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