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