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/vimage.h> 83 84 #include <net/if.h> 85 #include <net/route.h> 86 #include <net/vnet.h> 87 88 #include <netinet/in.h> 89 #include <netinet/in_var.h> 90 #include <netinet6/in6_var.h> 91 #include <netinet/ip6.h> 92 #include <netinet6/ip6_var.h> 93 #include <netinet6/scope6_var.h> 94 #include <netinet/icmp6.h> 95 #include <netinet6/mld6.h> 96 #include <netinet6/mld6_var.h> 97 #include <netinet/vinet.h> 98 #include <netinet6/vinet6.h> 99 100 #include <security/mac/mac_framework.h> 101 102 #ifndef KTR_MLD 103 #define KTR_MLD KTR_INET6 104 #endif 105 106 static struct mld_ifinfo * 107 mli_alloc_locked(struct ifnet *); 108 static void mli_delete_locked(const struct ifnet *); 109 static void mld_dispatch_packet(struct mbuf *); 110 static void mld_dispatch_queue(struct ifqueue *, int); 111 static void mld_final_leave(struct in6_multi *, struct mld_ifinfo *); 112 static void mld_fasttimo_vnet(void); 113 static int mld_handle_state_change(struct in6_multi *, 114 struct mld_ifinfo *); 115 static int mld_initial_join(struct in6_multi *, struct mld_ifinfo *, 116 const int); 117 #ifdef KTR 118 static char * mld_rec_type_to_str(const int); 119 #endif 120 static void mld_set_version(struct mld_ifinfo *, const int); 121 static void mld_slowtimo_vnet(void); 122 static void mld_sysinit(void); 123 static void mld_sysuninit(void); 124 static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *, 125 /*const*/ struct mld_hdr *); 126 static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *, 127 /*const*/ struct mld_hdr *); 128 static void mld_v1_process_group_timer(struct in6_multi *, const int); 129 static void mld_v1_process_querier_timers(struct mld_ifinfo *); 130 static int mld_v1_transmit_report(struct in6_multi *, const int); 131 static void mld_v1_update_group(struct in6_multi *, const int); 132 static void mld_v2_cancel_link_timers(struct mld_ifinfo *); 133 static void mld_v2_dispatch_general_query(struct mld_ifinfo *); 134 static struct mbuf * 135 mld_v2_encap_report(struct ifnet *, struct mbuf *); 136 static int mld_v2_enqueue_filter_change(struct ifqueue *, 137 struct in6_multi *); 138 static int mld_v2_enqueue_group_record(struct ifqueue *, 139 struct in6_multi *, const int, const int, const int); 140 static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *, 141 struct mbuf *, const int, const int); 142 static int mld_v2_merge_state_changes(struct in6_multi *, 143 struct ifqueue *); 144 static void mld_v2_process_group_timers(struct mld_ifinfo *, 145 struct ifqueue *, struct ifqueue *, 146 struct in6_multi *, const int); 147 static int mld_v2_process_group_query(struct in6_multi *, 148 struct mld_ifinfo *mli, int, struct mbuf *, const int); 149 static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); 150 static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS); 151 152 #ifdef VIMAGE 153 static vnet_attach_fn vnet_mld_iattach; 154 static vnet_detach_fn vnet_mld_idetach; 155 #else 156 static int vnet_mld_iattach(const void *); 157 static int vnet_mld_idetach(const void *); 158 #endif /* VIMAGE */ 159 160 /* 161 * Normative references: RFC 2710, RFC 3590, RFC 3810. 162 * 163 * Locking: 164 * * The MLD subsystem lock ends up being system-wide for the moment, 165 * but could be per-VIMAGE later on. 166 * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. 167 * Any may be taken independently; if any are held at the same 168 * time, the above lock order must be followed. 169 * * IN6_MULTI_LOCK covers in_multi. 170 * * MLD_LOCK covers per-link state and any global variables in this file. 171 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 172 * per-link state iterators. 173 * 174 * XXX LOR PREVENTION 175 * A special case for IPv6 is the in6_setscope() routine. ip6_output() 176 * will not accept an ifp; it wants an embedded scope ID, unlike 177 * ip_output(), which happily takes the ifp given to it. The embedded 178 * scope ID is only used by MLD to select the outgoing interface. 179 * 180 * During interface attach and detach, MLD will take MLD_LOCK *after* 181 * the IF_AFDATA_LOCK. 182 * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call 183 * it with MLD_LOCK held without triggering an LOR. A netisr with indirect 184 * dispatch could work around this, but we'd rather not do that, as it 185 * can introduce other races. 186 * 187 * As such, we exploit the fact that the scope ID is just the interface 188 * index, and embed it in the IPv6 destination address accordingly. 189 * This is potentially NOT VALID for MLDv1 reports, as they 190 * are always sent to the multicast group itself; as MLDv2 191 * reports are always sent to ff02::16, this is not an issue 192 * when MLDv2 is in use. 193 * 194 * This does not however eliminate the LOR when ip6_output() itself 195 * calls in6_setscope() internally whilst MLD_LOCK is held. This will 196 * trigger a LOR warning in WITNESS when the ifnet is detached. 197 * 198 * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given 199 * how it's used across the network stack. Here we're simply exploiting 200 * the fact that MLD runs at a similar layer in the stack to scope6.c. 201 * 202 * VIMAGE: 203 * * Each in6_multi corresponds to an ifp, and each ifp corresponds 204 * to a vnet in ifp->if_vnet. 205 */ 206 static struct mtx mld_mtx; 207 MALLOC_DEFINE(M_MLD, "mld", "mld state"); 208 209 #define MLD_EMBEDSCOPE(pin6, zoneid) \ 210 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) 211 212 /* 213 * VIMAGE-wide globals. 214 */ 215 #ifdef VIMAGE_GLOBALS 216 struct timeval mld_gsrdelay; 217 LIST_HEAD(, mld_ifinfo) mli_head; 218 int interface_timers_running6; 219 int state_change_timers_running6; 220 int current_state_timers_running6; 221 #endif /* VIMAGE_GLOBALS */ 222 223 SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ 224 225 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, 226 "IPv6 Multicast Listener Discovery"); 227 228 /* 229 * Virtualized sysctls. 230 */ 231 SYSCTL_V_PROC(V_NET, vnet_inet6, _net_inet6_mld, OID_AUTO, gsrdelay, 232 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, mld_gsrdelay.tv_sec, 0, 233 sysctl_mld_gsr, "I", 234 "Rate limit for MLDv2 Group-and-Source queries in seconds"); 235 236 /* 237 * Non-virtualized sysctls. 238 */ 239 SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE, 240 sysctl_mld_ifinfo, "Per-interface MLDv2 state"); 241 242 static int mld_v1enable = 1; 243 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW, 244 &mld_v1enable, 0, "Enable fallback to MLDv1"); 245 TUNABLE_INT("net.inet6.mld.v1enable", &mld_v1enable); 246 247 /* 248 * Packed Router Alert option structure declaration. 249 */ 250 struct mld_raopt { 251 struct ip6_hbh hbh; 252 struct ip6_opt pad; 253 struct ip6_opt_router ra; 254 } __packed; 255 256 /* 257 * Router Alert hop-by-hop option header. 258 */ 259 static struct mld_raopt mld_ra = { 260 .hbh = { 0, 0 }, 261 .pad = { .ip6o_type = IP6OPT_PADN, 0 }, 262 .ra = { 263 .ip6or_type = IP6OPT_ROUTER_ALERT, 264 .ip6or_len = IP6OPT_RTALERT_LEN - 2, 265 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), 266 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) 267 } 268 }; 269 static struct ip6_pktopts mld_po; 270 271 static __inline void 272 mld_save_context(struct mbuf *m, struct ifnet *ifp) 273 { 274 275 #ifdef VIMAGE 276 m->m_pkthdr.header = ifp->if_vnet; 277 #endif /* VIMAGE */ 278 m->m_pkthdr.flowid = ifp->if_index; 279 } 280 281 static __inline void 282 mld_scrub_context(struct mbuf *m) 283 { 284 285 m->m_pkthdr.header = NULL; 286 m->m_pkthdr.flowid = 0; 287 } 288 289 /* 290 * Restore context from a queued output chain. 291 * Return saved ifindex. 292 * 293 * VIMAGE: The assertion is there to make sure that we 294 * actually called CURVNET_SET() with what's in the mbuf chain. 295 */ 296 static __inline uint32_t 297 mld_restore_context(struct mbuf *m) 298 { 299 300 #if defined(VIMAGE) && defined(INVARIANTS) 301 KASSERT(curvnet == m->m_pkthdr.header, 302 ("%s: called when curvnet was not restored", __func__)); 303 #endif 304 return (m->m_pkthdr.flowid); 305 } 306 307 /* 308 * Retrieve or set threshold between group-source queries in seconds. 309 * 310 * VIMAGE: Assume curvnet set by caller. 311 * SMPng: NOTE: Serialized by MLD lock. 312 */ 313 static int 314 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS) 315 { 316 INIT_VNET_INET6(curvnet); 317 int error; 318 int i; 319 320 error = sysctl_wire_old_buffer(req, sizeof(int)); 321 if (error) 322 return (error); 323 324 MLD_LOCK(); 325 326 i = V_mld_gsrdelay.tv_sec; 327 328 error = sysctl_handle_int(oidp, &i, 0, req); 329 if (error || !req->newptr) 330 goto out_locked; 331 332 if (i < -1 || i >= 60) { 333 error = EINVAL; 334 goto out_locked; 335 } 336 337 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", 338 V_mld_gsrdelay.tv_sec, i); 339 V_mld_gsrdelay.tv_sec = i; 340 341 out_locked: 342 MLD_UNLOCK(); 343 return (error); 344 } 345 346 /* 347 * Expose struct mld_ifinfo to userland, keyed by ifindex. 348 * For use by ifmcstat(8). 349 * 350 * SMPng: NOTE: Does an unlocked ifindex space read. 351 * VIMAGE: Assume curvnet set by caller. The node handler itself 352 * is not directly virtualized. 353 */ 354 static int 355 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) 356 { 357 INIT_VNET_NET(curvnet); 358 INIT_VNET_INET6(curvnet); 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, ifp->if_xname); 469 470 MLD_LOCK(); 471 472 mli = mli_alloc_locked(ifp); 473 if (!(ifp->if_flags & IFF_MULTICAST)) 474 mli->mli_flags |= MLIF_SILENT; 475 476 MLD_UNLOCK(); 477 478 return (mli); 479 } 480 481 /* 482 * VIMAGE: assume curvnet set by caller. 483 */ 484 static struct mld_ifinfo * 485 mli_alloc_locked(/*const*/ struct ifnet *ifp) 486 { 487 INIT_VNET_INET6(ifp->if_vnet); 488 struct mld_ifinfo *mli; 489 490 MLD_LOCK_ASSERT(); 491 492 mli = malloc(sizeof(struct mld_ifinfo), M_MLD, M_NOWAIT|M_ZERO); 493 if (mli == NULL) 494 goto out; 495 496 mli->mli_ifp = ifp; 497 mli->mli_version = MLD_VERSION_2; 498 mli->mli_flags = 0; 499 mli->mli_rv = MLD_RV_INIT; 500 mli->mli_qi = MLD_QI_INIT; 501 mli->mli_qri = MLD_QRI_INIT; 502 mli->mli_uri = MLD_URI_INIT; 503 504 SLIST_INIT(&mli->mli_relinmhead); 505 506 /* 507 * Responses to general queries are subject to bounds. 508 */ 509 IFQ_SET_MAXLEN(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); 510 511 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); 512 513 CTR2(KTR_MLD, "allocate mld_ifinfo for ifp %p(%s)", 514 ifp, ifp->if_xname); 515 516 out: 517 return (mli); 518 } 519 520 /* 521 * Hook for ifdetach. 522 * 523 * NOTE: Some finalization tasks need to run before the protocol domain 524 * is detached, but also before the link layer does its cleanup. 525 * Run before link-layer cleanup; cleanup groups, but do not free MLD state. 526 * 527 * SMPng: Caller must hold IN6_MULTI_LOCK(). 528 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. 529 * XXX This routine is also bitten by unlocked ifma_protospec access. 530 */ 531 void 532 mld_ifdetach(struct ifnet *ifp) 533 { 534 struct mld_ifinfo *mli; 535 struct ifmultiaddr *ifma; 536 struct in6_multi *inm, *tinm; 537 538 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, 539 ifp->if_xname); 540 541 IN6_MULTI_LOCK_ASSERT(); 542 MLD_LOCK(); 543 544 mli = MLD_IFINFO(ifp); 545 if (mli->mli_version == MLD_VERSION_2) { 546 IF_ADDR_LOCK(ifp); 547 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 548 if (ifma->ifma_addr->sa_family != AF_INET6 || 549 ifma->ifma_protospec == NULL) 550 continue; 551 inm = (struct in6_multi *)ifma->ifma_protospec; 552 if (inm->in6m_state == MLD_LEAVING_MEMBER) { 553 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 554 inm, in6m_nrele); 555 } 556 in6m_clear_recorded(inm); 557 } 558 IF_ADDR_UNLOCK(ifp); 559 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, 560 tinm) { 561 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 562 in6m_release_locked(inm); 563 } 564 } 565 566 MLD_UNLOCK(); 567 } 568 569 /* 570 * Hook for domifdetach. 571 * Runs after link-layer cleanup; free MLD state. 572 * 573 * SMPng: Normally called with IF_AFDATA_LOCK held. 574 */ 575 void 576 mld_domifdetach(struct ifnet *ifp) 577 { 578 579 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 580 __func__, ifp, ifp->if_xname); 581 582 MLD_LOCK(); 583 mli_delete_locked(ifp); 584 MLD_UNLOCK(); 585 } 586 587 static void 588 mli_delete_locked(const struct ifnet *ifp) 589 { 590 INIT_VNET_INET6(ifp->if_vnet); 591 struct mld_ifinfo *mli, *tmli; 592 593 CTR3(KTR_MLD, "%s: freeing mld_ifinfo for ifp %p(%s)", 594 __func__, ifp, ifp->if_xname); 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, ifp->if_xname); 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, ifp->if_xname); 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 IF_ADDR_LOCK(ifp); 687 688 /* 689 * Switch to MLDv1 host compatibility mode. 690 */ 691 mli = MLD_IFINFO(ifp); 692 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 693 mld_set_version(mli, MLD_VERSION_1); 694 695 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; 696 if (timer == 0) 697 timer = 1; 698 699 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, ifp->if_xname); 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, ifp->if_xname); 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_UNLOCK(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 INIT_VNET_INET6(curvnet); 756 #ifdef KTR 757 char ip6tbuf[INET6_ADDRSTRLEN]; 758 #endif 759 760 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, 761 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 762 inm->in6m_ifp->if_xname, timer); 763 764 IN6_MULTI_LOCK_ASSERT(); 765 766 switch (inm->in6m_state) { 767 case MLD_NOT_MEMBER: 768 case MLD_SILENT_MEMBER: 769 break; 770 case MLD_REPORTING_MEMBER: 771 if (inm->in6m_timer != 0 && 772 inm->in6m_timer <= timer) { 773 CTR1(KTR_MLD, "%s: REPORTING and timer running, " 774 "skipping.", __func__); 775 break; 776 } 777 /* FALLTHROUGH */ 778 case MLD_SG_QUERY_PENDING_MEMBER: 779 case MLD_G_QUERY_PENDING_MEMBER: 780 case MLD_IDLE_MEMBER: 781 case MLD_LAZY_MEMBER: 782 case MLD_AWAKENING_MEMBER: 783 CTR1(KTR_MLD, "%s: ->REPORTING", __func__); 784 inm->in6m_state = MLD_REPORTING_MEMBER; 785 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 786 V_current_state_timers_running6 = 1; 787 break; 788 case MLD_SLEEPING_MEMBER: 789 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); 790 inm->in6m_state = MLD_AWAKENING_MEMBER; 791 break; 792 case MLD_LEAVING_MEMBER: 793 break; 794 } 795 } 796 797 /* 798 * Process a received MLDv2 general, group-specific or 799 * group-and-source-specific query. 800 * 801 * Assumes that the query header has been pulled up to sizeof(mldv2_query). 802 * 803 * Return 0 if successful, otherwise an appropriate error code is returned. 804 */ 805 static int 806 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 807 struct mbuf *m, const int off, const int icmp6len) 808 { 809 INIT_VNET_INET6(curvnet); 810 struct mld_ifinfo *mli; 811 struct mldv2_query *mld; 812 struct in6_multi *inm; 813 uint32_t maxdelay, nsrc, qqi; 814 int is_general_query; 815 uint16_t timer; 816 uint8_t qrv; 817 #ifdef KTR 818 char ip6tbuf[INET6_ADDRSTRLEN]; 819 #endif 820 821 is_general_query = 0; 822 823 /* 824 * RFC3810 Section 6.2: MLD queries must originate from 825 * a router's link-local address. 826 */ 827 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 828 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 829 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 830 ifp, ifp->if_xname); 831 return (0); 832 } 833 834 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, ifp->if_xname); 835 836 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off); 837 838 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ 839 if (maxdelay >= 32678) { 840 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << 841 (MLD_MRC_EXP(maxdelay) + 3); 842 } 843 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; 844 if (timer == 0) 845 timer = 1; 846 847 qrv = MLD_QRV(mld->mld_misc); 848 if (qrv < 2) { 849 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, 850 qrv, MLD_RV_INIT); 851 qrv = MLD_RV_INIT; 852 } 853 854 qqi = mld->mld_qqi; 855 if (qqi >= 128) { 856 qqi = MLD_QQIC_MANT(mld->mld_qqi) << 857 (MLD_QQIC_EXP(mld->mld_qqi) + 3); 858 } 859 860 nsrc = ntohs(mld->mld_numsrc); 861 if (nsrc > MLD_MAX_GS_SOURCES) 862 return (EMSGSIZE); 863 if (icmp6len < sizeof(struct mldv2_query) + 864 (nsrc * sizeof(struct in6_addr))) 865 return (EMSGSIZE); 866 867 /* 868 * Do further input validation upfront to avoid resetting timers 869 * should we need to discard this query. 870 */ 871 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 872 /* 873 * General Queries SHOULD be directed to ff02::1. 874 * A general query with a source list has undefined 875 * behaviour; discard it. 876 */ 877 struct in6_addr dst; 878 879 dst = ip6->ip6_dst; 880 in6_clearscope(&dst); 881 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes) || 882 nsrc > 0) 883 return (EINVAL); 884 is_general_query = 1; 885 } else { 886 /* 887 * Embed scope ID of receiving interface in MLD query for 888 * lookup whilst we don't hold other locks (due to KAME 889 * locking lameness). We own this mbuf chain just now. 890 */ 891 in6_setscope(&mld->mld_addr, ifp, NULL); 892 } 893 894 IN6_MULTI_LOCK(); 895 MLD_LOCK(); 896 IF_ADDR_LOCK(ifp); 897 898 mli = MLD_IFINFO(ifp); 899 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 900 901 /* 902 * Discard the v2 query if we're in Compatibility Mode. 903 * The RFC is pretty clear that hosts need to stay in MLDv1 mode 904 * until the Old Version Querier Present timer expires. 905 */ 906 if (mli->mli_version != MLD_VERSION_2) 907 goto out_locked; 908 909 mld_set_version(mli, MLD_VERSION_2); 910 mli->mli_rv = qrv; 911 mli->mli_qi = qqi; 912 mli->mli_qri = maxdelay; 913 914 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, 915 maxdelay); 916 917 if (is_general_query) { 918 /* 919 * MLDv2 General Query. 920 * 921 * Schedule a current-state report on this ifp for 922 * all groups, possibly containing source lists. 923 * 924 * If there is a pending General Query response 925 * scheduled earlier than the selected delay, do 926 * not schedule any other reports. 927 * Otherwise, reset the interface timer. 928 */ 929 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", 930 ifp, ifp->if_xname); 931 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { 932 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); 933 V_interface_timers_running6 = 1; 934 } 935 } else { 936 /* 937 * MLDv2 Group-specific or Group-and-source-specific Query. 938 * 939 * Group-source-specific queries are throttled on 940 * a per-group basis to defeat denial-of-service attempts. 941 * Queries for groups we are not a member of on this 942 * link are simply ignored. 943 */ 944 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 945 if (inm == NULL) 946 goto out_locked; 947 if (nsrc > 0) { 948 if (!ratecheck(&inm->in6m_lastgsrtv, 949 &V_mld_gsrdelay)) { 950 CTR1(KTR_MLD, "%s: GS query throttled.", 951 __func__); 952 goto out_locked; 953 } 954 } 955 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", 956 ifp, ifp->if_xname); 957 /* 958 * If there is a pending General Query response 959 * scheduled sooner than the selected delay, no 960 * further report need be scheduled. 961 * Otherwise, prepare to respond to the 962 * group-specific or group-and-source query. 963 */ 964 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) 965 mld_v2_process_group_query(inm, mli, timer, m, off); 966 967 /* XXX Clear embedded scope ID as userland won't expect it. */ 968 in6_clearscope(&mld->mld_addr); 969 } 970 971 out_locked: 972 IF_ADDR_UNLOCK(ifp); 973 MLD_UNLOCK(); 974 IN6_MULTI_UNLOCK(); 975 976 return (0); 977 } 978 979 /* 980 * Process a recieved MLDv2 group-specific or group-and-source-specific 981 * query. 982 * Return <0 if any error occured. Currently this is ignored. 983 */ 984 static int 985 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifinfo *mli, 986 int timer, struct mbuf *m0, const int off) 987 { 988 INIT_VNET_INET6(curvnet); 989 struct mldv2_query *mld; 990 int retval; 991 uint16_t nsrc; 992 993 IN6_MULTI_LOCK_ASSERT(); 994 MLD_LOCK_ASSERT(); 995 996 retval = 0; 997 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off); 998 999 switch (inm->in6m_state) { 1000 case MLD_NOT_MEMBER: 1001 case MLD_SILENT_MEMBER: 1002 case MLD_SLEEPING_MEMBER: 1003 case MLD_LAZY_MEMBER: 1004 case MLD_AWAKENING_MEMBER: 1005 case MLD_IDLE_MEMBER: 1006 case MLD_LEAVING_MEMBER: 1007 return (retval); 1008 break; 1009 case MLD_REPORTING_MEMBER: 1010 case MLD_G_QUERY_PENDING_MEMBER: 1011 case MLD_SG_QUERY_PENDING_MEMBER: 1012 break; 1013 } 1014 1015 nsrc = ntohs(mld->mld_numsrc); 1016 1017 /* 1018 * Deal with group-specific queries upfront. 1019 * If any group query is already pending, purge any recorded 1020 * source-list state if it exists, and schedule a query response 1021 * for this group-specific query. 1022 */ 1023 if (nsrc == 0) { 1024 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 1025 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { 1026 in6m_clear_recorded(inm); 1027 timer = min(inm->in6m_timer, timer); 1028 } 1029 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; 1030 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1031 V_current_state_timers_running6 = 1; 1032 return (retval); 1033 } 1034 1035 /* 1036 * Deal with the case where a group-and-source-specific query has 1037 * been received but a group-specific query is already pending. 1038 */ 1039 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { 1040 timer = min(inm->in6m_timer, timer); 1041 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1042 V_current_state_timers_running6 = 1; 1043 return (retval); 1044 } 1045 1046 /* 1047 * Finally, deal with the case where a group-and-source-specific 1048 * query has been received, where a response to a previous g-s-r 1049 * query exists, or none exists. 1050 * In this case, we need to parse the source-list which the Querier 1051 * has provided us with and check if we have any source list filter 1052 * entries at T1 for these sources. If we do not, there is no need 1053 * schedule a report and the query may be dropped. 1054 * If we do, we must record them and schedule a current-state 1055 * report for those sources. 1056 */ 1057 if (inm->in6m_nsrc > 0) { 1058 struct mbuf *m; 1059 uint8_t *sp; 1060 int i, nrecorded; 1061 int soff; 1062 1063 m = m0; 1064 soff = off + sizeof(struct mldv2_query); 1065 nrecorded = 0; 1066 for (i = 0; i < nsrc; i++) { 1067 sp = mtod(m, uint8_t *) + soff; 1068 retval = in6m_record_source(inm, 1069 (const struct in6_addr *)sp); 1070 if (retval < 0) 1071 break; 1072 nrecorded += retval; 1073 soff += sizeof(struct in6_addr); 1074 if (soff >= m->m_len) { 1075 soff = soff - m->m_len; 1076 m = m->m_next; 1077 if (m == NULL) 1078 break; 1079 } 1080 } 1081 if (nrecorded > 0) { 1082 CTR1(KTR_MLD, 1083 "%s: schedule response to SG query", __func__); 1084 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; 1085 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1086 V_current_state_timers_running6 = 1; 1087 } 1088 } 1089 1090 return (retval); 1091 } 1092 1093 /* 1094 * Process a received MLDv1 host membership report. 1095 * Assumes mld points to mld_hdr in pulled up mbuf chain. 1096 * 1097 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 1098 * mld_addr. This is OK as we own the mbuf chain. 1099 */ 1100 static int 1101 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, 1102 /*const*/ struct mld_hdr *mld) 1103 { 1104 struct in6_addr src, dst; 1105 struct in6_ifaddr *ia; 1106 struct in6_multi *inm; 1107 #ifdef KTR 1108 char ip6tbuf[INET6_ADDRSTRLEN]; 1109 #endif 1110 1111 if (!mld_v1enable) { 1112 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", 1113 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1114 ifp, ifp->if_xname); 1115 return (0); 1116 } 1117 1118 if (ifp->if_flags & IFF_LOOPBACK) 1119 return (0); 1120 1121 /* 1122 * MLDv1 reports must originate from a host's link-local address, 1123 * or the unspecified address (when booting). 1124 */ 1125 src = ip6->ip6_src; 1126 in6_clearscope(&src); 1127 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { 1128 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 1129 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 1130 ifp, ifp->if_xname); 1131 return (EINVAL); 1132 } 1133 1134 /* 1135 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast 1136 * group, and must be directed to the group itself. 1137 */ 1138 dst = ip6->ip6_dst; 1139 in6_clearscope(&dst); 1140 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || 1141 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { 1142 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", 1143 ip6_sprintf(ip6tbuf, &ip6->ip6_dst), 1144 ifp, ifp->if_xname); 1145 return (EINVAL); 1146 } 1147 1148 /* 1149 * Make sure we don't hear our own membership report, as fast 1150 * leave requires knowing that we are the only member of a 1151 * group. Assume we used the link-local address if available, 1152 * otherwise look for ::. 1153 * 1154 * XXX Note that scope ID comparison is needed for the address 1155 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be 1156 * performed for the on-wire address. 1157 */ 1158 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1159 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || 1160 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) 1161 return (0); 1162 1163 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", 1164 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, ifp->if_xname); 1165 1166 /* 1167 * Embed scope ID of receiving interface in MLD query for lookup 1168 * whilst we don't hold other locks (due to KAME locking lameness). 1169 */ 1170 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) 1171 in6_setscope(&mld->mld_addr, ifp, NULL); 1172 1173 IN6_MULTI_LOCK(); 1174 MLD_LOCK(); 1175 IF_ADDR_LOCK(ifp); 1176 1177 /* 1178 * MLDv1 report suppression. 1179 * If we are a member of this group, and our membership should be 1180 * reported, and our group timer is pending or about to be reset, 1181 * stop our group timer by transitioning to the 'lazy' state. 1182 */ 1183 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 1184 if (inm != NULL) { 1185 struct mld_ifinfo *mli; 1186 1187 mli = inm->in6m_mli; 1188 KASSERT(mli != NULL, 1189 ("%s: no mli for ifp %p", __func__, ifp)); 1190 1191 /* 1192 * If we are in MLDv2 host mode, do not allow the 1193 * other host's MLDv1 report to suppress our reports. 1194 */ 1195 if (mli->mli_version == MLD_VERSION_2) 1196 goto out_locked; 1197 1198 inm->in6m_timer = 0; 1199 1200 switch (inm->in6m_state) { 1201 case MLD_NOT_MEMBER: 1202 case MLD_SILENT_MEMBER: 1203 case MLD_SLEEPING_MEMBER: 1204 break; 1205 case MLD_REPORTING_MEMBER: 1206 case MLD_IDLE_MEMBER: 1207 case MLD_AWAKENING_MEMBER: 1208 CTR3(KTR_MLD, 1209 "report suppressed for %s on ifp %p(%s)", 1210 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1211 ifp, ifp->if_xname); 1212 case MLD_LAZY_MEMBER: 1213 inm->in6m_state = MLD_LAZY_MEMBER; 1214 break; 1215 case MLD_G_QUERY_PENDING_MEMBER: 1216 case MLD_SG_QUERY_PENDING_MEMBER: 1217 case MLD_LEAVING_MEMBER: 1218 break; 1219 } 1220 } 1221 1222 out_locked: 1223 MLD_UNLOCK(); 1224 IF_ADDR_UNLOCK(ifp); 1225 IN6_MULTI_UNLOCK(); 1226 1227 /* XXX Clear embedded scope ID as userland won't expect it. */ 1228 in6_clearscope(&mld->mld_addr); 1229 1230 return (0); 1231 } 1232 1233 /* 1234 * MLD input path. 1235 * 1236 * Assume query messages which fit in a single ICMPv6 message header 1237 * have been pulled up. 1238 * Assume that userland will want to see the message, even if it 1239 * otherwise fails kernel input validation; do not free it. 1240 * Pullup may however free the mbuf chain m if it fails. 1241 * 1242 * Return IPPROTO_DONE if we freed m. Otherwise, return 0. 1243 */ 1244 int 1245 mld_input(struct mbuf *m, int off, int icmp6len) 1246 { 1247 struct ifnet *ifp; 1248 struct ip6_hdr *ip6; 1249 struct mld_hdr *mld; 1250 int mldlen; 1251 1252 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); 1253 1254 ifp = m->m_pkthdr.rcvif; 1255 INIT_VNET_INET6(ifp->if_vnet); 1256 1257 ip6 = mtod(m, struct ip6_hdr *); 1258 1259 /* Pullup to appropriate size. */ 1260 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); 1261 if (mld->mld_type == MLD_LISTENER_QUERY && 1262 icmp6len >= sizeof(struct mldv2_query)) { 1263 mldlen = sizeof(struct mldv2_query); 1264 } else { 1265 mldlen = sizeof(struct mld_hdr); 1266 } 1267 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen); 1268 if (mld == NULL) { 1269 ICMP6STAT_INC(icp6s_badlen); 1270 return (IPPROTO_DONE); 1271 } 1272 1273 /* 1274 * Userland needs to see all of this traffic for implementing 1275 * the endpoint discovery portion of multicast routing. 1276 */ 1277 switch (mld->mld_type) { 1278 case MLD_LISTENER_QUERY: 1279 icmp6_ifstat_inc(ifp, ifs6_in_mldquery); 1280 if (icmp6len == sizeof(struct mld_hdr)) { 1281 if (mld_v1_input_query(ifp, ip6, mld) != 0) 1282 return (0); 1283 } else if (icmp6len >= sizeof(struct mldv2_query)) { 1284 if (mld_v2_input_query(ifp, ip6, m, off, 1285 icmp6len) != 0) 1286 return (0); 1287 } 1288 break; 1289 case MLD_LISTENER_REPORT: 1290 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1291 if (mld_v1_input_report(ifp, ip6, mld) != 0) 1292 return (0); 1293 break; 1294 case MLDV2_LISTENER_REPORT: 1295 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1296 break; 1297 case MLD_LISTENER_DONE: 1298 icmp6_ifstat_inc(ifp, ifs6_in_mlddone); 1299 break; 1300 default: 1301 break; 1302 } 1303 1304 return (0); 1305 } 1306 1307 /* 1308 * Fast timeout handler (global). 1309 * VIMAGE: Timeout handlers are expected to service all vimages. 1310 */ 1311 void 1312 mld_fasttimo(void) 1313 { 1314 VNET_ITERATOR_DECL(vnet_iter); 1315 1316 VNET_LIST_RLOCK(); 1317 VNET_FOREACH(vnet_iter) { 1318 CURVNET_SET(vnet_iter); 1319 mld_fasttimo_vnet(); 1320 CURVNET_RESTORE(); 1321 } 1322 VNET_LIST_RUNLOCK(); 1323 } 1324 1325 /* 1326 * Fast timeout handler (per-vnet). 1327 * 1328 * VIMAGE: Assume caller has set up our curvnet. 1329 */ 1330 static void 1331 mld_fasttimo_vnet(void) 1332 { 1333 INIT_VNET_INET6(curvnet); 1334 struct ifqueue scq; /* State-change packets */ 1335 struct ifqueue qrq; /* Query response packets */ 1336 struct ifnet *ifp; 1337 struct mld_ifinfo *mli; 1338 struct ifmultiaddr *ifma, *tifma; 1339 struct in6_multi *inm; 1340 int uri_fasthz; 1341 1342 uri_fasthz = 0; 1343 1344 /* 1345 * Quick check to see if any work needs to be done, in order to 1346 * minimize the overhead of fasttimo processing. 1347 * SMPng: XXX Unlocked reads. 1348 */ 1349 if (!V_current_state_timers_running6 && 1350 !V_interface_timers_running6 && 1351 !V_state_change_timers_running6) 1352 return; 1353 1354 IN6_MULTI_LOCK(); 1355 MLD_LOCK(); 1356 1357 /* 1358 * MLDv2 General Query response timer processing. 1359 */ 1360 if (V_interface_timers_running6) { 1361 CTR1(KTR_MLD, "%s: interface timers running", __func__); 1362 1363 V_interface_timers_running6 = 0; 1364 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1365 if (mli->mli_v2_timer == 0) { 1366 /* Do nothing. */ 1367 } else if (--mli->mli_v2_timer == 0) { 1368 mld_v2_dispatch_general_query(mli); 1369 } else { 1370 V_interface_timers_running6 = 1; 1371 } 1372 } 1373 } 1374 1375 if (!V_current_state_timers_running6 && 1376 !V_state_change_timers_running6) 1377 goto out_locked; 1378 1379 V_current_state_timers_running6 = 0; 1380 V_state_change_timers_running6 = 0; 1381 1382 CTR1(KTR_MLD, "%s: state change timers running", __func__); 1383 1384 /* 1385 * MLD host report and state-change timer processing. 1386 * Note: Processing a v2 group timer may remove a node. 1387 */ 1388 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1389 ifp = mli->mli_ifp; 1390 1391 if (mli->mli_version == MLD_VERSION_2) { 1392 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * 1393 PR_FASTHZ); 1394 1395 memset(&qrq, 0, sizeof(struct ifqueue)); 1396 IFQ_SET_MAXLEN(&qrq, MLD_MAX_G_GS_PACKETS); 1397 1398 memset(&scq, 0, sizeof(struct ifqueue)); 1399 IFQ_SET_MAXLEN(&scq, MLD_MAX_STATE_CHANGE_PACKETS); 1400 } 1401 1402 IF_ADDR_LOCK(ifp); 1403 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, 1404 tifma) { 1405 if (ifma->ifma_addr->sa_family != AF_INET6 || 1406 ifma->ifma_protospec == NULL) 1407 continue; 1408 inm = (struct in6_multi *)ifma->ifma_protospec; 1409 switch (mli->mli_version) { 1410 case MLD_VERSION_1: 1411 /* 1412 * XXX Drop IF_ADDR lock temporarily to 1413 * avoid recursion caused by a potential 1414 * call by in6ifa_ifpforlinklocal(). 1415 * rwlock candidate? 1416 */ 1417 IF_ADDR_UNLOCK(ifp); 1418 mld_v1_process_group_timer(inm, 1419 mli->mli_version); 1420 IF_ADDR_LOCK(ifp); 1421 break; 1422 case MLD_VERSION_2: 1423 mld_v2_process_group_timers(mli, &qrq, 1424 &scq, inm, uri_fasthz); 1425 break; 1426 } 1427 } 1428 IF_ADDR_UNLOCK(ifp); 1429 1430 if (mli->mli_version == MLD_VERSION_2) { 1431 struct in6_multi *tinm; 1432 1433 mld_dispatch_queue(&qrq, 0); 1434 mld_dispatch_queue(&scq, 0); 1435 1436 /* 1437 * Free the in_multi reference(s) for 1438 * this lifecycle. 1439 */ 1440 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1441 in6m_nrele, tinm) { 1442 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1443 in6m_nrele); 1444 in6m_release_locked(inm); 1445 } 1446 } 1447 } 1448 1449 out_locked: 1450 MLD_UNLOCK(); 1451 IN6_MULTI_UNLOCK(); 1452 } 1453 1454 /* 1455 * Update host report group timer. 1456 * Will update the global pending timer flags. 1457 */ 1458 static void 1459 mld_v1_process_group_timer(struct in6_multi *inm, const int version) 1460 { 1461 INIT_VNET_INET6(curvnet); 1462 int report_timer_expired; 1463 1464 IN6_MULTI_LOCK_ASSERT(); 1465 MLD_LOCK_ASSERT(); 1466 1467 if (inm->in6m_timer == 0) { 1468 report_timer_expired = 0; 1469 } else if (--inm->in6m_timer == 0) { 1470 report_timer_expired = 1; 1471 } else { 1472 V_current_state_timers_running6 = 1; 1473 return; 1474 } 1475 1476 switch (inm->in6m_state) { 1477 case MLD_NOT_MEMBER: 1478 case MLD_SILENT_MEMBER: 1479 case MLD_IDLE_MEMBER: 1480 case MLD_LAZY_MEMBER: 1481 case MLD_SLEEPING_MEMBER: 1482 case MLD_AWAKENING_MEMBER: 1483 break; 1484 case MLD_REPORTING_MEMBER: 1485 if (report_timer_expired) { 1486 inm->in6m_state = MLD_IDLE_MEMBER; 1487 (void)mld_v1_transmit_report(inm, 1488 MLD_LISTENER_REPORT); 1489 } 1490 break; 1491 case MLD_G_QUERY_PENDING_MEMBER: 1492 case MLD_SG_QUERY_PENDING_MEMBER: 1493 case MLD_LEAVING_MEMBER: 1494 break; 1495 } 1496 } 1497 1498 /* 1499 * Update a group's timers for MLDv2. 1500 * Will update the global pending timer flags. 1501 * Note: Unlocked read from mli. 1502 */ 1503 static void 1504 mld_v2_process_group_timers(struct mld_ifinfo *mli, 1505 struct ifqueue *qrq, struct ifqueue *scq, 1506 struct in6_multi *inm, const int uri_fasthz) 1507 { 1508 INIT_VNET_INET6(curvnet); 1509 int query_response_timer_expired; 1510 int state_change_retransmit_timer_expired; 1511 #ifdef KTR 1512 char ip6tbuf[INET6_ADDRSTRLEN]; 1513 #endif 1514 1515 IN6_MULTI_LOCK_ASSERT(); 1516 MLD_LOCK_ASSERT(); 1517 1518 query_response_timer_expired = 0; 1519 state_change_retransmit_timer_expired = 0; 1520 1521 /* 1522 * During a transition from compatibility mode back to MLDv2, 1523 * a group record in REPORTING state may still have its group 1524 * timer active. This is a no-op in this function; it is easier 1525 * to deal with it here than to complicate the slow-timeout path. 1526 */ 1527 if (inm->in6m_timer == 0) { 1528 query_response_timer_expired = 0; 1529 } else if (--inm->in6m_timer == 0) { 1530 query_response_timer_expired = 1; 1531 } else { 1532 V_current_state_timers_running6 = 1; 1533 } 1534 1535 if (inm->in6m_sctimer == 0) { 1536 state_change_retransmit_timer_expired = 0; 1537 } else if (--inm->in6m_sctimer == 0) { 1538 state_change_retransmit_timer_expired = 1; 1539 } else { 1540 V_state_change_timers_running6 = 1; 1541 } 1542 1543 /* We are in fasttimo, so be quick about it. */ 1544 if (!state_change_retransmit_timer_expired && 1545 !query_response_timer_expired) 1546 return; 1547 1548 switch (inm->in6m_state) { 1549 case MLD_NOT_MEMBER: 1550 case MLD_SILENT_MEMBER: 1551 case MLD_SLEEPING_MEMBER: 1552 case MLD_LAZY_MEMBER: 1553 case MLD_AWAKENING_MEMBER: 1554 case MLD_IDLE_MEMBER: 1555 break; 1556 case MLD_G_QUERY_PENDING_MEMBER: 1557 case MLD_SG_QUERY_PENDING_MEMBER: 1558 /* 1559 * Respond to a previously pending Group-Specific 1560 * or Group-and-Source-Specific query by enqueueing 1561 * the appropriate Current-State report for 1562 * immediate transmission. 1563 */ 1564 if (query_response_timer_expired) { 1565 int retval; 1566 1567 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, 1568 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)); 1569 CTR2(KTR_MLD, "%s: enqueue record = %d", 1570 __func__, retval); 1571 inm->in6m_state = MLD_REPORTING_MEMBER; 1572 in6m_clear_recorded(inm); 1573 } 1574 /* FALLTHROUGH */ 1575 case MLD_REPORTING_MEMBER: 1576 case MLD_LEAVING_MEMBER: 1577 if (state_change_retransmit_timer_expired) { 1578 /* 1579 * State-change retransmission timer fired. 1580 * If there are any further pending retransmissions, 1581 * set the global pending state-change flag, and 1582 * reset the timer. 1583 */ 1584 if (--inm->in6m_scrv > 0) { 1585 inm->in6m_sctimer = uri_fasthz; 1586 V_state_change_timers_running6 = 1; 1587 } 1588 /* 1589 * Retransmit the previously computed state-change 1590 * report. If there are no further pending 1591 * retransmissions, the mbuf queue will be consumed. 1592 * Update T0 state to T1 as we have now sent 1593 * a state-change. 1594 */ 1595 (void)mld_v2_merge_state_changes(inm, scq); 1596 1597 in6m_commit(inm); 1598 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 1599 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1600 inm->in6m_ifp->if_xname); 1601 1602 /* 1603 * If we are leaving the group for good, make sure 1604 * we release MLD's reference to it. 1605 * This release must be deferred using a SLIST, 1606 * as we are called from a loop which traverses 1607 * the in_ifmultiaddr TAILQ. 1608 */ 1609 if (inm->in6m_state == MLD_LEAVING_MEMBER && 1610 inm->in6m_scrv == 0) { 1611 inm->in6m_state = MLD_NOT_MEMBER; 1612 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 1613 inm, in6m_nrele); 1614 } 1615 } 1616 break; 1617 } 1618 } 1619 1620 /* 1621 * Switch to a different version on the given interface, 1622 * as per Section 9.12. 1623 */ 1624 static void 1625 mld_set_version(struct mld_ifinfo *mli, const int version) 1626 { 1627 int old_version_timer; 1628 1629 MLD_LOCK_ASSERT(); 1630 1631 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, 1632 version, mli->mli_ifp, mli->mli_ifp->if_xname); 1633 1634 if (version == MLD_VERSION_1) { 1635 /* 1636 * Compute the "Older Version Querier Present" timer as per 1637 * Section 9.12. 1638 */ 1639 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; 1640 old_version_timer *= PR_SLOWHZ; 1641 mli->mli_v1_timer = old_version_timer; 1642 } 1643 1644 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { 1645 mli->mli_version = MLD_VERSION_1; 1646 mld_v2_cancel_link_timers(mli); 1647 } 1648 } 1649 1650 /* 1651 * Cancel pending MLDv2 timers for the given link and all groups 1652 * joined on it; state-change, general-query, and group-query timers. 1653 */ 1654 static void 1655 mld_v2_cancel_link_timers(struct mld_ifinfo *mli) 1656 { 1657 INIT_VNET_INET6(curvnet); 1658 struct ifmultiaddr *ifma; 1659 struct ifnet *ifp; 1660 struct in6_multi *inm; 1661 1662 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, 1663 mli->mli_ifp, mli->mli_ifp->if_xname); 1664 1665 IN6_MULTI_LOCK_ASSERT(); 1666 MLD_LOCK_ASSERT(); 1667 1668 /* 1669 * Fast-track this potentially expensive operation 1670 * by checking all the global 'timer pending' flags. 1671 */ 1672 if (!V_interface_timers_running6 && 1673 !V_state_change_timers_running6 && 1674 !V_current_state_timers_running6) 1675 return; 1676 1677 mli->mli_v2_timer = 0; 1678 1679 ifp = mli->mli_ifp; 1680 1681 IF_ADDR_LOCK(ifp); 1682 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1683 if (ifma->ifma_addr->sa_family != AF_INET6) 1684 continue; 1685 inm = (struct in6_multi *)ifma->ifma_protospec; 1686 switch (inm->in6m_state) { 1687 case MLD_NOT_MEMBER: 1688 case MLD_SILENT_MEMBER: 1689 case MLD_IDLE_MEMBER: 1690 case MLD_LAZY_MEMBER: 1691 case MLD_SLEEPING_MEMBER: 1692 case MLD_AWAKENING_MEMBER: 1693 break; 1694 case MLD_LEAVING_MEMBER: 1695 /* 1696 * If we are leaving the group and switching 1697 * version, we need to release the final 1698 * reference held for issuing the INCLUDE {}. 1699 * 1700 * SMPNG: Must drop and re-acquire IF_ADDR_LOCK 1701 * around in6m_release_locked(), as it is not 1702 * a recursive mutex. 1703 */ 1704 IF_ADDR_UNLOCK(ifp); 1705 in6m_release_locked(inm); 1706 IF_ADDR_LOCK(ifp); 1707 /* FALLTHROUGH */ 1708 case MLD_G_QUERY_PENDING_MEMBER: 1709 case MLD_SG_QUERY_PENDING_MEMBER: 1710 in6m_clear_recorded(inm); 1711 /* FALLTHROUGH */ 1712 case MLD_REPORTING_MEMBER: 1713 inm->in6m_sctimer = 0; 1714 inm->in6m_timer = 0; 1715 inm->in6m_state = MLD_REPORTING_MEMBER; 1716 /* 1717 * Free any pending MLDv2 state-change records. 1718 */ 1719 _IF_DRAIN(&inm->in6m_scq); 1720 break; 1721 } 1722 } 1723 IF_ADDR_UNLOCK(ifp); 1724 } 1725 1726 /* 1727 * Global slowtimo handler. 1728 * VIMAGE: Timeout handlers are expected to service all vimages. 1729 */ 1730 void 1731 mld_slowtimo(void) 1732 { 1733 VNET_ITERATOR_DECL(vnet_iter); 1734 1735 VNET_LIST_RLOCK(); 1736 VNET_FOREACH(vnet_iter) { 1737 CURVNET_SET(vnet_iter); 1738 mld_slowtimo_vnet(); 1739 CURVNET_RESTORE(); 1740 } 1741 VNET_LIST_RUNLOCK(); 1742 } 1743 1744 /* 1745 * Per-vnet slowtimo handler. 1746 */ 1747 static void 1748 mld_slowtimo_vnet(void) 1749 { 1750 INIT_VNET_INET6(curvnet); 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, mli->mli_ifp->if_xname); 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 MGETHDR(mh, M_DONTWAIT, MT_HEADER); 1804 if (mh == NULL) 1805 return (ENOMEM); 1806 MGET(md, M_DONTWAIT, MT_DATA); 1807 if (md == NULL) { 1808 m_free(mh); 1809 return (ENOMEM); 1810 } 1811 mh->m_next = md; 1812 1813 /* 1814 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so 1815 * that ether_output() does not need to allocate another mbuf 1816 * for the header in the most common case. 1817 */ 1818 MH_ALIGN(mh, sizeof(struct ip6_hdr)); 1819 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr); 1820 mh->m_len = sizeof(struct ip6_hdr); 1821 1822 ip6 = mtod(mh, struct ip6_hdr *); 1823 ip6->ip6_flow = 0; 1824 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 1825 ip6->ip6_vfc |= IPV6_VERSION; 1826 ip6->ip6_nxt = IPPROTO_ICMPV6; 1827 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 1828 ip6->ip6_dst = in6m->in6m_addr; 1829 1830 md->m_len = sizeof(struct mld_hdr); 1831 mld = mtod(md, struct mld_hdr *); 1832 mld->mld_type = type; 1833 mld->mld_code = 0; 1834 mld->mld_cksum = 0; 1835 mld->mld_maxdelay = 0; 1836 mld->mld_reserved = 0; 1837 mld->mld_addr = in6m->in6m_addr; 1838 in6_clearscope(&mld->mld_addr); 1839 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 1840 sizeof(struct ip6_hdr), sizeof(struct mld_hdr)); 1841 1842 mld_save_context(mh, ifp); 1843 mh->m_flags |= M_MLDV1; 1844 1845 mld_dispatch_packet(mh); 1846 1847 return (0); 1848 } 1849 1850 /* 1851 * Process a state change from the upper layer for the given IPv6 group. 1852 * 1853 * Each socket holds a reference on the in_multi in its own ip_moptions. 1854 * The socket layer will have made the necessary updates to.the group 1855 * state, it is now up to MLD to issue a state change report if there 1856 * has been any change between T0 (when the last state-change was issued) 1857 * and T1 (now). 1858 * 1859 * We use the MLDv2 state machine at group level. The MLd module 1860 * however makes the decision as to which MLD protocol version to speak. 1861 * A state change *from* INCLUDE {} always means an initial join. 1862 * A state change *to* INCLUDE {} always means a final leave. 1863 * 1864 * If delay is non-zero, and the state change is an initial multicast 1865 * join, the state change report will be delayed by 'delay' ticks 1866 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise 1867 * the initial MLDv2 state change report will be delayed by whichever 1868 * is sooner, a pending state-change timer or delay itself. 1869 * 1870 * VIMAGE: curvnet should have been set by caller, as this routine 1871 * is called from the socket option handlers. 1872 */ 1873 int 1874 mld_change_state(struct in6_multi *inm, const int delay) 1875 { 1876 struct mld_ifinfo *mli; 1877 struct ifnet *ifp; 1878 int error; 1879 1880 IN6_MULTI_LOCK_ASSERT(); 1881 1882 error = 0; 1883 1884 /* 1885 * Try to detect if the upper layer just asked us to change state 1886 * for an interface which has now gone away. 1887 */ 1888 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 1889 ifp = inm->in6m_ifma->ifma_ifp; 1890 if (ifp != NULL) { 1891 /* 1892 * Sanity check that netinet6's notion of ifp is the 1893 * same as net's. 1894 */ 1895 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 1896 } 1897 1898 MLD_LOCK(); 1899 1900 mli = MLD_IFINFO(ifp); 1901 KASSERT(mli != NULL, ("%s: no mld_ifinfo for ifp %p", __func__, ifp)); 1902 1903 /* 1904 * If we detect a state transition to or from MCAST_UNDEFINED 1905 * for this group, then we are starting or finishing an MLD 1906 * life cycle for this group. 1907 */ 1908 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { 1909 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, 1910 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); 1911 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { 1912 CTR1(KTR_MLD, "%s: initial join", __func__); 1913 error = mld_initial_join(inm, mli, delay); 1914 goto out_locked; 1915 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { 1916 CTR1(KTR_MLD, "%s: final leave", __func__); 1917 mld_final_leave(inm, mli); 1918 goto out_locked; 1919 } 1920 } else { 1921 CTR1(KTR_MLD, "%s: filter set change", __func__); 1922 } 1923 1924 error = mld_handle_state_change(inm, mli); 1925 1926 out_locked: 1927 MLD_UNLOCK(); 1928 return (error); 1929 } 1930 1931 /* 1932 * Perform the initial join for an MLD group. 1933 * 1934 * When joining a group: 1935 * If the group should have its MLD traffic suppressed, do nothing. 1936 * MLDv1 starts sending MLDv1 host membership reports. 1937 * MLDv2 will schedule an MLDv2 state-change report containing the 1938 * initial state of the membership. 1939 * 1940 * If the delay argument is non-zero, then we must delay sending the 1941 * initial state change for delay ticks (in units of PR_FASTHZ). 1942 */ 1943 static int 1944 mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli, 1945 const int delay) 1946 { 1947 INIT_VNET_INET6(curvnet); 1948 struct ifnet *ifp; 1949 struct ifqueue *ifq; 1950 int error, retval, syncstates; 1951 int odelay; 1952 #ifdef KTR 1953 char ip6tbuf[INET6_ADDRSTRLEN]; 1954 #endif 1955 1956 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", 1957 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1958 inm->in6m_ifp, inm->in6m_ifp->if_xname); 1959 1960 error = 0; 1961 syncstates = 1; 1962 1963 ifp = inm->in6m_ifp; 1964 1965 IN6_MULTI_LOCK_ASSERT(); 1966 MLD_LOCK_ASSERT(); 1967 1968 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); 1969 1970 /* 1971 * Groups joined on loopback or marked as 'not reported', 1972 * enter the MLD_SILENT_MEMBER state and 1973 * are never reported in any protocol exchanges. 1974 * All other groups enter the appropriate state machine 1975 * for the version in use on this link. 1976 * A link marked as MLIF_SILENT causes MLD to be completely 1977 * disabled for the link. 1978 */ 1979 if ((ifp->if_flags & IFF_LOOPBACK) || 1980 (mli->mli_flags & MLIF_SILENT) || 1981 !mld_is_addr_reported(&inm->in6m_addr)) { 1982 CTR1(KTR_MLD, 1983 "%s: not kicking state machine for silent group", __func__); 1984 inm->in6m_state = MLD_SILENT_MEMBER; 1985 inm->in6m_timer = 0; 1986 } else { 1987 /* 1988 * Deal with overlapping in_multi lifecycle. 1989 * If this group was LEAVING, then make sure 1990 * we drop the reference we picked up to keep the 1991 * group around for the final INCLUDE {} enqueue. 1992 */ 1993 if (mli->mli_version == MLD_VERSION_2 && 1994 inm->in6m_state == MLD_LEAVING_MEMBER) 1995 in6m_release_locked(inm); 1996 1997 inm->in6m_state = MLD_REPORTING_MEMBER; 1998 1999 switch (mli->mli_version) { 2000 case MLD_VERSION_1: 2001 /* 2002 * If a delay was provided, only use it if 2003 * it is greater than the delay normally 2004 * used for an MLDv1 state change report, 2005 * and delay sending the initial MLDv1 report 2006 * by not transitioning to the IDLE state. 2007 */ 2008 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); 2009 if (delay) { 2010 inm->in6m_timer = max(delay, odelay); 2011 V_current_state_timers_running6 = 1; 2012 } else { 2013 inm->in6m_state = MLD_IDLE_MEMBER; 2014 error = mld_v1_transmit_report(inm, 2015 MLD_LISTENER_REPORT); 2016 if (error == 0) { 2017 inm->in6m_timer = odelay; 2018 V_current_state_timers_running6 = 1; 2019 } 2020 } 2021 break; 2022 2023 case MLD_VERSION_2: 2024 /* 2025 * Defer update of T0 to T1, until the first copy 2026 * of the state change has been transmitted. 2027 */ 2028 syncstates = 0; 2029 2030 /* 2031 * Immediately enqueue a State-Change Report for 2032 * this interface, freeing any previous reports. 2033 * Don't kick the timers if there is nothing to do, 2034 * or if an error occurred. 2035 */ 2036 ifq = &inm->in6m_scq; 2037 _IF_DRAIN(ifq); 2038 retval = mld_v2_enqueue_group_record(ifq, inm, 1, 2039 0, 0); 2040 CTR2(KTR_MLD, "%s: enqueue record = %d", 2041 __func__, retval); 2042 if (retval <= 0) { 2043 error = retval * -1; 2044 break; 2045 } 2046 2047 /* 2048 * Schedule transmission of pending state-change 2049 * report up to RV times for this link. The timer 2050 * will fire at the next mld_fasttimo (~200ms), 2051 * giving us an opportunity to merge the reports. 2052 * 2053 * If a delay was provided to this function, only 2054 * use this delay if sooner than the existing one. 2055 */ 2056 KASSERT(mli->mli_rv > 1, 2057 ("%s: invalid robustness %d", __func__, 2058 mli->mli_rv)); 2059 inm->in6m_scrv = mli->mli_rv; 2060 if (delay) { 2061 if (inm->in6m_sctimer > 1) { 2062 inm->in6m_sctimer = 2063 min(inm->in6m_sctimer, delay); 2064 } else 2065 inm->in6m_sctimer = delay; 2066 } else 2067 inm->in6m_sctimer = 1; 2068 V_state_change_timers_running6 = 1; 2069 2070 error = 0; 2071 break; 2072 } 2073 } 2074 2075 /* 2076 * Only update the T0 state if state change is atomic, 2077 * i.e. we don't need to wait for a timer to fire before we 2078 * can consider the state change to have been communicated. 2079 */ 2080 if (syncstates) { 2081 in6m_commit(inm); 2082 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2083 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2084 inm->in6m_ifp->if_xname); 2085 } 2086 2087 return (error); 2088 } 2089 2090 /* 2091 * Issue an intermediate state change during the life-cycle. 2092 */ 2093 static int 2094 mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli) 2095 { 2096 INIT_VNET_INET6(curvnet); 2097 struct ifnet *ifp; 2098 int retval; 2099 #ifdef KTR 2100 char ip6tbuf[INET6_ADDRSTRLEN]; 2101 #endif 2102 2103 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)", 2104 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2105 inm->in6m_ifp, inm->in6m_ifp->if_xname); 2106 2107 ifp = inm->in6m_ifp; 2108 2109 IN6_MULTI_LOCK_ASSERT(); 2110 MLD_LOCK_ASSERT(); 2111 2112 KASSERT(mli && mli->mli_ifp == ifp, 2113 ("%s: inconsistent ifp", __func__)); 2114 2115 if ((ifp->if_flags & IFF_LOOPBACK) || 2116 (mli->mli_flags & MLIF_SILENT) || 2117 !mld_is_addr_reported(&inm->in6m_addr) || 2118 (mli->mli_version != MLD_VERSION_2)) { 2119 if (!mld_is_addr_reported(&inm->in6m_addr)) { 2120 CTR1(KTR_MLD, 2121 "%s: not kicking state machine for silent group", __func__); 2122 } 2123 CTR1(KTR_MLD, "%s: nothing to do", __func__); 2124 in6m_commit(inm); 2125 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2126 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2127 inm->in6m_ifp->if_xname); 2128 return (0); 2129 } 2130 2131 _IF_DRAIN(&inm->in6m_scq); 2132 2133 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0); 2134 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); 2135 if (retval <= 0) 2136 return (-retval); 2137 2138 /* 2139 * If record(s) were enqueued, start the state-change 2140 * report timer for this group. 2141 */ 2142 inm->in6m_scrv = mli->mli_rv; 2143 inm->in6m_sctimer = 1; 2144 V_state_change_timers_running6 = 1; 2145 2146 return (0); 2147 } 2148 2149 /* 2150 * Perform the final leave for a multicast address. 2151 * 2152 * When leaving a group: 2153 * MLDv1 sends a DONE message, if and only if we are the reporter. 2154 * MLDv2 enqueues a state-change report containing a transition 2155 * to INCLUDE {} for immediate transmission. 2156 */ 2157 static void 2158 mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli) 2159 { 2160 INIT_VNET_INET6(curvnet); 2161 int syncstates; 2162 #ifdef KTR 2163 char ip6tbuf[INET6_ADDRSTRLEN]; 2164 #endif 2165 2166 syncstates = 1; 2167 2168 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)", 2169 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2170 inm->in6m_ifp, inm->in6m_ifp->if_xname); 2171 2172 IN6_MULTI_LOCK_ASSERT(); 2173 MLD_LOCK_ASSERT(); 2174 2175 switch (inm->in6m_state) { 2176 case MLD_NOT_MEMBER: 2177 case MLD_SILENT_MEMBER: 2178 case MLD_LEAVING_MEMBER: 2179 /* Already leaving or left; do nothing. */ 2180 CTR1(KTR_MLD, 2181 "%s: not kicking state machine for silent group", __func__); 2182 break; 2183 case MLD_REPORTING_MEMBER: 2184 case MLD_IDLE_MEMBER: 2185 case MLD_G_QUERY_PENDING_MEMBER: 2186 case MLD_SG_QUERY_PENDING_MEMBER: 2187 if (mli->mli_version == MLD_VERSION_1) { 2188 #ifdef INVARIANTS 2189 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 2190 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) 2191 panic("%s: MLDv2 state reached, not MLDv2 mode", 2192 __func__); 2193 #endif 2194 mld_v1_transmit_report(inm, MLD_LISTENER_DONE); 2195 inm->in6m_state = MLD_NOT_MEMBER; 2196 } else if (mli->mli_version == MLD_VERSION_2) { 2197 /* 2198 * Stop group timer and all pending reports. 2199 * Immediately enqueue a state-change report 2200 * TO_IN {} to be sent on the next fast timeout, 2201 * giving us an opportunity to merge reports. 2202 */ 2203 _IF_DRAIN(&inm->in6m_scq); 2204 inm->in6m_timer = 0; 2205 inm->in6m_scrv = mli->mli_rv; 2206 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " 2207 "pending retransmissions.", __func__, 2208 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2209 inm->in6m_ifp->if_xname, inm->in6m_scrv); 2210 if (inm->in6m_scrv == 0) { 2211 inm->in6m_state = MLD_NOT_MEMBER; 2212 inm->in6m_sctimer = 0; 2213 } else { 2214 int retval; 2215 2216 in6m_acquire_locked(inm); 2217 2218 retval = mld_v2_enqueue_group_record( 2219 &inm->in6m_scq, inm, 1, 0, 0); 2220 KASSERT(retval != 0, 2221 ("%s: enqueue record = %d", __func__, 2222 retval)); 2223 2224 inm->in6m_state = MLD_LEAVING_MEMBER; 2225 inm->in6m_sctimer = 1; 2226 V_state_change_timers_running6 = 1; 2227 syncstates = 0; 2228 } 2229 break; 2230 } 2231 break; 2232 case MLD_LAZY_MEMBER: 2233 case MLD_SLEEPING_MEMBER: 2234 case MLD_AWAKENING_MEMBER: 2235 /* Our reports are suppressed; do nothing. */ 2236 break; 2237 } 2238 2239 if (syncstates) { 2240 in6m_commit(inm); 2241 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2242 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2243 inm->in6m_ifp->if_xname); 2244 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 2245 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", 2246 __func__, &inm->in6m_addr, inm->in6m_ifp->if_xname); 2247 } 2248 } 2249 2250 /* 2251 * Enqueue an MLDv2 group record to the given output queue. 2252 * 2253 * If is_state_change is zero, a current-state record is appended. 2254 * If is_state_change is non-zero, a state-change report is appended. 2255 * 2256 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2257 * If is_group_query is zero, and if there is a packet with free space 2258 * at the tail of the queue, it will be appended to providing there 2259 * is enough free space. 2260 * Otherwise a new mbuf packet chain is allocated. 2261 * 2262 * If is_source_query is non-zero, each source is checked to see if 2263 * it was recorded for a Group-Source query, and will be omitted if 2264 * it is not both in-mode and recorded. 2265 * 2266 * The function will attempt to allocate leading space in the packet 2267 * for the IPv6+ICMP headers to be prepended without fragmenting the chain. 2268 * 2269 * If successful the size of all data appended to the queue is returned, 2270 * otherwise an error code less than zero is returned, or zero if 2271 * no record(s) were appended. 2272 */ 2273 static int 2274 mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm, 2275 const int is_state_change, const int is_group_query, 2276 const int is_source_query) 2277 { 2278 struct mldv2_record mr; 2279 struct mldv2_record *pmr; 2280 struct ifnet *ifp; 2281 struct ip6_msource *ims, *nims; 2282 struct mbuf *m0, *m, *md; 2283 int error, is_filter_list_change; 2284 int minrec0len, m0srcs, msrcs, nbytes, off; 2285 int record_has_sources; 2286 int now; 2287 int type; 2288 uint8_t mode; 2289 #ifdef KTR 2290 char ip6tbuf[INET6_ADDRSTRLEN]; 2291 #endif 2292 2293 IN6_MULTI_LOCK_ASSERT(); 2294 2295 error = 0; 2296 ifp = inm->in6m_ifp; 2297 is_filter_list_change = 0; 2298 m = NULL; 2299 m0 = NULL; 2300 m0srcs = 0; 2301 msrcs = 0; 2302 nbytes = 0; 2303 nims = NULL; 2304 record_has_sources = 1; 2305 pmr = NULL; 2306 type = MLD_DO_NOTHING; 2307 mode = inm->in6m_st[1].iss_fmode; 2308 2309 /* 2310 * If we did not transition out of ASM mode during t0->t1, 2311 * and there are no source nodes to process, we can skip 2312 * the generation of source records. 2313 */ 2314 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && 2315 inm->in6m_nsrc == 0) 2316 record_has_sources = 0; 2317 2318 if (is_state_change) { 2319 /* 2320 * Queue a state change record. 2321 * If the mode did not change, and there are non-ASM 2322 * listeners or source filters present, 2323 * we potentially need to issue two records for the group. 2324 * If we are transitioning to MCAST_UNDEFINED, we need 2325 * not send any sources. 2326 * If there are ASM listeners, and there was no filter 2327 * mode transition of any kind, do nothing. 2328 */ 2329 if (mode != inm->in6m_st[0].iss_fmode) { 2330 if (mode == MCAST_EXCLUDE) { 2331 CTR1(KTR_MLD, "%s: change to EXCLUDE", 2332 __func__); 2333 type = MLD_CHANGE_TO_EXCLUDE_MODE; 2334 } else { 2335 CTR1(KTR_MLD, "%s: change to INCLUDE", 2336 __func__); 2337 type = MLD_CHANGE_TO_INCLUDE_MODE; 2338 if (mode == MCAST_UNDEFINED) 2339 record_has_sources = 0; 2340 } 2341 } else { 2342 if (record_has_sources) { 2343 is_filter_list_change = 1; 2344 } else { 2345 type = MLD_DO_NOTHING; 2346 } 2347 } 2348 } else { 2349 /* 2350 * Queue a current state record. 2351 */ 2352 if (mode == MCAST_EXCLUDE) { 2353 type = MLD_MODE_IS_EXCLUDE; 2354 } else if (mode == MCAST_INCLUDE) { 2355 type = MLD_MODE_IS_INCLUDE; 2356 KASSERT(inm->in6m_st[1].iss_asm == 0, 2357 ("%s: inm %p is INCLUDE but ASM count is %d", 2358 __func__, inm, inm->in6m_st[1].iss_asm)); 2359 } 2360 } 2361 2362 /* 2363 * Generate the filter list changes using a separate function. 2364 */ 2365 if (is_filter_list_change) 2366 return (mld_v2_enqueue_filter_change(ifq, inm)); 2367 2368 if (type == MLD_DO_NOTHING) { 2369 CTR3(KTR_MLD, "%s: nothing to do for %s/%s", 2370 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2371 inm->in6m_ifp->if_xname); 2372 return (0); 2373 } 2374 2375 /* 2376 * If any sources are present, we must be able to fit at least 2377 * one in the trailing space of the tail packet's mbuf, 2378 * ideally more. 2379 */ 2380 minrec0len = sizeof(struct mldv2_record); 2381 if (record_has_sources) 2382 minrec0len += sizeof(struct in6_addr); 2383 2384 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, 2385 mld_rec_type_to_str(type), 2386 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2387 inm->in6m_ifp->if_xname); 2388 2389 /* 2390 * Check if we have a packet in the tail of the queue for this 2391 * group into which the first group record for this group will fit. 2392 * Otherwise allocate a new packet. 2393 * Always allocate leading space for IP6+RA+ICMPV6+REPORT. 2394 * Note: Group records for G/GSR query responses MUST be sent 2395 * in their own packet. 2396 */ 2397 m0 = ifq->ifq_tail; 2398 if (!is_group_query && 2399 m0 != NULL && 2400 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && 2401 (m0->m_pkthdr.len + minrec0len) < 2402 (ifp->if_mtu - MLD_MTUSPACE)) { 2403 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2404 sizeof(struct mldv2_record)) / 2405 sizeof(struct in6_addr); 2406 m = m0; 2407 CTR1(KTR_MLD, "%s: use existing packet", __func__); 2408 } else { 2409 if (_IF_QFULL(ifq)) { 2410 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2411 return (-ENOMEM); 2412 } 2413 m = NULL; 2414 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2415 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2416 if (!is_state_change && !is_group_query) 2417 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2418 if (m == NULL) 2419 m = m_gethdr(M_DONTWAIT, MT_DATA); 2420 if (m == NULL) 2421 return (-ENOMEM); 2422 2423 mld_save_context(m, ifp); 2424 2425 CTR1(KTR_MLD, "%s: allocated first packet", __func__); 2426 } 2427 2428 /* 2429 * Append group record. 2430 * If we have sources, we don't know how many yet. 2431 */ 2432 mr.mr_type = type; 2433 mr.mr_datalen = 0; 2434 mr.mr_numsrc = 0; 2435 mr.mr_addr = inm->in6m_addr; 2436 in6_clearscope(&mr.mr_addr); 2437 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2438 if (m != m0) 2439 m_freem(m); 2440 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2441 return (-ENOMEM); 2442 } 2443 nbytes += sizeof(struct mldv2_record); 2444 2445 /* 2446 * Append as many sources as will fit in the first packet. 2447 * If we are appending to a new packet, the chain allocation 2448 * may potentially use clusters; use m_getptr() in this case. 2449 * If we are appending to an existing packet, we need to obtain 2450 * a pointer to the group record after m_append(), in case a new 2451 * mbuf was allocated. 2452 * Only append sources which are in-mode at t1. If we are 2453 * transitioning to MCAST_UNDEFINED state on the group, do not 2454 * include source entries. 2455 * Only report recorded sources in our filter set when responding 2456 * to a group-source query. 2457 */ 2458 if (record_has_sources) { 2459 if (m == m0) { 2460 md = m_last(m); 2461 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2462 md->m_len - nbytes); 2463 } else { 2464 md = m_getptr(m, 0, &off); 2465 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2466 off); 2467 } 2468 msrcs = 0; 2469 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, 2470 nims) { 2471 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2472 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2473 now = im6s_get_mode(inm, ims, 1); 2474 CTR2(KTR_MLD, "%s: node is %d", __func__, now); 2475 if ((now != mode) || 2476 (now == mode && mode == MCAST_UNDEFINED)) { 2477 CTR1(KTR_MLD, "%s: skip node", __func__); 2478 continue; 2479 } 2480 if (is_source_query && ims->im6s_stp == 0) { 2481 CTR1(KTR_MLD, "%s: skip unrecorded node", 2482 __func__); 2483 continue; 2484 } 2485 CTR1(KTR_MLD, "%s: append node", __func__); 2486 if (!m_append(m, sizeof(struct in6_addr), 2487 (void *)&ims->im6s_addr)) { 2488 if (m != m0) 2489 m_freem(m); 2490 CTR1(KTR_MLD, "%s: m_append() failed.", 2491 __func__); 2492 return (-ENOMEM); 2493 } 2494 nbytes += sizeof(struct in6_addr); 2495 ++msrcs; 2496 if (msrcs == m0srcs) 2497 break; 2498 } 2499 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, 2500 msrcs); 2501 pmr->mr_numsrc = htons(msrcs); 2502 nbytes += (msrcs * sizeof(struct in6_addr)); 2503 } 2504 2505 if (is_source_query && msrcs == 0) { 2506 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); 2507 if (m != m0) 2508 m_freem(m); 2509 return (0); 2510 } 2511 2512 /* 2513 * We are good to go with first packet. 2514 */ 2515 if (m != m0) { 2516 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); 2517 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2518 _IF_ENQUEUE(ifq, m); 2519 } else 2520 m->m_pkthdr.PH_vt.vt_nrecs++; 2521 2522 /* 2523 * No further work needed if no source list in packet(s). 2524 */ 2525 if (!record_has_sources) 2526 return (nbytes); 2527 2528 /* 2529 * Whilst sources remain to be announced, we need to allocate 2530 * a new packet and fill out as many sources as will fit. 2531 * Always try for a cluster first. 2532 */ 2533 while (nims != NULL) { 2534 if (_IF_QFULL(ifq)) { 2535 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2536 return (-ENOMEM); 2537 } 2538 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2539 if (m == NULL) 2540 m = m_gethdr(M_DONTWAIT, MT_DATA); 2541 if (m == NULL) 2542 return (-ENOMEM); 2543 mld_save_context(m, ifp); 2544 md = m_getptr(m, 0, &off); 2545 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); 2546 CTR1(KTR_MLD, "%s: allocated next packet", __func__); 2547 2548 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2549 if (m != m0) 2550 m_freem(m); 2551 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2552 return (-ENOMEM); 2553 } 2554 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2555 nbytes += sizeof(struct mldv2_record); 2556 2557 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2558 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2559 2560 msrcs = 0; 2561 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2562 CTR2(KTR_MLD, "%s: visit node %s", 2563 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2564 now = im6s_get_mode(inm, ims, 1); 2565 if ((now != mode) || 2566 (now == mode && mode == MCAST_UNDEFINED)) { 2567 CTR1(KTR_MLD, "%s: skip node", __func__); 2568 continue; 2569 } 2570 if (is_source_query && ims->im6s_stp == 0) { 2571 CTR1(KTR_MLD, "%s: skip unrecorded node", 2572 __func__); 2573 continue; 2574 } 2575 CTR1(KTR_MLD, "%s: append node", __func__); 2576 if (!m_append(m, sizeof(struct in6_addr), 2577 (void *)&ims->im6s_addr)) { 2578 if (m != m0) 2579 m_freem(m); 2580 CTR1(KTR_MLD, "%s: m_append() failed.", 2581 __func__); 2582 return (-ENOMEM); 2583 } 2584 ++msrcs; 2585 if (msrcs == m0srcs) 2586 break; 2587 } 2588 pmr->mr_numsrc = htons(msrcs); 2589 nbytes += (msrcs * sizeof(struct in6_addr)); 2590 2591 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); 2592 _IF_ENQUEUE(ifq, m); 2593 } 2594 2595 return (nbytes); 2596 } 2597 2598 /* 2599 * Type used to mark record pass completion. 2600 * We exploit the fact we can cast to this easily from the 2601 * current filter modes on each ip_msource node. 2602 */ 2603 typedef enum { 2604 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2605 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2606 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2607 REC_FULL = REC_ALLOW | REC_BLOCK 2608 } rectype_t; 2609 2610 /* 2611 * Enqueue an MLDv2 filter list change to the given output queue. 2612 * 2613 * Source list filter state is held in an RB-tree. When the filter list 2614 * for a group is changed without changing its mode, we need to compute 2615 * the deltas between T0 and T1 for each source in the filter set, 2616 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2617 * 2618 * As we may potentially queue two record types, and the entire R-B tree 2619 * needs to be walked at once, we break this out into its own function 2620 * so we can generate a tightly packed queue of packets. 2621 * 2622 * XXX This could be written to only use one tree walk, although that makes 2623 * serializing into the mbuf chains a bit harder. For now we do two walks 2624 * which makes things easier on us, and it may or may not be harder on 2625 * the L2 cache. 2626 * 2627 * If successful the size of all data appended to the queue is returned, 2628 * otherwise an error code less than zero is returned, or zero if 2629 * no record(s) were appended. 2630 */ 2631 static int 2632 mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm) 2633 { 2634 static const int MINRECLEN = 2635 sizeof(struct mldv2_record) + sizeof(struct in6_addr); 2636 struct ifnet *ifp; 2637 struct mldv2_record mr; 2638 struct mldv2_record *pmr; 2639 struct ip6_msource *ims, *nims; 2640 struct mbuf *m, *m0, *md; 2641 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 2642 int nallow, nblock; 2643 uint8_t mode, now, then; 2644 rectype_t crt, drt, nrt; 2645 #ifdef KTR 2646 char ip6tbuf[INET6_ADDRSTRLEN]; 2647 #endif 2648 2649 IN6_MULTI_LOCK_ASSERT(); 2650 2651 if (inm->in6m_nsrc == 0 || 2652 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) 2653 return (0); 2654 2655 ifp = inm->in6m_ifp; /* interface */ 2656 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ 2657 crt = REC_NONE; /* current group record type */ 2658 drt = REC_NONE; /* mask of completed group record types */ 2659 nrt = REC_NONE; /* record type for current node */ 2660 m0srcs = 0; /* # source which will fit in current mbuf chain */ 2661 npbytes = 0; /* # of bytes appended this packet */ 2662 nbytes = 0; /* # of bytes appended to group's state-change queue */ 2663 rsrcs = 0; /* # sources encoded in current record */ 2664 schanged = 0; /* # nodes encoded in overall filter change */ 2665 nallow = 0; /* # of source entries in ALLOW_NEW */ 2666 nblock = 0; /* # of source entries in BLOCK_OLD */ 2667 nims = NULL; /* next tree node pointer */ 2668 2669 /* 2670 * For each possible filter record mode. 2671 * The first kind of source we encounter tells us which 2672 * is the first kind of record we start appending. 2673 * If a node transitioned to UNDEFINED at t1, its mode is treated 2674 * as the inverse of the group's filter mode. 2675 */ 2676 while (drt != REC_FULL) { 2677 do { 2678 m0 = ifq->ifq_tail; 2679 if (m0 != NULL && 2680 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 2681 MLD_V2_REPORT_MAXRECS) && 2682 (m0->m_pkthdr.len + MINRECLEN) < 2683 (ifp->if_mtu - MLD_MTUSPACE)) { 2684 m = m0; 2685 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2686 sizeof(struct mldv2_record)) / 2687 sizeof(struct in6_addr); 2688 CTR1(KTR_MLD, 2689 "%s: use previous packet", __func__); 2690 } else { 2691 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR); 2692 if (m == NULL) 2693 m = m_gethdr(M_DONTWAIT, MT_DATA); 2694 if (m == NULL) { 2695 CTR1(KTR_MLD, 2696 "%s: m_get*() failed", __func__); 2697 return (-ENOMEM); 2698 } 2699 m->m_pkthdr.PH_vt.vt_nrecs = 0; 2700 mld_save_context(m, ifp); 2701 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2702 sizeof(struct mldv2_record)) / 2703 sizeof(struct in6_addr); 2704 npbytes = 0; 2705 CTR1(KTR_MLD, 2706 "%s: allocated new packet", __func__); 2707 } 2708 /* 2709 * Append the MLD group record header to the 2710 * current packet's data area. 2711 * Recalculate pointer to free space for next 2712 * group record, in case m_append() allocated 2713 * a new mbuf or cluster. 2714 */ 2715 memset(&mr, 0, sizeof(mr)); 2716 mr.mr_addr = inm->in6m_addr; 2717 in6_clearscope(&mr.mr_addr); 2718 if (!m_append(m, sizeof(mr), (void *)&mr)) { 2719 if (m != m0) 2720 m_freem(m); 2721 CTR1(KTR_MLD, 2722 "%s: m_append() failed", __func__); 2723 return (-ENOMEM); 2724 } 2725 npbytes += sizeof(struct mldv2_record); 2726 if (m != m0) { 2727 /* new packet; offset in chain */ 2728 md = m_getptr(m, npbytes - 2729 sizeof(struct mldv2_record), &off); 2730 pmr = (struct mldv2_record *)(mtod(md, 2731 uint8_t *) + off); 2732 } else { 2733 /* current packet; offset from last append */ 2734 md = m_last(m); 2735 pmr = (struct mldv2_record *)(mtod(md, 2736 uint8_t *) + md->m_len - 2737 sizeof(struct mldv2_record)); 2738 } 2739 /* 2740 * Begin walking the tree for this record type 2741 * pass, or continue from where we left off 2742 * previously if we had to allocate a new packet. 2743 * Only report deltas in-mode at t1. 2744 * We need not report included sources as allowed 2745 * if we are in inclusive mode on the group, 2746 * however the converse is not true. 2747 */ 2748 rsrcs = 0; 2749 if (nims == NULL) { 2750 nims = RB_MIN(ip6_msource_tree, 2751 &inm->in6m_srcs); 2752 } 2753 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2754 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2755 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2756 now = im6s_get_mode(inm, ims, 1); 2757 then = im6s_get_mode(inm, ims, 0); 2758 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", 2759 __func__, then, now); 2760 if (now == then) { 2761 CTR1(KTR_MLD, 2762 "%s: skip unchanged", __func__); 2763 continue; 2764 } 2765 if (mode == MCAST_EXCLUDE && 2766 now == MCAST_INCLUDE) { 2767 CTR1(KTR_MLD, 2768 "%s: skip IN src on EX group", 2769 __func__); 2770 continue; 2771 } 2772 nrt = (rectype_t)now; 2773 if (nrt == REC_NONE) 2774 nrt = (rectype_t)(~mode & REC_FULL); 2775 if (schanged++ == 0) { 2776 crt = nrt; 2777 } else if (crt != nrt) 2778 continue; 2779 if (!m_append(m, sizeof(struct in6_addr), 2780 (void *)&ims->im6s_addr)) { 2781 if (m != m0) 2782 m_freem(m); 2783 CTR1(KTR_MLD, 2784 "%s: m_append() failed", __func__); 2785 return (-ENOMEM); 2786 } 2787 nallow += !!(crt == REC_ALLOW); 2788 nblock += !!(crt == REC_BLOCK); 2789 if (++rsrcs == m0srcs) 2790 break; 2791 } 2792 /* 2793 * If we did not append any tree nodes on this 2794 * pass, back out of allocations. 2795 */ 2796 if (rsrcs == 0) { 2797 npbytes -= sizeof(struct mldv2_record); 2798 if (m != m0) { 2799 CTR1(KTR_MLD, 2800 "%s: m_free(m)", __func__); 2801 m_freem(m); 2802 } else { 2803 CTR1(KTR_MLD, 2804 "%s: m_adj(m, -mr)", __func__); 2805 m_adj(m, -((int)sizeof( 2806 struct mldv2_record))); 2807 } 2808 continue; 2809 } 2810 npbytes += (rsrcs * sizeof(struct in6_addr)); 2811 if (crt == REC_ALLOW) 2812 pmr->mr_type = MLD_ALLOW_NEW_SOURCES; 2813 else if (crt == REC_BLOCK) 2814 pmr->mr_type = MLD_BLOCK_OLD_SOURCES; 2815 pmr->mr_numsrc = htons(rsrcs); 2816 /* 2817 * Count the new group record, and enqueue this 2818 * packet if it wasn't already queued. 2819 */ 2820 m->m_pkthdr.PH_vt.vt_nrecs++; 2821 if (m != m0) 2822 _IF_ENQUEUE(ifq, m); 2823 nbytes += npbytes; 2824 } while (nims != NULL); 2825 drt |= crt; 2826 crt = (~crt & REC_FULL); 2827 } 2828 2829 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 2830 nallow, nblock); 2831 2832 return (nbytes); 2833 } 2834 2835 static int 2836 mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq) 2837 { 2838 struct ifqueue *gq; 2839 struct mbuf *m; /* pending state-change */ 2840 struct mbuf *m0; /* copy of pending state-change */ 2841 struct mbuf *mt; /* last state-change in packet */ 2842 int docopy, domerge; 2843 u_int recslen; 2844 2845 docopy = 0; 2846 domerge = 0; 2847 recslen = 0; 2848 2849 IN6_MULTI_LOCK_ASSERT(); 2850 MLD_LOCK_ASSERT(); 2851 2852 /* 2853 * If there are further pending retransmissions, make a writable 2854 * copy of each queued state-change message before merging. 2855 */ 2856 if (inm->in6m_scrv > 0) 2857 docopy = 1; 2858 2859 gq = &inm->in6m_scq; 2860 #ifdef KTR 2861 if (gq->ifq_head == NULL) { 2862 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", 2863 __func__, inm); 2864 } 2865 #endif 2866 2867 m = gq->ifq_head; 2868 while (m != NULL) { 2869 /* 2870 * Only merge the report into the current packet if 2871 * there is sufficient space to do so; an MLDv2 report 2872 * packet may only contain 65,535 group records. 2873 * Always use a simple mbuf chain concatentation to do this, 2874 * as large state changes for single groups may have 2875 * allocated clusters. 2876 */ 2877 domerge = 0; 2878 mt = ifscq->ifq_tail; 2879 if (mt != NULL) { 2880 recslen = m_length(m, NULL); 2881 2882 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 2883 m->m_pkthdr.PH_vt.vt_nrecs <= 2884 MLD_V2_REPORT_MAXRECS) && 2885 (mt->m_pkthdr.len + recslen <= 2886 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) 2887 domerge = 1; 2888 } 2889 2890 if (!domerge && _IF_QFULL(gq)) { 2891 CTR2(KTR_MLD, 2892 "%s: outbound queue full, skipping whole packet %p", 2893 __func__, m); 2894 mt = m->m_nextpkt; 2895 if (!docopy) 2896 m_freem(m); 2897 m = mt; 2898 continue; 2899 } 2900 2901 if (!docopy) { 2902 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); 2903 _IF_DEQUEUE(gq, m0); 2904 m = m0->m_nextpkt; 2905 } else { 2906 CTR2(KTR_MLD, "%s: copying %p", __func__, m); 2907 m0 = m_dup(m, M_NOWAIT); 2908 if (m0 == NULL) 2909 return (ENOMEM); 2910 m0->m_nextpkt = NULL; 2911 m = m->m_nextpkt; 2912 } 2913 2914 if (!domerge) { 2915 CTR3(KTR_MLD, "%s: queueing %p to ifscq %p)", 2916 __func__, m0, ifscq); 2917 _IF_ENQUEUE(ifscq, m0); 2918 } else { 2919 struct mbuf *mtl; /* last mbuf of packet mt */ 2920 2921 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", 2922 __func__, m0, mt); 2923 2924 mtl = m_last(mt); 2925 m0->m_flags &= ~M_PKTHDR; 2926 mt->m_pkthdr.len += recslen; 2927 mt->m_pkthdr.PH_vt.vt_nrecs += 2928 m0->m_pkthdr.PH_vt.vt_nrecs; 2929 2930 mtl->m_next = m0; 2931 } 2932 } 2933 2934 return (0); 2935 } 2936 2937 /* 2938 * Respond to a pending MLDv2 General Query. 2939 */ 2940 static void 2941 mld_v2_dispatch_general_query(struct mld_ifinfo *mli) 2942 { 2943 INIT_VNET_INET6(curvnet); 2944 struct ifmultiaddr *ifma, *tifma; 2945 struct ifnet *ifp; 2946 struct in6_multi *inm; 2947 int retval; 2948 2949 IN6_MULTI_LOCK_ASSERT(); 2950 MLD_LOCK_ASSERT(); 2951 2952 KASSERT(mli->mli_version == MLD_VERSION_2, 2953 ("%s: called when version %d", __func__, mli->mli_version)); 2954 2955 ifp = mli->mli_ifp; 2956 2957 IF_ADDR_LOCK(ifp); 2958 TAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, tifma) { 2959 if (ifma->ifma_addr->sa_family != AF_INET6 || 2960 ifma->ifma_protospec == NULL) 2961 continue; 2962 2963 inm = (struct in6_multi *)ifma->ifma_protospec; 2964 KASSERT(ifp == inm->in6m_ifp, 2965 ("%s: inconsistent ifp", __func__)); 2966 2967 switch (inm->in6m_state) { 2968 case MLD_NOT_MEMBER: 2969 case MLD_SILENT_MEMBER: 2970 break; 2971 case MLD_REPORTING_MEMBER: 2972 case MLD_IDLE_MEMBER: 2973 case MLD_LAZY_MEMBER: 2974 case MLD_SLEEPING_MEMBER: 2975 case MLD_AWAKENING_MEMBER: 2976 inm->in6m_state = MLD_REPORTING_MEMBER; 2977 retval = mld_v2_enqueue_group_record(&mli->mli_gq, 2978 inm, 0, 0, 0); 2979 CTR2(KTR_MLD, "%s: enqueue record = %d", 2980 __func__, retval); 2981 break; 2982 case MLD_G_QUERY_PENDING_MEMBER: 2983 case MLD_SG_QUERY_PENDING_MEMBER: 2984 case MLD_LEAVING_MEMBER: 2985 break; 2986 } 2987 } 2988 IF_ADDR_UNLOCK(ifp); 2989 2990 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); 2991 2992 /* 2993 * Slew transmission of bursts over 500ms intervals. 2994 */ 2995 if (mli->mli_gq.ifq_head != NULL) { 2996 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( 2997 MLD_RESPONSE_BURST_INTERVAL); 2998 V_interface_timers_running6 = 1; 2999 } 3000 } 3001 3002 /* 3003 * Transmit the next pending message in the output queue. 3004 * 3005 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3006 * MRT: Nothing needs to be done, as MLD traffic is always local to 3007 * a link and uses a link-scope multicast address. 3008 */ 3009 static void 3010 mld_dispatch_packet(struct mbuf *m) 3011 { 3012 struct ip6_moptions im6o; 3013 struct ifnet *ifp; 3014 struct ifnet *oifp; 3015 struct mbuf *m0; 3016 struct mbuf *md; 3017 struct ip6_hdr *ip6; 3018 struct mld_hdr *mld; 3019 int error; 3020 int off; 3021 int type; 3022 uint32_t ifindex; 3023 3024 CTR2(KTR_MLD, "%s: transmit %p", __func__, m); 3025 3026 /* 3027 * Set VNET image pointer from enqueued mbuf chain 3028 * before doing anything else. Whilst we use interface 3029 * indexes to guard against interface detach, they are 3030 * unique to each VIMAGE and must be retrieved. 3031 */ 3032 INIT_VNET_NET(curvnet); 3033 INIT_VNET_INET6(curvnet); 3034 ifindex = mld_restore_context(m); 3035 3036 /* 3037 * Check if the ifnet still exists. This limits the scope of 3038 * any race in the absence of a global ifp lock for low cost 3039 * (an array lookup). 3040 */ 3041 ifp = ifnet_byindex(ifindex); 3042 if (ifp == NULL) { 3043 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", 3044 __func__, m, ifindex); 3045 m_freem(m); 3046 IP6STAT_INC(ip6s_noroute); 3047 goto out; 3048 } 3049 3050 im6o.im6o_multicast_hlim = 1; 3051 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); 3052 im6o.im6o_multicast_ifp = ifp; 3053 3054 if (m->m_flags & M_MLDV1) { 3055 m0 = m; 3056 } else { 3057 m0 = mld_v2_encap_report(ifp, m); 3058 if (m0 == NULL) { 3059 CTR2(KTR_MLD, "%s: dropped %p", __func__, m); 3060 m_freem(m); 3061 IP6STAT_INC(ip6s_odropped); 3062 goto out; 3063 } 3064 } 3065 3066 mld_scrub_context(m0); 3067 m->m_flags &= ~(M_PROTOFLAGS); 3068 m0->m_pkthdr.rcvif = V_loif; 3069 3070 ip6 = mtod(m0, struct ip6_hdr *); 3071 #if 0 3072 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */ 3073 #else 3074 /* 3075 * XXX XXX Break some KPI rules to prevent an LOR which would 3076 * occur if we called in6_setscope() at transmission. 3077 * See comments at top of file. 3078 */ 3079 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index); 3080 #endif 3081 3082 /* 3083 * Retrieve the ICMPv6 type before handoff to ip6_output(), 3084 * so we can bump the stats. 3085 */ 3086 md = m_getptr(m0, sizeof(struct ip6_hdr), &off); 3087 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off); 3088 type = mld->mld_type; 3089 3090 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o, 3091 &oifp, NULL); 3092 if (error) { 3093 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error); 3094 goto out; 3095 } 3096 ICMP6STAT_INC(icp6s_outhist[type]); 3097 if (oifp != NULL) { 3098 icmp6_ifstat_inc(oifp, ifs6_out_msg); 3099 switch (type) { 3100 case MLD_LISTENER_REPORT: 3101 case MLDV2_LISTENER_REPORT: 3102 icmp6_ifstat_inc(oifp, ifs6_out_mldreport); 3103 break; 3104 case MLD_LISTENER_DONE: 3105 icmp6_ifstat_inc(oifp, ifs6_out_mlddone); 3106 break; 3107 } 3108 } 3109 out: 3110 return; 3111 } 3112 3113 /* 3114 * Encapsulate an MLDv2 report. 3115 * 3116 * KAME IPv6 requires that hop-by-hop options be passed separately, 3117 * and that the IPv6 header be prepended in a separate mbuf. 3118 * 3119 * Returns a pointer to the new mbuf chain head, or NULL if the 3120 * allocation failed. 3121 */ 3122 static struct mbuf * 3123 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m) 3124 { 3125 struct mbuf *mh; 3126 struct mldv2_report *mld; 3127 struct ip6_hdr *ip6; 3128 struct in6_ifaddr *ia; 3129 int mldreclen; 3130 3131 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 3132 KASSERT((m->m_flags & M_PKTHDR), 3133 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3134 3135 /* 3136 * RFC3590: OK to send as :: or tentative during DAD. 3137 */ 3138 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 3139 if (ia == NULL) 3140 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__); 3141 3142 MGETHDR(mh, M_DONTWAIT, MT_HEADER); 3143 if (mh == NULL) { 3144 m_freem(m); 3145 return (NULL); 3146 } 3147 MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); 3148 3149 mldreclen = m_length(m, NULL); 3150 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); 3151 3152 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report); 3153 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + 3154 sizeof(struct mldv2_report) + mldreclen; 3155 3156 ip6 = mtod(mh, struct ip6_hdr *); 3157 ip6->ip6_flow = 0; 3158 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3159 ip6->ip6_vfc |= IPV6_VERSION; 3160 ip6->ip6_nxt = IPPROTO_ICMPV6; 3161 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 3162 ip6->ip6_dst = in6addr_linklocal_allv2routers; 3163 /* scope ID will be set in netisr */ 3164 3165 mld = (struct mldv2_report *)(ip6 + 1); 3166 mld->mld_type = MLDV2_LISTENER_REPORT; 3167 mld->mld_code = 0; 3168 mld->mld_cksum = 0; 3169 mld->mld_v2_reserved = 0; 3170 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs); 3171 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3172 3173 mh->m_next = m; 3174 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 3175 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen); 3176 3177 return (mh); 3178 } 3179 3180 #ifdef KTR 3181 static char * 3182 mld_rec_type_to_str(const int type) 3183 { 3184 3185 switch (type) { 3186 case MLD_CHANGE_TO_EXCLUDE_MODE: 3187 return "TO_EX"; 3188 break; 3189 case MLD_CHANGE_TO_INCLUDE_MODE: 3190 return "TO_IN"; 3191 break; 3192 case MLD_MODE_IS_EXCLUDE: 3193 return "MODE_EX"; 3194 break; 3195 case MLD_MODE_IS_INCLUDE: 3196 return "MODE_IN"; 3197 break; 3198 case MLD_ALLOW_NEW_SOURCES: 3199 return "ALLOW_NEW"; 3200 break; 3201 case MLD_BLOCK_OLD_SOURCES: 3202 return "BLOCK_OLD"; 3203 break; 3204 default: 3205 break; 3206 } 3207 return "unknown"; 3208 } 3209 #endif 3210 3211 static void 3212 mld_sysinit(void) 3213 { 3214 3215 CTR1(KTR_MLD, "%s: initializing", __func__); 3216 MLD_LOCK_INIT(); 3217 3218 ip6_initpktopts(&mld_po); 3219 mld_po.ip6po_hlim = 1; 3220 mld_po.ip6po_hbh = &mld_ra.hbh; 3221 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER; 3222 mld_po.ip6po_flags = IP6PO_DONTFRAG; 3223 } 3224 3225 static void 3226 mld_sysuninit(void) 3227 { 3228 3229 CTR1(KTR_MLD, "%s: tearing down", __func__); 3230 MLD_LOCK_DESTROY(); 3231 } 3232 3233 /* 3234 * Initialize an MLDv2 instance. 3235 * VIMAGE: Assumes curvnet set by caller and called per vimage. 3236 */ 3237 static int 3238 vnet_mld_iattach(const void *unused __unused) 3239 { 3240 INIT_VNET_INET6(curvnet); 3241 3242 CTR1(KTR_MLD, "%s: initializing", __func__); 3243 3244 LIST_INIT(&V_mli_head); 3245 3246 V_current_state_timers_running6 = 0; 3247 V_interface_timers_running6 = 0; 3248 V_state_change_timers_running6 = 0; 3249 3250 /* 3251 * Initialize sysctls to default values. 3252 */ 3253 V_mld_gsrdelay.tv_sec = 10; 3254 V_mld_gsrdelay.tv_usec = 0; 3255 3256 return (0); 3257 } 3258 3259 static int 3260 vnet_mld_idetach(const void *unused __unused) 3261 { 3262 #ifdef INVARIANTS 3263 INIT_VNET_INET6(curvnet); 3264 #endif 3265 3266 CTR1(KTR_MLD, "%s: tearing down", __func__); 3267 3268 KASSERT(LIST_EMPTY(&V_mli_head), 3269 ("%s: mli list not empty; ifnets not detached?", __func__)); 3270 3271 return (0); 3272 } 3273 3274 #ifndef VIMAGE_GLOBALS 3275 static vnet_modinfo_t vnet_mld_modinfo = { 3276 .vmi_id = VNET_MOD_MLD, 3277 .vmi_name = "mld", 3278 .vmi_dependson = VNET_MOD_INET6, 3279 .vmi_iattach = vnet_mld_iattach, 3280 .vmi_idetach = vnet_mld_idetach 3281 }; 3282 #endif 3283 3284 static int 3285 mld_modevent(module_t mod, int type, void *unused __unused) 3286 { 3287 3288 switch (type) { 3289 case MOD_LOAD: 3290 mld_sysinit(); 3291 #ifndef VIMAGE_GLOBALS 3292 vnet_mod_register(&vnet_mld_modinfo); 3293 #else 3294 vnet_mld_iattach(NULL); 3295 #endif 3296 break; 3297 case MOD_UNLOAD: 3298 #ifndef VIMAGE_GLOBALS 3299 #ifdef NOTYET 3300 vnet_mod_deregister(&vnet_mld_modinfo); 3301 #endif 3302 #else 3303 vnet_mld_idetach(NULL); 3304 #endif 3305 mld_sysuninit(); 3306 break; 3307 default: 3308 return (EOPNOTSUPP); 3309 } 3310 return (0); 3311 } 3312 3313 static moduledata_t mld_mod = { 3314 "mld", 3315 mld_modevent, 3316 0 3317 }; 3318 DECLARE_MODULE(mld, mld_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 3319