0, sysctl_igmp_stat, "S,igmpstat", 266 "IGMP statistics (struct igmpstat, netinet/igmp_var.h)"); 267 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW, 268 &VNET_NAME(igmp_recvifkludge), 0, 269 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address"); 270 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW, 271 &VNET_NAME(igmp_sendra), 0, 272 "Send IP Router Alert option in IGMPv2/v3 messages"); 273 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW, 274 &VNET_NAME(igmp_sendlocal), 0, 275 "Send IGMP membership reports for 224.0.0.0/24 groups"); 276 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW, 277 &VNET_NAME(igmp_v1enable), 0, 278 "Enable backwards compatibility with IGMPv1"); 279 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW, 280 &VNET_NAME(igmp_v2enable), 0, 281 "Enable backwards compatibility with IGMPv2"); 282 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW, 283 &VNET_NAME(igmp_legacysupp), 0, 284 "Allow v1/v2 reports to suppress v3 group responses"); 285 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version, 286 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 287 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I", 288 "Default version of IGMP to run on each interface"); 289 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay, 290 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 291 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I", 292 "Rate limit for IGMPv3 Group-and-Source queries in seconds"); 293 294 /* 295 * Non-virtualized sysctls. 296 */ 297 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, 298 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo, 299 "Per-interface IGMPv3 state"); 300 301 static __inline void 302 igmp_save_context(struct mbuf *m, struct ifnet *ifp) 303 { 304 305 #ifdef VIMAGE 306 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; 307 #endif /* VIMAGE */ 308 m->m_pkthdr.rcvif = ifp; 309 m->m_pkthdr.flowid = ifp->if_index; 310 } 311 312 static __inline void 313 igmp_scrub_context(struct mbuf *m) 314 { 315 316 m->m_pkthdr.PH_loc.ptr = NULL; 317 m->m_pkthdr.flowid = 0; 318 } 319 320 /* 321 * Restore context from a queued IGMP output chain. 322 * Return saved ifindex. 323 * 324 * VIMAGE: The assertion is there to make sure that we 325 * actually called CURVNET_SET() with what's in the mbuf chain. 326 */ 327 static __inline uint32_t 328 igmp_restore_context(struct mbuf *m) 329 { 330 331 #ifdef notyet 332 #if defined(VIMAGE) && defined(INVARIANTS) 333 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr), 334 ("%s: called when curvnet was not restored", __func__)); 335 #endif 336 #endif 337 return (m->m_pkthdr.flowid); 338 } 339 340 /* 341 * IGMP statistics. 342 */ 343 static int 344 sysctl_igmp_stat(SYSCTL_HANDLER_ARGS) 345 { 346 struct igmpstat igps0; 347 int error; 348 char *p; 349 350 error = sysctl_wire_old_buffer(req, sizeof(struct igmpstat)); 351 if (error) 352 return (error); 353 354 if (req->oldptr != NULL) { 355 if (req->oldlen < sizeof(struct igmpstat)) 356 error = ENOMEM; 357 else { 358 /* 359 * Copy the counters, and explicitly set the struct's 360 * version and length fields. 361 */ 362 COUNTER_ARRAY_COPY(VNET(igmpstat), &igps0, 363 sizeof(struct igmpstat) / sizeof(uint64_t)); 364 igps0.igps_version = IGPS_VERSION_3; 365 igps0.igps_len = IGPS_VERSION3_LEN; 366 error = SYSCTL_OUT(req, &igps0, 367 sizeof(struct igmpstat)); 368 } 369 } else 370 req->validlen = sizeof(struct igmpstat); 371 if (error) 372 goto out; 373 if (req->newptr != NULL) { 374 if (req->newlen < sizeof(struct igmpstat)) 375 error = ENOMEM; 376 else 377 error = SYSCTL_IN(req, &igps0, 378 sizeof(igps0)); 379 if (error) 380 goto out; 381 /* 382 * igps0 must be "all zero". 383 */ 384 p = (char *)&igps0; 385 while (p < (char *)&igps0 + sizeof(igps0) && *p == '\0') 386 p++; 387 if (p != (char *)&igps0 + sizeof(igps0)) { 388 error = EINVAL; 389 goto out; 390 } 391 COUNTER_ARRAY_ZERO(VNET(igmpstat), 392 sizeof(struct igmpstat) / sizeof(uint64_t)); 393 } 394 out: 395 return (error); 396 } 397 398 /* 399 * Retrieve or set default IGMP version. 400 * 401 * VIMAGE: Assume curvnet set by caller. 402 * SMPng: NOTE: Serialized by IGMP lock. 403 */ 404 static int 405 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS) 406 { 407 int error; 408 int new; 409 410 error = sysctl_wire_old_buffer(req, sizeof(int)); 411 if (error) 412 return (error); 413 414 IGMP_LOCK(); 415 416 new = V_igmp_default_version; 417 418 error = sysctl_handle_int(oidp, &new, 0, req); 419 if (error || !req->newptr) 420 goto out_locked; 421 422 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) { 423 error = EINVAL; 424 goto out_locked; 425 } 426 427 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d", 428 V_igmp_default_version, new); 429 430 V_igmp_default_version = new; 431 432 out_locked: 433 IGMP_UNLOCK(); 434 return (error); 435 } 436 437 /* 438 * Retrieve or set threshold between group-source queries in seconds. 439 * 440 * VIMAGE: Assume curvnet set by caller. 441 * SMPng: NOTE: Serialized by IGMP lock. 442 */ 443 static int 444 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS) 445 { 446 int error; 447 int i; 448 449 error = sysctl_wire_old_buffer(req, sizeof(int)); 450 if (error) 451 return (error); 452 453 IGMP_LOCK(); 454 455 i = V_igmp_gsrdelay.tv_sec; 456 457 error = sysctl_handle_int(oidp, &i, 0, req); 458 if (error || !req->newptr) 459 goto out_locked; 460 461 if (i < -1 || i >= 60) { 462 error = EINVAL; 463 goto out_locked; 464 } 465 466 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d", 467 V_igmp_gsrdelay.tv_sec, i); 468 V_igmp_gsrdelay.tv_sec = i; 469 470 out_locked: 471 IGMP_UNLOCK(); 472 return (error); 473 } 474 475 /* 476 * Expose struct igmp_ifsoftc to userland, keyed by ifindex. 477 * For use by ifmcstat(8). 478 * 479 * SMPng: NOTE: Does an unlocked ifindex space read. 480 * VIMAGE: Assume curvnet set by caller. The node handler itself 481 * is not directly virtualized. 482 */ 483 static int 484 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS) 485 { 486 struct epoch_tracker et; 487 int *name; 488 int error; 489 u_int namelen; 490 struct ifnet *ifp; 491 struct igmp_ifsoftc *igi; 492 493 name = (int *)arg1; 494 namelen = arg2; 495 496 if (req->newptr != NULL) 497 return (EPERM); 498 499 if (namelen != 1) 500 return (EINVAL); 501 502 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo)); 503 if (error) 504 return (error); 505 506 IN_MULTI_LIST_LOCK(); 507 IGMP_LOCK(); 508 509 error = ENOENT; 510 511 NET_EPOCH_ENTER(et); 512 ifp = ifnet_byindex(name[0]); 513 NET_EPOCH_EXIT(et); 514 if (ifp == NULL) 515 goto out_locked; 516 517 LIST_FOREACH(igi, &V_igi_head, igi_link) { 518 if (ifp == igi->igi_ifp) { 519 struct igmp_ifinfo info; 520 521 info.igi_version = igi->igi_version; 522 info.igi_v1_timer = igi->igi_v1_timer; 523 info.igi_v2_timer = igi->igi_v2_timer; 524 info.igi_v3_timer = igi->igi_v3_timer; 525 info.igi_flags = igi->igi_flags; 526 info.igi_rv = igi->igi_rv; 527 info.igi_qi = igi->igi_qi; 528 info.igi_qri = igi->igi_qri; 529 info.igi_uri = igi->igi_uri; 530 error = SYSCTL_OUT(req, &info, sizeof(info)); 531 break; 532 } 533 } 534 535 out_locked: 536 IGMP_UNLOCK(); 537 IN_MULTI_LIST_UNLOCK(); 538 return (error); 539 } 540 541 /* 542 * Dispatch an entire queue of pending packet chains 543 * using the netisr. 544 * VIMAGE: Assumes the vnet pointer has been set. 545 */ 546 static void 547 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop) 548 { 549 struct epoch_tracker et; 550 struct mbuf *m; 551 552 NET_EPOCH_ENTER(et); 553 while ((m = mbufq_dequeue(mq)) != NULL) { 554 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m); 555 if (loop) 556 m->m_flags |= M_IGMP_LOOP; 557 netisr_dispatch(NETISR_IGMP, m); 558 if (--limit == 0) 559 break; 560 } 561 NET_EPOCH_EXIT(et); 562 } 563 564 /* 565 * Filter outgoing IGMP report state by group. 566 * 567 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1). 568 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are 569 * disabled for all groups in the 224.0.0.0/24 link-local scope. However, 570 * this may break certain IGMP snooping switches which rely on the old 571 * report behaviour. 572 * 573 * Return zero if the given group is one for which IGMP reports 574 * should be suppressed, or non-zero if reports should be issued. 575 */ 576 static __inline int 577 igmp_isgroupreported(const struct in_addr addr) 578 { 579 580 if (in_allhosts(addr) || 581 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr))))) 582 return (0); 583 584 return (1); 585 } 586 587 /* 588 * Construct a Router Alert option to use in outgoing packets. 589 */ 590 static struct mbuf * 591 igmp_ra_alloc(void) 592 { 593 struct mbuf *m; 594 struct ipoption *p; 595 596 m = m_get(M_WAITOK, MT_DATA); 597 p = mtod(m, struct ipoption *); 598 p->ipopt_dst.s_addr = INADDR_ANY; 599 p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */ 600 p->ipopt_list[1] = 0x04; /* 4 bytes long */ 601 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */ 602 p->ipopt_list[3] = 0x00; /* pad byte */ 603 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1]; 604 605 return (m); 606 } 607 608 /* 609 * Attach IGMP when PF_INET is attached to an interface. 610 */ 611 struct igmp_ifsoftc * 612 igmp_domifattach(struct ifnet *ifp) 613 { 614 struct igmp_ifsoftc *igi; 615 616 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 617 __func__, ifp, ifp->if_xname); 618 619 IGMP_LOCK(); 620 621 igi = igi_alloc_locked(ifp); 622 if (!(ifp->if_flags & IFF_MULTICAST)) 623 igi->igi_flags |= IGIF_SILENT; 624 625 IGMP_UNLOCK(); 626 627 return (igi); 628 } 629 630 /* 631 * VIMAGE: assume curvnet set by caller. 632 */ 633 static struct igmp_ifsoftc * 634 igi_alloc_locked(/*const*/ struct ifnet *ifp) 635 { 636 struct igmp_ifsoftc *igi; 637 638 IGMP_LOCK_ASSERT(); 639 640 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO); 641 if (igi == NULL) 642 goto out; 643 644 igi->igi_ifp = ifp; 645 igi->igi_version = V_igmp_default_version; 646 igi->igi_flags = 0; 647 igi->igi_rv = IGMP_RV_INIT; 648 igi->igi_qi = IGMP_QI_INIT; 649 igi->igi_qri = IGMP_QRI_INIT; 650 igi->igi_uri = IGMP_URI_INIT; 651 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS); 652 653 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link); 654 655 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)", 656 ifp, ifp->if_xname); 657 658 out: 659 return (igi); 660 } 661 662 /* 663 * Hook for ifdetach. 664 * 665 * NOTE: Some finalization tasks need to run before the protocol domain 666 * is detached, but also before the link layer does its cleanup. 667 * 668 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK(). 669 * XXX This is also bitten by unlocked ifma_protospec access. 670 */ 671 void 672 igmp_ifdetach(struct ifnet *ifp) 673 { 674 struct epoch_tracker et; 675 struct igmp_ifsoftc *igi; 676 struct ifmultiaddr *ifma; 677 struct in_multi *inm; 678 struct in_multi_head inm_free_tmp; 679 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, 680 ifp->if_xname); 681 682 SLIST_INIT(&inm_free_tmp); 683 IGMP_LOCK(); 684 685 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 686 if (igi->igi_version == IGMP_VERSION_3) { 687 IF_ADDR_WLOCK(ifp); 688 NET_EPOCH_ENTER(et); 689 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 690 inm = inm_ifmultiaddr_get_inm(ifma); 691 if (inm == NULL) 692 continue; 693 if (inm->inm_state == IGMP_LEAVING_MEMBER) 694 inm_rele_locked(&inm_free_tmp, inm); 695 inm_clear_recorded(inm); 696 } 697 NET_EPOCH_EXIT(et); 698 IF_ADDR_WUNLOCK(ifp); 699 inm_release_list_deferred(&inm_free_tmp); 700 } 701 IGMP_UNLOCK(); 702 703 } 704 705 /* 706 * Hook for domifdetach. 707 */ 708 void 709 igmp_domifdetach(struct ifnet *ifp) 710 { 711 712 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 713 __func__, ifp, ifp->if_xname); 714 715 IGMP_LOCK(); 716 igi_delete_locked(ifp); 717 IGMP_UNLOCK(); 718 } 719 720 static void 721 igi_delete_locked(const struct ifnet *ifp) 722 { 723 struct igmp_ifsoftc *igi, *tigi; 724 725 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)", 726 __func__, ifp, ifp->if_xname); 727 728 IGMP_LOCK_ASSERT(); 729 730 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) { 731 if (igi->igi_ifp == ifp) { 732 /* 733 * Free deferred General Query responses. 734 */ 735 mbufq_drain(&igi->igi_gq); 736 737 LIST_REMOVE(igi, igi_link); 738 free(igi, M_IGMP); 739 return; 740 } 741 } 742 } 743 744 /* 745 * Process a received IGMPv1 query. 746 * Return non-zero if the message should be dropped. 747 * 748 * VIMAGE: The curvnet pointer is derived from the input ifp. 749 */ 750 static int 751 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip, 752 const struct igmp *igmp) 753 { 754 struct ifmultiaddr *ifma; 755 struct igmp_ifsoftc *igi; 756 struct in_multi *inm; 757 758 NET_EPOCH_ASSERT(); 759 760 /* 761 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to 762 * 224.0.0.1. They are always treated as General Queries. 763 * igmp_group is always ignored. Do not drop it as a userland 764 * daemon may wish to see it. 765 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 766 */ 767 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) { 768 IGMPSTAT_INC(igps_rcv_badqueries); 769 return (0); 770 } 771 IGMPSTAT_INC(igps_rcv_gen_queries); 772 773 IN_MULTI_LIST_LOCK(); 774 IGMP_LOCK(); 775 776 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 777 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 778 779 if (igi->igi_flags & IGIF_LOOPBACK) { 780 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)", 781 ifp, ifp->if_xname); 782 goto out_locked; 783 } 784 785 /* 786 * Switch to IGMPv1 host compatibility mode. 787 */ 788 igmp_set_version(igi, IGMP_VERSION_1); 789 790 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname); 791 792 /* 793 * Start the timers in all of our group records 794 * for the interface on which the query arrived, 795 * except those which are already running. 796 */ 797 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 798 inm = inm_ifmultiaddr_get_inm(ifma); 799 if (inm == NULL) 800 continue; 801 if (inm->inm_timer != 0) 802 continue; 803 switch (inm->inm_state) { 804 case IGMP_NOT_MEMBER: 805 case IGMP_SILENT_MEMBER: 806 break; 807 case IGMP_G_QUERY_PENDING_MEMBER: 808 case IGMP_SG_QUERY_PENDING_MEMBER: 809 case IGMP_REPORTING_MEMBER: 810 case IGMP_IDLE_MEMBER: 811 case IGMP_LAZY_MEMBER: 812 case IGMP_SLEEPING_MEMBER: 813 case IGMP_AWAKENING_MEMBER: 814 inm->inm_state = IGMP_REPORTING_MEMBER; 815 inm->inm_timer = IGMP_RANDOM_DELAY( 816 IGMP_V1V2_MAX_RI * IGMP_FASTHZ); 817 V_current_state_timers_running = 1; 818 break; 819 case IGMP_LEAVING_MEMBER: 820 break; 821 } 822 } 823 824 out_locked: 825 IGMP_UNLOCK(); 826 IN_MULTI_LIST_UNLOCK(); 827 828 return (0); 829 } 830 831 /* 832 * Process a received IGMPv2 general or group-specific query. 833 */ 834 static int 835 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip, 836 const struct igmp *igmp) 837 { 838 struct ifmultiaddr *ifma; 839 struct igmp_ifsoftc *igi; 840 struct in_multi *inm; 841 int is_general_query; 842 uint16_t timer; 843 844 NET_EPOCH_ASSERT(); 845 846 is_general_query = 0; 847 848 /* 849 * Validate address fields upfront. 850 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 851 */ 852 if (in_nullhost(igmp->igmp_group)) { 853 /* 854 * IGMPv2 General Query. 855 * If this was not sent to the all-hosts group, ignore it. 856 */ 857 if (!in_allhosts(ip->ip_dst)) 858 return (0); 859 IGMPSTAT_INC(igps_rcv_gen_queries); 860 is_general_query = 1; 861 } else { 862 /* IGMPv2 Group-Specific Query. */ 863 IGMPSTAT_INC(igps_rcv_group_queries); 864 } 865 866 IN_MULTI_LIST_LOCK(); 867 IGMP_LOCK(); 868 869 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 870 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 871 872 if (igi->igi_flags & IGIF_LOOPBACK) { 873 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)", 874 ifp, ifp->if_xname); 875 goto out_locked; 876 } 877 878 /* 879 * Ignore v2 query if in v1 Compatibility Mode. 880 */ 881 if (igi->igi_version == IGMP_VERSION_1) 882 goto out_locked; 883 884 igmp_set_version(igi, IGMP_VERSION_2); 885 886 timer = igmp->igmp_code * IGMP_FASTHZ / IGMP_TIMER_SCALE; 887 if (timer == 0) 888 timer = 1; 889 890 if (is_general_query) { 891 /* 892 * For each reporting group joined on this 893 * interface, kick the report timer. 894 */ 895 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)", 896 ifp, ifp->if_xname); 897 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 898 inm = inm_ifmultiaddr_get_inm(ifma); 899 if (inm == NULL) 900 continue; 901 igmp_v2_update_group(inm, timer); 902 } 903 } else { 904 /* 905 * Group-specific IGMPv2 query, we need only 906 * look up the single group to process it. 907 */ 908 inm = inm_lookup(ifp, igmp->igmp_group); 909 if (inm != NULL) { 910 CTR3(KTR_IGMPV3, 911 "process v2 query 0x%08x on ifp %p(%s)", 912 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 913 igmp_v2_update_group(inm, timer); 914 } 915 } 916 917 out_locked: 918 IGMP_UNLOCK(); 919 IN_MULTI_LIST_UNLOCK(); 920 921 return (0); 922 } 923 924 /* 925 * Update the report timer on a group in response to an IGMPv2 query. 926 * 927 * If we are becoming the reporting member for this group, start the timer. 928 * If we already are the reporting member for this group, and timer is 929 * below the threshold, reset it. 930 * 931 * We may be updating the group for the first time since we switched 932 * to IGMPv3. If we are, then we must clear any recorded source lists, 933 * and transition to REPORTING state; the group timer is overloaded 934 * for group and group-source query responses. 935 * 936 * Unlike IGMPv3, the delay per group should be jittered 937 * to avoid bursts of IGMPv2 reports. 938 */ 939 static void 940 igmp_v2_update_group(struct in_multi *inm, const int timer) 941 { 942 943 CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__, 944 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer); 945 946 IN_MULTI_LIST_LOCK_ASSERT(); 947 948 switch (inm->inm_state) { 949 case IGMP_NOT_MEMBER: 950 case IGMP_SILENT_MEMBER: 951 break; 952 case IGMP_REPORTING_MEMBER: 953 if (inm->inm_timer != 0 && 954 inm->inm_timer <= timer) { 955 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, " 956 "skipping.", __func__); 957 break; 958 } 959 /* FALLTHROUGH */ 960 case IGMP_SG_QUERY_PENDING_MEMBER: 961 case IGMP_G_QUERY_PENDING_MEMBER: 962 case IGMP_IDLE_MEMBER: 963 case IGMP_LAZY_MEMBER: 964 case IGMP_AWAKENING_MEMBER: 965 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__); 966 inm->inm_state = IGMP_REPORTING_MEMBER; 967 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 968 V_current_state_timers_running = 1; 969 break; 970 case IGMP_SLEEPING_MEMBER: 971 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__); 972 inm->inm_state = IGMP_AWAKENING_MEMBER; 973 break; 974 case IGMP_LEAVING_MEMBER: 975 break; 976 } 977 } 978 979 /* 980 * Process a received IGMPv3 general, group-specific or 981 * group-and-source-specific query. 982 * Assumes m has already been pulled up to the full IGMP message length. 983 * Return 0 if successful, otherwise an appropriate error code is returned. 984 */ 985 static int 986 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip, 987 /*const*/ struct igmpv3 *igmpv3) 988 { 989 struct igmp_ifsoftc *igi; 990 struct in_multi *inm; 991 int is_general_query; 992 uint32_t maxresp, nsrc, qqi; 993 uint16_t timer; 994 uint8_t qrv; 995 996 is_general_query = 0; 997 998 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname); 999 1000 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */ 1001 if (maxresp >= 128) { 1002 maxresp = IGMP_MANT(igmpv3->igmp_code) << 1003 (IGMP_EXP(igmpv3->igmp_code) + 3); 1004 } 1005 1006 /* 1007 * Robustness must never be less than 2 for on-wire IGMPv3. 1008 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make 1009 * an exception for interfaces whose IGMPv3 state changes 1010 * are redirected to loopback (e.g. MANET). 1011 */ 1012 qrv = IGMP_QRV(igmpv3->igmp_misc); 1013 if (qrv < 2) { 1014 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__, 1015 qrv, IGMP_RV_INIT); 1016 qrv = IGMP_RV_INIT; 1017 } 1018 1019 qqi = igmpv3->igmp_qqi; 1020 if (qqi >= 128) { 1021 qqi = IGMP_MANT(igmpv3->igmp_qqi) << 1022 (IGMP_EXP(igmpv3->igmp_qqi) + 3); 1023 } 1024 1025 timer = maxresp * IGMP_FASTHZ / IGMP_TIMER_SCALE; 1026 if (timer == 0) 1027 timer = 1; 1028 1029 nsrc = ntohs(igmpv3->igmp_numsrc); 1030 1031 /* 1032 * Validate address fields and versions upfront before 1033 * accepting v3 query. 1034 * XXX SMPng: Unlocked access to igmpstat counters here. 1035 */ 1036 if (in_nullhost(igmpv3->igmp_group)) { 1037 /* 1038 * IGMPv3 General Query. 1039 * 1040 * General Queries SHOULD be directed to 224.0.0.1. 1041 * A general query with a source list has undefined 1042 * behaviour; discard it. 1043 */ 1044 IGMPSTAT_INC(igps_rcv_gen_queries); 1045 if (!in_allhosts(ip->ip_dst) || nsrc > 0) { 1046 IGMPSTAT_INC(igps_rcv_badqueries); 1047 return (0); 1048 } 1049 is_general_query = 1; 1050 } else { 1051 /* Group or group-source specific query. */ 1052 if (nsrc == 0) 1053 IGMPSTAT_INC(igps_rcv_group_queries); 1054 else 1055 IGMPSTAT_INC(igps_rcv_gsr_queries); 1056 } 1057 1058 IN_MULTI_LIST_LOCK(); 1059 IGMP_LOCK(); 1060 1061 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 1062 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 1063 1064 if (igi->igi_flags & IGIF_LOOPBACK) { 1065 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)", 1066 ifp, ifp->if_xname); 1067 goto out_locked; 1068 } 1069 1070 /* 1071 * Discard the v3 query if we're in Compatibility Mode. 1072 * The RFC is not obviously worded that hosts need to stay in 1073 * compatibility mode until the Old Version Querier Present 1074 * timer expires. 1075 */ 1076 if (igi->igi_version != IGMP_VERSION_3) { 1077 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)", 1078 igi->igi_version, ifp, ifp->if_xname); 1079 goto out_locked; 1080 } 1081 1082 igmp_set_version(igi, IGMP_VERSION_3); 1083 igi->igi_rv = qrv; 1084 igi->igi_qi = qqi; 1085 igi->igi_qri = maxresp; 1086 1087 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi, 1088 maxresp); 1089 1090 if (is_general_query) { 1091 /* 1092 * Schedule a current-state report on this ifp for 1093 * all groups, possibly containing source lists. 1094 * If there is a pending General Query response 1095 * scheduled earlier than the selected delay, do 1096 * not schedule any other reports. 1097 * Otherwise, reset the interface timer. 1098 */ 1099 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)", 1100 ifp, ifp->if_xname); 1101 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) { 1102 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer); 1103 V_interface_timers_running = 1; 1104 } 1105 } else { 1106 /* 1107 * Group-source-specific queries are throttled on 1108 * a per-group basis to defeat denial-of-service attempts. 1109 * Queries for groups we are not a member of on this 1110 * link are simply ignored. 1111 */ 1112 inm = inm_lookup(ifp, igmpv3->igmp_group); 1113 if (inm == NULL) 1114 goto out_locked; 1115 if (nsrc > 0) { 1116 if (!ratecheck(&inm->inm_lastgsrtv, 1117 &V_igmp_gsrdelay)) { 1118 CTR1(KTR_IGMPV3, "%s: GS query throttled.", 1119 __func__); 1120 IGMPSTAT_INC(igps_drop_gsr_queries); 1121 goto out_locked; 1122 } 1123 } 1124 CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)", 1125 ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname); 1126 /* 1127 * If there is a pending General Query response 1128 * scheduled sooner than the selected delay, no 1129 * further report need be scheduled. 1130 * Otherwise, prepare to respond to the 1131 * group-specific or group-and-source query. 1132 */ 1133 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) 1134 igmp_input_v3_group_query(inm, igi, timer, igmpv3); 1135 } 1136 1137 out_locked: 1138 IGMP_UNLOCK(); 1139 IN_MULTI_LIST_UNLOCK(); 1140 1141 return (0); 1142 } 1143 1144 /* 1145 * Process a received IGMPv3 group-specific or group-and-source-specific 1146 * query. 1147 * Return <0 if any error occurred. Currently this is ignored. 1148 */ 1149 static int 1150 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi, 1151 int timer, /*const*/ struct igmpv3 *igmpv3) 1152 { 1153 int retval; 1154 uint16_t nsrc; 1155 1156 IN_MULTI_LIST_LOCK_ASSERT(); 1157 IGMP_LOCK_ASSERT(); 1158 1159 retval = 0; 1160 1161 switch (inm->inm_state) { 1162 case IGMP_NOT_MEMBER: 1163 case IGMP_SILENT_MEMBER: 1164 case IGMP_SLEEPING_MEMBER: 1165 case IGMP_LAZY_MEMBER: 1166 case IGMP_AWAKENING_MEMBER: 1167 case IGMP_IDLE_MEMBER: 1168 case IGMP_LEAVING_MEMBER: 1169 return (retval); 1170 break; 1171 case IGMP_REPORTING_MEMBER: 1172 case IGMP_G_QUERY_PENDING_MEMBER: 1173 case IGMP_SG_QUERY_PENDING_MEMBER: 1174 break; 1175 } 1176 1177 nsrc = ntohs(igmpv3->igmp_numsrc); 1178 1179 /* 1180 * Deal with group-specific queries upfront. 1181 * If any group query is already pending, purge any recorded 1182 * source-list state if it exists, and schedule a query response 1183 * for this group-specific query. 1184 */ 1185 if (nsrc == 0) { 1186 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || 1187 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) { 1188 inm_clear_recorded(inm); 1189 timer = min(inm->inm_timer, timer); 1190 } 1191 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER; 1192 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1193 V_current_state_timers_running = 1; 1194 return (retval); 1195 } 1196 1197 /* 1198 * Deal with the case where a group-and-source-specific query has 1199 * been received but a group-specific query is already pending. 1200 */ 1201 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) { 1202 timer = min(inm->inm_timer, timer); 1203 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1204 V_current_state_timers_running = 1; 1205 return (retval); 1206 } 1207 1208 /* 1209 * Finally, deal with the case where a group-and-source-specific 1210 * query has been received, where a response to a previous g-s-r 1211 * query exists, or none exists. 1212 * In this case, we need to parse the source-list which the Querier 1213 * has provided us with and check if we have any source list filter 1214 * entries at T1 for these sources. If we do not, there is no need 1215 * schedule a report and the query may be dropped. 1216 * If we do, we must record them and schedule a current-state 1217 * report for those sources. 1218 * FIXME: Handling source lists larger than 1 mbuf requires that 1219 * we pass the mbuf chain pointer down to this function, and use 1220 * m_getptr() to walk the chain. 1221 */ 1222 if (inm->inm_nsrc > 0) { 1223 const struct in_addr *ap; 1224 int i, nrecorded; 1225 1226 ap = (const struct in_addr *)(igmpv3 + 1); 1227 nrecorded = 0; 1228 for (i = 0; i < nsrc; i++, ap++) { 1229 retval = inm_record_source(inm, ap->s_addr); 1230 if (retval < 0) 1231 break; 1232 nrecorded += retval; 1233 } 1234 if (nrecorded > 0) { 1235 CTR1(KTR_IGMPV3, 1236 "%s: schedule response to SG query", __func__); 1237 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER; 1238 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1239 V_current_state_timers_running = 1; 1240 } 1241 } 1242 1243 return (retval); 1244 } 1245 1246 /* 1247 * Process a received IGMPv1 host membership report. 1248 * 1249 * NOTE: 0.0.0.0 workaround breaks const correctness. 1250 */ 1251 static int 1252 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1253 /*const*/ struct igmp *igmp) 1254 { 1255 struct in_ifaddr *ia; 1256 struct in_multi *inm; 1257 1258 IGMPSTAT_INC(igps_rcv_reports); 1259 1260 if (ifp->if_flags & IFF_LOOPBACK) 1261 return (0); 1262 1263 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || 1264 !in_hosteq(igmp->igmp_group, ip->ip_dst)) { 1265 IGMPSTAT_INC(igps_rcv_badreports); 1266 return (EINVAL); 1267 } 1268 1269 /* 1270 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1271 * Booting clients may use the source address 0.0.0.0. Some 1272 * IGMP daemons may not know how to use IP_RECVIF to determine 1273 * the interface upon which this message was received. 1274 * Replace 0.0.0.0 with the subnet address if told to do so. 1275 */ 1276 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1277 IFP_TO_IA(ifp, ia); 1278 if (ia != NULL) 1279 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1280 } 1281 1282 CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)", 1283 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 1284 1285 /* 1286 * IGMPv1 report suppression. 1287 * If we are a member of this group, and our membership should be 1288 * reported, stop our group timer and transition to the 'lazy' state. 1289 */ 1290 IN_MULTI_LIST_LOCK(); 1291 inm = inm_lookup(ifp, igmp->igmp_group); 1292 if (inm != NULL) { 1293 struct igmp_ifsoftc *igi; 1294 1295 igi = inm->inm_igi; 1296 if (igi == NULL) { 1297 KASSERT(igi != NULL, 1298 ("%s: no igi for ifp %p", __func__, ifp)); 1299 goto out_locked; 1300 } 1301 1302 IGMPSTAT_INC(igps_rcv_ourreports); 1303 1304 /* 1305 * If we are in IGMPv3 host mode, do not allow the 1306 * other host's IGMPv1 report to suppress our reports 1307 * unless explicitly configured to do so. 1308 */ 1309 if (igi->igi_version == IGMP_VERSION_3) { 1310 if (V_igmp_legacysupp) 1311 igmp_v3_suppress_group_record(inm); 1312 goto out_locked; 1313 } 1314 1315 inm->inm_timer = 0; 1316 1317 switch (inm->inm_state) { 1318 case IGMP_NOT_MEMBER: 1319 case IGMP_SILENT_MEMBER: 1320 break; 1321 case IGMP_IDLE_MEMBER: 1322 case IGMP_LAZY_MEMBER: 1323 case IGMP_AWAKENING_MEMBER: 1324 CTR3(KTR_IGMPV3, 1325 "report suppressed for 0x%08x on ifp %p(%s)", 1326 ntohl(igmp->igmp_group.s_addr), ifp, 1327 ifp->if_xname); 1328 case IGMP_SLEEPING_MEMBER: 1329 inm->inm_state = IGMP_SLEEPING_MEMBER; 1330 break; 1331 case IGMP_REPORTING_MEMBER: 1332 CTR3(KTR_IGMPV3, 1333 "report suppressed for 0x%08x on ifp %p(%s)", 1334 ntohl(igmp->igmp_group.s_addr), ifp, 1335 ifp->if_xname); 1336 if (igi->igi_version == IGMP_VERSION_1) 1337 inm->inm_state = IGMP_LAZY_MEMBER; 1338 else if (igi->igi_version == IGMP_VERSION_2) 1339 inm->inm_state = IGMP_SLEEPING_MEMBER; 1340 break; 1341 case IGMP_G_QUERY_PENDING_MEMBER: 1342 case IGMP_SG_QUERY_PENDING_MEMBER: 1343 case IGMP_LEAVING_MEMBER: 1344 break; 1345 } 1346 } 1347 1348 out_locked: 1349 IN_MULTI_LIST_UNLOCK(); 1350 1351 return (0); 1352 } 1353 1354 /* 1355 * Process a received IGMPv2 host membership report. 1356 * 1357 * NOTE: 0.0.0.0 workaround breaks const correctness. 1358 */ 1359 static int 1360 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1361 /*const*/ struct igmp *igmp) 1362 { 1363 struct in_ifaddr *ia; 1364 struct in_multi *inm; 1365 1366 /* 1367 * Make sure we don't hear our own membership report. Fast 1368 * leave requires knowing that we are the only member of a 1369 * group. 1370 */ 1371 IFP_TO_IA(ifp, ia); 1372 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) { 1373 return (0); 1374 } 1375 1376 IGMPSTAT_INC(igps_rcv_reports); 1377 1378 if (ifp->if_flags & IFF_LOOPBACK) { 1379 return (0); 1380 } 1381 1382 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || 1383 !in_hosteq(igmp->igmp_group, ip->ip_dst)) { 1384 IGMPSTAT_INC(igps_rcv_badreports); 1385 return (EINVAL); 1386 } 1387 1388 /* 1389 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1390 * Booting clients may use the source address 0.0.0.0. Some 1391 * IGMP daemons may not know how to use IP_RECVIF to determine 1392 * the interface upon which this message was received. 1393 * Replace 0.0.0.0 with the subnet address if told to do so. 1394 */ 1395 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1396 if (ia != NULL) 1397 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1398 } 1399 1400 CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)", 1401 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 1402 1403 /* 1404 * IGMPv2 report suppression. 1405 * If we are a member of this group, and our membership should be 1406 * reported, and our group timer is pending or about to be reset, 1407 * stop our group timer by transitioning to the 'lazy' state. 1408 */ 1409 IN_MULTI_LIST_LOCK(); 1410 inm = inm_lookup(ifp, igmp->igmp_group); 1411 if (inm != NULL) { 1412 struct igmp_ifsoftc *igi; 1413 1414 igi = inm->inm_igi; 1415 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp)); 1416 1417 IGMPSTAT_INC(igps_rcv_ourreports); 1418 1419 /* 1420 * If we are in IGMPv3 host mode, do not allow the 1421 * other host's IGMPv1 report to suppress our reports 1422 * unless explicitly configured to do so. 1423 */ 1424 if (igi->igi_version == IGMP_VERSION_3) { 1425 if (V_igmp_legacysupp) 1426 igmp_v3_suppress_group_record(inm); 1427 goto out_locked; 1428 } 1429 1430 inm->inm_timer = 0; 1431 1432 switch (inm->inm_state) { 1433 case IGMP_NOT_MEMBER: 1434 case IGMP_SILENT_MEMBER: 1435 case IGMP_SLEEPING_MEMBER: 1436 break; 1437 case IGMP_REPORTING_MEMBER: 1438 case IGMP_IDLE_MEMBER: 1439 case IGMP_AWAKENING_MEMBER: 1440 CTR3(KTR_IGMPV3, 1441 "report suppressed for 0x%08x on ifp %p(%s)", 1442 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 1443 case IGMP_LAZY_MEMBER: 1444 inm->inm_state = IGMP_LAZY_MEMBER; 1445 break; 1446 case IGMP_G_QUERY_PENDING_MEMBER: 1447 case IGMP_SG_QUERY_PENDING_MEMBER: 1448 case IGMP_LEAVING_MEMBER: 1449 break; 1450 } 1451 } 1452 1453 out_locked: 1454 IN_MULTI_LIST_UNLOCK(); 1455 1456 return (0); 1457 } 1458 1459 int 1460 igmp_input(struct mbuf **mp, int *offp, int proto) 1461 { 1462 int iphlen; 1463 struct ifnet *ifp; 1464 struct igmp *igmp; 1465 struct ip *ip; 1466 struct mbuf *m; 1467 int igmplen; 1468 int minlen; 1469 int queryver; 1470 1471 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp); 1472 1473 m = *mp; 1474 ifp = m->m_pkthdr.rcvif; 1475 *mp = NULL; 1476 1477 IGMPSTAT_INC(igps_rcv_total); 1478 1479 ip = mtod(m, struct ip *); 1480 iphlen = *offp; 1481 igmplen = ntohs(ip->ip_len) - iphlen; 1482 1483 /* 1484 * Validate lengths. 1485 */ 1486 if (igmplen < IGMP_MINLEN) { 1487 IGMPSTAT_INC(igps_rcv_tooshort); 1488 m_freem(m); 1489 return (IPPROTO_DONE); 1490 } 1491 1492 /* 1493 * Always pullup to the minimum size for v1/v2 or v3 1494 * to amortize calls to m_pullup(). 1495 */ 1496 minlen = iphlen; 1497 if (igmplen >= IGMP_V3_QUERY_MINLEN) 1498 minlen += IGMP_V3_QUERY_MINLEN; 1499 else 1500 minlen += IGMP_MINLEN; 1501 if ((!M_WRITABLE(m) || m->m_len < minlen) && 1502 (m = m_pullup(m, minlen)) == NULL) { 1503 IGMPSTAT_INC(igps_rcv_tooshort); 1504 return (IPPROTO_DONE); 1505 } 1506 ip = mtod(m, struct ip *); 1507 1508 /* 1509 * Validate checksum. 1510 */ 1511 m->m_data += iphlen; 1512 m->m_len -= iphlen; 1513 igmp = mtod(m, struct igmp *); 1514 if (in_cksum(m, igmplen)) { 1515 IGMPSTAT_INC(igps_rcv_badsum); 1516 m_freem(m); 1517 return (IPPROTO_DONE); 1518 } 1519 m->m_data -= iphlen; 1520 m->m_len += iphlen; 1521 1522 /* 1523 * IGMP control traffic is link-scope, and must have a TTL of 1. 1524 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception; 1525 * probe packets may come from beyond the LAN. 1526 */ 1527 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) { 1528 IGMPSTAT_INC(igps_rcv_badttl); 1529 m_freem(m); 1530 return (IPPROTO_DONE); 1531 } 1532 1533 switch (igmp->igmp_type) { 1534 case IGMP_HOST_MEMBERSHIP_QUERY: 1535 if (igmplen == IGMP_MINLEN) { 1536 if (igmp->igmp_code == 0) 1537 queryver = IGMP_VERSION_1; 1538 else 1539 queryver = IGMP_VERSION_2; 1540 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) { 1541 queryver = IGMP_VERSION_3; 1542 } else { 1543 IGMPSTAT_INC(igps_rcv_tooshort); 1544 m_freem(m); 1545 return (IPPROTO_DONE); 1546 } 1547 1548 switch (queryver) { 1549 case IGMP_VERSION_1: 1550 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1551 if (!V_igmp_v1enable) 1552 break; 1553 if (igmp_input_v1_query(ifp, ip, igmp) != 0) { 1554 m_freem(m); 1555 return (IPPROTO_DONE); 1556 } 1557 break; 1558 1559 case IGMP_VERSION_2: 1560 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1561 if (!V_igmp_v2enable) 1562 break; 1563 if (igmp_input_v2_query(ifp, ip, igmp) != 0) { 1564 m_freem(m); 1565 return (IPPROTO_DONE); 1566 } 1567 break; 1568 1569 case IGMP_VERSION_3: { 1570 struct igmpv3 *igmpv3; 1571 uint16_t igmpv3len; 1572 uint16_t nsrc; 1573 1574 IGMPSTAT_INC(igps_rcv_v3_queries); 1575 igmpv3 = (struct igmpv3 *)igmp; 1576 /* 1577 * Validate length based on source count. 1578 */ 1579 nsrc = ntohs(igmpv3->igmp_numsrc); 1580 if (nsrc * sizeof(in_addr_t) > 1581 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) { 1582 IGMPSTAT_INC(igps_rcv_tooshort); 1583 m_freem(m); 1584 return (IPPROTO_DONE); 1585 } 1586 /* 1587 * m_pullup() may modify m, so pullup in 1588 * this scope. 1589 */ 1590 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN + 1591 sizeof(struct in_addr) * nsrc; 1592 if ((!M_WRITABLE(m) || 1593 m->m_len < igmpv3len) && 1594 (m = m_pullup(m, igmpv3len)) == NULL) { 1595 IGMPSTAT_INC(igps_rcv_tooshort); 1596 return (IPPROTO_DONE); 1597 } 1598 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *) 1599 + iphlen); 1600 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) { 1601 m_freem(m); 1602 return (IPPROTO_DONE); 1603 } 1604 } 1605 break; 1606 } 1607 break; 1608 1609 case IGMP_v1_HOST_MEMBERSHIP_REPORT: 1610 if (!V_igmp_v1enable) 1611 break; 1612 if (igmp_input_v1_report(ifp, ip, igmp) != 0) { 1613 m_freem(m); 1614 return (IPPROTO_DONE); 1615 } 1616 break; 1617 1618 case IGMP_v2_HOST_MEMBERSHIP_REPORT: 1619 if (!V_igmp_v2enable) 1620 break; 1621 if (!ip_checkrouteralert(m)) 1622 IGMPSTAT_INC(igps_rcv_nora); 1623 if (igmp_input_v2_report(ifp, ip, igmp) != 0) { 1624 m_freem(m); 1625 return (IPPROTO_DONE); 1626 } 1627 break; 1628 1629 case IGMP_v3_HOST_MEMBERSHIP_REPORT: 1630 /* 1631 * Hosts do not need to process IGMPv3 membership reports, 1632 * as report suppression is no longer required. 1633 */ 1634 if (!ip_checkrouteralert(m)) 1635 IGMPSTAT_INC(igps_rcv_nora); 1636 break; 1637 1638 default: 1639 break; 1640 } 1641 1642 /* 1643 * Pass all valid IGMP packets up to any process(es) listening on a 1644 * raw IGMP socket. 1645 */ 1646 *mp = m; 1647 return (rip_input(mp, offp, proto)); 1648 } 1649 1650 /* 1651 * Fast timeout handler (global). 1652 * VIMAGE: Timeout handlers are expected to service all vimages. 1653 */ 1654 static struct callout igmpfast_callout; 1655 static void 1656 igmp_fasttimo(void *arg __unused) 1657 { 1658 struct epoch_tracker et; 1659 VNET_ITERATOR_DECL(vnet_iter); 1660 1661 NET_EPOCH_ENTER(et); 1662 VNET_LIST_RLOCK_NOSLEEP(); 1663 VNET_FOREACH(vnet_iter) { 1664 CURVNET_SET(vnet_iter); 1665 igmp_fasttimo_vnet(); 1666 CURVNET_RESTORE(); 1667 } 1668 VNET_LIST_RUNLOCK_NOSLEEP(); 1669 NET_EPOCH_EXIT(et); 1670 1671 callout_reset(&igmpfast_callout, hz / IGMP_FASTHZ, igmp_fasttimo, NULL); 1672 } 1673 1674 /* 1675 * Fast timeout handler (per-vnet). 1676 * 1677 * VIMAGE: Assume caller has set up our curvnet. 1678 */ 1679 static void 1680 igmp_fasttimo_vnet(void) 1681 { 1682 struct mbufq scq; /* State-change packets */ 1683 struct mbufq qrq; /* Query response packets */ 1684 struct ifnet *ifp; 1685 struct igmp_ifsoftc *igi; 1686 struct ifmultiaddr *ifma; 1687 struct in_multi *inm; 1688 struct in_multi_head inm_free_tmp; 1689 int loop, uri_fasthz; 1690 1691 loop = 0; 1692 uri_fasthz = 0; 1693 1694 /* 1695 * Quick check to see if any work needs to be done, in order to 1696 * minimize the overhead of fasttimo processing. 1697 * SMPng: XXX Unlocked reads. 1698 */ 1699 if (!V_current_state_timers_running && 1700 !V_interface_timers_running && 1701 !V_state_change_timers_running) 1702 return; 1703 1704 SLIST_INIT(&inm_free_tmp); 1705 IN_MULTI_LIST_LOCK(); 1706 IGMP_LOCK(); 1707 1708 /* 1709 * IGMPv3 General Query response timer processing. 1710 */ 1711 if (V_interface_timers_running) { 1712 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__); 1713 1714 V_interface_timers_running = 0; 1715 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1716 if (igi->igi_v3_timer == 0) { 1717 /* Do nothing. */ 1718 } else if (--igi->igi_v3_timer == 0) { 1719 igmp_v3_dispatch_general_query(igi); 1720 } else { 1721 V_interface_timers_running = 1; 1722 } 1723 } 1724 } 1725 1726 if (!V_current_state_timers_running && 1727 !V_state_change_timers_running) 1728 goto out_locked; 1729 1730 V_current_state_timers_running = 0; 1731 V_state_change_timers_running = 0; 1732 1733 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__); 1734 1735 /* 1736 * IGMPv1/v2/v3 host report and state-change timer processing. 1737 * Note: Processing a v3 group timer may remove a node. 1738 */ 1739 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1740 ifp = igi->igi_ifp; 1741 1742 if (igi->igi_version == IGMP_VERSION_3) { 1743 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; 1744 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri * 1745 IGMP_FASTHZ); 1746 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS); 1747 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS); 1748 } 1749 1750 IF_ADDR_WLOCK(ifp); 1751 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1752 inm = inm_ifmultiaddr_get_inm(ifma); 1753 if (inm == NULL) 1754 continue; 1755 switch (igi->igi_version) { 1756 case IGMP_VERSION_1: 1757 case IGMP_VERSION_2: 1758 igmp_v1v2_process_group_timer(inm, 1759 igi->igi_version); 1760 break; 1761 case IGMP_VERSION_3: 1762 igmp_v3_process_group_timers(&inm_free_tmp, &qrq, 1763 &scq, inm, uri_fasthz); 1764 break; 1765 } 1766 } 1767 IF_ADDR_WUNLOCK(ifp); 1768 1769 if (igi->igi_version == IGMP_VERSION_3) { 1770 igmp_dispatch_queue(&qrq, 0, loop); 1771 igmp_dispatch_queue(&scq, 0, loop); 1772 1773 /* 1774 * Free the in_multi reference(s) for this 1775 * IGMP lifecycle. 1776 */ 1777 inm_release_list_deferred(&inm_free_tmp); 1778 } 1779 } 1780 1781 out_locked: 1782 IGMP_UNLOCK(); 1783 IN_MULTI_LIST_UNLOCK(); 1784 } 1785 1786 /* 1787 * Update host report group timer for IGMPv1/v2. 1788 * Will update the global pending timer flags. 1789 */ 1790 static void 1791 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version) 1792 { 1793 int report_timer_expired; 1794 1795 IN_MULTI_LIST_LOCK_ASSERT(); 1796 IGMP_LOCK_ASSERT(); 1797 1798 if (inm->inm_timer == 0) { 1799 report_timer_expired = 0; 1800 } else if (--inm->inm_timer == 0) { 1801 report_timer_expired = 1; 1802 } else { 1803 V_current_state_timers_running = 1; 1804 return; 1805 } 1806 1807 switch (inm->inm_state) { 1808 case IGMP_NOT_MEMBER: 1809 case IGMP_SILENT_MEMBER: 1810 case IGMP_IDLE_MEMBER: 1811 case IGMP_LAZY_MEMBER: 1812 case IGMP_SLEEPING_MEMBER: 1813 case IGMP_AWAKENING_MEMBER: 1814 break; 1815 case IGMP_REPORTING_MEMBER: 1816 if (report_timer_expired) { 1817 inm->inm_state = IGMP_IDLE_MEMBER; 1818 (void)igmp_v1v2_queue_report(inm, 1819 (version == IGMP_VERSION_2) ? 1820 IGMP_v2_HOST_MEMBERSHIP_REPORT : 1821 IGMP_v1_HOST_MEMBERSHIP_REPORT); 1822 } 1823 break; 1824 case IGMP_G_QUERY_PENDING_MEMBER: 1825 case IGMP_SG_QUERY_PENDING_MEMBER: 1826 case IGMP_LEAVING_MEMBER: 1827 break; 1828 } 1829 } 1830 1831 /* 1832 * Update a group's timers for IGMPv3. 1833 * Will update the global pending timer flags. 1834 * Note: Unlocked read from igi. 1835 */ 1836 static void 1837 igmp_v3_process_group_timers(struct in_multi_head *inmh, 1838 struct mbufq *qrq, struct mbufq *scq, 1839 struct in_multi *inm, const int uri_fasthz) 1840 { 1841 int query_response_timer_expired; 1842 int state_change_retransmit_timer_expired; 1843 1844 IN_MULTI_LIST_LOCK_ASSERT(); 1845 IGMP_LOCK_ASSERT(); 1846 1847 query_response_timer_expired = 0; 1848 state_change_retransmit_timer_expired = 0; 1849 1850 /* 1851 * During a transition from v1/v2 compatibility mode back to v3, 1852 * a group record in REPORTING state may still have its group 1853 * timer active. This is a no-op in this function; it is easier 1854 * to deal with it here than to complicate the slow-timeout path. 1855 */ 1856 if (inm->inm_timer == 0) { 1857 query_response_timer_expired = 0; 1858 } else if (--inm->inm_timer == 0) { 1859 query_response_timer_expired = 1; 1860 } else { 1861 V_current_state_timers_running = 1; 1862 } 1863 1864 if (inm->inm_sctimer == 0) { 1865 state_change_retransmit_timer_expired = 0; 1866 } else if (--inm->inm_sctimer == 0) { 1867 state_change_retransmit_timer_expired = 1; 1868 } else { 1869 V_state_change_timers_running = 1; 1870 } 1871 1872 /* We are in fasttimo, so be quick about it. */ 1873 if (!state_change_retransmit_timer_expired && 1874 !query_response_timer_expired) 1875 return; 1876 1877 switch (inm->inm_state) { 1878 case IGMP_NOT_MEMBER: 1879 case IGMP_SILENT_MEMBER: 1880 case IGMP_SLEEPING_MEMBER: 1881 case IGMP_LAZY_MEMBER: 1882 case IGMP_AWAKENING_MEMBER: 1883 case IGMP_IDLE_MEMBER: 1884 break; 1885 case IGMP_G_QUERY_PENDING_MEMBER: 1886 case IGMP_SG_QUERY_PENDING_MEMBER: 1887 /* 1888 * Respond to a previously pending Group-Specific 1889 * or Group-and-Source-Specific query by enqueueing 1890 * the appropriate Current-State report for 1891 * immediate transmission. 1892 */ 1893 if (query_response_timer_expired) { 1894 int retval __unused; 1895 1896 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1, 1897 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)); 1898 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 1899 __func__, retval); 1900 inm->inm_state = IGMP_REPORTING_MEMBER; 1901 /* XXX Clear recorded sources for next time. */ 1902 inm_clear_recorded(inm); 1903 } 1904 /* FALLTHROUGH */ 1905 case IGMP_REPORTING_MEMBER: 1906 case IGMP_LEAVING_MEMBER: 1907 if (state_change_retransmit_timer_expired) { 1908 /* 1909 * State-change retransmission timer fired. 1910 * If there are any further pending retransmissions, 1911 * set the global pending state-change flag, and 1912 * reset the timer. 1913 */ 1914 if (--inm->inm_scrv > 0) { 1915 inm->inm_sctimer = uri_fasthz; 1916 V_state_change_timers_running = 1; 1917 } 1918 /* 1919 * Retransmit the previously computed state-change 1920 * report. If there are no further pending 1921 * retransmissions, the mbuf queue will be consumed. 1922 * Update T0 state to T1 as we have now sent 1923 * a state-change. 1924 */ 1925 (void)igmp_v3_merge_state_changes(inm, scq); 1926 1927 inm_commit(inm); 1928 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 1929 ntohl(inm->inm_addr.s_addr), 1930 inm->inm_ifp->if_xname); 1931 1932 /* 1933 * If we are leaving the group for good, make sure 1934 * we release IGMP's reference to it. 1935 * This release must be deferred using a SLIST, 1936 * as we are called from a loop which traverses 1937 * the in_ifmultiaddr TAILQ. 1938 */ 1939 if (inm->inm_state == IGMP_LEAVING_MEMBER && 1940 inm->inm_scrv == 0) { 1941 inm->inm_state = IGMP_NOT_MEMBER; 1942 inm_rele_locked(inmh, inm); 1943 } 1944 } 1945 break; 1946 } 1947 } 1948 1949 /* 1950 * Suppress a group's pending response to a group or source/group query. 1951 * 1952 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency. 1953 * Do NOT update ST1/ST0 as this operation merely suppresses 1954 * the currently pending group record. 1955 * Do NOT suppress the response to a general query. It is possible but 1956 * it would require adding another state or flag. 1957 */ 1958 static void 1959 igmp_v3_suppress_group_record(struct in_multi *inm) 1960 { 1961 1962 IN_MULTI_LIST_LOCK_ASSERT(); 1963 1964 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3, 1965 ("%s: not IGMPv3 mode on link", __func__)); 1966 1967 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER || 1968 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER) 1969 return; 1970 1971 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) 1972 inm_clear_recorded(inm); 1973 1974 inm->inm_timer = 0; 1975 inm->inm_state = IGMP_REPORTING_MEMBER; 1976 } 1977 1978 /* 1979 * Switch to a different IGMP version on the given interface, 1980 * as per Section 7.2.1. 1981 */ 1982 static void 1983 igmp_set_version(struct igmp_ifsoftc *igi, const int version) 1984 { 1985 int old_version_timer; 1986 1987 IGMP_LOCK_ASSERT(); 1988 1989 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__, 1990 version, igi->igi_ifp, igi->igi_ifp->if_xname); 1991 1992 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) { 1993 /* 1994 * Compute the "Older Version Querier Present" timer as per 1995 * Section 8.12. 1996 */ 1997 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri; 1998 old_version_timer *= IGMP_SLOWHZ; 1999 2000 if (version == IGMP_VERSION_1) { 2001 igi->igi_v1_timer = old_version_timer; 2002 igi->igi_v2_timer = 0; 2003 } else if (version == IGMP_VERSION_2) { 2004 igi->igi_v1_timer = 0; 2005 igi->igi_v2_timer = old_version_timer; 2006 } 2007 } 2008 2009 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 2010 if (igi->igi_version != IGMP_VERSION_2) { 2011 igi->igi_version = IGMP_VERSION_2; 2012 igmp_v3_cancel_link_timers(igi); 2013 } 2014 } else if (igi->igi_v1_timer > 0) { 2015 if (igi->igi_version != IGMP_VERSION_1) { 2016 igi->igi_version = IGMP_VERSION_1; 2017 igmp_v3_cancel_link_timers(igi); 2018 } 2019 } 2020 } 2021 2022 /* 2023 * Cancel pending IGMPv3 timers for the given link and all groups 2024 * joined on it; state-change, general-query, and group-query timers. 2025 * 2026 * Only ever called on a transition from v3 to Compatibility mode. Kill 2027 * the timers stone dead (this may be expensive for large N groups), they 2028 * will be restarted if Compatibility Mode deems that they must be due to 2029 * query processing. 2030 */ 2031 static void 2032 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi) 2033 { 2034 struct ifmultiaddr *ifma; 2035 struct ifnet *ifp; 2036 struct in_multi *inm; 2037 struct in_multi_head inm_free_tmp; 2038 2039 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__, 2040 igi->igi_ifp, igi->igi_ifp->if_xname); 2041 2042 IN_MULTI_LIST_LOCK_ASSERT(); 2043 IGMP_LOCK_ASSERT(); 2044 NET_EPOCH_ASSERT(); 2045 2046 SLIST_INIT(&inm_free_tmp); 2047 2048 /* 2049 * Stop the v3 General Query Response on this link stone dead. 2050 * If fasttimo is woken up due to V_interface_timers_running, 2051 * the flag will be cleared if there are no pending link timers. 2052 */ 2053 igi->igi_v3_timer = 0; 2054 2055 /* 2056 * Now clear the current-state and state-change report timers 2057 * for all memberships scoped to this link. 2058 */ 2059 ifp = igi->igi_ifp; 2060 IF_ADDR_WLOCK(ifp); 2061 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2062 inm = inm_ifmultiaddr_get_inm(ifma); 2063 if (inm == NULL) 2064 continue; 2065 switch (inm->inm_state) { 2066 case IGMP_NOT_MEMBER: 2067 case IGMP_SILENT_MEMBER: 2068 case IGMP_IDLE_MEMBER: 2069 case IGMP_LAZY_MEMBER: 2070 case IGMP_SLEEPING_MEMBER: 2071 case IGMP_AWAKENING_MEMBER: 2072 /* 2073 * These states are either not relevant in v3 mode, 2074 * or are unreported. Do nothing. 2075 */ 2076 break; 2077 case IGMP_LEAVING_MEMBER: 2078 /* 2079 * If we are leaving the group and switching to 2080 * compatibility mode, we need to release the final 2081 * reference held for issuing the INCLUDE {}, and 2082 * transition to REPORTING to ensure the host leave 2083 * message is sent upstream to the old querier -- 2084 * transition to NOT would lose the leave and race. 2085 */ 2086 inm_rele_locked(&inm_free_tmp, inm); 2087 /* FALLTHROUGH */ 2088 case IGMP_G_QUERY_PENDING_MEMBER: 2089 case IGMP_SG_QUERY_PENDING_MEMBER: 2090 inm_clear_recorded(inm); 2091 /* FALLTHROUGH */ 2092 case IGMP_REPORTING_MEMBER: 2093 inm->inm_state = IGMP_REPORTING_MEMBER; 2094 break; 2095 } 2096 /* 2097 * Always clear state-change and group report timers. 2098 * Free any pending IGMPv3 state-change records. 2099 */ 2100 inm->inm_sctimer = 0; 2101 inm->inm_timer = 0; 2102 mbufq_drain(&inm->inm_scq); 2103 } 2104 IF_ADDR_WUNLOCK(ifp); 2105 2106 inm_release_list_deferred(&inm_free_tmp); 2107 } 2108 2109 /* 2110 * Update the Older Version Querier Present timers for a link. 2111 * See Section 7.2.1 of RFC 3376. 2112 */ 2113 static void 2114 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi) 2115 { 2116 2117 IGMP_LOCK_ASSERT(); 2118 2119 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) { 2120 /* 2121 * IGMPv1 and IGMPv2 Querier Present timers expired. 2122 * 2123 * Revert to IGMPv3. 2124 */ 2125 if (igi->igi_version != IGMP_VERSION_3) { 2126 CTR5(KTR_IGMPV3, 2127 "%s: transition from v%d -> v%d on %p(%s)", 2128 __func__, igi->igi_version, IGMP_VERSION_3, 2129 igi->igi_ifp, igi->igi_ifp->if_xname); 2130 igi->igi_version = IGMP_VERSION_3; 2131 } 2132 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 2133 /* 2134 * IGMPv1 Querier Present timer expired, 2135 * IGMPv2 Querier Present timer running. 2136 * If IGMPv2 was disabled since last timeout, 2137 * revert to IGMPv3. 2138 * If IGMPv2 is enabled, revert to IGMPv2. 2139 */ 2140 if (!V_igmp_v2enable) { 2141 CTR5(KTR_IGMPV3, 2142 "%s: transition from v%d -> v%d on %p(%s)", 2143 __func__, igi->igi_version, IGMP_VERSION_3, 2144 igi->igi_ifp, igi->igi_ifp->if_xname); 2145 igi->igi_v2_timer = 0; 2146 igi->igi_version = IGMP_VERSION_3; 2147 } else { 2148 --igi->igi_v2_timer; 2149 if (igi->igi_version != IGMP_VERSION_2) { 2150 CTR5(KTR_IGMPV3, 2151 "%s: transition from v%d -> v%d on %p(%s)", 2152 __func__, igi->igi_version, IGMP_VERSION_2, 2153 igi->igi_ifp, igi->igi_ifp->if_xname); 2154 igi->igi_version = IGMP_VERSION_2; 2155 igmp_v3_cancel_link_timers(igi); 2156 } 2157 } 2158 } else if (igi->igi_v1_timer > 0) { 2159 /* 2160 * IGMPv1 Querier Present timer running. 2161 * Stop IGMPv2 timer if running. 2162 * 2163 * If IGMPv1 was disabled since last timeout, 2164 * revert to IGMPv3. 2165 * If IGMPv1 is enabled, reset IGMPv2 timer if running. 2166 */ 2167 if (!V_igmp_v1enable) { 2168 CTR5(KTR_IGMPV3, 2169 "%s: transition from v%d -> v%d on %p(%s)", 2170 __func__, igi->igi_version, IGMP_VERSION_3, 2171 igi->igi_ifp, igi->igi_ifp->if_xname); 2172 igi->igi_v1_timer = 0; 2173 igi->igi_version = IGMP_VERSION_3; 2174 } else { 2175 --igi->igi_v1_timer; 2176 } 2177 if (igi->igi_v2_timer > 0) { 2178 CTR3(KTR_IGMPV3, 2179 "%s: cancel v2 timer on %p(%s)", 2180 __func__, igi->igi_ifp, igi->igi_ifp->if_xname); 2181 igi->igi_v2_timer = 0; 2182 } 2183 } 2184 } 2185 2186 /* 2187 * Global slowtimo handler. 2188 * VIMAGE: Timeout handlers are expected to service all vimages. 2189 */ 2190 static struct callout igmpslow_callout; 2191 static void 2192 igmp_slowtimo(void *arg __unused) 2193 { 2194 struct epoch_tracker et; 2195 VNET_ITERATOR_DECL(vnet_iter); 2196 2197 NET_EPOCH_ENTER(et); 2198 VNET_LIST_RLOCK_NOSLEEP(); 2199 VNET_FOREACH(vnet_iter) { 2200 CURVNET_SET(vnet_iter); 2201 igmp_slowtimo_vnet(); 2202 CURVNET_RESTORE(); 2203 } 2204 VNET_LIST_RUNLOCK_NOSLEEP(); 2205 NET_EPOCH_EXIT(et); 2206 2207 callout_reset(&igmpslow_callout, hz / IGMP_SLOWHZ, igmp_slowtimo, NULL); 2208 } 2209 2210 /* 2211 * Per-vnet slowtimo handler. 2212 */ 2213 static void 2214 igmp_slowtimo_vnet(void) 2215 { 2216 struct igmp_ifsoftc *igi; 2217 2218 IGMP_LOCK(); 2219 2220 LIST_FOREACH(igi, &V_igi_head, igi_link) { 2221 igmp_v1v2_process_querier_timers(igi); 2222 } 2223 2224 IGMP_UNLOCK(); 2225 } 2226 2227 /* 2228 * Dispatch an IGMPv1/v2 host report or leave message. 2229 * These are always small enough to fit inside a single mbuf. 2230 */ 2231 static int 2232 igmp_v1v2_queue_report(struct in_multi *inm, const int type) 2233 { 2234 struct epoch_tracker et; 2235 struct ifnet *ifp; 2236 struct igmp *igmp; 2237 struct ip *ip; 2238 struct mbuf *m; 2239 2240 IN_MULTI_LIST_LOCK_ASSERT(); 2241 IGMP_LOCK_ASSERT(); 2242 2243 ifp = inm->inm_ifp; 2244 2245 m = m_gethdr(M_NOWAIT, MT_DATA); 2246 if (m == NULL) 2247 return (ENOMEM); 2248 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp)); 2249 2250 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp); 2251 2252 m->m_data += sizeof(struct ip); 2253 m->m_len = sizeof(struct igmp); 2254 2255 igmp = mtod(m, struct igmp *); 2256 igmp->igmp_type = type; 2257 igmp->igmp_code = 0; 2258 igmp->igmp_group = inm->inm_addr; 2259 igmp->igmp_cksum = 0; 2260 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp)); 2261 2262 m->m_data -= sizeof(struct ip); 2263 m->m_len += sizeof(struct ip); 2264 2265 ip = mtod(m, struct ip *); 2266 ip->ip_tos = 0; 2267 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp)); 2268 ip->ip_off = 0; 2269 ip->ip_p = IPPROTO_IGMP; 2270 ip->ip_src.s_addr = INADDR_ANY; 2271 2272 if (type == IGMP_HOST_LEAVE_MESSAGE) 2273 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP); 2274 else 2275 ip->ip_dst = inm->inm_addr; 2276 2277 igmp_save_context(m, ifp); 2278 2279 m->m_flags |= M_IGMPV2; 2280 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK) 2281 m->m_flags |= M_IGMP_LOOP; 2282 2283 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m); 2284 NET_EPOCH_ENTER(et); 2285 netisr_dispatch(NETISR_IGMP, m); 2286 NET_EPOCH_EXIT(et); 2287 2288 return (0); 2289 } 2290 2291 /* 2292 * Process a state change from the upper layer for the given IPv4 group. 2293 * 2294 * Each socket holds a reference on the in_multi in its own ip_moptions. 2295 * The socket layer will have made the necessary updates to.the group 2296 * state, it is now up to IGMP to issue a state change report if there 2297 * has been any change between T0 (when the last state-change was issued) 2298 * and T1 (now). 2299 * 2300 * We use the IGMPv3 state machine at group level. The IGMP module 2301 * however makes the decision as to which IGMP protocol version to speak. 2302 * A state change *from* INCLUDE {} always means an initial join. 2303 * A state change *to* INCLUDE {} always means a final leave. 2304 * 2305 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can 2306 * save ourselves a bunch of work; any exclusive mode groups need not 2307 * compute source filter lists. 2308 * 2309 * VIMAGE: curvnet should have been set by caller, as this routine 2310 * is called from the socket option handlers. 2311 */ 2312 int 2313 igmp_change_state(struct in_multi *inm) 2314 { 2315 struct igmp_ifsoftc *igi; 2316 struct ifnet *ifp; 2317 int error; 2318 2319 error = 0; 2320 IN_MULTI_LOCK_ASSERT(); 2321 /* 2322 * Try to detect if the upper layer just asked us to change state 2323 * for an interface which has now gone away. 2324 */ 2325 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 2326 ifp = inm->inm_ifma->ifma_ifp; 2327 if (ifp == NULL) 2328 return (0); 2329 /* 2330 * Sanity check that netinet's notion of ifp is the 2331 * same as net's. 2332 */ 2333 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 2334 2335 IGMP_LOCK(); 2336 2337 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 2338 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 2339 2340 /* 2341 * If we detect a state transition to or from MCAST_UNDEFINED 2342 * for this group, then we are starting or finishing an IGMP 2343 * life cycle for this group. 2344 */ 2345 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) { 2346 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__, 2347 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode); 2348 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) { 2349 CTR1(KTR_IGMPV3, "%s: initial join", __func__); 2350 error = igmp_initial_join(inm, igi); 2351 goto out_locked; 2352 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) { 2353 CTR1(KTR_IGMPV3, "%s: final leave", __func__); 2354 igmp_final_leave(inm, igi); 2355 goto out_locked; 2356 } 2357 } else { 2358 CTR1(KTR_IGMPV3, "%s: filter set change", __func__); 2359 } 2360 2361 error = igmp_handle_state_change(inm, igi); 2362 2363 out_locked: 2364 IGMP_UNLOCK(); 2365 return (error); 2366 } 2367 2368 /* 2369 * Perform the initial join for an IGMP group. 2370 * 2371 * When joining a group: 2372 * If the group should have its IGMP traffic suppressed, do nothing. 2373 * IGMPv1 starts sending IGMPv1 host membership reports. 2374 * IGMPv2 starts sending IGMPv2 host membership reports. 2375 * IGMPv3 will schedule an IGMPv3 state-change report containing the 2376 * initial state of the membership. 2377 */ 2378 static int 2379 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi) 2380 { 2381 struct ifnet *ifp; 2382 struct mbufq *mq; 2383 int error, retval, syncstates; 2384 2385 CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__, 2386 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); 2387 2388 error = 0; 2389 syncstates = 1; 2390 2391 ifp = inm->inm_ifp; 2392 2393 IN_MULTI_LOCK_ASSERT(); 2394 IGMP_LOCK_ASSERT(); 2395 2396 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); 2397 2398 /* 2399 * Groups joined on loopback or marked as 'not reported', 2400 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and 2401 * are never reported in any IGMP protocol exchanges. 2402 * All other groups enter the appropriate IGMP state machine 2403 * for the version in use on this link. 2404 * A link marked as IGIF_SILENT causes IGMP to be completely 2405 * disabled for the link. 2406 */ 2407 if ((ifp->if_flags & IFF_LOOPBACK) || 2408 (igi->igi_flags & IGIF_SILENT) || 2409 !igmp_isgroupreported(inm->inm_addr)) { 2410 CTR1(KTR_IGMPV3, 2411 "%s: not kicking state machine for silent group", __func__); 2412 inm->inm_state = IGMP_SILENT_MEMBER; 2413 inm->inm_timer = 0; 2414 } else { 2415 /* 2416 * Deal with overlapping in_multi lifecycle. 2417 * If this group was LEAVING, then make sure 2418 * we drop the reference we picked up to keep the 2419 * group around for the final INCLUDE {} enqueue. 2420 */ 2421 if (igi->igi_version == IGMP_VERSION_3 && 2422 inm->inm_state == IGMP_LEAVING_MEMBER) { 2423 MPASS(inm->inm_refcount > 1); 2424 inm_rele_locked(NULL, inm); 2425 } 2426 inm->inm_state = IGMP_REPORTING_MEMBER; 2427 2428 switch (igi->igi_version) { 2429 case IGMP_VERSION_1: 2430 case IGMP_VERSION_2: 2431 inm->inm_state = IGMP_IDLE_MEMBER; 2432 error = igmp_v1v2_queue_report(inm, 2433 (igi->igi_version == IGMP_VERSION_2) ? 2434 IGMP_v2_HOST_MEMBERSHIP_REPORT : 2435 IGMP_v1_HOST_MEMBERSHIP_REPORT); 2436 if (error == 0) { 2437 inm->inm_timer = IGMP_RANDOM_DELAY( 2438 IGMP_V1V2_MAX_RI * IGMP_FASTHZ); 2439 V_current_state_timers_running = 1; 2440 } 2441 break; 2442 2443 case IGMP_VERSION_3: 2444 /* 2445 * Defer update of T0 to T1, until the first copy 2446 * of the state change has been transmitted. 2447 */ 2448 syncstates = 0; 2449 2450 /* 2451 * Immediately enqueue a State-Change Report for 2452 * this interface, freeing any previous reports. 2453 * Don't kick the timers if there is nothing to do, 2454 * or if an error occurred. 2455 */ 2456 mq = &inm->inm_scq; 2457 mbufq_drain(mq); 2458 retval = igmp_v3_enqueue_group_record(mq, inm, 1, 2459 0, 0); 2460 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 2461 __func__, retval); 2462 if (retval <= 0) { 2463 error = retval * -1; 2464 break; 2465 } 2466 2467 /* 2468 * Schedule transmission of pending state-change 2469 * report up to RV times for this link. The timer 2470 * will fire at the next igmp_fasttimo (~200ms), 2471 * giving us an opportunity to merge the reports. 2472 */ 2473 if (igi->igi_flags & IGIF_LOOPBACK) { 2474 inm->inm_scrv = 1; 2475 } else { 2476 KASSERT(igi->igi_rv > 1, 2477 ("%s: invalid robustness %d", __func__, 2478 igi->igi_rv)); 2479 inm->inm_scrv = igi->igi_rv; 2480 } 2481 inm->inm_sctimer = 1; 2482 V_state_change_timers_running = 1; 2483 2484 error = 0; 2485 break; 2486 } 2487 } 2488 2489 /* 2490 * Only update the T0 state if state change is atomic, 2491 * i.e. we don't need to wait for a timer to fire before we 2492 * can consider the state change to have been communicated. 2493 */ 2494 if (syncstates) { 2495 inm_commit(inm); 2496 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 2497 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2498 } 2499 2500 return (error); 2501 } 2502 2503 /* 2504 * Issue an intermediate state change during the IGMP life-cycle. 2505 */ 2506 static int 2507 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi) 2508 { 2509 struct ifnet *ifp; 2510 int retval; 2511 2512 CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__, 2513 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); 2514 2515 ifp = inm->inm_ifp; 2516 2517 IN_MULTI_LIST_LOCK_ASSERT(); 2518 IGMP_LOCK_ASSERT(); 2519 2520 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); 2521 2522 if ((ifp->if_flags & IFF_LOOPBACK) || 2523 (igi->igi_flags & IGIF_SILENT) || 2524 !igmp_isgroupreported(inm->inm_addr) || 2525 (igi->igi_version != IGMP_VERSION_3)) { 2526 if (!igmp_isgroupreported(inm->inm_addr)) { 2527 CTR1(KTR_IGMPV3, 2528 "%s: not kicking state machine for silent group", __func__); 2529 } 2530 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__); 2531 inm_commit(inm); 2532 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 2533 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2534 return (0); 2535 } 2536 2537 mbufq_drain(&inm->inm_scq); 2538 2539 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0); 2540 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); 2541 if (retval <= 0) 2542 return (-retval); 2543 2544 /* 2545 * If record(s) were enqueued, start the state-change 2546 * report timer for this group. 2547 */ 2548 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv); 2549 inm->inm_sctimer = 1; 2550 V_state_change_timers_running = 1; 2551 2552 return (0); 2553 } 2554 2555 /* 2556 * Perform the final leave for an IGMP group. 2557 * 2558 * When leaving a group: 2559 * IGMPv1 does nothing. 2560 * IGMPv2 sends a host leave message, if and only if we are the reporter. 2561 * IGMPv3 enqueues a state-change report containing a transition 2562 * to INCLUDE {} for immediate transmission. 2563 */ 2564 static void 2565 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi) 2566 { 2567 int syncstates; 2568 2569 syncstates = 1; 2570 2571 CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)", 2572 __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp, 2573 inm->inm_ifp->if_xname); 2574 2575 IN_MULTI_LIST_LOCK_ASSERT(); 2576 IGMP_LOCK_ASSERT(); 2577 2578 switch (inm->inm_state) { 2579 case IGMP_NOT_MEMBER: 2580 case IGMP_SILENT_MEMBER: 2581 case IGMP_LEAVING_MEMBER: 2582 /* Already leaving or left; do nothing. */ 2583 CTR1(KTR_IGMPV3, 2584 "%s: not kicking state machine for silent group", __func__); 2585 break; 2586 case IGMP_REPORTING_MEMBER: 2587 case IGMP_IDLE_MEMBER: 2588 case IGMP_G_QUERY_PENDING_MEMBER: 2589 case IGMP_SG_QUERY_PENDING_MEMBER: 2590 if (igi->igi_version == IGMP_VERSION_2) { 2591 #ifdef INVARIANTS 2592 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || 2593 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) 2594 panic("%s: IGMPv3 state reached, not IGMPv3 mode", 2595 __func__); 2596 #endif 2597 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE); 2598 inm->inm_state = IGMP_NOT_MEMBER; 2599 } else if (igi->igi_version == IGMP_VERSION_3) { 2600 /* 2601 * Stop group timer and all pending reports. 2602 * Immediately enqueue a state-change report 2603 * TO_IN {} to be sent on the next fast timeout, 2604 * giving us an opportunity to merge reports. 2605 */ 2606 mbufq_drain(&inm->inm_scq); 2607 inm->inm_timer = 0; 2608 if (igi->igi_flags & IGIF_LOOPBACK) { 2609 inm->inm_scrv = 1; 2610 } else { 2611 inm->inm_scrv = igi->igi_rv; 2612 } 2613 CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d " 2614 "pending retransmissions.", __func__, 2615 ntohl(inm->inm_addr.s_addr), 2616 inm->inm_ifp->if_xname, inm->inm_scrv); 2617 if (inm->inm_scrv == 0) { 2618 inm->inm_state = IGMP_NOT_MEMBER; 2619 inm->inm_sctimer = 0; 2620 } else { 2621 int retval __unused; 2622 2623 inm_acquire_locked(inm); 2624 2625 retval = igmp_v3_enqueue_group_record( 2626 &inm->inm_scq, inm, 1, 0, 0); 2627 KASSERT(retval != 0, 2628 ("%s: enqueue record = %d", __func__, 2629 retval)); 2630 2631 inm->inm_state = IGMP_LEAVING_MEMBER; 2632 inm->inm_sctimer = 1; 2633 V_state_change_timers_running = 1; 2634 syncstates = 0; 2635 } 2636 break; 2637 } 2638 break; 2639 case IGMP_LAZY_MEMBER: 2640 case IGMP_SLEEPING_MEMBER: 2641 case IGMP_AWAKENING_MEMBER: 2642 /* Our reports are suppressed; do nothing. */ 2643 break; 2644 } 2645 2646 if (syncstates) { 2647 inm_commit(inm); 2648 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 2649 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2650 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 2651 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s", 2652 __func__, ntohl(inm->inm_addr.s_addr), 2653 inm->inm_ifp->if_xname); 2654 } 2655 } 2656 2657 /* 2658 * Enqueue an IGMPv3 group record to the given output queue. 2659 * 2660 * XXX This function could do with having the allocation code 2661 * split out, and the multiple-tree-walks coalesced into a single 2662 * routine as has been done in igmp_v3_enqueue_filter_change(). 2663 * 2664 * If is_state_change is zero, a current-state record is appended. 2665 * If is_state_change is non-zero, a state-change report is appended. 2666 * 2667 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2668 * If is_group_query is zero, and if there is a packet with free space 2669 * at the tail of the queue, it will be appended to providing there 2670 * is enough free space. 2671 * Otherwise a new mbuf packet chain is allocated. 2672 * 2673 * If is_source_query is non-zero, each source is checked to see if 2674 * it was recorded for a Group-Source query, and will be omitted if 2675 * it is not both in-mode and recorded. 2676 * 2677 * The function will attempt to allocate leading space in the packet 2678 * for the IP/IGMP header to be prepended without fragmenting the chain. 2679 * 2680 * If successful the size of all data appended to the queue is returned, 2681 * otherwise an error code less than zero is returned, or zero if 2682 * no record(s) were appended. 2683 */ 2684 static int 2685 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm, 2686 const int is_state_change, const int is_group_query, 2687 const int is_source_query) 2688 { 2689 struct igmp_grouprec ig; 2690 struct igmp_grouprec *pig; 2691 struct ifnet *ifp; 2692 struct ip_msource *ims, *nims; 2693 struct mbuf *m0, *m, *md; 2694 int is_filter_list_change; 2695 int minrec0len, m0srcs, msrcs, nbytes, off; 2696 int record_has_sources; 2697 int now; 2698 int type; 2699 in_addr_t naddr; 2700 uint8_t mode; 2701 2702 IN_MULTI_LIST_LOCK_ASSERT(); 2703 2704 ifp = inm->inm_ifp; 2705 is_filter_list_change = 0; 2706 m = NULL; 2707 m0 = NULL; 2708 m0srcs = 0; 2709 msrcs = 0; 2710 nbytes = 0; 2711 nims = NULL; 2712 record_has_sources = 1; 2713 pig = NULL; 2714 type = IGMP_DO_NOTHING; 2715 mode = inm->inm_st[1].iss_fmode; 2716 2717 /* 2718 * If we did not transition out of ASM mode during t0->t1, 2719 * and there are no source nodes to process, we can skip 2720 * the generation of source records. 2721 */ 2722 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 && 2723 inm->inm_nsrc == 0) 2724 record_has_sources = 0; 2725 2726 if (is_state_change) { 2727 /* 2728 * Queue a state change record. 2729 * If the mode did not change, and there are non-ASM 2730 * listeners or source filters present, 2731 * we potentially need to issue two records for the group. 2732 * If we are transitioning to MCAST_UNDEFINED, we need 2733 * not send any sources. 2734 * If there are ASM listeners, and there was no filter 2735 * mode transition of any kind, do nothing. 2736 */ 2737 if (mode != inm->inm_st[0].iss_fmode) { 2738 if (mode == MCAST_EXCLUDE) { 2739 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE", 2740 __func__); 2741 type = IGMP_CHANGE_TO_EXCLUDE_MODE; 2742 } else { 2743 CTR1(KTR_IGMPV3, "%s: change to INCLUDE", 2744 __func__); 2745 type = IGMP_CHANGE_TO_INCLUDE_MODE; 2746 if (mode == MCAST_UNDEFINED) 2747 record_has_sources = 0; 2748 } 2749 } else { 2750 if (record_has_sources) { 2751 is_filter_list_change = 1; 2752 } else { 2753 type = IGMP_DO_NOTHING; 2754 } 2755 } 2756 } else { 2757 /* 2758 * Queue a current state record. 2759 */ 2760 if (mode == MCAST_EXCLUDE) { 2761 type = IGMP_MODE_IS_EXCLUDE; 2762 } else if (mode == MCAST_INCLUDE) { 2763 type = IGMP_MODE_IS_INCLUDE; 2764 KASSERT(inm->inm_st[1].iss_asm == 0, 2765 ("%s: inm %p is INCLUDE but ASM count is %d", 2766 __func__, inm, inm->inm_st[1].iss_asm)); 2767 } 2768 } 2769 2770 /* 2771 * Generate the filter list changes using a separate function. 2772 */ 2773 if (is_filter_list_change) 2774 return (igmp_v3_enqueue_filter_change(mq, inm)); 2775 2776 if (type == IGMP_DO_NOTHING) { 2777 CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__, 2778 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2779 return (0); 2780 } 2781 2782 /* 2783 * If any sources are present, we must be able to fit at least 2784 * one in the trailing space of the tail packet's mbuf, 2785 * ideally more. 2786 */ 2787 minrec0len = sizeof(struct igmp_grouprec); 2788 if (record_has_sources) 2789 minrec0len += sizeof(in_addr_t); 2790 2791 CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__, 2792 igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr), 2793 inm->inm_ifp->if_xname); 2794 2795 /* 2796 * Check if we have a packet in the tail of the queue for this 2797 * group into which the first group record for this group will fit. 2798 * Otherwise allocate a new packet. 2799 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT. 2800 * Note: Group records for G/GSR query responses MUST be sent 2801 * in their own packet. 2802 */ 2803 m0 = mbufq_last(mq); 2804 if (!is_group_query && 2805 m0 != NULL && 2806 (m0->m_pkthdr.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) && 2807 (m0->m_pkthdr.len + minrec0len) < 2808 (ifp->if_mtu - IGMP_LEADINGSPACE)) { 2809 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2810 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2811 m = m0; 2812 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__); 2813 } else { 2814 if (mbufq_full(mq)) { 2815 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); 2816 return (-ENOMEM); 2817 } 2818 m = NULL; 2819 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 2820 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2821 if (!is_state_change && !is_group_query) { 2822 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2823 if (m) 2824 m->m_data += IGMP_LEADINGSPACE; 2825 } 2826 if (m == NULL) { 2827 m = m_gethdr(M_NOWAIT, MT_DATA); 2828 if (m) 2829 M_ALIGN(m, IGMP_LEADINGSPACE); 2830 } 2831 if (m == NULL) 2832 return (-ENOMEM); 2833 2834 igmp_save_context(m, ifp); 2835 2836 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__); 2837 } 2838 2839 /* 2840 * Append group record. 2841 * If we have sources, we don't know how many yet. 2842 */ 2843 ig.ig_type = type; 2844 ig.ig_datalen = 0; 2845 ig.ig_numsrc = 0; 2846 ig.ig_group = inm->inm_addr; 2847 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { 2848 if (m != m0) 2849 m_freem(m); 2850 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); 2851 return (-ENOMEM); 2852 } 2853 nbytes += sizeof(struct igmp_grouprec); 2854 2855 /* 2856 * Append as many sources as will fit in the first packet. 2857 * If we are appending to a new packet, the chain allocation 2858 * may potentially use clusters; use m_getptr() in this case. 2859 * If we are appending to an existing packet, we need to obtain 2860 * a pointer to the group record after m_append(), in case a new 2861 * mbuf was allocated. 2862 * Only append sources which are in-mode at t1. If we are 2863 * transitioning to MCAST_UNDEFINED state on the group, do not 2864 * include source entries. 2865 * Only report recorded sources in our filter set when responding 2866 * to a group-source query. 2867 */ 2868 if (record_has_sources) { 2869 if (m == m0) { 2870 md = m_last(m); 2871 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + 2872 md->m_len - nbytes); 2873 } else { 2874 md = m_getptr(m, 0, &off); 2875 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + 2876 off); 2877 } 2878 msrcs = 0; 2879 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) { 2880 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2881 ims->ims_haddr); 2882 now = ims_get_mode(inm, ims, 1); 2883 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now); 2884 if ((now != mode) || 2885 (now == mode && mode == MCAST_UNDEFINED)) { 2886 CTR1(KTR_IGMPV3, "%s: skip node", __func__); 2887 continue; 2888 } 2889 if (is_source_query && ims->ims_stp == 0) { 2890 CTR1(KTR_IGMPV3, "%s: skip unrecorded node", 2891 __func__); 2892 continue; 2893 } 2894 CTR1(KTR_IGMPV3, "%s: append node", __func__); 2895 naddr = htonl(ims->ims_haddr); 2896 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { 2897 if (m != m0) 2898 m_freem(m); 2899 CTR1(KTR_IGMPV3, "%s: m_append() failed.", 2900 __func__); 2901 return (-ENOMEM); 2902 } 2903 nbytes += sizeof(in_addr_t); 2904 ++msrcs; 2905 if (msrcs == m0srcs) 2906 break; 2907 } 2908 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__, 2909 msrcs); 2910 pig->ig_numsrc = htons(msrcs); 2911 nbytes += (msrcs * sizeof(in_addr_t)); 2912 } 2913 2914 if (is_source_query && msrcs == 0) { 2915 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__); 2916 if (m != m0) 2917 m_freem(m); 2918 return (0); 2919 } 2920 2921 /* 2922 * We are good to go with first packet. 2923 */ 2924 if (m != m0) { 2925 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__); 2926 m->m_pkthdr.vt_nrecs = 1; 2927 mbufq_enqueue(mq, m); 2928 } else 2929 m->m_pkthdr.vt_nrecs++; 2930 2931 /* 2932 * No further work needed if no source list in packet(s). 2933 */ 2934 if (!record_has_sources) 2935 return (nbytes); 2936 2937 /* 2938 * Whilst sources remain to be announced, we need to allocate 2939 * a new packet and fill out as many sources as will fit. 2940 * Always try for a cluster first. 2941 */ 2942 while (nims != NULL) { 2943 if (mbufq_full(mq)) { 2944 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); 2945 return (-ENOMEM); 2946 } 2947 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2948 if (m) 2949 m->m_data += IGMP_LEADINGSPACE; 2950 if (m == NULL) { 2951 m = m_gethdr(M_NOWAIT, MT_DATA); 2952 if (m) 2953 M_ALIGN(m, IGMP_LEADINGSPACE); 2954 } 2955 if (m == NULL) 2956 return (-ENOMEM); 2957 igmp_save_context(m, ifp); 2958 md = m_getptr(m, 0, &off); 2959 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); 2960 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__); 2961 2962 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { 2963 if (m != m0) 2964 m_freem(m); 2965 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); 2966 return (-ENOMEM); 2967 } 2968 m->m_pkthdr.vt_nrecs = 1; 2969 nbytes += sizeof(struct igmp_grouprec); 2970 2971 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 2972 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2973 2974 msrcs = 0; 2975 RB_FOREACH_FROM(ims, ip_msource_tree, nims) { 2976 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2977 ims->ims_haddr); 2978 now = ims_get_mode(inm, ims, 1); 2979 if ((now != mode) || 2980 (now == mode && mode == MCAST_UNDEFINED)) { 2981 CTR1(KTR_IGMPV3, "%s: skip node", __func__); 2982 continue; 2983 } 2984 if (is_source_query && ims->ims_stp == 0) { 2985 CTR1(KTR_IGMPV3, "%s: skip unrecorded node", 2986 __func__); 2987 continue; 2988 } 2989 CTR1(KTR_IGMPV3, "%s: append node", __func__); 2990 naddr = htonl(ims->ims_haddr); 2991 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { 2992 if (m != m0) 2993 m_freem(m); 2994 CTR1(KTR_IGMPV3, "%s: m_append() failed.", 2995 __func__); 2996 return (-ENOMEM); 2997 } 2998 ++msrcs; 2999 if (msrcs == m0srcs) 3000 break; 3001 } 3002 pig->ig_numsrc = htons(msrcs); 3003 nbytes += (msrcs * sizeof(in_addr_t)); 3004 3005 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__); 3006 mbufq_enqueue(mq, m); 3007 } 3008 3009 return (nbytes); 3010 } 3011 3012 /* 3013 * Type used to mark record pass completion. 3014 * We exploit the fact we can cast to this easily from the 3015 * current filter modes on each ip_msource node. 3016 */ 3017 typedef enum { 3018 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 3019 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 3020 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 3021 REC_FULL = REC_ALLOW | REC_BLOCK 3022 } rectype_t; 3023 3024 /* 3025 * Enqueue an IGMPv3 filter list change to the given output queue. 3026 * 3027 * Source list filter state is held in an RB-tree. When the filter list 3028 * for a group is changed without changing its mode, we need to compute 3029 * the deltas between T0 and T1 for each source in the filter set, 3030 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 3031 * 3032 * As we may potentially queue two record types, and the entire R-B tree 3033 * needs to be walked at once, we break this out into its own function 3034 * so we can generate a tightly packed queue of packets. 3035 * 3036 * XXX This could be written to only use one tree walk, although that makes 3037 * serializing into the mbuf chains a bit harder. For now we do two walks 3038 * which makes things easier on us, and it may or may not be harder on 3039 * the L2 cache. 3040 * 3041 * If successful the size of all data appended to the queue is returned, 3042 * otherwise an error code less than zero is returned, or zero if 3043 * no record(s) were appended. 3044 */ 3045 static int 3046 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm) 3047 { 3048 static const int MINRECLEN = 3049 sizeof(struct igmp_grouprec) + sizeof(in_addr_t); 3050 struct ifnet *ifp; 3051 struct igmp_grouprec ig; 3052 struct igmp_grouprec *pig; 3053 struct ip_msource *ims, *nims; 3054 struct mbuf *m, *m0, *md; 3055 in_addr_t naddr; 3056 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 3057 #ifdef KTR 3058 int nallow, nblock; 3059 #endif 3060 uint8_t mode, now, then; 3061 rectype_t crt, drt, nrt; 3062 3063 IN_MULTI_LIST_LOCK_ASSERT(); 3064 3065 if (inm->inm_nsrc == 0 || 3066 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0)) 3067 return (0); 3068 3069 ifp = inm->inm_ifp; /* interface */ 3070 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */ 3071 crt = REC_NONE; /* current group record type */ 3072 drt = REC_NONE; /* mask of completed group record types */ 3073 nrt = REC_NONE; /* record type for current node */ 3074 m0srcs = 0; /* # source which will fit in current mbuf chain */ 3075 nbytes = 0; /* # of bytes appended to group's state-change queue */ 3076 npbytes = 0; /* # of bytes appended this packet */ 3077 rsrcs = 0; /* # sources encoded in current record */ 3078 schanged = 0; /* # nodes encoded in overall filter change */ 3079 #ifdef KTR 3080 nallow = 0; /* # of source entries in ALLOW_NEW */ 3081 nblock = 0; /* # of source entries in BLOCK_OLD */ 3082 #endif 3083 nims = NULL; /* next tree node pointer */ 3084 3085 /* 3086 * For each possible filter record mode. 3087 * The first kind of source we encounter tells us which 3088 * is the first kind of record we start appending. 3089 * If a node transitioned to UNDEFINED at t1, its mode is treated 3090 * as the inverse of the group's filter mode. 3091 */ 3092 while (drt != REC_FULL) { 3093 do { 3094 m0 = mbufq_last(mq); 3095 if (m0 != NULL && 3096 (m0->m_pkthdr.vt_nrecs + 1 <= 3097 IGMP_V3_REPORT_MAXRECS) && 3098 (m0->m_pkthdr.len + MINRECLEN) < 3099 (ifp->if_mtu - IGMP_LEADINGSPACE)) { 3100 m = m0; 3101 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 3102 sizeof(struct igmp_grouprec)) / 3103 sizeof(in_addr_t); 3104 CTR1(KTR_IGMPV3, 3105 "%s: use previous packet", __func__); 3106 } else { 3107 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 3108 if (m) 3109 m->m_data += IGMP_LEADINGSPACE; 3110 if (m == NULL) { 3111 m = m_gethdr(M_NOWAIT, MT_DATA); 3112 if (m) 3113 M_ALIGN(m, IGMP_LEADINGSPACE); 3114 } 3115 if (m == NULL) { 3116 CTR1(KTR_IGMPV3, 3117 "%s: m_get*() failed", __func__); 3118 return (-ENOMEM); 3119 } 3120 m->m_pkthdr.vt_nrecs = 0; 3121 igmp_save_context(m, ifp); 3122 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 3123 sizeof(struct igmp_grouprec)) / 3124 sizeof(in_addr_t); 3125 npbytes = 0; 3126 CTR1(KTR_IGMPV3, 3127 "%s: allocated new packet", __func__); 3128 } 3129 /* 3130 * Append the IGMP group record header to the 3131 * current packet's data area. 3132 * Recalculate pointer to free space for next 3133 * group record, in case m_append() allocated 3134 * a new mbuf or cluster. 3135 */ 3136 memset(&ig, 0, sizeof(ig)); 3137 ig.ig_group = inm->inm_addr; 3138 if (!m_append(m, sizeof(ig), (void *)&ig)) { 3139 if (m != m0) 3140 m_freem(m); 3141 CTR1(KTR_IGMPV3, 3142 "%s: m_append() failed", __func__); 3143 return (-ENOMEM); 3144 } 3145 npbytes += sizeof(struct igmp_grouprec); 3146 if (m != m0) { 3147 /* new packet; offset in c hain */ 3148 md = m_getptr(m, npbytes - 3149 sizeof(struct igmp_grouprec), &off); 3150 pig = (struct igmp_grouprec *)(mtod(md, 3151 uint8_t *) + off); 3152 } else { 3153 /* current packet; offset from last append */ 3154 md = m_last(m); 3155 pig = (struct igmp_grouprec *)(mtod(md, 3156 uint8_t *) + md->m_len - 3157 sizeof(struct igmp_grouprec)); 3158 } 3159 /* 3160 * Begin walking the tree for this record type 3161 * pass, or continue from where we left off 3162 * previously if we had to allocate a new packet. 3163 * Only report deltas in-mode at t1. 3164 * We need not report included sources as allowed 3165 * if we are in inclusive mode on the group, 3166 * however the converse is not true. 3167 */ 3168 rsrcs = 0; 3169 if (nims == NULL) 3170 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs); 3171 RB_FOREACH_FROM(ims, ip_msource_tree, nims) { 3172 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", 3173 __func__, ims->ims_haddr); 3174 now = ims_get_mode(inm, ims, 1); 3175 then = ims_get_mode(inm, ims, 0); 3176 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d", 3177 __func__, then, now); 3178 if (now == then) { 3179 CTR1(KTR_IGMPV3, 3180 "%s: skip unchanged", __func__); 3181 continue; 3182 } 3183 if (mode == MCAST_EXCLUDE && 3184 now == MCAST_INCLUDE) { 3185 CTR1(KTR_IGMPV3, 3186 "%s: skip IN src on EX group", 3187 __func__); 3188 continue; 3189 } 3190 nrt = (rectype_t)now; 3191 if (nrt == REC_NONE) 3192 nrt = (rectype_t)(~mode & REC_FULL); 3193 if (schanged++ == 0) { 3194 crt = nrt; 3195 } else if (crt != nrt) 3196 continue; 3197 naddr = htonl(ims->ims_haddr); 3198 if (!m_append(m, sizeof(in_addr_t), 3199 (void *)&naddr)) { 3200 if (m != m0) 3201 m_freem(m); 3202 CTR1(KTR_IGMPV3, 3203 "%s: m_append() failed", __func__); 3204 return (-ENOMEM); 3205 } 3206 #ifdef KTR 3207 nallow += !!(crt == REC_ALLOW); 3208 nblock += !!(crt == REC_BLOCK); 3209 #endif 3210 if (++rsrcs == m0srcs) 3211 break; 3212 } 3213 /* 3214 * If we did not append any tree nodes on this 3215 * pass, back out of allocations. 3216 */ 3217 if (rsrcs == 0) { 3218 npbytes -= sizeof(struct igmp_grouprec); 3219 if (m != m0) { 3220 CTR1(KTR_IGMPV3, 3221 "%s: m_free(m)", __func__); 3222 m_freem(m); 3223 } else { 3224 CTR1(KTR_IGMPV3, 3225 "%s: m_adj(m, -ig)", __func__); 3226 m_adj(m, -((int)sizeof( 3227 struct igmp_grouprec))); 3228 } 3229 continue; 3230 } 3231 npbytes += (rsrcs * sizeof(in_addr_t)); 3232 if (crt == REC_ALLOW) 3233 pig->ig_type = IGMP_ALLOW_NEW_SOURCES; 3234 else if (crt == REC_BLOCK) 3235 pig->ig_type = IGMP_BLOCK_OLD_SOURCES; 3236 pig->ig_numsrc = htons(rsrcs); 3237 /* 3238 * Count the new group record, and enqueue this 3239 * packet if it wasn't already queued. 3240 */ 3241 m->m_pkthdr.vt_nrecs++; 3242 if (m != m0) 3243 mbufq_enqueue(mq, m); 3244 nbytes += npbytes; 3245 } while (nims != NULL); 3246 drt |= crt; 3247 crt = (~crt & REC_FULL); 3248 } 3249 3250 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 3251 nallow, nblock); 3252 3253 return (nbytes); 3254 } 3255 3256 static int 3257 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq) 3258 { 3259 struct mbufq *gq; 3260 struct mbuf *m; /* pending state-change */ 3261 struct mbuf *m0; /* copy of pending state-change */ 3262 struct mbuf *mt; /* last state-change in packet */ 3263 int docopy, domerge; 3264 u_int recslen; 3265 3266 docopy = 0; 3267 domerge = 0; 3268 recslen = 0; 3269 3270 IN_MULTI_LIST_LOCK_ASSERT(); 3271 IGMP_LOCK_ASSERT(); 3272 3273 /* 3274 * If there are further pending retransmissions, make a writable 3275 * copy of each queued state-change message before merging. 3276 */ 3277 if (inm->inm_scrv > 0) 3278 docopy = 1; 3279 3280 gq = &inm->inm_scq; 3281 #ifdef KTR 3282 if (mbufq_first(gq) == NULL) { 3283 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty", 3284 __func__, inm); 3285 } 3286 #endif 3287 3288 m = mbufq_first(gq); 3289 while (m != NULL) { 3290 /* 3291 * Only merge the report into the current packet if 3292 * there is sufficient space to do so; an IGMPv3 report 3293 * packet may only contain 65,535 group records. 3294 * Always use a simple mbuf chain concatentation to do this, 3295 * as large state changes for single groups may have 3296 * allocated clusters. 3297 */ 3298 domerge = 0; 3299 mt = mbufq_last(scq); 3300 if (mt != NULL) { 3301 recslen = m_length(m, NULL); 3302 3303 if ((mt->m_pkthdr.vt_nrecs + 3304 m->m_pkthdr.vt_nrecs <= 3305 IGMP_V3_REPORT_MAXRECS) && 3306 (mt->m_pkthdr.len + recslen <= 3307 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE))) 3308 domerge = 1; 3309 } 3310 3311 if (!domerge && mbufq_full(gq)) { 3312 CTR2(KTR_IGMPV3, 3313 "%s: outbound queue full, skipping whole packet %p", 3314 __func__, m); 3315 mt = m->m_nextpkt; 3316 if (!docopy) 3317 m_freem(m); 3318 m = mt; 3319 continue; 3320 } 3321 3322 if (!docopy) { 3323 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m); 3324 m0 = mbufq_dequeue(gq); 3325 m = m0->m_nextpkt; 3326 } else { 3327 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m); 3328 m0 = m_dup(m, M_NOWAIT); 3329 if (m0 == NULL) 3330 return (ENOMEM); 3331 m0->m_nextpkt = NULL; 3332 m = m->m_nextpkt; 3333 } 3334 3335 if (!domerge) { 3336 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)", 3337 __func__, m0, scq); 3338 mbufq_enqueue(scq, m0); 3339 } else { 3340 struct mbuf *mtl; /* last mbuf of packet mt */ 3341 3342 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)", 3343 __func__, m0, mt); 3344 3345 mtl = m_last(mt); 3346 m0->m_flags &= ~M_PKTHDR; 3347 mt->m_pkthdr.len += recslen; 3348 mt->m_pkthdr.vt_nrecs += 3349 m0->m_pkthdr.vt_nrecs; 3350 3351 mtl->m_next = m0; 3352 } 3353 } 3354 3355 return (0); 3356 } 3357 3358 /* 3359 * Respond to a pending IGMPv3 General Query. 3360 */ 3361 static void 3362 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi) 3363 { 3364 struct ifmultiaddr *ifma; 3365 struct ifnet *ifp; 3366 struct in_multi *inm; 3367 int retval __unused, loop; 3368 3369 IN_MULTI_LIST_LOCK_ASSERT(); 3370 IGMP_LOCK_ASSERT(); 3371 NET_EPOCH_ASSERT(); 3372 3373 KASSERT(igi->igi_version == IGMP_VERSION_3, 3374 ("%s: called when version %d", __func__, igi->igi_version)); 3375 3376 /* 3377 * Check that there are some packets queued. If so, send them first. 3378 * For large number of groups the reply to general query can take 3379 * many packets, we should finish sending them before starting of 3380 * queuing the new reply. 3381 */ 3382 if (!mbufq_empty(&igi->igi_gq)) 3383 goto send; 3384 3385 ifp = igi->igi_ifp; 3386 3387 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3388 inm = inm_ifmultiaddr_get_inm(ifma); 3389 if (inm == NULL) 3390 continue; 3391 KASSERT(ifp == inm->inm_ifp, 3392 ("%s: inconsistent ifp", __func__)); 3393 3394 switch (inm->inm_state) { 3395 case IGMP_NOT_MEMBER: 3396 case IGMP_SILENT_MEMBER: 3397 break; 3398 case IGMP_REPORTING_MEMBER: 3399 case IGMP_IDLE_MEMBER: 3400 case IGMP_LAZY_MEMBER: 3401 case IGMP_SLEEPING_MEMBER: 3402 case IGMP_AWAKENING_MEMBER: 3403 inm->inm_state = IGMP_REPORTING_MEMBER; 3404 retval = igmp_v3_enqueue_group_record(&igi->igi_gq, 3405 inm, 0, 0, 0); 3406 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 3407 __func__, retval); 3408 break; 3409 case IGMP_G_QUERY_PENDING_MEMBER: 3410 case IGMP_SG_QUERY_PENDING_MEMBER: 3411 case IGMP_LEAVING_MEMBER: 3412 break; 3413 } 3414 } 3415 3416 send: 3417 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; 3418 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop); 3419 3420 /* 3421 * Slew transmission of bursts over 500ms intervals. 3422 */ 3423 if (mbufq_first(&igi->igi_gq) != NULL) { 3424 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY( 3425 IGMP_RESPONSE_BURST_INTERVAL); 3426 V_interface_timers_running = 1; 3427 } 3428 } 3429 3430 /* 3431 * Transmit the next pending IGMP message in the output queue. 3432 * 3433 * We get called from netisr_processqueue(). A mutex private to igmpoq 3434 * will be acquired and released around this routine. 3435 * 3436 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3437 * MRT: Nothing needs to be done, as IGMP traffic is always local to 3438 * a link and uses a link-scope multicast address. 3439 */ 3440 static void 3441 igmp_intr(struct mbuf *m) 3442 { 3443 struct ip_moptions imo; 3444 struct ifnet *ifp; 3445 struct mbuf *ipopts, *m0; 3446 int error; 3447 uint32_t ifindex; 3448 3449 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m); 3450 3451 /* 3452 * Set VNET image pointer from enqueued mbuf chain 3453 * before doing anything else. Whilst we use interface 3454 * indexes to guard against interface detach, they are 3455 * unique to each VIMAGE and must be retrieved. 3456 */ 3457 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr)); 3458 ifindex = igmp_restore_context(m); 3459 3460 /* 3461 * Check if the ifnet still exists. This limits the scope of 3462 * any race in the absence of a global ifp lock for low cost 3463 * (an array lookup). 3464 */ 3465 ifp = ifnet_byindex(ifindex); 3466 if (ifp == NULL) { 3467 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.", 3468 __func__, m, ifindex); 3469 m_freem(m); 3470 IPSTAT_INC(ips_noroute); 3471 goto out; 3472 } 3473 3474 ipopts = V_igmp_sendra ? m_raopt : NULL; 3475 3476 imo.imo_multicast_ttl = 1; 3477 imo.imo_multicast_vif = -1; 3478 imo.imo_multicast_loop = (V_ip_mrouter != NULL); 3479 3480 /* 3481 * If the user requested that IGMP traffic be explicitly 3482 * redirected to the loopback interface (e.g. they are running a 3483 * MANET interface and the routing protocol needs to see the 3484 * updates), handle this now. 3485 */ 3486 if (m->m_flags & M_IGMP_LOOP) 3487 imo.imo_multicast_ifp = V_loif; 3488 else 3489 imo.imo_multicast_ifp = ifp; 3490 3491 if (m->m_flags & M_IGMPV2) { 3492 m0 = m; 3493 } else { 3494 m0 = igmp_v3_encap_report(ifp, m); 3495 if (m0 == NULL) { 3496 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m); 3497 m_freem(m); 3498 IPSTAT_INC(ips_odropped); 3499 goto out; 3500 } 3501 } 3502 3503 igmp_scrub_context(m0); 3504 m_clrprotoflags(m); 3505 m0->m_pkthdr.rcvif = V_loif; 3506 #ifdef MAC 3507 mac_netinet_igmp_send(ifp, m0); 3508 #endif 3509 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL); 3510 if (error) { 3511 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error); 3512 goto out; 3513 } 3514 3515 IGMPSTAT_INC(igps_snd_reports); 3516 3517 out: 3518 /* 3519 * We must restore the existing vnet pointer before 3520 * continuing as we are run from netisr context. 3521 */ 3522 CURVNET_RESTORE(); 3523 } 3524 3525 /* 3526 * Encapsulate an IGMPv3 report. 3527 * 3528 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf 3529 * chain has already had its IP/IGMPv3 header prepended. In this case 3530 * the function will not attempt to prepend; the lengths and checksums 3531 * will however be re-computed. 3532 * 3533 * Returns a pointer to the new mbuf chain head, or NULL if the 3534 * allocation failed. 3535 */ 3536 static struct mbuf * 3537 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m) 3538 { 3539 struct igmp_report *igmp; 3540 struct ip *ip; 3541 int hdrlen, igmpreclen; 3542 3543 KASSERT((m->m_flags & M_PKTHDR), 3544 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3545 3546 igmpreclen = m_length(m, NULL); 3547 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report); 3548 3549 if (m->m_flags & M_IGMPV3_HDR) { 3550 igmpreclen -= hdrlen; 3551 } else { 3552 M_PREPEND(m, hdrlen, M_NOWAIT); 3553 if (m == NULL) 3554 return (NULL); 3555 m->m_flags |= M_IGMPV3_HDR; 3556 } 3557 3558 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen); 3559 3560 m->m_data += sizeof(struct ip); 3561 m->m_len -= sizeof(struct ip); 3562 3563 igmp = mtod(m, struct igmp_report *); 3564 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT; 3565 igmp->ir_rsv1 = 0; 3566 igmp->ir_rsv2 = 0; 3567 igmp->ir_numgrps = htons(m->m_pkthdr.vt_nrecs); 3568 igmp->ir_cksum = 0; 3569 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen); 3570 m->m_pkthdr.vt_nrecs = 0; 3571 3572 m->m_data -= sizeof(struct ip); 3573 m->m_len += sizeof(struct ip); 3574 3575 ip = mtod(m, struct ip *); 3576 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL; 3577 ip->ip_len = htons(hdrlen + igmpreclen); 3578 ip->ip_off = htons(IP_DF); 3579 ip->ip_p = IPPROTO_IGMP; 3580 ip->ip_sum = 0; 3581 3582 ip->ip_src.s_addr = INADDR_ANY; 3583 3584 if (m->m_flags & M_IGMP_LOOP) { 3585 struct in_ifaddr *ia; 3586 3587 IFP_TO_IA(ifp, ia); 3588 if (ia != NULL) 3589 ip->ip_src = ia->ia_addr.sin_addr; 3590 } 3591 3592 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP); 3593 3594 return (m); 3595 } 3596 3597 #ifdef KTR 3598 static char * 3599 igmp_rec_type_to_str(const int type) 3600 { 3601 3602 switch (type) { 3603 case IGMP_CHANGE_TO_EXCLUDE_MODE: 3604 return "TO_EX"; 3605 break; 3606 case IGMP_CHANGE_TO_INCLUDE_MODE: 3607 return "TO_IN"; 3608 break; 3609 case IGMP_MODE_IS_EXCLUDE: 3610 return "MODE_EX"; 3611 break; 3612 case IGMP_MODE_IS_INCLUDE: 3613 return "MODE_IN"; 3614 break; 3615 case IGMP_ALLOW_NEW_SOURCES: 3616 return "ALLOW_NEW"; 3617 break; 3618 case IGMP_BLOCK_OLD_SOURCES: 3619 return "BLOCK_OLD"; 3620 break; 3621 default: 3622 break; 3623 } 3624 return "unknown"; 3625 } 3626 #endif 3627 3628 #ifdef VIMAGE 3629 static void 3630 vnet_igmp_init(const void *unused __unused) 3631 { 3632 3633 netisr_register_vnet(&igmp_nh); 3634 } 3635 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, 3636 vnet_igmp_init, NULL); 3637 3638 static void 3639 vnet_igmp_uninit(const void *unused __unused) 3640 { 3641 3642 /* This can happen when we shutdown the entire network stack. */ 3643 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3644 3645 netisr_unregister_vnet(&igmp_nh); 3646 } 3647 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, 3648 vnet_igmp_uninit, NULL); 3649 #endif 3650 3651 #ifdef DDB 3652 DB_SHOW_COMMAND(igi_list, db_show_igi_list) 3653 { 3654 struct igmp_ifsoftc *igi, *tigi; 3655 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head; 3656 3657 if (!have_addr) { 3658 db_printf("usage: show igi_list <addr>\n"); 3659 return; 3660 } 3661 igi_head = (struct _igi_list *)addr; 3662 3663 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) { 3664 db_printf("igmp_ifsoftc %p:\n", igi); 3665 db_printf(" ifp %p\n", igi->igi_ifp); 3666 db_printf(" version %u\n", igi->igi_version); 3667 db_printf(" v1_timer %u\n", igi->igi_v1_timer); 3668 db_printf(" v2_timer %u\n", igi->igi_v2_timer); 3669 db_printf(" v3_timer %u\n", igi->igi_v3_timer); 3670 db_printf(" flags %#x\n", igi->igi_flags); 3671 db_printf(" rv %u\n", igi->igi_rv); 3672 db_printf(" qi %u\n", igi->igi_qi); 3673 db_printf(" qri %u\n", igi->igi_qri); 3674 db_printf(" uri %u\n", igi->igi_uri); 3675 /* struct mbufq igi_gq; */ 3676 db_printf("\n"); 3677 } 3678 } 3679 #endif 3680 3681 static int 3682 igmp_modevent(module_t mod, int type, void *unused __unused) 3683 { 3684 3685 switch (type) { 3686 case MOD_LOAD: 3687 CTR1(KTR_IGMPV3, "%s: initializing", __func__); 3688 IGMP_LOCK_INIT(); 3689 m_raopt = igmp_ra_alloc(); 3690 netisr_register(&igmp_nh); 3691 callout_init(&igmpslow_callout, 1); 3692 callout_reset(&igmpslow_callout, hz / IGMP_SLOWHZ, 3693 igmp_slowtimo, NULL); 3694 callout_init(&igmpfast_callout, 1); 3695 callout_reset(&igmpfast_callout, hz / IGMP_FASTHZ, 3696 igmp_fasttimo, NULL); 3697 break; 3698 case MOD_UNLOAD: 3699 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3700 netisr_unregister(&igmp_nh); 3701 m_free(m_raopt); 3702 m_raopt = NULL; 3703 IGMP_LOCK_DESTROY(); 3704 break; 3705 default: 3706 return (EOPNOTSUPP); 3707 } 3708 return (0); 3709 } 3710 3711 static moduledata_t igmp_mod = { 3712 "igmp", 3713 igmp_modevent, 3714 0 3715 }; 3716 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE); 3717