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