1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2007-2009 Bruce Simpson. 5 * Copyright (c) 2005 Robert N. M. Watson. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. The name of the author may not be used to endorse or promote 17 * products derived from this software without specific prior written 18 * permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 */ 32 33 /* 34 * IPv4 multicast socket, group, and socket option processing module. 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/kernel.h> 43 #include <sys/lock.h> 44 #include <sys/malloc.h> 45 #include <sys/mbuf.h> 46 #include <sys/protosw.h> 47 #include <sys/socket.h> 48 #include <sys/socketvar.h> 49 #include <sys/protosw.h> 50 #include <sys/sysctl.h> 51 #include <sys/ktr.h> 52 #include <sys/taskqueue.h> 53 #include <sys/tree.h> 54 55 #include <net/if.h> 56 #include <net/if_var.h> 57 #include <net/if_dl.h> 58 #include <net/route.h> 59 #include <net/route/nhop.h> 60 #include <net/vnet.h> 61 62 #include <net/ethernet.h> 63 64 #include <netinet/in.h> 65 #include <netinet/in_systm.h> 66 #include <netinet/in_fib.h> 67 #include <netinet/in_pcb.h> 68 #include <netinet/in_var.h> 69 #include <netinet/ip_var.h> 70 #include <netinet/igmp_var.h> 71 72 #ifndef KTR_IGMPV3 73 #define KTR_IGMPV3 KTR_INET 74 #endif 75 76 #ifndef __SOCKUNION_DECLARED 77 union sockunion { 78 struct sockaddr_storage ss; 79 struct sockaddr sa; 80 struct sockaddr_dl sdl; 81 struct sockaddr_in sin; 82 }; 83 typedef union sockunion sockunion_t; 84 #define __SOCKUNION_DECLARED 85 #endif /* __SOCKUNION_DECLARED */ 86 87 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter", 88 "IPv4 multicast PCB-layer source filter"); 89 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group"); 90 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options"); 91 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource", 92 "IPv4 multicast IGMP-layer source filter"); 93 94 /* 95 * Locking: 96 * 97 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK, 98 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 99 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however 100 * it can be taken by code in net/if.c also. 101 * - ip_moptions and in_mfilter are covered by the INP_WLOCK. 102 * 103 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly 104 * any need for in_multi itself to be virtualized -- it is bound to an ifp 105 * anyway no matter what happens. 106 */ 107 struct mtx in_multi_list_mtx; 108 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF); 109 110 struct mtx in_multi_free_mtx; 111 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF); 112 113 struct sx in_multi_sx; 114 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx"); 115 116 int ifma_restart; 117 118 /* 119 * Functions with non-static linkage defined in this file should be 120 * declared in in_var.h: 121 * imo_multi_filter() 122 * in_joingroup() 123 * in_joingroup_locked() 124 * in_leavegroup() 125 * in_leavegroup_locked() 126 * and ip_var.h: 127 * inp_freemoptions() 128 * inp_getmoptions() 129 * inp_setmoptions() 130 */ 131 static void imf_commit(struct in_mfilter *); 132 static int imf_get_source(struct in_mfilter *imf, 133 const struct sockaddr_in *psin, 134 struct in_msource **); 135 static struct in_msource * 136 imf_graft(struct in_mfilter *, const uint8_t, 137 const struct sockaddr_in *); 138 static void imf_leave(struct in_mfilter *); 139 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *); 140 static void imf_purge(struct in_mfilter *); 141 static void imf_rollback(struct in_mfilter *); 142 static void imf_reap(struct in_mfilter *); 143 static struct in_mfilter * 144 imo_match_group(const struct ip_moptions *, 145 const struct ifnet *, const struct sockaddr *); 146 static struct in_msource * 147 imo_match_source(struct in_mfilter *, const struct sockaddr *); 148 static void ims_merge(struct ip_msource *ims, 149 const struct in_msource *lims, const int rollback); 150 static int in_getmulti(struct ifnet *, const struct in_addr *, 151 struct in_multi **); 152 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr, 153 const int noalloc, struct ip_msource **pims); 154 #ifdef KTR 155 static int inm_is_ifp_detached(const struct in_multi *); 156 #endif 157 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *); 158 static void inm_purge(struct in_multi *); 159 static void inm_reap(struct in_multi *); 160 static void inm_release(struct in_multi *); 161 static struct ip_moptions * 162 inp_findmoptions(struct inpcb *); 163 static int inp_get_source_filters(struct inpcb *, struct sockopt *); 164 static int inp_join_group(struct inpcb *, struct sockopt *); 165 static int inp_leave_group(struct inpcb *, struct sockopt *); 166 static struct ifnet * 167 inp_lookup_mcast_ifp(const struct inpcb *, 168 const struct sockaddr_in *, const struct in_addr); 169 static int inp_block_unblock_source(struct inpcb *, struct sockopt *); 170 static int inp_set_multicast_if(struct inpcb *, struct sockopt *); 171 static int inp_set_source_filters(struct inpcb *, struct sockopt *); 172 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS); 173 174 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, 175 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 176 "IPv4 multicast"); 177 178 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER; 179 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc, 180 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0, 181 "Max source filters per group"); 182 183 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER; 184 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc, 185 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0, 186 "Max source filters per socket"); 187 188 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP; 189 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN, 190 &in_mcast_loop, 0, "Loopback multicast datagrams by default"); 191 192 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters, 193 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters, 194 "Per-interface stack-wide source filters"); 195 196 #ifdef KTR 197 /* 198 * Inline function which wraps assertions for a valid ifp. 199 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 200 * is detached. 201 */ 202 static int __inline 203 inm_is_ifp_detached(const struct in_multi *inm) 204 { 205 struct ifnet *ifp; 206 207 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 208 ifp = inm->inm_ifma->ifma_ifp; 209 if (ifp != NULL) { 210 /* 211 * Sanity check that netinet's notion of ifp is the 212 * same as net's. 213 */ 214 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 215 } 216 217 return (ifp == NULL); 218 } 219 #endif 220 221 /* 222 * Interface detach can happen in a taskqueue thread context, so we must use a 223 * dedicated thread to avoid deadlocks when draining inm_release tasks. 224 */ 225 TASKQUEUE_DEFINE_THREAD(inm_free); 226 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER(); 227 static void inm_release_task(void *arg __unused, int pending __unused); 228 static struct task inm_free_task = TASK_INITIALIZER(0, inm_release_task, NULL); 229 230 void 231 inm_release_wait(void *arg __unused) 232 { 233 234 /* 235 * Make sure all pending multicast addresses are freed before 236 * the VNET or network device is destroyed: 237 */ 238 taskqueue_drain(taskqueue_inm_free, &inm_free_task); 239 } 240 #ifdef VIMAGE 241 /* XXX-BZ FIXME, see D24914. */ 242 VNET_SYSUNINIT(inm_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, inm_release_wait, NULL); 243 #endif 244 245 void 246 inm_release_list_deferred(struct in_multi_head *inmh) 247 { 248 249 if (SLIST_EMPTY(inmh)) 250 return; 251 mtx_lock(&in_multi_free_mtx); 252 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele); 253 mtx_unlock(&in_multi_free_mtx); 254 taskqueue_enqueue(taskqueue_inm_free, &inm_free_task); 255 } 256 257 void 258 inm_disconnect(struct in_multi *inm) 259 { 260 struct ifnet *ifp; 261 struct ifmultiaddr *ifma, *ll_ifma; 262 263 ifp = inm->inm_ifp; 264 IF_ADDR_WLOCK_ASSERT(ifp); 265 ifma = inm->inm_ifma; 266 267 if_ref(ifp); 268 if (ifma->ifma_flags & IFMA_F_ENQUEUED) { 269 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 270 ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 271 } 272 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname); 273 if ((ll_ifma = ifma->ifma_llifma) != NULL) { 274 MPASS(ifma != ll_ifma); 275 ifma->ifma_llifma = NULL; 276 MPASS(ll_ifma->ifma_llifma == NULL); 277 MPASS(ll_ifma->ifma_ifp == ifp); 278 if (--ll_ifma->ifma_refcount == 0) { 279 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) { 280 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link); 281 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 282 } 283 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname); 284 if_freemulti(ll_ifma); 285 ifma_restart = true; 286 } 287 } 288 } 289 290 void 291 inm_release_deferred(struct in_multi *inm) 292 { 293 struct in_multi_head tmp; 294 295 IN_MULTI_LIST_LOCK_ASSERT(); 296 MPASS(inm->inm_refcount > 0); 297 if (--inm->inm_refcount == 0) { 298 SLIST_INIT(&tmp); 299 inm_disconnect(inm); 300 inm->inm_ifma->ifma_protospec = NULL; 301 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele); 302 inm_release_list_deferred(&tmp); 303 } 304 } 305 306 static void 307 inm_release_task(void *arg __unused, int pending __unused) 308 { 309 struct in_multi_head inm_free_tmp; 310 struct in_multi *inm, *tinm; 311 312 SLIST_INIT(&inm_free_tmp); 313 mtx_lock(&in_multi_free_mtx); 314 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele); 315 mtx_unlock(&in_multi_free_mtx); 316 IN_MULTI_LOCK(); 317 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) { 318 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele); 319 MPASS(inm); 320 inm_release(inm); 321 } 322 IN_MULTI_UNLOCK(); 323 } 324 325 /* 326 * Initialize an in_mfilter structure to a known state at t0, t1 327 * with an empty source filter list. 328 */ 329 static __inline void 330 imf_init(struct in_mfilter *imf, const int st0, const int st1) 331 { 332 memset(imf, 0, sizeof(struct in_mfilter)); 333 RB_INIT(&imf->imf_sources); 334 imf->imf_st[0] = st0; 335 imf->imf_st[1] = st1; 336 } 337 338 struct in_mfilter * 339 ip_mfilter_alloc(const int mflags, const int st0, const int st1) 340 { 341 struct in_mfilter *imf; 342 343 imf = malloc(sizeof(*imf), M_INMFILTER, mflags); 344 if (imf != NULL) 345 imf_init(imf, st0, st1); 346 347 return (imf); 348 } 349 350 void 351 ip_mfilter_free(struct in_mfilter *imf) 352 { 353 354 imf_purge(imf); 355 free(imf, M_INMFILTER); 356 } 357 358 /* 359 * Function for looking up an in_multi record for an IPv4 multicast address 360 * on a given interface. ifp must be valid. If no record found, return NULL. 361 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held. 362 */ 363 struct in_multi * 364 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina) 365 { 366 struct ifmultiaddr *ifma; 367 struct in_multi *inm; 368 369 IN_MULTI_LIST_LOCK_ASSERT(); 370 IF_ADDR_LOCK_ASSERT(ifp); 371 372 inm = NULL; 373 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) { 374 if (ifma->ifma_addr->sa_family != AF_INET || 375 ifma->ifma_protospec == NULL) 376 continue; 377 inm = (struct in_multi *)ifma->ifma_protospec; 378 if (inm->inm_addr.s_addr == ina.s_addr) 379 break; 380 inm = NULL; 381 } 382 return (inm); 383 } 384 385 /* 386 * Wrapper for inm_lookup_locked(). 387 * The IF_ADDR_LOCK will be taken on ifp and released on return. 388 */ 389 struct in_multi * 390 inm_lookup(struct ifnet *ifp, const struct in_addr ina) 391 { 392 struct epoch_tracker et; 393 struct in_multi *inm; 394 395 IN_MULTI_LIST_LOCK_ASSERT(); 396 NET_EPOCH_ENTER(et); 397 398 inm = inm_lookup_locked(ifp, ina); 399 NET_EPOCH_EXIT(et); 400 401 return (inm); 402 } 403 404 /* 405 * Find an IPv4 multicast group entry for this ip_moptions instance 406 * which matches the specified group, and optionally an interface. 407 * Return its index into the array, or -1 if not found. 408 */ 409 static struct in_mfilter * 410 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp, 411 const struct sockaddr *group) 412 { 413 const struct sockaddr_in *gsin; 414 struct in_mfilter *imf; 415 struct in_multi *inm; 416 417 gsin = (const struct sockaddr_in *)group; 418 419 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 420 inm = imf->imf_inm; 421 if (inm == NULL) 422 continue; 423 if ((ifp == NULL || (inm->inm_ifp == ifp)) && 424 in_hosteq(inm->inm_addr, gsin->sin_addr)) { 425 break; 426 } 427 } 428 return (imf); 429 } 430 431 /* 432 * Find an IPv4 multicast source entry for this imo which matches 433 * the given group index for this socket, and source address. 434 * 435 * NOTE: This does not check if the entry is in-mode, merely if 436 * it exists, which may not be the desired behaviour. 437 */ 438 static struct in_msource * 439 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src) 440 { 441 struct ip_msource find; 442 struct ip_msource *ims; 443 const sockunion_t *psa; 444 445 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__)); 446 447 /* Source trees are keyed in host byte order. */ 448 psa = (const sockunion_t *)src; 449 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr); 450 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 451 452 return ((struct in_msource *)ims); 453 } 454 455 /* 456 * Perform filtering for multicast datagrams on a socket by group and source. 457 * 458 * Returns 0 if a datagram should be allowed through, or various error codes 459 * if the socket was not a member of the group, or the source was muted, etc. 460 */ 461 int 462 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp, 463 const struct sockaddr *group, const struct sockaddr *src) 464 { 465 struct in_mfilter *imf; 466 struct in_msource *ims; 467 int mode; 468 469 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 470 471 imf = imo_match_group(imo, ifp, group); 472 if (imf == NULL) 473 return (MCAST_NOTGMEMBER); 474 475 /* 476 * Check if the source was included in an (S,G) join. 477 * Allow reception on exclusive memberships by default, 478 * reject reception on inclusive memberships by default. 479 * Exclude source only if an in-mode exclude filter exists. 480 * Include source only if an in-mode include filter exists. 481 * NOTE: We are comparing group state here at IGMP t1 (now) 482 * with socket-layer t0 (since last downcall). 483 */ 484 mode = imf->imf_st[1]; 485 ims = imo_match_source(imf, src); 486 487 if ((ims == NULL && mode == MCAST_INCLUDE) || 488 (ims != NULL && ims->imsl_st[0] != mode)) 489 return (MCAST_NOTSMEMBER); 490 491 return (MCAST_PASS); 492 } 493 494 /* 495 * Find and return a reference to an in_multi record for (ifp, group), 496 * and bump its reference count. 497 * If one does not exist, try to allocate it, and update link-layer multicast 498 * filters on ifp to listen for group. 499 * Assumes the IN_MULTI lock is held across the call. 500 * Return 0 if successful, otherwise return an appropriate error code. 501 */ 502 static int 503 in_getmulti(struct ifnet *ifp, const struct in_addr *group, 504 struct in_multi **pinm) 505 { 506 struct sockaddr_in gsin; 507 struct ifmultiaddr *ifma; 508 struct in_ifinfo *ii; 509 struct in_multi *inm; 510 int error; 511 512 IN_MULTI_LOCK_ASSERT(); 513 514 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET]; 515 IN_MULTI_LIST_LOCK(); 516 inm = inm_lookup(ifp, *group); 517 if (inm != NULL) { 518 /* 519 * If we already joined this group, just bump the 520 * refcount and return it. 521 */ 522 KASSERT(inm->inm_refcount >= 1, 523 ("%s: bad refcount %d", __func__, inm->inm_refcount)); 524 inm_acquire_locked(inm); 525 *pinm = inm; 526 } 527 IN_MULTI_LIST_UNLOCK(); 528 if (inm != NULL) 529 return (0); 530 531 memset(&gsin, 0, sizeof(gsin)); 532 gsin.sin_family = AF_INET; 533 gsin.sin_len = sizeof(struct sockaddr_in); 534 gsin.sin_addr = *group; 535 536 /* 537 * Check if a link-layer group is already associated 538 * with this network-layer group on the given ifnet. 539 */ 540 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma); 541 if (error != 0) 542 return (error); 543 544 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */ 545 IN_MULTI_LIST_LOCK(); 546 IF_ADDR_WLOCK(ifp); 547 548 /* 549 * If something other than netinet is occupying the link-layer 550 * group, print a meaningful error message and back out of 551 * the allocation. 552 * Otherwise, bump the refcount on the existing network-layer 553 * group association and return it. 554 */ 555 if (ifma->ifma_protospec != NULL) { 556 inm = (struct in_multi *)ifma->ifma_protospec; 557 #ifdef INVARIANTS 558 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 559 __func__)); 560 KASSERT(ifma->ifma_addr->sa_family == AF_INET, 561 ("%s: ifma not AF_INET", __func__)); 562 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 563 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp || 564 !in_hosteq(inm->inm_addr, *group)) { 565 char addrbuf[INET_ADDRSTRLEN]; 566 567 panic("%s: ifma %p is inconsistent with %p (%s)", 568 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf)); 569 } 570 #endif 571 inm_acquire_locked(inm); 572 *pinm = inm; 573 goto out_locked; 574 } 575 576 IF_ADDR_WLOCK_ASSERT(ifp); 577 578 /* 579 * A new in_multi record is needed; allocate and initialize it. 580 * We DO NOT perform an IGMP join as the in_ layer may need to 581 * push an initial source list down to IGMP to support SSM. 582 * 583 * The initial source filter state is INCLUDE, {} as per the RFC. 584 */ 585 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO); 586 if (inm == NULL) { 587 IF_ADDR_WUNLOCK(ifp); 588 IN_MULTI_LIST_UNLOCK(); 589 if_delmulti_ifma(ifma); 590 return (ENOMEM); 591 } 592 inm->inm_addr = *group; 593 inm->inm_ifp = ifp; 594 inm->inm_igi = ii->ii_igmp; 595 inm->inm_ifma = ifma; 596 inm->inm_refcount = 1; 597 inm->inm_state = IGMP_NOT_MEMBER; 598 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES); 599 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED; 600 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 601 RB_INIT(&inm->inm_srcs); 602 603 ifma->ifma_protospec = inm; 604 605 *pinm = inm; 606 out_locked: 607 IF_ADDR_WUNLOCK(ifp); 608 IN_MULTI_LIST_UNLOCK(); 609 return (0); 610 } 611 612 /* 613 * Drop a reference to an in_multi record. 614 * 615 * If the refcount drops to 0, free the in_multi record and 616 * delete the underlying link-layer membership. 617 */ 618 static void 619 inm_release(struct in_multi *inm) 620 { 621 struct ifmultiaddr *ifma; 622 struct ifnet *ifp; 623 624 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount); 625 MPASS(inm->inm_refcount == 0); 626 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm); 627 628 ifma = inm->inm_ifma; 629 ifp = inm->inm_ifp; 630 631 /* XXX this access is not covered by IF_ADDR_LOCK */ 632 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma); 633 if (ifp != NULL) { 634 CURVNET_SET(ifp->if_vnet); 635 inm_purge(inm); 636 free(inm, M_IPMADDR); 637 if_delmulti_ifma_flags(ifma, 1); 638 CURVNET_RESTORE(); 639 if_rele(ifp); 640 } else { 641 inm_purge(inm); 642 free(inm, M_IPMADDR); 643 if_delmulti_ifma_flags(ifma, 1); 644 } 645 } 646 647 /* 648 * Clear recorded source entries for a group. 649 * Used by the IGMP code. Caller must hold the IN_MULTI lock. 650 * FIXME: Should reap. 651 */ 652 void 653 inm_clear_recorded(struct in_multi *inm) 654 { 655 struct ip_msource *ims; 656 657 IN_MULTI_LIST_LOCK_ASSERT(); 658 659 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 660 if (ims->ims_stp) { 661 ims->ims_stp = 0; 662 --inm->inm_st[1].iss_rec; 663 } 664 } 665 KASSERT(inm->inm_st[1].iss_rec == 0, 666 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec)); 667 } 668 669 /* 670 * Record a source as pending for a Source-Group IGMPv3 query. 671 * This lives here as it modifies the shared tree. 672 * 673 * inm is the group descriptor. 674 * naddr is the address of the source to record in network-byte order. 675 * 676 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will 677 * lazy-allocate a source node in response to an SG query. 678 * Otherwise, no allocation is performed. This saves some memory 679 * with the trade-off that the source will not be reported to the 680 * router if joined in the window between the query response and 681 * the group actually being joined on the local host. 682 * 683 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed. 684 * This turns off the allocation of a recorded source entry if 685 * the group has not been joined. 686 * 687 * Return 0 if the source didn't exist or was already marked as recorded. 688 * Return 1 if the source was marked as recorded by this function. 689 * Return <0 if any error occurred (negated errno code). 690 */ 691 int 692 inm_record_source(struct in_multi *inm, const in_addr_t naddr) 693 { 694 struct ip_msource find; 695 struct ip_msource *ims, *nims; 696 697 IN_MULTI_LIST_LOCK_ASSERT(); 698 699 find.ims_haddr = ntohl(naddr); 700 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 701 if (ims && ims->ims_stp) 702 return (0); 703 if (ims == NULL) { 704 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 705 return (-ENOSPC); 706 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 707 M_NOWAIT | M_ZERO); 708 if (nims == NULL) 709 return (-ENOMEM); 710 nims->ims_haddr = find.ims_haddr; 711 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 712 ++inm->inm_nsrc; 713 ims = nims; 714 } 715 716 /* 717 * Mark the source as recorded and update the recorded 718 * source count. 719 */ 720 ++ims->ims_stp; 721 ++inm->inm_st[1].iss_rec; 722 723 return (1); 724 } 725 726 /* 727 * Return a pointer to an in_msource owned by an in_mfilter, 728 * given its source address. 729 * Lazy-allocate if needed. If this is a new entry its filter state is 730 * undefined at t0. 731 * 732 * imf is the filter set being modified. 733 * haddr is the source address in *host* byte-order. 734 * 735 * SMPng: May be called with locks held; malloc must not block. 736 */ 737 static int 738 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin, 739 struct in_msource **plims) 740 { 741 struct ip_msource find; 742 struct ip_msource *ims, *nims; 743 struct in_msource *lims; 744 int error; 745 746 error = 0; 747 ims = NULL; 748 lims = NULL; 749 750 /* key is host byte order */ 751 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 752 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 753 lims = (struct in_msource *)ims; 754 if (lims == NULL) { 755 if (imf->imf_nsrc == in_mcast_maxsocksrc) 756 return (ENOSPC); 757 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 758 M_NOWAIT | M_ZERO); 759 if (nims == NULL) 760 return (ENOMEM); 761 lims = (struct in_msource *)nims; 762 lims->ims_haddr = find.ims_haddr; 763 lims->imsl_st[0] = MCAST_UNDEFINED; 764 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 765 ++imf->imf_nsrc; 766 } 767 768 *plims = lims; 769 770 return (error); 771 } 772 773 /* 774 * Graft a source entry into an existing socket-layer filter set, 775 * maintaining any required invariants and checking allocations. 776 * 777 * The source is marked as being in the new filter mode at t1. 778 * 779 * Return the pointer to the new node, otherwise return NULL. 780 */ 781 static struct in_msource * 782 imf_graft(struct in_mfilter *imf, const uint8_t st1, 783 const struct sockaddr_in *psin) 784 { 785 struct ip_msource *nims; 786 struct in_msource *lims; 787 788 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 789 M_NOWAIT | M_ZERO); 790 if (nims == NULL) 791 return (NULL); 792 lims = (struct in_msource *)nims; 793 lims->ims_haddr = ntohl(psin->sin_addr.s_addr); 794 lims->imsl_st[0] = MCAST_UNDEFINED; 795 lims->imsl_st[1] = st1; 796 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 797 ++imf->imf_nsrc; 798 799 return (lims); 800 } 801 802 /* 803 * Prune a source entry from an existing socket-layer filter set, 804 * maintaining any required invariants and checking allocations. 805 * 806 * The source is marked as being left at t1, it is not freed. 807 * 808 * Return 0 if no error occurred, otherwise return an errno value. 809 */ 810 static int 811 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin) 812 { 813 struct ip_msource find; 814 struct ip_msource *ims; 815 struct in_msource *lims; 816 817 /* key is host byte order */ 818 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 819 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 820 if (ims == NULL) 821 return (ENOENT); 822 lims = (struct in_msource *)ims; 823 lims->imsl_st[1] = MCAST_UNDEFINED; 824 return (0); 825 } 826 827 /* 828 * Revert socket-layer filter set deltas at t1 to t0 state. 829 */ 830 static void 831 imf_rollback(struct in_mfilter *imf) 832 { 833 struct ip_msource *ims, *tims; 834 struct in_msource *lims; 835 836 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 837 lims = (struct in_msource *)ims; 838 if (lims->imsl_st[0] == lims->imsl_st[1]) { 839 /* no change at t1 */ 840 continue; 841 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) { 842 /* revert change to existing source at t1 */ 843 lims->imsl_st[1] = lims->imsl_st[0]; 844 } else { 845 /* revert source added t1 */ 846 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 847 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 848 free(ims, M_INMFILTER); 849 imf->imf_nsrc--; 850 } 851 } 852 imf->imf_st[1] = imf->imf_st[0]; 853 } 854 855 /* 856 * Mark socket-layer filter set as INCLUDE {} at t1. 857 */ 858 static void 859 imf_leave(struct in_mfilter *imf) 860 { 861 struct ip_msource *ims; 862 struct in_msource *lims; 863 864 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 865 lims = (struct in_msource *)ims; 866 lims->imsl_st[1] = MCAST_UNDEFINED; 867 } 868 imf->imf_st[1] = MCAST_INCLUDE; 869 } 870 871 /* 872 * Mark socket-layer filter set deltas as committed. 873 */ 874 static void 875 imf_commit(struct in_mfilter *imf) 876 { 877 struct ip_msource *ims; 878 struct in_msource *lims; 879 880 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 881 lims = (struct in_msource *)ims; 882 lims->imsl_st[0] = lims->imsl_st[1]; 883 } 884 imf->imf_st[0] = imf->imf_st[1]; 885 } 886 887 /* 888 * Reap unreferenced sources from socket-layer filter set. 889 */ 890 static void 891 imf_reap(struct in_mfilter *imf) 892 { 893 struct ip_msource *ims, *tims; 894 struct in_msource *lims; 895 896 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 897 lims = (struct in_msource *)ims; 898 if ((lims->imsl_st[0] == MCAST_UNDEFINED) && 899 (lims->imsl_st[1] == MCAST_UNDEFINED)) { 900 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims); 901 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 902 free(ims, M_INMFILTER); 903 imf->imf_nsrc--; 904 } 905 } 906 } 907 908 /* 909 * Purge socket-layer filter set. 910 */ 911 static void 912 imf_purge(struct in_mfilter *imf) 913 { 914 struct ip_msource *ims, *tims; 915 916 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 917 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 918 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 919 free(ims, M_INMFILTER); 920 imf->imf_nsrc--; 921 } 922 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED; 923 KASSERT(RB_EMPTY(&imf->imf_sources), 924 ("%s: imf_sources not empty", __func__)); 925 } 926 927 /* 928 * Look up a source filter entry for a multicast group. 929 * 930 * inm is the group descriptor to work with. 931 * haddr is the host-byte-order IPv4 address to look up. 932 * noalloc may be non-zero to suppress allocation of sources. 933 * *pims will be set to the address of the retrieved or allocated source. 934 * 935 * SMPng: NOTE: may be called with locks held. 936 * Return 0 if successful, otherwise return a non-zero error code. 937 */ 938 static int 939 inm_get_source(struct in_multi *inm, const in_addr_t haddr, 940 const int noalloc, struct ip_msource **pims) 941 { 942 struct ip_msource find; 943 struct ip_msource *ims, *nims; 944 945 find.ims_haddr = haddr; 946 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 947 if (ims == NULL && !noalloc) { 948 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 949 return (ENOSPC); 950 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 951 M_NOWAIT | M_ZERO); 952 if (nims == NULL) 953 return (ENOMEM); 954 nims->ims_haddr = haddr; 955 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 956 ++inm->inm_nsrc; 957 ims = nims; 958 #ifdef KTR 959 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__, 960 haddr, ims); 961 #endif 962 } 963 964 *pims = ims; 965 return (0); 966 } 967 968 /* 969 * Merge socket-layer source into IGMP-layer source. 970 * If rollback is non-zero, perform the inverse of the merge. 971 */ 972 static void 973 ims_merge(struct ip_msource *ims, const struct in_msource *lims, 974 const int rollback) 975 { 976 int n = rollback ? -1 : 1; 977 978 if (lims->imsl_st[0] == MCAST_EXCLUDE) { 979 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x", 980 __func__, n, ims->ims_haddr); 981 ims->ims_st[1].ex -= n; 982 } else if (lims->imsl_st[0] == MCAST_INCLUDE) { 983 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x", 984 __func__, n, ims->ims_haddr); 985 ims->ims_st[1].in -= n; 986 } 987 988 if (lims->imsl_st[1] == MCAST_EXCLUDE) { 989 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x", 990 __func__, n, ims->ims_haddr); 991 ims->ims_st[1].ex += n; 992 } else if (lims->imsl_st[1] == MCAST_INCLUDE) { 993 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x", 994 __func__, n, ims->ims_haddr); 995 ims->ims_st[1].in += n; 996 } 997 } 998 999 /* 1000 * Atomically update the global in_multi state, when a membership's 1001 * filter list is being updated in any way. 1002 * 1003 * imf is the per-inpcb-membership group filter pointer. 1004 * A fake imf may be passed for in-kernel consumers. 1005 * 1006 * XXX This is a candidate for a set-symmetric-difference style loop 1007 * which would eliminate the repeated lookup from root of ims nodes, 1008 * as they share the same key space. 1009 * 1010 * If any error occurred this function will back out of refcounts 1011 * and return a non-zero value. 1012 */ 1013 static int 1014 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1015 { 1016 struct ip_msource *ims, *nims; 1017 struct in_msource *lims; 1018 int schanged, error; 1019 int nsrc0, nsrc1; 1020 1021 schanged = 0; 1022 error = 0; 1023 nsrc1 = nsrc0 = 0; 1024 IN_MULTI_LIST_LOCK_ASSERT(); 1025 1026 /* 1027 * Update the source filters first, as this may fail. 1028 * Maintain count of in-mode filters at t0, t1. These are 1029 * used to work out if we transition into ASM mode or not. 1030 * Maintain a count of source filters whose state was 1031 * actually modified by this operation. 1032 */ 1033 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1034 lims = (struct in_msource *)ims; 1035 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++; 1036 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++; 1037 if (lims->imsl_st[0] == lims->imsl_st[1]) continue; 1038 error = inm_get_source(inm, lims->ims_haddr, 0, &nims); 1039 ++schanged; 1040 if (error) 1041 break; 1042 ims_merge(nims, lims, 0); 1043 } 1044 if (error) { 1045 struct ip_msource *bims; 1046 1047 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) { 1048 lims = (struct in_msource *)ims; 1049 if (lims->imsl_st[0] == lims->imsl_st[1]) 1050 continue; 1051 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims); 1052 if (bims == NULL) 1053 continue; 1054 ims_merge(bims, lims, 1); 1055 } 1056 goto out_reap; 1057 } 1058 1059 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1", 1060 __func__, nsrc0, nsrc1); 1061 1062 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 1063 if (imf->imf_st[0] == imf->imf_st[1] && 1064 imf->imf_st[1] == MCAST_INCLUDE) { 1065 if (nsrc1 == 0) { 1066 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1067 --inm->inm_st[1].iss_in; 1068 } 1069 } 1070 1071 /* Handle filter mode transition on socket. */ 1072 if (imf->imf_st[0] != imf->imf_st[1]) { 1073 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d", 1074 __func__, imf->imf_st[0], imf->imf_st[1]); 1075 1076 if (imf->imf_st[0] == MCAST_EXCLUDE) { 1077 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__); 1078 --inm->inm_st[1].iss_ex; 1079 } else if (imf->imf_st[0] == MCAST_INCLUDE) { 1080 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1081 --inm->inm_st[1].iss_in; 1082 } 1083 1084 if (imf->imf_st[1] == MCAST_EXCLUDE) { 1085 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__); 1086 inm->inm_st[1].iss_ex++; 1087 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) { 1088 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__); 1089 inm->inm_st[1].iss_in++; 1090 } 1091 } 1092 1093 /* 1094 * Track inm filter state in terms of listener counts. 1095 * If there are any exclusive listeners, stack-wide 1096 * membership is exclusive. 1097 * Otherwise, if only inclusive listeners, stack-wide is inclusive. 1098 * If no listeners remain, state is undefined at t1, 1099 * and the IGMP lifecycle for this group should finish. 1100 */ 1101 if (inm->inm_st[1].iss_ex > 0) { 1102 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__); 1103 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE; 1104 } else if (inm->inm_st[1].iss_in > 0) { 1105 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__); 1106 inm->inm_st[1].iss_fmode = MCAST_INCLUDE; 1107 } else { 1108 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__); 1109 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 1110 } 1111 1112 /* Decrement ASM listener count on transition out of ASM mode. */ 1113 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1114 if ((imf->imf_st[1] != MCAST_EXCLUDE) || 1115 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) { 1116 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__); 1117 --inm->inm_st[1].iss_asm; 1118 } 1119 } 1120 1121 /* Increment ASM listener count on transition to ASM mode. */ 1122 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1123 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__); 1124 inm->inm_st[1].iss_asm++; 1125 } 1126 1127 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm); 1128 inm_print(inm); 1129 1130 out_reap: 1131 if (schanged > 0) { 1132 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__); 1133 inm_reap(inm); 1134 } 1135 return (error); 1136 } 1137 1138 /* 1139 * Mark an in_multi's filter set deltas as committed. 1140 * Called by IGMP after a state change has been enqueued. 1141 */ 1142 void 1143 inm_commit(struct in_multi *inm) 1144 { 1145 struct ip_msource *ims; 1146 1147 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm); 1148 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__); 1149 inm_print(inm); 1150 1151 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 1152 ims->ims_st[0] = ims->ims_st[1]; 1153 } 1154 inm->inm_st[0] = inm->inm_st[1]; 1155 } 1156 1157 /* 1158 * Reap unreferenced nodes from an in_multi's filter set. 1159 */ 1160 static void 1161 inm_reap(struct in_multi *inm) 1162 { 1163 struct ip_msource *ims, *tims; 1164 1165 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1166 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 || 1167 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 || 1168 ims->ims_stp != 0) 1169 continue; 1170 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1171 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1172 free(ims, M_IPMSOURCE); 1173 inm->inm_nsrc--; 1174 } 1175 } 1176 1177 /* 1178 * Purge all source nodes from an in_multi's filter set. 1179 */ 1180 static void 1181 inm_purge(struct in_multi *inm) 1182 { 1183 struct ip_msource *ims, *tims; 1184 1185 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1186 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1187 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1188 free(ims, M_IPMSOURCE); 1189 inm->inm_nsrc--; 1190 } 1191 } 1192 1193 /* 1194 * Join a multicast group; unlocked entry point. 1195 * 1196 * SMPng: XXX: in_joingroup() is called from in_control() when Giant 1197 * is not held. Fortunately, ifp is unlikely to have been detached 1198 * at this point, so we assume it's OK to recurse. 1199 */ 1200 int 1201 in_joingroup(struct ifnet *ifp, const struct in_addr *gina, 1202 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1203 { 1204 int error; 1205 1206 IN_MULTI_LOCK(); 1207 error = in_joingroup_locked(ifp, gina, imf, pinm); 1208 IN_MULTI_UNLOCK(); 1209 1210 return (error); 1211 } 1212 1213 /* 1214 * Join a multicast group; real entry point. 1215 * 1216 * Only preserves atomicity at inm level. 1217 * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1218 * 1219 * If the IGMP downcall fails, the group is not joined, and an error 1220 * code is returned. 1221 */ 1222 int 1223 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina, 1224 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1225 { 1226 struct in_mfilter timf; 1227 struct in_multi *inm; 1228 int error; 1229 1230 IN_MULTI_LOCK_ASSERT(); 1231 IN_MULTI_LIST_UNLOCK_ASSERT(); 1232 1233 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__, 1234 ntohl(gina->s_addr), ifp, ifp->if_xname); 1235 1236 error = 0; 1237 inm = NULL; 1238 1239 /* 1240 * If no imf was specified (i.e. kernel consumer), 1241 * fake one up and assume it is an ASM join. 1242 */ 1243 if (imf == NULL) { 1244 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1245 imf = &timf; 1246 } 1247 1248 error = in_getmulti(ifp, gina, &inm); 1249 if (error) { 1250 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__); 1251 return (error); 1252 } 1253 IN_MULTI_LIST_LOCK(); 1254 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1255 error = inm_merge(inm, imf); 1256 if (error) { 1257 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1258 goto out_inm_release; 1259 } 1260 1261 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1262 error = igmp_change_state(inm); 1263 if (error) { 1264 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__); 1265 goto out_inm_release; 1266 } 1267 1268 out_inm_release: 1269 if (error) { 1270 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1271 IF_ADDR_WLOCK(ifp); 1272 inm_release_deferred(inm); 1273 IF_ADDR_WUNLOCK(ifp); 1274 } else { 1275 *pinm = inm; 1276 } 1277 IN_MULTI_LIST_UNLOCK(); 1278 1279 return (error); 1280 } 1281 1282 /* 1283 * Leave a multicast group; unlocked entry point. 1284 */ 1285 int 1286 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1287 { 1288 int error; 1289 1290 IN_MULTI_LOCK(); 1291 error = in_leavegroup_locked(inm, imf); 1292 IN_MULTI_UNLOCK(); 1293 1294 return (error); 1295 } 1296 1297 /* 1298 * Leave a multicast group; real entry point. 1299 * All source filters will be expunged. 1300 * 1301 * Only preserves atomicity at inm level. 1302 * 1303 * Holding the write lock for the INP which contains imf 1304 * is highly advisable. We can't assert for it as imf does not 1305 * contain a back-pointer to the owning inp. 1306 * 1307 * Note: This is not the same as inm_release(*) as this function also 1308 * makes a state change downcall into IGMP. 1309 */ 1310 int 1311 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1312 { 1313 struct in_mfilter timf; 1314 int error; 1315 1316 IN_MULTI_LOCK_ASSERT(); 1317 IN_MULTI_LIST_UNLOCK_ASSERT(); 1318 1319 error = 0; 1320 1321 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__, 1322 inm, ntohl(inm->inm_addr.s_addr), 1323 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname), 1324 imf); 1325 1326 /* 1327 * If no imf was specified (i.e. kernel consumer), 1328 * fake one up and assume it is an ASM join. 1329 */ 1330 if (imf == NULL) { 1331 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1332 imf = &timf; 1333 } 1334 1335 /* 1336 * Begin state merge transaction at IGMP layer. 1337 * 1338 * As this particular invocation should not cause any memory 1339 * to be allocated, and there is no opportunity to roll back 1340 * the transaction, it MUST NOT fail. 1341 */ 1342 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1343 IN_MULTI_LIST_LOCK(); 1344 error = inm_merge(inm, imf); 1345 KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1346 1347 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1348 CURVNET_SET(inm->inm_ifp->if_vnet); 1349 error = igmp_change_state(inm); 1350 IF_ADDR_WLOCK(inm->inm_ifp); 1351 inm_release_deferred(inm); 1352 IF_ADDR_WUNLOCK(inm->inm_ifp); 1353 IN_MULTI_LIST_UNLOCK(); 1354 CURVNET_RESTORE(); 1355 if (error) 1356 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1357 1358 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1359 1360 return (error); 1361 } 1362 1363 /*#ifndef BURN_BRIDGES*/ 1364 1365 /* 1366 * Block or unblock an ASM multicast source on an inpcb. 1367 * This implements the delta-based API described in RFC 3678. 1368 * 1369 * The delta-based API applies only to exclusive-mode memberships. 1370 * An IGMP downcall will be performed. 1371 * 1372 * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1373 * 1374 * Return 0 if successful, otherwise return an appropriate error code. 1375 */ 1376 static int 1377 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1378 { 1379 struct epoch_tracker et; 1380 struct group_source_req gsr; 1381 sockunion_t *gsa, *ssa; 1382 struct ifnet *ifp; 1383 struct in_mfilter *imf; 1384 struct ip_moptions *imo; 1385 struct in_msource *ims; 1386 struct in_multi *inm; 1387 uint16_t fmode; 1388 int error, doblock; 1389 1390 ifp = NULL; 1391 error = 0; 1392 doblock = 0; 1393 1394 memset(&gsr, 0, sizeof(struct group_source_req)); 1395 gsa = (sockunion_t *)&gsr.gsr_group; 1396 ssa = (sockunion_t *)&gsr.gsr_source; 1397 1398 switch (sopt->sopt_name) { 1399 case IP_BLOCK_SOURCE: 1400 case IP_UNBLOCK_SOURCE: { 1401 struct ip_mreq_source mreqs; 1402 1403 error = sooptcopyin(sopt, &mreqs, 1404 sizeof(struct ip_mreq_source), 1405 sizeof(struct ip_mreq_source)); 1406 if (error) 1407 return (error); 1408 1409 gsa->sin.sin_family = AF_INET; 1410 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1411 gsa->sin.sin_addr = mreqs.imr_multiaddr; 1412 1413 ssa->sin.sin_family = AF_INET; 1414 ssa->sin.sin_len = sizeof(struct sockaddr_in); 1415 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1416 1417 if (!in_nullhost(mreqs.imr_interface)) { 1418 NET_EPOCH_ENTER(et); 1419 INADDR_TO_IFP(mreqs.imr_interface, ifp); 1420 /* XXXGL: ifref? */ 1421 NET_EPOCH_EXIT(et); 1422 } 1423 if (sopt->sopt_name == IP_BLOCK_SOURCE) 1424 doblock = 1; 1425 1426 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 1427 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 1428 break; 1429 } 1430 1431 case MCAST_BLOCK_SOURCE: 1432 case MCAST_UNBLOCK_SOURCE: 1433 error = sooptcopyin(sopt, &gsr, 1434 sizeof(struct group_source_req), 1435 sizeof(struct group_source_req)); 1436 if (error) 1437 return (error); 1438 1439 if (gsa->sin.sin_family != AF_INET || 1440 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 1441 return (EINVAL); 1442 1443 if (ssa->sin.sin_family != AF_INET || 1444 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 1445 return (EINVAL); 1446 1447 NET_EPOCH_ENTER(et); 1448 ifp = ifnet_byindex(gsr.gsr_interface); 1449 NET_EPOCH_EXIT(et); 1450 if (ifp == NULL) 1451 return (EADDRNOTAVAIL); 1452 1453 if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1454 doblock = 1; 1455 break; 1456 1457 default: 1458 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1459 __func__, sopt->sopt_name); 1460 return (EOPNOTSUPP); 1461 break; 1462 } 1463 1464 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1465 return (EINVAL); 1466 1467 IN_MULTI_LOCK(); 1468 1469 /* 1470 * Check if we are actually a member of this group. 1471 */ 1472 imo = inp_findmoptions(inp); 1473 imf = imo_match_group(imo, ifp, &gsa->sa); 1474 if (imf == NULL) { 1475 error = EADDRNOTAVAIL; 1476 goto out_inp_locked; 1477 } 1478 inm = imf->imf_inm; 1479 1480 /* 1481 * Attempting to use the delta-based API on an 1482 * non exclusive-mode membership is an error. 1483 */ 1484 fmode = imf->imf_st[0]; 1485 if (fmode != MCAST_EXCLUDE) { 1486 error = EINVAL; 1487 goto out_inp_locked; 1488 } 1489 1490 /* 1491 * Deal with error cases up-front: 1492 * Asked to block, but already blocked; or 1493 * Asked to unblock, but nothing to unblock. 1494 * If adding a new block entry, allocate it. 1495 */ 1496 ims = imo_match_source(imf, &ssa->sa); 1497 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1498 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__, 1499 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not "); 1500 error = EADDRNOTAVAIL; 1501 goto out_inp_locked; 1502 } 1503 1504 INP_WLOCK_ASSERT(inp); 1505 1506 /* 1507 * Begin state merge transaction at socket layer. 1508 */ 1509 if (doblock) { 1510 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 1511 ims = imf_graft(imf, fmode, &ssa->sin); 1512 if (ims == NULL) 1513 error = ENOMEM; 1514 } else { 1515 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 1516 error = imf_prune(imf, &ssa->sin); 1517 } 1518 1519 if (error) { 1520 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__); 1521 goto out_imf_rollback; 1522 } 1523 1524 /* 1525 * Begin state merge transaction at IGMP layer. 1526 */ 1527 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1528 IN_MULTI_LIST_LOCK(); 1529 error = inm_merge(inm, imf); 1530 if (error) { 1531 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1532 IN_MULTI_LIST_UNLOCK(); 1533 goto out_imf_rollback; 1534 } 1535 1536 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1537 error = igmp_change_state(inm); 1538 IN_MULTI_LIST_UNLOCK(); 1539 if (error) 1540 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1541 1542 out_imf_rollback: 1543 if (error) 1544 imf_rollback(imf); 1545 else 1546 imf_commit(imf); 1547 1548 imf_reap(imf); 1549 1550 out_inp_locked: 1551 INP_WUNLOCK(inp); 1552 IN_MULTI_UNLOCK(); 1553 return (error); 1554 } 1555 1556 /* 1557 * Given an inpcb, return its multicast options structure pointer. Accepts 1558 * an unlocked inpcb pointer, but will return it locked. May sleep. 1559 * 1560 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1561 * SMPng: NOTE: Returns with the INP write lock held. 1562 */ 1563 static struct ip_moptions * 1564 inp_findmoptions(struct inpcb *inp) 1565 { 1566 struct ip_moptions *imo; 1567 1568 INP_WLOCK(inp); 1569 if (inp->inp_moptions != NULL) 1570 return (inp->inp_moptions); 1571 1572 INP_WUNLOCK(inp); 1573 1574 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK); 1575 1576 imo->imo_multicast_ifp = NULL; 1577 imo->imo_multicast_addr.s_addr = INADDR_ANY; 1578 imo->imo_multicast_vif = -1; 1579 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL; 1580 imo->imo_multicast_loop = in_mcast_loop; 1581 STAILQ_INIT(&imo->imo_head); 1582 1583 INP_WLOCK(inp); 1584 if (inp->inp_moptions != NULL) { 1585 free(imo, M_IPMOPTS); 1586 return (inp->inp_moptions); 1587 } 1588 inp->inp_moptions = imo; 1589 return (imo); 1590 } 1591 1592 void 1593 inp_freemoptions(struct ip_moptions *imo) 1594 { 1595 struct in_mfilter *imf; 1596 struct in_multi *inm; 1597 struct ifnet *ifp; 1598 1599 if (imo == NULL) 1600 return; 1601 1602 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) { 1603 ip_mfilter_remove(&imo->imo_head, imf); 1604 1605 imf_leave(imf); 1606 if ((inm = imf->imf_inm) != NULL) { 1607 if ((ifp = inm->inm_ifp) != NULL) { 1608 CURVNET_SET(ifp->if_vnet); 1609 (void)in_leavegroup(inm, imf); 1610 CURVNET_RESTORE(); 1611 } else { 1612 (void)in_leavegroup(inm, imf); 1613 } 1614 } 1615 ip_mfilter_free(imf); 1616 } 1617 free(imo, M_IPMOPTS); 1618 } 1619 1620 /* 1621 * Atomically get source filters on a socket for an IPv4 multicast group. 1622 * Called with INP lock held; returns with lock released. 1623 */ 1624 static int 1625 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1626 { 1627 struct epoch_tracker et; 1628 struct __msfilterreq msfr; 1629 sockunion_t *gsa; 1630 struct ifnet *ifp; 1631 struct ip_moptions *imo; 1632 struct in_mfilter *imf; 1633 struct ip_msource *ims; 1634 struct in_msource *lims; 1635 struct sockaddr_in *psin; 1636 struct sockaddr_storage *ptss; 1637 struct sockaddr_storage *tss; 1638 int error; 1639 size_t nsrcs, ncsrcs; 1640 1641 INP_WLOCK_ASSERT(inp); 1642 1643 imo = inp->inp_moptions; 1644 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__)); 1645 1646 INP_WUNLOCK(inp); 1647 1648 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1649 sizeof(struct __msfilterreq)); 1650 if (error) 1651 return (error); 1652 1653 NET_EPOCH_ENTER(et); 1654 ifp = ifnet_byindex(msfr.msfr_ifindex); 1655 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifnet pointer left */ 1656 if (ifp == NULL) 1657 return (EINVAL); 1658 1659 INP_WLOCK(inp); 1660 1661 /* 1662 * Lookup group on the socket. 1663 */ 1664 gsa = (sockunion_t *)&msfr.msfr_group; 1665 imf = imo_match_group(imo, ifp, &gsa->sa); 1666 if (imf == NULL) { 1667 INP_WUNLOCK(inp); 1668 return (EADDRNOTAVAIL); 1669 } 1670 1671 /* 1672 * Ignore memberships which are in limbo. 1673 */ 1674 if (imf->imf_st[1] == MCAST_UNDEFINED) { 1675 INP_WUNLOCK(inp); 1676 return (EAGAIN); 1677 } 1678 msfr.msfr_fmode = imf->imf_st[1]; 1679 1680 /* 1681 * If the user specified a buffer, copy out the source filter 1682 * entries to userland gracefully. 1683 * We only copy out the number of entries which userland 1684 * has asked for, but we always tell userland how big the 1685 * buffer really needs to be. 1686 */ 1687 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 1688 msfr.msfr_nsrcs = in_mcast_maxsocksrc; 1689 tss = NULL; 1690 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1691 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1692 M_TEMP, M_NOWAIT | M_ZERO); 1693 if (tss == NULL) { 1694 INP_WUNLOCK(inp); 1695 return (ENOBUFS); 1696 } 1697 } 1698 1699 /* 1700 * Count number of sources in-mode at t0. 1701 * If buffer space exists and remains, copy out source entries. 1702 */ 1703 nsrcs = msfr.msfr_nsrcs; 1704 ncsrcs = 0; 1705 ptss = tss; 1706 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1707 lims = (struct in_msource *)ims; 1708 if (lims->imsl_st[0] == MCAST_UNDEFINED || 1709 lims->imsl_st[0] != imf->imf_st[0]) 1710 continue; 1711 ++ncsrcs; 1712 if (tss != NULL && nsrcs > 0) { 1713 psin = (struct sockaddr_in *)ptss; 1714 psin->sin_family = AF_INET; 1715 psin->sin_len = sizeof(struct sockaddr_in); 1716 psin->sin_addr.s_addr = htonl(lims->ims_haddr); 1717 psin->sin_port = 0; 1718 ++ptss; 1719 --nsrcs; 1720 } 1721 } 1722 1723 INP_WUNLOCK(inp); 1724 1725 if (tss != NULL) { 1726 error = copyout(tss, msfr.msfr_srcs, 1727 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1728 free(tss, M_TEMP); 1729 if (error) 1730 return (error); 1731 } 1732 1733 msfr.msfr_nsrcs = ncsrcs; 1734 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1735 1736 return (error); 1737 } 1738 1739 /* 1740 * Return the IP multicast options in response to user getsockopt(). 1741 */ 1742 int 1743 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1744 { 1745 struct ip_mreqn mreqn; 1746 struct ip_moptions *imo; 1747 struct ifnet *ifp; 1748 struct in_ifaddr *ia; 1749 int error, optval; 1750 u_char coptval; 1751 1752 INP_WLOCK(inp); 1753 imo = inp->inp_moptions; 1754 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM reject it. */ 1755 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1756 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM) { 1757 INP_WUNLOCK(inp); 1758 return (EOPNOTSUPP); 1759 } 1760 1761 error = 0; 1762 switch (sopt->sopt_name) { 1763 case IP_MULTICAST_VIF: 1764 if (imo != NULL) 1765 optval = imo->imo_multicast_vif; 1766 else 1767 optval = -1; 1768 INP_WUNLOCK(inp); 1769 error = sooptcopyout(sopt, &optval, sizeof(int)); 1770 break; 1771 1772 case IP_MULTICAST_IF: 1773 memset(&mreqn, 0, sizeof(struct ip_mreqn)); 1774 if (imo != NULL) { 1775 ifp = imo->imo_multicast_ifp; 1776 if (!in_nullhost(imo->imo_multicast_addr)) { 1777 mreqn.imr_address = imo->imo_multicast_addr; 1778 } else if (ifp != NULL) { 1779 struct epoch_tracker et; 1780 1781 mreqn.imr_ifindex = ifp->if_index; 1782 NET_EPOCH_ENTER(et); 1783 IFP_TO_IA(ifp, ia); 1784 if (ia != NULL) 1785 mreqn.imr_address = 1786 IA_SIN(ia)->sin_addr; 1787 NET_EPOCH_EXIT(et); 1788 } 1789 } 1790 INP_WUNLOCK(inp); 1791 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 1792 error = sooptcopyout(sopt, &mreqn, 1793 sizeof(struct ip_mreqn)); 1794 } else { 1795 error = sooptcopyout(sopt, &mreqn.imr_address, 1796 sizeof(struct in_addr)); 1797 } 1798 break; 1799 1800 case IP_MULTICAST_TTL: 1801 if (imo == NULL) 1802 optval = coptval = IP_DEFAULT_MULTICAST_TTL; 1803 else 1804 optval = coptval = imo->imo_multicast_ttl; 1805 INP_WUNLOCK(inp); 1806 if (sopt->sopt_valsize == sizeof(u_char)) 1807 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1808 else 1809 error = sooptcopyout(sopt, &optval, sizeof(int)); 1810 break; 1811 1812 case IP_MULTICAST_LOOP: 1813 if (imo == NULL) 1814 optval = coptval = IP_DEFAULT_MULTICAST_LOOP; 1815 else 1816 optval = coptval = imo->imo_multicast_loop; 1817 INP_WUNLOCK(inp); 1818 if (sopt->sopt_valsize == sizeof(u_char)) 1819 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1820 else 1821 error = sooptcopyout(sopt, &optval, sizeof(int)); 1822 break; 1823 1824 case IP_MSFILTER: 1825 if (imo == NULL) { 1826 error = EADDRNOTAVAIL; 1827 INP_WUNLOCK(inp); 1828 } else { 1829 error = inp_get_source_filters(inp, sopt); 1830 } 1831 break; 1832 1833 default: 1834 INP_WUNLOCK(inp); 1835 error = ENOPROTOOPT; 1836 break; 1837 } 1838 1839 INP_UNLOCK_ASSERT(inp); 1840 1841 return (error); 1842 } 1843 1844 /* 1845 * Look up the ifnet to use for a multicast group membership, 1846 * given the IPv4 address of an interface, and the IPv4 group address. 1847 * 1848 * This routine exists to support legacy multicast applications 1849 * which do not understand that multicast memberships are scoped to 1850 * specific physical links in the networking stack, or which need 1851 * to join link-scope groups before IPv4 addresses are configured. 1852 * 1853 * Use this socket's current FIB number for any required FIB lookup. 1854 * If ina is INADDR_ANY, look up the group address in the unicast FIB, 1855 * and use its ifp; usually, this points to the default next-hop. 1856 * 1857 * If the FIB lookup fails, attempt to use the first non-loopback 1858 * interface with multicast capability in the system as a 1859 * last resort. The legacy IPv4 ASM API requires that we do 1860 * this in order to allow groups to be joined when the routing 1861 * table has not yet been populated during boot. 1862 * 1863 * Returns NULL if no ifp could be found, otherwise return referenced ifp. 1864 * 1865 * FUTURE: Implement IPv4 source-address selection. 1866 */ 1867 static struct ifnet * 1868 inp_lookup_mcast_ifp(const struct inpcb *inp, 1869 const struct sockaddr_in *gsin, const struct in_addr ina) 1870 { 1871 struct ifnet *ifp; 1872 struct nhop_object *nh; 1873 1874 NET_EPOCH_ASSERT(); 1875 KASSERT(inp != NULL, ("%s: inp must not be NULL", __func__)); 1876 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__)); 1877 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)), 1878 ("%s: not multicast", __func__)); 1879 1880 ifp = NULL; 1881 if (!in_nullhost(ina)) { 1882 INADDR_TO_IFP(ina, ifp); 1883 if (ifp != NULL) 1884 if_ref(ifp); 1885 } else { 1886 nh = fib4_lookup(inp->inp_inc.inc_fibnum, gsin->sin_addr, 0, NHR_NONE, 0); 1887 if (nh != NULL) { 1888 ifp = nh->nh_ifp; 1889 if_ref(ifp); 1890 } else { 1891 struct in_ifaddr *ia; 1892 struct ifnet *mifp; 1893 1894 mifp = NULL; 1895 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1896 mifp = ia->ia_ifp; 1897 if (!(mifp->if_flags & IFF_LOOPBACK) && 1898 (mifp->if_flags & IFF_MULTICAST)) { 1899 ifp = mifp; 1900 if_ref(ifp); 1901 break; 1902 } 1903 } 1904 } 1905 } 1906 1907 return (ifp); 1908 } 1909 1910 /* 1911 * Join an IPv4 multicast group, possibly with a source. 1912 */ 1913 static int 1914 inp_join_group(struct inpcb *inp, struct sockopt *sopt) 1915 { 1916 struct group_source_req gsr; 1917 sockunion_t *gsa, *ssa; 1918 struct ifnet *ifp; 1919 struct in_mfilter *imf; 1920 struct ip_moptions *imo; 1921 struct in_multi *inm; 1922 struct in_msource *lims; 1923 struct epoch_tracker et; 1924 int error, is_new; 1925 1926 ifp = NULL; 1927 lims = NULL; 1928 error = 0; 1929 1930 memset(&gsr, 0, sizeof(struct group_source_req)); 1931 gsa = (sockunion_t *)&gsr.gsr_group; 1932 gsa->ss.ss_family = AF_UNSPEC; 1933 ssa = (sockunion_t *)&gsr.gsr_source; 1934 ssa->ss.ss_family = AF_UNSPEC; 1935 1936 switch (sopt->sopt_name) { 1937 case IP_ADD_MEMBERSHIP: { 1938 struct ip_mreqn mreqn; 1939 1940 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) 1941 error = sooptcopyin(sopt, &mreqn, 1942 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn)); 1943 else 1944 error = sooptcopyin(sopt, &mreqn, 1945 sizeof(struct ip_mreq), sizeof(struct ip_mreq)); 1946 if (error) 1947 return (error); 1948 1949 gsa->sin.sin_family = AF_INET; 1950 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1951 gsa->sin.sin_addr = mreqn.imr_multiaddr; 1952 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1953 return (EINVAL); 1954 1955 NET_EPOCH_ENTER(et); 1956 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) && 1957 mreqn.imr_ifindex != 0) 1958 ifp = ifnet_byindex_ref(mreqn.imr_ifindex); 1959 else 1960 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 1961 mreqn.imr_address); 1962 NET_EPOCH_EXIT(et); 1963 break; 1964 } 1965 case IP_ADD_SOURCE_MEMBERSHIP: { 1966 struct ip_mreq_source mreqs; 1967 1968 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source), 1969 sizeof(struct ip_mreq_source)); 1970 if (error) 1971 return (error); 1972 1973 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET; 1974 gsa->sin.sin_len = ssa->sin.sin_len = 1975 sizeof(struct sockaddr_in); 1976 1977 gsa->sin.sin_addr = mreqs.imr_multiaddr; 1978 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1979 return (EINVAL); 1980 1981 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1982 1983 NET_EPOCH_ENTER(et); 1984 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 1985 mreqs.imr_interface); 1986 NET_EPOCH_EXIT(et); 1987 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 1988 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 1989 break; 1990 } 1991 1992 case MCAST_JOIN_GROUP: 1993 case MCAST_JOIN_SOURCE_GROUP: 1994 if (sopt->sopt_name == MCAST_JOIN_GROUP) { 1995 error = sooptcopyin(sopt, &gsr, 1996 sizeof(struct group_req), 1997 sizeof(struct group_req)); 1998 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 1999 error = sooptcopyin(sopt, &gsr, 2000 sizeof(struct group_source_req), 2001 sizeof(struct group_source_req)); 2002 } 2003 if (error) 2004 return (error); 2005 2006 if (gsa->sin.sin_family != AF_INET || 2007 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2008 return (EINVAL); 2009 2010 /* 2011 * Overwrite the port field if present, as the sockaddr 2012 * being copied in may be matched with a binary comparison. 2013 */ 2014 gsa->sin.sin_port = 0; 2015 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2016 if (ssa->sin.sin_family != AF_INET || 2017 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2018 return (EINVAL); 2019 ssa->sin.sin_port = 0; 2020 } 2021 2022 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2023 return (EINVAL); 2024 2025 NET_EPOCH_ENTER(et); 2026 ifp = ifnet_byindex_ref(gsr.gsr_interface); 2027 NET_EPOCH_EXIT(et); 2028 if (ifp == NULL) 2029 return (EADDRNOTAVAIL); 2030 break; 2031 2032 default: 2033 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2034 __func__, sopt->sopt_name); 2035 return (EOPNOTSUPP); 2036 break; 2037 } 2038 2039 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) { 2040 if (ifp != NULL) 2041 if_rele(ifp); 2042 return (EADDRNOTAVAIL); 2043 } 2044 2045 IN_MULTI_LOCK(); 2046 2047 /* 2048 * Find the membership in the membership list. 2049 */ 2050 imo = inp_findmoptions(inp); 2051 imf = imo_match_group(imo, ifp, &gsa->sa); 2052 if (imf == NULL) { 2053 is_new = 1; 2054 inm = NULL; 2055 2056 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) { 2057 error = ENOMEM; 2058 goto out_inp_locked; 2059 } 2060 } else { 2061 is_new = 0; 2062 inm = imf->imf_inm; 2063 2064 if (ssa->ss.ss_family != AF_UNSPEC) { 2065 /* 2066 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership 2067 * is an error. On an existing inclusive membership, 2068 * it just adds the source to the filter list. 2069 */ 2070 if (imf->imf_st[1] != MCAST_INCLUDE) { 2071 error = EINVAL; 2072 goto out_inp_locked; 2073 } 2074 /* 2075 * Throw out duplicates. 2076 * 2077 * XXX FIXME: This makes a naive assumption that 2078 * even if entries exist for *ssa in this imf, 2079 * they will be rejected as dupes, even if they 2080 * are not valid in the current mode (in-mode). 2081 * 2082 * in_msource is transactioned just as for anything 2083 * else in SSM -- but note naive use of inm_graft() 2084 * below for allocating new filter entries. 2085 * 2086 * This is only an issue if someone mixes the 2087 * full-state SSM API with the delta-based API, 2088 * which is discouraged in the relevant RFCs. 2089 */ 2090 lims = imo_match_source(imf, &ssa->sa); 2091 if (lims != NULL /*&& 2092 lims->imsl_st[1] == MCAST_INCLUDE*/) { 2093 error = EADDRNOTAVAIL; 2094 goto out_inp_locked; 2095 } 2096 } else { 2097 /* 2098 * MCAST_JOIN_GROUP on an existing exclusive 2099 * membership is an error; return EADDRINUSE 2100 * to preserve 4.4BSD API idempotence, and 2101 * avoid tedious detour to code below. 2102 * NOTE: This is bending RFC 3678 a bit. 2103 * 2104 * On an existing inclusive membership, this is also 2105 * an error; if you want to change filter mode, 2106 * you must use the userland API setsourcefilter(). 2107 * XXX We don't reject this for imf in UNDEFINED 2108 * state at t1, because allocation of a filter 2109 * is atomic with allocation of a membership. 2110 */ 2111 error = EINVAL; 2112 if (imf->imf_st[1] == MCAST_EXCLUDE) 2113 error = EADDRINUSE; 2114 goto out_inp_locked; 2115 } 2116 } 2117 2118 /* 2119 * Begin state merge transaction at socket layer. 2120 */ 2121 INP_WLOCK_ASSERT(inp); 2122 2123 /* 2124 * Graft new source into filter list for this inpcb's 2125 * membership of the group. The in_multi may not have 2126 * been allocated yet if this is a new membership, however, 2127 * the in_mfilter slot will be allocated and must be initialized. 2128 * 2129 * Note: Grafting of exclusive mode filters doesn't happen 2130 * in this path. 2131 * XXX: Should check for non-NULL lims (node exists but may 2132 * not be in-mode) for interop with full-state API. 2133 */ 2134 if (ssa->ss.ss_family != AF_UNSPEC) { 2135 /* Membership starts in IN mode */ 2136 if (is_new) { 2137 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__); 2138 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE); 2139 if (imf == NULL) { 2140 error = ENOMEM; 2141 goto out_inp_locked; 2142 } 2143 } else { 2144 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 2145 } 2146 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin); 2147 if (lims == NULL) { 2148 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2149 __func__); 2150 error = ENOMEM; 2151 goto out_inp_locked; 2152 } 2153 } else { 2154 /* No address specified; Membership starts in EX mode */ 2155 if (is_new) { 2156 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__); 2157 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE); 2158 if (imf == NULL) { 2159 error = ENOMEM; 2160 goto out_inp_locked; 2161 } 2162 } 2163 } 2164 2165 /* 2166 * Begin state merge transaction at IGMP layer. 2167 */ 2168 if (is_new) { 2169 in_pcbref(inp); 2170 INP_WUNLOCK(inp); 2171 2172 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf, 2173 &imf->imf_inm); 2174 2175 INP_WLOCK(inp); 2176 if (in_pcbrele_wlocked(inp)) { 2177 error = ENXIO; 2178 goto out_inp_unlocked; 2179 } 2180 if (error) { 2181 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed", 2182 __func__); 2183 goto out_inp_locked; 2184 } 2185 /* 2186 * NOTE: Refcount from in_joingroup_locked() 2187 * is protecting membership. 2188 */ 2189 ip_mfilter_insert(&imo->imo_head, imf); 2190 } else { 2191 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2192 IN_MULTI_LIST_LOCK(); 2193 error = inm_merge(inm, imf); 2194 if (error) { 2195 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2196 __func__); 2197 IN_MULTI_LIST_UNLOCK(); 2198 imf_rollback(imf); 2199 imf_reap(imf); 2200 goto out_inp_locked; 2201 } 2202 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2203 error = igmp_change_state(inm); 2204 IN_MULTI_LIST_UNLOCK(); 2205 if (error) { 2206 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2207 __func__); 2208 imf_rollback(imf); 2209 imf_reap(imf); 2210 goto out_inp_locked; 2211 } 2212 } 2213 2214 imf_commit(imf); 2215 imf = NULL; 2216 2217 out_inp_locked: 2218 INP_WUNLOCK(inp); 2219 out_inp_unlocked: 2220 IN_MULTI_UNLOCK(); 2221 2222 if (is_new && imf) { 2223 if (imf->imf_inm != NULL) { 2224 IN_MULTI_LIST_LOCK(); 2225 IF_ADDR_WLOCK(ifp); 2226 inm_release_deferred(imf->imf_inm); 2227 IF_ADDR_WUNLOCK(ifp); 2228 IN_MULTI_LIST_UNLOCK(); 2229 } 2230 ip_mfilter_free(imf); 2231 } 2232 if_rele(ifp); 2233 return (error); 2234 } 2235 2236 /* 2237 * Leave an IPv4 multicast group on an inpcb, possibly with a source. 2238 */ 2239 static int 2240 inp_leave_group(struct inpcb *inp, struct sockopt *sopt) 2241 { 2242 struct epoch_tracker et; 2243 struct group_source_req gsr; 2244 struct ip_mreq_source mreqs; 2245 sockunion_t *gsa, *ssa; 2246 struct ifnet *ifp; 2247 struct in_mfilter *imf; 2248 struct ip_moptions *imo; 2249 struct in_msource *ims; 2250 struct in_multi *inm; 2251 int error; 2252 bool is_final; 2253 2254 ifp = NULL; 2255 error = 0; 2256 is_final = true; 2257 2258 memset(&gsr, 0, sizeof(struct group_source_req)); 2259 gsa = (sockunion_t *)&gsr.gsr_group; 2260 gsa->ss.ss_family = AF_UNSPEC; 2261 ssa = (sockunion_t *)&gsr.gsr_source; 2262 ssa->ss.ss_family = AF_UNSPEC; 2263 2264 switch (sopt->sopt_name) { 2265 case IP_DROP_MEMBERSHIP: 2266 case IP_DROP_SOURCE_MEMBERSHIP: 2267 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) { 2268 error = sooptcopyin(sopt, &mreqs, 2269 sizeof(struct ip_mreq), 2270 sizeof(struct ip_mreq)); 2271 /* 2272 * Swap interface and sourceaddr arguments, 2273 * as ip_mreq and ip_mreq_source are laid 2274 * out differently. 2275 */ 2276 mreqs.imr_interface = mreqs.imr_sourceaddr; 2277 mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 2278 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2279 error = sooptcopyin(sopt, &mreqs, 2280 sizeof(struct ip_mreq_source), 2281 sizeof(struct ip_mreq_source)); 2282 } 2283 if (error) 2284 return (error); 2285 2286 gsa->sin.sin_family = AF_INET; 2287 gsa->sin.sin_len = sizeof(struct sockaddr_in); 2288 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2289 2290 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2291 ssa->sin.sin_family = AF_INET; 2292 ssa->sin.sin_len = sizeof(struct sockaddr_in); 2293 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2294 } 2295 2296 /* 2297 * Attempt to look up hinted ifp from interface address. 2298 * Fallthrough with null ifp iff lookup fails, to 2299 * preserve 4.4BSD mcast API idempotence. 2300 * XXX NOTE WELL: The RFC 3678 API is preferred because 2301 * using an IPv4 address as a key is racy. 2302 */ 2303 if (!in_nullhost(mreqs.imr_interface)) { 2304 NET_EPOCH_ENTER(et); 2305 INADDR_TO_IFP(mreqs.imr_interface, ifp); 2306 /* XXXGL ifref? */ 2307 NET_EPOCH_EXIT(et); 2308 } 2309 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2310 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2311 2312 break; 2313 2314 case MCAST_LEAVE_GROUP: 2315 case MCAST_LEAVE_SOURCE_GROUP: 2316 if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2317 error = sooptcopyin(sopt, &gsr, 2318 sizeof(struct group_req), 2319 sizeof(struct group_req)); 2320 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2321 error = sooptcopyin(sopt, &gsr, 2322 sizeof(struct group_source_req), 2323 sizeof(struct group_source_req)); 2324 } 2325 if (error) 2326 return (error); 2327 2328 if (gsa->sin.sin_family != AF_INET || 2329 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2330 return (EINVAL); 2331 2332 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2333 if (ssa->sin.sin_family != AF_INET || 2334 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2335 return (EINVAL); 2336 } 2337 2338 NET_EPOCH_ENTER(et); 2339 ifp = ifnet_byindex(gsr.gsr_interface); 2340 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */ 2341 if (ifp == NULL) 2342 return (EADDRNOTAVAIL); 2343 break; 2344 2345 default: 2346 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2347 __func__, sopt->sopt_name); 2348 return (EOPNOTSUPP); 2349 break; 2350 } 2351 2352 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2353 return (EINVAL); 2354 2355 IN_MULTI_LOCK(); 2356 2357 /* 2358 * Find the membership in the membership list. 2359 */ 2360 imo = inp_findmoptions(inp); 2361 imf = imo_match_group(imo, ifp, &gsa->sa); 2362 if (imf == NULL) { 2363 error = EADDRNOTAVAIL; 2364 goto out_inp_locked; 2365 } 2366 inm = imf->imf_inm; 2367 2368 if (ssa->ss.ss_family != AF_UNSPEC) 2369 is_final = false; 2370 2371 /* 2372 * Begin state merge transaction at socket layer. 2373 */ 2374 INP_WLOCK_ASSERT(inp); 2375 2376 /* 2377 * If we were instructed only to leave a given source, do so. 2378 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2379 */ 2380 if (is_final) { 2381 ip_mfilter_remove(&imo->imo_head, imf); 2382 imf_leave(imf); 2383 2384 /* 2385 * Give up the multicast address record to which 2386 * the membership points. 2387 */ 2388 (void) in_leavegroup_locked(imf->imf_inm, imf); 2389 } else { 2390 if (imf->imf_st[0] == MCAST_EXCLUDE) { 2391 error = EADDRNOTAVAIL; 2392 goto out_inp_locked; 2393 } 2394 ims = imo_match_source(imf, &ssa->sa); 2395 if (ims == NULL) { 2396 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", 2397 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not "); 2398 error = EADDRNOTAVAIL; 2399 goto out_inp_locked; 2400 } 2401 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 2402 error = imf_prune(imf, &ssa->sin); 2403 if (error) { 2404 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2405 __func__); 2406 goto out_inp_locked; 2407 } 2408 } 2409 2410 /* 2411 * Begin state merge transaction at IGMP layer. 2412 */ 2413 if (!is_final) { 2414 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2415 IN_MULTI_LIST_LOCK(); 2416 error = inm_merge(inm, imf); 2417 if (error) { 2418 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2419 __func__); 2420 IN_MULTI_LIST_UNLOCK(); 2421 imf_rollback(imf); 2422 imf_reap(imf); 2423 goto out_inp_locked; 2424 } 2425 2426 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2427 error = igmp_change_state(inm); 2428 IN_MULTI_LIST_UNLOCK(); 2429 if (error) { 2430 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2431 __func__); 2432 imf_rollback(imf); 2433 imf_reap(imf); 2434 goto out_inp_locked; 2435 } 2436 } 2437 imf_commit(imf); 2438 imf_reap(imf); 2439 2440 out_inp_locked: 2441 INP_WUNLOCK(inp); 2442 2443 if (is_final && imf) 2444 ip_mfilter_free(imf); 2445 2446 IN_MULTI_UNLOCK(); 2447 return (error); 2448 } 2449 2450 /* 2451 * Select the interface for transmitting IPv4 multicast datagrams. 2452 * 2453 * Either an instance of struct in_addr or an instance of struct ip_mreqn 2454 * may be passed to this socket option. An address of INADDR_ANY or an 2455 * interface index of 0 is used to remove a previous selection. 2456 * When no interface is selected, one is chosen for every send. 2457 */ 2458 static int 2459 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2460 { 2461 struct in_addr addr; 2462 struct ip_mreqn mreqn; 2463 struct ifnet *ifp; 2464 struct ip_moptions *imo; 2465 int error; 2466 2467 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 2468 /* 2469 * An interface index was specified using the 2470 * Linux-derived ip_mreqn structure. 2471 */ 2472 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn), 2473 sizeof(struct ip_mreqn)); 2474 if (error) 2475 return (error); 2476 2477 if (mreqn.imr_ifindex < 0) 2478 return (EINVAL); 2479 2480 if (mreqn.imr_ifindex == 0) { 2481 ifp = NULL; 2482 } else { 2483 struct epoch_tracker et; 2484 2485 NET_EPOCH_ENTER(et); 2486 ifp = ifnet_byindex(mreqn.imr_ifindex); 2487 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */ 2488 if (ifp == NULL) 2489 return (EADDRNOTAVAIL); 2490 } 2491 } else { 2492 /* 2493 * An interface was specified by IPv4 address. 2494 * This is the traditional BSD usage. 2495 */ 2496 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr), 2497 sizeof(struct in_addr)); 2498 if (error) 2499 return (error); 2500 if (in_nullhost(addr)) { 2501 ifp = NULL; 2502 } else { 2503 struct epoch_tracker et; 2504 2505 NET_EPOCH_ENTER(et); 2506 INADDR_TO_IFP(addr, ifp); 2507 /* XXXGL ifref? */ 2508 NET_EPOCH_EXIT(et); 2509 if (ifp == NULL) 2510 return (EADDRNOTAVAIL); 2511 } 2512 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp, 2513 ntohl(addr.s_addr)); 2514 } 2515 2516 /* Reject interfaces which do not support multicast. */ 2517 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0) 2518 return (EOPNOTSUPP); 2519 2520 imo = inp_findmoptions(inp); 2521 imo->imo_multicast_ifp = ifp; 2522 imo->imo_multicast_addr.s_addr = INADDR_ANY; 2523 INP_WUNLOCK(inp); 2524 2525 return (0); 2526 } 2527 2528 /* 2529 * Atomically set source filters on a socket for an IPv4 multicast group. 2530 * 2531 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2532 */ 2533 static int 2534 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2535 { 2536 struct epoch_tracker et; 2537 struct __msfilterreq msfr; 2538 sockunion_t *gsa; 2539 struct ifnet *ifp; 2540 struct in_mfilter *imf; 2541 struct ip_moptions *imo; 2542 struct in_multi *inm; 2543 int error; 2544 2545 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2546 sizeof(struct __msfilterreq)); 2547 if (error) 2548 return (error); 2549 2550 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 2551 return (ENOBUFS); 2552 2553 if ((msfr.msfr_fmode != MCAST_EXCLUDE && 2554 msfr.msfr_fmode != MCAST_INCLUDE)) 2555 return (EINVAL); 2556 2557 if (msfr.msfr_group.ss_family != AF_INET || 2558 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in)) 2559 return (EINVAL); 2560 2561 gsa = (sockunion_t *)&msfr.msfr_group; 2562 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2563 return (EINVAL); 2564 2565 gsa->sin.sin_port = 0; /* ignore port */ 2566 2567 NET_EPOCH_ENTER(et); 2568 ifp = ifnet_byindex(msfr.msfr_ifindex); 2569 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */ 2570 if (ifp == NULL) 2571 return (EADDRNOTAVAIL); 2572 2573 IN_MULTI_LOCK(); 2574 2575 /* 2576 * Take the INP write lock. 2577 * Check if this socket is a member of this group. 2578 */ 2579 imo = inp_findmoptions(inp); 2580 imf = imo_match_group(imo, ifp, &gsa->sa); 2581 if (imf == NULL) { 2582 error = EADDRNOTAVAIL; 2583 goto out_inp_locked; 2584 } 2585 inm = imf->imf_inm; 2586 2587 /* 2588 * Begin state merge transaction at socket layer. 2589 */ 2590 INP_WLOCK_ASSERT(inp); 2591 2592 imf->imf_st[1] = msfr.msfr_fmode; 2593 2594 /* 2595 * Apply any new source filters, if present. 2596 * Make a copy of the user-space source vector so 2597 * that we may copy them with a single copyin. This 2598 * allows us to deal with page faults up-front. 2599 */ 2600 if (msfr.msfr_nsrcs > 0) { 2601 struct in_msource *lims; 2602 struct sockaddr_in *psin; 2603 struct sockaddr_storage *kss, *pkss; 2604 int i; 2605 2606 INP_WUNLOCK(inp); 2607 2608 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries", 2609 __func__, (unsigned long)msfr.msfr_nsrcs); 2610 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2611 M_TEMP, M_WAITOK); 2612 error = copyin(msfr.msfr_srcs, kss, 2613 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2614 if (error) { 2615 free(kss, M_TEMP); 2616 return (error); 2617 } 2618 2619 INP_WLOCK(inp); 2620 2621 /* 2622 * Mark all source filters as UNDEFINED at t1. 2623 * Restore new group filter mode, as imf_leave() 2624 * will set it to INCLUDE. 2625 */ 2626 imf_leave(imf); 2627 imf->imf_st[1] = msfr.msfr_fmode; 2628 2629 /* 2630 * Update socket layer filters at t1, lazy-allocating 2631 * new entries. This saves a bunch of memory at the 2632 * cost of one RB_FIND() per source entry; duplicate 2633 * entries in the msfr_nsrcs vector are ignored. 2634 * If we encounter an error, rollback transaction. 2635 * 2636 * XXX This too could be replaced with a set-symmetric 2637 * difference like loop to avoid walking from root 2638 * every time, as the key space is common. 2639 */ 2640 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2641 psin = (struct sockaddr_in *)pkss; 2642 if (psin->sin_family != AF_INET) { 2643 error = EAFNOSUPPORT; 2644 break; 2645 } 2646 if (psin->sin_len != sizeof(struct sockaddr_in)) { 2647 error = EINVAL; 2648 break; 2649 } 2650 error = imf_get_source(imf, psin, &lims); 2651 if (error) 2652 break; 2653 lims->imsl_st[1] = imf->imf_st[1]; 2654 } 2655 free(kss, M_TEMP); 2656 } 2657 2658 if (error) 2659 goto out_imf_rollback; 2660 2661 INP_WLOCK_ASSERT(inp); 2662 2663 /* 2664 * Begin state merge transaction at IGMP layer. 2665 */ 2666 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2667 IN_MULTI_LIST_LOCK(); 2668 error = inm_merge(inm, imf); 2669 if (error) { 2670 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 2671 IN_MULTI_LIST_UNLOCK(); 2672 goto out_imf_rollback; 2673 } 2674 2675 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2676 error = igmp_change_state(inm); 2677 IN_MULTI_LIST_UNLOCK(); 2678 if (error) 2679 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 2680 2681 out_imf_rollback: 2682 if (error) 2683 imf_rollback(imf); 2684 else 2685 imf_commit(imf); 2686 2687 imf_reap(imf); 2688 2689 out_inp_locked: 2690 INP_WUNLOCK(inp); 2691 IN_MULTI_UNLOCK(); 2692 return (error); 2693 } 2694 2695 /* 2696 * Set the IP multicast options in response to user setsockopt(). 2697 * 2698 * Many of the socket options handled in this function duplicate the 2699 * functionality of socket options in the regular unicast API. However, 2700 * it is not possible to merge the duplicate code, because the idempotence 2701 * of the IPv4 multicast part of the BSD Sockets API must be preserved; 2702 * the effects of these options must be treated as separate and distinct. 2703 * 2704 * SMPng: XXX: Unlocked read of inp_socket believed OK. 2705 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING 2706 * is refactored to no longer use vifs. 2707 */ 2708 int 2709 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2710 { 2711 struct ip_moptions *imo; 2712 int error; 2713 2714 error = 0; 2715 2716 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */ 2717 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2718 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM) 2719 return (EOPNOTSUPP); 2720 2721 switch (sopt->sopt_name) { 2722 case IP_MULTICAST_VIF: { 2723 int vifi; 2724 /* 2725 * Select a multicast VIF for transmission. 2726 * Only useful if multicast forwarding is active. 2727 */ 2728 if (legal_vif_num == NULL) { 2729 error = EOPNOTSUPP; 2730 break; 2731 } 2732 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int)); 2733 if (error) 2734 break; 2735 if (!legal_vif_num(vifi) && (vifi != -1)) { 2736 error = EINVAL; 2737 break; 2738 } 2739 imo = inp_findmoptions(inp); 2740 imo->imo_multicast_vif = vifi; 2741 INP_WUNLOCK(inp); 2742 break; 2743 } 2744 2745 case IP_MULTICAST_IF: 2746 error = inp_set_multicast_if(inp, sopt); 2747 break; 2748 2749 case IP_MULTICAST_TTL: { 2750 u_char ttl; 2751 2752 /* 2753 * Set the IP time-to-live for outgoing multicast packets. 2754 * The original multicast API required a char argument, 2755 * which is inconsistent with the rest of the socket API. 2756 * We allow either a char or an int. 2757 */ 2758 if (sopt->sopt_valsize == sizeof(u_char)) { 2759 error = sooptcopyin(sopt, &ttl, sizeof(u_char), 2760 sizeof(u_char)); 2761 if (error) 2762 break; 2763 } else { 2764 u_int ittl; 2765 2766 error = sooptcopyin(sopt, &ittl, sizeof(u_int), 2767 sizeof(u_int)); 2768 if (error) 2769 break; 2770 if (ittl > 255) { 2771 error = EINVAL; 2772 break; 2773 } 2774 ttl = (u_char)ittl; 2775 } 2776 imo = inp_findmoptions(inp); 2777 imo->imo_multicast_ttl = ttl; 2778 INP_WUNLOCK(inp); 2779 break; 2780 } 2781 2782 case IP_MULTICAST_LOOP: { 2783 u_char loop; 2784 2785 /* 2786 * Set the loopback flag for outgoing multicast packets. 2787 * Must be zero or one. The original multicast API required a 2788 * char argument, which is inconsistent with the rest 2789 * of the socket API. We allow either a char or an int. 2790 */ 2791 if (sopt->sopt_valsize == sizeof(u_char)) { 2792 error = sooptcopyin(sopt, &loop, sizeof(u_char), 2793 sizeof(u_char)); 2794 if (error) 2795 break; 2796 } else { 2797 u_int iloop; 2798 2799 error = sooptcopyin(sopt, &iloop, sizeof(u_int), 2800 sizeof(u_int)); 2801 if (error) 2802 break; 2803 loop = (u_char)iloop; 2804 } 2805 imo = inp_findmoptions(inp); 2806 imo->imo_multicast_loop = !!loop; 2807 INP_WUNLOCK(inp); 2808 break; 2809 } 2810 2811 case IP_ADD_MEMBERSHIP: 2812 case IP_ADD_SOURCE_MEMBERSHIP: 2813 case MCAST_JOIN_GROUP: 2814 case MCAST_JOIN_SOURCE_GROUP: 2815 error = inp_join_group(inp, sopt); 2816 break; 2817 2818 case IP_DROP_MEMBERSHIP: 2819 case IP_DROP_SOURCE_MEMBERSHIP: 2820 case MCAST_LEAVE_GROUP: 2821 case MCAST_LEAVE_SOURCE_GROUP: 2822 error = inp_leave_group(inp, sopt); 2823 break; 2824 2825 case IP_BLOCK_SOURCE: 2826 case IP_UNBLOCK_SOURCE: 2827 case MCAST_BLOCK_SOURCE: 2828 case MCAST_UNBLOCK_SOURCE: 2829 error = inp_block_unblock_source(inp, sopt); 2830 break; 2831 2832 case IP_MSFILTER: 2833 error = inp_set_source_filters(inp, sopt); 2834 break; 2835 2836 default: 2837 error = EOPNOTSUPP; 2838 break; 2839 } 2840 2841 INP_UNLOCK_ASSERT(inp); 2842 2843 return (error); 2844 } 2845 2846 /* 2847 * Expose IGMP's multicast filter mode and source list(s) to userland, 2848 * keyed by (ifindex, group). 2849 * The filter mode is written out as a uint32_t, followed by 2850 * 0..n of struct in_addr. 2851 * For use by ifmcstat(8). 2852 * SMPng: NOTE: unlocked read of ifindex space. 2853 */ 2854 static int 2855 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS) 2856 { 2857 struct in_addr src, group; 2858 struct epoch_tracker et; 2859 struct ifnet *ifp; 2860 struct ifmultiaddr *ifma; 2861 struct in_multi *inm; 2862 struct ip_msource *ims; 2863 int *name; 2864 int retval; 2865 u_int namelen; 2866 uint32_t fmode, ifindex; 2867 2868 name = (int *)arg1; 2869 namelen = arg2; 2870 2871 if (req->newptr != NULL) 2872 return (EPERM); 2873 2874 if (namelen != 2) 2875 return (EINVAL); 2876 2877 group.s_addr = name[1]; 2878 if (!IN_MULTICAST(ntohl(group.s_addr))) { 2879 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast", 2880 __func__, ntohl(group.s_addr)); 2881 return (EINVAL); 2882 } 2883 2884 ifindex = name[0]; 2885 NET_EPOCH_ENTER(et); 2886 ifp = ifnet_byindex(ifindex); 2887 if (ifp == NULL) { 2888 NET_EPOCH_EXIT(et); 2889 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u", 2890 __func__, ifindex); 2891 return (ENOENT); 2892 } 2893 2894 retval = sysctl_wire_old_buffer(req, 2895 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr))); 2896 if (retval) { 2897 NET_EPOCH_EXIT(et); 2898 return (retval); 2899 } 2900 2901 IN_MULTI_LIST_LOCK(); 2902 2903 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2904 if (ifma->ifma_addr->sa_family != AF_INET || 2905 ifma->ifma_protospec == NULL) 2906 continue; 2907 inm = (struct in_multi *)ifma->ifma_protospec; 2908 if (!in_hosteq(inm->inm_addr, group)) 2909 continue; 2910 fmode = inm->inm_st[1].iss_fmode; 2911 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 2912 if (retval != 0) 2913 break; 2914 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 2915 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2916 ims->ims_haddr); 2917 /* 2918 * Only copy-out sources which are in-mode. 2919 */ 2920 if (fmode != ims_get_mode(inm, ims, 1)) { 2921 CTR1(KTR_IGMPV3, "%s: skip non-in-mode", 2922 __func__); 2923 continue; 2924 } 2925 src.s_addr = htonl(ims->ims_haddr); 2926 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr)); 2927 if (retval != 0) 2928 break; 2929 } 2930 } 2931 2932 IN_MULTI_LIST_UNLOCK(); 2933 NET_EPOCH_EXIT(et); 2934 2935 return (retval); 2936 } 2937 2938 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3) 2939 2940 static const char *inm_modestrs[] = { 2941 [MCAST_UNDEFINED] = "un", 2942 [MCAST_INCLUDE] = "in", 2943 [MCAST_EXCLUDE] = "ex", 2944 }; 2945 _Static_assert(MCAST_UNDEFINED == 0 && 2946 MCAST_EXCLUDE + 1 == nitems(inm_modestrs), 2947 "inm_modestrs: no longer matches #defines"); 2948 2949 static const char * 2950 inm_mode_str(const int mode) 2951 { 2952 2953 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 2954 return (inm_modestrs[mode]); 2955 return ("??"); 2956 } 2957 2958 static const char *inm_statestrs[] = { 2959 [IGMP_NOT_MEMBER] = "not-member", 2960 [IGMP_SILENT_MEMBER] = "silent", 2961 [IGMP_REPORTING_MEMBER] = "reporting", 2962 [IGMP_IDLE_MEMBER] = "idle", 2963 [IGMP_LAZY_MEMBER] = "lazy", 2964 [IGMP_SLEEPING_MEMBER] = "sleeping", 2965 [IGMP_AWAKENING_MEMBER] = "awakening", 2966 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending", 2967 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending", 2968 [IGMP_LEAVING_MEMBER] = "leaving", 2969 }; 2970 _Static_assert(IGMP_NOT_MEMBER == 0 && 2971 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs), 2972 "inm_statetrs: no longer matches #defines"); 2973 2974 static const char * 2975 inm_state_str(const int state) 2976 { 2977 2978 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER) 2979 return (inm_statestrs[state]); 2980 return ("??"); 2981 } 2982 2983 /* 2984 * Dump an in_multi structure to the console. 2985 */ 2986 void 2987 inm_print(const struct in_multi *inm) 2988 { 2989 int t; 2990 char addrbuf[INET_ADDRSTRLEN]; 2991 2992 if ((ktr_mask & KTR_IGMPV3) == 0) 2993 return; 2994 2995 printf("%s: --- begin inm %p ---\n", __func__, inm); 2996 printf("addr %s ifp %p(%s) ifma %p\n", 2997 inet_ntoa_r(inm->inm_addr, addrbuf), 2998 inm->inm_ifp, 2999 inm->inm_ifp->if_xname, 3000 inm->inm_ifma); 3001 printf("timer %u state %s refcount %u scq.len %u\n", 3002 inm->inm_timer, 3003 inm_state_str(inm->inm_state), 3004 inm->inm_refcount, 3005 inm->inm_scq.mq_len); 3006 printf("igi %p nsrc %lu sctimer %u scrv %u\n", 3007 inm->inm_igi, 3008 inm->inm_nsrc, 3009 inm->inm_sctimer, 3010 inm->inm_scrv); 3011 for (t = 0; t < 2; t++) { 3012 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 3013 inm_mode_str(inm->inm_st[t].iss_fmode), 3014 inm->inm_st[t].iss_asm, 3015 inm->inm_st[t].iss_ex, 3016 inm->inm_st[t].iss_in, 3017 inm->inm_st[t].iss_rec); 3018 } 3019 printf("%s: --- end inm %p ---\n", __func__, inm); 3020 } 3021 3022 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */ 3023 3024 void 3025 inm_print(const struct in_multi *inm) 3026 { 3027 3028 } 3029 3030 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */ 3031 3032 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp); 3033