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