1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2009 Bruce Simpson. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. The name of the author may not be used to endorse or promote 15 * products derived from this software without specific prior written 16 * permission. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28 * SUCH DAMAGE. 29 * 30 * $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $ 31 */ 32 33 /*- 34 * Copyright (c) 1988 Stephen Deering. 35 * Copyright (c) 1992, 1993 36 * The Regents of the University of California. All rights reserved. 37 * 38 * This code is derived from software contributed to Berkeley by 39 * Stephen Deering of Stanford University. 40 * 41 * Redistribution and use in source and binary forms, with or without 42 * modification, are permitted provided that the following conditions 43 * are met: 44 * 1. Redistributions of source code must retain the above copyright 45 * notice, this list of conditions and the following disclaimer. 46 * 2. Redistributions in binary form must reproduce the above copyright 47 * notice, this list of conditions and the following disclaimer in the 48 * documentation and/or other materials provided with the distribution. 49 * 3. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * @(#)igmp.c 8.1 (Berkeley) 7/19/93 66 */ 67 68 #include <sys/cdefs.h> 69 __FBSDID("$FreeBSD$"); 70 71 #include "opt_inet.h" 72 #include "opt_inet6.h" 73 74 #include <sys/param.h> 75 #include <sys/systm.h> 76 #include <sys/mbuf.h> 77 #include <sys/socket.h> 78 #include <sys/protosw.h> 79 #include <sys/sysctl.h> 80 #include <sys/kernel.h> 81 #include <sys/callout.h> 82 #include <sys/malloc.h> 83 #include <sys/module.h> 84 #include <sys/ktr.h> 85 86 #include <net/if.h> 87 #include <net/if_var.h> 88 #include <net/route.h> 89 #include <net/vnet.h> 90 91 #include <netinet/in.h> 92 #include <netinet/in_var.h> 93 #include <netinet6/in6_var.h> 94 #include <netinet/ip6.h> 95 #include <netinet6/ip6_var.h> 96 #include <netinet6/scope6_var.h> 97 #include <netinet/icmp6.h> 98 #include <netinet6/mld6.h> 99 #include <netinet6/mld6_var.h> 100 101 #include <security/mac/mac_framework.h> 102 103 #ifndef KTR_MLD 104 #define KTR_MLD KTR_INET6 105 #endif 106 107 static struct mld_ifsoftc * 108 mli_alloc_locked(struct ifnet *); 109 static void mli_delete_locked(const struct ifnet *); 110 static void mld_dispatch_packet(struct mbuf *); 111 static void mld_dispatch_queue(struct mbufq *, int); 112 static void mld_final_leave(struct in6_multi *, struct mld_ifsoftc *); 113 static void mld_fasttimo_vnet(void); 114 static int mld_handle_state_change(struct in6_multi *, 115 struct mld_ifsoftc *); 116 static int mld_initial_join(struct in6_multi *, struct mld_ifsoftc *, 117 const int); 118 #ifdef KTR 119 static char * mld_rec_type_to_str(const int); 120 #endif 121 static void mld_set_version(struct mld_ifsoftc *, const int); 122 static void mld_slowtimo_vnet(void); 123 static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *, 124 /*const*/ struct mld_hdr *); 125 static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *, 126 /*const*/ struct mld_hdr *); 127 static void mld_v1_process_group_timer(struct mld_ifsoftc *, 128 struct in6_multi *); 129 static void mld_v1_process_querier_timers(struct mld_ifsoftc *); 130 static int mld_v1_transmit_report(struct in6_multi *, const int); 131 static void mld_v1_update_group(struct in6_multi *, const int); 132 static void mld_v2_cancel_link_timers(struct mld_ifsoftc *); 133 static void mld_v2_dispatch_general_query(struct mld_ifsoftc *); 134 static struct mbuf * 135 mld_v2_encap_report(struct ifnet *, struct mbuf *); 136 static int mld_v2_enqueue_filter_change(struct mbufq *, 137 struct in6_multi *); 138 static int mld_v2_enqueue_group_record(struct mbufq *, 139 struct in6_multi *, const int, const int, const int, 140 const int); 141 static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *, 142 struct mbuf *, const int, const int); 143 static int mld_v2_merge_state_changes(struct in6_multi *, 144 struct mbufq *); 145 static void mld_v2_process_group_timers(struct mld_ifsoftc *, 146 struct mbufq *, struct mbufq *, 147 struct in6_multi *, const int); 148 static int mld_v2_process_group_query(struct in6_multi *, 149 struct mld_ifsoftc *mli, int, struct mbuf *, const int); 150 static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS); 151 static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS); 152 153 /* 154 * Normative references: RFC 2710, RFC 3590, RFC 3810. 155 * 156 * Locking: 157 * * The MLD subsystem lock ends up being system-wide for the moment, 158 * but could be per-VIMAGE later on. 159 * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. 160 * Any may be taken independently; if any are held at the same 161 * time, the above lock order must be followed. 162 * * IN6_MULTI_LOCK covers in_multi. 163 * * MLD_LOCK covers per-link state and any global variables in this file. 164 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of 165 * per-link state iterators. 166 * 167 * XXX LOR PREVENTION 168 * A special case for IPv6 is the in6_setscope() routine. ip6_output() 169 * will not accept an ifp; it wants an embedded scope ID, unlike 170 * ip_output(), which happily takes the ifp given to it. The embedded 171 * scope ID is only used by MLD to select the outgoing interface. 172 * 173 * During interface attach and detach, MLD will take MLD_LOCK *after* 174 * the IF_AFDATA_LOCK. 175 * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call 176 * it with MLD_LOCK held without triggering an LOR. A netisr with indirect 177 * dispatch could work around this, but we'd rather not do that, as it 178 * can introduce other races. 179 * 180 * As such, we exploit the fact that the scope ID is just the interface 181 * index, and embed it in the IPv6 destination address accordingly. 182 * This is potentially NOT VALID for MLDv1 reports, as they 183 * are always sent to the multicast group itself; as MLDv2 184 * reports are always sent to ff02::16, this is not an issue 185 * when MLDv2 is in use. 186 * 187 * This does not however eliminate the LOR when ip6_output() itself 188 * calls in6_setscope() internally whilst MLD_LOCK is held. This will 189 * trigger a LOR warning in WITNESS when the ifnet is detached. 190 * 191 * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given 192 * how it's used across the network stack. Here we're simply exploiting 193 * the fact that MLD runs at a similar layer in the stack to scope6.c. 194 * 195 * VIMAGE: 196 * * Each in6_multi corresponds to an ifp, and each ifp corresponds 197 * to a vnet in ifp->if_vnet. 198 */ 199 static struct mtx mld_mtx; 200 static MALLOC_DEFINE(M_MLD, "mld", "mld state"); 201 202 #define MLD_EMBEDSCOPE(pin6, zoneid) \ 203 if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \ 204 IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \ 205 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \ 206 207 /* 208 * VIMAGE-wide globals. 209 */ 210 static VNET_DEFINE(struct timeval, mld_gsrdelay) = {10, 0}; 211 static VNET_DEFINE(LIST_HEAD(, mld_ifsoftc), mli_head); 212 static VNET_DEFINE(int, interface_timers_running6); 213 static VNET_DEFINE(int, state_change_timers_running6); 214 static VNET_DEFINE(int, current_state_timers_running6); 215 216 #define V_mld_gsrdelay VNET(mld_gsrdelay) 217 #define V_mli_head VNET(mli_head) 218 #define V_interface_timers_running6 VNET(interface_timers_running6) 219 #define V_state_change_timers_running6 VNET(state_change_timers_running6) 220 #define V_current_state_timers_running6 VNET(current_state_timers_running6) 221 222 SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */ 223 224 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0, 225 "IPv6 Multicast Listener Discovery"); 226 227 /* 228 * Virtualized sysctls. 229 */ 230 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay, 231 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, 232 &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I", 233 "Rate limit for MLDv2 Group-and-Source queries in seconds"); 234 235 /* 236 * Non-virtualized sysctls. 237 */ 238 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, 239 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo, 240 "Per-interface MLDv2 state"); 241 242 static int mld_v1enable = 1; 243 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN, 244 &mld_v1enable, 0, "Enable fallback to MLDv1"); 245 246 static int mld_use_allow = 1; 247 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN, 248 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves"); 249 250 /* 251 * Packed Router Alert option structure declaration. 252 */ 253 struct mld_raopt { 254 struct ip6_hbh hbh; 255 struct ip6_opt pad; 256 struct ip6_opt_router ra; 257 } __packed; 258 259 /* 260 * Router Alert hop-by-hop option header. 261 */ 262 static struct mld_raopt mld_ra = { 263 .hbh = { 0, 0 }, 264 .pad = { .ip6o_type = IP6OPT_PADN, 0 }, 265 .ra = { 266 .ip6or_type = IP6OPT_ROUTER_ALERT, 267 .ip6or_len = IP6OPT_RTALERT_LEN - 2, 268 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF), 269 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF) 270 } 271 }; 272 static struct ip6_pktopts mld_po; 273 274 static __inline void 275 mld_save_context(struct mbuf *m, struct ifnet *ifp) 276 { 277 278 #ifdef VIMAGE 279 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; 280 #endif /* VIMAGE */ 281 m->m_pkthdr.flowid = ifp->if_index; 282 } 283 284 static __inline void 285 mld_scrub_context(struct mbuf *m) 286 { 287 288 m->m_pkthdr.PH_loc.ptr = NULL; 289 m->m_pkthdr.flowid = 0; 290 } 291 292 /* 293 * Restore context from a queued output chain. 294 * Return saved ifindex. 295 * 296 * VIMAGE: The assertion is there to make sure that we 297 * actually called CURVNET_SET() with what's in the mbuf chain. 298 */ 299 static __inline uint32_t 300 mld_restore_context(struct mbuf *m) 301 { 302 303 #if defined(VIMAGE) && defined(INVARIANTS) 304 KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr, 305 ("%s: called when curvnet was not restored: cuvnet %p m ptr %p", 306 __func__, curvnet, m->m_pkthdr.PH_loc.ptr)); 307 #endif 308 return (m->m_pkthdr.flowid); 309 } 310 311 /* 312 * Retrieve or set threshold between group-source queries in seconds. 313 * 314 * VIMAGE: Assume curvnet set by caller. 315 * SMPng: NOTE: Serialized by MLD lock. 316 */ 317 static int 318 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS) 319 { 320 int error; 321 int i; 322 323 error = sysctl_wire_old_buffer(req, sizeof(int)); 324 if (error) 325 return (error); 326 327 MLD_LOCK(); 328 329 i = V_mld_gsrdelay.tv_sec; 330 331 error = sysctl_handle_int(oidp, &i, 0, req); 332 if (error || !req->newptr) 333 goto out_locked; 334 335 if (i < -1 || i >= 60) { 336 error = EINVAL; 337 goto out_locked; 338 } 339 340 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d", 341 V_mld_gsrdelay.tv_sec, i); 342 V_mld_gsrdelay.tv_sec = i; 343 344 out_locked: 345 MLD_UNLOCK(); 346 return (error); 347 } 348 349 /* 350 * Expose struct mld_ifsoftc to userland, keyed by ifindex. 351 * For use by ifmcstat(8). 352 * 353 * SMPng: NOTE: Does an unlocked ifindex space read. 354 * VIMAGE: Assume curvnet set by caller. The node handler itself 355 * is not directly virtualized. 356 */ 357 static int 358 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS) 359 { 360 int *name; 361 int error; 362 u_int namelen; 363 struct ifnet *ifp; 364 struct mld_ifsoftc *mli; 365 366 name = (int *)arg1; 367 namelen = arg2; 368 369 if (req->newptr != NULL) 370 return (EPERM); 371 372 if (namelen != 1) 373 return (EINVAL); 374 375 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo)); 376 if (error) 377 return (error); 378 379 IN6_MULTI_LOCK(); 380 MLD_LOCK(); 381 382 if (name[0] <= 0 || name[0] > V_if_index) { 383 error = ENOENT; 384 goto out_locked; 385 } 386 387 error = ENOENT; 388 389 ifp = ifnet_byindex(name[0]); 390 if (ifp == NULL) 391 goto out_locked; 392 393 LIST_FOREACH(mli, &V_mli_head, mli_link) { 394 if (ifp == mli->mli_ifp) { 395 struct mld_ifinfo info; 396 397 info.mli_version = mli->mli_version; 398 info.mli_v1_timer = mli->mli_v1_timer; 399 info.mli_v2_timer = mli->mli_v2_timer; 400 info.mli_flags = mli->mli_flags; 401 info.mli_rv = mli->mli_rv; 402 info.mli_qi = mli->mli_qi; 403 info.mli_qri = mli->mli_qri; 404 info.mli_uri = mli->mli_uri; 405 error = SYSCTL_OUT(req, &info, sizeof(info)); 406 break; 407 } 408 } 409 410 out_locked: 411 MLD_UNLOCK(); 412 IN6_MULTI_UNLOCK(); 413 return (error); 414 } 415 416 /* 417 * Dispatch an entire queue of pending packet chains. 418 * VIMAGE: Assumes the vnet pointer has been set. 419 */ 420 static void 421 mld_dispatch_queue(struct mbufq *mq, int limit) 422 { 423 struct mbuf *m; 424 425 while ((m = mbufq_dequeue(mq)) != NULL) { 426 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m); 427 mld_dispatch_packet(m); 428 if (--limit == 0) 429 break; 430 } 431 } 432 433 /* 434 * Filter outgoing MLD report state by group. 435 * 436 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1) 437 * and node-local addresses. However, kernel and socket consumers 438 * always embed the KAME scope ID in the address provided, so strip it 439 * when performing comparison. 440 * Note: This is not the same as the *multicast* scope. 441 * 442 * Return zero if the given group is one for which MLD reports 443 * should be suppressed, or non-zero if reports should be issued. 444 */ 445 static __inline int 446 mld_is_addr_reported(const struct in6_addr *addr) 447 { 448 449 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__)); 450 451 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) 452 return (0); 453 454 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) { 455 struct in6_addr tmp = *addr; 456 in6_clearscope(&tmp); 457 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) 458 return (0); 459 } 460 461 return (1); 462 } 463 464 /* 465 * Attach MLD when PF_INET6 is attached to an interface. 466 * 467 * SMPng: Normally called with IF_AFDATA_LOCK held. 468 */ 469 struct mld_ifsoftc * 470 mld_domifattach(struct ifnet *ifp) 471 { 472 struct mld_ifsoftc *mli; 473 474 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 475 __func__, ifp, if_name(ifp)); 476 477 MLD_LOCK(); 478 479 mli = mli_alloc_locked(ifp); 480 if (!(ifp->if_flags & IFF_MULTICAST)) 481 mli->mli_flags |= MLIF_SILENT; 482 if (mld_use_allow) 483 mli->mli_flags |= MLIF_USEALLOW; 484 485 MLD_UNLOCK(); 486 487 return (mli); 488 } 489 490 /* 491 * VIMAGE: assume curvnet set by caller. 492 */ 493 static struct mld_ifsoftc * 494 mli_alloc_locked(/*const*/ struct ifnet *ifp) 495 { 496 struct mld_ifsoftc *mli; 497 498 MLD_LOCK_ASSERT(); 499 500 mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO); 501 if (mli == NULL) 502 goto out; 503 504 mli->mli_ifp = ifp; 505 mli->mli_version = MLD_VERSION_2; 506 mli->mli_flags = 0; 507 mli->mli_rv = MLD_RV_INIT; 508 mli->mli_qi = MLD_QI_INIT; 509 mli->mli_qri = MLD_QRI_INIT; 510 mli->mli_uri = MLD_URI_INIT; 511 SLIST_INIT(&mli->mli_relinmhead); 512 mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS); 513 514 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link); 515 516 CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)", 517 ifp, if_name(ifp)); 518 519 out: 520 return (mli); 521 } 522 523 /* 524 * Hook for ifdetach. 525 * 526 * NOTE: Some finalization tasks need to run before the protocol domain 527 * is detached, but also before the link layer does its cleanup. 528 * Run before link-layer cleanup; cleanup groups, but do not free MLD state. 529 * 530 * SMPng: Caller must hold IN6_MULTI_LOCK(). 531 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator. 532 * XXX This routine is also bitten by unlocked ifma_protospec access. 533 */ 534 void 535 mld_ifdetach(struct ifnet *ifp) 536 { 537 struct mld_ifsoftc *mli; 538 struct ifmultiaddr *ifma; 539 struct in6_multi *inm, *tinm; 540 541 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp, 542 if_name(ifp)); 543 544 IN6_MULTI_LOCK_ASSERT(); 545 MLD_LOCK(); 546 547 mli = MLD_IFINFO(ifp); 548 if (mli->mli_version == MLD_VERSION_2) { 549 IF_ADDR_RLOCK(ifp); 550 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 551 if (ifma->ifma_addr->sa_family != AF_INET6 || 552 ifma->ifma_protospec == NULL) 553 continue; 554 inm = (struct in6_multi *)ifma->ifma_protospec; 555 if (inm->in6m_state == MLD_LEAVING_MEMBER) { 556 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 557 inm, in6m_nrele); 558 } 559 in6m_clear_recorded(inm); 560 } 561 IF_ADDR_RUNLOCK(ifp); 562 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, 563 tinm) { 564 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 565 in6m_release_locked(inm); 566 } 567 } 568 569 MLD_UNLOCK(); 570 } 571 572 /* 573 * Hook for domifdetach. 574 * Runs after link-layer cleanup; free MLD state. 575 * 576 * SMPng: Normally called with IF_AFDATA_LOCK held. 577 */ 578 void 579 mld_domifdetach(struct ifnet *ifp) 580 { 581 582 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", 583 __func__, ifp, if_name(ifp)); 584 585 MLD_LOCK(); 586 mli_delete_locked(ifp); 587 MLD_UNLOCK(); 588 } 589 590 static void 591 mli_delete_locked(const struct ifnet *ifp) 592 { 593 struct mld_ifsoftc *mli, *tmli; 594 595 CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)", 596 __func__, ifp, if_name(ifp)); 597 598 MLD_LOCK_ASSERT(); 599 600 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) { 601 if (mli->mli_ifp == ifp) { 602 /* 603 * Free deferred General Query responses. 604 */ 605 mbufq_drain(&mli->mli_gq); 606 607 LIST_REMOVE(mli, mli_link); 608 609 KASSERT(SLIST_EMPTY(&mli->mli_relinmhead), 610 ("%s: there are dangling in_multi references", 611 __func__)); 612 613 free(mli, M_MLD); 614 return; 615 } 616 } 617 } 618 619 /* 620 * Process a received MLDv1 general or address-specific query. 621 * Assumes that the query header has been pulled up to sizeof(mld_hdr). 622 * 623 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 624 * mld_addr. This is OK as we own the mbuf chain. 625 */ 626 static int 627 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 628 /*const*/ struct mld_hdr *mld) 629 { 630 struct ifmultiaddr *ifma; 631 struct mld_ifsoftc *mli; 632 struct in6_multi *inm; 633 int is_general_query; 634 uint16_t timer; 635 #ifdef KTR 636 char ip6tbuf[INET6_ADDRSTRLEN]; 637 #endif 638 639 is_general_query = 0; 640 641 if (!mld_v1enable) { 642 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)", 643 ip6_sprintf(ip6tbuf, &mld->mld_addr), 644 ifp, if_name(ifp)); 645 return (0); 646 } 647 648 /* 649 * RFC3810 Section 6.2: MLD queries must originate from 650 * a router's link-local address. 651 */ 652 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 653 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 654 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 655 ifp, if_name(ifp)); 656 return (0); 657 } 658 659 /* 660 * Do address field validation upfront before we accept 661 * the query. 662 */ 663 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 664 /* 665 * MLDv1 General Query. 666 * If this was not sent to the all-nodes group, ignore it. 667 */ 668 struct in6_addr dst; 669 670 dst = ip6->ip6_dst; 671 in6_clearscope(&dst); 672 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) 673 return (EINVAL); 674 is_general_query = 1; 675 } else { 676 /* 677 * Embed scope ID of receiving interface in MLD query for 678 * lookup whilst we don't hold other locks. 679 */ 680 in6_setscope(&mld->mld_addr, ifp, NULL); 681 } 682 683 IN6_MULTI_LOCK(); 684 MLD_LOCK(); 685 686 /* 687 * Switch to MLDv1 host compatibility mode. 688 */ 689 mli = MLD_IFINFO(ifp); 690 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); 691 mld_set_version(mli, MLD_VERSION_1); 692 693 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE; 694 if (timer == 0) 695 timer = 1; 696 697 IF_ADDR_RLOCK(ifp); 698 if (is_general_query) { 699 /* 700 * For each reporting group joined on this 701 * interface, kick the report timer. 702 */ 703 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)", 704 ifp, if_name(ifp)); 705 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 706 if (ifma->ifma_addr->sa_family != AF_INET6 || 707 ifma->ifma_protospec == NULL) 708 continue; 709 inm = (struct in6_multi *)ifma->ifma_protospec; 710 mld_v1_update_group(inm, timer); 711 } 712 } else { 713 /* 714 * MLDv1 Group-Specific Query. 715 * If this is a group-specific MLDv1 query, we need only 716 * look up the single group to process it. 717 */ 718 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 719 if (inm != NULL) { 720 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)", 721 ip6_sprintf(ip6tbuf, &mld->mld_addr), 722 ifp, if_name(ifp)); 723 mld_v1_update_group(inm, timer); 724 } 725 /* XXX Clear embedded scope ID as userland won't expect it. */ 726 in6_clearscope(&mld->mld_addr); 727 } 728 729 IF_ADDR_RUNLOCK(ifp); 730 MLD_UNLOCK(); 731 IN6_MULTI_UNLOCK(); 732 733 return (0); 734 } 735 736 /* 737 * Update the report timer on a group in response to an MLDv1 query. 738 * 739 * If we are becoming the reporting member for this group, start the timer. 740 * If we already are the reporting member for this group, and timer is 741 * below the threshold, reset it. 742 * 743 * We may be updating the group for the first time since we switched 744 * to MLDv2. If we are, then we must clear any recorded source lists, 745 * and transition to REPORTING state; the group timer is overloaded 746 * for group and group-source query responses. 747 * 748 * Unlike MLDv2, the delay per group should be jittered 749 * to avoid bursts of MLDv1 reports. 750 */ 751 static void 752 mld_v1_update_group(struct in6_multi *inm, const int timer) 753 { 754 #ifdef KTR 755 char ip6tbuf[INET6_ADDRSTRLEN]; 756 #endif 757 758 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__, 759 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 760 if_name(inm->in6m_ifp), timer); 761 762 IN6_MULTI_LOCK_ASSERT(); 763 764 switch (inm->in6m_state) { 765 case MLD_NOT_MEMBER: 766 case MLD_SILENT_MEMBER: 767 break; 768 case MLD_REPORTING_MEMBER: 769 if (inm->in6m_timer != 0 && 770 inm->in6m_timer <= timer) { 771 CTR1(KTR_MLD, "%s: REPORTING and timer running, " 772 "skipping.", __func__); 773 break; 774 } 775 /* FALLTHROUGH */ 776 case MLD_SG_QUERY_PENDING_MEMBER: 777 case MLD_G_QUERY_PENDING_MEMBER: 778 case MLD_IDLE_MEMBER: 779 case MLD_LAZY_MEMBER: 780 case MLD_AWAKENING_MEMBER: 781 CTR1(KTR_MLD, "%s: ->REPORTING", __func__); 782 inm->in6m_state = MLD_REPORTING_MEMBER; 783 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 784 V_current_state_timers_running6 = 1; 785 break; 786 case MLD_SLEEPING_MEMBER: 787 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__); 788 inm->in6m_state = MLD_AWAKENING_MEMBER; 789 break; 790 case MLD_LEAVING_MEMBER: 791 break; 792 } 793 } 794 795 /* 796 * Process a received MLDv2 general, group-specific or 797 * group-and-source-specific query. 798 * 799 * Assumes that the query header has been pulled up to sizeof(mldv2_query). 800 * 801 * Return 0 if successful, otherwise an appropriate error code is returned. 802 */ 803 static int 804 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6, 805 struct mbuf *m, const int off, const int icmp6len) 806 { 807 struct mld_ifsoftc *mli; 808 struct mldv2_query *mld; 809 struct in6_multi *inm; 810 uint32_t maxdelay, nsrc, qqi; 811 int is_general_query; 812 uint16_t timer; 813 uint8_t qrv; 814 #ifdef KTR 815 char ip6tbuf[INET6_ADDRSTRLEN]; 816 #endif 817 818 is_general_query = 0; 819 820 /* 821 * RFC3810 Section 6.2: MLD queries must originate from 822 * a router's link-local address. 823 */ 824 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) { 825 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 826 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 827 ifp, if_name(ifp)); 828 return (0); 829 } 830 831 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp)); 832 833 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off); 834 835 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */ 836 if (maxdelay >= 32768) { 837 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) << 838 (MLD_MRC_EXP(maxdelay) + 3); 839 } 840 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE; 841 if (timer == 0) 842 timer = 1; 843 844 qrv = MLD_QRV(mld->mld_misc); 845 if (qrv < 2) { 846 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__, 847 qrv, MLD_RV_INIT); 848 qrv = MLD_RV_INIT; 849 } 850 851 qqi = mld->mld_qqi; 852 if (qqi >= 128) { 853 qqi = MLD_QQIC_MANT(mld->mld_qqi) << 854 (MLD_QQIC_EXP(mld->mld_qqi) + 3); 855 } 856 857 nsrc = ntohs(mld->mld_numsrc); 858 if (nsrc > MLD_MAX_GS_SOURCES) 859 return (EMSGSIZE); 860 if (icmp6len < sizeof(struct mldv2_query) + 861 (nsrc * sizeof(struct in6_addr))) 862 return (EMSGSIZE); 863 864 /* 865 * Do further input validation upfront to avoid resetting timers 866 * should we need to discard this query. 867 */ 868 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) { 869 /* 870 * A general query with a source list has undefined 871 * behaviour; discard it. 872 */ 873 if (nsrc > 0) 874 return (EINVAL); 875 is_general_query = 1; 876 } else { 877 /* 878 * Embed scope ID of receiving interface in MLD query for 879 * lookup whilst we don't hold other locks (due to KAME 880 * locking lameness). We own this mbuf chain just now. 881 */ 882 in6_setscope(&mld->mld_addr, ifp, NULL); 883 } 884 885 IN6_MULTI_LOCK(); 886 MLD_LOCK(); 887 888 mli = MLD_IFINFO(ifp); 889 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); 890 891 /* 892 * Discard the v2 query if we're in Compatibility Mode. 893 * The RFC is pretty clear that hosts need to stay in MLDv1 mode 894 * until the Old Version Querier Present timer expires. 895 */ 896 if (mli->mli_version != MLD_VERSION_2) 897 goto out_locked; 898 899 mld_set_version(mli, MLD_VERSION_2); 900 mli->mli_rv = qrv; 901 mli->mli_qi = qqi; 902 mli->mli_qri = maxdelay; 903 904 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi, 905 maxdelay); 906 907 if (is_general_query) { 908 /* 909 * MLDv2 General Query. 910 * 911 * Schedule a current-state report on this ifp for 912 * all groups, possibly containing source lists. 913 * 914 * If there is a pending General Query response 915 * scheduled earlier than the selected delay, do 916 * not schedule any other reports. 917 * Otherwise, reset the interface timer. 918 */ 919 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)", 920 ifp, if_name(ifp)); 921 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) { 922 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer); 923 V_interface_timers_running6 = 1; 924 } 925 } else { 926 /* 927 * MLDv2 Group-specific or Group-and-source-specific Query. 928 * 929 * Group-source-specific queries are throttled on 930 * a per-group basis to defeat denial-of-service attempts. 931 * Queries for groups we are not a member of on this 932 * link are simply ignored. 933 */ 934 IF_ADDR_RLOCK(ifp); 935 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 936 if (inm == NULL) { 937 IF_ADDR_RUNLOCK(ifp); 938 goto out_locked; 939 } 940 if (nsrc > 0) { 941 if (!ratecheck(&inm->in6m_lastgsrtv, 942 &V_mld_gsrdelay)) { 943 CTR1(KTR_MLD, "%s: GS query throttled.", 944 __func__); 945 IF_ADDR_RUNLOCK(ifp); 946 goto out_locked; 947 } 948 } 949 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)", 950 ifp, if_name(ifp)); 951 /* 952 * If there is a pending General Query response 953 * scheduled sooner than the selected delay, no 954 * further report need be scheduled. 955 * Otherwise, prepare to respond to the 956 * group-specific or group-and-source query. 957 */ 958 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) 959 mld_v2_process_group_query(inm, mli, timer, m, off); 960 961 /* XXX Clear embedded scope ID as userland won't expect it. */ 962 in6_clearscope(&mld->mld_addr); 963 IF_ADDR_RUNLOCK(ifp); 964 } 965 966 out_locked: 967 MLD_UNLOCK(); 968 IN6_MULTI_UNLOCK(); 969 970 return (0); 971 } 972 973 /* 974 * Process a received MLDv2 group-specific or group-and-source-specific 975 * query. 976 * Return <0 if any error occurred. Currently this is ignored. 977 */ 978 static int 979 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli, 980 int timer, struct mbuf *m0, const int off) 981 { 982 struct mldv2_query *mld; 983 int retval; 984 uint16_t nsrc; 985 986 IN6_MULTI_LOCK_ASSERT(); 987 MLD_LOCK_ASSERT(); 988 989 retval = 0; 990 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off); 991 992 switch (inm->in6m_state) { 993 case MLD_NOT_MEMBER: 994 case MLD_SILENT_MEMBER: 995 case MLD_SLEEPING_MEMBER: 996 case MLD_LAZY_MEMBER: 997 case MLD_AWAKENING_MEMBER: 998 case MLD_IDLE_MEMBER: 999 case MLD_LEAVING_MEMBER: 1000 return (retval); 1001 break; 1002 case MLD_REPORTING_MEMBER: 1003 case MLD_G_QUERY_PENDING_MEMBER: 1004 case MLD_SG_QUERY_PENDING_MEMBER: 1005 break; 1006 } 1007 1008 nsrc = ntohs(mld->mld_numsrc); 1009 1010 /* 1011 * Deal with group-specific queries upfront. 1012 * If any group query is already pending, purge any recorded 1013 * source-list state if it exists, and schedule a query response 1014 * for this group-specific query. 1015 */ 1016 if (nsrc == 0) { 1017 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER || 1018 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) { 1019 in6m_clear_recorded(inm); 1020 timer = min(inm->in6m_timer, timer); 1021 } 1022 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER; 1023 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1024 V_current_state_timers_running6 = 1; 1025 return (retval); 1026 } 1027 1028 /* 1029 * Deal with the case where a group-and-source-specific query has 1030 * been received but a group-specific query is already pending. 1031 */ 1032 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) { 1033 timer = min(inm->in6m_timer, timer); 1034 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1035 V_current_state_timers_running6 = 1; 1036 return (retval); 1037 } 1038 1039 /* 1040 * Finally, deal with the case where a group-and-source-specific 1041 * query has been received, where a response to a previous g-s-r 1042 * query exists, or none exists. 1043 * In this case, we need to parse the source-list which the Querier 1044 * has provided us with and check if we have any source list filter 1045 * entries at T1 for these sources. If we do not, there is no need 1046 * schedule a report and the query may be dropped. 1047 * If we do, we must record them and schedule a current-state 1048 * report for those sources. 1049 */ 1050 if (inm->in6m_nsrc > 0) { 1051 struct mbuf *m; 1052 uint8_t *sp; 1053 int i, nrecorded; 1054 int soff; 1055 1056 m = m0; 1057 soff = off + sizeof(struct mldv2_query); 1058 nrecorded = 0; 1059 for (i = 0; i < nsrc; i++) { 1060 sp = mtod(m, uint8_t *) + soff; 1061 retval = in6m_record_source(inm, 1062 (const struct in6_addr *)sp); 1063 if (retval < 0) 1064 break; 1065 nrecorded += retval; 1066 soff += sizeof(struct in6_addr); 1067 if (soff >= m->m_len) { 1068 soff = soff - m->m_len; 1069 m = m->m_next; 1070 if (m == NULL) 1071 break; 1072 } 1073 } 1074 if (nrecorded > 0) { 1075 CTR1(KTR_MLD, 1076 "%s: schedule response to SG query", __func__); 1077 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER; 1078 inm->in6m_timer = MLD_RANDOM_DELAY(timer); 1079 V_current_state_timers_running6 = 1; 1080 } 1081 } 1082 1083 return (retval); 1084 } 1085 1086 /* 1087 * Process a received MLDv1 host membership report. 1088 * Assumes mld points to mld_hdr in pulled up mbuf chain. 1089 * 1090 * NOTE: Can't be fully const correct as we temporarily embed scope ID in 1091 * mld_addr. This is OK as we own the mbuf chain. 1092 */ 1093 static int 1094 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6, 1095 /*const*/ struct mld_hdr *mld) 1096 { 1097 struct in6_addr src, dst; 1098 struct in6_ifaddr *ia; 1099 struct in6_multi *inm; 1100 #ifdef KTR 1101 char ip6tbuf[INET6_ADDRSTRLEN]; 1102 #endif 1103 1104 if (!mld_v1enable) { 1105 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)", 1106 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1107 ifp, if_name(ifp)); 1108 return (0); 1109 } 1110 1111 if (ifp->if_flags & IFF_LOOPBACK) 1112 return (0); 1113 1114 /* 1115 * MLDv1 reports must originate from a host's link-local address, 1116 * or the unspecified address (when booting). 1117 */ 1118 src = ip6->ip6_src; 1119 in6_clearscope(&src); 1120 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) { 1121 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)", 1122 ip6_sprintf(ip6tbuf, &ip6->ip6_src), 1123 ifp, if_name(ifp)); 1124 return (EINVAL); 1125 } 1126 1127 /* 1128 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast 1129 * group, and must be directed to the group itself. 1130 */ 1131 dst = ip6->ip6_dst; 1132 in6_clearscope(&dst); 1133 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) || 1134 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) { 1135 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)", 1136 ip6_sprintf(ip6tbuf, &ip6->ip6_dst), 1137 ifp, if_name(ifp)); 1138 return (EINVAL); 1139 } 1140 1141 /* 1142 * Make sure we don't hear our own membership report, as fast 1143 * leave requires knowing that we are the only member of a 1144 * group. Assume we used the link-local address if available, 1145 * otherwise look for ::. 1146 * 1147 * XXX Note that scope ID comparison is needed for the address 1148 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be 1149 * performed for the on-wire address. 1150 */ 1151 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1152 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) || 1153 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) { 1154 if (ia != NULL) 1155 ifa_free(&ia->ia_ifa); 1156 return (0); 1157 } 1158 if (ia != NULL) 1159 ifa_free(&ia->ia_ifa); 1160 1161 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)", 1162 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp)); 1163 1164 /* 1165 * Embed scope ID of receiving interface in MLD query for lookup 1166 * whilst we don't hold other locks (due to KAME locking lameness). 1167 */ 1168 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) 1169 in6_setscope(&mld->mld_addr, ifp, NULL); 1170 1171 IN6_MULTI_LOCK(); 1172 MLD_LOCK(); 1173 IF_ADDR_RLOCK(ifp); 1174 1175 /* 1176 * MLDv1 report suppression. 1177 * If we are a member of this group, and our membership should be 1178 * reported, and our group timer is pending or about to be reset, 1179 * stop our group timer by transitioning to the 'lazy' state. 1180 */ 1181 inm = in6m_lookup_locked(ifp, &mld->mld_addr); 1182 if (inm != NULL) { 1183 struct mld_ifsoftc *mli; 1184 1185 mli = inm->in6m_mli; 1186 KASSERT(mli != NULL, 1187 ("%s: no mli for ifp %p", __func__, ifp)); 1188 1189 /* 1190 * If we are in MLDv2 host mode, do not allow the 1191 * other host's MLDv1 report to suppress our reports. 1192 */ 1193 if (mli->mli_version == MLD_VERSION_2) 1194 goto out_locked; 1195 1196 inm->in6m_timer = 0; 1197 1198 switch (inm->in6m_state) { 1199 case MLD_NOT_MEMBER: 1200 case MLD_SILENT_MEMBER: 1201 case MLD_SLEEPING_MEMBER: 1202 break; 1203 case MLD_REPORTING_MEMBER: 1204 case MLD_IDLE_MEMBER: 1205 case MLD_AWAKENING_MEMBER: 1206 CTR3(KTR_MLD, 1207 "report suppressed for %s on ifp %p(%s)", 1208 ip6_sprintf(ip6tbuf, &mld->mld_addr), 1209 ifp, if_name(ifp)); 1210 case MLD_LAZY_MEMBER: 1211 inm->in6m_state = MLD_LAZY_MEMBER; 1212 break; 1213 case MLD_G_QUERY_PENDING_MEMBER: 1214 case MLD_SG_QUERY_PENDING_MEMBER: 1215 case MLD_LEAVING_MEMBER: 1216 break; 1217 } 1218 } 1219 1220 out_locked: 1221 IF_ADDR_RUNLOCK(ifp); 1222 MLD_UNLOCK(); 1223 IN6_MULTI_UNLOCK(); 1224 1225 /* XXX Clear embedded scope ID as userland won't expect it. */ 1226 in6_clearscope(&mld->mld_addr); 1227 1228 return (0); 1229 } 1230 1231 /* 1232 * MLD input path. 1233 * 1234 * Assume query messages which fit in a single ICMPv6 message header 1235 * have been pulled up. 1236 * Assume that userland will want to see the message, even if it 1237 * otherwise fails kernel input validation; do not free it. 1238 * Pullup may however free the mbuf chain m if it fails. 1239 * 1240 * Return IPPROTO_DONE if we freed m. Otherwise, return 0. 1241 */ 1242 int 1243 mld_input(struct mbuf *m, int off, int icmp6len) 1244 { 1245 struct ifnet *ifp; 1246 struct ip6_hdr *ip6; 1247 struct mld_hdr *mld; 1248 int mldlen; 1249 1250 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off); 1251 1252 ifp = m->m_pkthdr.rcvif; 1253 1254 ip6 = mtod(m, struct ip6_hdr *); 1255 1256 /* Pullup to appropriate size. */ 1257 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off); 1258 if (mld->mld_type == MLD_LISTENER_QUERY && 1259 icmp6len >= sizeof(struct mldv2_query)) { 1260 mldlen = sizeof(struct mldv2_query); 1261 } else { 1262 mldlen = sizeof(struct mld_hdr); 1263 } 1264 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen); 1265 if (mld == NULL) { 1266 ICMP6STAT_INC(icp6s_badlen); 1267 return (IPPROTO_DONE); 1268 } 1269 1270 /* 1271 * Userland needs to see all of this traffic for implementing 1272 * the endpoint discovery portion of multicast routing. 1273 */ 1274 switch (mld->mld_type) { 1275 case MLD_LISTENER_QUERY: 1276 icmp6_ifstat_inc(ifp, ifs6_in_mldquery); 1277 if (icmp6len == sizeof(struct mld_hdr)) { 1278 if (mld_v1_input_query(ifp, ip6, mld) != 0) 1279 return (0); 1280 } else if (icmp6len >= sizeof(struct mldv2_query)) { 1281 if (mld_v2_input_query(ifp, ip6, m, off, 1282 icmp6len) != 0) 1283 return (0); 1284 } 1285 break; 1286 case MLD_LISTENER_REPORT: 1287 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1288 if (mld_v1_input_report(ifp, ip6, mld) != 0) 1289 return (0); 1290 break; 1291 case MLDV2_LISTENER_REPORT: 1292 icmp6_ifstat_inc(ifp, ifs6_in_mldreport); 1293 break; 1294 case MLD_LISTENER_DONE: 1295 icmp6_ifstat_inc(ifp, ifs6_in_mlddone); 1296 break; 1297 default: 1298 break; 1299 } 1300 1301 return (0); 1302 } 1303 1304 /* 1305 * Fast timeout handler (global). 1306 * VIMAGE: Timeout handlers are expected to service all vimages. 1307 */ 1308 void 1309 mld_fasttimo(void) 1310 { 1311 VNET_ITERATOR_DECL(vnet_iter); 1312 1313 VNET_LIST_RLOCK_NOSLEEP(); 1314 VNET_FOREACH(vnet_iter) { 1315 CURVNET_SET(vnet_iter); 1316 mld_fasttimo_vnet(); 1317 CURVNET_RESTORE(); 1318 } 1319 VNET_LIST_RUNLOCK_NOSLEEP(); 1320 } 1321 1322 /* 1323 * Fast timeout handler (per-vnet). 1324 * 1325 * VIMAGE: Assume caller has set up our curvnet. 1326 */ 1327 static void 1328 mld_fasttimo_vnet(void) 1329 { 1330 struct mbufq scq; /* State-change packets */ 1331 struct mbufq qrq; /* Query response packets */ 1332 struct ifnet *ifp; 1333 struct mld_ifsoftc *mli; 1334 struct ifmultiaddr *ifma; 1335 struct in6_multi *inm, *tinm; 1336 int uri_fasthz; 1337 1338 uri_fasthz = 0; 1339 1340 /* 1341 * Quick check to see if any work needs to be done, in order to 1342 * minimize the overhead of fasttimo processing. 1343 * SMPng: XXX Unlocked reads. 1344 */ 1345 if (!V_current_state_timers_running6 && 1346 !V_interface_timers_running6 && 1347 !V_state_change_timers_running6) 1348 return; 1349 1350 IN6_MULTI_LOCK(); 1351 MLD_LOCK(); 1352 1353 /* 1354 * MLDv2 General Query response timer processing. 1355 */ 1356 if (V_interface_timers_running6) { 1357 CTR1(KTR_MLD, "%s: interface timers running", __func__); 1358 1359 V_interface_timers_running6 = 0; 1360 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1361 if (mli->mli_v2_timer == 0) { 1362 /* Do nothing. */ 1363 } else if (--mli->mli_v2_timer == 0) { 1364 mld_v2_dispatch_general_query(mli); 1365 } else { 1366 V_interface_timers_running6 = 1; 1367 } 1368 } 1369 } 1370 1371 if (!V_current_state_timers_running6 && 1372 !V_state_change_timers_running6) 1373 goto out_locked; 1374 1375 V_current_state_timers_running6 = 0; 1376 V_state_change_timers_running6 = 0; 1377 1378 CTR1(KTR_MLD, "%s: state change timers running", __func__); 1379 1380 /* 1381 * MLD host report and state-change timer processing. 1382 * Note: Processing a v2 group timer may remove a node. 1383 */ 1384 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1385 ifp = mli->mli_ifp; 1386 1387 if (mli->mli_version == MLD_VERSION_2) { 1388 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri * 1389 PR_FASTHZ); 1390 mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS); 1391 mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS); 1392 } 1393 1394 IF_ADDR_RLOCK(ifp); 1395 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1396 if (ifma->ifma_addr->sa_family != AF_INET6 || 1397 ifma->ifma_protospec == NULL) 1398 continue; 1399 inm = (struct in6_multi *)ifma->ifma_protospec; 1400 switch (mli->mli_version) { 1401 case MLD_VERSION_1: 1402 mld_v1_process_group_timer(mli, inm); 1403 break; 1404 case MLD_VERSION_2: 1405 mld_v2_process_group_timers(mli, &qrq, 1406 &scq, inm, uri_fasthz); 1407 break; 1408 } 1409 } 1410 IF_ADDR_RUNLOCK(ifp); 1411 1412 switch (mli->mli_version) { 1413 case MLD_VERSION_1: 1414 /* 1415 * Transmit reports for this lifecycle. This 1416 * is done while not holding IF_ADDR_LOCK 1417 * since this can call 1418 * in6ifa_ifpforlinklocal() which locks 1419 * IF_ADDR_LOCK internally as well as 1420 * ip6_output() to transmit a packet. 1421 */ 1422 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1423 in6m_nrele, tinm) { 1424 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1425 in6m_nrele); 1426 (void)mld_v1_transmit_report(inm, 1427 MLD_LISTENER_REPORT); 1428 } 1429 break; 1430 case MLD_VERSION_2: 1431 mld_dispatch_queue(&qrq, 0); 1432 mld_dispatch_queue(&scq, 0); 1433 1434 /* 1435 * Free the in_multi reference(s) for 1436 * this lifecycle. 1437 */ 1438 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, 1439 in6m_nrele, tinm) { 1440 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, 1441 in6m_nrele); 1442 in6m_release_locked(inm); 1443 } 1444 break; 1445 } 1446 } 1447 1448 out_locked: 1449 MLD_UNLOCK(); 1450 IN6_MULTI_UNLOCK(); 1451 } 1452 1453 /* 1454 * Update host report group timer. 1455 * Will update the global pending timer flags. 1456 */ 1457 static void 1458 mld_v1_process_group_timer(struct mld_ifsoftc *mli, struct in6_multi *inm) 1459 { 1460 int report_timer_expired; 1461 1462 IN6_MULTI_LOCK_ASSERT(); 1463 MLD_LOCK_ASSERT(); 1464 1465 if (inm->in6m_timer == 0) { 1466 report_timer_expired = 0; 1467 } else if (--inm->in6m_timer == 0) { 1468 report_timer_expired = 1; 1469 } else { 1470 V_current_state_timers_running6 = 1; 1471 return; 1472 } 1473 1474 switch (inm->in6m_state) { 1475 case MLD_NOT_MEMBER: 1476 case MLD_SILENT_MEMBER: 1477 case MLD_IDLE_MEMBER: 1478 case MLD_LAZY_MEMBER: 1479 case MLD_SLEEPING_MEMBER: 1480 case MLD_AWAKENING_MEMBER: 1481 break; 1482 case MLD_REPORTING_MEMBER: 1483 if (report_timer_expired) { 1484 inm->in6m_state = MLD_IDLE_MEMBER; 1485 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1486 in6m_nrele); 1487 } 1488 break; 1489 case MLD_G_QUERY_PENDING_MEMBER: 1490 case MLD_SG_QUERY_PENDING_MEMBER: 1491 case MLD_LEAVING_MEMBER: 1492 break; 1493 } 1494 } 1495 1496 /* 1497 * Update a group's timers for MLDv2. 1498 * Will update the global pending timer flags. 1499 * Note: Unlocked read from mli. 1500 */ 1501 static void 1502 mld_v2_process_group_timers(struct mld_ifsoftc *mli, 1503 struct mbufq *qrq, struct mbufq *scq, 1504 struct in6_multi *inm, const int uri_fasthz) 1505 { 1506 int query_response_timer_expired; 1507 int state_change_retransmit_timer_expired; 1508 #ifdef KTR 1509 char ip6tbuf[INET6_ADDRSTRLEN]; 1510 #endif 1511 1512 IN6_MULTI_LOCK_ASSERT(); 1513 MLD_LOCK_ASSERT(); 1514 1515 query_response_timer_expired = 0; 1516 state_change_retransmit_timer_expired = 0; 1517 1518 /* 1519 * During a transition from compatibility mode back to MLDv2, 1520 * a group record in REPORTING state may still have its group 1521 * timer active. This is a no-op in this function; it is easier 1522 * to deal with it here than to complicate the slow-timeout path. 1523 */ 1524 if (inm->in6m_timer == 0) { 1525 query_response_timer_expired = 0; 1526 } else if (--inm->in6m_timer == 0) { 1527 query_response_timer_expired = 1; 1528 } else { 1529 V_current_state_timers_running6 = 1; 1530 } 1531 1532 if (inm->in6m_sctimer == 0) { 1533 state_change_retransmit_timer_expired = 0; 1534 } else if (--inm->in6m_sctimer == 0) { 1535 state_change_retransmit_timer_expired = 1; 1536 } else { 1537 V_state_change_timers_running6 = 1; 1538 } 1539 1540 /* We are in fasttimo, so be quick about it. */ 1541 if (!state_change_retransmit_timer_expired && 1542 !query_response_timer_expired) 1543 return; 1544 1545 switch (inm->in6m_state) { 1546 case MLD_NOT_MEMBER: 1547 case MLD_SILENT_MEMBER: 1548 case MLD_SLEEPING_MEMBER: 1549 case MLD_LAZY_MEMBER: 1550 case MLD_AWAKENING_MEMBER: 1551 case MLD_IDLE_MEMBER: 1552 break; 1553 case MLD_G_QUERY_PENDING_MEMBER: 1554 case MLD_SG_QUERY_PENDING_MEMBER: 1555 /* 1556 * Respond to a previously pending Group-Specific 1557 * or Group-and-Source-Specific query by enqueueing 1558 * the appropriate Current-State report for 1559 * immediate transmission. 1560 */ 1561 if (query_response_timer_expired) { 1562 int retval; 1563 1564 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1, 1565 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER), 1566 0); 1567 CTR2(KTR_MLD, "%s: enqueue record = %d", 1568 __func__, retval); 1569 inm->in6m_state = MLD_REPORTING_MEMBER; 1570 in6m_clear_recorded(inm); 1571 } 1572 /* FALLTHROUGH */ 1573 case MLD_REPORTING_MEMBER: 1574 case MLD_LEAVING_MEMBER: 1575 if (state_change_retransmit_timer_expired) { 1576 /* 1577 * State-change retransmission timer fired. 1578 * If there are any further pending retransmissions, 1579 * set the global pending state-change flag, and 1580 * reset the timer. 1581 */ 1582 if (--inm->in6m_scrv > 0) { 1583 inm->in6m_sctimer = uri_fasthz; 1584 V_state_change_timers_running6 = 1; 1585 } 1586 /* 1587 * Retransmit the previously computed state-change 1588 * report. If there are no further pending 1589 * retransmissions, the mbuf queue will be consumed. 1590 * Update T0 state to T1 as we have now sent 1591 * a state-change. 1592 */ 1593 (void)mld_v2_merge_state_changes(inm, scq); 1594 1595 in6m_commit(inm); 1596 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 1597 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1598 if_name(inm->in6m_ifp)); 1599 1600 /* 1601 * If we are leaving the group for good, make sure 1602 * we release MLD's reference to it. 1603 * This release must be deferred using a SLIST, 1604 * as we are called from a loop which traverses 1605 * the in_ifmultiaddr TAILQ. 1606 */ 1607 if (inm->in6m_state == MLD_LEAVING_MEMBER && 1608 inm->in6m_scrv == 0) { 1609 inm->in6m_state = MLD_NOT_MEMBER; 1610 SLIST_INSERT_HEAD(&mli->mli_relinmhead, 1611 inm, in6m_nrele); 1612 } 1613 } 1614 break; 1615 } 1616 } 1617 1618 /* 1619 * Switch to a different version on the given interface, 1620 * as per Section 9.12. 1621 */ 1622 static void 1623 mld_set_version(struct mld_ifsoftc *mli, const int version) 1624 { 1625 int old_version_timer; 1626 1627 MLD_LOCK_ASSERT(); 1628 1629 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__, 1630 version, mli->mli_ifp, if_name(mli->mli_ifp)); 1631 1632 if (version == MLD_VERSION_1) { 1633 /* 1634 * Compute the "Older Version Querier Present" timer as per 1635 * Section 9.12. 1636 */ 1637 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri; 1638 old_version_timer *= PR_SLOWHZ; 1639 mli->mli_v1_timer = old_version_timer; 1640 } 1641 1642 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) { 1643 mli->mli_version = MLD_VERSION_1; 1644 mld_v2_cancel_link_timers(mli); 1645 } 1646 } 1647 1648 /* 1649 * Cancel pending MLDv2 timers for the given link and all groups 1650 * joined on it; state-change, general-query, and group-query timers. 1651 */ 1652 static void 1653 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli) 1654 { 1655 struct ifmultiaddr *ifma; 1656 struct ifnet *ifp; 1657 struct in6_multi *inm, *tinm; 1658 1659 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__, 1660 mli->mli_ifp, if_name(mli->mli_ifp)); 1661 1662 IN6_MULTI_LOCK_ASSERT(); 1663 MLD_LOCK_ASSERT(); 1664 1665 /* 1666 * Fast-track this potentially expensive operation 1667 * by checking all the global 'timer pending' flags. 1668 */ 1669 if (!V_interface_timers_running6 && 1670 !V_state_change_timers_running6 && 1671 !V_current_state_timers_running6) 1672 return; 1673 1674 mli->mli_v2_timer = 0; 1675 1676 ifp = mli->mli_ifp; 1677 1678 IF_ADDR_RLOCK(ifp); 1679 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1680 if (ifma->ifma_addr->sa_family != AF_INET6) 1681 continue; 1682 inm = (struct in6_multi *)ifma->ifma_protospec; 1683 switch (inm->in6m_state) { 1684 case MLD_NOT_MEMBER: 1685 case MLD_SILENT_MEMBER: 1686 case MLD_IDLE_MEMBER: 1687 case MLD_LAZY_MEMBER: 1688 case MLD_SLEEPING_MEMBER: 1689 case MLD_AWAKENING_MEMBER: 1690 break; 1691 case MLD_LEAVING_MEMBER: 1692 /* 1693 * If we are leaving the group and switching 1694 * version, we need to release the final 1695 * reference held for issuing the INCLUDE {}. 1696 */ 1697 SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm, 1698 in6m_nrele); 1699 /* FALLTHROUGH */ 1700 case MLD_G_QUERY_PENDING_MEMBER: 1701 case MLD_SG_QUERY_PENDING_MEMBER: 1702 in6m_clear_recorded(inm); 1703 /* FALLTHROUGH */ 1704 case MLD_REPORTING_MEMBER: 1705 inm->in6m_sctimer = 0; 1706 inm->in6m_timer = 0; 1707 inm->in6m_state = MLD_REPORTING_MEMBER; 1708 /* 1709 * Free any pending MLDv2 state-change records. 1710 */ 1711 mbufq_drain(&inm->in6m_scq); 1712 break; 1713 } 1714 } 1715 IF_ADDR_RUNLOCK(ifp); 1716 SLIST_FOREACH_SAFE(inm, &mli->mli_relinmhead, in6m_nrele, tinm) { 1717 SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele); 1718 in6m_release_locked(inm); 1719 } 1720 } 1721 1722 /* 1723 * Global slowtimo handler. 1724 * VIMAGE: Timeout handlers are expected to service all vimages. 1725 */ 1726 void 1727 mld_slowtimo(void) 1728 { 1729 VNET_ITERATOR_DECL(vnet_iter); 1730 1731 VNET_LIST_RLOCK_NOSLEEP(); 1732 VNET_FOREACH(vnet_iter) { 1733 CURVNET_SET(vnet_iter); 1734 mld_slowtimo_vnet(); 1735 CURVNET_RESTORE(); 1736 } 1737 VNET_LIST_RUNLOCK_NOSLEEP(); 1738 } 1739 1740 /* 1741 * Per-vnet slowtimo handler. 1742 */ 1743 static void 1744 mld_slowtimo_vnet(void) 1745 { 1746 struct mld_ifsoftc *mli; 1747 1748 MLD_LOCK(); 1749 1750 LIST_FOREACH(mli, &V_mli_head, mli_link) { 1751 mld_v1_process_querier_timers(mli); 1752 } 1753 1754 MLD_UNLOCK(); 1755 } 1756 1757 /* 1758 * Update the Older Version Querier Present timers for a link. 1759 * See Section 9.12 of RFC 3810. 1760 */ 1761 static void 1762 mld_v1_process_querier_timers(struct mld_ifsoftc *mli) 1763 { 1764 1765 MLD_LOCK_ASSERT(); 1766 1767 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) { 1768 /* 1769 * MLDv1 Querier Present timer expired; revert to MLDv2. 1770 */ 1771 CTR5(KTR_MLD, 1772 "%s: transition from v%d -> v%d on %p(%s)", 1773 __func__, mli->mli_version, MLD_VERSION_2, 1774 mli->mli_ifp, if_name(mli->mli_ifp)); 1775 mli->mli_version = MLD_VERSION_2; 1776 } 1777 } 1778 1779 /* 1780 * Transmit an MLDv1 report immediately. 1781 */ 1782 static int 1783 mld_v1_transmit_report(struct in6_multi *in6m, const int type) 1784 { 1785 struct ifnet *ifp; 1786 struct in6_ifaddr *ia; 1787 struct ip6_hdr *ip6; 1788 struct mbuf *mh, *md; 1789 struct mld_hdr *mld; 1790 1791 IN6_MULTI_LOCK_ASSERT(); 1792 MLD_LOCK_ASSERT(); 1793 1794 ifp = in6m->in6m_ifp; 1795 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 1796 /* ia may be NULL if link-local address is tentative. */ 1797 1798 mh = m_gethdr(M_NOWAIT, MT_DATA); 1799 if (mh == NULL) { 1800 if (ia != NULL) 1801 ifa_free(&ia->ia_ifa); 1802 return (ENOMEM); 1803 } 1804 md = m_get(M_NOWAIT, MT_DATA); 1805 if (md == NULL) { 1806 m_free(mh); 1807 if (ia != NULL) 1808 ifa_free(&ia->ia_ifa); 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 M_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 if (ia != NULL) 1848 ifa_free(&ia->ia_ifa); 1849 return (0); 1850 } 1851 1852 /* 1853 * Process a state change from the upper layer for the given IPv6 group. 1854 * 1855 * Each socket holds a reference on the in_multi in its own ip_moptions. 1856 * The socket layer will have made the necessary updates to.the group 1857 * state, it is now up to MLD to issue a state change report if there 1858 * has been any change between T0 (when the last state-change was issued) 1859 * and T1 (now). 1860 * 1861 * We use the MLDv2 state machine at group level. The MLd module 1862 * however makes the decision as to which MLD protocol version to speak. 1863 * A state change *from* INCLUDE {} always means an initial join. 1864 * A state change *to* INCLUDE {} always means a final leave. 1865 * 1866 * If delay is non-zero, and the state change is an initial multicast 1867 * join, the state change report will be delayed by 'delay' ticks 1868 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise 1869 * the initial MLDv2 state change report will be delayed by whichever 1870 * is sooner, a pending state-change timer or delay itself. 1871 * 1872 * VIMAGE: curvnet should have been set by caller, as this routine 1873 * is called from the socket option handlers. 1874 */ 1875 int 1876 mld_change_state(struct in6_multi *inm, const int delay) 1877 { 1878 struct mld_ifsoftc *mli; 1879 struct ifnet *ifp; 1880 int error; 1881 1882 IN6_MULTI_LOCK_ASSERT(); 1883 1884 error = 0; 1885 1886 /* 1887 * Try to detect if the upper layer just asked us to change state 1888 * for an interface which has now gone away. 1889 */ 1890 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 1891 ifp = inm->in6m_ifma->ifma_ifp; 1892 if (ifp != NULL) { 1893 /* 1894 * Sanity check that netinet6's notion of ifp is the 1895 * same as net's. 1896 */ 1897 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 1898 } 1899 1900 MLD_LOCK(); 1901 1902 mli = MLD_IFINFO(ifp); 1903 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp)); 1904 1905 /* 1906 * If we detect a state transition to or from MCAST_UNDEFINED 1907 * for this group, then we are starting or finishing an MLD 1908 * life cycle for this group. 1909 */ 1910 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) { 1911 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__, 1912 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode); 1913 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) { 1914 CTR1(KTR_MLD, "%s: initial join", __func__); 1915 error = mld_initial_join(inm, mli, delay); 1916 goto out_locked; 1917 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) { 1918 CTR1(KTR_MLD, "%s: final leave", __func__); 1919 mld_final_leave(inm, mli); 1920 goto out_locked; 1921 } 1922 } else { 1923 CTR1(KTR_MLD, "%s: filter set change", __func__); 1924 } 1925 1926 error = mld_handle_state_change(inm, mli); 1927 1928 out_locked: 1929 MLD_UNLOCK(); 1930 return (error); 1931 } 1932 1933 /* 1934 * Perform the initial join for an MLD group. 1935 * 1936 * When joining a group: 1937 * If the group should have its MLD traffic suppressed, do nothing. 1938 * MLDv1 starts sending MLDv1 host membership reports. 1939 * MLDv2 will schedule an MLDv2 state-change report containing the 1940 * initial state of the membership. 1941 * 1942 * If the delay argument is non-zero, then we must delay sending the 1943 * initial state change for delay ticks (in units of PR_FASTHZ). 1944 */ 1945 static int 1946 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli, 1947 const int delay) 1948 { 1949 struct ifnet *ifp; 1950 struct mbufq *mq; 1951 int error, retval, syncstates; 1952 int odelay; 1953 #ifdef KTR 1954 char ip6tbuf[INET6_ADDRSTRLEN]; 1955 #endif 1956 1957 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)", 1958 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1959 inm->in6m_ifp, if_name(inm->in6m_ifp)); 1960 1961 error = 0; 1962 syncstates = 1; 1963 1964 ifp = inm->in6m_ifp; 1965 1966 IN6_MULTI_LOCK_ASSERT(); 1967 MLD_LOCK_ASSERT(); 1968 1969 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__)); 1970 1971 /* 1972 * Groups joined on loopback or marked as 'not reported', 1973 * enter the MLD_SILENT_MEMBER state and 1974 * are never reported in any protocol exchanges. 1975 * All other groups enter the appropriate state machine 1976 * for the version in use on this link. 1977 * A link marked as MLIF_SILENT causes MLD to be completely 1978 * disabled for the link. 1979 */ 1980 if ((ifp->if_flags & IFF_LOOPBACK) || 1981 (mli->mli_flags & MLIF_SILENT) || 1982 !mld_is_addr_reported(&inm->in6m_addr)) { 1983 CTR1(KTR_MLD, 1984 "%s: not kicking state machine for silent group", __func__); 1985 inm->in6m_state = MLD_SILENT_MEMBER; 1986 inm->in6m_timer = 0; 1987 } else { 1988 /* 1989 * Deal with overlapping in_multi lifecycle. 1990 * If this group was LEAVING, then make sure 1991 * we drop the reference we picked up to keep the 1992 * group around for the final INCLUDE {} enqueue. 1993 */ 1994 if (mli->mli_version == MLD_VERSION_2 && 1995 inm->in6m_state == MLD_LEAVING_MEMBER) 1996 in6m_release_locked(inm); 1997 1998 inm->in6m_state = MLD_REPORTING_MEMBER; 1999 2000 switch (mli->mli_version) { 2001 case MLD_VERSION_1: 2002 /* 2003 * If a delay was provided, only use it if 2004 * it is greater than the delay normally 2005 * used for an MLDv1 state change report, 2006 * and delay sending the initial MLDv1 report 2007 * by not transitioning to the IDLE state. 2008 */ 2009 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ); 2010 if (delay) { 2011 inm->in6m_timer = max(delay, odelay); 2012 V_current_state_timers_running6 = 1; 2013 } else { 2014 inm->in6m_state = MLD_IDLE_MEMBER; 2015 error = mld_v1_transmit_report(inm, 2016 MLD_LISTENER_REPORT); 2017 if (error == 0) { 2018 inm->in6m_timer = odelay; 2019 V_current_state_timers_running6 = 1; 2020 } 2021 } 2022 break; 2023 2024 case MLD_VERSION_2: 2025 /* 2026 * Defer update of T0 to T1, until the first copy 2027 * of the state change has been transmitted. 2028 */ 2029 syncstates = 0; 2030 2031 /* 2032 * Immediately enqueue a State-Change Report for 2033 * this interface, freeing any previous reports. 2034 * Don't kick the timers if there is nothing to do, 2035 * or if an error occurred. 2036 */ 2037 mq = &inm->in6m_scq; 2038 mbufq_drain(mq); 2039 retval = mld_v2_enqueue_group_record(mq, inm, 1, 2040 0, 0, (mli->mli_flags & MLIF_USEALLOW)); 2041 CTR2(KTR_MLD, "%s: enqueue record = %d", 2042 __func__, retval); 2043 if (retval <= 0) { 2044 error = retval * -1; 2045 break; 2046 } 2047 2048 /* 2049 * Schedule transmission of pending state-change 2050 * report up to RV times for this link. The timer 2051 * will fire at the next mld_fasttimo (~200ms), 2052 * giving us an opportunity to merge the reports. 2053 * 2054 * If a delay was provided to this function, only 2055 * use this delay if sooner than the existing one. 2056 */ 2057 KASSERT(mli->mli_rv > 1, 2058 ("%s: invalid robustness %d", __func__, 2059 mli->mli_rv)); 2060 inm->in6m_scrv = mli->mli_rv; 2061 if (delay) { 2062 if (inm->in6m_sctimer > 1) { 2063 inm->in6m_sctimer = 2064 min(inm->in6m_sctimer, delay); 2065 } else 2066 inm->in6m_sctimer = delay; 2067 } else 2068 inm->in6m_sctimer = 1; 2069 V_state_change_timers_running6 = 1; 2070 2071 error = 0; 2072 break; 2073 } 2074 } 2075 2076 /* 2077 * Only update the T0 state if state change is atomic, 2078 * i.e. we don't need to wait for a timer to fire before we 2079 * can consider the state change to have been communicated. 2080 */ 2081 if (syncstates) { 2082 in6m_commit(inm); 2083 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2084 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2085 if_name(inm->in6m_ifp)); 2086 } 2087 2088 return (error); 2089 } 2090 2091 /* 2092 * Issue an intermediate state change during the life-cycle. 2093 */ 2094 static int 2095 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli) 2096 { 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, if_name(inm->in6m_ifp)); 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 if_name(inm->in6m_ifp)); 2128 return (0); 2129 } 2130 2131 mbufq_drain(&inm->in6m_scq); 2132 2133 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0, 2134 (mli->mli_flags & MLIF_USEALLOW)); 2135 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval); 2136 if (retval <= 0) 2137 return (-retval); 2138 2139 /* 2140 * If record(s) were enqueued, start the state-change 2141 * report timer for this group. 2142 */ 2143 inm->in6m_scrv = mli->mli_rv; 2144 inm->in6m_sctimer = 1; 2145 V_state_change_timers_running6 = 1; 2146 2147 return (0); 2148 } 2149 2150 /* 2151 * Perform the final leave for a multicast address. 2152 * 2153 * When leaving a group: 2154 * MLDv1 sends a DONE message, if and only if we are the reporter. 2155 * MLDv2 enqueues a state-change report containing a transition 2156 * to INCLUDE {} for immediate transmission. 2157 */ 2158 static void 2159 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli) 2160 { 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, if_name(inm->in6m_ifp)); 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 V_current_state_timers_running6 = 1; 2197 } else if (mli->mli_version == MLD_VERSION_2) { 2198 /* 2199 * Stop group timer and all pending reports. 2200 * Immediately enqueue a state-change report 2201 * TO_IN {} to be sent on the next fast timeout, 2202 * giving us an opportunity to merge reports. 2203 */ 2204 mbufq_drain(&inm->in6m_scq); 2205 inm->in6m_timer = 0; 2206 inm->in6m_scrv = mli->mli_rv; 2207 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d " 2208 "pending retransmissions.", __func__, 2209 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2210 if_name(inm->in6m_ifp), inm->in6m_scrv); 2211 if (inm->in6m_scrv == 0) { 2212 inm->in6m_state = MLD_NOT_MEMBER; 2213 inm->in6m_sctimer = 0; 2214 } else { 2215 int retval; 2216 2217 in6m_acquire_locked(inm); 2218 2219 retval = mld_v2_enqueue_group_record( 2220 &inm->in6m_scq, inm, 1, 0, 0, 2221 (mli->mli_flags & MLIF_USEALLOW)); 2222 KASSERT(retval != 0, 2223 ("%s: enqueue record = %d", __func__, 2224 retval)); 2225 2226 inm->in6m_state = MLD_LEAVING_MEMBER; 2227 inm->in6m_sctimer = 1; 2228 V_state_change_timers_running6 = 1; 2229 syncstates = 0; 2230 } 2231 break; 2232 } 2233 break; 2234 case MLD_LAZY_MEMBER: 2235 case MLD_SLEEPING_MEMBER: 2236 case MLD_AWAKENING_MEMBER: 2237 /* Our reports are suppressed; do nothing. */ 2238 break; 2239 } 2240 2241 if (syncstates) { 2242 in6m_commit(inm); 2243 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__, 2244 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2245 if_name(inm->in6m_ifp)); 2246 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 2247 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s", 2248 __func__, &inm->in6m_addr, if_name(inm->in6m_ifp)); 2249 } 2250 } 2251 2252 /* 2253 * Enqueue an MLDv2 group record to the given output queue. 2254 * 2255 * If is_state_change is zero, a current-state record is appended. 2256 * If is_state_change is non-zero, a state-change report is appended. 2257 * 2258 * If is_group_query is non-zero, an mbuf packet chain is allocated. 2259 * If is_group_query is zero, and if there is a packet with free space 2260 * at the tail of the queue, it will be appended to providing there 2261 * is enough free space. 2262 * Otherwise a new mbuf packet chain is allocated. 2263 * 2264 * If is_source_query is non-zero, each source is checked to see if 2265 * it was recorded for a Group-Source query, and will be omitted if 2266 * it is not both in-mode and recorded. 2267 * 2268 * If use_block_allow is non-zero, state change reports for initial join 2269 * and final leave, on an inclusive mode group with a source list, will be 2270 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively. 2271 * 2272 * The function will attempt to allocate leading space in the packet 2273 * for the IPv6+ICMP headers to be prepended without fragmenting the chain. 2274 * 2275 * If successful the size of all data appended to the queue is returned, 2276 * otherwise an error code less than zero is returned, or zero if 2277 * no record(s) were appended. 2278 */ 2279 static int 2280 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm, 2281 const int is_state_change, const int is_group_query, 2282 const int is_source_query, const int use_block_allow) 2283 { 2284 struct mldv2_record mr; 2285 struct mldv2_record *pmr; 2286 struct ifnet *ifp; 2287 struct ip6_msource *ims, *nims; 2288 struct mbuf *m0, *m, *md; 2289 int is_filter_list_change; 2290 int minrec0len, m0srcs, msrcs, nbytes, off; 2291 int record_has_sources; 2292 int now; 2293 int type; 2294 uint8_t mode; 2295 #ifdef KTR 2296 char ip6tbuf[INET6_ADDRSTRLEN]; 2297 #endif 2298 2299 IN6_MULTI_LOCK_ASSERT(); 2300 2301 ifp = inm->in6m_ifp; 2302 is_filter_list_change = 0; 2303 m = NULL; 2304 m0 = NULL; 2305 m0srcs = 0; 2306 msrcs = 0; 2307 nbytes = 0; 2308 nims = NULL; 2309 record_has_sources = 1; 2310 pmr = NULL; 2311 type = MLD_DO_NOTHING; 2312 mode = inm->in6m_st[1].iss_fmode; 2313 2314 /* 2315 * If we did not transition out of ASM mode during t0->t1, 2316 * and there are no source nodes to process, we can skip 2317 * the generation of source records. 2318 */ 2319 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 && 2320 inm->in6m_nsrc == 0) 2321 record_has_sources = 0; 2322 2323 if (is_state_change) { 2324 /* 2325 * Queue a state change record. 2326 * If the mode did not change, and there are non-ASM 2327 * listeners or source filters present, 2328 * we potentially need to issue two records for the group. 2329 * If there are ASM listeners, and there was no filter 2330 * mode transition of any kind, do nothing. 2331 * 2332 * If we are transitioning to MCAST_UNDEFINED, we need 2333 * not send any sources. A transition to/from this state is 2334 * considered inclusive with some special treatment. 2335 * 2336 * If we are rewriting initial joins/leaves to use 2337 * ALLOW/BLOCK, and the group's membership is inclusive, 2338 * we need to send sources in all cases. 2339 */ 2340 if (mode != inm->in6m_st[0].iss_fmode) { 2341 if (mode == MCAST_EXCLUDE) { 2342 CTR1(KTR_MLD, "%s: change to EXCLUDE", 2343 __func__); 2344 type = MLD_CHANGE_TO_EXCLUDE_MODE; 2345 } else { 2346 CTR1(KTR_MLD, "%s: change to INCLUDE", 2347 __func__); 2348 if (use_block_allow) { 2349 /* 2350 * XXX 2351 * Here we're interested in state 2352 * edges either direction between 2353 * MCAST_UNDEFINED and MCAST_INCLUDE. 2354 * Perhaps we should just check 2355 * the group state, rather than 2356 * the filter mode. 2357 */ 2358 if (mode == MCAST_UNDEFINED) { 2359 type = MLD_BLOCK_OLD_SOURCES; 2360 } else { 2361 type = MLD_ALLOW_NEW_SOURCES; 2362 } 2363 } else { 2364 type = MLD_CHANGE_TO_INCLUDE_MODE; 2365 if (mode == MCAST_UNDEFINED) 2366 record_has_sources = 0; 2367 } 2368 } 2369 } else { 2370 if (record_has_sources) { 2371 is_filter_list_change = 1; 2372 } else { 2373 type = MLD_DO_NOTHING; 2374 } 2375 } 2376 } else { 2377 /* 2378 * Queue a current state record. 2379 */ 2380 if (mode == MCAST_EXCLUDE) { 2381 type = MLD_MODE_IS_EXCLUDE; 2382 } else if (mode == MCAST_INCLUDE) { 2383 type = MLD_MODE_IS_INCLUDE; 2384 KASSERT(inm->in6m_st[1].iss_asm == 0, 2385 ("%s: inm %p is INCLUDE but ASM count is %d", 2386 __func__, inm, inm->in6m_st[1].iss_asm)); 2387 } 2388 } 2389 2390 /* 2391 * Generate the filter list changes using a separate function. 2392 */ 2393 if (is_filter_list_change) 2394 return (mld_v2_enqueue_filter_change(mq, inm)); 2395 2396 if (type == MLD_DO_NOTHING) { 2397 CTR3(KTR_MLD, "%s: nothing to do for %s/%s", 2398 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2399 if_name(inm->in6m_ifp)); 2400 return (0); 2401 } 2402 2403 /* 2404 * If any sources are present, we must be able to fit at least 2405 * one in the trailing space of the tail packet's mbuf, 2406 * ideally more. 2407 */ 2408 minrec0len = sizeof(struct mldv2_record); 2409 if (record_has_sources) 2410 minrec0len += sizeof(struct in6_addr); 2411 2412 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__, 2413 mld_rec_type_to_str(type), 2414 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2415 if_name(inm->in6m_ifp)); 2416 2417 /* 2418 * Check if we have a packet in the tail of the queue for this 2419 * group into which the first group record for this group will fit. 2420 * Otherwise allocate a new packet. 2421 * Always allocate leading space for IP6+RA+ICMPV6+REPORT. 2422 * Note: Group records for G/GSR query responses MUST be sent 2423 * in their own packet. 2424 */ 2425 m0 = mbufq_last(mq); 2426 if (!is_group_query && 2427 m0 != NULL && 2428 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) && 2429 (m0->m_pkthdr.len + minrec0len) < 2430 (ifp->if_mtu - MLD_MTUSPACE)) { 2431 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2432 sizeof(struct mldv2_record)) / 2433 sizeof(struct in6_addr); 2434 m = m0; 2435 CTR1(KTR_MLD, "%s: use existing packet", __func__); 2436 } else { 2437 if (mbufq_full(mq)) { 2438 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2439 return (-ENOMEM); 2440 } 2441 m = NULL; 2442 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2443 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2444 if (!is_state_change && !is_group_query) 2445 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2446 if (m == NULL) 2447 m = m_gethdr(M_NOWAIT, MT_DATA); 2448 if (m == NULL) 2449 return (-ENOMEM); 2450 2451 mld_save_context(m, ifp); 2452 2453 CTR1(KTR_MLD, "%s: allocated first packet", __func__); 2454 } 2455 2456 /* 2457 * Append group record. 2458 * If we have sources, we don't know how many yet. 2459 */ 2460 mr.mr_type = type; 2461 mr.mr_datalen = 0; 2462 mr.mr_numsrc = 0; 2463 mr.mr_addr = inm->in6m_addr; 2464 in6_clearscope(&mr.mr_addr); 2465 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2466 if (m != m0) 2467 m_freem(m); 2468 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2469 return (-ENOMEM); 2470 } 2471 nbytes += sizeof(struct mldv2_record); 2472 2473 /* 2474 * Append as many sources as will fit in the first packet. 2475 * If we are appending to a new packet, the chain allocation 2476 * may potentially use clusters; use m_getptr() in this case. 2477 * If we are appending to an existing packet, we need to obtain 2478 * a pointer to the group record after m_append(), in case a new 2479 * mbuf was allocated. 2480 * 2481 * Only append sources which are in-mode at t1. If we are 2482 * transitioning to MCAST_UNDEFINED state on the group, and 2483 * use_block_allow is zero, do not include source entries. 2484 * Otherwise, we need to include this source in the report. 2485 * 2486 * Only report recorded sources in our filter set when responding 2487 * to a group-source query. 2488 */ 2489 if (record_has_sources) { 2490 if (m == m0) { 2491 md = m_last(m); 2492 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2493 md->m_len - nbytes); 2494 } else { 2495 md = m_getptr(m, 0, &off); 2496 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + 2497 off); 2498 } 2499 msrcs = 0; 2500 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, 2501 nims) { 2502 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2503 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2504 now = im6s_get_mode(inm, ims, 1); 2505 CTR2(KTR_MLD, "%s: node is %d", __func__, now); 2506 if ((now != mode) || 2507 (now == mode && 2508 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2509 CTR1(KTR_MLD, "%s: skip node", __func__); 2510 continue; 2511 } 2512 if (is_source_query && ims->im6s_stp == 0) { 2513 CTR1(KTR_MLD, "%s: skip unrecorded node", 2514 __func__); 2515 continue; 2516 } 2517 CTR1(KTR_MLD, "%s: append node", __func__); 2518 if (!m_append(m, sizeof(struct in6_addr), 2519 (void *)&ims->im6s_addr)) { 2520 if (m != m0) 2521 m_freem(m); 2522 CTR1(KTR_MLD, "%s: m_append() failed.", 2523 __func__); 2524 return (-ENOMEM); 2525 } 2526 nbytes += sizeof(struct in6_addr); 2527 ++msrcs; 2528 if (msrcs == m0srcs) 2529 break; 2530 } 2531 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__, 2532 msrcs); 2533 pmr->mr_numsrc = htons(msrcs); 2534 nbytes += (msrcs * sizeof(struct in6_addr)); 2535 } 2536 2537 if (is_source_query && msrcs == 0) { 2538 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__); 2539 if (m != m0) 2540 m_freem(m); 2541 return (0); 2542 } 2543 2544 /* 2545 * We are good to go with first packet. 2546 */ 2547 if (m != m0) { 2548 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__); 2549 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2550 mbufq_enqueue(mq, m); 2551 } else 2552 m->m_pkthdr.PH_vt.vt_nrecs++; 2553 2554 /* 2555 * No further work needed if no source list in packet(s). 2556 */ 2557 if (!record_has_sources) 2558 return (nbytes); 2559 2560 /* 2561 * Whilst sources remain to be announced, we need to allocate 2562 * a new packet and fill out as many sources as will fit. 2563 * Always try for a cluster first. 2564 */ 2565 while (nims != NULL) { 2566 if (mbufq_full(mq)) { 2567 CTR1(KTR_MLD, "%s: outbound queue full", __func__); 2568 return (-ENOMEM); 2569 } 2570 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2571 if (m == NULL) 2572 m = m_gethdr(M_NOWAIT, MT_DATA); 2573 if (m == NULL) 2574 return (-ENOMEM); 2575 mld_save_context(m, ifp); 2576 md = m_getptr(m, 0, &off); 2577 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off); 2578 CTR1(KTR_MLD, "%s: allocated next packet", __func__); 2579 2580 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) { 2581 if (m != m0) 2582 m_freem(m); 2583 CTR1(KTR_MLD, "%s: m_append() failed.", __func__); 2584 return (-ENOMEM); 2585 } 2586 m->m_pkthdr.PH_vt.vt_nrecs = 1; 2587 nbytes += sizeof(struct mldv2_record); 2588 2589 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2590 sizeof(struct mldv2_record)) / sizeof(struct in6_addr); 2591 2592 msrcs = 0; 2593 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2594 CTR2(KTR_MLD, "%s: visit node %s", 2595 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2596 now = im6s_get_mode(inm, ims, 1); 2597 if ((now != mode) || 2598 (now == mode && 2599 (!use_block_allow && mode == MCAST_UNDEFINED))) { 2600 CTR1(KTR_MLD, "%s: skip node", __func__); 2601 continue; 2602 } 2603 if (is_source_query && ims->im6s_stp == 0) { 2604 CTR1(KTR_MLD, "%s: skip unrecorded node", 2605 __func__); 2606 continue; 2607 } 2608 CTR1(KTR_MLD, "%s: append node", __func__); 2609 if (!m_append(m, sizeof(struct in6_addr), 2610 (void *)&ims->im6s_addr)) { 2611 if (m != m0) 2612 m_freem(m); 2613 CTR1(KTR_MLD, "%s: m_append() failed.", 2614 __func__); 2615 return (-ENOMEM); 2616 } 2617 ++msrcs; 2618 if (msrcs == m0srcs) 2619 break; 2620 } 2621 pmr->mr_numsrc = htons(msrcs); 2622 nbytes += (msrcs * sizeof(struct in6_addr)); 2623 2624 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__); 2625 mbufq_enqueue(mq, m); 2626 } 2627 2628 return (nbytes); 2629 } 2630 2631 /* 2632 * Type used to mark record pass completion. 2633 * We exploit the fact we can cast to this easily from the 2634 * current filter modes on each ip_msource node. 2635 */ 2636 typedef enum { 2637 REC_NONE = 0x00, /* MCAST_UNDEFINED */ 2638 REC_ALLOW = 0x01, /* MCAST_INCLUDE */ 2639 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ 2640 REC_FULL = REC_ALLOW | REC_BLOCK 2641 } rectype_t; 2642 2643 /* 2644 * Enqueue an MLDv2 filter list change to the given output queue. 2645 * 2646 * Source list filter state is held in an RB-tree. When the filter list 2647 * for a group is changed without changing its mode, we need to compute 2648 * the deltas between T0 and T1 for each source in the filter set, 2649 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. 2650 * 2651 * As we may potentially queue two record types, and the entire R-B tree 2652 * needs to be walked at once, we break this out into its own function 2653 * so we can generate a tightly packed queue of packets. 2654 * 2655 * XXX This could be written to only use one tree walk, although that makes 2656 * serializing into the mbuf chains a bit harder. For now we do two walks 2657 * which makes things easier on us, and it may or may not be harder on 2658 * the L2 cache. 2659 * 2660 * If successful the size of all data appended to the queue is returned, 2661 * otherwise an error code less than zero is returned, or zero if 2662 * no record(s) were appended. 2663 */ 2664 static int 2665 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm) 2666 { 2667 static const int MINRECLEN = 2668 sizeof(struct mldv2_record) + sizeof(struct in6_addr); 2669 struct ifnet *ifp; 2670 struct mldv2_record mr; 2671 struct mldv2_record *pmr; 2672 struct ip6_msource *ims, *nims; 2673 struct mbuf *m, *m0, *md; 2674 int m0srcs, nbytes, npbytes, off, rsrcs, schanged; 2675 int nallow, nblock; 2676 uint8_t mode, now, then; 2677 rectype_t crt, drt, nrt; 2678 #ifdef KTR 2679 char ip6tbuf[INET6_ADDRSTRLEN]; 2680 #endif 2681 2682 IN6_MULTI_LOCK_ASSERT(); 2683 2684 if (inm->in6m_nsrc == 0 || 2685 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) 2686 return (0); 2687 2688 ifp = inm->in6m_ifp; /* interface */ 2689 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */ 2690 crt = REC_NONE; /* current group record type */ 2691 drt = REC_NONE; /* mask of completed group record types */ 2692 nrt = REC_NONE; /* record type for current node */ 2693 m0srcs = 0; /* # source which will fit in current mbuf chain */ 2694 npbytes = 0; /* # of bytes appended this packet */ 2695 nbytes = 0; /* # of bytes appended to group's state-change queue */ 2696 rsrcs = 0; /* # sources encoded in current record */ 2697 schanged = 0; /* # nodes encoded in overall filter change */ 2698 nallow = 0; /* # of source entries in ALLOW_NEW */ 2699 nblock = 0; /* # of source entries in BLOCK_OLD */ 2700 nims = NULL; /* next tree node pointer */ 2701 2702 /* 2703 * For each possible filter record mode. 2704 * The first kind of source we encounter tells us which 2705 * is the first kind of record we start appending. 2706 * If a node transitioned to UNDEFINED at t1, its mode is treated 2707 * as the inverse of the group's filter mode. 2708 */ 2709 while (drt != REC_FULL) { 2710 do { 2711 m0 = mbufq_last(mq); 2712 if (m0 != NULL && 2713 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= 2714 MLD_V2_REPORT_MAXRECS) && 2715 (m0->m_pkthdr.len + MINRECLEN) < 2716 (ifp->if_mtu - MLD_MTUSPACE)) { 2717 m = m0; 2718 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - 2719 sizeof(struct mldv2_record)) / 2720 sizeof(struct in6_addr); 2721 CTR1(KTR_MLD, 2722 "%s: use previous packet", __func__); 2723 } else { 2724 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); 2725 if (m == NULL) 2726 m = m_gethdr(M_NOWAIT, MT_DATA); 2727 if (m == NULL) { 2728 CTR1(KTR_MLD, 2729 "%s: m_get*() failed", __func__); 2730 return (-ENOMEM); 2731 } 2732 m->m_pkthdr.PH_vt.vt_nrecs = 0; 2733 mld_save_context(m, ifp); 2734 m0srcs = (ifp->if_mtu - MLD_MTUSPACE - 2735 sizeof(struct mldv2_record)) / 2736 sizeof(struct in6_addr); 2737 npbytes = 0; 2738 CTR1(KTR_MLD, 2739 "%s: allocated new packet", __func__); 2740 } 2741 /* 2742 * Append the MLD group record header to the 2743 * current packet's data area. 2744 * Recalculate pointer to free space for next 2745 * group record, in case m_append() allocated 2746 * a new mbuf or cluster. 2747 */ 2748 memset(&mr, 0, sizeof(mr)); 2749 mr.mr_addr = inm->in6m_addr; 2750 in6_clearscope(&mr.mr_addr); 2751 if (!m_append(m, sizeof(mr), (void *)&mr)) { 2752 if (m != m0) 2753 m_freem(m); 2754 CTR1(KTR_MLD, 2755 "%s: m_append() failed", __func__); 2756 return (-ENOMEM); 2757 } 2758 npbytes += sizeof(struct mldv2_record); 2759 if (m != m0) { 2760 /* new packet; offset in chain */ 2761 md = m_getptr(m, npbytes - 2762 sizeof(struct mldv2_record), &off); 2763 pmr = (struct mldv2_record *)(mtod(md, 2764 uint8_t *) + off); 2765 } else { 2766 /* current packet; offset from last append */ 2767 md = m_last(m); 2768 pmr = (struct mldv2_record *)(mtod(md, 2769 uint8_t *) + md->m_len - 2770 sizeof(struct mldv2_record)); 2771 } 2772 /* 2773 * Begin walking the tree for this record type 2774 * pass, or continue from where we left off 2775 * previously if we had to allocate a new packet. 2776 * Only report deltas in-mode at t1. 2777 * We need not report included sources as allowed 2778 * if we are in inclusive mode on the group, 2779 * however the converse is not true. 2780 */ 2781 rsrcs = 0; 2782 if (nims == NULL) { 2783 nims = RB_MIN(ip6_msource_tree, 2784 &inm->in6m_srcs); 2785 } 2786 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) { 2787 CTR2(KTR_MLD, "%s: visit node %s", __func__, 2788 ip6_sprintf(ip6tbuf, &ims->im6s_addr)); 2789 now = im6s_get_mode(inm, ims, 1); 2790 then = im6s_get_mode(inm, ims, 0); 2791 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d", 2792 __func__, then, now); 2793 if (now == then) { 2794 CTR1(KTR_MLD, 2795 "%s: skip unchanged", __func__); 2796 continue; 2797 } 2798 if (mode == MCAST_EXCLUDE && 2799 now == MCAST_INCLUDE) { 2800 CTR1(KTR_MLD, 2801 "%s: skip IN src on EX group", 2802 __func__); 2803 continue; 2804 } 2805 nrt = (rectype_t)now; 2806 if (nrt == REC_NONE) 2807 nrt = (rectype_t)(~mode & REC_FULL); 2808 if (schanged++ == 0) { 2809 crt = nrt; 2810 } else if (crt != nrt) 2811 continue; 2812 if (!m_append(m, sizeof(struct in6_addr), 2813 (void *)&ims->im6s_addr)) { 2814 if (m != m0) 2815 m_freem(m); 2816 CTR1(KTR_MLD, 2817 "%s: m_append() failed", __func__); 2818 return (-ENOMEM); 2819 } 2820 nallow += !!(crt == REC_ALLOW); 2821 nblock += !!(crt == REC_BLOCK); 2822 if (++rsrcs == m0srcs) 2823 break; 2824 } 2825 /* 2826 * If we did not append any tree nodes on this 2827 * pass, back out of allocations. 2828 */ 2829 if (rsrcs == 0) { 2830 npbytes -= sizeof(struct mldv2_record); 2831 if (m != m0) { 2832 CTR1(KTR_MLD, 2833 "%s: m_free(m)", __func__); 2834 m_freem(m); 2835 } else { 2836 CTR1(KTR_MLD, 2837 "%s: m_adj(m, -mr)", __func__); 2838 m_adj(m, -((int)sizeof( 2839 struct mldv2_record))); 2840 } 2841 continue; 2842 } 2843 npbytes += (rsrcs * sizeof(struct in6_addr)); 2844 if (crt == REC_ALLOW) 2845 pmr->mr_type = MLD_ALLOW_NEW_SOURCES; 2846 else if (crt == REC_BLOCK) 2847 pmr->mr_type = MLD_BLOCK_OLD_SOURCES; 2848 pmr->mr_numsrc = htons(rsrcs); 2849 /* 2850 * Count the new group record, and enqueue this 2851 * packet if it wasn't already queued. 2852 */ 2853 m->m_pkthdr.PH_vt.vt_nrecs++; 2854 if (m != m0) 2855 mbufq_enqueue(mq, m); 2856 nbytes += npbytes; 2857 } while (nims != NULL); 2858 drt |= crt; 2859 crt = (~crt & REC_FULL); 2860 } 2861 2862 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, 2863 nallow, nblock); 2864 2865 return (nbytes); 2866 } 2867 2868 static int 2869 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq) 2870 { 2871 struct mbufq *gq; 2872 struct mbuf *m; /* pending state-change */ 2873 struct mbuf *m0; /* copy of pending state-change */ 2874 struct mbuf *mt; /* last state-change in packet */ 2875 int docopy, domerge; 2876 u_int recslen; 2877 2878 docopy = 0; 2879 domerge = 0; 2880 recslen = 0; 2881 2882 IN6_MULTI_LOCK_ASSERT(); 2883 MLD_LOCK_ASSERT(); 2884 2885 /* 2886 * If there are further pending retransmissions, make a writable 2887 * copy of each queued state-change message before merging. 2888 */ 2889 if (inm->in6m_scrv > 0) 2890 docopy = 1; 2891 2892 gq = &inm->in6m_scq; 2893 #ifdef KTR 2894 if (mbufq_first(gq) == NULL) { 2895 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty", 2896 __func__, inm); 2897 } 2898 #endif 2899 2900 m = mbufq_first(gq); 2901 while (m != NULL) { 2902 /* 2903 * Only merge the report into the current packet if 2904 * there is sufficient space to do so; an MLDv2 report 2905 * packet may only contain 65,535 group records. 2906 * Always use a simple mbuf chain concatentation to do this, 2907 * as large state changes for single groups may have 2908 * allocated clusters. 2909 */ 2910 domerge = 0; 2911 mt = mbufq_last(scq); 2912 if (mt != NULL) { 2913 recslen = m_length(m, NULL); 2914 2915 if ((mt->m_pkthdr.PH_vt.vt_nrecs + 2916 m->m_pkthdr.PH_vt.vt_nrecs <= 2917 MLD_V2_REPORT_MAXRECS) && 2918 (mt->m_pkthdr.len + recslen <= 2919 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) 2920 domerge = 1; 2921 } 2922 2923 if (!domerge && mbufq_full(gq)) { 2924 CTR2(KTR_MLD, 2925 "%s: outbound queue full, skipping whole packet %p", 2926 __func__, m); 2927 mt = m->m_nextpkt; 2928 if (!docopy) 2929 m_freem(m); 2930 m = mt; 2931 continue; 2932 } 2933 2934 if (!docopy) { 2935 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m); 2936 m0 = mbufq_dequeue(gq); 2937 m = m0->m_nextpkt; 2938 } else { 2939 CTR2(KTR_MLD, "%s: copying %p", __func__, m); 2940 m0 = m_dup(m, M_NOWAIT); 2941 if (m0 == NULL) 2942 return (ENOMEM); 2943 m0->m_nextpkt = NULL; 2944 m = m->m_nextpkt; 2945 } 2946 2947 if (!domerge) { 2948 CTR3(KTR_MLD, "%s: queueing %p to scq %p)", 2949 __func__, m0, scq); 2950 mbufq_enqueue(scq, m0); 2951 } else { 2952 struct mbuf *mtl; /* last mbuf of packet mt */ 2953 2954 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)", 2955 __func__, m0, mt); 2956 2957 mtl = m_last(mt); 2958 m0->m_flags &= ~M_PKTHDR; 2959 mt->m_pkthdr.len += recslen; 2960 mt->m_pkthdr.PH_vt.vt_nrecs += 2961 m0->m_pkthdr.PH_vt.vt_nrecs; 2962 2963 mtl->m_next = m0; 2964 } 2965 } 2966 2967 return (0); 2968 } 2969 2970 /* 2971 * Respond to a pending MLDv2 General Query. 2972 */ 2973 static void 2974 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli) 2975 { 2976 struct ifmultiaddr *ifma; 2977 struct ifnet *ifp; 2978 struct in6_multi *inm; 2979 int retval; 2980 2981 IN6_MULTI_LOCK_ASSERT(); 2982 MLD_LOCK_ASSERT(); 2983 2984 KASSERT(mli->mli_version == MLD_VERSION_2, 2985 ("%s: called when version %d", __func__, mli->mli_version)); 2986 2987 /* 2988 * Check that there are some packets queued. If so, send them first. 2989 * For large number of groups the reply to general query can take 2990 * many packets, we should finish sending them before starting of 2991 * queuing the new reply. 2992 */ 2993 if (mbufq_len(&mli->mli_gq) != 0) 2994 goto send; 2995 2996 ifp = mli->mli_ifp; 2997 2998 IF_ADDR_RLOCK(ifp); 2999 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3000 if (ifma->ifma_addr->sa_family != AF_INET6 || 3001 ifma->ifma_protospec == NULL) 3002 continue; 3003 3004 inm = (struct in6_multi *)ifma->ifma_protospec; 3005 KASSERT(ifp == inm->in6m_ifp, 3006 ("%s: inconsistent ifp", __func__)); 3007 3008 switch (inm->in6m_state) { 3009 case MLD_NOT_MEMBER: 3010 case MLD_SILENT_MEMBER: 3011 break; 3012 case MLD_REPORTING_MEMBER: 3013 case MLD_IDLE_MEMBER: 3014 case MLD_LAZY_MEMBER: 3015 case MLD_SLEEPING_MEMBER: 3016 case MLD_AWAKENING_MEMBER: 3017 inm->in6m_state = MLD_REPORTING_MEMBER; 3018 retval = mld_v2_enqueue_group_record(&mli->mli_gq, 3019 inm, 0, 0, 0, 0); 3020 CTR2(KTR_MLD, "%s: enqueue record = %d", 3021 __func__, retval); 3022 break; 3023 case MLD_G_QUERY_PENDING_MEMBER: 3024 case MLD_SG_QUERY_PENDING_MEMBER: 3025 case MLD_LEAVING_MEMBER: 3026 break; 3027 } 3028 } 3029 IF_ADDR_RUNLOCK(ifp); 3030 3031 send: 3032 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST); 3033 3034 /* 3035 * Slew transmission of bursts over 500ms intervals. 3036 */ 3037 if (mbufq_first(&mli->mli_gq) != NULL) { 3038 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY( 3039 MLD_RESPONSE_BURST_INTERVAL); 3040 V_interface_timers_running6 = 1; 3041 } 3042 } 3043 3044 /* 3045 * Transmit the next pending message in the output queue. 3046 * 3047 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. 3048 * MRT: Nothing needs to be done, as MLD traffic is always local to 3049 * a link and uses a link-scope multicast address. 3050 */ 3051 static void 3052 mld_dispatch_packet(struct mbuf *m) 3053 { 3054 struct ip6_moptions im6o; 3055 struct ifnet *ifp; 3056 struct ifnet *oifp; 3057 struct mbuf *m0; 3058 struct mbuf *md; 3059 struct ip6_hdr *ip6; 3060 struct mld_hdr *mld; 3061 int error; 3062 int off; 3063 int type; 3064 uint32_t ifindex; 3065 3066 CTR2(KTR_MLD, "%s: transmit %p", __func__, m); 3067 3068 /* 3069 * Set VNET image pointer from enqueued mbuf chain 3070 * before doing anything else. Whilst we use interface 3071 * indexes to guard against interface detach, they are 3072 * unique to each VIMAGE and must be retrieved. 3073 */ 3074 ifindex = mld_restore_context(m); 3075 3076 /* 3077 * Check if the ifnet still exists. This limits the scope of 3078 * any race in the absence of a global ifp lock for low cost 3079 * (an array lookup). 3080 */ 3081 ifp = ifnet_byindex(ifindex); 3082 if (ifp == NULL) { 3083 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.", 3084 __func__, m, ifindex); 3085 m_freem(m); 3086 IP6STAT_INC(ip6s_noroute); 3087 goto out; 3088 } 3089 3090 im6o.im6o_multicast_hlim = 1; 3091 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL); 3092 im6o.im6o_multicast_ifp = ifp; 3093 3094 if (m->m_flags & M_MLDV1) { 3095 m0 = m; 3096 } else { 3097 m0 = mld_v2_encap_report(ifp, m); 3098 if (m0 == NULL) { 3099 CTR2(KTR_MLD, "%s: dropped %p", __func__, m); 3100 IP6STAT_INC(ip6s_odropped); 3101 goto out; 3102 } 3103 } 3104 3105 mld_scrub_context(m0); 3106 m_clrprotoflags(m); 3107 m0->m_pkthdr.rcvif = V_loif; 3108 3109 ip6 = mtod(m0, struct ip6_hdr *); 3110 #if 0 3111 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */ 3112 #else 3113 /* 3114 * XXX XXX Break some KPI rules to prevent an LOR which would 3115 * occur if we called in6_setscope() at transmission. 3116 * See comments at top of file. 3117 */ 3118 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index); 3119 #endif 3120 3121 /* 3122 * Retrieve the ICMPv6 type before handoff to ip6_output(), 3123 * so we can bump the stats. 3124 */ 3125 md = m_getptr(m0, sizeof(struct ip6_hdr), &off); 3126 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off); 3127 type = mld->mld_type; 3128 3129 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o, 3130 &oifp, NULL); 3131 if (error) { 3132 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error); 3133 goto out; 3134 } 3135 ICMP6STAT_INC(icp6s_outhist[type]); 3136 if (oifp != NULL) { 3137 icmp6_ifstat_inc(oifp, ifs6_out_msg); 3138 switch (type) { 3139 case MLD_LISTENER_REPORT: 3140 case MLDV2_LISTENER_REPORT: 3141 icmp6_ifstat_inc(oifp, ifs6_out_mldreport); 3142 break; 3143 case MLD_LISTENER_DONE: 3144 icmp6_ifstat_inc(oifp, ifs6_out_mlddone); 3145 break; 3146 } 3147 } 3148 out: 3149 return; 3150 } 3151 3152 /* 3153 * Encapsulate an MLDv2 report. 3154 * 3155 * KAME IPv6 requires that hop-by-hop options be passed separately, 3156 * and that the IPv6 header be prepended in a separate mbuf. 3157 * 3158 * Returns a pointer to the new mbuf chain head, or NULL if the 3159 * allocation failed. 3160 */ 3161 static struct mbuf * 3162 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m) 3163 { 3164 struct mbuf *mh; 3165 struct mldv2_report *mld; 3166 struct ip6_hdr *ip6; 3167 struct in6_ifaddr *ia; 3168 int mldreclen; 3169 3170 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 3171 KASSERT((m->m_flags & M_PKTHDR), 3172 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); 3173 3174 /* 3175 * RFC3590: OK to send as :: or tentative during DAD. 3176 */ 3177 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST); 3178 if (ia == NULL) 3179 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__); 3180 3181 mh = m_gethdr(M_NOWAIT, MT_DATA); 3182 if (mh == NULL) { 3183 if (ia != NULL) 3184 ifa_free(&ia->ia_ifa); 3185 m_freem(m); 3186 return (NULL); 3187 } 3188 M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report)); 3189 3190 mldreclen = m_length(m, NULL); 3191 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen); 3192 3193 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report); 3194 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + 3195 sizeof(struct mldv2_report) + mldreclen; 3196 3197 ip6 = mtod(mh, struct ip6_hdr *); 3198 ip6->ip6_flow = 0; 3199 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 3200 ip6->ip6_vfc |= IPV6_VERSION; 3201 ip6->ip6_nxt = IPPROTO_ICMPV6; 3202 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any; 3203 if (ia != NULL) 3204 ifa_free(&ia->ia_ifa); 3205 ip6->ip6_dst = in6addr_linklocal_allv2routers; 3206 /* scope ID will be set in netisr */ 3207 3208 mld = (struct mldv2_report *)(ip6 + 1); 3209 mld->mld_type = MLDV2_LISTENER_REPORT; 3210 mld->mld_code = 0; 3211 mld->mld_cksum = 0; 3212 mld->mld_v2_reserved = 0; 3213 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs); 3214 m->m_pkthdr.PH_vt.vt_nrecs = 0; 3215 3216 mh->m_next = m; 3217 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6, 3218 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen); 3219 return (mh); 3220 } 3221 3222 #ifdef KTR 3223 static char * 3224 mld_rec_type_to_str(const int type) 3225 { 3226 3227 switch (type) { 3228 case MLD_CHANGE_TO_EXCLUDE_MODE: 3229 return "TO_EX"; 3230 break; 3231 case MLD_CHANGE_TO_INCLUDE_MODE: 3232 return "TO_IN"; 3233 break; 3234 case MLD_MODE_IS_EXCLUDE: 3235 return "MODE_EX"; 3236 break; 3237 case MLD_MODE_IS_INCLUDE: 3238 return "MODE_IN"; 3239 break; 3240 case MLD_ALLOW_NEW_SOURCES: 3241 return "ALLOW_NEW"; 3242 break; 3243 case MLD_BLOCK_OLD_SOURCES: 3244 return "BLOCK_OLD"; 3245 break; 3246 default: 3247 break; 3248 } 3249 return "unknown"; 3250 } 3251 #endif 3252 3253 static void 3254 mld_init(void *unused __unused) 3255 { 3256 3257 CTR1(KTR_MLD, "%s: initializing", __func__); 3258 MLD_LOCK_INIT(); 3259 3260 ip6_initpktopts(&mld_po); 3261 mld_po.ip6po_hlim = 1; 3262 mld_po.ip6po_hbh = &mld_ra.hbh; 3263 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER; 3264 mld_po.ip6po_flags = IP6PO_DONTFRAG; 3265 } 3266 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL); 3267 3268 static void 3269 mld_uninit(void *unused __unused) 3270 { 3271 3272 CTR1(KTR_MLD, "%s: tearing down", __func__); 3273 MLD_LOCK_DESTROY(); 3274 } 3275 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL); 3276 3277 static void 3278 vnet_mld_init(const void *unused __unused) 3279 { 3280 3281 CTR1(KTR_MLD, "%s: initializing", __func__); 3282 3283 LIST_INIT(&V_mli_head); 3284 } 3285 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init, 3286 NULL); 3287 3288 static void 3289 vnet_mld_uninit(const void *unused __unused) 3290 { 3291 3292 /* This can happen if we shutdown the network stack. */ 3293 CTR1(KTR_MLD, "%s: tearing down", __func__); 3294 } 3295 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit, 3296 NULL); 3297 3298 static int 3299 mld_modevent(module_t mod, int type, void *unused __unused) 3300 { 3301 3302 switch (type) { 3303 case MOD_LOAD: 3304 case MOD_UNLOAD: 3305 break; 3306 default: 3307 return (EOPNOTSUPP); 3308 } 3309 return (0); 3310 } 3311 3312 static moduledata_t mld_mod = { 3313 "mld", 3314 mld_modevent, 3315 0 3316 }; 3317 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY); 3318