1 /*- 2 * Copyright (c) 2009 Bruce Simpson. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 3. The name of the author may not be used to endorse or promote 13 * products derived from this software without specific prior written 14 * permission. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 * 28 * $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $ 29 */ 30 31 /*- 32 * Copyright (c) 1988 Stephen Deering. 33 * Copyright (c) 1992, 1993 34 * The Regents of the University of California. All rights reserved. 35 * 36 * This code is derived from software contributed to Berkeley by 37 * Stephen Deering of Stanford University. 38 * 39 * Redistribution and use in source and binary forms, with or without 40 * modification, are permitted provided that the following conditions 41 * are met: 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 4. Neither the name of the University nor the names of its contributors 48 * may be used to endorse or promote products derived from this software 49 * without specific prior written permission. 50 * 51 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 52 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 53 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 54 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 55 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 56 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 57 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 58 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 59 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 60 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 61 * SUCH DAMAGE. 62 * 63 * @(#)igmp.c 8.1 (Berkeley) 7/19/93 64 */ 65 66 #include <sys/cdefs.h> 67 __FBSDID("$FreeBSD$"); 68 69 #include "opt_inet.h" 70 #include "opt_inet6.h" 71 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/mbuf.h> 75 #include <sys/socket.h> 76 #include <sys/protosw.h> 77 #include <sys/sysctl.h> 78 #include <sys/kernel.h> 79 #include <sys/callout.h> 80 #include <sys/malloc.h> 81 #include <sys/module.h> 82 #include <sys/ktr.h> 83 84 #include <net/if.h> 85 #include <net/if_var.h> 86 #include <net/route.h> 87 #include <net/vnet.h> 88 89 #include <netinet/in.h> 90 #include <netinet/in_var.h> 91 #include <netinet6/in6_var.h> 92 #include <netinet/ip6.h> 93 #include <netinet6/ip6_var.h> 94 #include <netinet6/scope6_var.h> 95 #include <netinet/icmp6.h> 96 #include <netinet6/mld6.h> 97 #include <netinet6/mld6_var.h> 98 99 #include <security/mac/mac_framework.h> 100 101 #ifndef KTR_MLD 102 #define KTR_MLD KTR_INET6 103 #endif 104 105 static struct mld_ifinfo * 106 mli_alloc_locked(struct ifnet *); 107 static void mli_delete_locked(const struct ifnet *); 108 static void mld_dispatch_packet(struct mbuf *); 109 static void mld_dispatch_queue(struct ifqueue *, int); 110 static void mld_final_leave(struct in6_multi *, struct mld_ifinfo *); 111 static void mld_fasttimo_vnet(void); 112 static int mld_handle_state_change(struct in6_multi *, 113 struct mld_ifinfo *); 114 static int mld_initial_join(struct in6_multi *, struct mld_ifinfo *, 115 const int); 116 #ifdef KTR 117 static char * mld_rec_type_to_str(const int); 118 #endif 119 static void mld_set_version(struct mld_ifinfo *, const int); 120 static void mld_slowtimo_vnet(void); 121 static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *, 122 /*const*/ struct mld_hdr *); 123 static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *, 124 /*const*/ struct mld_hdr *); 125 static void mld_v1_process_group_timer(struct mld_ifinfo *, 126 struct in6_multi *); 127 static void mld_v1_process_querier_timers(struct mld_ifinfo *); 128 static int mld_v1_transmit_report(struct in6_multi *, const int); 129 static void mld_v1_update_group(struct in6_multi *, const int); 130 static void mld_v2_cancel_link_timers(struct mld_ifinfo *); 131 static void mld_v2_dispatch_general_query(struct mld_ifinfo *); 132 static struct mbuf * 133 mld_v2_encap_report(struct ifnet *, struct mbuf *); 134 static int mld_v2_enqueue_filter_change(struct ifqueue *, 135 struct in6_multi *); 136 static int mld_v2_enqueue_group_record(struct ifqueue *, 137 struct in6_multi *, const int, const int, const int, 138 const int); 139 static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *, 140 struct mbuf *, const int, const int); 141 static int mld_v2_merge_state_changes(struct in6_multi *, 142 struct ifqueue *); 143 static void mld_v2_process_group_timers(struct mld_ifinfo *, 144 struct ifqueue *, struct ifqueue *, 145 struct in6_multi *, const int); 146 static int mld_v2_process_group_query(struct in6_multi *, 147 struct mld_ifinfo *mli, int, struct mbuf *, const int); 148 static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); 149 static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS); 150 151 /* 152 * Normative references: RFC 2710, RFC 3590, RFC 3810. 153 * 154 * Locking: 155 * * The MLD subsystem lock ends up being system-wide for the moment, 156 * but could be per-VIMAGE later on. 157 * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. 158 * Any may be taken independently; if any are held at the same 159 * time, the above lock order must be followed. 160 * * IN6_MULTI_LOCK covers in_multi. 161 * * MLD_LOCK covers per-link state and any global variables in this file. 162 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 163 * per-link state iterators. 164 * 165 * XXX LOR PREVENTION 166 * A special case for IPv6 is the in6_setscope() routine. ip6_output() 167 * will not accept an ifp; it wants an embedded scope ID, unlike 168 * ip_output(), which happily takes the ifp given to it. The embedded 169 * scope ID is only used by MLD to select the outgoing interface. 170 * 171 * During interface attach and detach, MLD will take MLD_LOCK *after* 172 * the IF_AFDATA_LOCK. 173 * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call 174 * it with MLD_LOCK held without triggering an LOR. A netisr with indirect 175 * dispatch could work around this, but we'd rather not do that, as it 176 * can introduce other races. 177 * 178 * As such, we exploit the fact that the scope ID is just the interface 179 * index, and embed it in the IPv6 destination address accordingly. 180 * This is potentially NOT VALID for MLDv1 reports, as they 181 * are always sent to the multicast group itself; as MLDv2 182 * reports are always sent to ff02::16, this is not an issue 183 * when MLDv2 is in use. 184 * 185 * This does not however eliminate the LOR when ip6_output() itself 186 * calls in6_setscope() internally whilst MLD_LOCK is held. This will 187 * trigger a LOR warning in WITNESS when the ifnet is detached. 188 * 189 * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given 190 * how it's used across the network stack. Here we're simply exploiting 191 * the fact that MLD runs at a similar layer in the stack to scope6.c. 192 * 193 * VIMAGE: 194 * * Each in6_multi corresponds to an ifp, and each ifp corresponds 195 * to a vnet in ifp->if_vnet. 196 */ 197 static struct mtx mld_mtx; 198 static MALLOC_DEFINE(M_MLD, "mld", "mld state"); 199 200 #define MLD_EMBEDSCOPE(pin6, zoneid) \ 201 if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \ 202 IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \ 203 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \ 204 205 /* 206 * VIMAGE-wide globals. 207 */ 208 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0}; 209 static VNET_DEFINE(LIST_HEAD(, mld_ifinfo), mli_head); 210 static VNET_DEFINE(int, interface_timers_running6); 211 static VNET_DEFINE(int, state_change_timers_running6); 212 static VNET_DEFINE(int, current_state_timers_running6); 213 214 #define V_mld_gsrdelay VNET(mld_gsrdelay) 215 #define V_mli_head VNET(mli_head) 216 #define V_interface_timers_running6 VNET(interface_timers_running6) 217 #define V_state_change_timers_running6 VNET(state_change_timers_running6) 218 #define V_current_state_timers_running6 VNET(current_state_timers_running6) 219 220 SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ 221 222 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, 223 "IPv6 Multicast Listener Discovery"); 224 225 /* 226 * Virtualized sysctls. 227 */ 228 SYSCTL_VNET_PROC(_net_inet6_mld, OID_AUTO, gsrdelay, 229 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 230 &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I", 231 "Rate limit for MLDv2 Group-and-Source queries in seconds"); 232 233 /* 234 * Non-virtualized sysctls. 235 */ 236 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, 237 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo, 238 "Per-interface MLDv2 state"); 239 240 static int mld_v1enable = 1; 241 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW, 242 &mld_v1enable, 0, "Enable fallback to MLDv1"); 243 TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable); 244 245 static int mld_use_allow = 1; 246 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW, 247 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves"); 248 TUNABLE_INT("net.inet6.mld.use_allow", &mld_use_allow); 249 250 /* 251 * Packed Router Alert option structure declaration. 252 */ 253 struct mld_raopt { 254 struct ip6_hbh hbh; 255 struct ip6_opt pad; 256 struct ip6_opt_router ra; 257 } __packed; 258 259 /* 260 * Router Alert hop-by-hop option header. 261 */ 262 static struct mld_raopt mld_ra = { 263 .hbh = { 0, 0 }, 264 .pad = { .ip6o_type = IP6OPT_PADN, 0 }, 265 .ra = { 266 .ip6or_type = IP6OPT_ROUTER_ALERT, 267 .ip6or_len = IP6OPT_RTALERT_LEN - 2, 268 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), 269 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) 270 } 271 }; 272 static struct ip6_pktopts mld_po; 273 274 static __inline void 275 mld_save_context(struct mbuf *m, struct ifnet *ifp) 276 { 277 278 #ifdef VIMAGE 279 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; 280 #endif /* VIMAGE */ 281 m->m_pkthdr.flowid = ifp->if_index; 282 } 283 284 static __inline void 285 mld_scrub_context(struct mbuf *m) 286 { 287 288 m->m_pkthdr.PH_loc.ptr = NULL; 289 m->m_pkthdr.flowid = 0; 290 } 291 292 /* 293 * Restore context from a queued output chain. 294 * Return saved ifindex. 295 * 296 * VIMAGE: The assertion is there to make sure that we 297 * actually called CURVNET_SET() with what's in the mbuf chain. 298 */ 299 static __inline uint32_t 300 mld_restore_context(struct mbuf *m) 301 { 302 303 #if defined(VIMAGE) && defined(INVARIANTS) 304 KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr, 305 ("%s: called when curvnet was not restored", __func__)); 306 #endif 307 return (m->m_pkthdr.flowid); 308 } 309 310 /* 311 * Retrieve or set threshold between group-source queries in seconds. 312 * 313 * VIMAGE: Assume curvnet set by caller. 314 * SMPng: NOTE: Serialized by MLD lock. 315 */ 316 static int 317 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS) 318 { 319 int error; 320 int i; 321 322 error = sysctl_wire_old_buffer(req, sizeof(int)); 323 if (error) 324 return (error); 325 326 MLD_LOCK(); 327 328 i = V_mld_gsrdelay.tv_sec; 329 330 error = sysctl_handle_int(oidp, &i, 0, req); 331 if (error || !req->newptr) 332 goto out_locked; 333 334 if (i < -1 || i >= 60) { 335 error = EINVAL; 336 goto out_locked; 337 } 338 339 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", 340 V_mld_gsrdelay.tv_sec, i); 341 V_mld_gsrdelay.tv_sec = i; 342 343 out_locked: 344 MLD_UNLOCK(); 345 return (error); 346 } 347 348 /* 349 * Expose struct mld_ifinfo to userland, keyed by ifindex. 350 * For use by ifmcstat(8). 351 * 352 * SMPng: NOTE: Does an unlocked ifindex space read. 353 * VIMAGE: Assume curvnet set by caller. The node handler itself 354 * is not directly virtualized. 355 */ 356 static int 357 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) 358 { 359 int *name; 360 int error; 361 u_int namelen; 362 struct ifnet *ifp; 363 struct mld_ifinfo *mli; 364 365 name = (int *)arg1; 366 namelen = arg2; 367 368 if (req->newptr != NULL) 369 return (EPERM); 370 371 if (namelen != 1) 372 return (EINVAL); 373 374 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo)); 375 if (error) 376 return (error); 377 378 IN6_MULTI_LOCK(); 379 MLD_LOCK(); 380 381 if (name[0] <= 0 || name[0] > V_if_index) { 382 error = ENOENT; 383 goto out_locked; 384 } 385 386 error = ENOENT; 387 388 ifp = ifnet_byindex(name[0]); 389 if (ifp == NULL) 390 goto out_locked; 391 392 LIST_FOREACH(mli, &V_mli_head, mli_link) { 393 if (ifp == mli->mli_ifp) { 394 error = SYSCTL_OUT(req, mli, 395 sizeof(struct mld_ifinfo)); 396 break; 397 } 398 } 399 400 out_locked: 401 MLD_UNLOCK(); 402 IN6_MULTI_UNLOCK(); 403 return (error); 404 } 405 406 /* 407 * Dispatch an entire queue of pending packet chains. 408 * VIMAGE: Assumes the vnet pointer has been set. 409 */ 410 static void 411 mld_dispatch_queue(struct ifqueue *ifq, int limit) 412 { 413 struct mbuf *m; 414 415 for (;;) { 416 _IF_DEQUEUE(ifq, m); 417 if (m == NULL) 418 break; 419 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, ifq, m); 420 mld_dispatch_packet(m); 421 if (--limit == 0) 422 break; 423 } 424 } 425 426 /* 427 * Filter outgoing MLD report state by group. 428 * 429 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1) 430 * and node-local addresses. However, kernel and socket consumers 431 * always embed the KAME scope ID in the address provided, so strip it 432 * when performing comparison. 433 * Note: This is not the same as the *multicast* scope. 434 * 435 * Return zero if the given group is one for which MLD reports 436 * should be suppressed, or non-zero if reports should be issued. 437 */ 438 static __inline int 439 mld_is_addr_reported(const struct in6_addr *addr) 440 { 441 442 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__)); 443 444 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) 445 return (0); 446 447 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) { 448 struct in6_addr tmp = *addr; 449 in6_clearscope(&tmp); 450 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) 451 return (0); 452 } 453 454 return (1); 455 } 456 457 /* 458 * Attach MLD when PF_INET6 is attached to an interface. 459 * 460 * SMPng: Normally called with IF_AFDATA_LOCK held. 461 */ 462 struct mld_ifinfo * 463 mld_domifattach(struct ifnet *ifp) 464 { 465 struct mld_ifinfo *mli; 466 467 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 468 __func__, ifp, if_name(ifp)); 469 470 MLD_LOCK(); 471 472 mli = mli_alloc_locked(ifp); 473 if (!(ifp->if_flags & IFF_MULTICAST)) 474 mli->mli_flags |= MLIF_SILENT; 475 if (mld_use_allow) 476 mli->mli_flags |= MLIF_USEALLOW; 477 478 MLD_UNLOCK(); 479 480 return (mli); 481 } 482 483 /* 484 * VIMAGE: assume curvnet set by caller. 485 */ 486 static struct mld_ifinfo * 487 mli_alloc_locked(/*const*/ struct ifnet *ifp) 488 { 489 struct mld_ifinfo *mli; 490 491 MLD_LOCK_ASSERT(); 492 493 mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO); 494 if (mli == NULL) 495 goto out; 496 497 mli->mli_ifp = ifp; 498 mli->mli_version = MLD_VERSION_2; 499 mli->mli_flags = 0; 500 mli->mli_rv = MLD_RV_INIT; 501 mli->mli_qi = MLD_QI_INIT; 502 mli->mli_qri = MLD_QRI_INIT; 503 mli->mli_uri = MLD_URI_INIT; 504 505 SLIST_INIT(&mli->mli_relinmhead); 506 507 /* 508 * Responses to general queries are subject to bounds. 509 */ 510 IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); 511 512 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); 513 514 CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)", 515 ifp, if_name(ifp)); 516 517 out: 518 return (mli); 519 } 520 521 /* 522 * Hook for ifdetach. 523 * 524 * NOTE: Some finalization tasks need to run before the protocol domain 525 * is detached, but also before the link layer does its cleanup. 526 * Run before link-layer cleanup; cleanup groups, but do not free MLD state. 527 * 528 * SMPng: Caller must hold IN6_MULTI_LOCK(). 529 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. 530 * XXX This routine is also bitten by unlocked ifma_protospec access. 531 */ 532 void 533 mld_ifdetach(struct ifnet *ifp) 534 { 535 struct mld_ifinfo *mli; 536 struct ifmultiaddr *ifma; 537 struct in6_multi *inm, *tinm; 538 539 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, 540 if_name(ifp)); 541 542 IN6_MULTI_LOCK_ASSERT(); 543 MLD_LOCK(); 544 545 mli = MLD_IFINFO(ifp); 546 if (mli->mli_version == MLD_VERSION_2) { 547 IF_ADDR_RLOCK(ifp); 548 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 549 if (ifma->ifma_addr->sa_family != AF_INET6 || 550 ifma->ifma_protospec == NULL) 551 continue; 552 inm = (struct in6_multi *)ifma->ifma_protospec; 553 if (inm->in6m_state == MLD_LEAVING_MEMBER) { 554 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 555 inm, in6m_nrele); 556 } 557 in6m_clear_recorded(inm); 558 } 559 IF_ADDR_RUNLOCK(ifp); 560 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, 561 tinm) { 562 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 563 in6m_release_locked(inm); 564 } 565 } 566 567 MLD_UNLOCK(); 568 } 569 570 /* 571 * Hook for domifdetach. 572 * Runs after link-layer cleanup; free MLD state. 573 * 574 * SMPng: Normally called with IF_AFDATA_LOCK held. 575 */ 576 void 577 mld_domifdetach(struct ifnet *ifp) 578 { 579 580 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 581 __func__, ifp, if_name(ifp)); 582 583 MLD_LOCK(); 584 mli_delete_locked(ifp); 585 MLD_UNLOCK(); 586 } 587 588 static void 589 mli_delete_locked(const struct ifnet *ifp) 590 { 591 struct mld_ifinfo *mli, *tmli; 592 593 CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)", 594 __func__, ifp, if_name(ifp)); 595 596 MLD_LOCK_ASSERT(); 597 598 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) { 599 if (mli->mli_ifp == ifp) { 600 /* 601 * Free deferred General Query responses. 602 */ 603 _IF_DRAIN(&mli->mli_gq); 604 605 LIST_REMOVE(mli, mli_link); 606 607 KASSERT(SLIST_EMPTY(&mli->mli_relinmhead), 608 ("%s: there are dangling in_multi references", 609 __func__)); 610 611 free(mli, M_MLD); 612 return; 613 } 614 } 615 #ifdef INVARIANTS 616 panic("%s: mld_ifinfo not found for ifp %p\n", __func__, ifp); 617 #endif 618 } 619 620 /* 621 * Process a received MLDv1 general or address-specific query. 622 * Assumes that the query header has been pulled up to sizeof(mld_hdr). 623 * 624 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 625 * mld_addr. This is OK as we own the mbuf chain. 626 */ 627 static int 628 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 629 /*const*/ struct mld_hdr *mld) 630 { 631 struct ifmultiaddr *ifma; 632 struct mld_ifinfo *mli; 633 struct in6_multi *inm; 634 int is_general_query; 635 uint16_t timer; 636 #ifdef KTR 637 char ip6tbuf[INET6_ADDRSTRLEN]; 638 #endif 639 640 is_general_query = 0; 641 642 if (!mld_v1enable) { 643 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)", 644 ip6_sprintf(ip6tbuf, &mld->mld_addr), 645 ifp, if_name(ifp)); 646 return (0); 647 } 648 649 /* 650 * RFC3810 Section 6.2: MLD queries must originate from 651 * a router's link-local address. 652 */ 653 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 654 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 655 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 656 ifp, if_name(ifp)); 657 return (0); 658 } 659 660 /* 661 * Do address field validation upfront before we accept 662 * the query. 663 */ 664 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 665 /* 666 * MLDv1 General Query. 667 * If this was not sent to the all-nodes group, ignore it. 668 */ 669 struct in6_addr dst; 670 671 dst = ip6->ip6_dst; 672 in6_clearscope(&dst); 673 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) 674 return (EINVAL); 675 is_general_query = 1; 676 } else { 677 /* 678 * Embed scope ID of receiving interface in MLD query for 679 * lookup whilst we don't hold other locks. 680 */ 681 in6_setscope(&mld->mld_addr, ifp, NULL); 682 } 683 684 IN6_MULTI_LOCK(); 685 MLD_LOCK(); 686 687 /* 688 * Switch to MLDv1 host compatibility mode. 689 */ 690 mli = MLD_IFINFO(ifp); 691 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 692 mld_set_version(mli, MLD_VERSION_1); 693 694 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; 695 if (timer == 0) 696 timer = 1; 697 698 IF_ADDR_RLOCK(ifp); 699 if (is_general_query) { 700 /* 701 * For each reporting group joined on this 702 * interface, kick the report timer. 703 */ 704 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)", 705 ifp, if_name(ifp)); 706 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 707 if (ifma->ifma_addr->sa_family != AF_INET6 || 708 ifma->ifma_protospec == NULL) 709 continue; 710 inm = (struct in6_multi *)ifma->ifma_protospec; 711 mld_v1_update_group(inm, timer); 712 } 713 } else { 714 /* 715 * MLDv1 Group-Specific Query. 716 * If this is a group-specific MLDv1 query, we need only 717 * look up the single group to process it. 718 */ 719 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 720 if (inm != NULL) { 721 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)", 722 ip6_sprintf(ip6tbuf, &mld->mld_addr), 723 ifp, if_name(ifp)); 724 mld_v1_update_group(inm, timer); 725 } 726 /* XXX Clear embedded scope ID as userland won't expect it. */ 727 in6_clearscope(&mld->mld_addr); 728 } 729 730 IF_ADDR_RUNLOCK(ifp); 731 MLD_UNLOCK(); 732 IN6_MULTI_UNLOCK(); 733 734 return (0); 735 } 736 737 /* 738 * Update the report timer on a group in response to an MLDv1 query. 739 * 740 * If we are becoming the reporting member for this group, start the timer. 741 * If we already are the reporting member for this group, and timer is 742 * below the threshold, reset it. 743 * 744 * We may be updating the group for the first time since we switched 745 * to MLDv2. If we are, then we must clear any recorded source lists, 746 * and transition to REPORTING state; the group timer is overloaded 747 * for group and group-source query responses. 748 * 749 * Unlike MLDv2, the delay per group should be jittered 750 * to avoid bursts of MLDv1 reports. 751 */ 752 static void 753 mld_v1_update_group(struct in6_multi *inm, const int timer) 754 { 755 #ifdef KTR 756 char ip6tbuf[INET6_ADDRSTRLEN]; 757 #endif 758 759 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, 760 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 761 if_name(inm->in6m_ifp), timer); 762 763 IN6_MULTI_LOCK_ASSERT(); 764 765 switch (inm->in6m_state) { 766 case MLD_NOT_MEMBER: 767 case MLD_SILENT_MEMBER: 768 break; 769 case MLD_REPORTING_MEMBER: 770 if (inm->in6m_timer != 0 && 771 inm->in6m_timer <= timer) { 772 CTR1(KTR_MLD, "%s: REPORTING and timer running, " 773 "skipping.", __func__); 774 break; 775 } 776 /* FALLTHROUGH */ 777 case MLD_SG_QUERY_PENDING_MEMBER: 778 case MLD_G_QUERY_PENDING_MEMBER: 779 case MLD_IDLE_MEMBER: 780 case MLD_LAZY_MEMBER: 781 case MLD_AWAKENING_MEMBER: 782 CTR1(KTR_MLD, "%s: ->REPORTING", __func__); 783 inm->in6m_state = MLD_REPORTING_MEMBER; 784 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 785 V_current_state_timers_running6 = 1; 786 break; 787 case MLD_SLEEPING_MEMBER: 788 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); 789 inm->in6m_state = MLD_AWAKENING_MEMBER; 790 break; 791 case MLD_LEAVING_MEMBER: 792 break; 793 } 794 } 795 796 /* 797 * Process a received MLDv2 general, group-specific or 798 * group-and-source-specific query. 799 * 800 * Assumes that the query header has been pulled up to sizeof(mldv2_query). 801 * 802 * Return 0 if successful, otherwise an appropriate error code is returned. 803 */ 804 static int 805 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 806 struct mbuf *m, const int off, const int icmp6len) 807 { 808 struct mld_ifinfo *mli; 809 struct mldv2_query *mld; 810 struct in6_multi *inm; 811 uint32_t maxdelay, nsrc, qqi; 812 int is_general_query; 813 uint16_t timer; 814 uint8_t qrv; 815 #ifdef KTR 816 char ip6tbuf[INET6_ADDRSTRLEN]; 817 #endif 818 819 is_general_query = 0; 820 821 /* 822 * RFC3810 Section 6.2: MLD queries must originate from 823 * a router's link-local address. 824 */ 825 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 826 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 827 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 828 ifp, if_name(ifp)); 829 return (0); 830 } 831 832 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp)); 833 834 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off); 835 836 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ 837 if (maxdelay >= 32768) { 838 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << 839 (MLD_MRC_EXP(maxdelay) + 3); 840 } 841 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; 842 if (timer == 0) 843 timer = 1; 844 845 qrv = MLD_QRV(mld->mld_misc); 846 if (qrv < 2) { 847 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, 848 qrv, MLD_RV_INIT); 849 qrv = MLD_RV_INIT; 850 } 851 852 qqi = mld->mld_qqi; 853 if (qqi >= 128) { 854 qqi = MLD_QQIC_MANT(mld->mld_qqi) << 855 (MLD_QQIC_EXP(mld->mld_qqi) + 3); 856 } 857 858 nsrc = ntohs(mld->mld_numsrc); 859 if (nsrc > MLD_MAX_GS_SOURCES) 860 return (EMSGSIZE); 861 if (icmp6len < sizeof(struct mldv2_query) + 862 (nsrc * sizeof(struct in6_addr))) 863 return (EMSGSIZE); 864 865 /* 866 * Do further input validation upfront to avoid resetting timers 867 * should we need to discard this query. 868 */ 869 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 870 /* 871 * A general query with a source list has undefined 872 * behaviour; discard it. 873 */ 874 if (nsrc > 0) 875 return (EINVAL); 876 is_general_query = 1; 877 } else { 878 /* 879 * Embed scope ID of receiving interface in MLD query for 880 * lookup whilst we don't hold other locks (due to KAME 881 * locking lameness). We own this mbuf chain just now. 882 */ 883 in6_setscope(&mld->mld_addr, ifp, NULL); 884 } 885 886 IN6_MULTI_LOCK(); 887 MLD_LOCK(); 888 889 mli = MLD_IFINFO(ifp); 890 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 891 892 /* 893 * Discard the v2 query if we're in Compatibility Mode. 894 * The RFC is pretty clear that hosts need to stay in MLDv1 mode 895 * until the Old Version Querier Present timer expires. 896 */ 897 if (mli->mli_version != MLD_VERSION_2) 898 goto out_locked; 899 900 mld_set_version(mli, MLD_VERSION_2); 901 mli->mli_rv = qrv; 902 mli->mli_qi = qqi; 903 mli->mli_qri = maxdelay; 904 905 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, 906 maxdelay); 907 908 if (is_general_query) { 909 /* 910 * MLDv2 General Query. 911 * 912 * Schedule a current-state report on this ifp for 913 * all groups, possibly containing source lists. 914 * 915 * If there is a pending General Query response 916 * scheduled earlier than the selected delay, do 917 * not schedule any other reports. 918 * Otherwise, reset the interface timer. 919 */ 920 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", 921 ifp, if_name(ifp)); 922 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { 923 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); 924 V_interface_timers_running6 = 1; 925 } 926 } else { 927 /* 928 * MLDv2 Group-specific or Group-and-source-specific Query. 929 * 930 * Group-source-specific queries are throttled on 931 * a per-group basis to defeat denial-of-service attempts. 932 * Queries for groups we are not a member of on this 933 * link are simply ignored. 934 */ 935 IF_ADDR_RLOCK(ifp); 936 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 937 if (inm == NULL) { 938 IF_ADDR_RUNLOCK(ifp); 939 goto out_locked; 940 } 941 if (nsrc > 0) { 942 if (!ratecheck(&inm->in6m_lastgsrtv, 943 &V_mld_gsrdelay)) { 944 CTR1(KTR_MLD, "%s: GS query throttled.", 945 __func__); 946 IF_ADDR_RUNLOCK(ifp); 947 goto out_locked; 948 } 949 } 950 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", 951 ifp, if_name(ifp)); 952 /* 953 * If there is a pending General Query response 954 * scheduled sooner than the selected delay, no 955 * further report need be scheduled. 956 * Otherwise, prepare to respond to the 957 * group-specific or group-and-source query. 958 */ 959 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) 960 mld_v2_process_group_query(inm, mli, timer, m, off); 961 962 /* XXX Clear embedded scope ID as userland won't expect it. */ 963 in6_clearscope(&mld->mld_addr); 964 IF_ADDR_RUNLOCK(ifp); 965 } 966 967 out_locked: 968 MLD_UNLOCK(); 969 IN6_MULTI_UNLOCK(); 970 971 return (0); 972 } 973 974 /* 975 * Process a recieved MLDv2 group-specific or group-and-source-specific 976 * query. 977 * Return <0 if any error occured. Currently this is ignored. 978 */ 979 static int 980 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli, 981 int timer, struct mbuf *m0, const int off) 982 { 983 struct mldv2_query *mld; 984 int retval; 985 uint16_t nsrc; 986 987 IN6_MULTI_LOCK_ASSERT(); 988 MLD_LOCK_ASSERT(); 989 990 retval = 0; 991 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off); 992 993 switch (inm->in6m_state) { 994 case MLD_NOT_MEMBER: 995 case MLD_SILENT_MEMBER: 996 case MLD_SLEEPING_MEMBER: 997 case MLD_LAZY_MEMBER: 998 case MLD_AWAKENING_MEMBER: 999 case MLD_IDLE_MEMBER: 1000 case MLD_LEAVING_MEMBER: 1001 return (retval); 1002 break; 1003 case MLD_REPORTING_MEMBER: 1004 case MLD_G_QUERY_PENDING_MEMBER: 1005 case MLD_SG_QUERY_PENDING_MEMBER: 1006 break; 1007 } 1008 1009 nsrc = ntohs(mld->mld_numsrc); 1010 1011 /* 1012 * Deal with group-specific queries upfront. 1013 * If any group query is already pending, purge any recorded 1014 * source-list state if it exists, and schedule a query response 1015 * for this group-specific query. 1016 */ 1017 if (nsrc == 0) { 1018 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 1019 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { 1020 in6m_clear_recorded(inm); 1021 timer = min(inm->in6m_timer, timer); 1022 } 1023 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; 1024 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1025 V_current_state_timers_running6 = 1; 1026 return (retval); 1027 } 1028 1029 /* 1030 * Deal with the case where a group-and-source-specific query has 1031 * been received but a group-specific query is already pending. 1032 */ 1033 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { 1034 timer = min(inm->in6m_timer, timer); 1035 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1036 V_current_state_timers_running6 = 1; 1037 return (retval); 1038 } 1039 1040 /* 1041 * Finally, deal with the case where a group-and-source-specific 1042 * query has been received, where a response to a previous g-s-r 1043 * query exists, or none exists. 1044 * In this case, we need to parse the source-list which the Querier 1045 * has provided us with and check if we have any source list filter 1046 * entries at T1 for these sources. If we do not, there is no need 1047 * schedule a report and the query may be dropped. 1048 * If we do, we must record them and schedule a current-state 1049 * report for those sources. 1050 */ 1051 if (inm->in6m_nsrc > 0) { 1052 struct mbuf *m; 1053 uint8_t *sp; 1054 int i, nrecorded; 1055 int soff; 1056 1057 m = m0; 1058 soff = off + sizeof(struct mldv2_query); 1059 nrecorded = 0; 1060 for (i = 0; i < nsrc; i++) { 1061 sp = mtod(m, uint8_t *) + soff; 1062 retval = in6m_record_source(inm, 1063 (const struct in6_addr *)sp); 1064 if (retval < 0) 1065 break; 1066 nrecorded += retval; 1067 soff += sizeof(struct in6_addr); 1068 if (soff >= m->m_len) { 1069 soff = soff - m->m_len; 1070 m = m->m_next; 1071 if (m == NULL) 1072 break; 1073 } 1074 } 1075 if (nrecorded > 0) { 1076 CTR1(KTR_MLD, 1077 "%s: schedule response to SG query", __func__); 1078 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; 1079 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1080 V_current_state_timers_running6 = 1; 1081 } 1082 } 1083 1084 return (retval); 1085 } 1086 1087 /* 1088 * Process a received MLDv1 host membership report. 1089 * Assumes mld points to mld_hdr in pulled up mbuf chain. 1090 * 1091 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 1092 * mld_addr. This is OK as we own the mbuf chain. 1093 */ 1094 static int 1095 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, 1096 /*const*/ struct mld_hdr *mld) 1097 { 1098 struct in6_addr src, dst; 1099 struct in6_ifaddr *ia; 1100 struct in6_multi *inm; 1101 #ifdef KTR 1102 char ip6tbuf[INET6_ADDRSTRLEN]; 1103 #endif 1104 1105 if (!mld_v1enable) { 1106 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", 1107 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1108 ifp, if_name(ifp)); 1109 return (0); 1110 } 1111 1112 if (ifp->if_flags & IFF_LOOPBACK) 1113 return (0); 1114 1115 /* 1116 * MLDv1 reports must originate from a host's link-local address, 1117 * or the unspecified address (when booting). 1118 */ 1119 src = ip6->ip6_src; 1120 in6_clearscope(&src); 1121 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { 1122 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 1123 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 1124 ifp, if_name(ifp)); 1125 return (EINVAL); 1126 } 1127 1128 /* 1129 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast 1130 * group, and must be directed to the group itself. 1131 */ 1132 dst = ip6->ip6_dst; 1133 in6_clearscope(&dst); 1134 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || 1135 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { 1136 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", 1137 ip6_sprintf(ip6tbuf, &ip6->ip6_dst), 1138 ifp, if_name(ifp)); 1139 return (EINVAL); 1140 } 1141 1142 /* 1143 * Make sure we don't hear our own membership report, as fast 1144 * leave requires knowing that we are the only member of a 1145 * group. Assume we used the link-local address if available, 1146 * otherwise look for ::. 1147 * 1148 * XXX Note that scope ID comparison is needed for the address 1149 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be 1150 * performed for the on-wire address. 1151 */ 1152 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1153 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || 1154 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) { 1155 if (ia != NULL) 1156 ifa_free(&ia->ia_ifa); 1157 return (0); 1158 } 1159 if (ia != NULL) 1160 ifa_free(&ia->ia_ifa); 1161 1162 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", 1163 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); 1164 1165 /* 1166 * Embed scope ID of receiving interface in MLD query for lookup 1167 * whilst we don't hold other locks (due to KAME locking lameness). 1168 */ 1169 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) 1170 in6_setscope(&mld->mld_addr, ifp, NULL); 1171 1172 IN6_MULTI_LOCK(); 1173 MLD_LOCK(); 1174 IF_ADDR_RLOCK(ifp); 1175 1176 /* 1177 * MLDv1 report suppression. 1178 * If we are a member of this group, and our membership should be 1179 * reported, and our group timer is pending or about to be reset, 1180 * stop our group timer by transitioning to the 'lazy' state. 1181 */ 1182 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 1183 if (inm != NULL) { 1184 struct mld_ifinfo *mli; 1185 1186 mli = inm->in6m_mli; 1187 KASSERT(mli != NULL, 1188 ("%s: no mli for ifp %p", __func__, ifp)); 1189 1190 /* 1191 * If we are in MLDv2 host mode, do not allow the 1192 * other host's MLDv1 report to suppress our reports. 1193 */ 1194 if (mli->mli_version == MLD_VERSION_2) 1195 goto out_locked; 1196 1197 inm->in6m_timer = 0; 1198 1199 switch (inm->in6m_state) { 1200 case MLD_NOT_MEMBER: 1201 case MLD_SILENT_MEMBER: 1202 case MLD_SLEEPING_MEMBER: 1203 break; 1204 case MLD_REPORTING_MEMBER: 1205 case MLD_IDLE_MEMBER: 1206 case MLD_AWAKENING_MEMBER: 1207 CTR3(KTR_MLD, 1208 "report suppressed for %s on ifp %p(%s)", 1209 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1210 ifp, if_name(ifp)); 1211 case MLD_LAZY_MEMBER: 1212 inm->in6m_state = MLD_LAZY_MEMBER; 1213 break; 1214 case MLD_G_QUERY_PENDING_MEMBER: 1215 case MLD_SG_QUERY_PENDING_MEMBER: 1216 case MLD_LEAVING_MEMBER: 1217 break; 1218 } 1219 } 1220 1221 out_locked: 1222 IF_ADDR_RUNLOCK(ifp); 1223 MLD_UNLOCK(); 1224 IN6_MULTI_UNLOCK(); 1225 1226 /* XXX Clear embedded scope ID as userland won't expect it. */ 1227 in6_clearscope(&mld->mld_addr); 1228 1229 return (0); 1230 } 1231 1232 /* 1233 * MLD input path. 1234 * 1235 * Assume query messages which fit in a single ICMPv6 message header 1236 * have been pulled up. 1237 * Assume that userland will want to see the message, even if it 1238 * otherwise fails kernel input validation; do not free it. 1239 * Pullup may however free the mbuf chain m if it fails. 1240 * 1241 * Return IPPROTO_DONE if we freed m. Otherwise, return 0. 1242 */ 1243 int 1244 mld_input(struct mbuf *m, int off, int icmp6len) 1245 { 1246 struct ifnet *ifp; 1247 struct ip6_hdr *ip6; 1248 struct mld_hdr *mld; 1249 int mldlen; 1250 1251 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); 1252 1253 ifp = m->m_pkthdr.rcvif; 1254 1255 ip6 = mtod(m, struct ip6_hdr *); 1256 1257 /* Pullup to appropriate size. */ 1258 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); 1259 if (mld->mld_type == MLD_LISTENER_QUERY && 1260 icmp6len >= sizeof(struct mldv2_query)) { 1261 mldlen = sizeof(struct mldv2_query); 1262 } else { 1263 mldlen = sizeof(struct mld_hdr); 1264 } 1265 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen); 1266 if (mld == NULL) { 1267 ICMP6STAT_INC(icp6s_badlen); 1268 return (IPPROTO_DONE); 1269 } 1270 1271 /* 1272 * Userland needs to see all of this traffic for implementing 1273 * the endpoint discovery portion of multicast routing. 1274 */ 1275 switch (mld->mld_type) { 1276 case MLD_LISTENER_QUERY: 1277 icmp6_ifstat_inc(ifp, ifs6_in_mldquery); 1278 if (icmp6len == sizeof(struct mld_hdr)) { 1279 if (mld_v1_input_query(ifp, ip6, mld) != 0) 1280 return (0); 1281 } else if (icmp6len >= sizeof(struct mldv2_query)) { 1282 if (mld_v2_input_query(ifp, ip6, m, off, 1283 icmp6len) != 0) 1284 return (0); 1285 } 1286 break; 1287 case MLD_LISTENER_REPORT: 1288 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1289 if (mld_v1_input_report(ifp, ip6, mld) != 0) 1290 return (0); 1291 break; 1292 case MLDV2_LISTENER_REPORT: 1293 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1294 break; 1295 case MLD_LISTENER_DONE: 1296 icmp6_ifstat_inc(ifp, ifs6_in_mlddone); 1297 break; 1298 default: 1299 break; 1300 } 1301 1302 return (0); 1303 } 1304 1305 /* 1306 * Fast timeout handler (global). 1307 * VIMAGE: Timeout handlers are expected to service all vimages. 1308 */ 1309 void 1310 mld_fasttimo(void) 1311 { 1312 VNET_ITERATOR_DECL(vnet_iter); 1313 1314 VNET_LIST_RLOCK_NOSLEEP(); 1315 VNET_FOREACH(vnet_iter) { 1316 CURVNET_SET(vnet_iter); 1317 mld_fasttimo_vnet(); 1318 CURVNET_RESTORE(); 1319 } 1320 VNET_LIST_RUNLOCK_NOSLEEP(); 1321 } 1322 1323 /* 1324 * Fast timeout handler (per-vnet). 1325 * 1326 * VIMAGE: Assume caller has set up our curvnet. 1327 */ 1328 static void 1329 mld_fasttimo_vnet(void) 1330 { 1331 struct ifqueue scq; /* State-change packets */ 1332 struct ifqueue qrq; /* Query response packets */ 1333 struct ifnet *ifp; 1334 struct mld_ifinfo *mli; 1335 struct ifmultiaddr *ifma; 1336 struct in6_multi *inm, *tinm; 1337 int uri_fasthz; 1338 1339 uri_fasthz = 0; 1340 1341 /* 1342 * Quick check to see if any work needs to be done, in order to 1343 * minimize the overhead of fasttimo processing. 1344 * SMPng: XXX Unlocked reads. 1345 */ 1346 if (!V_current_state_timers_running6 && 1347 !V_interface_timers_running6 && 1348 !V_state_change_timers_running6) 1349 return; 1350 1351 IN6_MULTI_LOCK(); 1352 MLD_LOCK(); 1353 1354 /* 1355 * MLDv2 General Query response timer processing. 1356 */ 1357 if (V_interface_timers_running6) { 1358 CTR1(KTR_MLD, "%s: interface timers running", __func__); 1359 1360 V_interface_timers_running6 = 0; 1361 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1362 if (mli->mli_v2_timer == 0) { 1363 /* Do nothing. */ 1364 } else if (--mli->mli_v2_timer == 0) { 1365 mld_v2_dispatch_general_query(mli); 1366 } else { 1367 V_interface_timers_running6 = 1; 1368 } 1369 } 1370 } 1371 1372 if (!V_current_state_timers_running6 && 1373 !V_state_change_timers_running6) 1374 goto out_locked; 1375 1376 V_current_state_timers_running6 = 0; 1377 V_state_change_timers_running6 = 0; 1378 1379 CTR1(KTR_MLD, "%s: state change timers running", __func__); 1380 1381 /* 1382 * MLD host report and state-change timer processing. 1383 * Note: Processing a v2 group timer may remove a node. 1384 */ 1385 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1386 ifp = mli->mli_ifp; 1387 1388 if (mli->mli_version == MLD_VERSION_2) { 1389 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * 1390 PR_FASTHZ); 1391 1392 memset(&qrq, 0, sizeof(struct ifqueue)); 1393 IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS); 1394 1395 memset(&scq, 0, sizeof(struct ifqueue)); 1396 IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS); 1397 } 1398 1399 IF_ADDR_RLOCK(ifp); 1400 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1401 if (ifma->ifma_addr->sa_family != AF_INET6 || 1402 ifma->ifma_protospec == NULL) 1403 continue; 1404 inm = (struct in6_multi *)ifma->ifma_protospec; 1405 switch (mli->mli_version) { 1406 case MLD_VERSION_1: 1407 mld_v1_process_group_timer(mli, inm); 1408 break; 1409 case MLD_VERSION_2: 1410 mld_v2_process_group_timers(mli, &qrq, 1411 &scq, inm, uri_fasthz); 1412 break; 1413 } 1414 } 1415 IF_ADDR_RUNLOCK(ifp); 1416 1417 switch (mli->mli_version) { 1418 case MLD_VERSION_1: 1419 /* 1420 * Transmit reports for this lifecycle. This 1421 * is done while not holding IF_ADDR_LOCK 1422 * since this can call 1423 * in6ifa_ifpforlinklocal() which locks 1424 * IF_ADDR_LOCK internally as well as 1425 * ip6_output() to transmit a packet. 1426 */ 1427 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1428 in6m_nrele, tinm) { 1429 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1430 in6m_nrele); 1431 (void)mld_v1_transmit_report(inm, 1432 MLD_LISTENER_REPORT); 1433 } 1434 break; 1435 case MLD_VERSION_2: 1436 mld_dispatch_queue(&qrq, 0); 1437 mld_dispatch_queue(&scq, 0); 1438 1439 /* 1440 * Free the in_multi reference(s) for 1441 * this lifecycle. 1442 */ 1443 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1444 in6m_nrele, tinm) { 1445 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1446 in6m_nrele); 1447 in6m_release_locked(inm); 1448 } 1449 break; 1450 } 1451 } 1452 1453 out_locked: 1454 MLD_UNLOCK(); 1455 IN6_MULTI_UNLOCK(); 1456 } 1457 1458 /* 1459 * Update host report group timer. 1460 * Will update the global pending timer flags. 1461 */ 1462 static void 1463 mld_v1_process_group_timer(struct mld_ifinfo *mli, struct in6_multi *inm) 1464 { 1465 int report_timer_expired; 1466 1467 IN6_MULTI_LOCK_ASSERT(); 1468 MLD_LOCK_ASSERT(); 1469 1470 if (inm->in6m_timer == 0) { 1471 report_timer_expired = 0; 1472 } else if (--inm->in6m_timer == 0) { 1473 report_timer_expired = 1; 1474 } else { 1475 V_current_state_timers_running6 = 1; 1476 return; 1477 } 1478 1479 switch (inm->in6m_state) { 1480 case MLD_NOT_MEMBER: 1481 case MLD_SILENT_MEMBER: 1482 case MLD_IDLE_MEMBER: 1483 case MLD_LAZY_MEMBER: 1484 case MLD_SLEEPING_MEMBER: 1485 case MLD_AWAKENING_MEMBER: 1486 break; 1487 case MLD_REPORTING_MEMBER: 1488 if (report_timer_expired) { 1489 inm->in6m_state = MLD_IDLE_MEMBER; 1490 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1491 in6m_nrele); 1492 } 1493 break; 1494 case MLD_G_QUERY_PENDING_MEMBER: 1495 case MLD_SG_QUERY_PENDING_MEMBER: 1496 case MLD_LEAVING_MEMBER: 1497 break; 1498 } 1499 } 1500 1501 /* 1502 * Update a group's timers for MLDv2. 1503 * Will update the global pending timer flags. 1504 * Note: Unlocked read from mli. 1505 */ 1506 static void 1507 mld_v2_process_group_timers(struct mld_ifinfo *mli, 1508 struct ifqueue *qrq, struct ifqueue *scq, 1509 struct in6_multi *inm, const int uri_fasthz) 1510 { 1511 int query_response_timer_expired; 1512 int state_change_retransmit_timer_expired; 1513 #ifdef KTR 1514 char ip6tbuf[INET6_ADDRSTRLEN]; 1515 #endif 1516 1517 IN6_MULTI_LOCK_ASSERT(); 1518 MLD_LOCK_ASSERT(); 1519 1520 query_response_timer_expired = 0; 1521 state_change_retransmit_timer_expired = 0; 1522 1523 /* 1524 * During a transition from compatibility mode back to MLDv2, 1525 * a group record in REPORTING state may still have its group 1526 * timer active. This is a no-op in this function; it is easier 1527 * to deal with it here than to complicate the slow-timeout path. 1528 */ 1529 if (inm->in6m_timer == 0) { 1530 query_response_timer_expired = 0; 1531 } else if (--inm->in6m_timer == 0) { 1532 query_response_timer_expired = 1; 1533 } else { 1534 V_current_state_timers_running6 = 1; 1535 } 1536 1537 if (inm->in6m_sctimer == 0) { 1538 state_change_retransmit_timer_expired = 0; 1539 } else if (--inm->in6m_sctimer == 0) { 1540 state_change_retransmit_timer_expired = 1; 1541 } else { 1542 V_state_change_timers_running6 = 1; 1543 } 1544 1545 /* We are in fasttimo, so be quick about it. */ 1546 if (!state_change_retransmit_timer_expired && 1547 !query_response_timer_expired) 1548 return; 1549 1550 switch (inm->in6m_state) { 1551 case MLD_NOT_MEMBER: 1552 case MLD_SILENT_MEMBER: 1553 case MLD_SLEEPING_MEMBER: 1554 case MLD_LAZY_MEMBER: 1555 case MLD_AWAKENING_MEMBER: 1556 case MLD_IDLE_MEMBER: 1557 break; 1558 case MLD_G_QUERY_PENDING_MEMBER: 1559 case MLD_SG_QUERY_PENDING_MEMBER: 1560 /* 1561 * Respond to a previously pending Group-Specific 1562 * or Group-and-Source-Specific query by enqueueing 1563 * the appropriate Current-State report for 1564 * immediate transmission. 1565 */ 1566 if (query_response_timer_expired) { 1567 int retval; 1568 1569 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, 1570 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER), 1571 0); 1572 CTR2(KTR_MLD, "%s: enqueue record = %d", 1573 __func__, retval); 1574 inm->in6m_state = MLD_REPORTING_MEMBER; 1575 in6m_clear_recorded(inm); 1576 } 1577 /* FALLTHROUGH */ 1578 case MLD_REPORTING_MEMBER: 1579 case MLD_LEAVING_MEMBER: 1580 if (state_change_retransmit_timer_expired) { 1581 /* 1582 * State-change retransmission timer fired. 1583 * If there are any further pending retransmissions, 1584 * set the global pending state-change flag, and 1585 * reset the timer. 1586 */ 1587 if (--inm->in6m_scrv > 0) { 1588 inm->in6m_sctimer = uri_fasthz; 1589 V_state_change_timers_running6 = 1; 1590 } 1591 /* 1592 * Retransmit the previously computed state-change 1593 * report. If there are no further pending 1594 * retransmissions, the mbuf queue will be consumed. 1595 * Update T0 state to T1 as we have now sent 1596 * a state-change. 1597 */ 1598 (void)mld_v2_merge_state_changes(inm, scq); 1599 1600 in6m_commit(inm); 1601 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 1602 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1603 if_name(inm->in6m_ifp)); 1604 1605 /* 1606 * If we are leaving the group for good, make sure 1607 * we release MLD's reference to it. 1608 * This release must be deferred using a SLIST, 1609 * as we are called from a loop which traverses 1610 * the in_ifmultiaddr TAILQ. 1611 */ 1612 if (inm->in6m_state == MLD_LEAVING_MEMBER && 1613 inm->in6m_scrv == 0) { 1614 inm->in6m_state = MLD_NOT_MEMBER; 1615 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 1616 inm, in6m_nrele); 1617 } 1618 } 1619 break; 1620 } 1621 } 1622 1623 /* 1624 * Switch to a different version on the given interface, 1625 * as per Section 9.12. 1626 */ 1627 static void 1628 mld_set_version(struct mld_ifinfo *mli, const int version) 1629 { 1630 int old_version_timer; 1631 1632 MLD_LOCK_ASSERT(); 1633 1634 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, 1635 version, mli->mli_ifp, if_name(mli->mli_ifp)); 1636 1637 if (version == MLD_VERSION_1) { 1638 /* 1639 * Compute the "Older Version Querier Present" timer as per 1640 * Section 9.12. 1641 */ 1642 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; 1643 old_version_timer *= PR_SLOWHZ; 1644 mli->mli_v1_timer = old_version_timer; 1645 } 1646 1647 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { 1648 mli->mli_version = MLD_VERSION_1; 1649 mld_v2_cancel_link_timers(mli); 1650 } 1651 } 1652 1653 /* 1654 * Cancel pending MLDv2 timers for the given link and all groups 1655 * joined on it; state-change, general-query, and group-query timers. 1656 */ 1657 static void 1658 mld_v2_cancel_link_timers(struct mld_ifinfo *mli) 1659 { 1660 struct ifmultiaddr *ifma; 1661 struct ifnet *ifp; 1662 struct in6_multi *inm, *tinm; 1663 1664 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, 1665 mli->mli_ifp, if_name(mli->mli_ifp)); 1666 1667 IN6_MULTI_LOCK_ASSERT(); 1668 MLD_LOCK_ASSERT(); 1669 1670 /* 1671 * Fast-track this potentially expensive operation 1672 * by checking all the global 'timer pending' flags. 1673 */ 1674 if (!V_interface_timers_running6 && 1675 !V_state_change_timers_running6 && 1676 !V_current_state_timers_running6) 1677 return; 1678 1679 mli->mli_v2_timer = 0; 1680 1681 ifp = mli->mli_ifp; 1682 1683 IF_ADDR_RLOCK(ifp); 1684 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1685 if (ifma->ifma_addr->sa_family != AF_INET6) 1686 continue; 1687 inm = (struct in6_multi *)ifma->ifma_protospec; 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 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1703 in6m_nrele); 1704 /* FALLTHROUGH */ 1705 case MLD_G_QUERY_PENDING_MEMBER: 1706 case MLD_SG_QUERY_PENDING_MEMBER: 1707 in6m_clear_recorded(inm); 1708 /* FALLTHROUGH */ 1709 case MLD_REPORTING_MEMBER: 1710 inm->in6m_sctimer = 0; 1711 inm->in6m_timer = 0; 1712 inm->in6m_state = MLD_REPORTING_MEMBER; 1713 /* 1714 * Free any pending MLDv2 state-change records. 1715 */ 1716 _IF_DRAIN(&inm->in6m_scq); 1717 break; 1718 } 1719 } 1720 IF_ADDR_RUNLOCK(ifp); 1721 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, tinm) { 1722 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 1723 in6m_release_locked(inm); 1724 } 1725 } 1726 1727 /* 1728 * Global slowtimo handler. 1729 * VIMAGE: Timeout handlers are expected to service all vimages. 1730 */ 1731 void 1732 mld_slowtimo(void) 1733 { 1734 VNET_ITERATOR_DECL(vnet_iter); 1735 1736 VNET_LIST_RLOCK_NOSLEEP(); 1737 VNET_FOREACH(vnet_iter) { 1738 CURVNET_SET(vnet_iter); 1739 mld_slowtimo_vnet(); 1740 CURVNET_RESTORE(); 1741 } 1742 VNET_LIST_RUNLOCK_NOSLEEP(); 1743 } 1744 1745 /* 1746 * Per-vnet slowtimo handler. 1747 */ 1748 static void 1749 mld_slowtimo_vnet(void) 1750 { 1751 struct mld_ifinfo *mli; 1752 1753 MLD_LOCK(); 1754 1755 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1756 mld_v1_process_querier_timers(mli); 1757 } 1758 1759 MLD_UNLOCK(); 1760 } 1761 1762 /* 1763 * Update the Older Version Querier Present timers for a link. 1764 * See Section 9.12 of RFC 3810. 1765 */ 1766 static void 1767 mld_v1_process_querier_timers(struct mld_ifinfo *mli) 1768 { 1769 1770 MLD_LOCK_ASSERT(); 1771 1772 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) { 1773 /* 1774 * MLDv1 Querier Present timer expired; revert to MLDv2. 1775 */ 1776 CTR5(KTR_MLD, 1777 "%s: transition from v%d -> v%d on %p(%s)", 1778 __func__, mli->mli_version, MLD_VERSION_2, 1779 mli->mli_ifp, if_name(mli->mli_ifp)); 1780 mli->mli_version = MLD_VERSION_2; 1781 } 1782 } 1783 1784 /* 1785 * Transmit an MLDv1 report immediately. 1786 */ 1787 static int 1788 mld_v1_transmit_report(struct in6_multi *in6m, const int type) 1789 { 1790 struct ifnet *ifp; 1791 struct in6_ifaddr *ia; 1792 struct ip6_hdr *ip6; 1793 struct mbuf *mh, *md; 1794 struct mld_hdr *mld; 1795 1796 IN6_MULTI_LOCK_ASSERT(); 1797 MLD_LOCK_ASSERT(); 1798 1799 ifp = in6m->in6m_ifp; 1800 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1801 /* ia may be NULL if link-local address is tentative. */ 1802 1803 mh = m_gethdr(M_NOWAIT, MT_DATA); 1804 if (mh == NULL) { 1805 if (ia != NULL) 1806 ifa_free(&ia->ia_ifa); 1807 return (ENOMEM); 1808 } 1809 md = m_get(M_NOWAIT, MT_DATA); 1810 if (md == NULL) { 1811 m_free(mh); 1812 if (ia != NULL) 1813 ifa_free(&ia->ia_ifa); 1814 return (ENOMEM); 1815 } 1816 mh->m_next = md; 1817 1818 /* 1819 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so 1820 * that ether_output() does not need to allocate another mbuf 1821 * for the header in the most common case. 1822 */ 1823 MH_ALIGN(mh, sizeof(struct ip6_hdr)); 1824 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); 1825 mh->m_len = sizeof(struct ip6_hdr); 1826 1827 ip6 = mtod(mh, struct ip6_hdr *); 1828 ip6->ip6_flow = 0; 1829 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 1830 ip6->ip6_vfc |= IPV6_VERSION; 1831 ip6->ip6_nxt = IPPROTO_ICMPV6; 1832 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 1833 ip6->ip6_dst = in6m->in6m_addr; 1834 1835 md->m_len = sizeof(struct mld_hdr); 1836 mld = mtod(md, struct mld_hdr *); 1837 mld->mld_type = type; 1838 mld->mld_code = 0; 1839 mld->mld_cksum = 0; 1840 mld->mld_maxdelay = 0; 1841 mld->mld_reserved = 0; 1842 mld->mld_addr = in6m->in6m_addr; 1843 in6_clearscope(&mld->mld_addr); 1844 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 1845 sizeof(struct ip6_hdr), sizeof(struct mld_hdr)); 1846 1847 mld_save_context(mh, ifp); 1848 mh->m_flags |= M_MLDV1; 1849 1850 mld_dispatch_packet(mh); 1851 1852 if (ia != NULL) 1853 ifa_free(&ia->ia_ifa); 1854 return (0); 1855 } 1856 1857 /* 1858 * Process a state change from the upper layer for the given IPv6 group. 1859 * 1860 * Each socket holds a reference on the in_multi in its own ip_moptions. 1861 * The socket layer will have made the necessary updates to.the group 1862 * state, it is now up to MLD to issue a state change report if there 1863 * has been any change between T0 (when the last state-change was issued) 1864 * and T1 (now). 1865 * 1866 * We use the MLDv2 state machine at group level. The MLd module 1867 * however makes the decision as to which MLD protocol version to speak. 1868 * A state change *from* INCLUDE {} always means an initial join. 1869 * A state change *to* INCLUDE {} always means a final leave. 1870 * 1871 * If delay is non-zero, and the state change is an initial multicast 1872 * join, the state change report will be delayed by 'delay' ticks 1873 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise 1874 * the initial MLDv2 state change report will be delayed by whichever 1875 * is sooner, a pending state-change timer or delay itself. 1876 * 1877 * VIMAGE: curvnet should have been set by caller, as this routine 1878 * is called from the socket option handlers. 1879 */ 1880 int 1881 mld_change_state(struct in6_multi *inm, const int delay) 1882 { 1883 struct mld_ifinfo *mli; 1884 struct ifnet *ifp; 1885 int error; 1886 1887 IN6_MULTI_LOCK_ASSERT(); 1888 1889 error = 0; 1890 1891 /* 1892 * Try to detect if the upper layer just asked us to change state 1893 * for an interface which has now gone away. 1894 */ 1895 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 1896 ifp = inm->in6m_ifma->ifma_ifp; 1897 if (ifp != NULL) { 1898 /* 1899 * Sanity check that netinet6's notion of ifp is the 1900 * same as net's. 1901 */ 1902 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 1903 } 1904 1905 MLD_LOCK(); 1906 1907 mli = MLD_IFINFO(ifp); 1908 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 1909 1910 /* 1911 * If we detect a state transition to or from MCAST_UNDEFINED 1912 * for this group, then we are starting or finishing an MLD 1913 * life cycle for this group. 1914 */ 1915 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { 1916 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, 1917 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); 1918 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { 1919 CTR1(KTR_MLD, "%s: initial join", __func__); 1920 error = mld_initial_join(inm, mli, delay); 1921 goto out_locked; 1922 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { 1923 CTR1(KTR_MLD, "%s: final leave", __func__); 1924 mld_final_leave(inm, mli); 1925 goto out_locked; 1926 } 1927 } else { 1928 CTR1(KTR_MLD, "%s: filter set change", __func__); 1929 } 1930 1931 error = mld_handle_state_change(inm, mli); 1932 1933 out_locked: 1934 MLD_UNLOCK(); 1935 return (error); 1936 } 1937 1938 /* 1939 * Perform the initial join for an MLD group. 1940 * 1941 * When joining a group: 1942 * If the group should have its MLD traffic suppressed, do nothing. 1943 * MLDv1 starts sending MLDv1 host membership reports. 1944 * MLDv2 will schedule an MLDv2 state-change report containing the 1945 * initial state of the membership. 1946 * 1947 * If the delay argument is non-zero, then we must delay sending the 1948 * initial state change for delay ticks (in units of PR_FASTHZ). 1949 */ 1950 static int 1951 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli, 1952 const int delay) 1953 { 1954 struct ifnet *ifp; 1955 struct ifqueue *ifq; 1956 int error, retval, syncstates; 1957 int odelay; 1958 #ifdef KTR 1959 char ip6tbuf[INET6_ADDRSTRLEN]; 1960 #endif 1961 1962 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", 1963 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1964 inm->in6m_ifp, if_name(inm->in6m_ifp)); 1965 1966 error = 0; 1967 syncstates = 1; 1968 1969 ifp = inm->in6m_ifp; 1970 1971 IN6_MULTI_LOCK_ASSERT(); 1972 MLD_LOCK_ASSERT(); 1973 1974 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); 1975 1976 /* 1977 * Groups joined on loopback or marked as 'not reported', 1978 * enter the MLD_SILENT_MEMBER state and 1979 * are never reported in any protocol exchanges. 1980 * All other groups enter the appropriate state machine 1981 * for the version in use on this link. 1982 * A link marked as MLIF_SILENT causes MLD to be completely 1983 * disabled for the link. 1984 */ 1985 if ((ifp->if_flags & IFF_LOOPBACK) || 1986 (mli->mli_flags & MLIF_SILENT) || 1987 !mld_is_addr_reported(&inm->in6m_addr)) { 1988 CTR1(KTR_MLD, 1989 "%s: not kicking state machine for silent group", __func__); 1990 inm->in6m_state = MLD_SILENT_MEMBER; 1991 inm->in6m_timer = 0; 1992 } else { 1993 /* 1994 * Deal with overlapping in_multi lifecycle. 1995 * If this group was LEAVING, then make sure 1996 * we drop the reference we picked up to keep the 1997 * group around for the final INCLUDE {} enqueue. 1998 */ 1999 if (mli->mli_version == MLD_VERSION_2 && 2000 inm->in6m_state == MLD_LEAVING_MEMBER) 2001 in6m_release_locked(inm); 2002 2003 inm->in6m_state = MLD_REPORTING_MEMBER; 2004 2005 switch (mli->mli_version) { 2006 case MLD_VERSION_1: 2007 /* 2008 * If a delay was provided, only use it if 2009 * it is greater than the delay normally 2010 * used for an MLDv1 state change report, 2011 * and delay sending the initial MLDv1 report 2012 * by not transitioning to the IDLE state. 2013 */ 2014 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); 2015 if (delay) { 2016 inm->in6m_timer = max(delay, odelay); 2017 V_current_state_timers_running6 = 1; 2018 } else { 2019 inm->in6m_state = MLD_IDLE_MEMBER; 2020 error = mld_v1_transmit_report(inm, 2021 MLD_LISTENER_REPORT); 2022 if (error == 0) { 2023 inm->in6m_timer = odelay; 2024 V_current_state_timers_running6 = 1; 2025 } 2026 } 2027 break; 2028 2029 case MLD_VERSION_2: 2030 /* 2031 * Defer update of T0 to T1, until the first copy 2032 * of the state change has been transmitted. 2033 */ 2034 syncstates = 0; 2035 2036 /* 2037 * Immediately enqueue a State-Change Report for 2038 * this interface, freeing any previous reports. 2039 * Don't kick the timers if there is nothing to do, 2040 * or if an error occurred. 2041 */ 2042 ifq = &inm->in6m_scq; 2043 _IF_DRAIN(ifq); 2044 retval = mld_v2_enqueue_group_record(ifq, inm, 1, 2045 0, 0, (mli->mli_flags & MLIF_USEALLOW)); 2046 CTR2(KTR_MLD, "%s: enqueue record = %d", 2047 __func__, retval); 2048 if (retval <= 0) { 2049 error = retval * -1; 2050 break; 2051 } 2052 2053 /* 2054 * Schedule transmission of pending state-change 2055 * report up to RV times for this link. The timer 2056 * will fire at the next mld_fasttimo (~200ms), 2057 * giving us an opportunity to merge the reports. 2058 * 2059 * If a delay was provided to this function, only 2060 * use this delay if sooner than the existing one. 2061 */ 2062 KASSERT(mli->mli_rv > 1, 2063 ("%s: invalid robustness %d", __func__, 2064 mli->mli_rv)); 2065 inm->in6m_scrv = mli->mli_rv; 2066 if (delay) { 2067 if (inm->in6m_sctimer > 1) { 2068 inm->in6m_sctimer = 2069 min(inm->in6m_sctimer, delay); 2070 } else 2071 inm->in6m_sctimer = delay; 2072 } else 2073 inm->in6m_sctimer = 1; 2074 V_state_change_timers_running6 = 1; 2075 2076 error = 0; 2077 break; 2078 } 2079 } 2080 2081 /* 2082 * Only update the T0 state if state change is atomic, 2083 * i.e. we don't need to wait for a timer to fire before we 2084 * can consider the state change to have been communicated. 2085 */ 2086 if (syncstates) { 2087 in6m_commit(inm); 2088 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2089 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2090 if_name(inm->in6m_ifp)); 2091 } 2092 2093 return (error); 2094 } 2095 2096 /* 2097 * Issue an intermediate state change during the life-cycle. 2098 */ 2099 static int 2100 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli) 2101 { 2102 struct ifnet *ifp; 2103 int retval; 2104 #ifdef KTR 2105 char ip6tbuf[INET6_ADDRSTRLEN]; 2106 #endif 2107 2108 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)", 2109 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2110 inm->in6m_ifp, if_name(inm->in6m_ifp)); 2111 2112 ifp = inm->in6m_ifp; 2113 2114 IN6_MULTI_LOCK_ASSERT(); 2115 MLD_LOCK_ASSERT(); 2116 2117 KASSERT(mli && mli->mli_ifp == ifp, 2118 ("%s: inconsistent ifp", __func__)); 2119 2120 if ((ifp->if_flags & IFF_LOOPBACK) || 2121 (mli->mli_flags & MLIF_SILENT) || 2122 !mld_is_addr_reported(&inm->in6m_addr) || 2123 (mli->mli_version != MLD_VERSION_2)) { 2124 if (!mld_is_addr_reported(&inm->in6m_addr)) { 2125 CTR1(KTR_MLD, 2126 "%s: not kicking state machine for silent group", __func__); 2127 } 2128 CTR1(KTR_MLD, "%s: nothing to do", __func__); 2129 in6m_commit(inm); 2130 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2131 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2132 if_name(inm->in6m_ifp)); 2133 return (0); 2134 } 2135 2136 _IF_DRAIN(&inm->in6m_scq); 2137 2138 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0, 2139 (mli->mli_flags & MLIF_USEALLOW)); 2140 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); 2141 if (retval <= 0) 2142 return (-retval); 2143 2144 /* 2145 * If record(s) were enqueued, start the state-change 2146 * report timer for this group. 2147 */ 2148 inm->in6m_scrv = mli->mli_rv; 2149 inm->in6m_sctimer = 1; 2150 V_state_change_timers_running6 = 1; 2151 2152 return (0); 2153 } 2154 2155 /* 2156 * Perform the final leave for a multicast address. 2157 * 2158 * When leaving a group: 2159 * MLDv1 sends a DONE message, if and only if we are the reporter. 2160 * MLDv2 enqueues a state-change report containing a transition 2161 * to INCLUDE {} for immediate transmission. 2162 */ 2163 static void 2164 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli) 2165 { 2166 int syncstates; 2167 #ifdef KTR 2168 char ip6tbuf[INET6_ADDRSTRLEN]; 2169 #endif 2170 2171 syncstates = 1; 2172 2173 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)", 2174 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2175 inm->in6m_ifp, if_name(inm->in6m_ifp)); 2176 2177 IN6_MULTI_LOCK_ASSERT(); 2178 MLD_LOCK_ASSERT(); 2179 2180 switch (inm->in6m_state) { 2181 case MLD_NOT_MEMBER: 2182 case MLD_SILENT_MEMBER: 2183 case MLD_LEAVING_MEMBER: 2184 /* Already leaving or left; do nothing. */ 2185 CTR1(KTR_MLD, 2186 "%s: not kicking state machine for silent group", __func__); 2187 break; 2188 case MLD_REPORTING_MEMBER: 2189 case MLD_IDLE_MEMBER: 2190 case MLD_G_QUERY_PENDING_MEMBER: 2191 case MLD_SG_QUERY_PENDING_MEMBER: 2192 if (mli->mli_version == MLD_VERSION_1) { 2193 #ifdef INVARIANTS 2194 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 2195 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) 2196 panic("%s: MLDv2 state reached, not MLDv2 mode", 2197 __func__); 2198 #endif 2199 mld_v1_transmit_report(inm, MLD_LISTENER_DONE); 2200 inm->in6m_state = MLD_NOT_MEMBER; 2201 V_current_state_timers_running6 = 1; 2202 } else if (mli->mli_version == MLD_VERSION_2) { 2203 /* 2204 * Stop group timer and all pending reports. 2205 * Immediately enqueue a state-change report 2206 * TO_IN {} to be sent on the next fast timeout, 2207 * giving us an opportunity to merge reports. 2208 */ 2209 _IF_DRAIN(&inm->in6m_scq); 2210 inm->in6m_timer = 0; 2211 inm->in6m_scrv = mli->mli_rv; 2212 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " 2213 "pending retransmissions.", __func__, 2214 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2215 if_name(inm->in6m_ifp), inm->in6m_scrv); 2216 if (inm->in6m_scrv == 0) { 2217 inm->in6m_state = MLD_NOT_MEMBER; 2218 inm->in6m_sctimer = 0; 2219 } else { 2220 int retval; 2221 2222 in6m_acquire_locked(inm); 2223 2224 retval = mld_v2_enqueue_group_record( 2225 &inm->in6m_scq, inm, 1, 0, 0, 2226 (mli->mli_flags & MLIF_USEALLOW)); 2227 KASSERT(retval != 0, 2228 ("%s: enqueue record = %d", __func__, 2229 retval)); 2230 2231 inm->in6m_state = MLD_LEAVING_MEMBER; 2232 inm->in6m_sctimer = 1; 2233 V_state_change_timers_running6 = 1; 2234 syncstates = 0; 2235 } 2236 break; 2237 } 2238 break; 2239 case MLD_LAZY_MEMBER: 2240 case MLD_SLEEPING_MEMBER: 2241 case MLD_AWAKENING_MEMBER: 2242 /* Our reports are suppressed; do nothing. */ 2243 break; 2244 } 2245 2246 if (syncstates) { 2247 in6m_commit(inm); 2248 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2249 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2250 if_name(inm->in6m_ifp)); 2251 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 2252 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", 2253 __func__, &inm->in6m_addr, if_name(inm->in6m_ifp)); 2254 } 2255 } 2256 2257 /* 2258 * Enqueue an MLDv2 group record to the given output queue. 2259 * 2260 * If is_state_change is zero, a current-state record is appended. 2261 * If is_state_change is non-zero, a state-change report is appended. 2262 * 2263 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2264 * If is_group_query is zero, and if there is a packet with free space 2265 * at the tail of the queue, it will be appended to providing there 2266 * is enough free space. 2267 * Otherwise a new mbuf packet chain is allocated. 2268 * 2269 * If is_source_query is non-zero, each source is checked to see if 2270 * it was recorded for a Group-Source query, and will be omitted if 2271 * it is not both in-mode and recorded. 2272 * 2273 * If use_block_allow is non-zero, state change reports for initial join 2274 * and final leave, on an inclusive mode group with a source list, will be 2275 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively. 2276 * 2277 * The function will attempt to allocate leading space in the packet 2278 * for the IPv6+ICMP headers to be prepended without fragmenting the chain. 2279 * 2280 * If successful the size of all data appended to the queue is returned, 2281 * otherwise an error code less than zero is returned, or zero if 2282 * no record(s) were appended. 2283 */ 2284 static int 2285 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm, 2286 const int is_state_change, const int is_group_query, 2287 const int is_source_query, const int use_block_allow) 2288 { 2289 struct mldv2_record mr; 2290 struct mldv2_record *pmr; 2291 struct ifnet *ifp; 2292 struct ip6_msource *ims, *nims; 2293 struct mbuf *m0, *m, *md; 2294 int error, is_filter_list_change; 2295 int minrec0len, m0srcs, msrcs, nbytes, off; 2296 int record_has_sources; 2297 int now; 2298 int type; 2299 uint8_t mode; 2300 #ifdef KTR 2301 char ip6tbuf[INET6_ADDRSTRLEN]; 2302 #endif 2303 2304 IN6_MULTI_LOCK_ASSERT(); 2305 2306 error = 0; 2307 ifp = inm->in6m_ifp; 2308 is_filter_list_change = 0; 2309 m = NULL; 2310 m0 = NULL; 2311 m0srcs = 0; 2312 msrcs = 0; 2313 nbytes = 0; 2314 nims = NULL; 2315 record_has_sources = 1; 2316 pmr = NULL; 2317 type = MLD_DO_NOTHING; 2318 mode = inm->in6m_st[1].iss_fmode; 2319 2320 /* 2321 * If we did not transition out of ASM mode during t0->t1, 2322 * and there are no source nodes to process, we can skip 2323 * the generation of source records. 2324 */ 2325 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && 2326 inm->in6m_nsrc == 0) 2327 record_has_sources = 0; 2328 2329 if (is_state_change) { 2330 /* 2331 * Queue a state change record. 2332 * If the mode did not change, and there are non-ASM 2333 * listeners or source filters present, 2334 * we potentially need to issue two records for the group. 2335 * If there are ASM listeners, and there was no filter 2336 * mode transition of any kind, do nothing. 2337 * 2338 * If we are transitioning to MCAST_UNDEFINED, we need 2339 * not send any sources. A transition to/from this state is 2340 * considered inclusive with some special treatment. 2341 * 2342 * If we are rewriting initial joins/leaves to use 2343 * ALLOW/BLOCK, and the group's membership is inclusive, 2344 * we need to send sources in all cases. 2345 */ 2346 if (mode != inm->in6m_st[0].iss_fmode) { 2347 if (mode == MCAST_EXCLUDE) { 2348 CTR1(KTR_MLD, "%s: change to EXCLUDE", 2349 __func__); 2350 type = MLD_CHANGE_TO_EXCLUDE_MODE; 2351 } else { 2352 CTR1(KTR_MLD, "%s: change to INCLUDE", 2353 __func__); 2354 if (use_block_allow) { 2355 /* 2356 * XXX 2357 * Here we're interested in state 2358 * edges either direction between 2359 * MCAST_UNDEFINED and MCAST_INCLUDE. 2360 * Perhaps we should just check 2361 * the group state, rather than 2362 * the filter mode. 2363 */ 2364 if (mode == MCAST_UNDEFINED) { 2365 type = MLD_BLOCK_OLD_SOURCES; 2366 } else { 2367 type = MLD_ALLOW_NEW_SOURCES; 2368 } 2369 } else { 2370 type = MLD_CHANGE_TO_INCLUDE_MODE; 2371 if (mode == MCAST_UNDEFINED) 2372 record_has_sources = 0; 2373 } 2374 } 2375 } else { 2376 if (record_has_sources) { 2377 is_filter_list_change = 1; 2378 } else { 2379 type = MLD_DO_NOTHING; 2380 } 2381 } 2382 } else { 2383 /* 2384 * Queue a current state record. 2385 */ 2386 if (mode == MCAST_EXCLUDE) { 2387 type = MLD_MODE_IS_EXCLUDE; 2388 } else if (mode == MCAST_INCLUDE) { 2389 type = MLD_MODE_IS_INCLUDE; 2390 KASSERT(inm->in6m_st[1].iss_asm == 0, 2391 ("%s: inm %p is INCLUDE but ASM count is %d", 2392 __func__, inm, inm->in6m_st[1].iss_asm)); 2393 } 2394 } 2395 2396 /* 2397 * Generate the filter list changes using a separate function. 2398 */ 2399 if (is_filter_list_change) 2400 return (mld_v2_enqueue_filter_change(ifq, inm)); 2401 2402 if (type == MLD_DO_NOTHING) { 2403 CTR3(KTR_MLD, "%s: nothing to do for %s/%s", 2404 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2405 if_name(inm->in6m_ifp)); 2406 return (0); 2407 } 2408 2409 /* 2410 * If any sources are present, we must be able to fit at least 2411 * one in the trailing space of the tail packet's mbuf, 2412 * ideally more. 2413 */ 2414 minrec0len = sizeof(struct mldv2_record); 2415 if (record_has_sources) 2416 minrec0len += sizeof(struct in6_addr); 2417 2418 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, 2419 mld_rec_type_to_str(type), 2420 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2421 if_name(inm->in6m_ifp)); 2422 2423 /* 2424 * Check if we have a packet in the tail of the queue for this 2425 * group into which the first group record for this group will fit. 2426 * Otherwise allocate a new packet. 2427 * Always allocate leading space for IP6+RA+ICMPV6+REPORT. 2428 * Note: Group records for G/GSR query responses MUST be sent 2429 * in their own packet. 2430 */ 2431 m0 = ifq->ifq_tail; 2432 if (!is_group_query && 2433 m0 != NULL && 2434 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && 2435 (m0->m_pkthdr.len + minrec0len) < 2436 (ifp->if_mtu - MLD_MTUSPACE)) { 2437 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2438 sizeof(struct mldv2_record)) / 2439 sizeof(struct in6_addr); 2440 m = m0; 2441 CTR1(KTR_MLD, "%s: use existing packet", __func__); 2442 } else { 2443 if (_IF_QFULL(ifq)) { 2444 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2445 return (-ENOMEM); 2446 } 2447 m = NULL; 2448 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2449 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2450 if (!is_state_change && !is_group_query) 2451 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2452 if (m == NULL) 2453 m = m_gethdr(M_NOWAIT, MT_DATA); 2454 if (m == NULL) 2455 return (-ENOMEM); 2456 2457 mld_save_context(m, ifp); 2458 2459 CTR1(KTR_MLD, "%s: allocated first packet", __func__); 2460 } 2461 2462 /* 2463 * Append group record. 2464 * If we have sources, we don't know how many yet. 2465 */ 2466 mr.mr_type = type; 2467 mr.mr_datalen = 0; 2468 mr.mr_numsrc = 0; 2469 mr.mr_addr = inm->in6m_addr; 2470 in6_clearscope(&mr.mr_addr); 2471 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2472 if (m != m0) 2473 m_freem(m); 2474 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2475 return (-ENOMEM); 2476 } 2477 nbytes += sizeof(struct mldv2_record); 2478 2479 /* 2480 * Append as many sources as will fit in the first packet. 2481 * If we are appending to a new packet, the chain allocation 2482 * may potentially use clusters; use m_getptr() in this case. 2483 * If we are appending to an existing packet, we need to obtain 2484 * a pointer to the group record after m_append(), in case a new 2485 * mbuf was allocated. 2486 * 2487 * Only append sources which are in-mode at t1. If we are 2488 * transitioning to MCAST_UNDEFINED state on the group, and 2489 * use_block_allow is zero, do not include source entries. 2490 * Otherwise, we need to include this source in the report. 2491 * 2492 * Only report recorded sources in our filter set when responding 2493 * to a group-source query. 2494 */ 2495 if (record_has_sources) { 2496 if (m == m0) { 2497 md = m_last(m); 2498 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2499 md->m_len - nbytes); 2500 } else { 2501 md = m_getptr(m, 0, &off); 2502 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2503 off); 2504 } 2505 msrcs = 0; 2506 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, 2507 nims) { 2508 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2509 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2510 now = im6s_get_mode(inm, ims, 1); 2511 CTR2(KTR_MLD, "%s: node is %d", __func__, now); 2512 if ((now != mode) || 2513 (now == mode && 2514 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2515 CTR1(KTR_MLD, "%s: skip node", __func__); 2516 continue; 2517 } 2518 if (is_source_query && ims->im6s_stp == 0) { 2519 CTR1(KTR_MLD, "%s: skip unrecorded node", 2520 __func__); 2521 continue; 2522 } 2523 CTR1(KTR_MLD, "%s: append node", __func__); 2524 if (!m_append(m, sizeof(struct in6_addr), 2525 (void *)&ims->im6s_addr)) { 2526 if (m != m0) 2527 m_freem(m); 2528 CTR1(KTR_MLD, "%s: m_append() failed.", 2529 __func__); 2530 return (-ENOMEM); 2531 } 2532 nbytes += sizeof(struct in6_addr); 2533 ++msrcs; 2534 if (msrcs == m0srcs) 2535 break; 2536 } 2537 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, 2538 msrcs); 2539 pmr->mr_numsrc = htons(msrcs); 2540 nbytes += (msrcs * sizeof(struct in6_addr)); 2541 } 2542 2543 if (is_source_query && msrcs == 0) { 2544 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); 2545 if (m != m0) 2546 m_freem(m); 2547 return (0); 2548 } 2549 2550 /* 2551 * We are good to go with first packet. 2552 */ 2553 if (m != m0) { 2554 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); 2555 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2556 _IF_ENQUEUE(ifq, m); 2557 } else 2558 m->m_pkthdr.PH_vt.vt_nrecs++; 2559 2560 /* 2561 * No further work needed if no source list in packet(s). 2562 */ 2563 if (!record_has_sources) 2564 return (nbytes); 2565 2566 /* 2567 * Whilst sources remain to be announced, we need to allocate 2568 * a new packet and fill out as many sources as will fit. 2569 * Always try for a cluster first. 2570 */ 2571 while (nims != NULL) { 2572 if (_IF_QFULL(ifq)) { 2573 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2574 return (-ENOMEM); 2575 } 2576 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2577 if (m == NULL) 2578 m = m_gethdr(M_NOWAIT, MT_DATA); 2579 if (m == NULL) 2580 return (-ENOMEM); 2581 mld_save_context(m, ifp); 2582 md = m_getptr(m, 0, &off); 2583 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); 2584 CTR1(KTR_MLD, "%s: allocated next packet", __func__); 2585 2586 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2587 if (m != m0) 2588 m_freem(m); 2589 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2590 return (-ENOMEM); 2591 } 2592 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2593 nbytes += sizeof(struct mldv2_record); 2594 2595 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2596 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2597 2598 msrcs = 0; 2599 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2600 CTR2(KTR_MLD, "%s: visit node %s", 2601 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2602 now = im6s_get_mode(inm, ims, 1); 2603 if ((now != mode) || 2604 (now == mode && 2605 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2606 CTR1(KTR_MLD, "%s: skip node", __func__); 2607 continue; 2608 } 2609 if (is_source_query && ims->im6s_stp == 0) { 2610 CTR1(KTR_MLD, "%s: skip unrecorded node", 2611 __func__); 2612 continue; 2613 } 2614 CTR1(KTR_MLD, "%s: append node", __func__); 2615 if (!m_append(m, sizeof(struct in6_addr), 2616 (void *)&ims->im6s_addr)) { 2617 if (m != m0) 2618 m_freem(m); 2619 CTR1(KTR_MLD, "%s: m_append() failed.", 2620 __func__); 2621 return (-ENOMEM); 2622 } 2623 ++msrcs; 2624 if (msrcs == m0srcs) 2625 break; 2626 } 2627 pmr->mr_numsrc = htons(msrcs); 2628 nbytes += (msrcs * sizeof(struct in6_addr)); 2629 2630 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); 2631 _IF_ENQUEUE(ifq, m); 2632 } 2633 2634 return (nbytes); 2635 } 2636 2637 /* 2638 * Type used to mark record pass completion. 2639 * We exploit the fact we can cast to this easily from the 2640 * current filter modes on each ip_msource node. 2641 */ 2642 typedef enum { 2643 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2644 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2645 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2646 REC_FULL = REC_ALLOW | REC_BLOCK 2647 } rectype_t; 2648 2649 /* 2650 * Enqueue an MLDv2 filter list change to the given output queue. 2651 * 2652 * Source list filter state is held in an RB-tree. When the filter list 2653 * for a group is changed without changing its mode, we need to compute 2654 * the deltas between T0 and T1 for each source in the filter set, 2655 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2656 * 2657 * As we may potentially queue two record types, and the entire R-B tree 2658 * needs to be walked at once, we break this out into its own function 2659 * so we can generate a tightly packed queue of packets. 2660 * 2661 * XXX This could be written to only use one tree walk, although that makes 2662 * serializing into the mbuf chains a bit harder. For now we do two walks 2663 * which makes things easier on us, and it may or may not be harder on 2664 * the L2 cache. 2665 * 2666 * If successful the size of all data appended to the queue is returned, 2667 * otherwise an error code less than zero is returned, or zero if 2668 * no record(s) were appended. 2669 */ 2670 static int 2671 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm) 2672 { 2673 static const int MINRECLEN = 2674 sizeof(struct mldv2_record) + sizeof(struct in6_addr); 2675 struct ifnet *ifp; 2676 struct mldv2_record mr; 2677 struct mldv2_record *pmr; 2678 struct ip6_msource *ims, *nims; 2679 struct mbuf *m, *m0, *md; 2680 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 2681 int nallow, nblock; 2682 uint8_t mode, now, then; 2683 rectype_t crt, drt, nrt; 2684 #ifdef KTR 2685 char ip6tbuf[INET6_ADDRSTRLEN]; 2686 #endif 2687 2688 IN6_MULTI_LOCK_ASSERT(); 2689 2690 if (inm->in6m_nsrc == 0 || 2691 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) 2692 return (0); 2693 2694 ifp = inm->in6m_ifp; /* interface */ 2695 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ 2696 crt = REC_NONE; /* current group record type */ 2697 drt = REC_NONE; /* mask of completed group record types */ 2698 nrt = REC_NONE; /* record type for current node */ 2699 m0srcs = 0; /* # source which will fit in current mbuf chain */ 2700 npbytes = 0; /* # of bytes appended this packet */ 2701 nbytes = 0; /* # of bytes appended to group's state-change queue */ 2702 rsrcs = 0; /* # sources encoded in current record */ 2703 schanged = 0; /* # nodes encoded in overall filter change */ 2704 nallow = 0; /* # of source entries in ALLOW_NEW */ 2705 nblock = 0; /* # of source entries in BLOCK_OLD */ 2706 nims = NULL; /* next tree node pointer */ 2707 2708 /* 2709 * For each possible filter record mode. 2710 * The first kind of source we encounter tells us which 2711 * is the first kind of record we start appending. 2712 * If a node transitioned to UNDEFINED at t1, its mode is treated 2713 * as the inverse of the group's filter mode. 2714 */ 2715 while (drt != REC_FULL) { 2716 do { 2717 m0 = ifq->ifq_tail; 2718 if (m0 != NULL && 2719 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 2720 MLD_V2_REPORT_MAXRECS) && 2721 (m0->m_pkthdr.len + MINRECLEN) < 2722 (ifp->if_mtu - MLD_MTUSPACE)) { 2723 m = m0; 2724 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2725 sizeof(struct mldv2_record)) / 2726 sizeof(struct in6_addr); 2727 CTR1(KTR_MLD, 2728 "%s: use previous packet", __func__); 2729 } else { 2730 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2731 if (m == NULL) 2732 m = m_gethdr(M_NOWAIT, MT_DATA); 2733 if (m == NULL) { 2734 CTR1(KTR_MLD, 2735 "%s: m_get*() failed", __func__); 2736 return (-ENOMEM); 2737 } 2738 m->m_pkthdr.PH_vt.vt_nrecs = 0; 2739 mld_save_context(m, ifp); 2740 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2741 sizeof(struct mldv2_record)) / 2742 sizeof(struct in6_addr); 2743 npbytes = 0; 2744 CTR1(KTR_MLD, 2745 "%s: allocated new packet", __func__); 2746 } 2747 /* 2748 * Append the MLD group record header to the 2749 * current packet's data area. 2750 * Recalculate pointer to free space for next 2751 * group record, in case m_append() allocated 2752 * a new mbuf or cluster. 2753 */ 2754 memset(&mr, 0, sizeof(mr)); 2755 mr.mr_addr = inm->in6m_addr; 2756 in6_clearscope(&mr.mr_addr); 2757 if (!m_append(m, sizeof(mr), (void *)&mr)) { 2758 if (m != m0) 2759 m_freem(m); 2760 CTR1(KTR_MLD, 2761 "%s: m_append() failed", __func__); 2762 return (-ENOMEM); 2763 } 2764 npbytes += sizeof(struct mldv2_record); 2765 if (m != m0) { 2766 /* new packet; offset in chain */ 2767 md = m_getptr(m, npbytes - 2768 sizeof(struct mldv2_record), &off); 2769 pmr = (struct mldv2_record *)(mtod(md, 2770 uint8_t *) + off); 2771 } else { 2772 /* current packet; offset from last append */ 2773 md = m_last(m); 2774 pmr = (struct mldv2_record *)(mtod(md, 2775 uint8_t *) + md->m_len - 2776 sizeof(struct mldv2_record)); 2777 } 2778 /* 2779 * Begin walking the tree for this record type 2780 * pass, or continue from where we left off 2781 * previously if we had to allocate a new packet. 2782 * Only report deltas in-mode at t1. 2783 * We need not report included sources as allowed 2784 * if we are in inclusive mode on the group, 2785 * however the converse is not true. 2786 */ 2787 rsrcs = 0; 2788 if (nims == NULL) { 2789 nims = RB_MIN(ip6_msource_tree, 2790 &inm->in6m_srcs); 2791 } 2792 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2793 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2794 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2795 now = im6s_get_mode(inm, ims, 1); 2796 then = im6s_get_mode(inm, ims, 0); 2797 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", 2798 __func__, then, now); 2799 if (now == then) { 2800 CTR1(KTR_MLD, 2801 "%s: skip unchanged", __func__); 2802 continue; 2803 } 2804 if (mode == MCAST_EXCLUDE && 2805 now == MCAST_INCLUDE) { 2806 CTR1(KTR_MLD, 2807 "%s: skip IN src on EX group", 2808 __func__); 2809 continue; 2810 } 2811 nrt = (rectype_t)now; 2812 if (nrt == REC_NONE) 2813 nrt = (rectype_t)(~mode & REC_FULL); 2814 if (schanged++ == 0) { 2815 crt = nrt; 2816 } else if (crt != nrt) 2817 continue; 2818 if (!m_append(m, sizeof(struct in6_addr), 2819 (void *)&ims->im6s_addr)) { 2820 if (m != m0) 2821 m_freem(m); 2822 CTR1(KTR_MLD, 2823 "%s: m_append() failed", __func__); 2824 return (-ENOMEM); 2825 } 2826 nallow += !!(crt == REC_ALLOW); 2827 nblock += !!(crt == REC_BLOCK); 2828 if (++rsrcs == m0srcs) 2829 break; 2830 } 2831 /* 2832 * If we did not append any tree nodes on this 2833 * pass, back out of allocations. 2834 */ 2835 if (rsrcs == 0) { 2836 npbytes -= sizeof(struct mldv2_record); 2837 if (m != m0) { 2838 CTR1(KTR_MLD, 2839 "%s: m_free(m)", __func__); 2840 m_freem(m); 2841 } else { 2842 CTR1(KTR_MLD, 2843 "%s: m_adj(m, -mr)", __func__); 2844 m_adj(m, -((int)sizeof( 2845 struct mldv2_record))); 2846 } 2847 continue; 2848 } 2849 npbytes += (rsrcs * sizeof(struct in6_addr)); 2850 if (crt == REC_ALLOW) 2851 pmr->mr_type = MLD_ALLOW_NEW_SOURCES; 2852 else if (crt == REC_BLOCK) 2853 pmr->mr_type = MLD_BLOCK_OLD_SOURCES; 2854 pmr->mr_numsrc = htons(rsrcs); 2855 /* 2856 * Count the new group record, and enqueue this 2857 * packet if it wasn't already queued. 2858 */ 2859 m->m_pkthdr.PH_vt.vt_nrecs++; 2860 if (m != m0) 2861 _IF_ENQUEUE(ifq, m); 2862 nbytes += npbytes; 2863 } while (nims != NULL); 2864 drt |= crt; 2865 crt = (~crt & REC_FULL); 2866 } 2867 2868 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 2869 nallow, nblock); 2870 2871 return (nbytes); 2872 } 2873 2874 static int 2875 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq) 2876 { 2877 struct ifqueue *gq; 2878 struct mbuf *m; /* pending state-change */ 2879 struct mbuf *m0; /* copy of pending state-change */ 2880 struct mbuf *mt; /* last state-change in packet */ 2881 int docopy, domerge; 2882 u_int recslen; 2883 2884 docopy = 0; 2885 domerge = 0; 2886 recslen = 0; 2887 2888 IN6_MULTI_LOCK_ASSERT(); 2889 MLD_LOCK_ASSERT(); 2890 2891 /* 2892 * If there are further pending retransmissions, make a writable 2893 * copy of each queued state-change message before merging. 2894 */ 2895 if (inm->in6m_scrv > 0) 2896 docopy = 1; 2897 2898 gq = &inm->in6m_scq; 2899 #ifdef KTR 2900 if (gq->ifq_head == NULL) { 2901 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", 2902 __func__, inm); 2903 } 2904 #endif 2905 2906 m = gq->ifq_head; 2907 while (m != NULL) { 2908 /* 2909 * Only merge the report into the current packet if 2910 * there is sufficient space to do so; an MLDv2 report 2911 * packet may only contain 65,535 group records. 2912 * Always use a simple mbuf chain concatentation to do this, 2913 * as large state changes for single groups may have 2914 * allocated clusters. 2915 */ 2916 domerge = 0; 2917 mt = ifscq->ifq_tail; 2918 if (mt != NULL) { 2919 recslen = m_length(m, NULL); 2920 2921 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 2922 m->m_pkthdr.PH_vt.vt_nrecs <= 2923 MLD_V2_REPORT_MAXRECS) && 2924 (mt->m_pkthdr.len + recslen <= 2925 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) 2926 domerge = 1; 2927 } 2928 2929 if (!domerge && _IF_QFULL(gq)) { 2930 CTR2(KTR_MLD, 2931 "%s: outbound queue full, skipping whole packet %p", 2932 __func__, m); 2933 mt = m->m_nextpkt; 2934 if (!docopy) 2935 m_freem(m); 2936 m = mt; 2937 continue; 2938 } 2939 2940 if (!docopy) { 2941 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); 2942 _IF_DEQUEUE(gq, m0); 2943 m = m0->m_nextpkt; 2944 } else { 2945 CTR2(KTR_MLD, "%s: copying %p", __func__, m); 2946 m0 = m_dup(m, M_NOWAIT); 2947 if (m0 == NULL) 2948 return (ENOMEM); 2949 m0->m_nextpkt = NULL; 2950 m = m->m_nextpkt; 2951 } 2952 2953 if (!domerge) { 2954 CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)", 2955 __func__, m0, ifscq); 2956 _IF_ENQUEUE(ifscq, m0); 2957 } else { 2958 struct mbuf *mtl; /* last mbuf of packet mt */ 2959 2960 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", 2961 __func__, m0, mt); 2962 2963 mtl = m_last(mt); 2964 m0->m_flags &= ~M_PKTHDR; 2965 mt->m_pkthdr.len += recslen; 2966 mt->m_pkthdr.PH_vt.vt_nrecs += 2967 m0->m_pkthdr.PH_vt.vt_nrecs; 2968 2969 mtl->m_next = m0; 2970 } 2971 } 2972 2973 return (0); 2974 } 2975 2976 /* 2977 * Respond to a pending MLDv2 General Query. 2978 */ 2979 static void 2980 mld_v2_dispatch_general_query(struct mld_ifinfo *mli) 2981 { 2982 struct ifmultiaddr *ifma; 2983 struct ifnet *ifp; 2984 struct in6_multi *inm; 2985 int retval; 2986 2987 IN6_MULTI_LOCK_ASSERT(); 2988 MLD_LOCK_ASSERT(); 2989 2990 KASSERT(mli->mli_version == MLD_VERSION_2, 2991 ("%s: called when version %d", __func__, mli->mli_version)); 2992 2993 ifp = mli->mli_ifp; 2994 2995 IF_ADDR_RLOCK(ifp); 2996 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2997 if (ifma->ifma_addr->sa_family != AF_INET6 || 2998 ifma->ifma_protospec == NULL) 2999 continue; 3000 3001 inm = (struct in6_multi *)ifma->ifma_protospec; 3002 KASSERT(ifp == inm->in6m_ifp, 3003 ("%s: inconsistent ifp", __func__)); 3004 3005 switch (inm->in6m_state) { 3006 case MLD_NOT_MEMBER: 3007 case MLD_SILENT_MEMBER: 3008 break; 3009 case MLD_REPORTING_MEMBER: 3010 case MLD_IDLE_MEMBER: 3011 case MLD_LAZY_MEMBER: 3012 case MLD_SLEEPING_MEMBER: 3013 case MLD_AWAKENING_MEMBER: 3014 inm->in6m_state = MLD_REPORTING_MEMBER; 3015 retval = mld_v2_enqueue_group_record(&mli->mli_gq, 3016 inm, 0, 0, 0, 0); 3017 CTR2(KTR_MLD, "%s: enqueue record = %d", 3018 __func__, retval); 3019 break; 3020 case MLD_G_QUERY_PENDING_MEMBER: 3021 case MLD_SG_QUERY_PENDING_MEMBER: 3022 case MLD_LEAVING_MEMBER: 3023 break; 3024 } 3025 } 3026 IF_ADDR_RUNLOCK(ifp); 3027 3028 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); 3029 3030 /* 3031 * Slew transmission of bursts over 500ms intervals. 3032 */ 3033 if (mli->mli_gq.ifq_head != NULL) { 3034 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( 3035 MLD_RESPONSE_BURST_INTERVAL); 3036 V_interface_timers_running6 = 1; 3037 } 3038 } 3039 3040 /* 3041 * Transmit the next pending message in the output queue. 3042 * 3043 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3044 * MRT: Nothing needs to be done, as MLD traffic is always local to 3045 * a link and uses a link-scope multicast address. 3046 */ 3047 static void 3048 mld_dispatch_packet(struct mbuf *m) 3049 { 3050 struct ip6_moptions im6o; 3051 struct ifnet *ifp; 3052 struct ifnet *oifp; 3053 struct mbuf *m0; 3054 struct mbuf *md; 3055 struct ip6_hdr *ip6; 3056 struct mld_hdr *mld; 3057 int error; 3058 int off; 3059 int type; 3060 uint32_t ifindex; 3061 3062 CTR2(KTR_MLD, "%s: transmit %p", __func__, m); 3063 3064 /* 3065 * Set VNET image pointer from enqueued mbuf chain 3066 * before doing anything else. Whilst we use interface 3067 * indexes to guard against interface detach, they are 3068 * unique to each VIMAGE and must be retrieved. 3069 */ 3070 ifindex = mld_restore_context(m); 3071 3072 /* 3073 * Check if the ifnet still exists. This limits the scope of 3074 * any race in the absence of a global ifp lock for low cost 3075 * (an array lookup). 3076 */ 3077 ifp = ifnet_byindex(ifindex); 3078 if (ifp == NULL) { 3079 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", 3080 __func__, m, ifindex); 3081 m_freem(m); 3082 IP6STAT_INC(ip6s_noroute); 3083 goto out; 3084 } 3085 3086 im6o.im6o_multicast_hlim = 1; 3087 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); 3088 im6o.im6o_multicast_ifp = ifp; 3089 3090 if (m->m_flags & M_MLDV1) { 3091 m0 = m; 3092 } else { 3093 m0 = mld_v2_encap_report(ifp, m); 3094 if (m0 == NULL) { 3095 CTR2(KTR_MLD, "%s: dropped %p", __func__, m); 3096 IP6STAT_INC(ip6s_odropped); 3097 goto out; 3098 } 3099 } 3100 3101 mld_scrub_context(m0); 3102 m_clrprotoflags(m); 3103 m0->m_pkthdr.rcvif = V_loif; 3104 3105 ip6 = mtod(m0, struct ip6_hdr *); 3106 #if 0 3107 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */ 3108 #else 3109 /* 3110 * XXX XXX Break some KPI rules to prevent an LOR which would 3111 * occur if we called in6_setscope() at transmission. 3112 * See comments at top of file. 3113 */ 3114 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index); 3115 #endif 3116 3117 /* 3118 * Retrieve the ICMPv6 type before handoff to ip6_output(), 3119 * so we can bump the stats. 3120 */ 3121 md = m_getptr(m0, sizeof(struct ip6_hdr), &off); 3122 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off); 3123 type = mld->mld_type; 3124 3125 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o, 3126 &oifp, NULL); 3127 if (error) { 3128 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error); 3129 goto out; 3130 } 3131 ICMP6STAT_INC(icp6s_outhist[type]); 3132 if (oifp != NULL) { 3133 icmp6_ifstat_inc(oifp, ifs6_out_msg); 3134 switch (type) { 3135 case MLD_LISTENER_REPORT: 3136 case MLDV2_LISTENER_REPORT: 3137 icmp6_ifstat_inc(oifp, ifs6_out_mldreport); 3138 break; 3139 case MLD_LISTENER_DONE: 3140 icmp6_ifstat_inc(oifp, ifs6_out_mlddone); 3141 break; 3142 } 3143 } 3144 out: 3145 return; 3146 } 3147 3148 /* 3149 * Encapsulate an MLDv2 report. 3150 * 3151 * KAME IPv6 requires that hop-by-hop options be passed separately, 3152 * and that the IPv6 header be prepended in a separate mbuf. 3153 * 3154 * Returns a pointer to the new mbuf chain head, or NULL if the 3155 * allocation failed. 3156 */ 3157 static struct mbuf * 3158 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m) 3159 { 3160 struct mbuf *mh; 3161 struct mldv2_report *mld; 3162 struct ip6_hdr *ip6; 3163 struct in6_ifaddr *ia; 3164 int mldreclen; 3165 3166 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 3167 KASSERT((m->m_flags & M_PKTHDR), 3168 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3169 3170 /* 3171 * RFC3590: OK to send as :: or tentative during DAD. 3172 */ 3173 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 3174 if (ia == NULL) 3175 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__); 3176 3177 mh = m_gethdr(M_NOWAIT, MT_DATA); 3178 if (mh == NULL) { 3179 if (ia != NULL) 3180 ifa_free(&ia->ia_ifa); 3181 m_freem(m); 3182 return (NULL); 3183 } 3184 MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); 3185 3186 mldreclen = m_length(m, NULL); 3187 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); 3188 3189 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report); 3190 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + 3191 sizeof(struct mldv2_report) + mldreclen; 3192 3193 ip6 = mtod(mh, struct ip6_hdr *); 3194 ip6->ip6_flow = 0; 3195 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3196 ip6->ip6_vfc |= IPV6_VERSION; 3197 ip6->ip6_nxt = IPPROTO_ICMPV6; 3198 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 3199 if (ia != NULL) 3200 ifa_free(&ia->ia_ifa); 3201 ip6->ip6_dst = in6addr_linklocal_allv2routers; 3202 /* scope ID will be set in netisr */ 3203 3204 mld = (struct mldv2_report *)(ip6 + 1); 3205 mld->mld_type = MLDV2_LISTENER_REPORT; 3206 mld->mld_code = 0; 3207 mld->mld_cksum = 0; 3208 mld->mld_v2_reserved = 0; 3209 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs); 3210 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3211 3212 mh->m_next = m; 3213 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 3214 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen); 3215 return (mh); 3216 } 3217 3218 #ifdef KTR 3219 static char * 3220 mld_rec_type_to_str(const int type) 3221 { 3222 3223 switch (type) { 3224 case MLD_CHANGE_TO_EXCLUDE_MODE: 3225 return "TO_EX"; 3226 break; 3227 case MLD_CHANGE_TO_INCLUDE_MODE: 3228 return "TO_IN"; 3229 break; 3230 case MLD_MODE_IS_EXCLUDE: 3231 return "MODE_EX"; 3232 break; 3233 case MLD_MODE_IS_INCLUDE: 3234 return "MODE_IN"; 3235 break; 3236 case MLD_ALLOW_NEW_SOURCES: 3237 return "ALLOW_NEW"; 3238 break; 3239 case MLD_BLOCK_OLD_SOURCES: 3240 return "BLOCK_OLD"; 3241 break; 3242 default: 3243 break; 3244 } 3245 return "unknown"; 3246 } 3247 #endif 3248 3249 static void 3250 mld_init(void *unused __unused) 3251 { 3252 3253 CTR1(KTR_MLD, "%s: initializing", __func__); 3254 MLD_LOCK_INIT(); 3255 3256 ip6_initpktopts(&mld_po); 3257 mld_po.ip6po_hlim = 1; 3258 mld_po.ip6po_hbh = &mld_ra.hbh; 3259 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER; 3260 mld_po.ip6po_flags = IP6PO_DONTFRAG; 3261 } 3262 SYSINIT(mld_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_init, NULL); 3263 3264 static void 3265 mld_uninit(void *unused __unused) 3266 { 3267 3268 CTR1(KTR_MLD, "%s: tearing down", __func__); 3269 MLD_LOCK_DESTROY(); 3270 } 3271 SYSUNINIT(mld_uninit, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, mld_uninit, NULL); 3272 3273 static void 3274 vnet_mld_init(const void *unused __unused) 3275 { 3276 3277 CTR1(KTR_MLD, "%s: initializing", __func__); 3278 3279 LIST_INIT(&V_mli_head); 3280 } 3281 VNET_SYSINIT(vnet_mld_init, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_init, 3282 NULL); 3283 3284 static void 3285 vnet_mld_uninit(const void *unused __unused) 3286 { 3287 3288 CTR1(KTR_MLD, "%s: tearing down", __func__); 3289 3290 KASSERT(LIST_EMPTY(&V_mli_head), 3291 ("%s: mli list not empty; ifnets not detached?", __func__)); 3292 } 3293 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PSEUDO, SI_ORDER_ANY, vnet_mld_uninit, 3294 NULL); 3295 3296 static int 3297 mld_modevent(module_t mod, int type, void *unused __unused) 3298 { 3299 3300 switch (type) { 3301 case MOD_LOAD: 3302 case MOD_UNLOAD: 3303 break; 3304 default: 3305 return (EOPNOTSUPP); 3306 } 3307 return (0); 3308 } 3309 3310 static moduledata_t mld_mod = { 3311 "mld", 3312 mld_modevent, 3313 0 3314 }; 3315 DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3316