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 37 /* 38 * Internet Group Management Protocol (IGMP) routines. 39 * [RFC1112, RFC2236, RFC3376] 40 * 41 * Written by Steve Deering, Stanford, May 1988. 42 * Modified by Rosen Sharma, Stanford, Aug 1994. 43 * Modified by Bill Fenner, Xerox PARC, Feb 1995. 44 * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995. 45 * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson. 46 * 47 * MULTICAST Revision: 3.5.1.4 48 */ 49 50 #include "opt_ddb.h" 51 52 #include <sys/param.h> 53 #include <sys/systm.h> 54 #include <sys/module.h> 55 #include <sys/malloc.h> 56 #include <sys/mbuf.h> 57 #include <sys/socket.h> 58 #include <sys/kernel.h> 59 #include <sys/lock.h> 60 #include <sys/sysctl.h> 61 #include <sys/ktr.h> 62 #include <sys/condvar.h> 63 64 #ifdef DDB 65 #include <ddb/ddb.h> 66 #endif 67 68 #include <net/if.h> 69 #include <net/if_var.h> 70 #include <net/if_private.h> 71 #include <net/netisr.h> 72 #include <net/vnet.h> 73 74 #include <netinet/in.h> 75 #include <netinet/in_var.h> 76 #include <netinet/in_systm.h> 77 #include <netinet/ip.h> 78 #include <netinet/ip_var.h> 79 #include <netinet/ip_options.h> 80 #include <netinet/igmp.h> 81 #include <netinet/igmp_var.h> 82 83 #include <machine/in_cksum.h> 84 85 #include <security/mac/mac_framework.h> 86 87 #ifndef KTR_IGMPV3 88 #define KTR_IGMPV3 KTR_INET 89 #endif 90 91 #define IGMP_SLOWHZ 2 /* 2 slow timeouts per second */ 92 #define IGMP_FASTHZ 5 /* 5 fast timeouts per second */ 93 #define IGMP_RESPONSE_BURST_INTERVAL (IGMP_FASTHZ / 2) 94 95 static struct igmp_ifsoftc * 96 igi_alloc_locked(struct ifnet *); 97 static void igi_delete_locked(const struct ifnet *); 98 static void igmp_dispatch_queue(struct mbufq *, int, const int); 99 static void igmp_fasttimo_vnet(void); 100 static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *); 101 static int igmp_handle_state_change(struct in_multi *, 102 struct igmp_ifsoftc *); 103 static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *); 104 static int igmp_input_v1_query(struct ifnet *, const struct ip *, 105 const struct igmp *); 106 static int igmp_input_v2_query(struct ifnet *, const struct ip *, 107 const struct igmp *); 108 static int igmp_input_v3_query(struct ifnet *, const struct ip *, 109 /*const*/ struct igmpv3 *); 110 static int igmp_input_v3_group_query(struct in_multi *, 111 struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *); 112 static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *, 113 /*const*/ struct igmp *); 114 static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *, 115 /*const*/ struct igmp *); 116 static void igmp_intr(struct mbuf *); 117 static int igmp_isgroupreported(const struct in_addr); 118 static struct mbuf * 119 igmp_ra_alloc(void); 120 #ifdef KTR 121 static char * igmp_rec_type_to_str(const int); 122 #endif 123 static void igmp_set_version(struct igmp_ifsoftc *, const int); 124 static void igmp_slowtimo_vnet(void); 125 static int igmp_v1v2_queue_report(struct in_multi *, const int); 126 static void igmp_v1v2_process_group_timer(struct in_multi *, const int); 127 static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *); 128 static void igmp_v2_update_group(struct in_multi *, const int); 129 static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *); 130 static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *); 131 static struct mbuf * 132 igmp_v3_encap_report(struct ifnet *, struct mbuf *); 133 static int igmp_v3_enqueue_group_record(struct mbufq *, 134 struct in_multi *, const int, const int, const int); 135 static int igmp_v3_enqueue_filter_change(struct mbufq *, 136 struct in_multi *); 137 static void igmp_v3_process_group_timers(struct in_multi_head *, 138 struct mbufq *, struct mbufq *, struct in_multi *, 139 const int); 140 static int igmp_v3_merge_state_changes(struct in_multi *, 141 struct mbufq *); 142 static void igmp_v3_suppress_group_record(struct in_multi *); 143 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS); 144 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS); 145 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS); 146 static int sysctl_igmp_stat(SYSCTL_HANDLER_ARGS); 147 148 static const struct netisr_handler igmp_nh = { 149 .nh_name = "igmp", 150 .nh_handler = igmp_intr, 151 .nh_proto = NETISR_IGMP, 152 .nh_policy = NETISR_POLICY_SOURCE, 153 }; 154 155 /* 156 * System-wide globals. 157 * 158 * Unlocked access to these is OK, except for the global IGMP output 159 * queue. The IGMP subsystem lock ends up being system-wide for the moment, 160 * because all VIMAGEs have to share a global output queue, as netisrs 161 * themselves are not virtualized. 162 * 163 * Locking: 164 * * The permitted lock order is: IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 165 * Any may be taken independently; if any are held at the same 166 * time, the above lock order must be followed. 167 * * All output is delegated to the netisr. 168 * * IN_MULTI_LIST_LOCK covers in_multi. 169 * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file, 170 * including the output queue. 171 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 172 * per-link state iterators. 173 * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface, 174 * therefore it is not refcounted. 175 * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi. 176 * 177 * Reference counting 178 * * IGMP acquires its own reference every time an in_multi is passed to 179 * it and the group is being joined for the first time. 180 * * IGMP releases its reference(s) on in_multi in a deferred way, 181 * because the operations which process the release run as part of 182 * a loop whose control variables are directly affected by the release 183 * (that, and not recursing on the IF_ADDR_LOCK). 184 * 185 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds 186 * to a vnet in ifp->if_vnet. 187 * 188 * SMPng: XXX We may potentially race operations on ifma_protospec. 189 * The problem is that we currently lack a clean way of taking the 190 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing, 191 * as anything which modifies ifma needs to be covered by that lock. 192 * So check for ifma_protospec being NULL before proceeding. 193 */ 194 struct mtx igmp_mtx; 195 196 struct mbuf *m_raopt; /* Router Alert option */ 197 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state"); 198 199 /* 200 * VIMAGE-wide globals. 201 * 202 * The IGMPv3 timers themselves need to run per-image, however, for 203 * historical reasons, timers run globally. This needs to be improved. 204 * An ifnet can only be in one vimage at a time, and the loopback 205 * ifnet, loif, is itself virtualized. 206 * It would otherwise be possible to seriously hose IGMP state, 207 * and create inconsistencies in upstream multicast routing, if you have 208 * multiple VIMAGEs running on the same link joining different multicast 209 * groups, UNLESS the "primary IP address" is different. This is because 210 * IGMP for IPv4 does not force link-local addresses to be used for each 211 * node, unlike MLD for IPv6. 212 * Obviously the IGMPv3 per-interface state has per-vimage granularity 213 * also as a result. 214 * 215 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection 216 * policy to control the address used by IGMP on the link. 217 */ 218 VNET_DEFINE_STATIC(int, interface_timers_running); /* IGMPv3 general 219 * query response */ 220 VNET_DEFINE_STATIC(int, state_change_timers_running); /* IGMPv3 state-change 221 * retransmit */ 222 VNET_DEFINE_STATIC(int, current_state_timers_running); /* IGMPv1/v2 host 223 * report; IGMPv3 g/sg 224 * query response */ 225 226 #define V_interface_timers_running VNET(interface_timers_running) 227 #define V_state_change_timers_running VNET(state_change_timers_running) 228 #define V_current_state_timers_running VNET(current_state_timers_running) 229 230 VNET_PCPUSTAT_DEFINE(struct igmpstat, igmpstat); 231 VNET_PCPUSTAT_SYSINIT(igmpstat); 232 VNET_PCPUSTAT_SYSUNINIT(igmpstat); 233 234 VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) = 235 LIST_HEAD_INITIALIZER(igi_head); 236 VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0}; 237 238 #define V_igi_head VNET(igi_head) 239 #define V_igmp_gsrdelay VNET(igmp_gsrdelay) 240 241 VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1; 242 VNET_DEFINE_STATIC(int, igmp_sendra) = 1; 243 VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1; 244 VNET_DEFINE_STATIC(int, igmp_v1enable) = 1; 245 VNET_DEFINE_STATIC(int, igmp_v2enable) = 1; 246 VNET_DEFINE_STATIC(int, igmp_legacysupp); 247 VNET_DEFINE_STATIC(int, igmp_default_version) = IGMP_VERSION_3; 248 249 #define V_igmp_recvifkludge VNET(igmp_recvifkludge) 250 #define V_igmp_sendra VNET(igmp_sendra) 251 #define V_igmp_sendlocal VNET(igmp_sendlocal) 252 #define V_igmp_v1enable VNET(igmp_v1enable) 253 #define V_igmp_v2enable VNET(igmp_v2enable) 254 #define V_igmp_legacysupp VNET(igmp_legacysupp) 255 #define V_igmp_default_version VNET(igmp_default_version) 256 257 /* 258 * Virtualized sysctls. 259 */ 260 SYSCTL_PROC(_net_inet_igmp, IGMPCTL_STATS, stats, 261 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_MPSAFE, 262 &VNET_NAME(igmpstat), 0, sysctl_igmp_stat, "S,igmpstat", 263 "IGMP statistics (struct igmpstat, netinet/igmp_var.h)"); 264 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW, 265 &VNET_NAME(igmp_recvifkludge), 0, 266 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address"); 267 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW, 268 &VNET_NAME(igmp_sendra), 0, 269 "Send IP Router Alert option in IGMPv2/v3 messages"); 270 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW, 271 &VNET_NAME(igmp_sendlocal), 0, 272 "Send IGMP membership reports for 224.0.0.0/24 groups"); 273 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW, 274 &VNET_NAME(igmp_v1enable), 0, 275 "Enable backwards compatibility with IGMPv1"); 276 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW, 277 &VNET_NAME(igmp_v2enable), 0, 278 "Enable backwards compatibility with IGMPv2"); 279 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW, 280 &VNET_NAME(igmp_legacysupp), 0, 281 "Allow v1/v2 reports to suppress v3 group responses"); 282 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version, 283 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 284 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I", 285 "Default version of IGMP to run on each interface"); 286 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay, 287 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 288 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I", 289 "Rate limit for IGMPv3 Group-and-Source queries in seconds"); 290 291 /* 292 * Non-virtualized sysctls. 293 */ 294 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, 295 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo, 296 "Per-interface IGMPv3 state"); 297 298 static __inline void 299 igmp_save_context(struct mbuf *m, struct ifnet *ifp) 300 { 301 302 #ifdef VIMAGE 303 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; 304 #endif /* VIMAGE */ 305 m->m_pkthdr.rcvif = ifp; 306 m->m_pkthdr.flowid = ifp->if_index; 307 } 308 309 static __inline void 310 igmp_scrub_context(struct mbuf *m) 311 { 312 313 m->m_pkthdr.PH_loc.ptr = NULL; 314 m->m_pkthdr.flowid = 0; 315 } 316 317 /* 318 * Restore context from a queued IGMP output chain. 319 * Return saved ifindex. 320 * 321 * VIMAGE: The assertion is there to make sure that we 322 * actually called CURVNET_SET() with what's in the mbuf chain. 323 */ 324 static __inline uint32_t 325 igmp_restore_context(struct mbuf *m) 326 { 327 328 #ifdef notyet 329 #if defined(VIMAGE) && defined(INVARIANTS) 330 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr), 331 ("%s: called when curvnet was not restored", __func__)); 332 #endif 333 #endif 334 return (m->m_pkthdr.flowid); 335 } 336 337 /* 338 * IGMP statistics. 339 */ 340 static int 341 sysctl_igmp_stat(SYSCTL_HANDLER_ARGS) 342 { 343 struct igmpstat igps0; 344 int error; 345 char *p; 346 347 error = sysctl_wire_old_buffer(req, sizeof(struct igmpstat)); 348 if (error) 349 return (error); 350 351 if (req->oldptr != NULL) { 352 if (req->oldlen < sizeof(struct igmpstat)) 353 error = ENOMEM; 354 else { 355 /* 356 * Copy the counters, and explicitly set the struct's 357 * version and length fields. 358 */ 359 COUNTER_ARRAY_COPY(VNET(igmpstat), &igps0, 360 sizeof(struct igmpstat) / sizeof(uint64_t)); 361 igps0.igps_version = IGPS_VERSION_3; 362 igps0.igps_len = IGPS_VERSION3_LEN; 363 error = SYSCTL_OUT(req, &igps0, 364 sizeof(struct igmpstat)); 365 } 366 } else 367 req->validlen = sizeof(struct igmpstat); 368 if (error) 369 goto out; 370 if (req->newptr != NULL) { 371 if (req->newlen < sizeof(struct igmpstat)) 372 error = ENOMEM; 373 else 374 error = SYSCTL_IN(req, &igps0, 375 sizeof(igps0)); 376 if (error) 377 goto out; 378 /* 379 * igps0 must be "all zero". 380 */ 381 p = (char *)&igps0; 382 while (p < (char *)&igps0 + sizeof(igps0) && *p == '\0') 383 p++; 384 if (p != (char *)&igps0 + sizeof(igps0)) { 385 error = EINVAL; 386 goto out; 387 } 388 COUNTER_ARRAY_ZERO(VNET(igmpstat), 389 sizeof(struct igmpstat) / sizeof(uint64_t)); 390 } 391 out: 392 return (error); 393 } 394 395 /* 396 * Retrieve or set default IGMP version. 397 * 398 * VIMAGE: Assume curvnet set by caller. 399 * SMPng: NOTE: Serialized by IGMP lock. 400 */ 401 static int 402 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS) 403 { 404 struct epoch_tracker et; 405 int error; 406 int new; 407 struct igmp_ifsoftc *igi; 408 409 error = sysctl_wire_old_buffer(req, sizeof(int)); 410 if (error) 411 return (error); 412 413 new = V_igmp_default_version; 414 415 error = sysctl_handle_int(oidp, &new, 0, req); 416 if (error || !req->newptr) 417 return (error); 418 419 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) 420 return (EINVAL); 421 422 IN_MULTI_LIST_LOCK(); 423 IGMP_LOCK(); 424 NET_EPOCH_ENTER(et); 425 426 if (V_igmp_default_version != new) { 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 LIST_FOREACH(igi, &V_igi_head, igi_link) { 433 if (igi->igi_version > V_igmp_default_version){ 434 igmp_set_version(igi, V_igmp_default_version); 435 } 436 } 437 } 438 439 NET_EPOCH_EXIT(et); 440 IN_MULTI_LIST_UNLOCK(); 441 IGMP_UNLOCK(); 442 return (error); 443 } 444 445 /* 446 * Retrieve or set threshold between group-source queries in seconds. 447 * 448 * VIMAGE: Assume curvnet set by caller. 449 * SMPng: NOTE: Serialized by IGMP lock. 450 */ 451 static int 452 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS) 453 { 454 int error; 455 int i; 456 457 error = sysctl_wire_old_buffer(req, sizeof(int)); 458 if (error) 459 return (error); 460 461 IGMP_LOCK(); 462 463 i = V_igmp_gsrdelay.tv_sec; 464 465 error = sysctl_handle_int(oidp, &i, 0, req); 466 if (error || !req->newptr) 467 goto out_locked; 468 469 if (i < -1 || i >= 60) { 470 error = EINVAL; 471 goto out_locked; 472 } 473 474 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d", 475 V_igmp_gsrdelay.tv_sec, i); 476 V_igmp_gsrdelay.tv_sec = i; 477 478 out_locked: 479 IGMP_UNLOCK(); 480 return (error); 481 } 482 483 /* 484 * Expose struct igmp_ifsoftc to userland, keyed by ifindex. 485 * For use by ifmcstat(8). 486 * 487 * SMPng: NOTE: Does an unlocked ifindex space read. 488 * VIMAGE: Assume curvnet set by caller. The node handler itself 489 * is not directly virtualized. 490 */ 491 static int 492 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS) 493 { 494 struct epoch_tracker et; 495 int *name; 496 int error; 497 u_int namelen; 498 struct ifnet *ifp; 499 struct igmp_ifsoftc *igi; 500 501 name = (int *)arg1; 502 namelen = arg2; 503 504 if (req->newptr != NULL) 505 return (EPERM); 506 507 if (namelen != 1) 508 return (EINVAL); 509 510 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo)); 511 if (error) 512 return (error); 513 514 IN_MULTI_LIST_LOCK(); 515 IGMP_LOCK(); 516 517 error = ENOENT; 518 519 NET_EPOCH_ENTER(et); 520 ifp = ifnet_byindex(name[0]); 521 NET_EPOCH_EXIT(et); 522 if (ifp == NULL) 523 goto out_locked; 524 525 LIST_FOREACH(igi, &V_igi_head, igi_link) { 526 if (ifp == igi->igi_ifp) { 527 struct igmp_ifinfo info; 528 529 info.igi_version = igi->igi_version; 530 info.igi_v1_timer = igi->igi_v1_timer; 531 info.igi_v2_timer = igi->igi_v2_timer; 532 info.igi_v3_timer = igi->igi_v3_timer; 533 info.igi_flags = igi->igi_flags; 534 info.igi_rv = igi->igi_rv; 535 info.igi_qi = igi->igi_qi; 536 info.igi_qri = igi->igi_qri; 537 info.igi_uri = igi->igi_uri; 538 error = SYSCTL_OUT(req, &info, sizeof(info)); 539 break; 540 } 541 } 542 543 out_locked: 544 IGMP_UNLOCK(); 545 IN_MULTI_LIST_UNLOCK(); 546 return (error); 547 } 548 549 /* 550 * Dispatch an entire queue of pending packet chains 551 * using the netisr. 552 * VIMAGE: Assumes the vnet pointer has been set. 553 */ 554 static void 555 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop) 556 { 557 struct epoch_tracker et; 558 struct mbuf *m; 559 560 NET_EPOCH_ENTER(et); 561 while ((m = mbufq_dequeue(mq)) != NULL) { 562 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m); 563 if (loop) 564 m->m_flags |= M_IGMP_LOOP; 565 netisr_dispatch(NETISR_IGMP, m); 566 if (--limit == 0) 567 break; 568 } 569 NET_EPOCH_EXIT(et); 570 } 571 572 /* 573 * Filter outgoing IGMP report state by group. 574 * 575 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1). 576 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are 577 * disabled for all groups in the 224.0.0.0/24 link-local scope. However, 578 * this may break certain IGMP snooping switches which rely on the old 579 * report behaviour. 580 * 581 * Return zero if the given group is one for which IGMP reports 582 * should be suppressed, or non-zero if reports should be issued. 583 */ 584 static __inline int 585 igmp_isgroupreported(const struct in_addr addr) 586 { 587 588 if (in_allhosts(addr) || 589 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr))))) 590 return (0); 591 592 return (1); 593 } 594 595 /* 596 * Construct a Router Alert option to use in outgoing packets. 597 */ 598 static struct mbuf * 599 igmp_ra_alloc(void) 600 { 601 struct mbuf *m; 602 struct ipoption *p; 603 604 m = m_get(M_WAITOK, MT_DATA); 605 p = mtod(m, struct ipoption *); 606 p->ipopt_dst.s_addr = INADDR_ANY; 607 p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */ 608 p->ipopt_list[1] = 0x04; /* 4 bytes long */ 609 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */ 610 p->ipopt_list[3] = 0x00; /* pad byte */ 611 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1]; 612 613 return (m); 614 } 615 616 /* 617 * Attach IGMP when PF_INET is attached to an interface. 618 */ 619 struct igmp_ifsoftc * 620 igmp_domifattach(struct ifnet *ifp) 621 { 622 struct igmp_ifsoftc *igi; 623 624 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 625 __func__, ifp, ifp->if_xname); 626 627 IGMP_LOCK(); 628 629 igi = igi_alloc_locked(ifp); 630 if (!(ifp->if_flags & IFF_MULTICAST)) 631 igi->igi_flags |= IGIF_SILENT; 632 633 IGMP_UNLOCK(); 634 635 return (igi); 636 } 637 638 /* 639 * VIMAGE: assume curvnet set by caller. 640 */ 641 static struct igmp_ifsoftc * 642 igi_alloc_locked(/*const*/ struct ifnet *ifp) 643 { 644 struct igmp_ifsoftc *igi; 645 646 IGMP_LOCK_ASSERT(); 647 648 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO); 649 if (igi == NULL) 650 goto out; 651 652 igi->igi_ifp = ifp; 653 igi->igi_version = V_igmp_default_version; 654 igi->igi_flags = 0; 655 igi->igi_rv = IGMP_RV_INIT; 656 igi->igi_qi = IGMP_QI_INIT; 657 igi->igi_qri = IGMP_QRI_INIT; 658 igi->igi_uri = IGMP_URI_INIT; 659 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS); 660 661 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link); 662 663 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)", 664 ifp, ifp->if_xname); 665 666 out: 667 return (igi); 668 } 669 670 /* 671 * Hook for ifdetach. 672 * 673 * NOTE: Some finalization tasks need to run before the protocol domain 674 * is detached, but also before the link layer does its cleanup. 675 * 676 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK(). 677 * XXX This is also bitten by unlocked ifma_protospec access. 678 */ 679 void 680 igmp_ifdetach(struct ifnet *ifp) 681 { 682 struct epoch_tracker et; 683 struct igmp_ifsoftc *igi; 684 struct ifmultiaddr *ifma; 685 struct in_multi *inm; 686 struct in_multi_head inm_free_tmp; 687 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, 688 ifp->if_xname); 689 690 SLIST_INIT(&inm_free_tmp); 691 IGMP_LOCK(); 692 693 igi = ((struct in_ifinfo *)ifp->if_inet)->ii_igmp; 694 if (igi->igi_version == IGMP_VERSION_3) { 695 IF_ADDR_WLOCK(ifp); 696 NET_EPOCH_ENTER(et); 697 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 698 inm = inm_ifmultiaddr_get_inm(ifma); 699 if (inm == NULL) 700 continue; 701 if (inm->inm_state == IGMP_LEAVING_MEMBER) 702 inm_rele_locked(&inm_free_tmp, inm); 703 inm_clear_recorded(inm); 704 } 705 NET_EPOCH_EXIT(et); 706 IF_ADDR_WUNLOCK(ifp); 707 inm_release_list_deferred(&inm_free_tmp); 708 } 709 IGMP_UNLOCK(); 710 711 } 712 713 /* 714 * Hook for domifdetach. 715 */ 716 void 717 igmp_domifdetach(struct ifnet *ifp) 718 { 719 720 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 721 __func__, ifp, ifp->if_xname); 722 723 IGMP_LOCK(); 724 igi_delete_locked(ifp); 725 IGMP_UNLOCK(); 726 } 727 728 static void 729 igi_delete_locked(const struct ifnet *ifp) 730 { 731 struct igmp_ifsoftc *igi, *tigi; 732 733 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)", 734 __func__, ifp, ifp->if_xname); 735 736 IGMP_LOCK_ASSERT(); 737 738 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) { 739 if (igi->igi_ifp == ifp) { 740 /* 741 * Free deferred General Query responses. 742 */ 743 mbufq_drain(&igi->igi_gq); 744 745 LIST_REMOVE(igi, igi_link); 746 free(igi, M_IGMP); 747 return; 748 } 749 } 750 } 751 752 /* 753 * Process a received IGMPv1 query. 754 * Return non-zero if the message should be dropped. 755 * 756 * VIMAGE: The curvnet pointer is derived from the input ifp. 757 */ 758 static int 759 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip, 760 const struct igmp *igmp) 761 { 762 struct ifmultiaddr *ifma; 763 struct igmp_ifsoftc *igi; 764 struct in_multi *inm; 765 766 NET_EPOCH_ASSERT(); 767 768 /* 769 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to 770 * 224.0.0.1. They are always treated as General Queries. 771 * igmp_group is always ignored. Do not drop it as a userland 772 * daemon may wish to see it. 773 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 774 */ 775 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) { 776 IGMPSTAT_INC(igps_rcv_badqueries); 777 return (0); 778 } 779 IGMPSTAT_INC(igps_rcv_gen_queries); 780 781 IN_MULTI_LIST_LOCK(); 782 IGMP_LOCK(); 783 784 igi = ((struct in_ifinfo *)ifp->if_inet)->ii_igmp; 785 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 786 787 if (igi->igi_flags & IGIF_LOOPBACK) { 788 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)", 789 ifp, ifp->if_xname); 790 goto out_locked; 791 } 792 793 /* 794 * Switch to IGMPv1 host compatibility mode. 795 */ 796 igmp_set_version(igi, IGMP_VERSION_1); 797 798 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname); 799 800 /* 801 * Start the timers in all of our group records 802 * for the interface on which the query arrived, 803 * except those which are already running. 804 */ 805 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 806 inm = inm_ifmultiaddr_get_inm(ifma); 807 if (inm == NULL) 808 continue; 809 if (inm->inm_timer != 0) 810 continue; 811 switch (inm->inm_state) { 812 case IGMP_NOT_MEMBER: 813 case IGMP_SILENT_MEMBER: 814 break; 815 case IGMP_G_QUERY_PENDING_MEMBER: 816 case IGMP_SG_QUERY_PENDING_MEMBER: 817 case IGMP_REPORTING_MEMBER: 818 case IGMP_IDLE_MEMBER: 819 case IGMP_LAZY_MEMBER: 820 case IGMP_SLEEPING_MEMBER: 821 case IGMP_AWAKENING_MEMBER: 822 inm->inm_state = IGMP_REPORTING_MEMBER; 823 inm->inm_timer = IGMP_RANDOM_DELAY( 824 IGMP_V1V2_MAX_RI * IGMP_FASTHZ); 825 V_current_state_timers_running = 1; 826 break; 827 case IGMP_LEAVING_MEMBER: 828 break; 829 } 830 } 831 832 out_locked: 833 IGMP_UNLOCK(); 834 IN_MULTI_LIST_UNLOCK(); 835 836 return (0); 837 } 838 839 /* 840 * Process a received IGMPv2 general or group-specific query. 841 */ 842 static int 843 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip, 844 const struct igmp *igmp) 845 { 846 struct ifmultiaddr *ifma; 847 struct igmp_ifsoftc *igi; 848 struct in_multi *inm; 849 int is_general_query; 850 uint16_t timer; 851 852 NET_EPOCH_ASSERT(); 853 854 is_general_query = 0; 855 856 /* 857 * Validate address fields upfront. 858 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 859 */ 860 if (in_nullhost(igmp->igmp_group)) { 861 /* 862 * IGMPv2 General Query. 863 * If this was not sent to the all-hosts group, ignore it. 864 */ 865 if (!in_allhosts(ip->ip_dst)) 866 return (0); 867 IGMPSTAT_INC(igps_rcv_gen_queries); 868 is_general_query = 1; 869 } else { 870 /* IGMPv2 Group-Specific Query. */ 871 IGMPSTAT_INC(igps_rcv_group_queries); 872 } 873 874 IN_MULTI_LIST_LOCK(); 875 IGMP_LOCK(); 876 877 igi = ((struct in_ifinfo *)ifp->if_inet)->ii_igmp; 878 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 879 880 if (igi->igi_flags & IGIF_LOOPBACK) { 881 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)", 882 ifp, ifp->if_xname); 883 goto out_locked; 884 } 885 886 /* 887 * Ignore v2 query if in v1 Compatibility Mode. 888 */ 889 if (igi->igi_version == IGMP_VERSION_1) 890 goto out_locked; 891 892 igmp_set_version(igi, IGMP_VERSION_2); 893 894 timer = igmp->igmp_code * IGMP_FASTHZ / IGMP_TIMER_SCALE; 895 if (timer == 0) 896 timer = 1; 897 898 if (is_general_query) { 899 /* 900 * For each reporting group joined on this 901 * interface, kick the report timer. 902 */ 903 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)", 904 ifp, ifp->if_xname); 905 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 906 inm = inm_ifmultiaddr_get_inm(ifma); 907 if (inm == NULL) 908 continue; 909 igmp_v2_update_group(inm, timer); 910 } 911 } else { 912 /* 913 * Group-specific IGMPv2 query, we need only 914 * look up the single group to process it. 915 */ 916 inm = inm_lookup(ifp, igmp->igmp_group); 917 if (inm != NULL) { 918 CTR3(KTR_IGMPV3, 919 "process v2 query 0x%08x on ifp %p(%s)", 920 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 921 igmp_v2_update_group(inm, timer); 922 } 923 } 924 925 out_locked: 926 IGMP_UNLOCK(); 927 IN_MULTI_LIST_UNLOCK(); 928 929 return (0); 930 } 931 932 /* 933 * Update the report timer on a group in response to an IGMPv2 query. 934 * 935 * If we are becoming the reporting member for this group, start the timer. 936 * If we already are the reporting member for this group, and timer is 937 * below the threshold, reset it. 938 * 939 * We may be updating the group for the first time since we switched 940 * to IGMPv3. If we are, then we must clear any recorded source lists, 941 * and transition to REPORTING state; the group timer is overloaded 942 * for group and group-source query responses. 943 * 944 * Unlike IGMPv3, the delay per group should be jittered 945 * to avoid bursts of IGMPv2 reports. 946 */ 947 static void 948 igmp_v2_update_group(struct in_multi *inm, const int timer) 949 { 950 951 CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__, 952 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer); 953 954 IN_MULTI_LIST_LOCK_ASSERT(); 955 956 switch (inm->inm_state) { 957 case IGMP_NOT_MEMBER: 958 case IGMP_SILENT_MEMBER: 959 break; 960 case IGMP_REPORTING_MEMBER: 961 if (inm->inm_timer != 0 && 962 inm->inm_timer <= timer) { 963 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, " 964 "skipping.", __func__); 965 break; 966 } 967 /* FALLTHROUGH */ 968 case IGMP_SG_QUERY_PENDING_MEMBER: 969 case IGMP_G_QUERY_PENDING_MEMBER: 970 case IGMP_IDLE_MEMBER: 971 case IGMP_LAZY_MEMBER: 972 case IGMP_AWAKENING_MEMBER: 973 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__); 974 inm->inm_state = IGMP_REPORTING_MEMBER; 975 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 976 V_current_state_timers_running = 1; 977 break; 978 case IGMP_SLEEPING_MEMBER: 979 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__); 980 inm->inm_state = IGMP_AWAKENING_MEMBER; 981 break; 982 case IGMP_LEAVING_MEMBER: 983 break; 984 } 985 } 986 987 /* 988 * Process a received IGMPv3 general, group-specific or 989 * group-and-source-specific query. 990 * Assumes m has already been pulled up to the full IGMP message length. 991 * Return 0 if successful, otherwise an appropriate error code is returned. 992 */ 993 static int 994 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip, 995 /*const*/ struct igmpv3 *igmpv3) 996 { 997 struct igmp_ifsoftc *igi; 998 struct in_multi *inm; 999 int is_general_query; 1000 uint32_t maxresp, nsrc, qqi; 1001 uint16_t timer; 1002 uint8_t qrv; 1003 1004 is_general_query = 0; 1005 1006 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname); 1007 1008 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */ 1009 if (maxresp >= 128) { 1010 maxresp = IGMP_MANT(igmpv3->igmp_code) << 1011 (IGMP_EXP(igmpv3->igmp_code) + 3); 1012 } 1013 1014 /* 1015 * Robustness must never be less than 2 for on-wire IGMPv3. 1016 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make 1017 * an exception for interfaces whose IGMPv3 state changes 1018 * are redirected to loopback (e.g. MANET). 1019 */ 1020 qrv = IGMP_QRV(igmpv3->igmp_misc); 1021 if (qrv < 2) { 1022 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__, 1023 qrv, IGMP_RV_INIT); 1024 qrv = IGMP_RV_INIT; 1025 } 1026 1027 qqi = igmpv3->igmp_qqi; 1028 if (qqi >= 128) { 1029 qqi = IGMP_MANT(igmpv3->igmp_qqi) << 1030 (IGMP_EXP(igmpv3->igmp_qqi) + 3); 1031 } 1032 1033 timer = maxresp * IGMP_FASTHZ / IGMP_TIMER_SCALE; 1034 if (timer == 0) 1035 timer = 1; 1036 1037 nsrc = ntohs(igmpv3->igmp_numsrc); 1038 1039 /* 1040 * Validate address fields and versions upfront before 1041 * accepting v3 query. 1042 * XXX SMPng: Unlocked access to igmpstat counters here. 1043 */ 1044 if (in_nullhost(igmpv3->igmp_group)) { 1045 /* 1046 * IGMPv3 General Query. 1047 * 1048 * General Queries SHOULD be directed to 224.0.0.1. 1049 * A general query with a source list has undefined 1050 * behaviour; discard it. 1051 */ 1052 IGMPSTAT_INC(igps_rcv_gen_queries); 1053 if (!in_allhosts(ip->ip_dst) || nsrc > 0) { 1054 IGMPSTAT_INC(igps_rcv_badqueries); 1055 return (0); 1056 } 1057 is_general_query = 1; 1058 } else { 1059 /* Group or group-source specific query. */ 1060 if (nsrc == 0) 1061 IGMPSTAT_INC(igps_rcv_group_queries); 1062 else 1063 IGMPSTAT_INC(igps_rcv_gsr_queries); 1064 } 1065 1066 IN_MULTI_LIST_LOCK(); 1067 IGMP_LOCK(); 1068 1069 igi = ((struct in_ifinfo *)ifp->if_inet)->ii_igmp; 1070 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 1071 1072 if (igi->igi_flags & IGIF_LOOPBACK) { 1073 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)", 1074 ifp, ifp->if_xname); 1075 goto out_locked; 1076 } 1077 1078 /* 1079 * Discard the v3 query if we're in Compatibility Mode. 1080 * The RFC is not obviously worded that hosts need to stay in 1081 * compatibility mode until the Old Version Querier Present 1082 * timer expires. 1083 */ 1084 if (igi->igi_version != IGMP_VERSION_3) { 1085 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)", 1086 igi->igi_version, ifp, ifp->if_xname); 1087 goto out_locked; 1088 } 1089 1090 igmp_set_version(igi, IGMP_VERSION_3); 1091 igi->igi_rv = qrv; 1092 igi->igi_qi = qqi; 1093 igi->igi_qri = maxresp; 1094 1095 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi, 1096 maxresp); 1097 1098 if (is_general_query) { 1099 /* 1100 * Schedule a current-state report on this ifp for 1101 * all groups, possibly containing source lists. 1102 * If there is a pending General Query response 1103 * scheduled earlier than the selected delay, do 1104 * not schedule any other reports. 1105 * Otherwise, reset the interface timer. 1106 */ 1107 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)", 1108 ifp, ifp->if_xname); 1109 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) { 1110 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer); 1111 V_interface_timers_running = 1; 1112 } 1113 } else { 1114 /* 1115 * Group-source-specific queries are throttled on 1116 * a per-group basis to defeat denial-of-service attempts. 1117 * Queries for groups we are not a member of on this 1118 * link are simply ignored. 1119 */ 1120 inm = inm_lookup(ifp, igmpv3->igmp_group); 1121 if (inm == NULL) 1122 goto out_locked; 1123 if (nsrc > 0) { 1124 if (!ratecheck(&inm->inm_lastgsrtv, 1125 &V_igmp_gsrdelay)) { 1126 CTR1(KTR_IGMPV3, "%s: GS query throttled.", 1127 __func__); 1128 IGMPSTAT_INC(igps_drop_gsr_queries); 1129 goto out_locked; 1130 } 1131 } 1132 CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)", 1133 ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname); 1134 /* 1135 * If there is a pending General Query response 1136 * scheduled sooner than the selected delay, no 1137 * further report need be scheduled. 1138 * Otherwise, prepare to respond to the 1139 * group-specific or group-and-source query. 1140 */ 1141 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) 1142 igmp_input_v3_group_query(inm, igi, timer, igmpv3); 1143 } 1144 1145 out_locked: 1146 IGMP_UNLOCK(); 1147 IN_MULTI_LIST_UNLOCK(); 1148 1149 return (0); 1150 } 1151 1152 /* 1153 * Process a received IGMPv3 group-specific or group-and-source-specific 1154 * query. 1155 * Return <0 if any error occurred. Currently this is ignored. 1156 */ 1157 static int 1158 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi, 1159 int timer, /*const*/ struct igmpv3 *igmpv3) 1160 { 1161 int retval; 1162 uint16_t nsrc; 1163 1164 IN_MULTI_LIST_LOCK_ASSERT(); 1165 IGMP_LOCK_ASSERT(); 1166 1167 retval = 0; 1168 1169 switch (inm->inm_state) { 1170 case IGMP_NOT_MEMBER: 1171 case IGMP_SILENT_MEMBER: 1172 case IGMP_SLEEPING_MEMBER: 1173 case IGMP_LAZY_MEMBER: 1174 case IGMP_AWAKENING_MEMBER: 1175 case IGMP_IDLE_MEMBER: 1176 case IGMP_LEAVING_MEMBER: 1177 return (retval); 1178 break; 1179 case IGMP_REPORTING_MEMBER: 1180 case IGMP_G_QUERY_PENDING_MEMBER: 1181 case IGMP_SG_QUERY_PENDING_MEMBER: 1182 break; 1183 } 1184 1185 nsrc = ntohs(igmpv3->igmp_numsrc); 1186 1187 /* 1188 * Deal with group-specific queries upfront. 1189 * If any group query is already pending, purge any recorded 1190 * source-list state if it exists, and schedule a query response 1191 * for this group-specific query. 1192 */ 1193 if (nsrc == 0) { 1194 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || 1195 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) { 1196 inm_clear_recorded(inm); 1197 timer = min(inm->inm_timer, timer); 1198 } 1199 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER; 1200 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1201 V_current_state_timers_running = 1; 1202 return (retval); 1203 } 1204 1205 /* 1206 * Deal with the case where a group-and-source-specific query has 1207 * been received but a group-specific query is already pending. 1208 */ 1209 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) { 1210 timer = min(inm->inm_timer, timer); 1211 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1212 V_current_state_timers_running = 1; 1213 return (retval); 1214 } 1215 1216 /* 1217 * Finally, deal with the case where a group-and-source-specific 1218 * query has been received, where a response to a previous g-s-r 1219 * query exists, or none exists. 1220 * In this case, we need to parse the source-list which the Querier 1221 * has provided us with and check if we have any source list filter 1222 * entries at T1 for these sources. If we do not, there is no need 1223 * schedule a report and the query may be dropped. 1224 * If we do, we must record them and schedule a current-state 1225 * report for those sources. 1226 * FIXME: Handling source lists larger than 1 mbuf requires that 1227 * we pass the mbuf chain pointer down to this function, and use 1228 * m_getptr() to walk the chain. 1229 */ 1230 if (inm->inm_nsrc > 0) { 1231 const struct in_addr *ap; 1232 int i, nrecorded; 1233 1234 ap = (const struct in_addr *)(igmpv3 + 1); 1235 nrecorded = 0; 1236 for (i = 0; i < nsrc; i++, ap++) { 1237 retval = inm_record_source(inm, ap->s_addr); 1238 if (retval < 0) 1239 break; 1240 nrecorded += retval; 1241 } 1242 if (nrecorded > 0) { 1243 CTR1(KTR_IGMPV3, 1244 "%s: schedule response to SG query", __func__); 1245 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER; 1246 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1247 V_current_state_timers_running = 1; 1248 } 1249 } 1250 1251 return (retval); 1252 } 1253 1254 /* 1255 * Process a received IGMPv1 host membership report. 1256 * 1257 * NOTE: 0.0.0.0 workaround breaks const correctness. 1258 */ 1259 static int 1260 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1261 /*const*/ struct igmp *igmp) 1262 { 1263 struct in_ifaddr *ia; 1264 struct in_multi *inm; 1265 1266 IGMPSTAT_INC(igps_rcv_reports); 1267 1268 if (ifp->if_flags & IFF_LOOPBACK) 1269 return (0); 1270 1271 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || 1272 !in_hosteq(igmp->igmp_group, ip->ip_dst)) { 1273 IGMPSTAT_INC(igps_rcv_badreports); 1274 return (EINVAL); 1275 } 1276 1277 /* 1278 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1279 * Booting clients may use the source address 0.0.0.0. Some 1280 * IGMP daemons may not know how to use IP_RECVIF to determine 1281 * the interface upon which this message was received. 1282 * Replace 0.0.0.0 with the subnet address if told to do so. 1283 */ 1284 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1285 IFP_TO_IA(ifp, ia); 1286 if (ia != NULL) 1287 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1288 } 1289 1290 CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)", 1291 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 1292 1293 /* 1294 * IGMPv1 report suppression. 1295 * If we are a member of this group, and our membership should be 1296 * reported, stop our group timer and transition to the 'lazy' state. 1297 */ 1298 IN_MULTI_LIST_LOCK(); 1299 inm = inm_lookup(ifp, igmp->igmp_group); 1300 if (inm != NULL) { 1301 struct igmp_ifsoftc *igi; 1302 1303 igi = inm->inm_igi; 1304 if (igi == NULL) { 1305 KASSERT(igi != NULL, 1306 ("%s: no igi for ifp %p", __func__, ifp)); 1307 goto out_locked; 1308 } 1309 1310 IGMPSTAT_INC(igps_rcv_ourreports); 1311 1312 /* 1313 * If we are in IGMPv3 host mode, do not allow the 1314 * other host's IGMPv1 report to suppress our reports 1315 * unless explicitly configured to do so. 1316 */ 1317 if (igi->igi_version == IGMP_VERSION_3) { 1318 if (V_igmp_legacysupp) 1319 igmp_v3_suppress_group_record(inm); 1320 goto out_locked; 1321 } 1322 1323 inm->inm_timer = 0; 1324 1325 switch (inm->inm_state) { 1326 case IGMP_NOT_MEMBER: 1327 case IGMP_SILENT_MEMBER: 1328 break; 1329 case IGMP_IDLE_MEMBER: 1330 case IGMP_LAZY_MEMBER: 1331 case IGMP_AWAKENING_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 case IGMP_SLEEPING_MEMBER: 1337 inm->inm_state = IGMP_SLEEPING_MEMBER; 1338 break; 1339 case IGMP_REPORTING_MEMBER: 1340 CTR3(KTR_IGMPV3, 1341 "report suppressed for 0x%08x on ifp %p(%s)", 1342 ntohl(igmp->igmp_group.s_addr), ifp, 1343 ifp->if_xname); 1344 if (igi->igi_version == IGMP_VERSION_1) 1345 inm->inm_state = IGMP_LAZY_MEMBER; 1346 else if (igi->igi_version == IGMP_VERSION_2) 1347 inm->inm_state = IGMP_SLEEPING_MEMBER; 1348 break; 1349 case IGMP_G_QUERY_PENDING_MEMBER: 1350 case IGMP_SG_QUERY_PENDING_MEMBER: 1351 case IGMP_LEAVING_MEMBER: 1352 break; 1353 } 1354 } 1355 1356 out_locked: 1357 IN_MULTI_LIST_UNLOCK(); 1358 1359 return (0); 1360 } 1361 1362 /* 1363 * Process a received IGMPv2 host membership report. 1364 * 1365 * NOTE: 0.0.0.0 workaround breaks const correctness. 1366 */ 1367 static int 1368 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1369 /*const*/ struct igmp *igmp) 1370 { 1371 struct in_ifaddr *ia; 1372 struct in_multi *inm; 1373 1374 /* 1375 * Make sure we don't hear our own membership report. Fast 1376 * leave requires knowing that we are the only member of a 1377 * group. 1378 */ 1379 IFP_TO_IA(ifp, ia); 1380 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) { 1381 return (0); 1382 } 1383 1384 IGMPSTAT_INC(igps_rcv_reports); 1385 1386 if (ifp->if_flags & IFF_LOOPBACK) { 1387 return (0); 1388 } 1389 1390 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || 1391 !in_hosteq(igmp->igmp_group, ip->ip_dst)) { 1392 IGMPSTAT_INC(igps_rcv_badreports); 1393 return (EINVAL); 1394 } 1395 1396 /* 1397 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1398 * Booting clients may use the source address 0.0.0.0. Some 1399 * IGMP daemons may not know how to use IP_RECVIF to determine 1400 * the interface upon which this message was received. 1401 * Replace 0.0.0.0 with the subnet address if told to do so. 1402 */ 1403 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1404 if (ia != NULL) 1405 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1406 } 1407 1408 CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)", 1409 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 1410 1411 /* 1412 * IGMPv2 report suppression. 1413 * If we are a member of this group, and our membership should be 1414 * reported, and our group timer is pending or about to be reset, 1415 * stop our group timer by transitioning to the 'lazy' state. 1416 */ 1417 IN_MULTI_LIST_LOCK(); 1418 inm = inm_lookup(ifp, igmp->igmp_group); 1419 if (inm != NULL) { 1420 struct igmp_ifsoftc *igi; 1421 1422 igi = inm->inm_igi; 1423 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp)); 1424 1425 IGMPSTAT_INC(igps_rcv_ourreports); 1426 1427 /* 1428 * If we are in IGMPv3 host mode, do not allow the 1429 * other host's IGMPv1 report to suppress our reports 1430 * unless explicitly configured to do so. 1431 */ 1432 if (igi->igi_version == IGMP_VERSION_3) { 1433 if (V_igmp_legacysupp) 1434 igmp_v3_suppress_group_record(inm); 1435 goto out_locked; 1436 } 1437 1438 inm->inm_timer = 0; 1439 1440 switch (inm->inm_state) { 1441 case IGMP_NOT_MEMBER: 1442 case IGMP_SILENT_MEMBER: 1443 case IGMP_SLEEPING_MEMBER: 1444 break; 1445 case IGMP_REPORTING_MEMBER: 1446 case IGMP_IDLE_MEMBER: 1447 case IGMP_AWAKENING_MEMBER: 1448 CTR3(KTR_IGMPV3, 1449 "report suppressed for 0x%08x on ifp %p(%s)", 1450 ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); 1451 case IGMP_LAZY_MEMBER: 1452 inm->inm_state = IGMP_LAZY_MEMBER; 1453 break; 1454 case IGMP_G_QUERY_PENDING_MEMBER: 1455 case IGMP_SG_QUERY_PENDING_MEMBER: 1456 case IGMP_LEAVING_MEMBER: 1457 break; 1458 } 1459 } 1460 1461 out_locked: 1462 IN_MULTI_LIST_UNLOCK(); 1463 1464 return (0); 1465 } 1466 1467 int 1468 igmp_input(struct mbuf **mp, int *offp, int proto) 1469 { 1470 int iphlen; 1471 struct ifnet *ifp; 1472 struct igmp *igmp; 1473 struct ip *ip; 1474 struct mbuf *m; 1475 int igmplen; 1476 int minlen; 1477 int queryver; 1478 1479 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp); 1480 1481 m = *mp; 1482 ifp = m->m_pkthdr.rcvif; 1483 *mp = NULL; 1484 M_ASSERTMAPPED(m); 1485 1486 IGMPSTAT_INC(igps_rcv_total); 1487 1488 ip = mtod(m, struct ip *); 1489 iphlen = *offp; 1490 igmplen = ntohs(ip->ip_len) - iphlen; 1491 1492 /* 1493 * Validate lengths. 1494 */ 1495 if (igmplen < IGMP_MINLEN) { 1496 IGMPSTAT_INC(igps_rcv_tooshort); 1497 m_freem(m); 1498 return (IPPROTO_DONE); 1499 } 1500 1501 /* 1502 * Always pullup to the minimum size for v1/v2 or v3 1503 * to amortize calls to m_pullup(). 1504 */ 1505 minlen = iphlen; 1506 if (igmplen >= IGMP_V3_QUERY_MINLEN) 1507 minlen += IGMP_V3_QUERY_MINLEN; 1508 else 1509 minlen += IGMP_MINLEN; 1510 if ((!M_WRITABLE(m) || m->m_len < minlen) && 1511 (m = m_pullup(m, minlen)) == NULL) { 1512 IGMPSTAT_INC(igps_rcv_tooshort); 1513 return (IPPROTO_DONE); 1514 } 1515 ip = mtod(m, struct ip *); 1516 1517 /* 1518 * Validate checksum. 1519 */ 1520 m->m_data += iphlen; 1521 m->m_len -= iphlen; 1522 igmp = mtod(m, struct igmp *); 1523 if (in_cksum(m, igmplen)) { 1524 IGMPSTAT_INC(igps_rcv_badsum); 1525 m_freem(m); 1526 return (IPPROTO_DONE); 1527 } 1528 m->m_data -= iphlen; 1529 m->m_len += iphlen; 1530 1531 /* 1532 * IGMP control traffic is link-scope, and must have a TTL of 1. 1533 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception; 1534 * probe packets may come from beyond the LAN. 1535 */ 1536 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) { 1537 IGMPSTAT_INC(igps_rcv_badttl); 1538 m_freem(m); 1539 return (IPPROTO_DONE); 1540 } 1541 1542 switch (igmp->igmp_type) { 1543 case IGMP_HOST_MEMBERSHIP_QUERY: 1544 if (igmplen == IGMP_MINLEN) { 1545 if (igmp->igmp_code == 0) 1546 queryver = IGMP_VERSION_1; 1547 else 1548 queryver = IGMP_VERSION_2; 1549 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) { 1550 queryver = IGMP_VERSION_3; 1551 } else { 1552 IGMPSTAT_INC(igps_rcv_tooshort); 1553 m_freem(m); 1554 return (IPPROTO_DONE); 1555 } 1556 1557 switch (queryver) { 1558 case IGMP_VERSION_1: 1559 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1560 if (!V_igmp_v1enable) 1561 break; 1562 if (igmp_input_v1_query(ifp, ip, igmp) != 0) { 1563 m_freem(m); 1564 return (IPPROTO_DONE); 1565 } 1566 break; 1567 1568 case IGMP_VERSION_2: 1569 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1570 if (!V_igmp_v2enable) 1571 break; 1572 if (igmp_input_v2_query(ifp, ip, igmp) != 0) { 1573 m_freem(m); 1574 return (IPPROTO_DONE); 1575 } 1576 break; 1577 1578 case IGMP_VERSION_3: { 1579 struct igmpv3 *igmpv3; 1580 uint16_t igmpv3len; 1581 uint16_t nsrc; 1582 1583 IGMPSTAT_INC(igps_rcv_v3_queries); 1584 igmpv3 = (struct igmpv3 *)igmp; 1585 /* 1586 * Validate length based on source count. 1587 */ 1588 nsrc = ntohs(igmpv3->igmp_numsrc); 1589 if (nsrc * sizeof(in_addr_t) > 1590 UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) { 1591 IGMPSTAT_INC(igps_rcv_tooshort); 1592 m_freem(m); 1593 return (IPPROTO_DONE); 1594 } 1595 /* 1596 * m_pullup() may modify m, so pullup in 1597 * this scope. 1598 */ 1599 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN + 1600 sizeof(struct in_addr) * nsrc; 1601 if ((!M_WRITABLE(m) || 1602 m->m_len < igmpv3len) && 1603 (m = m_pullup(m, igmpv3len)) == NULL) { 1604 IGMPSTAT_INC(igps_rcv_tooshort); 1605 return (IPPROTO_DONE); 1606 } 1607 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *) 1608 + iphlen); 1609 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) { 1610 m_freem(m); 1611 return (IPPROTO_DONE); 1612 } 1613 } 1614 break; 1615 } 1616 break; 1617 1618 case IGMP_v1_HOST_MEMBERSHIP_REPORT: 1619 if (!V_igmp_v1enable) 1620 break; 1621 if (igmp_input_v1_report(ifp, ip, igmp) != 0) { 1622 m_freem(m); 1623 return (IPPROTO_DONE); 1624 } 1625 break; 1626 1627 case IGMP_v2_HOST_MEMBERSHIP_REPORT: 1628 if (!V_igmp_v2enable) 1629 break; 1630 if (!ip_checkrouteralert(m)) 1631 IGMPSTAT_INC(igps_rcv_nora); 1632 if (igmp_input_v2_report(ifp, ip, igmp) != 0) { 1633 m_freem(m); 1634 return (IPPROTO_DONE); 1635 } 1636 break; 1637 1638 case IGMP_v3_HOST_MEMBERSHIP_REPORT: 1639 /* 1640 * Hosts do not need to process IGMPv3 membership reports, 1641 * as report suppression is no longer required. 1642 */ 1643 if (!ip_checkrouteralert(m)) 1644 IGMPSTAT_INC(igps_rcv_nora); 1645 break; 1646 1647 default: 1648 break; 1649 } 1650 1651 /* 1652 * Pass all valid IGMP packets up to any process(es) listening on a 1653 * raw IGMP socket. 1654 */ 1655 *mp = m; 1656 return (rip_input(mp, offp, proto)); 1657 } 1658 1659 /* 1660 * Fast timeout handler (global). 1661 * VIMAGE: Timeout handlers are expected to service all vimages. 1662 */ 1663 static struct callout igmpfast_callout; 1664 static void 1665 igmp_fasttimo(void *arg __unused) 1666 { 1667 struct epoch_tracker et; 1668 VNET_ITERATOR_DECL(vnet_iter); 1669 1670 NET_EPOCH_ENTER(et); 1671 VNET_LIST_RLOCK_NOSLEEP(); 1672 VNET_FOREACH(vnet_iter) { 1673 CURVNET_SET(vnet_iter); 1674 igmp_fasttimo_vnet(); 1675 CURVNET_RESTORE(); 1676 } 1677 VNET_LIST_RUNLOCK_NOSLEEP(); 1678 NET_EPOCH_EXIT(et); 1679 1680 callout_reset(&igmpfast_callout, hz / IGMP_FASTHZ, igmp_fasttimo, NULL); 1681 } 1682 1683 /* 1684 * Fast timeout handler (per-vnet). 1685 * 1686 * VIMAGE: Assume caller has set up our curvnet. 1687 */ 1688 static void 1689 igmp_fasttimo_vnet(void) 1690 { 1691 struct mbufq scq; /* State-change packets */ 1692 struct mbufq qrq; /* Query response packets */ 1693 struct ifnet *ifp; 1694 struct igmp_ifsoftc *igi; 1695 struct ifmultiaddr *ifma; 1696 struct in_multi *inm; 1697 struct in_multi_head inm_free_tmp; 1698 int loop, uri_fasthz; 1699 1700 loop = 0; 1701 uri_fasthz = 0; 1702 1703 /* 1704 * Quick check to see if any work needs to be done, in order to 1705 * minimize the overhead of fasttimo processing. 1706 * SMPng: XXX Unlocked reads. 1707 */ 1708 if (!V_current_state_timers_running && 1709 !V_interface_timers_running && 1710 !V_state_change_timers_running) 1711 return; 1712 1713 SLIST_INIT(&inm_free_tmp); 1714 IN_MULTI_LIST_LOCK(); 1715 IGMP_LOCK(); 1716 1717 /* 1718 * IGMPv3 General Query response timer processing. 1719 */ 1720 if (V_interface_timers_running) { 1721 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__); 1722 1723 V_interface_timers_running = 0; 1724 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1725 if (igi->igi_v3_timer == 0) { 1726 /* Do nothing. */ 1727 } else if (--igi->igi_v3_timer == 0) { 1728 igmp_v3_dispatch_general_query(igi); 1729 } else { 1730 V_interface_timers_running = 1; 1731 } 1732 } 1733 } 1734 1735 if (!V_current_state_timers_running && 1736 !V_state_change_timers_running) 1737 goto out_locked; 1738 1739 V_current_state_timers_running = 0; 1740 V_state_change_timers_running = 0; 1741 1742 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__); 1743 1744 /* 1745 * IGMPv1/v2/v3 host report and state-change timer processing. 1746 * Note: Processing a v3 group timer may remove a node. 1747 */ 1748 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1749 ifp = igi->igi_ifp; 1750 1751 if (igi->igi_version == IGMP_VERSION_3) { 1752 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; 1753 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri * 1754 IGMP_FASTHZ); 1755 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS); 1756 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS); 1757 } 1758 1759 IF_ADDR_WLOCK(ifp); 1760 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1761 inm = inm_ifmultiaddr_get_inm(ifma); 1762 if (inm == NULL) 1763 continue; 1764 switch (igi->igi_version) { 1765 case IGMP_VERSION_1: 1766 case IGMP_VERSION_2: 1767 igmp_v1v2_process_group_timer(inm, 1768 igi->igi_version); 1769 break; 1770 case IGMP_VERSION_3: 1771 igmp_v3_process_group_timers(&inm_free_tmp, &qrq, 1772 &scq, inm, uri_fasthz); 1773 break; 1774 } 1775 } 1776 IF_ADDR_WUNLOCK(ifp); 1777 1778 if (igi->igi_version == IGMP_VERSION_3) { 1779 igmp_dispatch_queue(&qrq, 0, loop); 1780 igmp_dispatch_queue(&scq, 0, loop); 1781 1782 /* 1783 * Free the in_multi reference(s) for this 1784 * IGMP lifecycle. 1785 */ 1786 inm_release_list_deferred(&inm_free_tmp); 1787 } 1788 } 1789 1790 out_locked: 1791 IGMP_UNLOCK(); 1792 IN_MULTI_LIST_UNLOCK(); 1793 } 1794 1795 /* 1796 * Update host report group timer for IGMPv1/v2. 1797 * Will update the global pending timer flags. 1798 */ 1799 static void 1800 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version) 1801 { 1802 int report_timer_expired; 1803 1804 IN_MULTI_LIST_LOCK_ASSERT(); 1805 IGMP_LOCK_ASSERT(); 1806 1807 if (inm->inm_timer == 0) { 1808 report_timer_expired = 0; 1809 } else if (--inm->inm_timer == 0) { 1810 report_timer_expired = 1; 1811 } else { 1812 V_current_state_timers_running = 1; 1813 return; 1814 } 1815 1816 switch (inm->inm_state) { 1817 case IGMP_NOT_MEMBER: 1818 case IGMP_SILENT_MEMBER: 1819 case IGMP_IDLE_MEMBER: 1820 case IGMP_LAZY_MEMBER: 1821 case IGMP_SLEEPING_MEMBER: 1822 case IGMP_AWAKENING_MEMBER: 1823 break; 1824 case IGMP_REPORTING_MEMBER: 1825 if (report_timer_expired) { 1826 inm->inm_state = IGMP_IDLE_MEMBER; 1827 (void)igmp_v1v2_queue_report(inm, 1828 (version == IGMP_VERSION_2) ? 1829 IGMP_v2_HOST_MEMBERSHIP_REPORT : 1830 IGMP_v1_HOST_MEMBERSHIP_REPORT); 1831 } 1832 break; 1833 case IGMP_G_QUERY_PENDING_MEMBER: 1834 case IGMP_SG_QUERY_PENDING_MEMBER: 1835 case IGMP_LEAVING_MEMBER: 1836 break; 1837 } 1838 } 1839 1840 /* 1841 * Update a group's timers for IGMPv3. 1842 * Will update the global pending timer flags. 1843 * Note: Unlocked read from igi. 1844 */ 1845 static void 1846 igmp_v3_process_group_timers(struct in_multi_head *inmh, 1847 struct mbufq *qrq, struct mbufq *scq, 1848 struct in_multi *inm, const int uri_fasthz) 1849 { 1850 int query_response_timer_expired; 1851 int state_change_retransmit_timer_expired; 1852 1853 IN_MULTI_LIST_LOCK_ASSERT(); 1854 IGMP_LOCK_ASSERT(); 1855 1856 query_response_timer_expired = 0; 1857 state_change_retransmit_timer_expired = 0; 1858 1859 /* 1860 * During a transition from v1/v2 compatibility mode back to v3, 1861 * a group record in REPORTING state may still have its group 1862 * timer active. This is a no-op in this function; it is easier 1863 * to deal with it here than to complicate the slow-timeout path. 1864 */ 1865 if (inm->inm_timer == 0) { 1866 query_response_timer_expired = 0; 1867 } else if (--inm->inm_timer == 0) { 1868 query_response_timer_expired = 1; 1869 } else { 1870 V_current_state_timers_running = 1; 1871 } 1872 1873 if (inm->inm_sctimer == 0) { 1874 state_change_retransmit_timer_expired = 0; 1875 } else if (--inm->inm_sctimer == 0) { 1876 state_change_retransmit_timer_expired = 1; 1877 } else { 1878 V_state_change_timers_running = 1; 1879 } 1880 1881 /* We are in fasttimo, so be quick about it. */ 1882 if (!state_change_retransmit_timer_expired && 1883 !query_response_timer_expired) 1884 return; 1885 1886 switch (inm->inm_state) { 1887 case IGMP_NOT_MEMBER: 1888 case IGMP_SILENT_MEMBER: 1889 case IGMP_SLEEPING_MEMBER: 1890 case IGMP_LAZY_MEMBER: 1891 case IGMP_AWAKENING_MEMBER: 1892 case IGMP_IDLE_MEMBER: 1893 break; 1894 case IGMP_G_QUERY_PENDING_MEMBER: 1895 case IGMP_SG_QUERY_PENDING_MEMBER: 1896 /* 1897 * Respond to a previously pending Group-Specific 1898 * or Group-and-Source-Specific query by enqueueing 1899 * the appropriate Current-State report for 1900 * immediate transmission. 1901 */ 1902 if (query_response_timer_expired) { 1903 int retval __unused; 1904 1905 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1, 1906 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)); 1907 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 1908 __func__, retval); 1909 inm->inm_state = IGMP_REPORTING_MEMBER; 1910 /* XXX Clear recorded sources for next time. */ 1911 inm_clear_recorded(inm); 1912 } 1913 /* FALLTHROUGH */ 1914 case IGMP_REPORTING_MEMBER: 1915 case IGMP_LEAVING_MEMBER: 1916 if (state_change_retransmit_timer_expired) { 1917 /* 1918 * State-change retransmission timer fired. 1919 * If there are any further pending retransmissions, 1920 * set the global pending state-change flag, and 1921 * reset the timer. 1922 */ 1923 if (--inm->inm_scrv > 0) { 1924 inm->inm_sctimer = uri_fasthz; 1925 V_state_change_timers_running = 1; 1926 } 1927 /* 1928 * Retransmit the previously computed state-change 1929 * report. If there are no further pending 1930 * retransmissions, the mbuf queue will be consumed. 1931 * Update T0 state to T1 as we have now sent 1932 * a state-change. 1933 */ 1934 (void)igmp_v3_merge_state_changes(inm, scq); 1935 1936 inm_commit(inm); 1937 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 1938 ntohl(inm->inm_addr.s_addr), 1939 inm->inm_ifp->if_xname); 1940 1941 /* 1942 * If we are leaving the group for good, make sure 1943 * we release IGMP's reference to it. 1944 * This release must be deferred using a SLIST, 1945 * as we are called from a loop which traverses 1946 * the in_ifmultiaddr TAILQ. 1947 */ 1948 if (inm->inm_state == IGMP_LEAVING_MEMBER && 1949 inm->inm_scrv == 0) { 1950 inm->inm_state = IGMP_NOT_MEMBER; 1951 inm_rele_locked(inmh, inm); 1952 } 1953 } 1954 break; 1955 } 1956 } 1957 1958 /* 1959 * Suppress a group's pending response to a group or source/group query. 1960 * 1961 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency. 1962 * Do NOT update ST1/ST0 as this operation merely suppresses 1963 * the currently pending group record. 1964 * Do NOT suppress the response to a general query. It is possible but 1965 * it would require adding another state or flag. 1966 */ 1967 static void 1968 igmp_v3_suppress_group_record(struct in_multi *inm) 1969 { 1970 1971 IN_MULTI_LIST_LOCK_ASSERT(); 1972 1973 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3, 1974 ("%s: not IGMPv3 mode on link", __func__)); 1975 1976 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER || 1977 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER) 1978 return; 1979 1980 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) 1981 inm_clear_recorded(inm); 1982 1983 inm->inm_timer = 0; 1984 inm->inm_state = IGMP_REPORTING_MEMBER; 1985 } 1986 1987 /* 1988 * Switch to a different IGMP version on the given interface, 1989 * as per Section 7.2.1. 1990 */ 1991 static void 1992 igmp_set_version(struct igmp_ifsoftc *igi, const int version) 1993 { 1994 int old_version_timer; 1995 1996 IGMP_LOCK_ASSERT(); 1997 1998 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__, 1999 version, igi->igi_ifp, igi->igi_ifp->if_xname); 2000 2001 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) { 2002 /* 2003 * Compute the "Older Version Querier Present" timer as per 2004 * Section 8.12. 2005 */ 2006 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri; 2007 old_version_timer *= IGMP_SLOWHZ; 2008 2009 if (version == IGMP_VERSION_1) { 2010 igi->igi_v1_timer = old_version_timer; 2011 igi->igi_v2_timer = 0; 2012 } else if (version == IGMP_VERSION_2) { 2013 igi->igi_v1_timer = 0; 2014 igi->igi_v2_timer = old_version_timer; 2015 } 2016 } 2017 2018 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 2019 if (igi->igi_version != IGMP_VERSION_2) { 2020 igi->igi_version = IGMP_VERSION_2; 2021 igmp_v3_cancel_link_timers(igi); 2022 } 2023 } else if (igi->igi_v1_timer > 0) { 2024 if (igi->igi_version != IGMP_VERSION_1) { 2025 igi->igi_version = IGMP_VERSION_1; 2026 igmp_v3_cancel_link_timers(igi); 2027 } 2028 } 2029 } 2030 2031 /* 2032 * Cancel pending IGMPv3 timers for the given link and all groups 2033 * joined on it; state-change, general-query, and group-query timers. 2034 * 2035 * Only ever called on a transition from v3 to Compatibility mode. Kill 2036 * the timers stone dead (this may be expensive for large N groups), they 2037 * will be restarted if Compatibility Mode deems that they must be due to 2038 * query processing. 2039 */ 2040 static void 2041 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi) 2042 { 2043 struct ifmultiaddr *ifma; 2044 struct ifnet *ifp; 2045 struct in_multi *inm; 2046 struct in_multi_head inm_free_tmp; 2047 2048 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__, 2049 igi->igi_ifp, igi->igi_ifp->if_xname); 2050 2051 IN_MULTI_LIST_LOCK_ASSERT(); 2052 IGMP_LOCK_ASSERT(); 2053 NET_EPOCH_ASSERT(); 2054 2055 SLIST_INIT(&inm_free_tmp); 2056 2057 /* 2058 * Stop the v3 General Query Response on this link stone dead. 2059 * If fasttimo is woken up due to V_interface_timers_running, 2060 * the flag will be cleared if there are no pending link timers. 2061 */ 2062 igi->igi_v3_timer = 0; 2063 2064 /* 2065 * Now clear the current-state and state-change report timers 2066 * for all memberships scoped to this link. 2067 */ 2068 ifp = igi->igi_ifp; 2069 IF_ADDR_WLOCK(ifp); 2070 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2071 inm = inm_ifmultiaddr_get_inm(ifma); 2072 if (inm == NULL) 2073 continue; 2074 switch (inm->inm_state) { 2075 case IGMP_NOT_MEMBER: 2076 case IGMP_SILENT_MEMBER: 2077 case IGMP_IDLE_MEMBER: 2078 case IGMP_LAZY_MEMBER: 2079 case IGMP_SLEEPING_MEMBER: 2080 case IGMP_AWAKENING_MEMBER: 2081 /* 2082 * These states are either not relevant in v3 mode, 2083 * or are unreported. Do nothing. 2084 */ 2085 break; 2086 case IGMP_LEAVING_MEMBER: 2087 /* 2088 * If we are leaving the group and switching to 2089 * compatibility mode, we need to release the final 2090 * reference held for issuing the INCLUDE {}, and 2091 * transition to REPORTING to ensure the host leave 2092 * message is sent upstream to the old querier -- 2093 * transition to NOT would lose the leave and race. 2094 */ 2095 inm_rele_locked(&inm_free_tmp, inm); 2096 /* FALLTHROUGH */ 2097 case IGMP_G_QUERY_PENDING_MEMBER: 2098 case IGMP_SG_QUERY_PENDING_MEMBER: 2099 inm_clear_recorded(inm); 2100 /* FALLTHROUGH */ 2101 case IGMP_REPORTING_MEMBER: 2102 inm->inm_state = IGMP_REPORTING_MEMBER; 2103 break; 2104 } 2105 /* 2106 * Always clear state-change and group report timers. 2107 * Free any pending IGMPv3 state-change records. 2108 */ 2109 inm->inm_sctimer = 0; 2110 inm->inm_timer = 0; 2111 mbufq_drain(&inm->inm_scq); 2112 } 2113 IF_ADDR_WUNLOCK(ifp); 2114 2115 inm_release_list_deferred(&inm_free_tmp); 2116 } 2117 2118 /* 2119 * Update the Older Version Querier Present timers for a link. 2120 * See Section 7.2.1 of RFC 3376. 2121 */ 2122 static void 2123 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi) 2124 { 2125 2126 IGMP_LOCK_ASSERT(); 2127 2128 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) { 2129 /* 2130 * IGMPv1 and IGMPv2 Querier Present timers expired. 2131 * 2132 * Revert to IGMPv3. 2133 */ 2134 if (V_igmp_default_version == IGMP_VERSION_3 && 2135 igi->igi_version != IGMP_VERSION_3) { 2136 CTR5(KTR_IGMPV3, 2137 "%s: transition from v%d -> v%d on %p(%s)", 2138 __func__, igi->igi_version, IGMP_VERSION_3, 2139 igi->igi_ifp, igi->igi_ifp->if_xname); 2140 igi->igi_version = IGMP_VERSION_3; 2141 } 2142 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 2143 /* 2144 * IGMPv1 Querier Present timer expired, 2145 * IGMPv2 Querier Present timer running. 2146 * If IGMPv2 was disabled since last timeout, 2147 * revert to IGMPv3. 2148 * If IGMPv2 is enabled, revert to IGMPv2. 2149 */ 2150 if (V_igmp_default_version == IGMP_VERSION_3 && 2151 !V_igmp_v2enable) { 2152 CTR5(KTR_IGMPV3, 2153 "%s: transition from v%d -> v%d on %p(%s)", 2154 __func__, igi->igi_version, IGMP_VERSION_3, 2155 igi->igi_ifp, igi->igi_ifp->if_xname); 2156 igi->igi_v2_timer = 0; 2157 igi->igi_version = IGMP_VERSION_3; 2158 } else { 2159 --igi->igi_v2_timer; 2160 if (V_igmp_default_version == IGMP_VERSION_2 && 2161 igi->igi_version != IGMP_VERSION_2) { 2162 CTR5(KTR_IGMPV3, 2163 "%s: transition from v%d -> v%d on %p(%s)", 2164 __func__, igi->igi_version, IGMP_VERSION_2, 2165 igi->igi_ifp, igi->igi_ifp->if_xname); 2166 igi->igi_version = IGMP_VERSION_2; 2167 igmp_v3_cancel_link_timers(igi); 2168 } 2169 } 2170 } else if (igi->igi_v1_timer > 0) { 2171 /* 2172 * IGMPv1 Querier Present timer running. 2173 * Stop IGMPv2 timer if running. 2174 * 2175 * If IGMPv1 was disabled since last timeout, 2176 * revert to IGMPv3. 2177 * If IGMPv1 is enabled, reset IGMPv2 timer if running. 2178 */ 2179 if (V_igmp_default_version == IGMP_VERSION_3 && 2180 !V_igmp_v1enable) { 2181 CTR5(KTR_IGMPV3, 2182 "%s: transition from v%d -> v%d on %p(%s)", 2183 __func__, igi->igi_version, IGMP_VERSION_3, 2184 igi->igi_ifp, igi->igi_ifp->if_xname); 2185 igi->igi_v1_timer = 0; 2186 igi->igi_version = IGMP_VERSION_3; 2187 } else { 2188 --igi->igi_v1_timer; 2189 } 2190 if (igi->igi_v2_timer > 0) { 2191 CTR3(KTR_IGMPV3, 2192 "%s: cancel v2 timer on %p(%s)", 2193 __func__, igi->igi_ifp, igi->igi_ifp->if_xname); 2194 igi->igi_v2_timer = 0; 2195 } 2196 } 2197 } 2198 2199 /* 2200 * Global slowtimo handler. 2201 * VIMAGE: Timeout handlers are expected to service all vimages. 2202 */ 2203 static struct callout igmpslow_callout; 2204 static void 2205 igmp_slowtimo(void *arg __unused) 2206 { 2207 struct epoch_tracker et; 2208 VNET_ITERATOR_DECL(vnet_iter); 2209 2210 NET_EPOCH_ENTER(et); 2211 VNET_LIST_RLOCK_NOSLEEP(); 2212 VNET_FOREACH(vnet_iter) { 2213 CURVNET_SET(vnet_iter); 2214 igmp_slowtimo_vnet(); 2215 CURVNET_RESTORE(); 2216 } 2217 VNET_LIST_RUNLOCK_NOSLEEP(); 2218 NET_EPOCH_EXIT(et); 2219 2220 callout_reset(&igmpslow_callout, hz / IGMP_SLOWHZ, igmp_slowtimo, NULL); 2221 } 2222 2223 /* 2224 * Per-vnet slowtimo handler. 2225 */ 2226 static void 2227 igmp_slowtimo_vnet(void) 2228 { 2229 struct igmp_ifsoftc *igi; 2230 2231 IGMP_LOCK(); 2232 2233 LIST_FOREACH(igi, &V_igi_head, igi_link) { 2234 igmp_v1v2_process_querier_timers(igi); 2235 } 2236 2237 IGMP_UNLOCK(); 2238 } 2239 2240 /* 2241 * Dispatch an IGMPv1/v2 host report or leave message. 2242 * These are always small enough to fit inside a single mbuf. 2243 */ 2244 static int 2245 igmp_v1v2_queue_report(struct in_multi *inm, const int type) 2246 { 2247 struct epoch_tracker et; 2248 struct ifnet *ifp; 2249 struct igmp *igmp; 2250 struct ip *ip; 2251 struct mbuf *m; 2252 2253 IN_MULTI_LIST_LOCK_ASSERT(); 2254 IGMP_LOCK_ASSERT(); 2255 2256 ifp = inm->inm_ifp; 2257 2258 m = m_gethdr(M_NOWAIT, MT_DATA); 2259 if (m == NULL) 2260 return (ENOMEM); 2261 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp)); 2262 2263 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp); 2264 2265 m->m_data += sizeof(struct ip); 2266 m->m_len = sizeof(struct igmp); 2267 2268 igmp = mtod(m, struct igmp *); 2269 igmp->igmp_type = type; 2270 igmp->igmp_code = 0; 2271 igmp->igmp_group = inm->inm_addr; 2272 igmp->igmp_cksum = 0; 2273 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp)); 2274 2275 m->m_data -= sizeof(struct ip); 2276 m->m_len += sizeof(struct ip); 2277 2278 ip = mtod(m, struct ip *); 2279 ip->ip_tos = 0; 2280 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp)); 2281 ip->ip_off = 0; 2282 ip->ip_p = IPPROTO_IGMP; 2283 ip->ip_src.s_addr = INADDR_ANY; 2284 2285 if (type == IGMP_HOST_LEAVE_MESSAGE) 2286 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP); 2287 else 2288 ip->ip_dst = inm->inm_addr; 2289 2290 igmp_save_context(m, ifp); 2291 2292 m->m_flags |= M_IGMPV2; 2293 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK) 2294 m->m_flags |= M_IGMP_LOOP; 2295 2296 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m); 2297 NET_EPOCH_ENTER(et); 2298 netisr_dispatch(NETISR_IGMP, m); 2299 NET_EPOCH_EXIT(et); 2300 2301 return (0); 2302 } 2303 2304 /* 2305 * Process a state change from the upper layer for the given IPv4 group. 2306 * 2307 * Each socket holds a reference on the in_multi in its own ip_moptions. 2308 * The socket layer will have made the necessary updates to.the group 2309 * state, it is now up to IGMP to issue a state change report if there 2310 * has been any change between T0 (when the last state-change was issued) 2311 * and T1 (now). 2312 * 2313 * We use the IGMPv3 state machine at group level. The IGMP module 2314 * however makes the decision as to which IGMP protocol version to speak. 2315 * A state change *from* INCLUDE {} always means an initial join. 2316 * A state change *to* INCLUDE {} always means a final leave. 2317 * 2318 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can 2319 * save ourselves a bunch of work; any exclusive mode groups need not 2320 * compute source filter lists. 2321 * 2322 * VIMAGE: curvnet should have been set by caller, as this routine 2323 * is called from the socket option handlers. 2324 */ 2325 int 2326 igmp_change_state(struct in_multi *inm) 2327 { 2328 struct igmp_ifsoftc *igi; 2329 struct ifnet *ifp; 2330 int error; 2331 2332 error = 0; 2333 IN_MULTI_LOCK_ASSERT(); 2334 /* 2335 * Try to detect if the upper layer just asked us to change state 2336 * for an interface which has now gone away. 2337 */ 2338 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 2339 ifp = inm->inm_ifma->ifma_ifp; 2340 if (ifp == NULL) 2341 return (0); 2342 /* 2343 * Sanity check that netinet's notion of ifp is the 2344 * same as net's. 2345 */ 2346 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 2347 2348 IGMP_LOCK(); 2349 2350 igi = ((struct in_ifinfo *)ifp->if_inet)->ii_igmp; 2351 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 2352 2353 /* 2354 * If we detect a state transition to or from MCAST_UNDEFINED 2355 * for this group, then we are starting or finishing an IGMP 2356 * life cycle for this group. 2357 */ 2358 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) { 2359 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__, 2360 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode); 2361 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) { 2362 CTR1(KTR_IGMPV3, "%s: initial join", __func__); 2363 error = igmp_initial_join(inm, igi); 2364 goto out_locked; 2365 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) { 2366 CTR1(KTR_IGMPV3, "%s: final leave", __func__); 2367 igmp_final_leave(inm, igi); 2368 goto out_locked; 2369 } 2370 } else { 2371 CTR1(KTR_IGMPV3, "%s: filter set change", __func__); 2372 } 2373 2374 error = igmp_handle_state_change(inm, igi); 2375 2376 out_locked: 2377 IGMP_UNLOCK(); 2378 return (error); 2379 } 2380 2381 /* 2382 * Perform the initial join for an IGMP group. 2383 * 2384 * When joining a group: 2385 * If the group should have its IGMP traffic suppressed, do nothing. 2386 * IGMPv1 starts sending IGMPv1 host membership reports. 2387 * IGMPv2 starts sending IGMPv2 host membership reports. 2388 * IGMPv3 will schedule an IGMPv3 state-change report containing the 2389 * initial state of the membership. 2390 */ 2391 static int 2392 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi) 2393 { 2394 struct ifnet *ifp; 2395 struct mbufq *mq; 2396 int error, retval, syncstates; 2397 2398 CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__, 2399 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); 2400 2401 error = 0; 2402 syncstates = 1; 2403 2404 ifp = inm->inm_ifp; 2405 2406 IN_MULTI_LOCK_ASSERT(); 2407 IGMP_LOCK_ASSERT(); 2408 2409 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); 2410 2411 /* 2412 * Groups joined on loopback or marked as 'not reported', 2413 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and 2414 * are never reported in any IGMP protocol exchanges. 2415 * All other groups enter the appropriate IGMP state machine 2416 * for the version in use on this link. 2417 * A link marked as IGIF_SILENT causes IGMP to be completely 2418 * disabled for the link. 2419 */ 2420 if ((ifp->if_flags & IFF_LOOPBACK) || 2421 (igi->igi_flags & IGIF_SILENT) || 2422 !igmp_isgroupreported(inm->inm_addr)) { 2423 CTR1(KTR_IGMPV3, 2424 "%s: not kicking state machine for silent group", __func__); 2425 inm->inm_state = IGMP_SILENT_MEMBER; 2426 inm->inm_timer = 0; 2427 } else { 2428 /* 2429 * Deal with overlapping in_multi lifecycle. 2430 * If this group was LEAVING, then make sure 2431 * we drop the reference we picked up to keep the 2432 * group around for the final INCLUDE {} enqueue. 2433 */ 2434 if (igi->igi_version == IGMP_VERSION_3 && 2435 inm->inm_state == IGMP_LEAVING_MEMBER) { 2436 MPASS(inm->inm_refcount > 1); 2437 inm_rele_locked(NULL, inm); 2438 } 2439 inm->inm_state = IGMP_REPORTING_MEMBER; 2440 2441 switch (igi->igi_version) { 2442 case IGMP_VERSION_1: 2443 case IGMP_VERSION_2: 2444 inm->inm_state = IGMP_IDLE_MEMBER; 2445 error = igmp_v1v2_queue_report(inm, 2446 (igi->igi_version == IGMP_VERSION_2) ? 2447 IGMP_v2_HOST_MEMBERSHIP_REPORT : 2448 IGMP_v1_HOST_MEMBERSHIP_REPORT); 2449 if (error == 0) { 2450 inm->inm_timer = IGMP_RANDOM_DELAY( 2451 IGMP_V1V2_MAX_RI * IGMP_FASTHZ); 2452 V_current_state_timers_running = 1; 2453 } 2454 break; 2455 2456 case IGMP_VERSION_3: 2457 /* 2458 * Defer update of T0 to T1, until the first copy 2459 * of the state change has been transmitted. 2460 */ 2461 syncstates = 0; 2462 2463 /* 2464 * Immediately enqueue a State-Change Report for 2465 * this interface, freeing any previous reports. 2466 * Don't kick the timers if there is nothing to do, 2467 * or if an error occurred. 2468 */ 2469 mq = &inm->inm_scq; 2470 mbufq_drain(mq); 2471 retval = igmp_v3_enqueue_group_record(mq, inm, 1, 2472 0, 0); 2473 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 2474 __func__, retval); 2475 if (retval <= 0) { 2476 error = retval * -1; 2477 break; 2478 } 2479 2480 /* 2481 * Schedule transmission of pending state-change 2482 * report up to RV times for this link. The timer 2483 * will fire at the next igmp_fasttimo (~200ms), 2484 * giving us an opportunity to merge the reports. 2485 */ 2486 if (igi->igi_flags & IGIF_LOOPBACK) { 2487 inm->inm_scrv = 1; 2488 } else { 2489 KASSERT(igi->igi_rv > 1, 2490 ("%s: invalid robustness %d", __func__, 2491 igi->igi_rv)); 2492 inm->inm_scrv = igi->igi_rv; 2493 } 2494 inm->inm_sctimer = 1; 2495 V_state_change_timers_running = 1; 2496 2497 error = 0; 2498 break; 2499 } 2500 } 2501 2502 /* 2503 * Only update the T0 state if state change is atomic, 2504 * i.e. we don't need to wait for a timer to fire before we 2505 * can consider the state change to have been communicated. 2506 */ 2507 if (syncstates) { 2508 inm_commit(inm); 2509 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 2510 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2511 } 2512 2513 return (error); 2514 } 2515 2516 /* 2517 * Issue an intermediate state change during the IGMP life-cycle. 2518 */ 2519 static int 2520 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi) 2521 { 2522 struct ifnet *ifp; 2523 int retval; 2524 2525 CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__, 2526 ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); 2527 2528 ifp = inm->inm_ifp; 2529 2530 IN_MULTI_LIST_LOCK_ASSERT(); 2531 IGMP_LOCK_ASSERT(); 2532 2533 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); 2534 2535 if ((ifp->if_flags & IFF_LOOPBACK) || 2536 (igi->igi_flags & IGIF_SILENT) || 2537 !igmp_isgroupreported(inm->inm_addr) || 2538 (igi->igi_version != IGMP_VERSION_3)) { 2539 if (!igmp_isgroupreported(inm->inm_addr)) { 2540 CTR1(KTR_IGMPV3, 2541 "%s: not kicking state machine for silent group", __func__); 2542 } 2543 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__); 2544 inm_commit(inm); 2545 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 2546 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2547 return (0); 2548 } 2549 2550 mbufq_drain(&inm->inm_scq); 2551 2552 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0); 2553 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); 2554 if (retval <= 0) 2555 return (-retval); 2556 2557 /* 2558 * If record(s) were enqueued, start the state-change 2559 * report timer for this group. 2560 */ 2561 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv); 2562 inm->inm_sctimer = 1; 2563 V_state_change_timers_running = 1; 2564 2565 return (0); 2566 } 2567 2568 /* 2569 * Perform the final leave for an IGMP group. 2570 * 2571 * When leaving a group: 2572 * IGMPv1 does nothing. 2573 * IGMPv2 sends a host leave message, if and only if we are the reporter. 2574 * IGMPv3 enqueues a state-change report containing a transition 2575 * to INCLUDE {} for immediate transmission. 2576 */ 2577 static void 2578 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi) 2579 { 2580 int syncstates; 2581 2582 syncstates = 1; 2583 2584 CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)", 2585 __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp, 2586 inm->inm_ifp->if_xname); 2587 2588 IN_MULTI_LIST_LOCK_ASSERT(); 2589 IGMP_LOCK_ASSERT(); 2590 2591 switch (inm->inm_state) { 2592 case IGMP_NOT_MEMBER: 2593 case IGMP_SILENT_MEMBER: 2594 case IGMP_LEAVING_MEMBER: 2595 /* Already leaving or left; do nothing. */ 2596 CTR1(KTR_IGMPV3, 2597 "%s: not kicking state machine for silent group", __func__); 2598 break; 2599 case IGMP_REPORTING_MEMBER: 2600 case IGMP_IDLE_MEMBER: 2601 case IGMP_G_QUERY_PENDING_MEMBER: 2602 case IGMP_SG_QUERY_PENDING_MEMBER: 2603 if (igi->igi_version == IGMP_VERSION_2) { 2604 #ifdef INVARIANTS 2605 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || 2606 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) 2607 panic("%s: IGMPv3 state reached, not IGMPv3 mode", 2608 __func__); 2609 #endif 2610 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE); 2611 inm->inm_state = IGMP_NOT_MEMBER; 2612 } else if (igi->igi_version == IGMP_VERSION_3) { 2613 /* 2614 * Stop group timer and all pending reports. 2615 * Immediately enqueue a state-change report 2616 * TO_IN {} to be sent on the next fast timeout, 2617 * giving us an opportunity to merge reports. 2618 */ 2619 mbufq_drain(&inm->inm_scq); 2620 inm->inm_timer = 0; 2621 if (igi->igi_flags & IGIF_LOOPBACK) { 2622 inm->inm_scrv = 1; 2623 } else { 2624 inm->inm_scrv = igi->igi_rv; 2625 } 2626 CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d " 2627 "pending retransmissions.", __func__, 2628 ntohl(inm->inm_addr.s_addr), 2629 inm->inm_ifp->if_xname, inm->inm_scrv); 2630 if (inm->inm_scrv == 0) { 2631 inm->inm_state = IGMP_NOT_MEMBER; 2632 inm->inm_sctimer = 0; 2633 } else { 2634 int retval __unused; 2635 2636 inm_acquire_locked(inm); 2637 2638 retval = igmp_v3_enqueue_group_record( 2639 &inm->inm_scq, inm, 1, 0, 0); 2640 KASSERT(retval != 0, 2641 ("%s: enqueue record = %d", __func__, 2642 retval)); 2643 2644 inm->inm_state = IGMP_LEAVING_MEMBER; 2645 inm->inm_sctimer = 1; 2646 V_state_change_timers_running = 1; 2647 syncstates = 0; 2648 } 2649 break; 2650 } 2651 break; 2652 case IGMP_LAZY_MEMBER: 2653 case IGMP_SLEEPING_MEMBER: 2654 case IGMP_AWAKENING_MEMBER: 2655 /* Our reports are suppressed; do nothing. */ 2656 break; 2657 } 2658 2659 if (syncstates) { 2660 inm_commit(inm); 2661 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, 2662 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2663 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 2664 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s", 2665 __func__, ntohl(inm->inm_addr.s_addr), 2666 inm->inm_ifp->if_xname); 2667 } 2668 } 2669 2670 /* 2671 * Enqueue an IGMPv3 group record to the given output queue. 2672 * 2673 * XXX This function could do with having the allocation code 2674 * split out, and the multiple-tree-walks coalesced into a single 2675 * routine as has been done in igmp_v3_enqueue_filter_change(). 2676 * 2677 * If is_state_change is zero, a current-state record is appended. 2678 * If is_state_change is non-zero, a state-change report is appended. 2679 * 2680 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2681 * If is_group_query is zero, and if there is a packet with free space 2682 * at the tail of the queue, it will be appended to providing there 2683 * is enough free space. 2684 * Otherwise a new mbuf packet chain is allocated. 2685 * 2686 * If is_source_query is non-zero, each source is checked to see if 2687 * it was recorded for a Group-Source query, and will be omitted if 2688 * it is not both in-mode and recorded. 2689 * 2690 * The function will attempt to allocate leading space in the packet 2691 * for the IP/IGMP header to be prepended without fragmenting the chain. 2692 * 2693 * If successful the size of all data appended to the queue is returned, 2694 * otherwise an error code less than zero is returned, or zero if 2695 * no record(s) were appended. 2696 */ 2697 static int 2698 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm, 2699 const int is_state_change, const int is_group_query, 2700 const int is_source_query) 2701 { 2702 struct igmp_grouprec ig; 2703 struct igmp_grouprec *pig; 2704 struct ifnet *ifp; 2705 struct ip_msource *ims, *nims; 2706 struct mbuf *m0, *m, *md; 2707 int is_filter_list_change; 2708 int minrec0len, m0srcs, msrcs, nbytes, off; 2709 int record_has_sources; 2710 int now; 2711 int type; 2712 in_addr_t naddr; 2713 uint8_t mode; 2714 2715 IN_MULTI_LIST_LOCK_ASSERT(); 2716 2717 ifp = inm->inm_ifp; 2718 is_filter_list_change = 0; 2719 m = NULL; 2720 m0 = NULL; 2721 m0srcs = 0; 2722 msrcs = 0; 2723 nbytes = 0; 2724 nims = NULL; 2725 record_has_sources = 1; 2726 pig = NULL; 2727 type = IGMP_DO_NOTHING; 2728 mode = inm->inm_st[1].iss_fmode; 2729 2730 /* 2731 * If we did not transition out of ASM mode during t0->t1, 2732 * and there are no source nodes to process, we can skip 2733 * the generation of source records. 2734 */ 2735 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 && 2736 inm->inm_nsrc == 0) 2737 record_has_sources = 0; 2738 2739 if (is_state_change) { 2740 /* 2741 * Queue a state change record. 2742 * If the mode did not change, and there are non-ASM 2743 * listeners or source filters present, 2744 * we potentially need to issue two records for the group. 2745 * If we are transitioning to MCAST_UNDEFINED, we need 2746 * not send any sources. 2747 * If there are ASM listeners, and there was no filter 2748 * mode transition of any kind, do nothing. 2749 */ 2750 if (mode != inm->inm_st[0].iss_fmode) { 2751 if (mode == MCAST_EXCLUDE) { 2752 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE", 2753 __func__); 2754 type = IGMP_CHANGE_TO_EXCLUDE_MODE; 2755 } else { 2756 CTR1(KTR_IGMPV3, "%s: change to INCLUDE", 2757 __func__); 2758 type = IGMP_CHANGE_TO_INCLUDE_MODE; 2759 if (mode == MCAST_UNDEFINED) 2760 record_has_sources = 0; 2761 } 2762 } else { 2763 if (record_has_sources) { 2764 is_filter_list_change = 1; 2765 } else { 2766 type = IGMP_DO_NOTHING; 2767 } 2768 } 2769 } else { 2770 /* 2771 * Queue a current state record. 2772 */ 2773 if (mode == MCAST_EXCLUDE) { 2774 type = IGMP_MODE_IS_EXCLUDE; 2775 } else if (mode == MCAST_INCLUDE) { 2776 type = IGMP_MODE_IS_INCLUDE; 2777 KASSERT(inm->inm_st[1].iss_asm == 0, 2778 ("%s: inm %p is INCLUDE but ASM count is %d", 2779 __func__, inm, inm->inm_st[1].iss_asm)); 2780 } 2781 } 2782 2783 /* 2784 * Generate the filter list changes using a separate function. 2785 */ 2786 if (is_filter_list_change) 2787 return (igmp_v3_enqueue_filter_change(mq, inm)); 2788 2789 if (type == IGMP_DO_NOTHING) { 2790 CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__, 2791 ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); 2792 return (0); 2793 } 2794 2795 /* 2796 * If any sources are present, we must be able to fit at least 2797 * one in the trailing space of the tail packet's mbuf, 2798 * ideally more. 2799 */ 2800 minrec0len = sizeof(struct igmp_grouprec); 2801 if (record_has_sources) 2802 minrec0len += sizeof(in_addr_t); 2803 2804 CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__, 2805 igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr), 2806 inm->inm_ifp->if_xname); 2807 2808 /* 2809 * Check if we have a packet in the tail of the queue for this 2810 * group into which the first group record for this group will fit. 2811 * Otherwise allocate a new packet. 2812 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT. 2813 * Note: Group records for G/GSR query responses MUST be sent 2814 * in their own packet. 2815 */ 2816 m0 = mbufq_last(mq); 2817 if (!is_group_query && 2818 m0 != NULL && 2819 (m0->m_pkthdr.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) && 2820 (m0->m_pkthdr.len + minrec0len) < 2821 (ifp->if_mtu - IGMP_LEADINGSPACE)) { 2822 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2823 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2824 m = m0; 2825 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__); 2826 } else { 2827 if (mbufq_full(mq)) { 2828 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); 2829 return (-ENOMEM); 2830 } 2831 m = NULL; 2832 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 2833 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2834 if (!is_state_change && !is_group_query) { 2835 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2836 if (m) 2837 m->m_data += IGMP_LEADINGSPACE; 2838 } 2839 if (m == NULL) { 2840 m = m_gethdr(M_NOWAIT, MT_DATA); 2841 if (m) 2842 M_ALIGN(m, IGMP_LEADINGSPACE); 2843 } 2844 if (m == NULL) 2845 return (-ENOMEM); 2846 2847 igmp_save_context(m, ifp); 2848 2849 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__); 2850 } 2851 2852 /* 2853 * Append group record. 2854 * If we have sources, we don't know how many yet. 2855 */ 2856 ig.ig_type = type; 2857 ig.ig_datalen = 0; 2858 ig.ig_numsrc = 0; 2859 ig.ig_group = inm->inm_addr; 2860 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { 2861 if (m != m0) 2862 m_freem(m); 2863 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); 2864 return (-ENOMEM); 2865 } 2866 nbytes += sizeof(struct igmp_grouprec); 2867 2868 /* 2869 * Append as many sources as will fit in the first packet. 2870 * If we are appending to a new packet, the chain allocation 2871 * may potentially use clusters; use m_getptr() in this case. 2872 * If we are appending to an existing packet, we need to obtain 2873 * a pointer to the group record after m_append(), in case a new 2874 * mbuf was allocated. 2875 * Only append sources which are in-mode at t1. If we are 2876 * transitioning to MCAST_UNDEFINED state on the group, do not 2877 * include source entries. 2878 * Only report recorded sources in our filter set when responding 2879 * to a group-source query. 2880 */ 2881 if (record_has_sources) { 2882 if (m == m0) { 2883 md = m_last(m); 2884 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + 2885 md->m_len - nbytes); 2886 } else { 2887 md = m_getptr(m, 0, &off); 2888 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + 2889 off); 2890 } 2891 msrcs = 0; 2892 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) { 2893 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2894 ims->ims_haddr); 2895 now = ims_get_mode(inm, ims, 1); 2896 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now); 2897 if ((now != mode) || 2898 (now == mode && mode == MCAST_UNDEFINED)) { 2899 CTR1(KTR_IGMPV3, "%s: skip node", __func__); 2900 continue; 2901 } 2902 if (is_source_query && ims->ims_stp == 0) { 2903 CTR1(KTR_IGMPV3, "%s: skip unrecorded node", 2904 __func__); 2905 continue; 2906 } 2907 CTR1(KTR_IGMPV3, "%s: append node", __func__); 2908 naddr = htonl(ims->ims_haddr); 2909 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { 2910 if (m != m0) 2911 m_freem(m); 2912 CTR1(KTR_IGMPV3, "%s: m_append() failed.", 2913 __func__); 2914 return (-ENOMEM); 2915 } 2916 nbytes += sizeof(in_addr_t); 2917 ++msrcs; 2918 if (msrcs == m0srcs) 2919 break; 2920 } 2921 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__, 2922 msrcs); 2923 pig->ig_numsrc = htons(msrcs); 2924 nbytes += (msrcs * sizeof(in_addr_t)); 2925 } 2926 2927 if (is_source_query && msrcs == 0) { 2928 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__); 2929 if (m != m0) 2930 m_freem(m); 2931 return (0); 2932 } 2933 2934 /* 2935 * We are good to go with first packet. 2936 */ 2937 if (m != m0) { 2938 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__); 2939 m->m_pkthdr.vt_nrecs = 1; 2940 mbufq_enqueue(mq, m); 2941 } else 2942 m->m_pkthdr.vt_nrecs++; 2943 2944 /* 2945 * No further work needed if no source list in packet(s). 2946 */ 2947 if (!record_has_sources) 2948 return (nbytes); 2949 2950 /* 2951 * Whilst sources remain to be announced, we need to allocate 2952 * a new packet and fill out as many sources as will fit. 2953 * Always try for a cluster first. 2954 */ 2955 while (nims != NULL) { 2956 if (mbufq_full(mq)) { 2957 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); 2958 return (-ENOMEM); 2959 } 2960 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2961 if (m) 2962 m->m_data += IGMP_LEADINGSPACE; 2963 if (m == NULL) { 2964 m = m_gethdr(M_NOWAIT, MT_DATA); 2965 if (m) 2966 M_ALIGN(m, IGMP_LEADINGSPACE); 2967 } 2968 if (m == NULL) 2969 return (-ENOMEM); 2970 igmp_save_context(m, ifp); 2971 md = m_getptr(m, 0, &off); 2972 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); 2973 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__); 2974 2975 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { 2976 if (m != m0) 2977 m_freem(m); 2978 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); 2979 return (-ENOMEM); 2980 } 2981 m->m_pkthdr.vt_nrecs = 1; 2982 nbytes += sizeof(struct igmp_grouprec); 2983 2984 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 2985 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2986 2987 msrcs = 0; 2988 RB_FOREACH_FROM(ims, ip_msource_tree, nims) { 2989 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2990 ims->ims_haddr); 2991 now = ims_get_mode(inm, ims, 1); 2992 if ((now != mode) || 2993 (now == mode && mode == MCAST_UNDEFINED)) { 2994 CTR1(KTR_IGMPV3, "%s: skip node", __func__); 2995 continue; 2996 } 2997 if (is_source_query && ims->ims_stp == 0) { 2998 CTR1(KTR_IGMPV3, "%s: skip unrecorded node", 2999 __func__); 3000 continue; 3001 } 3002 CTR1(KTR_IGMPV3, "%s: append node", __func__); 3003 naddr = htonl(ims->ims_haddr); 3004 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { 3005 if (m != m0) 3006 m_freem(m); 3007 CTR1(KTR_IGMPV3, "%s: m_append() failed.", 3008 __func__); 3009 return (-ENOMEM); 3010 } 3011 ++msrcs; 3012 if (msrcs == m0srcs) 3013 break; 3014 } 3015 pig->ig_numsrc = htons(msrcs); 3016 nbytes += (msrcs * sizeof(in_addr_t)); 3017 3018 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__); 3019 mbufq_enqueue(mq, m); 3020 } 3021 3022 return (nbytes); 3023 } 3024 3025 /* 3026 * Type used to mark record pass completion. 3027 * We exploit the fact we can cast to this easily from the 3028 * current filter modes on each ip_msource node. 3029 */ 3030 typedef enum { 3031 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 3032 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 3033 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 3034 REC_FULL = REC_ALLOW | REC_BLOCK 3035 } rectype_t; 3036 3037 /* 3038 * Enqueue an IGMPv3 filter list change to the given output queue. 3039 * 3040 * Source list filter state is held in an RB-tree. When the filter list 3041 * for a group is changed without changing its mode, we need to compute 3042 * the deltas between T0 and T1 for each source in the filter set, 3043 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 3044 * 3045 * As we may potentially queue two record types, and the entire R-B tree 3046 * needs to be walked at once, we break this out into its own function 3047 * so we can generate a tightly packed queue of packets. 3048 * 3049 * XXX This could be written to only use one tree walk, although that makes 3050 * serializing into the mbuf chains a bit harder. For now we do two walks 3051 * which makes things easier on us, and it may or may not be harder on 3052 * the L2 cache. 3053 * 3054 * If successful the size of all data appended to the queue is returned, 3055 * otherwise an error code less than zero is returned, or zero if 3056 * no record(s) were appended. 3057 */ 3058 static int 3059 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm) 3060 { 3061 static const int MINRECLEN = 3062 sizeof(struct igmp_grouprec) + sizeof(in_addr_t); 3063 struct ifnet *ifp; 3064 struct igmp_grouprec ig; 3065 struct igmp_grouprec *pig; 3066 struct ip_msource *ims, *nims; 3067 struct mbuf *m, *m0, *md; 3068 in_addr_t naddr; 3069 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 3070 #ifdef KTR 3071 int nallow, nblock; 3072 #endif 3073 uint8_t mode, now, then; 3074 rectype_t crt, drt, nrt; 3075 3076 IN_MULTI_LIST_LOCK_ASSERT(); 3077 3078 if (inm->inm_nsrc == 0 || 3079 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0)) 3080 return (0); 3081 3082 ifp = inm->inm_ifp; /* interface */ 3083 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */ 3084 crt = REC_NONE; /* current group record type */ 3085 drt = REC_NONE; /* mask of completed group record types */ 3086 nrt = REC_NONE; /* record type for current node */ 3087 m0srcs = 0; /* # source which will fit in current mbuf chain */ 3088 nbytes = 0; /* # of bytes appended to group's state-change queue */ 3089 npbytes = 0; /* # of bytes appended this packet */ 3090 rsrcs = 0; /* # sources encoded in current record */ 3091 schanged = 0; /* # nodes encoded in overall filter change */ 3092 #ifdef KTR 3093 nallow = 0; /* # of source entries in ALLOW_NEW */ 3094 nblock = 0; /* # of source entries in BLOCK_OLD */ 3095 #endif 3096 nims = NULL; /* next tree node pointer */ 3097 3098 /* 3099 * For each possible filter record mode. 3100 * The first kind of source we encounter tells us which 3101 * is the first kind of record we start appending. 3102 * If a node transitioned to UNDEFINED at t1, its mode is treated 3103 * as the inverse of the group's filter mode. 3104 */ 3105 while (drt != REC_FULL) { 3106 do { 3107 m0 = mbufq_last(mq); 3108 if (m0 != NULL && 3109 (m0->m_pkthdr.vt_nrecs + 1 <= 3110 IGMP_V3_REPORT_MAXRECS) && 3111 (m0->m_pkthdr.len + MINRECLEN) < 3112 (ifp->if_mtu - IGMP_LEADINGSPACE)) { 3113 m = m0; 3114 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 3115 sizeof(struct igmp_grouprec)) / 3116 sizeof(in_addr_t); 3117 CTR1(KTR_IGMPV3, 3118 "%s: use previous packet", __func__); 3119 } else { 3120 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 3121 if (m) 3122 m->m_data += IGMP_LEADINGSPACE; 3123 if (m == NULL) { 3124 m = m_gethdr(M_NOWAIT, MT_DATA); 3125 if (m) 3126 M_ALIGN(m, IGMP_LEADINGSPACE); 3127 } 3128 if (m == NULL) { 3129 CTR1(KTR_IGMPV3, 3130 "%s: m_get*() failed", __func__); 3131 return (-ENOMEM); 3132 } 3133 m->m_pkthdr.vt_nrecs = 0; 3134 igmp_save_context(m, ifp); 3135 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 3136 sizeof(struct igmp_grouprec)) / 3137 sizeof(in_addr_t); 3138 npbytes = 0; 3139 CTR1(KTR_IGMPV3, 3140 "%s: allocated new packet", __func__); 3141 } 3142 /* 3143 * Append the IGMP group record header to the 3144 * current packet's data area. 3145 * Recalculate pointer to free space for next 3146 * group record, in case m_append() allocated 3147 * a new mbuf or cluster. 3148 */ 3149 memset(&ig, 0, sizeof(ig)); 3150 ig.ig_group = inm->inm_addr; 3151 if (!m_append(m, sizeof(ig), (void *)&ig)) { 3152 if (m != m0) 3153 m_freem(m); 3154 CTR1(KTR_IGMPV3, 3155 "%s: m_append() failed", __func__); 3156 return (-ENOMEM); 3157 } 3158 npbytes += sizeof(struct igmp_grouprec); 3159 if (m != m0) { 3160 /* new packet; offset in c hain */ 3161 md = m_getptr(m, npbytes - 3162 sizeof(struct igmp_grouprec), &off); 3163 pig = (struct igmp_grouprec *)(mtod(md, 3164 uint8_t *) + off); 3165 } else { 3166 /* current packet; offset from last append */ 3167 md = m_last(m); 3168 pig = (struct igmp_grouprec *)(mtod(md, 3169 uint8_t *) + md->m_len - 3170 sizeof(struct igmp_grouprec)); 3171 } 3172 /* 3173 * Begin walking the tree for this record type 3174 * pass, or continue from where we left off 3175 * previously if we had to allocate a new packet. 3176 * Only report deltas in-mode at t1. 3177 * We need not report included sources as allowed 3178 * if we are in inclusive mode on the group, 3179 * however the converse is not true. 3180 */ 3181 rsrcs = 0; 3182 if (nims == NULL) 3183 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs); 3184 RB_FOREACH_FROM(ims, ip_msource_tree, nims) { 3185 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", 3186 __func__, ims->ims_haddr); 3187 now = ims_get_mode(inm, ims, 1); 3188 then = ims_get_mode(inm, ims, 0); 3189 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d", 3190 __func__, then, now); 3191 if (now == then) { 3192 CTR1(KTR_IGMPV3, 3193 "%s: skip unchanged", __func__); 3194 continue; 3195 } 3196 if (mode == MCAST_EXCLUDE && 3197 now == MCAST_INCLUDE) { 3198 CTR1(KTR_IGMPV3, 3199 "%s: skip IN src on EX group", 3200 __func__); 3201 continue; 3202 } 3203 nrt = (rectype_t)now; 3204 if (nrt == REC_NONE) 3205 nrt = (rectype_t)(~mode & REC_FULL); 3206 if (schanged++ == 0) { 3207 crt = nrt; 3208 } else if (crt != nrt) 3209 continue; 3210 naddr = htonl(ims->ims_haddr); 3211 if (!m_append(m, sizeof(in_addr_t), 3212 (void *)&naddr)) { 3213 if (m != m0) 3214 m_freem(m); 3215 CTR1(KTR_IGMPV3, 3216 "%s: m_append() failed", __func__); 3217 return (-ENOMEM); 3218 } 3219 #ifdef KTR 3220 nallow += !!(crt == REC_ALLOW); 3221 nblock += !!(crt == REC_BLOCK); 3222 #endif 3223 if (++rsrcs == m0srcs) 3224 break; 3225 } 3226 /* 3227 * If we did not append any tree nodes on this 3228 * pass, back out of allocations. 3229 */ 3230 if (rsrcs == 0) { 3231 npbytes -= sizeof(struct igmp_grouprec); 3232 if (m != m0) { 3233 CTR1(KTR_IGMPV3, 3234 "%s: m_free(m)", __func__); 3235 m_freem(m); 3236 } else { 3237 CTR1(KTR_IGMPV3, 3238 "%s: m_adj(m, -ig)", __func__); 3239 m_adj(m, -((int)sizeof( 3240 struct igmp_grouprec))); 3241 } 3242 continue; 3243 } 3244 npbytes += (rsrcs * sizeof(in_addr_t)); 3245 if (crt == REC_ALLOW) 3246 pig->ig_type = IGMP_ALLOW_NEW_SOURCES; 3247 else if (crt == REC_BLOCK) 3248 pig->ig_type = IGMP_BLOCK_OLD_SOURCES; 3249 pig->ig_numsrc = htons(rsrcs); 3250 /* 3251 * Count the new group record, and enqueue this 3252 * packet if it wasn't already queued. 3253 */ 3254 m->m_pkthdr.vt_nrecs++; 3255 if (m != m0) 3256 mbufq_enqueue(mq, m); 3257 nbytes += npbytes; 3258 } while (nims != NULL); 3259 drt |= crt; 3260 crt = (~crt & REC_FULL); 3261 } 3262 3263 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 3264 nallow, nblock); 3265 3266 return (nbytes); 3267 } 3268 3269 static int 3270 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq) 3271 { 3272 struct mbufq *gq; 3273 struct mbuf *m; /* pending state-change */ 3274 struct mbuf *m0; /* copy of pending state-change */ 3275 struct mbuf *mt; /* last state-change in packet */ 3276 int docopy, domerge; 3277 u_int recslen; 3278 3279 docopy = 0; 3280 domerge = 0; 3281 recslen = 0; 3282 3283 IN_MULTI_LIST_LOCK_ASSERT(); 3284 IGMP_LOCK_ASSERT(); 3285 3286 /* 3287 * If there are further pending retransmissions, make a writable 3288 * copy of each queued state-change message before merging. 3289 */ 3290 if (inm->inm_scrv > 0) 3291 docopy = 1; 3292 3293 gq = &inm->inm_scq; 3294 #ifdef KTR 3295 if (mbufq_first(gq) == NULL) { 3296 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty", 3297 __func__, inm); 3298 } 3299 #endif 3300 3301 m = mbufq_first(gq); 3302 while (m != NULL) { 3303 /* 3304 * Only merge the report into the current packet if 3305 * there is sufficient space to do so; an IGMPv3 report 3306 * packet may only contain 65,535 group records. 3307 * Always use a simple mbuf chain concatentation to do this, 3308 * as large state changes for single groups may have 3309 * allocated clusters. 3310 */ 3311 domerge = 0; 3312 mt = mbufq_last(scq); 3313 if (mt != NULL) { 3314 recslen = m_length(m, NULL); 3315 3316 if ((mt->m_pkthdr.vt_nrecs + 3317 m->m_pkthdr.vt_nrecs <= 3318 IGMP_V3_REPORT_MAXRECS) && 3319 (mt->m_pkthdr.len + recslen <= 3320 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE))) 3321 domerge = 1; 3322 } 3323 3324 if (!domerge && mbufq_full(gq)) { 3325 CTR2(KTR_IGMPV3, 3326 "%s: outbound queue full, skipping whole packet %p", 3327 __func__, m); 3328 mt = m->m_nextpkt; 3329 if (!docopy) 3330 m_freem(m); 3331 m = mt; 3332 continue; 3333 } 3334 3335 if (!docopy) { 3336 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m); 3337 m0 = mbufq_dequeue(gq); 3338 m = m0->m_nextpkt; 3339 } else { 3340 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m); 3341 m0 = m_dup(m, M_NOWAIT); 3342 if (m0 == NULL) 3343 return (ENOMEM); 3344 m0->m_nextpkt = NULL; 3345 m = m->m_nextpkt; 3346 } 3347 3348 if (!domerge) { 3349 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)", 3350 __func__, m0, scq); 3351 mbufq_enqueue(scq, m0); 3352 } else { 3353 struct mbuf *mtl; /* last mbuf of packet mt */ 3354 3355 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)", 3356 __func__, m0, mt); 3357 3358 mtl = m_last(mt); 3359 m0->m_flags &= ~M_PKTHDR; 3360 mt->m_pkthdr.len += recslen; 3361 mt->m_pkthdr.vt_nrecs += 3362 m0->m_pkthdr.vt_nrecs; 3363 3364 mtl->m_next = m0; 3365 } 3366 } 3367 3368 return (0); 3369 } 3370 3371 /* 3372 * Respond to a pending IGMPv3 General Query. 3373 */ 3374 static void 3375 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi) 3376 { 3377 struct ifmultiaddr *ifma; 3378 struct ifnet *ifp; 3379 struct in_multi *inm; 3380 int retval __unused, loop; 3381 3382 IN_MULTI_LIST_LOCK_ASSERT(); 3383 IGMP_LOCK_ASSERT(); 3384 NET_EPOCH_ASSERT(); 3385 3386 KASSERT(igi->igi_version == IGMP_VERSION_3, 3387 ("%s: called when version %d", __func__, igi->igi_version)); 3388 3389 /* 3390 * Check that there are some packets queued. If so, send them first. 3391 * For large number of groups the reply to general query can take 3392 * many packets, we should finish sending them before starting of 3393 * queuing the new reply. 3394 */ 3395 if (!mbufq_empty(&igi->igi_gq)) 3396 goto send; 3397 3398 ifp = igi->igi_ifp; 3399 3400 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3401 inm = inm_ifmultiaddr_get_inm(ifma); 3402 if (inm == NULL) 3403 continue; 3404 KASSERT(ifp == inm->inm_ifp, 3405 ("%s: inconsistent ifp", __func__)); 3406 3407 switch (inm->inm_state) { 3408 case IGMP_NOT_MEMBER: 3409 case IGMP_SILENT_MEMBER: 3410 break; 3411 case IGMP_REPORTING_MEMBER: 3412 case IGMP_IDLE_MEMBER: 3413 case IGMP_LAZY_MEMBER: 3414 case IGMP_SLEEPING_MEMBER: 3415 case IGMP_AWAKENING_MEMBER: 3416 inm->inm_state = IGMP_REPORTING_MEMBER; 3417 retval = igmp_v3_enqueue_group_record(&igi->igi_gq, 3418 inm, 0, 0, 0); 3419 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 3420 __func__, retval); 3421 break; 3422 case IGMP_G_QUERY_PENDING_MEMBER: 3423 case IGMP_SG_QUERY_PENDING_MEMBER: 3424 case IGMP_LEAVING_MEMBER: 3425 break; 3426 } 3427 } 3428 3429 send: 3430 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; 3431 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop); 3432 3433 /* 3434 * Slew transmission of bursts over 500ms intervals. 3435 */ 3436 if (mbufq_first(&igi->igi_gq) != NULL) { 3437 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY( 3438 IGMP_RESPONSE_BURST_INTERVAL); 3439 V_interface_timers_running = 1; 3440 } 3441 } 3442 3443 /* 3444 * Transmit the next pending IGMP message in the output queue. 3445 * 3446 * We get called from netisr_processqueue(). A mutex private to igmpoq 3447 * will be acquired and released around this routine. 3448 * 3449 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3450 * MRT: Nothing needs to be done, as IGMP traffic is always local to 3451 * a link and uses a link-scope multicast address. 3452 */ 3453 static void 3454 igmp_intr(struct mbuf *m) 3455 { 3456 struct ip_moptions imo; 3457 struct ifnet *ifp; 3458 struct mbuf *ipopts, *m0; 3459 int error; 3460 uint32_t ifindex; 3461 3462 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m); 3463 3464 /* 3465 * Set VNET image pointer from enqueued mbuf chain 3466 * before doing anything else. Whilst we use interface 3467 * indexes to guard against interface detach, they are 3468 * unique to each VIMAGE and must be retrieved. 3469 */ 3470 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr)); 3471 ifindex = igmp_restore_context(m); 3472 3473 /* 3474 * Check if the ifnet still exists. This limits the scope of 3475 * any race in the absence of a global ifp lock for low cost 3476 * (an array lookup). 3477 */ 3478 ifp = ifnet_byindex(ifindex); 3479 if (ifp == NULL) { 3480 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.", 3481 __func__, m, ifindex); 3482 m_freem(m); 3483 IPSTAT_INC(ips_noroute); 3484 goto out; 3485 } 3486 3487 ipopts = V_igmp_sendra ? m_raopt : NULL; 3488 3489 imo.imo_multicast_ttl = 1; 3490 imo.imo_multicast_vif = -1; 3491 imo.imo_multicast_loop = (V_ip_mrouter != NULL); 3492 3493 /* 3494 * If the user requested that IGMP traffic be explicitly 3495 * redirected to the loopback interface (e.g. they are running a 3496 * MANET interface and the routing protocol needs to see the 3497 * updates), handle this now. 3498 */ 3499 if (m->m_flags & M_IGMP_LOOP) 3500 imo.imo_multicast_ifp = V_loif; 3501 else 3502 imo.imo_multicast_ifp = ifp; 3503 3504 if (m->m_flags & M_IGMPV2) { 3505 m0 = m; 3506 } else { 3507 m0 = igmp_v3_encap_report(ifp, m); 3508 if (m0 == NULL) { 3509 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m); 3510 m_freem(m); 3511 IPSTAT_INC(ips_odropped); 3512 goto out; 3513 } 3514 } 3515 3516 igmp_scrub_context(m0); 3517 m_clrprotoflags(m); 3518 m0->m_pkthdr.rcvif = V_loif; 3519 #ifdef MAC 3520 mac_netinet_igmp_send(ifp, m0); 3521 #endif 3522 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL); 3523 if (error) { 3524 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error); 3525 goto out; 3526 } 3527 3528 IGMPSTAT_INC(igps_snd_reports); 3529 3530 out: 3531 /* 3532 * We must restore the existing vnet pointer before 3533 * continuing as we are run from netisr context. 3534 */ 3535 CURVNET_RESTORE(); 3536 } 3537 3538 /* 3539 * Encapsulate an IGMPv3 report. 3540 * 3541 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf 3542 * chain has already had its IP/IGMPv3 header prepended. In this case 3543 * the function will not attempt to prepend; the lengths and checksums 3544 * will however be re-computed. 3545 * 3546 * Returns a pointer to the new mbuf chain head, or NULL if the 3547 * allocation failed. 3548 */ 3549 static struct mbuf * 3550 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m) 3551 { 3552 struct igmp_report *igmp; 3553 struct ip *ip; 3554 int hdrlen, igmpreclen; 3555 3556 KASSERT((m->m_flags & M_PKTHDR), 3557 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3558 3559 igmpreclen = m_length(m, NULL); 3560 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report); 3561 3562 if (m->m_flags & M_IGMPV3_HDR) { 3563 igmpreclen -= hdrlen; 3564 } else { 3565 M_PREPEND(m, hdrlen, M_NOWAIT); 3566 if (m == NULL) 3567 return (NULL); 3568 m->m_flags |= M_IGMPV3_HDR; 3569 } 3570 3571 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen); 3572 3573 m->m_data += sizeof(struct ip); 3574 m->m_len -= sizeof(struct ip); 3575 3576 igmp = mtod(m, struct igmp_report *); 3577 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT; 3578 igmp->ir_rsv1 = 0; 3579 igmp->ir_rsv2 = 0; 3580 igmp->ir_numgrps = htons(m->m_pkthdr.vt_nrecs); 3581 igmp->ir_cksum = 0; 3582 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen); 3583 m->m_pkthdr.vt_nrecs = 0; 3584 3585 m->m_data -= sizeof(struct ip); 3586 m->m_len += sizeof(struct ip); 3587 3588 ip = mtod(m, struct ip *); 3589 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL; 3590 ip->ip_len = htons(hdrlen + igmpreclen); 3591 ip->ip_off = htons(IP_DF); 3592 ip->ip_p = IPPROTO_IGMP; 3593 ip->ip_sum = 0; 3594 3595 ip->ip_src.s_addr = INADDR_ANY; 3596 3597 if (m->m_flags & M_IGMP_LOOP) { 3598 struct in_ifaddr *ia; 3599 3600 IFP_TO_IA(ifp, ia); 3601 if (ia != NULL) 3602 ip->ip_src = ia->ia_addr.sin_addr; 3603 } 3604 3605 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP); 3606 3607 return (m); 3608 } 3609 3610 #ifdef KTR 3611 static char * 3612 igmp_rec_type_to_str(const int type) 3613 { 3614 3615 switch (type) { 3616 case IGMP_CHANGE_TO_EXCLUDE_MODE: 3617 return "TO_EX"; 3618 break; 3619 case IGMP_CHANGE_TO_INCLUDE_MODE: 3620 return "TO_IN"; 3621 break; 3622 case IGMP_MODE_IS_EXCLUDE: 3623 return "MODE_EX"; 3624 break; 3625 case IGMP_MODE_IS_INCLUDE: 3626 return "MODE_IN"; 3627 break; 3628 case IGMP_ALLOW_NEW_SOURCES: 3629 return "ALLOW_NEW"; 3630 break; 3631 case IGMP_BLOCK_OLD_SOURCES: 3632 return "BLOCK_OLD"; 3633 break; 3634 default: 3635 break; 3636 } 3637 return "unknown"; 3638 } 3639 #endif 3640 3641 #ifdef VIMAGE 3642 static void 3643 vnet_igmp_init(const void *unused __unused) 3644 { 3645 3646 netisr_register_vnet(&igmp_nh); 3647 } 3648 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, 3649 vnet_igmp_init, NULL); 3650 3651 static void 3652 vnet_igmp_uninit(const void *unused __unused) 3653 { 3654 3655 /* This can happen when we shutdown the entire network stack. */ 3656 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3657 3658 netisr_unregister_vnet(&igmp_nh); 3659 } 3660 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, 3661 vnet_igmp_uninit, NULL); 3662 #endif 3663 3664 #ifdef DDB 3665 DB_SHOW_COMMAND(igi_list, db_show_igi_list) 3666 { 3667 struct igmp_ifsoftc *igi, *tigi; 3668 LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head; 3669 3670 if (!have_addr) { 3671 db_printf("usage: show igi_list <addr>\n"); 3672 return; 3673 } 3674 igi_head = (struct _igi_list *)addr; 3675 3676 LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) { 3677 db_printf("igmp_ifsoftc %p:\n", igi); 3678 db_printf(" ifp %p\n", igi->igi_ifp); 3679 db_printf(" version %u\n", igi->igi_version); 3680 db_printf(" v1_timer %u\n", igi->igi_v1_timer); 3681 db_printf(" v2_timer %u\n", igi->igi_v2_timer); 3682 db_printf(" v3_timer %u\n", igi->igi_v3_timer); 3683 db_printf(" flags %#x\n", igi->igi_flags); 3684 db_printf(" rv %u\n", igi->igi_rv); 3685 db_printf(" qi %u\n", igi->igi_qi); 3686 db_printf(" qri %u\n", igi->igi_qri); 3687 db_printf(" uri %u\n", igi->igi_uri); 3688 /* struct mbufq igi_gq; */ 3689 db_printf("\n"); 3690 } 3691 } 3692 #endif 3693 3694 static int 3695 igmp_modevent(module_t mod, int type, void *unused __unused) 3696 { 3697 3698 switch (type) { 3699 case MOD_LOAD: 3700 CTR1(KTR_IGMPV3, "%s: initializing", __func__); 3701 IGMP_LOCK_INIT(); 3702 m_raopt = igmp_ra_alloc(); 3703 netisr_register(&igmp_nh); 3704 callout_init(&igmpslow_callout, 1); 3705 callout_reset(&igmpslow_callout, hz / IGMP_SLOWHZ, 3706 igmp_slowtimo, NULL); 3707 callout_init(&igmpfast_callout, 1); 3708 callout_reset(&igmpfast_callout, hz / IGMP_FASTHZ, 3709 igmp_fasttimo, NULL); 3710 break; 3711 case MOD_UNLOAD: 3712 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3713 netisr_unregister(&igmp_nh); 3714 m_free(m_raopt); 3715 m_raopt = NULL; 3716 IGMP_LOCK_DESTROY(); 3717 break; 3718 default: 3719 return (EOPNOTSUPP); 3720 } 3721 return (0); 3722 } 3723 3724 static moduledata_t igmp_mod = { 3725 "igmp", 3726 igmp_modevent, 3727 0 3728 }; 3729 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE); 3730