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(struct in6_multi_head *inmh); 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 *, struct mldv2_query *, 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 *, 150 struct mldv2_query *, const int); 151 static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); 152 static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS); 153 154 /* 155 * Normative references: RFC 2710, RFC 3590, RFC 3810. 156 * 157 * Locking: 158 * * The MLD subsystem lock ends up being system-wide for the moment, 159 * but could be per-VIMAGE later on. 160 * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. 161 * Any may be taken independently; if any are held at the same 162 * time, the above lock order must be followed. 163 * * IN6_MULTI_LOCK covers in_multi. 164 * * MLD_LOCK covers per-link state and any global variables in this file. 165 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 166 * per-link state iterators. 167 * 168 * XXX LOR PREVENTION 169 * A special case for IPv6 is the in6_setscope() routine. ip6_output() 170 * will not accept an ifp; it wants an embedded scope ID, unlike 171 * ip_output(), which happily takes the ifp given to it. The embedded 172 * scope ID is only used by MLD to select the outgoing interface. 173 * 174 * During interface attach and detach, MLD will take MLD_LOCK *after* 175 * the IF_AFDATA_LOCK. 176 * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call 177 * it with MLD_LOCK held without triggering an LOR. A netisr with indirect 178 * dispatch could work around this, but we'd rather not do that, as it 179 * can introduce other races. 180 * 181 * As such, we exploit the fact that the scope ID is just the interface 182 * index, and embed it in the IPv6 destination address accordingly. 183 * This is potentially NOT VALID for MLDv1 reports, as they 184 * are always sent to the multicast group itself; as MLDv2 185 * reports are always sent to ff02::16, this is not an issue 186 * when MLDv2 is in use. 187 * 188 * This does not however eliminate the LOR when ip6_output() itself 189 * calls in6_setscope() internally whilst MLD_LOCK is held. This will 190 * trigger a LOR warning in WITNESS when the ifnet is detached. 191 * 192 * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given 193 * how it's used across the network stack. Here we're simply exploiting 194 * the fact that MLD runs at a similar layer in the stack to scope6.c. 195 * 196 * VIMAGE: 197 * * Each in6_multi corresponds to an ifp, and each ifp corresponds 198 * to a vnet in ifp->if_vnet. 199 */ 200 static struct mtx mld_mtx; 201 static MALLOC_DEFINE(M_MLD, "mld", "mld state"); 202 203 #define MLD_EMBEDSCOPE(pin6, zoneid) \ 204 if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \ 205 IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \ 206 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \ 207 208 /* 209 * VIMAGE-wide globals. 210 */ 211 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0}; 212 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head); 213 VNET_DEFINE_STATIC(int, interface_timers_running6); 214 VNET_DEFINE_STATIC(int, state_change_timers_running6); 215 VNET_DEFINE_STATIC(int, current_state_timers_running6); 216 217 #define V_mld_gsrdelay VNET(mld_gsrdelay) 218 #define V_mli_head VNET(mli_head) 219 #define V_interface_timers_running6 VNET(interface_timers_running6) 220 #define V_state_change_timers_running6 VNET(state_change_timers_running6) 221 #define V_current_state_timers_running6 VNET(current_state_timers_running6) 222 223 SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ 224 225 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, 226 "IPv6 Multicast Listener Discovery"); 227 228 /* 229 * Virtualized sysctls. 230 */ 231 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay, 232 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 233 &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I", 234 "Rate limit for MLDv2 Group-and-Source queries in seconds"); 235 236 /* 237 * Non-virtualized sysctls. 238 */ 239 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, 240 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo, 241 "Per-interface MLDv2 state"); 242 243 static int mld_v1enable = 1; 244 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN, 245 &mld_v1enable, 0, "Enable fallback to MLDv1"); 246 247 static int mld_v2enable = 1; 248 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN, 249 &mld_v2enable, 0, "Enable MLDv2"); 250 251 static int mld_use_allow = 1; 252 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN, 253 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves"); 254 255 /* 256 * Packed Router Alert option structure declaration. 257 */ 258 struct mld_raopt { 259 struct ip6_hbh hbh; 260 struct ip6_opt pad; 261 struct ip6_opt_router ra; 262 } __packed; 263 264 /* 265 * Router Alert hop-by-hop option header. 266 */ 267 static struct mld_raopt mld_ra = { 268 .hbh = { 0, 0 }, 269 .pad = { .ip6o_type = IP6OPT_PADN, 0 }, 270 .ra = { 271 .ip6or_type = IP6OPT_ROUTER_ALERT, 272 .ip6or_len = IP6OPT_RTALERT_LEN - 2, 273 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), 274 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) 275 } 276 }; 277 static struct ip6_pktopts mld_po; 278 279 static __inline void 280 mld_save_context(struct mbuf *m, struct ifnet *ifp) 281 { 282 283 #ifdef VIMAGE 284 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; 285 #endif /* VIMAGE */ 286 m->m_pkthdr.flowid = ifp->if_index; 287 } 288 289 static __inline void 290 mld_scrub_context(struct mbuf *m) 291 { 292 293 m->m_pkthdr.PH_loc.ptr = NULL; 294 m->m_pkthdr.flowid = 0; 295 } 296 297 /* 298 * Restore context from a queued output chain. 299 * Return saved ifindex. 300 * 301 * VIMAGE: The assertion is there to make sure that we 302 * actually called CURVNET_SET() with what's in the mbuf chain. 303 */ 304 static __inline uint32_t 305 mld_restore_context(struct mbuf *m) 306 { 307 308 #if defined(VIMAGE) && defined(INVARIANTS) 309 KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr, 310 ("%s: called when curvnet was not restored: cuvnet %p m ptr %p", 311 __func__, curvnet, m->m_pkthdr.PH_loc.ptr)); 312 #endif 313 return (m->m_pkthdr.flowid); 314 } 315 316 /* 317 * Retrieve or set threshold between group-source queries in seconds. 318 * 319 * VIMAGE: Assume curvnet set by caller. 320 * SMPng: NOTE: Serialized by MLD lock. 321 */ 322 static int 323 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS) 324 { 325 int error; 326 int i; 327 328 error = sysctl_wire_old_buffer(req, sizeof(int)); 329 if (error) 330 return (error); 331 332 MLD_LOCK(); 333 334 i = V_mld_gsrdelay.tv_sec; 335 336 error = sysctl_handle_int(oidp, &i, 0, req); 337 if (error || !req->newptr) 338 goto out_locked; 339 340 if (i < -1 || i >= 60) { 341 error = EINVAL; 342 goto out_locked; 343 } 344 345 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", 346 V_mld_gsrdelay.tv_sec, i); 347 V_mld_gsrdelay.tv_sec = i; 348 349 out_locked: 350 MLD_UNLOCK(); 351 return (error); 352 } 353 354 /* 355 * Expose struct mld_ifsoftc to userland, keyed by ifindex. 356 * For use by ifmcstat(8). 357 * 358 * SMPng: NOTE: Does an unlocked ifindex space read. 359 * VIMAGE: Assume curvnet set by caller. The node handler itself 360 * is not directly virtualized. 361 */ 362 static int 363 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) 364 { 365 int *name; 366 int error; 367 u_int namelen; 368 struct ifnet *ifp; 369 struct mld_ifsoftc *mli; 370 371 name = (int *)arg1; 372 namelen = arg2; 373 374 if (req->newptr != NULL) 375 return (EPERM); 376 377 if (namelen != 1) 378 return (EINVAL); 379 380 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo)); 381 if (error) 382 return (error); 383 384 IN6_MULTI_LOCK(); 385 IN6_MULTI_LIST_LOCK(); 386 MLD_LOCK(); 387 388 if (name[0] <= 0 || name[0] > V_if_index) { 389 error = ENOENT; 390 goto out_locked; 391 } 392 393 error = ENOENT; 394 395 ifp = ifnet_byindex(name[0]); 396 if (ifp == NULL) 397 goto out_locked; 398 399 LIST_FOREACH(mli, &V_mli_head, mli_link) { 400 if (ifp == mli->mli_ifp) { 401 struct mld_ifinfo info; 402 403 info.mli_version = mli->mli_version; 404 info.mli_v1_timer = mli->mli_v1_timer; 405 info.mli_v2_timer = mli->mli_v2_timer; 406 info.mli_flags = mli->mli_flags; 407 info.mli_rv = mli->mli_rv; 408 info.mli_qi = mli->mli_qi; 409 info.mli_qri = mli->mli_qri; 410 info.mli_uri = mli->mli_uri; 411 error = SYSCTL_OUT(req, &info, sizeof(info)); 412 break; 413 } 414 } 415 416 out_locked: 417 MLD_UNLOCK(); 418 IN6_MULTI_LIST_UNLOCK(); 419 IN6_MULTI_UNLOCK(); 420 return (error); 421 } 422 423 /* 424 * Dispatch an entire queue of pending packet chains. 425 * VIMAGE: Assumes the vnet pointer has been set. 426 */ 427 static void 428 mld_dispatch_queue(struct mbufq *mq, int limit) 429 { 430 struct mbuf *m; 431 432 while ((m = mbufq_dequeue(mq)) != NULL) { 433 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m); 434 mld_dispatch_packet(m); 435 if (--limit == 0) 436 break; 437 } 438 } 439 440 /* 441 * Filter outgoing MLD report state by group. 442 * 443 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1) 444 * and node-local addresses. However, kernel and socket consumers 445 * always embed the KAME scope ID in the address provided, so strip it 446 * when performing comparison. 447 * Note: This is not the same as the *multicast* scope. 448 * 449 * Return zero if the given group is one for which MLD reports 450 * should be suppressed, or non-zero if reports should be issued. 451 */ 452 static __inline int 453 mld_is_addr_reported(const struct in6_addr *addr) 454 { 455 456 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__)); 457 458 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) 459 return (0); 460 461 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) { 462 struct in6_addr tmp = *addr; 463 in6_clearscope(&tmp); 464 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) 465 return (0); 466 } 467 468 return (1); 469 } 470 471 /* 472 * Attach MLD when PF_INET6 is attached to an interface. 473 * 474 * SMPng: Normally called with IF_AFDATA_LOCK held. 475 */ 476 struct mld_ifsoftc * 477 mld_domifattach(struct ifnet *ifp) 478 { 479 struct mld_ifsoftc *mli; 480 481 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 482 __func__, ifp, if_name(ifp)); 483 484 MLD_LOCK(); 485 486 mli = mli_alloc_locked(ifp); 487 if (!(ifp->if_flags & IFF_MULTICAST)) 488 mli->mli_flags |= MLIF_SILENT; 489 if (mld_use_allow) 490 mli->mli_flags |= MLIF_USEALLOW; 491 492 MLD_UNLOCK(); 493 494 return (mli); 495 } 496 497 /* 498 * VIMAGE: assume curvnet set by caller. 499 */ 500 static struct mld_ifsoftc * 501 mli_alloc_locked(/*const*/ struct ifnet *ifp) 502 { 503 struct mld_ifsoftc *mli; 504 505 MLD_LOCK_ASSERT(); 506 507 mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO); 508 if (mli == NULL) 509 goto out; 510 511 mli->mli_ifp = ifp; 512 mli->mli_version = MLD_VERSION_2; 513 mli->mli_flags = 0; 514 mli->mli_rv = MLD_RV_INIT; 515 mli->mli_qi = MLD_QI_INIT; 516 mli->mli_qri = MLD_QRI_INIT; 517 mli->mli_uri = MLD_URI_INIT; 518 mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); 519 520 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); 521 522 CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)", 523 ifp, if_name(ifp)); 524 525 out: 526 return (mli); 527 } 528 529 /* 530 * Hook for ifdetach. 531 * 532 * NOTE: Some finalization tasks need to run before the protocol domain 533 * is detached, but also before the link layer does its cleanup. 534 * Run before link-layer cleanup; cleanup groups, but do not free MLD state. 535 * 536 * SMPng: Caller must hold IN6_MULTI_LOCK(). 537 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. 538 * XXX This routine is also bitten by unlocked ifma_protospec access. 539 */ 540 void 541 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh) 542 { 543 struct epoch_tracker et; 544 struct mld_ifsoftc *mli; 545 struct ifmultiaddr *ifma; 546 struct in6_multi *inm; 547 548 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, 549 if_name(ifp)); 550 551 IN6_MULTI_LIST_LOCK_ASSERT(); 552 MLD_LOCK(); 553 554 mli = MLD_IFINFO(ifp); 555 IF_ADDR_WLOCK(ifp); 556 /* 557 * Extract list of in6_multi associated with the detaching ifp 558 * which the PF_INET6 layer is about to release. 559 */ 560 NET_EPOCH_ENTER(et); 561 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 562 inm = in6m_ifmultiaddr_get_inm(ifma); 563 if (inm == NULL) 564 continue; 565 in6m_disconnect_locked(inmh, inm); 566 567 if (mli->mli_version == MLD_VERSION_2) { 568 in6m_clear_recorded(inm); 569 570 /* 571 * We need to release the final reference held 572 * for issuing the INCLUDE {}. 573 */ 574 if (inm->in6m_state == MLD_LEAVING_MEMBER) { 575 inm->in6m_state = MLD_NOT_MEMBER; 576 in6m_rele_locked(inmh, inm); 577 } 578 } 579 } 580 NET_EPOCH_EXIT(et); 581 IF_ADDR_WUNLOCK(ifp); 582 MLD_UNLOCK(); 583 } 584 585 /* 586 * Hook for domifdetach. 587 * Runs after link-layer cleanup; free MLD state. 588 * 589 * SMPng: Normally called with IF_AFDATA_LOCK held. 590 */ 591 void 592 mld_domifdetach(struct ifnet *ifp) 593 { 594 595 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 596 __func__, ifp, if_name(ifp)); 597 598 MLD_LOCK(); 599 mli_delete_locked(ifp); 600 MLD_UNLOCK(); 601 } 602 603 static void 604 mli_delete_locked(const struct ifnet *ifp) 605 { 606 struct mld_ifsoftc *mli, *tmli; 607 608 CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)", 609 __func__, ifp, if_name(ifp)); 610 611 MLD_LOCK_ASSERT(); 612 613 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) { 614 if (mli->mli_ifp == ifp) { 615 /* 616 * Free deferred General Query responses. 617 */ 618 mbufq_drain(&mli->mli_gq); 619 620 LIST_REMOVE(mli, mli_link); 621 622 free(mli, M_MLD); 623 return; 624 } 625 } 626 } 627 628 /* 629 * Process a received MLDv1 general or address-specific query. 630 * Assumes that the query header has been pulled up to sizeof(mld_hdr). 631 * 632 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 633 * mld_addr. This is OK as we own the mbuf chain. 634 */ 635 static int 636 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 637 /*const*/ struct mld_hdr *mld) 638 { 639 struct epoch_tracker et; 640 struct ifmultiaddr *ifma; 641 struct mld_ifsoftc *mli; 642 struct in6_multi *inm; 643 int is_general_query; 644 uint16_t timer; 645 #ifdef KTR 646 char ip6tbuf[INET6_ADDRSTRLEN]; 647 #endif 648 649 is_general_query = 0; 650 651 if (!mld_v1enable) { 652 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)", 653 ip6_sprintf(ip6tbuf, &mld->mld_addr), 654 ifp, if_name(ifp)); 655 return (0); 656 } 657 658 /* 659 * RFC3810 Section 6.2: MLD queries must originate from 660 * a router's link-local address. 661 */ 662 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 663 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 664 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 665 ifp, if_name(ifp)); 666 return (0); 667 } 668 669 /* 670 * Do address field validation upfront before we accept 671 * the query. 672 */ 673 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 674 /* 675 * MLDv1 General Query. 676 * If this was not sent to the all-nodes group, ignore it. 677 */ 678 struct in6_addr dst; 679 680 dst = ip6->ip6_dst; 681 in6_clearscope(&dst); 682 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) 683 return (EINVAL); 684 is_general_query = 1; 685 } else { 686 /* 687 * Embed scope ID of receiving interface in MLD query for 688 * lookup whilst we don't hold other locks. 689 */ 690 in6_setscope(&mld->mld_addr, ifp, NULL); 691 } 692 693 IN6_MULTI_LIST_LOCK(); 694 MLD_LOCK(); 695 696 /* 697 * Switch to MLDv1 host compatibility mode. 698 */ 699 mli = MLD_IFINFO(ifp); 700 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); 701 mld_set_version(mli, MLD_VERSION_1); 702 703 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; 704 if (timer == 0) 705 timer = 1; 706 707 NET_EPOCH_ENTER(et); 708 if (is_general_query) { 709 /* 710 * For each reporting group joined on this 711 * interface, kick the report timer. 712 */ 713 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)", 714 ifp, if_name(ifp)); 715 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 716 inm = in6m_ifmultiaddr_get_inm(ifma); 717 if (inm == NULL) 718 continue; 719 mld_v1_update_group(inm, timer); 720 } 721 } else { 722 /* 723 * MLDv1 Group-Specific Query. 724 * If this is a group-specific MLDv1 query, we need only 725 * look up the single group to process it. 726 */ 727 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 728 if (inm != NULL) { 729 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)", 730 ip6_sprintf(ip6tbuf, &mld->mld_addr), 731 ifp, if_name(ifp)); 732 mld_v1_update_group(inm, timer); 733 } 734 /* XXX Clear embedded scope ID as userland won't expect it. */ 735 in6_clearscope(&mld->mld_addr); 736 } 737 738 NET_EPOCH_EXIT(et); 739 MLD_UNLOCK(); 740 IN6_MULTI_LIST_UNLOCK(); 741 742 return (0); 743 } 744 745 /* 746 * Update the report timer on a group in response to an MLDv1 query. 747 * 748 * If we are becoming the reporting member for this group, start the timer. 749 * If we already are the reporting member for this group, and timer is 750 * below the threshold, reset it. 751 * 752 * We may be updating the group for the first time since we switched 753 * to MLDv2. If we are, then we must clear any recorded source lists, 754 * and transition to REPORTING state; the group timer is overloaded 755 * for group and group-source query responses. 756 * 757 * Unlike MLDv2, the delay per group should be jittered 758 * to avoid bursts of MLDv1 reports. 759 */ 760 static void 761 mld_v1_update_group(struct in6_multi *inm, const int timer) 762 { 763 #ifdef KTR 764 char ip6tbuf[INET6_ADDRSTRLEN]; 765 #endif 766 767 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, 768 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 769 if_name(inm->in6m_ifp), timer); 770 771 IN6_MULTI_LIST_LOCK_ASSERT(); 772 773 switch (inm->in6m_state) { 774 case MLD_NOT_MEMBER: 775 case MLD_SILENT_MEMBER: 776 break; 777 case MLD_REPORTING_MEMBER: 778 if (inm->in6m_timer != 0 && 779 inm->in6m_timer <= timer) { 780 CTR1(KTR_MLD, "%s: REPORTING and timer running, " 781 "skipping.", __func__); 782 break; 783 } 784 /* FALLTHROUGH */ 785 case MLD_SG_QUERY_PENDING_MEMBER: 786 case MLD_G_QUERY_PENDING_MEMBER: 787 case MLD_IDLE_MEMBER: 788 case MLD_LAZY_MEMBER: 789 case MLD_AWAKENING_MEMBER: 790 CTR1(KTR_MLD, "%s: ->REPORTING", __func__); 791 inm->in6m_state = MLD_REPORTING_MEMBER; 792 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 793 V_current_state_timers_running6 = 1; 794 break; 795 case MLD_SLEEPING_MEMBER: 796 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); 797 inm->in6m_state = MLD_AWAKENING_MEMBER; 798 break; 799 case MLD_LEAVING_MEMBER: 800 break; 801 } 802 } 803 804 /* 805 * Process a received MLDv2 general, group-specific or 806 * group-and-source-specific query. 807 * 808 * Assumes that mld points to a struct mldv2_query which is stored in 809 * contiguous memory. 810 * 811 * Return 0 if successful, otherwise an appropriate error code is returned. 812 */ 813 static int 814 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 815 struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len) 816 { 817 struct mld_ifsoftc *mli; 818 struct in6_multi *inm; 819 uint32_t maxdelay, nsrc, qqi; 820 int is_general_query; 821 uint16_t timer; 822 uint8_t qrv; 823 #ifdef KTR 824 char ip6tbuf[INET6_ADDRSTRLEN]; 825 #endif 826 827 if (!mld_v2enable) { 828 CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)", 829 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 830 ifp, if_name(ifp)); 831 return (0); 832 } 833 834 /* 835 * RFC3810 Section 6.2: MLD queries must originate from 836 * a router's link-local address. 837 */ 838 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 839 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 840 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 841 ifp, if_name(ifp)); 842 return (0); 843 } 844 845 is_general_query = 0; 846 847 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp)); 848 849 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ 850 if (maxdelay >= 32768) { 851 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << 852 (MLD_MRC_EXP(maxdelay) + 3); 853 } 854 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; 855 if (timer == 0) 856 timer = 1; 857 858 qrv = MLD_QRV(mld->mld_misc); 859 if (qrv < 2) { 860 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, 861 qrv, MLD_RV_INIT); 862 qrv = MLD_RV_INIT; 863 } 864 865 qqi = mld->mld_qqi; 866 if (qqi >= 128) { 867 qqi = MLD_QQIC_MANT(mld->mld_qqi) << 868 (MLD_QQIC_EXP(mld->mld_qqi) + 3); 869 } 870 871 nsrc = ntohs(mld->mld_numsrc); 872 if (nsrc > MLD_MAX_GS_SOURCES) 873 return (EMSGSIZE); 874 if (icmp6len < sizeof(struct mldv2_query) + 875 (nsrc * sizeof(struct in6_addr))) 876 return (EMSGSIZE); 877 878 /* 879 * Do further input validation upfront to avoid resetting timers 880 * should we need to discard this query. 881 */ 882 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 883 /* 884 * A general query with a source list has undefined 885 * behaviour; discard it. 886 */ 887 if (nsrc > 0) 888 return (EINVAL); 889 is_general_query = 1; 890 } else { 891 /* 892 * Embed scope ID of receiving interface in MLD query for 893 * lookup whilst we don't hold other locks (due to KAME 894 * locking lameness). We own this mbuf chain just now. 895 */ 896 in6_setscope(&mld->mld_addr, ifp, NULL); 897 } 898 899 IN6_MULTI_LIST_LOCK(); 900 MLD_LOCK(); 901 902 mli = MLD_IFINFO(ifp); 903 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); 904 905 /* 906 * Discard the v2 query if we're in Compatibility Mode. 907 * The RFC is pretty clear that hosts need to stay in MLDv1 mode 908 * until the Old Version Querier Present timer expires. 909 */ 910 if (mli->mli_version != MLD_VERSION_2) 911 goto out_locked; 912 913 mld_set_version(mli, MLD_VERSION_2); 914 mli->mli_rv = qrv; 915 mli->mli_qi = qqi; 916 mli->mli_qri = maxdelay; 917 918 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, 919 maxdelay); 920 921 if (is_general_query) { 922 /* 923 * MLDv2 General Query. 924 * 925 * Schedule a current-state report on this ifp for 926 * all groups, possibly containing source lists. 927 * 928 * If there is a pending General Query response 929 * scheduled earlier than the selected delay, do 930 * not schedule any other reports. 931 * Otherwise, reset the interface timer. 932 */ 933 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", 934 ifp, if_name(ifp)); 935 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { 936 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); 937 V_interface_timers_running6 = 1; 938 } 939 } else { 940 struct epoch_tracker et; 941 942 /* 943 * MLDv2 Group-specific or Group-and-source-specific Query. 944 * 945 * Group-source-specific queries are throttled on 946 * a per-group basis to defeat denial-of-service attempts. 947 * Queries for groups we are not a member of on this 948 * link are simply ignored. 949 */ 950 NET_EPOCH_ENTER(et); 951 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 952 if (inm == NULL) { 953 NET_EPOCH_EXIT(et); 954 goto out_locked; 955 } 956 if (nsrc > 0) { 957 if (!ratecheck(&inm->in6m_lastgsrtv, 958 &V_mld_gsrdelay)) { 959 CTR1(KTR_MLD, "%s: GS query throttled.", 960 __func__); 961 NET_EPOCH_EXIT(et); 962 goto out_locked; 963 } 964 } 965 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", 966 ifp, if_name(ifp)); 967 /* 968 * If there is a pending General Query response 969 * scheduled sooner than the selected delay, no 970 * further report need be scheduled. 971 * Otherwise, prepare to respond to the 972 * group-specific or group-and-source query. 973 */ 974 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) 975 mld_v2_process_group_query(inm, mli, timer, m, mld, off); 976 977 /* XXX Clear embedded scope ID as userland won't expect it. */ 978 in6_clearscope(&mld->mld_addr); 979 NET_EPOCH_EXIT(et); 980 } 981 982 out_locked: 983 MLD_UNLOCK(); 984 IN6_MULTI_LIST_UNLOCK(); 985 986 return (0); 987 } 988 989 /* 990 * Process a received MLDv2 group-specific or group-and-source-specific 991 * query. 992 * Return <0 if any error occurred. Currently this is ignored. 993 */ 994 static int 995 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli, 996 int timer, struct mbuf *m0, struct mldv2_query *mld, const int off) 997 { 998 int retval; 999 uint16_t nsrc; 1000 1001 IN6_MULTI_LIST_LOCK_ASSERT(); 1002 MLD_LOCK_ASSERT(); 1003 1004 retval = 0; 1005 1006 switch (inm->in6m_state) { 1007 case MLD_NOT_MEMBER: 1008 case MLD_SILENT_MEMBER: 1009 case MLD_SLEEPING_MEMBER: 1010 case MLD_LAZY_MEMBER: 1011 case MLD_AWAKENING_MEMBER: 1012 case MLD_IDLE_MEMBER: 1013 case MLD_LEAVING_MEMBER: 1014 return (retval); 1015 break; 1016 case MLD_REPORTING_MEMBER: 1017 case MLD_G_QUERY_PENDING_MEMBER: 1018 case MLD_SG_QUERY_PENDING_MEMBER: 1019 break; 1020 } 1021 1022 nsrc = ntohs(mld->mld_numsrc); 1023 1024 /* Length should be checked by calling function. */ 1025 KASSERT((m0->m_flags & M_PKTHDR) == 0 || 1026 m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) + 1027 nsrc * sizeof(struct in6_addr), 1028 ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)", 1029 m0->m_pkthdr.len, off + sizeof(struct mldv2_query) + 1030 nsrc * sizeof(struct in6_addr), m0)); 1031 1032 1033 /* 1034 * Deal with group-specific queries upfront. 1035 * If any group query is already pending, purge any recorded 1036 * source-list state if it exists, and schedule a query response 1037 * for this group-specific query. 1038 */ 1039 if (nsrc == 0) { 1040 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 1041 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { 1042 in6m_clear_recorded(inm); 1043 timer = min(inm->in6m_timer, timer); 1044 } 1045 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; 1046 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1047 V_current_state_timers_running6 = 1; 1048 return (retval); 1049 } 1050 1051 /* 1052 * Deal with the case where a group-and-source-specific query has 1053 * been received but a group-specific query is already pending. 1054 */ 1055 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { 1056 timer = min(inm->in6m_timer, timer); 1057 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1058 V_current_state_timers_running6 = 1; 1059 return (retval); 1060 } 1061 1062 /* 1063 * Finally, deal with the case where a group-and-source-specific 1064 * query has been received, where a response to a previous g-s-r 1065 * query exists, or none exists. 1066 * In this case, we need to parse the source-list which the Querier 1067 * has provided us with and check if we have any source list filter 1068 * entries at T1 for these sources. If we do not, there is no need 1069 * schedule a report and the query may be dropped. 1070 * If we do, we must record them and schedule a current-state 1071 * report for those sources. 1072 */ 1073 if (inm->in6m_nsrc > 0) { 1074 struct in6_addr srcaddr; 1075 int i, nrecorded; 1076 int soff; 1077 1078 soff = off + sizeof(struct mldv2_query); 1079 nrecorded = 0; 1080 for (i = 0; i < nsrc; i++) { 1081 m_copydata(m0, soff, sizeof(struct in6_addr), 1082 (caddr_t)&srcaddr); 1083 retval = in6m_record_source(inm, &srcaddr); 1084 if (retval < 0) 1085 break; 1086 nrecorded += retval; 1087 soff += sizeof(struct in6_addr); 1088 } 1089 if (nrecorded > 0) { 1090 CTR1(KTR_MLD, 1091 "%s: schedule response to SG query", __func__); 1092 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; 1093 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1094 V_current_state_timers_running6 = 1; 1095 } 1096 } 1097 1098 return (retval); 1099 } 1100 1101 /* 1102 * Process a received MLDv1 host membership report. 1103 * Assumes mld points to mld_hdr in pulled up mbuf chain. 1104 * 1105 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 1106 * mld_addr. This is OK as we own the mbuf chain. 1107 */ 1108 static int 1109 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, 1110 /*const*/ struct mld_hdr *mld) 1111 { 1112 struct in6_addr src, dst; 1113 struct epoch_tracker et; 1114 struct in6_ifaddr *ia; 1115 struct in6_multi *inm; 1116 #ifdef KTR 1117 char ip6tbuf[INET6_ADDRSTRLEN]; 1118 #endif 1119 1120 if (!mld_v1enable) { 1121 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", 1122 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1123 ifp, if_name(ifp)); 1124 return (0); 1125 } 1126 1127 if (ifp->if_flags & IFF_LOOPBACK) 1128 return (0); 1129 1130 /* 1131 * MLDv1 reports must originate from a host's link-local address, 1132 * or the unspecified address (when booting). 1133 */ 1134 src = ip6->ip6_src; 1135 in6_clearscope(&src); 1136 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { 1137 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 1138 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 1139 ifp, if_name(ifp)); 1140 return (EINVAL); 1141 } 1142 1143 /* 1144 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast 1145 * group, and must be directed to the group itself. 1146 */ 1147 dst = ip6->ip6_dst; 1148 in6_clearscope(&dst); 1149 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || 1150 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { 1151 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", 1152 ip6_sprintf(ip6tbuf, &ip6->ip6_dst), 1153 ifp, if_name(ifp)); 1154 return (EINVAL); 1155 } 1156 1157 /* 1158 * Make sure we don't hear our own membership report, as fast 1159 * leave requires knowing that we are the only member of a 1160 * group. Assume we used the link-local address if available, 1161 * otherwise look for ::. 1162 * 1163 * XXX Note that scope ID comparison is needed for the address 1164 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be 1165 * performed for the on-wire address. 1166 */ 1167 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1168 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || 1169 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) { 1170 if (ia != NULL) 1171 ifa_free(&ia->ia_ifa); 1172 return (0); 1173 } 1174 if (ia != NULL) 1175 ifa_free(&ia->ia_ifa); 1176 1177 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", 1178 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); 1179 1180 /* 1181 * Embed scope ID of receiving interface in MLD query for lookup 1182 * whilst we don't hold other locks (due to KAME locking lameness). 1183 */ 1184 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) 1185 in6_setscope(&mld->mld_addr, ifp, NULL); 1186 1187 IN6_MULTI_LIST_LOCK(); 1188 MLD_LOCK(); 1189 NET_EPOCH_ENTER(et); 1190 1191 /* 1192 * MLDv1 report suppression. 1193 * If we are a member of this group, and our membership should be 1194 * reported, and our group timer is pending or about to be reset, 1195 * stop our group timer by transitioning to the 'lazy' state. 1196 */ 1197 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 1198 if (inm != NULL) { 1199 struct mld_ifsoftc *mli; 1200 1201 mli = inm->in6m_mli; 1202 KASSERT(mli != NULL, 1203 ("%s: no mli for ifp %p", __func__, ifp)); 1204 1205 /* 1206 * If we are in MLDv2 host mode, do not allow the 1207 * other host's MLDv1 report to suppress our reports. 1208 */ 1209 if (mli->mli_version == MLD_VERSION_2) 1210 goto out_locked; 1211 1212 inm->in6m_timer = 0; 1213 1214 switch (inm->in6m_state) { 1215 case MLD_NOT_MEMBER: 1216 case MLD_SILENT_MEMBER: 1217 case MLD_SLEEPING_MEMBER: 1218 break; 1219 case MLD_REPORTING_MEMBER: 1220 case MLD_IDLE_MEMBER: 1221 case MLD_AWAKENING_MEMBER: 1222 CTR3(KTR_MLD, 1223 "report suppressed for %s on ifp %p(%s)", 1224 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1225 ifp, if_name(ifp)); 1226 case MLD_LAZY_MEMBER: 1227 inm->in6m_state = MLD_LAZY_MEMBER; 1228 break; 1229 case MLD_G_QUERY_PENDING_MEMBER: 1230 case MLD_SG_QUERY_PENDING_MEMBER: 1231 case MLD_LEAVING_MEMBER: 1232 break; 1233 } 1234 } 1235 1236 out_locked: 1237 NET_EPOCH_EXIT(et); 1238 MLD_UNLOCK(); 1239 IN6_MULTI_LIST_UNLOCK(); 1240 1241 /* XXX Clear embedded scope ID as userland won't expect it. */ 1242 in6_clearscope(&mld->mld_addr); 1243 1244 return (0); 1245 } 1246 1247 /* 1248 * MLD input path. 1249 * 1250 * Assume query messages which fit in a single ICMPv6 message header 1251 * have been pulled up. 1252 * Assume that userland will want to see the message, even if it 1253 * otherwise fails kernel input validation; do not free it. 1254 * Pullup may however free the mbuf chain m if it fails. 1255 * 1256 * Return IPPROTO_DONE if we freed m. Otherwise, return 0. 1257 */ 1258 int 1259 mld_input(struct mbuf *m, int off, int icmp6len) 1260 { 1261 struct ifnet *ifp; 1262 struct ip6_hdr *ip6; 1263 struct mld_hdr *mld; 1264 int mldlen; 1265 1266 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); 1267 1268 ifp = m->m_pkthdr.rcvif; 1269 1270 ip6 = mtod(m, struct ip6_hdr *); 1271 1272 /* Pullup to appropriate size. */ 1273 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); 1274 if (mld->mld_type == MLD_LISTENER_QUERY && 1275 icmp6len >= sizeof(struct mldv2_query)) { 1276 mldlen = sizeof(struct mldv2_query); 1277 } else { 1278 mldlen = sizeof(struct mld_hdr); 1279 } 1280 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen); 1281 if (mld == NULL) { 1282 ICMP6STAT_INC(icp6s_badlen); 1283 return (IPPROTO_DONE); 1284 } 1285 1286 /* 1287 * Userland needs to see all of this traffic for implementing 1288 * the endpoint discovery portion of multicast routing. 1289 */ 1290 switch (mld->mld_type) { 1291 case MLD_LISTENER_QUERY: 1292 icmp6_ifstat_inc(ifp, ifs6_in_mldquery); 1293 if (icmp6len == sizeof(struct mld_hdr)) { 1294 if (mld_v1_input_query(ifp, ip6, mld) != 0) 1295 return (0); 1296 } else if (icmp6len >= sizeof(struct mldv2_query)) { 1297 if (mld_v2_input_query(ifp, ip6, m, 1298 (struct mldv2_query *)mld, off, icmp6len) != 0) 1299 return (0); 1300 } 1301 break; 1302 case MLD_LISTENER_REPORT: 1303 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1304 if (mld_v1_input_report(ifp, ip6, mld) != 0) 1305 return (0); 1306 break; 1307 case MLDV2_LISTENER_REPORT: 1308 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1309 break; 1310 case MLD_LISTENER_DONE: 1311 icmp6_ifstat_inc(ifp, ifs6_in_mlddone); 1312 break; 1313 default: 1314 break; 1315 } 1316 1317 return (0); 1318 } 1319 1320 /* 1321 * Fast timeout handler (global). 1322 * VIMAGE: Timeout handlers are expected to service all vimages. 1323 */ 1324 void 1325 mld_fasttimo(void) 1326 { 1327 struct in6_multi_head inmh; 1328 VNET_ITERATOR_DECL(vnet_iter); 1329 1330 SLIST_INIT(&inmh); 1331 1332 VNET_LIST_RLOCK_NOSLEEP(); 1333 VNET_FOREACH(vnet_iter) { 1334 CURVNET_SET(vnet_iter); 1335 mld_fasttimo_vnet(&inmh); 1336 CURVNET_RESTORE(); 1337 } 1338 VNET_LIST_RUNLOCK_NOSLEEP(); 1339 in6m_release_list_deferred(&inmh); 1340 } 1341 1342 /* 1343 * Fast timeout handler (per-vnet). 1344 * 1345 * VIMAGE: Assume caller has set up our curvnet. 1346 */ 1347 static void 1348 mld_fasttimo_vnet(struct in6_multi_head *inmh) 1349 { 1350 struct epoch_tracker et; 1351 struct mbufq scq; /* State-change packets */ 1352 struct mbufq qrq; /* Query response packets */ 1353 struct ifnet *ifp; 1354 struct mld_ifsoftc *mli; 1355 struct ifmultiaddr *ifma; 1356 struct in6_multi *inm; 1357 int uri_fasthz; 1358 1359 uri_fasthz = 0; 1360 1361 /* 1362 * Quick check to see if any work needs to be done, in order to 1363 * minimize the overhead of fasttimo processing. 1364 * SMPng: XXX Unlocked reads. 1365 */ 1366 if (!V_current_state_timers_running6 && 1367 !V_interface_timers_running6 && 1368 !V_state_change_timers_running6) 1369 return; 1370 1371 IN6_MULTI_LIST_LOCK(); 1372 MLD_LOCK(); 1373 1374 /* 1375 * MLDv2 General Query response timer processing. 1376 */ 1377 if (V_interface_timers_running6) { 1378 CTR1(KTR_MLD, "%s: interface timers running", __func__); 1379 1380 V_interface_timers_running6 = 0; 1381 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1382 if (mli->mli_v2_timer == 0) { 1383 /* Do nothing. */ 1384 } else if (--mli->mli_v2_timer == 0) { 1385 mld_v2_dispatch_general_query(mli); 1386 } else { 1387 V_interface_timers_running6 = 1; 1388 } 1389 } 1390 } 1391 1392 if (!V_current_state_timers_running6 && 1393 !V_state_change_timers_running6) 1394 goto out_locked; 1395 1396 V_current_state_timers_running6 = 0; 1397 V_state_change_timers_running6 = 0; 1398 1399 CTR1(KTR_MLD, "%s: state change timers running", __func__); 1400 1401 /* 1402 * MLD host report and state-change timer processing. 1403 * Note: Processing a v2 group timer may remove a node. 1404 */ 1405 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1406 ifp = mli->mli_ifp; 1407 1408 if (mli->mli_version == MLD_VERSION_2) { 1409 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * 1410 PR_FASTHZ); 1411 mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS); 1412 mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS); 1413 } 1414 1415 IF_ADDR_WLOCK(ifp); 1416 NET_EPOCH_ENTER(et); 1417 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1418 inm = in6m_ifmultiaddr_get_inm(ifma); 1419 if (inm == NULL) 1420 continue; 1421 switch (mli->mli_version) { 1422 case MLD_VERSION_1: 1423 mld_v1_process_group_timer(inmh, inm); 1424 break; 1425 case MLD_VERSION_2: 1426 mld_v2_process_group_timers(inmh, &qrq, 1427 &scq, inm, uri_fasthz); 1428 break; 1429 } 1430 } 1431 IF_ADDR_WUNLOCK(ifp); 1432 1433 switch (mli->mli_version) { 1434 case MLD_VERSION_1: 1435 /* 1436 * Transmit reports for this lifecycle. This 1437 * is done while not holding IF_ADDR_LOCK 1438 * since this can call 1439 * in6ifa_ifpforlinklocal() which locks 1440 * IF_ADDR_LOCK internally as well as 1441 * ip6_output() to transmit a packet. 1442 */ 1443 while ((inm = SLIST_FIRST(inmh)) != NULL) { 1444 SLIST_REMOVE_HEAD(inmh, in6m_defer); 1445 (void)mld_v1_transmit_report(inm, 1446 MLD_LISTENER_REPORT); 1447 } 1448 break; 1449 case MLD_VERSION_2: 1450 mld_dispatch_queue(&qrq, 0); 1451 mld_dispatch_queue(&scq, 0); 1452 break; 1453 } 1454 NET_EPOCH_EXIT(et); 1455 } 1456 1457 out_locked: 1458 MLD_UNLOCK(); 1459 IN6_MULTI_LIST_UNLOCK(); 1460 } 1461 1462 /* 1463 * Update host report group timer. 1464 * Will update the global pending timer flags. 1465 */ 1466 static void 1467 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm) 1468 { 1469 int report_timer_expired; 1470 1471 IN6_MULTI_LIST_LOCK_ASSERT(); 1472 MLD_LOCK_ASSERT(); 1473 1474 if (inm->in6m_timer == 0) { 1475 report_timer_expired = 0; 1476 } else if (--inm->in6m_timer == 0) { 1477 report_timer_expired = 1; 1478 } else { 1479 V_current_state_timers_running6 = 1; 1480 return; 1481 } 1482 1483 switch (inm->in6m_state) { 1484 case MLD_NOT_MEMBER: 1485 case MLD_SILENT_MEMBER: 1486 case MLD_IDLE_MEMBER: 1487 case MLD_LAZY_MEMBER: 1488 case MLD_SLEEPING_MEMBER: 1489 case MLD_AWAKENING_MEMBER: 1490 break; 1491 case MLD_REPORTING_MEMBER: 1492 if (report_timer_expired) { 1493 inm->in6m_state = MLD_IDLE_MEMBER; 1494 SLIST_INSERT_HEAD(inmh, inm, in6m_defer); 1495 } 1496 break; 1497 case MLD_G_QUERY_PENDING_MEMBER: 1498 case MLD_SG_QUERY_PENDING_MEMBER: 1499 case MLD_LEAVING_MEMBER: 1500 break; 1501 } 1502 } 1503 1504 /* 1505 * Update a group's timers for MLDv2. 1506 * Will update the global pending timer flags. 1507 * Note: Unlocked read from mli. 1508 */ 1509 static void 1510 mld_v2_process_group_timers(struct in6_multi_head *inmh, 1511 struct mbufq *qrq, struct mbufq *scq, 1512 struct in6_multi *inm, const int uri_fasthz) 1513 { 1514 int query_response_timer_expired; 1515 int state_change_retransmit_timer_expired; 1516 #ifdef KTR 1517 char ip6tbuf[INET6_ADDRSTRLEN]; 1518 #endif 1519 1520 IN6_MULTI_LIST_LOCK_ASSERT(); 1521 MLD_LOCK_ASSERT(); 1522 1523 query_response_timer_expired = 0; 1524 state_change_retransmit_timer_expired = 0; 1525 1526 /* 1527 * During a transition from compatibility mode back to MLDv2, 1528 * a group record in REPORTING state may still have its group 1529 * timer active. This is a no-op in this function; it is easier 1530 * to deal with it here than to complicate the slow-timeout path. 1531 */ 1532 if (inm->in6m_timer == 0) { 1533 query_response_timer_expired = 0; 1534 } else if (--inm->in6m_timer == 0) { 1535 query_response_timer_expired = 1; 1536 } else { 1537 V_current_state_timers_running6 = 1; 1538 } 1539 1540 if (inm->in6m_sctimer == 0) { 1541 state_change_retransmit_timer_expired = 0; 1542 } else if (--inm->in6m_sctimer == 0) { 1543 state_change_retransmit_timer_expired = 1; 1544 } else { 1545 V_state_change_timers_running6 = 1; 1546 } 1547 1548 /* We are in fasttimo, so be quick about it. */ 1549 if (!state_change_retransmit_timer_expired && 1550 !query_response_timer_expired) 1551 return; 1552 1553 switch (inm->in6m_state) { 1554 case MLD_NOT_MEMBER: 1555 case MLD_SILENT_MEMBER: 1556 case MLD_SLEEPING_MEMBER: 1557 case MLD_LAZY_MEMBER: 1558 case MLD_AWAKENING_MEMBER: 1559 case MLD_IDLE_MEMBER: 1560 break; 1561 case MLD_G_QUERY_PENDING_MEMBER: 1562 case MLD_SG_QUERY_PENDING_MEMBER: 1563 /* 1564 * Respond to a previously pending Group-Specific 1565 * or Group-and-Source-Specific query by enqueueing 1566 * the appropriate Current-State report for 1567 * immediate transmission. 1568 */ 1569 if (query_response_timer_expired) { 1570 int retval; 1571 1572 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, 1573 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER), 1574 0); 1575 CTR2(KTR_MLD, "%s: enqueue record = %d", 1576 __func__, retval); 1577 inm->in6m_state = MLD_REPORTING_MEMBER; 1578 in6m_clear_recorded(inm); 1579 } 1580 /* FALLTHROUGH */ 1581 case MLD_REPORTING_MEMBER: 1582 case MLD_LEAVING_MEMBER: 1583 if (state_change_retransmit_timer_expired) { 1584 /* 1585 * State-change retransmission timer fired. 1586 * If there are any further pending retransmissions, 1587 * set the global pending state-change flag, and 1588 * reset the timer. 1589 */ 1590 if (--inm->in6m_scrv > 0) { 1591 inm->in6m_sctimer = uri_fasthz; 1592 V_state_change_timers_running6 = 1; 1593 } 1594 /* 1595 * Retransmit the previously computed state-change 1596 * report. If there are no further pending 1597 * retransmissions, the mbuf queue will be consumed. 1598 * Update T0 state to T1 as we have now sent 1599 * a state-change. 1600 */ 1601 (void)mld_v2_merge_state_changes(inm, scq); 1602 1603 in6m_commit(inm); 1604 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 1605 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1606 if_name(inm->in6m_ifp)); 1607 1608 /* 1609 * If we are leaving the group for good, make sure 1610 * we release MLD's reference to it. 1611 * This release must be deferred using a SLIST, 1612 * as we are called from a loop which traverses 1613 * the in_ifmultiaddr TAILQ. 1614 */ 1615 if (inm->in6m_state == MLD_LEAVING_MEMBER && 1616 inm->in6m_scrv == 0) { 1617 inm->in6m_state = MLD_NOT_MEMBER; 1618 in6m_disconnect_locked(inmh, inm); 1619 in6m_rele_locked(inmh, inm); 1620 } 1621 } 1622 break; 1623 } 1624 } 1625 1626 /* 1627 * Switch to a different version on the given interface, 1628 * as per Section 9.12. 1629 */ 1630 static void 1631 mld_set_version(struct mld_ifsoftc *mli, const int version) 1632 { 1633 int old_version_timer; 1634 1635 MLD_LOCK_ASSERT(); 1636 1637 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, 1638 version, mli->mli_ifp, if_name(mli->mli_ifp)); 1639 1640 if (version == MLD_VERSION_1) { 1641 /* 1642 * Compute the "Older Version Querier Present" timer as per 1643 * Section 9.12. 1644 */ 1645 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; 1646 old_version_timer *= PR_SLOWHZ; 1647 mli->mli_v1_timer = old_version_timer; 1648 } 1649 1650 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { 1651 mli->mli_version = MLD_VERSION_1; 1652 mld_v2_cancel_link_timers(mli); 1653 } 1654 } 1655 1656 /* 1657 * Cancel pending MLDv2 timers for the given link and all groups 1658 * joined on it; state-change, general-query, and group-query timers. 1659 */ 1660 static void 1661 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli) 1662 { 1663 struct epoch_tracker et; 1664 struct in6_multi_head inmh; 1665 struct ifmultiaddr *ifma; 1666 struct ifnet *ifp; 1667 struct in6_multi *inm; 1668 1669 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, 1670 mli->mli_ifp, if_name(mli->mli_ifp)); 1671 1672 SLIST_INIT(&inmh); 1673 IN6_MULTI_LIST_LOCK_ASSERT(); 1674 MLD_LOCK_ASSERT(); 1675 1676 /* 1677 * Fast-track this potentially expensive operation 1678 * by checking all the global 'timer pending' flags. 1679 */ 1680 if (!V_interface_timers_running6 && 1681 !V_state_change_timers_running6 && 1682 !V_current_state_timers_running6) 1683 return; 1684 1685 mli->mli_v2_timer = 0; 1686 1687 ifp = mli->mli_ifp; 1688 1689 IF_ADDR_WLOCK(ifp); 1690 NET_EPOCH_ENTER(et); 1691 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1692 inm = in6m_ifmultiaddr_get_inm(ifma); 1693 if (inm == NULL) 1694 continue; 1695 switch (inm->in6m_state) { 1696 case MLD_NOT_MEMBER: 1697 case MLD_SILENT_MEMBER: 1698 case MLD_IDLE_MEMBER: 1699 case MLD_LAZY_MEMBER: 1700 case MLD_SLEEPING_MEMBER: 1701 case MLD_AWAKENING_MEMBER: 1702 break; 1703 case MLD_LEAVING_MEMBER: 1704 /* 1705 * If we are leaving the group and switching 1706 * version, we need to release the final 1707 * reference held for issuing the INCLUDE {}. 1708 */ 1709 if (inm->in6m_refcount == 1) 1710 in6m_disconnect_locked(&inmh, inm); 1711 in6m_rele_locked(&inmh, inm); 1712 /* FALLTHROUGH */ 1713 case MLD_G_QUERY_PENDING_MEMBER: 1714 case MLD_SG_QUERY_PENDING_MEMBER: 1715 in6m_clear_recorded(inm); 1716 /* FALLTHROUGH */ 1717 case MLD_REPORTING_MEMBER: 1718 inm->in6m_sctimer = 0; 1719 inm->in6m_timer = 0; 1720 inm->in6m_state = MLD_REPORTING_MEMBER; 1721 /* 1722 * Free any pending MLDv2 state-change records. 1723 */ 1724 mbufq_drain(&inm->in6m_scq); 1725 break; 1726 } 1727 } 1728 NET_EPOCH_EXIT(et); 1729 IF_ADDR_WUNLOCK(ifp); 1730 in6m_release_list_deferred(&inmh); 1731 } 1732 1733 /* 1734 * Global slowtimo handler. 1735 * VIMAGE: Timeout handlers are expected to service all vimages. 1736 */ 1737 void 1738 mld_slowtimo(void) 1739 { 1740 VNET_ITERATOR_DECL(vnet_iter); 1741 1742 VNET_LIST_RLOCK_NOSLEEP(); 1743 VNET_FOREACH(vnet_iter) { 1744 CURVNET_SET(vnet_iter); 1745 mld_slowtimo_vnet(); 1746 CURVNET_RESTORE(); 1747 } 1748 VNET_LIST_RUNLOCK_NOSLEEP(); 1749 } 1750 1751 /* 1752 * Per-vnet slowtimo handler. 1753 */ 1754 static void 1755 mld_slowtimo_vnet(void) 1756 { 1757 struct mld_ifsoftc *mli; 1758 1759 MLD_LOCK(); 1760 1761 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1762 mld_v1_process_querier_timers(mli); 1763 } 1764 1765 MLD_UNLOCK(); 1766 } 1767 1768 /* 1769 * Update the Older Version Querier Present timers for a link. 1770 * See Section 9.12 of RFC 3810. 1771 */ 1772 static void 1773 mld_v1_process_querier_timers(struct mld_ifsoftc *mli) 1774 { 1775 1776 MLD_LOCK_ASSERT(); 1777 1778 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) { 1779 /* 1780 * MLDv1 Querier Present timer expired; revert to MLDv2. 1781 */ 1782 CTR5(KTR_MLD, 1783 "%s: transition from v%d -> v%d on %p(%s)", 1784 __func__, mli->mli_version, MLD_VERSION_2, 1785 mli->mli_ifp, if_name(mli->mli_ifp)); 1786 mli->mli_version = MLD_VERSION_2; 1787 } 1788 } 1789 1790 /* 1791 * Transmit an MLDv1 report immediately. 1792 */ 1793 static int 1794 mld_v1_transmit_report(struct in6_multi *in6m, const int type) 1795 { 1796 struct ifnet *ifp; 1797 struct in6_ifaddr *ia; 1798 struct ip6_hdr *ip6; 1799 struct mbuf *mh, *md; 1800 struct mld_hdr *mld; 1801 1802 IN6_MULTI_LIST_LOCK_ASSERT(); 1803 MLD_LOCK_ASSERT(); 1804 1805 ifp = in6m->in6m_ifp; 1806 /* in process of being freed */ 1807 if (ifp == NULL) 1808 return (0); 1809 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1810 /* ia may be NULL if link-local address is tentative. */ 1811 1812 mh = m_gethdr(M_NOWAIT, MT_DATA); 1813 if (mh == NULL) { 1814 if (ia != NULL) 1815 ifa_free(&ia->ia_ifa); 1816 return (ENOMEM); 1817 } 1818 md = m_get(M_NOWAIT, MT_DATA); 1819 if (md == NULL) { 1820 m_free(mh); 1821 if (ia != NULL) 1822 ifa_free(&ia->ia_ifa); 1823 return (ENOMEM); 1824 } 1825 mh->m_next = md; 1826 1827 /* 1828 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so 1829 * that ether_output() does not need to allocate another mbuf 1830 * for the header in the most common case. 1831 */ 1832 M_ALIGN(mh, sizeof(struct ip6_hdr)); 1833 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); 1834 mh->m_len = sizeof(struct ip6_hdr); 1835 1836 ip6 = mtod(mh, struct ip6_hdr *); 1837 ip6->ip6_flow = 0; 1838 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 1839 ip6->ip6_vfc |= IPV6_VERSION; 1840 ip6->ip6_nxt = IPPROTO_ICMPV6; 1841 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 1842 ip6->ip6_dst = in6m->in6m_addr; 1843 1844 md->m_len = sizeof(struct mld_hdr); 1845 mld = mtod(md, struct mld_hdr *); 1846 mld->mld_type = type; 1847 mld->mld_code = 0; 1848 mld->mld_cksum = 0; 1849 mld->mld_maxdelay = 0; 1850 mld->mld_reserved = 0; 1851 mld->mld_addr = in6m->in6m_addr; 1852 in6_clearscope(&mld->mld_addr); 1853 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 1854 sizeof(struct ip6_hdr), sizeof(struct mld_hdr)); 1855 1856 mld_save_context(mh, ifp); 1857 mh->m_flags |= M_MLDV1; 1858 1859 mld_dispatch_packet(mh); 1860 1861 if (ia != NULL) 1862 ifa_free(&ia->ia_ifa); 1863 return (0); 1864 } 1865 1866 /* 1867 * Process a state change from the upper layer for the given IPv6 group. 1868 * 1869 * Each socket holds a reference on the in_multi in its own ip_moptions. 1870 * The socket layer will have made the necessary updates to.the group 1871 * state, it is now up to MLD to issue a state change report if there 1872 * has been any change between T0 (when the last state-change was issued) 1873 * and T1 (now). 1874 * 1875 * We use the MLDv2 state machine at group level. The MLd module 1876 * however makes the decision as to which MLD protocol version to speak. 1877 * A state change *from* INCLUDE {} always means an initial join. 1878 * A state change *to* INCLUDE {} always means a final leave. 1879 * 1880 * If delay is non-zero, and the state change is an initial multicast 1881 * join, the state change report will be delayed by 'delay' ticks 1882 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise 1883 * the initial MLDv2 state change report will be delayed by whichever 1884 * is sooner, a pending state-change timer or delay itself. 1885 * 1886 * VIMAGE: curvnet should have been set by caller, as this routine 1887 * is called from the socket option handlers. 1888 */ 1889 int 1890 mld_change_state(struct in6_multi *inm, const int delay) 1891 { 1892 struct mld_ifsoftc *mli; 1893 struct ifnet *ifp; 1894 int error; 1895 1896 IN6_MULTI_LIST_LOCK_ASSERT(); 1897 1898 error = 0; 1899 1900 /* 1901 * Check if the in6_multi has already been disconnected. 1902 */ 1903 if (inm->in6m_ifp == NULL) { 1904 CTR1(KTR_MLD, "%s: inm is disconnected", __func__); 1905 return (0); 1906 } 1907 1908 /* 1909 * Try to detect if the upper layer just asked us to change state 1910 * for an interface which has now gone away. 1911 */ 1912 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 1913 ifp = inm->in6m_ifma->ifma_ifp; 1914 if (ifp == NULL) 1915 return (0); 1916 /* 1917 * Sanity check that netinet6's notion of ifp is the 1918 * same as net's. 1919 */ 1920 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 1921 1922 MLD_LOCK(); 1923 mli = MLD_IFINFO(ifp); 1924 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); 1925 1926 /* 1927 * If we detect a state transition to or from MCAST_UNDEFINED 1928 * for this group, then we are starting or finishing an MLD 1929 * life cycle for this group. 1930 */ 1931 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { 1932 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, 1933 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); 1934 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { 1935 CTR1(KTR_MLD, "%s: initial join", __func__); 1936 error = mld_initial_join(inm, mli, delay); 1937 goto out_locked; 1938 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { 1939 CTR1(KTR_MLD, "%s: final leave", __func__); 1940 mld_final_leave(inm, mli); 1941 goto out_locked; 1942 } 1943 } else { 1944 CTR1(KTR_MLD, "%s: filter set change", __func__); 1945 } 1946 1947 error = mld_handle_state_change(inm, mli); 1948 1949 out_locked: 1950 MLD_UNLOCK(); 1951 return (error); 1952 } 1953 1954 /* 1955 * Perform the initial join for an MLD group. 1956 * 1957 * When joining a group: 1958 * If the group should have its MLD traffic suppressed, do nothing. 1959 * MLDv1 starts sending MLDv1 host membership reports. 1960 * MLDv2 will schedule an MLDv2 state-change report containing the 1961 * initial state of the membership. 1962 * 1963 * If the delay argument is non-zero, then we must delay sending the 1964 * initial state change for delay ticks (in units of PR_FASTHZ). 1965 */ 1966 static int 1967 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli, 1968 const int delay) 1969 { 1970 struct ifnet *ifp; 1971 struct mbufq *mq; 1972 int error, retval, syncstates; 1973 int odelay; 1974 #ifdef KTR 1975 char ip6tbuf[INET6_ADDRSTRLEN]; 1976 #endif 1977 1978 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", 1979 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1980 inm->in6m_ifp, if_name(inm->in6m_ifp)); 1981 1982 error = 0; 1983 syncstates = 1; 1984 1985 ifp = inm->in6m_ifp; 1986 1987 IN6_MULTI_LIST_LOCK_ASSERT(); 1988 MLD_LOCK_ASSERT(); 1989 1990 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); 1991 1992 /* 1993 * Groups joined on loopback or marked as 'not reported', 1994 * enter the MLD_SILENT_MEMBER state and 1995 * are never reported in any protocol exchanges. 1996 * All other groups enter the appropriate state machine 1997 * for the version in use on this link. 1998 * A link marked as MLIF_SILENT causes MLD to be completely 1999 * disabled for the link. 2000 */ 2001 if ((ifp->if_flags & IFF_LOOPBACK) || 2002 (mli->mli_flags & MLIF_SILENT) || 2003 !mld_is_addr_reported(&inm->in6m_addr)) { 2004 CTR1(KTR_MLD, 2005 "%s: not kicking state machine for silent group", __func__); 2006 inm->in6m_state = MLD_SILENT_MEMBER; 2007 inm->in6m_timer = 0; 2008 } else { 2009 /* 2010 * Deal with overlapping in_multi lifecycle. 2011 * If this group was LEAVING, then make sure 2012 * we drop the reference we picked up to keep the 2013 * group around for the final INCLUDE {} enqueue. 2014 */ 2015 if (mli->mli_version == MLD_VERSION_2 && 2016 inm->in6m_state == MLD_LEAVING_MEMBER) { 2017 inm->in6m_refcount--; 2018 MPASS(inm->in6m_refcount > 0); 2019 } 2020 inm->in6m_state = MLD_REPORTING_MEMBER; 2021 2022 switch (mli->mli_version) { 2023 case MLD_VERSION_1: 2024 /* 2025 * If a delay was provided, only use it if 2026 * it is greater than the delay normally 2027 * used for an MLDv1 state change report, 2028 * and delay sending the initial MLDv1 report 2029 * by not transitioning to the IDLE state. 2030 */ 2031 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); 2032 if (delay) { 2033 inm->in6m_timer = max(delay, odelay); 2034 V_current_state_timers_running6 = 1; 2035 } else { 2036 inm->in6m_state = MLD_IDLE_MEMBER; 2037 error = mld_v1_transmit_report(inm, 2038 MLD_LISTENER_REPORT); 2039 if (error == 0) { 2040 inm->in6m_timer = odelay; 2041 V_current_state_timers_running6 = 1; 2042 } 2043 } 2044 break; 2045 2046 case MLD_VERSION_2: 2047 /* 2048 * Defer update of T0 to T1, until the first copy 2049 * of the state change has been transmitted. 2050 */ 2051 syncstates = 0; 2052 2053 /* 2054 * Immediately enqueue a State-Change Report for 2055 * this interface, freeing any previous reports. 2056 * Don't kick the timers if there is nothing to do, 2057 * or if an error occurred. 2058 */ 2059 mq = &inm->in6m_scq; 2060 mbufq_drain(mq); 2061 retval = mld_v2_enqueue_group_record(mq, inm, 1, 2062 0, 0, (mli->mli_flags & MLIF_USEALLOW)); 2063 CTR2(KTR_MLD, "%s: enqueue record = %d", 2064 __func__, retval); 2065 if (retval <= 0) { 2066 error = retval * -1; 2067 break; 2068 } 2069 2070 /* 2071 * Schedule transmission of pending state-change 2072 * report up to RV times for this link. The timer 2073 * will fire at the next mld_fasttimo (~200ms), 2074 * giving us an opportunity to merge the reports. 2075 * 2076 * If a delay was provided to this function, only 2077 * use this delay if sooner than the existing one. 2078 */ 2079 KASSERT(mli->mli_rv > 1, 2080 ("%s: invalid robustness %d", __func__, 2081 mli->mli_rv)); 2082 inm->in6m_scrv = mli->mli_rv; 2083 if (delay) { 2084 if (inm->in6m_sctimer > 1) { 2085 inm->in6m_sctimer = 2086 min(inm->in6m_sctimer, delay); 2087 } else 2088 inm->in6m_sctimer = delay; 2089 } else 2090 inm->in6m_sctimer = 1; 2091 V_state_change_timers_running6 = 1; 2092 2093 error = 0; 2094 break; 2095 } 2096 } 2097 2098 /* 2099 * Only update the T0 state if state change is atomic, 2100 * i.e. we don't need to wait for a timer to fire before we 2101 * can consider the state change to have been communicated. 2102 */ 2103 if (syncstates) { 2104 in6m_commit(inm); 2105 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2106 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2107 if_name(inm->in6m_ifp)); 2108 } 2109 2110 return (error); 2111 } 2112 2113 /* 2114 * Issue an intermediate state change during the life-cycle. 2115 */ 2116 static int 2117 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli) 2118 { 2119 struct ifnet *ifp; 2120 int retval; 2121 #ifdef KTR 2122 char ip6tbuf[INET6_ADDRSTRLEN]; 2123 #endif 2124 2125 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)", 2126 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2127 inm->in6m_ifp, if_name(inm->in6m_ifp)); 2128 2129 ifp = inm->in6m_ifp; 2130 2131 IN6_MULTI_LIST_LOCK_ASSERT(); 2132 MLD_LOCK_ASSERT(); 2133 2134 KASSERT(mli && mli->mli_ifp == ifp, 2135 ("%s: inconsistent ifp", __func__)); 2136 2137 if ((ifp->if_flags & IFF_LOOPBACK) || 2138 (mli->mli_flags & MLIF_SILENT) || 2139 !mld_is_addr_reported(&inm->in6m_addr) || 2140 (mli->mli_version != MLD_VERSION_2)) { 2141 if (!mld_is_addr_reported(&inm->in6m_addr)) { 2142 CTR1(KTR_MLD, 2143 "%s: not kicking state machine for silent group", __func__); 2144 } 2145 CTR1(KTR_MLD, "%s: nothing to do", __func__); 2146 in6m_commit(inm); 2147 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2148 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2149 if_name(inm->in6m_ifp)); 2150 return (0); 2151 } 2152 2153 mbufq_drain(&inm->in6m_scq); 2154 2155 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0, 2156 (mli->mli_flags & MLIF_USEALLOW)); 2157 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); 2158 if (retval <= 0) 2159 return (-retval); 2160 2161 /* 2162 * If record(s) were enqueued, start the state-change 2163 * report timer for this group. 2164 */ 2165 inm->in6m_scrv = mli->mli_rv; 2166 inm->in6m_sctimer = 1; 2167 V_state_change_timers_running6 = 1; 2168 2169 return (0); 2170 } 2171 2172 /* 2173 * Perform the final leave for a multicast address. 2174 * 2175 * When leaving a group: 2176 * MLDv1 sends a DONE message, if and only if we are the reporter. 2177 * MLDv2 enqueues a state-change report containing a transition 2178 * to INCLUDE {} for immediate transmission. 2179 */ 2180 static void 2181 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli) 2182 { 2183 int syncstates; 2184 #ifdef KTR 2185 char ip6tbuf[INET6_ADDRSTRLEN]; 2186 #endif 2187 2188 syncstates = 1; 2189 2190 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)", 2191 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2192 inm->in6m_ifp, if_name(inm->in6m_ifp)); 2193 2194 IN6_MULTI_LIST_LOCK_ASSERT(); 2195 MLD_LOCK_ASSERT(); 2196 2197 switch (inm->in6m_state) { 2198 case MLD_NOT_MEMBER: 2199 case MLD_SILENT_MEMBER: 2200 case MLD_LEAVING_MEMBER: 2201 /* Already leaving or left; do nothing. */ 2202 CTR1(KTR_MLD, 2203 "%s: not kicking state machine for silent group", __func__); 2204 break; 2205 case MLD_REPORTING_MEMBER: 2206 case MLD_IDLE_MEMBER: 2207 case MLD_G_QUERY_PENDING_MEMBER: 2208 case MLD_SG_QUERY_PENDING_MEMBER: 2209 if (mli->mli_version == MLD_VERSION_1) { 2210 #ifdef INVARIANTS 2211 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 2212 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) 2213 panic("%s: MLDv2 state reached, not MLDv2 mode", 2214 __func__); 2215 #endif 2216 mld_v1_transmit_report(inm, MLD_LISTENER_DONE); 2217 inm->in6m_state = MLD_NOT_MEMBER; 2218 V_current_state_timers_running6 = 1; 2219 } else if (mli->mli_version == MLD_VERSION_2) { 2220 /* 2221 * Stop group timer and all pending reports. 2222 * Immediately enqueue a state-change report 2223 * TO_IN {} to be sent on the next fast timeout, 2224 * giving us an opportunity to merge reports. 2225 */ 2226 mbufq_drain(&inm->in6m_scq); 2227 inm->in6m_timer = 0; 2228 inm->in6m_scrv = mli->mli_rv; 2229 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " 2230 "pending retransmissions.", __func__, 2231 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2232 if_name(inm->in6m_ifp), inm->in6m_scrv); 2233 if (inm->in6m_scrv == 0) { 2234 inm->in6m_state = MLD_NOT_MEMBER; 2235 inm->in6m_sctimer = 0; 2236 } else { 2237 int retval; 2238 2239 in6m_acquire_locked(inm); 2240 2241 retval = mld_v2_enqueue_group_record( 2242 &inm->in6m_scq, inm, 1, 0, 0, 2243 (mli->mli_flags & MLIF_USEALLOW)); 2244 KASSERT(retval != 0, 2245 ("%s: enqueue record = %d", __func__, 2246 retval)); 2247 2248 inm->in6m_state = MLD_LEAVING_MEMBER; 2249 inm->in6m_sctimer = 1; 2250 V_state_change_timers_running6 = 1; 2251 syncstates = 0; 2252 } 2253 break; 2254 } 2255 break; 2256 case MLD_LAZY_MEMBER: 2257 case MLD_SLEEPING_MEMBER: 2258 case MLD_AWAKENING_MEMBER: 2259 /* Our reports are suppressed; do nothing. */ 2260 break; 2261 } 2262 2263 if (syncstates) { 2264 in6m_commit(inm); 2265 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2266 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2267 if_name(inm->in6m_ifp)); 2268 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 2269 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", 2270 __func__, &inm->in6m_addr, if_name(inm->in6m_ifp)); 2271 } 2272 } 2273 2274 /* 2275 * Enqueue an MLDv2 group record to the given output queue. 2276 * 2277 * If is_state_change is zero, a current-state record is appended. 2278 * If is_state_change is non-zero, a state-change report is appended. 2279 * 2280 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2281 * If is_group_query is zero, and if there is a packet with free space 2282 * at the tail of the queue, it will be appended to providing there 2283 * is enough free space. 2284 * Otherwise a new mbuf packet chain is allocated. 2285 * 2286 * If is_source_query is non-zero, each source is checked to see if 2287 * it was recorded for a Group-Source query, and will be omitted if 2288 * it is not both in-mode and recorded. 2289 * 2290 * If use_block_allow is non-zero, state change reports for initial join 2291 * and final leave, on an inclusive mode group with a source list, will be 2292 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively. 2293 * 2294 * The function will attempt to allocate leading space in the packet 2295 * for the IPv6+ICMP headers to be prepended without fragmenting the chain. 2296 * 2297 * If successful the size of all data appended to the queue is returned, 2298 * otherwise an error code less than zero is returned, or zero if 2299 * no record(s) were appended. 2300 */ 2301 static int 2302 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm, 2303 const int is_state_change, const int is_group_query, 2304 const int is_source_query, const int use_block_allow) 2305 { 2306 struct mldv2_record mr; 2307 struct mldv2_record *pmr; 2308 struct ifnet *ifp; 2309 struct ip6_msource *ims, *nims; 2310 struct mbuf *m0, *m, *md; 2311 int is_filter_list_change; 2312 int minrec0len, m0srcs, msrcs, nbytes, off; 2313 int record_has_sources; 2314 int now; 2315 int type; 2316 uint8_t mode; 2317 #ifdef KTR 2318 char ip6tbuf[INET6_ADDRSTRLEN]; 2319 #endif 2320 2321 IN6_MULTI_LIST_LOCK_ASSERT(); 2322 2323 ifp = inm->in6m_ifp; 2324 is_filter_list_change = 0; 2325 m = NULL; 2326 m0 = NULL; 2327 m0srcs = 0; 2328 msrcs = 0; 2329 nbytes = 0; 2330 nims = NULL; 2331 record_has_sources = 1; 2332 pmr = NULL; 2333 type = MLD_DO_NOTHING; 2334 mode = inm->in6m_st[1].iss_fmode; 2335 2336 /* 2337 * If we did not transition out of ASM mode during t0->t1, 2338 * and there are no source nodes to process, we can skip 2339 * the generation of source records. 2340 */ 2341 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && 2342 inm->in6m_nsrc == 0) 2343 record_has_sources = 0; 2344 2345 if (is_state_change) { 2346 /* 2347 * Queue a state change record. 2348 * If the mode did not change, and there are non-ASM 2349 * listeners or source filters present, 2350 * we potentially need to issue two records for the group. 2351 * If there are ASM listeners, and there was no filter 2352 * mode transition of any kind, do nothing. 2353 * 2354 * If we are transitioning to MCAST_UNDEFINED, we need 2355 * not send any sources. A transition to/from this state is 2356 * considered inclusive with some special treatment. 2357 * 2358 * If we are rewriting initial joins/leaves to use 2359 * ALLOW/BLOCK, and the group's membership is inclusive, 2360 * we need to send sources in all cases. 2361 */ 2362 if (mode != inm->in6m_st[0].iss_fmode) { 2363 if (mode == MCAST_EXCLUDE) { 2364 CTR1(KTR_MLD, "%s: change to EXCLUDE", 2365 __func__); 2366 type = MLD_CHANGE_TO_EXCLUDE_MODE; 2367 } else { 2368 CTR1(KTR_MLD, "%s: change to INCLUDE", 2369 __func__); 2370 if (use_block_allow) { 2371 /* 2372 * XXX 2373 * Here we're interested in state 2374 * edges either direction between 2375 * MCAST_UNDEFINED and MCAST_INCLUDE. 2376 * Perhaps we should just check 2377 * the group state, rather than 2378 * the filter mode. 2379 */ 2380 if (mode == MCAST_UNDEFINED) { 2381 type = MLD_BLOCK_OLD_SOURCES; 2382 } else { 2383 type = MLD_ALLOW_NEW_SOURCES; 2384 } 2385 } else { 2386 type = MLD_CHANGE_TO_INCLUDE_MODE; 2387 if (mode == MCAST_UNDEFINED) 2388 record_has_sources = 0; 2389 } 2390 } 2391 } else { 2392 if (record_has_sources) { 2393 is_filter_list_change = 1; 2394 } else { 2395 type = MLD_DO_NOTHING; 2396 } 2397 } 2398 } else { 2399 /* 2400 * Queue a current state record. 2401 */ 2402 if (mode == MCAST_EXCLUDE) { 2403 type = MLD_MODE_IS_EXCLUDE; 2404 } else if (mode == MCAST_INCLUDE) { 2405 type = MLD_MODE_IS_INCLUDE; 2406 KASSERT(inm->in6m_st[1].iss_asm == 0, 2407 ("%s: inm %p is INCLUDE but ASM count is %d", 2408 __func__, inm, inm->in6m_st[1].iss_asm)); 2409 } 2410 } 2411 2412 /* 2413 * Generate the filter list changes using a separate function. 2414 */ 2415 if (is_filter_list_change) 2416 return (mld_v2_enqueue_filter_change(mq, inm)); 2417 2418 if (type == MLD_DO_NOTHING) { 2419 CTR3(KTR_MLD, "%s: nothing to do for %s/%s", 2420 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2421 if_name(inm->in6m_ifp)); 2422 return (0); 2423 } 2424 2425 /* 2426 * If any sources are present, we must be able to fit at least 2427 * one in the trailing space of the tail packet's mbuf, 2428 * ideally more. 2429 */ 2430 minrec0len = sizeof(struct mldv2_record); 2431 if (record_has_sources) 2432 minrec0len += sizeof(struct in6_addr); 2433 2434 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, 2435 mld_rec_type_to_str(type), 2436 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2437 if_name(inm->in6m_ifp)); 2438 2439 /* 2440 * Check if we have a packet in the tail of the queue for this 2441 * group into which the first group record for this group will fit. 2442 * Otherwise allocate a new packet. 2443 * Always allocate leading space for IP6+RA+ICMPV6+REPORT. 2444 * Note: Group records for G/GSR query responses MUST be sent 2445 * in their own packet. 2446 */ 2447 m0 = mbufq_last(mq); 2448 if (!is_group_query && 2449 m0 != NULL && 2450 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && 2451 (m0->m_pkthdr.len + minrec0len) < 2452 (ifp->if_mtu - MLD_MTUSPACE)) { 2453 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2454 sizeof(struct mldv2_record)) / 2455 sizeof(struct in6_addr); 2456 m = m0; 2457 CTR1(KTR_MLD, "%s: use existing packet", __func__); 2458 } else { 2459 if (mbufq_full(mq)) { 2460 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2461 return (-ENOMEM); 2462 } 2463 m = NULL; 2464 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2465 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2466 if (!is_state_change && !is_group_query) 2467 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2468 if (m == NULL) 2469 m = m_gethdr(M_NOWAIT, MT_DATA); 2470 if (m == NULL) 2471 return (-ENOMEM); 2472 2473 mld_save_context(m, ifp); 2474 2475 CTR1(KTR_MLD, "%s: allocated first packet", __func__); 2476 } 2477 2478 /* 2479 * Append group record. 2480 * If we have sources, we don't know how many yet. 2481 */ 2482 mr.mr_type = type; 2483 mr.mr_datalen = 0; 2484 mr.mr_numsrc = 0; 2485 mr.mr_addr = inm->in6m_addr; 2486 in6_clearscope(&mr.mr_addr); 2487 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2488 if (m != m0) 2489 m_freem(m); 2490 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2491 return (-ENOMEM); 2492 } 2493 nbytes += sizeof(struct mldv2_record); 2494 2495 /* 2496 * Append as many sources as will fit in the first packet. 2497 * If we are appending to a new packet, the chain allocation 2498 * may potentially use clusters; use m_getptr() in this case. 2499 * If we are appending to an existing packet, we need to obtain 2500 * a pointer to the group record after m_append(), in case a new 2501 * mbuf was allocated. 2502 * 2503 * Only append sources which are in-mode at t1. If we are 2504 * transitioning to MCAST_UNDEFINED state on the group, and 2505 * use_block_allow is zero, do not include source entries. 2506 * Otherwise, we need to include this source in the report. 2507 * 2508 * Only report recorded sources in our filter set when responding 2509 * to a group-source query. 2510 */ 2511 if (record_has_sources) { 2512 if (m == m0) { 2513 md = m_last(m); 2514 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2515 md->m_len - nbytes); 2516 } else { 2517 md = m_getptr(m, 0, &off); 2518 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2519 off); 2520 } 2521 msrcs = 0; 2522 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, 2523 nims) { 2524 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2525 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2526 now = im6s_get_mode(inm, ims, 1); 2527 CTR2(KTR_MLD, "%s: node is %d", __func__, now); 2528 if ((now != mode) || 2529 (now == mode && 2530 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2531 CTR1(KTR_MLD, "%s: skip node", __func__); 2532 continue; 2533 } 2534 if (is_source_query && ims->im6s_stp == 0) { 2535 CTR1(KTR_MLD, "%s: skip unrecorded node", 2536 __func__); 2537 continue; 2538 } 2539 CTR1(KTR_MLD, "%s: append node", __func__); 2540 if (!m_append(m, sizeof(struct in6_addr), 2541 (void *)&ims->im6s_addr)) { 2542 if (m != m0) 2543 m_freem(m); 2544 CTR1(KTR_MLD, "%s: m_append() failed.", 2545 __func__); 2546 return (-ENOMEM); 2547 } 2548 nbytes += sizeof(struct in6_addr); 2549 ++msrcs; 2550 if (msrcs == m0srcs) 2551 break; 2552 } 2553 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, 2554 msrcs); 2555 pmr->mr_numsrc = htons(msrcs); 2556 nbytes += (msrcs * sizeof(struct in6_addr)); 2557 } 2558 2559 if (is_source_query && msrcs == 0) { 2560 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); 2561 if (m != m0) 2562 m_freem(m); 2563 return (0); 2564 } 2565 2566 /* 2567 * We are good to go with first packet. 2568 */ 2569 if (m != m0) { 2570 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); 2571 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2572 mbufq_enqueue(mq, m); 2573 } else 2574 m->m_pkthdr.PH_vt.vt_nrecs++; 2575 2576 /* 2577 * No further work needed if no source list in packet(s). 2578 */ 2579 if (!record_has_sources) 2580 return (nbytes); 2581 2582 /* 2583 * Whilst sources remain to be announced, we need to allocate 2584 * a new packet and fill out as many sources as will fit. 2585 * Always try for a cluster first. 2586 */ 2587 while (nims != NULL) { 2588 if (mbufq_full(mq)) { 2589 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2590 return (-ENOMEM); 2591 } 2592 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2593 if (m == NULL) 2594 m = m_gethdr(M_NOWAIT, MT_DATA); 2595 if (m == NULL) 2596 return (-ENOMEM); 2597 mld_save_context(m, ifp); 2598 md = m_getptr(m, 0, &off); 2599 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); 2600 CTR1(KTR_MLD, "%s: allocated next packet", __func__); 2601 2602 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2603 if (m != m0) 2604 m_freem(m); 2605 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2606 return (-ENOMEM); 2607 } 2608 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2609 nbytes += sizeof(struct mldv2_record); 2610 2611 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2612 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2613 2614 msrcs = 0; 2615 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2616 CTR2(KTR_MLD, "%s: visit node %s", 2617 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2618 now = im6s_get_mode(inm, ims, 1); 2619 if ((now != mode) || 2620 (now == mode && 2621 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2622 CTR1(KTR_MLD, "%s: skip node", __func__); 2623 continue; 2624 } 2625 if (is_source_query && ims->im6s_stp == 0) { 2626 CTR1(KTR_MLD, "%s: skip unrecorded node", 2627 __func__); 2628 continue; 2629 } 2630 CTR1(KTR_MLD, "%s: append node", __func__); 2631 if (!m_append(m, sizeof(struct in6_addr), 2632 (void *)&ims->im6s_addr)) { 2633 if (m != m0) 2634 m_freem(m); 2635 CTR1(KTR_MLD, "%s: m_append() failed.", 2636 __func__); 2637 return (-ENOMEM); 2638 } 2639 ++msrcs; 2640 if (msrcs == m0srcs) 2641 break; 2642 } 2643 pmr->mr_numsrc = htons(msrcs); 2644 nbytes += (msrcs * sizeof(struct in6_addr)); 2645 2646 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); 2647 mbufq_enqueue(mq, m); 2648 } 2649 2650 return (nbytes); 2651 } 2652 2653 /* 2654 * Type used to mark record pass completion. 2655 * We exploit the fact we can cast to this easily from the 2656 * current filter modes on each ip_msource node. 2657 */ 2658 typedef enum { 2659 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2660 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2661 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2662 REC_FULL = REC_ALLOW | REC_BLOCK 2663 } rectype_t; 2664 2665 /* 2666 * Enqueue an MLDv2 filter list change to the given output queue. 2667 * 2668 * Source list filter state is held in an RB-tree. When the filter list 2669 * for a group is changed without changing its mode, we need to compute 2670 * the deltas between T0 and T1 for each source in the filter set, 2671 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2672 * 2673 * As we may potentially queue two record types, and the entire R-B tree 2674 * needs to be walked at once, we break this out into its own function 2675 * so we can generate a tightly packed queue of packets. 2676 * 2677 * XXX This could be written to only use one tree walk, although that makes 2678 * serializing into the mbuf chains a bit harder. For now we do two walks 2679 * which makes things easier on us, and it may or may not be harder on 2680 * the L2 cache. 2681 * 2682 * If successful the size of all data appended to the queue is returned, 2683 * otherwise an error code less than zero is returned, or zero if 2684 * no record(s) were appended. 2685 */ 2686 static int 2687 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm) 2688 { 2689 static const int MINRECLEN = 2690 sizeof(struct mldv2_record) + sizeof(struct in6_addr); 2691 struct ifnet *ifp; 2692 struct mldv2_record mr; 2693 struct mldv2_record *pmr; 2694 struct ip6_msource *ims, *nims; 2695 struct mbuf *m, *m0, *md; 2696 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 2697 int nallow, nblock; 2698 uint8_t mode, now, then; 2699 rectype_t crt, drt, nrt; 2700 #ifdef KTR 2701 char ip6tbuf[INET6_ADDRSTRLEN]; 2702 #endif 2703 2704 IN6_MULTI_LIST_LOCK_ASSERT(); 2705 2706 if (inm->in6m_nsrc == 0 || 2707 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) 2708 return (0); 2709 2710 ifp = inm->in6m_ifp; /* interface */ 2711 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ 2712 crt = REC_NONE; /* current group record type */ 2713 drt = REC_NONE; /* mask of completed group record types */ 2714 nrt = REC_NONE; /* record type for current node */ 2715 m0srcs = 0; /* # source which will fit in current mbuf chain */ 2716 npbytes = 0; /* # of bytes appended this packet */ 2717 nbytes = 0; /* # of bytes appended to group's state-change queue */ 2718 rsrcs = 0; /* # sources encoded in current record */ 2719 schanged = 0; /* # nodes encoded in overall filter change */ 2720 nallow = 0; /* # of source entries in ALLOW_NEW */ 2721 nblock = 0; /* # of source entries in BLOCK_OLD */ 2722 nims = NULL; /* next tree node pointer */ 2723 2724 /* 2725 * For each possible filter record mode. 2726 * The first kind of source we encounter tells us which 2727 * is the first kind of record we start appending. 2728 * If a node transitioned to UNDEFINED at t1, its mode is treated 2729 * as the inverse of the group's filter mode. 2730 */ 2731 while (drt != REC_FULL) { 2732 do { 2733 m0 = mbufq_last(mq); 2734 if (m0 != NULL && 2735 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 2736 MLD_V2_REPORT_MAXRECS) && 2737 (m0->m_pkthdr.len + MINRECLEN) < 2738 (ifp->if_mtu - MLD_MTUSPACE)) { 2739 m = m0; 2740 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2741 sizeof(struct mldv2_record)) / 2742 sizeof(struct in6_addr); 2743 CTR1(KTR_MLD, 2744 "%s: use previous packet", __func__); 2745 } else { 2746 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2747 if (m == NULL) 2748 m = m_gethdr(M_NOWAIT, MT_DATA); 2749 if (m == NULL) { 2750 CTR1(KTR_MLD, 2751 "%s: m_get*() failed", __func__); 2752 return (-ENOMEM); 2753 } 2754 m->m_pkthdr.PH_vt.vt_nrecs = 0; 2755 mld_save_context(m, ifp); 2756 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2757 sizeof(struct mldv2_record)) / 2758 sizeof(struct in6_addr); 2759 npbytes = 0; 2760 CTR1(KTR_MLD, 2761 "%s: allocated new packet", __func__); 2762 } 2763 /* 2764 * Append the MLD group record header to the 2765 * current packet's data area. 2766 * Recalculate pointer to free space for next 2767 * group record, in case m_append() allocated 2768 * a new mbuf or cluster. 2769 */ 2770 memset(&mr, 0, sizeof(mr)); 2771 mr.mr_addr = inm->in6m_addr; 2772 in6_clearscope(&mr.mr_addr); 2773 if (!m_append(m, sizeof(mr), (void *)&mr)) { 2774 if (m != m0) 2775 m_freem(m); 2776 CTR1(KTR_MLD, 2777 "%s: m_append() failed", __func__); 2778 return (-ENOMEM); 2779 } 2780 npbytes += sizeof(struct mldv2_record); 2781 if (m != m0) { 2782 /* new packet; offset in chain */ 2783 md = m_getptr(m, npbytes - 2784 sizeof(struct mldv2_record), &off); 2785 pmr = (struct mldv2_record *)(mtod(md, 2786 uint8_t *) + off); 2787 } else { 2788 /* current packet; offset from last append */ 2789 md = m_last(m); 2790 pmr = (struct mldv2_record *)(mtod(md, 2791 uint8_t *) + md->m_len - 2792 sizeof(struct mldv2_record)); 2793 } 2794 /* 2795 * Begin walking the tree for this record type 2796 * pass, or continue from where we left off 2797 * previously if we had to allocate a new packet. 2798 * Only report deltas in-mode at t1. 2799 * We need not report included sources as allowed 2800 * if we are in inclusive mode on the group, 2801 * however the converse is not true. 2802 */ 2803 rsrcs = 0; 2804 if (nims == NULL) { 2805 nims = RB_MIN(ip6_msource_tree, 2806 &inm->in6m_srcs); 2807 } 2808 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2809 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2810 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2811 now = im6s_get_mode(inm, ims, 1); 2812 then = im6s_get_mode(inm, ims, 0); 2813 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", 2814 __func__, then, now); 2815 if (now == then) { 2816 CTR1(KTR_MLD, 2817 "%s: skip unchanged", __func__); 2818 continue; 2819 } 2820 if (mode == MCAST_EXCLUDE && 2821 now == MCAST_INCLUDE) { 2822 CTR1(KTR_MLD, 2823 "%s: skip IN src on EX group", 2824 __func__); 2825 continue; 2826 } 2827 nrt = (rectype_t)now; 2828 if (nrt == REC_NONE) 2829 nrt = (rectype_t)(~mode & REC_FULL); 2830 if (schanged++ == 0) { 2831 crt = nrt; 2832 } else if (crt != nrt) 2833 continue; 2834 if (!m_append(m, sizeof(struct in6_addr), 2835 (void *)&ims->im6s_addr)) { 2836 if (m != m0) 2837 m_freem(m); 2838 CTR1(KTR_MLD, 2839 "%s: m_append() failed", __func__); 2840 return (-ENOMEM); 2841 } 2842 nallow += !!(crt == REC_ALLOW); 2843 nblock += !!(crt == REC_BLOCK); 2844 if (++rsrcs == m0srcs) 2845 break; 2846 } 2847 /* 2848 * If we did not append any tree nodes on this 2849 * pass, back out of allocations. 2850 */ 2851 if (rsrcs == 0) { 2852 npbytes -= sizeof(struct mldv2_record); 2853 if (m != m0) { 2854 CTR1(KTR_MLD, 2855 "%s: m_free(m)", __func__); 2856 m_freem(m); 2857 } else { 2858 CTR1(KTR_MLD, 2859 "%s: m_adj(m, -mr)", __func__); 2860 m_adj(m, -((int)sizeof( 2861 struct mldv2_record))); 2862 } 2863 continue; 2864 } 2865 npbytes += (rsrcs * sizeof(struct in6_addr)); 2866 if (crt == REC_ALLOW) 2867 pmr->mr_type = MLD_ALLOW_NEW_SOURCES; 2868 else if (crt == REC_BLOCK) 2869 pmr->mr_type = MLD_BLOCK_OLD_SOURCES; 2870 pmr->mr_numsrc = htons(rsrcs); 2871 /* 2872 * Count the new group record, and enqueue this 2873 * packet if it wasn't already queued. 2874 */ 2875 m->m_pkthdr.PH_vt.vt_nrecs++; 2876 if (m != m0) 2877 mbufq_enqueue(mq, m); 2878 nbytes += npbytes; 2879 } while (nims != NULL); 2880 drt |= crt; 2881 crt = (~crt & REC_FULL); 2882 } 2883 2884 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 2885 nallow, nblock); 2886 2887 return (nbytes); 2888 } 2889 2890 static int 2891 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq) 2892 { 2893 struct mbufq *gq; 2894 struct mbuf *m; /* pending state-change */ 2895 struct mbuf *m0; /* copy of pending state-change */ 2896 struct mbuf *mt; /* last state-change in packet */ 2897 int docopy, domerge; 2898 u_int recslen; 2899 2900 docopy = 0; 2901 domerge = 0; 2902 recslen = 0; 2903 2904 IN6_MULTI_LIST_LOCK_ASSERT(); 2905 MLD_LOCK_ASSERT(); 2906 2907 /* 2908 * If there are further pending retransmissions, make a writable 2909 * copy of each queued state-change message before merging. 2910 */ 2911 if (inm->in6m_scrv > 0) 2912 docopy = 1; 2913 2914 gq = &inm->in6m_scq; 2915 #ifdef KTR 2916 if (mbufq_first(gq) == NULL) { 2917 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", 2918 __func__, inm); 2919 } 2920 #endif 2921 2922 m = mbufq_first(gq); 2923 while (m != NULL) { 2924 /* 2925 * Only merge the report into the current packet if 2926 * there is sufficient space to do so; an MLDv2 report 2927 * packet may only contain 65,535 group records. 2928 * Always use a simple mbuf chain concatentation to do this, 2929 * as large state changes for single groups may have 2930 * allocated clusters. 2931 */ 2932 domerge = 0; 2933 mt = mbufq_last(scq); 2934 if (mt != NULL) { 2935 recslen = m_length(m, NULL); 2936 2937 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 2938 m->m_pkthdr.PH_vt.vt_nrecs <= 2939 MLD_V2_REPORT_MAXRECS) && 2940 (mt->m_pkthdr.len + recslen <= 2941 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) 2942 domerge = 1; 2943 } 2944 2945 if (!domerge && mbufq_full(gq)) { 2946 CTR2(KTR_MLD, 2947 "%s: outbound queue full, skipping whole packet %p", 2948 __func__, m); 2949 mt = m->m_nextpkt; 2950 if (!docopy) 2951 m_freem(m); 2952 m = mt; 2953 continue; 2954 } 2955 2956 if (!docopy) { 2957 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); 2958 m0 = mbufq_dequeue(gq); 2959 m = m0->m_nextpkt; 2960 } else { 2961 CTR2(KTR_MLD, "%s: copying %p", __func__, m); 2962 m0 = m_dup(m, M_NOWAIT); 2963 if (m0 == NULL) 2964 return (ENOMEM); 2965 m0->m_nextpkt = NULL; 2966 m = m->m_nextpkt; 2967 } 2968 2969 if (!domerge) { 2970 CTR3(KTR_MLD, "%s: queueing %p to scq %p)", 2971 __func__, m0, scq); 2972 mbufq_enqueue(scq, m0); 2973 } else { 2974 struct mbuf *mtl; /* last mbuf of packet mt */ 2975 2976 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", 2977 __func__, m0, mt); 2978 2979 mtl = m_last(mt); 2980 m0->m_flags &= ~M_PKTHDR; 2981 mt->m_pkthdr.len += recslen; 2982 mt->m_pkthdr.PH_vt.vt_nrecs += 2983 m0->m_pkthdr.PH_vt.vt_nrecs; 2984 2985 mtl->m_next = m0; 2986 } 2987 } 2988 2989 return (0); 2990 } 2991 2992 /* 2993 * Respond to a pending MLDv2 General Query. 2994 */ 2995 static void 2996 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli) 2997 { 2998 struct epoch_tracker et; 2999 struct ifmultiaddr *ifma; 3000 struct ifnet *ifp; 3001 struct in6_multi *inm; 3002 int retval; 3003 3004 IN6_MULTI_LIST_LOCK_ASSERT(); 3005 MLD_LOCK_ASSERT(); 3006 3007 KASSERT(mli->mli_version == MLD_VERSION_2, 3008 ("%s: called when version %d", __func__, mli->mli_version)); 3009 3010 /* 3011 * Check that there are some packets queued. If so, send them first. 3012 * For large number of groups the reply to general query can take 3013 * many packets, we should finish sending them before starting of 3014 * queuing the new reply. 3015 */ 3016 if (mbufq_len(&mli->mli_gq) != 0) 3017 goto send; 3018 3019 ifp = mli->mli_ifp; 3020 3021 NET_EPOCH_ENTER(et); 3022 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3023 inm = in6m_ifmultiaddr_get_inm(ifma); 3024 if (inm == NULL) 3025 continue; 3026 KASSERT(ifp == inm->in6m_ifp, 3027 ("%s: inconsistent ifp", __func__)); 3028 3029 switch (inm->in6m_state) { 3030 case MLD_NOT_MEMBER: 3031 case MLD_SILENT_MEMBER: 3032 break; 3033 case MLD_REPORTING_MEMBER: 3034 case MLD_IDLE_MEMBER: 3035 case MLD_LAZY_MEMBER: 3036 case MLD_SLEEPING_MEMBER: 3037 case MLD_AWAKENING_MEMBER: 3038 inm->in6m_state = MLD_REPORTING_MEMBER; 3039 retval = mld_v2_enqueue_group_record(&mli->mli_gq, 3040 inm, 0, 0, 0, 0); 3041 CTR2(KTR_MLD, "%s: enqueue record = %d", 3042 __func__, retval); 3043 break; 3044 case MLD_G_QUERY_PENDING_MEMBER: 3045 case MLD_SG_QUERY_PENDING_MEMBER: 3046 case MLD_LEAVING_MEMBER: 3047 break; 3048 } 3049 } 3050 NET_EPOCH_EXIT(et); 3051 3052 send: 3053 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); 3054 3055 /* 3056 * Slew transmission of bursts over 500ms intervals. 3057 */ 3058 if (mbufq_first(&mli->mli_gq) != NULL) { 3059 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( 3060 MLD_RESPONSE_BURST_INTERVAL); 3061 V_interface_timers_running6 = 1; 3062 } 3063 } 3064 3065 /* 3066 * Transmit the next pending message in the output queue. 3067 * 3068 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3069 * MRT: Nothing needs to be done, as MLD traffic is always local to 3070 * a link and uses a link-scope multicast address. 3071 */ 3072 static void 3073 mld_dispatch_packet(struct mbuf *m) 3074 { 3075 struct ip6_moptions im6o; 3076 struct ifnet *ifp; 3077 struct ifnet *oifp; 3078 struct mbuf *m0; 3079 struct mbuf *md; 3080 struct ip6_hdr *ip6; 3081 struct mld_hdr *mld; 3082 int error; 3083 int off; 3084 int type; 3085 uint32_t ifindex; 3086 3087 CTR2(KTR_MLD, "%s: transmit %p", __func__, m); 3088 3089 /* 3090 * Set VNET image pointer from enqueued mbuf chain 3091 * before doing anything else. Whilst we use interface 3092 * indexes to guard against interface detach, they are 3093 * unique to each VIMAGE and must be retrieved. 3094 */ 3095 ifindex = mld_restore_context(m); 3096 3097 /* 3098 * Check if the ifnet still exists. This limits the scope of 3099 * any race in the absence of a global ifp lock for low cost 3100 * (an array lookup). 3101 */ 3102 ifp = ifnet_byindex(ifindex); 3103 if (ifp == NULL) { 3104 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", 3105 __func__, m, ifindex); 3106 m_freem(m); 3107 IP6STAT_INC(ip6s_noroute); 3108 goto out; 3109 } 3110 3111 im6o.im6o_multicast_hlim = 1; 3112 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); 3113 im6o.im6o_multicast_ifp = ifp; 3114 3115 if (m->m_flags & M_MLDV1) { 3116 m0 = m; 3117 } else { 3118 m0 = mld_v2_encap_report(ifp, m); 3119 if (m0 == NULL) { 3120 CTR2(KTR_MLD, "%s: dropped %p", __func__, m); 3121 IP6STAT_INC(ip6s_odropped); 3122 goto out; 3123 } 3124 } 3125 3126 mld_scrub_context(m0); 3127 m_clrprotoflags(m); 3128 m0->m_pkthdr.rcvif = V_loif; 3129 3130 ip6 = mtod(m0, struct ip6_hdr *); 3131 #if 0 3132 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */ 3133 #else 3134 /* 3135 * XXX XXX Break some KPI rules to prevent an LOR which would 3136 * occur if we called in6_setscope() at transmission. 3137 * See comments at top of file. 3138 */ 3139 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index); 3140 #endif 3141 3142 /* 3143 * Retrieve the ICMPv6 type before handoff to ip6_output(), 3144 * so we can bump the stats. 3145 */ 3146 md = m_getptr(m0, sizeof(struct ip6_hdr), &off); 3147 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off); 3148 type = mld->mld_type; 3149 3150 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o, 3151 &oifp, NULL); 3152 if (error) { 3153 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error); 3154 goto out; 3155 } 3156 ICMP6STAT_INC(icp6s_outhist[type]); 3157 if (oifp != NULL) { 3158 icmp6_ifstat_inc(oifp, ifs6_out_msg); 3159 switch (type) { 3160 case MLD_LISTENER_REPORT: 3161 case MLDV2_LISTENER_REPORT: 3162 icmp6_ifstat_inc(oifp, ifs6_out_mldreport); 3163 break; 3164 case MLD_LISTENER_DONE: 3165 icmp6_ifstat_inc(oifp, ifs6_out_mlddone); 3166 break; 3167 } 3168 } 3169 out: 3170 return; 3171 } 3172 3173 /* 3174 * Encapsulate an MLDv2 report. 3175 * 3176 * KAME IPv6 requires that hop-by-hop options be passed separately, 3177 * and that the IPv6 header be prepended in a separate mbuf. 3178 * 3179 * Returns a pointer to the new mbuf chain head, or NULL if the 3180 * allocation failed. 3181 */ 3182 static struct mbuf * 3183 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m) 3184 { 3185 struct mbuf *mh; 3186 struct mldv2_report *mld; 3187 struct ip6_hdr *ip6; 3188 struct in6_ifaddr *ia; 3189 int mldreclen; 3190 3191 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 3192 KASSERT((m->m_flags & M_PKTHDR), 3193 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3194 3195 /* 3196 * RFC3590: OK to send as :: or tentative during DAD. 3197 */ 3198 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 3199 if (ia == NULL) 3200 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__); 3201 3202 mh = m_gethdr(M_NOWAIT, MT_DATA); 3203 if (mh == NULL) { 3204 if (ia != NULL) 3205 ifa_free(&ia->ia_ifa); 3206 m_freem(m); 3207 return (NULL); 3208 } 3209 M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); 3210 3211 mldreclen = m_length(m, NULL); 3212 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); 3213 3214 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report); 3215 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + 3216 sizeof(struct mldv2_report) + mldreclen; 3217 3218 ip6 = mtod(mh, struct ip6_hdr *); 3219 ip6->ip6_flow = 0; 3220 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3221 ip6->ip6_vfc |= IPV6_VERSION; 3222 ip6->ip6_nxt = IPPROTO_ICMPV6; 3223 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 3224 if (ia != NULL) 3225 ifa_free(&ia->ia_ifa); 3226 ip6->ip6_dst = in6addr_linklocal_allv2routers; 3227 /* scope ID will be set in netisr */ 3228 3229 mld = (struct mldv2_report *)(ip6 + 1); 3230 mld->mld_type = MLDV2_LISTENER_REPORT; 3231 mld->mld_code = 0; 3232 mld->mld_cksum = 0; 3233 mld->mld_v2_reserved = 0; 3234 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs); 3235 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3236 3237 mh->m_next = m; 3238 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 3239 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen); 3240 return (mh); 3241 } 3242 3243 #ifdef KTR 3244 static char * 3245 mld_rec_type_to_str(const int type) 3246 { 3247 3248 switch (type) { 3249 case MLD_CHANGE_TO_EXCLUDE_MODE: 3250 return "TO_EX"; 3251 break; 3252 case MLD_CHANGE_TO_INCLUDE_MODE: 3253 return "TO_IN"; 3254 break; 3255 case MLD_MODE_IS_EXCLUDE: 3256 return "MODE_EX"; 3257 break; 3258 case MLD_MODE_IS_INCLUDE: 3259 return "MODE_IN"; 3260 break; 3261 case MLD_ALLOW_NEW_SOURCES: 3262 return "ALLOW_NEW"; 3263 break; 3264 case MLD_BLOCK_OLD_SOURCES: 3265 return "BLOCK_OLD"; 3266 break; 3267 default: 3268 break; 3269 } 3270 return "unknown"; 3271 } 3272 #endif 3273 3274 static void 3275 mld_init(void *unused __unused) 3276 { 3277 3278 CTR1(KTR_MLD, "%s: initializing", __func__); 3279 MLD_LOCK_INIT(); 3280 3281 ip6_initpktopts(&mld_po); 3282 mld_po.ip6po_hlim = 1; 3283 mld_po.ip6po_hbh = &mld_ra.hbh; 3284 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER; 3285 mld_po.ip6po_flags = IP6PO_DONTFRAG; 3286 } 3287 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL); 3288 3289 static void 3290 mld_uninit(void *unused __unused) 3291 { 3292 3293 CTR1(KTR_MLD, "%s: tearing down", __func__); 3294 MLD_LOCK_DESTROY(); 3295 } 3296 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL); 3297 3298 static void 3299 vnet_mld_init(const void *unused __unused) 3300 { 3301 3302 CTR1(KTR_MLD, "%s: initializing", __func__); 3303 3304 LIST_INIT(&V_mli_head); 3305 } 3306 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init, 3307 NULL); 3308 3309 static void 3310 vnet_mld_uninit(const void *unused __unused) 3311 { 3312 3313 /* This can happen if we shutdown the network stack. */ 3314 CTR1(KTR_MLD, "%s: tearing down", __func__); 3315 } 3316 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit, 3317 NULL); 3318 3319 static int 3320 mld_modevent(module_t mod, int type, void *unused __unused) 3321 { 3322 3323 switch (type) { 3324 case MOD_LOAD: 3325 case MOD_UNLOAD: 3326 break; 3327 default: 3328 return (EOPNOTSUPP); 3329 } 3330 return (0); 3331 } 3332 3333 static moduledata_t mld_mod = { 3334 "mld", 3335 mld_modevent, 3336 0 3337 }; 3338 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY); 3339