1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2009 Bruce Simpson. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. The name of the author may not be used to endorse or promote 15 * products derived from this software without specific prior written 16 * permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $ 31 */ 32 33 /*- 34 * Copyright (c) 1988 Stephen Deering. 35 * Copyright (c) 1992, 1993 36 * The Regents of the University of California. All rights reserved. 37 * 38 * This code is derived from software contributed to Berkeley by 39 * Stephen Deering of Stanford University. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 3. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * @(#)igmp.c 8.1 (Berkeley) 7/19/93 66 */ 67 68 #include <sys/cdefs.h> 69 __FBSDID("$FreeBSD$"); 70 71 #include "opt_inet.h" 72 #include "opt_inet6.h" 73 74 #include <sys/param.h> 75 #include <sys/systm.h> 76 #include <sys/mbuf.h> 77 #include <sys/socket.h> 78 #include <sys/protosw.h> 79 #include <sys/sysctl.h> 80 #include <sys/kernel.h> 81 #include <sys/callout.h> 82 #include <sys/malloc.h> 83 #include <sys/module.h> 84 #include <sys/ktr.h> 85 86 #include <net/if.h> 87 #include <net/if_var.h> 88 #include <net/route.h> 89 #include <net/vnet.h> 90 91 #include <netinet/in.h> 92 #include <netinet/in_var.h> 93 #include <netinet6/in6_var.h> 94 #include <netinet/ip6.h> 95 #include <netinet6/ip6_var.h> 96 #include <netinet6/scope6_var.h> 97 #include <netinet/icmp6.h> 98 #include <netinet6/mld6.h> 99 #include <netinet6/mld6_var.h> 100 101 #include <security/mac/mac_framework.h> 102 103 #ifndef KTR_MLD 104 #define KTR_MLD KTR_INET6 105 #endif 106 107 static struct mld_ifsoftc * 108 mli_alloc_locked(struct ifnet *); 109 static void mli_delete_locked(const struct ifnet *); 110 static void mld_dispatch_packet(struct mbuf *); 111 static void mld_dispatch_queue(struct mbufq *, int); 112 static void mld_final_leave(struct in6_multi *, struct mld_ifsoftc *); 113 static void mld_fasttimo_vnet(void); 114 static int mld_handle_state_change(struct in6_multi *, 115 struct mld_ifsoftc *); 116 static int mld_initial_join(struct in6_multi *, struct mld_ifsoftc *, 117 const int); 118 #ifdef KTR 119 static char * mld_rec_type_to_str(const int); 120 #endif 121 static void mld_set_version(struct mld_ifsoftc *, const int); 122 static void mld_slowtimo_vnet(void); 123 static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *, 124 /*const*/ struct mld_hdr *); 125 static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *, 126 /*const*/ struct mld_hdr *); 127 static void mld_v1_process_group_timer(struct in6_multi_head *, 128 struct in6_multi *); 129 static void mld_v1_process_querier_timers(struct mld_ifsoftc *); 130 static int mld_v1_transmit_report(struct in6_multi *, const int); 131 static void mld_v1_update_group(struct in6_multi *, const int); 132 static void mld_v2_cancel_link_timers(struct mld_ifsoftc *); 133 static void mld_v2_dispatch_general_query(struct mld_ifsoftc *); 134 static struct mbuf * 135 mld_v2_encap_report(struct ifnet *, struct mbuf *); 136 static int mld_v2_enqueue_filter_change(struct mbufq *, 137 struct in6_multi *); 138 static int mld_v2_enqueue_group_record(struct mbufq *, 139 struct in6_multi *, const int, const int, const int, 140 const int); 141 static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *, 142 struct mbuf *, const int, const int); 143 static int mld_v2_merge_state_changes(struct in6_multi *, 144 struct mbufq *); 145 static void mld_v2_process_group_timers(struct in6_multi_head *, 146 struct mbufq *, struct mbufq *, 147 struct in6_multi *, const int); 148 static int mld_v2_process_group_query(struct in6_multi *, 149 struct mld_ifsoftc *mli, int, struct mbuf *, const int); 150 static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); 151 static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS); 152 153 /* 154 * Normative references: RFC 2710, RFC 3590, RFC 3810. 155 * 156 * Locking: 157 * * The MLD subsystem lock ends up being system-wide for the moment, 158 * but could be per-VIMAGE later on. 159 * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. 160 * Any may be taken independently; if any are held at the same 161 * time, the above lock order must be followed. 162 * * IN6_MULTI_LOCK covers in_multi. 163 * * MLD_LOCK covers per-link state and any global variables in this file. 164 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 165 * per-link state iterators. 166 * 167 * XXX LOR PREVENTION 168 * A special case for IPv6 is the in6_setscope() routine. ip6_output() 169 * will not accept an ifp; it wants an embedded scope ID, unlike 170 * ip_output(), which happily takes the ifp given to it. The embedded 171 * scope ID is only used by MLD to select the outgoing interface. 172 * 173 * During interface attach and detach, MLD will take MLD_LOCK *after* 174 * the IF_AFDATA_LOCK. 175 * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call 176 * it with MLD_LOCK held without triggering an LOR. A netisr with indirect 177 * dispatch could work around this, but we'd rather not do that, as it 178 * can introduce other races. 179 * 180 * As such, we exploit the fact that the scope ID is just the interface 181 * index, and embed it in the IPv6 destination address accordingly. 182 * This is potentially NOT VALID for MLDv1 reports, as they 183 * are always sent to the multicast group itself; as MLDv2 184 * reports are always sent to ff02::16, this is not an issue 185 * when MLDv2 is in use. 186 * 187 * This does not however eliminate the LOR when ip6_output() itself 188 * calls in6_setscope() internally whilst MLD_LOCK is held. This will 189 * trigger a LOR warning in WITNESS when the ifnet is detached. 190 * 191 * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given 192 * how it's used across the network stack. Here we're simply exploiting 193 * the fact that MLD runs at a similar layer in the stack to scope6.c. 194 * 195 * VIMAGE: 196 * * Each in6_multi corresponds to an ifp, and each ifp corresponds 197 * to a vnet in ifp->if_vnet. 198 */ 199 static struct mtx mld_mtx; 200 static MALLOC_DEFINE(M_MLD, "mld", "mld state"); 201 202 #define MLD_EMBEDSCOPE(pin6, zoneid) \ 203 if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \ 204 IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \ 205 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \ 206 207 /* 208 * VIMAGE-wide globals. 209 */ 210 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0}; 211 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head); 212 VNET_DEFINE_STATIC(int, interface_timers_running6); 213 VNET_DEFINE_STATIC(int, state_change_timers_running6); 214 VNET_DEFINE_STATIC(int, current_state_timers_running6); 215 216 #define V_mld_gsrdelay VNET(mld_gsrdelay) 217 #define V_mli_head VNET(mli_head) 218 #define V_interface_timers_running6 VNET(interface_timers_running6) 219 #define V_state_change_timers_running6 VNET(state_change_timers_running6) 220 #define V_current_state_timers_running6 VNET(current_state_timers_running6) 221 222 SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ 223 224 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, 225 "IPv6 Multicast Listener Discovery"); 226 227 /* 228 * Virtualized sysctls. 229 */ 230 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay, 231 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 232 &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I", 233 "Rate limit for MLDv2 Group-and-Source queries in seconds"); 234 235 /* 236 * Non-virtualized sysctls. 237 */ 238 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, 239 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo, 240 "Per-interface MLDv2 state"); 241 242 static int mld_v1enable = 1; 243 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN, 244 &mld_v1enable, 0, "Enable fallback to MLDv1"); 245 246 static int mld_use_allow = 1; 247 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN, 248 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves"); 249 250 /* 251 * Packed Router Alert option structure declaration. 252 */ 253 struct mld_raopt { 254 struct ip6_hbh hbh; 255 struct ip6_opt pad; 256 struct ip6_opt_router ra; 257 } __packed; 258 259 /* 260 * Router Alert hop-by-hop option header. 261 */ 262 static struct mld_raopt mld_ra = { 263 .hbh = { 0, 0 }, 264 .pad = { .ip6o_type = IP6OPT_PADN, 0 }, 265 .ra = { 266 .ip6or_type = IP6OPT_ROUTER_ALERT, 267 .ip6or_len = IP6OPT_RTALERT_LEN - 2, 268 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), 269 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) 270 } 271 }; 272 static struct ip6_pktopts mld_po; 273 274 static __inline void 275 mld_save_context(struct mbuf *m, struct ifnet *ifp) 276 { 277 278 #ifdef VIMAGE 279 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; 280 #endif /* VIMAGE */ 281 m->m_pkthdr.flowid = ifp->if_index; 282 } 283 284 static __inline void 285 mld_scrub_context(struct mbuf *m) 286 { 287 288 m->m_pkthdr.PH_loc.ptr = NULL; 289 m->m_pkthdr.flowid = 0; 290 } 291 292 /* 293 * Restore context from a queued output chain. 294 * Return saved ifindex. 295 * 296 * VIMAGE: The assertion is there to make sure that we 297 * actually called CURVNET_SET() with what's in the mbuf chain. 298 */ 299 static __inline uint32_t 300 mld_restore_context(struct mbuf *m) 301 { 302 303 #if defined(VIMAGE) && defined(INVARIANTS) 304 KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr, 305 ("%s: called when curvnet was not restored: cuvnet %p m ptr %p", 306 __func__, curvnet, m->m_pkthdr.PH_loc.ptr)); 307 #endif 308 return (m->m_pkthdr.flowid); 309 } 310 311 /* 312 * Retrieve or set threshold between group-source queries in seconds. 313 * 314 * VIMAGE: Assume curvnet set by caller. 315 * SMPng: NOTE: Serialized by MLD lock. 316 */ 317 static int 318 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS) 319 { 320 int error; 321 int i; 322 323 error = sysctl_wire_old_buffer(req, sizeof(int)); 324 if (error) 325 return (error); 326 327 MLD_LOCK(); 328 329 i = V_mld_gsrdelay.tv_sec; 330 331 error = sysctl_handle_int(oidp, &i, 0, req); 332 if (error || !req->newptr) 333 goto out_locked; 334 335 if (i < -1 || i >= 60) { 336 error = EINVAL; 337 goto out_locked; 338 } 339 340 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", 341 V_mld_gsrdelay.tv_sec, i); 342 V_mld_gsrdelay.tv_sec = i; 343 344 out_locked: 345 MLD_UNLOCK(); 346 return (error); 347 } 348 349 /* 350 * Expose struct mld_ifsoftc to userland, keyed by ifindex. 351 * For use by ifmcstat(8). 352 * 353 * SMPng: NOTE: Does an unlocked ifindex space read. 354 * VIMAGE: Assume curvnet set by caller. The node handler itself 355 * is not directly virtualized. 356 */ 357 static int 358 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) 359 { 360 int *name; 361 int error; 362 u_int namelen; 363 struct ifnet *ifp; 364 struct mld_ifsoftc *mli; 365 366 name = (int *)arg1; 367 namelen = arg2; 368 369 if (req->newptr != NULL) 370 return (EPERM); 371 372 if (namelen != 1) 373 return (EINVAL); 374 375 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo)); 376 if (error) 377 return (error); 378 379 IN6_MULTI_LOCK(); 380 IN6_MULTI_LIST_LOCK(); 381 MLD_LOCK(); 382 383 if (name[0] <= 0 || name[0] > V_if_index) { 384 error = ENOENT; 385 goto out_locked; 386 } 387 388 error = ENOENT; 389 390 ifp = ifnet_byindex(name[0]); 391 if (ifp == NULL) 392 goto out_locked; 393 394 LIST_FOREACH(mli, &V_mli_head, mli_link) { 395 if (ifp == mli->mli_ifp) { 396 struct mld_ifinfo info; 397 398 info.mli_version = mli->mli_version; 399 info.mli_v1_timer = mli->mli_v1_timer; 400 info.mli_v2_timer = mli->mli_v2_timer; 401 info.mli_flags = mli->mli_flags; 402 info.mli_rv = mli->mli_rv; 403 info.mli_qi = mli->mli_qi; 404 info.mli_qri = mli->mli_qri; 405 info.mli_uri = mli->mli_uri; 406 error = SYSCTL_OUT(req, &info, sizeof(info)); 407 break; 408 } 409 } 410 411 out_locked: 412 MLD_UNLOCK(); 413 IN6_MULTI_LIST_UNLOCK(); 414 IN6_MULTI_UNLOCK(); 415 return (error); 416 } 417 418 /* 419 * Dispatch an entire queue of pending packet chains. 420 * VIMAGE: Assumes the vnet pointer has been set. 421 */ 422 static void 423 mld_dispatch_queue(struct mbufq *mq, int limit) 424 { 425 struct mbuf *m; 426 427 while ((m = mbufq_dequeue(mq)) != NULL) { 428 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m); 429 mld_dispatch_packet(m); 430 if (--limit == 0) 431 break; 432 } 433 } 434 435 /* 436 * Filter outgoing MLD report state by group. 437 * 438 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1) 439 * and node-local addresses. However, kernel and socket consumers 440 * always embed the KAME scope ID in the address provided, so strip it 441 * when performing comparison. 442 * Note: This is not the same as the *multicast* scope. 443 * 444 * Return zero if the given group is one for which MLD reports 445 * should be suppressed, or non-zero if reports should be issued. 446 */ 447 static __inline int 448 mld_is_addr_reported(const struct in6_addr *addr) 449 { 450 451 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__)); 452 453 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) 454 return (0); 455 456 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) { 457 struct in6_addr tmp = *addr; 458 in6_clearscope(&tmp); 459 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) 460 return (0); 461 } 462 463 return (1); 464 } 465 466 /* 467 * Attach MLD when PF_INET6 is attached to an interface. 468 * 469 * SMPng: Normally called with IF_AFDATA_LOCK held. 470 */ 471 struct mld_ifsoftc * 472 mld_domifattach(struct ifnet *ifp) 473 { 474 struct mld_ifsoftc *mli; 475 476 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 477 __func__, ifp, if_name(ifp)); 478 479 MLD_LOCK(); 480 481 mli = mli_alloc_locked(ifp); 482 if (!(ifp->if_flags & IFF_MULTICAST)) 483 mli->mli_flags |= MLIF_SILENT; 484 if (mld_use_allow) 485 mli->mli_flags |= MLIF_USEALLOW; 486 487 MLD_UNLOCK(); 488 489 return (mli); 490 } 491 492 /* 493 * VIMAGE: assume curvnet set by caller. 494 */ 495 static struct mld_ifsoftc * 496 mli_alloc_locked(/*const*/ struct ifnet *ifp) 497 { 498 struct mld_ifsoftc *mli; 499 500 MLD_LOCK_ASSERT(); 501 502 mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO); 503 if (mli == NULL) 504 goto out; 505 506 mli->mli_ifp = ifp; 507 mli->mli_version = MLD_VERSION_2; 508 mli->mli_flags = 0; 509 mli->mli_rv = MLD_RV_INIT; 510 mli->mli_qi = MLD_QI_INIT; 511 mli->mli_qri = MLD_QRI_INIT; 512 mli->mli_uri = MLD_URI_INIT; 513 mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); 514 515 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); 516 517 CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)", 518 ifp, if_name(ifp)); 519 520 out: 521 return (mli); 522 } 523 524 /* 525 * Hook for ifdetach. 526 * 527 * NOTE: Some finalization tasks need to run before the protocol domain 528 * is detached, but also before the link layer does its cleanup. 529 * Run before link-layer cleanup; cleanup groups, but do not free MLD state. 530 * 531 * SMPng: Caller must hold IN6_MULTI_LOCK(). 532 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. 533 * XXX This routine is also bitten by unlocked ifma_protospec access. 534 */ 535 void 536 mld_ifdetach(struct ifnet *ifp) 537 { 538 struct mld_ifsoftc *mli; 539 struct ifmultiaddr *ifma, *next; 540 struct in6_multi *inm; 541 struct in6_multi_head inmh; 542 543 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, 544 if_name(ifp)); 545 546 SLIST_INIT(&inmh); 547 IN6_MULTI_LIST_LOCK_ASSERT(); 548 MLD_LOCK(); 549 550 mli = MLD_IFINFO(ifp); 551 if (mli->mli_version == MLD_VERSION_2) { 552 IF_ADDR_WLOCK(ifp); 553 restart: 554 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { 555 if (ifma->ifma_addr->sa_family != AF_INET6 || 556 ifma->ifma_protospec == NULL) 557 continue; 558 inm = (struct in6_multi *)ifma->ifma_protospec; 559 if (inm->in6m_state == MLD_LEAVING_MEMBER) { 560 in6m_disconnect(inm); 561 in6m_rele_locked(&inmh, inm); 562 ifma->ifma_protospec = NULL; 563 } 564 in6m_clear_recorded(inm); 565 if (__predict_false(ifma6_restart)) { 566 ifma6_restart = false; 567 goto restart; 568 } 569 } 570 IF_ADDR_WUNLOCK(ifp); 571 } 572 573 MLD_UNLOCK(); 574 in6m_release_list_deferred(&inmh); 575 } 576 577 /* 578 * Hook for domifdetach. 579 * Runs after link-layer cleanup; free MLD state. 580 * 581 * SMPng: Normally called with IF_AFDATA_LOCK held. 582 */ 583 void 584 mld_domifdetach(struct ifnet *ifp) 585 { 586 587 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 588 __func__, ifp, if_name(ifp)); 589 590 MLD_LOCK(); 591 mli_delete_locked(ifp); 592 MLD_UNLOCK(); 593 } 594 595 static void 596 mli_delete_locked(const struct ifnet *ifp) 597 { 598 struct mld_ifsoftc *mli, *tmli; 599 600 CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)", 601 __func__, ifp, if_name(ifp)); 602 603 MLD_LOCK_ASSERT(); 604 605 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) { 606 if (mli->mli_ifp == ifp) { 607 /* 608 * Free deferred General Query responses. 609 */ 610 mbufq_drain(&mli->mli_gq); 611 612 LIST_REMOVE(mli, mli_link); 613 614 free(mli, M_MLD); 615 return; 616 } 617 } 618 } 619 620 /* 621 * Process a received MLDv1 general or address-specific query. 622 * Assumes that the query header has been pulled up to sizeof(mld_hdr). 623 * 624 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 625 * mld_addr. This is OK as we own the mbuf chain. 626 */ 627 static int 628 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 629 /*const*/ struct mld_hdr *mld) 630 { 631 struct epoch_tracker et; 632 struct ifmultiaddr *ifma; 633 struct mld_ifsoftc *mli; 634 struct in6_multi *inm; 635 int is_general_query; 636 uint16_t timer; 637 #ifdef KTR 638 char ip6tbuf[INET6_ADDRSTRLEN]; 639 #endif 640 641 is_general_query = 0; 642 643 if (!mld_v1enable) { 644 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)", 645 ip6_sprintf(ip6tbuf, &mld->mld_addr), 646 ifp, if_name(ifp)); 647 return (0); 648 } 649 650 /* 651 * RFC3810 Section 6.2: MLD queries must originate from 652 * a router's link-local address. 653 */ 654 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 655 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 656 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 657 ifp, if_name(ifp)); 658 return (0); 659 } 660 661 /* 662 * Do address field validation upfront before we accept 663 * the query. 664 */ 665 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 666 /* 667 * MLDv1 General Query. 668 * If this was not sent to the all-nodes group, ignore it. 669 */ 670 struct in6_addr dst; 671 672 dst = ip6->ip6_dst; 673 in6_clearscope(&dst); 674 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) 675 return (EINVAL); 676 is_general_query = 1; 677 } else { 678 /* 679 * Embed scope ID of receiving interface in MLD query for 680 * lookup whilst we don't hold other locks. 681 */ 682 in6_setscope(&mld->mld_addr, ifp, NULL); 683 } 684 685 IN6_MULTI_LIST_LOCK(); 686 MLD_LOCK(); 687 688 /* 689 * Switch to MLDv1 host compatibility mode. 690 */ 691 mli = MLD_IFINFO(ifp); 692 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); 693 mld_set_version(mli, MLD_VERSION_1); 694 695 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; 696 if (timer == 0) 697 timer = 1; 698 699 NET_EPOCH_ENTER(et); 700 if (is_general_query) { 701 /* 702 * For each reporting group joined on this 703 * interface, kick the report timer. 704 */ 705 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)", 706 ifp, if_name(ifp)); 707 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 708 if (ifma->ifma_addr->sa_family != AF_INET6 || 709 ifma->ifma_protospec == NULL) 710 continue; 711 inm = (struct in6_multi *)ifma->ifma_protospec; 712 mld_v1_update_group(inm, timer); 713 } 714 } else { 715 /* 716 * MLDv1 Group-Specific Query. 717 * If this is a group-specific MLDv1 query, we need only 718 * look up the single group to process it. 719 */ 720 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 721 if (inm != NULL) { 722 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)", 723 ip6_sprintf(ip6tbuf, &mld->mld_addr), 724 ifp, if_name(ifp)); 725 mld_v1_update_group(inm, timer); 726 } 727 /* XXX Clear embedded scope ID as userland won't expect it. */ 728 in6_clearscope(&mld->mld_addr); 729 } 730 731 NET_EPOCH_EXIT(et); 732 MLD_UNLOCK(); 733 IN6_MULTI_LIST_UNLOCK(); 734 735 return (0); 736 } 737 738 /* 739 * Update the report timer on a group in response to an MLDv1 query. 740 * 741 * If we are becoming the reporting member for this group, start the timer. 742 * If we already are the reporting member for this group, and timer is 743 * below the threshold, reset it. 744 * 745 * We may be updating the group for the first time since we switched 746 * to MLDv2. If we are, then we must clear any recorded source lists, 747 * and transition to REPORTING state; the group timer is overloaded 748 * for group and group-source query responses. 749 * 750 * Unlike MLDv2, the delay per group should be jittered 751 * to avoid bursts of MLDv1 reports. 752 */ 753 static void 754 mld_v1_update_group(struct in6_multi *inm, const int timer) 755 { 756 #ifdef KTR 757 char ip6tbuf[INET6_ADDRSTRLEN]; 758 #endif 759 760 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, 761 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 762 if_name(inm->in6m_ifp), timer); 763 764 IN6_MULTI_LIST_LOCK_ASSERT(); 765 766 switch (inm->in6m_state) { 767 case MLD_NOT_MEMBER: 768 case MLD_SILENT_MEMBER: 769 break; 770 case MLD_REPORTING_MEMBER: 771 if (inm->in6m_timer != 0 && 772 inm->in6m_timer <= timer) { 773 CTR1(KTR_MLD, "%s: REPORTING and timer running, " 774 "skipping.", __func__); 775 break; 776 } 777 /* FALLTHROUGH */ 778 case MLD_SG_QUERY_PENDING_MEMBER: 779 case MLD_G_QUERY_PENDING_MEMBER: 780 case MLD_IDLE_MEMBER: 781 case MLD_LAZY_MEMBER: 782 case MLD_AWAKENING_MEMBER: 783 CTR1(KTR_MLD, "%s: ->REPORTING", __func__); 784 inm->in6m_state = MLD_REPORTING_MEMBER; 785 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 786 V_current_state_timers_running6 = 1; 787 break; 788 case MLD_SLEEPING_MEMBER: 789 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); 790 inm->in6m_state = MLD_AWAKENING_MEMBER; 791 break; 792 case MLD_LEAVING_MEMBER: 793 break; 794 } 795 } 796 797 /* 798 * Process a received MLDv2 general, group-specific or 799 * group-and-source-specific query. 800 * 801 * Assumes that the query header has been pulled up to sizeof(mldv2_query). 802 * 803 * Return 0 if successful, otherwise an appropriate error code is returned. 804 */ 805 static int 806 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 807 struct mbuf *m, const int off, const int icmp6len) 808 { 809 struct mld_ifsoftc *mli; 810 struct mldv2_query *mld; 811 struct in6_multi *inm; 812 uint32_t maxdelay, nsrc, qqi; 813 int is_general_query; 814 uint16_t timer; 815 uint8_t qrv; 816 #ifdef KTR 817 char ip6tbuf[INET6_ADDRSTRLEN]; 818 #endif 819 820 is_general_query = 0; 821 822 /* 823 * RFC3810 Section 6.2: MLD queries must originate from 824 * a router's link-local address. 825 */ 826 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 827 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 828 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 829 ifp, if_name(ifp)); 830 return (0); 831 } 832 833 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp)); 834 835 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off); 836 837 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ 838 if (maxdelay >= 32768) { 839 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << 840 (MLD_MRC_EXP(maxdelay) + 3); 841 } 842 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; 843 if (timer == 0) 844 timer = 1; 845 846 qrv = MLD_QRV(mld->mld_misc); 847 if (qrv < 2) { 848 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, 849 qrv, MLD_RV_INIT); 850 qrv = MLD_RV_INIT; 851 } 852 853 qqi = mld->mld_qqi; 854 if (qqi >= 128) { 855 qqi = MLD_QQIC_MANT(mld->mld_qqi) << 856 (MLD_QQIC_EXP(mld->mld_qqi) + 3); 857 } 858 859 nsrc = ntohs(mld->mld_numsrc); 860 if (nsrc > MLD_MAX_GS_SOURCES) 861 return (EMSGSIZE); 862 if (icmp6len < sizeof(struct mldv2_query) + 863 (nsrc * sizeof(struct in6_addr))) 864 return (EMSGSIZE); 865 866 /* 867 * Do further input validation upfront to avoid resetting timers 868 * should we need to discard this query. 869 */ 870 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 871 /* 872 * A general query with a source list has undefined 873 * behaviour; discard it. 874 */ 875 if (nsrc > 0) 876 return (EINVAL); 877 is_general_query = 1; 878 } else { 879 /* 880 * Embed scope ID of receiving interface in MLD query for 881 * lookup whilst we don't hold other locks (due to KAME 882 * locking lameness). We own this mbuf chain just now. 883 */ 884 in6_setscope(&mld->mld_addr, ifp, NULL); 885 } 886 887 IN6_MULTI_LIST_LOCK(); 888 MLD_LOCK(); 889 890 mli = MLD_IFINFO(ifp); 891 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); 892 893 /* 894 * Discard the v2 query if we're in Compatibility Mode. 895 * The RFC is pretty clear that hosts need to stay in MLDv1 mode 896 * until the Old Version Querier Present timer expires. 897 */ 898 if (mli->mli_version != MLD_VERSION_2) 899 goto out_locked; 900 901 mld_set_version(mli, MLD_VERSION_2); 902 mli->mli_rv = qrv; 903 mli->mli_qi = qqi; 904 mli->mli_qri = maxdelay; 905 906 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, 907 maxdelay); 908 909 if (is_general_query) { 910 /* 911 * MLDv2 General Query. 912 * 913 * Schedule a current-state report on this ifp for 914 * all groups, possibly containing source lists. 915 * 916 * If there is a pending General Query response 917 * scheduled earlier than the selected delay, do 918 * not schedule any other reports. 919 * Otherwise, reset the interface timer. 920 */ 921 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", 922 ifp, if_name(ifp)); 923 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { 924 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); 925 V_interface_timers_running6 = 1; 926 } 927 } else { 928 struct epoch_tracker et; 929 930 /* 931 * MLDv2 Group-specific or Group-and-source-specific Query. 932 * 933 * Group-source-specific queries are throttled on 934 * a per-group basis to defeat denial-of-service attempts. 935 * Queries for groups we are not a member of on this 936 * link are simply ignored. 937 */ 938 NET_EPOCH_ENTER(et); 939 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 940 if (inm == NULL) { 941 NET_EPOCH_EXIT(et); 942 goto out_locked; 943 } 944 if (nsrc > 0) { 945 if (!ratecheck(&inm->in6m_lastgsrtv, 946 &V_mld_gsrdelay)) { 947 CTR1(KTR_MLD, "%s: GS query throttled.", 948 __func__); 949 NET_EPOCH_EXIT(et); 950 goto out_locked; 951 } 952 } 953 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", 954 ifp, if_name(ifp)); 955 /* 956 * If there is a pending General Query response 957 * scheduled sooner than the selected delay, no 958 * further report need be scheduled. 959 * Otherwise, prepare to respond to the 960 * group-specific or group-and-source query. 961 */ 962 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) 963 mld_v2_process_group_query(inm, mli, timer, m, off); 964 965 /* XXX Clear embedded scope ID as userland won't expect it. */ 966 in6_clearscope(&mld->mld_addr); 967 NET_EPOCH_EXIT(et); 968 } 969 970 out_locked: 971 MLD_UNLOCK(); 972 IN6_MULTI_LIST_UNLOCK(); 973 974 return (0); 975 } 976 977 /* 978 * Process a received MLDv2 group-specific or group-and-source-specific 979 * query. 980 * Return <0 if any error occurred. Currently this is ignored. 981 */ 982 static int 983 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli, 984 int timer, struct mbuf *m0, const int off) 985 { 986 struct mldv2_query *mld; 987 int retval; 988 uint16_t nsrc; 989 990 IN6_MULTI_LIST_LOCK_ASSERT(); 991 MLD_LOCK_ASSERT(); 992 993 retval = 0; 994 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off); 995 996 switch (inm->in6m_state) { 997 case MLD_NOT_MEMBER: 998 case MLD_SILENT_MEMBER: 999 case MLD_SLEEPING_MEMBER: 1000 case MLD_LAZY_MEMBER: 1001 case MLD_AWAKENING_MEMBER: 1002 case MLD_IDLE_MEMBER: 1003 case MLD_LEAVING_MEMBER: 1004 return (retval); 1005 break; 1006 case MLD_REPORTING_MEMBER: 1007 case MLD_G_QUERY_PENDING_MEMBER: 1008 case MLD_SG_QUERY_PENDING_MEMBER: 1009 break; 1010 } 1011 1012 nsrc = ntohs(mld->mld_numsrc); 1013 1014 /* 1015 * Deal with group-specific queries upfront. 1016 * If any group query is already pending, purge any recorded 1017 * source-list state if it exists, and schedule a query response 1018 * for this group-specific query. 1019 */ 1020 if (nsrc == 0) { 1021 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 1022 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { 1023 in6m_clear_recorded(inm); 1024 timer = min(inm->in6m_timer, timer); 1025 } 1026 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; 1027 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1028 V_current_state_timers_running6 = 1; 1029 return (retval); 1030 } 1031 1032 /* 1033 * Deal with the case where a group-and-source-specific query has 1034 * been received but a group-specific query is already pending. 1035 */ 1036 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { 1037 timer = min(inm->in6m_timer, timer); 1038 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1039 V_current_state_timers_running6 = 1; 1040 return (retval); 1041 } 1042 1043 /* 1044 * Finally, deal with the case where a group-and-source-specific 1045 * query has been received, where a response to a previous g-s-r 1046 * query exists, or none exists. 1047 * In this case, we need to parse the source-list which the Querier 1048 * has provided us with and check if we have any source list filter 1049 * entries at T1 for these sources. If we do not, there is no need 1050 * schedule a report and the query may be dropped. 1051 * If we do, we must record them and schedule a current-state 1052 * report for those sources. 1053 */ 1054 if (inm->in6m_nsrc > 0) { 1055 struct mbuf *m; 1056 uint8_t *sp; 1057 int i, nrecorded; 1058 int soff; 1059 1060 m = m0; 1061 soff = off + sizeof(struct mldv2_query); 1062 nrecorded = 0; 1063 for (i = 0; i < nsrc; i++) { 1064 sp = mtod(m, uint8_t *) + soff; 1065 retval = in6m_record_source(inm, 1066 (const struct in6_addr *)sp); 1067 if (retval < 0) 1068 break; 1069 nrecorded += retval; 1070 soff += sizeof(struct in6_addr); 1071 if (soff >= m->m_len) { 1072 soff = soff - m->m_len; 1073 m = m->m_next; 1074 if (m == NULL) 1075 break; 1076 } 1077 } 1078 if (nrecorded > 0) { 1079 CTR1(KTR_MLD, 1080 "%s: schedule response to SG query", __func__); 1081 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; 1082 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1083 V_current_state_timers_running6 = 1; 1084 } 1085 } 1086 1087 return (retval); 1088 } 1089 1090 /* 1091 * Process a received MLDv1 host membership report. 1092 * Assumes mld points to mld_hdr in pulled up mbuf chain. 1093 * 1094 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 1095 * mld_addr. This is OK as we own the mbuf chain. 1096 */ 1097 static int 1098 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, 1099 /*const*/ struct mld_hdr *mld) 1100 { 1101 struct in6_addr src, dst; 1102 struct epoch_tracker et; 1103 struct in6_ifaddr *ia; 1104 struct in6_multi *inm; 1105 #ifdef KTR 1106 char ip6tbuf[INET6_ADDRSTRLEN]; 1107 #endif 1108 1109 if (!mld_v1enable) { 1110 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", 1111 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1112 ifp, if_name(ifp)); 1113 return (0); 1114 } 1115 1116 if (ifp->if_flags & IFF_LOOPBACK) 1117 return (0); 1118 1119 /* 1120 * MLDv1 reports must originate from a host's link-local address, 1121 * or the unspecified address (when booting). 1122 */ 1123 src = ip6->ip6_src; 1124 in6_clearscope(&src); 1125 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { 1126 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 1127 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 1128 ifp, if_name(ifp)); 1129 return (EINVAL); 1130 } 1131 1132 /* 1133 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast 1134 * group, and must be directed to the group itself. 1135 */ 1136 dst = ip6->ip6_dst; 1137 in6_clearscope(&dst); 1138 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || 1139 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { 1140 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", 1141 ip6_sprintf(ip6tbuf, &ip6->ip6_dst), 1142 ifp, if_name(ifp)); 1143 return (EINVAL); 1144 } 1145 1146 /* 1147 * Make sure we don't hear our own membership report, as fast 1148 * leave requires knowing that we are the only member of a 1149 * group. Assume we used the link-local address if available, 1150 * otherwise look for ::. 1151 * 1152 * XXX Note that scope ID comparison is needed for the address 1153 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be 1154 * performed for the on-wire address. 1155 */ 1156 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1157 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || 1158 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) { 1159 if (ia != NULL) 1160 ifa_free(&ia->ia_ifa); 1161 return (0); 1162 } 1163 if (ia != NULL) 1164 ifa_free(&ia->ia_ifa); 1165 1166 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", 1167 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); 1168 1169 /* 1170 * Embed scope ID of receiving interface in MLD query for lookup 1171 * whilst we don't hold other locks (due to KAME locking lameness). 1172 */ 1173 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) 1174 in6_setscope(&mld->mld_addr, ifp, NULL); 1175 1176 IN6_MULTI_LIST_LOCK(); 1177 MLD_LOCK(); 1178 NET_EPOCH_ENTER(et); 1179 1180 /* 1181 * MLDv1 report suppression. 1182 * If we are a member of this group, and our membership should be 1183 * reported, and our group timer is pending or about to be reset, 1184 * stop our group timer by transitioning to the 'lazy' state. 1185 */ 1186 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 1187 if (inm != NULL) { 1188 struct mld_ifsoftc *mli; 1189 1190 mli = inm->in6m_mli; 1191 KASSERT(mli != NULL, 1192 ("%s: no mli for ifp %p", __func__, ifp)); 1193 1194 /* 1195 * If we are in MLDv2 host mode, do not allow the 1196 * other host's MLDv1 report to suppress our reports. 1197 */ 1198 if (mli->mli_version == MLD_VERSION_2) 1199 goto out_locked; 1200 1201 inm->in6m_timer = 0; 1202 1203 switch (inm->in6m_state) { 1204 case MLD_NOT_MEMBER: 1205 case MLD_SILENT_MEMBER: 1206 case MLD_SLEEPING_MEMBER: 1207 break; 1208 case MLD_REPORTING_MEMBER: 1209 case MLD_IDLE_MEMBER: 1210 case MLD_AWAKENING_MEMBER: 1211 CTR3(KTR_MLD, 1212 "report suppressed for %s on ifp %p(%s)", 1213 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1214 ifp, if_name(ifp)); 1215 case MLD_LAZY_MEMBER: 1216 inm->in6m_state = MLD_LAZY_MEMBER; 1217 break; 1218 case MLD_G_QUERY_PENDING_MEMBER: 1219 case MLD_SG_QUERY_PENDING_MEMBER: 1220 case MLD_LEAVING_MEMBER: 1221 break; 1222 } 1223 } 1224 1225 out_locked: 1226 NET_EPOCH_EXIT(et); 1227 MLD_UNLOCK(); 1228 IN6_MULTI_LIST_UNLOCK(); 1229 1230 /* XXX Clear embedded scope ID as userland won't expect it. */ 1231 in6_clearscope(&mld->mld_addr); 1232 1233 return (0); 1234 } 1235 1236 /* 1237 * MLD input path. 1238 * 1239 * Assume query messages which fit in a single ICMPv6 message header 1240 * have been pulled up. 1241 * Assume that userland will want to see the message, even if it 1242 * otherwise fails kernel input validation; do not free it. 1243 * Pullup may however free the mbuf chain m if it fails. 1244 * 1245 * Return IPPROTO_DONE if we freed m. Otherwise, return 0. 1246 */ 1247 int 1248 mld_input(struct mbuf *m, int off, int icmp6len) 1249 { 1250 struct ifnet *ifp; 1251 struct ip6_hdr *ip6; 1252 struct mld_hdr *mld; 1253 int mldlen; 1254 1255 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); 1256 1257 ifp = m->m_pkthdr.rcvif; 1258 1259 ip6 = mtod(m, struct ip6_hdr *); 1260 1261 /* Pullup to appropriate size. */ 1262 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); 1263 if (mld->mld_type == MLD_LISTENER_QUERY && 1264 icmp6len >= sizeof(struct mldv2_query)) { 1265 mldlen = sizeof(struct mldv2_query); 1266 } else { 1267 mldlen = sizeof(struct mld_hdr); 1268 } 1269 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen); 1270 if (mld == NULL) { 1271 ICMP6STAT_INC(icp6s_badlen); 1272 return (IPPROTO_DONE); 1273 } 1274 1275 /* 1276 * Userland needs to see all of this traffic for implementing 1277 * the endpoint discovery portion of multicast routing. 1278 */ 1279 switch (mld->mld_type) { 1280 case MLD_LISTENER_QUERY: 1281 icmp6_ifstat_inc(ifp, ifs6_in_mldquery); 1282 if (icmp6len == sizeof(struct mld_hdr)) { 1283 if (mld_v1_input_query(ifp, ip6, mld) != 0) 1284 return (0); 1285 } else if (icmp6len >= sizeof(struct mldv2_query)) { 1286 if (mld_v2_input_query(ifp, ip6, m, off, 1287 icmp6len) != 0) 1288 return (0); 1289 } 1290 break; 1291 case MLD_LISTENER_REPORT: 1292 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1293 if (mld_v1_input_report(ifp, ip6, mld) != 0) 1294 return (0); 1295 break; 1296 case MLDV2_LISTENER_REPORT: 1297 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1298 break; 1299 case MLD_LISTENER_DONE: 1300 icmp6_ifstat_inc(ifp, ifs6_in_mlddone); 1301 break; 1302 default: 1303 break; 1304 } 1305 1306 return (0); 1307 } 1308 1309 /* 1310 * Fast timeout handler (global). 1311 * VIMAGE: Timeout handlers are expected to service all vimages. 1312 */ 1313 void 1314 mld_fasttimo(void) 1315 { 1316 VNET_ITERATOR_DECL(vnet_iter); 1317 1318 VNET_LIST_RLOCK_NOSLEEP(); 1319 VNET_FOREACH(vnet_iter) { 1320 CURVNET_SET(vnet_iter); 1321 mld_fasttimo_vnet(); 1322 CURVNET_RESTORE(); 1323 } 1324 VNET_LIST_RUNLOCK_NOSLEEP(); 1325 } 1326 1327 /* 1328 * Fast timeout handler (per-vnet). 1329 * 1330 * VIMAGE: Assume caller has set up our curvnet. 1331 */ 1332 static void 1333 mld_fasttimo_vnet(void) 1334 { 1335 struct mbufq scq; /* State-change packets */ 1336 struct mbufq qrq; /* Query response packets */ 1337 struct ifnet *ifp; 1338 struct mld_ifsoftc *mli; 1339 struct ifmultiaddr *ifma, *next; 1340 struct in6_multi *inm, *tinm; 1341 struct in6_multi_head inmh; 1342 int uri_fasthz; 1343 1344 uri_fasthz = 0; 1345 1346 /* 1347 * Quick check to see if any work needs to be done, in order to 1348 * minimize the overhead of fasttimo processing. 1349 * SMPng: XXX Unlocked reads. 1350 */ 1351 if (!V_current_state_timers_running6 && 1352 !V_interface_timers_running6 && 1353 !V_state_change_timers_running6) 1354 return; 1355 1356 SLIST_INIT(&inmh); 1357 IN6_MULTI_LIST_LOCK(); 1358 MLD_LOCK(); 1359 1360 /* 1361 * MLDv2 General Query response timer processing. 1362 */ 1363 if (V_interface_timers_running6) { 1364 CTR1(KTR_MLD, "%s: interface timers running", __func__); 1365 1366 V_interface_timers_running6 = 0; 1367 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1368 if (mli->mli_v2_timer == 0) { 1369 /* Do nothing. */ 1370 } else if (--mli->mli_v2_timer == 0) { 1371 mld_v2_dispatch_general_query(mli); 1372 } else { 1373 V_interface_timers_running6 = 1; 1374 } 1375 } 1376 } 1377 1378 if (!V_current_state_timers_running6 && 1379 !V_state_change_timers_running6) 1380 goto out_locked; 1381 1382 V_current_state_timers_running6 = 0; 1383 V_state_change_timers_running6 = 0; 1384 1385 CTR1(KTR_MLD, "%s: state change timers running", __func__); 1386 1387 /* 1388 * MLD host report and state-change timer processing. 1389 * Note: Processing a v2 group timer may remove a node. 1390 */ 1391 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1392 ifp = mli->mli_ifp; 1393 1394 if (mli->mli_version == MLD_VERSION_2) { 1395 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * 1396 PR_FASTHZ); 1397 mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS); 1398 mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS); 1399 } 1400 1401 IF_ADDR_WLOCK(ifp); 1402 restart: 1403 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { 1404 if (ifma->ifma_addr->sa_family != AF_INET6 || 1405 ifma->ifma_protospec == NULL) 1406 continue; 1407 inm = (struct in6_multi *)ifma->ifma_protospec; 1408 switch (mli->mli_version) { 1409 case MLD_VERSION_1: 1410 mld_v1_process_group_timer(&inmh, inm); 1411 break; 1412 case MLD_VERSION_2: 1413 mld_v2_process_group_timers(&inmh, &qrq, 1414 &scq, inm, uri_fasthz); 1415 break; 1416 } 1417 if (__predict_false(ifma6_restart)) { 1418 ifma6_restart = false; 1419 goto restart; 1420 } 1421 } 1422 IF_ADDR_WUNLOCK(ifp); 1423 1424 switch (mli->mli_version) { 1425 case MLD_VERSION_1: 1426 /* 1427 * Transmit reports for this lifecycle. This 1428 * is done while not holding IF_ADDR_LOCK 1429 * since this can call 1430 * in6ifa_ifpforlinklocal() which locks 1431 * IF_ADDR_LOCK internally as well as 1432 * ip6_output() to transmit a packet. 1433 */ 1434 SLIST_FOREACH_SAFE(inm, &inmh, in6m_nrele, tinm) { 1435 SLIST_REMOVE_HEAD(&inmh, 1436 in6m_nrele); 1437 (void)mld_v1_transmit_report(inm, 1438 MLD_LISTENER_REPORT); 1439 } 1440 break; 1441 case MLD_VERSION_2: 1442 mld_dispatch_queue(&qrq, 0); 1443 mld_dispatch_queue(&scq, 0); 1444 1445 /* 1446 * Free the in_multi reference(s) for 1447 * this lifecycle. 1448 */ 1449 in6m_release_list_deferred(&inmh); 1450 break; 1451 } 1452 } 1453 1454 out_locked: 1455 MLD_UNLOCK(); 1456 IN6_MULTI_LIST_UNLOCK(); 1457 } 1458 1459 /* 1460 * Update host report group timer. 1461 * Will update the global pending timer flags. 1462 */ 1463 static void 1464 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm) 1465 { 1466 int report_timer_expired; 1467 1468 IN6_MULTI_LIST_LOCK_ASSERT(); 1469 MLD_LOCK_ASSERT(); 1470 1471 if (inm->in6m_timer == 0) { 1472 report_timer_expired = 0; 1473 } else if (--inm->in6m_timer == 0) { 1474 report_timer_expired = 1; 1475 } else { 1476 V_current_state_timers_running6 = 1; 1477 return; 1478 } 1479 1480 switch (inm->in6m_state) { 1481 case MLD_NOT_MEMBER: 1482 case MLD_SILENT_MEMBER: 1483 case MLD_IDLE_MEMBER: 1484 case MLD_LAZY_MEMBER: 1485 case MLD_SLEEPING_MEMBER: 1486 case MLD_AWAKENING_MEMBER: 1487 break; 1488 case MLD_REPORTING_MEMBER: 1489 if (report_timer_expired) { 1490 inm->in6m_state = MLD_IDLE_MEMBER; 1491 in6m_disconnect(inm); 1492 in6m_rele_locked(inmh, inm); 1493 } 1494 break; 1495 case MLD_G_QUERY_PENDING_MEMBER: 1496 case MLD_SG_QUERY_PENDING_MEMBER: 1497 case MLD_LEAVING_MEMBER: 1498 break; 1499 } 1500 } 1501 1502 /* 1503 * Update a group's timers for MLDv2. 1504 * Will update the global pending timer flags. 1505 * Note: Unlocked read from mli. 1506 */ 1507 static void 1508 mld_v2_process_group_timers(struct in6_multi_head *inmh, 1509 struct mbufq *qrq, struct mbufq *scq, 1510 struct in6_multi *inm, const int uri_fasthz) 1511 { 1512 int query_response_timer_expired; 1513 int state_change_retransmit_timer_expired; 1514 #ifdef KTR 1515 char ip6tbuf[INET6_ADDRSTRLEN]; 1516 #endif 1517 1518 IN6_MULTI_LIST_LOCK_ASSERT(); 1519 MLD_LOCK_ASSERT(); 1520 1521 query_response_timer_expired = 0; 1522 state_change_retransmit_timer_expired = 0; 1523 1524 /* 1525 * During a transition from compatibility mode back to MLDv2, 1526 * a group record in REPORTING state may still have its group 1527 * timer active. This is a no-op in this function; it is easier 1528 * to deal with it here than to complicate the slow-timeout path. 1529 */ 1530 if (inm->in6m_timer == 0) { 1531 query_response_timer_expired = 0; 1532 } else if (--inm->in6m_timer == 0) { 1533 query_response_timer_expired = 1; 1534 } else { 1535 V_current_state_timers_running6 = 1; 1536 } 1537 1538 if (inm->in6m_sctimer == 0) { 1539 state_change_retransmit_timer_expired = 0; 1540 } else if (--inm->in6m_sctimer == 0) { 1541 state_change_retransmit_timer_expired = 1; 1542 } else { 1543 V_state_change_timers_running6 = 1; 1544 } 1545 1546 /* We are in fasttimo, so be quick about it. */ 1547 if (!state_change_retransmit_timer_expired && 1548 !query_response_timer_expired) 1549 return; 1550 1551 switch (inm->in6m_state) { 1552 case MLD_NOT_MEMBER: 1553 case MLD_SILENT_MEMBER: 1554 case MLD_SLEEPING_MEMBER: 1555 case MLD_LAZY_MEMBER: 1556 case MLD_AWAKENING_MEMBER: 1557 case MLD_IDLE_MEMBER: 1558 break; 1559 case MLD_G_QUERY_PENDING_MEMBER: 1560 case MLD_SG_QUERY_PENDING_MEMBER: 1561 /* 1562 * Respond to a previously pending Group-Specific 1563 * or Group-and-Source-Specific query by enqueueing 1564 * the appropriate Current-State report for 1565 * immediate transmission. 1566 */ 1567 if (query_response_timer_expired) { 1568 int retval; 1569 1570 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, 1571 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER), 1572 0); 1573 CTR2(KTR_MLD, "%s: enqueue record = %d", 1574 __func__, retval); 1575 inm->in6m_state = MLD_REPORTING_MEMBER; 1576 in6m_clear_recorded(inm); 1577 } 1578 /* FALLTHROUGH */ 1579 case MLD_REPORTING_MEMBER: 1580 case MLD_LEAVING_MEMBER: 1581 if (state_change_retransmit_timer_expired) { 1582 /* 1583 * State-change retransmission timer fired. 1584 * If there are any further pending retransmissions, 1585 * set the global pending state-change flag, and 1586 * reset the timer. 1587 */ 1588 if (--inm->in6m_scrv > 0) { 1589 inm->in6m_sctimer = uri_fasthz; 1590 V_state_change_timers_running6 = 1; 1591 } 1592 /* 1593 * Retransmit the previously computed state-change 1594 * report. If there are no further pending 1595 * retransmissions, the mbuf queue will be consumed. 1596 * Update T0 state to T1 as we have now sent 1597 * a state-change. 1598 */ 1599 (void)mld_v2_merge_state_changes(inm, scq); 1600 1601 in6m_commit(inm); 1602 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 1603 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1604 if_name(inm->in6m_ifp)); 1605 1606 /* 1607 * If we are leaving the group for good, make sure 1608 * we release MLD's reference to it. 1609 * This release must be deferred using a SLIST, 1610 * as we are called from a loop which traverses 1611 * the in_ifmultiaddr TAILQ. 1612 */ 1613 if (inm->in6m_state == MLD_LEAVING_MEMBER && 1614 inm->in6m_scrv == 0) { 1615 inm->in6m_state = MLD_NOT_MEMBER; 1616 in6m_disconnect(inm); 1617 in6m_rele_locked(inmh, inm); 1618 } 1619 } 1620 break; 1621 } 1622 } 1623 1624 /* 1625 * Switch to a different version on the given interface, 1626 * as per Section 9.12. 1627 */ 1628 static void 1629 mld_set_version(struct mld_ifsoftc *mli, const int version) 1630 { 1631 int old_version_timer; 1632 1633 MLD_LOCK_ASSERT(); 1634 1635 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, 1636 version, mli->mli_ifp, if_name(mli->mli_ifp)); 1637 1638 if (version == MLD_VERSION_1) { 1639 /* 1640 * Compute the "Older Version Querier Present" timer as per 1641 * Section 9.12. 1642 */ 1643 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; 1644 old_version_timer *= PR_SLOWHZ; 1645 mli->mli_v1_timer = old_version_timer; 1646 } 1647 1648 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { 1649 mli->mli_version = MLD_VERSION_1; 1650 mld_v2_cancel_link_timers(mli); 1651 } 1652 } 1653 1654 /* 1655 * Cancel pending MLDv2 timers for the given link and all groups 1656 * joined on it; state-change, general-query, and group-query timers. 1657 */ 1658 static void 1659 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli) 1660 { 1661 struct ifmultiaddr *ifma, *next; 1662 struct ifnet *ifp; 1663 struct in6_multi *inm; 1664 struct in6_multi_head inmh; 1665 1666 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, 1667 mli->mli_ifp, if_name(mli->mli_ifp)); 1668 1669 SLIST_INIT(&inmh); 1670 IN6_MULTI_LIST_LOCK_ASSERT(); 1671 MLD_LOCK_ASSERT(); 1672 1673 /* 1674 * Fast-track this potentially expensive operation 1675 * by checking all the global 'timer pending' flags. 1676 */ 1677 if (!V_interface_timers_running6 && 1678 !V_state_change_timers_running6 && 1679 !V_current_state_timers_running6) 1680 return; 1681 1682 mli->mli_v2_timer = 0; 1683 1684 ifp = mli->mli_ifp; 1685 1686 IF_ADDR_WLOCK(ifp); 1687 restart: 1688 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { 1689 if (ifma->ifma_addr->sa_family != AF_INET6 || 1690 ifma->ifma_protospec == NULL) 1691 continue; 1692 inm = (struct in6_multi *)ifma->ifma_protospec; 1693 switch (inm->in6m_state) { 1694 case MLD_NOT_MEMBER: 1695 case MLD_SILENT_MEMBER: 1696 case MLD_IDLE_MEMBER: 1697 case MLD_LAZY_MEMBER: 1698 case MLD_SLEEPING_MEMBER: 1699 case MLD_AWAKENING_MEMBER: 1700 break; 1701 case MLD_LEAVING_MEMBER: 1702 /* 1703 * If we are leaving the group and switching 1704 * version, we need to release the final 1705 * reference held for issuing the INCLUDE {}. 1706 */ 1707 in6m_disconnect(inm); 1708 in6m_rele_locked(&inmh, inm); 1709 ifma->ifma_protospec = NULL; 1710 /* FALLTHROUGH */ 1711 case MLD_G_QUERY_PENDING_MEMBER: 1712 case MLD_SG_QUERY_PENDING_MEMBER: 1713 in6m_clear_recorded(inm); 1714 /* FALLTHROUGH */ 1715 case MLD_REPORTING_MEMBER: 1716 inm->in6m_sctimer = 0; 1717 inm->in6m_timer = 0; 1718 inm->in6m_state = MLD_REPORTING_MEMBER; 1719 /* 1720 * Free any pending MLDv2 state-change records. 1721 */ 1722 mbufq_drain(&inm->in6m_scq); 1723 break; 1724 } 1725 if (__predict_false(ifma6_restart)) { 1726 ifma6_restart = false; 1727 goto restart; 1728 } 1729 } 1730 IF_ADDR_WUNLOCK(ifp); 1731 in6m_release_list_deferred(&inmh); 1732 } 1733 1734 /* 1735 * Global slowtimo handler. 1736 * VIMAGE: Timeout handlers are expected to service all vimages. 1737 */ 1738 void 1739 mld_slowtimo(void) 1740 { 1741 VNET_ITERATOR_DECL(vnet_iter); 1742 1743 VNET_LIST_RLOCK_NOSLEEP(); 1744 VNET_FOREACH(vnet_iter) { 1745 CURVNET_SET(vnet_iter); 1746 mld_slowtimo_vnet(); 1747 CURVNET_RESTORE(); 1748 } 1749 VNET_LIST_RUNLOCK_NOSLEEP(); 1750 } 1751 1752 /* 1753 * Per-vnet slowtimo handler. 1754 */ 1755 static void 1756 mld_slowtimo_vnet(void) 1757 { 1758 struct mld_ifsoftc *mli; 1759 1760 MLD_LOCK(); 1761 1762 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1763 mld_v1_process_querier_timers(mli); 1764 } 1765 1766 MLD_UNLOCK(); 1767 } 1768 1769 /* 1770 * Update the Older Version Querier Present timers for a link. 1771 * See Section 9.12 of RFC 3810. 1772 */ 1773 static void 1774 mld_v1_process_querier_timers(struct mld_ifsoftc *mli) 1775 { 1776 1777 MLD_LOCK_ASSERT(); 1778 1779 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) { 1780 /* 1781 * MLDv1 Querier Present timer expired; revert to MLDv2. 1782 */ 1783 CTR5(KTR_MLD, 1784 "%s: transition from v%d -> v%d on %p(%s)", 1785 __func__, mli->mli_version, MLD_VERSION_2, 1786 mli->mli_ifp, if_name(mli->mli_ifp)); 1787 mli->mli_version = MLD_VERSION_2; 1788 } 1789 } 1790 1791 /* 1792 * Transmit an MLDv1 report immediately. 1793 */ 1794 static int 1795 mld_v1_transmit_report(struct in6_multi *in6m, const int type) 1796 { 1797 struct ifnet *ifp; 1798 struct in6_ifaddr *ia; 1799 struct ip6_hdr *ip6; 1800 struct mbuf *mh, *md; 1801 struct mld_hdr *mld; 1802 1803 IN6_MULTI_LIST_LOCK_ASSERT(); 1804 MLD_LOCK_ASSERT(); 1805 1806 ifp = in6m->in6m_ifp; 1807 /* in process of being freed */ 1808 if (ifp == NULL) 1809 return (0); 1810 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1811 /* ia may be NULL if link-local address is tentative. */ 1812 1813 mh = m_gethdr(M_NOWAIT, MT_DATA); 1814 if (mh == NULL) { 1815 if (ia != NULL) 1816 ifa_free(&ia->ia_ifa); 1817 return (ENOMEM); 1818 } 1819 md = m_get(M_NOWAIT, MT_DATA); 1820 if (md == NULL) { 1821 m_free(mh); 1822 if (ia != NULL) 1823 ifa_free(&ia->ia_ifa); 1824 return (ENOMEM); 1825 } 1826 mh->m_next = md; 1827 1828 /* 1829 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so 1830 * that ether_output() does not need to allocate another mbuf 1831 * for the header in the most common case. 1832 */ 1833 M_ALIGN(mh, sizeof(struct ip6_hdr)); 1834 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); 1835 mh->m_len = sizeof(struct ip6_hdr); 1836 1837 ip6 = mtod(mh, struct ip6_hdr *); 1838 ip6->ip6_flow = 0; 1839 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 1840 ip6->ip6_vfc |= IPV6_VERSION; 1841 ip6->ip6_nxt = IPPROTO_ICMPV6; 1842 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 1843 ip6->ip6_dst = in6m->in6m_addr; 1844 1845 md->m_len = sizeof(struct mld_hdr); 1846 mld = mtod(md, struct mld_hdr *); 1847 mld->mld_type = type; 1848 mld->mld_code = 0; 1849 mld->mld_cksum = 0; 1850 mld->mld_maxdelay = 0; 1851 mld->mld_reserved = 0; 1852 mld->mld_addr = in6m->in6m_addr; 1853 in6_clearscope(&mld->mld_addr); 1854 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 1855 sizeof(struct ip6_hdr), sizeof(struct mld_hdr)); 1856 1857 mld_save_context(mh, ifp); 1858 mh->m_flags |= M_MLDV1; 1859 1860 mld_dispatch_packet(mh); 1861 1862 if (ia != NULL) 1863 ifa_free(&ia->ia_ifa); 1864 return (0); 1865 } 1866 1867 /* 1868 * Process a state change from the upper layer for the given IPv6 group. 1869 * 1870 * Each socket holds a reference on the in_multi in its own ip_moptions. 1871 * The socket layer will have made the necessary updates to.the group 1872 * state, it is now up to MLD to issue a state change report if there 1873 * has been any change between T0 (when the last state-change was issued) 1874 * and T1 (now). 1875 * 1876 * We use the MLDv2 state machine at group level. The MLd module 1877 * however makes the decision as to which MLD protocol version to speak. 1878 * A state change *from* INCLUDE {} always means an initial join. 1879 * A state change *to* INCLUDE {} always means a final leave. 1880 * 1881 * If delay is non-zero, and the state change is an initial multicast 1882 * join, the state change report will be delayed by 'delay' ticks 1883 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise 1884 * the initial MLDv2 state change report will be delayed by whichever 1885 * is sooner, a pending state-change timer or delay itself. 1886 * 1887 * VIMAGE: curvnet should have been set by caller, as this routine 1888 * is called from the socket option handlers. 1889 */ 1890 int 1891 mld_change_state(struct in6_multi *inm, const int delay) 1892 { 1893 struct mld_ifsoftc *mli; 1894 struct ifnet *ifp; 1895 int error; 1896 1897 IN6_MULTI_LIST_LOCK_ASSERT(); 1898 1899 error = 0; 1900 1901 /* 1902 * Try to detect if the upper layer just asked us to change state 1903 * for an interface which has now gone away. 1904 */ 1905 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 1906 ifp = inm->in6m_ifma->ifma_ifp; 1907 if (ifp == NULL) 1908 return (0); 1909 /* 1910 * Sanity check that netinet6's notion of ifp is the 1911 * same as net's. 1912 */ 1913 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 1914 1915 MLD_LOCK(); 1916 mli = MLD_IFINFO(ifp); 1917 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); 1918 1919 /* 1920 * If we detect a state transition to or from MCAST_UNDEFINED 1921 * for this group, then we are starting or finishing an MLD 1922 * life cycle for this group. 1923 */ 1924 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { 1925 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, 1926 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); 1927 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { 1928 CTR1(KTR_MLD, "%s: initial join", __func__); 1929 error = mld_initial_join(inm, mli, delay); 1930 goto out_locked; 1931 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { 1932 CTR1(KTR_MLD, "%s: final leave", __func__); 1933 mld_final_leave(inm, mli); 1934 goto out_locked; 1935 } 1936 } else { 1937 CTR1(KTR_MLD, "%s: filter set change", __func__); 1938 } 1939 1940 error = mld_handle_state_change(inm, mli); 1941 1942 out_locked: 1943 MLD_UNLOCK(); 1944 return (error); 1945 } 1946 1947 /* 1948 * Perform the initial join for an MLD group. 1949 * 1950 * When joining a group: 1951 * If the group should have its MLD traffic suppressed, do nothing. 1952 * MLDv1 starts sending MLDv1 host membership reports. 1953 * MLDv2 will schedule an MLDv2 state-change report containing the 1954 * initial state of the membership. 1955 * 1956 * If the delay argument is non-zero, then we must delay sending the 1957 * initial state change for delay ticks (in units of PR_FASTHZ). 1958 */ 1959 static int 1960 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli, 1961 const int delay) 1962 { 1963 struct ifnet *ifp; 1964 struct mbufq *mq; 1965 int error, retval, syncstates; 1966 int odelay; 1967 #ifdef KTR 1968 char ip6tbuf[INET6_ADDRSTRLEN]; 1969 #endif 1970 1971 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", 1972 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1973 inm->in6m_ifp, if_name(inm->in6m_ifp)); 1974 1975 error = 0; 1976 syncstates = 1; 1977 1978 ifp = inm->in6m_ifp; 1979 1980 IN6_MULTI_LIST_LOCK_ASSERT(); 1981 MLD_LOCK_ASSERT(); 1982 1983 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); 1984 1985 /* 1986 * Groups joined on loopback or marked as 'not reported', 1987 * enter the MLD_SILENT_MEMBER state and 1988 * are never reported in any protocol exchanges. 1989 * All other groups enter the appropriate state machine 1990 * for the version in use on this link. 1991 * A link marked as MLIF_SILENT causes MLD to be completely 1992 * disabled for the link. 1993 */ 1994 if ((ifp->if_flags & IFF_LOOPBACK) || 1995 (mli->mli_flags & MLIF_SILENT) || 1996 !mld_is_addr_reported(&inm->in6m_addr)) { 1997 CTR1(KTR_MLD, 1998 "%s: not kicking state machine for silent group", __func__); 1999 inm->in6m_state = MLD_SILENT_MEMBER; 2000 inm->in6m_timer = 0; 2001 } else { 2002 /* 2003 * Deal with overlapping in_multi lifecycle. 2004 * If this group was LEAVING, then make sure 2005 * we drop the reference we picked up to keep the 2006 * group around for the final INCLUDE {} enqueue. 2007 */ 2008 if (mli->mli_version == MLD_VERSION_2 && 2009 inm->in6m_state == MLD_LEAVING_MEMBER) { 2010 inm->in6m_refcount--; 2011 } 2012 inm->in6m_state = MLD_REPORTING_MEMBER; 2013 2014 switch (mli->mli_version) { 2015 case MLD_VERSION_1: 2016 /* 2017 * If a delay was provided, only use it if 2018 * it is greater than the delay normally 2019 * used for an MLDv1 state change report, 2020 * and delay sending the initial MLDv1 report 2021 * by not transitioning to the IDLE state. 2022 */ 2023 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); 2024 if (delay) { 2025 inm->in6m_timer = max(delay, odelay); 2026 V_current_state_timers_running6 = 1; 2027 } else { 2028 inm->in6m_state = MLD_IDLE_MEMBER; 2029 error = mld_v1_transmit_report(inm, 2030 MLD_LISTENER_REPORT); 2031 if (error == 0) { 2032 inm->in6m_timer = odelay; 2033 V_current_state_timers_running6 = 1; 2034 } 2035 } 2036 break; 2037 2038 case MLD_VERSION_2: 2039 /* 2040 * Defer update of T0 to T1, until the first copy 2041 * of the state change has been transmitted. 2042 */ 2043 syncstates = 0; 2044 2045 /* 2046 * Immediately enqueue a State-Change Report for 2047 * this interface, freeing any previous reports. 2048 * Don't kick the timers if there is nothing to do, 2049 * or if an error occurred. 2050 */ 2051 mq = &inm->in6m_scq; 2052 mbufq_drain(mq); 2053 retval = mld_v2_enqueue_group_record(mq, inm, 1, 2054 0, 0, (mli->mli_flags & MLIF_USEALLOW)); 2055 CTR2(KTR_MLD, "%s: enqueue record = %d", 2056 __func__, retval); 2057 if (retval <= 0) { 2058 error = retval * -1; 2059 break; 2060 } 2061 2062 /* 2063 * Schedule transmission of pending state-change 2064 * report up to RV times for this link. The timer 2065 * will fire at the next mld_fasttimo (~200ms), 2066 * giving us an opportunity to merge the reports. 2067 * 2068 * If a delay was provided to this function, only 2069 * use this delay if sooner than the existing one. 2070 */ 2071 KASSERT(mli->mli_rv > 1, 2072 ("%s: invalid robustness %d", __func__, 2073 mli->mli_rv)); 2074 inm->in6m_scrv = mli->mli_rv; 2075 if (delay) { 2076 if (inm->in6m_sctimer > 1) { 2077 inm->in6m_sctimer = 2078 min(inm->in6m_sctimer, delay); 2079 } else 2080 inm->in6m_sctimer = delay; 2081 } else 2082 inm->in6m_sctimer = 1; 2083 V_state_change_timers_running6 = 1; 2084 2085 error = 0; 2086 break; 2087 } 2088 } 2089 2090 /* 2091 * Only update the T0 state if state change is atomic, 2092 * i.e. we don't need to wait for a timer to fire before we 2093 * can consider the state change to have been communicated. 2094 */ 2095 if (syncstates) { 2096 in6m_commit(inm); 2097 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2098 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2099 if_name(inm->in6m_ifp)); 2100 } 2101 2102 return (error); 2103 } 2104 2105 /* 2106 * Issue an intermediate state change during the life-cycle. 2107 */ 2108 static int 2109 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli) 2110 { 2111 struct ifnet *ifp; 2112 int retval; 2113 #ifdef KTR 2114 char ip6tbuf[INET6_ADDRSTRLEN]; 2115 #endif 2116 2117 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)", 2118 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2119 inm->in6m_ifp, if_name(inm->in6m_ifp)); 2120 2121 ifp = inm->in6m_ifp; 2122 2123 IN6_MULTI_LIST_LOCK_ASSERT(); 2124 MLD_LOCK_ASSERT(); 2125 2126 KASSERT(mli && mli->mli_ifp == ifp, 2127 ("%s: inconsistent ifp", __func__)); 2128 2129 if ((ifp->if_flags & IFF_LOOPBACK) || 2130 (mli->mli_flags & MLIF_SILENT) || 2131 !mld_is_addr_reported(&inm->in6m_addr) || 2132 (mli->mli_version != MLD_VERSION_2)) { 2133 if (!mld_is_addr_reported(&inm->in6m_addr)) { 2134 CTR1(KTR_MLD, 2135 "%s: not kicking state machine for silent group", __func__); 2136 } 2137 CTR1(KTR_MLD, "%s: nothing to do", __func__); 2138 in6m_commit(inm); 2139 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2140 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2141 if_name(inm->in6m_ifp)); 2142 return (0); 2143 } 2144 2145 mbufq_drain(&inm->in6m_scq); 2146 2147 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0, 2148 (mli->mli_flags & MLIF_USEALLOW)); 2149 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); 2150 if (retval <= 0) 2151 return (-retval); 2152 2153 /* 2154 * If record(s) were enqueued, start the state-change 2155 * report timer for this group. 2156 */ 2157 inm->in6m_scrv = mli->mli_rv; 2158 inm->in6m_sctimer = 1; 2159 V_state_change_timers_running6 = 1; 2160 2161 return (0); 2162 } 2163 2164 /* 2165 * Perform the final leave for a multicast address. 2166 * 2167 * When leaving a group: 2168 * MLDv1 sends a DONE message, if and only if we are the reporter. 2169 * MLDv2 enqueues a state-change report containing a transition 2170 * to INCLUDE {} for immediate transmission. 2171 */ 2172 static void 2173 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli) 2174 { 2175 int syncstates; 2176 #ifdef KTR 2177 char ip6tbuf[INET6_ADDRSTRLEN]; 2178 #endif 2179 2180 syncstates = 1; 2181 2182 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)", 2183 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2184 inm->in6m_ifp, if_name(inm->in6m_ifp)); 2185 2186 IN6_MULTI_LIST_LOCK_ASSERT(); 2187 MLD_LOCK_ASSERT(); 2188 2189 switch (inm->in6m_state) { 2190 case MLD_NOT_MEMBER: 2191 case MLD_SILENT_MEMBER: 2192 case MLD_LEAVING_MEMBER: 2193 /* Already leaving or left; do nothing. */ 2194 CTR1(KTR_MLD, 2195 "%s: not kicking state machine for silent group", __func__); 2196 break; 2197 case MLD_REPORTING_MEMBER: 2198 case MLD_IDLE_MEMBER: 2199 case MLD_G_QUERY_PENDING_MEMBER: 2200 case MLD_SG_QUERY_PENDING_MEMBER: 2201 if (mli->mli_version == MLD_VERSION_1) { 2202 #ifdef INVARIANTS 2203 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 2204 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) 2205 panic("%s: MLDv2 state reached, not MLDv2 mode", 2206 __func__); 2207 #endif 2208 mld_v1_transmit_report(inm, MLD_LISTENER_DONE); 2209 inm->in6m_state = MLD_NOT_MEMBER; 2210 V_current_state_timers_running6 = 1; 2211 } else if (mli->mli_version == MLD_VERSION_2) { 2212 /* 2213 * Stop group timer and all pending reports. 2214 * Immediately enqueue a state-change report 2215 * TO_IN {} to be sent on the next fast timeout, 2216 * giving us an opportunity to merge reports. 2217 */ 2218 mbufq_drain(&inm->in6m_scq); 2219 inm->in6m_timer = 0; 2220 inm->in6m_scrv = mli->mli_rv; 2221 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " 2222 "pending retransmissions.", __func__, 2223 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2224 if_name(inm->in6m_ifp), inm->in6m_scrv); 2225 if (inm->in6m_scrv == 0) { 2226 inm->in6m_state = MLD_NOT_MEMBER; 2227 inm->in6m_sctimer = 0; 2228 } else { 2229 int retval; 2230 2231 in6m_acquire_locked(inm); 2232 2233 retval = mld_v2_enqueue_group_record( 2234 &inm->in6m_scq, inm, 1, 0, 0, 2235 (mli->mli_flags & MLIF_USEALLOW)); 2236 KASSERT(retval != 0, 2237 ("%s: enqueue record = %d", __func__, 2238 retval)); 2239 2240 inm->in6m_state = MLD_LEAVING_MEMBER; 2241 inm->in6m_sctimer = 1; 2242 V_state_change_timers_running6 = 1; 2243 syncstates = 0; 2244 } 2245 break; 2246 } 2247 break; 2248 case MLD_LAZY_MEMBER: 2249 case MLD_SLEEPING_MEMBER: 2250 case MLD_AWAKENING_MEMBER: 2251 /* Our reports are suppressed; do nothing. */ 2252 break; 2253 } 2254 2255 if (syncstates) { 2256 in6m_commit(inm); 2257 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2258 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2259 if_name(inm->in6m_ifp)); 2260 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 2261 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", 2262 __func__, &inm->in6m_addr, if_name(inm->in6m_ifp)); 2263 } 2264 } 2265 2266 /* 2267 * Enqueue an MLDv2 group record to the given output queue. 2268 * 2269 * If is_state_change is zero, a current-state record is appended. 2270 * If is_state_change is non-zero, a state-change report is appended. 2271 * 2272 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2273 * If is_group_query is zero, and if there is a packet with free space 2274 * at the tail of the queue, it will be appended to providing there 2275 * is enough free space. 2276 * Otherwise a new mbuf packet chain is allocated. 2277 * 2278 * If is_source_query is non-zero, each source is checked to see if 2279 * it was recorded for a Group-Source query, and will be omitted if 2280 * it is not both in-mode and recorded. 2281 * 2282 * If use_block_allow is non-zero, state change reports for initial join 2283 * and final leave, on an inclusive mode group with a source list, will be 2284 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively. 2285 * 2286 * The function will attempt to allocate leading space in the packet 2287 * for the IPv6+ICMP headers to be prepended without fragmenting the chain. 2288 * 2289 * If successful the size of all data appended to the queue is returned, 2290 * otherwise an error code less than zero is returned, or zero if 2291 * no record(s) were appended. 2292 */ 2293 static int 2294 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm, 2295 const int is_state_change, const int is_group_query, 2296 const int is_source_query, const int use_block_allow) 2297 { 2298 struct mldv2_record mr; 2299 struct mldv2_record *pmr; 2300 struct ifnet *ifp; 2301 struct ip6_msource *ims, *nims; 2302 struct mbuf *m0, *m, *md; 2303 int is_filter_list_change; 2304 int minrec0len, m0srcs, msrcs, nbytes, off; 2305 int record_has_sources; 2306 int now; 2307 int type; 2308 uint8_t mode; 2309 #ifdef KTR 2310 char ip6tbuf[INET6_ADDRSTRLEN]; 2311 #endif 2312 2313 IN6_MULTI_LIST_LOCK_ASSERT(); 2314 2315 ifp = inm->in6m_ifp; 2316 is_filter_list_change = 0; 2317 m = NULL; 2318 m0 = NULL; 2319 m0srcs = 0; 2320 msrcs = 0; 2321 nbytes = 0; 2322 nims = NULL; 2323 record_has_sources = 1; 2324 pmr = NULL; 2325 type = MLD_DO_NOTHING; 2326 mode = inm->in6m_st[1].iss_fmode; 2327 2328 /* 2329 * If we did not transition out of ASM mode during t0->t1, 2330 * and there are no source nodes to process, we can skip 2331 * the generation of source records. 2332 */ 2333 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && 2334 inm->in6m_nsrc == 0) 2335 record_has_sources = 0; 2336 2337 if (is_state_change) { 2338 /* 2339 * Queue a state change record. 2340 * If the mode did not change, and there are non-ASM 2341 * listeners or source filters present, 2342 * we potentially need to issue two records for the group. 2343 * If there are ASM listeners, and there was no filter 2344 * mode transition of any kind, do nothing. 2345 * 2346 * If we are transitioning to MCAST_UNDEFINED, we need 2347 * not send any sources. A transition to/from this state is 2348 * considered inclusive with some special treatment. 2349 * 2350 * If we are rewriting initial joins/leaves to use 2351 * ALLOW/BLOCK, and the group's membership is inclusive, 2352 * we need to send sources in all cases. 2353 */ 2354 if (mode != inm->in6m_st[0].iss_fmode) { 2355 if (mode == MCAST_EXCLUDE) { 2356 CTR1(KTR_MLD, "%s: change to EXCLUDE", 2357 __func__); 2358 type = MLD_CHANGE_TO_EXCLUDE_MODE; 2359 } else { 2360 CTR1(KTR_MLD, "%s: change to INCLUDE", 2361 __func__); 2362 if (use_block_allow) { 2363 /* 2364 * XXX 2365 * Here we're interested in state 2366 * edges either direction between 2367 * MCAST_UNDEFINED and MCAST_INCLUDE. 2368 * Perhaps we should just check 2369 * the group state, rather than 2370 * the filter mode. 2371 */ 2372 if (mode == MCAST_UNDEFINED) { 2373 type = MLD_BLOCK_OLD_SOURCES; 2374 } else { 2375 type = MLD_ALLOW_NEW_SOURCES; 2376 } 2377 } else { 2378 type = MLD_CHANGE_TO_INCLUDE_MODE; 2379 if (mode == MCAST_UNDEFINED) 2380 record_has_sources = 0; 2381 } 2382 } 2383 } else { 2384 if (record_has_sources) { 2385 is_filter_list_change = 1; 2386 } else { 2387 type = MLD_DO_NOTHING; 2388 } 2389 } 2390 } else { 2391 /* 2392 * Queue a current state record. 2393 */ 2394 if (mode == MCAST_EXCLUDE) { 2395 type = MLD_MODE_IS_EXCLUDE; 2396 } else if (mode == MCAST_INCLUDE) { 2397 type = MLD_MODE_IS_INCLUDE; 2398 KASSERT(inm->in6m_st[1].iss_asm == 0, 2399 ("%s: inm %p is INCLUDE but ASM count is %d", 2400 __func__, inm, inm->in6m_st[1].iss_asm)); 2401 } 2402 } 2403 2404 /* 2405 * Generate the filter list changes using a separate function. 2406 */ 2407 if (is_filter_list_change) 2408 return (mld_v2_enqueue_filter_change(mq, inm)); 2409 2410 if (type == MLD_DO_NOTHING) { 2411 CTR3(KTR_MLD, "%s: nothing to do for %s/%s", 2412 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2413 if_name(inm->in6m_ifp)); 2414 return (0); 2415 } 2416 2417 /* 2418 * If any sources are present, we must be able to fit at least 2419 * one in the trailing space of the tail packet's mbuf, 2420 * ideally more. 2421 */ 2422 minrec0len = sizeof(struct mldv2_record); 2423 if (record_has_sources) 2424 minrec0len += sizeof(struct in6_addr); 2425 2426 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, 2427 mld_rec_type_to_str(type), 2428 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2429 if_name(inm->in6m_ifp)); 2430 2431 /* 2432 * Check if we have a packet in the tail of the queue for this 2433 * group into which the first group record for this group will fit. 2434 * Otherwise allocate a new packet. 2435 * Always allocate leading space for IP6+RA+ICMPV6+REPORT. 2436 * Note: Group records for G/GSR query responses MUST be sent 2437 * in their own packet. 2438 */ 2439 m0 = mbufq_last(mq); 2440 if (!is_group_query && 2441 m0 != NULL && 2442 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && 2443 (m0->m_pkthdr.len + minrec0len) < 2444 (ifp->if_mtu - MLD_MTUSPACE)) { 2445 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2446 sizeof(struct mldv2_record)) / 2447 sizeof(struct in6_addr); 2448 m = m0; 2449 CTR1(KTR_MLD, "%s: use existing packet", __func__); 2450 } else { 2451 if (mbufq_full(mq)) { 2452 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2453 return (-ENOMEM); 2454 } 2455 m = NULL; 2456 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2457 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2458 if (!is_state_change && !is_group_query) 2459 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2460 if (m == NULL) 2461 m = m_gethdr(M_NOWAIT, MT_DATA); 2462 if (m == NULL) 2463 return (-ENOMEM); 2464 2465 mld_save_context(m, ifp); 2466 2467 CTR1(KTR_MLD, "%s: allocated first packet", __func__); 2468 } 2469 2470 /* 2471 * Append group record. 2472 * If we have sources, we don't know how many yet. 2473 */ 2474 mr.mr_type = type; 2475 mr.mr_datalen = 0; 2476 mr.mr_numsrc = 0; 2477 mr.mr_addr = inm->in6m_addr; 2478 in6_clearscope(&mr.mr_addr); 2479 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2480 if (m != m0) 2481 m_freem(m); 2482 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2483 return (-ENOMEM); 2484 } 2485 nbytes += sizeof(struct mldv2_record); 2486 2487 /* 2488 * Append as many sources as will fit in the first packet. 2489 * If we are appending to a new packet, the chain allocation 2490 * may potentially use clusters; use m_getptr() in this case. 2491 * If we are appending to an existing packet, we need to obtain 2492 * a pointer to the group record after m_append(), in case a new 2493 * mbuf was allocated. 2494 * 2495 * Only append sources which are in-mode at t1. If we are 2496 * transitioning to MCAST_UNDEFINED state on the group, and 2497 * use_block_allow is zero, do not include source entries. 2498 * Otherwise, we need to include this source in the report. 2499 * 2500 * Only report recorded sources in our filter set when responding 2501 * to a group-source query. 2502 */ 2503 if (record_has_sources) { 2504 if (m == m0) { 2505 md = m_last(m); 2506 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2507 md->m_len - nbytes); 2508 } else { 2509 md = m_getptr(m, 0, &off); 2510 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2511 off); 2512 } 2513 msrcs = 0; 2514 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, 2515 nims) { 2516 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2517 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2518 now = im6s_get_mode(inm, ims, 1); 2519 CTR2(KTR_MLD, "%s: node is %d", __func__, now); 2520 if ((now != mode) || 2521 (now == mode && 2522 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2523 CTR1(KTR_MLD, "%s: skip node", __func__); 2524 continue; 2525 } 2526 if (is_source_query && ims->im6s_stp == 0) { 2527 CTR1(KTR_MLD, "%s: skip unrecorded node", 2528 __func__); 2529 continue; 2530 } 2531 CTR1(KTR_MLD, "%s: append node", __func__); 2532 if (!m_append(m, sizeof(struct in6_addr), 2533 (void *)&ims->im6s_addr)) { 2534 if (m != m0) 2535 m_freem(m); 2536 CTR1(KTR_MLD, "%s: m_append() failed.", 2537 __func__); 2538 return (-ENOMEM); 2539 } 2540 nbytes += sizeof(struct in6_addr); 2541 ++msrcs; 2542 if (msrcs == m0srcs) 2543 break; 2544 } 2545 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, 2546 msrcs); 2547 pmr->mr_numsrc = htons(msrcs); 2548 nbytes += (msrcs * sizeof(struct in6_addr)); 2549 } 2550 2551 if (is_source_query && msrcs == 0) { 2552 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); 2553 if (m != m0) 2554 m_freem(m); 2555 return (0); 2556 } 2557 2558 /* 2559 * We are good to go with first packet. 2560 */ 2561 if (m != m0) { 2562 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); 2563 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2564 mbufq_enqueue(mq, m); 2565 } else 2566 m->m_pkthdr.PH_vt.vt_nrecs++; 2567 2568 /* 2569 * No further work needed if no source list in packet(s). 2570 */ 2571 if (!record_has_sources) 2572 return (nbytes); 2573 2574 /* 2575 * Whilst sources remain to be announced, we need to allocate 2576 * a new packet and fill out as many sources as will fit. 2577 * Always try for a cluster first. 2578 */ 2579 while (nims != NULL) { 2580 if (mbufq_full(mq)) { 2581 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2582 return (-ENOMEM); 2583 } 2584 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2585 if (m == NULL) 2586 m = m_gethdr(M_NOWAIT, MT_DATA); 2587 if (m == NULL) 2588 return (-ENOMEM); 2589 mld_save_context(m, ifp); 2590 md = m_getptr(m, 0, &off); 2591 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); 2592 CTR1(KTR_MLD, "%s: allocated next packet", __func__); 2593 2594 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2595 if (m != m0) 2596 m_freem(m); 2597 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2598 return (-ENOMEM); 2599 } 2600 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2601 nbytes += sizeof(struct mldv2_record); 2602 2603 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2604 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2605 2606 msrcs = 0; 2607 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2608 CTR2(KTR_MLD, "%s: visit node %s", 2609 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2610 now = im6s_get_mode(inm, ims, 1); 2611 if ((now != mode) || 2612 (now == mode && 2613 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2614 CTR1(KTR_MLD, "%s: skip node", __func__); 2615 continue; 2616 } 2617 if (is_source_query && ims->im6s_stp == 0) { 2618 CTR1(KTR_MLD, "%s: skip unrecorded node", 2619 __func__); 2620 continue; 2621 } 2622 CTR1(KTR_MLD, "%s: append node", __func__); 2623 if (!m_append(m, sizeof(struct in6_addr), 2624 (void *)&ims->im6s_addr)) { 2625 if (m != m0) 2626 m_freem(m); 2627 CTR1(KTR_MLD, "%s: m_append() failed.", 2628 __func__); 2629 return (-ENOMEM); 2630 } 2631 ++msrcs; 2632 if (msrcs == m0srcs) 2633 break; 2634 } 2635 pmr->mr_numsrc = htons(msrcs); 2636 nbytes += (msrcs * sizeof(struct in6_addr)); 2637 2638 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); 2639 mbufq_enqueue(mq, m); 2640 } 2641 2642 return (nbytes); 2643 } 2644 2645 /* 2646 * Type used to mark record pass completion. 2647 * We exploit the fact we can cast to this easily from the 2648 * current filter modes on each ip_msource node. 2649 */ 2650 typedef enum { 2651 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2652 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2653 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2654 REC_FULL = REC_ALLOW | REC_BLOCK 2655 } rectype_t; 2656 2657 /* 2658 * Enqueue an MLDv2 filter list change to the given output queue. 2659 * 2660 * Source list filter state is held in an RB-tree. When the filter list 2661 * for a group is changed without changing its mode, we need to compute 2662 * the deltas between T0 and T1 for each source in the filter set, 2663 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2664 * 2665 * As we may potentially queue two record types, and the entire R-B tree 2666 * needs to be walked at once, we break this out into its own function 2667 * so we can generate a tightly packed queue of packets. 2668 * 2669 * XXX This could be written to only use one tree walk, although that makes 2670 * serializing into the mbuf chains a bit harder. For now we do two walks 2671 * which makes things easier on us, and it may or may not be harder on 2672 * the L2 cache. 2673 * 2674 * If successful the size of all data appended to the queue is returned, 2675 * otherwise an error code less than zero is returned, or zero if 2676 * no record(s) were appended. 2677 */ 2678 static int 2679 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm) 2680 { 2681 static const int MINRECLEN = 2682 sizeof(struct mldv2_record) + sizeof(struct in6_addr); 2683 struct ifnet *ifp; 2684 struct mldv2_record mr; 2685 struct mldv2_record *pmr; 2686 struct ip6_msource *ims, *nims; 2687 struct mbuf *m, *m0, *md; 2688 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 2689 int nallow, nblock; 2690 uint8_t mode, now, then; 2691 rectype_t crt, drt, nrt; 2692 #ifdef KTR 2693 char ip6tbuf[INET6_ADDRSTRLEN]; 2694 #endif 2695 2696 IN6_MULTI_LIST_LOCK_ASSERT(); 2697 2698 if (inm->in6m_nsrc == 0 || 2699 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) 2700 return (0); 2701 2702 ifp = inm->in6m_ifp; /* interface */ 2703 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ 2704 crt = REC_NONE; /* current group record type */ 2705 drt = REC_NONE; /* mask of completed group record types */ 2706 nrt = REC_NONE; /* record type for current node */ 2707 m0srcs = 0; /* # source which will fit in current mbuf chain */ 2708 npbytes = 0; /* # of bytes appended this packet */ 2709 nbytes = 0; /* # of bytes appended to group's state-change queue */ 2710 rsrcs = 0; /* # sources encoded in current record */ 2711 schanged = 0; /* # nodes encoded in overall filter change */ 2712 nallow = 0; /* # of source entries in ALLOW_NEW */ 2713 nblock = 0; /* # of source entries in BLOCK_OLD */ 2714 nims = NULL; /* next tree node pointer */ 2715 2716 /* 2717 * For each possible filter record mode. 2718 * The first kind of source we encounter tells us which 2719 * is the first kind of record we start appending. 2720 * If a node transitioned to UNDEFINED at t1, its mode is treated 2721 * as the inverse of the group's filter mode. 2722 */ 2723 while (drt != REC_FULL) { 2724 do { 2725 m0 = mbufq_last(mq); 2726 if (m0 != NULL && 2727 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 2728 MLD_V2_REPORT_MAXRECS) && 2729 (m0->m_pkthdr.len + MINRECLEN) < 2730 (ifp->if_mtu - MLD_MTUSPACE)) { 2731 m = m0; 2732 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2733 sizeof(struct mldv2_record)) / 2734 sizeof(struct in6_addr); 2735 CTR1(KTR_MLD, 2736 "%s: use previous packet", __func__); 2737 } else { 2738 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2739 if (m == NULL) 2740 m = m_gethdr(M_NOWAIT, MT_DATA); 2741 if (m == NULL) { 2742 CTR1(KTR_MLD, 2743 "%s: m_get*() failed", __func__); 2744 return (-ENOMEM); 2745 } 2746 m->m_pkthdr.PH_vt.vt_nrecs = 0; 2747 mld_save_context(m, ifp); 2748 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2749 sizeof(struct mldv2_record)) / 2750 sizeof(struct in6_addr); 2751 npbytes = 0; 2752 CTR1(KTR_MLD, 2753 "%s: allocated new packet", __func__); 2754 } 2755 /* 2756 * Append the MLD group record header to the 2757 * current packet's data area. 2758 * Recalculate pointer to free space for next 2759 * group record, in case m_append() allocated 2760 * a new mbuf or cluster. 2761 */ 2762 memset(&mr, 0, sizeof(mr)); 2763 mr.mr_addr = inm->in6m_addr; 2764 in6_clearscope(&mr.mr_addr); 2765 if (!m_append(m, sizeof(mr), (void *)&mr)) { 2766 if (m != m0) 2767 m_freem(m); 2768 CTR1(KTR_MLD, 2769 "%s: m_append() failed", __func__); 2770 return (-ENOMEM); 2771 } 2772 npbytes += sizeof(struct mldv2_record); 2773 if (m != m0) { 2774 /* new packet; offset in chain */ 2775 md = m_getptr(m, npbytes - 2776 sizeof(struct mldv2_record), &off); 2777 pmr = (struct mldv2_record *)(mtod(md, 2778 uint8_t *) + off); 2779 } else { 2780 /* current packet; offset from last append */ 2781 md = m_last(m); 2782 pmr = (struct mldv2_record *)(mtod(md, 2783 uint8_t *) + md->m_len - 2784 sizeof(struct mldv2_record)); 2785 } 2786 /* 2787 * Begin walking the tree for this record type 2788 * pass, or continue from where we left off 2789 * previously if we had to allocate a new packet. 2790 * Only report deltas in-mode at t1. 2791 * We need not report included sources as allowed 2792 * if we are in inclusive mode on the group, 2793 * however the converse is not true. 2794 */ 2795 rsrcs = 0; 2796 if (nims == NULL) { 2797 nims = RB_MIN(ip6_msource_tree, 2798 &inm->in6m_srcs); 2799 } 2800 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2801 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2802 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2803 now = im6s_get_mode(inm, ims, 1); 2804 then = im6s_get_mode(inm, ims, 0); 2805 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", 2806 __func__, then, now); 2807 if (now == then) { 2808 CTR1(KTR_MLD, 2809 "%s: skip unchanged", __func__); 2810 continue; 2811 } 2812 if (mode == MCAST_EXCLUDE && 2813 now == MCAST_INCLUDE) { 2814 CTR1(KTR_MLD, 2815 "%s: skip IN src on EX group", 2816 __func__); 2817 continue; 2818 } 2819 nrt = (rectype_t)now; 2820 if (nrt == REC_NONE) 2821 nrt = (rectype_t)(~mode & REC_FULL); 2822 if (schanged++ == 0) { 2823 crt = nrt; 2824 } else if (crt != nrt) 2825 continue; 2826 if (!m_append(m, sizeof(struct in6_addr), 2827 (void *)&ims->im6s_addr)) { 2828 if (m != m0) 2829 m_freem(m); 2830 CTR1(KTR_MLD, 2831 "%s: m_append() failed", __func__); 2832 return (-ENOMEM); 2833 } 2834 nallow += !!(crt == REC_ALLOW); 2835 nblock += !!(crt == REC_BLOCK); 2836 if (++rsrcs == m0srcs) 2837 break; 2838 } 2839 /* 2840 * If we did not append any tree nodes on this 2841 * pass, back out of allocations. 2842 */ 2843 if (rsrcs == 0) { 2844 npbytes -= sizeof(struct mldv2_record); 2845 if (m != m0) { 2846 CTR1(KTR_MLD, 2847 "%s: m_free(m)", __func__); 2848 m_freem(m); 2849 } else { 2850 CTR1(KTR_MLD, 2851 "%s: m_adj(m, -mr)", __func__); 2852 m_adj(m, -((int)sizeof( 2853 struct mldv2_record))); 2854 } 2855 continue; 2856 } 2857 npbytes += (rsrcs * sizeof(struct in6_addr)); 2858 if (crt == REC_ALLOW) 2859 pmr->mr_type = MLD_ALLOW_NEW_SOURCES; 2860 else if (crt == REC_BLOCK) 2861 pmr->mr_type = MLD_BLOCK_OLD_SOURCES; 2862 pmr->mr_numsrc = htons(rsrcs); 2863 /* 2864 * Count the new group record, and enqueue this 2865 * packet if it wasn't already queued. 2866 */ 2867 m->m_pkthdr.PH_vt.vt_nrecs++; 2868 if (m != m0) 2869 mbufq_enqueue(mq, m); 2870 nbytes += npbytes; 2871 } while (nims != NULL); 2872 drt |= crt; 2873 crt = (~crt & REC_FULL); 2874 } 2875 2876 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 2877 nallow, nblock); 2878 2879 return (nbytes); 2880 } 2881 2882 static int 2883 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq) 2884 { 2885 struct mbufq *gq; 2886 struct mbuf *m; /* pending state-change */ 2887 struct mbuf *m0; /* copy of pending state-change */ 2888 struct mbuf *mt; /* last state-change in packet */ 2889 int docopy, domerge; 2890 u_int recslen; 2891 2892 docopy = 0; 2893 domerge = 0; 2894 recslen = 0; 2895 2896 IN6_MULTI_LIST_LOCK_ASSERT(); 2897 MLD_LOCK_ASSERT(); 2898 2899 /* 2900 * If there are further pending retransmissions, make a writable 2901 * copy of each queued state-change message before merging. 2902 */ 2903 if (inm->in6m_scrv > 0) 2904 docopy = 1; 2905 2906 gq = &inm->in6m_scq; 2907 #ifdef KTR 2908 if (mbufq_first(gq) == NULL) { 2909 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", 2910 __func__, inm); 2911 } 2912 #endif 2913 2914 m = mbufq_first(gq); 2915 while (m != NULL) { 2916 /* 2917 * Only merge the report into the current packet if 2918 * there is sufficient space to do so; an MLDv2 report 2919 * packet may only contain 65,535 group records. 2920 * Always use a simple mbuf chain concatentation to do this, 2921 * as large state changes for single groups may have 2922 * allocated clusters. 2923 */ 2924 domerge = 0; 2925 mt = mbufq_last(scq); 2926 if (mt != NULL) { 2927 recslen = m_length(m, NULL); 2928 2929 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 2930 m->m_pkthdr.PH_vt.vt_nrecs <= 2931 MLD_V2_REPORT_MAXRECS) && 2932 (mt->m_pkthdr.len + recslen <= 2933 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) 2934 domerge = 1; 2935 } 2936 2937 if (!domerge && mbufq_full(gq)) { 2938 CTR2(KTR_MLD, 2939 "%s: outbound queue full, skipping whole packet %p", 2940 __func__, m); 2941 mt = m->m_nextpkt; 2942 if (!docopy) 2943 m_freem(m); 2944 m = mt; 2945 continue; 2946 } 2947 2948 if (!docopy) { 2949 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); 2950 m0 = mbufq_dequeue(gq); 2951 m = m0->m_nextpkt; 2952 } else { 2953 CTR2(KTR_MLD, "%s: copying %p", __func__, m); 2954 m0 = m_dup(m, M_NOWAIT); 2955 if (m0 == NULL) 2956 return (ENOMEM); 2957 m0->m_nextpkt = NULL; 2958 m = m->m_nextpkt; 2959 } 2960 2961 if (!domerge) { 2962 CTR3(KTR_MLD, "%s: queueing %p to scq %p)", 2963 __func__, m0, scq); 2964 mbufq_enqueue(scq, m0); 2965 } else { 2966 struct mbuf *mtl; /* last mbuf of packet mt */ 2967 2968 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", 2969 __func__, m0, mt); 2970 2971 mtl = m_last(mt); 2972 m0->m_flags &= ~M_PKTHDR; 2973 mt->m_pkthdr.len += recslen; 2974 mt->m_pkthdr.PH_vt.vt_nrecs += 2975 m0->m_pkthdr.PH_vt.vt_nrecs; 2976 2977 mtl->m_next = m0; 2978 } 2979 } 2980 2981 return (0); 2982 } 2983 2984 /* 2985 * Respond to a pending MLDv2 General Query. 2986 */ 2987 static void 2988 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli) 2989 { 2990 struct epoch_tracker et; 2991 struct ifmultiaddr *ifma; 2992 struct ifnet *ifp; 2993 struct in6_multi *inm; 2994 int retval; 2995 2996 IN6_MULTI_LIST_LOCK_ASSERT(); 2997 MLD_LOCK_ASSERT(); 2998 2999 KASSERT(mli->mli_version == MLD_VERSION_2, 3000 ("%s: called when version %d", __func__, mli->mli_version)); 3001 3002 /* 3003 * Check that there are some packets queued. If so, send them first. 3004 * For large number of groups the reply to general query can take 3005 * many packets, we should finish sending them before starting of 3006 * queuing the new reply. 3007 */ 3008 if (mbufq_len(&mli->mli_gq) != 0) 3009 goto send; 3010 3011 ifp = mli->mli_ifp; 3012 3013 NET_EPOCH_ENTER(et); 3014 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3015 if (ifma->ifma_addr->sa_family != AF_INET6 || 3016 ifma->ifma_protospec == NULL) 3017 continue; 3018 3019 inm = (struct in6_multi *)ifma->ifma_protospec; 3020 KASSERT(ifp == inm->in6m_ifp, 3021 ("%s: inconsistent ifp", __func__)); 3022 3023 switch (inm->in6m_state) { 3024 case MLD_NOT_MEMBER: 3025 case MLD_SILENT_MEMBER: 3026 break; 3027 case MLD_REPORTING_MEMBER: 3028 case MLD_IDLE_MEMBER: 3029 case MLD_LAZY_MEMBER: 3030 case MLD_SLEEPING_MEMBER: 3031 case MLD_AWAKENING_MEMBER: 3032 inm->in6m_state = MLD_REPORTING_MEMBER; 3033 retval = mld_v2_enqueue_group_record(&mli->mli_gq, 3034 inm, 0, 0, 0, 0); 3035 CTR2(KTR_MLD, "%s: enqueue record = %d", 3036 __func__, retval); 3037 break; 3038 case MLD_G_QUERY_PENDING_MEMBER: 3039 case MLD_SG_QUERY_PENDING_MEMBER: 3040 case MLD_LEAVING_MEMBER: 3041 break; 3042 } 3043 } 3044 NET_EPOCH_EXIT(et); 3045 3046 send: 3047 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); 3048 3049 /* 3050 * Slew transmission of bursts over 500ms intervals. 3051 */ 3052 if (mbufq_first(&mli->mli_gq) != NULL) { 3053 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( 3054 MLD_RESPONSE_BURST_INTERVAL); 3055 V_interface_timers_running6 = 1; 3056 } 3057 } 3058 3059 /* 3060 * Transmit the next pending message in the output queue. 3061 * 3062 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3063 * MRT: Nothing needs to be done, as MLD traffic is always local to 3064 * a link and uses a link-scope multicast address. 3065 */ 3066 static void 3067 mld_dispatch_packet(struct mbuf *m) 3068 { 3069 struct ip6_moptions im6o; 3070 struct ifnet *ifp; 3071 struct ifnet *oifp; 3072 struct mbuf *m0; 3073 struct mbuf *md; 3074 struct ip6_hdr *ip6; 3075 struct mld_hdr *mld; 3076 int error; 3077 int off; 3078 int type; 3079 uint32_t ifindex; 3080 3081 CTR2(KTR_MLD, "%s: transmit %p", __func__, m); 3082 3083 /* 3084 * Set VNET image pointer from enqueued mbuf chain 3085 * before doing anything else. Whilst we use interface 3086 * indexes to guard against interface detach, they are 3087 * unique to each VIMAGE and must be retrieved. 3088 */ 3089 ifindex = mld_restore_context(m); 3090 3091 /* 3092 * Check if the ifnet still exists. This limits the scope of 3093 * any race in the absence of a global ifp lock for low cost 3094 * (an array lookup). 3095 */ 3096 ifp = ifnet_byindex(ifindex); 3097 if (ifp == NULL) { 3098 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", 3099 __func__, m, ifindex); 3100 m_freem(m); 3101 IP6STAT_INC(ip6s_noroute); 3102 goto out; 3103 } 3104 3105 im6o.im6o_multicast_hlim = 1; 3106 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); 3107 im6o.im6o_multicast_ifp = ifp; 3108 3109 if (m->m_flags & M_MLDV1) { 3110 m0 = m; 3111 } else { 3112 m0 = mld_v2_encap_report(ifp, m); 3113 if (m0 == NULL) { 3114 CTR2(KTR_MLD, "%s: dropped %p", __func__, m); 3115 IP6STAT_INC(ip6s_odropped); 3116 goto out; 3117 } 3118 } 3119 3120 mld_scrub_context(m0); 3121 m_clrprotoflags(m); 3122 m0->m_pkthdr.rcvif = V_loif; 3123 3124 ip6 = mtod(m0, struct ip6_hdr *); 3125 #if 0 3126 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */ 3127 #else 3128 /* 3129 * XXX XXX Break some KPI rules to prevent an LOR which would 3130 * occur if we called in6_setscope() at transmission. 3131 * See comments at top of file. 3132 */ 3133 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index); 3134 #endif 3135 3136 /* 3137 * Retrieve the ICMPv6 type before handoff to ip6_output(), 3138 * so we can bump the stats. 3139 */ 3140 md = m_getptr(m0, sizeof(struct ip6_hdr), &off); 3141 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off); 3142 type = mld->mld_type; 3143 3144 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o, 3145 &oifp, NULL); 3146 if (error) { 3147 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error); 3148 goto out; 3149 } 3150 ICMP6STAT_INC(icp6s_outhist[type]); 3151 if (oifp != NULL) { 3152 icmp6_ifstat_inc(oifp, ifs6_out_msg); 3153 switch (type) { 3154 case MLD_LISTENER_REPORT: 3155 case MLDV2_LISTENER_REPORT: 3156 icmp6_ifstat_inc(oifp, ifs6_out_mldreport); 3157 break; 3158 case MLD_LISTENER_DONE: 3159 icmp6_ifstat_inc(oifp, ifs6_out_mlddone); 3160 break; 3161 } 3162 } 3163 out: 3164 return; 3165 } 3166 3167 /* 3168 * Encapsulate an MLDv2 report. 3169 * 3170 * KAME IPv6 requires that hop-by-hop options be passed separately, 3171 * and that the IPv6 header be prepended in a separate mbuf. 3172 * 3173 * Returns a pointer to the new mbuf chain head, or NULL if the 3174 * allocation failed. 3175 */ 3176 static struct mbuf * 3177 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m) 3178 { 3179 struct mbuf *mh; 3180 struct mldv2_report *mld; 3181 struct ip6_hdr *ip6; 3182 struct in6_ifaddr *ia; 3183 int mldreclen; 3184 3185 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 3186 KASSERT((m->m_flags & M_PKTHDR), 3187 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3188 3189 /* 3190 * RFC3590: OK to send as :: or tentative during DAD. 3191 */ 3192 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 3193 if (ia == NULL) 3194 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__); 3195 3196 mh = m_gethdr(M_NOWAIT, MT_DATA); 3197 if (mh == NULL) { 3198 if (ia != NULL) 3199 ifa_free(&ia->ia_ifa); 3200 m_freem(m); 3201 return (NULL); 3202 } 3203 M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); 3204 3205 mldreclen = m_length(m, NULL); 3206 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); 3207 3208 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report); 3209 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + 3210 sizeof(struct mldv2_report) + mldreclen; 3211 3212 ip6 = mtod(mh, struct ip6_hdr *); 3213 ip6->ip6_flow = 0; 3214 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3215 ip6->ip6_vfc |= IPV6_VERSION; 3216 ip6->ip6_nxt = IPPROTO_ICMPV6; 3217 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 3218 if (ia != NULL) 3219 ifa_free(&ia->ia_ifa); 3220 ip6->ip6_dst = in6addr_linklocal_allv2routers; 3221 /* scope ID will be set in netisr */ 3222 3223 mld = (struct mldv2_report *)(ip6 + 1); 3224 mld->mld_type = MLDV2_LISTENER_REPORT; 3225 mld->mld_code = 0; 3226 mld->mld_cksum = 0; 3227 mld->mld_v2_reserved = 0; 3228 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs); 3229 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3230 3231 mh->m_next = m; 3232 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 3233 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen); 3234 return (mh); 3235 } 3236 3237 #ifdef KTR 3238 static char * 3239 mld_rec_type_to_str(const int type) 3240 { 3241 3242 switch (type) { 3243 case MLD_CHANGE_TO_EXCLUDE_MODE: 3244 return "TO_EX"; 3245 break; 3246 case MLD_CHANGE_TO_INCLUDE_MODE: 3247 return "TO_IN"; 3248 break; 3249 case MLD_MODE_IS_EXCLUDE: 3250 return "MODE_EX"; 3251 break; 3252 case MLD_MODE_IS_INCLUDE: 3253 return "MODE_IN"; 3254 break; 3255 case MLD_ALLOW_NEW_SOURCES: 3256 return "ALLOW_NEW"; 3257 break; 3258 case MLD_BLOCK_OLD_SOURCES: 3259 return "BLOCK_OLD"; 3260 break; 3261 default: 3262 break; 3263 } 3264 return "unknown"; 3265 } 3266 #endif 3267 3268 static void 3269 mld_init(void *unused __unused) 3270 { 3271 3272 CTR1(KTR_MLD, "%s: initializing", __func__); 3273 MLD_LOCK_INIT(); 3274 3275 ip6_initpktopts(&mld_po); 3276 mld_po.ip6po_hlim = 1; 3277 mld_po.ip6po_hbh = &mld_ra.hbh; 3278 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER; 3279 mld_po.ip6po_flags = IP6PO_DONTFRAG; 3280 } 3281 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL); 3282 3283 static void 3284 mld_uninit(void *unused __unused) 3285 { 3286 3287 CTR1(KTR_MLD, "%s: tearing down", __func__); 3288 MLD_LOCK_DESTROY(); 3289 } 3290 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL); 3291 3292 static void 3293 vnet_mld_init(const void *unused __unused) 3294 { 3295 3296 CTR1(KTR_MLD, "%s: initializing", __func__); 3297 3298 LIST_INIT(&V_mli_head); 3299 } 3300 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init, 3301 NULL); 3302 3303 static void 3304 vnet_mld_uninit(const void *unused __unused) 3305 { 3306 3307 /* This can happen if we shutdown the network stack. */ 3308 CTR1(KTR_MLD, "%s: tearing down", __func__); 3309 } 3310 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit, 3311 NULL); 3312 3313 static int 3314 mld_modevent(module_t mod, int type, void *unused __unused) 3315 { 3316 3317 switch (type) { 3318 case MOD_LOAD: 3319 case MOD_UNLOAD: 3320 break; 3321 default: 3322 return (EOPNOTSUPP); 3323 } 3324 return (0); 3325 } 3326 3327 static moduledata_t mld_mod = { 3328 "mld", 3329 mld_modevent, 3330 0 3331 }; 3332 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY); 3333