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/rmlock.h> 48 #include <sys/socket.h> 49 #include <sys/socketvar.h> 50 #include <sys/protosw.h> 51 #include <sys/sysctl.h> 52 #include <sys/ktr.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 <net/ethernet.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_systm.h> 67 #include <netinet/in_fib.h> 68 #include <netinet/in_pcb.h> 69 #include <netinet/in_var.h> 70 #include <netinet/ip_var.h> 71 #include <netinet/igmp_var.h> 72 73 #ifndef KTR_IGMPV3 74 #define KTR_IGMPV3 KTR_INET 75 #endif 76 77 #ifndef __SOCKUNION_DECLARED 78 union sockunion { 79 struct sockaddr_storage ss; 80 struct sockaddr sa; 81 struct sockaddr_dl sdl; 82 struct sockaddr_in sin; 83 }; 84 typedef union sockunion sockunion_t; 85 #define __SOCKUNION_DECLARED 86 #endif /* __SOCKUNION_DECLARED */ 87 88 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter", 89 "IPv4 multicast PCB-layer source filter"); 90 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group"); 91 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options"); 92 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource", 93 "IPv4 multicast IGMP-layer source filter"); 94 95 /* 96 * Locking: 97 * 98 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK, 99 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 100 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however 101 * it can be taken by code in net/if.c also. 102 * - ip_moptions and in_mfilter are covered by the INP_WLOCK. 103 * 104 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly 105 * any need for in_multi itself to be virtualized -- it is bound to an ifp 106 * anyway no matter what happens. 107 */ 108 struct mtx in_multi_list_mtx; 109 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF); 110 111 struct mtx in_multi_free_mtx; 112 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF); 113 114 struct sx in_multi_sx; 115 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx"); 116 117 int ifma_restart; 118 119 /* 120 * Functions with non-static linkage defined in this file should be 121 * declared in in_var.h: 122 * imo_multi_filter() 123 * in_joingroup() 124 * in_joingroup_locked() 125 * in_leavegroup() 126 * in_leavegroup_locked() 127 * and ip_var.h: 128 * inp_freemoptions() 129 * inp_getmoptions() 130 * inp_setmoptions() 131 */ 132 static void imf_commit(struct in_mfilter *); 133 static int imf_get_source(struct in_mfilter *imf, 134 const struct sockaddr_in *psin, 135 struct in_msource **); 136 static struct in_msource * 137 imf_graft(struct in_mfilter *, const uint8_t, 138 const struct sockaddr_in *); 139 static void imf_leave(struct in_mfilter *); 140 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *); 141 static void imf_purge(struct in_mfilter *); 142 static void imf_rollback(struct in_mfilter *); 143 static void imf_reap(struct in_mfilter *); 144 static struct in_mfilter * 145 imo_match_group(const struct ip_moptions *, 146 const struct ifnet *, const struct sockaddr *); 147 static struct in_msource * 148 imo_match_source(struct in_mfilter *, const struct sockaddr *); 149 static void ims_merge(struct ip_msource *ims, 150 const struct in_msource *lims, const int rollback); 151 static int in_getmulti(struct ifnet *, const struct in_addr *, 152 struct in_multi **); 153 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr, 154 const int noalloc, struct ip_msource **pims); 155 #ifdef KTR 156 static int inm_is_ifp_detached(const struct in_multi *); 157 #endif 158 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *); 159 static void inm_purge(struct in_multi *); 160 static void inm_reap(struct in_multi *); 161 static void inm_release(struct in_multi *); 162 static struct ip_moptions * 163 inp_findmoptions(struct inpcb *); 164 static int inp_get_source_filters(struct inpcb *, struct sockopt *); 165 static int inp_join_group(struct inpcb *, struct sockopt *); 166 static int inp_leave_group(struct inpcb *, struct sockopt *); 167 static struct ifnet * 168 inp_lookup_mcast_ifp(const struct inpcb *, 169 const struct sockaddr_in *, const struct in_addr); 170 static int inp_block_unblock_source(struct inpcb *, struct sockopt *); 171 static int inp_set_multicast_if(struct inpcb *, struct sockopt *); 172 static int inp_set_source_filters(struct inpcb *, struct sockopt *); 173 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS); 174 175 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, 176 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 177 "IPv4 multicast"); 178 179 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER; 180 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc, 181 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0, 182 "Max source filters per group"); 183 184 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER; 185 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc, 186 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0, 187 "Max source filters per socket"); 188 189 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP; 190 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN, 191 &in_mcast_loop, 0, "Loopback multicast datagrams by default"); 192 193 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters, 194 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters, 195 "Per-interface stack-wide source filters"); 196 197 #ifdef KTR 198 /* 199 * Inline function which wraps assertions for a valid ifp. 200 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 201 * is detached. 202 */ 203 static int __inline 204 inm_is_ifp_detached(const struct in_multi *inm) 205 { 206 struct ifnet *ifp; 207 208 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 209 ifp = inm->inm_ifma->ifma_ifp; 210 if (ifp != NULL) { 211 /* 212 * Sanity check that netinet's notion of ifp is the 213 * same as net's. 214 */ 215 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 216 } 217 218 return (ifp == NULL); 219 } 220 #endif 221 222 /* 223 * Interface detach can happen in a taskqueue thread context, so we must use a 224 * dedicated thread to avoid deadlocks when draining inm_release tasks. 225 */ 226 TASKQUEUE_DEFINE_THREAD(inm_free); 227 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER(); 228 static void inm_release_task(void *arg __unused, int pending __unused); 229 static struct task inm_free_task = TASK_INITIALIZER(0, inm_release_task, NULL); 230 231 void 232 inm_release_wait(void *arg __unused) 233 { 234 235 /* 236 * Make sure all pending multicast addresses are freed before 237 * the VNET or network device is destroyed: 238 */ 239 taskqueue_drain(taskqueue_inm_free, &inm_free_task); 240 } 241 #ifdef VIMAGE 242 /* XXX-BZ FIXME, see D24914. */ 243 VNET_SYSUNINIT(inm_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, inm_release_wait, NULL); 244 #endif 245 246 void 247 inm_release_list_deferred(struct in_multi_head *inmh) 248 { 249 250 if (SLIST_EMPTY(inmh)) 251 return; 252 mtx_lock(&in_multi_free_mtx); 253 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele); 254 mtx_unlock(&in_multi_free_mtx); 255 taskqueue_enqueue(taskqueue_inm_free, &inm_free_task); 256 } 257 258 void 259 inm_disconnect(struct in_multi *inm) 260 { 261 struct ifnet *ifp; 262 struct ifmultiaddr *ifma, *ll_ifma; 263 264 ifp = inm->inm_ifp; 265 IF_ADDR_WLOCK_ASSERT(ifp); 266 ifma = inm->inm_ifma; 267 268 if_ref(ifp); 269 if (ifma->ifma_flags & IFMA_F_ENQUEUED) { 270 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 271 ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 272 } 273 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname); 274 if ((ll_ifma = ifma->ifma_llifma) != NULL) { 275 MPASS(ifma != ll_ifma); 276 ifma->ifma_llifma = NULL; 277 MPASS(ll_ifma->ifma_llifma == NULL); 278 MPASS(ll_ifma->ifma_ifp == ifp); 279 if (--ll_ifma->ifma_refcount == 0) { 280 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) { 281 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link); 282 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 283 } 284 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname); 285 if_freemulti(ll_ifma); 286 ifma_restart = true; 287 } 288 } 289 } 290 291 void 292 inm_release_deferred(struct in_multi *inm) 293 { 294 struct in_multi_head tmp; 295 296 IN_MULTI_LIST_LOCK_ASSERT(); 297 MPASS(inm->inm_refcount > 0); 298 if (--inm->inm_refcount == 0) { 299 SLIST_INIT(&tmp); 300 inm_disconnect(inm); 301 inm->inm_ifma->ifma_protospec = NULL; 302 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele); 303 inm_release_list_deferred(&tmp); 304 } 305 } 306 307 static void 308 inm_release_task(void *arg __unused, int pending __unused) 309 { 310 struct in_multi_head inm_free_tmp; 311 struct in_multi *inm, *tinm; 312 313 SLIST_INIT(&inm_free_tmp); 314 mtx_lock(&in_multi_free_mtx); 315 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele); 316 mtx_unlock(&in_multi_free_mtx); 317 IN_MULTI_LOCK(); 318 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) { 319 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele); 320 MPASS(inm); 321 inm_release(inm); 322 } 323 IN_MULTI_UNLOCK(); 324 } 325 326 /* 327 * Initialize an in_mfilter structure to a known state at t0, t1 328 * with an empty source filter list. 329 */ 330 static __inline void 331 imf_init(struct in_mfilter *imf, const int st0, const int st1) 332 { 333 memset(imf, 0, sizeof(struct in_mfilter)); 334 RB_INIT(&imf->imf_sources); 335 imf->imf_st[0] = st0; 336 imf->imf_st[1] = st1; 337 } 338 339 struct in_mfilter * 340 ip_mfilter_alloc(const int mflags, const int st0, const int st1) 341 { 342 struct in_mfilter *imf; 343 344 imf = malloc(sizeof(*imf), M_INMFILTER, mflags); 345 if (imf != NULL) 346 imf_init(imf, st0, st1); 347 348 return (imf); 349 } 350 351 void 352 ip_mfilter_free(struct in_mfilter *imf) 353 { 354 355 imf_purge(imf); 356 free(imf, M_INMFILTER); 357 } 358 359 /* 360 * Function for looking up an in_multi record for an IPv4 multicast address 361 * on a given interface. ifp must be valid. If no record found, return NULL. 362 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held. 363 */ 364 struct in_multi * 365 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina) 366 { 367 struct ifmultiaddr *ifma; 368 struct in_multi *inm; 369 370 IN_MULTI_LIST_LOCK_ASSERT(); 371 IF_ADDR_LOCK_ASSERT(ifp); 372 373 inm = NULL; 374 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) { 375 if (ifma->ifma_addr->sa_family != AF_INET || 376 ifma->ifma_protospec == NULL) 377 continue; 378 inm = (struct in_multi *)ifma->ifma_protospec; 379 if (inm->inm_addr.s_addr == ina.s_addr) 380 break; 381 inm = NULL; 382 } 383 return (inm); 384 } 385 386 /* 387 * Wrapper for inm_lookup_locked(). 388 * The IF_ADDR_LOCK will be taken on ifp and released on return. 389 */ 390 struct in_multi * 391 inm_lookup(struct ifnet *ifp, const struct in_addr ina) 392 { 393 struct epoch_tracker et; 394 struct in_multi *inm; 395 396 IN_MULTI_LIST_LOCK_ASSERT(); 397 NET_EPOCH_ENTER(et); 398 399 inm = inm_lookup_locked(ifp, ina); 400 NET_EPOCH_EXIT(et); 401 402 return (inm); 403 } 404 405 /* 406 * Find an IPv4 multicast group entry for this ip_moptions instance 407 * which matches the specified group, and optionally an interface. 408 * Return its index into the array, or -1 if not found. 409 */ 410 static struct in_mfilter * 411 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp, 412 const struct sockaddr *group) 413 { 414 const struct sockaddr_in *gsin; 415 struct in_mfilter *imf; 416 struct in_multi *inm; 417 418 gsin = (const struct sockaddr_in *)group; 419 420 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 421 inm = imf->imf_inm; 422 if (inm == NULL) 423 continue; 424 if ((ifp == NULL || (inm->inm_ifp == ifp)) && 425 in_hosteq(inm->inm_addr, gsin->sin_addr)) { 426 break; 427 } 428 } 429 return (imf); 430 } 431 432 /* 433 * Find an IPv4 multicast source entry for this imo which matches 434 * the given group index for this socket, and source address. 435 * 436 * NOTE: This does not check if the entry is in-mode, merely if 437 * it exists, which may not be the desired behaviour. 438 */ 439 static struct in_msource * 440 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src) 441 { 442 struct ip_msource find; 443 struct ip_msource *ims; 444 const sockunion_t *psa; 445 446 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__)); 447 448 /* Source trees are keyed in host byte order. */ 449 psa = (const sockunion_t *)src; 450 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr); 451 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 452 453 return ((struct in_msource *)ims); 454 } 455 456 /* 457 * Perform filtering for multicast datagrams on a socket by group and source. 458 * 459 * Returns 0 if a datagram should be allowed through, or various error codes 460 * if the socket was not a member of the group, or the source was muted, etc. 461 */ 462 int 463 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp, 464 const struct sockaddr *group, const struct sockaddr *src) 465 { 466 struct in_mfilter *imf; 467 struct in_msource *ims; 468 int mode; 469 470 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 471 472 imf = imo_match_group(imo, ifp, group); 473 if (imf == NULL) 474 return (MCAST_NOTGMEMBER); 475 476 /* 477 * Check if the source was included in an (S,G) join. 478 * Allow reception on exclusive memberships by default, 479 * reject reception on inclusive memberships by default. 480 * Exclude source only if an in-mode exclude filter exists. 481 * Include source only if an in-mode include filter exists. 482 * NOTE: We are comparing group state here at IGMP t1 (now) 483 * with socket-layer t0 (since last downcall). 484 */ 485 mode = imf->imf_st[1]; 486 ims = imo_match_source(imf, src); 487 488 if ((ims == NULL && mode == MCAST_INCLUDE) || 489 (ims != NULL && ims->imsl_st[0] != mode)) 490 return (MCAST_NOTSMEMBER); 491 492 return (MCAST_PASS); 493 } 494 495 /* 496 * Find and return a reference to an in_multi record for (ifp, group), 497 * and bump its reference count. 498 * If one does not exist, try to allocate it, and update link-layer multicast 499 * filters on ifp to listen for group. 500 * Assumes the IN_MULTI lock is held across the call. 501 * Return 0 if successful, otherwise return an appropriate error code. 502 */ 503 static int 504 in_getmulti(struct ifnet *ifp, const struct in_addr *group, 505 struct in_multi **pinm) 506 { 507 struct sockaddr_in gsin; 508 struct ifmultiaddr *ifma; 509 struct in_ifinfo *ii; 510 struct in_multi *inm; 511 int error; 512 513 IN_MULTI_LOCK_ASSERT(); 514 515 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET]; 516 IN_MULTI_LIST_LOCK(); 517 inm = inm_lookup(ifp, *group); 518 if (inm != NULL) { 519 /* 520 * If we already joined this group, just bump the 521 * refcount and return it. 522 */ 523 KASSERT(inm->inm_refcount >= 1, 524 ("%s: bad refcount %d", __func__, inm->inm_refcount)); 525 inm_acquire_locked(inm); 526 *pinm = inm; 527 } 528 IN_MULTI_LIST_UNLOCK(); 529 if (inm != NULL) 530 return (0); 531 532 memset(&gsin, 0, sizeof(gsin)); 533 gsin.sin_family = AF_INET; 534 gsin.sin_len = sizeof(struct sockaddr_in); 535 gsin.sin_addr = *group; 536 537 /* 538 * Check if a link-layer group is already associated 539 * with this network-layer group on the given ifnet. 540 */ 541 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma); 542 if (error != 0) 543 return (error); 544 545 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */ 546 IN_MULTI_LIST_LOCK(); 547 IF_ADDR_WLOCK(ifp); 548 549 /* 550 * If something other than netinet is occupying the link-layer 551 * group, print a meaningful error message and back out of 552 * the allocation. 553 * Otherwise, bump the refcount on the existing network-layer 554 * group association and return it. 555 */ 556 if (ifma->ifma_protospec != NULL) { 557 inm = (struct in_multi *)ifma->ifma_protospec; 558 #ifdef INVARIANTS 559 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 560 __func__)); 561 KASSERT(ifma->ifma_addr->sa_family == AF_INET, 562 ("%s: ifma not AF_INET", __func__)); 563 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 564 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp || 565 !in_hosteq(inm->inm_addr, *group)) { 566 char addrbuf[INET_ADDRSTRLEN]; 567 568 panic("%s: ifma %p is inconsistent with %p (%s)", 569 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf)); 570 } 571 #endif 572 inm_acquire_locked(inm); 573 *pinm = inm; 574 goto out_locked; 575 } 576 577 IF_ADDR_WLOCK_ASSERT(ifp); 578 579 /* 580 * A new in_multi record is needed; allocate and initialize it. 581 * We DO NOT perform an IGMP join as the in_ layer may need to 582 * push an initial source list down to IGMP to support SSM. 583 * 584 * The initial source filter state is INCLUDE, {} as per the RFC. 585 */ 586 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO); 587 if (inm == NULL) { 588 IF_ADDR_WUNLOCK(ifp); 589 IN_MULTI_LIST_UNLOCK(); 590 if_delmulti_ifma(ifma); 591 return (ENOMEM); 592 } 593 inm->inm_addr = *group; 594 inm->inm_ifp = ifp; 595 inm->inm_igi = ii->ii_igmp; 596 inm->inm_ifma = ifma; 597 inm->inm_refcount = 1; 598 inm->inm_state = IGMP_NOT_MEMBER; 599 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES); 600 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED; 601 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 602 RB_INIT(&inm->inm_srcs); 603 604 ifma->ifma_protospec = inm; 605 606 *pinm = inm; 607 out_locked: 608 IF_ADDR_WUNLOCK(ifp); 609 IN_MULTI_LIST_UNLOCK(); 610 return (0); 611 } 612 613 /* 614 * Drop a reference to an in_multi record. 615 * 616 * If the refcount drops to 0, free the in_multi record and 617 * delete the underlying link-layer membership. 618 */ 619 static void 620 inm_release(struct in_multi *inm) 621 { 622 struct ifmultiaddr *ifma; 623 struct ifnet *ifp; 624 625 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount); 626 MPASS(inm->inm_refcount == 0); 627 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm); 628 629 ifma = inm->inm_ifma; 630 ifp = inm->inm_ifp; 631 632 /* XXX this access is not covered by IF_ADDR_LOCK */ 633 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma); 634 if (ifp != NULL) { 635 CURVNET_SET(ifp->if_vnet); 636 inm_purge(inm); 637 free(inm, M_IPMADDR); 638 if_delmulti_ifma_flags(ifma, 1); 639 CURVNET_RESTORE(); 640 if_rele(ifp); 641 } else { 642 inm_purge(inm); 643 free(inm, M_IPMADDR); 644 if_delmulti_ifma_flags(ifma, 1); 645 } 646 } 647 648 /* 649 * Clear recorded source entries for a group. 650 * Used by the IGMP code. Caller must hold the IN_MULTI lock. 651 * FIXME: Should reap. 652 */ 653 void 654 inm_clear_recorded(struct in_multi *inm) 655 { 656 struct ip_msource *ims; 657 658 IN_MULTI_LIST_LOCK_ASSERT(); 659 660 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 661 if (ims->ims_stp) { 662 ims->ims_stp = 0; 663 --inm->inm_st[1].iss_rec; 664 } 665 } 666 KASSERT(inm->inm_st[1].iss_rec == 0, 667 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec)); 668 } 669 670 /* 671 * Record a source as pending for a Source-Group IGMPv3 query. 672 * This lives here as it modifies the shared tree. 673 * 674 * inm is the group descriptor. 675 * naddr is the address of the source to record in network-byte order. 676 * 677 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will 678 * lazy-allocate a source node in response to an SG query. 679 * Otherwise, no allocation is performed. This saves some memory 680 * with the trade-off that the source will not be reported to the 681 * router if joined in the window between the query response and 682 * the group actually being joined on the local host. 683 * 684 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed. 685 * This turns off the allocation of a recorded source entry if 686 * the group has not been joined. 687 * 688 * Return 0 if the source didn't exist or was already marked as recorded. 689 * Return 1 if the source was marked as recorded by this function. 690 * Return <0 if any error occurred (negated errno code). 691 */ 692 int 693 inm_record_source(struct in_multi *inm, const in_addr_t naddr) 694 { 695 struct ip_msource find; 696 struct ip_msource *ims, *nims; 697 698 IN_MULTI_LIST_LOCK_ASSERT(); 699 700 find.ims_haddr = ntohl(naddr); 701 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 702 if (ims && ims->ims_stp) 703 return (0); 704 if (ims == NULL) { 705 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 706 return (-ENOSPC); 707 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 708 M_NOWAIT | M_ZERO); 709 if (nims == NULL) 710 return (-ENOMEM); 711 nims->ims_haddr = find.ims_haddr; 712 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 713 ++inm->inm_nsrc; 714 ims = nims; 715 } 716 717 /* 718 * Mark the source as recorded and update the recorded 719 * source count. 720 */ 721 ++ims->ims_stp; 722 ++inm->inm_st[1].iss_rec; 723 724 return (1); 725 } 726 727 /* 728 * Return a pointer to an in_msource owned by an in_mfilter, 729 * given its source address. 730 * Lazy-allocate if needed. If this is a new entry its filter state is 731 * undefined at t0. 732 * 733 * imf is the filter set being modified. 734 * haddr is the source address in *host* byte-order. 735 * 736 * SMPng: May be called with locks held; malloc must not block. 737 */ 738 static int 739 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin, 740 struct in_msource **plims) 741 { 742 struct ip_msource find; 743 struct ip_msource *ims, *nims; 744 struct in_msource *lims; 745 int error; 746 747 error = 0; 748 ims = NULL; 749 lims = NULL; 750 751 /* key is host byte order */ 752 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 753 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 754 lims = (struct in_msource *)ims; 755 if (lims == NULL) { 756 if (imf->imf_nsrc == in_mcast_maxsocksrc) 757 return (ENOSPC); 758 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 759 M_NOWAIT | M_ZERO); 760 if (nims == NULL) 761 return (ENOMEM); 762 lims = (struct in_msource *)nims; 763 lims->ims_haddr = find.ims_haddr; 764 lims->imsl_st[0] = MCAST_UNDEFINED; 765 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 766 ++imf->imf_nsrc; 767 } 768 769 *plims = lims; 770 771 return (error); 772 } 773 774 /* 775 * Graft a source entry into an existing socket-layer filter set, 776 * maintaining any required invariants and checking allocations. 777 * 778 * The source is marked as being in the new filter mode at t1. 779 * 780 * Return the pointer to the new node, otherwise return NULL. 781 */ 782 static struct in_msource * 783 imf_graft(struct in_mfilter *imf, const uint8_t st1, 784 const struct sockaddr_in *psin) 785 { 786 struct ip_msource *nims; 787 struct in_msource *lims; 788 789 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 790 M_NOWAIT | M_ZERO); 791 if (nims == NULL) 792 return (NULL); 793 lims = (struct in_msource *)nims; 794 lims->ims_haddr = ntohl(psin->sin_addr.s_addr); 795 lims->imsl_st[0] = MCAST_UNDEFINED; 796 lims->imsl_st[1] = st1; 797 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 798 ++imf->imf_nsrc; 799 800 return (lims); 801 } 802 803 /* 804 * Prune a source entry from an existing socket-layer filter set, 805 * maintaining any required invariants and checking allocations. 806 * 807 * The source is marked as being left at t1, it is not freed. 808 * 809 * Return 0 if no error occurred, otherwise return an errno value. 810 */ 811 static int 812 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin) 813 { 814 struct ip_msource find; 815 struct ip_msource *ims; 816 struct in_msource *lims; 817 818 /* key is host byte order */ 819 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 820 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 821 if (ims == NULL) 822 return (ENOENT); 823 lims = (struct in_msource *)ims; 824 lims->imsl_st[1] = MCAST_UNDEFINED; 825 return (0); 826 } 827 828 /* 829 * Revert socket-layer filter set deltas at t1 to t0 state. 830 */ 831 static void 832 imf_rollback(struct in_mfilter *imf) 833 { 834 struct ip_msource *ims, *tims; 835 struct in_msource *lims; 836 837 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 838 lims = (struct in_msource *)ims; 839 if (lims->imsl_st[0] == lims->imsl_st[1]) { 840 /* no change at t1 */ 841 continue; 842 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) { 843 /* revert change to existing source at t1 */ 844 lims->imsl_st[1] = lims->imsl_st[0]; 845 } else { 846 /* revert source added t1 */ 847 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 848 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 849 free(ims, M_INMFILTER); 850 imf->imf_nsrc--; 851 } 852 } 853 imf->imf_st[1] = imf->imf_st[0]; 854 } 855 856 /* 857 * Mark socket-layer filter set as INCLUDE {} at t1. 858 */ 859 static void 860 imf_leave(struct in_mfilter *imf) 861 { 862 struct ip_msource *ims; 863 struct in_msource *lims; 864 865 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 866 lims = (struct in_msource *)ims; 867 lims->imsl_st[1] = MCAST_UNDEFINED; 868 } 869 imf->imf_st[1] = MCAST_INCLUDE; 870 } 871 872 /* 873 * Mark socket-layer filter set deltas as committed. 874 */ 875 static void 876 imf_commit(struct in_mfilter *imf) 877 { 878 struct ip_msource *ims; 879 struct in_msource *lims; 880 881 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 882 lims = (struct in_msource *)ims; 883 lims->imsl_st[0] = lims->imsl_st[1]; 884 } 885 imf->imf_st[0] = imf->imf_st[1]; 886 } 887 888 /* 889 * Reap unreferenced sources from socket-layer filter set. 890 */ 891 static void 892 imf_reap(struct in_mfilter *imf) 893 { 894 struct ip_msource *ims, *tims; 895 struct in_msource *lims; 896 897 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 898 lims = (struct in_msource *)ims; 899 if ((lims->imsl_st[0] == MCAST_UNDEFINED) && 900 (lims->imsl_st[1] == MCAST_UNDEFINED)) { 901 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims); 902 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 903 free(ims, M_INMFILTER); 904 imf->imf_nsrc--; 905 } 906 } 907 } 908 909 /* 910 * Purge socket-layer filter set. 911 */ 912 static void 913 imf_purge(struct in_mfilter *imf) 914 { 915 struct ip_msource *ims, *tims; 916 917 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 918 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 919 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 920 free(ims, M_INMFILTER); 921 imf->imf_nsrc--; 922 } 923 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED; 924 KASSERT(RB_EMPTY(&imf->imf_sources), 925 ("%s: imf_sources not empty", __func__)); 926 } 927 928 /* 929 * Look up a source filter entry for a multicast group. 930 * 931 * inm is the group descriptor to work with. 932 * haddr is the host-byte-order IPv4 address to look up. 933 * noalloc may be non-zero to suppress allocation of sources. 934 * *pims will be set to the address of the retrieved or allocated source. 935 * 936 * SMPng: NOTE: may be called with locks held. 937 * Return 0 if successful, otherwise return a non-zero error code. 938 */ 939 static int 940 inm_get_source(struct in_multi *inm, const in_addr_t haddr, 941 const int noalloc, struct ip_msource **pims) 942 { 943 struct ip_msource find; 944 struct ip_msource *ims, *nims; 945 946 find.ims_haddr = haddr; 947 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 948 if (ims == NULL && !noalloc) { 949 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 950 return (ENOSPC); 951 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 952 M_NOWAIT | M_ZERO); 953 if (nims == NULL) 954 return (ENOMEM); 955 nims->ims_haddr = haddr; 956 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 957 ++inm->inm_nsrc; 958 ims = nims; 959 #ifdef KTR 960 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__, 961 haddr, ims); 962 #endif 963 } 964 965 *pims = ims; 966 return (0); 967 } 968 969 /* 970 * Merge socket-layer source into IGMP-layer source. 971 * If rollback is non-zero, perform the inverse of the merge. 972 */ 973 static void 974 ims_merge(struct ip_msource *ims, const struct in_msource *lims, 975 const int rollback) 976 { 977 int n = rollback ? -1 : 1; 978 979 if (lims->imsl_st[0] == MCAST_EXCLUDE) { 980 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x", 981 __func__, n, ims->ims_haddr); 982 ims->ims_st[1].ex -= n; 983 } else if (lims->imsl_st[0] == MCAST_INCLUDE) { 984 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x", 985 __func__, n, ims->ims_haddr); 986 ims->ims_st[1].in -= n; 987 } 988 989 if (lims->imsl_st[1] == MCAST_EXCLUDE) { 990 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x", 991 __func__, n, ims->ims_haddr); 992 ims->ims_st[1].ex += n; 993 } else if (lims->imsl_st[1] == MCAST_INCLUDE) { 994 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x", 995 __func__, n, ims->ims_haddr); 996 ims->ims_st[1].in += n; 997 } 998 } 999 1000 /* 1001 * Atomically update the global in_multi state, when a membership's 1002 * filter list is being updated in any way. 1003 * 1004 * imf is the per-inpcb-membership group filter pointer. 1005 * A fake imf may be passed for in-kernel consumers. 1006 * 1007 * XXX This is a candidate for a set-symmetric-difference style loop 1008 * which would eliminate the repeated lookup from root of ims nodes, 1009 * as they share the same key space. 1010 * 1011 * If any error occurred this function will back out of refcounts 1012 * and return a non-zero value. 1013 */ 1014 static int 1015 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1016 { 1017 struct ip_msource *ims, *nims; 1018 struct in_msource *lims; 1019 int schanged, error; 1020 int nsrc0, nsrc1; 1021 1022 schanged = 0; 1023 error = 0; 1024 nsrc1 = nsrc0 = 0; 1025 IN_MULTI_LIST_LOCK_ASSERT(); 1026 1027 /* 1028 * Update the source filters first, as this may fail. 1029 * Maintain count of in-mode filters at t0, t1. These are 1030 * used to work out if we transition into ASM mode or not. 1031 * Maintain a count of source filters whose state was 1032 * actually modified by this operation. 1033 */ 1034 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1035 lims = (struct in_msource *)ims; 1036 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++; 1037 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++; 1038 if (lims->imsl_st[0] == lims->imsl_st[1]) continue; 1039 error = inm_get_source(inm, lims->ims_haddr, 0, &nims); 1040 ++schanged; 1041 if (error) 1042 break; 1043 ims_merge(nims, lims, 0); 1044 } 1045 if (error) { 1046 struct ip_msource *bims; 1047 1048 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) { 1049 lims = (struct in_msource *)ims; 1050 if (lims->imsl_st[0] == lims->imsl_st[1]) 1051 continue; 1052 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims); 1053 if (bims == NULL) 1054 continue; 1055 ims_merge(bims, lims, 1); 1056 } 1057 goto out_reap; 1058 } 1059 1060 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1", 1061 __func__, nsrc0, nsrc1); 1062 1063 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 1064 if (imf->imf_st[0] == imf->imf_st[1] && 1065 imf->imf_st[1] == MCAST_INCLUDE) { 1066 if (nsrc1 == 0) { 1067 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1068 --inm->inm_st[1].iss_in; 1069 } 1070 } 1071 1072 /* Handle filter mode transition on socket. */ 1073 if (imf->imf_st[0] != imf->imf_st[1]) { 1074 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d", 1075 __func__, imf->imf_st[0], imf->imf_st[1]); 1076 1077 if (imf->imf_st[0] == MCAST_EXCLUDE) { 1078 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__); 1079 --inm->inm_st[1].iss_ex; 1080 } else if (imf->imf_st[0] == MCAST_INCLUDE) { 1081 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1082 --inm->inm_st[1].iss_in; 1083 } 1084 1085 if (imf->imf_st[1] == MCAST_EXCLUDE) { 1086 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__); 1087 inm->inm_st[1].iss_ex++; 1088 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) { 1089 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__); 1090 inm->inm_st[1].iss_in++; 1091 } 1092 } 1093 1094 /* 1095 * Track inm filter state in terms of listener counts. 1096 * If there are any exclusive listeners, stack-wide 1097 * membership is exclusive. 1098 * Otherwise, if only inclusive listeners, stack-wide is inclusive. 1099 * If no listeners remain, state is undefined at t1, 1100 * and the IGMP lifecycle for this group should finish. 1101 */ 1102 if (inm->inm_st[1].iss_ex > 0) { 1103 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__); 1104 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE; 1105 } else if (inm->inm_st[1].iss_in > 0) { 1106 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__); 1107 inm->inm_st[1].iss_fmode = MCAST_INCLUDE; 1108 } else { 1109 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__); 1110 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 1111 } 1112 1113 /* Decrement ASM listener count on transition out of ASM mode. */ 1114 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1115 if ((imf->imf_st[1] != MCAST_EXCLUDE) || 1116 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) { 1117 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__); 1118 --inm->inm_st[1].iss_asm; 1119 } 1120 } 1121 1122 /* Increment ASM listener count on transition to ASM mode. */ 1123 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1124 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__); 1125 inm->inm_st[1].iss_asm++; 1126 } 1127 1128 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm); 1129 inm_print(inm); 1130 1131 out_reap: 1132 if (schanged > 0) { 1133 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__); 1134 inm_reap(inm); 1135 } 1136 return (error); 1137 } 1138 1139 /* 1140 * Mark an in_multi's filter set deltas as committed. 1141 * Called by IGMP after a state change has been enqueued. 1142 */ 1143 void 1144 inm_commit(struct in_multi *inm) 1145 { 1146 struct ip_msource *ims; 1147 1148 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm); 1149 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__); 1150 inm_print(inm); 1151 1152 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 1153 ims->ims_st[0] = ims->ims_st[1]; 1154 } 1155 inm->inm_st[0] = inm->inm_st[1]; 1156 } 1157 1158 /* 1159 * Reap unreferenced nodes from an in_multi's filter set. 1160 */ 1161 static void 1162 inm_reap(struct in_multi *inm) 1163 { 1164 struct ip_msource *ims, *tims; 1165 1166 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1167 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 || 1168 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 || 1169 ims->ims_stp != 0) 1170 continue; 1171 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1172 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1173 free(ims, M_IPMSOURCE); 1174 inm->inm_nsrc--; 1175 } 1176 } 1177 1178 /* 1179 * Purge all source nodes from an in_multi's filter set. 1180 */ 1181 static void 1182 inm_purge(struct in_multi *inm) 1183 { 1184 struct ip_msource *ims, *tims; 1185 1186 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1187 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1188 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1189 free(ims, M_IPMSOURCE); 1190 inm->inm_nsrc--; 1191 } 1192 } 1193 1194 /* 1195 * Join a multicast group; unlocked entry point. 1196 * 1197 * SMPng: XXX: in_joingroup() is called from in_control() when Giant 1198 * is not held. Fortunately, ifp is unlikely to have been detached 1199 * at this point, so we assume it's OK to recurse. 1200 */ 1201 int 1202 in_joingroup(struct ifnet *ifp, const struct in_addr *gina, 1203 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1204 { 1205 int error; 1206 1207 IN_MULTI_LOCK(); 1208 error = in_joingroup_locked(ifp, gina, imf, pinm); 1209 IN_MULTI_UNLOCK(); 1210 1211 return (error); 1212 } 1213 1214 /* 1215 * Join a multicast group; real entry point. 1216 * 1217 * Only preserves atomicity at inm level. 1218 * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1219 * 1220 * If the IGMP downcall fails, the group is not joined, and an error 1221 * code is returned. 1222 */ 1223 int 1224 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina, 1225 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1226 { 1227 struct in_mfilter timf; 1228 struct in_multi *inm; 1229 int error; 1230 1231 IN_MULTI_LOCK_ASSERT(); 1232 IN_MULTI_LIST_UNLOCK_ASSERT(); 1233 1234 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__, 1235 ntohl(gina->s_addr), ifp, ifp->if_xname); 1236 1237 error = 0; 1238 inm = NULL; 1239 1240 /* 1241 * If no imf was specified (i.e. kernel consumer), 1242 * fake one up and assume it is an ASM join. 1243 */ 1244 if (imf == NULL) { 1245 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1246 imf = &timf; 1247 } 1248 1249 error = in_getmulti(ifp, gina, &inm); 1250 if (error) { 1251 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__); 1252 return (error); 1253 } 1254 IN_MULTI_LIST_LOCK(); 1255 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1256 error = inm_merge(inm, imf); 1257 if (error) { 1258 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1259 goto out_inm_release; 1260 } 1261 1262 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1263 error = igmp_change_state(inm); 1264 if (error) { 1265 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__); 1266 goto out_inm_release; 1267 } 1268 1269 out_inm_release: 1270 if (error) { 1271 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1272 IF_ADDR_WLOCK(ifp); 1273 inm_release_deferred(inm); 1274 IF_ADDR_WUNLOCK(ifp); 1275 } else { 1276 *pinm = inm; 1277 } 1278 IN_MULTI_LIST_UNLOCK(); 1279 1280 return (error); 1281 } 1282 1283 /* 1284 * Leave a multicast group; unlocked entry point. 1285 */ 1286 int 1287 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1288 { 1289 int error; 1290 1291 IN_MULTI_LOCK(); 1292 error = in_leavegroup_locked(inm, imf); 1293 IN_MULTI_UNLOCK(); 1294 1295 return (error); 1296 } 1297 1298 /* 1299 * Leave a multicast group; real entry point. 1300 * All source filters will be expunged. 1301 * 1302 * Only preserves atomicity at inm level. 1303 * 1304 * Holding the write lock for the INP which contains imf 1305 * is highly advisable. We can't assert for it as imf does not 1306 * contain a back-pointer to the owning inp. 1307 * 1308 * Note: This is not the same as inm_release(*) as this function also 1309 * makes a state change downcall into IGMP. 1310 */ 1311 int 1312 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1313 { 1314 struct in_mfilter timf; 1315 int error; 1316 1317 IN_MULTI_LOCK_ASSERT(); 1318 IN_MULTI_LIST_UNLOCK_ASSERT(); 1319 1320 error = 0; 1321 1322 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__, 1323 inm, ntohl(inm->inm_addr.s_addr), 1324 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname), 1325 imf); 1326 1327 /* 1328 * If no imf was specified (i.e. kernel consumer), 1329 * fake one up and assume it is an ASM join. 1330 */ 1331 if (imf == NULL) { 1332 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1333 imf = &timf; 1334 } 1335 1336 /* 1337 * Begin state merge transaction at IGMP layer. 1338 * 1339 * As this particular invocation should not cause any memory 1340 * to be allocated, and there is no opportunity to roll back 1341 * the transaction, it MUST NOT fail. 1342 */ 1343 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1344 IN_MULTI_LIST_LOCK(); 1345 error = inm_merge(inm, imf); 1346 KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1347 1348 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1349 CURVNET_SET(inm->inm_ifp->if_vnet); 1350 error = igmp_change_state(inm); 1351 IF_ADDR_WLOCK(inm->inm_ifp); 1352 inm_release_deferred(inm); 1353 IF_ADDR_WUNLOCK(inm->inm_ifp); 1354 IN_MULTI_LIST_UNLOCK(); 1355 CURVNET_RESTORE(); 1356 if (error) 1357 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1358 1359 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1360 1361 return (error); 1362 } 1363 1364 /*#ifndef BURN_BRIDGES*/ 1365 1366 /* 1367 * Block or unblock an ASM multicast source on an inpcb. 1368 * This implements the delta-based API described in RFC 3678. 1369 * 1370 * The delta-based API applies only to exclusive-mode memberships. 1371 * An IGMP downcall will be performed. 1372 * 1373 * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1374 * 1375 * Return 0 if successful, otherwise return an appropriate error code. 1376 */ 1377 static int 1378 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1379 { 1380 struct group_source_req gsr; 1381 struct rm_priotracker in_ifa_tracker; 1382 sockunion_t *gsa, *ssa; 1383 struct ifnet *ifp; 1384 struct in_mfilter *imf; 1385 struct ip_moptions *imo; 1386 struct in_msource *ims; 1387 struct in_multi *inm; 1388 uint16_t fmode; 1389 int error, doblock; 1390 1391 ifp = NULL; 1392 error = 0; 1393 doblock = 0; 1394 1395 memset(&gsr, 0, sizeof(struct group_source_req)); 1396 gsa = (sockunion_t *)&gsr.gsr_group; 1397 ssa = (sockunion_t *)&gsr.gsr_source; 1398 1399 switch (sopt->sopt_name) { 1400 case IP_BLOCK_SOURCE: 1401 case IP_UNBLOCK_SOURCE: { 1402 struct ip_mreq_source mreqs; 1403 1404 error = sooptcopyin(sopt, &mreqs, 1405 sizeof(struct ip_mreq_source), 1406 sizeof(struct ip_mreq_source)); 1407 if (error) 1408 return (error); 1409 1410 gsa->sin.sin_family = AF_INET; 1411 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1412 gsa->sin.sin_addr = mreqs.imr_multiaddr; 1413 1414 ssa->sin.sin_family = AF_INET; 1415 ssa->sin.sin_len = sizeof(struct sockaddr_in); 1416 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1417 1418 if (!in_nullhost(mreqs.imr_interface)) { 1419 IN_IFADDR_RLOCK(&in_ifa_tracker); 1420 INADDR_TO_IFP(mreqs.imr_interface, ifp); 1421 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 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 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1448 return (EADDRNOTAVAIL); 1449 1450 ifp = ifnet_byindex(gsr.gsr_interface); 1451 1452 if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1453 doblock = 1; 1454 break; 1455 1456 default: 1457 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1458 __func__, sopt->sopt_name); 1459 return (EOPNOTSUPP); 1460 break; 1461 } 1462 1463 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1464 return (EINVAL); 1465 1466 IN_MULTI_LOCK(); 1467 1468 /* 1469 * Check if we are actually a member of this group. 1470 */ 1471 imo = inp_findmoptions(inp); 1472 imf = imo_match_group(imo, ifp, &gsa->sa); 1473 if (imf == NULL) { 1474 error = EADDRNOTAVAIL; 1475 goto out_inp_locked; 1476 } 1477 inm = imf->imf_inm; 1478 1479 /* 1480 * Attempting to use the delta-based API on an 1481 * non exclusive-mode membership is an error. 1482 */ 1483 fmode = imf->imf_st[0]; 1484 if (fmode != MCAST_EXCLUDE) { 1485 error = EINVAL; 1486 goto out_inp_locked; 1487 } 1488 1489 /* 1490 * Deal with error cases up-front: 1491 * Asked to block, but already blocked; or 1492 * Asked to unblock, but nothing to unblock. 1493 * If adding a new block entry, allocate it. 1494 */ 1495 ims = imo_match_source(imf, &ssa->sa); 1496 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1497 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__, 1498 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not "); 1499 error = EADDRNOTAVAIL; 1500 goto out_inp_locked; 1501 } 1502 1503 INP_WLOCK_ASSERT(inp); 1504 1505 /* 1506 * Begin state merge transaction at socket layer. 1507 */ 1508 if (doblock) { 1509 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 1510 ims = imf_graft(imf, fmode, &ssa->sin); 1511 if (ims == NULL) 1512 error = ENOMEM; 1513 } else { 1514 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 1515 error = imf_prune(imf, &ssa->sin); 1516 } 1517 1518 if (error) { 1519 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__); 1520 goto out_imf_rollback; 1521 } 1522 1523 /* 1524 * Begin state merge transaction at IGMP layer. 1525 */ 1526 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1527 IN_MULTI_LIST_LOCK(); 1528 error = inm_merge(inm, imf); 1529 if (error) { 1530 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1531 IN_MULTI_LIST_UNLOCK(); 1532 goto out_imf_rollback; 1533 } 1534 1535 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1536 error = igmp_change_state(inm); 1537 IN_MULTI_LIST_UNLOCK(); 1538 if (error) 1539 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1540 1541 out_imf_rollback: 1542 if (error) 1543 imf_rollback(imf); 1544 else 1545 imf_commit(imf); 1546 1547 imf_reap(imf); 1548 1549 out_inp_locked: 1550 INP_WUNLOCK(inp); 1551 IN_MULTI_UNLOCK(); 1552 return (error); 1553 } 1554 1555 /* 1556 * Given an inpcb, return its multicast options structure pointer. Accepts 1557 * an unlocked inpcb pointer, but will return it locked. May sleep. 1558 * 1559 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1560 * SMPng: NOTE: Returns with the INP write lock held. 1561 */ 1562 static struct ip_moptions * 1563 inp_findmoptions(struct inpcb *inp) 1564 { 1565 struct ip_moptions *imo; 1566 1567 INP_WLOCK(inp); 1568 if (inp->inp_moptions != NULL) 1569 return (inp->inp_moptions); 1570 1571 INP_WUNLOCK(inp); 1572 1573 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK); 1574 1575 imo->imo_multicast_ifp = NULL; 1576 imo->imo_multicast_addr.s_addr = INADDR_ANY; 1577 imo->imo_multicast_vif = -1; 1578 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL; 1579 imo->imo_multicast_loop = in_mcast_loop; 1580 STAILQ_INIT(&imo->imo_head); 1581 1582 INP_WLOCK(inp); 1583 if (inp->inp_moptions != NULL) { 1584 free(imo, M_IPMOPTS); 1585 return (inp->inp_moptions); 1586 } 1587 inp->inp_moptions = imo; 1588 return (imo); 1589 } 1590 1591 static void 1592 inp_gcmoptions(struct ip_moptions *imo) 1593 { 1594 struct in_mfilter *imf; 1595 struct in_multi *inm; 1596 struct ifnet *ifp; 1597 1598 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) { 1599 ip_mfilter_remove(&imo->imo_head, imf); 1600 1601 imf_leave(imf); 1602 if ((inm = imf->imf_inm) != NULL) { 1603 if ((ifp = inm->inm_ifp) != NULL) { 1604 CURVNET_SET(ifp->if_vnet); 1605 (void)in_leavegroup(inm, imf); 1606 CURVNET_RESTORE(); 1607 } else { 1608 (void)in_leavegroup(inm, imf); 1609 } 1610 } 1611 ip_mfilter_free(imf); 1612 } 1613 free(imo, M_IPMOPTS); 1614 } 1615 1616 /* 1617 * Discard the IP multicast options (and source filters). To minimize 1618 * the amount of work done while holding locks such as the INP's 1619 * pcbinfo lock (which is used in the receive path), the free 1620 * operation is deferred to the epoch callback task. 1621 */ 1622 void 1623 inp_freemoptions(struct ip_moptions *imo) 1624 { 1625 if (imo == NULL) 1626 return; 1627 inp_gcmoptions(imo); 1628 } 1629 1630 /* 1631 * Atomically get source filters on a socket for an IPv4 multicast group. 1632 * Called with INP lock held; returns with lock released. 1633 */ 1634 static int 1635 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1636 { 1637 struct __msfilterreq msfr; 1638 sockunion_t *gsa; 1639 struct ifnet *ifp; 1640 struct ip_moptions *imo; 1641 struct in_mfilter *imf; 1642 struct ip_msource *ims; 1643 struct in_msource *lims; 1644 struct sockaddr_in *psin; 1645 struct sockaddr_storage *ptss; 1646 struct sockaddr_storage *tss; 1647 int error; 1648 size_t nsrcs, ncsrcs; 1649 1650 INP_WLOCK_ASSERT(inp); 1651 1652 imo = inp->inp_moptions; 1653 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__)); 1654 1655 INP_WUNLOCK(inp); 1656 1657 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1658 sizeof(struct __msfilterreq)); 1659 if (error) 1660 return (error); 1661 1662 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 1663 return (EINVAL); 1664 1665 ifp = ifnet_byindex(msfr.msfr_ifindex); 1666 if (ifp == NULL) 1667 return (EINVAL); 1668 1669 INP_WLOCK(inp); 1670 1671 /* 1672 * Lookup group on the socket. 1673 */ 1674 gsa = (sockunion_t *)&msfr.msfr_group; 1675 imf = imo_match_group(imo, ifp, &gsa->sa); 1676 if (imf == NULL) { 1677 INP_WUNLOCK(inp); 1678 return (EADDRNOTAVAIL); 1679 } 1680 1681 /* 1682 * Ignore memberships which are in limbo. 1683 */ 1684 if (imf->imf_st[1] == MCAST_UNDEFINED) { 1685 INP_WUNLOCK(inp); 1686 return (EAGAIN); 1687 } 1688 msfr.msfr_fmode = imf->imf_st[1]; 1689 1690 /* 1691 * If the user specified a buffer, copy out the source filter 1692 * entries to userland gracefully. 1693 * We only copy out the number of entries which userland 1694 * has asked for, but we always tell userland how big the 1695 * buffer really needs to be. 1696 */ 1697 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 1698 msfr.msfr_nsrcs = in_mcast_maxsocksrc; 1699 tss = NULL; 1700 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1701 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1702 M_TEMP, M_NOWAIT | M_ZERO); 1703 if (tss == NULL) { 1704 INP_WUNLOCK(inp); 1705 return (ENOBUFS); 1706 } 1707 } 1708 1709 /* 1710 * Count number of sources in-mode at t0. 1711 * If buffer space exists and remains, copy out source entries. 1712 */ 1713 nsrcs = msfr.msfr_nsrcs; 1714 ncsrcs = 0; 1715 ptss = tss; 1716 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1717 lims = (struct in_msource *)ims; 1718 if (lims->imsl_st[0] == MCAST_UNDEFINED || 1719 lims->imsl_st[0] != imf->imf_st[0]) 1720 continue; 1721 ++ncsrcs; 1722 if (tss != NULL && nsrcs > 0) { 1723 psin = (struct sockaddr_in *)ptss; 1724 psin->sin_family = AF_INET; 1725 psin->sin_len = sizeof(struct sockaddr_in); 1726 psin->sin_addr.s_addr = htonl(lims->ims_haddr); 1727 psin->sin_port = 0; 1728 ++ptss; 1729 --nsrcs; 1730 } 1731 } 1732 1733 INP_WUNLOCK(inp); 1734 1735 if (tss != NULL) { 1736 error = copyout(tss, msfr.msfr_srcs, 1737 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1738 free(tss, M_TEMP); 1739 if (error) 1740 return (error); 1741 } 1742 1743 msfr.msfr_nsrcs = ncsrcs; 1744 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1745 1746 return (error); 1747 } 1748 1749 /* 1750 * Return the IP multicast options in response to user getsockopt(). 1751 */ 1752 int 1753 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1754 { 1755 struct rm_priotracker in_ifa_tracker; 1756 struct ip_mreqn mreqn; 1757 struct ip_moptions *imo; 1758 struct ifnet *ifp; 1759 struct in_ifaddr *ia; 1760 int error, optval; 1761 u_char coptval; 1762 1763 INP_WLOCK(inp); 1764 imo = inp->inp_moptions; 1765 /* 1766 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 1767 * or is a divert socket, reject it. 1768 */ 1769 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 1770 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1771 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) { 1772 INP_WUNLOCK(inp); 1773 return (EOPNOTSUPP); 1774 } 1775 1776 error = 0; 1777 switch (sopt->sopt_name) { 1778 case IP_MULTICAST_VIF: 1779 if (imo != NULL) 1780 optval = imo->imo_multicast_vif; 1781 else 1782 optval = -1; 1783 INP_WUNLOCK(inp); 1784 error = sooptcopyout(sopt, &optval, sizeof(int)); 1785 break; 1786 1787 case IP_MULTICAST_IF: 1788 memset(&mreqn, 0, sizeof(struct ip_mreqn)); 1789 if (imo != NULL) { 1790 ifp = imo->imo_multicast_ifp; 1791 if (!in_nullhost(imo->imo_multicast_addr)) { 1792 mreqn.imr_address = imo->imo_multicast_addr; 1793 } else if (ifp != NULL) { 1794 struct epoch_tracker et; 1795 1796 mreqn.imr_ifindex = ifp->if_index; 1797 NET_EPOCH_ENTER(et); 1798 IFP_TO_IA(ifp, ia, &in_ifa_tracker); 1799 if (ia != NULL) 1800 mreqn.imr_address = 1801 IA_SIN(ia)->sin_addr; 1802 NET_EPOCH_EXIT(et); 1803 } 1804 } 1805 INP_WUNLOCK(inp); 1806 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 1807 error = sooptcopyout(sopt, &mreqn, 1808 sizeof(struct ip_mreqn)); 1809 } else { 1810 error = sooptcopyout(sopt, &mreqn.imr_address, 1811 sizeof(struct in_addr)); 1812 } 1813 break; 1814 1815 case IP_MULTICAST_TTL: 1816 if (imo == NULL) 1817 optval = coptval = IP_DEFAULT_MULTICAST_TTL; 1818 else 1819 optval = coptval = imo->imo_multicast_ttl; 1820 INP_WUNLOCK(inp); 1821 if (sopt->sopt_valsize == sizeof(u_char)) 1822 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1823 else 1824 error = sooptcopyout(sopt, &optval, sizeof(int)); 1825 break; 1826 1827 case IP_MULTICAST_LOOP: 1828 if (imo == NULL) 1829 optval = coptval = IP_DEFAULT_MULTICAST_LOOP; 1830 else 1831 optval = coptval = imo->imo_multicast_loop; 1832 INP_WUNLOCK(inp); 1833 if (sopt->sopt_valsize == sizeof(u_char)) 1834 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1835 else 1836 error = sooptcopyout(sopt, &optval, sizeof(int)); 1837 break; 1838 1839 case IP_MSFILTER: 1840 if (imo == NULL) { 1841 error = EADDRNOTAVAIL; 1842 INP_WUNLOCK(inp); 1843 } else { 1844 error = inp_get_source_filters(inp, sopt); 1845 } 1846 break; 1847 1848 default: 1849 INP_WUNLOCK(inp); 1850 error = ENOPROTOOPT; 1851 break; 1852 } 1853 1854 INP_UNLOCK_ASSERT(inp); 1855 1856 return (error); 1857 } 1858 1859 /* 1860 * Look up the ifnet to use for a multicast group membership, 1861 * given the IPv4 address of an interface, and the IPv4 group address. 1862 * 1863 * This routine exists to support legacy multicast applications 1864 * which do not understand that multicast memberships are scoped to 1865 * specific physical links in the networking stack, or which need 1866 * to join link-scope groups before IPv4 addresses are configured. 1867 * 1868 * Use this socket's current FIB number for any required FIB lookup. 1869 * If ina is INADDR_ANY, look up the group address in the unicast FIB, 1870 * and use its ifp; usually, this points to the default next-hop. 1871 * 1872 * If the FIB lookup fails, attempt to use the first non-loopback 1873 * interface with multicast capability in the system as a 1874 * last resort. The legacy IPv4 ASM API requires that we do 1875 * this in order to allow groups to be joined when the routing 1876 * table has not yet been populated during boot. 1877 * 1878 * Returns NULL if no ifp could be found, otherwise return referenced ifp. 1879 * 1880 * FUTURE: Implement IPv4 source-address selection. 1881 */ 1882 static struct ifnet * 1883 inp_lookup_mcast_ifp(const struct inpcb *inp, 1884 const struct sockaddr_in *gsin, const struct in_addr ina) 1885 { 1886 struct rm_priotracker in_ifa_tracker; 1887 struct ifnet *ifp; 1888 struct nhop_object *nh; 1889 1890 KASSERT(inp != NULL, ("%s: inp must not be NULL", __func__)); 1891 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__)); 1892 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)), 1893 ("%s: not multicast", __func__)); 1894 1895 ifp = NULL; 1896 if (!in_nullhost(ina)) { 1897 IN_IFADDR_RLOCK(&in_ifa_tracker); 1898 INADDR_TO_IFP(ina, ifp); 1899 if (ifp != NULL) 1900 if_ref(ifp); 1901 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1902 } else { 1903 nh = fib4_lookup(inp->inp_inc.inc_fibnum, gsin->sin_addr, 0, NHR_NONE, 0); 1904 if (nh != NULL) { 1905 ifp = nh->nh_ifp; 1906 if_ref(ifp); 1907 } else { 1908 struct in_ifaddr *ia; 1909 struct ifnet *mifp; 1910 1911 mifp = NULL; 1912 IN_IFADDR_RLOCK(&in_ifa_tracker); 1913 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1914 mifp = ia->ia_ifp; 1915 if (!(mifp->if_flags & IFF_LOOPBACK) && 1916 (mifp->if_flags & IFF_MULTICAST)) { 1917 ifp = mifp; 1918 if_ref(ifp); 1919 break; 1920 } 1921 } 1922 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1923 } 1924 } 1925 1926 return (ifp); 1927 } 1928 1929 /* 1930 * Join an IPv4 multicast group, possibly with a source. 1931 */ 1932 static int 1933 inp_join_group(struct inpcb *inp, struct sockopt *sopt) 1934 { 1935 struct group_source_req gsr; 1936 sockunion_t *gsa, *ssa; 1937 struct ifnet *ifp; 1938 struct in_mfilter *imf; 1939 struct ip_moptions *imo; 1940 struct in_multi *inm; 1941 struct in_msource *lims; 1942 struct epoch_tracker et; 1943 int error, is_new; 1944 1945 ifp = NULL; 1946 lims = NULL; 1947 error = 0; 1948 1949 memset(&gsr, 0, sizeof(struct group_source_req)); 1950 gsa = (sockunion_t *)&gsr.gsr_group; 1951 gsa->ss.ss_family = AF_UNSPEC; 1952 ssa = (sockunion_t *)&gsr.gsr_source; 1953 ssa->ss.ss_family = AF_UNSPEC; 1954 1955 switch (sopt->sopt_name) { 1956 case IP_ADD_MEMBERSHIP: { 1957 struct ip_mreqn mreqn; 1958 1959 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) 1960 error = sooptcopyin(sopt, &mreqn, 1961 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn)); 1962 else 1963 error = sooptcopyin(sopt, &mreqn, 1964 sizeof(struct ip_mreq), sizeof(struct ip_mreq)); 1965 if (error) 1966 return (error); 1967 1968 gsa->sin.sin_family = AF_INET; 1969 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1970 gsa->sin.sin_addr = mreqn.imr_multiaddr; 1971 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1972 return (EINVAL); 1973 1974 NET_EPOCH_ENTER(et); 1975 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) && 1976 mreqn.imr_ifindex != 0) 1977 ifp = ifnet_byindex_ref(mreqn.imr_ifindex); 1978 else 1979 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 1980 mreqn.imr_address); 1981 NET_EPOCH_EXIT(et); 1982 break; 1983 } 1984 case IP_ADD_SOURCE_MEMBERSHIP: { 1985 struct ip_mreq_source mreqs; 1986 1987 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source), 1988 sizeof(struct ip_mreq_source)); 1989 if (error) 1990 return (error); 1991 1992 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET; 1993 gsa->sin.sin_len = ssa->sin.sin_len = 1994 sizeof(struct sockaddr_in); 1995 1996 gsa->sin.sin_addr = mreqs.imr_multiaddr; 1997 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1998 return (EINVAL); 1999 2000 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2001 2002 NET_EPOCH_ENTER(et); 2003 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 2004 mreqs.imr_interface); 2005 NET_EPOCH_EXIT(et); 2006 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2007 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2008 break; 2009 } 2010 2011 case MCAST_JOIN_GROUP: 2012 case MCAST_JOIN_SOURCE_GROUP: 2013 if (sopt->sopt_name == MCAST_JOIN_GROUP) { 2014 error = sooptcopyin(sopt, &gsr, 2015 sizeof(struct group_req), 2016 sizeof(struct group_req)); 2017 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2018 error = sooptcopyin(sopt, &gsr, 2019 sizeof(struct group_source_req), 2020 sizeof(struct group_source_req)); 2021 } 2022 if (error) 2023 return (error); 2024 2025 if (gsa->sin.sin_family != AF_INET || 2026 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2027 return (EINVAL); 2028 2029 /* 2030 * Overwrite the port field if present, as the sockaddr 2031 * being copied in may be matched with a binary comparison. 2032 */ 2033 gsa->sin.sin_port = 0; 2034 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2035 if (ssa->sin.sin_family != AF_INET || 2036 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2037 return (EINVAL); 2038 ssa->sin.sin_port = 0; 2039 } 2040 2041 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2042 return (EINVAL); 2043 2044 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2045 return (EADDRNOTAVAIL); 2046 NET_EPOCH_ENTER(et); 2047 ifp = ifnet_byindex_ref(gsr.gsr_interface); 2048 NET_EPOCH_EXIT(et); 2049 break; 2050 2051 default: 2052 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2053 __func__, sopt->sopt_name); 2054 return (EOPNOTSUPP); 2055 break; 2056 } 2057 2058 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) { 2059 if (ifp != NULL) 2060 if_rele(ifp); 2061 return (EADDRNOTAVAIL); 2062 } 2063 2064 IN_MULTI_LOCK(); 2065 2066 /* 2067 * Find the membership in the membership list. 2068 */ 2069 imo = inp_findmoptions(inp); 2070 imf = imo_match_group(imo, ifp, &gsa->sa); 2071 if (imf == NULL) { 2072 is_new = 1; 2073 inm = NULL; 2074 2075 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) { 2076 error = ENOMEM; 2077 goto out_inp_locked; 2078 } 2079 } else { 2080 is_new = 0; 2081 inm = imf->imf_inm; 2082 2083 if (ssa->ss.ss_family != AF_UNSPEC) { 2084 /* 2085 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership 2086 * is an error. On an existing inclusive membership, 2087 * it just adds the source to the filter list. 2088 */ 2089 if (imf->imf_st[1] != MCAST_INCLUDE) { 2090 error = EINVAL; 2091 goto out_inp_locked; 2092 } 2093 /* 2094 * Throw out duplicates. 2095 * 2096 * XXX FIXME: This makes a naive assumption that 2097 * even if entries exist for *ssa in this imf, 2098 * they will be rejected as dupes, even if they 2099 * are not valid in the current mode (in-mode). 2100 * 2101 * in_msource is transactioned just as for anything 2102 * else in SSM -- but note naive use of inm_graft() 2103 * below for allocating new filter entries. 2104 * 2105 * This is only an issue if someone mixes the 2106 * full-state SSM API with the delta-based API, 2107 * which is discouraged in the relevant RFCs. 2108 */ 2109 lims = imo_match_source(imf, &ssa->sa); 2110 if (lims != NULL /*&& 2111 lims->imsl_st[1] == MCAST_INCLUDE*/) { 2112 error = EADDRNOTAVAIL; 2113 goto out_inp_locked; 2114 } 2115 } else { 2116 /* 2117 * MCAST_JOIN_GROUP on an existing exclusive 2118 * membership is an error; return EADDRINUSE 2119 * to preserve 4.4BSD API idempotence, and 2120 * avoid tedious detour to code below. 2121 * NOTE: This is bending RFC 3678 a bit. 2122 * 2123 * On an existing inclusive membership, this is also 2124 * an error; if you want to change filter mode, 2125 * you must use the userland API setsourcefilter(). 2126 * XXX We don't reject this for imf in UNDEFINED 2127 * state at t1, because allocation of a filter 2128 * is atomic with allocation of a membership. 2129 */ 2130 error = EINVAL; 2131 if (imf->imf_st[1] == MCAST_EXCLUDE) 2132 error = EADDRINUSE; 2133 goto out_inp_locked; 2134 } 2135 } 2136 2137 /* 2138 * Begin state merge transaction at socket layer. 2139 */ 2140 INP_WLOCK_ASSERT(inp); 2141 2142 /* 2143 * Graft new source into filter list for this inpcb's 2144 * membership of the group. The in_multi may not have 2145 * been allocated yet if this is a new membership, however, 2146 * the in_mfilter slot will be allocated and must be initialized. 2147 * 2148 * Note: Grafting of exclusive mode filters doesn't happen 2149 * in this path. 2150 * XXX: Should check for non-NULL lims (node exists but may 2151 * not be in-mode) for interop with full-state API. 2152 */ 2153 if (ssa->ss.ss_family != AF_UNSPEC) { 2154 /* Membership starts in IN mode */ 2155 if (is_new) { 2156 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__); 2157 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE); 2158 if (imf == NULL) { 2159 error = ENOMEM; 2160 goto out_inp_locked; 2161 } 2162 } else { 2163 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 2164 } 2165 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin); 2166 if (lims == NULL) { 2167 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2168 __func__); 2169 error = ENOMEM; 2170 goto out_inp_locked; 2171 } 2172 } else { 2173 /* No address specified; Membership starts in EX mode */ 2174 if (is_new) { 2175 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__); 2176 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE); 2177 if (imf == NULL) { 2178 error = ENOMEM; 2179 goto out_inp_locked; 2180 } 2181 } 2182 } 2183 2184 /* 2185 * Begin state merge transaction at IGMP layer. 2186 */ 2187 if (is_new) { 2188 in_pcbref(inp); 2189 INP_WUNLOCK(inp); 2190 2191 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf, 2192 &imf->imf_inm); 2193 2194 INP_WLOCK(inp); 2195 if (in_pcbrele_wlocked(inp)) { 2196 error = ENXIO; 2197 goto out_inp_unlocked; 2198 } 2199 if (error) { 2200 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed", 2201 __func__); 2202 goto out_inp_locked; 2203 } 2204 /* 2205 * NOTE: Refcount from in_joingroup_locked() 2206 * is protecting membership. 2207 */ 2208 ip_mfilter_insert(&imo->imo_head, imf); 2209 } else { 2210 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2211 IN_MULTI_LIST_LOCK(); 2212 error = inm_merge(inm, imf); 2213 if (error) { 2214 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2215 __func__); 2216 IN_MULTI_LIST_UNLOCK(); 2217 imf_rollback(imf); 2218 imf_reap(imf); 2219 goto out_inp_locked; 2220 } 2221 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2222 error = igmp_change_state(inm); 2223 IN_MULTI_LIST_UNLOCK(); 2224 if (error) { 2225 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2226 __func__); 2227 imf_rollback(imf); 2228 imf_reap(imf); 2229 goto out_inp_locked; 2230 } 2231 } 2232 2233 imf_commit(imf); 2234 imf = NULL; 2235 2236 out_inp_locked: 2237 INP_WUNLOCK(inp); 2238 out_inp_unlocked: 2239 IN_MULTI_UNLOCK(); 2240 2241 if (is_new && imf) { 2242 if (imf->imf_inm != NULL) { 2243 IN_MULTI_LIST_LOCK(); 2244 IF_ADDR_WLOCK(ifp); 2245 inm_release_deferred(imf->imf_inm); 2246 IF_ADDR_WUNLOCK(ifp); 2247 IN_MULTI_LIST_UNLOCK(); 2248 } 2249 ip_mfilter_free(imf); 2250 } 2251 if_rele(ifp); 2252 return (error); 2253 } 2254 2255 /* 2256 * Leave an IPv4 multicast group on an inpcb, possibly with a source. 2257 */ 2258 static int 2259 inp_leave_group(struct inpcb *inp, struct sockopt *sopt) 2260 { 2261 struct group_source_req gsr; 2262 struct ip_mreq_source mreqs; 2263 struct rm_priotracker in_ifa_tracker; 2264 sockunion_t *gsa, *ssa; 2265 struct ifnet *ifp; 2266 struct in_mfilter *imf; 2267 struct ip_moptions *imo; 2268 struct in_msource *ims; 2269 struct in_multi *inm; 2270 int error; 2271 bool is_final; 2272 2273 ifp = NULL; 2274 error = 0; 2275 is_final = true; 2276 2277 memset(&gsr, 0, sizeof(struct group_source_req)); 2278 gsa = (sockunion_t *)&gsr.gsr_group; 2279 gsa->ss.ss_family = AF_UNSPEC; 2280 ssa = (sockunion_t *)&gsr.gsr_source; 2281 ssa->ss.ss_family = AF_UNSPEC; 2282 2283 switch (sopt->sopt_name) { 2284 case IP_DROP_MEMBERSHIP: 2285 case IP_DROP_SOURCE_MEMBERSHIP: 2286 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) { 2287 error = sooptcopyin(sopt, &mreqs, 2288 sizeof(struct ip_mreq), 2289 sizeof(struct ip_mreq)); 2290 /* 2291 * Swap interface and sourceaddr arguments, 2292 * as ip_mreq and ip_mreq_source are laid 2293 * out differently. 2294 */ 2295 mreqs.imr_interface = mreqs.imr_sourceaddr; 2296 mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 2297 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2298 error = sooptcopyin(sopt, &mreqs, 2299 sizeof(struct ip_mreq_source), 2300 sizeof(struct ip_mreq_source)); 2301 } 2302 if (error) 2303 return (error); 2304 2305 gsa->sin.sin_family = AF_INET; 2306 gsa->sin.sin_len = sizeof(struct sockaddr_in); 2307 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2308 2309 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2310 ssa->sin.sin_family = AF_INET; 2311 ssa->sin.sin_len = sizeof(struct sockaddr_in); 2312 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2313 } 2314 2315 /* 2316 * Attempt to look up hinted ifp from interface address. 2317 * Fallthrough with null ifp iff lookup fails, to 2318 * preserve 4.4BSD mcast API idempotence. 2319 * XXX NOTE WELL: The RFC 3678 API is preferred because 2320 * using an IPv4 address as a key is racy. 2321 */ 2322 if (!in_nullhost(mreqs.imr_interface)) { 2323 IN_IFADDR_RLOCK(&in_ifa_tracker); 2324 INADDR_TO_IFP(mreqs.imr_interface, ifp); 2325 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2326 } 2327 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2328 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2329 2330 break; 2331 2332 case MCAST_LEAVE_GROUP: 2333 case MCAST_LEAVE_SOURCE_GROUP: 2334 if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2335 error = sooptcopyin(sopt, &gsr, 2336 sizeof(struct group_req), 2337 sizeof(struct group_req)); 2338 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2339 error = sooptcopyin(sopt, &gsr, 2340 sizeof(struct group_source_req), 2341 sizeof(struct group_source_req)); 2342 } 2343 if (error) 2344 return (error); 2345 2346 if (gsa->sin.sin_family != AF_INET || 2347 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2348 return (EINVAL); 2349 2350 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2351 if (ssa->sin.sin_family != AF_INET || 2352 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2353 return (EINVAL); 2354 } 2355 2356 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2357 return (EADDRNOTAVAIL); 2358 2359 ifp = ifnet_byindex(gsr.gsr_interface); 2360 2361 if (ifp == NULL) 2362 return (EADDRNOTAVAIL); 2363 break; 2364 2365 default: 2366 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2367 __func__, sopt->sopt_name); 2368 return (EOPNOTSUPP); 2369 break; 2370 } 2371 2372 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2373 return (EINVAL); 2374 2375 IN_MULTI_LOCK(); 2376 2377 /* 2378 * Find the membership in the membership list. 2379 */ 2380 imo = inp_findmoptions(inp); 2381 imf = imo_match_group(imo, ifp, &gsa->sa); 2382 if (imf == NULL) { 2383 error = EADDRNOTAVAIL; 2384 goto out_inp_locked; 2385 } 2386 inm = imf->imf_inm; 2387 2388 if (ssa->ss.ss_family != AF_UNSPEC) 2389 is_final = false; 2390 2391 /* 2392 * Begin state merge transaction at socket layer. 2393 */ 2394 INP_WLOCK_ASSERT(inp); 2395 2396 /* 2397 * If we were instructed only to leave a given source, do so. 2398 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2399 */ 2400 if (is_final) { 2401 ip_mfilter_remove(&imo->imo_head, imf); 2402 imf_leave(imf); 2403 2404 /* 2405 * Give up the multicast address record to which 2406 * the membership points. 2407 */ 2408 (void) in_leavegroup_locked(imf->imf_inm, imf); 2409 } else { 2410 if (imf->imf_st[0] == MCAST_EXCLUDE) { 2411 error = EADDRNOTAVAIL; 2412 goto out_inp_locked; 2413 } 2414 ims = imo_match_source(imf, &ssa->sa); 2415 if (ims == NULL) { 2416 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", 2417 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not "); 2418 error = EADDRNOTAVAIL; 2419 goto out_inp_locked; 2420 } 2421 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 2422 error = imf_prune(imf, &ssa->sin); 2423 if (error) { 2424 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2425 __func__); 2426 goto out_inp_locked; 2427 } 2428 } 2429 2430 /* 2431 * Begin state merge transaction at IGMP layer. 2432 */ 2433 if (!is_final) { 2434 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2435 IN_MULTI_LIST_LOCK(); 2436 error = inm_merge(inm, imf); 2437 if (error) { 2438 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2439 __func__); 2440 IN_MULTI_LIST_UNLOCK(); 2441 imf_rollback(imf); 2442 imf_reap(imf); 2443 goto out_inp_locked; 2444 } 2445 2446 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2447 error = igmp_change_state(inm); 2448 IN_MULTI_LIST_UNLOCK(); 2449 if (error) { 2450 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2451 __func__); 2452 imf_rollback(imf); 2453 imf_reap(imf); 2454 goto out_inp_locked; 2455 } 2456 } 2457 imf_commit(imf); 2458 imf_reap(imf); 2459 2460 out_inp_locked: 2461 INP_WUNLOCK(inp); 2462 2463 if (is_final && imf) 2464 ip_mfilter_free(imf); 2465 2466 IN_MULTI_UNLOCK(); 2467 return (error); 2468 } 2469 2470 /* 2471 * Select the interface for transmitting IPv4 multicast datagrams. 2472 * 2473 * Either an instance of struct in_addr or an instance of struct ip_mreqn 2474 * may be passed to this socket option. An address of INADDR_ANY or an 2475 * interface index of 0 is used to remove a previous selection. 2476 * When no interface is selected, one is chosen for every send. 2477 */ 2478 static int 2479 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2480 { 2481 struct rm_priotracker in_ifa_tracker; 2482 struct in_addr addr; 2483 struct ip_mreqn mreqn; 2484 struct ifnet *ifp; 2485 struct ip_moptions *imo; 2486 int error; 2487 2488 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 2489 /* 2490 * An interface index was specified using the 2491 * Linux-derived ip_mreqn structure. 2492 */ 2493 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn), 2494 sizeof(struct ip_mreqn)); 2495 if (error) 2496 return (error); 2497 2498 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex) 2499 return (EINVAL); 2500 2501 if (mreqn.imr_ifindex == 0) { 2502 ifp = NULL; 2503 } else { 2504 ifp = ifnet_byindex(mreqn.imr_ifindex); 2505 if (ifp == NULL) 2506 return (EADDRNOTAVAIL); 2507 } 2508 } else { 2509 /* 2510 * An interface was specified by IPv4 address. 2511 * This is the traditional BSD usage. 2512 */ 2513 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr), 2514 sizeof(struct in_addr)); 2515 if (error) 2516 return (error); 2517 if (in_nullhost(addr)) { 2518 ifp = NULL; 2519 } else { 2520 IN_IFADDR_RLOCK(&in_ifa_tracker); 2521 INADDR_TO_IFP(addr, ifp); 2522 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2523 if (ifp == NULL) 2524 return (EADDRNOTAVAIL); 2525 } 2526 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp, 2527 ntohl(addr.s_addr)); 2528 } 2529 2530 /* Reject interfaces which do not support multicast. */ 2531 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0) 2532 return (EOPNOTSUPP); 2533 2534 imo = inp_findmoptions(inp); 2535 imo->imo_multicast_ifp = ifp; 2536 imo->imo_multicast_addr.s_addr = INADDR_ANY; 2537 INP_WUNLOCK(inp); 2538 2539 return (0); 2540 } 2541 2542 /* 2543 * Atomically set source filters on a socket for an IPv4 multicast group. 2544 * 2545 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2546 */ 2547 static int 2548 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2549 { 2550 struct __msfilterreq msfr; 2551 sockunion_t *gsa; 2552 struct ifnet *ifp; 2553 struct in_mfilter *imf; 2554 struct ip_moptions *imo; 2555 struct in_multi *inm; 2556 int error; 2557 2558 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2559 sizeof(struct __msfilterreq)); 2560 if (error) 2561 return (error); 2562 2563 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 2564 return (ENOBUFS); 2565 2566 if ((msfr.msfr_fmode != MCAST_EXCLUDE && 2567 msfr.msfr_fmode != MCAST_INCLUDE)) 2568 return (EINVAL); 2569 2570 if (msfr.msfr_group.ss_family != AF_INET || 2571 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in)) 2572 return (EINVAL); 2573 2574 gsa = (sockunion_t *)&msfr.msfr_group; 2575 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2576 return (EINVAL); 2577 2578 gsa->sin.sin_port = 0; /* ignore port */ 2579 2580 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 2581 return (EADDRNOTAVAIL); 2582 2583 ifp = ifnet_byindex(msfr.msfr_ifindex); 2584 if (ifp == NULL) 2585 return (EADDRNOTAVAIL); 2586 2587 IN_MULTI_LOCK(); 2588 2589 /* 2590 * Take the INP write lock. 2591 * Check if this socket is a member of this group. 2592 */ 2593 imo = inp_findmoptions(inp); 2594 imf = imo_match_group(imo, ifp, &gsa->sa); 2595 if (imf == NULL) { 2596 error = EADDRNOTAVAIL; 2597 goto out_inp_locked; 2598 } 2599 inm = imf->imf_inm; 2600 2601 /* 2602 * Begin state merge transaction at socket layer. 2603 */ 2604 INP_WLOCK_ASSERT(inp); 2605 2606 imf->imf_st[1] = msfr.msfr_fmode; 2607 2608 /* 2609 * Apply any new source filters, if present. 2610 * Make a copy of the user-space source vector so 2611 * that we may copy them with a single copyin. This 2612 * allows us to deal with page faults up-front. 2613 */ 2614 if (msfr.msfr_nsrcs > 0) { 2615 struct in_msource *lims; 2616 struct sockaddr_in *psin; 2617 struct sockaddr_storage *kss, *pkss; 2618 int i; 2619 2620 INP_WUNLOCK(inp); 2621 2622 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries", 2623 __func__, (unsigned long)msfr.msfr_nsrcs); 2624 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2625 M_TEMP, M_WAITOK); 2626 error = copyin(msfr.msfr_srcs, kss, 2627 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2628 if (error) { 2629 free(kss, M_TEMP); 2630 return (error); 2631 } 2632 2633 INP_WLOCK(inp); 2634 2635 /* 2636 * Mark all source filters as UNDEFINED at t1. 2637 * Restore new group filter mode, as imf_leave() 2638 * will set it to INCLUDE. 2639 */ 2640 imf_leave(imf); 2641 imf->imf_st[1] = msfr.msfr_fmode; 2642 2643 /* 2644 * Update socket layer filters at t1, lazy-allocating 2645 * new entries. This saves a bunch of memory at the 2646 * cost of one RB_FIND() per source entry; duplicate 2647 * entries in the msfr_nsrcs vector are ignored. 2648 * If we encounter an error, rollback transaction. 2649 * 2650 * XXX This too could be replaced with a set-symmetric 2651 * difference like loop to avoid walking from root 2652 * every time, as the key space is common. 2653 */ 2654 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2655 psin = (struct sockaddr_in *)pkss; 2656 if (psin->sin_family != AF_INET) { 2657 error = EAFNOSUPPORT; 2658 break; 2659 } 2660 if (psin->sin_len != sizeof(struct sockaddr_in)) { 2661 error = EINVAL; 2662 break; 2663 } 2664 error = imf_get_source(imf, psin, &lims); 2665 if (error) 2666 break; 2667 lims->imsl_st[1] = imf->imf_st[1]; 2668 } 2669 free(kss, M_TEMP); 2670 } 2671 2672 if (error) 2673 goto out_imf_rollback; 2674 2675 INP_WLOCK_ASSERT(inp); 2676 2677 /* 2678 * Begin state merge transaction at IGMP layer. 2679 */ 2680 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2681 IN_MULTI_LIST_LOCK(); 2682 error = inm_merge(inm, imf); 2683 if (error) { 2684 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 2685 IN_MULTI_LIST_UNLOCK(); 2686 goto out_imf_rollback; 2687 } 2688 2689 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2690 error = igmp_change_state(inm); 2691 IN_MULTI_LIST_UNLOCK(); 2692 if (error) 2693 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 2694 2695 out_imf_rollback: 2696 if (error) 2697 imf_rollback(imf); 2698 else 2699 imf_commit(imf); 2700 2701 imf_reap(imf); 2702 2703 out_inp_locked: 2704 INP_WUNLOCK(inp); 2705 IN_MULTI_UNLOCK(); 2706 return (error); 2707 } 2708 2709 /* 2710 * Set the IP multicast options in response to user setsockopt(). 2711 * 2712 * Many of the socket options handled in this function duplicate the 2713 * functionality of socket options in the regular unicast API. However, 2714 * it is not possible to merge the duplicate code, because the idempotence 2715 * of the IPv4 multicast part of the BSD Sockets API must be preserved; 2716 * the effects of these options must be treated as separate and distinct. 2717 * 2718 * SMPng: XXX: Unlocked read of inp_socket believed OK. 2719 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING 2720 * is refactored to no longer use vifs. 2721 */ 2722 int 2723 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2724 { 2725 struct ip_moptions *imo; 2726 int error; 2727 2728 error = 0; 2729 2730 /* 2731 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 2732 * or is a divert socket, reject it. 2733 */ 2734 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 2735 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2736 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) 2737 return (EOPNOTSUPP); 2738 2739 switch (sopt->sopt_name) { 2740 case IP_MULTICAST_VIF: { 2741 int vifi; 2742 /* 2743 * Select a multicast VIF for transmission. 2744 * Only useful if multicast forwarding is active. 2745 */ 2746 if (legal_vif_num == NULL) { 2747 error = EOPNOTSUPP; 2748 break; 2749 } 2750 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int)); 2751 if (error) 2752 break; 2753 if (!legal_vif_num(vifi) && (vifi != -1)) { 2754 error = EINVAL; 2755 break; 2756 } 2757 imo = inp_findmoptions(inp); 2758 imo->imo_multicast_vif = vifi; 2759 INP_WUNLOCK(inp); 2760 break; 2761 } 2762 2763 case IP_MULTICAST_IF: 2764 error = inp_set_multicast_if(inp, sopt); 2765 break; 2766 2767 case IP_MULTICAST_TTL: { 2768 u_char ttl; 2769 2770 /* 2771 * Set the IP time-to-live for outgoing multicast packets. 2772 * The original multicast API required a char argument, 2773 * which is inconsistent with the rest of the socket API. 2774 * We allow either a char or an int. 2775 */ 2776 if (sopt->sopt_valsize == sizeof(u_char)) { 2777 error = sooptcopyin(sopt, &ttl, sizeof(u_char), 2778 sizeof(u_char)); 2779 if (error) 2780 break; 2781 } else { 2782 u_int ittl; 2783 2784 error = sooptcopyin(sopt, &ittl, sizeof(u_int), 2785 sizeof(u_int)); 2786 if (error) 2787 break; 2788 if (ittl > 255) { 2789 error = EINVAL; 2790 break; 2791 } 2792 ttl = (u_char)ittl; 2793 } 2794 imo = inp_findmoptions(inp); 2795 imo->imo_multicast_ttl = ttl; 2796 INP_WUNLOCK(inp); 2797 break; 2798 } 2799 2800 case IP_MULTICAST_LOOP: { 2801 u_char loop; 2802 2803 /* 2804 * Set the loopback flag for outgoing multicast packets. 2805 * Must be zero or one. The original multicast API required a 2806 * char argument, which is inconsistent with the rest 2807 * of the socket API. We allow either a char or an int. 2808 */ 2809 if (sopt->sopt_valsize == sizeof(u_char)) { 2810 error = sooptcopyin(sopt, &loop, sizeof(u_char), 2811 sizeof(u_char)); 2812 if (error) 2813 break; 2814 } else { 2815 u_int iloop; 2816 2817 error = sooptcopyin(sopt, &iloop, sizeof(u_int), 2818 sizeof(u_int)); 2819 if (error) 2820 break; 2821 loop = (u_char)iloop; 2822 } 2823 imo = inp_findmoptions(inp); 2824 imo->imo_multicast_loop = !!loop; 2825 INP_WUNLOCK(inp); 2826 break; 2827 } 2828 2829 case IP_ADD_MEMBERSHIP: 2830 case IP_ADD_SOURCE_MEMBERSHIP: 2831 case MCAST_JOIN_GROUP: 2832 case MCAST_JOIN_SOURCE_GROUP: 2833 error = inp_join_group(inp, sopt); 2834 break; 2835 2836 case IP_DROP_MEMBERSHIP: 2837 case IP_DROP_SOURCE_MEMBERSHIP: 2838 case MCAST_LEAVE_GROUP: 2839 case MCAST_LEAVE_SOURCE_GROUP: 2840 error = inp_leave_group(inp, sopt); 2841 break; 2842 2843 case IP_BLOCK_SOURCE: 2844 case IP_UNBLOCK_SOURCE: 2845 case MCAST_BLOCK_SOURCE: 2846 case MCAST_UNBLOCK_SOURCE: 2847 error = inp_block_unblock_source(inp, sopt); 2848 break; 2849 2850 case IP_MSFILTER: 2851 error = inp_set_source_filters(inp, sopt); 2852 break; 2853 2854 default: 2855 error = EOPNOTSUPP; 2856 break; 2857 } 2858 2859 INP_UNLOCK_ASSERT(inp); 2860 2861 return (error); 2862 } 2863 2864 /* 2865 * Expose IGMP's multicast filter mode and source list(s) to userland, 2866 * keyed by (ifindex, group). 2867 * The filter mode is written out as a uint32_t, followed by 2868 * 0..n of struct in_addr. 2869 * For use by ifmcstat(8). 2870 * SMPng: NOTE: unlocked read of ifindex space. 2871 */ 2872 static int 2873 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS) 2874 { 2875 struct in_addr src, group; 2876 struct epoch_tracker et; 2877 struct ifnet *ifp; 2878 struct ifmultiaddr *ifma; 2879 struct in_multi *inm; 2880 struct ip_msource *ims; 2881 int *name; 2882 int retval; 2883 u_int namelen; 2884 uint32_t fmode, ifindex; 2885 2886 name = (int *)arg1; 2887 namelen = arg2; 2888 2889 if (req->newptr != NULL) 2890 return (EPERM); 2891 2892 if (namelen != 2) 2893 return (EINVAL); 2894 2895 ifindex = name[0]; 2896 if (ifindex <= 0 || ifindex > V_if_index) { 2897 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range", 2898 __func__, ifindex); 2899 return (ENOENT); 2900 } 2901 2902 group.s_addr = name[1]; 2903 if (!IN_MULTICAST(ntohl(group.s_addr))) { 2904 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast", 2905 __func__, ntohl(group.s_addr)); 2906 return (EINVAL); 2907 } 2908 2909 NET_EPOCH_ENTER(et); 2910 ifp = ifnet_byindex(ifindex); 2911 if (ifp == NULL) { 2912 NET_EPOCH_EXIT(et); 2913 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u", 2914 __func__, ifindex); 2915 return (ENOENT); 2916 } 2917 2918 retval = sysctl_wire_old_buffer(req, 2919 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr))); 2920 if (retval) { 2921 NET_EPOCH_EXIT(et); 2922 return (retval); 2923 } 2924 2925 IN_MULTI_LIST_LOCK(); 2926 2927 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2928 if (ifma->ifma_addr->sa_family != AF_INET || 2929 ifma->ifma_protospec == NULL) 2930 continue; 2931 inm = (struct in_multi *)ifma->ifma_protospec; 2932 if (!in_hosteq(inm->inm_addr, group)) 2933 continue; 2934 fmode = inm->inm_st[1].iss_fmode; 2935 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 2936 if (retval != 0) 2937 break; 2938 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 2939 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2940 ims->ims_haddr); 2941 /* 2942 * Only copy-out sources which are in-mode. 2943 */ 2944 if (fmode != ims_get_mode(inm, ims, 1)) { 2945 CTR1(KTR_IGMPV3, "%s: skip non-in-mode", 2946 __func__); 2947 continue; 2948 } 2949 src.s_addr = htonl(ims->ims_haddr); 2950 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr)); 2951 if (retval != 0) 2952 break; 2953 } 2954 } 2955 2956 IN_MULTI_LIST_UNLOCK(); 2957 NET_EPOCH_EXIT(et); 2958 2959 return (retval); 2960 } 2961 2962 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3) 2963 2964 static const char *inm_modestrs[] = { 2965 [MCAST_UNDEFINED] = "un", 2966 [MCAST_INCLUDE] = "in", 2967 [MCAST_EXCLUDE] = "ex", 2968 }; 2969 _Static_assert(MCAST_UNDEFINED == 0 && 2970 MCAST_EXCLUDE + 1 == nitems(inm_modestrs), 2971 "inm_modestrs: no longer matches #defines"); 2972 2973 static const char * 2974 inm_mode_str(const int mode) 2975 { 2976 2977 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 2978 return (inm_modestrs[mode]); 2979 return ("??"); 2980 } 2981 2982 static const char *inm_statestrs[] = { 2983 [IGMP_NOT_MEMBER] = "not-member", 2984 [IGMP_SILENT_MEMBER] = "silent", 2985 [IGMP_REPORTING_MEMBER] = "reporting", 2986 [IGMP_IDLE_MEMBER] = "idle", 2987 [IGMP_LAZY_MEMBER] = "lazy", 2988 [IGMP_SLEEPING_MEMBER] = "sleeping", 2989 [IGMP_AWAKENING_MEMBER] = "awakening", 2990 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending", 2991 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending", 2992 [IGMP_LEAVING_MEMBER] = "leaving", 2993 }; 2994 _Static_assert(IGMP_NOT_MEMBER == 0 && 2995 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs), 2996 "inm_statetrs: no longer matches #defines"); 2997 2998 static const char * 2999 inm_state_str(const int state) 3000 { 3001 3002 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER) 3003 return (inm_statestrs[state]); 3004 return ("??"); 3005 } 3006 3007 /* 3008 * Dump an in_multi structure to the console. 3009 */ 3010 void 3011 inm_print(const struct in_multi *inm) 3012 { 3013 int t; 3014 char addrbuf[INET_ADDRSTRLEN]; 3015 3016 if ((ktr_mask & KTR_IGMPV3) == 0) 3017 return; 3018 3019 printf("%s: --- begin inm %p ---\n", __func__, inm); 3020 printf("addr %s ifp %p(%s) ifma %p\n", 3021 inet_ntoa_r(inm->inm_addr, addrbuf), 3022 inm->inm_ifp, 3023 inm->inm_ifp->if_xname, 3024 inm->inm_ifma); 3025 printf("timer %u state %s refcount %u scq.len %u\n", 3026 inm->inm_timer, 3027 inm_state_str(inm->inm_state), 3028 inm->inm_refcount, 3029 inm->inm_scq.mq_len); 3030 printf("igi %p nsrc %lu sctimer %u scrv %u\n", 3031 inm->inm_igi, 3032 inm->inm_nsrc, 3033 inm->inm_sctimer, 3034 inm->inm_scrv); 3035 for (t = 0; t < 2; t++) { 3036 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 3037 inm_mode_str(inm->inm_st[t].iss_fmode), 3038 inm->inm_st[t].iss_asm, 3039 inm->inm_st[t].iss_ex, 3040 inm->inm_st[t].iss_in, 3041 inm->inm_st[t].iss_rec); 3042 } 3043 printf("%s: --- end inm %p ---\n", __func__, inm); 3044 } 3045 3046 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */ 3047 3048 void 3049 inm_print(const struct in_multi *inm) 3050 { 3051 3052 } 3053 3054 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */ 3055 3056 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp); 3057