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/gtaskqueue.h> 55 #include <sys/tree.h> 56 57 #include <net/if.h> 58 #include <net/if_var.h> 59 #include <net/if_dl.h> 60 #include <net/route.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 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 98 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however 99 * it can be taken by code in net/if.c also. 100 * - ip_moptions and in_mfilter are covered by the INP_WLOCK. 101 * 102 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly 103 * any need for in_multi itself to be virtualized -- it is bound to an ifp 104 * anyway no matter what happens. 105 */ 106 struct mtx in_multi_list_mtx; 107 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF); 108 109 struct mtx in_multi_free_mtx; 110 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF); 111 112 struct sx in_multi_sx; 113 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx"); 114 115 int ifma_restart; 116 117 /* 118 * Functions with non-static linkage defined in this file should be 119 * declared in in_var.h: 120 * imo_multi_filter() 121 * in_addmulti() 122 * in_delmulti() 123 * in_joingroup() 124 * in_joingroup_locked() 125 * in_leavegroup() 126 * in_leavegroup_locked() 127 * and ip_var.h: 128 * inp_freemoptions() 129 * inp_getmoptions() 130 * inp_setmoptions() 131 * 132 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti() 133 * and in_delmulti(). 134 */ 135 static void imf_commit(struct in_mfilter *); 136 static int imf_get_source(struct in_mfilter *imf, 137 const struct sockaddr_in *psin, 138 struct in_msource **); 139 static struct in_msource * 140 imf_graft(struct in_mfilter *, const uint8_t, 141 const struct sockaddr_in *); 142 static void imf_leave(struct in_mfilter *); 143 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *); 144 static void imf_purge(struct in_mfilter *); 145 static void imf_rollback(struct in_mfilter *); 146 static void imf_reap(struct in_mfilter *); 147 static int imo_grow(struct ip_moptions *); 148 static size_t imo_match_group(const struct ip_moptions *, 149 const struct ifnet *, const struct sockaddr *); 150 static struct in_msource * 151 imo_match_source(const struct ip_moptions *, const size_t, 152 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, CTLFLAG_RW, 0, 180 "IPv4 multicast"); 181 182 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER; 183 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc, 184 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0, 185 "Max source filters per group"); 186 187 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER; 188 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc, 189 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0, 190 "Max source filters per socket"); 191 192 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP; 193 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN, 194 &in_mcast_loop, 0, "Loopback multicast datagrams by default"); 195 196 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters, 197 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters, 198 "Per-interface stack-wide source filters"); 199 200 #ifdef KTR 201 /* 202 * Inline function which wraps assertions for a valid ifp. 203 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 204 * is detached. 205 */ 206 static int __inline 207 inm_is_ifp_detached(const struct in_multi *inm) 208 { 209 struct ifnet *ifp; 210 211 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 212 ifp = inm->inm_ifma->ifma_ifp; 213 if (ifp != NULL) { 214 /* 215 * Sanity check that netinet's notion of ifp is the 216 * same as net's. 217 */ 218 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 219 } 220 221 return (ifp == NULL); 222 } 223 #endif 224 225 static struct grouptask free_gtask; 226 static struct in_multi_head inm_free_list; 227 static void inm_release_task(void *arg __unused); 228 static void inm_init(void) 229 { 230 SLIST_INIT(&inm_free_list); 231 taskqgroup_config_gtask_init(NULL, &free_gtask, inm_release_task, "inm release task"); 232 } 233 234 #ifdef EARLY_AP_STARTUP 235 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST, 236 inm_init, NULL); 237 #else 238 SYSINIT(inm_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_FIRST, 239 inm_init, NULL); 240 #endif 241 242 243 void 244 inm_release_list_deferred(struct in_multi_head *inmh) 245 { 246 247 if (SLIST_EMPTY(inmh)) 248 return; 249 mtx_lock(&in_multi_free_mtx); 250 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele); 251 mtx_unlock(&in_multi_free_mtx); 252 GROUPTASK_ENQUEUE(&free_gtask); 253 } 254 255 void 256 inm_disconnect(struct in_multi *inm) 257 { 258 struct ifnet *ifp; 259 struct ifmultiaddr *ifma, *ll_ifma; 260 261 ifp = inm->inm_ifp; 262 IF_ADDR_WLOCK_ASSERT(ifp); 263 ifma = inm->inm_ifma; 264 265 if_ref(ifp); 266 if (ifma->ifma_flags & IFMA_F_ENQUEUED) { 267 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 268 ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 269 } 270 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname); 271 if ((ll_ifma = ifma->ifma_llifma) != NULL) { 272 MPASS(ifma != ll_ifma); 273 ifma->ifma_llifma = NULL; 274 MPASS(ll_ifma->ifma_llifma == NULL); 275 MPASS(ll_ifma->ifma_ifp == ifp); 276 if (--ll_ifma->ifma_refcount == 0) { 277 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) { 278 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link); 279 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 280 } 281 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname); 282 if_freemulti(ll_ifma); 283 ifma_restart = true; 284 } 285 } 286 } 287 288 void 289 inm_release_deferred(struct in_multi *inm) 290 { 291 struct in_multi_head tmp; 292 293 IN_MULTI_LIST_LOCK_ASSERT(); 294 MPASS(inm->inm_refcount > 0); 295 if (--inm->inm_refcount == 0) { 296 SLIST_INIT(&tmp); 297 inm_disconnect(inm); 298 inm->inm_ifma->ifma_protospec = NULL; 299 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele); 300 inm_release_list_deferred(&tmp); 301 } 302 } 303 304 static void 305 inm_release_task(void *arg __unused) 306 { 307 struct in_multi_head inm_free_tmp; 308 struct in_multi *inm, *tinm; 309 310 SLIST_INIT(&inm_free_tmp); 311 mtx_lock(&in_multi_free_mtx); 312 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele); 313 mtx_unlock(&in_multi_free_mtx); 314 IN_MULTI_LOCK(); 315 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) { 316 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele); 317 MPASS(inm); 318 inm_release(inm); 319 } 320 IN_MULTI_UNLOCK(); 321 } 322 323 /* 324 * Initialize an in_mfilter structure to a known state at t0, t1 325 * with an empty source filter list. 326 */ 327 static __inline void 328 imf_init(struct in_mfilter *imf, const int st0, const int st1) 329 { 330 memset(imf, 0, sizeof(struct in_mfilter)); 331 RB_INIT(&imf->imf_sources); 332 imf->imf_st[0] = st0; 333 imf->imf_st[1] = st1; 334 } 335 336 /* 337 * Function for looking up an in_multi record for an IPv4 multicast address 338 * on a given interface. ifp must be valid. If no record found, return NULL. 339 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held. 340 */ 341 struct in_multi * 342 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina) 343 { 344 struct ifmultiaddr *ifma; 345 struct in_multi *inm; 346 347 IN_MULTI_LIST_LOCK_ASSERT(); 348 IF_ADDR_LOCK_ASSERT(ifp); 349 350 inm = NULL; 351 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) { 352 if (ifma->ifma_addr->sa_family != AF_INET || 353 ifma->ifma_protospec == NULL) 354 continue; 355 inm = (struct in_multi *)ifma->ifma_protospec; 356 if (inm->inm_addr.s_addr == ina.s_addr) 357 break; 358 inm = NULL; 359 } 360 return (inm); 361 } 362 363 /* 364 * Wrapper for inm_lookup_locked(). 365 * The IF_ADDR_LOCK will be taken on ifp and released on return. 366 */ 367 struct in_multi * 368 inm_lookup(struct ifnet *ifp, const struct in_addr ina) 369 { 370 struct epoch_tracker et; 371 struct in_multi *inm; 372 373 IN_MULTI_LIST_LOCK_ASSERT(); 374 NET_EPOCH_ENTER(et); 375 inm = inm_lookup_locked(ifp, ina); 376 NET_EPOCH_EXIT(et); 377 378 return (inm); 379 } 380 381 /* 382 * Resize the ip_moptions vector to the next power-of-two minus 1. 383 * May be called with locks held; do not sleep. 384 */ 385 static int 386 imo_grow(struct ip_moptions *imo) 387 { 388 struct in_multi **nmships; 389 struct in_multi **omships; 390 struct in_mfilter *nmfilters; 391 struct in_mfilter *omfilters; 392 size_t idx; 393 size_t newmax; 394 size_t oldmax; 395 396 nmships = NULL; 397 nmfilters = NULL; 398 omships = imo->imo_membership; 399 omfilters = imo->imo_mfilters; 400 oldmax = imo->imo_max_memberships; 401 newmax = ((oldmax + 1) * 2) - 1; 402 403 if (newmax <= IP_MAX_MEMBERSHIPS) { 404 nmships = (struct in_multi **)realloc(omships, 405 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT); 406 nmfilters = (struct in_mfilter *)realloc(omfilters, 407 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT); 408 if (nmships != NULL && nmfilters != NULL) { 409 /* Initialize newly allocated source filter heads. */ 410 for (idx = oldmax; idx < newmax; idx++) { 411 imf_init(&nmfilters[idx], MCAST_UNDEFINED, 412 MCAST_EXCLUDE); 413 } 414 imo->imo_max_memberships = newmax; 415 imo->imo_membership = nmships; 416 imo->imo_mfilters = nmfilters; 417 } 418 } 419 420 if (nmships == NULL || nmfilters == NULL) { 421 if (nmships != NULL) 422 free(nmships, M_IPMOPTS); 423 if (nmfilters != NULL) 424 free(nmfilters, M_INMFILTER); 425 return (ETOOMANYREFS); 426 } 427 428 return (0); 429 } 430 431 /* 432 * Find an IPv4 multicast group entry for this ip_moptions instance 433 * which matches the specified group, and optionally an interface. 434 * Return its index into the array, or -1 if not found. 435 */ 436 static size_t 437 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp, 438 const struct sockaddr *group) 439 { 440 const struct sockaddr_in *gsin; 441 struct in_multi **pinm; 442 int idx; 443 int nmships; 444 445 gsin = (const struct sockaddr_in *)group; 446 447 /* The imo_membership array may be lazy allocated. */ 448 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0) 449 return (-1); 450 451 nmships = imo->imo_num_memberships; 452 pinm = &imo->imo_membership[0]; 453 for (idx = 0; idx < nmships; idx++, pinm++) { 454 if (*pinm == NULL) 455 continue; 456 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) && 457 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) { 458 break; 459 } 460 } 461 if (idx >= nmships) 462 idx = -1; 463 464 return (idx); 465 } 466 467 /* 468 * Find an IPv4 multicast source entry for this imo which matches 469 * the given group index for this socket, and source address. 470 * 471 * NOTE: This does not check if the entry is in-mode, merely if 472 * it exists, which may not be the desired behaviour. 473 */ 474 static struct in_msource * 475 imo_match_source(const struct ip_moptions *imo, const size_t gidx, 476 const struct sockaddr *src) 477 { 478 struct ip_msource find; 479 struct in_mfilter *imf; 480 struct ip_msource *ims; 481 const sockunion_t *psa; 482 483 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__)); 484 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships, 485 ("%s: invalid index %d\n", __func__, (int)gidx)); 486 487 /* The imo_mfilters array may be lazy allocated. */ 488 if (imo->imo_mfilters == NULL) 489 return (NULL); 490 imf = &imo->imo_mfilters[gidx]; 491 492 /* Source trees are keyed in host byte order. */ 493 psa = (const sockunion_t *)src; 494 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr); 495 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 496 497 return ((struct in_msource *)ims); 498 } 499 500 /* 501 * Perform filtering for multicast datagrams on a socket by group and source. 502 * 503 * Returns 0 if a datagram should be allowed through, or various error codes 504 * if the socket was not a member of the group, or the source was muted, etc. 505 */ 506 int 507 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp, 508 const struct sockaddr *group, const struct sockaddr *src) 509 { 510 size_t gidx; 511 struct in_msource *ims; 512 int mode; 513 514 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 515 516 gidx = imo_match_group(imo, ifp, group); 517 if (gidx == -1) 518 return (MCAST_NOTGMEMBER); 519 520 /* 521 * Check if the source was included in an (S,G) join. 522 * Allow reception on exclusive memberships by default, 523 * reject reception on inclusive memberships by default. 524 * Exclude source only if an in-mode exclude filter exists. 525 * Include source only if an in-mode include filter exists. 526 * NOTE: We are comparing group state here at IGMP t1 (now) 527 * with socket-layer t0 (since last downcall). 528 */ 529 mode = imo->imo_mfilters[gidx].imf_st[1]; 530 ims = imo_match_source(imo, gidx, src); 531 532 if ((ims == NULL && mode == MCAST_INCLUDE) || 533 (ims != NULL && ims->imsl_st[0] != mode)) 534 return (MCAST_NOTSMEMBER); 535 536 return (MCAST_PASS); 537 } 538 539 /* 540 * Find and return a reference to an in_multi record for (ifp, group), 541 * and bump its reference count. 542 * If one does not exist, try to allocate it, and update link-layer multicast 543 * filters on ifp to listen for group. 544 * Assumes the IN_MULTI lock is held across the call. 545 * Return 0 if successful, otherwise return an appropriate error code. 546 */ 547 static int 548 in_getmulti(struct ifnet *ifp, const struct in_addr *group, 549 struct in_multi **pinm) 550 { 551 struct sockaddr_in gsin; 552 struct ifmultiaddr *ifma; 553 struct in_ifinfo *ii; 554 struct in_multi *inm; 555 int error; 556 557 IN_MULTI_LOCK_ASSERT(); 558 559 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET]; 560 IN_MULTI_LIST_LOCK(); 561 inm = inm_lookup(ifp, *group); 562 if (inm != NULL) { 563 /* 564 * If we already joined this group, just bump the 565 * refcount and return it. 566 */ 567 KASSERT(inm->inm_refcount >= 1, 568 ("%s: bad refcount %d", __func__, inm->inm_refcount)); 569 inm_acquire_locked(inm); 570 *pinm = inm; 571 } 572 IN_MULTI_LIST_UNLOCK(); 573 if (inm != NULL) 574 return (0); 575 576 memset(&gsin, 0, sizeof(gsin)); 577 gsin.sin_family = AF_INET; 578 gsin.sin_len = sizeof(struct sockaddr_in); 579 gsin.sin_addr = *group; 580 581 /* 582 * Check if a link-layer group is already associated 583 * with this network-layer group on the given ifnet. 584 */ 585 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma); 586 if (error != 0) 587 return (error); 588 589 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */ 590 IN_MULTI_LIST_LOCK(); 591 IF_ADDR_WLOCK(ifp); 592 593 /* 594 * If something other than netinet is occupying the link-layer 595 * group, print a meaningful error message and back out of 596 * the allocation. 597 * Otherwise, bump the refcount on the existing network-layer 598 * group association and return it. 599 */ 600 if (ifma->ifma_protospec != NULL) { 601 inm = (struct in_multi *)ifma->ifma_protospec; 602 #ifdef INVARIANTS 603 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 604 __func__)); 605 KASSERT(ifma->ifma_addr->sa_family == AF_INET, 606 ("%s: ifma not AF_INET", __func__)); 607 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 608 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp || 609 !in_hosteq(inm->inm_addr, *group)) { 610 char addrbuf[INET_ADDRSTRLEN]; 611 612 panic("%s: ifma %p is inconsistent with %p (%s)", 613 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf)); 614 } 615 #endif 616 inm_acquire_locked(inm); 617 *pinm = inm; 618 goto out_locked; 619 } 620 621 IF_ADDR_WLOCK_ASSERT(ifp); 622 623 /* 624 * A new in_multi record is needed; allocate and initialize it. 625 * We DO NOT perform an IGMP join as the in_ layer may need to 626 * push an initial source list down to IGMP to support SSM. 627 * 628 * The initial source filter state is INCLUDE, {} as per the RFC. 629 */ 630 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO); 631 if (inm == NULL) { 632 IF_ADDR_WUNLOCK(ifp); 633 IN_MULTI_LIST_UNLOCK(); 634 if_delmulti_ifma(ifma); 635 return (ENOMEM); 636 } 637 inm->inm_addr = *group; 638 inm->inm_ifp = ifp; 639 inm->inm_igi = ii->ii_igmp; 640 inm->inm_ifma = ifma; 641 inm->inm_refcount = 1; 642 inm->inm_state = IGMP_NOT_MEMBER; 643 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES); 644 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED; 645 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 646 RB_INIT(&inm->inm_srcs); 647 648 ifma->ifma_protospec = inm; 649 650 *pinm = inm; 651 out_locked: 652 IF_ADDR_WUNLOCK(ifp); 653 IN_MULTI_LIST_UNLOCK(); 654 return (0); 655 } 656 657 /* 658 * Drop a reference to an in_multi record. 659 * 660 * If the refcount drops to 0, free the in_multi record and 661 * delete the underlying link-layer membership. 662 */ 663 static void 664 inm_release(struct in_multi *inm) 665 { 666 struct ifmultiaddr *ifma; 667 struct ifnet *ifp; 668 669 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount); 670 MPASS(inm->inm_refcount == 0); 671 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm); 672 673 ifma = inm->inm_ifma; 674 ifp = inm->inm_ifp; 675 676 /* XXX this access is not covered by IF_ADDR_LOCK */ 677 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma); 678 if (ifp != NULL) { 679 CURVNET_SET(ifp->if_vnet); 680 inm_purge(inm); 681 free(inm, M_IPMADDR); 682 if_delmulti_ifma_flags(ifma, 1); 683 CURVNET_RESTORE(); 684 if_rele(ifp); 685 } else { 686 inm_purge(inm); 687 free(inm, M_IPMADDR); 688 if_delmulti_ifma_flags(ifma, 1); 689 } 690 } 691 692 /* 693 * Clear recorded source entries for a group. 694 * Used by the IGMP code. Caller must hold the IN_MULTI lock. 695 * FIXME: Should reap. 696 */ 697 void 698 inm_clear_recorded(struct in_multi *inm) 699 { 700 struct ip_msource *ims; 701 702 IN_MULTI_LIST_LOCK_ASSERT(); 703 704 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 705 if (ims->ims_stp) { 706 ims->ims_stp = 0; 707 --inm->inm_st[1].iss_rec; 708 } 709 } 710 KASSERT(inm->inm_st[1].iss_rec == 0, 711 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec)); 712 } 713 714 /* 715 * Record a source as pending for a Source-Group IGMPv3 query. 716 * This lives here as it modifies the shared tree. 717 * 718 * inm is the group descriptor. 719 * naddr is the address of the source to record in network-byte order. 720 * 721 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will 722 * lazy-allocate a source node in response to an SG query. 723 * Otherwise, no allocation is performed. This saves some memory 724 * with the trade-off that the source will not be reported to the 725 * router if joined in the window between the query response and 726 * the group actually being joined on the local host. 727 * 728 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed. 729 * This turns off the allocation of a recorded source entry if 730 * the group has not been joined. 731 * 732 * Return 0 if the source didn't exist or was already marked as recorded. 733 * Return 1 if the source was marked as recorded by this function. 734 * Return <0 if any error occurred (negated errno code). 735 */ 736 int 737 inm_record_source(struct in_multi *inm, const in_addr_t naddr) 738 { 739 struct ip_msource find; 740 struct ip_msource *ims, *nims; 741 742 IN_MULTI_LIST_LOCK_ASSERT(); 743 744 find.ims_haddr = ntohl(naddr); 745 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 746 if (ims && ims->ims_stp) 747 return (0); 748 if (ims == NULL) { 749 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 750 return (-ENOSPC); 751 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 752 M_NOWAIT | M_ZERO); 753 if (nims == NULL) 754 return (-ENOMEM); 755 nims->ims_haddr = find.ims_haddr; 756 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 757 ++inm->inm_nsrc; 758 ims = nims; 759 } 760 761 /* 762 * Mark the source as recorded and update the recorded 763 * source count. 764 */ 765 ++ims->ims_stp; 766 ++inm->inm_st[1].iss_rec; 767 768 return (1); 769 } 770 771 /* 772 * Return a pointer to an in_msource owned by an in_mfilter, 773 * given its source address. 774 * Lazy-allocate if needed. If this is a new entry its filter state is 775 * undefined at t0. 776 * 777 * imf is the filter set being modified. 778 * haddr is the source address in *host* byte-order. 779 * 780 * SMPng: May be called with locks held; malloc must not block. 781 */ 782 static int 783 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin, 784 struct in_msource **plims) 785 { 786 struct ip_msource find; 787 struct ip_msource *ims, *nims; 788 struct in_msource *lims; 789 int error; 790 791 error = 0; 792 ims = NULL; 793 lims = NULL; 794 795 /* key is host byte order */ 796 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 797 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 798 lims = (struct in_msource *)ims; 799 if (lims == NULL) { 800 if (imf->imf_nsrc == in_mcast_maxsocksrc) 801 return (ENOSPC); 802 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 803 M_NOWAIT | M_ZERO); 804 if (nims == NULL) 805 return (ENOMEM); 806 lims = (struct in_msource *)nims; 807 lims->ims_haddr = find.ims_haddr; 808 lims->imsl_st[0] = MCAST_UNDEFINED; 809 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 810 ++imf->imf_nsrc; 811 } 812 813 *plims = lims; 814 815 return (error); 816 } 817 818 /* 819 * Graft a source entry into an existing socket-layer filter set, 820 * maintaining any required invariants and checking allocations. 821 * 822 * The source is marked as being in the new filter mode at t1. 823 * 824 * Return the pointer to the new node, otherwise return NULL. 825 */ 826 static struct in_msource * 827 imf_graft(struct in_mfilter *imf, const uint8_t st1, 828 const struct sockaddr_in *psin) 829 { 830 struct ip_msource *nims; 831 struct in_msource *lims; 832 833 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 834 M_NOWAIT | M_ZERO); 835 if (nims == NULL) 836 return (NULL); 837 lims = (struct in_msource *)nims; 838 lims->ims_haddr = ntohl(psin->sin_addr.s_addr); 839 lims->imsl_st[0] = MCAST_UNDEFINED; 840 lims->imsl_st[1] = st1; 841 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 842 ++imf->imf_nsrc; 843 844 return (lims); 845 } 846 847 /* 848 * Prune a source entry from an existing socket-layer filter set, 849 * maintaining any required invariants and checking allocations. 850 * 851 * The source is marked as being left at t1, it is not freed. 852 * 853 * Return 0 if no error occurred, otherwise return an errno value. 854 */ 855 static int 856 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin) 857 { 858 struct ip_msource find; 859 struct ip_msource *ims; 860 struct in_msource *lims; 861 862 /* key is host byte order */ 863 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 864 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 865 if (ims == NULL) 866 return (ENOENT); 867 lims = (struct in_msource *)ims; 868 lims->imsl_st[1] = MCAST_UNDEFINED; 869 return (0); 870 } 871 872 /* 873 * Revert socket-layer filter set deltas at t1 to t0 state. 874 */ 875 static void 876 imf_rollback(struct in_mfilter *imf) 877 { 878 struct ip_msource *ims, *tims; 879 struct in_msource *lims; 880 881 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 882 lims = (struct in_msource *)ims; 883 if (lims->imsl_st[0] == lims->imsl_st[1]) { 884 /* no change at t1 */ 885 continue; 886 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) { 887 /* revert change to existing source at t1 */ 888 lims->imsl_st[1] = lims->imsl_st[0]; 889 } else { 890 /* revert source added t1 */ 891 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 892 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 893 free(ims, M_INMFILTER); 894 imf->imf_nsrc--; 895 } 896 } 897 imf->imf_st[1] = imf->imf_st[0]; 898 } 899 900 /* 901 * Mark socket-layer filter set as INCLUDE {} at t1. 902 */ 903 static void 904 imf_leave(struct in_mfilter *imf) 905 { 906 struct ip_msource *ims; 907 struct in_msource *lims; 908 909 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 910 lims = (struct in_msource *)ims; 911 lims->imsl_st[1] = MCAST_UNDEFINED; 912 } 913 imf->imf_st[1] = MCAST_INCLUDE; 914 } 915 916 /* 917 * Mark socket-layer filter set deltas as committed. 918 */ 919 static void 920 imf_commit(struct in_mfilter *imf) 921 { 922 struct ip_msource *ims; 923 struct in_msource *lims; 924 925 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 926 lims = (struct in_msource *)ims; 927 lims->imsl_st[0] = lims->imsl_st[1]; 928 } 929 imf->imf_st[0] = imf->imf_st[1]; 930 } 931 932 /* 933 * Reap unreferenced sources from socket-layer filter set. 934 */ 935 static void 936 imf_reap(struct in_mfilter *imf) 937 { 938 struct ip_msource *ims, *tims; 939 struct in_msource *lims; 940 941 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 942 lims = (struct in_msource *)ims; 943 if ((lims->imsl_st[0] == MCAST_UNDEFINED) && 944 (lims->imsl_st[1] == MCAST_UNDEFINED)) { 945 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims); 946 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 947 free(ims, M_INMFILTER); 948 imf->imf_nsrc--; 949 } 950 } 951 } 952 953 /* 954 * Purge socket-layer filter set. 955 */ 956 static void 957 imf_purge(struct in_mfilter *imf) 958 { 959 struct ip_msource *ims, *tims; 960 961 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 962 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 963 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 964 free(ims, M_INMFILTER); 965 imf->imf_nsrc--; 966 } 967 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED; 968 KASSERT(RB_EMPTY(&imf->imf_sources), 969 ("%s: imf_sources not empty", __func__)); 970 } 971 972 /* 973 * Look up a source filter entry for a multicast group. 974 * 975 * inm is the group descriptor to work with. 976 * haddr is the host-byte-order IPv4 address to look up. 977 * noalloc may be non-zero to suppress allocation of sources. 978 * *pims will be set to the address of the retrieved or allocated source. 979 * 980 * SMPng: NOTE: may be called with locks held. 981 * Return 0 if successful, otherwise return a non-zero error code. 982 */ 983 static int 984 inm_get_source(struct in_multi *inm, const in_addr_t haddr, 985 const int noalloc, struct ip_msource **pims) 986 { 987 struct ip_msource find; 988 struct ip_msource *ims, *nims; 989 990 find.ims_haddr = haddr; 991 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 992 if (ims == NULL && !noalloc) { 993 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 994 return (ENOSPC); 995 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 996 M_NOWAIT | M_ZERO); 997 if (nims == NULL) 998 return (ENOMEM); 999 nims->ims_haddr = haddr; 1000 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 1001 ++inm->inm_nsrc; 1002 ims = nims; 1003 #ifdef KTR 1004 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__, 1005 haddr, ims); 1006 #endif 1007 } 1008 1009 *pims = ims; 1010 return (0); 1011 } 1012 1013 /* 1014 * Merge socket-layer source into IGMP-layer source. 1015 * If rollback is non-zero, perform the inverse of the merge. 1016 */ 1017 static void 1018 ims_merge(struct ip_msource *ims, const struct in_msource *lims, 1019 const int rollback) 1020 { 1021 int n = rollback ? -1 : 1; 1022 1023 if (lims->imsl_st[0] == MCAST_EXCLUDE) { 1024 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x", 1025 __func__, n, ims->ims_haddr); 1026 ims->ims_st[1].ex -= n; 1027 } else if (lims->imsl_st[0] == MCAST_INCLUDE) { 1028 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x", 1029 __func__, n, ims->ims_haddr); 1030 ims->ims_st[1].in -= n; 1031 } 1032 1033 if (lims->imsl_st[1] == MCAST_EXCLUDE) { 1034 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x", 1035 __func__, n, ims->ims_haddr); 1036 ims->ims_st[1].ex += n; 1037 } else if (lims->imsl_st[1] == MCAST_INCLUDE) { 1038 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x", 1039 __func__, n, ims->ims_haddr); 1040 ims->ims_st[1].in += n; 1041 } 1042 } 1043 1044 /* 1045 * Atomically update the global in_multi state, when a membership's 1046 * filter list is being updated in any way. 1047 * 1048 * imf is the per-inpcb-membership group filter pointer. 1049 * A fake imf may be passed for in-kernel consumers. 1050 * 1051 * XXX This is a candidate for a set-symmetric-difference style loop 1052 * which would eliminate the repeated lookup from root of ims nodes, 1053 * as they share the same key space. 1054 * 1055 * If any error occurred this function will back out of refcounts 1056 * and return a non-zero value. 1057 */ 1058 static int 1059 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1060 { 1061 struct ip_msource *ims, *nims; 1062 struct in_msource *lims; 1063 int schanged, error; 1064 int nsrc0, nsrc1; 1065 1066 schanged = 0; 1067 error = 0; 1068 nsrc1 = nsrc0 = 0; 1069 IN_MULTI_LIST_LOCK_ASSERT(); 1070 1071 /* 1072 * Update the source filters first, as this may fail. 1073 * Maintain count of in-mode filters at t0, t1. These are 1074 * used to work out if we transition into ASM mode or not. 1075 * Maintain a count of source filters whose state was 1076 * actually modified by this operation. 1077 */ 1078 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1079 lims = (struct in_msource *)ims; 1080 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++; 1081 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++; 1082 if (lims->imsl_st[0] == lims->imsl_st[1]) continue; 1083 error = inm_get_source(inm, lims->ims_haddr, 0, &nims); 1084 ++schanged; 1085 if (error) 1086 break; 1087 ims_merge(nims, lims, 0); 1088 } 1089 if (error) { 1090 struct ip_msource *bims; 1091 1092 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) { 1093 lims = (struct in_msource *)ims; 1094 if (lims->imsl_st[0] == lims->imsl_st[1]) 1095 continue; 1096 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims); 1097 if (bims == NULL) 1098 continue; 1099 ims_merge(bims, lims, 1); 1100 } 1101 goto out_reap; 1102 } 1103 1104 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1", 1105 __func__, nsrc0, nsrc1); 1106 1107 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 1108 if (imf->imf_st[0] == imf->imf_st[1] && 1109 imf->imf_st[1] == MCAST_INCLUDE) { 1110 if (nsrc1 == 0) { 1111 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1112 --inm->inm_st[1].iss_in; 1113 } 1114 } 1115 1116 /* Handle filter mode transition on socket. */ 1117 if (imf->imf_st[0] != imf->imf_st[1]) { 1118 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d", 1119 __func__, imf->imf_st[0], imf->imf_st[1]); 1120 1121 if (imf->imf_st[0] == MCAST_EXCLUDE) { 1122 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__); 1123 --inm->inm_st[1].iss_ex; 1124 } else if (imf->imf_st[0] == MCAST_INCLUDE) { 1125 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1126 --inm->inm_st[1].iss_in; 1127 } 1128 1129 if (imf->imf_st[1] == MCAST_EXCLUDE) { 1130 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__); 1131 inm->inm_st[1].iss_ex++; 1132 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) { 1133 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__); 1134 inm->inm_st[1].iss_in++; 1135 } 1136 } 1137 1138 /* 1139 * Track inm filter state in terms of listener counts. 1140 * If there are any exclusive listeners, stack-wide 1141 * membership is exclusive. 1142 * Otherwise, if only inclusive listeners, stack-wide is inclusive. 1143 * If no listeners remain, state is undefined at t1, 1144 * and the IGMP lifecycle for this group should finish. 1145 */ 1146 if (inm->inm_st[1].iss_ex > 0) { 1147 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__); 1148 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE; 1149 } else if (inm->inm_st[1].iss_in > 0) { 1150 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__); 1151 inm->inm_st[1].iss_fmode = MCAST_INCLUDE; 1152 } else { 1153 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__); 1154 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 1155 } 1156 1157 /* Decrement ASM listener count on transition out of ASM mode. */ 1158 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1159 if ((imf->imf_st[1] != MCAST_EXCLUDE) || 1160 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) { 1161 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__); 1162 --inm->inm_st[1].iss_asm; 1163 } 1164 } 1165 1166 /* Increment ASM listener count on transition to ASM mode. */ 1167 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1168 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__); 1169 inm->inm_st[1].iss_asm++; 1170 } 1171 1172 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm); 1173 inm_print(inm); 1174 1175 out_reap: 1176 if (schanged > 0) { 1177 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__); 1178 inm_reap(inm); 1179 } 1180 return (error); 1181 } 1182 1183 /* 1184 * Mark an in_multi's filter set deltas as committed. 1185 * Called by IGMP after a state change has been enqueued. 1186 */ 1187 void 1188 inm_commit(struct in_multi *inm) 1189 { 1190 struct ip_msource *ims; 1191 1192 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm); 1193 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__); 1194 inm_print(inm); 1195 1196 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 1197 ims->ims_st[0] = ims->ims_st[1]; 1198 } 1199 inm->inm_st[0] = inm->inm_st[1]; 1200 } 1201 1202 /* 1203 * Reap unreferenced nodes from an in_multi's filter set. 1204 */ 1205 static void 1206 inm_reap(struct in_multi *inm) 1207 { 1208 struct ip_msource *ims, *tims; 1209 1210 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1211 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 || 1212 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 || 1213 ims->ims_stp != 0) 1214 continue; 1215 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1216 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1217 free(ims, M_IPMSOURCE); 1218 inm->inm_nsrc--; 1219 } 1220 } 1221 1222 /* 1223 * Purge all source nodes from an in_multi's filter set. 1224 */ 1225 static void 1226 inm_purge(struct in_multi *inm) 1227 { 1228 struct ip_msource *ims, *tims; 1229 1230 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1231 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1232 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1233 free(ims, M_IPMSOURCE); 1234 inm->inm_nsrc--; 1235 } 1236 } 1237 1238 /* 1239 * Join a multicast group; unlocked entry point. 1240 * 1241 * SMPng: XXX: in_joingroup() is called from in_control() when Giant 1242 * is not held. Fortunately, ifp is unlikely to have been detached 1243 * at this point, so we assume it's OK to recurse. 1244 */ 1245 int 1246 in_joingroup(struct ifnet *ifp, const struct in_addr *gina, 1247 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1248 { 1249 int error; 1250 1251 IN_MULTI_LOCK(); 1252 error = in_joingroup_locked(ifp, gina, imf, pinm); 1253 IN_MULTI_UNLOCK(); 1254 1255 return (error); 1256 } 1257 1258 /* 1259 * Join a multicast group; real entry point. 1260 * 1261 * Only preserves atomicity at inm level. 1262 * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1263 * 1264 * If the IGMP downcall fails, the group is not joined, and an error 1265 * code is returned. 1266 */ 1267 int 1268 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina, 1269 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1270 { 1271 struct in_mfilter timf; 1272 struct in_multi *inm; 1273 int error; 1274 1275 IN_MULTI_LOCK_ASSERT(); 1276 IN_MULTI_LIST_UNLOCK_ASSERT(); 1277 1278 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__, 1279 ntohl(gina->s_addr), ifp, ifp->if_xname); 1280 1281 error = 0; 1282 inm = NULL; 1283 1284 /* 1285 * If no imf was specified (i.e. kernel consumer), 1286 * fake one up and assume it is an ASM join. 1287 */ 1288 if (imf == NULL) { 1289 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1290 imf = &timf; 1291 } 1292 1293 error = in_getmulti(ifp, gina, &inm); 1294 if (error) { 1295 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__); 1296 return (error); 1297 } 1298 IN_MULTI_LIST_LOCK(); 1299 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1300 error = inm_merge(inm, imf); 1301 if (error) { 1302 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1303 goto out_inm_release; 1304 } 1305 1306 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1307 error = igmp_change_state(inm); 1308 if (error) { 1309 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__); 1310 goto out_inm_release; 1311 } 1312 1313 out_inm_release: 1314 if (error) { 1315 1316 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1317 inm_release_deferred(inm); 1318 } else { 1319 *pinm = inm; 1320 } 1321 IN_MULTI_LIST_UNLOCK(); 1322 1323 return (error); 1324 } 1325 1326 /* 1327 * Leave a multicast group; unlocked entry point. 1328 */ 1329 int 1330 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1331 { 1332 int error; 1333 1334 IN_MULTI_LOCK(); 1335 error = in_leavegroup_locked(inm, imf); 1336 IN_MULTI_UNLOCK(); 1337 1338 return (error); 1339 } 1340 1341 /* 1342 * Leave a multicast group; real entry point. 1343 * All source filters will be expunged. 1344 * 1345 * Only preserves atomicity at inm level. 1346 * 1347 * Holding the write lock for the INP which contains imf 1348 * is highly advisable. We can't assert for it as imf does not 1349 * contain a back-pointer to the owning inp. 1350 * 1351 * Note: This is not the same as inm_release(*) as this function also 1352 * makes a state change downcall into IGMP. 1353 */ 1354 int 1355 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1356 { 1357 struct in_mfilter timf; 1358 int error; 1359 1360 error = 0; 1361 1362 IN_MULTI_LOCK_ASSERT(); 1363 IN_MULTI_LIST_UNLOCK_ASSERT(); 1364 1365 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__, 1366 inm, ntohl(inm->inm_addr.s_addr), 1367 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname), 1368 imf); 1369 1370 /* 1371 * If no imf was specified (i.e. kernel consumer), 1372 * fake one up and assume it is an ASM join. 1373 */ 1374 if (imf == NULL) { 1375 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1376 imf = &timf; 1377 } 1378 1379 /* 1380 * Begin state merge transaction at IGMP layer. 1381 * 1382 * As this particular invocation should not cause any memory 1383 * to be allocated, and there is no opportunity to roll back 1384 * the transaction, it MUST NOT fail. 1385 */ 1386 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1387 IN_MULTI_LIST_LOCK(); 1388 error = inm_merge(inm, imf); 1389 KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1390 1391 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1392 CURVNET_SET(inm->inm_ifp->if_vnet); 1393 error = igmp_change_state(inm); 1394 IF_ADDR_WLOCK(inm->inm_ifp); 1395 inm_release_deferred(inm); 1396 IF_ADDR_WUNLOCK(inm->inm_ifp); 1397 IN_MULTI_LIST_UNLOCK(); 1398 CURVNET_RESTORE(); 1399 if (error) 1400 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1401 1402 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1403 1404 return (error); 1405 } 1406 1407 /*#ifndef BURN_BRIDGES*/ 1408 /* 1409 * Join an IPv4 multicast group in (*,G) exclusive mode. 1410 * The group must be a 224.0.0.0/24 link-scope group. 1411 * This KPI is for legacy kernel consumers only. 1412 */ 1413 struct in_multi * 1414 in_addmulti(struct in_addr *ap, struct ifnet *ifp) 1415 { 1416 struct in_multi *pinm; 1417 int error; 1418 #ifdef INVARIANTS 1419 char addrbuf[INET_ADDRSTRLEN]; 1420 #endif 1421 1422 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)), 1423 ("%s: %s not in 224.0.0.0/24", __func__, 1424 inet_ntoa_r(*ap, addrbuf))); 1425 1426 error = in_joingroup(ifp, ap, NULL, &pinm); 1427 if (error != 0) 1428 pinm = NULL; 1429 1430 return (pinm); 1431 } 1432 1433 /* 1434 * Block or unblock an ASM multicast source on an inpcb. 1435 * This implements the delta-based API described in RFC 3678. 1436 * 1437 * The delta-based API applies only to exclusive-mode memberships. 1438 * An IGMP downcall will be performed. 1439 * 1440 * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1441 * 1442 * Return 0 if successful, otherwise return an appropriate error code. 1443 */ 1444 static int 1445 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1446 { 1447 struct group_source_req gsr; 1448 struct rm_priotracker in_ifa_tracker; 1449 sockunion_t *gsa, *ssa; 1450 struct ifnet *ifp; 1451 struct in_mfilter *imf; 1452 struct ip_moptions *imo; 1453 struct in_msource *ims; 1454 struct in_multi *inm; 1455 size_t idx; 1456 uint16_t fmode; 1457 int error, doblock; 1458 1459 ifp = NULL; 1460 error = 0; 1461 doblock = 0; 1462 1463 memset(&gsr, 0, sizeof(struct group_source_req)); 1464 gsa = (sockunion_t *)&gsr.gsr_group; 1465 ssa = (sockunion_t *)&gsr.gsr_source; 1466 1467 switch (sopt->sopt_name) { 1468 case IP_BLOCK_SOURCE: 1469 case IP_UNBLOCK_SOURCE: { 1470 struct ip_mreq_source mreqs; 1471 1472 error = sooptcopyin(sopt, &mreqs, 1473 sizeof(struct ip_mreq_source), 1474 sizeof(struct ip_mreq_source)); 1475 if (error) 1476 return (error); 1477 1478 gsa->sin.sin_family = AF_INET; 1479 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1480 gsa->sin.sin_addr = mreqs.imr_multiaddr; 1481 1482 ssa->sin.sin_family = AF_INET; 1483 ssa->sin.sin_len = sizeof(struct sockaddr_in); 1484 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1485 1486 if (!in_nullhost(mreqs.imr_interface)) { 1487 IN_IFADDR_RLOCK(&in_ifa_tracker); 1488 INADDR_TO_IFP(mreqs.imr_interface, ifp); 1489 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1490 } 1491 if (sopt->sopt_name == IP_BLOCK_SOURCE) 1492 doblock = 1; 1493 1494 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 1495 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 1496 break; 1497 } 1498 1499 case MCAST_BLOCK_SOURCE: 1500 case MCAST_UNBLOCK_SOURCE: 1501 error = sooptcopyin(sopt, &gsr, 1502 sizeof(struct group_source_req), 1503 sizeof(struct group_source_req)); 1504 if (error) 1505 return (error); 1506 1507 if (gsa->sin.sin_family != AF_INET || 1508 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 1509 return (EINVAL); 1510 1511 if (ssa->sin.sin_family != AF_INET || 1512 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 1513 return (EINVAL); 1514 1515 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1516 return (EADDRNOTAVAIL); 1517 1518 ifp = ifnet_byindex(gsr.gsr_interface); 1519 1520 if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1521 doblock = 1; 1522 break; 1523 1524 default: 1525 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1526 __func__, sopt->sopt_name); 1527 return (EOPNOTSUPP); 1528 break; 1529 } 1530 1531 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1532 return (EINVAL); 1533 1534 /* 1535 * Check if we are actually a member of this group. 1536 */ 1537 imo = inp_findmoptions(inp); 1538 idx = imo_match_group(imo, ifp, &gsa->sa); 1539 if (idx == -1 || imo->imo_mfilters == NULL) { 1540 error = EADDRNOTAVAIL; 1541 goto out_inp_locked; 1542 } 1543 1544 KASSERT(imo->imo_mfilters != NULL, 1545 ("%s: imo_mfilters not allocated", __func__)); 1546 imf = &imo->imo_mfilters[idx]; 1547 inm = imo->imo_membership[idx]; 1548 1549 /* 1550 * Attempting to use the delta-based API on an 1551 * non exclusive-mode membership is an error. 1552 */ 1553 fmode = imf->imf_st[0]; 1554 if (fmode != MCAST_EXCLUDE) { 1555 error = EINVAL; 1556 goto out_inp_locked; 1557 } 1558 1559 /* 1560 * Deal with error cases up-front: 1561 * Asked to block, but already blocked; or 1562 * Asked to unblock, but nothing to unblock. 1563 * If adding a new block entry, allocate it. 1564 */ 1565 ims = imo_match_source(imo, idx, &ssa->sa); 1566 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1567 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__, 1568 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not "); 1569 error = EADDRNOTAVAIL; 1570 goto out_inp_locked; 1571 } 1572 1573 INP_WLOCK_ASSERT(inp); 1574 1575 /* 1576 * Begin state merge transaction at socket layer. 1577 */ 1578 if (doblock) { 1579 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 1580 ims = imf_graft(imf, fmode, &ssa->sin); 1581 if (ims == NULL) 1582 error = ENOMEM; 1583 } else { 1584 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 1585 error = imf_prune(imf, &ssa->sin); 1586 } 1587 1588 if (error) { 1589 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__); 1590 goto out_imf_rollback; 1591 } 1592 1593 /* 1594 * Begin state merge transaction at IGMP layer. 1595 */ 1596 IN_MULTI_LOCK(); 1597 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1598 IN_MULTI_LIST_LOCK(); 1599 error = inm_merge(inm, imf); 1600 if (error) { 1601 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1602 IN_MULTI_LIST_UNLOCK(); 1603 goto out_in_multi_locked; 1604 } 1605 1606 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1607 error = igmp_change_state(inm); 1608 IN_MULTI_LIST_UNLOCK(); 1609 if (error) 1610 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1611 1612 out_in_multi_locked: 1613 1614 IN_MULTI_UNLOCK(); 1615 out_imf_rollback: 1616 if (error) 1617 imf_rollback(imf); 1618 else 1619 imf_commit(imf); 1620 1621 imf_reap(imf); 1622 1623 out_inp_locked: 1624 INP_WUNLOCK(inp); 1625 return (error); 1626 } 1627 1628 /* 1629 * Given an inpcb, return its multicast options structure pointer. Accepts 1630 * an unlocked inpcb pointer, but will return it locked. May sleep. 1631 * 1632 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1633 * SMPng: NOTE: Returns with the INP write lock held. 1634 */ 1635 static struct ip_moptions * 1636 inp_findmoptions(struct inpcb *inp) 1637 { 1638 struct ip_moptions *imo; 1639 struct in_multi **immp; 1640 struct in_mfilter *imfp; 1641 size_t idx; 1642 1643 INP_WLOCK(inp); 1644 if (inp->inp_moptions != NULL) 1645 return (inp->inp_moptions); 1646 1647 INP_WUNLOCK(inp); 1648 1649 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK); 1650 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS, 1651 M_WAITOK | M_ZERO); 1652 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS, 1653 M_INMFILTER, M_WAITOK); 1654 1655 imo->imo_multicast_ifp = NULL; 1656 imo->imo_multicast_addr.s_addr = INADDR_ANY; 1657 imo->imo_multicast_vif = -1; 1658 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL; 1659 imo->imo_multicast_loop = in_mcast_loop; 1660 imo->imo_num_memberships = 0; 1661 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; 1662 imo->imo_membership = immp; 1663 1664 /* Initialize per-group source filters. */ 1665 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++) 1666 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE); 1667 imo->imo_mfilters = imfp; 1668 1669 INP_WLOCK(inp); 1670 if (inp->inp_moptions != NULL) { 1671 free(imfp, M_INMFILTER); 1672 free(immp, M_IPMOPTS); 1673 free(imo, M_IPMOPTS); 1674 return (inp->inp_moptions); 1675 } 1676 inp->inp_moptions = imo; 1677 return (imo); 1678 } 1679 1680 static void 1681 inp_gcmoptions(struct ip_moptions *imo) 1682 { 1683 struct in_mfilter *imf; 1684 struct in_multi *inm; 1685 struct ifnet *ifp; 1686 size_t idx, nmships; 1687 1688 nmships = imo->imo_num_memberships; 1689 for (idx = 0; idx < nmships; ++idx) { 1690 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL; 1691 if (imf) 1692 imf_leave(imf); 1693 inm = imo->imo_membership[idx]; 1694 ifp = inm->inm_ifp; 1695 if (ifp != NULL) { 1696 CURVNET_SET(ifp->if_vnet); 1697 (void)in_leavegroup(inm, imf); 1698 CURVNET_RESTORE(); 1699 } else { 1700 (void)in_leavegroup(inm, imf); 1701 } 1702 if (imf) 1703 imf_purge(imf); 1704 } 1705 1706 if (imo->imo_mfilters) 1707 free(imo->imo_mfilters, M_INMFILTER); 1708 free(imo->imo_membership, M_IPMOPTS); 1709 free(imo, M_IPMOPTS); 1710 } 1711 1712 /* 1713 * Discard the IP multicast options (and source filters). To minimize 1714 * the amount of work done while holding locks such as the INP's 1715 * pcbinfo lock (which is used in the receive path), the free 1716 * operation is deferred to the epoch callback task. 1717 */ 1718 void 1719 inp_freemoptions(struct ip_moptions *imo) 1720 { 1721 if (imo == NULL) 1722 return; 1723 inp_gcmoptions(imo); 1724 } 1725 1726 /* 1727 * Atomically get source filters on a socket for an IPv4 multicast group. 1728 * Called with INP lock held; returns with lock released. 1729 */ 1730 static int 1731 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1732 { 1733 struct __msfilterreq msfr; 1734 sockunion_t *gsa; 1735 struct ifnet *ifp; 1736 struct ip_moptions *imo; 1737 struct in_mfilter *imf; 1738 struct ip_msource *ims; 1739 struct in_msource *lims; 1740 struct sockaddr_in *psin; 1741 struct sockaddr_storage *ptss; 1742 struct sockaddr_storage *tss; 1743 int error; 1744 size_t idx, nsrcs, ncsrcs; 1745 1746 INP_WLOCK_ASSERT(inp); 1747 1748 imo = inp->inp_moptions; 1749 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__)); 1750 1751 INP_WUNLOCK(inp); 1752 1753 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1754 sizeof(struct __msfilterreq)); 1755 if (error) 1756 return (error); 1757 1758 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 1759 return (EINVAL); 1760 1761 ifp = ifnet_byindex(msfr.msfr_ifindex); 1762 if (ifp == NULL) 1763 return (EINVAL); 1764 1765 INP_WLOCK(inp); 1766 1767 /* 1768 * Lookup group on the socket. 1769 */ 1770 gsa = (sockunion_t *)&msfr.msfr_group; 1771 idx = imo_match_group(imo, ifp, &gsa->sa); 1772 if (idx == -1 || imo->imo_mfilters == NULL) { 1773 INP_WUNLOCK(inp); 1774 return (EADDRNOTAVAIL); 1775 } 1776 imf = &imo->imo_mfilters[idx]; 1777 1778 /* 1779 * Ignore memberships which are in limbo. 1780 */ 1781 if (imf->imf_st[1] == MCAST_UNDEFINED) { 1782 INP_WUNLOCK(inp); 1783 return (EAGAIN); 1784 } 1785 msfr.msfr_fmode = imf->imf_st[1]; 1786 1787 /* 1788 * If the user specified a buffer, copy out the source filter 1789 * entries to userland gracefully. 1790 * We only copy out the number of entries which userland 1791 * has asked for, but we always tell userland how big the 1792 * buffer really needs to be. 1793 */ 1794 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 1795 msfr.msfr_nsrcs = in_mcast_maxsocksrc; 1796 tss = NULL; 1797 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1798 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1799 M_TEMP, M_NOWAIT | M_ZERO); 1800 if (tss == NULL) { 1801 INP_WUNLOCK(inp); 1802 return (ENOBUFS); 1803 } 1804 } 1805 1806 /* 1807 * Count number of sources in-mode at t0. 1808 * If buffer space exists and remains, copy out source entries. 1809 */ 1810 nsrcs = msfr.msfr_nsrcs; 1811 ncsrcs = 0; 1812 ptss = tss; 1813 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1814 lims = (struct in_msource *)ims; 1815 if (lims->imsl_st[0] == MCAST_UNDEFINED || 1816 lims->imsl_st[0] != imf->imf_st[0]) 1817 continue; 1818 ++ncsrcs; 1819 if (tss != NULL && nsrcs > 0) { 1820 psin = (struct sockaddr_in *)ptss; 1821 psin->sin_family = AF_INET; 1822 psin->sin_len = sizeof(struct sockaddr_in); 1823 psin->sin_addr.s_addr = htonl(lims->ims_haddr); 1824 psin->sin_port = 0; 1825 ++ptss; 1826 --nsrcs; 1827 } 1828 } 1829 1830 INP_WUNLOCK(inp); 1831 1832 if (tss != NULL) { 1833 error = copyout(tss, msfr.msfr_srcs, 1834 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1835 free(tss, M_TEMP); 1836 if (error) 1837 return (error); 1838 } 1839 1840 msfr.msfr_nsrcs = ncsrcs; 1841 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1842 1843 return (error); 1844 } 1845 1846 /* 1847 * Return the IP multicast options in response to user getsockopt(). 1848 */ 1849 int 1850 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1851 { 1852 struct rm_priotracker in_ifa_tracker; 1853 struct ip_mreqn mreqn; 1854 struct ip_moptions *imo; 1855 struct ifnet *ifp; 1856 struct in_ifaddr *ia; 1857 int error, optval; 1858 u_char coptval; 1859 1860 INP_WLOCK(inp); 1861 imo = inp->inp_moptions; 1862 /* 1863 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 1864 * or is a divert socket, reject it. 1865 */ 1866 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 1867 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1868 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) { 1869 INP_WUNLOCK(inp); 1870 return (EOPNOTSUPP); 1871 } 1872 1873 error = 0; 1874 switch (sopt->sopt_name) { 1875 case IP_MULTICAST_VIF: 1876 if (imo != NULL) 1877 optval = imo->imo_multicast_vif; 1878 else 1879 optval = -1; 1880 INP_WUNLOCK(inp); 1881 error = sooptcopyout(sopt, &optval, sizeof(int)); 1882 break; 1883 1884 case IP_MULTICAST_IF: 1885 memset(&mreqn, 0, sizeof(struct ip_mreqn)); 1886 if (imo != NULL) { 1887 ifp = imo->imo_multicast_ifp; 1888 if (!in_nullhost(imo->imo_multicast_addr)) { 1889 mreqn.imr_address = imo->imo_multicast_addr; 1890 } else if (ifp != NULL) { 1891 struct epoch_tracker et; 1892 1893 mreqn.imr_ifindex = ifp->if_index; 1894 NET_EPOCH_ENTER(et); 1895 IFP_TO_IA(ifp, ia, &in_ifa_tracker); 1896 if (ia != NULL) 1897 mreqn.imr_address = 1898 IA_SIN(ia)->sin_addr; 1899 NET_EPOCH_EXIT(et); 1900 } 1901 } 1902 INP_WUNLOCK(inp); 1903 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 1904 error = sooptcopyout(sopt, &mreqn, 1905 sizeof(struct ip_mreqn)); 1906 } else { 1907 error = sooptcopyout(sopt, &mreqn.imr_address, 1908 sizeof(struct in_addr)); 1909 } 1910 break; 1911 1912 case IP_MULTICAST_TTL: 1913 if (imo == NULL) 1914 optval = coptval = IP_DEFAULT_MULTICAST_TTL; 1915 else 1916 optval = coptval = imo->imo_multicast_ttl; 1917 INP_WUNLOCK(inp); 1918 if (sopt->sopt_valsize == sizeof(u_char)) 1919 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1920 else 1921 error = sooptcopyout(sopt, &optval, sizeof(int)); 1922 break; 1923 1924 case IP_MULTICAST_LOOP: 1925 if (imo == NULL) 1926 optval = coptval = IP_DEFAULT_MULTICAST_LOOP; 1927 else 1928 optval = coptval = imo->imo_multicast_loop; 1929 INP_WUNLOCK(inp); 1930 if (sopt->sopt_valsize == sizeof(u_char)) 1931 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1932 else 1933 error = sooptcopyout(sopt, &optval, sizeof(int)); 1934 break; 1935 1936 case IP_MSFILTER: 1937 if (imo == NULL) { 1938 error = EADDRNOTAVAIL; 1939 INP_WUNLOCK(inp); 1940 } else { 1941 error = inp_get_source_filters(inp, sopt); 1942 } 1943 break; 1944 1945 default: 1946 INP_WUNLOCK(inp); 1947 error = ENOPROTOOPT; 1948 break; 1949 } 1950 1951 INP_UNLOCK_ASSERT(inp); 1952 1953 return (error); 1954 } 1955 1956 /* 1957 * Look up the ifnet to use for a multicast group membership, 1958 * given the IPv4 address of an interface, and the IPv4 group address. 1959 * 1960 * This routine exists to support legacy multicast applications 1961 * which do not understand that multicast memberships are scoped to 1962 * specific physical links in the networking stack, or which need 1963 * to join link-scope groups before IPv4 addresses are configured. 1964 * 1965 * If inp is non-NULL, use this socket's current FIB number for any 1966 * required FIB lookup. 1967 * If ina is INADDR_ANY, look up the group address in the unicast FIB, 1968 * and use its ifp; usually, this points to the default next-hop. 1969 * 1970 * If the FIB lookup fails, attempt to use the first non-loopback 1971 * interface with multicast capability in the system as a 1972 * last resort. The legacy IPv4 ASM API requires that we do 1973 * this in order to allow groups to be joined when the routing 1974 * table has not yet been populated during boot. 1975 * 1976 * Returns NULL if no ifp could be found. 1977 * 1978 * FUTURE: Implement IPv4 source-address selection. 1979 */ 1980 static struct ifnet * 1981 inp_lookup_mcast_ifp(const struct inpcb *inp, 1982 const struct sockaddr_in *gsin, const struct in_addr ina) 1983 { 1984 struct rm_priotracker in_ifa_tracker; 1985 struct ifnet *ifp; 1986 struct nhop4_basic nh4; 1987 uint32_t fibnum; 1988 1989 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__)); 1990 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)), 1991 ("%s: not multicast", __func__)); 1992 1993 ifp = NULL; 1994 if (!in_nullhost(ina)) { 1995 IN_IFADDR_RLOCK(&in_ifa_tracker); 1996 INADDR_TO_IFP(ina, ifp); 1997 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1998 } else { 1999 fibnum = inp ? inp->inp_inc.inc_fibnum : 0; 2000 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0) 2001 ifp = nh4.nh_ifp; 2002 else { 2003 struct in_ifaddr *ia; 2004 struct ifnet *mifp; 2005 2006 mifp = NULL; 2007 IN_IFADDR_RLOCK(&in_ifa_tracker); 2008 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 2009 mifp = ia->ia_ifp; 2010 if (!(mifp->if_flags & IFF_LOOPBACK) && 2011 (mifp->if_flags & IFF_MULTICAST)) { 2012 ifp = mifp; 2013 break; 2014 } 2015 } 2016 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2017 } 2018 } 2019 2020 return (ifp); 2021 } 2022 2023 /* 2024 * Join an IPv4 multicast group, possibly with a source. 2025 */ 2026 static int 2027 inp_join_group(struct inpcb *inp, struct sockopt *sopt) 2028 { 2029 struct group_source_req gsr; 2030 sockunion_t *gsa, *ssa; 2031 struct ifnet *ifp; 2032 struct in_mfilter *imf; 2033 struct ip_moptions *imo; 2034 struct in_multi *inm; 2035 struct in_msource *lims; 2036 size_t idx; 2037 int error, is_new; 2038 2039 ifp = NULL; 2040 imf = NULL; 2041 lims = NULL; 2042 error = 0; 2043 is_new = 0; 2044 2045 memset(&gsr, 0, sizeof(struct group_source_req)); 2046 gsa = (sockunion_t *)&gsr.gsr_group; 2047 gsa->ss.ss_family = AF_UNSPEC; 2048 ssa = (sockunion_t *)&gsr.gsr_source; 2049 ssa->ss.ss_family = AF_UNSPEC; 2050 2051 switch (sopt->sopt_name) { 2052 case IP_ADD_MEMBERSHIP: { 2053 struct ip_mreqn mreqn; 2054 2055 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) 2056 error = sooptcopyin(sopt, &mreqn, 2057 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn)); 2058 else 2059 error = sooptcopyin(sopt, &mreqn, 2060 sizeof(struct ip_mreq), sizeof(struct ip_mreq)); 2061 if (error) 2062 return (error); 2063 2064 gsa->sin.sin_family = AF_INET; 2065 gsa->sin.sin_len = sizeof(struct sockaddr_in); 2066 gsa->sin.sin_addr = mreqn.imr_multiaddr; 2067 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2068 return (EINVAL); 2069 2070 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) && 2071 mreqn.imr_ifindex != 0) 2072 ifp = ifnet_byindex(mreqn.imr_ifindex); 2073 else 2074 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 2075 mreqn.imr_address); 2076 break; 2077 } 2078 case IP_ADD_SOURCE_MEMBERSHIP: { 2079 struct ip_mreq_source mreqs; 2080 2081 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source), 2082 sizeof(struct ip_mreq_source)); 2083 if (error) 2084 return (error); 2085 2086 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET; 2087 gsa->sin.sin_len = ssa->sin.sin_len = 2088 sizeof(struct sockaddr_in); 2089 2090 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2091 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2092 return (EINVAL); 2093 2094 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2095 2096 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 2097 mreqs.imr_interface); 2098 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2099 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2100 break; 2101 } 2102 2103 case MCAST_JOIN_GROUP: 2104 case MCAST_JOIN_SOURCE_GROUP: 2105 if (sopt->sopt_name == MCAST_JOIN_GROUP) { 2106 error = sooptcopyin(sopt, &gsr, 2107 sizeof(struct group_req), 2108 sizeof(struct group_req)); 2109 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2110 error = sooptcopyin(sopt, &gsr, 2111 sizeof(struct group_source_req), 2112 sizeof(struct group_source_req)); 2113 } 2114 if (error) 2115 return (error); 2116 2117 if (gsa->sin.sin_family != AF_INET || 2118 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2119 return (EINVAL); 2120 2121 /* 2122 * Overwrite the port field if present, as the sockaddr 2123 * being copied in may be matched with a binary comparison. 2124 */ 2125 gsa->sin.sin_port = 0; 2126 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2127 if (ssa->sin.sin_family != AF_INET || 2128 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2129 return (EINVAL); 2130 ssa->sin.sin_port = 0; 2131 } 2132 2133 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2134 return (EINVAL); 2135 2136 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2137 return (EADDRNOTAVAIL); 2138 ifp = ifnet_byindex(gsr.gsr_interface); 2139 break; 2140 2141 default: 2142 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2143 __func__, sopt->sopt_name); 2144 return (EOPNOTSUPP); 2145 break; 2146 } 2147 2148 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) 2149 return (EADDRNOTAVAIL); 2150 2151 imo = inp_findmoptions(inp); 2152 idx = imo_match_group(imo, ifp, &gsa->sa); 2153 if (idx == -1) { 2154 is_new = 1; 2155 } else { 2156 inm = imo->imo_membership[idx]; 2157 imf = &imo->imo_mfilters[idx]; 2158 if (ssa->ss.ss_family != AF_UNSPEC) { 2159 /* 2160 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership 2161 * is an error. On an existing inclusive membership, 2162 * it just adds the source to the filter list. 2163 */ 2164 if (imf->imf_st[1] != MCAST_INCLUDE) { 2165 error = EINVAL; 2166 goto out_inp_locked; 2167 } 2168 /* 2169 * Throw out duplicates. 2170 * 2171 * XXX FIXME: This makes a naive assumption that 2172 * even if entries exist for *ssa in this imf, 2173 * they will be rejected as dupes, even if they 2174 * are not valid in the current mode (in-mode). 2175 * 2176 * in_msource is transactioned just as for anything 2177 * else in SSM -- but note naive use of inm_graft() 2178 * below for allocating new filter entries. 2179 * 2180 * This is only an issue if someone mixes the 2181 * full-state SSM API with the delta-based API, 2182 * which is discouraged in the relevant RFCs. 2183 */ 2184 lims = imo_match_source(imo, idx, &ssa->sa); 2185 if (lims != NULL /*&& 2186 lims->imsl_st[1] == MCAST_INCLUDE*/) { 2187 error = EADDRNOTAVAIL; 2188 goto out_inp_locked; 2189 } 2190 } else { 2191 /* 2192 * MCAST_JOIN_GROUP on an existing exclusive 2193 * membership is an error; return EADDRINUSE 2194 * to preserve 4.4BSD API idempotence, and 2195 * avoid tedious detour to code below. 2196 * NOTE: This is bending RFC 3678 a bit. 2197 * 2198 * On an existing inclusive membership, this is also 2199 * an error; if you want to change filter mode, 2200 * you must use the userland API setsourcefilter(). 2201 * XXX We don't reject this for imf in UNDEFINED 2202 * state at t1, because allocation of a filter 2203 * is atomic with allocation of a membership. 2204 */ 2205 error = EINVAL; 2206 if (imf->imf_st[1] == MCAST_EXCLUDE) 2207 error = EADDRINUSE; 2208 goto out_inp_locked; 2209 } 2210 } 2211 2212 /* 2213 * Begin state merge transaction at socket layer. 2214 */ 2215 INP_WLOCK_ASSERT(inp); 2216 2217 if (is_new) { 2218 if (imo->imo_num_memberships == imo->imo_max_memberships) { 2219 error = imo_grow(imo); 2220 if (error) 2221 goto out_inp_locked; 2222 } 2223 /* 2224 * Allocate the new slot upfront so we can deal with 2225 * grafting the new source filter in same code path 2226 * as for join-source on existing membership. 2227 */ 2228 idx = imo->imo_num_memberships; 2229 imo->imo_membership[idx] = NULL; 2230 imo->imo_num_memberships++; 2231 KASSERT(imo->imo_mfilters != NULL, 2232 ("%s: imf_mfilters vector was not allocated", __func__)); 2233 imf = &imo->imo_mfilters[idx]; 2234 KASSERT(RB_EMPTY(&imf->imf_sources), 2235 ("%s: imf_sources not empty", __func__)); 2236 } 2237 2238 /* 2239 * Graft new source into filter list for this inpcb's 2240 * membership of the group. The in_multi may not have 2241 * been allocated yet if this is a new membership, however, 2242 * the in_mfilter slot will be allocated and must be initialized. 2243 * 2244 * Note: Grafting of exclusive mode filters doesn't happen 2245 * in this path. 2246 * XXX: Should check for non-NULL lims (node exists but may 2247 * not be in-mode) for interop with full-state API. 2248 */ 2249 if (ssa->ss.ss_family != AF_UNSPEC) { 2250 /* Membership starts in IN mode */ 2251 if (is_new) { 2252 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__); 2253 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE); 2254 } else { 2255 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 2256 } 2257 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin); 2258 if (lims == NULL) { 2259 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2260 __func__); 2261 error = ENOMEM; 2262 goto out_imo_free; 2263 } 2264 } else { 2265 /* No address specified; Membership starts in EX mode */ 2266 if (is_new) { 2267 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__); 2268 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE); 2269 } 2270 } 2271 2272 /* 2273 * Begin state merge transaction at IGMP layer. 2274 */ 2275 in_pcbref(inp); 2276 INP_WUNLOCK(inp); 2277 IN_MULTI_LOCK(); 2278 2279 if (is_new) { 2280 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf, 2281 &inm); 2282 if (error) { 2283 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed", 2284 __func__); 2285 IN_MULTI_LIST_UNLOCK(); 2286 goto out_imo_free; 2287 } 2288 inm_acquire(inm); 2289 imo->imo_membership[idx] = inm; 2290 } else { 2291 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2292 IN_MULTI_LIST_LOCK(); 2293 error = inm_merge(inm, imf); 2294 if (error) { 2295 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2296 __func__); 2297 IN_MULTI_LIST_UNLOCK(); 2298 goto out_in_multi_locked; 2299 } 2300 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2301 error = igmp_change_state(inm); 2302 IN_MULTI_LIST_UNLOCK(); 2303 if (error) { 2304 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2305 __func__); 2306 goto out_in_multi_locked; 2307 } 2308 } 2309 2310 out_in_multi_locked: 2311 2312 IN_MULTI_UNLOCK(); 2313 INP_WLOCK(inp); 2314 if (in_pcbrele_wlocked(inp)) 2315 return (ENXIO); 2316 if (error) { 2317 imf_rollback(imf); 2318 if (is_new) 2319 imf_purge(imf); 2320 else 2321 imf_reap(imf); 2322 } else { 2323 imf_commit(imf); 2324 } 2325 2326 out_imo_free: 2327 if (error && is_new) { 2328 inm = imo->imo_membership[idx]; 2329 if (inm != NULL) { 2330 IN_MULTI_LIST_LOCK(); 2331 inm_release_deferred(inm); 2332 IN_MULTI_LIST_UNLOCK(); 2333 } 2334 imo->imo_membership[idx] = NULL; 2335 --imo->imo_num_memberships; 2336 } 2337 2338 out_inp_locked: 2339 INP_WUNLOCK(inp); 2340 return (error); 2341 } 2342 2343 /* 2344 * Leave an IPv4 multicast group on an inpcb, possibly with a source. 2345 */ 2346 static int 2347 inp_leave_group(struct inpcb *inp, struct sockopt *sopt) 2348 { 2349 struct group_source_req gsr; 2350 struct ip_mreq_source mreqs; 2351 struct rm_priotracker in_ifa_tracker; 2352 sockunion_t *gsa, *ssa; 2353 struct ifnet *ifp; 2354 struct in_mfilter *imf; 2355 struct ip_moptions *imo; 2356 struct in_msource *ims; 2357 struct in_multi *inm; 2358 size_t idx; 2359 int error, is_final; 2360 2361 ifp = NULL; 2362 error = 0; 2363 is_final = 1; 2364 2365 memset(&gsr, 0, sizeof(struct group_source_req)); 2366 gsa = (sockunion_t *)&gsr.gsr_group; 2367 gsa->ss.ss_family = AF_UNSPEC; 2368 ssa = (sockunion_t *)&gsr.gsr_source; 2369 ssa->ss.ss_family = AF_UNSPEC; 2370 2371 switch (sopt->sopt_name) { 2372 case IP_DROP_MEMBERSHIP: 2373 case IP_DROP_SOURCE_MEMBERSHIP: 2374 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) { 2375 error = sooptcopyin(sopt, &mreqs, 2376 sizeof(struct ip_mreq), 2377 sizeof(struct ip_mreq)); 2378 /* 2379 * Swap interface and sourceaddr arguments, 2380 * as ip_mreq and ip_mreq_source are laid 2381 * out differently. 2382 */ 2383 mreqs.imr_interface = mreqs.imr_sourceaddr; 2384 mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 2385 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2386 error = sooptcopyin(sopt, &mreqs, 2387 sizeof(struct ip_mreq_source), 2388 sizeof(struct ip_mreq_source)); 2389 } 2390 if (error) 2391 return (error); 2392 2393 gsa->sin.sin_family = AF_INET; 2394 gsa->sin.sin_len = sizeof(struct sockaddr_in); 2395 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2396 2397 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2398 ssa->sin.sin_family = AF_INET; 2399 ssa->sin.sin_len = sizeof(struct sockaddr_in); 2400 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2401 } 2402 2403 /* 2404 * Attempt to look up hinted ifp from interface address. 2405 * Fallthrough with null ifp iff lookup fails, to 2406 * preserve 4.4BSD mcast API idempotence. 2407 * XXX NOTE WELL: The RFC 3678 API is preferred because 2408 * using an IPv4 address as a key is racy. 2409 */ 2410 if (!in_nullhost(mreqs.imr_interface)) { 2411 IN_IFADDR_RLOCK(&in_ifa_tracker); 2412 INADDR_TO_IFP(mreqs.imr_interface, ifp); 2413 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2414 } 2415 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2416 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2417 2418 break; 2419 2420 case MCAST_LEAVE_GROUP: 2421 case MCAST_LEAVE_SOURCE_GROUP: 2422 if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2423 error = sooptcopyin(sopt, &gsr, 2424 sizeof(struct group_req), 2425 sizeof(struct group_req)); 2426 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2427 error = sooptcopyin(sopt, &gsr, 2428 sizeof(struct group_source_req), 2429 sizeof(struct group_source_req)); 2430 } 2431 if (error) 2432 return (error); 2433 2434 if (gsa->sin.sin_family != AF_INET || 2435 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2436 return (EINVAL); 2437 2438 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2439 if (ssa->sin.sin_family != AF_INET || 2440 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2441 return (EINVAL); 2442 } 2443 2444 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2445 return (EADDRNOTAVAIL); 2446 2447 ifp = ifnet_byindex(gsr.gsr_interface); 2448 2449 if (ifp == NULL) 2450 return (EADDRNOTAVAIL); 2451 break; 2452 2453 default: 2454 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2455 __func__, sopt->sopt_name); 2456 return (EOPNOTSUPP); 2457 break; 2458 } 2459 2460 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2461 return (EINVAL); 2462 2463 /* 2464 * Find the membership in the membership array. 2465 */ 2466 imo = inp_findmoptions(inp); 2467 idx = imo_match_group(imo, ifp, &gsa->sa); 2468 if (idx == -1) { 2469 error = EADDRNOTAVAIL; 2470 goto out_inp_locked; 2471 } 2472 inm = imo->imo_membership[idx]; 2473 imf = &imo->imo_mfilters[idx]; 2474 2475 if (ssa->ss.ss_family != AF_UNSPEC) 2476 is_final = 0; 2477 2478 /* 2479 * Begin state merge transaction at socket layer. 2480 */ 2481 INP_WLOCK_ASSERT(inp); 2482 2483 /* 2484 * If we were instructed only to leave a given source, do so. 2485 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2486 */ 2487 if (is_final) { 2488 imf_leave(imf); 2489 } else { 2490 if (imf->imf_st[0] == MCAST_EXCLUDE) { 2491 error = EADDRNOTAVAIL; 2492 goto out_inp_locked; 2493 } 2494 ims = imo_match_source(imo, idx, &ssa->sa); 2495 if (ims == NULL) { 2496 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", 2497 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not "); 2498 error = EADDRNOTAVAIL; 2499 goto out_inp_locked; 2500 } 2501 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 2502 error = imf_prune(imf, &ssa->sin); 2503 if (error) { 2504 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2505 __func__); 2506 goto out_inp_locked; 2507 } 2508 } 2509 2510 /* 2511 * Begin state merge transaction at IGMP layer. 2512 */ 2513 in_pcbref(inp); 2514 INP_WUNLOCK(inp); 2515 IN_MULTI_LOCK(); 2516 2517 if (is_final) { 2518 /* 2519 * Give up the multicast address record to which 2520 * the membership points. 2521 */ 2522 (void)in_leavegroup_locked(inm, imf); 2523 } else { 2524 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2525 IN_MULTI_LIST_LOCK(); 2526 error = inm_merge(inm, imf); 2527 if (error) { 2528 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2529 __func__); 2530 IN_MULTI_LIST_UNLOCK(); 2531 goto out_in_multi_locked; 2532 } 2533 2534 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2535 error = igmp_change_state(inm); 2536 IN_MULTI_LIST_UNLOCK(); 2537 if (error) { 2538 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2539 __func__); 2540 } 2541 } 2542 2543 out_in_multi_locked: 2544 2545 IN_MULTI_UNLOCK(); 2546 INP_WLOCK(inp); 2547 if (in_pcbrele_wlocked(inp)) 2548 return (ENXIO); 2549 2550 if (error) 2551 imf_rollback(imf); 2552 else 2553 imf_commit(imf); 2554 2555 imf_reap(imf); 2556 2557 if (is_final) { 2558 /* Remove the gap in the membership and filter array. */ 2559 KASSERT(RB_EMPTY(&imf->imf_sources), 2560 ("%s: imf_sources not empty", __func__)); 2561 for (++idx; idx < imo->imo_num_memberships; ++idx) { 2562 imo->imo_membership[idx - 1] = imo->imo_membership[idx]; 2563 imo->imo_mfilters[idx - 1] = imo->imo_mfilters[idx]; 2564 } 2565 imf_init(&imo->imo_mfilters[idx - 1], MCAST_UNDEFINED, 2566 MCAST_EXCLUDE); 2567 imo->imo_num_memberships--; 2568 } 2569 2570 out_inp_locked: 2571 INP_WUNLOCK(inp); 2572 return (error); 2573 } 2574 2575 /* 2576 * Select the interface for transmitting IPv4 multicast datagrams. 2577 * 2578 * Either an instance of struct in_addr or an instance of struct ip_mreqn 2579 * may be passed to this socket option. An address of INADDR_ANY or an 2580 * interface index of 0 is used to remove a previous selection. 2581 * When no interface is selected, one is chosen for every send. 2582 */ 2583 static int 2584 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2585 { 2586 struct rm_priotracker in_ifa_tracker; 2587 struct in_addr addr; 2588 struct ip_mreqn mreqn; 2589 struct ifnet *ifp; 2590 struct ip_moptions *imo; 2591 int error; 2592 2593 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 2594 /* 2595 * An interface index was specified using the 2596 * Linux-derived ip_mreqn structure. 2597 */ 2598 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn), 2599 sizeof(struct ip_mreqn)); 2600 if (error) 2601 return (error); 2602 2603 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex) 2604 return (EINVAL); 2605 2606 if (mreqn.imr_ifindex == 0) { 2607 ifp = NULL; 2608 } else { 2609 ifp = ifnet_byindex(mreqn.imr_ifindex); 2610 if (ifp == NULL) 2611 return (EADDRNOTAVAIL); 2612 } 2613 } else { 2614 /* 2615 * An interface was specified by IPv4 address. 2616 * This is the traditional BSD usage. 2617 */ 2618 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr), 2619 sizeof(struct in_addr)); 2620 if (error) 2621 return (error); 2622 if (in_nullhost(addr)) { 2623 ifp = NULL; 2624 } else { 2625 IN_IFADDR_RLOCK(&in_ifa_tracker); 2626 INADDR_TO_IFP(addr, ifp); 2627 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2628 if (ifp == NULL) 2629 return (EADDRNOTAVAIL); 2630 } 2631 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp, 2632 ntohl(addr.s_addr)); 2633 } 2634 2635 /* Reject interfaces which do not support multicast. */ 2636 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0) 2637 return (EOPNOTSUPP); 2638 2639 imo = inp_findmoptions(inp); 2640 imo->imo_multicast_ifp = ifp; 2641 imo->imo_multicast_addr.s_addr = INADDR_ANY; 2642 INP_WUNLOCK(inp); 2643 2644 return (0); 2645 } 2646 2647 /* 2648 * Atomically set source filters on a socket for an IPv4 multicast group. 2649 * 2650 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2651 */ 2652 static int 2653 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2654 { 2655 struct __msfilterreq msfr; 2656 sockunion_t *gsa; 2657 struct ifnet *ifp; 2658 struct in_mfilter *imf; 2659 struct ip_moptions *imo; 2660 struct in_multi *inm; 2661 size_t idx; 2662 int error; 2663 2664 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2665 sizeof(struct __msfilterreq)); 2666 if (error) 2667 return (error); 2668 2669 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 2670 return (ENOBUFS); 2671 2672 if ((msfr.msfr_fmode != MCAST_EXCLUDE && 2673 msfr.msfr_fmode != MCAST_INCLUDE)) 2674 return (EINVAL); 2675 2676 if (msfr.msfr_group.ss_family != AF_INET || 2677 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in)) 2678 return (EINVAL); 2679 2680 gsa = (sockunion_t *)&msfr.msfr_group; 2681 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2682 return (EINVAL); 2683 2684 gsa->sin.sin_port = 0; /* ignore port */ 2685 2686 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 2687 return (EADDRNOTAVAIL); 2688 2689 ifp = ifnet_byindex(msfr.msfr_ifindex); 2690 if (ifp == NULL) 2691 return (EADDRNOTAVAIL); 2692 2693 /* 2694 * Take the INP write lock. 2695 * Check if this socket is a member of this group. 2696 */ 2697 imo = inp_findmoptions(inp); 2698 idx = imo_match_group(imo, ifp, &gsa->sa); 2699 if (idx == -1 || imo->imo_mfilters == NULL) { 2700 error = EADDRNOTAVAIL; 2701 goto out_inp_locked; 2702 } 2703 inm = imo->imo_membership[idx]; 2704 imf = &imo->imo_mfilters[idx]; 2705 2706 /* 2707 * Begin state merge transaction at socket layer. 2708 */ 2709 INP_WLOCK_ASSERT(inp); 2710 2711 imf->imf_st[1] = msfr.msfr_fmode; 2712 2713 /* 2714 * Apply any new source filters, if present. 2715 * Make a copy of the user-space source vector so 2716 * that we may copy them with a single copyin. This 2717 * allows us to deal with page faults up-front. 2718 */ 2719 if (msfr.msfr_nsrcs > 0) { 2720 struct in_msource *lims; 2721 struct sockaddr_in *psin; 2722 struct sockaddr_storage *kss, *pkss; 2723 int i; 2724 2725 INP_WUNLOCK(inp); 2726 2727 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries", 2728 __func__, (unsigned long)msfr.msfr_nsrcs); 2729 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2730 M_TEMP, M_WAITOK); 2731 error = copyin(msfr.msfr_srcs, kss, 2732 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2733 if (error) { 2734 free(kss, M_TEMP); 2735 return (error); 2736 } 2737 2738 INP_WLOCK(inp); 2739 2740 /* 2741 * Mark all source filters as UNDEFINED at t1. 2742 * Restore new group filter mode, as imf_leave() 2743 * will set it to INCLUDE. 2744 */ 2745 imf_leave(imf); 2746 imf->imf_st[1] = msfr.msfr_fmode; 2747 2748 /* 2749 * Update socket layer filters at t1, lazy-allocating 2750 * new entries. This saves a bunch of memory at the 2751 * cost of one RB_FIND() per source entry; duplicate 2752 * entries in the msfr_nsrcs vector are ignored. 2753 * If we encounter an error, rollback transaction. 2754 * 2755 * XXX This too could be replaced with a set-symmetric 2756 * difference like loop to avoid walking from root 2757 * every time, as the key space is common. 2758 */ 2759 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2760 psin = (struct sockaddr_in *)pkss; 2761 if (psin->sin_family != AF_INET) { 2762 error = EAFNOSUPPORT; 2763 break; 2764 } 2765 if (psin->sin_len != sizeof(struct sockaddr_in)) { 2766 error = EINVAL; 2767 break; 2768 } 2769 error = imf_get_source(imf, psin, &lims); 2770 if (error) 2771 break; 2772 lims->imsl_st[1] = imf->imf_st[1]; 2773 } 2774 free(kss, M_TEMP); 2775 } 2776 2777 if (error) 2778 goto out_imf_rollback; 2779 2780 INP_WLOCK_ASSERT(inp); 2781 IN_MULTI_LOCK(); 2782 2783 /* 2784 * Begin state merge transaction at IGMP layer. 2785 */ 2786 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2787 IN_MULTI_LIST_LOCK(); 2788 error = inm_merge(inm, imf); 2789 if (error) { 2790 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 2791 IN_MULTI_LIST_UNLOCK(); 2792 goto out_in_multi_locked; 2793 } 2794 2795 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2796 error = igmp_change_state(inm); 2797 IN_MULTI_LIST_UNLOCK(); 2798 if (error) 2799 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 2800 2801 out_in_multi_locked: 2802 2803 IN_MULTI_UNLOCK(); 2804 2805 out_imf_rollback: 2806 if (error) 2807 imf_rollback(imf); 2808 else 2809 imf_commit(imf); 2810 2811 imf_reap(imf); 2812 2813 out_inp_locked: 2814 INP_WUNLOCK(inp); 2815 return (error); 2816 } 2817 2818 /* 2819 * Set the IP multicast options in response to user setsockopt(). 2820 * 2821 * Many of the socket options handled in this function duplicate the 2822 * functionality of socket options in the regular unicast API. However, 2823 * it is not possible to merge the duplicate code, because the idempotence 2824 * of the IPv4 multicast part of the BSD Sockets API must be preserved; 2825 * the effects of these options must be treated as separate and distinct. 2826 * 2827 * SMPng: XXX: Unlocked read of inp_socket believed OK. 2828 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING 2829 * is refactored to no longer use vifs. 2830 */ 2831 int 2832 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2833 { 2834 struct ip_moptions *imo; 2835 int error; 2836 2837 error = 0; 2838 2839 /* 2840 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 2841 * or is a divert socket, reject it. 2842 */ 2843 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 2844 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2845 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) 2846 return (EOPNOTSUPP); 2847 2848 switch (sopt->sopt_name) { 2849 case IP_MULTICAST_VIF: { 2850 int vifi; 2851 /* 2852 * Select a multicast VIF for transmission. 2853 * Only useful if multicast forwarding is active. 2854 */ 2855 if (legal_vif_num == NULL) { 2856 error = EOPNOTSUPP; 2857 break; 2858 } 2859 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int)); 2860 if (error) 2861 break; 2862 if (!legal_vif_num(vifi) && (vifi != -1)) { 2863 error = EINVAL; 2864 break; 2865 } 2866 imo = inp_findmoptions(inp); 2867 imo->imo_multicast_vif = vifi; 2868 INP_WUNLOCK(inp); 2869 break; 2870 } 2871 2872 case IP_MULTICAST_IF: 2873 error = inp_set_multicast_if(inp, sopt); 2874 break; 2875 2876 case IP_MULTICAST_TTL: { 2877 u_char ttl; 2878 2879 /* 2880 * Set the IP time-to-live for outgoing multicast packets. 2881 * The original multicast API required a char argument, 2882 * which is inconsistent with the rest of the socket API. 2883 * We allow either a char or an int. 2884 */ 2885 if (sopt->sopt_valsize == sizeof(u_char)) { 2886 error = sooptcopyin(sopt, &ttl, sizeof(u_char), 2887 sizeof(u_char)); 2888 if (error) 2889 break; 2890 } else { 2891 u_int ittl; 2892 2893 error = sooptcopyin(sopt, &ittl, sizeof(u_int), 2894 sizeof(u_int)); 2895 if (error) 2896 break; 2897 if (ittl > 255) { 2898 error = EINVAL; 2899 break; 2900 } 2901 ttl = (u_char)ittl; 2902 } 2903 imo = inp_findmoptions(inp); 2904 imo->imo_multicast_ttl = ttl; 2905 INP_WUNLOCK(inp); 2906 break; 2907 } 2908 2909 case IP_MULTICAST_LOOP: { 2910 u_char loop; 2911 2912 /* 2913 * Set the loopback flag for outgoing multicast packets. 2914 * Must be zero or one. The original multicast API required a 2915 * char argument, which is inconsistent with the rest 2916 * of the socket API. We allow either a char or an int. 2917 */ 2918 if (sopt->sopt_valsize == sizeof(u_char)) { 2919 error = sooptcopyin(sopt, &loop, sizeof(u_char), 2920 sizeof(u_char)); 2921 if (error) 2922 break; 2923 } else { 2924 u_int iloop; 2925 2926 error = sooptcopyin(sopt, &iloop, sizeof(u_int), 2927 sizeof(u_int)); 2928 if (error) 2929 break; 2930 loop = (u_char)iloop; 2931 } 2932 imo = inp_findmoptions(inp); 2933 imo->imo_multicast_loop = !!loop; 2934 INP_WUNLOCK(inp); 2935 break; 2936 } 2937 2938 case IP_ADD_MEMBERSHIP: 2939 case IP_ADD_SOURCE_MEMBERSHIP: 2940 case MCAST_JOIN_GROUP: 2941 case MCAST_JOIN_SOURCE_GROUP: 2942 error = inp_join_group(inp, sopt); 2943 break; 2944 2945 case IP_DROP_MEMBERSHIP: 2946 case IP_DROP_SOURCE_MEMBERSHIP: 2947 case MCAST_LEAVE_GROUP: 2948 case MCAST_LEAVE_SOURCE_GROUP: 2949 error = inp_leave_group(inp, sopt); 2950 break; 2951 2952 case IP_BLOCK_SOURCE: 2953 case IP_UNBLOCK_SOURCE: 2954 case MCAST_BLOCK_SOURCE: 2955 case MCAST_UNBLOCK_SOURCE: 2956 error = inp_block_unblock_source(inp, sopt); 2957 break; 2958 2959 case IP_MSFILTER: 2960 error = inp_set_source_filters(inp, sopt); 2961 break; 2962 2963 default: 2964 error = EOPNOTSUPP; 2965 break; 2966 } 2967 2968 INP_UNLOCK_ASSERT(inp); 2969 2970 return (error); 2971 } 2972 2973 /* 2974 * Expose IGMP's multicast filter mode and source list(s) to userland, 2975 * keyed by (ifindex, group). 2976 * The filter mode is written out as a uint32_t, followed by 2977 * 0..n of struct in_addr. 2978 * For use by ifmcstat(8). 2979 * SMPng: NOTE: unlocked read of ifindex space. 2980 */ 2981 static int 2982 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS) 2983 { 2984 struct in_addr src, group; 2985 struct epoch_tracker et; 2986 struct ifnet *ifp; 2987 struct ifmultiaddr *ifma; 2988 struct in_multi *inm; 2989 struct ip_msource *ims; 2990 int *name; 2991 int retval; 2992 u_int namelen; 2993 uint32_t fmode, ifindex; 2994 2995 name = (int *)arg1; 2996 namelen = arg2; 2997 2998 if (req->newptr != NULL) 2999 return (EPERM); 3000 3001 if (namelen != 2) 3002 return (EINVAL); 3003 3004 ifindex = name[0]; 3005 if (ifindex <= 0 || ifindex > V_if_index) { 3006 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range", 3007 __func__, ifindex); 3008 return (ENOENT); 3009 } 3010 3011 group.s_addr = name[1]; 3012 if (!IN_MULTICAST(ntohl(group.s_addr))) { 3013 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast", 3014 __func__, ntohl(group.s_addr)); 3015 return (EINVAL); 3016 } 3017 3018 ifp = ifnet_byindex(ifindex); 3019 if (ifp == NULL) { 3020 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u", 3021 __func__, ifindex); 3022 return (ENOENT); 3023 } 3024 3025 retval = sysctl_wire_old_buffer(req, 3026 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr))); 3027 if (retval) 3028 return (retval); 3029 3030 IN_MULTI_LIST_LOCK(); 3031 3032 NET_EPOCH_ENTER(et); 3033 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3034 if (ifma->ifma_addr->sa_family != AF_INET || 3035 ifma->ifma_protospec == NULL) 3036 continue; 3037 inm = (struct in_multi *)ifma->ifma_protospec; 3038 if (!in_hosteq(inm->inm_addr, group)) 3039 continue; 3040 fmode = inm->inm_st[1].iss_fmode; 3041 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 3042 if (retval != 0) 3043 break; 3044 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 3045 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 3046 ims->ims_haddr); 3047 /* 3048 * Only copy-out sources which are in-mode. 3049 */ 3050 if (fmode != ims_get_mode(inm, ims, 1)) { 3051 CTR1(KTR_IGMPV3, "%s: skip non-in-mode", 3052 __func__); 3053 continue; 3054 } 3055 src.s_addr = htonl(ims->ims_haddr); 3056 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr)); 3057 if (retval != 0) 3058 break; 3059 } 3060 } 3061 NET_EPOCH_EXIT(et); 3062 3063 IN_MULTI_LIST_UNLOCK(); 3064 3065 return (retval); 3066 } 3067 3068 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3) 3069 3070 static const char *inm_modestrs[] = { "un", "in", "ex" }; 3071 3072 static const char * 3073 inm_mode_str(const int mode) 3074 { 3075 3076 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 3077 return (inm_modestrs[mode]); 3078 return ("??"); 3079 } 3080 3081 static const char *inm_statestrs[] = { 3082 "not-member", 3083 "silent", 3084 "idle", 3085 "lazy", 3086 "sleeping", 3087 "awakening", 3088 "query-pending", 3089 "sg-query-pending", 3090 "leaving" 3091 }; 3092 3093 static const char * 3094 inm_state_str(const int state) 3095 { 3096 3097 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER) 3098 return (inm_statestrs[state]); 3099 return ("??"); 3100 } 3101 3102 /* 3103 * Dump an in_multi structure to the console. 3104 */ 3105 void 3106 inm_print(const struct in_multi *inm) 3107 { 3108 int t; 3109 char addrbuf[INET_ADDRSTRLEN]; 3110 3111 if ((ktr_mask & KTR_IGMPV3) == 0) 3112 return; 3113 3114 printf("%s: --- begin inm %p ---\n", __func__, inm); 3115 printf("addr %s ifp %p(%s) ifma %p\n", 3116 inet_ntoa_r(inm->inm_addr, addrbuf), 3117 inm->inm_ifp, 3118 inm->inm_ifp->if_xname, 3119 inm->inm_ifma); 3120 printf("timer %u state %s refcount %u scq.len %u\n", 3121 inm->inm_timer, 3122 inm_state_str(inm->inm_state), 3123 inm->inm_refcount, 3124 inm->inm_scq.mq_len); 3125 printf("igi %p nsrc %lu sctimer %u scrv %u\n", 3126 inm->inm_igi, 3127 inm->inm_nsrc, 3128 inm->inm_sctimer, 3129 inm->inm_scrv); 3130 for (t = 0; t < 2; t++) { 3131 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 3132 inm_mode_str(inm->inm_st[t].iss_fmode), 3133 inm->inm_st[t].iss_asm, 3134 inm->inm_st[t].iss_ex, 3135 inm->inm_st[t].iss_in, 3136 inm->inm_st[t].iss_rec); 3137 } 3138 printf("%s: --- end inm %p ---\n", __func__, inm); 3139 } 3140 3141 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */ 3142 3143 void 3144 inm_print(const struct in_multi *inm) 3145 { 3146 3147 } 3148 3149 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */ 3150 3151 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp); 3152