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