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/vimage.h> 63 #include <sys/ktr.h> 64 #include <sys/condvar.h> 65 66 #include <net/if.h> 67 #include <net/netisr.h> 68 #include <net/vnet.h> 69 70 #include <netinet/in.h> 71 #include <netinet/in_var.h> 72 #include <netinet/in_systm.h> 73 #include <netinet/ip.h> 74 #include <netinet/ip_var.h> 75 #include <netinet/ip_options.h> 76 #include <netinet/igmp.h> 77 #include <netinet/igmp_var.h> 78 79 #include <machine/in_cksum.h> 80 81 #include <security/mac/mac_framework.h> 82 83 #ifndef KTR_IGMPV3 84 #define KTR_IGMPV3 KTR_INET 85 #endif 86 87 static struct igmp_ifinfo * 88 igi_alloc_locked(struct ifnet *); 89 static void igi_delete_locked(const struct ifnet *); 90 static void igmp_dispatch_queue(struct ifqueue *, int, const int); 91 static void igmp_fasttimo_vnet(void); 92 static void igmp_final_leave(struct in_multi *, struct igmp_ifinfo *); 93 static int igmp_handle_state_change(struct in_multi *, 94 struct igmp_ifinfo *); 95 static int igmp_initial_join(struct in_multi *, struct igmp_ifinfo *); 96 static int igmp_input_v1_query(struct ifnet *, const struct ip *, 97 const struct igmp *); 98 static int igmp_input_v2_query(struct ifnet *, const struct ip *, 99 const struct igmp *); 100 static int igmp_input_v3_query(struct ifnet *, const struct ip *, 101 /*const*/ struct igmpv3 *); 102 static int igmp_input_v3_group_query(struct in_multi *, 103 struct igmp_ifinfo *, int, /*const*/ struct igmpv3 *); 104 static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *, 105 /*const*/ struct igmp *); 106 static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *, 107 /*const*/ struct igmp *); 108 static void igmp_intr(struct mbuf *); 109 static int igmp_isgroupreported(const struct in_addr); 110 static struct mbuf * 111 igmp_ra_alloc(void); 112 #ifdef KTR 113 static char * igmp_rec_type_to_str(const int); 114 #endif 115 static void igmp_set_version(struct igmp_ifinfo *, const int); 116 static void igmp_slowtimo_vnet(void); 117 static void igmp_sysinit(void); 118 static int igmp_v1v2_queue_report(struct in_multi *, const int); 119 static void igmp_v1v2_process_group_timer(struct in_multi *, const int); 120 static void igmp_v1v2_process_querier_timers(struct igmp_ifinfo *); 121 static void igmp_v2_update_group(struct in_multi *, const int); 122 static void igmp_v3_cancel_link_timers(struct igmp_ifinfo *); 123 static void igmp_v3_dispatch_general_query(struct igmp_ifinfo *); 124 static struct mbuf * 125 igmp_v3_encap_report(struct ifnet *, struct mbuf *); 126 static int igmp_v3_enqueue_group_record(struct ifqueue *, 127 struct in_multi *, const int, const int, const int); 128 static int igmp_v3_enqueue_filter_change(struct ifqueue *, 129 struct in_multi *); 130 static void igmp_v3_process_group_timers(struct igmp_ifinfo *, 131 struct ifqueue *, struct ifqueue *, struct in_multi *, 132 const int); 133 static int igmp_v3_merge_state_changes(struct in_multi *, 134 struct ifqueue *); 135 static void igmp_v3_suppress_group_record(struct in_multi *); 136 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS); 137 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS); 138 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS); 139 140 static vnet_attach_fn vnet_igmp_iattach; 141 static vnet_detach_fn vnet_igmp_idetach; 142 143 static const struct netisr_handler igmp_nh = { 144 .nh_name = "igmp", 145 .nh_handler = igmp_intr, 146 .nh_proto = NETISR_IGMP, 147 .nh_policy = NETISR_POLICY_SOURCE, 148 }; 149 150 /* 151 * System-wide globals. 152 * 153 * Unlocked access to these is OK, except for the global IGMP output 154 * queue. The IGMP subsystem lock ends up being system-wide for the moment, 155 * because all VIMAGEs have to share a global output queue, as netisrs 156 * themselves are not virtualized. 157 * 158 * Locking: 159 * * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 160 * Any may be taken independently; if any are held at the same 161 * time, the above lock order must be followed. 162 * * All output is delegated to the netisr. 163 * Now that Giant has been eliminated, the netisr may be inlined. 164 * * IN_MULTI_LOCK covers in_multi. 165 * * IGMP_LOCK covers igmp_ifinfo and any global variables in this file, 166 * including the output queue. 167 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 168 * per-link state iterators. 169 * * igmp_ifinfo is valid as long as PF_INET is attached to the interface, 170 * therefore it is not refcounted. 171 * We allow unlocked reads of igmp_ifinfo when accessed via in_multi. 172 * 173 * Reference counting 174 * * IGMP acquires its own reference every time an in_multi is passed to 175 * it and the group is being joined for the first time. 176 * * IGMP releases its reference(s) on in_multi in a deferred way, 177 * because the operations which process the release run as part of 178 * a loop whose control variables are directly affected by the release 179 * (that, and not recursing on the IF_ADDR_LOCK). 180 * 181 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds 182 * to a vnet in ifp->if_vnet. 183 * 184 * SMPng: XXX We may potentially race operations on ifma_protospec. 185 * The problem is that we currently lack a clean way of taking the 186 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing, 187 * as anything which modifies ifma needs to be covered by that lock. 188 * So check for ifma_protospec being NULL before proceeding. 189 */ 190 struct mtx igmp_mtx; 191 192 struct mbuf *m_raopt; /* Router Alert option */ 193 MALLOC_DEFINE(M_IGMP, "igmp", "igmp state"); 194 195 /* 196 * VIMAGE-wide globals. 197 * 198 * The IGMPv3 timers themselves need to run per-image, however, 199 * protosw timers run globally (see tcp). 200 * An ifnet can only be in one vimage at a time, and the loopback 201 * ifnet, loif, is itself virtualized. 202 * It would otherwise be possible to seriously hose IGMP state, 203 * and create inconsistencies in upstream multicast routing, if you have 204 * multiple VIMAGEs running on the same link joining different multicast 205 * groups, UNLESS the "primary IP address" is different. This is because 206 * IGMP for IPv4 does not force link-local addresses to be used for each 207 * node, unlike MLD for IPv6. 208 * Obviously the IGMPv3 per-interface state has per-vimage granularity 209 * also as a result. 210 * 211 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection 212 * policy to control the address used by IGMP on the link. 213 */ 214 static VNET_DEFINE(int, interface_timers_running); /* IGMPv3 general 215 * query response */ 216 static VNET_DEFINE(int, state_change_timers_running); /* IGMPv3 state-change 217 * retransmit */ 218 static VNET_DEFINE(int, current_state_timers_running); /* IGMPv1/v2 host 219 * report; IGMPv3 g/sg 220 * query response */ 221 222 #define V_interface_timers_running VNET_GET(interface_timers_running) 223 #define V_state_change_timers_running VNET_GET(state_change_timers_running) 224 #define V_current_state_timers_running VNET_GET(current_state_timers_running) 225 226 static VNET_DEFINE(LIST_HEAD(, igmp_ifinfo), igi_head); 227 static VNET_DEFINE(struct igmpstat, igmpstat); 228 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0}; 229 230 #define V_igi_head VNET_GET(igi_head) 231 #define V_igmpstat VNET_GET(igmpstat) 232 #define V_igmp_gsrdelay VNET_GET(igmp_gsrdelay) 233 234 static VNET_DEFINE(int, igmp_recvifkludge) = 1; 235 static VNET_DEFINE(int, igmp_sendra) = 1; 236 static VNET_DEFINE(int, igmp_sendlocal) = 1; 237 static VNET_DEFINE(int, igmp_v1enable) = 1; 238 static VNET_DEFINE(int, igmp_v2enable) = 1; 239 static VNET_DEFINE(int, igmp_legacysupp); 240 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3; 241 242 #define V_igmp_recvifkludge VNET_GET(igmp_recvifkludge) 243 #define V_igmp_sendra VNET_GET(igmp_sendra) 244 #define V_igmp_sendlocal VNET_GET(igmp_sendlocal) 245 #define V_igmp_v1enable VNET_GET(igmp_v1enable) 246 #define V_igmp_v2enable VNET_GET(igmp_v2enable) 247 #define V_igmp_legacysupp VNET_GET(igmp_legacysupp) 248 #define V_igmp_default_version VNET_GET(igmp_default_version) 249 250 /* 251 * Virtualized sysctls. 252 */ 253 SYSCTL_VNET_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_RW, 254 &VNET_NAME(igmpstat), igmpstat, ""); 255 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_RW, 256 &VNET_NAME(igmp_recvifkludge), 0, 257 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address"); 258 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_RW, 259 &VNET_NAME(igmp_sendra), 0, 260 "Send IP Router Alert option in IGMPv2/v3 messages"); 261 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_RW, 262 &VNET_NAME(igmp_sendlocal), 0, 263 "Send IGMP membership reports for 224.0.0.0/24 groups"); 264 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_RW, 265 &VNET_NAME(igmp_v1enable), 0, 266 "Enable backwards compatibility with IGMPv1"); 267 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_RW, 268 &VNET_NAME(igmp_v2enable), 0, 269 "Enable backwards compatibility with IGMPv2"); 270 SYSCTL_VNET_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_RW, 271 &VNET_NAME(igmp_legacysupp), 0, 272 "Allow v1/v2 reports to suppress v3 group responses"); 273 SYSCTL_VNET_PROC(_net_inet_igmp, OID_AUTO, default_version, 274 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 275 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I", 276 "Default version of IGMP to run on each interface"); 277 SYSCTL_VNET_PROC(_net_inet_igmp, OID_AUTO, gsrdelay, 278 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 279 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I", 280 "Rate limit for IGMPv3 Group-and-Source queries in seconds"); 281 282 /* 283 * Non-virtualized sysctls. 284 */ 285 SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE, 286 sysctl_igmp_ifinfo, "Per-interface IGMPv3 state"); 287 288 static __inline void 289 igmp_save_context(struct mbuf *m, struct ifnet *ifp) 290 { 291 292 #ifdef VIMAGE 293 m->m_pkthdr.header = ifp->if_vnet; 294 #endif /* VIMAGE */ 295 m->m_pkthdr.flowid = ifp->if_index; 296 } 297 298 static __inline void 299 igmp_scrub_context(struct mbuf *m) 300 { 301 302 m->m_pkthdr.header = NULL; 303 m->m_pkthdr.flowid = 0; 304 } 305 306 #ifdef KTR 307 static __inline char * 308 inet_ntoa_haddr(in_addr_t haddr) 309 { 310 struct in_addr ia; 311 312 ia.s_addr = htonl(haddr); 313 return (inet_ntoa(ia)); 314 } 315 #endif 316 317 /* 318 * Restore context from a queued IGMP output chain. 319 * Return saved ifindex. 320 * 321 * VIMAGE: The assertion is there to make sure that we 322 * actually called CURVNET_SET() with what's in the mbuf chain. 323 */ 324 static __inline uint32_t 325 igmp_restore_context(struct mbuf *m) 326 { 327 328 #ifdef notyet 329 #if defined(VIMAGE) && defined(INVARIANTS) 330 KASSERT(curvnet == (m->m_pkthdr.header), 331 ("%s: called when curvnet was not restored", __func__)); 332 #endif 333 #endif 334 return (m->m_pkthdr.flowid); 335 } 336 337 /* 338 * Retrieve or set default IGMP version. 339 * 340 * VIMAGE: Assume curvnet set by caller. 341 * SMPng: NOTE: Serialized by IGMP lock. 342 */ 343 static int 344 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS) 345 { 346 int error; 347 int new; 348 349 error = sysctl_wire_old_buffer(req, sizeof(int)); 350 if (error) 351 return (error); 352 353 IGMP_LOCK(); 354 355 new = V_igmp_default_version; 356 357 error = sysctl_handle_int(oidp, &new, 0, req); 358 if (error || !req->newptr) 359 goto out_locked; 360 361 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) { 362 error = EINVAL; 363 goto out_locked; 364 } 365 366 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d", 367 V_igmp_default_version, new); 368 369 V_igmp_default_version = new; 370 371 out_locked: 372 IGMP_UNLOCK(); 373 return (error); 374 } 375 376 /* 377 * Retrieve or set threshold between group-source queries in seconds. 378 * 379 * VIMAGE: Assume curvnet set by caller. 380 * SMPng: NOTE: Serialized by IGMP lock. 381 */ 382 static int 383 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS) 384 { 385 int error; 386 int i; 387 388 error = sysctl_wire_old_buffer(req, sizeof(int)); 389 if (error) 390 return (error); 391 392 IGMP_LOCK(); 393 394 i = V_igmp_gsrdelay.tv_sec; 395 396 error = sysctl_handle_int(oidp, &i, 0, req); 397 if (error || !req->newptr) 398 goto out_locked; 399 400 if (i < -1 || i >= 60) { 401 error = EINVAL; 402 goto out_locked; 403 } 404 405 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d", 406 V_igmp_gsrdelay.tv_sec, i); 407 V_igmp_gsrdelay.tv_sec = i; 408 409 out_locked: 410 IGMP_UNLOCK(); 411 return (error); 412 } 413 414 /* 415 * Expose struct igmp_ifinfo to userland, keyed by ifindex. 416 * For use by ifmcstat(8). 417 * 418 * SMPng: NOTE: Does an unlocked ifindex space read. 419 * VIMAGE: Assume curvnet set by caller. The node handler itself 420 * is not directly virtualized. 421 */ 422 static int 423 sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS) 424 { 425 int *name; 426 int error; 427 u_int namelen; 428 struct ifnet *ifp; 429 struct igmp_ifinfo *igi; 430 431 name = (int *)arg1; 432 namelen = arg2; 433 434 if (req->newptr != NULL) 435 return (EPERM); 436 437 if (namelen != 1) 438 return (EINVAL); 439 440 error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo)); 441 if (error) 442 return (error); 443 444 IN_MULTI_LOCK(); 445 IGMP_LOCK(); 446 447 if (name[0] <= 0 || name[0] > V_if_index) { 448 error = ENOENT; 449 goto out_locked; 450 } 451 452 error = ENOENT; 453 454 ifp = ifnet_byindex(name[0]); 455 if (ifp == NULL) 456 goto out_locked; 457 458 LIST_FOREACH(igi, &V_igi_head, igi_link) { 459 if (ifp == igi->igi_ifp) { 460 error = SYSCTL_OUT(req, igi, 461 sizeof(struct igmp_ifinfo)); 462 break; 463 } 464 } 465 466 out_locked: 467 IGMP_UNLOCK(); 468 IN_MULTI_UNLOCK(); 469 return (error); 470 } 471 472 /* 473 * Dispatch an entire queue of pending packet chains 474 * using the netisr. 475 * VIMAGE: Assumes the vnet pointer has been set. 476 */ 477 static void 478 igmp_dispatch_queue(struct ifqueue *ifq, int limit, const int loop) 479 { 480 struct mbuf *m; 481 482 for (;;) { 483 _IF_DEQUEUE(ifq, m); 484 if (m == NULL) 485 break; 486 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, ifq, m); 487 if (loop) 488 m->m_flags |= M_IGMP_LOOP; 489 netisr_dispatch(NETISR_IGMP, m); 490 if (--limit == 0) 491 break; 492 } 493 } 494 495 /* 496 * Filter outgoing IGMP report state by group. 497 * 498 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1). 499 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are 500 * disabled for all groups in the 224.0.0.0/24 link-local scope. However, 501 * this may break certain IGMP snooping switches which rely on the old 502 * report behaviour. 503 * 504 * Return zero if the given group is one for which IGMP reports 505 * should be suppressed, or non-zero if reports should be issued. 506 */ 507 static __inline int 508 igmp_isgroupreported(const struct in_addr addr) 509 { 510 511 if (in_allhosts(addr) || 512 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr))))) 513 return (0); 514 515 return (1); 516 } 517 518 /* 519 * Construct a Router Alert option to use in outgoing packets. 520 */ 521 static struct mbuf * 522 igmp_ra_alloc(void) 523 { 524 struct mbuf *m; 525 struct ipoption *p; 526 527 MGET(m, M_DONTWAIT, MT_DATA); 528 p = mtod(m, struct ipoption *); 529 p->ipopt_dst.s_addr = INADDR_ANY; 530 p->ipopt_list[0] = IPOPT_RA; /* Router Alert Option */ 531 p->ipopt_list[1] = 0x04; /* 4 bytes long */ 532 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */ 533 p->ipopt_list[3] = 0x00; /* pad byte */ 534 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1]; 535 536 return (m); 537 } 538 539 /* 540 * Attach IGMP when PF_INET is attached to an interface. 541 */ 542 struct igmp_ifinfo * 543 igmp_domifattach(struct ifnet *ifp) 544 { 545 struct igmp_ifinfo *igi; 546 547 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 548 __func__, ifp, ifp->if_xname); 549 550 IGMP_LOCK(); 551 552 igi = igi_alloc_locked(ifp); 553 if (!(ifp->if_flags & IFF_MULTICAST)) 554 igi->igi_flags |= IGIF_SILENT; 555 556 IGMP_UNLOCK(); 557 558 return (igi); 559 } 560 561 /* 562 * VIMAGE: assume curvnet set by caller. 563 */ 564 static struct igmp_ifinfo * 565 igi_alloc_locked(/*const*/ struct ifnet *ifp) 566 { 567 struct igmp_ifinfo *igi; 568 569 IGMP_LOCK_ASSERT(); 570 571 igi = malloc(sizeof(struct igmp_ifinfo), M_IGMP, M_NOWAIT|M_ZERO); 572 if (igi == NULL) 573 goto out; 574 575 igi->igi_ifp = ifp; 576 igi->igi_version = V_igmp_default_version; 577 igi->igi_flags = 0; 578 igi->igi_rv = IGMP_RV_INIT; 579 igi->igi_qi = IGMP_QI_INIT; 580 igi->igi_qri = IGMP_QRI_INIT; 581 igi->igi_uri = IGMP_URI_INIT; 582 583 SLIST_INIT(&igi->igi_relinmhead); 584 585 /* 586 * Responses to general queries are subject to bounds. 587 */ 588 IFQ_SET_MAXLEN(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS); 589 590 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link); 591 592 CTR2(KTR_IGMPV3, "allocate igmp_ifinfo for ifp %p(%s)", 593 ifp, ifp->if_xname); 594 595 out: 596 return (igi); 597 } 598 599 /* 600 * Hook for ifdetach. 601 * 602 * NOTE: Some finalization tasks need to run before the protocol domain 603 * is detached, but also before the link layer does its cleanup. 604 * 605 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK(). 606 * XXX This is also bitten by unlocked ifma_protospec access. 607 */ 608 void 609 igmp_ifdetach(struct ifnet *ifp) 610 { 611 struct igmp_ifinfo *igi; 612 struct ifmultiaddr *ifma; 613 struct in_multi *inm, *tinm; 614 615 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, 616 ifp->if_xname); 617 618 IGMP_LOCK(); 619 620 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 621 if (igi->igi_version == IGMP_VERSION_3) { 622 IF_ADDR_LOCK(ifp); 623 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 624 if (ifma->ifma_addr->sa_family != AF_INET || 625 ifma->ifma_protospec == NULL) 626 continue; 627 #if 0 628 KASSERT(ifma->ifma_protospec != NULL, 629 ("%s: ifma_protospec is NULL", __func__)); 630 #endif 631 inm = (struct in_multi *)ifma->ifma_protospec; 632 if (inm->inm_state == IGMP_LEAVING_MEMBER) { 633 SLIST_INSERT_HEAD(&igi->igi_relinmhead, 634 inm, inm_nrele); 635 } 636 inm_clear_recorded(inm); 637 } 638 IF_ADDR_UNLOCK(ifp); 639 /* 640 * Free the in_multi reference(s) for this IGMP lifecycle. 641 */ 642 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele, 643 tinm) { 644 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele); 645 inm_release_locked(inm); 646 } 647 } 648 649 IGMP_UNLOCK(); 650 } 651 652 /* 653 * Hook for domifdetach. 654 */ 655 void 656 igmp_domifdetach(struct ifnet *ifp) 657 { 658 struct igmp_ifinfo *igi; 659 660 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 661 __func__, ifp, ifp->if_xname); 662 663 IGMP_LOCK(); 664 665 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 666 igi_delete_locked(ifp); 667 668 IGMP_UNLOCK(); 669 } 670 671 static void 672 igi_delete_locked(const struct ifnet *ifp) 673 { 674 struct igmp_ifinfo *igi, *tigi; 675 676 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifinfo for ifp %p(%s)", 677 __func__, ifp, ifp->if_xname); 678 679 IGMP_LOCK_ASSERT(); 680 681 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) { 682 if (igi->igi_ifp == ifp) { 683 /* 684 * Free deferred General Query responses. 685 */ 686 _IF_DRAIN(&igi->igi_gq); 687 688 LIST_REMOVE(igi, igi_link); 689 690 KASSERT(SLIST_EMPTY(&igi->igi_relinmhead), 691 ("%s: there are dangling in_multi references", 692 __func__)); 693 694 free(igi, M_IGMP); 695 return; 696 } 697 } 698 699 #ifdef INVARIANTS 700 panic("%s: igmp_ifinfo not found for ifp %p\n", __func__, ifp); 701 #endif 702 } 703 704 /* 705 * Process a received IGMPv1 query. 706 * Return non-zero if the message should be dropped. 707 * 708 * VIMAGE: The curvnet pointer is derived from the input ifp. 709 */ 710 static int 711 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip, 712 const struct igmp *igmp) 713 { 714 struct ifmultiaddr *ifma; 715 struct igmp_ifinfo *igi; 716 struct in_multi *inm; 717 718 /* 719 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to 720 * 224.0.0.1. They are always treated as General Queries. 721 * igmp_group is always ignored. Do not drop it as a userland 722 * daemon may wish to see it. 723 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 724 */ 725 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) { 726 IGMPSTAT_INC(igps_rcv_badqueries); 727 return (0); 728 } 729 IGMPSTAT_INC(igps_rcv_gen_queries); 730 731 IN_MULTI_LOCK(); 732 IGMP_LOCK(); 733 734 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 735 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp)); 736 737 if (igi->igi_flags & IGIF_LOOPBACK) { 738 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)", 739 ifp, ifp->if_xname); 740 goto out_locked; 741 } 742 743 /* 744 * Switch to IGMPv1 host compatibility mode. 745 */ 746 igmp_set_version(igi, IGMP_VERSION_1); 747 748 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname); 749 750 /* 751 * Start the timers in all of our group records 752 * for the interface on which the query arrived, 753 * except those which are already running. 754 */ 755 IF_ADDR_LOCK(ifp); 756 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 757 if (ifma->ifma_addr->sa_family != AF_INET || 758 ifma->ifma_protospec == NULL) 759 continue; 760 inm = (struct in_multi *)ifma->ifma_protospec; 761 if (inm->inm_timer != 0) 762 continue; 763 switch (inm->inm_state) { 764 case IGMP_NOT_MEMBER: 765 case IGMP_SILENT_MEMBER: 766 break; 767 case IGMP_G_QUERY_PENDING_MEMBER: 768 case IGMP_SG_QUERY_PENDING_MEMBER: 769 case IGMP_REPORTING_MEMBER: 770 case IGMP_IDLE_MEMBER: 771 case IGMP_LAZY_MEMBER: 772 case IGMP_SLEEPING_MEMBER: 773 case IGMP_AWAKENING_MEMBER: 774 inm->inm_state = IGMP_REPORTING_MEMBER; 775 inm->inm_timer = IGMP_RANDOM_DELAY( 776 IGMP_V1V2_MAX_RI * PR_FASTHZ); 777 V_current_state_timers_running = 1; 778 break; 779 case IGMP_LEAVING_MEMBER: 780 break; 781 } 782 } 783 IF_ADDR_UNLOCK(ifp); 784 785 out_locked: 786 IGMP_UNLOCK(); 787 IN_MULTI_UNLOCK(); 788 789 return (0); 790 } 791 792 /* 793 * Process a received IGMPv2 general or group-specific query. 794 */ 795 static int 796 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip, 797 const struct igmp *igmp) 798 { 799 struct ifmultiaddr *ifma; 800 struct igmp_ifinfo *igi; 801 struct in_multi *inm; 802 int is_general_query; 803 uint16_t timer; 804 805 is_general_query = 0; 806 807 /* 808 * Validate address fields upfront. 809 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 810 */ 811 if (in_nullhost(igmp->igmp_group)) { 812 /* 813 * IGMPv2 General Query. 814 * If this was not sent to the all-hosts group, ignore it. 815 */ 816 if (!in_allhosts(ip->ip_dst)) 817 return (0); 818 IGMPSTAT_INC(igps_rcv_gen_queries); 819 is_general_query = 1; 820 } else { 821 /* IGMPv2 Group-Specific Query. */ 822 IGMPSTAT_INC(igps_rcv_group_queries); 823 } 824 825 IN_MULTI_LOCK(); 826 IGMP_LOCK(); 827 828 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 829 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp)); 830 831 if (igi->igi_flags & IGIF_LOOPBACK) { 832 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)", 833 ifp, ifp->if_xname); 834 goto out_locked; 835 } 836 837 /* 838 * Ignore v2 query if in v1 Compatibility Mode. 839 */ 840 if (igi->igi_version == IGMP_VERSION_1) 841 goto out_locked; 842 843 igmp_set_version(igi, IGMP_VERSION_2); 844 845 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE; 846 if (timer == 0) 847 timer = 1; 848 849 if (is_general_query) { 850 /* 851 * For each reporting group joined on this 852 * interface, kick the report timer. 853 */ 854 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)", 855 ifp, ifp->if_xname); 856 IF_ADDR_LOCK(ifp); 857 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 858 if (ifma->ifma_addr->sa_family != AF_INET || 859 ifma->ifma_protospec == NULL) 860 continue; 861 inm = (struct in_multi *)ifma->ifma_protospec; 862 igmp_v2_update_group(inm, timer); 863 } 864 IF_ADDR_UNLOCK(ifp); 865 } else { 866 /* 867 * Group-specific IGMPv2 query, we need only 868 * look up the single group to process it. 869 */ 870 inm = inm_lookup(ifp, igmp->igmp_group); 871 if (inm != NULL) { 872 CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)", 873 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 874 igmp_v2_update_group(inm, timer); 875 } 876 } 877 878 out_locked: 879 IGMP_UNLOCK(); 880 IN_MULTI_UNLOCK(); 881 882 return (0); 883 } 884 885 /* 886 * Update the report timer on a group in response to an IGMPv2 query. 887 * 888 * If we are becoming the reporting member for this group, start the timer. 889 * If we already are the reporting member for this group, and timer is 890 * below the threshold, reset it. 891 * 892 * We may be updating the group for the first time since we switched 893 * to IGMPv3. If we are, then we must clear any recorded source lists, 894 * and transition to REPORTING state; the group timer is overloaded 895 * for group and group-source query responses. 896 * 897 * Unlike IGMPv3, the delay per group should be jittered 898 * to avoid bursts of IGMPv2 reports. 899 */ 900 static void 901 igmp_v2_update_group(struct in_multi *inm, const int timer) 902 { 903 904 CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__, 905 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer); 906 907 IN_MULTI_LOCK_ASSERT(); 908 909 switch (inm->inm_state) { 910 case IGMP_NOT_MEMBER: 911 case IGMP_SILENT_MEMBER: 912 break; 913 case IGMP_REPORTING_MEMBER: 914 if (inm->inm_timer != 0 && 915 inm->inm_timer <= timer) { 916 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, " 917 "skipping.", __func__); 918 break; 919 } 920 /* FALLTHROUGH */ 921 case IGMP_SG_QUERY_PENDING_MEMBER: 922 case IGMP_G_QUERY_PENDING_MEMBER: 923 case IGMP_IDLE_MEMBER: 924 case IGMP_LAZY_MEMBER: 925 case IGMP_AWAKENING_MEMBER: 926 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__); 927 inm->inm_state = IGMP_REPORTING_MEMBER; 928 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 929 V_current_state_timers_running = 1; 930 break; 931 case IGMP_SLEEPING_MEMBER: 932 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__); 933 inm->inm_state = IGMP_AWAKENING_MEMBER; 934 break; 935 case IGMP_LEAVING_MEMBER: 936 break; 937 } 938 } 939 940 /* 941 * Process a received IGMPv3 general, group-specific or 942 * group-and-source-specific query. 943 * Assumes m has already been pulled up to the full IGMP message length. 944 * Return 0 if successful, otherwise an appropriate error code is returned. 945 */ 946 static int 947 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip, 948 /*const*/ struct igmpv3 *igmpv3) 949 { 950 struct igmp_ifinfo *igi; 951 struct in_multi *inm; 952 int is_general_query; 953 uint32_t maxresp, nsrc, qqi; 954 uint16_t timer; 955 uint8_t qrv; 956 957 is_general_query = 0; 958 959 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname); 960 961 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */ 962 if (maxresp >= 128) { 963 maxresp = IGMP_MANT(igmpv3->igmp_code) << 964 (IGMP_EXP(igmpv3->igmp_code) + 3); 965 } 966 967 /* 968 * Robustness must never be less than 2 for on-wire IGMPv3. 969 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make 970 * an exception for interfaces whose IGMPv3 state changes 971 * are redirected to loopback (e.g. MANET). 972 */ 973 qrv = IGMP_QRV(igmpv3->igmp_misc); 974 if (qrv < 2) { 975 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__, 976 qrv, IGMP_RV_INIT); 977 qrv = IGMP_RV_INIT; 978 } 979 980 qqi = igmpv3->igmp_qqi; 981 if (qqi >= 128) { 982 qqi = IGMP_MANT(igmpv3->igmp_qqi) << 983 (IGMP_EXP(igmpv3->igmp_qqi) + 3); 984 } 985 986 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE; 987 if (timer == 0) 988 timer = 1; 989 990 nsrc = ntohs(igmpv3->igmp_numsrc); 991 992 /* 993 * Validate address fields and versions upfront before 994 * accepting v3 query. 995 * XXX SMPng: Unlocked access to igmpstat counters here. 996 */ 997 if (in_nullhost(igmpv3->igmp_group)) { 998 /* 999 * IGMPv3 General Query. 1000 * 1001 * General Queries SHOULD be directed to 224.0.0.1. 1002 * A general query with a source list has undefined 1003 * behaviour; discard it. 1004 */ 1005 IGMPSTAT_INC(igps_rcv_gen_queries); 1006 if (!in_allhosts(ip->ip_dst) || nsrc > 0) { 1007 IGMPSTAT_INC(igps_rcv_badqueries); 1008 return (0); 1009 } 1010 is_general_query = 1; 1011 } else { 1012 /* Group or group-source specific query. */ 1013 if (nsrc == 0) 1014 IGMPSTAT_INC(igps_rcv_group_queries); 1015 else 1016 IGMPSTAT_INC(igps_rcv_gsr_queries); 1017 } 1018 1019 IN_MULTI_LOCK(); 1020 IGMP_LOCK(); 1021 1022 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 1023 KASSERT(igi != NULL, ("%s: no igmp_ifinfo for ifp %p", __func__, ifp)); 1024 1025 if (igi->igi_flags & IGIF_LOOPBACK) { 1026 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)", 1027 ifp, ifp->if_xname); 1028 goto out_locked; 1029 } 1030 1031 /* 1032 * Discard the v3 query if we're in Compatibility Mode. 1033 * The RFC is not obviously worded that hosts need to stay in 1034 * compatibility mode until the Old Version Querier Present 1035 * timer expires. 1036 */ 1037 if (igi->igi_version != IGMP_VERSION_3) { 1038 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)", 1039 igi->igi_version, ifp, ifp->if_xname); 1040 goto out_locked; 1041 } 1042 1043 igmp_set_version(igi, IGMP_VERSION_3); 1044 igi->igi_rv = qrv; 1045 igi->igi_qi = qqi; 1046 igi->igi_qri = maxresp; 1047 1048 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi, 1049 maxresp); 1050 1051 if (is_general_query) { 1052 /* 1053 * Schedule a current-state report on this ifp for 1054 * all groups, possibly containing source lists. 1055 * If there is a pending General Query response 1056 * scheduled earlier than the selected delay, do 1057 * not schedule any other reports. 1058 * Otherwise, reset the interface timer. 1059 */ 1060 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)", 1061 ifp, ifp->if_xname); 1062 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) { 1063 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer); 1064 V_interface_timers_running = 1; 1065 } 1066 } else { 1067 /* 1068 * Group-source-specific queries are throttled on 1069 * a per-group basis to defeat denial-of-service attempts. 1070 * Queries for groups we are not a member of on this 1071 * link are simply ignored. 1072 */ 1073 inm = inm_lookup(ifp, igmpv3->igmp_group); 1074 if (inm == NULL) 1075 goto out_locked; 1076 if (nsrc > 0) { 1077 if (!ratecheck(&inm->inm_lastgsrtv, 1078 &V_igmp_gsrdelay)) { 1079 CTR1(KTR_IGMPV3, "%s: GS query throttled.", 1080 __func__); 1081 IGMPSTAT_INC(igps_drop_gsr_queries); 1082 goto out_locked; 1083 } 1084 } 1085 CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)", 1086 inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname); 1087 /* 1088 * If there is a pending General Query response 1089 * scheduled sooner than the selected delay, no 1090 * further report need be scheduled. 1091 * Otherwise, prepare to respond to the 1092 * group-specific or group-and-source query. 1093 */ 1094 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) 1095 igmp_input_v3_group_query(inm, igi, timer, igmpv3); 1096 } 1097 1098 out_locked: 1099 IGMP_UNLOCK(); 1100 IN_MULTI_UNLOCK(); 1101 1102 return (0); 1103 } 1104 1105 /* 1106 * Process a recieved IGMPv3 group-specific or group-and-source-specific 1107 * query. 1108 * Return <0 if any error occured. Currently this is ignored. 1109 */ 1110 static int 1111 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifinfo *igi, 1112 int timer, /*const*/ struct igmpv3 *igmpv3) 1113 { 1114 int retval; 1115 uint16_t nsrc; 1116 1117 IN_MULTI_LOCK_ASSERT(); 1118 IGMP_LOCK_ASSERT(); 1119 1120 retval = 0; 1121 1122 switch (inm->inm_state) { 1123 case IGMP_NOT_MEMBER: 1124 case IGMP_SILENT_MEMBER: 1125 case IGMP_SLEEPING_MEMBER: 1126 case IGMP_LAZY_MEMBER: 1127 case IGMP_AWAKENING_MEMBER: 1128 case IGMP_IDLE_MEMBER: 1129 case IGMP_LEAVING_MEMBER: 1130 return (retval); 1131 break; 1132 case IGMP_REPORTING_MEMBER: 1133 case IGMP_G_QUERY_PENDING_MEMBER: 1134 case IGMP_SG_QUERY_PENDING_MEMBER: 1135 break; 1136 } 1137 1138 nsrc = ntohs(igmpv3->igmp_numsrc); 1139 1140 /* 1141 * Deal with group-specific queries upfront. 1142 * If any group query is already pending, purge any recorded 1143 * source-list state if it exists, and schedule a query response 1144 * for this group-specific query. 1145 */ 1146 if (nsrc == 0) { 1147 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || 1148 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) { 1149 inm_clear_recorded(inm); 1150 timer = min(inm->inm_timer, timer); 1151 } 1152 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER; 1153 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1154 V_current_state_timers_running = 1; 1155 return (retval); 1156 } 1157 1158 /* 1159 * Deal with the case where a group-and-source-specific query has 1160 * been received but a group-specific query is already pending. 1161 */ 1162 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) { 1163 timer = min(inm->inm_timer, timer); 1164 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1165 V_current_state_timers_running = 1; 1166 return (retval); 1167 } 1168 1169 /* 1170 * Finally, deal with the case where a group-and-source-specific 1171 * query has been received, where a response to a previous g-s-r 1172 * query exists, or none exists. 1173 * In this case, we need to parse the source-list which the Querier 1174 * has provided us with and check if we have any source list filter 1175 * entries at T1 for these sources. If we do not, there is no need 1176 * schedule a report and the query may be dropped. 1177 * If we do, we must record them and schedule a current-state 1178 * report for those sources. 1179 * FIXME: Handling source lists larger than 1 mbuf requires that 1180 * we pass the mbuf chain pointer down to this function, and use 1181 * m_getptr() to walk the chain. 1182 */ 1183 if (inm->inm_nsrc > 0) { 1184 const struct in_addr *ap; 1185 int i, nrecorded; 1186 1187 ap = (const struct in_addr *)(igmpv3 + 1); 1188 nrecorded = 0; 1189 for (i = 0; i < nsrc; i++, ap++) { 1190 retval = inm_record_source(inm, ap->s_addr); 1191 if (retval < 0) 1192 break; 1193 nrecorded += retval; 1194 } 1195 if (nrecorded > 0) { 1196 CTR1(KTR_IGMPV3, 1197 "%s: schedule response to SG query", __func__); 1198 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER; 1199 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1200 V_current_state_timers_running = 1; 1201 } 1202 } 1203 1204 return (retval); 1205 } 1206 1207 /* 1208 * Process a received IGMPv1 host membership report. 1209 * 1210 * NOTE: 0.0.0.0 workaround breaks const correctness. 1211 */ 1212 static int 1213 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1214 /*const*/ struct igmp *igmp) 1215 { 1216 struct in_ifaddr *ia; 1217 struct in_multi *inm; 1218 1219 IGMPSTAT_INC(igps_rcv_reports); 1220 1221 if (ifp->if_flags & IFF_LOOPBACK) 1222 return (0); 1223 1224 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr) || 1225 !in_hosteq(igmp->igmp_group, ip->ip_dst))) { 1226 IGMPSTAT_INC(igps_rcv_badreports); 1227 return (EINVAL); 1228 } 1229 1230 /* 1231 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1232 * Booting clients may use the source address 0.0.0.0. Some 1233 * IGMP daemons may not know how to use IP_RECVIF to determine 1234 * the interface upon which this message was received. 1235 * Replace 0.0.0.0 with the subnet address if told to do so. 1236 */ 1237 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1238 IFP_TO_IA(ifp, ia); 1239 if (ia != NULL) { 1240 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1241 ifa_free(&ia->ia_ifa); 1242 } 1243 } 1244 1245 CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)", 1246 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1247 1248 /* 1249 * IGMPv1 report suppression. 1250 * If we are a member of this group, and our membership should be 1251 * reported, stop our group timer and transition to the 'lazy' state. 1252 */ 1253 IN_MULTI_LOCK(); 1254 inm = inm_lookup(ifp, igmp->igmp_group); 1255 if (inm != NULL) { 1256 struct igmp_ifinfo *igi; 1257 1258 igi = inm->inm_igi; 1259 if (igi == NULL) { 1260 KASSERT(igi != NULL, 1261 ("%s: no igi for ifp %p", __func__, ifp)); 1262 goto out_locked; 1263 } 1264 1265 IGMPSTAT_INC(igps_rcv_ourreports); 1266 1267 /* 1268 * If we are in IGMPv3 host mode, do not allow the 1269 * other host's IGMPv1 report to suppress our reports 1270 * unless explicitly configured to do so. 1271 */ 1272 if (igi->igi_version == IGMP_VERSION_3) { 1273 if (V_igmp_legacysupp) 1274 igmp_v3_suppress_group_record(inm); 1275 goto out_locked; 1276 } 1277 1278 inm->inm_timer = 0; 1279 1280 switch (inm->inm_state) { 1281 case IGMP_NOT_MEMBER: 1282 case IGMP_SILENT_MEMBER: 1283 break; 1284 case IGMP_IDLE_MEMBER: 1285 case IGMP_LAZY_MEMBER: 1286 case IGMP_AWAKENING_MEMBER: 1287 CTR3(KTR_IGMPV3, 1288 "report suppressed for %s on ifp %p(%s)", 1289 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1290 case IGMP_SLEEPING_MEMBER: 1291 inm->inm_state = IGMP_SLEEPING_MEMBER; 1292 break; 1293 case IGMP_REPORTING_MEMBER: 1294 CTR3(KTR_IGMPV3, 1295 "report suppressed for %s on ifp %p(%s)", 1296 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1297 if (igi->igi_version == IGMP_VERSION_1) 1298 inm->inm_state = IGMP_LAZY_MEMBER; 1299 else if (igi->igi_version == IGMP_VERSION_2) 1300 inm->inm_state = IGMP_SLEEPING_MEMBER; 1301 break; 1302 case IGMP_G_QUERY_PENDING_MEMBER: 1303 case IGMP_SG_QUERY_PENDING_MEMBER: 1304 case IGMP_LEAVING_MEMBER: 1305 break; 1306 } 1307 } 1308 1309 out_locked: 1310 IN_MULTI_UNLOCK(); 1311 1312 return (0); 1313 } 1314 1315 /* 1316 * Process a received IGMPv2 host membership report. 1317 * 1318 * NOTE: 0.0.0.0 workaround breaks const correctness. 1319 */ 1320 static int 1321 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1322 /*const*/ struct igmp *igmp) 1323 { 1324 struct in_ifaddr *ia; 1325 struct in_multi *inm; 1326 1327 /* 1328 * Make sure we don't hear our own membership report. Fast 1329 * leave requires knowing that we are the only member of a 1330 * group. 1331 */ 1332 IFP_TO_IA(ifp, ia); 1333 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) { 1334 ifa_free(&ia->ia_ifa); 1335 return (0); 1336 } 1337 1338 IGMPSTAT_INC(igps_rcv_reports); 1339 1340 if (ifp->if_flags & IFF_LOOPBACK) { 1341 if (ia != NULL) 1342 ifa_free(&ia->ia_ifa); 1343 return (0); 1344 } 1345 1346 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || 1347 !in_hosteq(igmp->igmp_group, ip->ip_dst)) { 1348 if (ia != NULL) 1349 ifa_free(&ia->ia_ifa); 1350 IGMPSTAT_INC(igps_rcv_badreports); 1351 return (EINVAL); 1352 } 1353 1354 /* 1355 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1356 * Booting clients may use the source address 0.0.0.0. Some 1357 * IGMP daemons may not know how to use IP_RECVIF to determine 1358 * the interface upon which this message was received. 1359 * Replace 0.0.0.0 with the subnet address if told to do so. 1360 */ 1361 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1362 if (ia != NULL) 1363 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1364 } 1365 if (ia != NULL) 1366 ifa_free(&ia->ia_ifa); 1367 1368 CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)", 1369 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1370 1371 /* 1372 * IGMPv2 report suppression. 1373 * If we are a member of this group, and our membership should be 1374 * reported, and our group timer is pending or about to be reset, 1375 * stop our group timer by transitioning to the 'lazy' state. 1376 */ 1377 IN_MULTI_LOCK(); 1378 inm = inm_lookup(ifp, igmp->igmp_group); 1379 if (inm != NULL) { 1380 struct igmp_ifinfo *igi; 1381 1382 igi = inm->inm_igi; 1383 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp)); 1384 1385 IGMPSTAT_INC(igps_rcv_ourreports); 1386 1387 /* 1388 * If we are in IGMPv3 host mode, do not allow the 1389 * other host's IGMPv1 report to suppress our reports 1390 * unless explicitly configured to do so. 1391 */ 1392 if (igi->igi_version == IGMP_VERSION_3) { 1393 if (V_igmp_legacysupp) 1394 igmp_v3_suppress_group_record(inm); 1395 goto out_locked; 1396 } 1397 1398 inm->inm_timer = 0; 1399 1400 switch (inm->inm_state) { 1401 case IGMP_NOT_MEMBER: 1402 case IGMP_SILENT_MEMBER: 1403 case IGMP_SLEEPING_MEMBER: 1404 break; 1405 case IGMP_REPORTING_MEMBER: 1406 case IGMP_IDLE_MEMBER: 1407 case IGMP_AWAKENING_MEMBER: 1408 CTR3(KTR_IGMPV3, 1409 "report suppressed for %s on ifp %p(%s)", 1410 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1411 case IGMP_LAZY_MEMBER: 1412 inm->inm_state = IGMP_LAZY_MEMBER; 1413 break; 1414 case IGMP_G_QUERY_PENDING_MEMBER: 1415 case IGMP_SG_QUERY_PENDING_MEMBER: 1416 case IGMP_LEAVING_MEMBER: 1417 break; 1418 } 1419 } 1420 1421 out_locked: 1422 IN_MULTI_UNLOCK(); 1423 1424 return (0); 1425 } 1426 1427 void 1428 igmp_input(struct mbuf *m, int off) 1429 { 1430 int iphlen; 1431 struct ifnet *ifp; 1432 struct igmp *igmp; 1433 struct ip *ip; 1434 int igmplen; 1435 int minlen; 1436 int queryver; 1437 1438 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, m, off); 1439 1440 ifp = m->m_pkthdr.rcvif; 1441 1442 IGMPSTAT_INC(igps_rcv_total); 1443 1444 ip = mtod(m, struct ip *); 1445 iphlen = off; 1446 igmplen = ip->ip_len; 1447 1448 /* 1449 * Validate lengths. 1450 */ 1451 if (igmplen < IGMP_MINLEN) { 1452 IGMPSTAT_INC(igps_rcv_tooshort); 1453 m_freem(m); 1454 return; 1455 } 1456 1457 /* 1458 * Always pullup to the minimum size for v1/v2 or v3 1459 * to amortize calls to m_pullup(). 1460 */ 1461 minlen = iphlen; 1462 if (igmplen >= IGMP_V3_QUERY_MINLEN) 1463 minlen += IGMP_V3_QUERY_MINLEN; 1464 else 1465 minlen += IGMP_MINLEN; 1466 if ((m->m_flags & M_EXT || m->m_len < minlen) && 1467 (m = m_pullup(m, minlen)) == 0) { 1468 IGMPSTAT_INC(igps_rcv_tooshort); 1469 return; 1470 } 1471 ip = mtod(m, struct ip *); 1472 1473 if (ip->ip_ttl != 1) { 1474 IGMPSTAT_INC(igps_rcv_badttl); 1475 m_freem(m); 1476 return; 1477 } 1478 1479 /* 1480 * Validate checksum. 1481 */ 1482 m->m_data += iphlen; 1483 m->m_len -= iphlen; 1484 igmp = mtod(m, struct igmp *); 1485 if (in_cksum(m, igmplen)) { 1486 IGMPSTAT_INC(igps_rcv_badsum); 1487 m_freem(m); 1488 return; 1489 } 1490 m->m_data -= iphlen; 1491 m->m_len += iphlen; 1492 1493 switch (igmp->igmp_type) { 1494 case IGMP_HOST_MEMBERSHIP_QUERY: 1495 if (igmplen == IGMP_MINLEN) { 1496 if (igmp->igmp_code == 0) 1497 queryver = IGMP_VERSION_1; 1498 else 1499 queryver = IGMP_VERSION_2; 1500 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) { 1501 queryver = IGMP_VERSION_3; 1502 } else { 1503 IGMPSTAT_INC(igps_rcv_tooshort); 1504 m_freem(m); 1505 return; 1506 } 1507 1508 switch (queryver) { 1509 case IGMP_VERSION_1: 1510 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1511 if (!V_igmp_v1enable) 1512 break; 1513 if (igmp_input_v1_query(ifp, ip, igmp) != 0) { 1514 m_freem(m); 1515 return; 1516 } 1517 break; 1518 1519 case IGMP_VERSION_2: 1520 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1521 if (!V_igmp_v2enable) 1522 break; 1523 if (igmp_input_v2_query(ifp, ip, igmp) != 0) { 1524 m_freem(m); 1525 return; 1526 } 1527 break; 1528 1529 case IGMP_VERSION_3: { 1530 struct igmpv3 *igmpv3; 1531 uint16_t igmpv3len; 1532 uint16_t srclen; 1533 int nsrc; 1534 1535 IGMPSTAT_INC(igps_rcv_v3_queries); 1536 igmpv3 = (struct igmpv3 *)igmp; 1537 /* 1538 * Validate length based on source count. 1539 */ 1540 nsrc = ntohs(igmpv3->igmp_numsrc); 1541 srclen = sizeof(struct in_addr) * nsrc; 1542 if (nsrc * sizeof(in_addr_t) > srclen) { 1543 IGMPSTAT_INC(igps_rcv_tooshort); 1544 return; 1545 } 1546 /* 1547 * m_pullup() may modify m, so pullup in 1548 * this scope. 1549 */ 1550 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN + 1551 srclen; 1552 if ((m->m_flags & M_EXT || 1553 m->m_len < igmpv3len) && 1554 (m = m_pullup(m, igmpv3len)) == NULL) { 1555 IGMPSTAT_INC(igps_rcv_tooshort); 1556 return; 1557 } 1558 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *) 1559 + iphlen); 1560 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) { 1561 m_freem(m); 1562 return; 1563 } 1564 } 1565 break; 1566 } 1567 break; 1568 1569 case IGMP_v1_HOST_MEMBERSHIP_REPORT: 1570 if (!V_igmp_v1enable) 1571 break; 1572 if (igmp_input_v1_report(ifp, ip, igmp) != 0) { 1573 m_freem(m); 1574 return; 1575 } 1576 break; 1577 1578 case IGMP_v2_HOST_MEMBERSHIP_REPORT: 1579 if (!V_igmp_v2enable) 1580 break; 1581 if (!ip_checkrouteralert(m)) 1582 IGMPSTAT_INC(igps_rcv_nora); 1583 if (igmp_input_v2_report(ifp, ip, igmp) != 0) { 1584 m_freem(m); 1585 return; 1586 } 1587 break; 1588 1589 case IGMP_v3_HOST_MEMBERSHIP_REPORT: 1590 /* 1591 * Hosts do not need to process IGMPv3 membership reports, 1592 * as report suppression is no longer required. 1593 */ 1594 if (!ip_checkrouteralert(m)) 1595 IGMPSTAT_INC(igps_rcv_nora); 1596 break; 1597 1598 default: 1599 break; 1600 } 1601 1602 /* 1603 * Pass all valid IGMP packets up to any process(es) listening on a 1604 * raw IGMP socket. 1605 */ 1606 rip_input(m, off); 1607 } 1608 1609 1610 /* 1611 * Fast timeout handler (global). 1612 * VIMAGE: Timeout handlers are expected to service all vimages. 1613 */ 1614 void 1615 igmp_fasttimo(void) 1616 { 1617 VNET_ITERATOR_DECL(vnet_iter); 1618 1619 VNET_LIST_RLOCK(); 1620 VNET_FOREACH(vnet_iter) { 1621 CURVNET_SET(vnet_iter); 1622 igmp_fasttimo_vnet(); 1623 CURVNET_RESTORE(); 1624 } 1625 VNET_LIST_RUNLOCK(); 1626 } 1627 1628 /* 1629 * Fast timeout handler (per-vnet). 1630 * Sends are shuffled off to a netisr to deal with Giant. 1631 * 1632 * VIMAGE: Assume caller has set up our curvnet. 1633 */ 1634 static void 1635 igmp_fasttimo_vnet(void) 1636 { 1637 struct ifqueue scq; /* State-change packets */ 1638 struct ifqueue qrq; /* Query response packets */ 1639 struct ifnet *ifp; 1640 struct igmp_ifinfo *igi; 1641 struct ifmultiaddr *ifma, *tifma; 1642 struct in_multi *inm; 1643 int loop, uri_fasthz; 1644 1645 loop = 0; 1646 uri_fasthz = 0; 1647 1648 /* 1649 * Quick check to see if any work needs to be done, in order to 1650 * minimize the overhead of fasttimo processing. 1651 * SMPng: XXX Unlocked reads. 1652 */ 1653 if (!V_current_state_timers_running && 1654 !V_interface_timers_running && 1655 !V_state_change_timers_running) 1656 return; 1657 1658 IN_MULTI_LOCK(); 1659 IGMP_LOCK(); 1660 1661 /* 1662 * IGMPv3 General Query response timer processing. 1663 */ 1664 if (V_interface_timers_running) { 1665 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__); 1666 1667 V_interface_timers_running = 0; 1668 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1669 if (igi->igi_v3_timer == 0) { 1670 /* Do nothing. */ 1671 } else if (--igi->igi_v3_timer == 0) { 1672 igmp_v3_dispatch_general_query(igi); 1673 } else { 1674 V_interface_timers_running = 1; 1675 } 1676 } 1677 } 1678 1679 if (!V_current_state_timers_running && 1680 !V_state_change_timers_running) 1681 goto out_locked; 1682 1683 V_current_state_timers_running = 0; 1684 V_state_change_timers_running = 0; 1685 1686 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__); 1687 1688 /* 1689 * IGMPv1/v2/v3 host report and state-change timer processing. 1690 * Note: Processing a v3 group timer may remove a node. 1691 */ 1692 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1693 ifp = igi->igi_ifp; 1694 1695 if (igi->igi_version == IGMP_VERSION_3) { 1696 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; 1697 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri * 1698 PR_FASTHZ); 1699 1700 memset(&qrq, 0, sizeof(struct ifqueue)); 1701 IFQ_SET_MAXLEN(&qrq, IGMP_MAX_G_GS_PACKETS); 1702 1703 memset(&scq, 0, sizeof(struct ifqueue)); 1704 IFQ_SET_MAXLEN(&scq, IGMP_MAX_STATE_CHANGE_PACKETS); 1705 } 1706 1707 IF_ADDR_LOCK(ifp); 1708 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, 1709 tifma) { 1710 if (ifma->ifma_addr->sa_family != AF_INET || 1711 ifma->ifma_protospec == NULL) 1712 continue; 1713 inm = (struct in_multi *)ifma->ifma_protospec; 1714 switch (igi->igi_version) { 1715 case IGMP_VERSION_1: 1716 case IGMP_VERSION_2: 1717 igmp_v1v2_process_group_timer(inm, 1718 igi->igi_version); 1719 break; 1720 case IGMP_VERSION_3: 1721 igmp_v3_process_group_timers(igi, &qrq, 1722 &scq, inm, uri_fasthz); 1723 break; 1724 } 1725 } 1726 IF_ADDR_UNLOCK(ifp); 1727 1728 if (igi->igi_version == IGMP_VERSION_3) { 1729 struct in_multi *tinm; 1730 1731 igmp_dispatch_queue(&qrq, 0, loop); 1732 igmp_dispatch_queue(&scq, 0, loop); 1733 1734 /* 1735 * Free the in_multi reference(s) for this 1736 * IGMP lifecycle. 1737 */ 1738 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, 1739 inm_nrele, tinm) { 1740 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, 1741 inm_nrele); 1742 inm_release_locked(inm); 1743 } 1744 } 1745 } 1746 1747 out_locked: 1748 IGMP_UNLOCK(); 1749 IN_MULTI_UNLOCK(); 1750 } 1751 1752 /* 1753 * Update host report group timer for IGMPv1/v2. 1754 * Will update the global pending timer flags. 1755 */ 1756 static void 1757 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version) 1758 { 1759 int report_timer_expired; 1760 1761 IN_MULTI_LOCK_ASSERT(); 1762 IGMP_LOCK_ASSERT(); 1763 1764 if (inm->inm_timer == 0) { 1765 report_timer_expired = 0; 1766 } else if (--inm->inm_timer == 0) { 1767 report_timer_expired = 1; 1768 } else { 1769 V_current_state_timers_running = 1; 1770 return; 1771 } 1772 1773 switch (inm->inm_state) { 1774 case IGMP_NOT_MEMBER: 1775 case IGMP_SILENT_MEMBER: 1776 case IGMP_IDLE_MEMBER: 1777 case IGMP_LAZY_MEMBER: 1778 case IGMP_SLEEPING_MEMBER: 1779 case IGMP_AWAKENING_MEMBER: 1780 break; 1781 case IGMP_REPORTING_MEMBER: 1782 if (report_timer_expired) { 1783 inm->inm_state = IGMP_IDLE_MEMBER; 1784 (void)igmp_v1v2_queue_report(inm, 1785 (version == IGMP_VERSION_2) ? 1786 IGMP_v2_HOST_MEMBERSHIP_REPORT : 1787 IGMP_v1_HOST_MEMBERSHIP_REPORT); 1788 } 1789 break; 1790 case IGMP_G_QUERY_PENDING_MEMBER: 1791 case IGMP_SG_QUERY_PENDING_MEMBER: 1792 case IGMP_LEAVING_MEMBER: 1793 break; 1794 } 1795 } 1796 1797 /* 1798 * Update a group's timers for IGMPv3. 1799 * Will update the global pending timer flags. 1800 * Note: Unlocked read from igi. 1801 */ 1802 static void 1803 igmp_v3_process_group_timers(struct igmp_ifinfo *igi, 1804 struct ifqueue *qrq, struct ifqueue *scq, 1805 struct in_multi *inm, const int uri_fasthz) 1806 { 1807 int query_response_timer_expired; 1808 int state_change_retransmit_timer_expired; 1809 1810 IN_MULTI_LOCK_ASSERT(); 1811 IGMP_LOCK_ASSERT(); 1812 1813 query_response_timer_expired = 0; 1814 state_change_retransmit_timer_expired = 0; 1815 1816 /* 1817 * During a transition from v1/v2 compatibility mode back to v3, 1818 * a group record in REPORTING state may still have its group 1819 * timer active. This is a no-op in this function; it is easier 1820 * to deal with it here than to complicate the slow-timeout path. 1821 */ 1822 if (inm->inm_timer == 0) { 1823 query_response_timer_expired = 0; 1824 } else if (--inm->inm_timer == 0) { 1825 query_response_timer_expired = 1; 1826 } else { 1827 V_current_state_timers_running = 1; 1828 } 1829 1830 if (inm->inm_sctimer == 0) { 1831 state_change_retransmit_timer_expired = 0; 1832 } else if (--inm->inm_sctimer == 0) { 1833 state_change_retransmit_timer_expired = 1; 1834 } else { 1835 V_state_change_timers_running = 1; 1836 } 1837 1838 /* We are in fasttimo, so be quick about it. */ 1839 if (!state_change_retransmit_timer_expired && 1840 !query_response_timer_expired) 1841 return; 1842 1843 switch (inm->inm_state) { 1844 case IGMP_NOT_MEMBER: 1845 case IGMP_SILENT_MEMBER: 1846 case IGMP_SLEEPING_MEMBER: 1847 case IGMP_LAZY_MEMBER: 1848 case IGMP_AWAKENING_MEMBER: 1849 case IGMP_IDLE_MEMBER: 1850 break; 1851 case IGMP_G_QUERY_PENDING_MEMBER: 1852 case IGMP_SG_QUERY_PENDING_MEMBER: 1853 /* 1854 * Respond to a previously pending Group-Specific 1855 * or Group-and-Source-Specific query by enqueueing 1856 * the appropriate Current-State report for 1857 * immediate transmission. 1858 */ 1859 if (query_response_timer_expired) { 1860 int retval; 1861 1862 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1, 1863 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)); 1864 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 1865 __func__, retval); 1866 inm->inm_state = IGMP_REPORTING_MEMBER; 1867 /* XXX Clear recorded sources for next time. */ 1868 inm_clear_recorded(inm); 1869 } 1870 /* FALLTHROUGH */ 1871 case IGMP_REPORTING_MEMBER: 1872 case IGMP_LEAVING_MEMBER: 1873 if (state_change_retransmit_timer_expired) { 1874 /* 1875 * State-change retransmission timer fired. 1876 * If there are any further pending retransmissions, 1877 * set the global pending state-change flag, and 1878 * reset the timer. 1879 */ 1880 if (--inm->inm_scrv > 0) { 1881 inm->inm_sctimer = uri_fasthz; 1882 V_state_change_timers_running = 1; 1883 } 1884 /* 1885 * Retransmit the previously computed state-change 1886 * report. If there are no further pending 1887 * retransmissions, the mbuf queue will be consumed. 1888 * Update T0 state to T1 as we have now sent 1889 * a state-change. 1890 */ 1891 (void)igmp_v3_merge_state_changes(inm, scq); 1892 1893 inm_commit(inm); 1894 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__, 1895 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname); 1896 1897 /* 1898 * If we are leaving the group for good, make sure 1899 * we release IGMP's reference to it. 1900 * This release must be deferred using a SLIST, 1901 * as we are called from a loop which traverses 1902 * the in_ifmultiaddr TAILQ. 1903 */ 1904 if (inm->inm_state == IGMP_LEAVING_MEMBER && 1905 inm->inm_scrv == 0) { 1906 inm->inm_state = IGMP_NOT_MEMBER; 1907 SLIST_INSERT_HEAD(&igi->igi_relinmhead, 1908 inm, inm_nrele); 1909 } 1910 } 1911 break; 1912 } 1913 } 1914 1915 1916 /* 1917 * Suppress a group's pending response to a group or source/group query. 1918 * 1919 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency. 1920 * Do NOT update ST1/ST0 as this operation merely suppresses 1921 * the currently pending group record. 1922 * Do NOT suppress the response to a general query. It is possible but 1923 * it would require adding another state or flag. 1924 */ 1925 static void 1926 igmp_v3_suppress_group_record(struct in_multi *inm) 1927 { 1928 1929 IN_MULTI_LOCK_ASSERT(); 1930 1931 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3, 1932 ("%s: not IGMPv3 mode on link", __func__)); 1933 1934 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER || 1935 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER) 1936 return; 1937 1938 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) 1939 inm_clear_recorded(inm); 1940 1941 inm->inm_timer = 0; 1942 inm->inm_state = IGMP_REPORTING_MEMBER; 1943 } 1944 1945 /* 1946 * Switch to a different IGMP version on the given interface, 1947 * as per Section 7.2.1. 1948 */ 1949 static void 1950 igmp_set_version(struct igmp_ifinfo *igi, const int version) 1951 { 1952 int old_version_timer; 1953 1954 IGMP_LOCK_ASSERT(); 1955 1956 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__, 1957 version, igi->igi_ifp, igi->igi_ifp->if_xname); 1958 1959 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) { 1960 /* 1961 * Compute the "Older Version Querier Present" timer as per 1962 * Section 8.12. 1963 */ 1964 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri; 1965 old_version_timer *= PR_SLOWHZ; 1966 1967 if (version == IGMP_VERSION_1) { 1968 igi->igi_v1_timer = old_version_timer; 1969 igi->igi_v2_timer = 0; 1970 } else if (version == IGMP_VERSION_2) { 1971 igi->igi_v1_timer = 0; 1972 igi->igi_v2_timer = old_version_timer; 1973 } 1974 } 1975 1976 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 1977 if (igi->igi_version != IGMP_VERSION_2) { 1978 igi->igi_version = IGMP_VERSION_2; 1979 igmp_v3_cancel_link_timers(igi); 1980 } 1981 } else if (igi->igi_v1_timer > 0) { 1982 if (igi->igi_version != IGMP_VERSION_1) { 1983 igi->igi_version = IGMP_VERSION_1; 1984 igmp_v3_cancel_link_timers(igi); 1985 } 1986 } 1987 } 1988 1989 /* 1990 * Cancel pending IGMPv3 timers for the given link and all groups 1991 * joined on it; state-change, general-query, and group-query timers. 1992 * 1993 * Only ever called on a transition from v3 to Compatibility mode. Kill 1994 * the timers stone dead (this may be expensive for large N groups), they 1995 * will be restarted if Compatibility Mode deems that they must be due to 1996 * query processing. 1997 */ 1998 static void 1999 igmp_v3_cancel_link_timers(struct igmp_ifinfo *igi) 2000 { 2001 struct ifmultiaddr *ifma; 2002 struct ifnet *ifp; 2003 struct in_multi *inm; 2004 2005 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__, 2006 igi->igi_ifp, igi->igi_ifp->if_xname); 2007 2008 IN_MULTI_LOCK_ASSERT(); 2009 IGMP_LOCK_ASSERT(); 2010 2011 /* 2012 * Stop the v3 General Query Response on this link stone dead. 2013 * If fasttimo is woken up due to V_interface_timers_running, 2014 * the flag will be cleared if there are no pending link timers. 2015 */ 2016 igi->igi_v3_timer = 0; 2017 2018 /* 2019 * Now clear the current-state and state-change report timers 2020 * for all memberships scoped to this link. 2021 */ 2022 ifp = igi->igi_ifp; 2023 IF_ADDR_LOCK(ifp); 2024 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2025 if (ifma->ifma_addr->sa_family != AF_INET || 2026 ifma->ifma_protospec == NULL) 2027 continue; 2028 inm = (struct in_multi *)ifma->ifma_protospec; 2029 switch (inm->inm_state) { 2030 case IGMP_NOT_MEMBER: 2031 case IGMP_SILENT_MEMBER: 2032 case IGMP_IDLE_MEMBER: 2033 case IGMP_LAZY_MEMBER: 2034 case IGMP_SLEEPING_MEMBER: 2035 case IGMP_AWAKENING_MEMBER: 2036 /* 2037 * These states are either not relevant in v3 mode, 2038 * or are unreported. Do nothing. 2039 */ 2040 break; 2041 case IGMP_LEAVING_MEMBER: 2042 /* 2043 * If we are leaving the group and switching to 2044 * compatibility mode, we need to release the final 2045 * reference held for issuing the INCLUDE {}, and 2046 * transition to REPORTING to ensure the host leave 2047 * message is sent upstream to the old querier -- 2048 * transition to NOT would lose the leave and race. 2049 * 2050 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK 2051 * around inm_release_locked(), as it is not 2052 * a recursive mutex. 2053 */ 2054 IF_ADDR_UNLOCK(ifp); 2055 inm_release_locked(inm); 2056 IF_ADDR_LOCK(ifp); 2057 /* FALLTHROUGH */ 2058 case IGMP_G_QUERY_PENDING_MEMBER: 2059 case IGMP_SG_QUERY_PENDING_MEMBER: 2060 inm_clear_recorded(inm); 2061 /* FALLTHROUGH */ 2062 case IGMP_REPORTING_MEMBER: 2063 inm->inm_state = IGMP_REPORTING_MEMBER; 2064 break; 2065 } 2066 /* 2067 * Always clear state-change and group report timers. 2068 * Free any pending IGMPv3 state-change records. 2069 */ 2070 inm->inm_sctimer = 0; 2071 inm->inm_timer = 0; 2072 _IF_DRAIN(&inm->inm_scq); 2073 } 2074 IF_ADDR_UNLOCK(ifp); 2075 } 2076 2077 /* 2078 * Update the Older Version Querier Present timers for a link. 2079 * See Section 7.2.1 of RFC 3376. 2080 */ 2081 static void 2082 igmp_v1v2_process_querier_timers(struct igmp_ifinfo *igi) 2083 { 2084 2085 IGMP_LOCK_ASSERT(); 2086 2087 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) { 2088 /* 2089 * IGMPv1 and IGMPv2 Querier Present timers expired. 2090 * 2091 * Revert to IGMPv3. 2092 */ 2093 if (igi->igi_version != IGMP_VERSION_3) { 2094 CTR5(KTR_IGMPV3, 2095 "%s: transition from v%d -> v%d on %p(%s)", 2096 __func__, igi->igi_version, IGMP_VERSION_3, 2097 igi->igi_ifp, igi->igi_ifp->if_xname); 2098 igi->igi_version = IGMP_VERSION_3; 2099 } 2100 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 2101 /* 2102 * IGMPv1 Querier Present timer expired, 2103 * IGMPv2 Querier Present timer running. 2104 * If IGMPv2 was disabled since last timeout, 2105 * revert to IGMPv3. 2106 * If IGMPv2 is enabled, revert to IGMPv2. 2107 */ 2108 if (!V_igmp_v2enable) { 2109 CTR5(KTR_IGMPV3, 2110 "%s: transition from v%d -> v%d on %p(%s)", 2111 __func__, igi->igi_version, IGMP_VERSION_3, 2112 igi->igi_ifp, igi->igi_ifp->if_xname); 2113 igi->igi_v2_timer = 0; 2114 igi->igi_version = IGMP_VERSION_3; 2115 } else { 2116 --igi->igi_v2_timer; 2117 if (igi->igi_version != IGMP_VERSION_2) { 2118 CTR5(KTR_IGMPV3, 2119 "%s: transition from v%d -> v%d on %p(%s)", 2120 __func__, igi->igi_version, IGMP_VERSION_2, 2121 igi->igi_ifp, igi->igi_ifp->if_xname); 2122 igi->igi_version = IGMP_VERSION_2; 2123 } 2124 } 2125 } else if (igi->igi_v1_timer > 0) { 2126 /* 2127 * IGMPv1 Querier Present timer running. 2128 * Stop IGMPv2 timer if running. 2129 * 2130 * If IGMPv1 was disabled since last timeout, 2131 * revert to IGMPv3. 2132 * If IGMPv1 is enabled, reset IGMPv2 timer if running. 2133 */ 2134 if (!V_igmp_v1enable) { 2135 CTR5(KTR_IGMPV3, 2136 "%s: transition from v%d -> v%d on %p(%s)", 2137 __func__, igi->igi_version, IGMP_VERSION_3, 2138 igi->igi_ifp, igi->igi_ifp->if_xname); 2139 igi->igi_v1_timer = 0; 2140 igi->igi_version = IGMP_VERSION_3; 2141 } else { 2142 --igi->igi_v1_timer; 2143 } 2144 if (igi->igi_v2_timer > 0) { 2145 CTR3(KTR_IGMPV3, 2146 "%s: cancel v2 timer on %p(%s)", 2147 __func__, igi->igi_ifp, igi->igi_ifp->if_xname); 2148 igi->igi_v2_timer = 0; 2149 } 2150 } 2151 } 2152 2153 /* 2154 * Global slowtimo handler. 2155 * VIMAGE: Timeout handlers are expected to service all vimages. 2156 */ 2157 void 2158 igmp_slowtimo(void) 2159 { 2160 VNET_ITERATOR_DECL(vnet_iter); 2161 2162 VNET_LIST_RLOCK(); 2163 VNET_FOREACH(vnet_iter) { 2164 CURVNET_SET(vnet_iter); 2165 igmp_slowtimo_vnet(); 2166 CURVNET_RESTORE(); 2167 } 2168 VNET_LIST_RUNLOCK(); 2169 } 2170 2171 /* 2172 * Per-vnet slowtimo handler. 2173 */ 2174 static void 2175 igmp_slowtimo_vnet(void) 2176 { 2177 struct igmp_ifinfo *igi; 2178 2179 IGMP_LOCK(); 2180 2181 LIST_FOREACH(igi, &V_igi_head, igi_link) { 2182 igmp_v1v2_process_querier_timers(igi); 2183 } 2184 2185 IGMP_UNLOCK(); 2186 } 2187 2188 /* 2189 * Dispatch an IGMPv1/v2 host report or leave message. 2190 * These are always small enough to fit inside a single mbuf. 2191 */ 2192 static int 2193 igmp_v1v2_queue_report(struct in_multi *inm, const int type) 2194 { 2195 struct ifnet *ifp; 2196 struct igmp *igmp; 2197 struct ip *ip; 2198 struct mbuf *m; 2199 2200 IN_MULTI_LOCK_ASSERT(); 2201 IGMP_LOCK_ASSERT(); 2202 2203 ifp = inm->inm_ifp; 2204 2205 MGETHDR(m, M_DONTWAIT, MT_DATA); 2206 if (m == NULL) 2207 return (ENOMEM); 2208 MH_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp)); 2209 2210 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp); 2211 2212 m->m_data += sizeof(struct ip); 2213 m->m_len = sizeof(struct igmp); 2214 2215 igmp = mtod(m, struct igmp *); 2216 igmp->igmp_type = type; 2217 igmp->igmp_code = 0; 2218 igmp->igmp_group = inm->inm_addr; 2219 igmp->igmp_cksum = 0; 2220 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp)); 2221 2222 m->m_data -= sizeof(struct ip); 2223 m->m_len += sizeof(struct ip); 2224 2225 ip = mtod(m, struct ip *); 2226 ip->ip_tos = 0; 2227 ip->ip_len = sizeof(struct ip) + sizeof(struct igmp); 2228 ip->ip_off = 0; 2229 ip->ip_p = IPPROTO_IGMP; 2230 ip->ip_src.s_addr = INADDR_ANY; 2231 2232 if (type == IGMP_HOST_LEAVE_MESSAGE) 2233 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP); 2234 else 2235 ip->ip_dst = inm->inm_addr; 2236 2237 igmp_save_context(m, ifp); 2238 2239 m->m_flags |= M_IGMPV2; 2240 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK) 2241 m->m_flags |= M_IGMP_LOOP; 2242 2243 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m); 2244 netisr_dispatch(NETISR_IGMP, m); 2245 2246 return (0); 2247 } 2248 2249 /* 2250 * Process a state change from the upper layer for the given IPv4 group. 2251 * 2252 * Each socket holds a reference on the in_multi in its own ip_moptions. 2253 * The socket layer will have made the necessary updates to.the group 2254 * state, it is now up to IGMP to issue a state change report if there 2255 * has been any change between T0 (when the last state-change was issued) 2256 * and T1 (now). 2257 * 2258 * We use the IGMPv3 state machine at group level. The IGMP module 2259 * however makes the decision as to which IGMP protocol version to speak. 2260 * A state change *from* INCLUDE {} always means an initial join. 2261 * A state change *to* INCLUDE {} always means a final leave. 2262 * 2263 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can 2264 * save ourselves a bunch of work; any exclusive mode groups need not 2265 * compute source filter lists. 2266 * 2267 * VIMAGE: curvnet should have been set by caller, as this routine 2268 * is called from the socket option handlers. 2269 */ 2270 int 2271 igmp_change_state(struct in_multi *inm) 2272 { 2273 struct igmp_ifinfo *igi; 2274 struct ifnet *ifp; 2275 int error; 2276 2277 IN_MULTI_LOCK_ASSERT(); 2278 2279 error = 0; 2280 2281 /* 2282 * Try to detect if the upper layer just asked us to change state 2283 * for an interface which has now gone away. 2284 */ 2285 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 2286 ifp = inm->inm_ifma->ifma_ifp; 2287 if (ifp != NULL) { 2288 /* 2289 * Sanity check that netinet's notion of ifp is the 2290 * same as net's. 2291 */ 2292 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 2293 } 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_DONTWAIT, MT_DATA, M_PKTHDR); 2785 if (m) 2786 m->m_data += IGMP_LEADINGSPACE; 2787 } 2788 if (m == NULL) { 2789 m = m_gethdr(M_DONTWAIT, 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_DONTWAIT, MT_DATA, M_PKTHDR); 2910 if (m) 2911 m->m_data += IGMP_LEADINGSPACE; 2912 if (m == NULL) { 2913 m = m_gethdr(M_DONTWAIT, 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_DONTWAIT, MT_DATA, M_PKTHDR); 3066 if (m) 3067 m->m_data += IGMP_LEADINGSPACE; 3068 if (m == NULL) { 3069 m = m_gethdr(M_DONTWAIT, 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, *tifma; 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_LOCK(ifp); 3334 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) { 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_UNLOCK(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_DONTWAIT); 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 = hdrlen + igmpreclen; 3527 ip->ip_off = 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_sysinit(void) 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 3592 static void 3593 igmp_sysuninit(void) 3594 { 3595 3596 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3597 3598 netisr_unregister(&igmp_nh); 3599 3600 m_free(m_raopt); 3601 m_raopt = NULL; 3602 3603 IGMP_LOCK_DESTROY(); 3604 } 3605 3606 /* 3607 * Initialize an IGMPv3 instance. 3608 * VIMAGE: Assumes curvnet set by caller and called per vimage. 3609 */ 3610 static int 3611 vnet_igmp_iattach(const void *unused __unused) 3612 { 3613 3614 CTR1(KTR_IGMPV3, "%s: initializing", __func__); 3615 3616 LIST_INIT(&V_igi_head); 3617 3618 /* 3619 * Initialize sysctls to default values. 3620 */ 3621 V_igmpstat.igps_version = IGPS_VERSION_3; 3622 V_igmpstat.igps_len = sizeof(struct igmpstat); 3623 3624 return (0); 3625 } 3626 3627 static int 3628 vnet_igmp_idetach(const void *unused __unused) 3629 { 3630 3631 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3632 3633 KASSERT(LIST_EMPTY(&V_igi_head), 3634 ("%s: igi list not empty; ifnets not detached?", __func__)); 3635 3636 return (0); 3637 } 3638 3639 #ifdef VIMAGE 3640 static vnet_modinfo_t vnet_igmp_modinfo = { 3641 .vmi_id = VNET_MOD_IGMP, 3642 .vmi_name = "igmp", 3643 .vmi_dependson = VNET_MOD_INET, 3644 .vmi_iattach = vnet_igmp_iattach, 3645 .vmi_idetach = vnet_igmp_idetach 3646 }; 3647 #endif 3648 3649 static int 3650 igmp_modevent(module_t mod, int type, void *unused __unused) 3651 { 3652 3653 switch (type) { 3654 case MOD_LOAD: 3655 igmp_sysinit(); 3656 #ifdef VIMAGE 3657 vnet_mod_register(&vnet_igmp_modinfo); 3658 #else 3659 vnet_igmp_iattach(NULL); 3660 #endif 3661 break; 3662 case MOD_UNLOAD: 3663 #ifdef VIMAGE 3664 vnet_mod_deregister(&vnet_igmp_modinfo); 3665 #else 3666 vnet_igmp_idetach(NULL); 3667 #endif 3668 igmp_sysuninit(); 3669 break; 3670 default: 3671 return (EOPNOTSUPP); 3672 } 3673 return (0); 3674 } 3675 3676 static moduledata_t igmp_mod = { 3677 "igmp", 3678 igmp_modevent, 3679 0 3680 }; 3681 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3682