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