1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2009 Bruce Simpson. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. The name of the author may not be used to endorse or promote 16 * products derived from this software without specific prior written 17 * permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * IPv6 multicast socket, group, and socket option processing module. 34 * Normative references: RFC 2292, RFC 3492, RFC 3542, RFC 3678, RFC 3810. 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_inet6.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/kernel.h> 45 #include <sys/ktr.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/protosw.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/sysctl.h> 52 #include <sys/priv.h> 53 #include <sys/taskqueue.h> 54 #include <sys/tree.h> 55 56 #include <net/if.h> 57 #include <net/if_var.h> 58 #include <net/if_dl.h> 59 #include <net/route.h> 60 #include <net/route/nhop.h> 61 #include <net/vnet.h> 62 63 #include <netinet/in.h> 64 #include <netinet/udp.h> 65 #include <netinet/in_var.h> 66 #include <netinet/ip_var.h> 67 #include <netinet/udp_var.h> 68 #include <netinet6/in6_fib.h> 69 #include <netinet6/in6_var.h> 70 #include <netinet/ip6.h> 71 #include <netinet/icmp6.h> 72 #include <netinet6/ip6_var.h> 73 #include <netinet/in_pcb.h> 74 #include <netinet/tcp_var.h> 75 #include <netinet6/nd6.h> 76 #include <netinet6/mld6_var.h> 77 #include <netinet6/scope6_var.h> 78 79 #ifndef KTR_MLD 80 #define KTR_MLD KTR_INET6 81 #endif 82 83 #ifndef __SOCKUNION_DECLARED 84 union sockunion { 85 struct sockaddr_storage ss; 86 struct sockaddr sa; 87 struct sockaddr_dl sdl; 88 struct sockaddr_in6 sin6; 89 }; 90 typedef union sockunion sockunion_t; 91 #define __SOCKUNION_DECLARED 92 #endif /* __SOCKUNION_DECLARED */ 93 94 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter", 95 "IPv6 multicast PCB-layer source filter"); 96 MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group"); 97 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options"); 98 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource", 99 "IPv6 multicast MLD-layer source filter"); 100 101 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp); 102 103 /* 104 * Locking: 105 * - Lock order is: Giant, IN6_MULTI_LOCK, INP_WLOCK, 106 * IN6_MULTI_LIST_LOCK, MLD_LOCK, IF_ADDR_LOCK. 107 * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however 108 * it can be taken by code in net/if.c also. 109 * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK. 110 * 111 * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly 112 * any need for in6_multi itself to be virtualized -- it is bound to an ifp 113 * anyway no matter what happens. 114 */ 115 struct mtx in6_multi_list_mtx; 116 MTX_SYSINIT(in6_multi_mtx, &in6_multi_list_mtx, "in6_multi_list_mtx", MTX_DEF); 117 118 struct mtx in6_multi_free_mtx; 119 MTX_SYSINIT(in6_multi_free_mtx, &in6_multi_free_mtx, "in6_multi_free_mtx", MTX_DEF); 120 121 struct sx in6_multi_sx; 122 SX_SYSINIT(in6_multi_sx, &in6_multi_sx, "in6_multi_sx"); 123 124 static void im6f_commit(struct in6_mfilter *); 125 static int im6f_get_source(struct in6_mfilter *imf, 126 const struct sockaddr_in6 *psin, 127 struct in6_msource **); 128 static struct in6_msource * 129 im6f_graft(struct in6_mfilter *, const uint8_t, 130 const struct sockaddr_in6 *); 131 static void im6f_leave(struct in6_mfilter *); 132 static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *); 133 static void im6f_purge(struct in6_mfilter *); 134 static void im6f_rollback(struct in6_mfilter *); 135 static void im6f_reap(struct in6_mfilter *); 136 static struct in6_mfilter * 137 im6o_match_group(const struct ip6_moptions *, 138 const struct ifnet *, const struct sockaddr *); 139 static struct in6_msource * 140 im6o_match_source(struct in6_mfilter *, const struct sockaddr *); 141 static void im6s_merge(struct ip6_msource *ims, 142 const struct in6_msource *lims, const int rollback); 143 static int in6_getmulti(struct ifnet *, const struct in6_addr *, 144 struct in6_multi **); 145 static int in6_joingroup_locked(struct ifnet *, const struct in6_addr *, 146 struct in6_mfilter *, struct in6_multi **, int); 147 static int in6m_get_source(struct in6_multi *inm, 148 const struct in6_addr *addr, const int noalloc, 149 struct ip6_msource **pims); 150 #ifdef KTR 151 static int in6m_is_ifp_detached(const struct in6_multi *); 152 #endif 153 static int in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *); 154 static void in6m_purge(struct in6_multi *); 155 static void in6m_reap(struct in6_multi *); 156 static struct ip6_moptions * 157 in6p_findmoptions(struct inpcb *); 158 static int in6p_get_source_filters(struct inpcb *, struct sockopt *); 159 static int in6p_join_group(struct inpcb *, struct sockopt *); 160 static int in6p_leave_group(struct inpcb *, struct sockopt *); 161 static struct ifnet * 162 in6p_lookup_mcast_ifp(const struct inpcb *, 163 const struct sockaddr_in6 *); 164 static int in6p_block_unblock_source(struct inpcb *, struct sockopt *); 165 static int in6p_set_multicast_if(struct inpcb *, struct sockopt *); 166 static int in6p_set_source_filters(struct inpcb *, struct sockopt *); 167 static int sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS); 168 169 SYSCTL_DECL(_net_inet6_ip6); /* XXX Not in any common header. */ 170 171 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, 172 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 173 "IPv6 multicast"); 174 175 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER; 176 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc, 177 CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0, 178 "Max source filters per group"); 179 180 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER; 181 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc, 182 CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0, 183 "Max source filters per socket"); 184 185 /* TODO Virtualize this switch. */ 186 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP; 187 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN, 188 &in6_mcast_loop, 0, "Loopback multicast datagrams by default"); 189 190 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters, 191 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters, 192 "Per-interface stack-wide source filters"); 193 194 #ifdef KTR 195 /* 196 * Inline function which wraps assertions for a valid ifp. 197 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 198 * is detached. 199 */ 200 static int __inline 201 in6m_is_ifp_detached(const struct in6_multi *inm) 202 { 203 struct ifnet *ifp; 204 205 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 206 ifp = inm->in6m_ifma->ifma_ifp; 207 if (ifp != NULL) { 208 /* 209 * Sanity check that network-layer notion of ifp is the 210 * same as that of link-layer. 211 */ 212 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 213 } 214 215 return (ifp == NULL); 216 } 217 #endif 218 219 /* 220 * Initialize an in6_mfilter structure to a known state at t0, t1 221 * with an empty source filter list. 222 */ 223 static __inline void 224 im6f_init(struct in6_mfilter *imf, const int st0, const int st1) 225 { 226 memset(imf, 0, sizeof(struct in6_mfilter)); 227 RB_INIT(&imf->im6f_sources); 228 imf->im6f_st[0] = st0; 229 imf->im6f_st[1] = st1; 230 } 231 232 struct in6_mfilter * 233 ip6_mfilter_alloc(const int mflags, const int st0, const int st1) 234 { 235 struct in6_mfilter *imf; 236 237 imf = malloc(sizeof(*imf), M_IN6MFILTER, mflags); 238 239 if (imf != NULL) 240 im6f_init(imf, st0, st1); 241 242 return (imf); 243 } 244 245 void 246 ip6_mfilter_free(struct in6_mfilter *imf) 247 { 248 249 im6f_purge(imf); 250 free(imf, M_IN6MFILTER); 251 } 252 253 /* 254 * Find an IPv6 multicast group entry for this ip6_moptions instance 255 * which matches the specified group, and optionally an interface. 256 * Return its index into the array, or -1 if not found. 257 */ 258 static struct in6_mfilter * 259 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp, 260 const struct sockaddr *group) 261 { 262 const struct sockaddr_in6 *gsin6; 263 struct in6_mfilter *imf; 264 struct in6_multi *inm; 265 266 gsin6 = (const struct sockaddr_in6 *)group; 267 268 IP6_MFILTER_FOREACH(imf, &imo->im6o_head) { 269 inm = imf->im6f_in6m; 270 if (inm == NULL) 271 continue; 272 if ((ifp == NULL || (inm->in6m_ifp == ifp)) && 273 IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, 274 &gsin6->sin6_addr)) { 275 break; 276 } 277 } 278 return (imf); 279 } 280 281 /* 282 * Find an IPv6 multicast source entry for this imo which matches 283 * the given group index for this socket, and source address. 284 * 285 * XXX TODO: The scope ID, if present in src, is stripped before 286 * any comparison. We SHOULD enforce scope/zone checks where the source 287 * filter entry has a link scope. 288 * 289 * NOTE: This does not check if the entry is in-mode, merely if 290 * it exists, which may not be the desired behaviour. 291 */ 292 static struct in6_msource * 293 im6o_match_source(struct in6_mfilter *imf, const struct sockaddr *src) 294 { 295 struct ip6_msource find; 296 struct ip6_msource *ims; 297 const sockunion_t *psa; 298 299 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__)); 300 301 psa = (const sockunion_t *)src; 302 find.im6s_addr = psa->sin6.sin6_addr; 303 in6_clearscope(&find.im6s_addr); /* XXX */ 304 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find); 305 306 return ((struct in6_msource *)ims); 307 } 308 309 /* 310 * Perform filtering for multicast datagrams on a socket by group and source. 311 * 312 * Returns 0 if a datagram should be allowed through, or various error codes 313 * if the socket was not a member of the group, or the source was muted, etc. 314 */ 315 int 316 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp, 317 const struct sockaddr *group, const struct sockaddr *src) 318 { 319 struct in6_mfilter *imf; 320 struct in6_msource *ims; 321 int mode; 322 323 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 324 325 imf = im6o_match_group(imo, ifp, group); 326 if (imf == NULL) 327 return (MCAST_NOTGMEMBER); 328 329 /* 330 * Check if the source was included in an (S,G) join. 331 * Allow reception on exclusive memberships by default, 332 * reject reception on inclusive memberships by default. 333 * Exclude source only if an in-mode exclude filter exists. 334 * Include source only if an in-mode include filter exists. 335 * NOTE: We are comparing group state here at MLD t1 (now) 336 * with socket-layer t0 (since last downcall). 337 */ 338 mode = imf->im6f_st[1]; 339 ims = im6o_match_source(imf, src); 340 341 if ((ims == NULL && mode == MCAST_INCLUDE) || 342 (ims != NULL && ims->im6sl_st[0] != mode)) 343 return (MCAST_NOTSMEMBER); 344 345 return (MCAST_PASS); 346 } 347 348 /* 349 * Find and return a reference to an in6_multi record for (ifp, group), 350 * and bump its reference count. 351 * If one does not exist, try to allocate it, and update link-layer multicast 352 * filters on ifp to listen for group. 353 * Assumes the IN6_MULTI lock is held across the call. 354 * Return 0 if successful, otherwise return an appropriate error code. 355 */ 356 static int 357 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group, 358 struct in6_multi **pinm) 359 { 360 struct epoch_tracker et; 361 struct sockaddr_in6 gsin6; 362 struct ifmultiaddr *ifma; 363 struct in6_multi *inm; 364 int error; 365 366 error = 0; 367 368 /* 369 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK; 370 * if_addmulti() takes this mutex itself, so we must drop and 371 * re-acquire around the call. 372 */ 373 IN6_MULTI_LOCK_ASSERT(); 374 IN6_MULTI_LIST_LOCK(); 375 IF_ADDR_WLOCK(ifp); 376 NET_EPOCH_ENTER(et); 377 inm = in6m_lookup_locked(ifp, group); 378 NET_EPOCH_EXIT(et); 379 380 if (inm != NULL) { 381 /* 382 * If we already joined this group, just bump the 383 * refcount and return it. 384 */ 385 KASSERT(inm->in6m_refcount >= 1, 386 ("%s: bad refcount %d", __func__, inm->in6m_refcount)); 387 in6m_acquire_locked(inm); 388 *pinm = inm; 389 goto out_locked; 390 } 391 392 memset(&gsin6, 0, sizeof(gsin6)); 393 gsin6.sin6_family = AF_INET6; 394 gsin6.sin6_len = sizeof(struct sockaddr_in6); 395 gsin6.sin6_addr = *group; 396 397 /* 398 * Check if a link-layer group is already associated 399 * with this network-layer group on the given ifnet. 400 */ 401 IN6_MULTI_LIST_UNLOCK(); 402 IF_ADDR_WUNLOCK(ifp); 403 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma); 404 if (error != 0) 405 return (error); 406 IN6_MULTI_LIST_LOCK(); 407 IF_ADDR_WLOCK(ifp); 408 409 /* 410 * If something other than netinet6 is occupying the link-layer 411 * group, print a meaningful error message and back out of 412 * the allocation. 413 * Otherwise, bump the refcount on the existing network-layer 414 * group association and return it. 415 */ 416 if (ifma->ifma_protospec != NULL) { 417 inm = (struct in6_multi *)ifma->ifma_protospec; 418 #ifdef INVARIANTS 419 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 420 __func__)); 421 KASSERT(ifma->ifma_addr->sa_family == AF_INET6, 422 ("%s: ifma not AF_INET6", __func__)); 423 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 424 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp || 425 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group)) 426 panic("%s: ifma %p is inconsistent with %p (%p)", 427 __func__, ifma, inm, group); 428 #endif 429 in6m_acquire_locked(inm); 430 *pinm = inm; 431 goto out_locked; 432 } 433 434 IF_ADDR_WLOCK_ASSERT(ifp); 435 436 /* 437 * A new in6_multi record is needed; allocate and initialize it. 438 * We DO NOT perform an MLD join as the in6_ layer may need to 439 * push an initial source list down to MLD to support SSM. 440 * 441 * The initial source filter state is INCLUDE, {} as per the RFC. 442 * Pending state-changes per group are subject to a bounds check. 443 */ 444 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO); 445 if (inm == NULL) { 446 IN6_MULTI_LIST_UNLOCK(); 447 IF_ADDR_WUNLOCK(ifp); 448 if_delmulti_ifma(ifma); 449 return (ENOMEM); 450 } 451 inm->in6m_addr = *group; 452 inm->in6m_ifp = ifp; 453 inm->in6m_mli = MLD_IFINFO(ifp); 454 inm->in6m_ifma = ifma; 455 inm->in6m_refcount = 1; 456 inm->in6m_state = MLD_NOT_MEMBER; 457 mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES); 458 459 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED; 460 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 461 RB_INIT(&inm->in6m_srcs); 462 463 ifma->ifma_protospec = inm; 464 *pinm = inm; 465 466 out_locked: 467 IN6_MULTI_LIST_UNLOCK(); 468 IF_ADDR_WUNLOCK(ifp); 469 return (error); 470 } 471 472 /* 473 * Drop a reference to an in6_multi record. 474 * 475 * If the refcount drops to 0, free the in6_multi record and 476 * delete the underlying link-layer membership. 477 */ 478 static void 479 in6m_release(struct in6_multi *inm) 480 { 481 struct ifmultiaddr *ifma; 482 struct ifnet *ifp; 483 484 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount); 485 486 MPASS(inm->in6m_refcount == 0); 487 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm); 488 489 ifma = inm->in6m_ifma; 490 ifp = inm->in6m_ifp; 491 MPASS(ifma->ifma_llifma == NULL); 492 493 /* XXX this access is not covered by IF_ADDR_LOCK */ 494 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma); 495 KASSERT(ifma->ifma_protospec == NULL, 496 ("%s: ifma_protospec != NULL", __func__)); 497 if (ifp == NULL) 498 ifp = ifma->ifma_ifp; 499 500 if (ifp != NULL) { 501 CURVNET_SET(ifp->if_vnet); 502 in6m_purge(inm); 503 free(inm, M_IP6MADDR); 504 if_delmulti_ifma_flags(ifma, 1); 505 CURVNET_RESTORE(); 506 if_rele(ifp); 507 } else { 508 in6m_purge(inm); 509 free(inm, M_IP6MADDR); 510 if_delmulti_ifma_flags(ifma, 1); 511 } 512 } 513 514 /* 515 * Interface detach can happen in a taskqueue thread context, so we must use a 516 * dedicated thread to avoid deadlocks when draining in6m_release tasks. 517 */ 518 TASKQUEUE_DEFINE_THREAD(in6m_free); 519 static struct task in6m_free_task; 520 static struct in6_multi_head in6m_free_list = SLIST_HEAD_INITIALIZER(); 521 static void in6m_release_task(void *arg __unused, int pending __unused); 522 523 static void 524 in6m_init(void) 525 { 526 TASK_INIT(&in6m_free_task, 0, in6m_release_task, NULL); 527 } 528 SYSINIT(in6m_init, SI_SUB_TASKQ, SI_ORDER_ANY, in6m_init, NULL); 529 530 void 531 in6m_release_list_deferred(struct in6_multi_head *inmh) 532 { 533 if (SLIST_EMPTY(inmh)) 534 return; 535 mtx_lock(&in6_multi_free_mtx); 536 SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele); 537 mtx_unlock(&in6_multi_free_mtx); 538 taskqueue_enqueue(taskqueue_in6m_free, &in6m_free_task); 539 } 540 541 void 542 in6m_release_wait(void) 543 { 544 taskqueue_drain_all(taskqueue_in6m_free); 545 } 546 547 void 548 in6m_disconnect_locked(struct in6_multi_head *inmh, struct in6_multi *inm) 549 { 550 struct ifnet *ifp; 551 struct ifaddr *ifa; 552 struct in6_ifaddr *ifa6; 553 struct in6_multi_mship *imm, *imm_tmp; 554 struct ifmultiaddr *ifma, *ll_ifma; 555 556 IN6_MULTI_LIST_LOCK_ASSERT(); 557 558 ifp = inm->in6m_ifp; 559 if (ifp == NULL) 560 return; /* already called */ 561 562 inm->in6m_ifp = NULL; 563 IF_ADDR_WLOCK_ASSERT(ifp); 564 ifma = inm->in6m_ifma; 565 if (ifma == NULL) 566 return; 567 568 if_ref(ifp); 569 if (ifma->ifma_flags & IFMA_F_ENQUEUED) { 570 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 571 ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 572 } 573 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname); 574 if ((ll_ifma = ifma->ifma_llifma) != NULL) { 575 MPASS(ifma != ll_ifma); 576 ifma->ifma_llifma = NULL; 577 MPASS(ll_ifma->ifma_llifma == NULL); 578 MPASS(ll_ifma->ifma_ifp == ifp); 579 if (--ll_ifma->ifma_refcount == 0) { 580 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) { 581 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link); 582 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 583 } 584 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname); 585 if_freemulti(ll_ifma); 586 } 587 } 588 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 589 if (ifa->ifa_addr->sa_family != AF_INET6) 590 continue; 591 ifa6 = (void *)ifa; 592 LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships, 593 i6mm_chain, imm_tmp) { 594 if (inm == imm->i6mm_maddr) { 595 LIST_REMOVE(imm, i6mm_chain); 596 free(imm, M_IP6MADDR); 597 in6m_rele_locked(inmh, inm); 598 } 599 } 600 } 601 } 602 603 static void 604 in6m_release_task(void *arg __unused, int pending __unused) 605 { 606 struct in6_multi_head in6m_free_tmp; 607 struct in6_multi *inm, *tinm; 608 609 SLIST_INIT(&in6m_free_tmp); 610 mtx_lock(&in6_multi_free_mtx); 611 SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele); 612 mtx_unlock(&in6_multi_free_mtx); 613 IN6_MULTI_LOCK(); 614 SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) { 615 SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele); 616 in6m_release(inm); 617 } 618 IN6_MULTI_UNLOCK(); 619 } 620 621 /* 622 * Clear recorded source entries for a group. 623 * Used by the MLD code. Caller must hold the IN6_MULTI lock. 624 * FIXME: Should reap. 625 */ 626 void 627 in6m_clear_recorded(struct in6_multi *inm) 628 { 629 struct ip6_msource *ims; 630 631 IN6_MULTI_LIST_LOCK_ASSERT(); 632 633 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) { 634 if (ims->im6s_stp) { 635 ims->im6s_stp = 0; 636 --inm->in6m_st[1].iss_rec; 637 } 638 } 639 KASSERT(inm->in6m_st[1].iss_rec == 0, 640 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec)); 641 } 642 643 /* 644 * Record a source as pending for a Source-Group MLDv2 query. 645 * This lives here as it modifies the shared tree. 646 * 647 * inm is the group descriptor. 648 * naddr is the address of the source to record in network-byte order. 649 * 650 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will 651 * lazy-allocate a source node in response to an SG query. 652 * Otherwise, no allocation is performed. This saves some memory 653 * with the trade-off that the source will not be reported to the 654 * router if joined in the window between the query response and 655 * the group actually being joined on the local host. 656 * 657 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed. 658 * This turns off the allocation of a recorded source entry if 659 * the group has not been joined. 660 * 661 * Return 0 if the source didn't exist or was already marked as recorded. 662 * Return 1 if the source was marked as recorded by this function. 663 * Return <0 if any error occurred (negated errno code). 664 */ 665 int 666 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr) 667 { 668 struct ip6_msource find; 669 struct ip6_msource *ims, *nims; 670 671 IN6_MULTI_LIST_LOCK_ASSERT(); 672 673 find.im6s_addr = *addr; 674 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find); 675 if (ims && ims->im6s_stp) 676 return (0); 677 if (ims == NULL) { 678 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc) 679 return (-ENOSPC); 680 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE, 681 M_NOWAIT | M_ZERO); 682 if (nims == NULL) 683 return (-ENOMEM); 684 nims->im6s_addr = find.im6s_addr; 685 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims); 686 ++inm->in6m_nsrc; 687 ims = nims; 688 } 689 690 /* 691 * Mark the source as recorded and update the recorded 692 * source count. 693 */ 694 ++ims->im6s_stp; 695 ++inm->in6m_st[1].iss_rec; 696 697 return (1); 698 } 699 700 /* 701 * Return a pointer to an in6_msource owned by an in6_mfilter, 702 * given its source address. 703 * Lazy-allocate if needed. If this is a new entry its filter state is 704 * undefined at t0. 705 * 706 * imf is the filter set being modified. 707 * addr is the source address. 708 * 709 * SMPng: May be called with locks held; malloc must not block. 710 */ 711 static int 712 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin, 713 struct in6_msource **plims) 714 { 715 struct ip6_msource find; 716 struct ip6_msource *ims, *nims; 717 struct in6_msource *lims; 718 int error; 719 720 error = 0; 721 ims = NULL; 722 lims = NULL; 723 724 find.im6s_addr = psin->sin6_addr; 725 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find); 726 lims = (struct in6_msource *)ims; 727 if (lims == NULL) { 728 if (imf->im6f_nsrc == in6_mcast_maxsocksrc) 729 return (ENOSPC); 730 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER, 731 M_NOWAIT | M_ZERO); 732 if (nims == NULL) 733 return (ENOMEM); 734 lims = (struct in6_msource *)nims; 735 lims->im6s_addr = find.im6s_addr; 736 lims->im6sl_st[0] = MCAST_UNDEFINED; 737 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims); 738 ++imf->im6f_nsrc; 739 } 740 741 *plims = lims; 742 743 return (error); 744 } 745 746 /* 747 * Graft a source entry into an existing socket-layer filter set, 748 * maintaining any required invariants and checking allocations. 749 * 750 * The source is marked as being in the new filter mode at t1. 751 * 752 * Return the pointer to the new node, otherwise return NULL. 753 */ 754 static struct in6_msource * 755 im6f_graft(struct in6_mfilter *imf, const uint8_t st1, 756 const struct sockaddr_in6 *psin) 757 { 758 struct ip6_msource *nims; 759 struct in6_msource *lims; 760 761 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER, 762 M_NOWAIT | M_ZERO); 763 if (nims == NULL) 764 return (NULL); 765 lims = (struct in6_msource *)nims; 766 lims->im6s_addr = psin->sin6_addr; 767 lims->im6sl_st[0] = MCAST_UNDEFINED; 768 lims->im6sl_st[1] = st1; 769 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims); 770 ++imf->im6f_nsrc; 771 772 return (lims); 773 } 774 775 /* 776 * Prune a source entry from an existing socket-layer filter set, 777 * maintaining any required invariants and checking allocations. 778 * 779 * The source is marked as being left at t1, it is not freed. 780 * 781 * Return 0 if no error occurred, otherwise return an errno value. 782 */ 783 static int 784 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin) 785 { 786 struct ip6_msource find; 787 struct ip6_msource *ims; 788 struct in6_msource *lims; 789 790 find.im6s_addr = psin->sin6_addr; 791 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find); 792 if (ims == NULL) 793 return (ENOENT); 794 lims = (struct in6_msource *)ims; 795 lims->im6sl_st[1] = MCAST_UNDEFINED; 796 return (0); 797 } 798 799 /* 800 * Revert socket-layer filter set deltas at t1 to t0 state. 801 */ 802 static void 803 im6f_rollback(struct in6_mfilter *imf) 804 { 805 struct ip6_msource *ims, *tims; 806 struct in6_msource *lims; 807 808 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) { 809 lims = (struct in6_msource *)ims; 810 if (lims->im6sl_st[0] == lims->im6sl_st[1]) { 811 /* no change at t1 */ 812 continue; 813 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) { 814 /* revert change to existing source at t1 */ 815 lims->im6sl_st[1] = lims->im6sl_st[0]; 816 } else { 817 /* revert source added t1 */ 818 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims); 819 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims); 820 free(ims, M_IN6MFILTER); 821 imf->im6f_nsrc--; 822 } 823 } 824 imf->im6f_st[1] = imf->im6f_st[0]; 825 } 826 827 /* 828 * Mark socket-layer filter set as INCLUDE {} at t1. 829 */ 830 static void 831 im6f_leave(struct in6_mfilter *imf) 832 { 833 struct ip6_msource *ims; 834 struct in6_msource *lims; 835 836 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) { 837 lims = (struct in6_msource *)ims; 838 lims->im6sl_st[1] = MCAST_UNDEFINED; 839 } 840 imf->im6f_st[1] = MCAST_INCLUDE; 841 } 842 843 /* 844 * Mark socket-layer filter set deltas as committed. 845 */ 846 static void 847 im6f_commit(struct in6_mfilter *imf) 848 { 849 struct ip6_msource *ims; 850 struct in6_msource *lims; 851 852 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) { 853 lims = (struct in6_msource *)ims; 854 lims->im6sl_st[0] = lims->im6sl_st[1]; 855 } 856 imf->im6f_st[0] = imf->im6f_st[1]; 857 } 858 859 /* 860 * Reap unreferenced sources from socket-layer filter set. 861 */ 862 static void 863 im6f_reap(struct in6_mfilter *imf) 864 { 865 struct ip6_msource *ims, *tims; 866 struct in6_msource *lims; 867 868 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) { 869 lims = (struct in6_msource *)ims; 870 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) && 871 (lims->im6sl_st[1] == MCAST_UNDEFINED)) { 872 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims); 873 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims); 874 free(ims, M_IN6MFILTER); 875 imf->im6f_nsrc--; 876 } 877 } 878 } 879 880 /* 881 * Purge socket-layer filter set. 882 */ 883 static void 884 im6f_purge(struct in6_mfilter *imf) 885 { 886 struct ip6_msource *ims, *tims; 887 888 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) { 889 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims); 890 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims); 891 free(ims, M_IN6MFILTER); 892 imf->im6f_nsrc--; 893 } 894 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED; 895 KASSERT(RB_EMPTY(&imf->im6f_sources), 896 ("%s: im6f_sources not empty", __func__)); 897 } 898 899 /* 900 * Look up a source filter entry for a multicast group. 901 * 902 * inm is the group descriptor to work with. 903 * addr is the IPv6 address to look up. 904 * noalloc may be non-zero to suppress allocation of sources. 905 * *pims will be set to the address of the retrieved or allocated source. 906 * 907 * SMPng: NOTE: may be called with locks held. 908 * Return 0 if successful, otherwise return a non-zero error code. 909 */ 910 static int 911 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr, 912 const int noalloc, struct ip6_msource **pims) 913 { 914 struct ip6_msource find; 915 struct ip6_msource *ims, *nims; 916 #ifdef KTR 917 char ip6tbuf[INET6_ADDRSTRLEN]; 918 #endif 919 920 find.im6s_addr = *addr; 921 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find); 922 if (ims == NULL && !noalloc) { 923 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc) 924 return (ENOSPC); 925 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE, 926 M_NOWAIT | M_ZERO); 927 if (nims == NULL) 928 return (ENOMEM); 929 nims->im6s_addr = *addr; 930 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims); 931 ++inm->in6m_nsrc; 932 ims = nims; 933 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__, 934 ip6_sprintf(ip6tbuf, addr), ims); 935 } 936 937 *pims = ims; 938 return (0); 939 } 940 941 /* 942 * Merge socket-layer source into MLD-layer source. 943 * If rollback is non-zero, perform the inverse of the merge. 944 */ 945 static void 946 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims, 947 const int rollback) 948 { 949 int n = rollback ? -1 : 1; 950 #ifdef KTR 951 char ip6tbuf[INET6_ADDRSTRLEN]; 952 953 ip6_sprintf(ip6tbuf, &lims->im6s_addr); 954 #endif 955 956 if (lims->im6sl_st[0] == MCAST_EXCLUDE) { 957 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf); 958 ims->im6s_st[1].ex -= n; 959 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) { 960 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf); 961 ims->im6s_st[1].in -= n; 962 } 963 964 if (lims->im6sl_st[1] == MCAST_EXCLUDE) { 965 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf); 966 ims->im6s_st[1].ex += n; 967 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) { 968 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf); 969 ims->im6s_st[1].in += n; 970 } 971 } 972 973 /* 974 * Atomically update the global in6_multi state, when a membership's 975 * filter list is being updated in any way. 976 * 977 * imf is the per-inpcb-membership group filter pointer. 978 * A fake imf may be passed for in-kernel consumers. 979 * 980 * XXX This is a candidate for a set-symmetric-difference style loop 981 * which would eliminate the repeated lookup from root of ims nodes, 982 * as they share the same key space. 983 * 984 * If any error occurred this function will back out of refcounts 985 * and return a non-zero value. 986 */ 987 static int 988 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf) 989 { 990 struct ip6_msource *ims, *nims; 991 struct in6_msource *lims; 992 int schanged, error; 993 int nsrc0, nsrc1; 994 995 schanged = 0; 996 error = 0; 997 nsrc1 = nsrc0 = 0; 998 IN6_MULTI_LIST_LOCK_ASSERT(); 999 1000 /* 1001 * Update the source filters first, as this may fail. 1002 * Maintain count of in-mode filters at t0, t1. These are 1003 * used to work out if we transition into ASM mode or not. 1004 * Maintain a count of source filters whose state was 1005 * actually modified by this operation. 1006 */ 1007 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) { 1008 lims = (struct in6_msource *)ims; 1009 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++; 1010 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++; 1011 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue; 1012 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims); 1013 ++schanged; 1014 if (error) 1015 break; 1016 im6s_merge(nims, lims, 0); 1017 } 1018 if (error) { 1019 struct ip6_msource *bims; 1020 1021 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) { 1022 lims = (struct in6_msource *)ims; 1023 if (lims->im6sl_st[0] == lims->im6sl_st[1]) 1024 continue; 1025 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims); 1026 if (bims == NULL) 1027 continue; 1028 im6s_merge(bims, lims, 1); 1029 } 1030 goto out_reap; 1031 } 1032 1033 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1", 1034 __func__, nsrc0, nsrc1); 1035 1036 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 1037 if (imf->im6f_st[0] == imf->im6f_st[1] && 1038 imf->im6f_st[1] == MCAST_INCLUDE) { 1039 if (nsrc1 == 0) { 1040 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__); 1041 --inm->in6m_st[1].iss_in; 1042 } 1043 } 1044 1045 /* Handle filter mode transition on socket. */ 1046 if (imf->im6f_st[0] != imf->im6f_st[1]) { 1047 CTR3(KTR_MLD, "%s: imf transition %d to %d", 1048 __func__, imf->im6f_st[0], imf->im6f_st[1]); 1049 1050 if (imf->im6f_st[0] == MCAST_EXCLUDE) { 1051 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__); 1052 --inm->in6m_st[1].iss_ex; 1053 } else if (imf->im6f_st[0] == MCAST_INCLUDE) { 1054 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__); 1055 --inm->in6m_st[1].iss_in; 1056 } 1057 1058 if (imf->im6f_st[1] == MCAST_EXCLUDE) { 1059 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__); 1060 inm->in6m_st[1].iss_ex++; 1061 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) { 1062 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__); 1063 inm->in6m_st[1].iss_in++; 1064 } 1065 } 1066 1067 /* 1068 * Track inm filter state in terms of listener counts. 1069 * If there are any exclusive listeners, stack-wide 1070 * membership is exclusive. 1071 * Otherwise, if only inclusive listeners, stack-wide is inclusive. 1072 * If no listeners remain, state is undefined at t1, 1073 * and the MLD lifecycle for this group should finish. 1074 */ 1075 if (inm->in6m_st[1].iss_ex > 0) { 1076 CTR1(KTR_MLD, "%s: transition to EX", __func__); 1077 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE; 1078 } else if (inm->in6m_st[1].iss_in > 0) { 1079 CTR1(KTR_MLD, "%s: transition to IN", __func__); 1080 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE; 1081 } else { 1082 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__); 1083 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 1084 } 1085 1086 /* Decrement ASM listener count on transition out of ASM mode. */ 1087 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1088 if ((imf->im6f_st[1] != MCAST_EXCLUDE) || 1089 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) { 1090 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__); 1091 --inm->in6m_st[1].iss_asm; 1092 } 1093 } 1094 1095 /* Increment ASM listener count on transition to ASM mode. */ 1096 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1097 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__); 1098 inm->in6m_st[1].iss_asm++; 1099 } 1100 1101 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm); 1102 in6m_print(inm); 1103 1104 out_reap: 1105 if (schanged > 0) { 1106 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__); 1107 in6m_reap(inm); 1108 } 1109 return (error); 1110 } 1111 1112 /* 1113 * Mark an in6_multi's filter set deltas as committed. 1114 * Called by MLD after a state change has been enqueued. 1115 */ 1116 void 1117 in6m_commit(struct in6_multi *inm) 1118 { 1119 struct ip6_msource *ims; 1120 1121 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm); 1122 CTR1(KTR_MLD, "%s: pre commit:", __func__); 1123 in6m_print(inm); 1124 1125 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) { 1126 ims->im6s_st[0] = ims->im6s_st[1]; 1127 } 1128 inm->in6m_st[0] = inm->in6m_st[1]; 1129 } 1130 1131 /* 1132 * Reap unreferenced nodes from an in6_multi's filter set. 1133 */ 1134 static void 1135 in6m_reap(struct in6_multi *inm) 1136 { 1137 struct ip6_msource *ims, *tims; 1138 1139 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) { 1140 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 || 1141 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 || 1142 ims->im6s_stp != 0) 1143 continue; 1144 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims); 1145 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims); 1146 free(ims, M_IP6MSOURCE); 1147 inm->in6m_nsrc--; 1148 } 1149 } 1150 1151 /* 1152 * Purge all source nodes from an in6_multi's filter set. 1153 */ 1154 static void 1155 in6m_purge(struct in6_multi *inm) 1156 { 1157 struct ip6_msource *ims, *tims; 1158 1159 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) { 1160 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims); 1161 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims); 1162 free(ims, M_IP6MSOURCE); 1163 inm->in6m_nsrc--; 1164 } 1165 /* Free state-change requests that might be queued. */ 1166 mbufq_drain(&inm->in6m_scq); 1167 } 1168 1169 /* 1170 * Join a multicast address w/o sources. 1171 * KAME compatibility entry point. 1172 * 1173 * SMPng: Assume no mc locks held by caller. 1174 */ 1175 int 1176 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr, 1177 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm, 1178 const int delay) 1179 { 1180 int error; 1181 1182 IN6_MULTI_LOCK(); 1183 error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay); 1184 IN6_MULTI_UNLOCK(); 1185 return (error); 1186 } 1187 1188 /* 1189 * Join a multicast group; real entry point. 1190 * 1191 * Only preserves atomicity at inm level. 1192 * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1193 * 1194 * If the MLD downcall fails, the group is not joined, and an error 1195 * code is returned. 1196 */ 1197 static int 1198 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr, 1199 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm, 1200 const int delay) 1201 { 1202 struct in6_multi_head inmh; 1203 struct in6_mfilter timf; 1204 struct in6_multi *inm; 1205 struct ifmultiaddr *ifma; 1206 int error; 1207 #ifdef KTR 1208 char ip6tbuf[INET6_ADDRSTRLEN]; 1209 #endif 1210 1211 /* 1212 * Sanity: Check scope zone ID was set for ifp, if and 1213 * only if group is scoped to an interface. 1214 */ 1215 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr), 1216 ("%s: not a multicast address", __func__)); 1217 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) || 1218 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) { 1219 KASSERT(mcaddr->s6_addr16[1] != 0, 1220 ("%s: scope zone ID not set", __func__)); 1221 } 1222 1223 IN6_MULTI_LOCK_ASSERT(); 1224 IN6_MULTI_LIST_UNLOCK_ASSERT(); 1225 1226 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__, 1227 ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp)); 1228 1229 error = 0; 1230 inm = NULL; 1231 1232 /* 1233 * If no imf was specified (i.e. kernel consumer), 1234 * fake one up and assume it is an ASM join. 1235 */ 1236 if (imf == NULL) { 1237 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1238 imf = &timf; 1239 } 1240 error = in6_getmulti(ifp, mcaddr, &inm); 1241 if (error) { 1242 CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__); 1243 return (error); 1244 } 1245 1246 IN6_MULTI_LIST_LOCK(); 1247 CTR1(KTR_MLD, "%s: merge inm state", __func__); 1248 error = in6m_merge(inm, imf); 1249 if (error) { 1250 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__); 1251 goto out_in6m_release; 1252 } 1253 1254 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 1255 error = mld_change_state(inm, delay); 1256 if (error) { 1257 CTR1(KTR_MLD, "%s: failed to update source", __func__); 1258 goto out_in6m_release; 1259 } 1260 1261 out_in6m_release: 1262 SLIST_INIT(&inmh); 1263 if (error) { 1264 struct epoch_tracker et; 1265 1266 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm); 1267 IF_ADDR_WLOCK(ifp); 1268 NET_EPOCH_ENTER(et); 1269 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1270 if (ifma->ifma_protospec == inm) { 1271 ifma->ifma_protospec = NULL; 1272 break; 1273 } 1274 } 1275 in6m_disconnect_locked(&inmh, inm); 1276 in6m_rele_locked(&inmh, inm); 1277 NET_EPOCH_EXIT(et); 1278 IF_ADDR_WUNLOCK(ifp); 1279 } else { 1280 *pinm = inm; 1281 } 1282 IN6_MULTI_LIST_UNLOCK(); 1283 in6m_release_list_deferred(&inmh); 1284 return (error); 1285 } 1286 1287 /* 1288 * Leave a multicast group; unlocked entry point. 1289 */ 1290 int 1291 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf) 1292 { 1293 int error; 1294 1295 IN6_MULTI_LOCK(); 1296 error = in6_leavegroup_locked(inm, imf); 1297 IN6_MULTI_UNLOCK(); 1298 return (error); 1299 } 1300 1301 /* 1302 * Leave a multicast group; real entry point. 1303 * All source filters will be expunged. 1304 * 1305 * Only preserves atomicity at inm level. 1306 * 1307 * Holding the write lock for the INP which contains imf 1308 * is highly advisable. We can't assert for it as imf does not 1309 * contain a back-pointer to the owning inp. 1310 * 1311 * Note: This is not the same as in6m_release(*) as this function also 1312 * makes a state change downcall into MLD. 1313 */ 1314 int 1315 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf) 1316 { 1317 struct in6_multi_head inmh; 1318 struct in6_mfilter timf; 1319 struct ifnet *ifp; 1320 int error; 1321 #ifdef KTR 1322 char ip6tbuf[INET6_ADDRSTRLEN]; 1323 #endif 1324 1325 error = 0; 1326 1327 IN6_MULTI_LOCK_ASSERT(); 1328 1329 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__, 1330 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1331 (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)), 1332 imf); 1333 1334 /* 1335 * If no imf was specified (i.e. kernel consumer), 1336 * fake one up and assume it is an ASM join. 1337 */ 1338 if (imf == NULL) { 1339 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1340 imf = &timf; 1341 } 1342 1343 /* 1344 * Begin state merge transaction at MLD layer. 1345 * 1346 * As this particular invocation should not cause any memory 1347 * to be allocated, and there is no opportunity to roll back 1348 * the transaction, it MUST NOT fail. 1349 */ 1350 1351 ifp = inm->in6m_ifp; 1352 IN6_MULTI_LIST_LOCK(); 1353 CTR1(KTR_MLD, "%s: merge inm state", __func__); 1354 error = in6m_merge(inm, imf); 1355 KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1356 1357 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 1358 error = 0; 1359 if (ifp) 1360 error = mld_change_state(inm, 0); 1361 if (error) 1362 CTR1(KTR_MLD, "%s: failed mld downcall", __func__); 1363 1364 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm); 1365 if (ifp) 1366 IF_ADDR_WLOCK(ifp); 1367 1368 SLIST_INIT(&inmh); 1369 if (inm->in6m_refcount == 1) 1370 in6m_disconnect_locked(&inmh, inm); 1371 in6m_rele_locked(&inmh, inm); 1372 if (ifp) 1373 IF_ADDR_WUNLOCK(ifp); 1374 IN6_MULTI_LIST_UNLOCK(); 1375 in6m_release_list_deferred(&inmh); 1376 return (error); 1377 } 1378 1379 1380 /* 1381 * Block or unblock an ASM multicast source on an inpcb. 1382 * This implements the delta-based API described in RFC 3678. 1383 * 1384 * The delta-based API applies only to exclusive-mode memberships. 1385 * An MLD downcall will be performed. 1386 * 1387 * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1388 * 1389 * Return 0 if successful, otherwise return an appropriate error code. 1390 */ 1391 static int 1392 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1393 { 1394 struct group_source_req gsr; 1395 sockunion_t *gsa, *ssa; 1396 struct ifnet *ifp; 1397 struct in6_mfilter *imf; 1398 struct ip6_moptions *imo; 1399 struct in6_msource *ims; 1400 struct in6_multi *inm; 1401 uint16_t fmode; 1402 int error, doblock; 1403 #ifdef KTR 1404 char ip6tbuf[INET6_ADDRSTRLEN]; 1405 #endif 1406 1407 ifp = NULL; 1408 error = 0; 1409 doblock = 0; 1410 1411 memset(&gsr, 0, sizeof(struct group_source_req)); 1412 gsa = (sockunion_t *)&gsr.gsr_group; 1413 ssa = (sockunion_t *)&gsr.gsr_source; 1414 1415 switch (sopt->sopt_name) { 1416 case MCAST_BLOCK_SOURCE: 1417 case MCAST_UNBLOCK_SOURCE: 1418 error = sooptcopyin(sopt, &gsr, 1419 sizeof(struct group_source_req), 1420 sizeof(struct group_source_req)); 1421 if (error) 1422 return (error); 1423 1424 if (gsa->sin6.sin6_family != AF_INET6 || 1425 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 1426 return (EINVAL); 1427 1428 if (ssa->sin6.sin6_family != AF_INET6 || 1429 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 1430 return (EINVAL); 1431 1432 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1433 return (EADDRNOTAVAIL); 1434 1435 ifp = ifnet_byindex(gsr.gsr_interface); 1436 1437 if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1438 doblock = 1; 1439 break; 1440 1441 default: 1442 CTR2(KTR_MLD, "%s: unknown sopt_name %d", 1443 __func__, sopt->sopt_name); 1444 return (EOPNOTSUPP); 1445 break; 1446 } 1447 1448 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 1449 return (EINVAL); 1450 1451 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 1452 1453 /* 1454 * Check if we are actually a member of this group. 1455 */ 1456 imo = in6p_findmoptions(inp); 1457 imf = im6o_match_group(imo, ifp, &gsa->sa); 1458 if (imf == NULL) { 1459 error = EADDRNOTAVAIL; 1460 goto out_in6p_locked; 1461 } 1462 inm = imf->im6f_in6m; 1463 1464 /* 1465 * Attempting to use the delta-based API on an 1466 * non exclusive-mode membership is an error. 1467 */ 1468 fmode = imf->im6f_st[0]; 1469 if (fmode != MCAST_EXCLUDE) { 1470 error = EINVAL; 1471 goto out_in6p_locked; 1472 } 1473 1474 /* 1475 * Deal with error cases up-front: 1476 * Asked to block, but already blocked; or 1477 * Asked to unblock, but nothing to unblock. 1478 * If adding a new block entry, allocate it. 1479 */ 1480 ims = im6o_match_source(imf, &ssa->sa); 1481 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1482 CTR3(KTR_MLD, "%s: source %s %spresent", __func__, 1483 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr), 1484 doblock ? "" : "not "); 1485 error = EADDRNOTAVAIL; 1486 goto out_in6p_locked; 1487 } 1488 1489 INP_WLOCK_ASSERT(inp); 1490 1491 /* 1492 * Begin state merge transaction at socket layer. 1493 */ 1494 if (doblock) { 1495 CTR2(KTR_MLD, "%s: %s source", __func__, "block"); 1496 ims = im6f_graft(imf, fmode, &ssa->sin6); 1497 if (ims == NULL) 1498 error = ENOMEM; 1499 } else { 1500 CTR2(KTR_MLD, "%s: %s source", __func__, "allow"); 1501 error = im6f_prune(imf, &ssa->sin6); 1502 } 1503 1504 if (error) { 1505 CTR1(KTR_MLD, "%s: merge imf state failed", __func__); 1506 goto out_im6f_rollback; 1507 } 1508 1509 /* 1510 * Begin state merge transaction at MLD layer. 1511 */ 1512 IN6_MULTI_LIST_LOCK(); 1513 CTR1(KTR_MLD, "%s: merge inm state", __func__); 1514 error = in6m_merge(inm, imf); 1515 if (error) 1516 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__); 1517 else { 1518 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 1519 error = mld_change_state(inm, 0); 1520 if (error) 1521 CTR1(KTR_MLD, "%s: failed mld downcall", __func__); 1522 } 1523 1524 IN6_MULTI_LIST_UNLOCK(); 1525 1526 out_im6f_rollback: 1527 if (error) 1528 im6f_rollback(imf); 1529 else 1530 im6f_commit(imf); 1531 1532 im6f_reap(imf); 1533 1534 out_in6p_locked: 1535 INP_WUNLOCK(inp); 1536 return (error); 1537 } 1538 1539 /* 1540 * Given an inpcb, return its multicast options structure pointer. Accepts 1541 * an unlocked inpcb pointer, but will return it locked. May sleep. 1542 * 1543 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1544 * SMPng: NOTE: Returns with the INP write lock held. 1545 */ 1546 static struct ip6_moptions * 1547 in6p_findmoptions(struct inpcb *inp) 1548 { 1549 struct ip6_moptions *imo; 1550 1551 INP_WLOCK(inp); 1552 if (inp->in6p_moptions != NULL) 1553 return (inp->in6p_moptions); 1554 1555 INP_WUNLOCK(inp); 1556 1557 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK); 1558 1559 imo->im6o_multicast_ifp = NULL; 1560 imo->im6o_multicast_hlim = V_ip6_defmcasthlim; 1561 imo->im6o_multicast_loop = in6_mcast_loop; 1562 STAILQ_INIT(&imo->im6o_head); 1563 1564 INP_WLOCK(inp); 1565 if (inp->in6p_moptions != NULL) { 1566 free(imo, M_IP6MOPTS); 1567 return (inp->in6p_moptions); 1568 } 1569 inp->in6p_moptions = imo; 1570 return (imo); 1571 } 1572 1573 /* 1574 * Discard the IPv6 multicast options (and source filters). 1575 * 1576 * SMPng: NOTE: assumes INP write lock is held. 1577 * 1578 * XXX can all be safely deferred to epoch_call 1579 * 1580 */ 1581 1582 static void 1583 inp_gcmoptions(struct ip6_moptions *imo) 1584 { 1585 struct in6_mfilter *imf; 1586 struct in6_multi *inm; 1587 struct ifnet *ifp; 1588 1589 while ((imf = ip6_mfilter_first(&imo->im6o_head)) != NULL) { 1590 ip6_mfilter_remove(&imo->im6o_head, imf); 1591 1592 im6f_leave(imf); 1593 if ((inm = imf->im6f_in6m) != NULL) { 1594 if ((ifp = inm->in6m_ifp) != NULL) { 1595 CURVNET_SET(ifp->if_vnet); 1596 (void)in6_leavegroup(inm, imf); 1597 CURVNET_RESTORE(); 1598 } else { 1599 (void)in6_leavegroup(inm, imf); 1600 } 1601 } 1602 ip6_mfilter_free(imf); 1603 } 1604 free(imo, M_IP6MOPTS); 1605 } 1606 1607 void 1608 ip6_freemoptions(struct ip6_moptions *imo) 1609 { 1610 if (imo == NULL) 1611 return; 1612 inp_gcmoptions(imo); 1613 } 1614 1615 /* 1616 * Atomically get source filters on a socket for an IPv6 multicast group. 1617 * Called with INP lock held; returns with lock released. 1618 */ 1619 static int 1620 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1621 { 1622 struct __msfilterreq msfr; 1623 sockunion_t *gsa; 1624 struct ifnet *ifp; 1625 struct ip6_moptions *imo; 1626 struct in6_mfilter *imf; 1627 struct ip6_msource *ims; 1628 struct in6_msource *lims; 1629 struct sockaddr_in6 *psin; 1630 struct sockaddr_storage *ptss; 1631 struct sockaddr_storage *tss; 1632 int error; 1633 size_t nsrcs, ncsrcs; 1634 1635 INP_WLOCK_ASSERT(inp); 1636 1637 imo = inp->in6p_moptions; 1638 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__)); 1639 1640 INP_WUNLOCK(inp); 1641 1642 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1643 sizeof(struct __msfilterreq)); 1644 if (error) 1645 return (error); 1646 1647 if (msfr.msfr_group.ss_family != AF_INET6 || 1648 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6)) 1649 return (EINVAL); 1650 1651 gsa = (sockunion_t *)&msfr.msfr_group; 1652 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 1653 return (EINVAL); 1654 1655 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 1656 return (EADDRNOTAVAIL); 1657 ifp = ifnet_byindex(msfr.msfr_ifindex); 1658 if (ifp == NULL) 1659 return (EADDRNOTAVAIL); 1660 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 1661 1662 INP_WLOCK(inp); 1663 1664 /* 1665 * Lookup group on the socket. 1666 */ 1667 imf = im6o_match_group(imo, ifp, &gsa->sa); 1668 if (imf == NULL) { 1669 INP_WUNLOCK(inp); 1670 return (EADDRNOTAVAIL); 1671 } 1672 1673 /* 1674 * Ignore memberships which are in limbo. 1675 */ 1676 if (imf->im6f_st[1] == MCAST_UNDEFINED) { 1677 INP_WUNLOCK(inp); 1678 return (EAGAIN); 1679 } 1680 msfr.msfr_fmode = imf->im6f_st[1]; 1681 1682 /* 1683 * If the user specified a buffer, copy out the source filter 1684 * entries to userland gracefully. 1685 * We only copy out the number of entries which userland 1686 * has asked for, but we always tell userland how big the 1687 * buffer really needs to be. 1688 */ 1689 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc) 1690 msfr.msfr_nsrcs = in6_mcast_maxsocksrc; 1691 tss = NULL; 1692 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1693 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1694 M_TEMP, M_NOWAIT | M_ZERO); 1695 if (tss == NULL) { 1696 INP_WUNLOCK(inp); 1697 return (ENOBUFS); 1698 } 1699 } 1700 1701 /* 1702 * Count number of sources in-mode at t0. 1703 * If buffer space exists and remains, copy out source entries. 1704 */ 1705 nsrcs = msfr.msfr_nsrcs; 1706 ncsrcs = 0; 1707 ptss = tss; 1708 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) { 1709 lims = (struct in6_msource *)ims; 1710 if (lims->im6sl_st[0] == MCAST_UNDEFINED || 1711 lims->im6sl_st[0] != imf->im6f_st[0]) 1712 continue; 1713 ++ncsrcs; 1714 if (tss != NULL && nsrcs > 0) { 1715 psin = (struct sockaddr_in6 *)ptss; 1716 psin->sin6_family = AF_INET6; 1717 psin->sin6_len = sizeof(struct sockaddr_in6); 1718 psin->sin6_addr = lims->im6s_addr; 1719 psin->sin6_port = 0; 1720 --nsrcs; 1721 ++ptss; 1722 } 1723 } 1724 1725 INP_WUNLOCK(inp); 1726 1727 if (tss != NULL) { 1728 error = copyout(tss, msfr.msfr_srcs, 1729 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1730 free(tss, M_TEMP); 1731 if (error) 1732 return (error); 1733 } 1734 1735 msfr.msfr_nsrcs = ncsrcs; 1736 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1737 1738 return (error); 1739 } 1740 1741 /* 1742 * Return the IP multicast options in response to user getsockopt(). 1743 */ 1744 int 1745 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1746 { 1747 struct ip6_moptions *im6o; 1748 int error; 1749 u_int optval; 1750 1751 INP_WLOCK(inp); 1752 im6o = inp->in6p_moptions; 1753 /* 1754 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 1755 * or is a divert socket, reject it. 1756 */ 1757 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 1758 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1759 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) { 1760 INP_WUNLOCK(inp); 1761 return (EOPNOTSUPP); 1762 } 1763 1764 error = 0; 1765 switch (sopt->sopt_name) { 1766 case IPV6_MULTICAST_IF: 1767 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) { 1768 optval = 0; 1769 } else { 1770 optval = im6o->im6o_multicast_ifp->if_index; 1771 } 1772 INP_WUNLOCK(inp); 1773 error = sooptcopyout(sopt, &optval, sizeof(u_int)); 1774 break; 1775 1776 case IPV6_MULTICAST_HOPS: 1777 if (im6o == NULL) 1778 optval = V_ip6_defmcasthlim; 1779 else 1780 optval = im6o->im6o_multicast_hlim; 1781 INP_WUNLOCK(inp); 1782 error = sooptcopyout(sopt, &optval, sizeof(u_int)); 1783 break; 1784 1785 case IPV6_MULTICAST_LOOP: 1786 if (im6o == NULL) 1787 optval = in6_mcast_loop; /* XXX VIMAGE */ 1788 else 1789 optval = im6o->im6o_multicast_loop; 1790 INP_WUNLOCK(inp); 1791 error = sooptcopyout(sopt, &optval, sizeof(u_int)); 1792 break; 1793 1794 case IPV6_MSFILTER: 1795 if (im6o == NULL) { 1796 error = EADDRNOTAVAIL; 1797 INP_WUNLOCK(inp); 1798 } else { 1799 error = in6p_get_source_filters(inp, sopt); 1800 } 1801 break; 1802 1803 default: 1804 INP_WUNLOCK(inp); 1805 error = ENOPROTOOPT; 1806 break; 1807 } 1808 1809 INP_UNLOCK_ASSERT(inp); 1810 1811 return (error); 1812 } 1813 1814 /* 1815 * Look up the ifnet to use for a multicast group membership, 1816 * given the address of an IPv6 group. 1817 * 1818 * This routine exists to support legacy IPv6 multicast applications. 1819 * 1820 * If inp is non-NULL, use this socket's current FIB number for any 1821 * required FIB lookup. Look up the group address in the unicast FIB, 1822 * and use its ifp; usually, this points to the default next-hop. 1823 * If the FIB lookup fails, return NULL. 1824 * 1825 * FUTURE: Support multiple forwarding tables for IPv6. 1826 * 1827 * Returns NULL if no ifp could be found. 1828 */ 1829 static struct ifnet * 1830 in6p_lookup_mcast_ifp(const struct inpcb *inp, 1831 const struct sockaddr_in6 *gsin6) 1832 { 1833 struct nhop_object *nh; 1834 struct in6_addr dst; 1835 uint32_t scopeid; 1836 uint32_t fibnum; 1837 1838 KASSERT(inp->inp_vflag & INP_IPV6, 1839 ("%s: not INP_IPV6 inpcb", __func__)); 1840 KASSERT(gsin6->sin6_family == AF_INET6, 1841 ("%s: not AF_INET6 group", __func__)); 1842 1843 in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid); 1844 fibnum = inp ? inp->inp_inc.inc_fibnum : RT_DEFAULT_FIB; 1845 nh = fib6_lookup(fibnum, &dst, scopeid, 0, 0); 1846 1847 return (nh ? nh->nh_ifp : NULL); 1848 } 1849 1850 /* 1851 * Join an IPv6 multicast group, possibly with a source. 1852 * 1853 * FIXME: The KAME use of the unspecified address (::) 1854 * to join *all* multicast groups is currently unsupported. 1855 */ 1856 static int 1857 in6p_join_group(struct inpcb *inp, struct sockopt *sopt) 1858 { 1859 struct in6_multi_head inmh; 1860 struct group_source_req gsr; 1861 sockunion_t *gsa, *ssa; 1862 struct ifnet *ifp; 1863 struct in6_mfilter *imf; 1864 struct ip6_moptions *imo; 1865 struct in6_multi *inm; 1866 struct in6_msource *lims; 1867 int error, is_new; 1868 1869 SLIST_INIT(&inmh); 1870 ifp = NULL; 1871 lims = NULL; 1872 error = 0; 1873 1874 memset(&gsr, 0, sizeof(struct group_source_req)); 1875 gsa = (sockunion_t *)&gsr.gsr_group; 1876 gsa->ss.ss_family = AF_UNSPEC; 1877 ssa = (sockunion_t *)&gsr.gsr_source; 1878 ssa->ss.ss_family = AF_UNSPEC; 1879 1880 /* 1881 * Chew everything into struct group_source_req. 1882 * Overwrite the port field if present, as the sockaddr 1883 * being copied in may be matched with a binary comparison. 1884 * Ignore passed-in scope ID. 1885 */ 1886 switch (sopt->sopt_name) { 1887 case IPV6_JOIN_GROUP: { 1888 struct ipv6_mreq mreq; 1889 1890 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq), 1891 sizeof(struct ipv6_mreq)); 1892 if (error) 1893 return (error); 1894 1895 gsa->sin6.sin6_family = AF_INET6; 1896 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6); 1897 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr; 1898 1899 if (mreq.ipv6mr_interface == 0) { 1900 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6); 1901 } else { 1902 if (V_if_index < mreq.ipv6mr_interface) 1903 return (EADDRNOTAVAIL); 1904 ifp = ifnet_byindex(mreq.ipv6mr_interface); 1905 } 1906 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p", 1907 __func__, mreq.ipv6mr_interface, ifp); 1908 } break; 1909 1910 case MCAST_JOIN_GROUP: 1911 case MCAST_JOIN_SOURCE_GROUP: 1912 if (sopt->sopt_name == MCAST_JOIN_GROUP) { 1913 error = sooptcopyin(sopt, &gsr, 1914 sizeof(struct group_req), 1915 sizeof(struct group_req)); 1916 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 1917 error = sooptcopyin(sopt, &gsr, 1918 sizeof(struct group_source_req), 1919 sizeof(struct group_source_req)); 1920 } 1921 if (error) 1922 return (error); 1923 1924 if (gsa->sin6.sin6_family != AF_INET6 || 1925 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 1926 return (EINVAL); 1927 1928 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 1929 if (ssa->sin6.sin6_family != AF_INET6 || 1930 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 1931 return (EINVAL); 1932 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr)) 1933 return (EINVAL); 1934 /* 1935 * TODO: Validate embedded scope ID in source 1936 * list entry against passed-in ifp, if and only 1937 * if source list filter entry is iface or node local. 1938 */ 1939 in6_clearscope(&ssa->sin6.sin6_addr); 1940 ssa->sin6.sin6_port = 0; 1941 ssa->sin6.sin6_scope_id = 0; 1942 } 1943 1944 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1945 return (EADDRNOTAVAIL); 1946 ifp = ifnet_byindex(gsr.gsr_interface); 1947 break; 1948 1949 default: 1950 CTR2(KTR_MLD, "%s: unknown sopt_name %d", 1951 __func__, sopt->sopt_name); 1952 return (EOPNOTSUPP); 1953 break; 1954 } 1955 1956 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 1957 return (EINVAL); 1958 1959 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) 1960 return (EADDRNOTAVAIL); 1961 1962 gsa->sin6.sin6_port = 0; 1963 gsa->sin6.sin6_scope_id = 0; 1964 1965 /* 1966 * Always set the scope zone ID on memberships created from userland. 1967 * Use the passed-in ifp to do this. 1968 * XXX The in6_setscope() return value is meaningless. 1969 * XXX SCOPE6_LOCK() is taken by in6_setscope(). 1970 */ 1971 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 1972 1973 IN6_MULTI_LOCK(); 1974 1975 /* 1976 * Find the membership in the membership list. 1977 */ 1978 imo = in6p_findmoptions(inp); 1979 imf = im6o_match_group(imo, ifp, &gsa->sa); 1980 if (imf == NULL) { 1981 is_new = 1; 1982 inm = NULL; 1983 1984 if (ip6_mfilter_count(&imo->im6o_head) >= IPV6_MAX_MEMBERSHIPS) { 1985 error = ENOMEM; 1986 goto out_in6p_locked; 1987 } 1988 } else { 1989 is_new = 0; 1990 inm = imf->im6f_in6m; 1991 1992 if (ssa->ss.ss_family != AF_UNSPEC) { 1993 /* 1994 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership 1995 * is an error. On an existing inclusive membership, 1996 * it just adds the source to the filter list. 1997 */ 1998 if (imf->im6f_st[1] != MCAST_INCLUDE) { 1999 error = EINVAL; 2000 goto out_in6p_locked; 2001 } 2002 /* 2003 * Throw out duplicates. 2004 * 2005 * XXX FIXME: This makes a naive assumption that 2006 * even if entries exist for *ssa in this imf, 2007 * they will be rejected as dupes, even if they 2008 * are not valid in the current mode (in-mode). 2009 * 2010 * in6_msource is transactioned just as for anything 2011 * else in SSM -- but note naive use of in6m_graft() 2012 * below for allocating new filter entries. 2013 * 2014 * This is only an issue if someone mixes the 2015 * full-state SSM API with the delta-based API, 2016 * which is discouraged in the relevant RFCs. 2017 */ 2018 lims = im6o_match_source(imf, &ssa->sa); 2019 if (lims != NULL /*&& 2020 lims->im6sl_st[1] == MCAST_INCLUDE*/) { 2021 error = EADDRNOTAVAIL; 2022 goto out_in6p_locked; 2023 } 2024 } else { 2025 /* 2026 * MCAST_JOIN_GROUP alone, on any existing membership, 2027 * is rejected, to stop the same inpcb tying up 2028 * multiple refs to the in_multi. 2029 * On an existing inclusive membership, this is also 2030 * an error; if you want to change filter mode, 2031 * you must use the userland API setsourcefilter(). 2032 * XXX We don't reject this for imf in UNDEFINED 2033 * state at t1, because allocation of a filter 2034 * is atomic with allocation of a membership. 2035 */ 2036 error = EINVAL; 2037 goto out_in6p_locked; 2038 } 2039 } 2040 2041 /* 2042 * Begin state merge transaction at socket layer. 2043 */ 2044 INP_WLOCK_ASSERT(inp); 2045 2046 /* 2047 * Graft new source into filter list for this inpcb's 2048 * membership of the group. The in6_multi may not have 2049 * been allocated yet if this is a new membership, however, 2050 * the in_mfilter slot will be allocated and must be initialized. 2051 * 2052 * Note: Grafting of exclusive mode filters doesn't happen 2053 * in this path. 2054 * XXX: Should check for non-NULL lims (node exists but may 2055 * not be in-mode) for interop with full-state API. 2056 */ 2057 if (ssa->ss.ss_family != AF_UNSPEC) { 2058 /* Membership starts in IN mode */ 2059 if (is_new) { 2060 CTR1(KTR_MLD, "%s: new join w/source", __func__); 2061 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE); 2062 if (imf == NULL) { 2063 error = ENOMEM; 2064 goto out_in6p_locked; 2065 } 2066 } else { 2067 CTR2(KTR_MLD, "%s: %s source", __func__, "allow"); 2068 } 2069 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6); 2070 if (lims == NULL) { 2071 CTR1(KTR_MLD, "%s: merge imf state failed", 2072 __func__); 2073 error = ENOMEM; 2074 goto out_in6p_locked; 2075 } 2076 } else { 2077 /* No address specified; Membership starts in EX mode */ 2078 if (is_new) { 2079 CTR1(KTR_MLD, "%s: new join w/o source", __func__); 2080 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE); 2081 if (imf == NULL) { 2082 error = ENOMEM; 2083 goto out_in6p_locked; 2084 } 2085 } 2086 } 2087 2088 /* 2089 * Begin state merge transaction at MLD layer. 2090 */ 2091 if (is_new) { 2092 in_pcbref(inp); 2093 INP_WUNLOCK(inp); 2094 2095 error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf, 2096 &imf->im6f_in6m, 0); 2097 2098 INP_WLOCK(inp); 2099 if (in_pcbrele_wlocked(inp)) { 2100 error = ENXIO; 2101 goto out_in6p_unlocked; 2102 } 2103 if (error) { 2104 goto out_in6p_locked; 2105 } 2106 /* 2107 * NOTE: Refcount from in6_joingroup_locked() 2108 * is protecting membership. 2109 */ 2110 ip6_mfilter_insert(&imo->im6o_head, imf); 2111 } else { 2112 CTR1(KTR_MLD, "%s: merge inm state", __func__); 2113 IN6_MULTI_LIST_LOCK(); 2114 error = in6m_merge(inm, imf); 2115 if (error) { 2116 CTR1(KTR_MLD, "%s: failed to merge inm state", 2117 __func__); 2118 IN6_MULTI_LIST_UNLOCK(); 2119 im6f_rollback(imf); 2120 im6f_reap(imf); 2121 goto out_in6p_locked; 2122 } 2123 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 2124 error = mld_change_state(inm, 0); 2125 IN6_MULTI_LIST_UNLOCK(); 2126 2127 if (error) { 2128 CTR1(KTR_MLD, "%s: failed mld downcall", 2129 __func__); 2130 im6f_rollback(imf); 2131 im6f_reap(imf); 2132 goto out_in6p_locked; 2133 } 2134 } 2135 2136 im6f_commit(imf); 2137 imf = NULL; 2138 2139 out_in6p_locked: 2140 INP_WUNLOCK(inp); 2141 out_in6p_unlocked: 2142 IN6_MULTI_UNLOCK(); 2143 2144 if (is_new && imf) { 2145 if (imf->im6f_in6m != NULL) { 2146 struct in6_multi_head inmh; 2147 2148 SLIST_INIT(&inmh); 2149 SLIST_INSERT_HEAD(&inmh, imf->im6f_in6m, in6m_defer); 2150 in6m_release_list_deferred(&inmh); 2151 } 2152 ip6_mfilter_free(imf); 2153 } 2154 return (error); 2155 } 2156 2157 /* 2158 * Leave an IPv6 multicast group on an inpcb, possibly with a source. 2159 */ 2160 static int 2161 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt) 2162 { 2163 struct ipv6_mreq mreq; 2164 struct group_source_req gsr; 2165 sockunion_t *gsa, *ssa; 2166 struct ifnet *ifp; 2167 struct in6_mfilter *imf; 2168 struct ip6_moptions *imo; 2169 struct in6_msource *ims; 2170 struct in6_multi *inm; 2171 uint32_t ifindex; 2172 int error; 2173 bool is_final; 2174 #ifdef KTR 2175 char ip6tbuf[INET6_ADDRSTRLEN]; 2176 #endif 2177 2178 ifp = NULL; 2179 ifindex = 0; 2180 error = 0; 2181 is_final = true; 2182 2183 memset(&gsr, 0, sizeof(struct group_source_req)); 2184 gsa = (sockunion_t *)&gsr.gsr_group; 2185 gsa->ss.ss_family = AF_UNSPEC; 2186 ssa = (sockunion_t *)&gsr.gsr_source; 2187 ssa->ss.ss_family = AF_UNSPEC; 2188 2189 /* 2190 * Chew everything passed in up into a struct group_source_req 2191 * as that is easier to process. 2192 * Note: Any embedded scope ID in the multicast group passed 2193 * in by userland is ignored, the interface index is the recommended 2194 * mechanism to specify an interface; see below. 2195 */ 2196 switch (sopt->sopt_name) { 2197 case IPV6_LEAVE_GROUP: 2198 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq), 2199 sizeof(struct ipv6_mreq)); 2200 if (error) 2201 return (error); 2202 gsa->sin6.sin6_family = AF_INET6; 2203 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6); 2204 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr; 2205 gsa->sin6.sin6_port = 0; 2206 gsa->sin6.sin6_scope_id = 0; 2207 ifindex = mreq.ipv6mr_interface; 2208 break; 2209 2210 case MCAST_LEAVE_GROUP: 2211 case MCAST_LEAVE_SOURCE_GROUP: 2212 if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2213 error = sooptcopyin(sopt, &gsr, 2214 sizeof(struct group_req), 2215 sizeof(struct group_req)); 2216 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2217 error = sooptcopyin(sopt, &gsr, 2218 sizeof(struct group_source_req), 2219 sizeof(struct group_source_req)); 2220 } 2221 if (error) 2222 return (error); 2223 2224 if (gsa->sin6.sin6_family != AF_INET6 || 2225 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 2226 return (EINVAL); 2227 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2228 if (ssa->sin6.sin6_family != AF_INET6 || 2229 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 2230 return (EINVAL); 2231 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr)) 2232 return (EINVAL); 2233 /* 2234 * TODO: Validate embedded scope ID in source 2235 * list entry against passed-in ifp, if and only 2236 * if source list filter entry is iface or node local. 2237 */ 2238 in6_clearscope(&ssa->sin6.sin6_addr); 2239 } 2240 gsa->sin6.sin6_port = 0; 2241 gsa->sin6.sin6_scope_id = 0; 2242 ifindex = gsr.gsr_interface; 2243 break; 2244 2245 default: 2246 CTR2(KTR_MLD, "%s: unknown sopt_name %d", 2247 __func__, sopt->sopt_name); 2248 return (EOPNOTSUPP); 2249 break; 2250 } 2251 2252 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 2253 return (EINVAL); 2254 2255 /* 2256 * Validate interface index if provided. If no interface index 2257 * was provided separately, attempt to look the membership up 2258 * from the default scope as a last resort to disambiguate 2259 * the membership we are being asked to leave. 2260 * XXX SCOPE6 lock potentially taken here. 2261 */ 2262 if (ifindex != 0) { 2263 if (V_if_index < ifindex) 2264 return (EADDRNOTAVAIL); 2265 ifp = ifnet_byindex(ifindex); 2266 if (ifp == NULL) 2267 return (EADDRNOTAVAIL); 2268 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 2269 } else { 2270 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone); 2271 if (error) 2272 return (EADDRNOTAVAIL); 2273 /* 2274 * Some badly behaved applications don't pass an ifindex 2275 * or a scope ID, which is an API violation. In this case, 2276 * perform a lookup as per a v6 join. 2277 * 2278 * XXX For now, stomp on zone ID for the corner case. 2279 * This is not the 'KAME way', but we need to see the ifp 2280 * directly until such time as this implementation is 2281 * refactored, assuming the scope IDs are the way to go. 2282 */ 2283 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]); 2284 if (ifindex == 0) { 2285 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up " 2286 "ifp for group %s.", __func__, 2287 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr)); 2288 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6); 2289 } else { 2290 ifp = ifnet_byindex(ifindex); 2291 } 2292 if (ifp == NULL) 2293 return (EADDRNOTAVAIL); 2294 } 2295 2296 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp); 2297 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__)); 2298 2299 IN6_MULTI_LOCK(); 2300 2301 /* 2302 * Find the membership in the membership list. 2303 */ 2304 imo = in6p_findmoptions(inp); 2305 imf = im6o_match_group(imo, ifp, &gsa->sa); 2306 if (imf == NULL) { 2307 error = EADDRNOTAVAIL; 2308 goto out_in6p_locked; 2309 } 2310 inm = imf->im6f_in6m; 2311 2312 if (ssa->ss.ss_family != AF_UNSPEC) 2313 is_final = false; 2314 2315 /* 2316 * Begin state merge transaction at socket layer. 2317 */ 2318 INP_WLOCK_ASSERT(inp); 2319 2320 /* 2321 * If we were instructed only to leave a given source, do so. 2322 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2323 */ 2324 if (is_final) { 2325 ip6_mfilter_remove(&imo->im6o_head, imf); 2326 im6f_leave(imf); 2327 2328 /* 2329 * Give up the multicast address record to which 2330 * the membership points. 2331 */ 2332 (void)in6_leavegroup_locked(inm, imf); 2333 } else { 2334 if (imf->im6f_st[0] == MCAST_EXCLUDE) { 2335 error = EADDRNOTAVAIL; 2336 goto out_in6p_locked; 2337 } 2338 ims = im6o_match_source(imf, &ssa->sa); 2339 if (ims == NULL) { 2340 CTR3(KTR_MLD, "%s: source %p %spresent", __func__, 2341 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr), 2342 "not "); 2343 error = EADDRNOTAVAIL; 2344 goto out_in6p_locked; 2345 } 2346 CTR2(KTR_MLD, "%s: %s source", __func__, "block"); 2347 error = im6f_prune(imf, &ssa->sin6); 2348 if (error) { 2349 CTR1(KTR_MLD, "%s: merge imf state failed", 2350 __func__); 2351 goto out_in6p_locked; 2352 } 2353 } 2354 2355 /* 2356 * Begin state merge transaction at MLD layer. 2357 */ 2358 if (!is_final) { 2359 CTR1(KTR_MLD, "%s: merge inm state", __func__); 2360 IN6_MULTI_LIST_LOCK(); 2361 error = in6m_merge(inm, imf); 2362 if (error) { 2363 CTR1(KTR_MLD, "%s: failed to merge inm state", 2364 __func__); 2365 IN6_MULTI_LIST_UNLOCK(); 2366 im6f_rollback(imf); 2367 im6f_reap(imf); 2368 goto out_in6p_locked; 2369 } 2370 2371 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 2372 error = mld_change_state(inm, 0); 2373 IN6_MULTI_LIST_UNLOCK(); 2374 if (error) { 2375 CTR1(KTR_MLD, "%s: failed mld downcall", 2376 __func__); 2377 im6f_rollback(imf); 2378 im6f_reap(imf); 2379 goto out_in6p_locked; 2380 } 2381 } 2382 2383 im6f_commit(imf); 2384 im6f_reap(imf); 2385 2386 out_in6p_locked: 2387 INP_WUNLOCK(inp); 2388 2389 if (is_final && imf) 2390 ip6_mfilter_free(imf); 2391 2392 IN6_MULTI_UNLOCK(); 2393 return (error); 2394 } 2395 2396 /* 2397 * Select the interface for transmitting IPv6 multicast datagrams. 2398 * 2399 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn 2400 * may be passed to this socket option. An address of in6addr_any or an 2401 * interface index of 0 is used to remove a previous selection. 2402 * When no interface is selected, one is chosen for every send. 2403 */ 2404 static int 2405 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2406 { 2407 struct ifnet *ifp; 2408 struct ip6_moptions *imo; 2409 u_int ifindex; 2410 int error; 2411 2412 if (sopt->sopt_valsize != sizeof(u_int)) 2413 return (EINVAL); 2414 2415 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int)); 2416 if (error) 2417 return (error); 2418 if (V_if_index < ifindex) 2419 return (EINVAL); 2420 if (ifindex == 0) 2421 ifp = NULL; 2422 else { 2423 ifp = ifnet_byindex(ifindex); 2424 if (ifp == NULL) 2425 return (EINVAL); 2426 if ((ifp->if_flags & IFF_MULTICAST) == 0) 2427 return (EADDRNOTAVAIL); 2428 } 2429 imo = in6p_findmoptions(inp); 2430 imo->im6o_multicast_ifp = ifp; 2431 INP_WUNLOCK(inp); 2432 2433 return (0); 2434 } 2435 2436 /* 2437 * Atomically set source filters on a socket for an IPv6 multicast group. 2438 * 2439 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2440 */ 2441 static int 2442 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2443 { 2444 struct __msfilterreq msfr; 2445 sockunion_t *gsa; 2446 struct ifnet *ifp; 2447 struct in6_mfilter *imf; 2448 struct ip6_moptions *imo; 2449 struct in6_multi *inm; 2450 int error; 2451 2452 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2453 sizeof(struct __msfilterreq)); 2454 if (error) 2455 return (error); 2456 2457 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc) 2458 return (ENOBUFS); 2459 2460 if (msfr.msfr_fmode != MCAST_EXCLUDE && 2461 msfr.msfr_fmode != MCAST_INCLUDE) 2462 return (EINVAL); 2463 2464 if (msfr.msfr_group.ss_family != AF_INET6 || 2465 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6)) 2466 return (EINVAL); 2467 2468 gsa = (sockunion_t *)&msfr.msfr_group; 2469 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 2470 return (EINVAL); 2471 2472 gsa->sin6.sin6_port = 0; /* ignore port */ 2473 2474 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 2475 return (EADDRNOTAVAIL); 2476 ifp = ifnet_byindex(msfr.msfr_ifindex); 2477 if (ifp == NULL) 2478 return (EADDRNOTAVAIL); 2479 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 2480 2481 /* 2482 * Take the INP write lock. 2483 * Check if this socket is a member of this group. 2484 */ 2485 imo = in6p_findmoptions(inp); 2486 imf = im6o_match_group(imo, ifp, &gsa->sa); 2487 if (imf == NULL) { 2488 error = EADDRNOTAVAIL; 2489 goto out_in6p_locked; 2490 } 2491 inm = imf->im6f_in6m; 2492 2493 /* 2494 * Begin state merge transaction at socket layer. 2495 */ 2496 INP_WLOCK_ASSERT(inp); 2497 2498 imf->im6f_st[1] = msfr.msfr_fmode; 2499 2500 /* 2501 * Apply any new source filters, if present. 2502 * Make a copy of the user-space source vector so 2503 * that we may copy them with a single copyin. This 2504 * allows us to deal with page faults up-front. 2505 */ 2506 if (msfr.msfr_nsrcs > 0) { 2507 struct in6_msource *lims; 2508 struct sockaddr_in6 *psin; 2509 struct sockaddr_storage *kss, *pkss; 2510 int i; 2511 2512 INP_WUNLOCK(inp); 2513 2514 CTR2(KTR_MLD, "%s: loading %lu source list entries", 2515 __func__, (unsigned long)msfr.msfr_nsrcs); 2516 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2517 M_TEMP, M_WAITOK); 2518 error = copyin(msfr.msfr_srcs, kss, 2519 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2520 if (error) { 2521 free(kss, M_TEMP); 2522 return (error); 2523 } 2524 2525 INP_WLOCK(inp); 2526 2527 /* 2528 * Mark all source filters as UNDEFINED at t1. 2529 * Restore new group filter mode, as im6f_leave() 2530 * will set it to INCLUDE. 2531 */ 2532 im6f_leave(imf); 2533 imf->im6f_st[1] = msfr.msfr_fmode; 2534 2535 /* 2536 * Update socket layer filters at t1, lazy-allocating 2537 * new entries. This saves a bunch of memory at the 2538 * cost of one RB_FIND() per source entry; duplicate 2539 * entries in the msfr_nsrcs vector are ignored. 2540 * If we encounter an error, rollback transaction. 2541 * 2542 * XXX This too could be replaced with a set-symmetric 2543 * difference like loop to avoid walking from root 2544 * every time, as the key space is common. 2545 */ 2546 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2547 psin = (struct sockaddr_in6 *)pkss; 2548 if (psin->sin6_family != AF_INET6) { 2549 error = EAFNOSUPPORT; 2550 break; 2551 } 2552 if (psin->sin6_len != sizeof(struct sockaddr_in6)) { 2553 error = EINVAL; 2554 break; 2555 } 2556 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) { 2557 error = EINVAL; 2558 break; 2559 } 2560 /* 2561 * TODO: Validate embedded scope ID in source 2562 * list entry against passed-in ifp, if and only 2563 * if source list filter entry is iface or node local. 2564 */ 2565 in6_clearscope(&psin->sin6_addr); 2566 error = im6f_get_source(imf, psin, &lims); 2567 if (error) 2568 break; 2569 lims->im6sl_st[1] = imf->im6f_st[1]; 2570 } 2571 free(kss, M_TEMP); 2572 } 2573 2574 if (error) 2575 goto out_im6f_rollback; 2576 2577 INP_WLOCK_ASSERT(inp); 2578 IN6_MULTI_LIST_LOCK(); 2579 2580 /* 2581 * Begin state merge transaction at MLD layer. 2582 */ 2583 CTR1(KTR_MLD, "%s: merge inm state", __func__); 2584 error = in6m_merge(inm, imf); 2585 if (error) 2586 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__); 2587 else { 2588 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 2589 error = mld_change_state(inm, 0); 2590 if (error) 2591 CTR1(KTR_MLD, "%s: failed mld downcall", __func__); 2592 } 2593 2594 IN6_MULTI_LIST_UNLOCK(); 2595 2596 out_im6f_rollback: 2597 if (error) 2598 im6f_rollback(imf); 2599 else 2600 im6f_commit(imf); 2601 2602 im6f_reap(imf); 2603 2604 out_in6p_locked: 2605 INP_WUNLOCK(inp); 2606 return (error); 2607 } 2608 2609 /* 2610 * Set the IP multicast options in response to user setsockopt(). 2611 * 2612 * Many of the socket options handled in this function duplicate the 2613 * functionality of socket options in the regular unicast API. However, 2614 * it is not possible to merge the duplicate code, because the idempotence 2615 * of the IPv6 multicast part of the BSD Sockets API must be preserved; 2616 * the effects of these options must be treated as separate and distinct. 2617 * 2618 * SMPng: XXX: Unlocked read of inp_socket believed OK. 2619 */ 2620 int 2621 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2622 { 2623 struct ip6_moptions *im6o; 2624 int error; 2625 2626 error = 0; 2627 2628 /* 2629 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 2630 * or is a divert socket, reject it. 2631 */ 2632 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 2633 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2634 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) 2635 return (EOPNOTSUPP); 2636 2637 switch (sopt->sopt_name) { 2638 case IPV6_MULTICAST_IF: 2639 error = in6p_set_multicast_if(inp, sopt); 2640 break; 2641 2642 case IPV6_MULTICAST_HOPS: { 2643 int hlim; 2644 2645 if (sopt->sopt_valsize != sizeof(int)) { 2646 error = EINVAL; 2647 break; 2648 } 2649 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int)); 2650 if (error) 2651 break; 2652 if (hlim < -1 || hlim > 255) { 2653 error = EINVAL; 2654 break; 2655 } else if (hlim == -1) { 2656 hlim = V_ip6_defmcasthlim; 2657 } 2658 im6o = in6p_findmoptions(inp); 2659 im6o->im6o_multicast_hlim = hlim; 2660 INP_WUNLOCK(inp); 2661 break; 2662 } 2663 2664 case IPV6_MULTICAST_LOOP: { 2665 u_int loop; 2666 2667 /* 2668 * Set the loopback flag for outgoing multicast packets. 2669 * Must be zero or one. 2670 */ 2671 if (sopt->sopt_valsize != sizeof(u_int)) { 2672 error = EINVAL; 2673 break; 2674 } 2675 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int)); 2676 if (error) 2677 break; 2678 if (loop > 1) { 2679 error = EINVAL; 2680 break; 2681 } 2682 im6o = in6p_findmoptions(inp); 2683 im6o->im6o_multicast_loop = loop; 2684 INP_WUNLOCK(inp); 2685 break; 2686 } 2687 2688 case IPV6_JOIN_GROUP: 2689 case MCAST_JOIN_GROUP: 2690 case MCAST_JOIN_SOURCE_GROUP: 2691 error = in6p_join_group(inp, sopt); 2692 break; 2693 2694 case IPV6_LEAVE_GROUP: 2695 case MCAST_LEAVE_GROUP: 2696 case MCAST_LEAVE_SOURCE_GROUP: 2697 error = in6p_leave_group(inp, sopt); 2698 break; 2699 2700 case MCAST_BLOCK_SOURCE: 2701 case MCAST_UNBLOCK_SOURCE: 2702 error = in6p_block_unblock_source(inp, sopt); 2703 break; 2704 2705 case IPV6_MSFILTER: 2706 error = in6p_set_source_filters(inp, sopt); 2707 break; 2708 2709 default: 2710 error = EOPNOTSUPP; 2711 break; 2712 } 2713 2714 INP_UNLOCK_ASSERT(inp); 2715 2716 return (error); 2717 } 2718 2719 /* 2720 * Expose MLD's multicast filter mode and source list(s) to userland, 2721 * keyed by (ifindex, group). 2722 * The filter mode is written out as a uint32_t, followed by 2723 * 0..n of struct in6_addr. 2724 * For use by ifmcstat(8). 2725 * SMPng: NOTE: unlocked read of ifindex space. 2726 */ 2727 static int 2728 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS) 2729 { 2730 struct in6_addr mcaddr; 2731 struct in6_addr src; 2732 struct epoch_tracker et; 2733 struct ifnet *ifp; 2734 struct ifmultiaddr *ifma; 2735 struct in6_multi *inm; 2736 struct ip6_msource *ims; 2737 int *name; 2738 int retval; 2739 u_int namelen; 2740 uint32_t fmode, ifindex; 2741 #ifdef KTR 2742 char ip6tbuf[INET6_ADDRSTRLEN]; 2743 #endif 2744 2745 name = (int *)arg1; 2746 namelen = arg2; 2747 2748 if (req->newptr != NULL) 2749 return (EPERM); 2750 2751 /* int: ifindex + 4 * 32 bits of IPv6 address */ 2752 if (namelen != 5) 2753 return (EINVAL); 2754 2755 ifindex = name[0]; 2756 if (ifindex <= 0 || ifindex > V_if_index) { 2757 CTR2(KTR_MLD, "%s: ifindex %u out of range", 2758 __func__, ifindex); 2759 return (ENOENT); 2760 } 2761 2762 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr)); 2763 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) { 2764 CTR2(KTR_MLD, "%s: group %s is not multicast", 2765 __func__, ip6_sprintf(ip6tbuf, &mcaddr)); 2766 return (EINVAL); 2767 } 2768 2769 NET_EPOCH_ENTER(et); 2770 ifp = ifnet_byindex(ifindex); 2771 if (ifp == NULL) { 2772 NET_EPOCH_EXIT(et); 2773 CTR2(KTR_MLD, "%s: no ifp for ifindex %u", 2774 __func__, ifindex); 2775 return (ENOENT); 2776 } 2777 /* 2778 * Internal MLD lookups require that scope/zone ID is set. 2779 */ 2780 (void)in6_setscope(&mcaddr, ifp, NULL); 2781 2782 retval = sysctl_wire_old_buffer(req, 2783 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr))); 2784 if (retval) { 2785 NET_EPOCH_EXIT(et); 2786 return (retval); 2787 } 2788 2789 IN6_MULTI_LOCK(); 2790 IN6_MULTI_LIST_LOCK(); 2791 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2792 inm = in6m_ifmultiaddr_get_inm(ifma); 2793 if (inm == NULL) 2794 continue; 2795 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr)) 2796 continue; 2797 fmode = inm->in6m_st[1].iss_fmode; 2798 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 2799 if (retval != 0) 2800 break; 2801 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) { 2802 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims); 2803 /* 2804 * Only copy-out sources which are in-mode. 2805 */ 2806 if (fmode != im6s_get_mode(inm, ims, 1)) { 2807 CTR1(KTR_MLD, "%s: skip non-in-mode", 2808 __func__); 2809 continue; 2810 } 2811 src = ims->im6s_addr; 2812 retval = SYSCTL_OUT(req, &src, 2813 sizeof(struct in6_addr)); 2814 if (retval != 0) 2815 break; 2816 } 2817 } 2818 IN6_MULTI_LIST_UNLOCK(); 2819 IN6_MULTI_UNLOCK(); 2820 NET_EPOCH_EXIT(et); 2821 2822 return (retval); 2823 } 2824 2825 #ifdef KTR 2826 2827 static const char *in6m_modestrs[] = { "un", "in", "ex" }; 2828 2829 static const char * 2830 in6m_mode_str(const int mode) 2831 { 2832 2833 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 2834 return (in6m_modestrs[mode]); 2835 return ("??"); 2836 } 2837 2838 static const char *in6m_statestrs[] = { 2839 "not-member", 2840 "silent", 2841 "idle", 2842 "lazy", 2843 "sleeping", 2844 "awakening", 2845 "query-pending", 2846 "sg-query-pending", 2847 "leaving" 2848 }; 2849 2850 static const char * 2851 in6m_state_str(const int state) 2852 { 2853 2854 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER) 2855 return (in6m_statestrs[state]); 2856 return ("??"); 2857 } 2858 2859 /* 2860 * Dump an in6_multi structure to the console. 2861 */ 2862 void 2863 in6m_print(const struct in6_multi *inm) 2864 { 2865 int t; 2866 char ip6tbuf[INET6_ADDRSTRLEN]; 2867 2868 if ((ktr_mask & KTR_MLD) == 0) 2869 return; 2870 2871 printf("%s: --- begin in6m %p ---\n", __func__, inm); 2872 printf("addr %s ifp %p(%s) ifma %p\n", 2873 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2874 inm->in6m_ifp, 2875 if_name(inm->in6m_ifp), 2876 inm->in6m_ifma); 2877 printf("timer %u state %s refcount %u scq.len %u\n", 2878 inm->in6m_timer, 2879 in6m_state_str(inm->in6m_state), 2880 inm->in6m_refcount, 2881 mbufq_len(&inm->in6m_scq)); 2882 printf("mli %p nsrc %lu sctimer %u scrv %u\n", 2883 inm->in6m_mli, 2884 inm->in6m_nsrc, 2885 inm->in6m_sctimer, 2886 inm->in6m_scrv); 2887 for (t = 0; t < 2; t++) { 2888 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 2889 in6m_mode_str(inm->in6m_st[t].iss_fmode), 2890 inm->in6m_st[t].iss_asm, 2891 inm->in6m_st[t].iss_ex, 2892 inm->in6m_st[t].iss_in, 2893 inm->in6m_st[t].iss_rec); 2894 } 2895 printf("%s: --- end in6m %p ---\n", __func__, inm); 2896 } 2897 2898 #else /* !KTR */ 2899 2900 void 2901 in6m_print(const struct in6_multi *inm) 2902 { 2903 2904 } 2905 2906 #endif /* KTR */ 2907