1 /*- 2 * Copyright (c) 2007-2009 Bruce Simpson. 3 * Copyright (c) 1988 Stephen Deering. 4 * Copyright (c) 1992, 1993 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Stephen Deering of Stanford University. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)igmp.c 8.1 (Berkeley) 7/19/93 35 */ 36 37 /* 38 * Internet Group Management Protocol (IGMP) routines. 39 * [RFC1112, RFC2236, RFC3376] 40 * 41 * Written by Steve Deering, Stanford, May 1988. 42 * Modified by Rosen Sharma, Stanford, Aug 1994. 43 * Modified by Bill Fenner, Xerox PARC, Feb 1995. 44 * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995. 45 * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson. 46 * 47 * MULTICAST Revision: 3.5.1.4 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include <sys/param.h> 54 #include <sys/systm.h> 55 #include <sys/module.h> 56 #include <sys/malloc.h> 57 #include <sys/mbuf.h> 58 #include <sys/socket.h> 59 #include <sys/protosw.h> 60 #include <sys/kernel.h> 61 #include <sys/sysctl.h> 62 #include <sys/ktr.h> 63 #include <sys/condvar.h> 64 65 #include <net/if.h> 66 #include <net/if_var.h> 67 #include <net/netisr.h> 68 #include <net/vnet.h> 69 70 #include <netinet/in.h> 71 #include <netinet/in_var.h> 72 #include <netinet/in_systm.h> 73 #include <netinet/ip.h> 74 #include <netinet/ip_var.h> 75 #include <netinet/ip_options.h> 76 #include <netinet/igmp.h> 77 #include <netinet/igmp_var.h> 78 79 #include <machine/in_cksum.h> 80 81 #include <security/mac/mac_framework.h> 82 83 #ifndef KTR_IGMPV3 84 #define KTR_IGMPV3 KTR_INET 85 #endif 86 87 static struct igmp_ifsoftc * 88 igi_alloc_locked(struct ifnet *); 89 static void igi_delete_locked(const struct ifnet *); 90 static void igmp_dispatch_queue(struct mbufq *, int, const int); 91 static void igmp_fasttimo_vnet(void); 92 static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *); 93 static int igmp_handle_state_change(struct in_multi *, 94 struct igmp_ifsoftc *); 95 static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *); 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_ifsoftc *, 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_ifsoftc *, const int); 116 static void igmp_slowtimo_vnet(void); 117 static int igmp_v1v2_queue_report(struct in_multi *, const int); 118 static void igmp_v1v2_process_group_timer(struct in_multi *, const int); 119 static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *); 120 static void igmp_v2_update_group(struct in_multi *, const int); 121 static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *); 122 static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *); 123 static struct mbuf * 124 igmp_v3_encap_report(struct ifnet *, struct mbuf *); 125 static int igmp_v3_enqueue_group_record(struct mbufq *, 126 struct in_multi *, const int, const int, const int); 127 static int igmp_v3_enqueue_filter_change(struct mbufq *, 128 struct in_multi *); 129 static void igmp_v3_process_group_timers(struct igmp_ifsoftc *, 130 struct mbufq *, struct mbufq *, struct in_multi *, 131 const int); 132 static int igmp_v3_merge_state_changes(struct in_multi *, 133 struct mbufq *); 134 static void igmp_v3_suppress_group_record(struct in_multi *); 135 static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS); 136 static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS); 137 static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS); 138 139 static const struct netisr_handler igmp_nh = { 140 .nh_name = "igmp", 141 .nh_handler = igmp_intr, 142 .nh_proto = NETISR_IGMP, 143 .nh_policy = NETISR_POLICY_SOURCE, 144 }; 145 146 /* 147 * System-wide globals. 148 * 149 * Unlocked access to these is OK, except for the global IGMP output 150 * queue. The IGMP subsystem lock ends up being system-wide for the moment, 151 * because all VIMAGEs have to share a global output queue, as netisrs 152 * themselves are not virtualized. 153 * 154 * Locking: 155 * * The permitted lock order is: IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 156 * Any may be taken independently; if any are held at the same 157 * time, the above lock order must be followed. 158 * * All output is delegated to the netisr. 159 * Now that Giant has been eliminated, the netisr may be inlined. 160 * * IN_MULTI_LOCK covers in_multi. 161 * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file, 162 * including the output queue. 163 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 164 * per-link state iterators. 165 * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface, 166 * therefore it is not refcounted. 167 * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi. 168 * 169 * Reference counting 170 * * IGMP acquires its own reference every time an in_multi is passed to 171 * it and the group is being joined for the first time. 172 * * IGMP releases its reference(s) on in_multi in a deferred way, 173 * because the operations which process the release run as part of 174 * a loop whose control variables are directly affected by the release 175 * (that, and not recursing on the IF_ADDR_LOCK). 176 * 177 * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds 178 * to a vnet in ifp->if_vnet. 179 * 180 * SMPng: XXX We may potentially race operations on ifma_protospec. 181 * The problem is that we currently lack a clean way of taking the 182 * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing, 183 * as anything which modifies ifma needs to be covered by that lock. 184 * So check for ifma_protospec being NULL before proceeding. 185 */ 186 struct mtx igmp_mtx; 187 188 struct mbuf *m_raopt; /* Router Alert option */ 189 static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state"); 190 191 /* 192 * VIMAGE-wide globals. 193 * 194 * The IGMPv3 timers themselves need to run per-image, however, 195 * protosw timers run globally (see tcp). 196 * An ifnet can only be in one vimage at a time, and the loopback 197 * ifnet, loif, is itself virtualized. 198 * It would otherwise be possible to seriously hose IGMP state, 199 * and create inconsistencies in upstream multicast routing, if you have 200 * multiple VIMAGEs running on the same link joining different multicast 201 * groups, UNLESS the "primary IP address" is different. This is because 202 * IGMP for IPv4 does not force link-local addresses to be used for each 203 * node, unlike MLD for IPv6. 204 * Obviously the IGMPv3 per-interface state has per-vimage granularity 205 * also as a result. 206 * 207 * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection 208 * policy to control the address used by IGMP on the link. 209 */ 210 static VNET_DEFINE(int, interface_timers_running); /* IGMPv3 general 211 * query response */ 212 static VNET_DEFINE(int, state_change_timers_running); /* IGMPv3 state-change 213 * retransmit */ 214 static VNET_DEFINE(int, current_state_timers_running); /* IGMPv1/v2 host 215 * report; IGMPv3 g/sg 216 * query response */ 217 218 #define V_interface_timers_running VNET(interface_timers_running) 219 #define V_state_change_timers_running VNET(state_change_timers_running) 220 #define V_current_state_timers_running VNET(current_state_timers_running) 221 222 static VNET_DEFINE(LIST_HEAD(, igmp_ifsoftc), igi_head); 223 static VNET_DEFINE(struct igmpstat, igmpstat) = { 224 .igps_version = IGPS_VERSION_3, 225 .igps_len = sizeof(struct igmpstat), 226 }; 227 static VNET_DEFINE(struct timeval, igmp_gsrdelay) = {10, 0}; 228 229 #define V_igi_head VNET(igi_head) 230 #define V_igmpstat VNET(igmpstat) 231 #define V_igmp_gsrdelay VNET(igmp_gsrdelay) 232 233 static VNET_DEFINE(int, igmp_recvifkludge) = 1; 234 static VNET_DEFINE(int, igmp_sendra) = 1; 235 static VNET_DEFINE(int, igmp_sendlocal) = 1; 236 static VNET_DEFINE(int, igmp_v1enable) = 1; 237 static VNET_DEFINE(int, igmp_v2enable) = 1; 238 static VNET_DEFINE(int, igmp_legacysupp); 239 static VNET_DEFINE(int, igmp_default_version) = IGMP_VERSION_3; 240 241 #define V_igmp_recvifkludge VNET(igmp_recvifkludge) 242 #define V_igmp_sendra VNET(igmp_sendra) 243 #define V_igmp_sendlocal VNET(igmp_sendlocal) 244 #define V_igmp_v1enable VNET(igmp_v1enable) 245 #define V_igmp_v2enable VNET(igmp_v2enable) 246 #define V_igmp_legacysupp VNET(igmp_legacysupp) 247 #define V_igmp_default_version VNET(igmp_default_version) 248 249 /* 250 * Virtualized sysctls. 251 */ 252 SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW, 253 &VNET_NAME(igmpstat), igmpstat, ""); 254 SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW, 255 &VNET_NAME(igmp_recvifkludge), 0, 256 "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address"); 257 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW, 258 &VNET_NAME(igmp_sendra), 0, 259 "Send IP Router Alert option in IGMPv2/v3 messages"); 260 SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW, 261 &VNET_NAME(igmp_sendlocal), 0, 262 "Send IGMP membership reports for 224.0.0.0/24 groups"); 263 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW, 264 &VNET_NAME(igmp_v1enable), 0, 265 "Enable backwards compatibility with IGMPv1"); 266 SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW, 267 &VNET_NAME(igmp_v2enable), 0, 268 "Enable backwards compatibility with IGMPv2"); 269 SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW, 270 &VNET_NAME(igmp_legacysupp), 0, 271 "Allow v1/v2 reports to suppress v3 group responses"); 272 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version, 273 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 274 &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I", 275 "Default version of IGMP to run on each interface"); 276 SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay, 277 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 278 &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I", 279 "Rate limit for IGMPv3 Group-and-Source queries in seconds"); 280 281 /* 282 * Non-virtualized sysctls. 283 */ 284 static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, 285 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo, 286 "Per-interface IGMPv3 state"); 287 288 static __inline void 289 igmp_save_context(struct mbuf *m, struct ifnet *ifp) 290 { 291 292 #ifdef VIMAGE 293 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; 294 #endif /* VIMAGE */ 295 m->m_pkthdr.flowid = ifp->if_index; 296 } 297 298 static __inline void 299 igmp_scrub_context(struct mbuf *m) 300 { 301 302 m->m_pkthdr.PH_loc.ptr = NULL; 303 m->m_pkthdr.flowid = 0; 304 } 305 306 #ifdef KTR 307 static __inline char * 308 inet_ntoa_haddr(in_addr_t haddr) 309 { 310 struct in_addr ia; 311 312 ia.s_addr = htonl(haddr); 313 return (inet_ntoa(ia)); 314 } 315 #endif 316 317 /* 318 * Restore context from a queued IGMP output chain. 319 * Return saved ifindex. 320 * 321 * VIMAGE: The assertion is there to make sure that we 322 * actually called CURVNET_SET() with what's in the mbuf chain. 323 */ 324 static __inline uint32_t 325 igmp_restore_context(struct mbuf *m) 326 { 327 328 #ifdef notyet 329 #if defined(VIMAGE) && defined(INVARIANTS) 330 KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr), 331 ("%s: called when curvnet was not restored", __func__)); 332 #endif 333 #endif 334 return (m->m_pkthdr.flowid); 335 } 336 337 /* 338 * Retrieve or set default IGMP version. 339 * 340 * VIMAGE: Assume curvnet set by caller. 341 * SMPng: NOTE: Serialized by IGMP lock. 342 */ 343 static int 344 sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS) 345 { 346 int error; 347 int new; 348 349 error = sysctl_wire_old_buffer(req, sizeof(int)); 350 if (error) 351 return (error); 352 353 IGMP_LOCK(); 354 355 new = V_igmp_default_version; 356 357 error = sysctl_handle_int(oidp, &new, 0, req); 358 if (error || !req->newptr) 359 goto out_locked; 360 361 if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) { 362 error = EINVAL; 363 goto out_locked; 364 } 365 366 CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d", 367 V_igmp_default_version, new); 368 369 V_igmp_default_version = new; 370 371 out_locked: 372 IGMP_UNLOCK(); 373 return (error); 374 } 375 376 /* 377 * Retrieve or set threshold between group-source queries in seconds. 378 * 379 * VIMAGE: Assume curvnet set by caller. 380 * SMPng: NOTE: Serialized by IGMP lock. 381 */ 382 static int 383 sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS) 384 { 385 int error; 386 int i; 387 388 error = sysctl_wire_old_buffer(req, sizeof(int)); 389 if (error) 390 return (error); 391 392 IGMP_LOCK(); 393 394 i = V_igmp_gsrdelay.tv_sec; 395 396 error = sysctl_handle_int(oidp, &i, 0, req); 397 if (error || !req->newptr) 398 goto out_locked; 399 400 if (i < -1 || i >= 60) { 401 error = EINVAL; 402 goto out_locked; 403 } 404 405 CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d", 406 V_igmp_gsrdelay.tv_sec, i); 407 V_igmp_gsrdelay.tv_sec = i; 408 409 out_locked: 410 IGMP_UNLOCK(); 411 return (error); 412 } 413 414 /* 415 * Expose struct igmp_ifsoftc 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_ifsoftc *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 struct igmp_ifinfo info; 461 462 info.igi_version = igi->igi_version; 463 info.igi_v1_timer = igi->igi_v1_timer; 464 info.igi_v2_timer = igi->igi_v2_timer; 465 info.igi_v3_timer = igi->igi_v3_timer; 466 info.igi_flags = igi->igi_flags; 467 info.igi_rv = igi->igi_rv; 468 info.igi_qi = igi->igi_qi; 469 info.igi_qri = igi->igi_qri; 470 info.igi_uri = igi->igi_uri; 471 error = SYSCTL_OUT(req, &info, sizeof(info)); 472 break; 473 } 474 } 475 476 out_locked: 477 IGMP_UNLOCK(); 478 IN_MULTI_UNLOCK(); 479 return (error); 480 } 481 482 /* 483 * Dispatch an entire queue of pending packet chains 484 * using the netisr. 485 * VIMAGE: Assumes the vnet pointer has been set. 486 */ 487 static void 488 igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop) 489 { 490 struct mbuf *m; 491 492 while ((m = mbufq_dequeue(mq)) != NULL) { 493 CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m); 494 if (loop) 495 m->m_flags |= M_IGMP_LOOP; 496 netisr_dispatch(NETISR_IGMP, m); 497 if (--limit == 0) 498 break; 499 } 500 } 501 502 /* 503 * Filter outgoing IGMP report state by group. 504 * 505 * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1). 506 * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are 507 * disabled for all groups in the 224.0.0.0/24 link-local scope. However, 508 * this may break certain IGMP snooping switches which rely on the old 509 * report behaviour. 510 * 511 * Return zero if the given group is one for which IGMP reports 512 * should be suppressed, or non-zero if reports should be issued. 513 */ 514 static __inline int 515 igmp_isgroupreported(const struct in_addr addr) 516 { 517 518 if (in_allhosts(addr) || 519 ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr))))) 520 return (0); 521 522 return (1); 523 } 524 525 /* 526 * Construct a Router Alert option to use in outgoing packets. 527 */ 528 static struct mbuf * 529 igmp_ra_alloc(void) 530 { 531 struct mbuf *m; 532 struct ipoption *p; 533 534 m = m_get(M_WAITOK, MT_DATA); 535 p = mtod(m, struct ipoption *); 536 p->ipopt_dst.s_addr = INADDR_ANY; 537 p->ipopt_list[0] = IPOPT_RA; /* Router Alert Option */ 538 p->ipopt_list[1] = 0x04; /* 4 bytes long */ 539 p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */ 540 p->ipopt_list[3] = 0x00; /* pad byte */ 541 m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1]; 542 543 return (m); 544 } 545 546 /* 547 * Attach IGMP when PF_INET is attached to an interface. 548 */ 549 struct igmp_ifsoftc * 550 igmp_domifattach(struct ifnet *ifp) 551 { 552 struct igmp_ifsoftc *igi; 553 554 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 555 __func__, ifp, ifp->if_xname); 556 557 IGMP_LOCK(); 558 559 igi = igi_alloc_locked(ifp); 560 if (!(ifp->if_flags & IFF_MULTICAST)) 561 igi->igi_flags |= IGIF_SILENT; 562 563 IGMP_UNLOCK(); 564 565 return (igi); 566 } 567 568 /* 569 * VIMAGE: assume curvnet set by caller. 570 */ 571 static struct igmp_ifsoftc * 572 igi_alloc_locked(/*const*/ struct ifnet *ifp) 573 { 574 struct igmp_ifsoftc *igi; 575 576 IGMP_LOCK_ASSERT(); 577 578 igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO); 579 if (igi == NULL) 580 goto out; 581 582 igi->igi_ifp = ifp; 583 igi->igi_version = V_igmp_default_version; 584 igi->igi_flags = 0; 585 igi->igi_rv = IGMP_RV_INIT; 586 igi->igi_qi = IGMP_QI_INIT; 587 igi->igi_qri = IGMP_QRI_INIT; 588 igi->igi_uri = IGMP_URI_INIT; 589 SLIST_INIT(&igi->igi_relinmhead); 590 mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS); 591 592 LIST_INSERT_HEAD(&V_igi_head, igi, igi_link); 593 594 CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)", 595 ifp, ifp->if_xname); 596 597 out: 598 return (igi); 599 } 600 601 /* 602 * Hook for ifdetach. 603 * 604 * NOTE: Some finalization tasks need to run before the protocol domain 605 * is detached, but also before the link layer does its cleanup. 606 * 607 * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK(). 608 * XXX This is also bitten by unlocked ifma_protospec access. 609 */ 610 void 611 igmp_ifdetach(struct ifnet *ifp) 612 { 613 struct igmp_ifsoftc *igi; 614 struct ifmultiaddr *ifma; 615 struct in_multi *inm, *tinm; 616 617 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, 618 ifp->if_xname); 619 620 IGMP_LOCK(); 621 622 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 623 if (igi->igi_version == IGMP_VERSION_3) { 624 IF_ADDR_RLOCK(ifp); 625 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 626 if (ifma->ifma_addr->sa_family != AF_INET || 627 ifma->ifma_protospec == NULL) 628 continue; 629 #if 0 630 KASSERT(ifma->ifma_protospec != NULL, 631 ("%s: ifma_protospec is NULL", __func__)); 632 #endif 633 inm = (struct in_multi *)ifma->ifma_protospec; 634 if (inm->inm_state == IGMP_LEAVING_MEMBER) { 635 SLIST_INSERT_HEAD(&igi->igi_relinmhead, 636 inm, inm_nrele); 637 } 638 inm_clear_recorded(inm); 639 } 640 IF_ADDR_RUNLOCK(ifp); 641 /* 642 * Free the in_multi reference(s) for this IGMP lifecycle. 643 */ 644 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele, 645 tinm) { 646 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele); 647 inm_release_locked(inm); 648 } 649 } 650 651 IGMP_UNLOCK(); 652 } 653 654 /* 655 * Hook for domifdetach. 656 */ 657 void 658 igmp_domifdetach(struct ifnet *ifp) 659 { 660 struct igmp_ifsoftc *igi; 661 662 CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", 663 __func__, ifp, ifp->if_xname); 664 665 IGMP_LOCK(); 666 667 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 668 igi_delete_locked(ifp); 669 670 IGMP_UNLOCK(); 671 } 672 673 static void 674 igi_delete_locked(const struct ifnet *ifp) 675 { 676 struct igmp_ifsoftc *igi, *tigi; 677 678 CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)", 679 __func__, ifp, ifp->if_xname); 680 681 IGMP_LOCK_ASSERT(); 682 683 LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) { 684 if (igi->igi_ifp == ifp) { 685 /* 686 * Free deferred General Query responses. 687 */ 688 mbufq_drain(&igi->igi_gq); 689 690 LIST_REMOVE(igi, igi_link); 691 692 KASSERT(SLIST_EMPTY(&igi->igi_relinmhead), 693 ("%s: there are dangling in_multi references", 694 __func__)); 695 696 free(igi, M_IGMP); 697 return; 698 } 699 } 700 701 #ifdef INVARIANTS 702 panic("%s: igmp_ifsoftc not found for ifp %p\n", __func__, ifp); 703 #endif 704 } 705 706 /* 707 * Process a received IGMPv1 query. 708 * Return non-zero if the message should be dropped. 709 * 710 * VIMAGE: The curvnet pointer is derived from the input ifp. 711 */ 712 static int 713 igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip, 714 const struct igmp *igmp) 715 { 716 struct ifmultiaddr *ifma; 717 struct igmp_ifsoftc *igi; 718 struct in_multi *inm; 719 720 /* 721 * IGMPv1 Host Mmembership Queries SHOULD always be addressed to 722 * 224.0.0.1. They are always treated as General Queries. 723 * igmp_group is always ignored. Do not drop it as a userland 724 * daemon may wish to see it. 725 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 726 */ 727 if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) { 728 IGMPSTAT_INC(igps_rcv_badqueries); 729 return (0); 730 } 731 IGMPSTAT_INC(igps_rcv_gen_queries); 732 733 IN_MULTI_LOCK(); 734 IGMP_LOCK(); 735 736 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 737 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 738 739 if (igi->igi_flags & IGIF_LOOPBACK) { 740 CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)", 741 ifp, ifp->if_xname); 742 goto out_locked; 743 } 744 745 /* 746 * Switch to IGMPv1 host compatibility mode. 747 */ 748 igmp_set_version(igi, IGMP_VERSION_1); 749 750 CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname); 751 752 /* 753 * Start the timers in all of our group records 754 * for the interface on which the query arrived, 755 * except those which are already running. 756 */ 757 IF_ADDR_RLOCK(ifp); 758 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 759 if (ifma->ifma_addr->sa_family != AF_INET || 760 ifma->ifma_protospec == NULL) 761 continue; 762 inm = (struct in_multi *)ifma->ifma_protospec; 763 if (inm->inm_timer != 0) 764 continue; 765 switch (inm->inm_state) { 766 case IGMP_NOT_MEMBER: 767 case IGMP_SILENT_MEMBER: 768 break; 769 case IGMP_G_QUERY_PENDING_MEMBER: 770 case IGMP_SG_QUERY_PENDING_MEMBER: 771 case IGMP_REPORTING_MEMBER: 772 case IGMP_IDLE_MEMBER: 773 case IGMP_LAZY_MEMBER: 774 case IGMP_SLEEPING_MEMBER: 775 case IGMP_AWAKENING_MEMBER: 776 inm->inm_state = IGMP_REPORTING_MEMBER; 777 inm->inm_timer = IGMP_RANDOM_DELAY( 778 IGMP_V1V2_MAX_RI * PR_FASTHZ); 779 V_current_state_timers_running = 1; 780 break; 781 case IGMP_LEAVING_MEMBER: 782 break; 783 } 784 } 785 IF_ADDR_RUNLOCK(ifp); 786 787 out_locked: 788 IGMP_UNLOCK(); 789 IN_MULTI_UNLOCK(); 790 791 return (0); 792 } 793 794 /* 795 * Process a received IGMPv2 general or group-specific query. 796 */ 797 static int 798 igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip, 799 const struct igmp *igmp) 800 { 801 struct ifmultiaddr *ifma; 802 struct igmp_ifsoftc *igi; 803 struct in_multi *inm; 804 int is_general_query; 805 uint16_t timer; 806 807 is_general_query = 0; 808 809 /* 810 * Validate address fields upfront. 811 * XXX SMPng: unlocked increments in igmpstat assumed atomic. 812 */ 813 if (in_nullhost(igmp->igmp_group)) { 814 /* 815 * IGMPv2 General Query. 816 * If this was not sent to the all-hosts group, ignore it. 817 */ 818 if (!in_allhosts(ip->ip_dst)) 819 return (0); 820 IGMPSTAT_INC(igps_rcv_gen_queries); 821 is_general_query = 1; 822 } else { 823 /* IGMPv2 Group-Specific Query. */ 824 IGMPSTAT_INC(igps_rcv_group_queries); 825 } 826 827 IN_MULTI_LOCK(); 828 IGMP_LOCK(); 829 830 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 831 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 832 833 if (igi->igi_flags & IGIF_LOOPBACK) { 834 CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)", 835 ifp, ifp->if_xname); 836 goto out_locked; 837 } 838 839 /* 840 * Ignore v2 query if in v1 Compatibility Mode. 841 */ 842 if (igi->igi_version == IGMP_VERSION_1) 843 goto out_locked; 844 845 igmp_set_version(igi, IGMP_VERSION_2); 846 847 timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE; 848 if (timer == 0) 849 timer = 1; 850 851 if (is_general_query) { 852 /* 853 * For each reporting group joined on this 854 * interface, kick the report timer. 855 */ 856 CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)", 857 ifp, ifp->if_xname); 858 IF_ADDR_RLOCK(ifp); 859 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 860 if (ifma->ifma_addr->sa_family != AF_INET || 861 ifma->ifma_protospec == NULL) 862 continue; 863 inm = (struct in_multi *)ifma->ifma_protospec; 864 igmp_v2_update_group(inm, timer); 865 } 866 IF_ADDR_RUNLOCK(ifp); 867 } else { 868 /* 869 * Group-specific IGMPv2 query, we need only 870 * look up the single group to process it. 871 */ 872 inm = inm_lookup(ifp, igmp->igmp_group); 873 if (inm != NULL) { 874 CTR3(KTR_IGMPV3, "process v2 query %s on ifp %p(%s)", 875 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 876 igmp_v2_update_group(inm, timer); 877 } 878 } 879 880 out_locked: 881 IGMP_UNLOCK(); 882 IN_MULTI_UNLOCK(); 883 884 return (0); 885 } 886 887 /* 888 * Update the report timer on a group in response to an IGMPv2 query. 889 * 890 * If we are becoming the reporting member for this group, start the timer. 891 * If we already are the reporting member for this group, and timer is 892 * below the threshold, reset it. 893 * 894 * We may be updating the group for the first time since we switched 895 * to IGMPv3. If we are, then we must clear any recorded source lists, 896 * and transition to REPORTING state; the group timer is overloaded 897 * for group and group-source query responses. 898 * 899 * Unlike IGMPv3, the delay per group should be jittered 900 * to avoid bursts of IGMPv2 reports. 901 */ 902 static void 903 igmp_v2_update_group(struct in_multi *inm, const int timer) 904 { 905 906 CTR4(KTR_IGMPV3, "%s: %s/%s timer=%d", __func__, 907 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname, timer); 908 909 IN_MULTI_LOCK_ASSERT(); 910 911 switch (inm->inm_state) { 912 case IGMP_NOT_MEMBER: 913 case IGMP_SILENT_MEMBER: 914 break; 915 case IGMP_REPORTING_MEMBER: 916 if (inm->inm_timer != 0 && 917 inm->inm_timer <= timer) { 918 CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, " 919 "skipping.", __func__); 920 break; 921 } 922 /* FALLTHROUGH */ 923 case IGMP_SG_QUERY_PENDING_MEMBER: 924 case IGMP_G_QUERY_PENDING_MEMBER: 925 case IGMP_IDLE_MEMBER: 926 case IGMP_LAZY_MEMBER: 927 case IGMP_AWAKENING_MEMBER: 928 CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__); 929 inm->inm_state = IGMP_REPORTING_MEMBER; 930 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 931 V_current_state_timers_running = 1; 932 break; 933 case IGMP_SLEEPING_MEMBER: 934 CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__); 935 inm->inm_state = IGMP_AWAKENING_MEMBER; 936 break; 937 case IGMP_LEAVING_MEMBER: 938 break; 939 } 940 } 941 942 /* 943 * Process a received IGMPv3 general, group-specific or 944 * group-and-source-specific query. 945 * Assumes m has already been pulled up to the full IGMP message length. 946 * Return 0 if successful, otherwise an appropriate error code is returned. 947 */ 948 static int 949 igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip, 950 /*const*/ struct igmpv3 *igmpv3) 951 { 952 struct igmp_ifsoftc *igi; 953 struct in_multi *inm; 954 int is_general_query; 955 uint32_t maxresp, nsrc, qqi; 956 uint16_t timer; 957 uint8_t qrv; 958 959 is_general_query = 0; 960 961 CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname); 962 963 maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */ 964 if (maxresp >= 128) { 965 maxresp = IGMP_MANT(igmpv3->igmp_code) << 966 (IGMP_EXP(igmpv3->igmp_code) + 3); 967 } 968 969 /* 970 * Robustness must never be less than 2 for on-wire IGMPv3. 971 * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make 972 * an exception for interfaces whose IGMPv3 state changes 973 * are redirected to loopback (e.g. MANET). 974 */ 975 qrv = IGMP_QRV(igmpv3->igmp_misc); 976 if (qrv < 2) { 977 CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__, 978 qrv, IGMP_RV_INIT); 979 qrv = IGMP_RV_INIT; 980 } 981 982 qqi = igmpv3->igmp_qqi; 983 if (qqi >= 128) { 984 qqi = IGMP_MANT(igmpv3->igmp_qqi) << 985 (IGMP_EXP(igmpv3->igmp_qqi) + 3); 986 } 987 988 timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE; 989 if (timer == 0) 990 timer = 1; 991 992 nsrc = ntohs(igmpv3->igmp_numsrc); 993 994 /* 995 * Validate address fields and versions upfront before 996 * accepting v3 query. 997 * XXX SMPng: Unlocked access to igmpstat counters here. 998 */ 999 if (in_nullhost(igmpv3->igmp_group)) { 1000 /* 1001 * IGMPv3 General Query. 1002 * 1003 * General Queries SHOULD be directed to 224.0.0.1. 1004 * A general query with a source list has undefined 1005 * behaviour; discard it. 1006 */ 1007 IGMPSTAT_INC(igps_rcv_gen_queries); 1008 if (!in_allhosts(ip->ip_dst) || nsrc > 0) { 1009 IGMPSTAT_INC(igps_rcv_badqueries); 1010 return (0); 1011 } 1012 is_general_query = 1; 1013 } else { 1014 /* Group or group-source specific query. */ 1015 if (nsrc == 0) 1016 IGMPSTAT_INC(igps_rcv_group_queries); 1017 else 1018 IGMPSTAT_INC(igps_rcv_gsr_queries); 1019 } 1020 1021 IN_MULTI_LOCK(); 1022 IGMP_LOCK(); 1023 1024 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 1025 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 1026 1027 if (igi->igi_flags & IGIF_LOOPBACK) { 1028 CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)", 1029 ifp, ifp->if_xname); 1030 goto out_locked; 1031 } 1032 1033 /* 1034 * Discard the v3 query if we're in Compatibility Mode. 1035 * The RFC is not obviously worded that hosts need to stay in 1036 * compatibility mode until the Old Version Querier Present 1037 * timer expires. 1038 */ 1039 if (igi->igi_version != IGMP_VERSION_3) { 1040 CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)", 1041 igi->igi_version, ifp, ifp->if_xname); 1042 goto out_locked; 1043 } 1044 1045 igmp_set_version(igi, IGMP_VERSION_3); 1046 igi->igi_rv = qrv; 1047 igi->igi_qi = qqi; 1048 igi->igi_qri = maxresp; 1049 1050 CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi, 1051 maxresp); 1052 1053 if (is_general_query) { 1054 /* 1055 * Schedule a current-state report on this ifp for 1056 * all groups, possibly containing source lists. 1057 * If there is a pending General Query response 1058 * scheduled earlier than the selected delay, do 1059 * not schedule any other reports. 1060 * Otherwise, reset the interface timer. 1061 */ 1062 CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)", 1063 ifp, ifp->if_xname); 1064 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) { 1065 igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer); 1066 V_interface_timers_running = 1; 1067 } 1068 } else { 1069 /* 1070 * Group-source-specific queries are throttled on 1071 * a per-group basis to defeat denial-of-service attempts. 1072 * Queries for groups we are not a member of on this 1073 * link are simply ignored. 1074 */ 1075 inm = inm_lookup(ifp, igmpv3->igmp_group); 1076 if (inm == NULL) 1077 goto out_locked; 1078 if (nsrc > 0) { 1079 if (!ratecheck(&inm->inm_lastgsrtv, 1080 &V_igmp_gsrdelay)) { 1081 CTR1(KTR_IGMPV3, "%s: GS query throttled.", 1082 __func__); 1083 IGMPSTAT_INC(igps_drop_gsr_queries); 1084 goto out_locked; 1085 } 1086 } 1087 CTR3(KTR_IGMPV3, "process v3 %s query on ifp %p(%s)", 1088 inet_ntoa(igmpv3->igmp_group), ifp, ifp->if_xname); 1089 /* 1090 * If there is a pending General Query response 1091 * scheduled sooner than the selected delay, no 1092 * further report need be scheduled. 1093 * Otherwise, prepare to respond to the 1094 * group-specific or group-and-source query. 1095 */ 1096 if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) 1097 igmp_input_v3_group_query(inm, igi, timer, igmpv3); 1098 } 1099 1100 out_locked: 1101 IGMP_UNLOCK(); 1102 IN_MULTI_UNLOCK(); 1103 1104 return (0); 1105 } 1106 1107 /* 1108 * Process a recieved IGMPv3 group-specific or group-and-source-specific 1109 * query. 1110 * Return <0 if any error occured. Currently this is ignored. 1111 */ 1112 static int 1113 igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi, 1114 int timer, /*const*/ struct igmpv3 *igmpv3) 1115 { 1116 int retval; 1117 uint16_t nsrc; 1118 1119 IN_MULTI_LOCK_ASSERT(); 1120 IGMP_LOCK_ASSERT(); 1121 1122 retval = 0; 1123 1124 switch (inm->inm_state) { 1125 case IGMP_NOT_MEMBER: 1126 case IGMP_SILENT_MEMBER: 1127 case IGMP_SLEEPING_MEMBER: 1128 case IGMP_LAZY_MEMBER: 1129 case IGMP_AWAKENING_MEMBER: 1130 case IGMP_IDLE_MEMBER: 1131 case IGMP_LEAVING_MEMBER: 1132 return (retval); 1133 break; 1134 case IGMP_REPORTING_MEMBER: 1135 case IGMP_G_QUERY_PENDING_MEMBER: 1136 case IGMP_SG_QUERY_PENDING_MEMBER: 1137 break; 1138 } 1139 1140 nsrc = ntohs(igmpv3->igmp_numsrc); 1141 1142 /* 1143 * Deal with group-specific queries upfront. 1144 * If any group query is already pending, purge any recorded 1145 * source-list state if it exists, and schedule a query response 1146 * for this group-specific query. 1147 */ 1148 if (nsrc == 0) { 1149 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || 1150 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) { 1151 inm_clear_recorded(inm); 1152 timer = min(inm->inm_timer, timer); 1153 } 1154 inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER; 1155 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1156 V_current_state_timers_running = 1; 1157 return (retval); 1158 } 1159 1160 /* 1161 * Deal with the case where a group-and-source-specific query has 1162 * been received but a group-specific query is already pending. 1163 */ 1164 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) { 1165 timer = min(inm->inm_timer, timer); 1166 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1167 V_current_state_timers_running = 1; 1168 return (retval); 1169 } 1170 1171 /* 1172 * Finally, deal with the case where a group-and-source-specific 1173 * query has been received, where a response to a previous g-s-r 1174 * query exists, or none exists. 1175 * In this case, we need to parse the source-list which the Querier 1176 * has provided us with and check if we have any source list filter 1177 * entries at T1 for these sources. If we do not, there is no need 1178 * schedule a report and the query may be dropped. 1179 * If we do, we must record them and schedule a current-state 1180 * report for those sources. 1181 * FIXME: Handling source lists larger than 1 mbuf requires that 1182 * we pass the mbuf chain pointer down to this function, and use 1183 * m_getptr() to walk the chain. 1184 */ 1185 if (inm->inm_nsrc > 0) { 1186 const struct in_addr *ap; 1187 int i, nrecorded; 1188 1189 ap = (const struct in_addr *)(igmpv3 + 1); 1190 nrecorded = 0; 1191 for (i = 0; i < nsrc; i++, ap++) { 1192 retval = inm_record_source(inm, ap->s_addr); 1193 if (retval < 0) 1194 break; 1195 nrecorded += retval; 1196 } 1197 if (nrecorded > 0) { 1198 CTR1(KTR_IGMPV3, 1199 "%s: schedule response to SG query", __func__); 1200 inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER; 1201 inm->inm_timer = IGMP_RANDOM_DELAY(timer); 1202 V_current_state_timers_running = 1; 1203 } 1204 } 1205 1206 return (retval); 1207 } 1208 1209 /* 1210 * Process a received IGMPv1 host membership report. 1211 * 1212 * NOTE: 0.0.0.0 workaround breaks const correctness. 1213 */ 1214 static int 1215 igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1216 /*const*/ struct igmp *igmp) 1217 { 1218 struct in_ifaddr *ia; 1219 struct in_multi *inm; 1220 1221 IGMPSTAT_INC(igps_rcv_reports); 1222 1223 if (ifp->if_flags & IFF_LOOPBACK) 1224 return (0); 1225 1226 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || 1227 !in_hosteq(igmp->igmp_group, ip->ip_dst)) { 1228 IGMPSTAT_INC(igps_rcv_badreports); 1229 return (EINVAL); 1230 } 1231 1232 /* 1233 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1234 * Booting clients may use the source address 0.0.0.0. Some 1235 * IGMP daemons may not know how to use IP_RECVIF to determine 1236 * the interface upon which this message was received. 1237 * Replace 0.0.0.0 with the subnet address if told to do so. 1238 */ 1239 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1240 IFP_TO_IA(ifp, ia); 1241 if (ia != NULL) { 1242 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1243 ifa_free(&ia->ia_ifa); 1244 } 1245 } 1246 1247 CTR3(KTR_IGMPV3, "process v1 report %s on ifp %p(%s)", 1248 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1249 1250 /* 1251 * IGMPv1 report suppression. 1252 * If we are a member of this group, and our membership should be 1253 * reported, stop our group timer and transition to the 'lazy' state. 1254 */ 1255 IN_MULTI_LOCK(); 1256 inm = inm_lookup(ifp, igmp->igmp_group); 1257 if (inm != NULL) { 1258 struct igmp_ifsoftc *igi; 1259 1260 igi = inm->inm_igi; 1261 if (igi == NULL) { 1262 KASSERT(igi != NULL, 1263 ("%s: no igi for ifp %p", __func__, ifp)); 1264 goto out_locked; 1265 } 1266 1267 IGMPSTAT_INC(igps_rcv_ourreports); 1268 1269 /* 1270 * If we are in IGMPv3 host mode, do not allow the 1271 * other host's IGMPv1 report to suppress our reports 1272 * unless explicitly configured to do so. 1273 */ 1274 if (igi->igi_version == IGMP_VERSION_3) { 1275 if (V_igmp_legacysupp) 1276 igmp_v3_suppress_group_record(inm); 1277 goto out_locked; 1278 } 1279 1280 inm->inm_timer = 0; 1281 1282 switch (inm->inm_state) { 1283 case IGMP_NOT_MEMBER: 1284 case IGMP_SILENT_MEMBER: 1285 break; 1286 case IGMP_IDLE_MEMBER: 1287 case IGMP_LAZY_MEMBER: 1288 case IGMP_AWAKENING_MEMBER: 1289 CTR3(KTR_IGMPV3, 1290 "report suppressed for %s on ifp %p(%s)", 1291 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1292 case IGMP_SLEEPING_MEMBER: 1293 inm->inm_state = IGMP_SLEEPING_MEMBER; 1294 break; 1295 case IGMP_REPORTING_MEMBER: 1296 CTR3(KTR_IGMPV3, 1297 "report suppressed for %s on ifp %p(%s)", 1298 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1299 if (igi->igi_version == IGMP_VERSION_1) 1300 inm->inm_state = IGMP_LAZY_MEMBER; 1301 else if (igi->igi_version == IGMP_VERSION_2) 1302 inm->inm_state = IGMP_SLEEPING_MEMBER; 1303 break; 1304 case IGMP_G_QUERY_PENDING_MEMBER: 1305 case IGMP_SG_QUERY_PENDING_MEMBER: 1306 case IGMP_LEAVING_MEMBER: 1307 break; 1308 } 1309 } 1310 1311 out_locked: 1312 IN_MULTI_UNLOCK(); 1313 1314 return (0); 1315 } 1316 1317 /* 1318 * Process a received IGMPv2 host membership report. 1319 * 1320 * NOTE: 0.0.0.0 workaround breaks const correctness. 1321 */ 1322 static int 1323 igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip, 1324 /*const*/ struct igmp *igmp) 1325 { 1326 struct in_ifaddr *ia; 1327 struct in_multi *inm; 1328 1329 /* 1330 * Make sure we don't hear our own membership report. Fast 1331 * leave requires knowing that we are the only member of a 1332 * group. 1333 */ 1334 IFP_TO_IA(ifp, ia); 1335 if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) { 1336 ifa_free(&ia->ia_ifa); 1337 return (0); 1338 } 1339 1340 IGMPSTAT_INC(igps_rcv_reports); 1341 1342 if (ifp->if_flags & IFF_LOOPBACK) { 1343 if (ia != NULL) 1344 ifa_free(&ia->ia_ifa); 1345 return (0); 1346 } 1347 1348 if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || 1349 !in_hosteq(igmp->igmp_group, ip->ip_dst)) { 1350 if (ia != NULL) 1351 ifa_free(&ia->ia_ifa); 1352 IGMPSTAT_INC(igps_rcv_badreports); 1353 return (EINVAL); 1354 } 1355 1356 /* 1357 * RFC 3376, Section 4.2.13, 9.2, 9.3: 1358 * Booting clients may use the source address 0.0.0.0. Some 1359 * IGMP daemons may not know how to use IP_RECVIF to determine 1360 * the interface upon which this message was received. 1361 * Replace 0.0.0.0 with the subnet address if told to do so. 1362 */ 1363 if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { 1364 if (ia != NULL) 1365 ip->ip_src.s_addr = htonl(ia->ia_subnet); 1366 } 1367 if (ia != NULL) 1368 ifa_free(&ia->ia_ifa); 1369 1370 CTR3(KTR_IGMPV3, "process v2 report %s on ifp %p(%s)", 1371 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1372 1373 /* 1374 * IGMPv2 report suppression. 1375 * If we are a member of this group, and our membership should be 1376 * reported, and our group timer is pending or about to be reset, 1377 * stop our group timer by transitioning to the 'lazy' state. 1378 */ 1379 IN_MULTI_LOCK(); 1380 inm = inm_lookup(ifp, igmp->igmp_group); 1381 if (inm != NULL) { 1382 struct igmp_ifsoftc *igi; 1383 1384 igi = inm->inm_igi; 1385 KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp)); 1386 1387 IGMPSTAT_INC(igps_rcv_ourreports); 1388 1389 /* 1390 * If we are in IGMPv3 host mode, do not allow the 1391 * other host's IGMPv1 report to suppress our reports 1392 * unless explicitly configured to do so. 1393 */ 1394 if (igi->igi_version == IGMP_VERSION_3) { 1395 if (V_igmp_legacysupp) 1396 igmp_v3_suppress_group_record(inm); 1397 goto out_locked; 1398 } 1399 1400 inm->inm_timer = 0; 1401 1402 switch (inm->inm_state) { 1403 case IGMP_NOT_MEMBER: 1404 case IGMP_SILENT_MEMBER: 1405 case IGMP_SLEEPING_MEMBER: 1406 break; 1407 case IGMP_REPORTING_MEMBER: 1408 case IGMP_IDLE_MEMBER: 1409 case IGMP_AWAKENING_MEMBER: 1410 CTR3(KTR_IGMPV3, 1411 "report suppressed for %s on ifp %p(%s)", 1412 inet_ntoa(igmp->igmp_group), ifp, ifp->if_xname); 1413 case IGMP_LAZY_MEMBER: 1414 inm->inm_state = IGMP_LAZY_MEMBER; 1415 break; 1416 case IGMP_G_QUERY_PENDING_MEMBER: 1417 case IGMP_SG_QUERY_PENDING_MEMBER: 1418 case IGMP_LEAVING_MEMBER: 1419 break; 1420 } 1421 } 1422 1423 out_locked: 1424 IN_MULTI_UNLOCK(); 1425 1426 return (0); 1427 } 1428 1429 int 1430 igmp_input(struct mbuf **mp, int *offp, int proto) 1431 { 1432 int iphlen; 1433 struct ifnet *ifp; 1434 struct igmp *igmp; 1435 struct ip *ip; 1436 struct mbuf *m; 1437 int igmplen; 1438 int minlen; 1439 int queryver; 1440 1441 CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp); 1442 1443 m = *mp; 1444 ifp = m->m_pkthdr.rcvif; 1445 *mp = NULL; 1446 1447 IGMPSTAT_INC(igps_rcv_total); 1448 1449 ip = mtod(m, struct ip *); 1450 iphlen = *offp; 1451 igmplen = ntohs(ip->ip_len) - iphlen; 1452 1453 /* 1454 * Validate lengths. 1455 */ 1456 if (igmplen < IGMP_MINLEN) { 1457 IGMPSTAT_INC(igps_rcv_tooshort); 1458 m_freem(m); 1459 return (IPPROTO_DONE); 1460 } 1461 1462 /* 1463 * Always pullup to the minimum size for v1/v2 or v3 1464 * to amortize calls to m_pullup(). 1465 */ 1466 minlen = iphlen; 1467 if (igmplen >= IGMP_V3_QUERY_MINLEN) 1468 minlen += IGMP_V3_QUERY_MINLEN; 1469 else 1470 minlen += IGMP_MINLEN; 1471 if ((!M_WRITABLE(m) || m->m_len < minlen) && 1472 (m = m_pullup(m, minlen)) == 0) { 1473 IGMPSTAT_INC(igps_rcv_tooshort); 1474 return (IPPROTO_DONE); 1475 } 1476 ip = mtod(m, struct ip *); 1477 1478 /* 1479 * Validate checksum. 1480 */ 1481 m->m_data += iphlen; 1482 m->m_len -= iphlen; 1483 igmp = mtod(m, struct igmp *); 1484 if (in_cksum(m, igmplen)) { 1485 IGMPSTAT_INC(igps_rcv_badsum); 1486 m_freem(m); 1487 return (IPPROTO_DONE); 1488 } 1489 m->m_data -= iphlen; 1490 m->m_len += iphlen; 1491 1492 /* 1493 * IGMP control traffic is link-scope, and must have a TTL of 1. 1494 * DVMRP traffic (e.g. mrinfo, mtrace) is an exception; 1495 * probe packets may come from beyond the LAN. 1496 */ 1497 if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) { 1498 IGMPSTAT_INC(igps_rcv_badttl); 1499 m_freem(m); 1500 return (IPPROTO_DONE); 1501 } 1502 1503 switch (igmp->igmp_type) { 1504 case IGMP_HOST_MEMBERSHIP_QUERY: 1505 if (igmplen == IGMP_MINLEN) { 1506 if (igmp->igmp_code == 0) 1507 queryver = IGMP_VERSION_1; 1508 else 1509 queryver = IGMP_VERSION_2; 1510 } else if (igmplen >= IGMP_V3_QUERY_MINLEN) { 1511 queryver = IGMP_VERSION_3; 1512 } else { 1513 IGMPSTAT_INC(igps_rcv_tooshort); 1514 m_freem(m); 1515 return (IPPROTO_DONE); 1516 } 1517 1518 switch (queryver) { 1519 case IGMP_VERSION_1: 1520 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1521 if (!V_igmp_v1enable) 1522 break; 1523 if (igmp_input_v1_query(ifp, ip, igmp) != 0) { 1524 m_freem(m); 1525 return (IPPROTO_DONE); 1526 } 1527 break; 1528 1529 case IGMP_VERSION_2: 1530 IGMPSTAT_INC(igps_rcv_v1v2_queries); 1531 if (!V_igmp_v2enable) 1532 break; 1533 if (igmp_input_v2_query(ifp, ip, igmp) != 0) { 1534 m_freem(m); 1535 return (IPPROTO_DONE); 1536 } 1537 break; 1538 1539 case IGMP_VERSION_3: { 1540 struct igmpv3 *igmpv3; 1541 uint16_t igmpv3len; 1542 uint16_t nsrc; 1543 int srclen; 1544 1545 IGMPSTAT_INC(igps_rcv_v3_queries); 1546 igmpv3 = (struct igmpv3 *)igmp; 1547 /* 1548 * Validate length based on source count. 1549 */ 1550 nsrc = ntohs(igmpv3->igmp_numsrc); 1551 srclen = sizeof(struct in_addr) * nsrc; 1552 if (nsrc * sizeof(in_addr_t) > srclen) { 1553 IGMPSTAT_INC(igps_rcv_tooshort); 1554 return (IPPROTO_DONE); 1555 } 1556 /* 1557 * m_pullup() may modify m, so pullup in 1558 * this scope. 1559 */ 1560 igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN + 1561 srclen; 1562 if ((!M_WRITABLE(m) || 1563 m->m_len < igmpv3len) && 1564 (m = m_pullup(m, igmpv3len)) == NULL) { 1565 IGMPSTAT_INC(igps_rcv_tooshort); 1566 return (IPPROTO_DONE); 1567 } 1568 igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *) 1569 + iphlen); 1570 if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) { 1571 m_freem(m); 1572 return (IPPROTO_DONE); 1573 } 1574 } 1575 break; 1576 } 1577 break; 1578 1579 case IGMP_v1_HOST_MEMBERSHIP_REPORT: 1580 if (!V_igmp_v1enable) 1581 break; 1582 if (igmp_input_v1_report(ifp, ip, igmp) != 0) { 1583 m_freem(m); 1584 return (IPPROTO_DONE); 1585 } 1586 break; 1587 1588 case IGMP_v2_HOST_MEMBERSHIP_REPORT: 1589 if (!V_igmp_v2enable) 1590 break; 1591 if (!ip_checkrouteralert(m)) 1592 IGMPSTAT_INC(igps_rcv_nora); 1593 if (igmp_input_v2_report(ifp, ip, igmp) != 0) { 1594 m_freem(m); 1595 return (IPPROTO_DONE); 1596 } 1597 break; 1598 1599 case IGMP_v3_HOST_MEMBERSHIP_REPORT: 1600 /* 1601 * Hosts do not need to process IGMPv3 membership reports, 1602 * as report suppression is no longer required. 1603 */ 1604 if (!ip_checkrouteralert(m)) 1605 IGMPSTAT_INC(igps_rcv_nora); 1606 break; 1607 1608 default: 1609 break; 1610 } 1611 1612 /* 1613 * Pass all valid IGMP packets up to any process(es) listening on a 1614 * raw IGMP socket. 1615 */ 1616 *mp = m; 1617 return (rip_input(mp, offp, proto)); 1618 } 1619 1620 1621 /* 1622 * Fast timeout handler (global). 1623 * VIMAGE: Timeout handlers are expected to service all vimages. 1624 */ 1625 void 1626 igmp_fasttimo(void) 1627 { 1628 VNET_ITERATOR_DECL(vnet_iter); 1629 1630 VNET_LIST_RLOCK_NOSLEEP(); 1631 VNET_FOREACH(vnet_iter) { 1632 CURVNET_SET(vnet_iter); 1633 igmp_fasttimo_vnet(); 1634 CURVNET_RESTORE(); 1635 } 1636 VNET_LIST_RUNLOCK_NOSLEEP(); 1637 } 1638 1639 /* 1640 * Fast timeout handler (per-vnet). 1641 * Sends are shuffled off to a netisr to deal with Giant. 1642 * 1643 * VIMAGE: Assume caller has set up our curvnet. 1644 */ 1645 static void 1646 igmp_fasttimo_vnet(void) 1647 { 1648 struct mbufq scq; /* State-change packets */ 1649 struct mbufq qrq; /* Query response packets */ 1650 struct ifnet *ifp; 1651 struct igmp_ifsoftc *igi; 1652 struct ifmultiaddr *ifma; 1653 struct in_multi *inm; 1654 int loop, uri_fasthz; 1655 1656 loop = 0; 1657 uri_fasthz = 0; 1658 1659 /* 1660 * Quick check to see if any work needs to be done, in order to 1661 * minimize the overhead of fasttimo processing. 1662 * SMPng: XXX Unlocked reads. 1663 */ 1664 if (!V_current_state_timers_running && 1665 !V_interface_timers_running && 1666 !V_state_change_timers_running) 1667 return; 1668 1669 IN_MULTI_LOCK(); 1670 IGMP_LOCK(); 1671 1672 /* 1673 * IGMPv3 General Query response timer processing. 1674 */ 1675 if (V_interface_timers_running) { 1676 CTR1(KTR_IGMPV3, "%s: interface timers running", __func__); 1677 1678 V_interface_timers_running = 0; 1679 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1680 if (igi->igi_v3_timer == 0) { 1681 /* Do nothing. */ 1682 } else if (--igi->igi_v3_timer == 0) { 1683 igmp_v3_dispatch_general_query(igi); 1684 } else { 1685 V_interface_timers_running = 1; 1686 } 1687 } 1688 } 1689 1690 if (!V_current_state_timers_running && 1691 !V_state_change_timers_running) 1692 goto out_locked; 1693 1694 V_current_state_timers_running = 0; 1695 V_state_change_timers_running = 0; 1696 1697 CTR1(KTR_IGMPV3, "%s: state change timers running", __func__); 1698 1699 /* 1700 * IGMPv1/v2/v3 host report and state-change timer processing. 1701 * Note: Processing a v3 group timer may remove a node. 1702 */ 1703 LIST_FOREACH(igi, &V_igi_head, igi_link) { 1704 ifp = igi->igi_ifp; 1705 1706 if (igi->igi_version == IGMP_VERSION_3) { 1707 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; 1708 uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri * 1709 PR_FASTHZ); 1710 mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS); 1711 mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS); 1712 } 1713 1714 IF_ADDR_RLOCK(ifp); 1715 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1716 if (ifma->ifma_addr->sa_family != AF_INET || 1717 ifma->ifma_protospec == NULL) 1718 continue; 1719 inm = (struct in_multi *)ifma->ifma_protospec; 1720 switch (igi->igi_version) { 1721 case IGMP_VERSION_1: 1722 case IGMP_VERSION_2: 1723 igmp_v1v2_process_group_timer(inm, 1724 igi->igi_version); 1725 break; 1726 case IGMP_VERSION_3: 1727 igmp_v3_process_group_timers(igi, &qrq, 1728 &scq, inm, uri_fasthz); 1729 break; 1730 } 1731 } 1732 IF_ADDR_RUNLOCK(ifp); 1733 1734 if (igi->igi_version == IGMP_VERSION_3) { 1735 struct in_multi *tinm; 1736 1737 igmp_dispatch_queue(&qrq, 0, loop); 1738 igmp_dispatch_queue(&scq, 0, loop); 1739 1740 /* 1741 * Free the in_multi reference(s) for this 1742 * IGMP lifecycle. 1743 */ 1744 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, 1745 inm_nrele, tinm) { 1746 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, 1747 inm_nrele); 1748 inm_release_locked(inm); 1749 } 1750 } 1751 } 1752 1753 out_locked: 1754 IGMP_UNLOCK(); 1755 IN_MULTI_UNLOCK(); 1756 } 1757 1758 /* 1759 * Update host report group timer for IGMPv1/v2. 1760 * Will update the global pending timer flags. 1761 */ 1762 static void 1763 igmp_v1v2_process_group_timer(struct in_multi *inm, const int version) 1764 { 1765 int report_timer_expired; 1766 1767 IN_MULTI_LOCK_ASSERT(); 1768 IGMP_LOCK_ASSERT(); 1769 1770 if (inm->inm_timer == 0) { 1771 report_timer_expired = 0; 1772 } else if (--inm->inm_timer == 0) { 1773 report_timer_expired = 1; 1774 } else { 1775 V_current_state_timers_running = 1; 1776 return; 1777 } 1778 1779 switch (inm->inm_state) { 1780 case IGMP_NOT_MEMBER: 1781 case IGMP_SILENT_MEMBER: 1782 case IGMP_IDLE_MEMBER: 1783 case IGMP_LAZY_MEMBER: 1784 case IGMP_SLEEPING_MEMBER: 1785 case IGMP_AWAKENING_MEMBER: 1786 break; 1787 case IGMP_REPORTING_MEMBER: 1788 if (report_timer_expired) { 1789 inm->inm_state = IGMP_IDLE_MEMBER; 1790 (void)igmp_v1v2_queue_report(inm, 1791 (version == IGMP_VERSION_2) ? 1792 IGMP_v2_HOST_MEMBERSHIP_REPORT : 1793 IGMP_v1_HOST_MEMBERSHIP_REPORT); 1794 } 1795 break; 1796 case IGMP_G_QUERY_PENDING_MEMBER: 1797 case IGMP_SG_QUERY_PENDING_MEMBER: 1798 case IGMP_LEAVING_MEMBER: 1799 break; 1800 } 1801 } 1802 1803 /* 1804 * Update a group's timers for IGMPv3. 1805 * Will update the global pending timer flags. 1806 * Note: Unlocked read from igi. 1807 */ 1808 static void 1809 igmp_v3_process_group_timers(struct igmp_ifsoftc *igi, 1810 struct mbufq *qrq, struct mbufq *scq, 1811 struct in_multi *inm, const int uri_fasthz) 1812 { 1813 int query_response_timer_expired; 1814 int state_change_retransmit_timer_expired; 1815 1816 IN_MULTI_LOCK_ASSERT(); 1817 IGMP_LOCK_ASSERT(); 1818 1819 query_response_timer_expired = 0; 1820 state_change_retransmit_timer_expired = 0; 1821 1822 /* 1823 * During a transition from v1/v2 compatibility mode back to v3, 1824 * a group record in REPORTING state may still have its group 1825 * timer active. This is a no-op in this function; it is easier 1826 * to deal with it here than to complicate the slow-timeout path. 1827 */ 1828 if (inm->inm_timer == 0) { 1829 query_response_timer_expired = 0; 1830 } else if (--inm->inm_timer == 0) { 1831 query_response_timer_expired = 1; 1832 } else { 1833 V_current_state_timers_running = 1; 1834 } 1835 1836 if (inm->inm_sctimer == 0) { 1837 state_change_retransmit_timer_expired = 0; 1838 } else if (--inm->inm_sctimer == 0) { 1839 state_change_retransmit_timer_expired = 1; 1840 } else { 1841 V_state_change_timers_running = 1; 1842 } 1843 1844 /* We are in fasttimo, so be quick about it. */ 1845 if (!state_change_retransmit_timer_expired && 1846 !query_response_timer_expired) 1847 return; 1848 1849 switch (inm->inm_state) { 1850 case IGMP_NOT_MEMBER: 1851 case IGMP_SILENT_MEMBER: 1852 case IGMP_SLEEPING_MEMBER: 1853 case IGMP_LAZY_MEMBER: 1854 case IGMP_AWAKENING_MEMBER: 1855 case IGMP_IDLE_MEMBER: 1856 break; 1857 case IGMP_G_QUERY_PENDING_MEMBER: 1858 case IGMP_SG_QUERY_PENDING_MEMBER: 1859 /* 1860 * Respond to a previously pending Group-Specific 1861 * or Group-and-Source-Specific query by enqueueing 1862 * the appropriate Current-State report for 1863 * immediate transmission. 1864 */ 1865 if (query_response_timer_expired) { 1866 int retval; 1867 1868 retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1, 1869 (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)); 1870 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 1871 __func__, retval); 1872 inm->inm_state = IGMP_REPORTING_MEMBER; 1873 /* XXX Clear recorded sources for next time. */ 1874 inm_clear_recorded(inm); 1875 } 1876 /* FALLTHROUGH */ 1877 case IGMP_REPORTING_MEMBER: 1878 case IGMP_LEAVING_MEMBER: 1879 if (state_change_retransmit_timer_expired) { 1880 /* 1881 * State-change retransmission timer fired. 1882 * If there are any further pending retransmissions, 1883 * set the global pending state-change flag, and 1884 * reset the timer. 1885 */ 1886 if (--inm->inm_scrv > 0) { 1887 inm->inm_sctimer = uri_fasthz; 1888 V_state_change_timers_running = 1; 1889 } 1890 /* 1891 * Retransmit the previously computed state-change 1892 * report. If there are no further pending 1893 * retransmissions, the mbuf queue will be consumed. 1894 * Update T0 state to T1 as we have now sent 1895 * a state-change. 1896 */ 1897 (void)igmp_v3_merge_state_changes(inm, scq); 1898 1899 inm_commit(inm); 1900 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__, 1901 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname); 1902 1903 /* 1904 * If we are leaving the group for good, make sure 1905 * we release IGMP's reference to it. 1906 * This release must be deferred using a SLIST, 1907 * as we are called from a loop which traverses 1908 * the in_ifmultiaddr TAILQ. 1909 */ 1910 if (inm->inm_state == IGMP_LEAVING_MEMBER && 1911 inm->inm_scrv == 0) { 1912 inm->inm_state = IGMP_NOT_MEMBER; 1913 SLIST_INSERT_HEAD(&igi->igi_relinmhead, 1914 inm, inm_nrele); 1915 } 1916 } 1917 break; 1918 } 1919 } 1920 1921 1922 /* 1923 * Suppress a group's pending response to a group or source/group query. 1924 * 1925 * Do NOT suppress state changes. This leads to IGMPv3 inconsistency. 1926 * Do NOT update ST1/ST0 as this operation merely suppresses 1927 * the currently pending group record. 1928 * Do NOT suppress the response to a general query. It is possible but 1929 * it would require adding another state or flag. 1930 */ 1931 static void 1932 igmp_v3_suppress_group_record(struct in_multi *inm) 1933 { 1934 1935 IN_MULTI_LOCK_ASSERT(); 1936 1937 KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3, 1938 ("%s: not IGMPv3 mode on link", __func__)); 1939 1940 if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER || 1941 inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER) 1942 return; 1943 1944 if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) 1945 inm_clear_recorded(inm); 1946 1947 inm->inm_timer = 0; 1948 inm->inm_state = IGMP_REPORTING_MEMBER; 1949 } 1950 1951 /* 1952 * Switch to a different IGMP version on the given interface, 1953 * as per Section 7.2.1. 1954 */ 1955 static void 1956 igmp_set_version(struct igmp_ifsoftc *igi, const int version) 1957 { 1958 int old_version_timer; 1959 1960 IGMP_LOCK_ASSERT(); 1961 1962 CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__, 1963 version, igi->igi_ifp, igi->igi_ifp->if_xname); 1964 1965 if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) { 1966 /* 1967 * Compute the "Older Version Querier Present" timer as per 1968 * Section 8.12. 1969 */ 1970 old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri; 1971 old_version_timer *= PR_SLOWHZ; 1972 1973 if (version == IGMP_VERSION_1) { 1974 igi->igi_v1_timer = old_version_timer; 1975 igi->igi_v2_timer = 0; 1976 } else if (version == IGMP_VERSION_2) { 1977 igi->igi_v1_timer = 0; 1978 igi->igi_v2_timer = old_version_timer; 1979 } 1980 } 1981 1982 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 1983 if (igi->igi_version != IGMP_VERSION_2) { 1984 igi->igi_version = IGMP_VERSION_2; 1985 igmp_v3_cancel_link_timers(igi); 1986 } 1987 } else if (igi->igi_v1_timer > 0) { 1988 if (igi->igi_version != IGMP_VERSION_1) { 1989 igi->igi_version = IGMP_VERSION_1; 1990 igmp_v3_cancel_link_timers(igi); 1991 } 1992 } 1993 } 1994 1995 /* 1996 * Cancel pending IGMPv3 timers for the given link and all groups 1997 * joined on it; state-change, general-query, and group-query timers. 1998 * 1999 * Only ever called on a transition from v3 to Compatibility mode. Kill 2000 * the timers stone dead (this may be expensive for large N groups), they 2001 * will be restarted if Compatibility Mode deems that they must be due to 2002 * query processing. 2003 */ 2004 static void 2005 igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi) 2006 { 2007 struct ifmultiaddr *ifma; 2008 struct ifnet *ifp; 2009 struct in_multi *inm, *tinm; 2010 2011 CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__, 2012 igi->igi_ifp, igi->igi_ifp->if_xname); 2013 2014 IN_MULTI_LOCK_ASSERT(); 2015 IGMP_LOCK_ASSERT(); 2016 2017 /* 2018 * Stop the v3 General Query Response on this link stone dead. 2019 * If fasttimo is woken up due to V_interface_timers_running, 2020 * the flag will be cleared if there are no pending link timers. 2021 */ 2022 igi->igi_v3_timer = 0; 2023 2024 /* 2025 * Now clear the current-state and state-change report timers 2026 * for all memberships scoped to this link. 2027 */ 2028 ifp = igi->igi_ifp; 2029 IF_ADDR_RLOCK(ifp); 2030 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2031 if (ifma->ifma_addr->sa_family != AF_INET || 2032 ifma->ifma_protospec == NULL) 2033 continue; 2034 inm = (struct in_multi *)ifma->ifma_protospec; 2035 switch (inm->inm_state) { 2036 case IGMP_NOT_MEMBER: 2037 case IGMP_SILENT_MEMBER: 2038 case IGMP_IDLE_MEMBER: 2039 case IGMP_LAZY_MEMBER: 2040 case IGMP_SLEEPING_MEMBER: 2041 case IGMP_AWAKENING_MEMBER: 2042 /* 2043 * These states are either not relevant in v3 mode, 2044 * or are unreported. Do nothing. 2045 */ 2046 break; 2047 case IGMP_LEAVING_MEMBER: 2048 /* 2049 * If we are leaving the group and switching to 2050 * compatibility mode, we need to release the final 2051 * reference held for issuing the INCLUDE {}, and 2052 * transition to REPORTING to ensure the host leave 2053 * message is sent upstream to the old querier -- 2054 * transition to NOT would lose the leave and race. 2055 */ 2056 SLIST_INSERT_HEAD(&igi->igi_relinmhead, inm, inm_nrele); 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 mbufq_drain(&inm->inm_scq); 2073 } 2074 IF_ADDR_RUNLOCK(ifp); 2075 SLIST_FOREACH_SAFE(inm, &igi->igi_relinmhead, inm_nrele, tinm) { 2076 SLIST_REMOVE_HEAD(&igi->igi_relinmhead, inm_nrele); 2077 inm_release_locked(inm); 2078 } 2079 } 2080 2081 /* 2082 * Update the Older Version Querier Present timers for a link. 2083 * See Section 7.2.1 of RFC 3376. 2084 */ 2085 static void 2086 igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi) 2087 { 2088 2089 IGMP_LOCK_ASSERT(); 2090 2091 if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) { 2092 /* 2093 * IGMPv1 and IGMPv2 Querier Present timers expired. 2094 * 2095 * Revert to IGMPv3. 2096 */ 2097 if (igi->igi_version != IGMP_VERSION_3) { 2098 CTR5(KTR_IGMPV3, 2099 "%s: transition from v%d -> v%d on %p(%s)", 2100 __func__, igi->igi_version, IGMP_VERSION_3, 2101 igi->igi_ifp, igi->igi_ifp->if_xname); 2102 igi->igi_version = IGMP_VERSION_3; 2103 } 2104 } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { 2105 /* 2106 * IGMPv1 Querier Present timer expired, 2107 * IGMPv2 Querier Present timer running. 2108 * If IGMPv2 was disabled since last timeout, 2109 * revert to IGMPv3. 2110 * If IGMPv2 is enabled, revert to IGMPv2. 2111 */ 2112 if (!V_igmp_v2enable) { 2113 CTR5(KTR_IGMPV3, 2114 "%s: transition from v%d -> v%d on %p(%s)", 2115 __func__, igi->igi_version, IGMP_VERSION_3, 2116 igi->igi_ifp, igi->igi_ifp->if_xname); 2117 igi->igi_v2_timer = 0; 2118 igi->igi_version = IGMP_VERSION_3; 2119 } else { 2120 --igi->igi_v2_timer; 2121 if (igi->igi_version != IGMP_VERSION_2) { 2122 CTR5(KTR_IGMPV3, 2123 "%s: transition from v%d -> v%d on %p(%s)", 2124 __func__, igi->igi_version, IGMP_VERSION_2, 2125 igi->igi_ifp, igi->igi_ifp->if_xname); 2126 igi->igi_version = IGMP_VERSION_2; 2127 igmp_v3_cancel_link_timers(igi); 2128 } 2129 } 2130 } else if (igi->igi_v1_timer > 0) { 2131 /* 2132 * IGMPv1 Querier Present timer running. 2133 * Stop IGMPv2 timer if running. 2134 * 2135 * If IGMPv1 was disabled since last timeout, 2136 * revert to IGMPv3. 2137 * If IGMPv1 is enabled, reset IGMPv2 timer if running. 2138 */ 2139 if (!V_igmp_v1enable) { 2140 CTR5(KTR_IGMPV3, 2141 "%s: transition from v%d -> v%d on %p(%s)", 2142 __func__, igi->igi_version, IGMP_VERSION_3, 2143 igi->igi_ifp, igi->igi_ifp->if_xname); 2144 igi->igi_v1_timer = 0; 2145 igi->igi_version = IGMP_VERSION_3; 2146 } else { 2147 --igi->igi_v1_timer; 2148 } 2149 if (igi->igi_v2_timer > 0) { 2150 CTR3(KTR_IGMPV3, 2151 "%s: cancel v2 timer on %p(%s)", 2152 __func__, igi->igi_ifp, igi->igi_ifp->if_xname); 2153 igi->igi_v2_timer = 0; 2154 } 2155 } 2156 } 2157 2158 /* 2159 * Global slowtimo handler. 2160 * VIMAGE: Timeout handlers are expected to service all vimages. 2161 */ 2162 void 2163 igmp_slowtimo(void) 2164 { 2165 VNET_ITERATOR_DECL(vnet_iter); 2166 2167 VNET_LIST_RLOCK_NOSLEEP(); 2168 VNET_FOREACH(vnet_iter) { 2169 CURVNET_SET(vnet_iter); 2170 igmp_slowtimo_vnet(); 2171 CURVNET_RESTORE(); 2172 } 2173 VNET_LIST_RUNLOCK_NOSLEEP(); 2174 } 2175 2176 /* 2177 * Per-vnet slowtimo handler. 2178 */ 2179 static void 2180 igmp_slowtimo_vnet(void) 2181 { 2182 struct igmp_ifsoftc *igi; 2183 2184 IGMP_LOCK(); 2185 2186 LIST_FOREACH(igi, &V_igi_head, igi_link) { 2187 igmp_v1v2_process_querier_timers(igi); 2188 } 2189 2190 IGMP_UNLOCK(); 2191 } 2192 2193 /* 2194 * Dispatch an IGMPv1/v2 host report or leave message. 2195 * These are always small enough to fit inside a single mbuf. 2196 */ 2197 static int 2198 igmp_v1v2_queue_report(struct in_multi *inm, const int type) 2199 { 2200 struct ifnet *ifp; 2201 struct igmp *igmp; 2202 struct ip *ip; 2203 struct mbuf *m; 2204 2205 IN_MULTI_LOCK_ASSERT(); 2206 IGMP_LOCK_ASSERT(); 2207 2208 ifp = inm->inm_ifp; 2209 2210 m = m_gethdr(M_NOWAIT, MT_DATA); 2211 if (m == NULL) 2212 return (ENOMEM); 2213 M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp)); 2214 2215 m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp); 2216 2217 m->m_data += sizeof(struct ip); 2218 m->m_len = sizeof(struct igmp); 2219 2220 igmp = mtod(m, struct igmp *); 2221 igmp->igmp_type = type; 2222 igmp->igmp_code = 0; 2223 igmp->igmp_group = inm->inm_addr; 2224 igmp->igmp_cksum = 0; 2225 igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp)); 2226 2227 m->m_data -= sizeof(struct ip); 2228 m->m_len += sizeof(struct ip); 2229 2230 ip = mtod(m, struct ip *); 2231 ip->ip_tos = 0; 2232 ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp)); 2233 ip->ip_off = 0; 2234 ip->ip_p = IPPROTO_IGMP; 2235 ip->ip_src.s_addr = INADDR_ANY; 2236 2237 if (type == IGMP_HOST_LEAVE_MESSAGE) 2238 ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP); 2239 else 2240 ip->ip_dst = inm->inm_addr; 2241 2242 igmp_save_context(m, ifp); 2243 2244 m->m_flags |= M_IGMPV2; 2245 if (inm->inm_igi->igi_flags & IGIF_LOOPBACK) 2246 m->m_flags |= M_IGMP_LOOP; 2247 2248 CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m); 2249 netisr_dispatch(NETISR_IGMP, m); 2250 2251 return (0); 2252 } 2253 2254 /* 2255 * Process a state change from the upper layer for the given IPv4 group. 2256 * 2257 * Each socket holds a reference on the in_multi in its own ip_moptions. 2258 * The socket layer will have made the necessary updates to.the group 2259 * state, it is now up to IGMP to issue a state change report if there 2260 * has been any change between T0 (when the last state-change was issued) 2261 * and T1 (now). 2262 * 2263 * We use the IGMPv3 state machine at group level. The IGMP module 2264 * however makes the decision as to which IGMP protocol version to speak. 2265 * A state change *from* INCLUDE {} always means an initial join. 2266 * A state change *to* INCLUDE {} always means a final leave. 2267 * 2268 * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can 2269 * save ourselves a bunch of work; any exclusive mode groups need not 2270 * compute source filter lists. 2271 * 2272 * VIMAGE: curvnet should have been set by caller, as this routine 2273 * is called from the socket option handlers. 2274 */ 2275 int 2276 igmp_change_state(struct in_multi *inm) 2277 { 2278 struct igmp_ifsoftc *igi; 2279 struct ifnet *ifp; 2280 int error; 2281 2282 IN_MULTI_LOCK_ASSERT(); 2283 2284 error = 0; 2285 2286 /* 2287 * Try to detect if the upper layer just asked us to change state 2288 * for an interface which has now gone away. 2289 */ 2290 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 2291 ifp = inm->inm_ifma->ifma_ifp; 2292 /* 2293 * Sanity check that netinet's notion of ifp is the 2294 * same as net's. 2295 */ 2296 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 2297 2298 IGMP_LOCK(); 2299 2300 igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; 2301 KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); 2302 2303 /* 2304 * If we detect a state transition to or from MCAST_UNDEFINED 2305 * for this group, then we are starting or finishing an IGMP 2306 * life cycle for this group. 2307 */ 2308 if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) { 2309 CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__, 2310 inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode); 2311 if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) { 2312 CTR1(KTR_IGMPV3, "%s: initial join", __func__); 2313 error = igmp_initial_join(inm, igi); 2314 goto out_locked; 2315 } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) { 2316 CTR1(KTR_IGMPV3, "%s: final leave", __func__); 2317 igmp_final_leave(inm, igi); 2318 goto out_locked; 2319 } 2320 } else { 2321 CTR1(KTR_IGMPV3, "%s: filter set change", __func__); 2322 } 2323 2324 error = igmp_handle_state_change(inm, igi); 2325 2326 out_locked: 2327 IGMP_UNLOCK(); 2328 return (error); 2329 } 2330 2331 /* 2332 * Perform the initial join for an IGMP group. 2333 * 2334 * When joining a group: 2335 * If the group should have its IGMP traffic suppressed, do nothing. 2336 * IGMPv1 starts sending IGMPv1 host membership reports. 2337 * IGMPv2 starts sending IGMPv2 host membership reports. 2338 * IGMPv3 will schedule an IGMPv3 state-change report containing the 2339 * initial state of the membership. 2340 */ 2341 static int 2342 igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi) 2343 { 2344 struct ifnet *ifp; 2345 struct mbufq *mq; 2346 int error, retval, syncstates; 2347 2348 CTR4(KTR_IGMPV3, "%s: initial join %s on ifp %p(%s)", 2349 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp, 2350 inm->inm_ifp->if_xname); 2351 2352 error = 0; 2353 syncstates = 1; 2354 2355 ifp = inm->inm_ifp; 2356 2357 IN_MULTI_LOCK_ASSERT(); 2358 IGMP_LOCK_ASSERT(); 2359 2360 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); 2361 2362 /* 2363 * Groups joined on loopback or marked as 'not reported', 2364 * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and 2365 * are never reported in any IGMP protocol exchanges. 2366 * All other groups enter the appropriate IGMP state machine 2367 * for the version in use on this link. 2368 * A link marked as IGIF_SILENT causes IGMP to be completely 2369 * disabled for the link. 2370 */ 2371 if ((ifp->if_flags & IFF_LOOPBACK) || 2372 (igi->igi_flags & IGIF_SILENT) || 2373 !igmp_isgroupreported(inm->inm_addr)) { 2374 CTR1(KTR_IGMPV3, 2375 "%s: not kicking state machine for silent group", __func__); 2376 inm->inm_state = IGMP_SILENT_MEMBER; 2377 inm->inm_timer = 0; 2378 } else { 2379 /* 2380 * Deal with overlapping in_multi lifecycle. 2381 * If this group was LEAVING, then make sure 2382 * we drop the reference we picked up to keep the 2383 * group around for the final INCLUDE {} enqueue. 2384 */ 2385 if (igi->igi_version == IGMP_VERSION_3 && 2386 inm->inm_state == IGMP_LEAVING_MEMBER) 2387 inm_release_locked(inm); 2388 2389 inm->inm_state = IGMP_REPORTING_MEMBER; 2390 2391 switch (igi->igi_version) { 2392 case IGMP_VERSION_1: 2393 case IGMP_VERSION_2: 2394 inm->inm_state = IGMP_IDLE_MEMBER; 2395 error = igmp_v1v2_queue_report(inm, 2396 (igi->igi_version == IGMP_VERSION_2) ? 2397 IGMP_v2_HOST_MEMBERSHIP_REPORT : 2398 IGMP_v1_HOST_MEMBERSHIP_REPORT); 2399 if (error == 0) { 2400 inm->inm_timer = IGMP_RANDOM_DELAY( 2401 IGMP_V1V2_MAX_RI * PR_FASTHZ); 2402 V_current_state_timers_running = 1; 2403 } 2404 break; 2405 2406 case IGMP_VERSION_3: 2407 /* 2408 * Defer update of T0 to T1, until the first copy 2409 * of the state change has been transmitted. 2410 */ 2411 syncstates = 0; 2412 2413 /* 2414 * Immediately enqueue a State-Change Report for 2415 * this interface, freeing any previous reports. 2416 * Don't kick the timers if there is nothing to do, 2417 * or if an error occurred. 2418 */ 2419 mq = &inm->inm_scq; 2420 mbufq_drain(mq); 2421 retval = igmp_v3_enqueue_group_record(mq, inm, 1, 2422 0, 0); 2423 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 2424 __func__, retval); 2425 if (retval <= 0) { 2426 error = retval * -1; 2427 break; 2428 } 2429 2430 /* 2431 * Schedule transmission of pending state-change 2432 * report up to RV times for this link. The timer 2433 * will fire at the next igmp_fasttimo (~200ms), 2434 * giving us an opportunity to merge the reports. 2435 */ 2436 if (igi->igi_flags & IGIF_LOOPBACK) { 2437 inm->inm_scrv = 1; 2438 } else { 2439 KASSERT(igi->igi_rv > 1, 2440 ("%s: invalid robustness %d", __func__, 2441 igi->igi_rv)); 2442 inm->inm_scrv = igi->igi_rv; 2443 } 2444 inm->inm_sctimer = 1; 2445 V_state_change_timers_running = 1; 2446 2447 error = 0; 2448 break; 2449 } 2450 } 2451 2452 /* 2453 * Only update the T0 state if state change is atomic, 2454 * i.e. we don't need to wait for a timer to fire before we 2455 * can consider the state change to have been communicated. 2456 */ 2457 if (syncstates) { 2458 inm_commit(inm); 2459 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__, 2460 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname); 2461 } 2462 2463 return (error); 2464 } 2465 2466 /* 2467 * Issue an intermediate state change during the IGMP life-cycle. 2468 */ 2469 static int 2470 igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi) 2471 { 2472 struct ifnet *ifp; 2473 int retval; 2474 2475 CTR4(KTR_IGMPV3, "%s: state change for %s on ifp %p(%s)", 2476 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp, 2477 inm->inm_ifp->if_xname); 2478 2479 ifp = inm->inm_ifp; 2480 2481 IN_MULTI_LOCK_ASSERT(); 2482 IGMP_LOCK_ASSERT(); 2483 2484 KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); 2485 2486 if ((ifp->if_flags & IFF_LOOPBACK) || 2487 (igi->igi_flags & IGIF_SILENT) || 2488 !igmp_isgroupreported(inm->inm_addr) || 2489 (igi->igi_version != IGMP_VERSION_3)) { 2490 if (!igmp_isgroupreported(inm->inm_addr)) { 2491 CTR1(KTR_IGMPV3, 2492 "%s: not kicking state machine for silent group", __func__); 2493 } 2494 CTR1(KTR_IGMPV3, "%s: nothing to do", __func__); 2495 inm_commit(inm); 2496 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__, 2497 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname); 2498 return (0); 2499 } 2500 2501 mbufq_drain(&inm->inm_scq); 2502 2503 retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0); 2504 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); 2505 if (retval <= 0) 2506 return (-retval); 2507 2508 /* 2509 * If record(s) were enqueued, start the state-change 2510 * report timer for this group. 2511 */ 2512 inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv); 2513 inm->inm_sctimer = 1; 2514 V_state_change_timers_running = 1; 2515 2516 return (0); 2517 } 2518 2519 /* 2520 * Perform the final leave for an IGMP group. 2521 * 2522 * When leaving a group: 2523 * IGMPv1 does nothing. 2524 * IGMPv2 sends a host leave message, if and only if we are the reporter. 2525 * IGMPv3 enqueues a state-change report containing a transition 2526 * to INCLUDE {} for immediate transmission. 2527 */ 2528 static void 2529 igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi) 2530 { 2531 int syncstates; 2532 2533 syncstates = 1; 2534 2535 CTR4(KTR_IGMPV3, "%s: final leave %s on ifp %p(%s)", 2536 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp, 2537 inm->inm_ifp->if_xname); 2538 2539 IN_MULTI_LOCK_ASSERT(); 2540 IGMP_LOCK_ASSERT(); 2541 2542 switch (inm->inm_state) { 2543 case IGMP_NOT_MEMBER: 2544 case IGMP_SILENT_MEMBER: 2545 case IGMP_LEAVING_MEMBER: 2546 /* Already leaving or left; do nothing. */ 2547 CTR1(KTR_IGMPV3, 2548 "%s: not kicking state machine for silent group", __func__); 2549 break; 2550 case IGMP_REPORTING_MEMBER: 2551 case IGMP_IDLE_MEMBER: 2552 case IGMP_G_QUERY_PENDING_MEMBER: 2553 case IGMP_SG_QUERY_PENDING_MEMBER: 2554 if (igi->igi_version == IGMP_VERSION_2) { 2555 #ifdef INVARIANTS 2556 if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || 2557 inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) 2558 panic("%s: IGMPv3 state reached, not IGMPv3 mode", 2559 __func__); 2560 #endif 2561 igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE); 2562 inm->inm_state = IGMP_NOT_MEMBER; 2563 } else if (igi->igi_version == IGMP_VERSION_3) { 2564 /* 2565 * Stop group timer and all pending reports. 2566 * Immediately enqueue a state-change report 2567 * TO_IN {} to be sent on the next fast timeout, 2568 * giving us an opportunity to merge reports. 2569 */ 2570 mbufq_drain(&inm->inm_scq); 2571 inm->inm_timer = 0; 2572 if (igi->igi_flags & IGIF_LOOPBACK) { 2573 inm->inm_scrv = 1; 2574 } else { 2575 inm->inm_scrv = igi->igi_rv; 2576 } 2577 CTR4(KTR_IGMPV3, "%s: Leaving %s/%s with %d " 2578 "pending retransmissions.", __func__, 2579 inet_ntoa(inm->inm_addr), 2580 inm->inm_ifp->if_xname, inm->inm_scrv); 2581 if (inm->inm_scrv == 0) { 2582 inm->inm_state = IGMP_NOT_MEMBER; 2583 inm->inm_sctimer = 0; 2584 } else { 2585 int retval; 2586 2587 inm_acquire_locked(inm); 2588 2589 retval = igmp_v3_enqueue_group_record( 2590 &inm->inm_scq, inm, 1, 0, 0); 2591 KASSERT(retval != 0, 2592 ("%s: enqueue record = %d", __func__, 2593 retval)); 2594 2595 inm->inm_state = IGMP_LEAVING_MEMBER; 2596 inm->inm_sctimer = 1; 2597 V_state_change_timers_running = 1; 2598 syncstates = 0; 2599 } 2600 break; 2601 } 2602 break; 2603 case IGMP_LAZY_MEMBER: 2604 case IGMP_SLEEPING_MEMBER: 2605 case IGMP_AWAKENING_MEMBER: 2606 /* Our reports are suppressed; do nothing. */ 2607 break; 2608 } 2609 2610 if (syncstates) { 2611 inm_commit(inm); 2612 CTR3(KTR_IGMPV3, "%s: T1 -> T0 for %s/%s", __func__, 2613 inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname); 2614 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 2615 CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for %s/%s", 2616 __func__, inet_ntoa(inm->inm_addr), inm->inm_ifp->if_xname); 2617 } 2618 } 2619 2620 /* 2621 * Enqueue an IGMPv3 group record to the given output queue. 2622 * 2623 * XXX This function could do with having the allocation code 2624 * split out, and the multiple-tree-walks coalesced into a single 2625 * routine as has been done in igmp_v3_enqueue_filter_change(). 2626 * 2627 * If is_state_change is zero, a current-state record is appended. 2628 * If is_state_change is non-zero, a state-change report is appended. 2629 * 2630 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2631 * If is_group_query is zero, and if there is a packet with free space 2632 * at the tail of the queue, it will be appended to providing there 2633 * is enough free space. 2634 * Otherwise a new mbuf packet chain is allocated. 2635 * 2636 * If is_source_query is non-zero, each source is checked to see if 2637 * it was recorded for a Group-Source query, and will be omitted if 2638 * it is not both in-mode and recorded. 2639 * 2640 * The function will attempt to allocate leading space in the packet 2641 * for the IP/IGMP header to be prepended without fragmenting the chain. 2642 * 2643 * If successful the size of all data appended to the queue is returned, 2644 * otherwise an error code less than zero is returned, or zero if 2645 * no record(s) were appended. 2646 */ 2647 static int 2648 igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm, 2649 const int is_state_change, const int is_group_query, 2650 const int is_source_query) 2651 { 2652 struct igmp_grouprec ig; 2653 struct igmp_grouprec *pig; 2654 struct ifnet *ifp; 2655 struct ip_msource *ims, *nims; 2656 struct mbuf *m0, *m, *md; 2657 int error, is_filter_list_change; 2658 int minrec0len, m0srcs, msrcs, nbytes, off; 2659 int record_has_sources; 2660 int now; 2661 int type; 2662 in_addr_t naddr; 2663 uint8_t mode; 2664 2665 IN_MULTI_LOCK_ASSERT(); 2666 2667 error = 0; 2668 ifp = inm->inm_ifp; 2669 is_filter_list_change = 0; 2670 m = NULL; 2671 m0 = NULL; 2672 m0srcs = 0; 2673 msrcs = 0; 2674 nbytes = 0; 2675 nims = NULL; 2676 record_has_sources = 1; 2677 pig = NULL; 2678 type = IGMP_DO_NOTHING; 2679 mode = inm->inm_st[1].iss_fmode; 2680 2681 /* 2682 * If we did not transition out of ASM mode during t0->t1, 2683 * and there are no source nodes to process, we can skip 2684 * the generation of source records. 2685 */ 2686 if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 && 2687 inm->inm_nsrc == 0) 2688 record_has_sources = 0; 2689 2690 if (is_state_change) { 2691 /* 2692 * Queue a state change record. 2693 * If the mode did not change, and there are non-ASM 2694 * listeners or source filters present, 2695 * we potentially need to issue two records for the group. 2696 * If we are transitioning to MCAST_UNDEFINED, we need 2697 * not send any sources. 2698 * If there are ASM listeners, and there was no filter 2699 * mode transition of any kind, do nothing. 2700 */ 2701 if (mode != inm->inm_st[0].iss_fmode) { 2702 if (mode == MCAST_EXCLUDE) { 2703 CTR1(KTR_IGMPV3, "%s: change to EXCLUDE", 2704 __func__); 2705 type = IGMP_CHANGE_TO_EXCLUDE_MODE; 2706 } else { 2707 CTR1(KTR_IGMPV3, "%s: change to INCLUDE", 2708 __func__); 2709 type = IGMP_CHANGE_TO_INCLUDE_MODE; 2710 if (mode == MCAST_UNDEFINED) 2711 record_has_sources = 0; 2712 } 2713 } else { 2714 if (record_has_sources) { 2715 is_filter_list_change = 1; 2716 } else { 2717 type = IGMP_DO_NOTHING; 2718 } 2719 } 2720 } else { 2721 /* 2722 * Queue a current state record. 2723 */ 2724 if (mode == MCAST_EXCLUDE) { 2725 type = IGMP_MODE_IS_EXCLUDE; 2726 } else if (mode == MCAST_INCLUDE) { 2727 type = IGMP_MODE_IS_INCLUDE; 2728 KASSERT(inm->inm_st[1].iss_asm == 0, 2729 ("%s: inm %p is INCLUDE but ASM count is %d", 2730 __func__, inm, inm->inm_st[1].iss_asm)); 2731 } 2732 } 2733 2734 /* 2735 * Generate the filter list changes using a separate function. 2736 */ 2737 if (is_filter_list_change) 2738 return (igmp_v3_enqueue_filter_change(mq, inm)); 2739 2740 if (type == IGMP_DO_NOTHING) { 2741 CTR3(KTR_IGMPV3, "%s: nothing to do for %s/%s", 2742 __func__, inet_ntoa(inm->inm_addr), 2743 inm->inm_ifp->if_xname); 2744 return (0); 2745 } 2746 2747 /* 2748 * If any sources are present, we must be able to fit at least 2749 * one in the trailing space of the tail packet's mbuf, 2750 * ideally more. 2751 */ 2752 minrec0len = sizeof(struct igmp_grouprec); 2753 if (record_has_sources) 2754 minrec0len += sizeof(in_addr_t); 2755 2756 CTR4(KTR_IGMPV3, "%s: queueing %s for %s/%s", __func__, 2757 igmp_rec_type_to_str(type), inet_ntoa(inm->inm_addr), 2758 inm->inm_ifp->if_xname); 2759 2760 /* 2761 * Check if we have a packet in the tail of the queue for this 2762 * group into which the first group record for this group will fit. 2763 * Otherwise allocate a new packet. 2764 * Always allocate leading space for IP+RA_OPT+IGMP+REPORT. 2765 * Note: Group records for G/GSR query responses MUST be sent 2766 * in their own packet. 2767 */ 2768 m0 = mbufq_last(mq); 2769 if (!is_group_query && 2770 m0 != NULL && 2771 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) && 2772 (m0->m_pkthdr.len + minrec0len) < 2773 (ifp->if_mtu - IGMP_LEADINGSPACE)) { 2774 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2775 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2776 m = m0; 2777 CTR1(KTR_IGMPV3, "%s: use existing packet", __func__); 2778 } else { 2779 if (mbufq_full(mq)) { 2780 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); 2781 return (-ENOMEM); 2782 } 2783 m = NULL; 2784 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 2785 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2786 if (!is_state_change && !is_group_query) { 2787 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2788 if (m) 2789 m->m_data += IGMP_LEADINGSPACE; 2790 } 2791 if (m == NULL) { 2792 m = m_gethdr(M_NOWAIT, MT_DATA); 2793 if (m) 2794 M_ALIGN(m, IGMP_LEADINGSPACE); 2795 } 2796 if (m == NULL) 2797 return (-ENOMEM); 2798 2799 igmp_save_context(m, ifp); 2800 2801 CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__); 2802 } 2803 2804 /* 2805 * Append group record. 2806 * If we have sources, we don't know how many yet. 2807 */ 2808 ig.ig_type = type; 2809 ig.ig_datalen = 0; 2810 ig.ig_numsrc = 0; 2811 ig.ig_group = inm->inm_addr; 2812 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { 2813 if (m != m0) 2814 m_freem(m); 2815 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); 2816 return (-ENOMEM); 2817 } 2818 nbytes += sizeof(struct igmp_grouprec); 2819 2820 /* 2821 * Append as many sources as will fit in the first packet. 2822 * If we are appending to a new packet, the chain allocation 2823 * may potentially use clusters; use m_getptr() in this case. 2824 * If we are appending to an existing packet, we need to obtain 2825 * a pointer to the group record after m_append(), in case a new 2826 * mbuf was allocated. 2827 * Only append sources which are in-mode at t1. If we are 2828 * transitioning to MCAST_UNDEFINED state on the group, do not 2829 * include source entries. 2830 * Only report recorded sources in our filter set when responding 2831 * to a group-source query. 2832 */ 2833 if (record_has_sources) { 2834 if (m == m0) { 2835 md = m_last(m); 2836 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + 2837 md->m_len - nbytes); 2838 } else { 2839 md = m_getptr(m, 0, &off); 2840 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + 2841 off); 2842 } 2843 msrcs = 0; 2844 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) { 2845 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__, 2846 inet_ntoa_haddr(ims->ims_haddr)); 2847 now = ims_get_mode(inm, ims, 1); 2848 CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now); 2849 if ((now != mode) || 2850 (now == mode && mode == MCAST_UNDEFINED)) { 2851 CTR1(KTR_IGMPV3, "%s: skip node", __func__); 2852 continue; 2853 } 2854 if (is_source_query && ims->ims_stp == 0) { 2855 CTR1(KTR_IGMPV3, "%s: skip unrecorded node", 2856 __func__); 2857 continue; 2858 } 2859 CTR1(KTR_IGMPV3, "%s: append node", __func__); 2860 naddr = htonl(ims->ims_haddr); 2861 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { 2862 if (m != m0) 2863 m_freem(m); 2864 CTR1(KTR_IGMPV3, "%s: m_append() failed.", 2865 __func__); 2866 return (-ENOMEM); 2867 } 2868 nbytes += sizeof(in_addr_t); 2869 ++msrcs; 2870 if (msrcs == m0srcs) 2871 break; 2872 } 2873 CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__, 2874 msrcs); 2875 pig->ig_numsrc = htons(msrcs); 2876 nbytes += (msrcs * sizeof(in_addr_t)); 2877 } 2878 2879 if (is_source_query && msrcs == 0) { 2880 CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__); 2881 if (m != m0) 2882 m_freem(m); 2883 return (0); 2884 } 2885 2886 /* 2887 * We are good to go with first packet. 2888 */ 2889 if (m != m0) { 2890 CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__); 2891 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2892 mbufq_enqueue(mq, m); 2893 } else 2894 m->m_pkthdr.PH_vt.vt_nrecs++; 2895 2896 /* 2897 * No further work needed if no source list in packet(s). 2898 */ 2899 if (!record_has_sources) 2900 return (nbytes); 2901 2902 /* 2903 * Whilst sources remain to be announced, we need to allocate 2904 * a new packet and fill out as many sources as will fit. 2905 * Always try for a cluster first. 2906 */ 2907 while (nims != NULL) { 2908 if (mbufq_full(mq)) { 2909 CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); 2910 return (-ENOMEM); 2911 } 2912 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2913 if (m) 2914 m->m_data += IGMP_LEADINGSPACE; 2915 if (m == NULL) { 2916 m = m_gethdr(M_NOWAIT, MT_DATA); 2917 if (m) 2918 M_ALIGN(m, IGMP_LEADINGSPACE); 2919 } 2920 if (m == NULL) 2921 return (-ENOMEM); 2922 igmp_save_context(m, ifp); 2923 md = m_getptr(m, 0, &off); 2924 pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); 2925 CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__); 2926 2927 if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { 2928 if (m != m0) 2929 m_freem(m); 2930 CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); 2931 return (-ENOMEM); 2932 } 2933 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2934 nbytes += sizeof(struct igmp_grouprec); 2935 2936 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 2937 sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); 2938 2939 msrcs = 0; 2940 RB_FOREACH_FROM(ims, ip_msource_tree, nims) { 2941 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__, 2942 inet_ntoa_haddr(ims->ims_haddr)); 2943 now = ims_get_mode(inm, ims, 1); 2944 if ((now != mode) || 2945 (now == mode && mode == MCAST_UNDEFINED)) { 2946 CTR1(KTR_IGMPV3, "%s: skip node", __func__); 2947 continue; 2948 } 2949 if (is_source_query && ims->ims_stp == 0) { 2950 CTR1(KTR_IGMPV3, "%s: skip unrecorded node", 2951 __func__); 2952 continue; 2953 } 2954 CTR1(KTR_IGMPV3, "%s: append node", __func__); 2955 naddr = htonl(ims->ims_haddr); 2956 if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { 2957 if (m != m0) 2958 m_freem(m); 2959 CTR1(KTR_IGMPV3, "%s: m_append() failed.", 2960 __func__); 2961 return (-ENOMEM); 2962 } 2963 ++msrcs; 2964 if (msrcs == m0srcs) 2965 break; 2966 } 2967 pig->ig_numsrc = htons(msrcs); 2968 nbytes += (msrcs * sizeof(in_addr_t)); 2969 2970 CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__); 2971 mbufq_enqueue(mq, m); 2972 } 2973 2974 return (nbytes); 2975 } 2976 2977 /* 2978 * Type used to mark record pass completion. 2979 * We exploit the fact we can cast to this easily from the 2980 * current filter modes on each ip_msource node. 2981 */ 2982 typedef enum { 2983 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2984 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2985 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2986 REC_FULL = REC_ALLOW | REC_BLOCK 2987 } rectype_t; 2988 2989 /* 2990 * Enqueue an IGMPv3 filter list change to the given output queue. 2991 * 2992 * Source list filter state is held in an RB-tree. When the filter list 2993 * for a group is changed without changing its mode, we need to compute 2994 * the deltas between T0 and T1 for each source in the filter set, 2995 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2996 * 2997 * As we may potentially queue two record types, and the entire R-B tree 2998 * needs to be walked at once, we break this out into its own function 2999 * so we can generate a tightly packed queue of packets. 3000 * 3001 * XXX This could be written to only use one tree walk, although that makes 3002 * serializing into the mbuf chains a bit harder. For now we do two walks 3003 * which makes things easier on us, and it may or may not be harder on 3004 * the L2 cache. 3005 * 3006 * If successful the size of all data appended to the queue is returned, 3007 * otherwise an error code less than zero is returned, or zero if 3008 * no record(s) were appended. 3009 */ 3010 static int 3011 igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm) 3012 { 3013 static const int MINRECLEN = 3014 sizeof(struct igmp_grouprec) + sizeof(in_addr_t); 3015 struct ifnet *ifp; 3016 struct igmp_grouprec ig; 3017 struct igmp_grouprec *pig; 3018 struct ip_msource *ims, *nims; 3019 struct mbuf *m, *m0, *md; 3020 in_addr_t naddr; 3021 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 3022 int nallow, nblock; 3023 uint8_t mode, now, then; 3024 rectype_t crt, drt, nrt; 3025 3026 IN_MULTI_LOCK_ASSERT(); 3027 3028 if (inm->inm_nsrc == 0 || 3029 (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0)) 3030 return (0); 3031 3032 ifp = inm->inm_ifp; /* interface */ 3033 mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */ 3034 crt = REC_NONE; /* current group record type */ 3035 drt = REC_NONE; /* mask of completed group record types */ 3036 nrt = REC_NONE; /* record type for current node */ 3037 m0srcs = 0; /* # source which will fit in current mbuf chain */ 3038 nbytes = 0; /* # of bytes appended to group's state-change queue */ 3039 npbytes = 0; /* # of bytes appended this packet */ 3040 rsrcs = 0; /* # sources encoded in current record */ 3041 schanged = 0; /* # nodes encoded in overall filter change */ 3042 nallow = 0; /* # of source entries in ALLOW_NEW */ 3043 nblock = 0; /* # of source entries in BLOCK_OLD */ 3044 nims = NULL; /* next tree node pointer */ 3045 3046 /* 3047 * For each possible filter record mode. 3048 * The first kind of source we encounter tells us which 3049 * is the first kind of record we start appending. 3050 * If a node transitioned to UNDEFINED at t1, its mode is treated 3051 * as the inverse of the group's filter mode. 3052 */ 3053 while (drt != REC_FULL) { 3054 do { 3055 m0 = mbufq_last(mq); 3056 if (m0 != NULL && 3057 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 3058 IGMP_V3_REPORT_MAXRECS) && 3059 (m0->m_pkthdr.len + MINRECLEN) < 3060 (ifp->if_mtu - IGMP_LEADINGSPACE)) { 3061 m = m0; 3062 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 3063 sizeof(struct igmp_grouprec)) / 3064 sizeof(in_addr_t); 3065 CTR1(KTR_IGMPV3, 3066 "%s: use previous packet", __func__); 3067 } else { 3068 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 3069 if (m) 3070 m->m_data += IGMP_LEADINGSPACE; 3071 if (m == NULL) { 3072 m = m_gethdr(M_NOWAIT, MT_DATA); 3073 if (m) 3074 M_ALIGN(m, IGMP_LEADINGSPACE); 3075 } 3076 if (m == NULL) { 3077 CTR1(KTR_IGMPV3, 3078 "%s: m_get*() failed", __func__); 3079 return (-ENOMEM); 3080 } 3081 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3082 igmp_save_context(m, ifp); 3083 m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - 3084 sizeof(struct igmp_grouprec)) / 3085 sizeof(in_addr_t); 3086 npbytes = 0; 3087 CTR1(KTR_IGMPV3, 3088 "%s: allocated new packet", __func__); 3089 } 3090 /* 3091 * Append the IGMP group record header to the 3092 * current packet's data area. 3093 * Recalculate pointer to free space for next 3094 * group record, in case m_append() allocated 3095 * a new mbuf or cluster. 3096 */ 3097 memset(&ig, 0, sizeof(ig)); 3098 ig.ig_group = inm->inm_addr; 3099 if (!m_append(m, sizeof(ig), (void *)&ig)) { 3100 if (m != m0) 3101 m_freem(m); 3102 CTR1(KTR_IGMPV3, 3103 "%s: m_append() failed", __func__); 3104 return (-ENOMEM); 3105 } 3106 npbytes += sizeof(struct igmp_grouprec); 3107 if (m != m0) { 3108 /* new packet; offset in c hain */ 3109 md = m_getptr(m, npbytes - 3110 sizeof(struct igmp_grouprec), &off); 3111 pig = (struct igmp_grouprec *)(mtod(md, 3112 uint8_t *) + off); 3113 } else { 3114 /* current packet; offset from last append */ 3115 md = m_last(m); 3116 pig = (struct igmp_grouprec *)(mtod(md, 3117 uint8_t *) + md->m_len - 3118 sizeof(struct igmp_grouprec)); 3119 } 3120 /* 3121 * Begin walking the tree for this record type 3122 * pass, or continue from where we left off 3123 * previously if we had to allocate a new packet. 3124 * Only report deltas in-mode at t1. 3125 * We need not report included sources as allowed 3126 * if we are in inclusive mode on the group, 3127 * however the converse is not true. 3128 */ 3129 rsrcs = 0; 3130 if (nims == NULL) 3131 nims = RB_MIN(ip_msource_tree, &inm->inm_srcs); 3132 RB_FOREACH_FROM(ims, ip_msource_tree, nims) { 3133 CTR2(KTR_IGMPV3, "%s: visit node %s", 3134 __func__, inet_ntoa_haddr(ims->ims_haddr)); 3135 now = ims_get_mode(inm, ims, 1); 3136 then = ims_get_mode(inm, ims, 0); 3137 CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d", 3138 __func__, then, now); 3139 if (now == then) { 3140 CTR1(KTR_IGMPV3, 3141 "%s: skip unchanged", __func__); 3142 continue; 3143 } 3144 if (mode == MCAST_EXCLUDE && 3145 now == MCAST_INCLUDE) { 3146 CTR1(KTR_IGMPV3, 3147 "%s: skip IN src on EX group", 3148 __func__); 3149 continue; 3150 } 3151 nrt = (rectype_t)now; 3152 if (nrt == REC_NONE) 3153 nrt = (rectype_t)(~mode & REC_FULL); 3154 if (schanged++ == 0) { 3155 crt = nrt; 3156 } else if (crt != nrt) 3157 continue; 3158 naddr = htonl(ims->ims_haddr); 3159 if (!m_append(m, sizeof(in_addr_t), 3160 (void *)&naddr)) { 3161 if (m != m0) 3162 m_freem(m); 3163 CTR1(KTR_IGMPV3, 3164 "%s: m_append() failed", __func__); 3165 return (-ENOMEM); 3166 } 3167 nallow += !!(crt == REC_ALLOW); 3168 nblock += !!(crt == REC_BLOCK); 3169 if (++rsrcs == m0srcs) 3170 break; 3171 } 3172 /* 3173 * If we did not append any tree nodes on this 3174 * pass, back out of allocations. 3175 */ 3176 if (rsrcs == 0) { 3177 npbytes -= sizeof(struct igmp_grouprec); 3178 if (m != m0) { 3179 CTR1(KTR_IGMPV3, 3180 "%s: m_free(m)", __func__); 3181 m_freem(m); 3182 } else { 3183 CTR1(KTR_IGMPV3, 3184 "%s: m_adj(m, -ig)", __func__); 3185 m_adj(m, -((int)sizeof( 3186 struct igmp_grouprec))); 3187 } 3188 continue; 3189 } 3190 npbytes += (rsrcs * sizeof(in_addr_t)); 3191 if (crt == REC_ALLOW) 3192 pig->ig_type = IGMP_ALLOW_NEW_SOURCES; 3193 else if (crt == REC_BLOCK) 3194 pig->ig_type = IGMP_BLOCK_OLD_SOURCES; 3195 pig->ig_numsrc = htons(rsrcs); 3196 /* 3197 * Count the new group record, and enqueue this 3198 * packet if it wasn't already queued. 3199 */ 3200 m->m_pkthdr.PH_vt.vt_nrecs++; 3201 if (m != m0) 3202 mbufq_enqueue(mq, m); 3203 nbytes += npbytes; 3204 } while (nims != NULL); 3205 drt |= crt; 3206 crt = (~crt & REC_FULL); 3207 } 3208 3209 CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 3210 nallow, nblock); 3211 3212 return (nbytes); 3213 } 3214 3215 static int 3216 igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq) 3217 { 3218 struct mbufq *gq; 3219 struct mbuf *m; /* pending state-change */ 3220 struct mbuf *m0; /* copy of pending state-change */ 3221 struct mbuf *mt; /* last state-change in packet */ 3222 int docopy, domerge; 3223 u_int recslen; 3224 3225 docopy = 0; 3226 domerge = 0; 3227 recslen = 0; 3228 3229 IN_MULTI_LOCK_ASSERT(); 3230 IGMP_LOCK_ASSERT(); 3231 3232 /* 3233 * If there are further pending retransmissions, make a writable 3234 * copy of each queued state-change message before merging. 3235 */ 3236 if (inm->inm_scrv > 0) 3237 docopy = 1; 3238 3239 gq = &inm->inm_scq; 3240 #ifdef KTR 3241 if (mbufq_first(gq) == NULL) { 3242 CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty", 3243 __func__, inm); 3244 } 3245 #endif 3246 3247 m = mbufq_first(gq); 3248 while (m != NULL) { 3249 /* 3250 * Only merge the report into the current packet if 3251 * there is sufficient space to do so; an IGMPv3 report 3252 * packet may only contain 65,535 group records. 3253 * Always use a simple mbuf chain concatentation to do this, 3254 * as large state changes for single groups may have 3255 * allocated clusters. 3256 */ 3257 domerge = 0; 3258 mt = mbufq_last(scq); 3259 if (mt != NULL) { 3260 recslen = m_length(m, NULL); 3261 3262 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 3263 m->m_pkthdr.PH_vt.vt_nrecs <= 3264 IGMP_V3_REPORT_MAXRECS) && 3265 (mt->m_pkthdr.len + recslen <= 3266 (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE))) 3267 domerge = 1; 3268 } 3269 3270 if (!domerge && mbufq_full(gq)) { 3271 CTR2(KTR_IGMPV3, 3272 "%s: outbound queue full, skipping whole packet %p", 3273 __func__, m); 3274 mt = m->m_nextpkt; 3275 if (!docopy) 3276 m_freem(m); 3277 m = mt; 3278 continue; 3279 } 3280 3281 if (!docopy) { 3282 CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m); 3283 m0 = mbufq_dequeue(gq); 3284 m = m0->m_nextpkt; 3285 } else { 3286 CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m); 3287 m0 = m_dup(m, M_NOWAIT); 3288 if (m0 == NULL) 3289 return (ENOMEM); 3290 m0->m_nextpkt = NULL; 3291 m = m->m_nextpkt; 3292 } 3293 3294 if (!domerge) { 3295 CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)", 3296 __func__, m0, scq); 3297 mbufq_enqueue(scq, m0); 3298 } else { 3299 struct mbuf *mtl; /* last mbuf of packet mt */ 3300 3301 CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)", 3302 __func__, m0, mt); 3303 3304 mtl = m_last(mt); 3305 m0->m_flags &= ~M_PKTHDR; 3306 mt->m_pkthdr.len += recslen; 3307 mt->m_pkthdr.PH_vt.vt_nrecs += 3308 m0->m_pkthdr.PH_vt.vt_nrecs; 3309 3310 mtl->m_next = m0; 3311 } 3312 } 3313 3314 return (0); 3315 } 3316 3317 /* 3318 * Respond to a pending IGMPv3 General Query. 3319 */ 3320 static void 3321 igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi) 3322 { 3323 struct ifmultiaddr *ifma; 3324 struct ifnet *ifp; 3325 struct in_multi *inm; 3326 int retval, loop; 3327 3328 IN_MULTI_LOCK_ASSERT(); 3329 IGMP_LOCK_ASSERT(); 3330 3331 KASSERT(igi->igi_version == IGMP_VERSION_3, 3332 ("%s: called when version %d", __func__, igi->igi_version)); 3333 3334 ifp = igi->igi_ifp; 3335 3336 IF_ADDR_RLOCK(ifp); 3337 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3338 if (ifma->ifma_addr->sa_family != AF_INET || 3339 ifma->ifma_protospec == NULL) 3340 continue; 3341 3342 inm = (struct in_multi *)ifma->ifma_protospec; 3343 KASSERT(ifp == inm->inm_ifp, 3344 ("%s: inconsistent ifp", __func__)); 3345 3346 switch (inm->inm_state) { 3347 case IGMP_NOT_MEMBER: 3348 case IGMP_SILENT_MEMBER: 3349 break; 3350 case IGMP_REPORTING_MEMBER: 3351 case IGMP_IDLE_MEMBER: 3352 case IGMP_LAZY_MEMBER: 3353 case IGMP_SLEEPING_MEMBER: 3354 case IGMP_AWAKENING_MEMBER: 3355 inm->inm_state = IGMP_REPORTING_MEMBER; 3356 retval = igmp_v3_enqueue_group_record(&igi->igi_gq, 3357 inm, 0, 0, 0); 3358 CTR2(KTR_IGMPV3, "%s: enqueue record = %d", 3359 __func__, retval); 3360 break; 3361 case IGMP_G_QUERY_PENDING_MEMBER: 3362 case IGMP_SG_QUERY_PENDING_MEMBER: 3363 case IGMP_LEAVING_MEMBER: 3364 break; 3365 } 3366 } 3367 IF_ADDR_RUNLOCK(ifp); 3368 3369 loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; 3370 igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop); 3371 3372 /* 3373 * Slew transmission of bursts over 500ms intervals. 3374 */ 3375 if (mbufq_first(&igi->igi_gq) != NULL) { 3376 igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY( 3377 IGMP_RESPONSE_BURST_INTERVAL); 3378 V_interface_timers_running = 1; 3379 } 3380 } 3381 3382 /* 3383 * Transmit the next pending IGMP message in the output queue. 3384 * 3385 * We get called from netisr_processqueue(). A mutex private to igmpoq 3386 * will be acquired and released around this routine. 3387 * 3388 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3389 * MRT: Nothing needs to be done, as IGMP traffic is always local to 3390 * a link and uses a link-scope multicast address. 3391 */ 3392 static void 3393 igmp_intr(struct mbuf *m) 3394 { 3395 struct ip_moptions imo; 3396 struct ifnet *ifp; 3397 struct mbuf *ipopts, *m0; 3398 int error; 3399 uint32_t ifindex; 3400 3401 CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m); 3402 3403 /* 3404 * Set VNET image pointer from enqueued mbuf chain 3405 * before doing anything else. Whilst we use interface 3406 * indexes to guard against interface detach, they are 3407 * unique to each VIMAGE and must be retrieved. 3408 */ 3409 CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr)); 3410 ifindex = igmp_restore_context(m); 3411 3412 /* 3413 * Check if the ifnet still exists. This limits the scope of 3414 * any race in the absence of a global ifp lock for low cost 3415 * (an array lookup). 3416 */ 3417 ifp = ifnet_byindex(ifindex); 3418 if (ifp == NULL) { 3419 CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.", 3420 __func__, m, ifindex); 3421 m_freem(m); 3422 IPSTAT_INC(ips_noroute); 3423 goto out; 3424 } 3425 3426 ipopts = V_igmp_sendra ? m_raopt : NULL; 3427 3428 imo.imo_multicast_ttl = 1; 3429 imo.imo_multicast_vif = -1; 3430 imo.imo_multicast_loop = (V_ip_mrouter != NULL); 3431 3432 /* 3433 * If the user requested that IGMP traffic be explicitly 3434 * redirected to the loopback interface (e.g. they are running a 3435 * MANET interface and the routing protocol needs to see the 3436 * updates), handle this now. 3437 */ 3438 if (m->m_flags & M_IGMP_LOOP) 3439 imo.imo_multicast_ifp = V_loif; 3440 else 3441 imo.imo_multicast_ifp = ifp; 3442 3443 if (m->m_flags & M_IGMPV2) { 3444 m0 = m; 3445 } else { 3446 m0 = igmp_v3_encap_report(ifp, m); 3447 if (m0 == NULL) { 3448 CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m); 3449 m_freem(m); 3450 IPSTAT_INC(ips_odropped); 3451 goto out; 3452 } 3453 } 3454 3455 igmp_scrub_context(m0); 3456 m_clrprotoflags(m); 3457 m0->m_pkthdr.rcvif = V_loif; 3458 #ifdef MAC 3459 mac_netinet_igmp_send(ifp, m0); 3460 #endif 3461 error = ip_output(m0, ipopts, NULL, 0, &imo, NULL); 3462 if (error) { 3463 CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error); 3464 goto out; 3465 } 3466 3467 IGMPSTAT_INC(igps_snd_reports); 3468 3469 out: 3470 /* 3471 * We must restore the existing vnet pointer before 3472 * continuing as we are run from netisr context. 3473 */ 3474 CURVNET_RESTORE(); 3475 } 3476 3477 /* 3478 * Encapsulate an IGMPv3 report. 3479 * 3480 * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf 3481 * chain has already had its IP/IGMPv3 header prepended. In this case 3482 * the function will not attempt to prepend; the lengths and checksums 3483 * will however be re-computed. 3484 * 3485 * Returns a pointer to the new mbuf chain head, or NULL if the 3486 * allocation failed. 3487 */ 3488 static struct mbuf * 3489 igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m) 3490 { 3491 struct igmp_report *igmp; 3492 struct ip *ip; 3493 int hdrlen, igmpreclen; 3494 3495 KASSERT((m->m_flags & M_PKTHDR), 3496 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3497 3498 igmpreclen = m_length(m, NULL); 3499 hdrlen = sizeof(struct ip) + sizeof(struct igmp_report); 3500 3501 if (m->m_flags & M_IGMPV3_HDR) { 3502 igmpreclen -= hdrlen; 3503 } else { 3504 M_PREPEND(m, hdrlen, M_NOWAIT); 3505 if (m == NULL) 3506 return (NULL); 3507 m->m_flags |= M_IGMPV3_HDR; 3508 } 3509 3510 CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen); 3511 3512 m->m_data += sizeof(struct ip); 3513 m->m_len -= sizeof(struct ip); 3514 3515 igmp = mtod(m, struct igmp_report *); 3516 igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT; 3517 igmp->ir_rsv1 = 0; 3518 igmp->ir_rsv2 = 0; 3519 igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs); 3520 igmp->ir_cksum = 0; 3521 igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen); 3522 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3523 3524 m->m_data -= sizeof(struct ip); 3525 m->m_len += sizeof(struct ip); 3526 3527 ip = mtod(m, struct ip *); 3528 ip->ip_tos = IPTOS_PREC_INTERNETCONTROL; 3529 ip->ip_len = htons(hdrlen + igmpreclen); 3530 ip->ip_off = htons(IP_DF); 3531 ip->ip_p = IPPROTO_IGMP; 3532 ip->ip_sum = 0; 3533 3534 ip->ip_src.s_addr = INADDR_ANY; 3535 3536 if (m->m_flags & M_IGMP_LOOP) { 3537 struct in_ifaddr *ia; 3538 3539 IFP_TO_IA(ifp, ia); 3540 if (ia != NULL) { 3541 ip->ip_src = ia->ia_addr.sin_addr; 3542 ifa_free(&ia->ia_ifa); 3543 } 3544 } 3545 3546 ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP); 3547 3548 return (m); 3549 } 3550 3551 #ifdef KTR 3552 static char * 3553 igmp_rec_type_to_str(const int type) 3554 { 3555 3556 switch (type) { 3557 case IGMP_CHANGE_TO_EXCLUDE_MODE: 3558 return "TO_EX"; 3559 break; 3560 case IGMP_CHANGE_TO_INCLUDE_MODE: 3561 return "TO_IN"; 3562 break; 3563 case IGMP_MODE_IS_EXCLUDE: 3564 return "MODE_EX"; 3565 break; 3566 case IGMP_MODE_IS_INCLUDE: 3567 return "MODE_IN"; 3568 break; 3569 case IGMP_ALLOW_NEW_SOURCES: 3570 return "ALLOW_NEW"; 3571 break; 3572 case IGMP_BLOCK_OLD_SOURCES: 3573 return "BLOCK_OLD"; 3574 break; 3575 default: 3576 break; 3577 } 3578 return "unknown"; 3579 } 3580 #endif 3581 3582 static void 3583 igmp_init(void *unused __unused) 3584 { 3585 3586 CTR1(KTR_IGMPV3, "%s: initializing", __func__); 3587 3588 IGMP_LOCK_INIT(); 3589 3590 m_raopt = igmp_ra_alloc(); 3591 3592 netisr_register(&igmp_nh); 3593 } 3594 SYSINIT(igmp_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_init, NULL); 3595 3596 static void 3597 igmp_uninit(void *unused __unused) 3598 { 3599 3600 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3601 3602 netisr_unregister(&igmp_nh); 3603 3604 m_free(m_raopt); 3605 m_raopt = NULL; 3606 3607 IGMP_LOCK_DESTROY(); 3608 } 3609 SYSUNINIT(igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, igmp_uninit, NULL); 3610 3611 static void 3612 vnet_igmp_init(const void *unused __unused) 3613 { 3614 3615 CTR1(KTR_IGMPV3, "%s: initializing", __func__); 3616 3617 LIST_INIT(&V_igi_head); 3618 } 3619 VNET_SYSINIT(vnet_igmp_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_igmp_init, 3620 NULL); 3621 3622 static void 3623 vnet_igmp_uninit(const void *unused __unused) 3624 { 3625 3626 CTR1(KTR_IGMPV3, "%s: tearing down", __func__); 3627 3628 KASSERT(LIST_EMPTY(&V_igi_head), 3629 ("%s: igi list not empty; ifnets not detached?", __func__)); 3630 } 3631 VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, 3632 vnet_igmp_uninit, NULL); 3633 3634 static int 3635 igmp_modevent(module_t mod, int type, void *unused __unused) 3636 { 3637 3638 switch (type) { 3639 case MOD_LOAD: 3640 case MOD_UNLOAD: 3641 break; 3642 default: 3643 return (EOPNOTSUPP); 3644 } 3645 return (0); 3646 } 3647 3648 static moduledata_t igmp_mod = { 3649 "igmp", 3650 igmp_modevent, 3651 0 3652 }; 3653 DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3654