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/vnet.h> 61 62 #include <net/ethernet.h> 63 64 #include <netinet/in.h> 65 #include <netinet/in_systm.h> 66 #include <netinet/in_fib.h> 67 #include <netinet/in_pcb.h> 68 #include <netinet/in_var.h> 69 #include <netinet/ip_var.h> 70 #include <netinet/igmp_var.h> 71 72 #ifndef KTR_IGMPV3 73 #define KTR_IGMPV3 KTR_INET 74 #endif 75 76 #ifndef __SOCKUNION_DECLARED 77 union sockunion { 78 struct sockaddr_storage ss; 79 struct sockaddr sa; 80 struct sockaddr_dl sdl; 81 struct sockaddr_in sin; 82 }; 83 typedef union sockunion sockunion_t; 84 #define __SOCKUNION_DECLARED 85 #endif /* __SOCKUNION_DECLARED */ 86 87 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter", 88 "IPv4 multicast PCB-layer source filter"); 89 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group"); 90 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options"); 91 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource", 92 "IPv4 multicast IGMP-layer source filter"); 93 94 /* 95 * Locking: 96 * 97 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK, 98 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 99 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however 100 * it can be taken by code in net/if.c also. 101 * - ip_moptions and in_mfilter are covered by the INP_WLOCK. 102 * 103 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly 104 * any need for in_multi itself to be virtualized -- it is bound to an ifp 105 * anyway no matter what happens. 106 */ 107 struct mtx in_multi_list_mtx; 108 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF); 109 110 struct mtx in_multi_free_mtx; 111 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF); 112 113 struct sx in_multi_sx; 114 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx"); 115 116 int ifma_restart; 117 118 /* 119 * Functions with non-static linkage defined in this file should be 120 * declared in in_var.h: 121 * imo_multi_filter() 122 * in_addmulti() 123 * in_delmulti() 124 * in_joingroup() 125 * in_joingroup_locked() 126 * in_leavegroup() 127 * in_leavegroup_locked() 128 * and ip_var.h: 129 * inp_freemoptions() 130 * inp_getmoptions() 131 * inp_setmoptions() 132 * 133 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti() 134 * and in_delmulti(). 135 */ 136 static void imf_commit(struct in_mfilter *); 137 static int imf_get_source(struct in_mfilter *imf, 138 const struct sockaddr_in *psin, 139 struct in_msource **); 140 static struct in_msource * 141 imf_graft(struct in_mfilter *, const uint8_t, 142 const struct sockaddr_in *); 143 static void imf_leave(struct in_mfilter *); 144 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *); 145 static void imf_purge(struct in_mfilter *); 146 static void imf_rollback(struct in_mfilter *); 147 static void imf_reap(struct in_mfilter *); 148 static struct in_mfilter * 149 imo_match_group(const struct ip_moptions *, 150 const struct ifnet *, const struct sockaddr *); 151 static struct in_msource * 152 imo_match_source(struct in_mfilter *, const struct sockaddr *); 153 static void ims_merge(struct ip_msource *ims, 154 const struct in_msource *lims, const int rollback); 155 static int in_getmulti(struct ifnet *, const struct in_addr *, 156 struct in_multi **); 157 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr, 158 const int noalloc, struct ip_msource **pims); 159 #ifdef KTR 160 static int inm_is_ifp_detached(const struct in_multi *); 161 #endif 162 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *); 163 static void inm_purge(struct in_multi *); 164 static void inm_reap(struct in_multi *); 165 static void inm_release(struct in_multi *); 166 static struct ip_moptions * 167 inp_findmoptions(struct inpcb *); 168 static int inp_get_source_filters(struct inpcb *, struct sockopt *); 169 static int inp_join_group(struct inpcb *, struct sockopt *); 170 static int inp_leave_group(struct inpcb *, struct sockopt *); 171 static struct ifnet * 172 inp_lookup_mcast_ifp(const struct inpcb *, 173 const struct sockaddr_in *, const struct in_addr); 174 static int inp_block_unblock_source(struct inpcb *, struct sockopt *); 175 static int inp_set_multicast_if(struct inpcb *, struct sockopt *); 176 static int inp_set_source_filters(struct inpcb *, struct sockopt *); 177 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS); 178 179 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, 180 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 181 "IPv4 multicast"); 182 183 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER; 184 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc, 185 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0, 186 "Max source filters per group"); 187 188 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER; 189 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc, 190 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0, 191 "Max source filters per socket"); 192 193 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP; 194 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN, 195 &in_mcast_loop, 0, "Loopback multicast datagrams by default"); 196 197 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters, 198 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters, 199 "Per-interface stack-wide source filters"); 200 201 #ifdef KTR 202 /* 203 * Inline function which wraps assertions for a valid ifp. 204 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 205 * is detached. 206 */ 207 static int __inline 208 inm_is_ifp_detached(const struct in_multi *inm) 209 { 210 struct ifnet *ifp; 211 212 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 213 ifp = inm->inm_ifma->ifma_ifp; 214 if (ifp != NULL) { 215 /* 216 * Sanity check that netinet's notion of ifp is the 217 * same as net's. 218 */ 219 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 220 } 221 222 return (ifp == NULL); 223 } 224 #endif 225 226 static struct task free_task; 227 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER(); 228 static void inm_release_task(void *arg __unused, int pending __unused); 229 230 static void 231 inm_init(void) 232 { 233 TASK_INIT(&free_task, 0, inm_release_task, NULL); 234 } 235 SYSINIT(inm_init, SI_SUB_TASKQ, SI_ORDER_ANY, inm_init, NULL); 236 237 void 238 inm_release_list_deferred(struct in_multi_head *inmh) 239 { 240 241 if (SLIST_EMPTY(inmh)) 242 return; 243 mtx_lock(&in_multi_free_mtx); 244 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele); 245 mtx_unlock(&in_multi_free_mtx); 246 taskqueue_enqueue(taskqueue_thread, &free_task); 247 } 248 249 void 250 inm_disconnect(struct in_multi *inm) 251 { 252 struct ifnet *ifp; 253 struct ifmultiaddr *ifma, *ll_ifma; 254 255 ifp = inm->inm_ifp; 256 IF_ADDR_WLOCK_ASSERT(ifp); 257 ifma = inm->inm_ifma; 258 259 if_ref(ifp); 260 if (ifma->ifma_flags & IFMA_F_ENQUEUED) { 261 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 262 ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 263 } 264 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname); 265 if ((ll_ifma = ifma->ifma_llifma) != NULL) { 266 MPASS(ifma != ll_ifma); 267 ifma->ifma_llifma = NULL; 268 MPASS(ll_ifma->ifma_llifma == NULL); 269 MPASS(ll_ifma->ifma_ifp == ifp); 270 if (--ll_ifma->ifma_refcount == 0) { 271 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) { 272 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link); 273 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 274 } 275 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname); 276 if_freemulti(ll_ifma); 277 ifma_restart = true; 278 } 279 } 280 } 281 282 void 283 inm_release_deferred(struct in_multi *inm) 284 { 285 struct in_multi_head tmp; 286 287 IN_MULTI_LIST_LOCK_ASSERT(); 288 MPASS(inm->inm_refcount > 0); 289 if (--inm->inm_refcount == 0) { 290 SLIST_INIT(&tmp); 291 inm_disconnect(inm); 292 inm->inm_ifma->ifma_protospec = NULL; 293 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele); 294 inm_release_list_deferred(&tmp); 295 } 296 } 297 298 static void 299 inm_release_task(void *arg __unused, int pending __unused) 300 { 301 struct in_multi_head inm_free_tmp; 302 struct in_multi *inm, *tinm; 303 304 SLIST_INIT(&inm_free_tmp); 305 mtx_lock(&in_multi_free_mtx); 306 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele); 307 mtx_unlock(&in_multi_free_mtx); 308 IN_MULTI_LOCK(); 309 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) { 310 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele); 311 MPASS(inm); 312 inm_release(inm); 313 } 314 IN_MULTI_UNLOCK(); 315 } 316 317 /* 318 * Initialize an in_mfilter structure to a known state at t0, t1 319 * with an empty source filter list. 320 */ 321 static __inline void 322 imf_init(struct in_mfilter *imf, const int st0, const int st1) 323 { 324 memset(imf, 0, sizeof(struct in_mfilter)); 325 RB_INIT(&imf->imf_sources); 326 imf->imf_st[0] = st0; 327 imf->imf_st[1] = st1; 328 } 329 330 struct in_mfilter * 331 ip_mfilter_alloc(const int mflags, const int st0, const int st1) 332 { 333 struct in_mfilter *imf; 334 335 imf = malloc(sizeof(*imf), M_INMFILTER, mflags); 336 if (imf != NULL) 337 imf_init(imf, st0, st1); 338 339 return (imf); 340 } 341 342 void 343 ip_mfilter_free(struct in_mfilter *imf) 344 { 345 346 imf_purge(imf); 347 free(imf, M_INMFILTER); 348 } 349 350 /* 351 * Function for looking up an in_multi record for an IPv4 multicast address 352 * on a given interface. ifp must be valid. If no record found, return NULL. 353 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held. 354 */ 355 struct in_multi * 356 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina) 357 { 358 struct ifmultiaddr *ifma; 359 struct in_multi *inm; 360 361 IN_MULTI_LIST_LOCK_ASSERT(); 362 IF_ADDR_LOCK_ASSERT(ifp); 363 364 inm = NULL; 365 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) { 366 if (ifma->ifma_addr->sa_family != AF_INET || 367 ifma->ifma_protospec == NULL) 368 continue; 369 inm = (struct in_multi *)ifma->ifma_protospec; 370 if (inm->inm_addr.s_addr == ina.s_addr) 371 break; 372 inm = NULL; 373 } 374 return (inm); 375 } 376 377 /* 378 * Wrapper for inm_lookup_locked(). 379 * The IF_ADDR_LOCK will be taken on ifp and released on return. 380 */ 381 struct in_multi * 382 inm_lookup(struct ifnet *ifp, const struct in_addr ina) 383 { 384 struct epoch_tracker et; 385 struct in_multi *inm; 386 387 IN_MULTI_LIST_LOCK_ASSERT(); 388 NET_EPOCH_ENTER(et); 389 390 inm = inm_lookup_locked(ifp, ina); 391 NET_EPOCH_EXIT(et); 392 393 return (inm); 394 } 395 396 /* 397 * Find an IPv4 multicast group entry for this ip_moptions instance 398 * which matches the specified group, and optionally an interface. 399 * Return its index into the array, or -1 if not found. 400 */ 401 static struct in_mfilter * 402 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp, 403 const struct sockaddr *group) 404 { 405 const struct sockaddr_in *gsin; 406 struct in_mfilter *imf; 407 struct in_multi *inm; 408 409 gsin = (const struct sockaddr_in *)group; 410 411 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 412 inm = imf->imf_inm; 413 if (inm == NULL) 414 continue; 415 if ((ifp == NULL || (inm->inm_ifp == ifp)) && 416 in_hosteq(inm->inm_addr, gsin->sin_addr)) { 417 break; 418 } 419 } 420 return (imf); 421 } 422 423 /* 424 * Find an IPv4 multicast source entry for this imo which matches 425 * the given group index for this socket, and source address. 426 * 427 * NOTE: This does not check if the entry is in-mode, merely if 428 * it exists, which may not be the desired behaviour. 429 */ 430 static struct in_msource * 431 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src) 432 { 433 struct ip_msource find; 434 struct ip_msource *ims; 435 const sockunion_t *psa; 436 437 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__)); 438 439 /* Source trees are keyed in host byte order. */ 440 psa = (const sockunion_t *)src; 441 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr); 442 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 443 444 return ((struct in_msource *)ims); 445 } 446 447 /* 448 * Perform filtering for multicast datagrams on a socket by group and source. 449 * 450 * Returns 0 if a datagram should be allowed through, or various error codes 451 * if the socket was not a member of the group, or the source was muted, etc. 452 */ 453 int 454 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp, 455 const struct sockaddr *group, const struct sockaddr *src) 456 { 457 struct in_mfilter *imf; 458 struct in_msource *ims; 459 int mode; 460 461 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 462 463 imf = imo_match_group(imo, ifp, group); 464 if (imf == NULL) 465 return (MCAST_NOTGMEMBER); 466 467 /* 468 * Check if the source was included in an (S,G) join. 469 * Allow reception on exclusive memberships by default, 470 * reject reception on inclusive memberships by default. 471 * Exclude source only if an in-mode exclude filter exists. 472 * Include source only if an in-mode include filter exists. 473 * NOTE: We are comparing group state here at IGMP t1 (now) 474 * with socket-layer t0 (since last downcall). 475 */ 476 mode = imf->imf_st[1]; 477 ims = imo_match_source(imf, src); 478 479 if ((ims == NULL && mode == MCAST_INCLUDE) || 480 (ims != NULL && ims->imsl_st[0] != mode)) 481 return (MCAST_NOTSMEMBER); 482 483 return (MCAST_PASS); 484 } 485 486 /* 487 * Find and return a reference to an in_multi record for (ifp, group), 488 * and bump its reference count. 489 * If one does not exist, try to allocate it, and update link-layer multicast 490 * filters on ifp to listen for group. 491 * Assumes the IN_MULTI lock is held across the call. 492 * Return 0 if successful, otherwise return an appropriate error code. 493 */ 494 static int 495 in_getmulti(struct ifnet *ifp, const struct in_addr *group, 496 struct in_multi **pinm) 497 { 498 struct sockaddr_in gsin; 499 struct ifmultiaddr *ifma; 500 struct in_ifinfo *ii; 501 struct in_multi *inm; 502 int error; 503 504 IN_MULTI_LOCK_ASSERT(); 505 506 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET]; 507 IN_MULTI_LIST_LOCK(); 508 inm = inm_lookup(ifp, *group); 509 if (inm != NULL) { 510 /* 511 * If we already joined this group, just bump the 512 * refcount and return it. 513 */ 514 KASSERT(inm->inm_refcount >= 1, 515 ("%s: bad refcount %d", __func__, inm->inm_refcount)); 516 inm_acquire_locked(inm); 517 *pinm = inm; 518 } 519 IN_MULTI_LIST_UNLOCK(); 520 if (inm != NULL) 521 return (0); 522 523 memset(&gsin, 0, sizeof(gsin)); 524 gsin.sin_family = AF_INET; 525 gsin.sin_len = sizeof(struct sockaddr_in); 526 gsin.sin_addr = *group; 527 528 /* 529 * Check if a link-layer group is already associated 530 * with this network-layer group on the given ifnet. 531 */ 532 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma); 533 if (error != 0) 534 return (error); 535 536 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */ 537 IN_MULTI_LIST_LOCK(); 538 IF_ADDR_WLOCK(ifp); 539 540 /* 541 * If something other than netinet is occupying the link-layer 542 * group, print a meaningful error message and back out of 543 * the allocation. 544 * Otherwise, bump the refcount on the existing network-layer 545 * group association and return it. 546 */ 547 if (ifma->ifma_protospec != NULL) { 548 inm = (struct in_multi *)ifma->ifma_protospec; 549 #ifdef INVARIANTS 550 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 551 __func__)); 552 KASSERT(ifma->ifma_addr->sa_family == AF_INET, 553 ("%s: ifma not AF_INET", __func__)); 554 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 555 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp || 556 !in_hosteq(inm->inm_addr, *group)) { 557 char addrbuf[INET_ADDRSTRLEN]; 558 559 panic("%s: ifma %p is inconsistent with %p (%s)", 560 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf)); 561 } 562 #endif 563 inm_acquire_locked(inm); 564 *pinm = inm; 565 goto out_locked; 566 } 567 568 IF_ADDR_WLOCK_ASSERT(ifp); 569 570 /* 571 * A new in_multi record is needed; allocate and initialize it. 572 * We DO NOT perform an IGMP join as the in_ layer may need to 573 * push an initial source list down to IGMP to support SSM. 574 * 575 * The initial source filter state is INCLUDE, {} as per the RFC. 576 */ 577 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO); 578 if (inm == NULL) { 579 IF_ADDR_WUNLOCK(ifp); 580 IN_MULTI_LIST_UNLOCK(); 581 if_delmulti_ifma(ifma); 582 return (ENOMEM); 583 } 584 inm->inm_addr = *group; 585 inm->inm_ifp = ifp; 586 inm->inm_igi = ii->ii_igmp; 587 inm->inm_ifma = ifma; 588 inm->inm_refcount = 1; 589 inm->inm_state = IGMP_NOT_MEMBER; 590 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES); 591 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED; 592 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 593 RB_INIT(&inm->inm_srcs); 594 595 ifma->ifma_protospec = inm; 596 597 *pinm = inm; 598 out_locked: 599 IF_ADDR_WUNLOCK(ifp); 600 IN_MULTI_LIST_UNLOCK(); 601 return (0); 602 } 603 604 /* 605 * Drop a reference to an in_multi record. 606 * 607 * If the refcount drops to 0, free the in_multi record and 608 * delete the underlying link-layer membership. 609 */ 610 static void 611 inm_release(struct in_multi *inm) 612 { 613 struct ifmultiaddr *ifma; 614 struct ifnet *ifp; 615 616 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount); 617 MPASS(inm->inm_refcount == 0); 618 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm); 619 620 ifma = inm->inm_ifma; 621 ifp = inm->inm_ifp; 622 623 /* XXX this access is not covered by IF_ADDR_LOCK */ 624 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma); 625 if (ifp != NULL) { 626 CURVNET_SET(ifp->if_vnet); 627 inm_purge(inm); 628 free(inm, M_IPMADDR); 629 if_delmulti_ifma_flags(ifma, 1); 630 CURVNET_RESTORE(); 631 if_rele(ifp); 632 } else { 633 inm_purge(inm); 634 free(inm, M_IPMADDR); 635 if_delmulti_ifma_flags(ifma, 1); 636 } 637 } 638 639 /* 640 * Clear recorded source entries for a group. 641 * Used by the IGMP code. Caller must hold the IN_MULTI lock. 642 * FIXME: Should reap. 643 */ 644 void 645 inm_clear_recorded(struct in_multi *inm) 646 { 647 struct ip_msource *ims; 648 649 IN_MULTI_LIST_LOCK_ASSERT(); 650 651 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 652 if (ims->ims_stp) { 653 ims->ims_stp = 0; 654 --inm->inm_st[1].iss_rec; 655 } 656 } 657 KASSERT(inm->inm_st[1].iss_rec == 0, 658 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec)); 659 } 660 661 /* 662 * Record a source as pending for a Source-Group IGMPv3 query. 663 * This lives here as it modifies the shared tree. 664 * 665 * inm is the group descriptor. 666 * naddr is the address of the source to record in network-byte order. 667 * 668 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will 669 * lazy-allocate a source node in response to an SG query. 670 * Otherwise, no allocation is performed. This saves some memory 671 * with the trade-off that the source will not be reported to the 672 * router if joined in the window between the query response and 673 * the group actually being joined on the local host. 674 * 675 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed. 676 * This turns off the allocation of a recorded source entry if 677 * the group has not been joined. 678 * 679 * Return 0 if the source didn't exist or was already marked as recorded. 680 * Return 1 if the source was marked as recorded by this function. 681 * Return <0 if any error occurred (negated errno code). 682 */ 683 int 684 inm_record_source(struct in_multi *inm, const in_addr_t naddr) 685 { 686 struct ip_msource find; 687 struct ip_msource *ims, *nims; 688 689 IN_MULTI_LIST_LOCK_ASSERT(); 690 691 find.ims_haddr = ntohl(naddr); 692 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 693 if (ims && ims->ims_stp) 694 return (0); 695 if (ims == NULL) { 696 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 697 return (-ENOSPC); 698 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 699 M_NOWAIT | M_ZERO); 700 if (nims == NULL) 701 return (-ENOMEM); 702 nims->ims_haddr = find.ims_haddr; 703 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 704 ++inm->inm_nsrc; 705 ims = nims; 706 } 707 708 /* 709 * Mark the source as recorded and update the recorded 710 * source count. 711 */ 712 ++ims->ims_stp; 713 ++inm->inm_st[1].iss_rec; 714 715 return (1); 716 } 717 718 /* 719 * Return a pointer to an in_msource owned by an in_mfilter, 720 * given its source address. 721 * Lazy-allocate if needed. If this is a new entry its filter state is 722 * undefined at t0. 723 * 724 * imf is the filter set being modified. 725 * haddr is the source address in *host* byte-order. 726 * 727 * SMPng: May be called with locks held; malloc must not block. 728 */ 729 static int 730 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin, 731 struct in_msource **plims) 732 { 733 struct ip_msource find; 734 struct ip_msource *ims, *nims; 735 struct in_msource *lims; 736 int error; 737 738 error = 0; 739 ims = NULL; 740 lims = NULL; 741 742 /* key is host byte order */ 743 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 744 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 745 lims = (struct in_msource *)ims; 746 if (lims == NULL) { 747 if (imf->imf_nsrc == in_mcast_maxsocksrc) 748 return (ENOSPC); 749 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 750 M_NOWAIT | M_ZERO); 751 if (nims == NULL) 752 return (ENOMEM); 753 lims = (struct in_msource *)nims; 754 lims->ims_haddr = find.ims_haddr; 755 lims->imsl_st[0] = MCAST_UNDEFINED; 756 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 757 ++imf->imf_nsrc; 758 } 759 760 *plims = lims; 761 762 return (error); 763 } 764 765 /* 766 * Graft a source entry into an existing socket-layer filter set, 767 * maintaining any required invariants and checking allocations. 768 * 769 * The source is marked as being in the new filter mode at t1. 770 * 771 * Return the pointer to the new node, otherwise return NULL. 772 */ 773 static struct in_msource * 774 imf_graft(struct in_mfilter *imf, const uint8_t st1, 775 const struct sockaddr_in *psin) 776 { 777 struct ip_msource *nims; 778 struct in_msource *lims; 779 780 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 781 M_NOWAIT | M_ZERO); 782 if (nims == NULL) 783 return (NULL); 784 lims = (struct in_msource *)nims; 785 lims->ims_haddr = ntohl(psin->sin_addr.s_addr); 786 lims->imsl_st[0] = MCAST_UNDEFINED; 787 lims->imsl_st[1] = st1; 788 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 789 ++imf->imf_nsrc; 790 791 return (lims); 792 } 793 794 /* 795 * Prune a source entry from an existing socket-layer filter set, 796 * maintaining any required invariants and checking allocations. 797 * 798 * The source is marked as being left at t1, it is not freed. 799 * 800 * Return 0 if no error occurred, otherwise return an errno value. 801 */ 802 static int 803 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin) 804 { 805 struct ip_msource find; 806 struct ip_msource *ims; 807 struct in_msource *lims; 808 809 /* key is host byte order */ 810 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 811 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 812 if (ims == NULL) 813 return (ENOENT); 814 lims = (struct in_msource *)ims; 815 lims->imsl_st[1] = MCAST_UNDEFINED; 816 return (0); 817 } 818 819 /* 820 * Revert socket-layer filter set deltas at t1 to t0 state. 821 */ 822 static void 823 imf_rollback(struct in_mfilter *imf) 824 { 825 struct ip_msource *ims, *tims; 826 struct in_msource *lims; 827 828 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 829 lims = (struct in_msource *)ims; 830 if (lims->imsl_st[0] == lims->imsl_st[1]) { 831 /* no change at t1 */ 832 continue; 833 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) { 834 /* revert change to existing source at t1 */ 835 lims->imsl_st[1] = lims->imsl_st[0]; 836 } else { 837 /* revert source added t1 */ 838 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 839 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 840 free(ims, M_INMFILTER); 841 imf->imf_nsrc--; 842 } 843 } 844 imf->imf_st[1] = imf->imf_st[0]; 845 } 846 847 /* 848 * Mark socket-layer filter set as INCLUDE {} at t1. 849 */ 850 static void 851 imf_leave(struct in_mfilter *imf) 852 { 853 struct ip_msource *ims; 854 struct in_msource *lims; 855 856 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 857 lims = (struct in_msource *)ims; 858 lims->imsl_st[1] = MCAST_UNDEFINED; 859 } 860 imf->imf_st[1] = MCAST_INCLUDE; 861 } 862 863 /* 864 * Mark socket-layer filter set deltas as committed. 865 */ 866 static void 867 imf_commit(struct in_mfilter *imf) 868 { 869 struct ip_msource *ims; 870 struct in_msource *lims; 871 872 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 873 lims = (struct in_msource *)ims; 874 lims->imsl_st[0] = lims->imsl_st[1]; 875 } 876 imf->imf_st[0] = imf->imf_st[1]; 877 } 878 879 /* 880 * Reap unreferenced sources from socket-layer filter set. 881 */ 882 static void 883 imf_reap(struct in_mfilter *imf) 884 { 885 struct ip_msource *ims, *tims; 886 struct in_msource *lims; 887 888 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 889 lims = (struct in_msource *)ims; 890 if ((lims->imsl_st[0] == MCAST_UNDEFINED) && 891 (lims->imsl_st[1] == MCAST_UNDEFINED)) { 892 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims); 893 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 894 free(ims, M_INMFILTER); 895 imf->imf_nsrc--; 896 } 897 } 898 } 899 900 /* 901 * Purge socket-layer filter set. 902 */ 903 static void 904 imf_purge(struct in_mfilter *imf) 905 { 906 struct ip_msource *ims, *tims; 907 908 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 909 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 910 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 911 free(ims, M_INMFILTER); 912 imf->imf_nsrc--; 913 } 914 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED; 915 KASSERT(RB_EMPTY(&imf->imf_sources), 916 ("%s: imf_sources not empty", __func__)); 917 } 918 919 /* 920 * Look up a source filter entry for a multicast group. 921 * 922 * inm is the group descriptor to work with. 923 * haddr is the host-byte-order IPv4 address to look up. 924 * noalloc may be non-zero to suppress allocation of sources. 925 * *pims will be set to the address of the retrieved or allocated source. 926 * 927 * SMPng: NOTE: may be called with locks held. 928 * Return 0 if successful, otherwise return a non-zero error code. 929 */ 930 static int 931 inm_get_source(struct in_multi *inm, const in_addr_t haddr, 932 const int noalloc, struct ip_msource **pims) 933 { 934 struct ip_msource find; 935 struct ip_msource *ims, *nims; 936 937 find.ims_haddr = haddr; 938 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 939 if (ims == NULL && !noalloc) { 940 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 941 return (ENOSPC); 942 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 943 M_NOWAIT | M_ZERO); 944 if (nims == NULL) 945 return (ENOMEM); 946 nims->ims_haddr = haddr; 947 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 948 ++inm->inm_nsrc; 949 ims = nims; 950 #ifdef KTR 951 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__, 952 haddr, ims); 953 #endif 954 } 955 956 *pims = ims; 957 return (0); 958 } 959 960 /* 961 * Merge socket-layer source into IGMP-layer source. 962 * If rollback is non-zero, perform the inverse of the merge. 963 */ 964 static void 965 ims_merge(struct ip_msource *ims, const struct in_msource *lims, 966 const int rollback) 967 { 968 int n = rollback ? -1 : 1; 969 970 if (lims->imsl_st[0] == MCAST_EXCLUDE) { 971 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x", 972 __func__, n, ims->ims_haddr); 973 ims->ims_st[1].ex -= n; 974 } else if (lims->imsl_st[0] == MCAST_INCLUDE) { 975 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x", 976 __func__, n, ims->ims_haddr); 977 ims->ims_st[1].in -= n; 978 } 979 980 if (lims->imsl_st[1] == MCAST_EXCLUDE) { 981 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x", 982 __func__, n, ims->ims_haddr); 983 ims->ims_st[1].ex += n; 984 } else if (lims->imsl_st[1] == MCAST_INCLUDE) { 985 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x", 986 __func__, n, ims->ims_haddr); 987 ims->ims_st[1].in += n; 988 } 989 } 990 991 /* 992 * Atomically update the global in_multi state, when a membership's 993 * filter list is being updated in any way. 994 * 995 * imf is the per-inpcb-membership group filter pointer. 996 * A fake imf may be passed for in-kernel consumers. 997 * 998 * XXX This is a candidate for a set-symmetric-difference style loop 999 * which would eliminate the repeated lookup from root of ims nodes, 1000 * as they share the same key space. 1001 * 1002 * If any error occurred this function will back out of refcounts 1003 * and return a non-zero value. 1004 */ 1005 static int 1006 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1007 { 1008 struct ip_msource *ims, *nims; 1009 struct in_msource *lims; 1010 int schanged, error; 1011 int nsrc0, nsrc1; 1012 1013 schanged = 0; 1014 error = 0; 1015 nsrc1 = nsrc0 = 0; 1016 IN_MULTI_LIST_LOCK_ASSERT(); 1017 1018 /* 1019 * Update the source filters first, as this may fail. 1020 * Maintain count of in-mode filters at t0, t1. These are 1021 * used to work out if we transition into ASM mode or not. 1022 * Maintain a count of source filters whose state was 1023 * actually modified by this operation. 1024 */ 1025 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1026 lims = (struct in_msource *)ims; 1027 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++; 1028 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++; 1029 if (lims->imsl_st[0] == lims->imsl_st[1]) continue; 1030 error = inm_get_source(inm, lims->ims_haddr, 0, &nims); 1031 ++schanged; 1032 if (error) 1033 break; 1034 ims_merge(nims, lims, 0); 1035 } 1036 if (error) { 1037 struct ip_msource *bims; 1038 1039 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) { 1040 lims = (struct in_msource *)ims; 1041 if (lims->imsl_st[0] == lims->imsl_st[1]) 1042 continue; 1043 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims); 1044 if (bims == NULL) 1045 continue; 1046 ims_merge(bims, lims, 1); 1047 } 1048 goto out_reap; 1049 } 1050 1051 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1", 1052 __func__, nsrc0, nsrc1); 1053 1054 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 1055 if (imf->imf_st[0] == imf->imf_st[1] && 1056 imf->imf_st[1] == MCAST_INCLUDE) { 1057 if (nsrc1 == 0) { 1058 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1059 --inm->inm_st[1].iss_in; 1060 } 1061 } 1062 1063 /* Handle filter mode transition on socket. */ 1064 if (imf->imf_st[0] != imf->imf_st[1]) { 1065 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d", 1066 __func__, imf->imf_st[0], imf->imf_st[1]); 1067 1068 if (imf->imf_st[0] == MCAST_EXCLUDE) { 1069 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__); 1070 --inm->inm_st[1].iss_ex; 1071 } else if (imf->imf_st[0] == MCAST_INCLUDE) { 1072 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1073 --inm->inm_st[1].iss_in; 1074 } 1075 1076 if (imf->imf_st[1] == MCAST_EXCLUDE) { 1077 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__); 1078 inm->inm_st[1].iss_ex++; 1079 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) { 1080 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__); 1081 inm->inm_st[1].iss_in++; 1082 } 1083 } 1084 1085 /* 1086 * Track inm filter state in terms of listener counts. 1087 * If there are any exclusive listeners, stack-wide 1088 * membership is exclusive. 1089 * Otherwise, if only inclusive listeners, stack-wide is inclusive. 1090 * If no listeners remain, state is undefined at t1, 1091 * and the IGMP lifecycle for this group should finish. 1092 */ 1093 if (inm->inm_st[1].iss_ex > 0) { 1094 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__); 1095 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE; 1096 } else if (inm->inm_st[1].iss_in > 0) { 1097 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__); 1098 inm->inm_st[1].iss_fmode = MCAST_INCLUDE; 1099 } else { 1100 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__); 1101 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 1102 } 1103 1104 /* Decrement ASM listener count on transition out of ASM mode. */ 1105 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1106 if ((imf->imf_st[1] != MCAST_EXCLUDE) || 1107 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) { 1108 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__); 1109 --inm->inm_st[1].iss_asm; 1110 } 1111 } 1112 1113 /* Increment ASM listener count on transition to ASM mode. */ 1114 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1115 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__); 1116 inm->inm_st[1].iss_asm++; 1117 } 1118 1119 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm); 1120 inm_print(inm); 1121 1122 out_reap: 1123 if (schanged > 0) { 1124 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__); 1125 inm_reap(inm); 1126 } 1127 return (error); 1128 } 1129 1130 /* 1131 * Mark an in_multi's filter set deltas as committed. 1132 * Called by IGMP after a state change has been enqueued. 1133 */ 1134 void 1135 inm_commit(struct in_multi *inm) 1136 { 1137 struct ip_msource *ims; 1138 1139 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm); 1140 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__); 1141 inm_print(inm); 1142 1143 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 1144 ims->ims_st[0] = ims->ims_st[1]; 1145 } 1146 inm->inm_st[0] = inm->inm_st[1]; 1147 } 1148 1149 /* 1150 * Reap unreferenced nodes from an in_multi's filter set. 1151 */ 1152 static void 1153 inm_reap(struct in_multi *inm) 1154 { 1155 struct ip_msource *ims, *tims; 1156 1157 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1158 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 || 1159 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 || 1160 ims->ims_stp != 0) 1161 continue; 1162 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1163 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1164 free(ims, M_IPMSOURCE); 1165 inm->inm_nsrc--; 1166 } 1167 } 1168 1169 /* 1170 * Purge all source nodes from an in_multi's filter set. 1171 */ 1172 static void 1173 inm_purge(struct in_multi *inm) 1174 { 1175 struct ip_msource *ims, *tims; 1176 1177 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1178 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1179 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1180 free(ims, M_IPMSOURCE); 1181 inm->inm_nsrc--; 1182 } 1183 } 1184 1185 /* 1186 * Join a multicast group; unlocked entry point. 1187 * 1188 * SMPng: XXX: in_joingroup() is called from in_control() when Giant 1189 * is not held. Fortunately, ifp is unlikely to have been detached 1190 * at this point, so we assume it's OK to recurse. 1191 */ 1192 int 1193 in_joingroup(struct ifnet *ifp, const struct in_addr *gina, 1194 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1195 { 1196 int error; 1197 1198 IN_MULTI_LOCK(); 1199 error = in_joingroup_locked(ifp, gina, imf, pinm); 1200 IN_MULTI_UNLOCK(); 1201 1202 return (error); 1203 } 1204 1205 /* 1206 * Join a multicast group; real entry point. 1207 * 1208 * Only preserves atomicity at inm level. 1209 * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1210 * 1211 * If the IGMP downcall fails, the group is not joined, and an error 1212 * code is returned. 1213 */ 1214 int 1215 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina, 1216 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1217 { 1218 struct in_mfilter timf; 1219 struct in_multi *inm; 1220 int error; 1221 1222 IN_MULTI_LOCK_ASSERT(); 1223 IN_MULTI_LIST_UNLOCK_ASSERT(); 1224 1225 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__, 1226 ntohl(gina->s_addr), ifp, ifp->if_xname); 1227 1228 error = 0; 1229 inm = NULL; 1230 1231 /* 1232 * If no imf was specified (i.e. kernel consumer), 1233 * fake one up and assume it is an ASM join. 1234 */ 1235 if (imf == NULL) { 1236 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1237 imf = &timf; 1238 } 1239 1240 error = in_getmulti(ifp, gina, &inm); 1241 if (error) { 1242 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__); 1243 return (error); 1244 } 1245 IN_MULTI_LIST_LOCK(); 1246 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1247 error = inm_merge(inm, imf); 1248 if (error) { 1249 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1250 goto out_inm_release; 1251 } 1252 1253 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1254 error = igmp_change_state(inm); 1255 if (error) { 1256 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__); 1257 goto out_inm_release; 1258 } 1259 1260 out_inm_release: 1261 if (error) { 1262 1263 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1264 IF_ADDR_WLOCK(ifp); 1265 inm_release_deferred(inm); 1266 IF_ADDR_WUNLOCK(ifp); 1267 } else { 1268 *pinm = inm; 1269 } 1270 IN_MULTI_LIST_UNLOCK(); 1271 1272 return (error); 1273 } 1274 1275 /* 1276 * Leave a multicast group; unlocked entry point. 1277 */ 1278 int 1279 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1280 { 1281 int error; 1282 1283 IN_MULTI_LOCK(); 1284 error = in_leavegroup_locked(inm, imf); 1285 IN_MULTI_UNLOCK(); 1286 1287 return (error); 1288 } 1289 1290 /* 1291 * Leave a multicast group; real entry point. 1292 * All source filters will be expunged. 1293 * 1294 * Only preserves atomicity at inm level. 1295 * 1296 * Holding the write lock for the INP which contains imf 1297 * is highly advisable. We can't assert for it as imf does not 1298 * contain a back-pointer to the owning inp. 1299 * 1300 * Note: This is not the same as inm_release(*) as this function also 1301 * makes a state change downcall into IGMP. 1302 */ 1303 int 1304 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1305 { 1306 struct in_mfilter timf; 1307 int error; 1308 1309 IN_MULTI_LOCK_ASSERT(); 1310 IN_MULTI_LIST_UNLOCK_ASSERT(); 1311 1312 error = 0; 1313 1314 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__, 1315 inm, ntohl(inm->inm_addr.s_addr), 1316 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname), 1317 imf); 1318 1319 /* 1320 * If no imf was specified (i.e. kernel consumer), 1321 * fake one up and assume it is an ASM join. 1322 */ 1323 if (imf == NULL) { 1324 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1325 imf = &timf; 1326 } 1327 1328 /* 1329 * Begin state merge transaction at IGMP layer. 1330 * 1331 * As this particular invocation should not cause any memory 1332 * to be allocated, and there is no opportunity to roll back 1333 * the transaction, it MUST NOT fail. 1334 */ 1335 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1336 IN_MULTI_LIST_LOCK(); 1337 error = inm_merge(inm, imf); 1338 KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1339 1340 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1341 CURVNET_SET(inm->inm_ifp->if_vnet); 1342 error = igmp_change_state(inm); 1343 IF_ADDR_WLOCK(inm->inm_ifp); 1344 inm_release_deferred(inm); 1345 IF_ADDR_WUNLOCK(inm->inm_ifp); 1346 IN_MULTI_LIST_UNLOCK(); 1347 CURVNET_RESTORE(); 1348 if (error) 1349 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1350 1351 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1352 1353 return (error); 1354 } 1355 1356 /*#ifndef BURN_BRIDGES*/ 1357 /* 1358 * Join an IPv4 multicast group in (*,G) exclusive mode. 1359 * The group must be a 224.0.0.0/24 link-scope group. 1360 * This KPI is for legacy kernel consumers only. 1361 */ 1362 struct in_multi * 1363 in_addmulti(struct in_addr *ap, struct ifnet *ifp) 1364 { 1365 struct in_multi *pinm; 1366 int error; 1367 #ifdef INVARIANTS 1368 char addrbuf[INET_ADDRSTRLEN]; 1369 #endif 1370 1371 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)), 1372 ("%s: %s not in 224.0.0.0/24", __func__, 1373 inet_ntoa_r(*ap, addrbuf))); 1374 1375 error = in_joingroup(ifp, ap, NULL, &pinm); 1376 if (error != 0) 1377 pinm = NULL; 1378 1379 return (pinm); 1380 } 1381 1382 /* 1383 * Block or unblock an ASM multicast source on an inpcb. 1384 * This implements the delta-based API described in RFC 3678. 1385 * 1386 * The delta-based API applies only to exclusive-mode memberships. 1387 * An IGMP downcall will be performed. 1388 * 1389 * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1390 * 1391 * Return 0 if successful, otherwise return an appropriate error code. 1392 */ 1393 static int 1394 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1395 { 1396 struct group_source_req gsr; 1397 struct rm_priotracker in_ifa_tracker; 1398 sockunion_t *gsa, *ssa; 1399 struct ifnet *ifp; 1400 struct in_mfilter *imf; 1401 struct ip_moptions *imo; 1402 struct in_msource *ims; 1403 struct in_multi *inm; 1404 uint16_t fmode; 1405 int error, doblock; 1406 1407 ifp = NULL; 1408 error = 0; 1409 doblock = 0; 1410 1411 memset(&gsr, 0, sizeof(struct group_source_req)); 1412 gsa = (sockunion_t *)&gsr.gsr_group; 1413 ssa = (sockunion_t *)&gsr.gsr_source; 1414 1415 switch (sopt->sopt_name) { 1416 case IP_BLOCK_SOURCE: 1417 case IP_UNBLOCK_SOURCE: { 1418 struct ip_mreq_source mreqs; 1419 1420 error = sooptcopyin(sopt, &mreqs, 1421 sizeof(struct ip_mreq_source), 1422 sizeof(struct ip_mreq_source)); 1423 if (error) 1424 return (error); 1425 1426 gsa->sin.sin_family = AF_INET; 1427 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1428 gsa->sin.sin_addr = mreqs.imr_multiaddr; 1429 1430 ssa->sin.sin_family = AF_INET; 1431 ssa->sin.sin_len = sizeof(struct sockaddr_in); 1432 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1433 1434 if (!in_nullhost(mreqs.imr_interface)) { 1435 IN_IFADDR_RLOCK(&in_ifa_tracker); 1436 INADDR_TO_IFP(mreqs.imr_interface, ifp); 1437 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1438 } 1439 if (sopt->sopt_name == IP_BLOCK_SOURCE) 1440 doblock = 1; 1441 1442 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 1443 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 1444 break; 1445 } 1446 1447 case MCAST_BLOCK_SOURCE: 1448 case MCAST_UNBLOCK_SOURCE: 1449 error = sooptcopyin(sopt, &gsr, 1450 sizeof(struct group_source_req), 1451 sizeof(struct group_source_req)); 1452 if (error) 1453 return (error); 1454 1455 if (gsa->sin.sin_family != AF_INET || 1456 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 1457 return (EINVAL); 1458 1459 if (ssa->sin.sin_family != AF_INET || 1460 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 1461 return (EINVAL); 1462 1463 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1464 return (EADDRNOTAVAIL); 1465 1466 ifp = ifnet_byindex(gsr.gsr_interface); 1467 1468 if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1469 doblock = 1; 1470 break; 1471 1472 default: 1473 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1474 __func__, sopt->sopt_name); 1475 return (EOPNOTSUPP); 1476 break; 1477 } 1478 1479 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1480 return (EINVAL); 1481 1482 IN_MULTI_LOCK(); 1483 1484 /* 1485 * Check if we are actually a member of this group. 1486 */ 1487 imo = inp_findmoptions(inp); 1488 imf = imo_match_group(imo, ifp, &gsa->sa); 1489 if (imf == NULL) { 1490 error = EADDRNOTAVAIL; 1491 goto out_inp_locked; 1492 } 1493 inm = imf->imf_inm; 1494 1495 /* 1496 * Attempting to use the delta-based API on an 1497 * non exclusive-mode membership is an error. 1498 */ 1499 fmode = imf->imf_st[0]; 1500 if (fmode != MCAST_EXCLUDE) { 1501 error = EINVAL; 1502 goto out_inp_locked; 1503 } 1504 1505 /* 1506 * Deal with error cases up-front: 1507 * Asked to block, but already blocked; or 1508 * Asked to unblock, but nothing to unblock. 1509 * If adding a new block entry, allocate it. 1510 */ 1511 ims = imo_match_source(imf, &ssa->sa); 1512 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1513 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__, 1514 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not "); 1515 error = EADDRNOTAVAIL; 1516 goto out_inp_locked; 1517 } 1518 1519 INP_WLOCK_ASSERT(inp); 1520 1521 /* 1522 * Begin state merge transaction at socket layer. 1523 */ 1524 if (doblock) { 1525 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 1526 ims = imf_graft(imf, fmode, &ssa->sin); 1527 if (ims == NULL) 1528 error = ENOMEM; 1529 } else { 1530 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 1531 error = imf_prune(imf, &ssa->sin); 1532 } 1533 1534 if (error) { 1535 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__); 1536 goto out_imf_rollback; 1537 } 1538 1539 /* 1540 * Begin state merge transaction at IGMP layer. 1541 */ 1542 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1543 IN_MULTI_LIST_LOCK(); 1544 error = inm_merge(inm, imf); 1545 if (error) { 1546 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1547 IN_MULTI_LIST_UNLOCK(); 1548 goto out_imf_rollback; 1549 } 1550 1551 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1552 error = igmp_change_state(inm); 1553 IN_MULTI_LIST_UNLOCK(); 1554 if (error) 1555 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1556 1557 out_imf_rollback: 1558 if (error) 1559 imf_rollback(imf); 1560 else 1561 imf_commit(imf); 1562 1563 imf_reap(imf); 1564 1565 out_inp_locked: 1566 INP_WUNLOCK(inp); 1567 IN_MULTI_UNLOCK(); 1568 return (error); 1569 } 1570 1571 /* 1572 * Given an inpcb, return its multicast options structure pointer. Accepts 1573 * an unlocked inpcb pointer, but will return it locked. May sleep. 1574 * 1575 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1576 * SMPng: NOTE: Returns with the INP write lock held. 1577 */ 1578 static struct ip_moptions * 1579 inp_findmoptions(struct inpcb *inp) 1580 { 1581 struct ip_moptions *imo; 1582 1583 INP_WLOCK(inp); 1584 if (inp->inp_moptions != NULL) 1585 return (inp->inp_moptions); 1586 1587 INP_WUNLOCK(inp); 1588 1589 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK); 1590 1591 imo->imo_multicast_ifp = NULL; 1592 imo->imo_multicast_addr.s_addr = INADDR_ANY; 1593 imo->imo_multicast_vif = -1; 1594 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL; 1595 imo->imo_multicast_loop = in_mcast_loop; 1596 STAILQ_INIT(&imo->imo_head); 1597 1598 INP_WLOCK(inp); 1599 if (inp->inp_moptions != NULL) { 1600 free(imo, M_IPMOPTS); 1601 return (inp->inp_moptions); 1602 } 1603 inp->inp_moptions = imo; 1604 return (imo); 1605 } 1606 1607 static void 1608 inp_gcmoptions(struct ip_moptions *imo) 1609 { 1610 struct in_mfilter *imf; 1611 struct in_multi *inm; 1612 struct ifnet *ifp; 1613 1614 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) { 1615 ip_mfilter_remove(&imo->imo_head, imf); 1616 1617 imf_leave(imf); 1618 if ((inm = imf->imf_inm) != NULL) { 1619 if ((ifp = inm->inm_ifp) != NULL) { 1620 CURVNET_SET(ifp->if_vnet); 1621 (void)in_leavegroup(inm, imf); 1622 CURVNET_RESTORE(); 1623 } else { 1624 (void)in_leavegroup(inm, imf); 1625 } 1626 } 1627 ip_mfilter_free(imf); 1628 } 1629 free(imo, M_IPMOPTS); 1630 } 1631 1632 /* 1633 * Discard the IP multicast options (and source filters). To minimize 1634 * the amount of work done while holding locks such as the INP's 1635 * pcbinfo lock (which is used in the receive path), the free 1636 * operation is deferred to the epoch callback task. 1637 */ 1638 void 1639 inp_freemoptions(struct ip_moptions *imo) 1640 { 1641 if (imo == NULL) 1642 return; 1643 inp_gcmoptions(imo); 1644 } 1645 1646 /* 1647 * Atomically get source filters on a socket for an IPv4 multicast group. 1648 * Called with INP lock held; returns with lock released. 1649 */ 1650 static int 1651 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1652 { 1653 struct __msfilterreq msfr; 1654 sockunion_t *gsa; 1655 struct ifnet *ifp; 1656 struct ip_moptions *imo; 1657 struct in_mfilter *imf; 1658 struct ip_msource *ims; 1659 struct in_msource *lims; 1660 struct sockaddr_in *psin; 1661 struct sockaddr_storage *ptss; 1662 struct sockaddr_storage *tss; 1663 int error; 1664 size_t nsrcs, ncsrcs; 1665 1666 INP_WLOCK_ASSERT(inp); 1667 1668 imo = inp->inp_moptions; 1669 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__)); 1670 1671 INP_WUNLOCK(inp); 1672 1673 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1674 sizeof(struct __msfilterreq)); 1675 if (error) 1676 return (error); 1677 1678 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 1679 return (EINVAL); 1680 1681 ifp = ifnet_byindex(msfr.msfr_ifindex); 1682 if (ifp == NULL) 1683 return (EINVAL); 1684 1685 INP_WLOCK(inp); 1686 1687 /* 1688 * Lookup group on the socket. 1689 */ 1690 gsa = (sockunion_t *)&msfr.msfr_group; 1691 imf = imo_match_group(imo, ifp, &gsa->sa); 1692 if (imf == NULL) { 1693 INP_WUNLOCK(inp); 1694 return (EADDRNOTAVAIL); 1695 } 1696 1697 /* 1698 * Ignore memberships which are in limbo. 1699 */ 1700 if (imf->imf_st[1] == MCAST_UNDEFINED) { 1701 INP_WUNLOCK(inp); 1702 return (EAGAIN); 1703 } 1704 msfr.msfr_fmode = imf->imf_st[1]; 1705 1706 /* 1707 * If the user specified a buffer, copy out the source filter 1708 * entries to userland gracefully. 1709 * We only copy out the number of entries which userland 1710 * has asked for, but we always tell userland how big the 1711 * buffer really needs to be. 1712 */ 1713 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 1714 msfr.msfr_nsrcs = in_mcast_maxsocksrc; 1715 tss = NULL; 1716 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1717 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1718 M_TEMP, M_NOWAIT | M_ZERO); 1719 if (tss == NULL) { 1720 INP_WUNLOCK(inp); 1721 return (ENOBUFS); 1722 } 1723 } 1724 1725 /* 1726 * Count number of sources in-mode at t0. 1727 * If buffer space exists and remains, copy out source entries. 1728 */ 1729 nsrcs = msfr.msfr_nsrcs; 1730 ncsrcs = 0; 1731 ptss = tss; 1732 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1733 lims = (struct in_msource *)ims; 1734 if (lims->imsl_st[0] == MCAST_UNDEFINED || 1735 lims->imsl_st[0] != imf->imf_st[0]) 1736 continue; 1737 ++ncsrcs; 1738 if (tss != NULL && nsrcs > 0) { 1739 psin = (struct sockaddr_in *)ptss; 1740 psin->sin_family = AF_INET; 1741 psin->sin_len = sizeof(struct sockaddr_in); 1742 psin->sin_addr.s_addr = htonl(lims->ims_haddr); 1743 psin->sin_port = 0; 1744 ++ptss; 1745 --nsrcs; 1746 } 1747 } 1748 1749 INP_WUNLOCK(inp); 1750 1751 if (tss != NULL) { 1752 error = copyout(tss, msfr.msfr_srcs, 1753 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1754 free(tss, M_TEMP); 1755 if (error) 1756 return (error); 1757 } 1758 1759 msfr.msfr_nsrcs = ncsrcs; 1760 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1761 1762 return (error); 1763 } 1764 1765 /* 1766 * Return the IP multicast options in response to user getsockopt(). 1767 */ 1768 int 1769 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1770 { 1771 struct rm_priotracker in_ifa_tracker; 1772 struct ip_mreqn mreqn; 1773 struct ip_moptions *imo; 1774 struct ifnet *ifp; 1775 struct in_ifaddr *ia; 1776 int error, optval; 1777 u_char coptval; 1778 1779 INP_WLOCK(inp); 1780 imo = inp->inp_moptions; 1781 /* 1782 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 1783 * or is a divert socket, reject it. 1784 */ 1785 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 1786 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1787 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) { 1788 INP_WUNLOCK(inp); 1789 return (EOPNOTSUPP); 1790 } 1791 1792 error = 0; 1793 switch (sopt->sopt_name) { 1794 case IP_MULTICAST_VIF: 1795 if (imo != NULL) 1796 optval = imo->imo_multicast_vif; 1797 else 1798 optval = -1; 1799 INP_WUNLOCK(inp); 1800 error = sooptcopyout(sopt, &optval, sizeof(int)); 1801 break; 1802 1803 case IP_MULTICAST_IF: 1804 memset(&mreqn, 0, sizeof(struct ip_mreqn)); 1805 if (imo != NULL) { 1806 ifp = imo->imo_multicast_ifp; 1807 if (!in_nullhost(imo->imo_multicast_addr)) { 1808 mreqn.imr_address = imo->imo_multicast_addr; 1809 } else if (ifp != NULL) { 1810 struct epoch_tracker et; 1811 1812 mreqn.imr_ifindex = ifp->if_index; 1813 NET_EPOCH_ENTER(et); 1814 IFP_TO_IA(ifp, ia, &in_ifa_tracker); 1815 if (ia != NULL) 1816 mreqn.imr_address = 1817 IA_SIN(ia)->sin_addr; 1818 NET_EPOCH_EXIT(et); 1819 } 1820 } 1821 INP_WUNLOCK(inp); 1822 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 1823 error = sooptcopyout(sopt, &mreqn, 1824 sizeof(struct ip_mreqn)); 1825 } else { 1826 error = sooptcopyout(sopt, &mreqn.imr_address, 1827 sizeof(struct in_addr)); 1828 } 1829 break; 1830 1831 case IP_MULTICAST_TTL: 1832 if (imo == NULL) 1833 optval = coptval = IP_DEFAULT_MULTICAST_TTL; 1834 else 1835 optval = coptval = imo->imo_multicast_ttl; 1836 INP_WUNLOCK(inp); 1837 if (sopt->sopt_valsize == sizeof(u_char)) 1838 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1839 else 1840 error = sooptcopyout(sopt, &optval, sizeof(int)); 1841 break; 1842 1843 case IP_MULTICAST_LOOP: 1844 if (imo == NULL) 1845 optval = coptval = IP_DEFAULT_MULTICAST_LOOP; 1846 else 1847 optval = coptval = imo->imo_multicast_loop; 1848 INP_WUNLOCK(inp); 1849 if (sopt->sopt_valsize == sizeof(u_char)) 1850 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1851 else 1852 error = sooptcopyout(sopt, &optval, sizeof(int)); 1853 break; 1854 1855 case IP_MSFILTER: 1856 if (imo == NULL) { 1857 error = EADDRNOTAVAIL; 1858 INP_WUNLOCK(inp); 1859 } else { 1860 error = inp_get_source_filters(inp, sopt); 1861 } 1862 break; 1863 1864 default: 1865 INP_WUNLOCK(inp); 1866 error = ENOPROTOOPT; 1867 break; 1868 } 1869 1870 INP_UNLOCK_ASSERT(inp); 1871 1872 return (error); 1873 } 1874 1875 /* 1876 * Look up the ifnet to use for a multicast group membership, 1877 * given the IPv4 address of an interface, and the IPv4 group address. 1878 * 1879 * This routine exists to support legacy multicast applications 1880 * which do not understand that multicast memberships are scoped to 1881 * specific physical links in the networking stack, or which need 1882 * to join link-scope groups before IPv4 addresses are configured. 1883 * 1884 * If inp is non-NULL, use this socket's current FIB number for any 1885 * required FIB lookup. 1886 * If ina is INADDR_ANY, look up the group address in the unicast FIB, 1887 * and use its ifp; usually, this points to the default next-hop. 1888 * 1889 * If the FIB lookup fails, attempt to use the first non-loopback 1890 * interface with multicast capability in the system as a 1891 * last resort. The legacy IPv4 ASM API requires that we do 1892 * this in order to allow groups to be joined when the routing 1893 * table has not yet been populated during boot. 1894 * 1895 * Returns NULL if no ifp could be found. 1896 * 1897 * FUTURE: Implement IPv4 source-address selection. 1898 */ 1899 static struct ifnet * 1900 inp_lookup_mcast_ifp(const struct inpcb *inp, 1901 const struct sockaddr_in *gsin, const struct in_addr ina) 1902 { 1903 struct rm_priotracker in_ifa_tracker; 1904 struct ifnet *ifp; 1905 struct nhop4_basic nh4; 1906 uint32_t fibnum; 1907 1908 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__)); 1909 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)), 1910 ("%s: not multicast", __func__)); 1911 1912 ifp = NULL; 1913 if (!in_nullhost(ina)) { 1914 IN_IFADDR_RLOCK(&in_ifa_tracker); 1915 INADDR_TO_IFP(ina, ifp); 1916 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1917 } else { 1918 fibnum = inp ? inp->inp_inc.inc_fibnum : 0; 1919 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0) 1920 ifp = nh4.nh_ifp; 1921 else { 1922 struct in_ifaddr *ia; 1923 struct ifnet *mifp; 1924 1925 mifp = NULL; 1926 IN_IFADDR_RLOCK(&in_ifa_tracker); 1927 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1928 mifp = ia->ia_ifp; 1929 if (!(mifp->if_flags & IFF_LOOPBACK) && 1930 (mifp->if_flags & IFF_MULTICAST)) { 1931 ifp = mifp; 1932 break; 1933 } 1934 } 1935 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1936 } 1937 } 1938 1939 return (ifp); 1940 } 1941 1942 /* 1943 * Join an IPv4 multicast group, possibly with a source. 1944 */ 1945 static int 1946 inp_join_group(struct inpcb *inp, struct sockopt *sopt) 1947 { 1948 struct group_source_req gsr; 1949 sockunion_t *gsa, *ssa; 1950 struct ifnet *ifp; 1951 struct in_mfilter *imf; 1952 struct ip_moptions *imo; 1953 struct in_multi *inm; 1954 struct in_msource *lims; 1955 int error, is_new; 1956 1957 ifp = NULL; 1958 lims = NULL; 1959 error = 0; 1960 1961 memset(&gsr, 0, sizeof(struct group_source_req)); 1962 gsa = (sockunion_t *)&gsr.gsr_group; 1963 gsa->ss.ss_family = AF_UNSPEC; 1964 ssa = (sockunion_t *)&gsr.gsr_source; 1965 ssa->ss.ss_family = AF_UNSPEC; 1966 1967 switch (sopt->sopt_name) { 1968 case IP_ADD_MEMBERSHIP: { 1969 struct ip_mreqn mreqn; 1970 1971 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) 1972 error = sooptcopyin(sopt, &mreqn, 1973 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn)); 1974 else 1975 error = sooptcopyin(sopt, &mreqn, 1976 sizeof(struct ip_mreq), sizeof(struct ip_mreq)); 1977 if (error) 1978 return (error); 1979 1980 gsa->sin.sin_family = AF_INET; 1981 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1982 gsa->sin.sin_addr = mreqn.imr_multiaddr; 1983 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1984 return (EINVAL); 1985 1986 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) && 1987 mreqn.imr_ifindex != 0) 1988 ifp = ifnet_byindex(mreqn.imr_ifindex); 1989 else 1990 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 1991 mreqn.imr_address); 1992 break; 1993 } 1994 case IP_ADD_SOURCE_MEMBERSHIP: { 1995 struct ip_mreq_source mreqs; 1996 1997 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source), 1998 sizeof(struct ip_mreq_source)); 1999 if (error) 2000 return (error); 2001 2002 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET; 2003 gsa->sin.sin_len = ssa->sin.sin_len = 2004 sizeof(struct sockaddr_in); 2005 2006 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2007 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2008 return (EINVAL); 2009 2010 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2011 2012 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 2013 mreqs.imr_interface); 2014 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2015 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2016 break; 2017 } 2018 2019 case MCAST_JOIN_GROUP: 2020 case MCAST_JOIN_SOURCE_GROUP: 2021 if (sopt->sopt_name == MCAST_JOIN_GROUP) { 2022 error = sooptcopyin(sopt, &gsr, 2023 sizeof(struct group_req), 2024 sizeof(struct group_req)); 2025 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2026 error = sooptcopyin(sopt, &gsr, 2027 sizeof(struct group_source_req), 2028 sizeof(struct group_source_req)); 2029 } 2030 if (error) 2031 return (error); 2032 2033 if (gsa->sin.sin_family != AF_INET || 2034 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2035 return (EINVAL); 2036 2037 /* 2038 * Overwrite the port field if present, as the sockaddr 2039 * being copied in may be matched with a binary comparison. 2040 */ 2041 gsa->sin.sin_port = 0; 2042 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2043 if (ssa->sin.sin_family != AF_INET || 2044 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2045 return (EINVAL); 2046 ssa->sin.sin_port = 0; 2047 } 2048 2049 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2050 return (EINVAL); 2051 2052 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2053 return (EADDRNOTAVAIL); 2054 ifp = ifnet_byindex(gsr.gsr_interface); 2055 break; 2056 2057 default: 2058 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2059 __func__, sopt->sopt_name); 2060 return (EOPNOTSUPP); 2061 break; 2062 } 2063 2064 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) 2065 return (EADDRNOTAVAIL); 2066 2067 IN_MULTI_LOCK(); 2068 2069 /* 2070 * Find the membership in the membership list. 2071 */ 2072 imo = inp_findmoptions(inp); 2073 imf = imo_match_group(imo, ifp, &gsa->sa); 2074 if (imf == NULL) { 2075 is_new = 1; 2076 inm = NULL; 2077 2078 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) { 2079 error = ENOMEM; 2080 goto out_inp_locked; 2081 } 2082 } else { 2083 is_new = 0; 2084 inm = imf->imf_inm; 2085 2086 if (ssa->ss.ss_family != AF_UNSPEC) { 2087 /* 2088 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership 2089 * is an error. On an existing inclusive membership, 2090 * it just adds the source to the filter list. 2091 */ 2092 if (imf->imf_st[1] != MCAST_INCLUDE) { 2093 error = EINVAL; 2094 goto out_inp_locked; 2095 } 2096 /* 2097 * Throw out duplicates. 2098 * 2099 * XXX FIXME: This makes a naive assumption that 2100 * even if entries exist for *ssa in this imf, 2101 * they will be rejected as dupes, even if they 2102 * are not valid in the current mode (in-mode). 2103 * 2104 * in_msource is transactioned just as for anything 2105 * else in SSM -- but note naive use of inm_graft() 2106 * below for allocating new filter entries. 2107 * 2108 * This is only an issue if someone mixes the 2109 * full-state SSM API with the delta-based API, 2110 * which is discouraged in the relevant RFCs. 2111 */ 2112 lims = imo_match_source(imf, &ssa->sa); 2113 if (lims != NULL /*&& 2114 lims->imsl_st[1] == MCAST_INCLUDE*/) { 2115 error = EADDRNOTAVAIL; 2116 goto out_inp_locked; 2117 } 2118 } else { 2119 /* 2120 * MCAST_JOIN_GROUP on an existing exclusive 2121 * membership is an error; return EADDRINUSE 2122 * to preserve 4.4BSD API idempotence, and 2123 * avoid tedious detour to code below. 2124 * NOTE: This is bending RFC 3678 a bit. 2125 * 2126 * On an existing inclusive membership, this is also 2127 * an error; if you want to change filter mode, 2128 * you must use the userland API setsourcefilter(). 2129 * XXX We don't reject this for imf in UNDEFINED 2130 * state at t1, because allocation of a filter 2131 * is atomic with allocation of a membership. 2132 */ 2133 error = EINVAL; 2134 if (imf->imf_st[1] == MCAST_EXCLUDE) 2135 error = EADDRINUSE; 2136 goto out_inp_locked; 2137 } 2138 } 2139 2140 /* 2141 * Begin state merge transaction at socket layer. 2142 */ 2143 INP_WLOCK_ASSERT(inp); 2144 2145 /* 2146 * Graft new source into filter list for this inpcb's 2147 * membership of the group. The in_multi may not have 2148 * been allocated yet if this is a new membership, however, 2149 * the in_mfilter slot will be allocated and must be initialized. 2150 * 2151 * Note: Grafting of exclusive mode filters doesn't happen 2152 * in this path. 2153 * XXX: Should check for non-NULL lims (node exists but may 2154 * not be in-mode) for interop with full-state API. 2155 */ 2156 if (ssa->ss.ss_family != AF_UNSPEC) { 2157 /* Membership starts in IN mode */ 2158 if (is_new) { 2159 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__); 2160 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE); 2161 if (imf == NULL) { 2162 error = ENOMEM; 2163 goto out_inp_locked; 2164 } 2165 } else { 2166 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 2167 } 2168 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin); 2169 if (lims == NULL) { 2170 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2171 __func__); 2172 error = ENOMEM; 2173 goto out_inp_locked; 2174 } 2175 } else { 2176 /* No address specified; Membership starts in EX mode */ 2177 if (is_new) { 2178 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__); 2179 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE); 2180 if (imf == NULL) { 2181 error = ENOMEM; 2182 goto out_inp_locked; 2183 } 2184 } 2185 } 2186 2187 /* 2188 * Begin state merge transaction at IGMP layer. 2189 */ 2190 if (is_new) { 2191 in_pcbref(inp); 2192 INP_WUNLOCK(inp); 2193 2194 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf, 2195 &imf->imf_inm); 2196 2197 INP_WLOCK(inp); 2198 if (in_pcbrele_wlocked(inp)) { 2199 error = ENXIO; 2200 goto out_inp_unlocked; 2201 } 2202 if (error) { 2203 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed", 2204 __func__); 2205 goto out_inp_locked; 2206 } 2207 /* 2208 * NOTE: Refcount from in_joingroup_locked() 2209 * is protecting membership. 2210 */ 2211 ip_mfilter_insert(&imo->imo_head, imf); 2212 } else { 2213 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2214 IN_MULTI_LIST_LOCK(); 2215 error = inm_merge(inm, imf); 2216 if (error) { 2217 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2218 __func__); 2219 IN_MULTI_LIST_UNLOCK(); 2220 imf_rollback(imf); 2221 imf_reap(imf); 2222 goto out_inp_locked; 2223 } 2224 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2225 error = igmp_change_state(inm); 2226 IN_MULTI_LIST_UNLOCK(); 2227 if (error) { 2228 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2229 __func__); 2230 imf_rollback(imf); 2231 imf_reap(imf); 2232 goto out_inp_locked; 2233 } 2234 } 2235 2236 imf_commit(imf); 2237 imf = NULL; 2238 2239 out_inp_locked: 2240 INP_WUNLOCK(inp); 2241 out_inp_unlocked: 2242 IN_MULTI_UNLOCK(); 2243 2244 if (is_new && imf) { 2245 if (imf->imf_inm != NULL) { 2246 IN_MULTI_LIST_LOCK(); 2247 IF_ADDR_WLOCK(ifp); 2248 inm_release_deferred(imf->imf_inm); 2249 IF_ADDR_WUNLOCK(ifp); 2250 IN_MULTI_LIST_UNLOCK(); 2251 } 2252 ip_mfilter_free(imf); 2253 } 2254 return (error); 2255 } 2256 2257 /* 2258 * Leave an IPv4 multicast group on an inpcb, possibly with a source. 2259 */ 2260 static int 2261 inp_leave_group(struct inpcb *inp, struct sockopt *sopt) 2262 { 2263 struct group_source_req gsr; 2264 struct ip_mreq_source mreqs; 2265 struct rm_priotracker in_ifa_tracker; 2266 sockunion_t *gsa, *ssa; 2267 struct ifnet *ifp; 2268 struct in_mfilter *imf; 2269 struct ip_moptions *imo; 2270 struct in_msource *ims; 2271 struct in_multi *inm; 2272 int error; 2273 bool is_final; 2274 2275 ifp = NULL; 2276 error = 0; 2277 is_final = true; 2278 2279 memset(&gsr, 0, sizeof(struct group_source_req)); 2280 gsa = (sockunion_t *)&gsr.gsr_group; 2281 gsa->ss.ss_family = AF_UNSPEC; 2282 ssa = (sockunion_t *)&gsr.gsr_source; 2283 ssa->ss.ss_family = AF_UNSPEC; 2284 2285 switch (sopt->sopt_name) { 2286 case IP_DROP_MEMBERSHIP: 2287 case IP_DROP_SOURCE_MEMBERSHIP: 2288 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) { 2289 error = sooptcopyin(sopt, &mreqs, 2290 sizeof(struct ip_mreq), 2291 sizeof(struct ip_mreq)); 2292 /* 2293 * Swap interface and sourceaddr arguments, 2294 * as ip_mreq and ip_mreq_source are laid 2295 * out differently. 2296 */ 2297 mreqs.imr_interface = mreqs.imr_sourceaddr; 2298 mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 2299 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2300 error = sooptcopyin(sopt, &mreqs, 2301 sizeof(struct ip_mreq_source), 2302 sizeof(struct ip_mreq_source)); 2303 } 2304 if (error) 2305 return (error); 2306 2307 gsa->sin.sin_family = AF_INET; 2308 gsa->sin.sin_len = sizeof(struct sockaddr_in); 2309 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2310 2311 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2312 ssa->sin.sin_family = AF_INET; 2313 ssa->sin.sin_len = sizeof(struct sockaddr_in); 2314 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2315 } 2316 2317 /* 2318 * Attempt to look up hinted ifp from interface address. 2319 * Fallthrough with null ifp iff lookup fails, to 2320 * preserve 4.4BSD mcast API idempotence. 2321 * XXX NOTE WELL: The RFC 3678 API is preferred because 2322 * using an IPv4 address as a key is racy. 2323 */ 2324 if (!in_nullhost(mreqs.imr_interface)) { 2325 IN_IFADDR_RLOCK(&in_ifa_tracker); 2326 INADDR_TO_IFP(mreqs.imr_interface, ifp); 2327 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2328 } 2329 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2330 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2331 2332 break; 2333 2334 case MCAST_LEAVE_GROUP: 2335 case MCAST_LEAVE_SOURCE_GROUP: 2336 if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2337 error = sooptcopyin(sopt, &gsr, 2338 sizeof(struct group_req), 2339 sizeof(struct group_req)); 2340 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2341 error = sooptcopyin(sopt, &gsr, 2342 sizeof(struct group_source_req), 2343 sizeof(struct group_source_req)); 2344 } 2345 if (error) 2346 return (error); 2347 2348 if (gsa->sin.sin_family != AF_INET || 2349 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2350 return (EINVAL); 2351 2352 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2353 if (ssa->sin.sin_family != AF_INET || 2354 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2355 return (EINVAL); 2356 } 2357 2358 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2359 return (EADDRNOTAVAIL); 2360 2361 ifp = ifnet_byindex(gsr.gsr_interface); 2362 2363 if (ifp == NULL) 2364 return (EADDRNOTAVAIL); 2365 break; 2366 2367 default: 2368 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2369 __func__, sopt->sopt_name); 2370 return (EOPNOTSUPP); 2371 break; 2372 } 2373 2374 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2375 return (EINVAL); 2376 2377 IN_MULTI_LOCK(); 2378 2379 /* 2380 * Find the membership in the membership list. 2381 */ 2382 imo = inp_findmoptions(inp); 2383 imf = imo_match_group(imo, ifp, &gsa->sa); 2384 if (imf == NULL) { 2385 error = EADDRNOTAVAIL; 2386 goto out_inp_locked; 2387 } 2388 inm = imf->imf_inm; 2389 2390 if (ssa->ss.ss_family != AF_UNSPEC) 2391 is_final = false; 2392 2393 /* 2394 * Begin state merge transaction at socket layer. 2395 */ 2396 INP_WLOCK_ASSERT(inp); 2397 2398 /* 2399 * If we were instructed only to leave a given source, do so. 2400 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2401 */ 2402 if (is_final) { 2403 ip_mfilter_remove(&imo->imo_head, imf); 2404 imf_leave(imf); 2405 2406 /* 2407 * Give up the multicast address record to which 2408 * the membership points. 2409 */ 2410 (void) in_leavegroup_locked(imf->imf_inm, imf); 2411 } else { 2412 if (imf->imf_st[0] == MCAST_EXCLUDE) { 2413 error = EADDRNOTAVAIL; 2414 goto out_inp_locked; 2415 } 2416 ims = imo_match_source(imf, &ssa->sa); 2417 if (ims == NULL) { 2418 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", 2419 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not "); 2420 error = EADDRNOTAVAIL; 2421 goto out_inp_locked; 2422 } 2423 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 2424 error = imf_prune(imf, &ssa->sin); 2425 if (error) { 2426 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2427 __func__); 2428 goto out_inp_locked; 2429 } 2430 } 2431 2432 /* 2433 * Begin state merge transaction at IGMP layer. 2434 */ 2435 if (!is_final) { 2436 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2437 IN_MULTI_LIST_LOCK(); 2438 error = inm_merge(inm, imf); 2439 if (error) { 2440 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2441 __func__); 2442 IN_MULTI_LIST_UNLOCK(); 2443 imf_rollback(imf); 2444 imf_reap(imf); 2445 goto out_inp_locked; 2446 } 2447 2448 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2449 error = igmp_change_state(inm); 2450 IN_MULTI_LIST_UNLOCK(); 2451 if (error) { 2452 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2453 __func__); 2454 imf_rollback(imf); 2455 imf_reap(imf); 2456 goto out_inp_locked; 2457 } 2458 } 2459 imf_commit(imf); 2460 imf_reap(imf); 2461 2462 out_inp_locked: 2463 INP_WUNLOCK(inp); 2464 2465 if (is_final && imf) 2466 ip_mfilter_free(imf); 2467 2468 IN_MULTI_UNLOCK(); 2469 return (error); 2470 } 2471 2472 /* 2473 * Select the interface for transmitting IPv4 multicast datagrams. 2474 * 2475 * Either an instance of struct in_addr or an instance of struct ip_mreqn 2476 * may be passed to this socket option. An address of INADDR_ANY or an 2477 * interface index of 0 is used to remove a previous selection. 2478 * When no interface is selected, one is chosen for every send. 2479 */ 2480 static int 2481 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2482 { 2483 struct rm_priotracker in_ifa_tracker; 2484 struct in_addr addr; 2485 struct ip_mreqn mreqn; 2486 struct ifnet *ifp; 2487 struct ip_moptions *imo; 2488 int error; 2489 2490 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 2491 /* 2492 * An interface index was specified using the 2493 * Linux-derived ip_mreqn structure. 2494 */ 2495 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn), 2496 sizeof(struct ip_mreqn)); 2497 if (error) 2498 return (error); 2499 2500 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex) 2501 return (EINVAL); 2502 2503 if (mreqn.imr_ifindex == 0) { 2504 ifp = NULL; 2505 } else { 2506 ifp = ifnet_byindex(mreqn.imr_ifindex); 2507 if (ifp == NULL) 2508 return (EADDRNOTAVAIL); 2509 } 2510 } else { 2511 /* 2512 * An interface was specified by IPv4 address. 2513 * This is the traditional BSD usage. 2514 */ 2515 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr), 2516 sizeof(struct in_addr)); 2517 if (error) 2518 return (error); 2519 if (in_nullhost(addr)) { 2520 ifp = NULL; 2521 } else { 2522 IN_IFADDR_RLOCK(&in_ifa_tracker); 2523 INADDR_TO_IFP(addr, ifp); 2524 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2525 if (ifp == NULL) 2526 return (EADDRNOTAVAIL); 2527 } 2528 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp, 2529 ntohl(addr.s_addr)); 2530 } 2531 2532 /* Reject interfaces which do not support multicast. */ 2533 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0) 2534 return (EOPNOTSUPP); 2535 2536 imo = inp_findmoptions(inp); 2537 imo->imo_multicast_ifp = ifp; 2538 imo->imo_multicast_addr.s_addr = INADDR_ANY; 2539 INP_WUNLOCK(inp); 2540 2541 return (0); 2542 } 2543 2544 /* 2545 * Atomically set source filters on a socket for an IPv4 multicast group. 2546 * 2547 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2548 */ 2549 static int 2550 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2551 { 2552 struct __msfilterreq msfr; 2553 sockunion_t *gsa; 2554 struct ifnet *ifp; 2555 struct in_mfilter *imf; 2556 struct ip_moptions *imo; 2557 struct in_multi *inm; 2558 int error; 2559 2560 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2561 sizeof(struct __msfilterreq)); 2562 if (error) 2563 return (error); 2564 2565 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 2566 return (ENOBUFS); 2567 2568 if ((msfr.msfr_fmode != MCAST_EXCLUDE && 2569 msfr.msfr_fmode != MCAST_INCLUDE)) 2570 return (EINVAL); 2571 2572 if (msfr.msfr_group.ss_family != AF_INET || 2573 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in)) 2574 return (EINVAL); 2575 2576 gsa = (sockunion_t *)&msfr.msfr_group; 2577 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2578 return (EINVAL); 2579 2580 gsa->sin.sin_port = 0; /* ignore port */ 2581 2582 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 2583 return (EADDRNOTAVAIL); 2584 2585 ifp = ifnet_byindex(msfr.msfr_ifindex); 2586 if (ifp == NULL) 2587 return (EADDRNOTAVAIL); 2588 2589 IN_MULTI_LOCK(); 2590 2591 /* 2592 * Take the INP write lock. 2593 * Check if this socket is a member of this group. 2594 */ 2595 imo = inp_findmoptions(inp); 2596 imf = imo_match_group(imo, ifp, &gsa->sa); 2597 if (imf == NULL) { 2598 error = EADDRNOTAVAIL; 2599 goto out_inp_locked; 2600 } 2601 inm = imf->imf_inm; 2602 2603 /* 2604 * Begin state merge transaction at socket layer. 2605 */ 2606 INP_WLOCK_ASSERT(inp); 2607 2608 imf->imf_st[1] = msfr.msfr_fmode; 2609 2610 /* 2611 * Apply any new source filters, if present. 2612 * Make a copy of the user-space source vector so 2613 * that we may copy them with a single copyin. This 2614 * allows us to deal with page faults up-front. 2615 */ 2616 if (msfr.msfr_nsrcs > 0) { 2617 struct in_msource *lims; 2618 struct sockaddr_in *psin; 2619 struct sockaddr_storage *kss, *pkss; 2620 int i; 2621 2622 INP_WUNLOCK(inp); 2623 2624 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries", 2625 __func__, (unsigned long)msfr.msfr_nsrcs); 2626 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2627 M_TEMP, M_WAITOK); 2628 error = copyin(msfr.msfr_srcs, kss, 2629 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2630 if (error) { 2631 free(kss, M_TEMP); 2632 return (error); 2633 } 2634 2635 INP_WLOCK(inp); 2636 2637 /* 2638 * Mark all source filters as UNDEFINED at t1. 2639 * Restore new group filter mode, as imf_leave() 2640 * will set it to INCLUDE. 2641 */ 2642 imf_leave(imf); 2643 imf->imf_st[1] = msfr.msfr_fmode; 2644 2645 /* 2646 * Update socket layer filters at t1, lazy-allocating 2647 * new entries. This saves a bunch of memory at the 2648 * cost of one RB_FIND() per source entry; duplicate 2649 * entries in the msfr_nsrcs vector are ignored. 2650 * If we encounter an error, rollback transaction. 2651 * 2652 * XXX This too could be replaced with a set-symmetric 2653 * difference like loop to avoid walking from root 2654 * every time, as the key space is common. 2655 */ 2656 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2657 psin = (struct sockaddr_in *)pkss; 2658 if (psin->sin_family != AF_INET) { 2659 error = EAFNOSUPPORT; 2660 break; 2661 } 2662 if (psin->sin_len != sizeof(struct sockaddr_in)) { 2663 error = EINVAL; 2664 break; 2665 } 2666 error = imf_get_source(imf, psin, &lims); 2667 if (error) 2668 break; 2669 lims->imsl_st[1] = imf->imf_st[1]; 2670 } 2671 free(kss, M_TEMP); 2672 } 2673 2674 if (error) 2675 goto out_imf_rollback; 2676 2677 INP_WLOCK_ASSERT(inp); 2678 2679 /* 2680 * Begin state merge transaction at IGMP layer. 2681 */ 2682 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2683 IN_MULTI_LIST_LOCK(); 2684 error = inm_merge(inm, imf); 2685 if (error) { 2686 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 2687 IN_MULTI_LIST_UNLOCK(); 2688 goto out_imf_rollback; 2689 } 2690 2691 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2692 error = igmp_change_state(inm); 2693 IN_MULTI_LIST_UNLOCK(); 2694 if (error) 2695 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 2696 2697 out_imf_rollback: 2698 if (error) 2699 imf_rollback(imf); 2700 else 2701 imf_commit(imf); 2702 2703 imf_reap(imf); 2704 2705 out_inp_locked: 2706 INP_WUNLOCK(inp); 2707 IN_MULTI_UNLOCK(); 2708 return (error); 2709 } 2710 2711 /* 2712 * Set the IP multicast options in response to user setsockopt(). 2713 * 2714 * Many of the socket options handled in this function duplicate the 2715 * functionality of socket options in the regular unicast API. However, 2716 * it is not possible to merge the duplicate code, because the idempotence 2717 * of the IPv4 multicast part of the BSD Sockets API must be preserved; 2718 * the effects of these options must be treated as separate and distinct. 2719 * 2720 * SMPng: XXX: Unlocked read of inp_socket believed OK. 2721 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING 2722 * is refactored to no longer use vifs. 2723 */ 2724 int 2725 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2726 { 2727 struct ip_moptions *imo; 2728 int error; 2729 2730 error = 0; 2731 2732 /* 2733 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 2734 * or is a divert socket, reject it. 2735 */ 2736 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 2737 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2738 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) 2739 return (EOPNOTSUPP); 2740 2741 switch (sopt->sopt_name) { 2742 case IP_MULTICAST_VIF: { 2743 int vifi; 2744 /* 2745 * Select a multicast VIF for transmission. 2746 * Only useful if multicast forwarding is active. 2747 */ 2748 if (legal_vif_num == NULL) { 2749 error = EOPNOTSUPP; 2750 break; 2751 } 2752 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int)); 2753 if (error) 2754 break; 2755 if (!legal_vif_num(vifi) && (vifi != -1)) { 2756 error = EINVAL; 2757 break; 2758 } 2759 imo = inp_findmoptions(inp); 2760 imo->imo_multicast_vif = vifi; 2761 INP_WUNLOCK(inp); 2762 break; 2763 } 2764 2765 case IP_MULTICAST_IF: 2766 error = inp_set_multicast_if(inp, sopt); 2767 break; 2768 2769 case IP_MULTICAST_TTL: { 2770 u_char ttl; 2771 2772 /* 2773 * Set the IP time-to-live for outgoing multicast packets. 2774 * The original multicast API required a char argument, 2775 * which is inconsistent with the rest of the socket API. 2776 * We allow either a char or an int. 2777 */ 2778 if (sopt->sopt_valsize == sizeof(u_char)) { 2779 error = sooptcopyin(sopt, &ttl, sizeof(u_char), 2780 sizeof(u_char)); 2781 if (error) 2782 break; 2783 } else { 2784 u_int ittl; 2785 2786 error = sooptcopyin(sopt, &ittl, sizeof(u_int), 2787 sizeof(u_int)); 2788 if (error) 2789 break; 2790 if (ittl > 255) { 2791 error = EINVAL; 2792 break; 2793 } 2794 ttl = (u_char)ittl; 2795 } 2796 imo = inp_findmoptions(inp); 2797 imo->imo_multicast_ttl = ttl; 2798 INP_WUNLOCK(inp); 2799 break; 2800 } 2801 2802 case IP_MULTICAST_LOOP: { 2803 u_char loop; 2804 2805 /* 2806 * Set the loopback flag for outgoing multicast packets. 2807 * Must be zero or one. The original multicast API required a 2808 * char argument, which is inconsistent with the rest 2809 * of the socket API. We allow either a char or an int. 2810 */ 2811 if (sopt->sopt_valsize == sizeof(u_char)) { 2812 error = sooptcopyin(sopt, &loop, sizeof(u_char), 2813 sizeof(u_char)); 2814 if (error) 2815 break; 2816 } else { 2817 u_int iloop; 2818 2819 error = sooptcopyin(sopt, &iloop, sizeof(u_int), 2820 sizeof(u_int)); 2821 if (error) 2822 break; 2823 loop = (u_char)iloop; 2824 } 2825 imo = inp_findmoptions(inp); 2826 imo->imo_multicast_loop = !!loop; 2827 INP_WUNLOCK(inp); 2828 break; 2829 } 2830 2831 case IP_ADD_MEMBERSHIP: 2832 case IP_ADD_SOURCE_MEMBERSHIP: 2833 case MCAST_JOIN_GROUP: 2834 case MCAST_JOIN_SOURCE_GROUP: 2835 error = inp_join_group(inp, sopt); 2836 break; 2837 2838 case IP_DROP_MEMBERSHIP: 2839 case IP_DROP_SOURCE_MEMBERSHIP: 2840 case MCAST_LEAVE_GROUP: 2841 case MCAST_LEAVE_SOURCE_GROUP: 2842 error = inp_leave_group(inp, sopt); 2843 break; 2844 2845 case IP_BLOCK_SOURCE: 2846 case IP_UNBLOCK_SOURCE: 2847 case MCAST_BLOCK_SOURCE: 2848 case MCAST_UNBLOCK_SOURCE: 2849 error = inp_block_unblock_source(inp, sopt); 2850 break; 2851 2852 case IP_MSFILTER: 2853 error = inp_set_source_filters(inp, sopt); 2854 break; 2855 2856 default: 2857 error = EOPNOTSUPP; 2858 break; 2859 } 2860 2861 INP_UNLOCK_ASSERT(inp); 2862 2863 return (error); 2864 } 2865 2866 /* 2867 * Expose IGMP's multicast filter mode and source list(s) to userland, 2868 * keyed by (ifindex, group). 2869 * The filter mode is written out as a uint32_t, followed by 2870 * 0..n of struct in_addr. 2871 * For use by ifmcstat(8). 2872 * SMPng: NOTE: unlocked read of ifindex space. 2873 */ 2874 static int 2875 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS) 2876 { 2877 struct in_addr src, group; 2878 struct epoch_tracker et; 2879 struct ifnet *ifp; 2880 struct ifmultiaddr *ifma; 2881 struct in_multi *inm; 2882 struct ip_msource *ims; 2883 int *name; 2884 int retval; 2885 u_int namelen; 2886 uint32_t fmode, ifindex; 2887 2888 name = (int *)arg1; 2889 namelen = arg2; 2890 2891 if (req->newptr != NULL) 2892 return (EPERM); 2893 2894 if (namelen != 2) 2895 return (EINVAL); 2896 2897 ifindex = name[0]; 2898 if (ifindex <= 0 || ifindex > V_if_index) { 2899 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range", 2900 __func__, ifindex); 2901 return (ENOENT); 2902 } 2903 2904 group.s_addr = name[1]; 2905 if (!IN_MULTICAST(ntohl(group.s_addr))) { 2906 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast", 2907 __func__, ntohl(group.s_addr)); 2908 return (EINVAL); 2909 } 2910 2911 NET_EPOCH_ENTER(et); 2912 ifp = ifnet_byindex(ifindex); 2913 if (ifp == NULL) { 2914 NET_EPOCH_EXIT(et); 2915 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u", 2916 __func__, ifindex); 2917 return (ENOENT); 2918 } 2919 2920 retval = sysctl_wire_old_buffer(req, 2921 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr))); 2922 if (retval) { 2923 NET_EPOCH_EXIT(et); 2924 return (retval); 2925 } 2926 2927 IN_MULTI_LIST_LOCK(); 2928 2929 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2930 if (ifma->ifma_addr->sa_family != AF_INET || 2931 ifma->ifma_protospec == NULL) 2932 continue; 2933 inm = (struct in_multi *)ifma->ifma_protospec; 2934 if (!in_hosteq(inm->inm_addr, group)) 2935 continue; 2936 fmode = inm->inm_st[1].iss_fmode; 2937 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 2938 if (retval != 0) 2939 break; 2940 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 2941 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2942 ims->ims_haddr); 2943 /* 2944 * Only copy-out sources which are in-mode. 2945 */ 2946 if (fmode != ims_get_mode(inm, ims, 1)) { 2947 CTR1(KTR_IGMPV3, "%s: skip non-in-mode", 2948 __func__); 2949 continue; 2950 } 2951 src.s_addr = htonl(ims->ims_haddr); 2952 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr)); 2953 if (retval != 0) 2954 break; 2955 } 2956 } 2957 2958 IN_MULTI_LIST_UNLOCK(); 2959 NET_EPOCH_EXIT(et); 2960 2961 return (retval); 2962 } 2963 2964 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3) 2965 2966 static const char *inm_modestrs[] = { 2967 [MCAST_UNDEFINED] = "un", 2968 [MCAST_INCLUDE] = "in", 2969 [MCAST_EXCLUDE] = "ex", 2970 }; 2971 _Static_assert(MCAST_UNDEFINED == 0 && 2972 MCAST_EXCLUDE + 1 == nitems(inm_modestrs), 2973 "inm_modestrs: no longer matches #defines"); 2974 2975 static const char * 2976 inm_mode_str(const int mode) 2977 { 2978 2979 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 2980 return (inm_modestrs[mode]); 2981 return ("??"); 2982 } 2983 2984 static const char *inm_statestrs[] = { 2985 [IGMP_NOT_MEMBER] = "not-member", 2986 [IGMP_SILENT_MEMBER] = "silent", 2987 [IGMP_REPORTING_MEMBER] = "reporting", 2988 [IGMP_IDLE_MEMBER] = "idle", 2989 [IGMP_LAZY_MEMBER] = "lazy", 2990 [IGMP_SLEEPING_MEMBER] = "sleeping", 2991 [IGMP_AWAKENING_MEMBER] = "awakening", 2992 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending", 2993 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending", 2994 [IGMP_LEAVING_MEMBER] = "leaving", 2995 }; 2996 _Static_assert(IGMP_NOT_MEMBER == 0 && 2997 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs), 2998 "inm_statetrs: no longer matches #defines"); 2999 3000 static const char * 3001 inm_state_str(const int state) 3002 { 3003 3004 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER) 3005 return (inm_statestrs[state]); 3006 return ("??"); 3007 } 3008 3009 /* 3010 * Dump an in_multi structure to the console. 3011 */ 3012 void 3013 inm_print(const struct in_multi *inm) 3014 { 3015 int t; 3016 char addrbuf[INET_ADDRSTRLEN]; 3017 3018 if ((ktr_mask & KTR_IGMPV3) == 0) 3019 return; 3020 3021 printf("%s: --- begin inm %p ---\n", __func__, inm); 3022 printf("addr %s ifp %p(%s) ifma %p\n", 3023 inet_ntoa_r(inm->inm_addr, addrbuf), 3024 inm->inm_ifp, 3025 inm->inm_ifp->if_xname, 3026 inm->inm_ifma); 3027 printf("timer %u state %s refcount %u scq.len %u\n", 3028 inm->inm_timer, 3029 inm_state_str(inm->inm_state), 3030 inm->inm_refcount, 3031 inm->inm_scq.mq_len); 3032 printf("igi %p nsrc %lu sctimer %u scrv %u\n", 3033 inm->inm_igi, 3034 inm->inm_nsrc, 3035 inm->inm_sctimer, 3036 inm->inm_scrv); 3037 for (t = 0; t < 2; t++) { 3038 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 3039 inm_mode_str(inm->inm_st[t].iss_fmode), 3040 inm->inm_st[t].iss_asm, 3041 inm->inm_st[t].iss_ex, 3042 inm->inm_st[t].iss_in, 3043 inm->inm_st[t].iss_rec); 3044 } 3045 printf("%s: --- end inm %p ---\n", __func__, inm); 3046 } 3047 3048 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */ 3049 3050 void 3051 inm_print(const struct in_multi *inm) 3052 { 3053 3054 } 3055 3056 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */ 3057 3058 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp); 3059