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 * 98 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK, 99 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 100 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however 101 * it can be taken by code in net/if.c also. 102 * - ip_moptions and in_mfilter are covered by the INP_WLOCK. 103 * 104 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly 105 * any need for in_multi itself to be virtualized -- it is bound to an ifp 106 * anyway no matter what happens. 107 */ 108 struct mtx in_multi_list_mtx; 109 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF); 110 111 struct mtx in_multi_free_mtx; 112 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF); 113 114 struct sx in_multi_sx; 115 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx"); 116 117 int ifma_restart; 118 119 /* 120 * Functions with non-static linkage defined in this file should be 121 * declared in in_var.h: 122 * imo_multi_filter() 123 * in_addmulti() 124 * in_delmulti() 125 * in_joingroup() 126 * in_joingroup_locked() 127 * in_leavegroup() 128 * in_leavegroup_locked() 129 * and ip_var.h: 130 * inp_freemoptions() 131 * inp_getmoptions() 132 * inp_setmoptions() 133 * 134 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti() 135 * and in_delmulti(). 136 */ 137 static void imf_commit(struct in_mfilter *); 138 static int imf_get_source(struct in_mfilter *imf, 139 const struct sockaddr_in *psin, 140 struct in_msource **); 141 static struct in_msource * 142 imf_graft(struct in_mfilter *, const uint8_t, 143 const struct sockaddr_in *); 144 static void imf_leave(struct in_mfilter *); 145 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *); 146 static void imf_purge(struct in_mfilter *); 147 static void imf_rollback(struct in_mfilter *); 148 static void imf_reap(struct in_mfilter *); 149 static struct in_mfilter * 150 imo_match_group(const struct ip_moptions *, 151 const struct ifnet *, const struct sockaddr *); 152 static struct in_msource * 153 imo_match_source(struct in_mfilter *, const struct sockaddr *); 154 static void ims_merge(struct ip_msource *ims, 155 const struct in_msource *lims, const int rollback); 156 static int in_getmulti(struct ifnet *, const struct in_addr *, 157 struct in_multi **); 158 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr, 159 const int noalloc, struct ip_msource **pims); 160 #ifdef KTR 161 static int inm_is_ifp_detached(const struct in_multi *); 162 #endif 163 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *); 164 static void inm_purge(struct in_multi *); 165 static void inm_reap(struct in_multi *); 166 static void inm_release(struct in_multi *); 167 static struct ip_moptions * 168 inp_findmoptions(struct inpcb *); 169 static int inp_get_source_filters(struct inpcb *, struct sockopt *); 170 static int inp_join_group(struct inpcb *, struct sockopt *); 171 static int inp_leave_group(struct inpcb *, struct sockopt *); 172 static struct ifnet * 173 inp_lookup_mcast_ifp(const struct inpcb *, 174 const struct sockaddr_in *, const struct in_addr); 175 static int inp_block_unblock_source(struct inpcb *, struct sockopt *); 176 static int inp_set_multicast_if(struct inpcb *, struct sockopt *); 177 static int inp_set_source_filters(struct inpcb *, struct sockopt *); 178 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS); 179 180 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, 181 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 182 "IPv4 multicast"); 183 184 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER; 185 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc, 186 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0, 187 "Max source filters per group"); 188 189 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER; 190 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc, 191 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0, 192 "Max source filters per socket"); 193 194 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP; 195 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN, 196 &in_mcast_loop, 0, "Loopback multicast datagrams by default"); 197 198 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters, 199 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters, 200 "Per-interface stack-wide source filters"); 201 202 #ifdef KTR 203 /* 204 * Inline function which wraps assertions for a valid ifp. 205 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 206 * is detached. 207 */ 208 static int __inline 209 inm_is_ifp_detached(const struct in_multi *inm) 210 { 211 struct ifnet *ifp; 212 213 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 214 ifp = inm->inm_ifma->ifma_ifp; 215 if (ifp != NULL) { 216 /* 217 * Sanity check that netinet's notion of ifp is the 218 * same as net's. 219 */ 220 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 221 } 222 223 return (ifp == NULL); 224 } 225 #endif 226 227 static struct grouptask free_gtask; 228 static struct in_multi_head inm_free_list; 229 static void inm_release_task(void *arg __unused); 230 static void inm_init(void) 231 { 232 SLIST_INIT(&inm_free_list); 233 taskqgroup_config_gtask_init(NULL, &free_gtask, inm_release_task, "inm release task"); 234 } 235 236 #ifdef EARLY_AP_STARTUP 237 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST, 238 inm_init, NULL); 239 #else 240 SYSINIT(inm_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_FIRST, 241 inm_init, NULL); 242 #endif 243 244 245 void 246 inm_release_list_deferred(struct in_multi_head *inmh) 247 { 248 249 if (SLIST_EMPTY(inmh)) 250 return; 251 mtx_lock(&in_multi_free_mtx); 252 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele); 253 mtx_unlock(&in_multi_free_mtx); 254 GROUPTASK_ENQUEUE(&free_gtask); 255 } 256 257 void 258 inm_disconnect(struct in_multi *inm) 259 { 260 struct ifnet *ifp; 261 struct ifmultiaddr *ifma, *ll_ifma; 262 263 ifp = inm->inm_ifp; 264 IF_ADDR_WLOCK_ASSERT(ifp); 265 ifma = inm->inm_ifma; 266 267 if_ref(ifp); 268 if (ifma->ifma_flags & IFMA_F_ENQUEUED) { 269 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 270 ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 271 } 272 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname); 273 if ((ll_ifma = ifma->ifma_llifma) != NULL) { 274 MPASS(ifma != ll_ifma); 275 ifma->ifma_llifma = NULL; 276 MPASS(ll_ifma->ifma_llifma == NULL); 277 MPASS(ll_ifma->ifma_ifp == ifp); 278 if (--ll_ifma->ifma_refcount == 0) { 279 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) { 280 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link); 281 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED; 282 } 283 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname); 284 if_freemulti(ll_ifma); 285 ifma_restart = true; 286 } 287 } 288 } 289 290 void 291 inm_release_deferred(struct in_multi *inm) 292 { 293 struct in_multi_head tmp; 294 295 IN_MULTI_LIST_LOCK_ASSERT(); 296 MPASS(inm->inm_refcount > 0); 297 if (--inm->inm_refcount == 0) { 298 SLIST_INIT(&tmp); 299 inm_disconnect(inm); 300 inm->inm_ifma->ifma_protospec = NULL; 301 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele); 302 inm_release_list_deferred(&tmp); 303 } 304 } 305 306 static void 307 inm_release_task(void *arg __unused) 308 { 309 struct in_multi_head inm_free_tmp; 310 struct in_multi *inm, *tinm; 311 312 SLIST_INIT(&inm_free_tmp); 313 mtx_lock(&in_multi_free_mtx); 314 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele); 315 mtx_unlock(&in_multi_free_mtx); 316 IN_MULTI_LOCK(); 317 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) { 318 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele); 319 MPASS(inm); 320 inm_release(inm); 321 } 322 IN_MULTI_UNLOCK(); 323 } 324 325 /* 326 * Initialize an in_mfilter structure to a known state at t0, t1 327 * with an empty source filter list. 328 */ 329 static __inline void 330 imf_init(struct in_mfilter *imf, const int st0, const int st1) 331 { 332 memset(imf, 0, sizeof(struct in_mfilter)); 333 RB_INIT(&imf->imf_sources); 334 imf->imf_st[0] = st0; 335 imf->imf_st[1] = st1; 336 } 337 338 struct in_mfilter * 339 ip_mfilter_alloc(const int mflags, const int st0, const int st1) 340 { 341 struct in_mfilter *imf; 342 343 imf = malloc(sizeof(*imf), M_INMFILTER, mflags); 344 if (imf != NULL) 345 imf_init(imf, st0, st1); 346 347 return (imf); 348 } 349 350 void 351 ip_mfilter_free(struct in_mfilter *imf) 352 { 353 354 imf_purge(imf); 355 free(imf, M_INMFILTER); 356 } 357 358 /* 359 * Function for looking up an in_multi record for an IPv4 multicast address 360 * on a given interface. ifp must be valid. If no record found, return NULL. 361 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held. 362 */ 363 struct in_multi * 364 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina) 365 { 366 struct ifmultiaddr *ifma; 367 struct in_multi *inm; 368 369 IN_MULTI_LIST_LOCK_ASSERT(); 370 IF_ADDR_LOCK_ASSERT(ifp); 371 372 inm = NULL; 373 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) { 374 if (ifma->ifma_addr->sa_family != AF_INET || 375 ifma->ifma_protospec == NULL) 376 continue; 377 inm = (struct in_multi *)ifma->ifma_protospec; 378 if (inm->inm_addr.s_addr == ina.s_addr) 379 break; 380 inm = NULL; 381 } 382 return (inm); 383 } 384 385 /* 386 * Wrapper for inm_lookup_locked(). 387 * The IF_ADDR_LOCK will be taken on ifp and released on return. 388 */ 389 struct in_multi * 390 inm_lookup(struct ifnet *ifp, const struct in_addr ina) 391 { 392 struct epoch_tracker et; 393 struct in_multi *inm; 394 395 IN_MULTI_LIST_LOCK_ASSERT(); 396 NET_EPOCH_ENTER(et); 397 398 inm = inm_lookup_locked(ifp, ina); 399 NET_EPOCH_EXIT(et); 400 401 return (inm); 402 } 403 404 /* 405 * Find an IPv4 multicast group entry for this ip_moptions instance 406 * which matches the specified group, and optionally an interface. 407 * Return its index into the array, or -1 if not found. 408 */ 409 static struct in_mfilter * 410 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp, 411 const struct sockaddr *group) 412 { 413 const struct sockaddr_in *gsin; 414 struct in_mfilter *imf; 415 struct in_multi *inm; 416 417 gsin = (const struct sockaddr_in *)group; 418 419 IP_MFILTER_FOREACH(imf, &imo->imo_head) { 420 inm = imf->imf_inm; 421 if (inm == NULL) 422 continue; 423 if ((ifp == NULL || (inm->inm_ifp == ifp)) && 424 in_hosteq(inm->inm_addr, gsin->sin_addr)) { 425 break; 426 } 427 } 428 return (imf); 429 } 430 431 /* 432 * Find an IPv4 multicast source entry for this imo which matches 433 * the given group index for this socket, and source address. 434 * 435 * NOTE: This does not check if the entry is in-mode, merely if 436 * it exists, which may not be the desired behaviour. 437 */ 438 static struct in_msource * 439 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src) 440 { 441 struct ip_msource find; 442 struct ip_msource *ims; 443 const sockunion_t *psa; 444 445 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__)); 446 447 /* Source trees are keyed in host byte order. */ 448 psa = (const sockunion_t *)src; 449 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr); 450 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 451 452 return ((struct in_msource *)ims); 453 } 454 455 /* 456 * Perform filtering for multicast datagrams on a socket by group and source. 457 * 458 * Returns 0 if a datagram should be allowed through, or various error codes 459 * if the socket was not a member of the group, or the source was muted, etc. 460 */ 461 int 462 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp, 463 const struct sockaddr *group, const struct sockaddr *src) 464 { 465 struct in_mfilter *imf; 466 struct in_msource *ims; 467 int mode; 468 469 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 470 471 imf = imo_match_group(imo, ifp, group); 472 if (imf == NULL) 473 return (MCAST_NOTGMEMBER); 474 475 /* 476 * Check if the source was included in an (S,G) join. 477 * Allow reception on exclusive memberships by default, 478 * reject reception on inclusive memberships by default. 479 * Exclude source only if an in-mode exclude filter exists. 480 * Include source only if an in-mode include filter exists. 481 * NOTE: We are comparing group state here at IGMP t1 (now) 482 * with socket-layer t0 (since last downcall). 483 */ 484 mode = imf->imf_st[1]; 485 ims = imo_match_source(imf, src); 486 487 if ((ims == NULL && mode == MCAST_INCLUDE) || 488 (ims != NULL && ims->imsl_st[0] != mode)) 489 return (MCAST_NOTSMEMBER); 490 491 return (MCAST_PASS); 492 } 493 494 /* 495 * Find and return a reference to an in_multi record for (ifp, group), 496 * and bump its reference count. 497 * If one does not exist, try to allocate it, and update link-layer multicast 498 * filters on ifp to listen for group. 499 * Assumes the IN_MULTI lock is held across the call. 500 * Return 0 if successful, otherwise return an appropriate error code. 501 */ 502 static int 503 in_getmulti(struct ifnet *ifp, const struct in_addr *group, 504 struct in_multi **pinm) 505 { 506 struct sockaddr_in gsin; 507 struct ifmultiaddr *ifma; 508 struct in_ifinfo *ii; 509 struct in_multi *inm; 510 int error; 511 512 IN_MULTI_LOCK_ASSERT(); 513 514 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET]; 515 IN_MULTI_LIST_LOCK(); 516 inm = inm_lookup(ifp, *group); 517 if (inm != NULL) { 518 /* 519 * If we already joined this group, just bump the 520 * refcount and return it. 521 */ 522 KASSERT(inm->inm_refcount >= 1, 523 ("%s: bad refcount %d", __func__, inm->inm_refcount)); 524 inm_acquire_locked(inm); 525 *pinm = inm; 526 } 527 IN_MULTI_LIST_UNLOCK(); 528 if (inm != NULL) 529 return (0); 530 531 memset(&gsin, 0, sizeof(gsin)); 532 gsin.sin_family = AF_INET; 533 gsin.sin_len = sizeof(struct sockaddr_in); 534 gsin.sin_addr = *group; 535 536 /* 537 * Check if a link-layer group is already associated 538 * with this network-layer group on the given ifnet. 539 */ 540 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma); 541 if (error != 0) 542 return (error); 543 544 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */ 545 IN_MULTI_LIST_LOCK(); 546 IF_ADDR_WLOCK(ifp); 547 548 /* 549 * If something other than netinet is occupying the link-layer 550 * group, print a meaningful error message and back out of 551 * the allocation. 552 * Otherwise, bump the refcount on the existing network-layer 553 * group association and return it. 554 */ 555 if (ifma->ifma_protospec != NULL) { 556 inm = (struct in_multi *)ifma->ifma_protospec; 557 #ifdef INVARIANTS 558 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 559 __func__)); 560 KASSERT(ifma->ifma_addr->sa_family == AF_INET, 561 ("%s: ifma not AF_INET", __func__)); 562 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 563 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp || 564 !in_hosteq(inm->inm_addr, *group)) { 565 char addrbuf[INET_ADDRSTRLEN]; 566 567 panic("%s: ifma %p is inconsistent with %p (%s)", 568 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf)); 569 } 570 #endif 571 inm_acquire_locked(inm); 572 *pinm = inm; 573 goto out_locked; 574 } 575 576 IF_ADDR_WLOCK_ASSERT(ifp); 577 578 /* 579 * A new in_multi record is needed; allocate and initialize it. 580 * We DO NOT perform an IGMP join as the in_ layer may need to 581 * push an initial source list down to IGMP to support SSM. 582 * 583 * The initial source filter state is INCLUDE, {} as per the RFC. 584 */ 585 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO); 586 if (inm == NULL) { 587 IF_ADDR_WUNLOCK(ifp); 588 IN_MULTI_LIST_UNLOCK(); 589 if_delmulti_ifma(ifma); 590 return (ENOMEM); 591 } 592 inm->inm_addr = *group; 593 inm->inm_ifp = ifp; 594 inm->inm_igi = ii->ii_igmp; 595 inm->inm_ifma = ifma; 596 inm->inm_refcount = 1; 597 inm->inm_state = IGMP_NOT_MEMBER; 598 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES); 599 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED; 600 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 601 RB_INIT(&inm->inm_srcs); 602 603 ifma->ifma_protospec = inm; 604 605 *pinm = inm; 606 out_locked: 607 IF_ADDR_WUNLOCK(ifp); 608 IN_MULTI_LIST_UNLOCK(); 609 return (0); 610 } 611 612 /* 613 * Drop a reference to an in_multi record. 614 * 615 * If the refcount drops to 0, free the in_multi record and 616 * delete the underlying link-layer membership. 617 */ 618 static void 619 inm_release(struct in_multi *inm) 620 { 621 struct ifmultiaddr *ifma; 622 struct ifnet *ifp; 623 624 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount); 625 MPASS(inm->inm_refcount == 0); 626 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm); 627 628 ifma = inm->inm_ifma; 629 ifp = inm->inm_ifp; 630 631 /* XXX this access is not covered by IF_ADDR_LOCK */ 632 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma); 633 if (ifp != NULL) { 634 CURVNET_SET(ifp->if_vnet); 635 inm_purge(inm); 636 free(inm, M_IPMADDR); 637 if_delmulti_ifma_flags(ifma, 1); 638 CURVNET_RESTORE(); 639 if_rele(ifp); 640 } else { 641 inm_purge(inm); 642 free(inm, M_IPMADDR); 643 if_delmulti_ifma_flags(ifma, 1); 644 } 645 } 646 647 /* 648 * Clear recorded source entries for a group. 649 * Used by the IGMP code. Caller must hold the IN_MULTI lock. 650 * FIXME: Should reap. 651 */ 652 void 653 inm_clear_recorded(struct in_multi *inm) 654 { 655 struct ip_msource *ims; 656 657 IN_MULTI_LIST_LOCK_ASSERT(); 658 659 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 660 if (ims->ims_stp) { 661 ims->ims_stp = 0; 662 --inm->inm_st[1].iss_rec; 663 } 664 } 665 KASSERT(inm->inm_st[1].iss_rec == 0, 666 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec)); 667 } 668 669 /* 670 * Record a source as pending for a Source-Group IGMPv3 query. 671 * This lives here as it modifies the shared tree. 672 * 673 * inm is the group descriptor. 674 * naddr is the address of the source to record in network-byte order. 675 * 676 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will 677 * lazy-allocate a source node in response to an SG query. 678 * Otherwise, no allocation is performed. This saves some memory 679 * with the trade-off that the source will not be reported to the 680 * router if joined in the window between the query response and 681 * the group actually being joined on the local host. 682 * 683 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed. 684 * This turns off the allocation of a recorded source entry if 685 * the group has not been joined. 686 * 687 * Return 0 if the source didn't exist or was already marked as recorded. 688 * Return 1 if the source was marked as recorded by this function. 689 * Return <0 if any error occurred (negated errno code). 690 */ 691 int 692 inm_record_source(struct in_multi *inm, const in_addr_t naddr) 693 { 694 struct ip_msource find; 695 struct ip_msource *ims, *nims; 696 697 IN_MULTI_LIST_LOCK_ASSERT(); 698 699 find.ims_haddr = ntohl(naddr); 700 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 701 if (ims && ims->ims_stp) 702 return (0); 703 if (ims == NULL) { 704 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 705 return (-ENOSPC); 706 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 707 M_NOWAIT | M_ZERO); 708 if (nims == NULL) 709 return (-ENOMEM); 710 nims->ims_haddr = find.ims_haddr; 711 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 712 ++inm->inm_nsrc; 713 ims = nims; 714 } 715 716 /* 717 * Mark the source as recorded and update the recorded 718 * source count. 719 */ 720 ++ims->ims_stp; 721 ++inm->inm_st[1].iss_rec; 722 723 return (1); 724 } 725 726 /* 727 * Return a pointer to an in_msource owned by an in_mfilter, 728 * given its source address. 729 * Lazy-allocate if needed. If this is a new entry its filter state is 730 * undefined at t0. 731 * 732 * imf is the filter set being modified. 733 * haddr is the source address in *host* byte-order. 734 * 735 * SMPng: May be called with locks held; malloc must not block. 736 */ 737 static int 738 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin, 739 struct in_msource **plims) 740 { 741 struct ip_msource find; 742 struct ip_msource *ims, *nims; 743 struct in_msource *lims; 744 int error; 745 746 error = 0; 747 ims = NULL; 748 lims = NULL; 749 750 /* key is host byte order */ 751 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 752 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 753 lims = (struct in_msource *)ims; 754 if (lims == NULL) { 755 if (imf->imf_nsrc == in_mcast_maxsocksrc) 756 return (ENOSPC); 757 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 758 M_NOWAIT | M_ZERO); 759 if (nims == NULL) 760 return (ENOMEM); 761 lims = (struct in_msource *)nims; 762 lims->ims_haddr = find.ims_haddr; 763 lims->imsl_st[0] = MCAST_UNDEFINED; 764 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 765 ++imf->imf_nsrc; 766 } 767 768 *plims = lims; 769 770 return (error); 771 } 772 773 /* 774 * Graft a source entry into an existing socket-layer filter set, 775 * maintaining any required invariants and checking allocations. 776 * 777 * The source is marked as being in the new filter mode at t1. 778 * 779 * Return the pointer to the new node, otherwise return NULL. 780 */ 781 static struct in_msource * 782 imf_graft(struct in_mfilter *imf, const uint8_t st1, 783 const struct sockaddr_in *psin) 784 { 785 struct ip_msource *nims; 786 struct in_msource *lims; 787 788 nims = malloc(sizeof(struct in_msource), M_INMFILTER, 789 M_NOWAIT | M_ZERO); 790 if (nims == NULL) 791 return (NULL); 792 lims = (struct in_msource *)nims; 793 lims->ims_haddr = ntohl(psin->sin_addr.s_addr); 794 lims->imsl_st[0] = MCAST_UNDEFINED; 795 lims->imsl_st[1] = st1; 796 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 797 ++imf->imf_nsrc; 798 799 return (lims); 800 } 801 802 /* 803 * Prune a source entry from an existing socket-layer filter set, 804 * maintaining any required invariants and checking allocations. 805 * 806 * The source is marked as being left at t1, it is not freed. 807 * 808 * Return 0 if no error occurred, otherwise return an errno value. 809 */ 810 static int 811 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin) 812 { 813 struct ip_msource find; 814 struct ip_msource *ims; 815 struct in_msource *lims; 816 817 /* key is host byte order */ 818 find.ims_haddr = ntohl(psin->sin_addr.s_addr); 819 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 820 if (ims == NULL) 821 return (ENOENT); 822 lims = (struct in_msource *)ims; 823 lims->imsl_st[1] = MCAST_UNDEFINED; 824 return (0); 825 } 826 827 /* 828 * Revert socket-layer filter set deltas at t1 to t0 state. 829 */ 830 static void 831 imf_rollback(struct in_mfilter *imf) 832 { 833 struct ip_msource *ims, *tims; 834 struct in_msource *lims; 835 836 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 837 lims = (struct in_msource *)ims; 838 if (lims->imsl_st[0] == lims->imsl_st[1]) { 839 /* no change at t1 */ 840 continue; 841 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) { 842 /* revert change to existing source at t1 */ 843 lims->imsl_st[1] = lims->imsl_st[0]; 844 } else { 845 /* revert source added t1 */ 846 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 847 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 848 free(ims, M_INMFILTER); 849 imf->imf_nsrc--; 850 } 851 } 852 imf->imf_st[1] = imf->imf_st[0]; 853 } 854 855 /* 856 * Mark socket-layer filter set as INCLUDE {} at t1. 857 */ 858 static void 859 imf_leave(struct in_mfilter *imf) 860 { 861 struct ip_msource *ims; 862 struct in_msource *lims; 863 864 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 865 lims = (struct in_msource *)ims; 866 lims->imsl_st[1] = MCAST_UNDEFINED; 867 } 868 imf->imf_st[1] = MCAST_INCLUDE; 869 } 870 871 /* 872 * Mark socket-layer filter set deltas as committed. 873 */ 874 static void 875 imf_commit(struct in_mfilter *imf) 876 { 877 struct ip_msource *ims; 878 struct in_msource *lims; 879 880 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 881 lims = (struct in_msource *)ims; 882 lims->imsl_st[0] = lims->imsl_st[1]; 883 } 884 imf->imf_st[0] = imf->imf_st[1]; 885 } 886 887 /* 888 * Reap unreferenced sources from socket-layer filter set. 889 */ 890 static void 891 imf_reap(struct in_mfilter *imf) 892 { 893 struct ip_msource *ims, *tims; 894 struct in_msource *lims; 895 896 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 897 lims = (struct in_msource *)ims; 898 if ((lims->imsl_st[0] == MCAST_UNDEFINED) && 899 (lims->imsl_st[1] == MCAST_UNDEFINED)) { 900 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims); 901 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 902 free(ims, M_INMFILTER); 903 imf->imf_nsrc--; 904 } 905 } 906 } 907 908 /* 909 * Purge socket-layer filter set. 910 */ 911 static void 912 imf_purge(struct in_mfilter *imf) 913 { 914 struct ip_msource *ims, *tims; 915 916 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 917 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 918 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 919 free(ims, M_INMFILTER); 920 imf->imf_nsrc--; 921 } 922 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED; 923 KASSERT(RB_EMPTY(&imf->imf_sources), 924 ("%s: imf_sources not empty", __func__)); 925 } 926 927 /* 928 * Look up a source filter entry for a multicast group. 929 * 930 * inm is the group descriptor to work with. 931 * haddr is the host-byte-order IPv4 address to look up. 932 * noalloc may be non-zero to suppress allocation of sources. 933 * *pims will be set to the address of the retrieved or allocated source. 934 * 935 * SMPng: NOTE: may be called with locks held. 936 * Return 0 if successful, otherwise return a non-zero error code. 937 */ 938 static int 939 inm_get_source(struct in_multi *inm, const in_addr_t haddr, 940 const int noalloc, struct ip_msource **pims) 941 { 942 struct ip_msource find; 943 struct ip_msource *ims, *nims; 944 945 find.ims_haddr = haddr; 946 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 947 if (ims == NULL && !noalloc) { 948 if (inm->inm_nsrc == in_mcast_maxgrpsrc) 949 return (ENOSPC); 950 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 951 M_NOWAIT | M_ZERO); 952 if (nims == NULL) 953 return (ENOMEM); 954 nims->ims_haddr = haddr; 955 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 956 ++inm->inm_nsrc; 957 ims = nims; 958 #ifdef KTR 959 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__, 960 haddr, ims); 961 #endif 962 } 963 964 *pims = ims; 965 return (0); 966 } 967 968 /* 969 * Merge socket-layer source into IGMP-layer source. 970 * If rollback is non-zero, perform the inverse of the merge. 971 */ 972 static void 973 ims_merge(struct ip_msource *ims, const struct in_msource *lims, 974 const int rollback) 975 { 976 int n = rollback ? -1 : 1; 977 978 if (lims->imsl_st[0] == MCAST_EXCLUDE) { 979 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x", 980 __func__, n, ims->ims_haddr); 981 ims->ims_st[1].ex -= n; 982 } else if (lims->imsl_st[0] == MCAST_INCLUDE) { 983 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x", 984 __func__, n, ims->ims_haddr); 985 ims->ims_st[1].in -= n; 986 } 987 988 if (lims->imsl_st[1] == MCAST_EXCLUDE) { 989 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x", 990 __func__, n, ims->ims_haddr); 991 ims->ims_st[1].ex += n; 992 } else if (lims->imsl_st[1] == MCAST_INCLUDE) { 993 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x", 994 __func__, n, ims->ims_haddr); 995 ims->ims_st[1].in += n; 996 } 997 } 998 999 /* 1000 * Atomically update the global in_multi state, when a membership's 1001 * filter list is being updated in any way. 1002 * 1003 * imf is the per-inpcb-membership group filter pointer. 1004 * A fake imf may be passed for in-kernel consumers. 1005 * 1006 * XXX This is a candidate for a set-symmetric-difference style loop 1007 * which would eliminate the repeated lookup from root of ims nodes, 1008 * as they share the same key space. 1009 * 1010 * If any error occurred this function will back out of refcounts 1011 * and return a non-zero value. 1012 */ 1013 static int 1014 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1015 { 1016 struct ip_msource *ims, *nims; 1017 struct in_msource *lims; 1018 int schanged, error; 1019 int nsrc0, nsrc1; 1020 1021 schanged = 0; 1022 error = 0; 1023 nsrc1 = nsrc0 = 0; 1024 IN_MULTI_LIST_LOCK_ASSERT(); 1025 1026 /* 1027 * Update the source filters first, as this may fail. 1028 * Maintain count of in-mode filters at t0, t1. These are 1029 * used to work out if we transition into ASM mode or not. 1030 * Maintain a count of source filters whose state was 1031 * actually modified by this operation. 1032 */ 1033 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1034 lims = (struct in_msource *)ims; 1035 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++; 1036 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++; 1037 if (lims->imsl_st[0] == lims->imsl_st[1]) continue; 1038 error = inm_get_source(inm, lims->ims_haddr, 0, &nims); 1039 ++schanged; 1040 if (error) 1041 break; 1042 ims_merge(nims, lims, 0); 1043 } 1044 if (error) { 1045 struct ip_msource *bims; 1046 1047 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) { 1048 lims = (struct in_msource *)ims; 1049 if (lims->imsl_st[0] == lims->imsl_st[1]) 1050 continue; 1051 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims); 1052 if (bims == NULL) 1053 continue; 1054 ims_merge(bims, lims, 1); 1055 } 1056 goto out_reap; 1057 } 1058 1059 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1", 1060 __func__, nsrc0, nsrc1); 1061 1062 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 1063 if (imf->imf_st[0] == imf->imf_st[1] && 1064 imf->imf_st[1] == MCAST_INCLUDE) { 1065 if (nsrc1 == 0) { 1066 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1067 --inm->inm_st[1].iss_in; 1068 } 1069 } 1070 1071 /* Handle filter mode transition on socket. */ 1072 if (imf->imf_st[0] != imf->imf_st[1]) { 1073 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d", 1074 __func__, imf->imf_st[0], imf->imf_st[1]); 1075 1076 if (imf->imf_st[0] == MCAST_EXCLUDE) { 1077 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__); 1078 --inm->inm_st[1].iss_ex; 1079 } else if (imf->imf_st[0] == MCAST_INCLUDE) { 1080 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 1081 --inm->inm_st[1].iss_in; 1082 } 1083 1084 if (imf->imf_st[1] == MCAST_EXCLUDE) { 1085 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__); 1086 inm->inm_st[1].iss_ex++; 1087 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) { 1088 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__); 1089 inm->inm_st[1].iss_in++; 1090 } 1091 } 1092 1093 /* 1094 * Track inm filter state in terms of listener counts. 1095 * If there are any exclusive listeners, stack-wide 1096 * membership is exclusive. 1097 * Otherwise, if only inclusive listeners, stack-wide is inclusive. 1098 * If no listeners remain, state is undefined at t1, 1099 * and the IGMP lifecycle for this group should finish. 1100 */ 1101 if (inm->inm_st[1].iss_ex > 0) { 1102 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__); 1103 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE; 1104 } else if (inm->inm_st[1].iss_in > 0) { 1105 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__); 1106 inm->inm_st[1].iss_fmode = MCAST_INCLUDE; 1107 } else { 1108 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__); 1109 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 1110 } 1111 1112 /* Decrement ASM listener count on transition out of ASM mode. */ 1113 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1114 if ((imf->imf_st[1] != MCAST_EXCLUDE) || 1115 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) { 1116 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__); 1117 --inm->inm_st[1].iss_asm; 1118 } 1119 } 1120 1121 /* Increment ASM listener count on transition to ASM mode. */ 1122 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1123 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__); 1124 inm->inm_st[1].iss_asm++; 1125 } 1126 1127 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm); 1128 inm_print(inm); 1129 1130 out_reap: 1131 if (schanged > 0) { 1132 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__); 1133 inm_reap(inm); 1134 } 1135 return (error); 1136 } 1137 1138 /* 1139 * Mark an in_multi's filter set deltas as committed. 1140 * Called by IGMP after a state change has been enqueued. 1141 */ 1142 void 1143 inm_commit(struct in_multi *inm) 1144 { 1145 struct ip_msource *ims; 1146 1147 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm); 1148 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__); 1149 inm_print(inm); 1150 1151 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 1152 ims->ims_st[0] = ims->ims_st[1]; 1153 } 1154 inm->inm_st[0] = inm->inm_st[1]; 1155 } 1156 1157 /* 1158 * Reap unreferenced nodes from an in_multi's filter set. 1159 */ 1160 static void 1161 inm_reap(struct in_multi *inm) 1162 { 1163 struct ip_msource *ims, *tims; 1164 1165 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1166 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 || 1167 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 || 1168 ims->ims_stp != 0) 1169 continue; 1170 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1171 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1172 free(ims, M_IPMSOURCE); 1173 inm->inm_nsrc--; 1174 } 1175 } 1176 1177 /* 1178 * Purge all source nodes from an in_multi's filter set. 1179 */ 1180 static void 1181 inm_purge(struct in_multi *inm) 1182 { 1183 struct ip_msource *ims, *tims; 1184 1185 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1186 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1187 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1188 free(ims, M_IPMSOURCE); 1189 inm->inm_nsrc--; 1190 } 1191 } 1192 1193 /* 1194 * Join a multicast group; unlocked entry point. 1195 * 1196 * SMPng: XXX: in_joingroup() is called from in_control() when Giant 1197 * is not held. Fortunately, ifp is unlikely to have been detached 1198 * at this point, so we assume it's OK to recurse. 1199 */ 1200 int 1201 in_joingroup(struct ifnet *ifp, const struct in_addr *gina, 1202 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1203 { 1204 int error; 1205 1206 IN_MULTI_LOCK(); 1207 error = in_joingroup_locked(ifp, gina, imf, pinm); 1208 IN_MULTI_UNLOCK(); 1209 1210 return (error); 1211 } 1212 1213 /* 1214 * Join a multicast group; real entry point. 1215 * 1216 * Only preserves atomicity at inm level. 1217 * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1218 * 1219 * If the IGMP downcall fails, the group is not joined, and an error 1220 * code is returned. 1221 */ 1222 int 1223 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina, 1224 /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1225 { 1226 struct in_mfilter timf; 1227 struct in_multi *inm; 1228 int error; 1229 1230 IN_MULTI_LOCK_ASSERT(); 1231 IN_MULTI_LIST_UNLOCK_ASSERT(); 1232 1233 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__, 1234 ntohl(gina->s_addr), ifp, ifp->if_xname); 1235 1236 error = 0; 1237 inm = NULL; 1238 1239 /* 1240 * If no imf was specified (i.e. kernel consumer), 1241 * fake one up and assume it is an ASM join. 1242 */ 1243 if (imf == NULL) { 1244 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1245 imf = &timf; 1246 } 1247 1248 error = in_getmulti(ifp, gina, &inm); 1249 if (error) { 1250 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__); 1251 return (error); 1252 } 1253 IN_MULTI_LIST_LOCK(); 1254 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1255 error = inm_merge(inm, imf); 1256 if (error) { 1257 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1258 goto out_inm_release; 1259 } 1260 1261 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1262 error = igmp_change_state(inm); 1263 if (error) { 1264 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__); 1265 goto out_inm_release; 1266 } 1267 1268 out_inm_release: 1269 if (error) { 1270 1271 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1272 IF_ADDR_WLOCK(ifp); 1273 inm_release_deferred(inm); 1274 IF_ADDR_WUNLOCK(ifp); 1275 } else { 1276 *pinm = inm; 1277 } 1278 IN_MULTI_LIST_UNLOCK(); 1279 1280 return (error); 1281 } 1282 1283 /* 1284 * Leave a multicast group; unlocked entry point. 1285 */ 1286 int 1287 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1288 { 1289 int error; 1290 1291 IN_MULTI_LOCK(); 1292 error = in_leavegroup_locked(inm, imf); 1293 IN_MULTI_UNLOCK(); 1294 1295 return (error); 1296 } 1297 1298 /* 1299 * Leave a multicast group; real entry point. 1300 * All source filters will be expunged. 1301 * 1302 * Only preserves atomicity at inm level. 1303 * 1304 * Holding the write lock for the INP which contains imf 1305 * is highly advisable. We can't assert for it as imf does not 1306 * contain a back-pointer to the owning inp. 1307 * 1308 * Note: This is not the same as inm_release(*) as this function also 1309 * makes a state change downcall into IGMP. 1310 */ 1311 int 1312 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1313 { 1314 struct in_mfilter timf; 1315 int error; 1316 1317 IN_MULTI_LOCK_ASSERT(); 1318 IN_MULTI_LIST_UNLOCK_ASSERT(); 1319 1320 error = 0; 1321 1322 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__, 1323 inm, ntohl(inm->inm_addr.s_addr), 1324 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname), 1325 imf); 1326 1327 /* 1328 * If no imf was specified (i.e. kernel consumer), 1329 * fake one up and assume it is an ASM join. 1330 */ 1331 if (imf == NULL) { 1332 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1333 imf = &timf; 1334 } 1335 1336 /* 1337 * Begin state merge transaction at IGMP layer. 1338 * 1339 * As this particular invocation should not cause any memory 1340 * to be allocated, and there is no opportunity to roll back 1341 * the transaction, it MUST NOT fail. 1342 */ 1343 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1344 IN_MULTI_LIST_LOCK(); 1345 error = inm_merge(inm, imf); 1346 KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1347 1348 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1349 CURVNET_SET(inm->inm_ifp->if_vnet); 1350 error = igmp_change_state(inm); 1351 IF_ADDR_WLOCK(inm->inm_ifp); 1352 inm_release_deferred(inm); 1353 IF_ADDR_WUNLOCK(inm->inm_ifp); 1354 IN_MULTI_LIST_UNLOCK(); 1355 CURVNET_RESTORE(); 1356 if (error) 1357 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1358 1359 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1360 1361 return (error); 1362 } 1363 1364 /*#ifndef BURN_BRIDGES*/ 1365 /* 1366 * Join an IPv4 multicast group in (*,G) exclusive mode. 1367 * The group must be a 224.0.0.0/24 link-scope group. 1368 * This KPI is for legacy kernel consumers only. 1369 */ 1370 struct in_multi * 1371 in_addmulti(struct in_addr *ap, struct ifnet *ifp) 1372 { 1373 struct in_multi *pinm; 1374 int error; 1375 #ifdef INVARIANTS 1376 char addrbuf[INET_ADDRSTRLEN]; 1377 #endif 1378 1379 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)), 1380 ("%s: %s not in 224.0.0.0/24", __func__, 1381 inet_ntoa_r(*ap, addrbuf))); 1382 1383 error = in_joingroup(ifp, ap, NULL, &pinm); 1384 if (error != 0) 1385 pinm = NULL; 1386 1387 return (pinm); 1388 } 1389 1390 /* 1391 * Block or unblock an ASM multicast source on an inpcb. 1392 * This implements the delta-based API described in RFC 3678. 1393 * 1394 * The delta-based API applies only to exclusive-mode memberships. 1395 * An IGMP downcall will be performed. 1396 * 1397 * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1398 * 1399 * Return 0 if successful, otherwise return an appropriate error code. 1400 */ 1401 static int 1402 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1403 { 1404 struct group_source_req gsr; 1405 struct rm_priotracker in_ifa_tracker; 1406 sockunion_t *gsa, *ssa; 1407 struct ifnet *ifp; 1408 struct in_mfilter *imf; 1409 struct ip_moptions *imo; 1410 struct in_msource *ims; 1411 struct in_multi *inm; 1412 uint16_t fmode; 1413 int error, doblock; 1414 1415 ifp = NULL; 1416 error = 0; 1417 doblock = 0; 1418 1419 memset(&gsr, 0, sizeof(struct group_source_req)); 1420 gsa = (sockunion_t *)&gsr.gsr_group; 1421 ssa = (sockunion_t *)&gsr.gsr_source; 1422 1423 switch (sopt->sopt_name) { 1424 case IP_BLOCK_SOURCE: 1425 case IP_UNBLOCK_SOURCE: { 1426 struct ip_mreq_source mreqs; 1427 1428 error = sooptcopyin(sopt, &mreqs, 1429 sizeof(struct ip_mreq_source), 1430 sizeof(struct ip_mreq_source)); 1431 if (error) 1432 return (error); 1433 1434 gsa->sin.sin_family = AF_INET; 1435 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1436 gsa->sin.sin_addr = mreqs.imr_multiaddr; 1437 1438 ssa->sin.sin_family = AF_INET; 1439 ssa->sin.sin_len = sizeof(struct sockaddr_in); 1440 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1441 1442 if (!in_nullhost(mreqs.imr_interface)) { 1443 IN_IFADDR_RLOCK(&in_ifa_tracker); 1444 INADDR_TO_IFP(mreqs.imr_interface, ifp); 1445 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1446 } 1447 if (sopt->sopt_name == IP_BLOCK_SOURCE) 1448 doblock = 1; 1449 1450 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 1451 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 1452 break; 1453 } 1454 1455 case MCAST_BLOCK_SOURCE: 1456 case MCAST_UNBLOCK_SOURCE: 1457 error = sooptcopyin(sopt, &gsr, 1458 sizeof(struct group_source_req), 1459 sizeof(struct group_source_req)); 1460 if (error) 1461 return (error); 1462 1463 if (gsa->sin.sin_family != AF_INET || 1464 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 1465 return (EINVAL); 1466 1467 if (ssa->sin.sin_family != AF_INET || 1468 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 1469 return (EINVAL); 1470 1471 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1472 return (EADDRNOTAVAIL); 1473 1474 ifp = ifnet_byindex(gsr.gsr_interface); 1475 1476 if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1477 doblock = 1; 1478 break; 1479 1480 default: 1481 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1482 __func__, sopt->sopt_name); 1483 return (EOPNOTSUPP); 1484 break; 1485 } 1486 1487 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1488 return (EINVAL); 1489 1490 IN_MULTI_LOCK(); 1491 1492 /* 1493 * Check if we are actually a member of this group. 1494 */ 1495 imo = inp_findmoptions(inp); 1496 imf = imo_match_group(imo, ifp, &gsa->sa); 1497 if (imf == NULL) { 1498 error = EADDRNOTAVAIL; 1499 goto out_inp_locked; 1500 } 1501 inm = imf->imf_inm; 1502 1503 /* 1504 * Attempting to use the delta-based API on an 1505 * non exclusive-mode membership is an error. 1506 */ 1507 fmode = imf->imf_st[0]; 1508 if (fmode != MCAST_EXCLUDE) { 1509 error = EINVAL; 1510 goto out_inp_locked; 1511 } 1512 1513 /* 1514 * Deal with error cases up-front: 1515 * Asked to block, but already blocked; or 1516 * Asked to unblock, but nothing to unblock. 1517 * If adding a new block entry, allocate it. 1518 */ 1519 ims = imo_match_source(imf, &ssa->sa); 1520 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1521 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__, 1522 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not "); 1523 error = EADDRNOTAVAIL; 1524 goto out_inp_locked; 1525 } 1526 1527 INP_WLOCK_ASSERT(inp); 1528 1529 /* 1530 * Begin state merge transaction at socket layer. 1531 */ 1532 if (doblock) { 1533 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 1534 ims = imf_graft(imf, fmode, &ssa->sin); 1535 if (ims == NULL) 1536 error = ENOMEM; 1537 } else { 1538 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 1539 error = imf_prune(imf, &ssa->sin); 1540 } 1541 1542 if (error) { 1543 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__); 1544 goto out_imf_rollback; 1545 } 1546 1547 /* 1548 * Begin state merge transaction at IGMP layer. 1549 */ 1550 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1551 IN_MULTI_LIST_LOCK(); 1552 error = inm_merge(inm, imf); 1553 if (error) { 1554 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1555 IN_MULTI_LIST_UNLOCK(); 1556 goto out_imf_rollback; 1557 } 1558 1559 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1560 error = igmp_change_state(inm); 1561 IN_MULTI_LIST_UNLOCK(); 1562 if (error) 1563 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1564 1565 out_imf_rollback: 1566 if (error) 1567 imf_rollback(imf); 1568 else 1569 imf_commit(imf); 1570 1571 imf_reap(imf); 1572 1573 out_inp_locked: 1574 INP_WUNLOCK(inp); 1575 IN_MULTI_UNLOCK(); 1576 return (error); 1577 } 1578 1579 /* 1580 * Given an inpcb, return its multicast options structure pointer. Accepts 1581 * an unlocked inpcb pointer, but will return it locked. May sleep. 1582 * 1583 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1584 * SMPng: NOTE: Returns with the INP write lock held. 1585 */ 1586 static struct ip_moptions * 1587 inp_findmoptions(struct inpcb *inp) 1588 { 1589 struct ip_moptions *imo; 1590 1591 INP_WLOCK(inp); 1592 if (inp->inp_moptions != NULL) 1593 return (inp->inp_moptions); 1594 1595 INP_WUNLOCK(inp); 1596 1597 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK); 1598 1599 imo->imo_multicast_ifp = NULL; 1600 imo->imo_multicast_addr.s_addr = INADDR_ANY; 1601 imo->imo_multicast_vif = -1; 1602 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL; 1603 imo->imo_multicast_loop = in_mcast_loop; 1604 STAILQ_INIT(&imo->imo_head); 1605 1606 INP_WLOCK(inp); 1607 if (inp->inp_moptions != NULL) { 1608 free(imo, M_IPMOPTS); 1609 return (inp->inp_moptions); 1610 } 1611 inp->inp_moptions = imo; 1612 return (imo); 1613 } 1614 1615 static void 1616 inp_gcmoptions(struct ip_moptions *imo) 1617 { 1618 struct in_mfilter *imf; 1619 struct in_multi *inm; 1620 struct ifnet *ifp; 1621 1622 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) { 1623 ip_mfilter_remove(&imo->imo_head, imf); 1624 1625 imf_leave(imf); 1626 if ((inm = imf->imf_inm) != NULL) { 1627 if ((ifp = inm->inm_ifp) != NULL) { 1628 CURVNET_SET(ifp->if_vnet); 1629 (void)in_leavegroup(inm, imf); 1630 CURVNET_RESTORE(); 1631 } else { 1632 (void)in_leavegroup(inm, imf); 1633 } 1634 } 1635 ip_mfilter_free(imf); 1636 } 1637 free(imo, M_IPMOPTS); 1638 } 1639 1640 /* 1641 * Discard the IP multicast options (and source filters). To minimize 1642 * the amount of work done while holding locks such as the INP's 1643 * pcbinfo lock (which is used in the receive path), the free 1644 * operation is deferred to the epoch callback task. 1645 */ 1646 void 1647 inp_freemoptions(struct ip_moptions *imo) 1648 { 1649 if (imo == NULL) 1650 return; 1651 inp_gcmoptions(imo); 1652 } 1653 1654 /* 1655 * Atomically get source filters on a socket for an IPv4 multicast group. 1656 * Called with INP lock held; returns with lock released. 1657 */ 1658 static int 1659 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1660 { 1661 struct __msfilterreq msfr; 1662 sockunion_t *gsa; 1663 struct ifnet *ifp; 1664 struct ip_moptions *imo; 1665 struct in_mfilter *imf; 1666 struct ip_msource *ims; 1667 struct in_msource *lims; 1668 struct sockaddr_in *psin; 1669 struct sockaddr_storage *ptss; 1670 struct sockaddr_storage *tss; 1671 int error; 1672 size_t nsrcs, ncsrcs; 1673 1674 INP_WLOCK_ASSERT(inp); 1675 1676 imo = inp->inp_moptions; 1677 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__)); 1678 1679 INP_WUNLOCK(inp); 1680 1681 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1682 sizeof(struct __msfilterreq)); 1683 if (error) 1684 return (error); 1685 1686 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 1687 return (EINVAL); 1688 1689 ifp = ifnet_byindex(msfr.msfr_ifindex); 1690 if (ifp == NULL) 1691 return (EINVAL); 1692 1693 INP_WLOCK(inp); 1694 1695 /* 1696 * Lookup group on the socket. 1697 */ 1698 gsa = (sockunion_t *)&msfr.msfr_group; 1699 imf = imo_match_group(imo, ifp, &gsa->sa); 1700 if (imf == NULL) { 1701 INP_WUNLOCK(inp); 1702 return (EADDRNOTAVAIL); 1703 } 1704 1705 /* 1706 * Ignore memberships which are in limbo. 1707 */ 1708 if (imf->imf_st[1] == MCAST_UNDEFINED) { 1709 INP_WUNLOCK(inp); 1710 return (EAGAIN); 1711 } 1712 msfr.msfr_fmode = imf->imf_st[1]; 1713 1714 /* 1715 * If the user specified a buffer, copy out the source filter 1716 * entries to userland gracefully. 1717 * We only copy out the number of entries which userland 1718 * has asked for, but we always tell userland how big the 1719 * buffer really needs to be. 1720 */ 1721 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 1722 msfr.msfr_nsrcs = in_mcast_maxsocksrc; 1723 tss = NULL; 1724 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1725 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1726 M_TEMP, M_NOWAIT | M_ZERO); 1727 if (tss == NULL) { 1728 INP_WUNLOCK(inp); 1729 return (ENOBUFS); 1730 } 1731 } 1732 1733 /* 1734 * Count number of sources in-mode at t0. 1735 * If buffer space exists and remains, copy out source entries. 1736 */ 1737 nsrcs = msfr.msfr_nsrcs; 1738 ncsrcs = 0; 1739 ptss = tss; 1740 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1741 lims = (struct in_msource *)ims; 1742 if (lims->imsl_st[0] == MCAST_UNDEFINED || 1743 lims->imsl_st[0] != imf->imf_st[0]) 1744 continue; 1745 ++ncsrcs; 1746 if (tss != NULL && nsrcs > 0) { 1747 psin = (struct sockaddr_in *)ptss; 1748 psin->sin_family = AF_INET; 1749 psin->sin_len = sizeof(struct sockaddr_in); 1750 psin->sin_addr.s_addr = htonl(lims->ims_haddr); 1751 psin->sin_port = 0; 1752 ++ptss; 1753 --nsrcs; 1754 } 1755 } 1756 1757 INP_WUNLOCK(inp); 1758 1759 if (tss != NULL) { 1760 error = copyout(tss, msfr.msfr_srcs, 1761 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1762 free(tss, M_TEMP); 1763 if (error) 1764 return (error); 1765 } 1766 1767 msfr.msfr_nsrcs = ncsrcs; 1768 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1769 1770 return (error); 1771 } 1772 1773 /* 1774 * Return the IP multicast options in response to user getsockopt(). 1775 */ 1776 int 1777 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1778 { 1779 struct rm_priotracker in_ifa_tracker; 1780 struct ip_mreqn mreqn; 1781 struct ip_moptions *imo; 1782 struct ifnet *ifp; 1783 struct in_ifaddr *ia; 1784 int error, optval; 1785 u_char coptval; 1786 1787 INP_WLOCK(inp); 1788 imo = inp->inp_moptions; 1789 /* 1790 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 1791 * or is a divert socket, reject it. 1792 */ 1793 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 1794 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1795 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) { 1796 INP_WUNLOCK(inp); 1797 return (EOPNOTSUPP); 1798 } 1799 1800 error = 0; 1801 switch (sopt->sopt_name) { 1802 case IP_MULTICAST_VIF: 1803 if (imo != NULL) 1804 optval = imo->imo_multicast_vif; 1805 else 1806 optval = -1; 1807 INP_WUNLOCK(inp); 1808 error = sooptcopyout(sopt, &optval, sizeof(int)); 1809 break; 1810 1811 case IP_MULTICAST_IF: 1812 memset(&mreqn, 0, sizeof(struct ip_mreqn)); 1813 if (imo != NULL) { 1814 ifp = imo->imo_multicast_ifp; 1815 if (!in_nullhost(imo->imo_multicast_addr)) { 1816 mreqn.imr_address = imo->imo_multicast_addr; 1817 } else if (ifp != NULL) { 1818 struct epoch_tracker et; 1819 1820 mreqn.imr_ifindex = ifp->if_index; 1821 NET_EPOCH_ENTER(et); 1822 IFP_TO_IA(ifp, ia, &in_ifa_tracker); 1823 if (ia != NULL) 1824 mreqn.imr_address = 1825 IA_SIN(ia)->sin_addr; 1826 NET_EPOCH_EXIT(et); 1827 } 1828 } 1829 INP_WUNLOCK(inp); 1830 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 1831 error = sooptcopyout(sopt, &mreqn, 1832 sizeof(struct ip_mreqn)); 1833 } else { 1834 error = sooptcopyout(sopt, &mreqn.imr_address, 1835 sizeof(struct in_addr)); 1836 } 1837 break; 1838 1839 case IP_MULTICAST_TTL: 1840 if (imo == NULL) 1841 optval = coptval = IP_DEFAULT_MULTICAST_TTL; 1842 else 1843 optval = coptval = imo->imo_multicast_ttl; 1844 INP_WUNLOCK(inp); 1845 if (sopt->sopt_valsize == sizeof(u_char)) 1846 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1847 else 1848 error = sooptcopyout(sopt, &optval, sizeof(int)); 1849 break; 1850 1851 case IP_MULTICAST_LOOP: 1852 if (imo == NULL) 1853 optval = coptval = IP_DEFAULT_MULTICAST_LOOP; 1854 else 1855 optval = coptval = imo->imo_multicast_loop; 1856 INP_WUNLOCK(inp); 1857 if (sopt->sopt_valsize == sizeof(u_char)) 1858 error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1859 else 1860 error = sooptcopyout(sopt, &optval, sizeof(int)); 1861 break; 1862 1863 case IP_MSFILTER: 1864 if (imo == NULL) { 1865 error = EADDRNOTAVAIL; 1866 INP_WUNLOCK(inp); 1867 } else { 1868 error = inp_get_source_filters(inp, sopt); 1869 } 1870 break; 1871 1872 default: 1873 INP_WUNLOCK(inp); 1874 error = ENOPROTOOPT; 1875 break; 1876 } 1877 1878 INP_UNLOCK_ASSERT(inp); 1879 1880 return (error); 1881 } 1882 1883 /* 1884 * Look up the ifnet to use for a multicast group membership, 1885 * given the IPv4 address of an interface, and the IPv4 group address. 1886 * 1887 * This routine exists to support legacy multicast applications 1888 * which do not understand that multicast memberships are scoped to 1889 * specific physical links in the networking stack, or which need 1890 * to join link-scope groups before IPv4 addresses are configured. 1891 * 1892 * If inp is non-NULL, use this socket's current FIB number for any 1893 * required FIB lookup. 1894 * If ina is INADDR_ANY, look up the group address in the unicast FIB, 1895 * and use its ifp; usually, this points to the default next-hop. 1896 * 1897 * If the FIB lookup fails, attempt to use the first non-loopback 1898 * interface with multicast capability in the system as a 1899 * last resort. The legacy IPv4 ASM API requires that we do 1900 * this in order to allow groups to be joined when the routing 1901 * table has not yet been populated during boot. 1902 * 1903 * Returns NULL if no ifp could be found. 1904 * 1905 * FUTURE: Implement IPv4 source-address selection. 1906 */ 1907 static struct ifnet * 1908 inp_lookup_mcast_ifp(const struct inpcb *inp, 1909 const struct sockaddr_in *gsin, const struct in_addr ina) 1910 { 1911 struct rm_priotracker in_ifa_tracker; 1912 struct ifnet *ifp; 1913 struct nhop4_basic nh4; 1914 uint32_t fibnum; 1915 1916 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__)); 1917 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)), 1918 ("%s: not multicast", __func__)); 1919 1920 ifp = NULL; 1921 if (!in_nullhost(ina)) { 1922 IN_IFADDR_RLOCK(&in_ifa_tracker); 1923 INADDR_TO_IFP(ina, ifp); 1924 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1925 } else { 1926 fibnum = inp ? inp->inp_inc.inc_fibnum : 0; 1927 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0) 1928 ifp = nh4.nh_ifp; 1929 else { 1930 struct in_ifaddr *ia; 1931 struct ifnet *mifp; 1932 1933 mifp = NULL; 1934 IN_IFADDR_RLOCK(&in_ifa_tracker); 1935 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1936 mifp = ia->ia_ifp; 1937 if (!(mifp->if_flags & IFF_LOOPBACK) && 1938 (mifp->if_flags & IFF_MULTICAST)) { 1939 ifp = mifp; 1940 break; 1941 } 1942 } 1943 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 1944 } 1945 } 1946 1947 return (ifp); 1948 } 1949 1950 /* 1951 * Join an IPv4 multicast group, possibly with a source. 1952 */ 1953 static int 1954 inp_join_group(struct inpcb *inp, struct sockopt *sopt) 1955 { 1956 struct group_source_req gsr; 1957 sockunion_t *gsa, *ssa; 1958 struct ifnet *ifp; 1959 struct in_mfilter *imf; 1960 struct ip_moptions *imo; 1961 struct in_multi *inm; 1962 struct in_msource *lims; 1963 int error, is_new; 1964 1965 ifp = NULL; 1966 lims = NULL; 1967 error = 0; 1968 1969 memset(&gsr, 0, sizeof(struct group_source_req)); 1970 gsa = (sockunion_t *)&gsr.gsr_group; 1971 gsa->ss.ss_family = AF_UNSPEC; 1972 ssa = (sockunion_t *)&gsr.gsr_source; 1973 ssa->ss.ss_family = AF_UNSPEC; 1974 1975 switch (sopt->sopt_name) { 1976 case IP_ADD_MEMBERSHIP: { 1977 struct ip_mreqn mreqn; 1978 1979 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) 1980 error = sooptcopyin(sopt, &mreqn, 1981 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn)); 1982 else 1983 error = sooptcopyin(sopt, &mreqn, 1984 sizeof(struct ip_mreq), sizeof(struct ip_mreq)); 1985 if (error) 1986 return (error); 1987 1988 gsa->sin.sin_family = AF_INET; 1989 gsa->sin.sin_len = sizeof(struct sockaddr_in); 1990 gsa->sin.sin_addr = mreqn.imr_multiaddr; 1991 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1992 return (EINVAL); 1993 1994 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) && 1995 mreqn.imr_ifindex != 0) 1996 ifp = ifnet_byindex(mreqn.imr_ifindex); 1997 else 1998 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 1999 mreqn.imr_address); 2000 break; 2001 } 2002 case IP_ADD_SOURCE_MEMBERSHIP: { 2003 struct ip_mreq_source mreqs; 2004 2005 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source), 2006 sizeof(struct ip_mreq_source)); 2007 if (error) 2008 return (error); 2009 2010 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET; 2011 gsa->sin.sin_len = ssa->sin.sin_len = 2012 sizeof(struct sockaddr_in); 2013 2014 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2015 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2016 return (EINVAL); 2017 2018 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2019 2020 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 2021 mreqs.imr_interface); 2022 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2023 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2024 break; 2025 } 2026 2027 case MCAST_JOIN_GROUP: 2028 case MCAST_JOIN_SOURCE_GROUP: 2029 if (sopt->sopt_name == MCAST_JOIN_GROUP) { 2030 error = sooptcopyin(sopt, &gsr, 2031 sizeof(struct group_req), 2032 sizeof(struct group_req)); 2033 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2034 error = sooptcopyin(sopt, &gsr, 2035 sizeof(struct group_source_req), 2036 sizeof(struct group_source_req)); 2037 } 2038 if (error) 2039 return (error); 2040 2041 if (gsa->sin.sin_family != AF_INET || 2042 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2043 return (EINVAL); 2044 2045 /* 2046 * Overwrite the port field if present, as the sockaddr 2047 * being copied in may be matched with a binary comparison. 2048 */ 2049 gsa->sin.sin_port = 0; 2050 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 2051 if (ssa->sin.sin_family != AF_INET || 2052 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2053 return (EINVAL); 2054 ssa->sin.sin_port = 0; 2055 } 2056 2057 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2058 return (EINVAL); 2059 2060 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2061 return (EADDRNOTAVAIL); 2062 ifp = ifnet_byindex(gsr.gsr_interface); 2063 break; 2064 2065 default: 2066 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2067 __func__, sopt->sopt_name); 2068 return (EOPNOTSUPP); 2069 break; 2070 } 2071 2072 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) 2073 return (EADDRNOTAVAIL); 2074 2075 IN_MULTI_LOCK(); 2076 2077 /* 2078 * Find the membership in the membership list. 2079 */ 2080 imo = inp_findmoptions(inp); 2081 imf = imo_match_group(imo, ifp, &gsa->sa); 2082 if (imf == NULL) { 2083 is_new = 1; 2084 inm = NULL; 2085 2086 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) { 2087 error = ENOMEM; 2088 goto out_inp_locked; 2089 } 2090 } else { 2091 is_new = 0; 2092 inm = imf->imf_inm; 2093 2094 if (ssa->ss.ss_family != AF_UNSPEC) { 2095 /* 2096 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership 2097 * is an error. On an existing inclusive membership, 2098 * it just adds the source to the filter list. 2099 */ 2100 if (imf->imf_st[1] != MCAST_INCLUDE) { 2101 error = EINVAL; 2102 goto out_inp_locked; 2103 } 2104 /* 2105 * Throw out duplicates. 2106 * 2107 * XXX FIXME: This makes a naive assumption that 2108 * even if entries exist for *ssa in this imf, 2109 * they will be rejected as dupes, even if they 2110 * are not valid in the current mode (in-mode). 2111 * 2112 * in_msource is transactioned just as for anything 2113 * else in SSM -- but note naive use of inm_graft() 2114 * below for allocating new filter entries. 2115 * 2116 * This is only an issue if someone mixes the 2117 * full-state SSM API with the delta-based API, 2118 * which is discouraged in the relevant RFCs. 2119 */ 2120 lims = imo_match_source(imf, &ssa->sa); 2121 if (lims != NULL /*&& 2122 lims->imsl_st[1] == MCAST_INCLUDE*/) { 2123 error = EADDRNOTAVAIL; 2124 goto out_inp_locked; 2125 } 2126 } else { 2127 /* 2128 * MCAST_JOIN_GROUP on an existing exclusive 2129 * membership is an error; return EADDRINUSE 2130 * to preserve 4.4BSD API idempotence, and 2131 * avoid tedious detour to code below. 2132 * NOTE: This is bending RFC 3678 a bit. 2133 * 2134 * On an existing inclusive membership, this is also 2135 * an error; if you want to change filter mode, 2136 * you must use the userland API setsourcefilter(). 2137 * XXX We don't reject this for imf in UNDEFINED 2138 * state at t1, because allocation of a filter 2139 * is atomic with allocation of a membership. 2140 */ 2141 error = EINVAL; 2142 if (imf->imf_st[1] == MCAST_EXCLUDE) 2143 error = EADDRINUSE; 2144 goto out_inp_locked; 2145 } 2146 } 2147 2148 /* 2149 * Begin state merge transaction at socket layer. 2150 */ 2151 INP_WLOCK_ASSERT(inp); 2152 2153 /* 2154 * Graft new source into filter list for this inpcb's 2155 * membership of the group. The in_multi may not have 2156 * been allocated yet if this is a new membership, however, 2157 * the in_mfilter slot will be allocated and must be initialized. 2158 * 2159 * Note: Grafting of exclusive mode filters doesn't happen 2160 * in this path. 2161 * XXX: Should check for non-NULL lims (node exists but may 2162 * not be in-mode) for interop with full-state API. 2163 */ 2164 if (ssa->ss.ss_family != AF_UNSPEC) { 2165 /* Membership starts in IN mode */ 2166 if (is_new) { 2167 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__); 2168 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE); 2169 if (imf == NULL) { 2170 error = ENOMEM; 2171 goto out_inp_locked; 2172 } 2173 } else { 2174 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 2175 } 2176 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin); 2177 if (lims == NULL) { 2178 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2179 __func__); 2180 error = ENOMEM; 2181 goto out_inp_locked; 2182 } 2183 } else { 2184 /* No address specified; Membership starts in EX mode */ 2185 if (is_new) { 2186 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__); 2187 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE); 2188 if (imf == NULL) { 2189 error = ENOMEM; 2190 goto out_inp_locked; 2191 } 2192 } 2193 } 2194 2195 /* 2196 * Begin state merge transaction at IGMP layer. 2197 */ 2198 if (is_new) { 2199 in_pcbref(inp); 2200 INP_WUNLOCK(inp); 2201 2202 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf, 2203 &imf->imf_inm); 2204 2205 INP_WLOCK(inp); 2206 if (in_pcbrele_wlocked(inp)) { 2207 error = ENXIO; 2208 goto out_inp_unlocked; 2209 } 2210 if (error) { 2211 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed", 2212 __func__); 2213 goto out_inp_locked; 2214 } 2215 /* 2216 * NOTE: Refcount from in_joingroup_locked() 2217 * is protecting membership. 2218 */ 2219 ip_mfilter_insert(&imo->imo_head, imf); 2220 } else { 2221 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2222 IN_MULTI_LIST_LOCK(); 2223 error = inm_merge(inm, imf); 2224 if (error) { 2225 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2226 __func__); 2227 IN_MULTI_LIST_UNLOCK(); 2228 imf_rollback(imf); 2229 imf_reap(imf); 2230 goto out_inp_locked; 2231 } 2232 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2233 error = igmp_change_state(inm); 2234 IN_MULTI_LIST_UNLOCK(); 2235 if (error) { 2236 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2237 __func__); 2238 imf_rollback(imf); 2239 imf_reap(imf); 2240 goto out_inp_locked; 2241 } 2242 } 2243 2244 imf_commit(imf); 2245 imf = NULL; 2246 2247 out_inp_locked: 2248 INP_WUNLOCK(inp); 2249 out_inp_unlocked: 2250 IN_MULTI_UNLOCK(); 2251 2252 if (is_new && imf) { 2253 if (imf->imf_inm != NULL) { 2254 IN_MULTI_LIST_LOCK(); 2255 IF_ADDR_WLOCK(ifp); 2256 inm_release_deferred(imf->imf_inm); 2257 IF_ADDR_WUNLOCK(ifp); 2258 IN_MULTI_LIST_UNLOCK(); 2259 } 2260 ip_mfilter_free(imf); 2261 } 2262 return (error); 2263 } 2264 2265 /* 2266 * Leave an IPv4 multicast group on an inpcb, possibly with a source. 2267 */ 2268 static int 2269 inp_leave_group(struct inpcb *inp, struct sockopt *sopt) 2270 { 2271 struct group_source_req gsr; 2272 struct ip_mreq_source mreqs; 2273 struct rm_priotracker in_ifa_tracker; 2274 sockunion_t *gsa, *ssa; 2275 struct ifnet *ifp; 2276 struct in_mfilter *imf; 2277 struct ip_moptions *imo; 2278 struct in_msource *ims; 2279 struct in_multi *inm; 2280 int error; 2281 bool is_final; 2282 2283 ifp = NULL; 2284 error = 0; 2285 is_final = true; 2286 2287 memset(&gsr, 0, sizeof(struct group_source_req)); 2288 gsa = (sockunion_t *)&gsr.gsr_group; 2289 gsa->ss.ss_family = AF_UNSPEC; 2290 ssa = (sockunion_t *)&gsr.gsr_source; 2291 ssa->ss.ss_family = AF_UNSPEC; 2292 2293 switch (sopt->sopt_name) { 2294 case IP_DROP_MEMBERSHIP: 2295 case IP_DROP_SOURCE_MEMBERSHIP: 2296 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) { 2297 error = sooptcopyin(sopt, &mreqs, 2298 sizeof(struct ip_mreq), 2299 sizeof(struct ip_mreq)); 2300 /* 2301 * Swap interface and sourceaddr arguments, 2302 * as ip_mreq and ip_mreq_source are laid 2303 * out differently. 2304 */ 2305 mreqs.imr_interface = mreqs.imr_sourceaddr; 2306 mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 2307 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2308 error = sooptcopyin(sopt, &mreqs, 2309 sizeof(struct ip_mreq_source), 2310 sizeof(struct ip_mreq_source)); 2311 } 2312 if (error) 2313 return (error); 2314 2315 gsa->sin.sin_family = AF_INET; 2316 gsa->sin.sin_len = sizeof(struct sockaddr_in); 2317 gsa->sin.sin_addr = mreqs.imr_multiaddr; 2318 2319 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2320 ssa->sin.sin_family = AF_INET; 2321 ssa->sin.sin_len = sizeof(struct sockaddr_in); 2322 ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2323 } 2324 2325 /* 2326 * Attempt to look up hinted ifp from interface address. 2327 * Fallthrough with null ifp iff lookup fails, to 2328 * preserve 4.4BSD mcast API idempotence. 2329 * XXX NOTE WELL: The RFC 3678 API is preferred because 2330 * using an IPv4 address as a key is racy. 2331 */ 2332 if (!in_nullhost(mreqs.imr_interface)) { 2333 IN_IFADDR_RLOCK(&in_ifa_tracker); 2334 INADDR_TO_IFP(mreqs.imr_interface, ifp); 2335 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2336 } 2337 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p", 2338 __func__, ntohl(mreqs.imr_interface.s_addr), ifp); 2339 2340 break; 2341 2342 case MCAST_LEAVE_GROUP: 2343 case MCAST_LEAVE_SOURCE_GROUP: 2344 if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2345 error = sooptcopyin(sopt, &gsr, 2346 sizeof(struct group_req), 2347 sizeof(struct group_req)); 2348 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2349 error = sooptcopyin(sopt, &gsr, 2350 sizeof(struct group_source_req), 2351 sizeof(struct group_source_req)); 2352 } 2353 if (error) 2354 return (error); 2355 2356 if (gsa->sin.sin_family != AF_INET || 2357 gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2358 return (EINVAL); 2359 2360 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2361 if (ssa->sin.sin_family != AF_INET || 2362 ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2363 return (EINVAL); 2364 } 2365 2366 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2367 return (EADDRNOTAVAIL); 2368 2369 ifp = ifnet_byindex(gsr.gsr_interface); 2370 2371 if (ifp == NULL) 2372 return (EADDRNOTAVAIL); 2373 break; 2374 2375 default: 2376 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2377 __func__, sopt->sopt_name); 2378 return (EOPNOTSUPP); 2379 break; 2380 } 2381 2382 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2383 return (EINVAL); 2384 2385 IN_MULTI_LOCK(); 2386 2387 /* 2388 * Find the membership in the membership list. 2389 */ 2390 imo = inp_findmoptions(inp); 2391 imf = imo_match_group(imo, ifp, &gsa->sa); 2392 if (imf == NULL) { 2393 error = EADDRNOTAVAIL; 2394 goto out_inp_locked; 2395 } 2396 inm = imf->imf_inm; 2397 2398 if (ssa->ss.ss_family != AF_UNSPEC) 2399 is_final = false; 2400 2401 /* 2402 * Begin state merge transaction at socket layer. 2403 */ 2404 INP_WLOCK_ASSERT(inp); 2405 2406 /* 2407 * If we were instructed only to leave a given source, do so. 2408 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2409 */ 2410 if (is_final) { 2411 ip_mfilter_remove(&imo->imo_head, imf); 2412 imf_leave(imf); 2413 2414 /* 2415 * Give up the multicast address record to which 2416 * the membership points. 2417 */ 2418 (void) in_leavegroup_locked(imf->imf_inm, imf); 2419 } else { 2420 if (imf->imf_st[0] == MCAST_EXCLUDE) { 2421 error = EADDRNOTAVAIL; 2422 goto out_inp_locked; 2423 } 2424 ims = imo_match_source(imf, &ssa->sa); 2425 if (ims == NULL) { 2426 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", 2427 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not "); 2428 error = EADDRNOTAVAIL; 2429 goto out_inp_locked; 2430 } 2431 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 2432 error = imf_prune(imf, &ssa->sin); 2433 if (error) { 2434 CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2435 __func__); 2436 goto out_inp_locked; 2437 } 2438 } 2439 2440 /* 2441 * Begin state merge transaction at IGMP layer. 2442 */ 2443 if (!is_final) { 2444 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2445 IN_MULTI_LIST_LOCK(); 2446 error = inm_merge(inm, imf); 2447 if (error) { 2448 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2449 __func__); 2450 IN_MULTI_LIST_UNLOCK(); 2451 imf_rollback(imf); 2452 imf_reap(imf); 2453 goto out_inp_locked; 2454 } 2455 2456 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2457 error = igmp_change_state(inm); 2458 IN_MULTI_LIST_UNLOCK(); 2459 if (error) { 2460 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2461 __func__); 2462 imf_rollback(imf); 2463 imf_reap(imf); 2464 goto out_inp_locked; 2465 } 2466 } 2467 imf_commit(imf); 2468 imf_reap(imf); 2469 2470 out_inp_locked: 2471 INP_WUNLOCK(inp); 2472 2473 if (is_final && imf) 2474 ip_mfilter_free(imf); 2475 2476 IN_MULTI_UNLOCK(); 2477 return (error); 2478 } 2479 2480 /* 2481 * Select the interface for transmitting IPv4 multicast datagrams. 2482 * 2483 * Either an instance of struct in_addr or an instance of struct ip_mreqn 2484 * may be passed to this socket option. An address of INADDR_ANY or an 2485 * interface index of 0 is used to remove a previous selection. 2486 * When no interface is selected, one is chosen for every send. 2487 */ 2488 static int 2489 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2490 { 2491 struct rm_priotracker in_ifa_tracker; 2492 struct in_addr addr; 2493 struct ip_mreqn mreqn; 2494 struct ifnet *ifp; 2495 struct ip_moptions *imo; 2496 int error; 2497 2498 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 2499 /* 2500 * An interface index was specified using the 2501 * Linux-derived ip_mreqn structure. 2502 */ 2503 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn), 2504 sizeof(struct ip_mreqn)); 2505 if (error) 2506 return (error); 2507 2508 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex) 2509 return (EINVAL); 2510 2511 if (mreqn.imr_ifindex == 0) { 2512 ifp = NULL; 2513 } else { 2514 ifp = ifnet_byindex(mreqn.imr_ifindex); 2515 if (ifp == NULL) 2516 return (EADDRNOTAVAIL); 2517 } 2518 } else { 2519 /* 2520 * An interface was specified by IPv4 address. 2521 * This is the traditional BSD usage. 2522 */ 2523 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr), 2524 sizeof(struct in_addr)); 2525 if (error) 2526 return (error); 2527 if (in_nullhost(addr)) { 2528 ifp = NULL; 2529 } else { 2530 IN_IFADDR_RLOCK(&in_ifa_tracker); 2531 INADDR_TO_IFP(addr, ifp); 2532 IN_IFADDR_RUNLOCK(&in_ifa_tracker); 2533 if (ifp == NULL) 2534 return (EADDRNOTAVAIL); 2535 } 2536 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp, 2537 ntohl(addr.s_addr)); 2538 } 2539 2540 /* Reject interfaces which do not support multicast. */ 2541 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0) 2542 return (EOPNOTSUPP); 2543 2544 imo = inp_findmoptions(inp); 2545 imo->imo_multicast_ifp = ifp; 2546 imo->imo_multicast_addr.s_addr = INADDR_ANY; 2547 INP_WUNLOCK(inp); 2548 2549 return (0); 2550 } 2551 2552 /* 2553 * Atomically set source filters on a socket for an IPv4 multicast group. 2554 * 2555 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2556 */ 2557 static int 2558 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2559 { 2560 struct __msfilterreq msfr; 2561 sockunion_t *gsa; 2562 struct ifnet *ifp; 2563 struct in_mfilter *imf; 2564 struct ip_moptions *imo; 2565 struct in_multi *inm; 2566 int error; 2567 2568 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2569 sizeof(struct __msfilterreq)); 2570 if (error) 2571 return (error); 2572 2573 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc) 2574 return (ENOBUFS); 2575 2576 if ((msfr.msfr_fmode != MCAST_EXCLUDE && 2577 msfr.msfr_fmode != MCAST_INCLUDE)) 2578 return (EINVAL); 2579 2580 if (msfr.msfr_group.ss_family != AF_INET || 2581 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in)) 2582 return (EINVAL); 2583 2584 gsa = (sockunion_t *)&msfr.msfr_group; 2585 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2586 return (EINVAL); 2587 2588 gsa->sin.sin_port = 0; /* ignore port */ 2589 2590 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 2591 return (EADDRNOTAVAIL); 2592 2593 ifp = ifnet_byindex(msfr.msfr_ifindex); 2594 if (ifp == NULL) 2595 return (EADDRNOTAVAIL); 2596 2597 IN_MULTI_LOCK(); 2598 2599 /* 2600 * Take the INP write lock. 2601 * Check if this socket is a member of this group. 2602 */ 2603 imo = inp_findmoptions(inp); 2604 imf = imo_match_group(imo, ifp, &gsa->sa); 2605 if (imf == NULL) { 2606 error = EADDRNOTAVAIL; 2607 goto out_inp_locked; 2608 } 2609 inm = imf->imf_inm; 2610 2611 /* 2612 * Begin state merge transaction at socket layer. 2613 */ 2614 INP_WLOCK_ASSERT(inp); 2615 2616 imf->imf_st[1] = msfr.msfr_fmode; 2617 2618 /* 2619 * Apply any new source filters, if present. 2620 * Make a copy of the user-space source vector so 2621 * that we may copy them with a single copyin. This 2622 * allows us to deal with page faults up-front. 2623 */ 2624 if (msfr.msfr_nsrcs > 0) { 2625 struct in_msource *lims; 2626 struct sockaddr_in *psin; 2627 struct sockaddr_storage *kss, *pkss; 2628 int i; 2629 2630 INP_WUNLOCK(inp); 2631 2632 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries", 2633 __func__, (unsigned long)msfr.msfr_nsrcs); 2634 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2635 M_TEMP, M_WAITOK); 2636 error = copyin(msfr.msfr_srcs, kss, 2637 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2638 if (error) { 2639 free(kss, M_TEMP); 2640 return (error); 2641 } 2642 2643 INP_WLOCK(inp); 2644 2645 /* 2646 * Mark all source filters as UNDEFINED at t1. 2647 * Restore new group filter mode, as imf_leave() 2648 * will set it to INCLUDE. 2649 */ 2650 imf_leave(imf); 2651 imf->imf_st[1] = msfr.msfr_fmode; 2652 2653 /* 2654 * Update socket layer filters at t1, lazy-allocating 2655 * new entries. This saves a bunch of memory at the 2656 * cost of one RB_FIND() per source entry; duplicate 2657 * entries in the msfr_nsrcs vector are ignored. 2658 * If we encounter an error, rollback transaction. 2659 * 2660 * XXX This too could be replaced with a set-symmetric 2661 * difference like loop to avoid walking from root 2662 * every time, as the key space is common. 2663 */ 2664 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2665 psin = (struct sockaddr_in *)pkss; 2666 if (psin->sin_family != AF_INET) { 2667 error = EAFNOSUPPORT; 2668 break; 2669 } 2670 if (psin->sin_len != sizeof(struct sockaddr_in)) { 2671 error = EINVAL; 2672 break; 2673 } 2674 error = imf_get_source(imf, psin, &lims); 2675 if (error) 2676 break; 2677 lims->imsl_st[1] = imf->imf_st[1]; 2678 } 2679 free(kss, M_TEMP); 2680 } 2681 2682 if (error) 2683 goto out_imf_rollback; 2684 2685 INP_WLOCK_ASSERT(inp); 2686 2687 /* 2688 * Begin state merge transaction at IGMP layer. 2689 */ 2690 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2691 IN_MULTI_LIST_LOCK(); 2692 error = inm_merge(inm, imf); 2693 if (error) { 2694 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 2695 IN_MULTI_LIST_UNLOCK(); 2696 goto out_imf_rollback; 2697 } 2698 2699 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2700 error = igmp_change_state(inm); 2701 IN_MULTI_LIST_UNLOCK(); 2702 if (error) 2703 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 2704 2705 out_imf_rollback: 2706 if (error) 2707 imf_rollback(imf); 2708 else 2709 imf_commit(imf); 2710 2711 imf_reap(imf); 2712 2713 out_inp_locked: 2714 INP_WUNLOCK(inp); 2715 IN_MULTI_UNLOCK(); 2716 return (error); 2717 } 2718 2719 /* 2720 * Set the IP multicast options in response to user setsockopt(). 2721 * 2722 * Many of the socket options handled in this function duplicate the 2723 * functionality of socket options in the regular unicast API. However, 2724 * it is not possible to merge the duplicate code, because the idempotence 2725 * of the IPv4 multicast part of the BSD Sockets API must be preserved; 2726 * the effects of these options must be treated as separate and distinct. 2727 * 2728 * SMPng: XXX: Unlocked read of inp_socket believed OK. 2729 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING 2730 * is refactored to no longer use vifs. 2731 */ 2732 int 2733 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2734 { 2735 struct ip_moptions *imo; 2736 int error; 2737 2738 error = 0; 2739 2740 /* 2741 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 2742 * or is a divert socket, reject it. 2743 */ 2744 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 2745 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2746 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) 2747 return (EOPNOTSUPP); 2748 2749 switch (sopt->sopt_name) { 2750 case IP_MULTICAST_VIF: { 2751 int vifi; 2752 /* 2753 * Select a multicast VIF for transmission. 2754 * Only useful if multicast forwarding is active. 2755 */ 2756 if (legal_vif_num == NULL) { 2757 error = EOPNOTSUPP; 2758 break; 2759 } 2760 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int)); 2761 if (error) 2762 break; 2763 if (!legal_vif_num(vifi) && (vifi != -1)) { 2764 error = EINVAL; 2765 break; 2766 } 2767 imo = inp_findmoptions(inp); 2768 imo->imo_multicast_vif = vifi; 2769 INP_WUNLOCK(inp); 2770 break; 2771 } 2772 2773 case IP_MULTICAST_IF: 2774 error = inp_set_multicast_if(inp, sopt); 2775 break; 2776 2777 case IP_MULTICAST_TTL: { 2778 u_char ttl; 2779 2780 /* 2781 * Set the IP time-to-live for outgoing multicast packets. 2782 * The original multicast API required a char argument, 2783 * which is inconsistent with the rest of the socket API. 2784 * We allow either a char or an int. 2785 */ 2786 if (sopt->sopt_valsize == sizeof(u_char)) { 2787 error = sooptcopyin(sopt, &ttl, sizeof(u_char), 2788 sizeof(u_char)); 2789 if (error) 2790 break; 2791 } else { 2792 u_int ittl; 2793 2794 error = sooptcopyin(sopt, &ittl, sizeof(u_int), 2795 sizeof(u_int)); 2796 if (error) 2797 break; 2798 if (ittl > 255) { 2799 error = EINVAL; 2800 break; 2801 } 2802 ttl = (u_char)ittl; 2803 } 2804 imo = inp_findmoptions(inp); 2805 imo->imo_multicast_ttl = ttl; 2806 INP_WUNLOCK(inp); 2807 break; 2808 } 2809 2810 case IP_MULTICAST_LOOP: { 2811 u_char loop; 2812 2813 /* 2814 * Set the loopback flag for outgoing multicast packets. 2815 * Must be zero or one. The original multicast API required a 2816 * char argument, which is inconsistent with the rest 2817 * of the socket API. We allow either a char or an int. 2818 */ 2819 if (sopt->sopt_valsize == sizeof(u_char)) { 2820 error = sooptcopyin(sopt, &loop, sizeof(u_char), 2821 sizeof(u_char)); 2822 if (error) 2823 break; 2824 } else { 2825 u_int iloop; 2826 2827 error = sooptcopyin(sopt, &iloop, sizeof(u_int), 2828 sizeof(u_int)); 2829 if (error) 2830 break; 2831 loop = (u_char)iloop; 2832 } 2833 imo = inp_findmoptions(inp); 2834 imo->imo_multicast_loop = !!loop; 2835 INP_WUNLOCK(inp); 2836 break; 2837 } 2838 2839 case IP_ADD_MEMBERSHIP: 2840 case IP_ADD_SOURCE_MEMBERSHIP: 2841 case MCAST_JOIN_GROUP: 2842 case MCAST_JOIN_SOURCE_GROUP: 2843 error = inp_join_group(inp, sopt); 2844 break; 2845 2846 case IP_DROP_MEMBERSHIP: 2847 case IP_DROP_SOURCE_MEMBERSHIP: 2848 case MCAST_LEAVE_GROUP: 2849 case MCAST_LEAVE_SOURCE_GROUP: 2850 error = inp_leave_group(inp, sopt); 2851 break; 2852 2853 case IP_BLOCK_SOURCE: 2854 case IP_UNBLOCK_SOURCE: 2855 case MCAST_BLOCK_SOURCE: 2856 case MCAST_UNBLOCK_SOURCE: 2857 error = inp_block_unblock_source(inp, sopt); 2858 break; 2859 2860 case IP_MSFILTER: 2861 error = inp_set_source_filters(inp, sopt); 2862 break; 2863 2864 default: 2865 error = EOPNOTSUPP; 2866 break; 2867 } 2868 2869 INP_UNLOCK_ASSERT(inp); 2870 2871 return (error); 2872 } 2873 2874 /* 2875 * Expose IGMP's multicast filter mode and source list(s) to userland, 2876 * keyed by (ifindex, group). 2877 * The filter mode is written out as a uint32_t, followed by 2878 * 0..n of struct in_addr. 2879 * For use by ifmcstat(8). 2880 * SMPng: NOTE: unlocked read of ifindex space. 2881 */ 2882 static int 2883 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS) 2884 { 2885 struct in_addr src, group; 2886 struct epoch_tracker et; 2887 struct ifnet *ifp; 2888 struct ifmultiaddr *ifma; 2889 struct in_multi *inm; 2890 struct ip_msource *ims; 2891 int *name; 2892 int retval; 2893 u_int namelen; 2894 uint32_t fmode, ifindex; 2895 2896 name = (int *)arg1; 2897 namelen = arg2; 2898 2899 if (req->newptr != NULL) 2900 return (EPERM); 2901 2902 if (namelen != 2) 2903 return (EINVAL); 2904 2905 ifindex = name[0]; 2906 if (ifindex <= 0 || ifindex > V_if_index) { 2907 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range", 2908 __func__, ifindex); 2909 return (ENOENT); 2910 } 2911 2912 group.s_addr = name[1]; 2913 if (!IN_MULTICAST(ntohl(group.s_addr))) { 2914 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast", 2915 __func__, ntohl(group.s_addr)); 2916 return (EINVAL); 2917 } 2918 2919 NET_EPOCH_ENTER(et); 2920 ifp = ifnet_byindex(ifindex); 2921 if (ifp == NULL) { 2922 NET_EPOCH_EXIT(et); 2923 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u", 2924 __func__, ifindex); 2925 return (ENOENT); 2926 } 2927 2928 retval = sysctl_wire_old_buffer(req, 2929 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr))); 2930 if (retval) { 2931 NET_EPOCH_EXIT(et); 2932 return (retval); 2933 } 2934 2935 IN_MULTI_LIST_LOCK(); 2936 2937 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2938 if (ifma->ifma_addr->sa_family != AF_INET || 2939 ifma->ifma_protospec == NULL) 2940 continue; 2941 inm = (struct in_multi *)ifma->ifma_protospec; 2942 if (!in_hosteq(inm->inm_addr, group)) 2943 continue; 2944 fmode = inm->inm_st[1].iss_fmode; 2945 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 2946 if (retval != 0) 2947 break; 2948 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 2949 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, 2950 ims->ims_haddr); 2951 /* 2952 * Only copy-out sources which are in-mode. 2953 */ 2954 if (fmode != ims_get_mode(inm, ims, 1)) { 2955 CTR1(KTR_IGMPV3, "%s: skip non-in-mode", 2956 __func__); 2957 continue; 2958 } 2959 src.s_addr = htonl(ims->ims_haddr); 2960 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr)); 2961 if (retval != 0) 2962 break; 2963 } 2964 } 2965 2966 IN_MULTI_LIST_UNLOCK(); 2967 NET_EPOCH_EXIT(et); 2968 2969 return (retval); 2970 } 2971 2972 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3) 2973 2974 static const char *inm_modestrs[] = { 2975 [MCAST_UNDEFINED] = "un", 2976 [MCAST_INCLUDE] = "in", 2977 [MCAST_EXCLUDE] = "ex", 2978 }; 2979 _Static_assert(MCAST_UNDEFINED == 0 && 2980 MCAST_EXCLUDE + 1 == nitems(inm_modestrs), 2981 "inm_modestrs: no longer matches #defines"); 2982 2983 static const char * 2984 inm_mode_str(const int mode) 2985 { 2986 2987 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 2988 return (inm_modestrs[mode]); 2989 return ("??"); 2990 } 2991 2992 static const char *inm_statestrs[] = { 2993 [IGMP_NOT_MEMBER] = "not-member", 2994 [IGMP_SILENT_MEMBER] = "silent", 2995 [IGMP_REPORTING_MEMBER] = "reporting", 2996 [IGMP_IDLE_MEMBER] = "idle", 2997 [IGMP_LAZY_MEMBER] = "lazy", 2998 [IGMP_SLEEPING_MEMBER] = "sleeping", 2999 [IGMP_AWAKENING_MEMBER] = "awakening", 3000 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending", 3001 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending", 3002 [IGMP_LEAVING_MEMBER] = "leaving", 3003 }; 3004 _Static_assert(IGMP_NOT_MEMBER == 0 && 3005 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs), 3006 "inm_statetrs: no longer matches #defines"); 3007 3008 static const char * 3009 inm_state_str(const int state) 3010 { 3011 3012 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER) 3013 return (inm_statestrs[state]); 3014 return ("??"); 3015 } 3016 3017 /* 3018 * Dump an in_multi structure to the console. 3019 */ 3020 void 3021 inm_print(const struct in_multi *inm) 3022 { 3023 int t; 3024 char addrbuf[INET_ADDRSTRLEN]; 3025 3026 if ((ktr_mask & KTR_IGMPV3) == 0) 3027 return; 3028 3029 printf("%s: --- begin inm %p ---\n", __func__, inm); 3030 printf("addr %s ifp %p(%s) ifma %p\n", 3031 inet_ntoa_r(inm->inm_addr, addrbuf), 3032 inm->inm_ifp, 3033 inm->inm_ifp->if_xname, 3034 inm->inm_ifma); 3035 printf("timer %u state %s refcount %u scq.len %u\n", 3036 inm->inm_timer, 3037 inm_state_str(inm->inm_state), 3038 inm->inm_refcount, 3039 inm->inm_scq.mq_len); 3040 printf("igi %p nsrc %lu sctimer %u scrv %u\n", 3041 inm->inm_igi, 3042 inm->inm_nsrc, 3043 inm->inm_sctimer, 3044 inm->inm_scrv); 3045 for (t = 0; t < 2; t++) { 3046 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 3047 inm_mode_str(inm->inm_st[t].iss_fmode), 3048 inm->inm_st[t].iss_asm, 3049 inm->inm_st[t].iss_ex, 3050 inm->inm_st[t].iss_in, 3051 inm->inm_st[t].iss_rec); 3052 } 3053 printf("%s: --- end inm %p ---\n", __func__, inm); 3054 } 3055 3056 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */ 3057 3058 void 3059 inm_print(const struct in_multi *inm) 3060 { 3061 3062 } 3063 3064 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */ 3065 3066 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp); 3067