1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2007-2009 Bruce Simpson. 5 * Copyright (c) 2005 Robert N. M. Watson. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. The name of the author may not be used to endorse or promote 17 * products derived from this software without specific prior written 18 * permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 */ 32 33 /* 34 * IPv4 multicast socket, group, and socket option processing module. 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include <sys/param.h> 41 #include <sys/systm.h> 42 #include <sys/kernel.h> 43 #include <sys/lock.h> 44 #include <sys/malloc.h> 45 #include <sys/mbuf.h> 46 #include <sys/protosw.h> 47 #include <sys/rmlock.h> 48 #include <sys/socket.h> 49 #include <sys/socketvar.h> 50 #include <sys/protosw.h> 51 #include <sys/sysctl.h> 52 #include <sys/ktr.h> 53 #include <sys/taskqueue.h> 54 #include <sys/tree.h> 55 56 #include <net/if.h> 57 #include <net/if_var.h> 58 #include <net/if_dl.h> 59 #include <net/route.h> 60 #include <net/route/nhop.h> 61 #include <net/vnet.h> 62 63 #include <net/ethernet.h> 64 65 #include <netinet/in.h> 66 #include <netinet/in_systm.h> 67 #include <netinet/in_fib.h> 68 #include <netinet/in_pcb.h> 69 #include <netinet/in_var.h> 70 #include <netinet/ip_var.h> 71 #include <netinet/igmp_var.h> 72 73 #ifndef KTR_IGMPV3 74 #define KTR_IGMPV3 KTR_INET 75 #endif 76 77 #ifndef __SOCKUNION_DECLARED 78 union sockunion { 79 struct sockaddr_storage ss; 80 struct sockaddr sa; 81 struct sockaddr_dl sdl; 82 struct sockaddr_in sin; 83 }; 84 typedef union sockunion sockunion_t; 85 #define __SOCKUNION_DECLARED 86 #endif /* __SOCKUNION_DECLARED */ 87 88 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter", 89 "IPv4 multicast PCB-layer source filter"); 90 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group"); 91 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options"); 92 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource", 93 "IPv4 multicast IGMP-layer source filter"); 94 95 /* 96 * Locking: 97 * 98 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK, 99 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 100 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however 101 * it can be taken by code in net/if.c also. 102 * - ip_moptions and in_mfilter are covered by the INP_WLOCK. 103 * 104 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly 105 * any need for in_multi itself to be virtualized -- it is bound to an ifp 106 * anyway no matter what happens. 107 */ 108 struct mtx in_multi_list_mtx; 109 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF); 110 111 struct mtx in_multi_free_mtx; 112 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF); 113 114 struct sx in_multi_sx; 115 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx"); 116 117 int ifma_restart; 118 119 /* 120 * Functions with non-static linkage defined in this file should be 121 * declared in in_var.h: 122 * imo_multi_filter() 123 * in_addmulti() 124 * in_delmulti() 125 * in_joingroup() 126 * in_joingroup_locked() 127 * in_leavegroup() 128 * in_leavegroup_locked() 129 * and ip_var.h: 130 * inp_freemoptions() 131 * inp_getmoptions() 132 * inp_setmoptions() 133 * 134 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti() 135 * and in_delmulti(). 136 */ 137 static void imf_commit(struct in_mfilter *); 138 static int imf_get_source(struct in_mfilter *imf, 139 const struct sockaddr_in *psin, 140 struct in_msource **); 141 static struct in_msource * 142 imf_graft(struct in_mfilter *, const uint8_t, 143 const struct sockaddr_in *); 144 static void imf_leave(struct in_mfilter *); 145 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *); 146 static void imf_purge(struct in_mfilter *); 147 static void imf_rollback(struct in_mfilter *); 148 static void imf_reap(struct in_mfilter *); 149 static struct in_mfilter * 150 imo_match_group(const struct ip_moptions *, 151 const struct ifnet *, const struct sockaddr *); 152 static struct in_msource * 153 imo_match_source(struct in_mfilter *, const struct sockaddr *); 154 static void ims_merge(struct ip_msource *ims, 155 const struct in_msource *lims, const int rollback); 156 static int in_getmulti(struct ifnet *, const struct in_addr *, 157 struct in_multi **); 158 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr, 159 const int noalloc, struct ip_msource **pims); 160 #ifdef KTR 161 static int inm_is_ifp_detached(const struct in_multi *); 162 #endif 163 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *); 164 static void inm_purge(struct in_multi *); 165 static void inm_reap(struct in_multi *); 166 static void inm_release(struct in_multi *); 167 static struct ip_moptions * 168 inp_findmoptions(struct inpcb *); 169 static int inp_get_source_filters(struct inpcb *, struct sockopt *); 170 static int inp_join_group(struct inpcb *, struct sockopt *); 171 static int inp_leave_group(struct inpcb *, struct sockopt *); 172 static struct ifnet * 173 inp_lookup_mcast_ifp(const struct inpcb *, 174 const struct sockaddr_in *, const struct in_addr); 175 static int inp_block_unblock_source(struct inpcb *, struct sockopt *); 176 static int inp_set_multicast_if(struct inpcb *, struct sockopt *); 177 static int inp_set_source_filters(struct inpcb *, struct sockopt *); 178 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS); 179 180 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, 181 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 182 "IPv4 multicast"); 183 184 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER; 185 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc, 186 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0, 187 "Max source filters per group"); 188 189 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER; 190 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc, 191 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0, 192 "Max source filters per socket"); 193 194 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP; 195 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN, 196 &in_mcast_loop, 0, "Loopback multicast datagrams by default"); 197 198 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters, 199 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters, 200 "Per-interface stack-wide source filters"); 201 202 #ifdef KTR 203 /* 204 * Inline function which wraps assertions for a valid ifp. 205 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 206 * is detached. 207 */ 208 static int __inline 209 inm_is_ifp_detached(const struct in_multi *inm) 210 { 211 struct ifnet *ifp; 212 213 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 214 ifp = inm->inm_ifma->ifma_ifp; 215 if (ifp != NULL) { 216 /* 217 * Sanity check that netinet's notion of ifp is the 218 * same as net's. 219 */ 220 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 221 } 222 223 return (ifp == NULL); 224 } 225 #endif 226 227 /* 228 * Interface detach can happen in a taskqueue thread context, so we must use a 229 * dedicated thread to avoid deadlocks when draining inm_release tasks. 230 */ 231 TASKQUEUE_DEFINE_THREAD(inm_free); 232 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER(); 233 static void inm_release_task(void *arg __unused, int pending __unused); 234 static struct task inm_free_task = TASK_INITIALIZER(0, inm_release_task, NULL); 235 236 void 237 inm_release_wait(void *arg __unused) 238 { 239 240 /* 241 * Make sure all pending multicast addresses are freed before 242 * the VNET or network device is destroyed: 243 */ 244 taskqueue_drain(taskqueue_inm_free, &inm_free_task); 245 } 246 #ifdef VIMAGE 247 /* XXX-BZ FIXME, see D24914. */ 248 VNET_SYSUNINIT(inm_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, inm_release_wait, NULL); 249 #endif 250 251 void 252 inm_release_list_deferred(struct in_multi_head *inmh) 253 { 254 255 if (SLIST_EMPTY(inmh)) 256 return; 257 mtx_lock(&in_multi_free_mtx); 258 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele); 259 mtx_unlock(&in_multi_free_mtx); 260 taskqueue_enqueue(taskqueue_inm_free, &inm_free_task); 261 } 262 263 void 264 inm_disconnect(struct in_multi *inm) 265 { 266 struct ifnet *ifp; 267 struct ifmultiaddr *ifma, *ll_ifma; 268 269 ifp = inm->inm_ifp; 270 IF_ADDR_WLOCK_ASSERT(ifp); 271 ifma = inm->inm_ifma; 272 273 if_ref(ifp); 274 if (ifma->ifma_flags & IFMA_F_ENQUEUED) { 275 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 276 ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 277 } 278 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname); 279 if ((ll_ifma = ifma->ifma_llifma) != NULL) { 280 MPASS(ifma != ll_ifma); 281 ifma->ifma_llifma = NULL; 282 MPASS(ll_ifma->ifma_llifma == NULL); 283 MPASS(ll_ifma->ifma_ifp == ifp); 284 if (--ll_ifma->ifma_refcount == 0) { 285 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) { 286 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link); 287 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 288 } 289 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname); 290 if_freemulti(ll_ifma); 291 ifma_restart = true; 292 } 293 } 294 } 295 296 void 297 inm_release_deferred(struct in_multi *inm) 298 { 299 struct in_multi_head tmp; 300 301 IN_MULTI_LIST_LOCK_ASSERT(); 302 MPASS(inm->inm_refcount > 0); 303 if (--inm->inm_refcount == 0) { 304 SLIST_INIT(&tmp); 305 inm_disconnect(inm); 306 inm->inm_ifma->ifma_protospec = NULL; 307 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele); 308 inm_release_list_deferred(&tmp); 309 } 310 } 311 312 static void 313 inm_release_task(void *arg __unused, int pending __unused) 314 { 315 struct in_multi_head inm_free_tmp; 316 struct in_multi *inm, *tinm; 317 318 SLIST_INIT(&inm_free_tmp); 319 mtx_lock(&in_multi_free_mtx); 320 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele); 321 mtx_unlock(&in_multi_free_mtx); 322 IN_MULTI_LOCK(); 323 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) { 324 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele); 325 MPASS(inm); 326 inm_release(inm); 327 } 328 IN_MULTI_UNLOCK(); 329 } 330 331 /* 332 * Initialize an in_mfilter structure to a known state at t0, t1 333 * with an empty source filter list. 334 */ 335 static __inline void 336 imf_init(struct in_mfilter *imf, const int st0, const int st1) 337 { 338 memset(imf, 0, sizeof(struct in_mfilter)); 339 RB_INIT(&imf->imf_sources); 340 imf->imf_st[0] = st0; 341 imf->imf_st[1] = st1; 342 } 343 344 struct in_mfilter * 345 ip_mfilter_alloc(const int mflags, const int st0, const int st1) 346 { 347 struct in_mfilter *imf; 348 349 imf = malloc(sizeof(*imf), M_INMFILTER, mflags); 350 if (imf != NULL) 351 imf_init(imf, st0, st1); 352 353 return (imf); 354 } 355 356 void 357 ip_mfilter_free(struct in_mfilter *imf) 358 { 359 360 imf_purge(imf); 361 free(imf, M_INMFILTER); 362 } 363 364 /* 365 * Function for looking up an in_multi record for an IPv4 multicast address 366 * on a given interface. ifp must be valid. If no record found, return NULL. 367 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held. 368 */ 369 struct in_multi * 370 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina) 371 { 372 struct ifmultiaddr *ifma; 373 struct in_multi *inm; 374 375 IN_MULTI_LIST_LOCK_ASSERT(); 376 IF_ADDR_LOCK_ASSERT(ifp); 377 378 inm = NULL; 379 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) { 380 if (ifma->ifma_addr->sa_family != AF_INET || 381 ifma->ifma_protospec == NULL) 382 continue; 383 inm = (struct in_multi *)ifma->ifma_protospec; 384 if (inm->inm_addr.s_addr == ina.s_addr) 385 break; 386 inm = NULL; 387 } 388 return (inm); 389 } 390 391 /* 392 * Wrapper for inm_lookup_locked(). 393 * The IF_ADDR_LOCK will be taken on ifp and released on return. 394 */ 395 struct in_multi * 396 inm_lookup(struct ifnet *ifp, const struct in_addr ina) 397 { 398 struct epoch_tracker et; 399 struct in_multi *inm; 400 401 IN_MULTI_LIST_LOCK_ASSERT(); 402 NET_EPOCH_ENTER(et); 403 404 inm = inm_lookup_locked(ifp, ina); 405 NET_EPOCH_EXIT(et); 406 407 return (inm); 408 } 409 410 /* 411 * Find an IPv4 multicast group entry for this ip_moptions instance 412 * which matches the specified group, and optionally an interface. 413 * Return its index into the array, or -1 if not found. 414 */ 415 static struct in_mfilter * 416 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp, 417 const struct sockaddr *group) 418 { 419 const struct sockaddr_in *gsin; 420 struct in_mfilter *imf; 421 struct in_multi *inm; 422 423 gsin = (const struct sockaddr_in *)group; 424 425 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 426 inm = imf->imf_inm; 427 if (inm == NULL) 428 continue; 429 if ((ifp == NULL || (inm->inm_ifp == ifp)) && 430 in_hosteq(inm->inm_addr, gsin->sin_addr)) { 431 break; 432 } 433 } 434 return (imf); 435 } 436 437 /* 438 * Find an IPv4 multicast source entry for this imo which matches 439 * the given group index for this socket, and source address. 440 * 441 * NOTE: This does not check if the entry is in-mode, merely if 442 * it exists, which may not be the desired behaviour. 443 */ 444 static struct in_msource * 445 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src) 446 { 447 struct ip_msource find; 448 struct ip_msource *ims; 449 const sockunion_t *psa; 450 451 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__)); 452 453 /* Source trees are keyed in host byte order. */ 454 psa = (const sockunion_t *)src; 455 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr); 456 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 457 458 return ((struct in_msource *)ims); 459 } 460 461 /* 462 * Perform filtering for multicast datagrams on a socket by group and source. 463 * 464 * Returns 0 if a datagram should be allowed through, or various error codes 465 * if the socket was not a member of the group, or the source was muted, etc. 466 */ 467 int 468 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp, 469 const struct sockaddr *group, const struct sockaddr *src) 470 { 471 struct in_mfilter *imf; 472 struct in_msource *ims; 473 int mode; 474 475 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 476 477 imf = imo_match_group(imo, ifp, group); 478 if (imf == NULL) 479 return (MCAST_NOTGMEMBER); 480 481 /* 482 * Check if the source was included in an (S,G) join. 483 * Allow reception on exclusive memberships by default, 484 * reject reception on inclusive memberships by default. 485 * Exclude source only if an in-mode exclude filter exists. 486 * Include source only if an in-mode include filter exists. 487 * NOTE: We are comparing group state here at IGMP t1 (now) 488 * with socket-layer t0 (since last downcall). 489 */ 490 mode = imf->imf_st[1]; 491 ims = imo_match_source(imf, src); 492 493 if ((ims == NULL && mode == MCAST_INCLUDE) || 494 (ims != NULL && ims->imsl_st[0] != mode)) 495 return (MCAST_NOTSMEMBER); 496 497 return (MCAST_PASS); 498 } 499 500 /* 501 * Find and return a reference to an in_multi record for (ifp, group), 502 * and bump its reference count. 503 * If one does not exist, try to allocate it, and update link-layer multicast 504 * filters on ifp to listen for group. 505 * Assumes the IN_MULTI lock is held across the call. 506 * Return 0 if successful, otherwise return an appropriate error code. 507 */ 508 static int 509 in_getmulti(struct ifnet *ifp, const struct in_addr *group, 510 struct in_multi **pinm) 511 { 512 struct sockaddr_in gsin; 513 struct ifmultiaddr *ifma; 514 struct in_ifinfo *ii; 515 struct in_multi *inm; 516 int error; 517 518 IN_MULTI_LOCK_ASSERT(); 519 520 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET]; 521 IN_MULTI_LIST_LOCK(); 522 inm = inm_lookup(ifp, *group); 523 if (inm != NULL) { 524 /* 525 * If we already joined this group, just bump the 526 * refcount and return it. 527 */ 528 KASSERT(inm->inm_refcount >= 1, 529 ("%s: bad refcount %d", __func__, inm->inm_refcount)); 530 inm_acquire_locked(inm); 531 *pinm = inm; 532 } 533 IN_MULTI_LIST_UNLOCK(); 534 if (inm != NULL) 535 return (0); 536 537 memset(&gsin, 0, sizeof(gsin)); 538 gsin.sin_family = AF_INET; 539 gsin.sin_len = sizeof(struct sockaddr_in); 540 gsin.sin_addr = *group; 541 542 /* 543 * Check if a link-layer group is already associated 544 * with this network-layer group on the given ifnet. 545 */ 546 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma); 547 if (error != 0) 548 return (error); 549 550 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */ 551 IN_MULTI_LIST_LOCK(); 552 IF_ADDR_WLOCK(ifp); 553 554 /* 555 * If something other than netinet is occupying the link-layer 556 * group, print a meaningful error message and back out of 557 * the allocation. 558 * Otherwise, bump the refcount on the existing network-layer 559 * group association and return it. 560 */ 561 if (ifma->ifma_protospec != NULL) { 562 inm = (struct in_multi *)ifma->ifma_protospec; 563 #ifdef INVARIANTS 564 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 565 __func__)); 566 KASSERT(ifma->ifma_addr->sa_family == AF_INET, 567 ("%s: ifma not AF_INET", __func__)); 568 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 569 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp || 570 !in_hosteq(inm->inm_addr, *group)) { 571 char addrbuf[INET_ADDRSTRLEN]; 572 573 panic("%s: ifma %p is inconsistent with %p (%s)", 574 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf)); 575 } 576 #endif 577 inm_acquire_locked(inm); 578 *pinm = inm; 579 goto out_locked; 580 } 581 582 IF_ADDR_WLOCK_ASSERT(ifp); 583 584 /* 585 * A new in_multi record is needed; allocate and initialize it. 586 * We DO NOT perform an IGMP join as the in_ layer may need to 587 * push an initial source list down to IGMP to support SSM. 588 * 589 * The initial source filter state is INCLUDE, {} as per the RFC. 590 */ 591 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO); 592 if (inm == NULL) { 593 IF_ADDR_WUNLOCK(ifp); 594 IN_MULTI_LIST_UNLOCK(); 595 if_delmulti_ifma(ifma); 596 return (ENOMEM); 597 } 598 inm->inm_addr = *group; 599 inm->inm_ifp = ifp; 600 inm->inm_igi = ii->ii_igmp; 601 inm->inm_ifma = ifma; 602 inm->inm_refcount = 1; 603 inm->inm_state = IGMP_NOT_MEMBER; 604 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES); 605 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED; 606 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 607 RB_INIT(&inm->inm_srcs); 608 609 ifma->ifma_protospec = inm; 610 611 *pinm = inm; 612 out_locked: 613 IF_ADDR_WUNLOCK(ifp); 614 IN_MULTI_LIST_UNLOCK(); 615 return (0); 616 } 617 618 /* 619 * Drop a reference to an in_multi record. 620 * 621 * If the refcount drops to 0, free the in_multi record and 622 * delete the underlying link-layer membership. 623 */ 624 static void 625 inm_release(struct in_multi *inm) 626 { 627 struct ifmultiaddr *ifma; 628 struct ifnet *ifp; 629 630 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount); 631 MPASS(inm->inm_refcount == 0); 632 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm); 633 634 ifma = inm->inm_ifma; 635 ifp = inm->inm_ifp; 636 637 /* XXX this access is not covered by IF_ADDR_LOCK */ 638 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma); 639 if (ifp != NULL) { 640 CURVNET_SET(ifp->if_vnet); 641 inm_purge(inm); 642 free(inm, M_IPMADDR); 643 if_delmulti_ifma_flags(ifma, 1); 644 CURVNET_RESTORE(); 645 if_rele(ifp); 646 } else { 647 inm_purge(inm); 648 free(inm, M_IPMADDR); 649 if_delmulti_ifma_flags(ifma, 1); 650 } 651 } 652 653 /* 654 * Clear recorded source entries for a group. 655 * Used by the IGMP code. Caller must hold the IN_MULTI lock. 656 * FIXME: Should reap. 657 */ 658 void 659 inm_clear_recorded(struct in_multi *inm) 660 { 661 struct ip_msource *ims; 662 663 IN_MULTI_LIST_LOCK_ASSERT(); 664 665 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 666 if (ims->ims_stp) { 667 ims->ims_stp = 0; 668 --inm->inm_st[1].iss_rec; 669 } 670 } 671 KASSERT(inm->inm_st[1].iss_rec == 0, 672 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec)); 673 } 674 675 /* 676 * Record a source as pending for a Source-Group IGMPv3 query. 677 * This lives here as it modifies the shared tree. 678 * 679 * inm is the group descriptor. 680 * naddr is the address of the source to record in network-byte order. 681 * 682 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will 683 * lazy-allocate a source node in response to an SG query. 684 * Otherwise, no allocation is performed. This saves some memory 685 * with the trade-off that the source will not be reported to the 686 * router if joined in the window between the query response and 687 * the group actually being joined on the local host. 688 * 689 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed. 690 * This turns off the allocation of a recorded source entry if 691 * the group has not been joined. 692 * 693 * Return 0 if the source didn't exist or was already marked as recorded. 694 * Return 1 if the source was marked as recorded by this function. 695 * Return <0 if any error occurred (negated errno code). 696 */ 697 int 698 inm_record_source(struct in_multi *inm, const in_addr_t naddr) 699 { 700 struct ip_msource find; 701 struct ip_msource *ims, *nims; 702 703 IN_MULTI_LIST_LOCK_ASSERT(); 704 705 find.ims_haddr = ntohl(naddr); 706 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 707 if (ims && ims->ims_stp) 708 return (0); 709 if (ims == NULL) { 710 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 711 return (-ENOSPC); 712 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 713 M_NOWAIT | M_ZERO); 714 if (nims == NULL) 715 return (-ENOMEM); 716 nims->ims_haddr = find.ims_haddr; 717 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 718 ++inm->inm_nsrc; 719 ims = nims; 720 } 721 722 /* 723 * Mark the source as recorded and update the recorded 724 * source count. 725 */ 726 ++ims->ims_stp; 727 ++inm->inm_st[1].iss_rec; 728 729 return (1); 730 } 731 732 /* 733 * Return a pointer to an in_msource owned by an in_mfilter, 734 * given its source address. 735 * Lazy-allocate if needed. If this is a new entry its filter state is 736 * undefined at t0. 737 * 738 * imf is the filter set being modified. 739 * haddr is the source address in *host* byte-order. 740 * 741 * SMPng: May be called with locks held; malloc must not block. 742 */ 743 static int 744 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin, 745 struct in_msource **plims) 746 { 747 struct ip_msource find; 748 struct ip_msource *ims, *nims; 749 struct in_msource *lims; 750 int error; 751 752 error = 0; 753 ims = NULL; 754 lims = NULL; 755 756 /* key is host byte order */ 757 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 758 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 759 lims = (struct in_msource *)ims; 760 if (lims == NULL) { 761 if (imf->imf_nsrc == in_mcast_maxsocksrc) 762 return (ENOSPC); 763 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 764 M_NOWAIT | M_ZERO); 765 if (nims == NULL) 766 return (ENOMEM); 767 lims = (struct in_msource *)nims; 768 lims->ims_haddr = find.ims_haddr; 769 lims->imsl_st[0] = MCAST_UNDEFINED; 770 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 771 ++imf->imf_nsrc; 772 } 773 774 *plims = lims; 775 776 return (error); 777 } 778 779 /* 780 * Graft a source entry into an existing socket-layer filter set, 781 * maintaining any required invariants and checking allocations. 782 * 783 * The source is marked as being in the new filter mode at t1. 784 * 785 * Return the pointer to the new node, otherwise return NULL. 786 */ 787 static struct in_msource * 788 imf_graft(struct in_mfilter *imf, const uint8_t st1, 789 const struct sockaddr_in *psin) 790 { 791 struct ip_msource *nims; 792 struct in_msource *lims; 793 794 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 795 M_NOWAIT | M_ZERO); 796 if (nims == NULL) 797 return (NULL); 798 lims = (struct in_msource *)nims; 799 lims->ims_haddr = ntohl(psin->sin_addr.s_addr); 800 lims->imsl_st[0] = MCAST_UNDEFINED; 801 lims->imsl_st[1] = st1; 802 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 803 ++imf->imf_nsrc; 804 805 return (lims); 806 } 807 808 /* 809 * Prune a source entry from an existing socket-layer filter set, 810 * maintaining any required invariants and checking allocations. 811 * 812 * The source is marked as being left at t1, it is not freed. 813 * 814 * Return 0 if no error occurred, otherwise return an errno value. 815 */ 816 static int 817 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin) 818 { 819 struct ip_msource find; 820 struct ip_msource *ims; 821 struct in_msource *lims; 822 823 /* key is host byte order */ 824 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 825 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 826 if (ims == NULL) 827 return (ENOENT); 828 lims = (struct in_msource *)ims; 829 lims->imsl_st[1] = MCAST_UNDEFINED; 830 return (0); 831 } 832 833 /* 834 * Revert socket-layer filter set deltas at t1 to t0 state. 835 */ 836 static void 837 imf_rollback(struct in_mfilter *imf) 838 { 839 struct ip_msource *ims, *tims; 840 struct in_msource *lims; 841 842 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 843 lims = (struct in_msource *)ims; 844 if (lims->imsl_st[0] == lims->imsl_st[1]) { 845 /* no change at t1 */ 846 continue; 847 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) { 848 /* revert change to existing source at t1 */ 849 lims->imsl_st[1] = lims->imsl_st[0]; 850 } else { 851 /* revert source added t1 */ 852 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 853 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 854 free(ims, M_INMFILTER); 855 imf->imf_nsrc--; 856 } 857 } 858 imf->imf_st[1] = imf->imf_st[0]; 859 } 860 861 /* 862 * Mark socket-layer filter set as INCLUDE {} at t1. 863 */ 864 static void 865 imf_leave(struct in_mfilter *imf) 866 { 867 struct ip_msource *ims; 868 struct in_msource *lims; 869 870 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 871 lims = (struct in_msource *)ims; 872 lims->imsl_st[1] = MCAST_UNDEFINED; 873 } 874 imf->imf_st[1] = MCAST_INCLUDE; 875 } 876 877 /* 878 * Mark socket-layer filter set deltas as committed. 879 */ 880 static void 881 imf_commit(struct in_mfilter *imf) 882 { 883 struct ip_msource *ims; 884 struct in_msource *lims; 885 886 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 887 lims = (struct in_msource *)ims; 888 lims->imsl_st[0] = lims->imsl_st[1]; 889 } 890 imf->imf_st[0] = imf->imf_st[1]; 891 } 892 893 /* 894 * Reap unreferenced sources from socket-layer filter set. 895 */ 896 static void 897 imf_reap(struct in_mfilter *imf) 898 { 899 struct ip_msource *ims, *tims; 900 struct in_msource *lims; 901 902 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 903 lims = (struct in_msource *)ims; 904 if ((lims->imsl_st[0] == MCAST_UNDEFINED) && 905 (lims->imsl_st[1] == MCAST_UNDEFINED)) { 906 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims); 907 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 908 free(ims, M_INMFILTER); 909 imf->imf_nsrc--; 910 } 911 } 912 } 913 914 /* 915 * Purge socket-layer filter set. 916 */ 917 static void 918 imf_purge(struct in_mfilter *imf) 919 { 920 struct ip_msource *ims, *tims; 921 922 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 923 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 924 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 925 free(ims, M_INMFILTER); 926 imf->imf_nsrc--; 927 } 928 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED; 929 KASSERT(RB_EMPTY(&imf->imf_sources), 930 ("%s: imf_sources not empty", __func__)); 931 } 932 933 /* 934 * Look up a source filter entry for a multicast group. 935 * 936 * inm is the group descriptor to work with. 937 * haddr is the host-byte-order IPv4 address to look up. 938 * noalloc may be non-zero to suppress allocation of sources. 939 * *pims will be set to the address of the retrieved or allocated source. 940 * 941 * SMPng: NOTE: may be called with locks held. 942 * Return 0 if successful, otherwise return a non-zero error code. 943 */ 944 static int 945 inm_get_source(struct in_multi *inm, const in_addr_t haddr, 946 const int noalloc, struct ip_msource **pims) 947 { 948 struct ip_msource find; 949 struct ip_msource *ims, *nims; 950 951 find.ims_haddr = haddr; 952 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 953 if (ims == NULL && !noalloc) { 954 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 955 return (ENOSPC); 956 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 957 M_NOWAIT | M_ZERO); 958 if (nims == NULL) 959 return (ENOMEM); 960 nims->ims_haddr = haddr; 961 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 962 ++inm->inm_nsrc; 963 ims = nims; 964 #ifdef KTR 965 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__, 966 haddr, ims); 967 #endif 968 } 969 970 *pims = ims; 971 return (0); 972 } 973 974 /* 975 * Merge socket-layer source into IGMP-layer source. 976 * If rollback is non-zero, perform the inverse of the merge. 977 */ 978 static void 979 ims_merge(struct ip_msource *ims, const struct in_msource *lims, 980 const int rollback) 981 { 982 int n = rollback ? -1 : 1; 983 984 if (lims->imsl_st[0] == MCAST_EXCLUDE) { 985 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x", 986 __func__, n, ims->ims_haddr); 987 ims->ims_st[1].ex -= n; 988 } else if (lims->imsl_st[0] == MCAST_INCLUDE) { 989 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x", 990 __func__, n, ims->ims_haddr); 991 ims->ims_st[1].in -= n; 992 } 993 994 if (lims->imsl_st[1] == MCAST_EXCLUDE) { 995 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x", 996 __func__, n, ims->ims_haddr); 997 ims->ims_st[1].ex += n; 998 } else if (lims->imsl_st[1] == MCAST_INCLUDE) { 999 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x", 1000 __func__, n, ims->ims_haddr); 1001 ims->ims_st[1].in += n; 1002 } 1003 } 1004 1005 /* 1006 * Atomically update the global in_multi state, when a membership's 1007 * filter list is being updated in any way. 1008 * 1009 * imf is the per-inpcb-membership group filter pointer. 1010 * A fake imf may be passed for in-kernel consumers. 1011 * 1012 * XXX This is a candidate for a set-symmetric-difference style loop 1013 * which would eliminate the repeated lookup from root of ims nodes, 1014 * as they share the same key space. 1015 * 1016 * If any error occurred this function will back out of refcounts 1017 * and return a non-zero value. 1018 */ 1019 static int 1020 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1021 { 1022 struct ip_msource *ims, *nims; 1023 struct in_msource *lims; 1024 int schanged, error; 1025 int nsrc0, nsrc1; 1026 1027 schanged = 0; 1028 error = 0; 1029 nsrc1 = nsrc0 = 0; 1030 IN_MULTI_LIST_LOCK_ASSERT(); 1031 1032 /* 1033 * Update the source filters first, as this may fail. 1034 * Maintain count of in-mode filters at t0, t1. These are 1035 * used to work out if we transition into ASM mode or not. 1036 * Maintain a count of source filters whose state was 1037 * actually modified by this operation. 1038 */ 1039 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1040 lims = (struct in_msource *)ims; 1041 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++; 1042 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++; 1043 if (lims->imsl_st[0] == lims->imsl_st[1]) continue; 1044 error = inm_get_source(inm, lims->ims_haddr, 0, &nims); 1045 ++schanged; 1046 if (error) 1047 break; 1048 ims_merge(nims, lims, 0); 1049 } 1050 if (error) { 1051 struct ip_msource *bims; 1052 1053 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) { 1054 lims = (struct in_msource *)ims; 1055 if (lims->imsl_st[0] == lims->imsl_st[1]) 1056 continue; 1057 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims); 1058 if (bims == NULL) 1059 continue; 1060 ims_merge(bims, lims, 1); 1061 } 1062 goto out_reap; 1063 } 1064 1065 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1", 1066 __func__, nsrc0, nsrc1); 1067 1068 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 1069 if (imf->imf_st[0] == imf->imf_st[1] && 1070 imf->imf_st[1] == MCAST_INCLUDE) { 1071 if (nsrc1 == 0) { 1072 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1073 --inm->inm_st[1].iss_in; 1074 } 1075 } 1076 1077 /* Handle filter mode transition on socket. */ 1078 if (imf->imf_st[0] != imf->imf_st[1]) { 1079 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d", 1080 __func__, imf->imf_st[0], imf->imf_st[1]); 1081 1082 if (imf->imf_st[0] == MCAST_EXCLUDE) { 1083 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__); 1084 --inm->inm_st[1].iss_ex; 1085 } else if (imf->imf_st[0] == MCAST_INCLUDE) { 1086 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1087 --inm->inm_st[1].iss_in; 1088 } 1089 1090 if (imf->imf_st[1] == MCAST_EXCLUDE) { 1091 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__); 1092 inm->inm_st[1].iss_ex++; 1093 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) { 1094 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__); 1095 inm->inm_st[1].iss_in++; 1096 } 1097 } 1098 1099 /* 1100 * Track inm filter state in terms of listener counts. 1101 * If there are any exclusive listeners, stack-wide 1102 * membership is exclusive. 1103 * Otherwise, if only inclusive listeners, stack-wide is inclusive. 1104 * If no listeners remain, state is undefined at t1, 1105 * and the IGMP lifecycle for this group should finish. 1106 */ 1107 if (inm->inm_st[1].iss_ex > 0) { 1108 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__); 1109 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE; 1110 } else if (inm->inm_st[1].iss_in > 0) { 1111 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__); 1112 inm->inm_st[1].iss_fmode = MCAST_INCLUDE; 1113 } else { 1114 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__); 1115 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 1116 } 1117 1118 /* Decrement ASM listener count on transition out of ASM mode. */ 1119 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1120 if ((imf->imf_st[1] != MCAST_EXCLUDE) || 1121 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) { 1122 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__); 1123 --inm->inm_st[1].iss_asm; 1124 } 1125 } 1126 1127 /* Increment ASM listener count on transition to ASM mode. */ 1128 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1129 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__); 1130 inm->inm_st[1].iss_asm++; 1131 } 1132 1133 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm); 1134 inm_print(inm); 1135 1136 out_reap: 1137 if (schanged > 0) { 1138 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__); 1139 inm_reap(inm); 1140 } 1141 return (error); 1142 } 1143 1144 /* 1145 * Mark an in_multi's filter set deltas as committed. 1146 * Called by IGMP after a state change has been enqueued. 1147 */ 1148 void 1149 inm_commit(struct in_multi *inm) 1150 { 1151 struct ip_msource *ims; 1152 1153 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm); 1154 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__); 1155 inm_print(inm); 1156 1157 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 1158 ims->ims_st[0] = ims->ims_st[1]; 1159 } 1160 inm->inm_st[0] = inm->inm_st[1]; 1161 } 1162 1163 /* 1164 * Reap unreferenced nodes from an in_multi's filter set. 1165 */ 1166 static void 1167 inm_reap(struct in_multi *inm) 1168 { 1169 struct ip_msource *ims, *tims; 1170 1171 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1172 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 || 1173 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 || 1174 ims->ims_stp != 0) 1175 continue; 1176 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1177 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1178 free(ims, M_IPMSOURCE); 1179 inm->inm_nsrc--; 1180 } 1181 } 1182 1183 /* 1184 * Purge all source nodes from an in_multi's filter set. 1185 */ 1186 static void 1187 inm_purge(struct in_multi *inm) 1188 { 1189 struct ip_msource *ims, *tims; 1190 1191 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1192 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1193 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1194 free(ims, M_IPMSOURCE); 1195 inm->inm_nsrc--; 1196 } 1197 } 1198 1199 /* 1200 * Join a multicast group; unlocked entry point. 1201 * 1202 * SMPng: XXX: in_joingroup() is called from in_control() when Giant 1203 * is not held. Fortunately, ifp is unlikely to have been detached 1204 * at this point, so we assume it's OK to recurse. 1205 */ 1206 int 1207 in_joingroup(struct ifnet *ifp, const struct in_addr *gina, 1208 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1209 { 1210 int error; 1211 1212 IN_MULTI_LOCK(); 1213 error = in_joingroup_locked(ifp, gina, imf, pinm); 1214 IN_MULTI_UNLOCK(); 1215 1216 return (error); 1217 } 1218 1219 /* 1220 * Join a multicast group; real entry point. 1221 * 1222 * Only preserves atomicity at inm level. 1223 * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1224 * 1225 * If the IGMP downcall fails, the group is not joined, and an error 1226 * code is returned. 1227 */ 1228 int 1229 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina, 1230 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1231 { 1232 struct in_mfilter timf; 1233 struct in_multi *inm; 1234 int error; 1235 1236 IN_MULTI_LOCK_ASSERT(); 1237 IN_MULTI_LIST_UNLOCK_ASSERT(); 1238 1239 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__, 1240 ntohl(gina->s_addr), ifp, ifp->if_xname); 1241 1242 error = 0; 1243 inm = NULL; 1244 1245 /* 1246 * If no imf was specified (i.e. kernel consumer), 1247 * fake one up and assume it is an ASM join. 1248 */ 1249 if (imf == NULL) { 1250 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1251 imf = &timf; 1252 } 1253 1254 error = in_getmulti(ifp, gina, &inm); 1255 if (error) { 1256 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__); 1257 return (error); 1258 } 1259 IN_MULTI_LIST_LOCK(); 1260 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1261 error = inm_merge(inm, imf); 1262 if (error) { 1263 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1264 goto out_inm_release; 1265 } 1266 1267 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1268 error = igmp_change_state(inm); 1269 if (error) { 1270 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__); 1271 goto out_inm_release; 1272 } 1273 1274 out_inm_release: 1275 if (error) { 1276 1277 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1278 IF_ADDR_WLOCK(ifp); 1279 inm_release_deferred(inm); 1280 IF_ADDR_WUNLOCK(ifp); 1281 } else { 1282 *pinm = inm; 1283 } 1284 IN_MULTI_LIST_UNLOCK(); 1285 1286 return (error); 1287 } 1288 1289 /* 1290 * Leave a multicast group; unlocked entry point. 1291 */ 1292 int 1293 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1294 { 1295 int error; 1296 1297 IN_MULTI_LOCK(); 1298 error = in_leavegroup_locked(inm, imf); 1299 IN_MULTI_UNLOCK(); 1300 1301 return (error); 1302 } 1303 1304 /* 1305 * Leave a multicast group; real entry point. 1306 * All source filters will be expunged. 1307 * 1308 * Only preserves atomicity at inm level. 1309 * 1310 * Holding the write lock for the INP which contains imf 1311 * is highly advisable. We can't assert for it as imf does not 1312 * contain a back-pointer to the owning inp. 1313 * 1314 * Note: This is not the same as inm_release(*) as this function also 1315 * makes a state change downcall into IGMP. 1316 */ 1317 int 1318 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1319 { 1320 struct in_mfilter timf; 1321 int error; 1322 1323 IN_MULTI_LOCK_ASSERT(); 1324 IN_MULTI_LIST_UNLOCK_ASSERT(); 1325 1326 error = 0; 1327 1328 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__, 1329 inm, ntohl(inm->inm_addr.s_addr), 1330 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname), 1331 imf); 1332 1333 /* 1334 * If no imf was specified (i.e. kernel consumer), 1335 * fake one up and assume it is an ASM join. 1336 */ 1337 if (imf == NULL) { 1338 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1339 imf = &timf; 1340 } 1341 1342 /* 1343 * Begin state merge transaction at IGMP layer. 1344 * 1345 * As this particular invocation should not cause any memory 1346 * to be allocated, and there is no opportunity to roll back 1347 * the transaction, it MUST NOT fail. 1348 */ 1349 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1350 IN_MULTI_LIST_LOCK(); 1351 error = inm_merge(inm, imf); 1352 KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1353 1354 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1355 CURVNET_SET(inm->inm_ifp->if_vnet); 1356 error = igmp_change_state(inm); 1357 IF_ADDR_WLOCK(inm->inm_ifp); 1358 inm_release_deferred(inm); 1359 IF_ADDR_WUNLOCK(inm->inm_ifp); 1360 IN_MULTI_LIST_UNLOCK(); 1361 CURVNET_RESTORE(); 1362 if (error) 1363 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1364 1365 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1366 1367 return (error); 1368 } 1369 1370 /*#ifndef BURN_BRIDGES*/ 1371 /* 1372 * Join an IPv4 multicast group in (*,G) exclusive mode. 1373 * The group must be a 224.0.0.0/24 link-scope group. 1374 * This KPI is for legacy kernel consumers only. 1375 */ 1376 struct in_multi * 1377 in_addmulti(struct in_addr *ap, struct ifnet *ifp) 1378 { 1379 struct in_multi *pinm; 1380 int error; 1381 #ifdef INVARIANTS 1382 char addrbuf[INET_ADDRSTRLEN]; 1383 #endif 1384 1385 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)), 1386 ("%s: %s not in 224.0.0.0/24", __func__, 1387 inet_ntoa_r(*ap, addrbuf))); 1388 1389 error = in_joingroup(ifp, ap, NULL, &pinm); 1390 if (error != 0) 1391 pinm = NULL; 1392 1393 return (pinm); 1394 } 1395 1396 /* 1397 * Block or unblock an ASM multicast source on an inpcb. 1398 * This implements the delta-based API described in RFC 3678. 1399 * 1400 * The delta-based API applies only to exclusive-mode memberships. 1401 * An IGMP downcall will be performed. 1402 * 1403 * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1404 * 1405 * Return 0 if successful, otherwise return an appropriate error code. 1406 */ 1407 static int 1408 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1409 { 1410 struct group_source_req gsr; 1411 struct rm_priotracker in_ifa_tracker; 1412 sockunion_t *gsa, *ssa; 1413 struct ifnet *ifp; 1414 struct in_mfilter *imf; 1415 struct ip_moptions *imo; 1416 struct in_msource *ims; 1417 struct in_multi *inm; 1418 uint16_t fmode; 1419 int error, doblock; 1420 1421 ifp = NULL; 1422 error = 0; 1423 doblock = 0; 1424 1425 memset(&gsr, 0, sizeof(struct group_source_req)); 1426 gsa = (sockunion_t *)&gsr.gsr_group; 1427 ssa = (sockunion_t *)&gsr.gsr_source; 1428 1429 switch (sopt->sopt_name) { 1430 case IP_BLOCK_SOURCE: 1431 case IP_UNBLOCK_SOURCE: { 1432 struct ip_mreq_source mreqs; 1433 1434 error = sooptcopyin(sopt, &mreqs, 1435 sizeof(struct ip_mreq_source), 1436 sizeof(struct ip_mreq_source)); 1437 if (error) 1438 return (error); 1439 1440 gsa->sin.sin_family = AF_INET; 1441 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1442 gsa->sin.sin_addr = mreqs.imr_multiaddr; 1443 1444 ssa->sin.sin_family = AF_INET; 1445 ssa->sin.sin_len = sizeof(struct sockaddr_in); 1446 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1447 1448 if (!in_nullhost(mreqs.imr_interface)) { 1449 IN_IFADDR_RLOCK(&in_ifa_tracker); 1450 INADDR_TO_IFP(mreqs.imr_interface, ifp); 1451 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1452 } 1453 if (sopt->sopt_name == IP_BLOCK_SOURCE) 1454 doblock = 1; 1455 1456 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 1457 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 1458 break; 1459 } 1460 1461 case MCAST_BLOCK_SOURCE: 1462 case MCAST_UNBLOCK_SOURCE: 1463 error = sooptcopyin(sopt, &gsr, 1464 sizeof(struct group_source_req), 1465 sizeof(struct group_source_req)); 1466 if (error) 1467 return (error); 1468 1469 if (gsa->sin.sin_family != AF_INET || 1470 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 1471 return (EINVAL); 1472 1473 if (ssa->sin.sin_family != AF_INET || 1474 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 1475 return (EINVAL); 1476 1477 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1478 return (EADDRNOTAVAIL); 1479 1480 ifp = ifnet_byindex(gsr.gsr_interface); 1481 1482 if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1483 doblock = 1; 1484 break; 1485 1486 default: 1487 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1488 __func__, sopt->sopt_name); 1489 return (EOPNOTSUPP); 1490 break; 1491 } 1492 1493 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1494 return (EINVAL); 1495 1496 IN_MULTI_LOCK(); 1497 1498 /* 1499 * Check if we are actually a member of this group. 1500 */ 1501 imo = inp_findmoptions(inp); 1502 imf = imo_match_group(imo, ifp, &gsa->sa); 1503 if (imf == NULL) { 1504 error = EADDRNOTAVAIL; 1505 goto out_inp_locked; 1506 } 1507 inm = imf->imf_inm; 1508 1509 /* 1510 * Attempting to use the delta-based API on an 1511 * non exclusive-mode membership is an error. 1512 */ 1513 fmode = imf->imf_st[0]; 1514 if (fmode != MCAST_EXCLUDE) { 1515 error = EINVAL; 1516 goto out_inp_locked; 1517 } 1518 1519 /* 1520 * Deal with error cases up-front: 1521 * Asked to block, but already blocked; or 1522 * Asked to unblock, but nothing to unblock. 1523 * If adding a new block entry, allocate it. 1524 */ 1525 ims = imo_match_source(imf, &ssa->sa); 1526 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1527 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__, 1528 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not "); 1529 error = EADDRNOTAVAIL; 1530 goto out_inp_locked; 1531 } 1532 1533 INP_WLOCK_ASSERT(inp); 1534 1535 /* 1536 * Begin state merge transaction at socket layer. 1537 */ 1538 if (doblock) { 1539 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 1540 ims = imf_graft(imf, fmode, &ssa->sin); 1541 if (ims == NULL) 1542 error = ENOMEM; 1543 } else { 1544 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 1545 error = imf_prune(imf, &ssa->sin); 1546 } 1547 1548 if (error) { 1549 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__); 1550 goto out_imf_rollback; 1551 } 1552 1553 /* 1554 * Begin state merge transaction at IGMP layer. 1555 */ 1556 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1557 IN_MULTI_LIST_LOCK(); 1558 error = inm_merge(inm, imf); 1559 if (error) { 1560 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1561 IN_MULTI_LIST_UNLOCK(); 1562 goto out_imf_rollback; 1563 } 1564 1565 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1566 error = igmp_change_state(inm); 1567 IN_MULTI_LIST_UNLOCK(); 1568 if (error) 1569 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1570 1571 out_imf_rollback: 1572 if (error) 1573 imf_rollback(imf); 1574 else 1575 imf_commit(imf); 1576 1577 imf_reap(imf); 1578 1579 out_inp_locked: 1580 INP_WUNLOCK(inp); 1581 IN_MULTI_UNLOCK(); 1582 return (error); 1583 } 1584 1585 /* 1586 * Given an inpcb, return its multicast options structure pointer. Accepts 1587 * an unlocked inpcb pointer, but will return it locked. May sleep. 1588 * 1589 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1590 * SMPng: NOTE: Returns with the INP write lock held. 1591 */ 1592 static struct ip_moptions * 1593 inp_findmoptions(struct inpcb *inp) 1594 { 1595 struct ip_moptions *imo; 1596 1597 INP_WLOCK(inp); 1598 if (inp->inp_moptions != NULL) 1599 return (inp->inp_moptions); 1600 1601 INP_WUNLOCK(inp); 1602 1603 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK); 1604 1605 imo->imo_multicast_ifp = NULL; 1606 imo->imo_multicast_addr.s_addr = INADDR_ANY; 1607 imo->imo_multicast_vif = -1; 1608 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL; 1609 imo->imo_multicast_loop = in_mcast_loop; 1610 STAILQ_INIT(&imo->imo_head); 1611 1612 INP_WLOCK(inp); 1613 if (inp->inp_moptions != NULL) { 1614 free(imo, M_IPMOPTS); 1615 return (inp->inp_moptions); 1616 } 1617 inp->inp_moptions = imo; 1618 return (imo); 1619 } 1620 1621 static void 1622 inp_gcmoptions(struct ip_moptions *imo) 1623 { 1624 struct in_mfilter *imf; 1625 struct in_multi *inm; 1626 struct ifnet *ifp; 1627 1628 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) { 1629 ip_mfilter_remove(&imo->imo_head, imf); 1630 1631 imf_leave(imf); 1632 if ((inm = imf->imf_inm) != NULL) { 1633 if ((ifp = inm->inm_ifp) != NULL) { 1634 CURVNET_SET(ifp->if_vnet); 1635 (void)in_leavegroup(inm, imf); 1636 CURVNET_RESTORE(); 1637 } else { 1638 (void)in_leavegroup(inm, imf); 1639 } 1640 } 1641 ip_mfilter_free(imf); 1642 } 1643 free(imo, M_IPMOPTS); 1644 } 1645 1646 /* 1647 * Discard the IP multicast options (and source filters). To minimize 1648 * the amount of work done while holding locks such as the INP's 1649 * pcbinfo lock (which is used in the receive path), the free 1650 * operation is deferred to the epoch callback task. 1651 */ 1652 void 1653 inp_freemoptions(struct ip_moptions *imo) 1654 { 1655 if (imo == NULL) 1656 return; 1657 inp_gcmoptions(imo); 1658 } 1659 1660 /* 1661 * Atomically get source filters on a socket for an IPv4 multicast group. 1662 * Called with INP lock held; returns with lock released. 1663 */ 1664 static int 1665 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1666 { 1667 struct __msfilterreq msfr; 1668 sockunion_t *gsa; 1669 struct ifnet *ifp; 1670 struct ip_moptions *imo; 1671 struct in_mfilter *imf; 1672 struct ip_msource *ims; 1673 struct in_msource *lims; 1674 struct sockaddr_in *psin; 1675 struct sockaddr_storage *ptss; 1676 struct sockaddr_storage *tss; 1677 int error; 1678 size_t nsrcs, ncsrcs; 1679 1680 INP_WLOCK_ASSERT(inp); 1681 1682 imo = inp->inp_moptions; 1683 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__)); 1684 1685 INP_WUNLOCK(inp); 1686 1687 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1688 sizeof(struct __msfilterreq)); 1689 if (error) 1690 return (error); 1691 1692 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 1693 return (EINVAL); 1694 1695 ifp = ifnet_byindex(msfr.msfr_ifindex); 1696 if (ifp == NULL) 1697 return (EINVAL); 1698 1699 INP_WLOCK(inp); 1700 1701 /* 1702 * Lookup group on the socket. 1703 */ 1704 gsa = (sockunion_t *)&msfr.msfr_group; 1705 imf = imo_match_group(imo, ifp, &gsa->sa); 1706 if (imf == NULL) { 1707 INP_WUNLOCK(inp); 1708 return (EADDRNOTAVAIL); 1709 } 1710 1711 /* 1712 * Ignore memberships which are in limbo. 1713 */ 1714 if (imf->imf_st[1] == MCAST_UNDEFINED) { 1715 INP_WUNLOCK(inp); 1716 return (EAGAIN); 1717 } 1718 msfr.msfr_fmode = imf->imf_st[1]; 1719 1720 /* 1721 * If the user specified a buffer, copy out the source filter 1722 * entries to userland gracefully. 1723 * We only copy out the number of entries which userland 1724 * has asked for, but we always tell userland how big the 1725 * buffer really needs to be. 1726 */ 1727 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 1728 msfr.msfr_nsrcs = in_mcast_maxsocksrc; 1729 tss = NULL; 1730 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1731 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1732 M_TEMP, M_NOWAIT | M_ZERO); 1733 if (tss == NULL) { 1734 INP_WUNLOCK(inp); 1735 return (ENOBUFS); 1736 } 1737 } 1738 1739 /* 1740 * Count number of sources in-mode at t0. 1741 * If buffer space exists and remains, copy out source entries. 1742 */ 1743 nsrcs = msfr.msfr_nsrcs; 1744 ncsrcs = 0; 1745 ptss = tss; 1746 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1747 lims = (struct in_msource *)ims; 1748 if (lims->imsl_st[0] == MCAST_UNDEFINED || 1749 lims->imsl_st[0] != imf->imf_st[0]) 1750 continue; 1751 ++ncsrcs; 1752 if (tss != NULL && nsrcs > 0) { 1753 psin = (struct sockaddr_in *)ptss; 1754 psin->sin_family = AF_INET; 1755 psin->sin_len = sizeof(struct sockaddr_in); 1756 psin->sin_addr.s_addr = htonl(lims->ims_haddr); 1757 psin->sin_port = 0; 1758 ++ptss; 1759 --nsrcs; 1760 } 1761 } 1762 1763 INP_WUNLOCK(inp); 1764 1765 if (tss != NULL) { 1766 error = copyout(tss, msfr.msfr_srcs, 1767 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1768 free(tss, M_TEMP); 1769 if (error) 1770 return (error); 1771 } 1772 1773 msfr.msfr_nsrcs = ncsrcs; 1774 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1775 1776 return (error); 1777 } 1778 1779 /* 1780 * Return the IP multicast options in response to user getsockopt(). 1781 */ 1782 int 1783 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1784 { 1785 struct rm_priotracker in_ifa_tracker; 1786 struct ip_mreqn mreqn; 1787 struct ip_moptions *imo; 1788 struct ifnet *ifp; 1789 struct in_ifaddr *ia; 1790 int error, optval; 1791 u_char coptval; 1792 1793 INP_WLOCK(inp); 1794 imo = inp->inp_moptions; 1795 /* 1796 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 1797 * or is a divert socket, reject it. 1798 */ 1799 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 1800 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1801 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) { 1802 INP_WUNLOCK(inp); 1803 return (EOPNOTSUPP); 1804 } 1805 1806 error = 0; 1807 switch (sopt->sopt_name) { 1808 case IP_MULTICAST_VIF: 1809 if (imo != NULL) 1810 optval = imo->imo_multicast_vif; 1811 else 1812 optval = -1; 1813 INP_WUNLOCK(inp); 1814 error = sooptcopyout(sopt, &optval, sizeof(int)); 1815 break; 1816 1817 case IP_MULTICAST_IF: 1818 memset(&mreqn, 0, sizeof(struct ip_mreqn)); 1819 if (imo != NULL) { 1820 ifp = imo->imo_multicast_ifp; 1821 if (!in_nullhost(imo->imo_multicast_addr)) { 1822 mreqn.imr_address = imo->imo_multicast_addr; 1823 } else if (ifp != NULL) { 1824 struct epoch_tracker et; 1825 1826 mreqn.imr_ifindex = ifp->if_index; 1827 NET_EPOCH_ENTER(et); 1828 IFP_TO_IA(ifp, ia, &in_ifa_tracker); 1829 if (ia != NULL) 1830 mreqn.imr_address = 1831 IA_SIN(ia)->sin_addr; 1832 NET_EPOCH_EXIT(et); 1833 } 1834 } 1835 INP_WUNLOCK(inp); 1836 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 1837 error = sooptcopyout(sopt, &mreqn, 1838 sizeof(struct ip_mreqn)); 1839 } else { 1840 error = sooptcopyout(sopt, &mreqn.imr_address, 1841 sizeof(struct in_addr)); 1842 } 1843 break; 1844 1845 case IP_MULTICAST_TTL: 1846 if (imo == NULL) 1847 optval = coptval = IP_DEFAULT_MULTICAST_TTL; 1848 else 1849 optval = coptval = imo->imo_multicast_ttl; 1850 INP_WUNLOCK(inp); 1851 if (sopt->sopt_valsize == sizeof(u_char)) 1852 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1853 else 1854 error = sooptcopyout(sopt, &optval, sizeof(int)); 1855 break; 1856 1857 case IP_MULTICAST_LOOP: 1858 if (imo == NULL) 1859 optval = coptval = IP_DEFAULT_MULTICAST_LOOP; 1860 else 1861 optval = coptval = imo->imo_multicast_loop; 1862 INP_WUNLOCK(inp); 1863 if (sopt->sopt_valsize == sizeof(u_char)) 1864 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1865 else 1866 error = sooptcopyout(sopt, &optval, sizeof(int)); 1867 break; 1868 1869 case IP_MSFILTER: 1870 if (imo == NULL) { 1871 error = EADDRNOTAVAIL; 1872 INP_WUNLOCK(inp); 1873 } else { 1874 error = inp_get_source_filters(inp, sopt); 1875 } 1876 break; 1877 1878 default: 1879 INP_WUNLOCK(inp); 1880 error = ENOPROTOOPT; 1881 break; 1882 } 1883 1884 INP_UNLOCK_ASSERT(inp); 1885 1886 return (error); 1887 } 1888 1889 /* 1890 * Look up the ifnet to use for a multicast group membership, 1891 * given the IPv4 address of an interface, and the IPv4 group address. 1892 * 1893 * This routine exists to support legacy multicast applications 1894 * which do not understand that multicast memberships are scoped to 1895 * specific physical links in the networking stack, or which need 1896 * to join link-scope groups before IPv4 addresses are configured. 1897 * 1898 * If inp is non-NULL, use this socket's current FIB number for any 1899 * required FIB lookup. 1900 * If ina is INADDR_ANY, look up the group address in the unicast FIB, 1901 * and use its ifp; usually, this points to the default next-hop. 1902 * 1903 * If the FIB lookup fails, attempt to use the first non-loopback 1904 * interface with multicast capability in the system as a 1905 * last resort. The legacy IPv4 ASM API requires that we do 1906 * this in order to allow groups to be joined when the routing 1907 * table has not yet been populated during boot. 1908 * 1909 * Returns NULL if no ifp could be found. 1910 * 1911 * FUTURE: Implement IPv4 source-address selection. 1912 */ 1913 static struct ifnet * 1914 inp_lookup_mcast_ifp(const struct inpcb *inp, 1915 const struct sockaddr_in *gsin, const struct in_addr ina) 1916 { 1917 struct rm_priotracker in_ifa_tracker; 1918 struct ifnet *ifp; 1919 struct nhop_object *nh; 1920 uint32_t fibnum; 1921 1922 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__)); 1923 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)), 1924 ("%s: not multicast", __func__)); 1925 1926 ifp = NULL; 1927 if (!in_nullhost(ina)) { 1928 IN_IFADDR_RLOCK(&in_ifa_tracker); 1929 INADDR_TO_IFP(ina, ifp); 1930 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1931 } else { 1932 fibnum = inp ? inp->inp_inc.inc_fibnum : 0; 1933 nh = fib4_lookup(fibnum, gsin->sin_addr, 0, 0, 0); 1934 if (nh != NULL) 1935 ifp = nh->nh_ifp; 1936 else { 1937 struct in_ifaddr *ia; 1938 struct ifnet *mifp; 1939 1940 mifp = NULL; 1941 IN_IFADDR_RLOCK(&in_ifa_tracker); 1942 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1943 mifp = ia->ia_ifp; 1944 if (!(mifp->if_flags & IFF_LOOPBACK) && 1945 (mifp->if_flags & IFF_MULTICAST)) { 1946 ifp = mifp; 1947 break; 1948 } 1949 } 1950 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1951 } 1952 } 1953 1954 return (ifp); 1955 } 1956 1957 /* 1958 * Join an IPv4 multicast group, possibly with a source. 1959 */ 1960 static int 1961 inp_join_group(struct inpcb *inp, struct sockopt *sopt) 1962 { 1963 struct group_source_req gsr; 1964 sockunion_t *gsa, *ssa; 1965 struct ifnet *ifp; 1966 struct in_mfilter *imf; 1967 struct ip_moptions *imo; 1968 struct in_multi *inm; 1969 struct in_msource *lims; 1970 int error, is_new; 1971 1972 ifp = NULL; 1973 lims = NULL; 1974 error = 0; 1975 1976 memset(&gsr, 0, sizeof(struct group_source_req)); 1977 gsa = (sockunion_t *)&gsr.gsr_group; 1978 gsa->ss.ss_family = AF_UNSPEC; 1979 ssa = (sockunion_t *)&gsr.gsr_source; 1980 ssa->ss.ss_family = AF_UNSPEC; 1981 1982 switch (sopt->sopt_name) { 1983 case IP_ADD_MEMBERSHIP: { 1984 struct ip_mreqn mreqn; 1985 1986 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) 1987 error = sooptcopyin(sopt, &mreqn, 1988 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn)); 1989 else 1990 error = sooptcopyin(sopt, &mreqn, 1991 sizeof(struct ip_mreq), sizeof(struct ip_mreq)); 1992 if (error) 1993 return (error); 1994 1995 gsa->sin.sin_family = AF_INET; 1996 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1997 gsa->sin.sin_addr = mreqn.imr_multiaddr; 1998 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1999 return (EINVAL); 2000 2001 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) && 2002 mreqn.imr_ifindex != 0) 2003 ifp = ifnet_byindex(mreqn.imr_ifindex); 2004 else 2005 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 2006 mreqn.imr_address); 2007 break; 2008 } 2009 case IP_ADD_SOURCE_MEMBERSHIP: { 2010 struct ip_mreq_source mreqs; 2011 2012 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source), 2013 sizeof(struct ip_mreq_source)); 2014 if (error) 2015 return (error); 2016 2017 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET; 2018 gsa->sin.sin_len = ssa->sin.sin_len = 2019 sizeof(struct sockaddr_in); 2020 2021 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2022 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2023 return (EINVAL); 2024 2025 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2026 2027 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 2028 mreqs.imr_interface); 2029 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2030 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2031 break; 2032 } 2033 2034 case MCAST_JOIN_GROUP: 2035 case MCAST_JOIN_SOURCE_GROUP: 2036 if (sopt->sopt_name == MCAST_JOIN_GROUP) { 2037 error = sooptcopyin(sopt, &gsr, 2038 sizeof(struct group_req), 2039 sizeof(struct group_req)); 2040 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2041 error = sooptcopyin(sopt, &gsr, 2042 sizeof(struct group_source_req), 2043 sizeof(struct group_source_req)); 2044 } 2045 if (error) 2046 return (error); 2047 2048 if (gsa->sin.sin_family != AF_INET || 2049 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2050 return (EINVAL); 2051 2052 /* 2053 * Overwrite the port field if present, as the sockaddr 2054 * being copied in may be matched with a binary comparison. 2055 */ 2056 gsa->sin.sin_port = 0; 2057 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2058 if (ssa->sin.sin_family != AF_INET || 2059 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2060 return (EINVAL); 2061 ssa->sin.sin_port = 0; 2062 } 2063 2064 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2065 return (EINVAL); 2066 2067 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2068 return (EADDRNOTAVAIL); 2069 ifp = ifnet_byindex(gsr.gsr_interface); 2070 break; 2071 2072 default: 2073 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2074 __func__, sopt->sopt_name); 2075 return (EOPNOTSUPP); 2076 break; 2077 } 2078 2079 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) 2080 return (EADDRNOTAVAIL); 2081 2082 IN_MULTI_LOCK(); 2083 2084 /* 2085 * Find the membership in the membership list. 2086 */ 2087 imo = inp_findmoptions(inp); 2088 imf = imo_match_group(imo, ifp, &gsa->sa); 2089 if (imf == NULL) { 2090 is_new = 1; 2091 inm = NULL; 2092 2093 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) { 2094 error = ENOMEM; 2095 goto out_inp_locked; 2096 } 2097 } else { 2098 is_new = 0; 2099 inm = imf->imf_inm; 2100 2101 if (ssa->ss.ss_family != AF_UNSPEC) { 2102 /* 2103 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership 2104 * is an error. On an existing inclusive membership, 2105 * it just adds the source to the filter list. 2106 */ 2107 if (imf->imf_st[1] != MCAST_INCLUDE) { 2108 error = EINVAL; 2109 goto out_inp_locked; 2110 } 2111 /* 2112 * Throw out duplicates. 2113 * 2114 * XXX FIXME: This makes a naive assumption that 2115 * even if entries exist for *ssa in this imf, 2116 * they will be rejected as dupes, even if they 2117 * are not valid in the current mode (in-mode). 2118 * 2119 * in_msource is transactioned just as for anything 2120 * else in SSM -- but note naive use of inm_graft() 2121 * below for allocating new filter entries. 2122 * 2123 * This is only an issue if someone mixes the 2124 * full-state SSM API with the delta-based API, 2125 * which is discouraged in the relevant RFCs. 2126 */ 2127 lims = imo_match_source(imf, &ssa->sa); 2128 if (lims != NULL /*&& 2129 lims->imsl_st[1] == MCAST_INCLUDE*/) { 2130 error = EADDRNOTAVAIL; 2131 goto out_inp_locked; 2132 } 2133 } else { 2134 /* 2135 * MCAST_JOIN_GROUP on an existing exclusive 2136 * membership is an error; return EADDRINUSE 2137 * to preserve 4.4BSD API idempotence, and 2138 * avoid tedious detour to code below. 2139 * NOTE: This is bending RFC 3678 a bit. 2140 * 2141 * On an existing inclusive membership, this is also 2142 * an error; if you want to change filter mode, 2143 * you must use the userland API setsourcefilter(). 2144 * XXX We don't reject this for imf in UNDEFINED 2145 * state at t1, because allocation of a filter 2146 * is atomic with allocation of a membership. 2147 */ 2148 error = EINVAL; 2149 if (imf->imf_st[1] == MCAST_EXCLUDE) 2150 error = EADDRINUSE; 2151 goto out_inp_locked; 2152 } 2153 } 2154 2155 /* 2156 * Begin state merge transaction at socket layer. 2157 */ 2158 INP_WLOCK_ASSERT(inp); 2159 2160 /* 2161 * Graft new source into filter list for this inpcb's 2162 * membership of the group. The in_multi may not have 2163 * been allocated yet if this is a new membership, however, 2164 * the in_mfilter slot will be allocated and must be initialized. 2165 * 2166 * Note: Grafting of exclusive mode filters doesn't happen 2167 * in this path. 2168 * XXX: Should check for non-NULL lims (node exists but may 2169 * not be in-mode) for interop with full-state API. 2170 */ 2171 if (ssa->ss.ss_family != AF_UNSPEC) { 2172 /* Membership starts in IN mode */ 2173 if (is_new) { 2174 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__); 2175 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE); 2176 if (imf == NULL) { 2177 error = ENOMEM; 2178 goto out_inp_locked; 2179 } 2180 } else { 2181 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 2182 } 2183 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin); 2184 if (lims == NULL) { 2185 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2186 __func__); 2187 error = ENOMEM; 2188 goto out_inp_locked; 2189 } 2190 } else { 2191 /* No address specified; Membership starts in EX mode */ 2192 if (is_new) { 2193 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__); 2194 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE); 2195 if (imf == NULL) { 2196 error = ENOMEM; 2197 goto out_inp_locked; 2198 } 2199 } 2200 } 2201 2202 /* 2203 * Begin state merge transaction at IGMP layer. 2204 */ 2205 if (is_new) { 2206 in_pcbref(inp); 2207 INP_WUNLOCK(inp); 2208 2209 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf, 2210 &imf->imf_inm); 2211 2212 INP_WLOCK(inp); 2213 if (in_pcbrele_wlocked(inp)) { 2214 error = ENXIO; 2215 goto out_inp_unlocked; 2216 } 2217 if (error) { 2218 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed", 2219 __func__); 2220 goto out_inp_locked; 2221 } 2222 /* 2223 * NOTE: Refcount from in_joingroup_locked() 2224 * is protecting membership. 2225 */ 2226 ip_mfilter_insert(&imo->imo_head, imf); 2227 } else { 2228 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2229 IN_MULTI_LIST_LOCK(); 2230 error = inm_merge(inm, imf); 2231 if (error) { 2232 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2233 __func__); 2234 IN_MULTI_LIST_UNLOCK(); 2235 imf_rollback(imf); 2236 imf_reap(imf); 2237 goto out_inp_locked; 2238 } 2239 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2240 error = igmp_change_state(inm); 2241 IN_MULTI_LIST_UNLOCK(); 2242 if (error) { 2243 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2244 __func__); 2245 imf_rollback(imf); 2246 imf_reap(imf); 2247 goto out_inp_locked; 2248 } 2249 } 2250 2251 imf_commit(imf); 2252 imf = NULL; 2253 2254 out_inp_locked: 2255 INP_WUNLOCK(inp); 2256 out_inp_unlocked: 2257 IN_MULTI_UNLOCK(); 2258 2259 if (is_new && imf) { 2260 if (imf->imf_inm != NULL) { 2261 IN_MULTI_LIST_LOCK(); 2262 IF_ADDR_WLOCK(ifp); 2263 inm_release_deferred(imf->imf_inm); 2264 IF_ADDR_WUNLOCK(ifp); 2265 IN_MULTI_LIST_UNLOCK(); 2266 } 2267 ip_mfilter_free(imf); 2268 } 2269 return (error); 2270 } 2271 2272 /* 2273 * Leave an IPv4 multicast group on an inpcb, possibly with a source. 2274 */ 2275 static int 2276 inp_leave_group(struct inpcb *inp, struct sockopt *sopt) 2277 { 2278 struct group_source_req gsr; 2279 struct ip_mreq_source mreqs; 2280 struct rm_priotracker in_ifa_tracker; 2281 sockunion_t *gsa, *ssa; 2282 struct ifnet *ifp; 2283 struct in_mfilter *imf; 2284 struct ip_moptions *imo; 2285 struct in_msource *ims; 2286 struct in_multi *inm; 2287 int error; 2288 bool is_final; 2289 2290 ifp = NULL; 2291 error = 0; 2292 is_final = true; 2293 2294 memset(&gsr, 0, sizeof(struct group_source_req)); 2295 gsa = (sockunion_t *)&gsr.gsr_group; 2296 gsa->ss.ss_family = AF_UNSPEC; 2297 ssa = (sockunion_t *)&gsr.gsr_source; 2298 ssa->ss.ss_family = AF_UNSPEC; 2299 2300 switch (sopt->sopt_name) { 2301 case IP_DROP_MEMBERSHIP: 2302 case IP_DROP_SOURCE_MEMBERSHIP: 2303 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) { 2304 error = sooptcopyin(sopt, &mreqs, 2305 sizeof(struct ip_mreq), 2306 sizeof(struct ip_mreq)); 2307 /* 2308 * Swap interface and sourceaddr arguments, 2309 * as ip_mreq and ip_mreq_source are laid 2310 * out differently. 2311 */ 2312 mreqs.imr_interface = mreqs.imr_sourceaddr; 2313 mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 2314 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2315 error = sooptcopyin(sopt, &mreqs, 2316 sizeof(struct ip_mreq_source), 2317 sizeof(struct ip_mreq_source)); 2318 } 2319 if (error) 2320 return (error); 2321 2322 gsa->sin.sin_family = AF_INET; 2323 gsa->sin.sin_len = sizeof(struct sockaddr_in); 2324 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2325 2326 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2327 ssa->sin.sin_family = AF_INET; 2328 ssa->sin.sin_len = sizeof(struct sockaddr_in); 2329 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2330 } 2331 2332 /* 2333 * Attempt to look up hinted ifp from interface address. 2334 * Fallthrough with null ifp iff lookup fails, to 2335 * preserve 4.4BSD mcast API idempotence. 2336 * XXX NOTE WELL: The RFC 3678 API is preferred because 2337 * using an IPv4 address as a key is racy. 2338 */ 2339 if (!in_nullhost(mreqs.imr_interface)) { 2340 IN_IFADDR_RLOCK(&in_ifa_tracker); 2341 INADDR_TO_IFP(mreqs.imr_interface, ifp); 2342 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2343 } 2344 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2345 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2346 2347 break; 2348 2349 case MCAST_LEAVE_GROUP: 2350 case MCAST_LEAVE_SOURCE_GROUP: 2351 if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2352 error = sooptcopyin(sopt, &gsr, 2353 sizeof(struct group_req), 2354 sizeof(struct group_req)); 2355 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2356 error = sooptcopyin(sopt, &gsr, 2357 sizeof(struct group_source_req), 2358 sizeof(struct group_source_req)); 2359 } 2360 if (error) 2361 return (error); 2362 2363 if (gsa->sin.sin_family != AF_INET || 2364 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2365 return (EINVAL); 2366 2367 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2368 if (ssa->sin.sin_family != AF_INET || 2369 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2370 return (EINVAL); 2371 } 2372 2373 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2374 return (EADDRNOTAVAIL); 2375 2376 ifp = ifnet_byindex(gsr.gsr_interface); 2377 2378 if (ifp == NULL) 2379 return (EADDRNOTAVAIL); 2380 break; 2381 2382 default: 2383 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2384 __func__, sopt->sopt_name); 2385 return (EOPNOTSUPP); 2386 break; 2387 } 2388 2389 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2390 return (EINVAL); 2391 2392 IN_MULTI_LOCK(); 2393 2394 /* 2395 * Find the membership in the membership list. 2396 */ 2397 imo = inp_findmoptions(inp); 2398 imf = imo_match_group(imo, ifp, &gsa->sa); 2399 if (imf == NULL) { 2400 error = EADDRNOTAVAIL; 2401 goto out_inp_locked; 2402 } 2403 inm = imf->imf_inm; 2404 2405 if (ssa->ss.ss_family != AF_UNSPEC) 2406 is_final = false; 2407 2408 /* 2409 * Begin state merge transaction at socket layer. 2410 */ 2411 INP_WLOCK_ASSERT(inp); 2412 2413 /* 2414 * If we were instructed only to leave a given source, do so. 2415 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2416 */ 2417 if (is_final) { 2418 ip_mfilter_remove(&imo->imo_head, imf); 2419 imf_leave(imf); 2420 2421 /* 2422 * Give up the multicast address record to which 2423 * the membership points. 2424 */ 2425 (void) in_leavegroup_locked(imf->imf_inm, imf); 2426 } else { 2427 if (imf->imf_st[0] == MCAST_EXCLUDE) { 2428 error = EADDRNOTAVAIL; 2429 goto out_inp_locked; 2430 } 2431 ims = imo_match_source(imf, &ssa->sa); 2432 if (ims == NULL) { 2433 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", 2434 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not "); 2435 error = EADDRNOTAVAIL; 2436 goto out_inp_locked; 2437 } 2438 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 2439 error = imf_prune(imf, &ssa->sin); 2440 if (error) { 2441 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2442 __func__); 2443 goto out_inp_locked; 2444 } 2445 } 2446 2447 /* 2448 * Begin state merge transaction at IGMP layer. 2449 */ 2450 if (!is_final) { 2451 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2452 IN_MULTI_LIST_LOCK(); 2453 error = inm_merge(inm, imf); 2454 if (error) { 2455 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2456 __func__); 2457 IN_MULTI_LIST_UNLOCK(); 2458 imf_rollback(imf); 2459 imf_reap(imf); 2460 goto out_inp_locked; 2461 } 2462 2463 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2464 error = igmp_change_state(inm); 2465 IN_MULTI_LIST_UNLOCK(); 2466 if (error) { 2467 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2468 __func__); 2469 imf_rollback(imf); 2470 imf_reap(imf); 2471 goto out_inp_locked; 2472 } 2473 } 2474 imf_commit(imf); 2475 imf_reap(imf); 2476 2477 out_inp_locked: 2478 INP_WUNLOCK(inp); 2479 2480 if (is_final && imf) 2481 ip_mfilter_free(imf); 2482 2483 IN_MULTI_UNLOCK(); 2484 return (error); 2485 } 2486 2487 /* 2488 * Select the interface for transmitting IPv4 multicast datagrams. 2489 * 2490 * Either an instance of struct in_addr or an instance of struct ip_mreqn 2491 * may be passed to this socket option. An address of INADDR_ANY or an 2492 * interface index of 0 is used to remove a previous selection. 2493 * When no interface is selected, one is chosen for every send. 2494 */ 2495 static int 2496 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2497 { 2498 struct rm_priotracker in_ifa_tracker; 2499 struct in_addr addr; 2500 struct ip_mreqn mreqn; 2501 struct ifnet *ifp; 2502 struct ip_moptions *imo; 2503 int error; 2504 2505 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 2506 /* 2507 * An interface index was specified using the 2508 * Linux-derived ip_mreqn structure. 2509 */ 2510 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn), 2511 sizeof(struct ip_mreqn)); 2512 if (error) 2513 return (error); 2514 2515 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex) 2516 return (EINVAL); 2517 2518 if (mreqn.imr_ifindex == 0) { 2519 ifp = NULL; 2520 } else { 2521 ifp = ifnet_byindex(mreqn.imr_ifindex); 2522 if (ifp == NULL) 2523 return (EADDRNOTAVAIL); 2524 } 2525 } else { 2526 /* 2527 * An interface was specified by IPv4 address. 2528 * This is the traditional BSD usage. 2529 */ 2530 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr), 2531 sizeof(struct in_addr)); 2532 if (error) 2533 return (error); 2534 if (in_nullhost(addr)) { 2535 ifp = NULL; 2536 } else { 2537 IN_IFADDR_RLOCK(&in_ifa_tracker); 2538 INADDR_TO_IFP(addr, ifp); 2539 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2540 if (ifp == NULL) 2541 return (EADDRNOTAVAIL); 2542 } 2543 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp, 2544 ntohl(addr.s_addr)); 2545 } 2546 2547 /* Reject interfaces which do not support multicast. */ 2548 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0) 2549 return (EOPNOTSUPP); 2550 2551 imo = inp_findmoptions(inp); 2552 imo->imo_multicast_ifp = ifp; 2553 imo->imo_multicast_addr.s_addr = INADDR_ANY; 2554 INP_WUNLOCK(inp); 2555 2556 return (0); 2557 } 2558 2559 /* 2560 * Atomically set source filters on a socket for an IPv4 multicast group. 2561 * 2562 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2563 */ 2564 static int 2565 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2566 { 2567 struct __msfilterreq msfr; 2568 sockunion_t *gsa; 2569 struct ifnet *ifp; 2570 struct in_mfilter *imf; 2571 struct ip_moptions *imo; 2572 struct in_multi *inm; 2573 int error; 2574 2575 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2576 sizeof(struct __msfilterreq)); 2577 if (error) 2578 return (error); 2579 2580 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 2581 return (ENOBUFS); 2582 2583 if ((msfr.msfr_fmode != MCAST_EXCLUDE && 2584 msfr.msfr_fmode != MCAST_INCLUDE)) 2585 return (EINVAL); 2586 2587 if (msfr.msfr_group.ss_family != AF_INET || 2588 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in)) 2589 return (EINVAL); 2590 2591 gsa = (sockunion_t *)&msfr.msfr_group; 2592 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2593 return (EINVAL); 2594 2595 gsa->sin.sin_port = 0; /* ignore port */ 2596 2597 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 2598 return (EADDRNOTAVAIL); 2599 2600 ifp = ifnet_byindex(msfr.msfr_ifindex); 2601 if (ifp == NULL) 2602 return (EADDRNOTAVAIL); 2603 2604 IN_MULTI_LOCK(); 2605 2606 /* 2607 * Take the INP write lock. 2608 * Check if this socket is a member of this group. 2609 */ 2610 imo = inp_findmoptions(inp); 2611 imf = imo_match_group(imo, ifp, &gsa->sa); 2612 if (imf == NULL) { 2613 error = EADDRNOTAVAIL; 2614 goto out_inp_locked; 2615 } 2616 inm = imf->imf_inm; 2617 2618 /* 2619 * Begin state merge transaction at socket layer. 2620 */ 2621 INP_WLOCK_ASSERT(inp); 2622 2623 imf->imf_st[1] = msfr.msfr_fmode; 2624 2625 /* 2626 * Apply any new source filters, if present. 2627 * Make a copy of the user-space source vector so 2628 * that we may copy them with a single copyin. This 2629 * allows us to deal with page faults up-front. 2630 */ 2631 if (msfr.msfr_nsrcs > 0) { 2632 struct in_msource *lims; 2633 struct sockaddr_in *psin; 2634 struct sockaddr_storage *kss, *pkss; 2635 int i; 2636 2637 INP_WUNLOCK(inp); 2638 2639 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries", 2640 __func__, (unsigned long)msfr.msfr_nsrcs); 2641 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2642 M_TEMP, M_WAITOK); 2643 error = copyin(msfr.msfr_srcs, kss, 2644 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2645 if (error) { 2646 free(kss, M_TEMP); 2647 return (error); 2648 } 2649 2650 INP_WLOCK(inp); 2651 2652 /* 2653 * Mark all source filters as UNDEFINED at t1. 2654 * Restore new group filter mode, as imf_leave() 2655 * will set it to INCLUDE. 2656 */ 2657 imf_leave(imf); 2658 imf->imf_st[1] = msfr.msfr_fmode; 2659 2660 /* 2661 * Update socket layer filters at t1, lazy-allocating 2662 * new entries. This saves a bunch of memory at the 2663 * cost of one RB_FIND() per source entry; duplicate 2664 * entries in the msfr_nsrcs vector are ignored. 2665 * If we encounter an error, rollback transaction. 2666 * 2667 * XXX This too could be replaced with a set-symmetric 2668 * difference like loop to avoid walking from root 2669 * every time, as the key space is common. 2670 */ 2671 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2672 psin = (struct sockaddr_in *)pkss; 2673 if (psin->sin_family != AF_INET) { 2674 error = EAFNOSUPPORT; 2675 break; 2676 } 2677 if (psin->sin_len != sizeof(struct sockaddr_in)) { 2678 error = EINVAL; 2679 break; 2680 } 2681 error = imf_get_source(imf, psin, &lims); 2682 if (error) 2683 break; 2684 lims->imsl_st[1] = imf->imf_st[1]; 2685 } 2686 free(kss, M_TEMP); 2687 } 2688 2689 if (error) 2690 goto out_imf_rollback; 2691 2692 INP_WLOCK_ASSERT(inp); 2693 2694 /* 2695 * Begin state merge transaction at IGMP layer. 2696 */ 2697 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2698 IN_MULTI_LIST_LOCK(); 2699 error = inm_merge(inm, imf); 2700 if (error) { 2701 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 2702 IN_MULTI_LIST_UNLOCK(); 2703 goto out_imf_rollback; 2704 } 2705 2706 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2707 error = igmp_change_state(inm); 2708 IN_MULTI_LIST_UNLOCK(); 2709 if (error) 2710 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 2711 2712 out_imf_rollback: 2713 if (error) 2714 imf_rollback(imf); 2715 else 2716 imf_commit(imf); 2717 2718 imf_reap(imf); 2719 2720 out_inp_locked: 2721 INP_WUNLOCK(inp); 2722 IN_MULTI_UNLOCK(); 2723 return (error); 2724 } 2725 2726 /* 2727 * Set the IP multicast options in response to user setsockopt(). 2728 * 2729 * Many of the socket options handled in this function duplicate the 2730 * functionality of socket options in the regular unicast API. However, 2731 * it is not possible to merge the duplicate code, because the idempotence 2732 * of the IPv4 multicast part of the BSD Sockets API must be preserved; 2733 * the effects of these options must be treated as separate and distinct. 2734 * 2735 * SMPng: XXX: Unlocked read of inp_socket believed OK. 2736 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING 2737 * is refactored to no longer use vifs. 2738 */ 2739 int 2740 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2741 { 2742 struct ip_moptions *imo; 2743 int error; 2744 struct epoch_tracker et; 2745 2746 error = 0; 2747 2748 /* 2749 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 2750 * or is a divert socket, reject it. 2751 */ 2752 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 2753 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2754 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) 2755 return (EOPNOTSUPP); 2756 2757 switch (sopt->sopt_name) { 2758 case IP_MULTICAST_VIF: { 2759 int vifi; 2760 /* 2761 * Select a multicast VIF for transmission. 2762 * Only useful if multicast forwarding is active. 2763 */ 2764 if (legal_vif_num == NULL) { 2765 error = EOPNOTSUPP; 2766 break; 2767 } 2768 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int)); 2769 if (error) 2770 break; 2771 if (!legal_vif_num(vifi) && (vifi != -1)) { 2772 error = EINVAL; 2773 break; 2774 } 2775 imo = inp_findmoptions(inp); 2776 imo->imo_multicast_vif = vifi; 2777 INP_WUNLOCK(inp); 2778 break; 2779 } 2780 2781 case IP_MULTICAST_IF: 2782 error = inp_set_multicast_if(inp, sopt); 2783 break; 2784 2785 case IP_MULTICAST_TTL: { 2786 u_char ttl; 2787 2788 /* 2789 * Set the IP time-to-live for outgoing multicast packets. 2790 * The original multicast API required a char argument, 2791 * which is inconsistent with the rest of the socket API. 2792 * We allow either a char or an int. 2793 */ 2794 if (sopt->sopt_valsize == sizeof(u_char)) { 2795 error = sooptcopyin(sopt, &ttl, sizeof(u_char), 2796 sizeof(u_char)); 2797 if (error) 2798 break; 2799 } else { 2800 u_int ittl; 2801 2802 error = sooptcopyin(sopt, &ittl, sizeof(u_int), 2803 sizeof(u_int)); 2804 if (error) 2805 break; 2806 if (ittl > 255) { 2807 error = EINVAL; 2808 break; 2809 } 2810 ttl = (u_char)ittl; 2811 } 2812 imo = inp_findmoptions(inp); 2813 imo->imo_multicast_ttl = ttl; 2814 INP_WUNLOCK(inp); 2815 break; 2816 } 2817 2818 case IP_MULTICAST_LOOP: { 2819 u_char loop; 2820 2821 /* 2822 * Set the loopback flag for outgoing multicast packets. 2823 * Must be zero or one. The original multicast API required a 2824 * char argument, which is inconsistent with the rest 2825 * of the socket API. We allow either a char or an int. 2826 */ 2827 if (sopt->sopt_valsize == sizeof(u_char)) { 2828 error = sooptcopyin(sopt, &loop, sizeof(u_char), 2829 sizeof(u_char)); 2830 if (error) 2831 break; 2832 } else { 2833 u_int iloop; 2834 2835 error = sooptcopyin(sopt, &iloop, sizeof(u_int), 2836 sizeof(u_int)); 2837 if (error) 2838 break; 2839 loop = (u_char)iloop; 2840 } 2841 imo = inp_findmoptions(inp); 2842 imo->imo_multicast_loop = !!loop; 2843 INP_WUNLOCK(inp); 2844 break; 2845 } 2846 2847 case IP_ADD_MEMBERSHIP: 2848 case IP_ADD_SOURCE_MEMBERSHIP: 2849 case MCAST_JOIN_GROUP: 2850 case MCAST_JOIN_SOURCE_GROUP: 2851 NET_EPOCH_ENTER(et); 2852 error = inp_join_group(inp, sopt); 2853 NET_EPOCH_EXIT(et); 2854 break; 2855 2856 case IP_DROP_MEMBERSHIP: 2857 case IP_DROP_SOURCE_MEMBERSHIP: 2858 case MCAST_LEAVE_GROUP: 2859 case MCAST_LEAVE_SOURCE_GROUP: 2860 error = inp_leave_group(inp, sopt); 2861 break; 2862 2863 case IP_BLOCK_SOURCE: 2864 case IP_UNBLOCK_SOURCE: 2865 case MCAST_BLOCK_SOURCE: 2866 case MCAST_UNBLOCK_SOURCE: 2867 error = inp_block_unblock_source(inp, sopt); 2868 break; 2869 2870 case IP_MSFILTER: 2871 error = inp_set_source_filters(inp, sopt); 2872 break; 2873 2874 default: 2875 error = EOPNOTSUPP; 2876 break; 2877 } 2878 2879 INP_UNLOCK_ASSERT(inp); 2880 2881 return (error); 2882 } 2883 2884 /* 2885 * Expose IGMP's multicast filter mode and source list(s) to userland, 2886 * keyed by (ifindex, group). 2887 * The filter mode is written out as a uint32_t, followed by 2888 * 0..n of struct in_addr. 2889 * For use by ifmcstat(8). 2890 * SMPng: NOTE: unlocked read of ifindex space. 2891 */ 2892 static int 2893 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS) 2894 { 2895 struct in_addr src, group; 2896 struct epoch_tracker et; 2897 struct ifnet *ifp; 2898 struct ifmultiaddr *ifma; 2899 struct in_multi *inm; 2900 struct ip_msource *ims; 2901 int *name; 2902 int retval; 2903 u_int namelen; 2904 uint32_t fmode, ifindex; 2905 2906 name = (int *)arg1; 2907 namelen = arg2; 2908 2909 if (req->newptr != NULL) 2910 return (EPERM); 2911 2912 if (namelen != 2) 2913 return (EINVAL); 2914 2915 ifindex = name[0]; 2916 if (ifindex <= 0 || ifindex > V_if_index) { 2917 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range", 2918 __func__, ifindex); 2919 return (ENOENT); 2920 } 2921 2922 group.s_addr = name[1]; 2923 if (!IN_MULTICAST(ntohl(group.s_addr))) { 2924 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast", 2925 __func__, ntohl(group.s_addr)); 2926 return (EINVAL); 2927 } 2928 2929 NET_EPOCH_ENTER(et); 2930 ifp = ifnet_byindex(ifindex); 2931 if (ifp == NULL) { 2932 NET_EPOCH_EXIT(et); 2933 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u", 2934 __func__, ifindex); 2935 return (ENOENT); 2936 } 2937 2938 retval = sysctl_wire_old_buffer(req, 2939 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr))); 2940 if (retval) { 2941 NET_EPOCH_EXIT(et); 2942 return (retval); 2943 } 2944 2945 IN_MULTI_LIST_LOCK(); 2946 2947 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2948 if (ifma->ifma_addr->sa_family != AF_INET || 2949 ifma->ifma_protospec == NULL) 2950 continue; 2951 inm = (struct in_multi *)ifma->ifma_protospec; 2952 if (!in_hosteq(inm->inm_addr, group)) 2953 continue; 2954 fmode = inm->inm_st[1].iss_fmode; 2955 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 2956 if (retval != 0) 2957 break; 2958 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 2959 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2960 ims->ims_haddr); 2961 /* 2962 * Only copy-out sources which are in-mode. 2963 */ 2964 if (fmode != ims_get_mode(inm, ims, 1)) { 2965 CTR1(KTR_IGMPV3, "%s: skip non-in-mode", 2966 __func__); 2967 continue; 2968 } 2969 src.s_addr = htonl(ims->ims_haddr); 2970 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr)); 2971 if (retval != 0) 2972 break; 2973 } 2974 } 2975 2976 IN_MULTI_LIST_UNLOCK(); 2977 NET_EPOCH_EXIT(et); 2978 2979 return (retval); 2980 } 2981 2982 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3) 2983 2984 static const char *inm_modestrs[] = { 2985 [MCAST_UNDEFINED] = "un", 2986 [MCAST_INCLUDE] = "in", 2987 [MCAST_EXCLUDE] = "ex", 2988 }; 2989 _Static_assert(MCAST_UNDEFINED == 0 && 2990 MCAST_EXCLUDE + 1 == nitems(inm_modestrs), 2991 "inm_modestrs: no longer matches #defines"); 2992 2993 static const char * 2994 inm_mode_str(const int mode) 2995 { 2996 2997 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 2998 return (inm_modestrs[mode]); 2999 return ("??"); 3000 } 3001 3002 static const char *inm_statestrs[] = { 3003 [IGMP_NOT_MEMBER] = "not-member", 3004 [IGMP_SILENT_MEMBER] = "silent", 3005 [IGMP_REPORTING_MEMBER] = "reporting", 3006 [IGMP_IDLE_MEMBER] = "idle", 3007 [IGMP_LAZY_MEMBER] = "lazy", 3008 [IGMP_SLEEPING_MEMBER] = "sleeping", 3009 [IGMP_AWAKENING_MEMBER] = "awakening", 3010 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending", 3011 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending", 3012 [IGMP_LEAVING_MEMBER] = "leaving", 3013 }; 3014 _Static_assert(IGMP_NOT_MEMBER == 0 && 3015 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs), 3016 "inm_statetrs: no longer matches #defines"); 3017 3018 static const char * 3019 inm_state_str(const int state) 3020 { 3021 3022 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER) 3023 return (inm_statestrs[state]); 3024 return ("??"); 3025 } 3026 3027 /* 3028 * Dump an in_multi structure to the console. 3029 */ 3030 void 3031 inm_print(const struct in_multi *inm) 3032 { 3033 int t; 3034 char addrbuf[INET_ADDRSTRLEN]; 3035 3036 if ((ktr_mask & KTR_IGMPV3) == 0) 3037 return; 3038 3039 printf("%s: --- begin inm %p ---\n", __func__, inm); 3040 printf("addr %s ifp %p(%s) ifma %p\n", 3041 inet_ntoa_r(inm->inm_addr, addrbuf), 3042 inm->inm_ifp, 3043 inm->inm_ifp->if_xname, 3044 inm->inm_ifma); 3045 printf("timer %u state %s refcount %u scq.len %u\n", 3046 inm->inm_timer, 3047 inm_state_str(inm->inm_state), 3048 inm->inm_refcount, 3049 inm->inm_scq.mq_len); 3050 printf("igi %p nsrc %lu sctimer %u scrv %u\n", 3051 inm->inm_igi, 3052 inm->inm_nsrc, 3053 inm->inm_sctimer, 3054 inm->inm_scrv); 3055 for (t = 0; t < 2; t++) { 3056 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 3057 inm_mode_str(inm->inm_st[t].iss_fmode), 3058 inm->inm_st[t].iss_asm, 3059 inm->inm_st[t].iss_ex, 3060 inm->inm_st[t].iss_in, 3061 inm->inm_st[t].iss_rec); 3062 } 3063 printf("%s: --- end inm %p ---\n", __func__, inm); 3064 } 3065 3066 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */ 3067 3068 void 3069 inm_print(const struct in_multi *inm) 3070 { 3071 3072 } 3073 3074 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */ 3075 3076 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp); 3077