1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 2009 Bruce Simpson. 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. The name of the author may not be used to endorse or promote 16 * products derived from this software without specific prior written 17 * permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 */ 31 32 /* 33 * IPv6 multicast socket, group, and socket option processing module. 34 * Normative references: RFC 2292, RFC 3492, RFC 3542, RFC 3678, RFC 3810. 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_inet6.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/gtaskqueue.h> 45 #include <sys/kernel.h> 46 #include <sys/malloc.h> 47 #include <sys/mbuf.h> 48 #include <sys/protosw.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/sysctl.h> 52 #include <sys/priv.h> 53 #include <sys/ktr.h> 54 #include <sys/tree.h> 55 56 #include <net/if.h> 57 #include <net/if_var.h> 58 #include <net/if_dl.h> 59 #include <net/route.h> 60 #include <net/vnet.h> 61 62 63 #include <netinet/in.h> 64 #include <netinet/udp.h> 65 #include <netinet/in_var.h> 66 #include <netinet/ip_var.h> 67 #include <netinet/udp_var.h> 68 #include <netinet6/in6_fib.h> 69 #include <netinet6/in6_var.h> 70 #include <netinet/ip6.h> 71 #include <netinet/icmp6.h> 72 #include <netinet6/ip6_var.h> 73 #include <netinet/in_pcb.h> 74 #include <netinet/tcp_var.h> 75 #include <netinet6/nd6.h> 76 #include <netinet6/mld6_var.h> 77 #include <netinet6/scope6_var.h> 78 79 #ifndef KTR_MLD 80 #define KTR_MLD KTR_INET6 81 #endif 82 83 #ifndef __SOCKUNION_DECLARED 84 union sockunion { 85 struct sockaddr_storage ss; 86 struct sockaddr sa; 87 struct sockaddr_dl sdl; 88 struct sockaddr_in6 sin6; 89 }; 90 typedef union sockunion sockunion_t; 91 #define __SOCKUNION_DECLARED 92 #endif /* __SOCKUNION_DECLARED */ 93 94 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter", 95 "IPv6 multicast PCB-layer source filter"); 96 MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group"); 97 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options"); 98 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource", 99 "IPv6 multicast MLD-layer source filter"); 100 101 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp); 102 103 /* 104 * Locking: 105 * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK. 106 * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however 107 * it can be taken by code in net/if.c also. 108 * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK. 109 * 110 * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly 111 * any need for in6_multi itself to be virtualized -- it is bound to an ifp 112 * anyway no matter what happens. 113 */ 114 struct mtx in6_multi_list_mtx; 115 MTX_SYSINIT(in6_multi_mtx, &in6_multi_list_mtx, "in6_multi_list_mtx", MTX_DEF); 116 117 struct mtx in6_multi_free_mtx; 118 MTX_SYSINIT(in6_multi_free_mtx, &in6_multi_free_mtx, "in6_multi_free_mtx", MTX_DEF); 119 120 struct sx in6_multi_sx; 121 SX_SYSINIT(in6_multi_sx, &in6_multi_sx, "in6_multi_sx"); 122 123 124 125 static void im6f_commit(struct in6_mfilter *); 126 static int im6f_get_source(struct in6_mfilter *imf, 127 const struct sockaddr_in6 *psin, 128 struct in6_msource **); 129 static struct in6_msource * 130 im6f_graft(struct in6_mfilter *, const uint8_t, 131 const struct sockaddr_in6 *); 132 static void im6f_leave(struct in6_mfilter *); 133 static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *); 134 static void im6f_purge(struct in6_mfilter *); 135 static void im6f_rollback(struct in6_mfilter *); 136 static void im6f_reap(struct in6_mfilter *); 137 static int im6o_grow(struct ip6_moptions *); 138 static size_t im6o_match_group(const struct ip6_moptions *, 139 const struct ifnet *, const struct sockaddr *); 140 static struct in6_msource * 141 im6o_match_source(const struct ip6_moptions *, const size_t, 142 const struct sockaddr *); 143 static void im6s_merge(struct ip6_msource *ims, 144 const struct in6_msource *lims, const int rollback); 145 static int in6_getmulti(struct ifnet *, const struct in6_addr *, 146 struct in6_multi **); 147 static int in6m_get_source(struct in6_multi *inm, 148 const struct in6_addr *addr, const int noalloc, 149 struct ip6_msource **pims); 150 #ifdef KTR 151 static int in6m_is_ifp_detached(const struct in6_multi *); 152 #endif 153 static int in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *); 154 static void in6m_purge(struct in6_multi *); 155 static void in6m_reap(struct in6_multi *); 156 static struct ip6_moptions * 157 in6p_findmoptions(struct inpcb *); 158 static int in6p_get_source_filters(struct inpcb *, struct sockopt *); 159 static int in6p_join_group(struct inpcb *, struct sockopt *); 160 static int in6p_leave_group(struct inpcb *, struct sockopt *); 161 static struct ifnet * 162 in6p_lookup_mcast_ifp(const struct inpcb *, 163 const struct sockaddr_in6 *); 164 static int in6p_block_unblock_source(struct inpcb *, struct sockopt *); 165 static int in6p_set_multicast_if(struct inpcb *, struct sockopt *); 166 static int in6p_set_source_filters(struct inpcb *, struct sockopt *); 167 static int sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS); 168 169 SYSCTL_DECL(_net_inet6_ip6); /* XXX Not in any common header. */ 170 171 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0, 172 "IPv6 multicast"); 173 174 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER; 175 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc, 176 CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0, 177 "Max source filters per group"); 178 179 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER; 180 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc, 181 CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0, 182 "Max source filters per socket"); 183 184 /* TODO Virtualize this switch. */ 185 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP; 186 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN, 187 &in6_mcast_loop, 0, "Loopback multicast datagrams by default"); 188 189 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters, 190 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters, 191 "Per-interface stack-wide source filters"); 192 193 int ifma6_restart = 0; 194 #ifdef KTR 195 /* 196 * Inline function which wraps assertions for a valid ifp. 197 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 198 * is detached. 199 */ 200 static int __inline 201 in6m_is_ifp_detached(const struct in6_multi *inm) 202 { 203 struct ifnet *ifp; 204 205 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__)); 206 ifp = inm->in6m_ifma->ifma_ifp; 207 if (ifp != NULL) { 208 /* 209 * Sanity check that network-layer notion of ifp is the 210 * same as that of link-layer. 211 */ 212 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__)); 213 } 214 215 return (ifp == NULL); 216 } 217 #endif 218 219 /* 220 * Initialize an in6_mfilter structure to a known state at t0, t1 221 * with an empty source filter list. 222 */ 223 static __inline void 224 im6f_init(struct in6_mfilter *imf, const int st0, const int st1) 225 { 226 memset(imf, 0, sizeof(struct in6_mfilter)); 227 RB_INIT(&imf->im6f_sources); 228 imf->im6f_st[0] = st0; 229 imf->im6f_st[1] = st1; 230 } 231 232 /* 233 * Resize the ip6_moptions vector to the next power-of-two minus 1. 234 * May be called with locks held; do not sleep. 235 */ 236 static int 237 im6o_grow(struct ip6_moptions *imo) 238 { 239 struct in6_multi **nmships; 240 struct in6_multi **omships; 241 struct in6_mfilter *nmfilters; 242 struct in6_mfilter *omfilters; 243 size_t idx; 244 size_t newmax; 245 size_t oldmax; 246 247 nmships = NULL; 248 nmfilters = NULL; 249 omships = imo->im6o_membership; 250 omfilters = imo->im6o_mfilters; 251 oldmax = imo->im6o_max_memberships; 252 newmax = ((oldmax + 1) * 2) - 1; 253 254 if (newmax <= IPV6_MAX_MEMBERSHIPS) { 255 nmships = (struct in6_multi **)realloc(omships, 256 sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT); 257 nmfilters = (struct in6_mfilter *)realloc(omfilters, 258 sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER, 259 M_NOWAIT); 260 if (nmships != NULL && nmfilters != NULL) { 261 /* Initialize newly allocated source filter heads. */ 262 for (idx = oldmax; idx < newmax; idx++) { 263 im6f_init(&nmfilters[idx], MCAST_UNDEFINED, 264 MCAST_EXCLUDE); 265 } 266 imo->im6o_max_memberships = newmax; 267 imo->im6o_membership = nmships; 268 imo->im6o_mfilters = nmfilters; 269 } 270 } 271 272 if (nmships == NULL || nmfilters == NULL) { 273 if (nmships != NULL) 274 free(nmships, M_IP6MOPTS); 275 if (nmfilters != NULL) 276 free(nmfilters, M_IN6MFILTER); 277 return (ETOOMANYREFS); 278 } 279 280 return (0); 281 } 282 283 /* 284 * Find an IPv6 multicast group entry for this ip6_moptions instance 285 * which matches the specified group, and optionally an interface. 286 * Return its index into the array, or -1 if not found. 287 */ 288 static size_t 289 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp, 290 const struct sockaddr *group) 291 { 292 const struct sockaddr_in6 *gsin6; 293 struct in6_multi **pinm; 294 int idx; 295 int nmships; 296 297 gsin6 = (const struct sockaddr_in6 *)group; 298 299 /* The im6o_membership array may be lazy allocated. */ 300 if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0) 301 return (-1); 302 303 nmships = imo->im6o_num_memberships; 304 pinm = &imo->im6o_membership[0]; 305 for (idx = 0; idx < nmships; idx++, pinm++) { 306 if (*pinm == NULL) 307 continue; 308 if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) && 309 IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr, 310 &gsin6->sin6_addr)) { 311 break; 312 } 313 } 314 if (idx >= nmships) 315 idx = -1; 316 317 return (idx); 318 } 319 320 /* 321 * Find an IPv6 multicast source entry for this imo which matches 322 * the given group index for this socket, and source address. 323 * 324 * XXX TODO: The scope ID, if present in src, is stripped before 325 * any comparison. We SHOULD enforce scope/zone checks where the source 326 * filter entry has a link scope. 327 * 328 * NOTE: This does not check if the entry is in-mode, merely if 329 * it exists, which may not be the desired behaviour. 330 */ 331 static struct in6_msource * 332 im6o_match_source(const struct ip6_moptions *imo, const size_t gidx, 333 const struct sockaddr *src) 334 { 335 struct ip6_msource find; 336 struct in6_mfilter *imf; 337 struct ip6_msource *ims; 338 const sockunion_t *psa; 339 340 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__)); 341 KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships, 342 ("%s: invalid index %d\n", __func__, (int)gidx)); 343 344 /* The im6o_mfilters array may be lazy allocated. */ 345 if (imo->im6o_mfilters == NULL) 346 return (NULL); 347 imf = &imo->im6o_mfilters[gidx]; 348 349 psa = (const sockunion_t *)src; 350 find.im6s_addr = psa->sin6.sin6_addr; 351 in6_clearscope(&find.im6s_addr); /* XXX */ 352 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find); 353 354 return ((struct in6_msource *)ims); 355 } 356 357 /* 358 * Perform filtering for multicast datagrams on a socket by group and source. 359 * 360 * Returns 0 if a datagram should be allowed through, or various error codes 361 * if the socket was not a member of the group, or the source was muted, etc. 362 */ 363 int 364 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp, 365 const struct sockaddr *group, const struct sockaddr *src) 366 { 367 size_t gidx; 368 struct in6_msource *ims; 369 int mode; 370 371 KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 372 373 gidx = im6o_match_group(imo, ifp, group); 374 if (gidx == -1) 375 return (MCAST_NOTGMEMBER); 376 377 /* 378 * Check if the source was included in an (S,G) join. 379 * Allow reception on exclusive memberships by default, 380 * reject reception on inclusive memberships by default. 381 * Exclude source only if an in-mode exclude filter exists. 382 * Include source only if an in-mode include filter exists. 383 * NOTE: We are comparing group state here at MLD t1 (now) 384 * with socket-layer t0 (since last downcall). 385 */ 386 mode = imo->im6o_mfilters[gidx].im6f_st[1]; 387 ims = im6o_match_source(imo, gidx, src); 388 389 if ((ims == NULL && mode == MCAST_INCLUDE) || 390 (ims != NULL && ims->im6sl_st[0] != mode)) 391 return (MCAST_NOTSMEMBER); 392 393 return (MCAST_PASS); 394 } 395 396 /* 397 * Find and return a reference to an in6_multi record for (ifp, group), 398 * and bump its reference count. 399 * If one does not exist, try to allocate it, and update link-layer multicast 400 * filters on ifp to listen for group. 401 * Assumes the IN6_MULTI lock is held across the call. 402 * Return 0 if successful, otherwise return an appropriate error code. 403 */ 404 static int 405 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group, 406 struct in6_multi **pinm) 407 { 408 struct sockaddr_in6 gsin6; 409 struct ifmultiaddr *ifma; 410 struct in6_multi *inm; 411 int error; 412 413 error = 0; 414 415 /* 416 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK; 417 * if_addmulti() takes this mutex itself, so we must drop and 418 * re-acquire around the call. 419 */ 420 IN6_MULTI_LOCK_ASSERT(); 421 IN6_MULTI_LIST_LOCK(); 422 IF_ADDR_WLOCK(ifp); 423 inm = in6m_lookup_locked(ifp, group); 424 if (inm != NULL) { 425 /* 426 * If we already joined this group, just bump the 427 * refcount and return it. 428 */ 429 KASSERT(inm->in6m_refcount >= 1, 430 ("%s: bad refcount %d", __func__, inm->in6m_refcount)); 431 in6m_acquire_locked(inm); 432 *pinm = inm; 433 goto out_locked; 434 } 435 436 memset(&gsin6, 0, sizeof(gsin6)); 437 gsin6.sin6_family = AF_INET6; 438 gsin6.sin6_len = sizeof(struct sockaddr_in6); 439 gsin6.sin6_addr = *group; 440 441 /* 442 * Check if a link-layer group is already associated 443 * with this network-layer group on the given ifnet. 444 */ 445 IN6_MULTI_LIST_UNLOCK(); 446 IF_ADDR_WUNLOCK(ifp); 447 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma); 448 if (error != 0) 449 return (error); 450 IN6_MULTI_LIST_LOCK(); 451 IF_ADDR_WLOCK(ifp); 452 453 /* 454 * If something other than netinet6 is occupying the link-layer 455 * group, print a meaningful error message and back out of 456 * the allocation. 457 * Otherwise, bump the refcount on the existing network-layer 458 * group association and return it. 459 */ 460 if (ifma->ifma_protospec != NULL) { 461 inm = (struct in6_multi *)ifma->ifma_protospec; 462 #ifdef INVARIANTS 463 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 464 __func__)); 465 KASSERT(ifma->ifma_addr->sa_family == AF_INET6, 466 ("%s: ifma not AF_INET6", __func__)); 467 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 468 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp || 469 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group)) 470 panic("%s: ifma %p is inconsistent with %p (%p)", 471 __func__, ifma, inm, group); 472 #endif 473 in6m_acquire_locked(inm); 474 *pinm = inm; 475 goto out_locked; 476 } 477 478 IF_ADDR_WLOCK_ASSERT(ifp); 479 480 /* 481 * A new in6_multi record is needed; allocate and initialize it. 482 * We DO NOT perform an MLD join as the in6_ layer may need to 483 * push an initial source list down to MLD to support SSM. 484 * 485 * The initial source filter state is INCLUDE, {} as per the RFC. 486 * Pending state-changes per group are subject to a bounds check. 487 */ 488 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO); 489 if (inm == NULL) { 490 IN6_MULTI_LIST_UNLOCK(); 491 IF_ADDR_WUNLOCK(ifp); 492 if_delmulti_ifma(ifma); 493 return (ENOMEM); 494 } 495 inm->in6m_addr = *group; 496 inm->in6m_ifp = ifp; 497 inm->in6m_mli = MLD_IFINFO(ifp); 498 inm->in6m_ifma = ifma; 499 inm->in6m_refcount = 1; 500 inm->in6m_state = MLD_NOT_MEMBER; 501 mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES); 502 503 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED; 504 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 505 RB_INIT(&inm->in6m_srcs); 506 507 ifma->ifma_protospec = inm; 508 *pinm = inm; 509 510 out_locked: 511 IN6_MULTI_LIST_UNLOCK(); 512 IF_ADDR_WUNLOCK(ifp); 513 return (error); 514 } 515 516 /* 517 * Drop a reference to an in6_multi record. 518 * 519 * If the refcount drops to 0, free the in6_multi record and 520 * delete the underlying link-layer membership. 521 */ 522 static void 523 in6m_release(struct in6_multi *inm) 524 { 525 struct ifmultiaddr *ifma; 526 struct ifnet *ifp; 527 528 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount); 529 530 MPASS(inm->in6m_refcount == 0); 531 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm); 532 533 ifma = inm->in6m_ifma; 534 ifp = inm->in6m_ifp; 535 MPASS(ifma->ifma_llifma == NULL); 536 537 /* XXX this access is not covered by IF_ADDR_LOCK */ 538 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma); 539 KASSERT(ifma->ifma_protospec == NULL, 540 ("%s: ifma_protospec != NULL", __func__)); 541 542 if (ifp != NULL) { 543 CURVNET_SET(ifp->if_vnet); 544 in6m_purge(inm); 545 free(inm, M_IP6MADDR); 546 if_delmulti_ifma_flags(ifma, 1); 547 CURVNET_RESTORE(); 548 if_rele(ifp); 549 } else { 550 in6m_purge(inm); 551 free(inm, M_IP6MADDR); 552 if_delmulti_ifma_flags(ifma, 1); 553 } 554 } 555 556 static struct grouptask free_gtask; 557 static struct in6_multi_head in6m_free_list; 558 static void in6m_release_task(void *arg __unused); 559 static void in6m_init(void) 560 { 561 SLIST_INIT(&in6m_free_list); 562 taskqgroup_config_gtask_init(NULL, &free_gtask, in6m_release_task, "in6m release task"); 563 } 564 565 #ifdef EARLY_AP_STARTUP 566 SYSINIT(in6m_init, SI_SUB_SMP + 1, SI_ORDER_FIRST, 567 in6m_init, NULL); 568 #else 569 SYSINIT(in6m_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_SECOND, 570 in6m_init, NULL); 571 #endif 572 573 574 void 575 in6m_release_list_deferred(struct in6_multi_head *inmh) 576 { 577 if (SLIST_EMPTY(inmh)) 578 return; 579 mtx_lock(&in6_multi_free_mtx); 580 SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele); 581 mtx_unlock(&in6_multi_free_mtx); 582 GROUPTASK_ENQUEUE(&free_gtask); 583 } 584 585 void 586 in6m_disconnect(struct in6_multi *inm) 587 { 588 struct ifnet *ifp; 589 struct ifaddr *ifa; 590 struct in6_ifaddr *ifa6; 591 struct in6_multi_mship *imm, *imm_tmp; 592 struct ifmultiaddr *ifma, *ll_ifma; 593 594 ifp = inm->in6m_ifp; 595 IF_ADDR_WLOCK_ASSERT(ifp); 596 ifma = inm->in6m_ifma; 597 598 if_ref(ifp); 599 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link); 600 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname); 601 if ((ll_ifma = ifma->ifma_llifma) != NULL) { 602 MPASS(ifma != ll_ifma); 603 ifma->ifma_llifma = NULL; 604 MPASS(ll_ifma->ifma_llifma == NULL); 605 MPASS(ll_ifma->ifma_ifp == ifp); 606 if (--ll_ifma->ifma_refcount == 0) { 607 ifma6_restart = true; 608 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link); 609 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname); 610 if_freemulti(ll_ifma); 611 } 612 } 613 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 614 if (ifa->ifa_addr->sa_family != AF_INET6) 615 continue; 616 ifa6 = (void *)ifa; 617 LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships, 618 i6mm_chain, imm_tmp) { 619 if (inm == imm->i6mm_maddr) { 620 LIST_REMOVE(imm, i6mm_chain); 621 free(imm, M_IP6MADDR); 622 } 623 } 624 } 625 } 626 627 void 628 in6m_release_deferred(struct in6_multi *inm) 629 { 630 struct in6_multi_head tmp; 631 632 IN6_MULTI_LIST_LOCK_ASSERT(); 633 KASSERT(inm->in6m_refcount > 0, ("refcount == %d inm: %p", inm->in6m_refcount, inm)); 634 if (--inm->in6m_refcount == 0) { 635 in6m_disconnect(inm); 636 SLIST_INIT(&tmp); 637 inm->in6m_ifma->ifma_protospec = NULL; 638 MPASS(inm->in6m_ifma->ifma_llifma == NULL); 639 SLIST_INSERT_HEAD(&tmp, inm, in6m_nrele); 640 in6m_release_list_deferred(&tmp); 641 } 642 } 643 644 static void 645 in6m_release_task(void *arg __unused) 646 { 647 struct in6_multi_head in6m_free_tmp; 648 struct in6_multi *inm, *tinm; 649 650 SLIST_INIT(&in6m_free_tmp); 651 mtx_lock(&in6_multi_free_mtx); 652 SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele); 653 mtx_unlock(&in6_multi_free_mtx); 654 IN6_MULTI_LOCK(); 655 SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) { 656 SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele); 657 in6m_release(inm); 658 } 659 IN6_MULTI_UNLOCK(); 660 } 661 662 /* 663 * Clear recorded source entries for a group. 664 * Used by the MLD code. Caller must hold the IN6_MULTI lock. 665 * FIXME: Should reap. 666 */ 667 void 668 in6m_clear_recorded(struct in6_multi *inm) 669 { 670 struct ip6_msource *ims; 671 672 IN6_MULTI_LIST_LOCK_ASSERT(); 673 674 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) { 675 if (ims->im6s_stp) { 676 ims->im6s_stp = 0; 677 --inm->in6m_st[1].iss_rec; 678 } 679 } 680 KASSERT(inm->in6m_st[1].iss_rec == 0, 681 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec)); 682 } 683 684 /* 685 * Record a source as pending for a Source-Group MLDv2 query. 686 * This lives here as it modifies the shared tree. 687 * 688 * inm is the group descriptor. 689 * naddr is the address of the source to record in network-byte order. 690 * 691 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will 692 * lazy-allocate a source node in response to an SG query. 693 * Otherwise, no allocation is performed. This saves some memory 694 * with the trade-off that the source will not be reported to the 695 * router if joined in the window between the query response and 696 * the group actually being joined on the local host. 697 * 698 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed. 699 * This turns off the allocation of a recorded source entry if 700 * the group has not been joined. 701 * 702 * Return 0 if the source didn't exist or was already marked as recorded. 703 * Return 1 if the source was marked as recorded by this function. 704 * Return <0 if any error occurred (negated errno code). 705 */ 706 int 707 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr) 708 { 709 struct ip6_msource find; 710 struct ip6_msource *ims, *nims; 711 712 IN6_MULTI_LIST_LOCK_ASSERT(); 713 714 find.im6s_addr = *addr; 715 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find); 716 if (ims && ims->im6s_stp) 717 return (0); 718 if (ims == NULL) { 719 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc) 720 return (-ENOSPC); 721 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE, 722 M_NOWAIT | M_ZERO); 723 if (nims == NULL) 724 return (-ENOMEM); 725 nims->im6s_addr = find.im6s_addr; 726 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims); 727 ++inm->in6m_nsrc; 728 ims = nims; 729 } 730 731 /* 732 * Mark the source as recorded and update the recorded 733 * source count. 734 */ 735 ++ims->im6s_stp; 736 ++inm->in6m_st[1].iss_rec; 737 738 return (1); 739 } 740 741 /* 742 * Return a pointer to an in6_msource owned by an in6_mfilter, 743 * given its source address. 744 * Lazy-allocate if needed. If this is a new entry its filter state is 745 * undefined at t0. 746 * 747 * imf is the filter set being modified. 748 * addr is the source address. 749 * 750 * SMPng: May be called with locks held; malloc must not block. 751 */ 752 static int 753 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin, 754 struct in6_msource **plims) 755 { 756 struct ip6_msource find; 757 struct ip6_msource *ims, *nims; 758 struct in6_msource *lims; 759 int error; 760 761 error = 0; 762 ims = NULL; 763 lims = NULL; 764 765 find.im6s_addr = psin->sin6_addr; 766 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find); 767 lims = (struct in6_msource *)ims; 768 if (lims == NULL) { 769 if (imf->im6f_nsrc == in6_mcast_maxsocksrc) 770 return (ENOSPC); 771 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER, 772 M_NOWAIT | M_ZERO); 773 if (nims == NULL) 774 return (ENOMEM); 775 lims = (struct in6_msource *)nims; 776 lims->im6s_addr = find.im6s_addr; 777 lims->im6sl_st[0] = MCAST_UNDEFINED; 778 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims); 779 ++imf->im6f_nsrc; 780 } 781 782 *plims = lims; 783 784 return (error); 785 } 786 787 /* 788 * Graft a source entry into an existing socket-layer filter set, 789 * maintaining any required invariants and checking allocations. 790 * 791 * The source is marked as being in the new filter mode at t1. 792 * 793 * Return the pointer to the new node, otherwise return NULL. 794 */ 795 static struct in6_msource * 796 im6f_graft(struct in6_mfilter *imf, const uint8_t st1, 797 const struct sockaddr_in6 *psin) 798 { 799 struct ip6_msource *nims; 800 struct in6_msource *lims; 801 802 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER, 803 M_NOWAIT | M_ZERO); 804 if (nims == NULL) 805 return (NULL); 806 lims = (struct in6_msource *)nims; 807 lims->im6s_addr = psin->sin6_addr; 808 lims->im6sl_st[0] = MCAST_UNDEFINED; 809 lims->im6sl_st[1] = st1; 810 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims); 811 ++imf->im6f_nsrc; 812 813 return (lims); 814 } 815 816 /* 817 * Prune a source entry from an existing socket-layer filter set, 818 * maintaining any required invariants and checking allocations. 819 * 820 * The source is marked as being left at t1, it is not freed. 821 * 822 * Return 0 if no error occurred, otherwise return an errno value. 823 */ 824 static int 825 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin) 826 { 827 struct ip6_msource find; 828 struct ip6_msource *ims; 829 struct in6_msource *lims; 830 831 find.im6s_addr = psin->sin6_addr; 832 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find); 833 if (ims == NULL) 834 return (ENOENT); 835 lims = (struct in6_msource *)ims; 836 lims->im6sl_st[1] = MCAST_UNDEFINED; 837 return (0); 838 } 839 840 /* 841 * Revert socket-layer filter set deltas at t1 to t0 state. 842 */ 843 static void 844 im6f_rollback(struct in6_mfilter *imf) 845 { 846 struct ip6_msource *ims, *tims; 847 struct in6_msource *lims; 848 849 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) { 850 lims = (struct in6_msource *)ims; 851 if (lims->im6sl_st[0] == lims->im6sl_st[1]) { 852 /* no change at t1 */ 853 continue; 854 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) { 855 /* revert change to existing source at t1 */ 856 lims->im6sl_st[1] = lims->im6sl_st[0]; 857 } else { 858 /* revert source added t1 */ 859 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims); 860 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims); 861 free(ims, M_IN6MFILTER); 862 imf->im6f_nsrc--; 863 } 864 } 865 imf->im6f_st[1] = imf->im6f_st[0]; 866 } 867 868 /* 869 * Mark socket-layer filter set as INCLUDE {} at t1. 870 */ 871 static void 872 im6f_leave(struct in6_mfilter *imf) 873 { 874 struct ip6_msource *ims; 875 struct in6_msource *lims; 876 877 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) { 878 lims = (struct in6_msource *)ims; 879 lims->im6sl_st[1] = MCAST_UNDEFINED; 880 } 881 imf->im6f_st[1] = MCAST_INCLUDE; 882 } 883 884 /* 885 * Mark socket-layer filter set deltas as committed. 886 */ 887 static void 888 im6f_commit(struct in6_mfilter *imf) 889 { 890 struct ip6_msource *ims; 891 struct in6_msource *lims; 892 893 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) { 894 lims = (struct in6_msource *)ims; 895 lims->im6sl_st[0] = lims->im6sl_st[1]; 896 } 897 imf->im6f_st[0] = imf->im6f_st[1]; 898 } 899 900 /* 901 * Reap unreferenced sources from socket-layer filter set. 902 */ 903 static void 904 im6f_reap(struct in6_mfilter *imf) 905 { 906 struct ip6_msource *ims, *tims; 907 struct in6_msource *lims; 908 909 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) { 910 lims = (struct in6_msource *)ims; 911 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) && 912 (lims->im6sl_st[1] == MCAST_UNDEFINED)) { 913 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims); 914 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims); 915 free(ims, M_IN6MFILTER); 916 imf->im6f_nsrc--; 917 } 918 } 919 } 920 921 /* 922 * Purge socket-layer filter set. 923 */ 924 static void 925 im6f_purge(struct in6_mfilter *imf) 926 { 927 struct ip6_msource *ims, *tims; 928 929 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) { 930 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims); 931 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims); 932 free(ims, M_IN6MFILTER); 933 imf->im6f_nsrc--; 934 } 935 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED; 936 KASSERT(RB_EMPTY(&imf->im6f_sources), 937 ("%s: im6f_sources not empty", __func__)); 938 } 939 940 /* 941 * Look up a source filter entry for a multicast group. 942 * 943 * inm is the group descriptor to work with. 944 * addr is the IPv6 address to look up. 945 * noalloc may be non-zero to suppress allocation of sources. 946 * *pims will be set to the address of the retrieved or allocated source. 947 * 948 * SMPng: NOTE: may be called with locks held. 949 * Return 0 if successful, otherwise return a non-zero error code. 950 */ 951 static int 952 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr, 953 const int noalloc, struct ip6_msource **pims) 954 { 955 struct ip6_msource find; 956 struct ip6_msource *ims, *nims; 957 #ifdef KTR 958 char ip6tbuf[INET6_ADDRSTRLEN]; 959 #endif 960 961 find.im6s_addr = *addr; 962 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find); 963 if (ims == NULL && !noalloc) { 964 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc) 965 return (ENOSPC); 966 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE, 967 M_NOWAIT | M_ZERO); 968 if (nims == NULL) 969 return (ENOMEM); 970 nims->im6s_addr = *addr; 971 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims); 972 ++inm->in6m_nsrc; 973 ims = nims; 974 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__, 975 ip6_sprintf(ip6tbuf, addr), ims); 976 } 977 978 *pims = ims; 979 return (0); 980 } 981 982 /* 983 * Merge socket-layer source into MLD-layer source. 984 * If rollback is non-zero, perform the inverse of the merge. 985 */ 986 static void 987 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims, 988 const int rollback) 989 { 990 int n = rollback ? -1 : 1; 991 #ifdef KTR 992 char ip6tbuf[INET6_ADDRSTRLEN]; 993 994 ip6_sprintf(ip6tbuf, &lims->im6s_addr); 995 #endif 996 997 if (lims->im6sl_st[0] == MCAST_EXCLUDE) { 998 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf); 999 ims->im6s_st[1].ex -= n; 1000 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) { 1001 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf); 1002 ims->im6s_st[1].in -= n; 1003 } 1004 1005 if (lims->im6sl_st[1] == MCAST_EXCLUDE) { 1006 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf); 1007 ims->im6s_st[1].ex += n; 1008 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) { 1009 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf); 1010 ims->im6s_st[1].in += n; 1011 } 1012 } 1013 1014 /* 1015 * Atomically update the global in6_multi state, when a membership's 1016 * filter list is being updated in any way. 1017 * 1018 * imf is the per-inpcb-membership group filter pointer. 1019 * A fake imf may be passed for in-kernel consumers. 1020 * 1021 * XXX This is a candidate for a set-symmetric-difference style loop 1022 * which would eliminate the repeated lookup from root of ims nodes, 1023 * as they share the same key space. 1024 * 1025 * If any error occurred this function will back out of refcounts 1026 * and return a non-zero value. 1027 */ 1028 static int 1029 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf) 1030 { 1031 struct ip6_msource *ims, *nims; 1032 struct in6_msource *lims; 1033 int schanged, error; 1034 int nsrc0, nsrc1; 1035 1036 schanged = 0; 1037 error = 0; 1038 nsrc1 = nsrc0 = 0; 1039 IN6_MULTI_LIST_LOCK_ASSERT(); 1040 1041 /* 1042 * Update the source filters first, as this may fail. 1043 * Maintain count of in-mode filters at t0, t1. These are 1044 * used to work out if we transition into ASM mode or not. 1045 * Maintain a count of source filters whose state was 1046 * actually modified by this operation. 1047 */ 1048 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) { 1049 lims = (struct in6_msource *)ims; 1050 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++; 1051 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++; 1052 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue; 1053 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims); 1054 ++schanged; 1055 if (error) 1056 break; 1057 im6s_merge(nims, lims, 0); 1058 } 1059 if (error) { 1060 struct ip6_msource *bims; 1061 1062 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) { 1063 lims = (struct in6_msource *)ims; 1064 if (lims->im6sl_st[0] == lims->im6sl_st[1]) 1065 continue; 1066 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims); 1067 if (bims == NULL) 1068 continue; 1069 im6s_merge(bims, lims, 1); 1070 } 1071 goto out_reap; 1072 } 1073 1074 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1", 1075 __func__, nsrc0, nsrc1); 1076 1077 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 1078 if (imf->im6f_st[0] == imf->im6f_st[1] && 1079 imf->im6f_st[1] == MCAST_INCLUDE) { 1080 if (nsrc1 == 0) { 1081 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__); 1082 --inm->in6m_st[1].iss_in; 1083 } 1084 } 1085 1086 /* Handle filter mode transition on socket. */ 1087 if (imf->im6f_st[0] != imf->im6f_st[1]) { 1088 CTR3(KTR_MLD, "%s: imf transition %d to %d", 1089 __func__, imf->im6f_st[0], imf->im6f_st[1]); 1090 1091 if (imf->im6f_st[0] == MCAST_EXCLUDE) { 1092 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__); 1093 --inm->in6m_st[1].iss_ex; 1094 } else if (imf->im6f_st[0] == MCAST_INCLUDE) { 1095 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__); 1096 --inm->in6m_st[1].iss_in; 1097 } 1098 1099 if (imf->im6f_st[1] == MCAST_EXCLUDE) { 1100 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__); 1101 inm->in6m_st[1].iss_ex++; 1102 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) { 1103 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__); 1104 inm->in6m_st[1].iss_in++; 1105 } 1106 } 1107 1108 /* 1109 * Track inm filter state in terms of listener counts. 1110 * If there are any exclusive listeners, stack-wide 1111 * membership is exclusive. 1112 * Otherwise, if only inclusive listeners, stack-wide is inclusive. 1113 * If no listeners remain, state is undefined at t1, 1114 * and the MLD lifecycle for this group should finish. 1115 */ 1116 if (inm->in6m_st[1].iss_ex > 0) { 1117 CTR1(KTR_MLD, "%s: transition to EX", __func__); 1118 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE; 1119 } else if (inm->in6m_st[1].iss_in > 0) { 1120 CTR1(KTR_MLD, "%s: transition to IN", __func__); 1121 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE; 1122 } else { 1123 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__); 1124 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED; 1125 } 1126 1127 /* Decrement ASM listener count on transition out of ASM mode. */ 1128 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1129 if ((imf->im6f_st[1] != MCAST_EXCLUDE) || 1130 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) { 1131 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__); 1132 --inm->in6m_st[1].iss_asm; 1133 } 1134 } 1135 1136 /* Increment ASM listener count on transition to ASM mode. */ 1137 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1138 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__); 1139 inm->in6m_st[1].iss_asm++; 1140 } 1141 1142 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm); 1143 in6m_print(inm); 1144 1145 out_reap: 1146 if (schanged > 0) { 1147 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__); 1148 in6m_reap(inm); 1149 } 1150 return (error); 1151 } 1152 1153 /* 1154 * Mark an in6_multi's filter set deltas as committed. 1155 * Called by MLD after a state change has been enqueued. 1156 */ 1157 void 1158 in6m_commit(struct in6_multi *inm) 1159 { 1160 struct ip6_msource *ims; 1161 1162 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm); 1163 CTR1(KTR_MLD, "%s: pre commit:", __func__); 1164 in6m_print(inm); 1165 1166 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) { 1167 ims->im6s_st[0] = ims->im6s_st[1]; 1168 } 1169 inm->in6m_st[0] = inm->in6m_st[1]; 1170 } 1171 1172 /* 1173 * Reap unreferenced nodes from an in6_multi's filter set. 1174 */ 1175 static void 1176 in6m_reap(struct in6_multi *inm) 1177 { 1178 struct ip6_msource *ims, *tims; 1179 1180 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) { 1181 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 || 1182 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 || 1183 ims->im6s_stp != 0) 1184 continue; 1185 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims); 1186 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims); 1187 free(ims, M_IP6MSOURCE); 1188 inm->in6m_nsrc--; 1189 } 1190 } 1191 1192 /* 1193 * Purge all source nodes from an in6_multi's filter set. 1194 */ 1195 static void 1196 in6m_purge(struct in6_multi *inm) 1197 { 1198 struct ip6_msource *ims, *tims; 1199 1200 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) { 1201 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims); 1202 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims); 1203 free(ims, M_IP6MSOURCE); 1204 inm->in6m_nsrc--; 1205 } 1206 /* Free state-change requests that might be queued. */ 1207 mbufq_drain(&inm->in6m_scq); 1208 } 1209 1210 /* 1211 * Join a multicast address w/o sources. 1212 * KAME compatibility entry point. 1213 * 1214 * SMPng: Assume no mc locks held by caller. 1215 */ 1216 int 1217 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr, 1218 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm, 1219 const int delay) 1220 { 1221 int error; 1222 1223 IN6_MULTI_LOCK(); 1224 error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay); 1225 IN6_MULTI_UNLOCK(); 1226 return (error); 1227 } 1228 1229 /* 1230 * Join a multicast group; real entry point. 1231 * 1232 * Only preserves atomicity at inm level. 1233 * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1234 * 1235 * If the MLD downcall fails, the group is not joined, and an error 1236 * code is returned. 1237 */ 1238 int 1239 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr, 1240 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm, 1241 const int delay) 1242 { 1243 struct in6_mfilter timf; 1244 struct in6_multi *inm; 1245 struct ifmultiaddr *ifma; 1246 int error; 1247 #ifdef KTR 1248 char ip6tbuf[INET6_ADDRSTRLEN]; 1249 #endif 1250 1251 #ifdef INVARIANTS 1252 /* 1253 * Sanity: Check scope zone ID was set for ifp, if and 1254 * only if group is scoped to an interface. 1255 */ 1256 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr), 1257 ("%s: not a multicast address", __func__)); 1258 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) || 1259 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) { 1260 KASSERT(mcaddr->s6_addr16[1] != 0, 1261 ("%s: scope zone ID not set", __func__)); 1262 } 1263 #endif 1264 1265 IN6_MULTI_LOCK_ASSERT(); 1266 IN6_MULTI_LIST_UNLOCK_ASSERT(); 1267 1268 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__, 1269 ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp)); 1270 1271 error = 0; 1272 inm = NULL; 1273 1274 /* 1275 * If no imf was specified (i.e. kernel consumer), 1276 * fake one up and assume it is an ASM join. 1277 */ 1278 if (imf == NULL) { 1279 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1280 imf = &timf; 1281 } 1282 error = in6_getmulti(ifp, mcaddr, &inm); 1283 if (error) { 1284 CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__); 1285 return (error); 1286 } 1287 1288 IN6_MULTI_LIST_LOCK(); 1289 CTR1(KTR_MLD, "%s: merge inm state", __func__); 1290 error = in6m_merge(inm, imf); 1291 if (error) { 1292 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__); 1293 goto out_in6m_release; 1294 } 1295 1296 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 1297 error = mld_change_state(inm, delay); 1298 if (error) { 1299 CTR1(KTR_MLD, "%s: failed to update source", __func__); 1300 goto out_in6m_release; 1301 } 1302 1303 out_in6m_release: 1304 if (error) { 1305 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm); 1306 IF_ADDR_RLOCK(ifp); 1307 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 1308 if (ifma->ifma_protospec == inm) { 1309 ifma->ifma_protospec = NULL; 1310 break; 1311 } 1312 } 1313 in6m_release_deferred(inm); 1314 IF_ADDR_RUNLOCK(ifp); 1315 } else { 1316 *pinm = inm; 1317 } 1318 IN6_MULTI_LIST_UNLOCK(); 1319 return (error); 1320 } 1321 1322 /* 1323 * Leave a multicast group; unlocked entry point. 1324 */ 1325 int 1326 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf) 1327 { 1328 int error; 1329 1330 IN6_MULTI_LOCK(); 1331 error = in6_leavegroup_locked(inm, imf); 1332 IN6_MULTI_UNLOCK(); 1333 return (error); 1334 } 1335 1336 /* 1337 * Leave a multicast group; real entry point. 1338 * All source filters will be expunged. 1339 * 1340 * Only preserves atomicity at inm level. 1341 * 1342 * Holding the write lock for the INP which contains imf 1343 * is highly advisable. We can't assert for it as imf does not 1344 * contain a back-pointer to the owning inp. 1345 * 1346 * Note: This is not the same as in6m_release(*) as this function also 1347 * makes a state change downcall into MLD. 1348 */ 1349 int 1350 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf) 1351 { 1352 struct in6_mfilter timf; 1353 struct ifnet *ifp; 1354 int error; 1355 #ifdef KTR 1356 char ip6tbuf[INET6_ADDRSTRLEN]; 1357 #endif 1358 1359 error = 0; 1360 1361 IN6_MULTI_LOCK_ASSERT(); 1362 1363 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__, 1364 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr), 1365 (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)), 1366 imf); 1367 1368 /* 1369 * If no imf was specified (i.e. kernel consumer), 1370 * fake one up and assume it is an ASM join. 1371 */ 1372 if (imf == NULL) { 1373 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1374 imf = &timf; 1375 } 1376 1377 /* 1378 * Begin state merge transaction at MLD layer. 1379 * 1380 * As this particular invocation should not cause any memory 1381 * to be allocated, and there is no opportunity to roll back 1382 * the transaction, it MUST NOT fail. 1383 */ 1384 1385 ifp = inm->in6m_ifp; 1386 IN6_MULTI_LIST_LOCK(); 1387 CTR1(KTR_MLD, "%s: merge inm state", __func__); 1388 error = in6m_merge(inm, imf); 1389 KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1390 1391 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 1392 error = mld_change_state(inm, 0); 1393 if (error) 1394 CTR1(KTR_MLD, "%s: failed mld downcall", __func__); 1395 1396 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm); 1397 if (ifp) 1398 IF_ADDR_WLOCK(ifp); 1399 in6m_release_deferred(inm); 1400 if (ifp) 1401 IF_ADDR_WUNLOCK(ifp); 1402 IN6_MULTI_LIST_UNLOCK(); 1403 1404 return (error); 1405 } 1406 1407 1408 /* 1409 * Block or unblock an ASM multicast source on an inpcb. 1410 * This implements the delta-based API described in RFC 3678. 1411 * 1412 * The delta-based API applies only to exclusive-mode memberships. 1413 * An MLD downcall will be performed. 1414 * 1415 * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1416 * 1417 * Return 0 if successful, otherwise return an appropriate error code. 1418 */ 1419 static int 1420 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1421 { 1422 struct group_source_req gsr; 1423 sockunion_t *gsa, *ssa; 1424 struct ifnet *ifp; 1425 struct in6_mfilter *imf; 1426 struct ip6_moptions *imo; 1427 struct in6_msource *ims; 1428 struct in6_multi *inm; 1429 size_t idx; 1430 uint16_t fmode; 1431 int error, doblock; 1432 #ifdef KTR 1433 char ip6tbuf[INET6_ADDRSTRLEN]; 1434 #endif 1435 1436 ifp = NULL; 1437 error = 0; 1438 doblock = 0; 1439 1440 memset(&gsr, 0, sizeof(struct group_source_req)); 1441 gsa = (sockunion_t *)&gsr.gsr_group; 1442 ssa = (sockunion_t *)&gsr.gsr_source; 1443 1444 switch (sopt->sopt_name) { 1445 case MCAST_BLOCK_SOURCE: 1446 case MCAST_UNBLOCK_SOURCE: 1447 error = sooptcopyin(sopt, &gsr, 1448 sizeof(struct group_source_req), 1449 sizeof(struct group_source_req)); 1450 if (error) 1451 return (error); 1452 1453 if (gsa->sin6.sin6_family != AF_INET6 || 1454 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 1455 return (EINVAL); 1456 1457 if (ssa->sin6.sin6_family != AF_INET6 || 1458 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 1459 return (EINVAL); 1460 1461 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1462 return (EADDRNOTAVAIL); 1463 1464 ifp = ifnet_byindex(gsr.gsr_interface); 1465 1466 if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1467 doblock = 1; 1468 break; 1469 1470 default: 1471 CTR2(KTR_MLD, "%s: unknown sopt_name %d", 1472 __func__, sopt->sopt_name); 1473 return (EOPNOTSUPP); 1474 break; 1475 } 1476 1477 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 1478 return (EINVAL); 1479 1480 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 1481 1482 /* 1483 * Check if we are actually a member of this group. 1484 */ 1485 imo = in6p_findmoptions(inp); 1486 idx = im6o_match_group(imo, ifp, &gsa->sa); 1487 if (idx == -1 || imo->im6o_mfilters == NULL) { 1488 error = EADDRNOTAVAIL; 1489 goto out_in6p_locked; 1490 } 1491 1492 KASSERT(imo->im6o_mfilters != NULL, 1493 ("%s: im6o_mfilters not allocated", __func__)); 1494 imf = &imo->im6o_mfilters[idx]; 1495 inm = imo->im6o_membership[idx]; 1496 1497 /* 1498 * Attempting to use the delta-based API on an 1499 * non exclusive-mode membership is an error. 1500 */ 1501 fmode = imf->im6f_st[0]; 1502 if (fmode != MCAST_EXCLUDE) { 1503 error = EINVAL; 1504 goto out_in6p_locked; 1505 } 1506 1507 /* 1508 * Deal with error cases up-front: 1509 * Asked to block, but already blocked; or 1510 * Asked to unblock, but nothing to unblock. 1511 * If adding a new block entry, allocate it. 1512 */ 1513 ims = im6o_match_source(imo, idx, &ssa->sa); 1514 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1515 CTR3(KTR_MLD, "%s: source %s %spresent", __func__, 1516 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr), 1517 doblock ? "" : "not "); 1518 error = EADDRNOTAVAIL; 1519 goto out_in6p_locked; 1520 } 1521 1522 INP_WLOCK_ASSERT(inp); 1523 1524 /* 1525 * Begin state merge transaction at socket layer. 1526 */ 1527 if (doblock) { 1528 CTR2(KTR_MLD, "%s: %s source", __func__, "block"); 1529 ims = im6f_graft(imf, fmode, &ssa->sin6); 1530 if (ims == NULL) 1531 error = ENOMEM; 1532 } else { 1533 CTR2(KTR_MLD, "%s: %s source", __func__, "allow"); 1534 error = im6f_prune(imf, &ssa->sin6); 1535 } 1536 1537 if (error) { 1538 CTR1(KTR_MLD, "%s: merge imf state failed", __func__); 1539 goto out_im6f_rollback; 1540 } 1541 1542 /* 1543 * Begin state merge transaction at MLD layer. 1544 */ 1545 IN6_MULTI_LIST_LOCK(); 1546 CTR1(KTR_MLD, "%s: merge inm state", __func__); 1547 error = in6m_merge(inm, imf); 1548 if (error) 1549 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__); 1550 else { 1551 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 1552 error = mld_change_state(inm, 0); 1553 if (error) 1554 CTR1(KTR_MLD, "%s: failed mld downcall", __func__); 1555 } 1556 1557 IN6_MULTI_LIST_UNLOCK(); 1558 1559 out_im6f_rollback: 1560 if (error) 1561 im6f_rollback(imf); 1562 else 1563 im6f_commit(imf); 1564 1565 im6f_reap(imf); 1566 1567 out_in6p_locked: 1568 INP_WUNLOCK(inp); 1569 return (error); 1570 } 1571 1572 /* 1573 * Given an inpcb, return its multicast options structure pointer. Accepts 1574 * an unlocked inpcb pointer, but will return it locked. May sleep. 1575 * 1576 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1577 * SMPng: NOTE: Returns with the INP write lock held. 1578 */ 1579 static struct ip6_moptions * 1580 in6p_findmoptions(struct inpcb *inp) 1581 { 1582 struct ip6_moptions *imo; 1583 struct in6_multi **immp; 1584 struct in6_mfilter *imfp; 1585 size_t idx; 1586 1587 INP_WLOCK(inp); 1588 if (inp->in6p_moptions != NULL) 1589 return (inp->in6p_moptions); 1590 1591 INP_WUNLOCK(inp); 1592 1593 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK); 1594 immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS, 1595 M_WAITOK | M_ZERO); 1596 imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS, 1597 M_IN6MFILTER, M_WAITOK); 1598 1599 imo->im6o_multicast_ifp = NULL; 1600 imo->im6o_multicast_hlim = V_ip6_defmcasthlim; 1601 imo->im6o_multicast_loop = in6_mcast_loop; 1602 imo->im6o_num_memberships = 0; 1603 imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS; 1604 imo->im6o_membership = immp; 1605 1606 /* Initialize per-group source filters. */ 1607 for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++) 1608 im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE); 1609 imo->im6o_mfilters = imfp; 1610 1611 INP_WLOCK(inp); 1612 if (inp->in6p_moptions != NULL) { 1613 free(imfp, M_IN6MFILTER); 1614 free(immp, M_IP6MOPTS); 1615 free(imo, M_IP6MOPTS); 1616 return (inp->in6p_moptions); 1617 } 1618 inp->in6p_moptions = imo; 1619 return (imo); 1620 } 1621 1622 /* 1623 * Discard the IPv6 multicast options (and source filters). 1624 * 1625 * SMPng: NOTE: assumes INP write lock is held. 1626 * 1627 * XXX can all be safely deferred to epoch_call 1628 * 1629 */ 1630 1631 static void 1632 inp_gcmoptions(epoch_context_t ctx) 1633 { 1634 struct ip6_moptions *imo; 1635 struct in6_mfilter *imf; 1636 struct in6_multi *inm; 1637 struct ifnet *ifp; 1638 size_t idx, nmships; 1639 1640 imo = __containerof(ctx, struct ip6_moptions, imo6_epoch_ctx); 1641 1642 nmships = imo->im6o_num_memberships; 1643 for (idx = 0; idx < nmships; ++idx) { 1644 imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL; 1645 if (imf) 1646 im6f_leave(imf); 1647 inm = imo->im6o_membership[idx]; 1648 ifp = inm->in6m_ifp; 1649 if (ifp != NULL) { 1650 CURVNET_SET(ifp->if_vnet); 1651 (void)in6_leavegroup(inm, imf); 1652 CURVNET_RESTORE(); 1653 } else { 1654 (void)in6_leavegroup(inm, imf); 1655 } 1656 if (imf) 1657 im6f_purge(imf); 1658 } 1659 1660 if (imo->im6o_mfilters) 1661 free(imo->im6o_mfilters, M_IN6MFILTER); 1662 free(imo->im6o_membership, M_IP6MOPTS); 1663 free(imo, M_IP6MOPTS); 1664 } 1665 1666 void 1667 ip6_freemoptions(struct ip6_moptions *imo) 1668 { 1669 if (imo == NULL) 1670 return; 1671 epoch_call(net_epoch_preempt, &imo->imo6_epoch_ctx, inp_gcmoptions); 1672 } 1673 1674 /* 1675 * Atomically get source filters on a socket for an IPv6 multicast group. 1676 * Called with INP lock held; returns with lock released. 1677 */ 1678 static int 1679 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1680 { 1681 struct __msfilterreq msfr; 1682 sockunion_t *gsa; 1683 struct ifnet *ifp; 1684 struct ip6_moptions *imo; 1685 struct in6_mfilter *imf; 1686 struct ip6_msource *ims; 1687 struct in6_msource *lims; 1688 struct sockaddr_in6 *psin; 1689 struct sockaddr_storage *ptss; 1690 struct sockaddr_storage *tss; 1691 int error; 1692 size_t idx, nsrcs, ncsrcs; 1693 1694 INP_WLOCK_ASSERT(inp); 1695 1696 imo = inp->in6p_moptions; 1697 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__)); 1698 1699 INP_WUNLOCK(inp); 1700 1701 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1702 sizeof(struct __msfilterreq)); 1703 if (error) 1704 return (error); 1705 1706 if (msfr.msfr_group.ss_family != AF_INET6 || 1707 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6)) 1708 return (EINVAL); 1709 1710 gsa = (sockunion_t *)&msfr.msfr_group; 1711 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 1712 return (EINVAL); 1713 1714 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 1715 return (EADDRNOTAVAIL); 1716 ifp = ifnet_byindex(msfr.msfr_ifindex); 1717 if (ifp == NULL) 1718 return (EADDRNOTAVAIL); 1719 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 1720 1721 INP_WLOCK(inp); 1722 1723 /* 1724 * Lookup group on the socket. 1725 */ 1726 idx = im6o_match_group(imo, ifp, &gsa->sa); 1727 if (idx == -1 || imo->im6o_mfilters == NULL) { 1728 INP_WUNLOCK(inp); 1729 return (EADDRNOTAVAIL); 1730 } 1731 imf = &imo->im6o_mfilters[idx]; 1732 1733 /* 1734 * Ignore memberships which are in limbo. 1735 */ 1736 if (imf->im6f_st[1] == MCAST_UNDEFINED) { 1737 INP_WUNLOCK(inp); 1738 return (EAGAIN); 1739 } 1740 msfr.msfr_fmode = imf->im6f_st[1]; 1741 1742 /* 1743 * If the user specified a buffer, copy out the source filter 1744 * entries to userland gracefully. 1745 * We only copy out the number of entries which userland 1746 * has asked for, but we always tell userland how big the 1747 * buffer really needs to be. 1748 */ 1749 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc) 1750 msfr.msfr_nsrcs = in6_mcast_maxsocksrc; 1751 tss = NULL; 1752 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1753 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1754 M_TEMP, M_NOWAIT | M_ZERO); 1755 if (tss == NULL) { 1756 INP_WUNLOCK(inp); 1757 return (ENOBUFS); 1758 } 1759 } 1760 1761 /* 1762 * Count number of sources in-mode at t0. 1763 * If buffer space exists and remains, copy out source entries. 1764 */ 1765 nsrcs = msfr.msfr_nsrcs; 1766 ncsrcs = 0; 1767 ptss = tss; 1768 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) { 1769 lims = (struct in6_msource *)ims; 1770 if (lims->im6sl_st[0] == MCAST_UNDEFINED || 1771 lims->im6sl_st[0] != imf->im6f_st[0]) 1772 continue; 1773 ++ncsrcs; 1774 if (tss != NULL && nsrcs > 0) { 1775 psin = (struct sockaddr_in6 *)ptss; 1776 psin->sin6_family = AF_INET6; 1777 psin->sin6_len = sizeof(struct sockaddr_in6); 1778 psin->sin6_addr = lims->im6s_addr; 1779 psin->sin6_port = 0; 1780 --nsrcs; 1781 ++ptss; 1782 } 1783 } 1784 1785 INP_WUNLOCK(inp); 1786 1787 if (tss != NULL) { 1788 error = copyout(tss, msfr.msfr_srcs, 1789 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1790 free(tss, M_TEMP); 1791 if (error) 1792 return (error); 1793 } 1794 1795 msfr.msfr_nsrcs = ncsrcs; 1796 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1797 1798 return (error); 1799 } 1800 1801 /* 1802 * Return the IP multicast options in response to user getsockopt(). 1803 */ 1804 int 1805 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1806 { 1807 struct ip6_moptions *im6o; 1808 int error; 1809 u_int optval; 1810 1811 INP_WLOCK(inp); 1812 im6o = inp->in6p_moptions; 1813 /* 1814 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 1815 * or is a divert socket, reject it. 1816 */ 1817 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 1818 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1819 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) { 1820 INP_WUNLOCK(inp); 1821 return (EOPNOTSUPP); 1822 } 1823 1824 error = 0; 1825 switch (sopt->sopt_name) { 1826 case IPV6_MULTICAST_IF: 1827 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) { 1828 optval = 0; 1829 } else { 1830 optval = im6o->im6o_multicast_ifp->if_index; 1831 } 1832 INP_WUNLOCK(inp); 1833 error = sooptcopyout(sopt, &optval, sizeof(u_int)); 1834 break; 1835 1836 case IPV6_MULTICAST_HOPS: 1837 if (im6o == NULL) 1838 optval = V_ip6_defmcasthlim; 1839 else 1840 optval = im6o->im6o_multicast_hlim; 1841 INP_WUNLOCK(inp); 1842 error = sooptcopyout(sopt, &optval, sizeof(u_int)); 1843 break; 1844 1845 case IPV6_MULTICAST_LOOP: 1846 if (im6o == NULL) 1847 optval = in6_mcast_loop; /* XXX VIMAGE */ 1848 else 1849 optval = im6o->im6o_multicast_loop; 1850 INP_WUNLOCK(inp); 1851 error = sooptcopyout(sopt, &optval, sizeof(u_int)); 1852 break; 1853 1854 case IPV6_MSFILTER: 1855 if (im6o == NULL) { 1856 error = EADDRNOTAVAIL; 1857 INP_WUNLOCK(inp); 1858 } else { 1859 error = in6p_get_source_filters(inp, sopt); 1860 } 1861 break; 1862 1863 default: 1864 INP_WUNLOCK(inp); 1865 error = ENOPROTOOPT; 1866 break; 1867 } 1868 1869 INP_UNLOCK_ASSERT(inp); 1870 1871 return (error); 1872 } 1873 1874 /* 1875 * Look up the ifnet to use for a multicast group membership, 1876 * given the address of an IPv6 group. 1877 * 1878 * This routine exists to support legacy IPv6 multicast applications. 1879 * 1880 * If inp is non-NULL, use this socket's current FIB number for any 1881 * required FIB lookup. Look up the group address in the unicast FIB, 1882 * and use its ifp; usually, this points to the default next-hop. 1883 * If the FIB lookup fails, return NULL. 1884 * 1885 * FUTURE: Support multiple forwarding tables for IPv6. 1886 * 1887 * Returns NULL if no ifp could be found. 1888 */ 1889 static struct ifnet * 1890 in6p_lookup_mcast_ifp(const struct inpcb *in6p, 1891 const struct sockaddr_in6 *gsin6) 1892 { 1893 struct nhop6_basic nh6; 1894 struct in6_addr dst; 1895 uint32_t scopeid; 1896 uint32_t fibnum; 1897 1898 KASSERT(in6p->inp_vflag & INP_IPV6, 1899 ("%s: not INP_IPV6 inpcb", __func__)); 1900 KASSERT(gsin6->sin6_family == AF_INET6, 1901 ("%s: not AF_INET6 group", __func__)); 1902 1903 in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid); 1904 fibnum = in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB; 1905 if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6) != 0) 1906 return (NULL); 1907 1908 return (nh6.nh_ifp); 1909 } 1910 1911 /* 1912 * Join an IPv6 multicast group, possibly with a source. 1913 * 1914 * FIXME: The KAME use of the unspecified address (::) 1915 * to join *all* multicast groups is currently unsupported. 1916 */ 1917 static int 1918 in6p_join_group(struct inpcb *inp, struct sockopt *sopt) 1919 { 1920 struct group_source_req gsr; 1921 sockunion_t *gsa, *ssa; 1922 struct ifnet *ifp; 1923 struct in6_mfilter *imf; 1924 struct ip6_moptions *imo; 1925 struct in6_multi *inm; 1926 struct in6_msource *lims; 1927 size_t idx; 1928 int error, is_new; 1929 1930 ifp = NULL; 1931 imf = NULL; 1932 lims = NULL; 1933 error = 0; 1934 is_new = 0; 1935 1936 memset(&gsr, 0, sizeof(struct group_source_req)); 1937 gsa = (sockunion_t *)&gsr.gsr_group; 1938 gsa->ss.ss_family = AF_UNSPEC; 1939 ssa = (sockunion_t *)&gsr.gsr_source; 1940 ssa->ss.ss_family = AF_UNSPEC; 1941 1942 /* 1943 * Chew everything into struct group_source_req. 1944 * Overwrite the port field if present, as the sockaddr 1945 * being copied in may be matched with a binary comparison. 1946 * Ignore passed-in scope ID. 1947 */ 1948 switch (sopt->sopt_name) { 1949 case IPV6_JOIN_GROUP: { 1950 struct ipv6_mreq mreq; 1951 1952 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq), 1953 sizeof(struct ipv6_mreq)); 1954 if (error) 1955 return (error); 1956 1957 gsa->sin6.sin6_family = AF_INET6; 1958 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6); 1959 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr; 1960 1961 if (mreq.ipv6mr_interface == 0) { 1962 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6); 1963 } else { 1964 if (V_if_index < mreq.ipv6mr_interface) 1965 return (EADDRNOTAVAIL); 1966 ifp = ifnet_byindex(mreq.ipv6mr_interface); 1967 } 1968 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p", 1969 __func__, mreq.ipv6mr_interface, ifp); 1970 } break; 1971 1972 case MCAST_JOIN_GROUP: 1973 case MCAST_JOIN_SOURCE_GROUP: 1974 if (sopt->sopt_name == MCAST_JOIN_GROUP) { 1975 error = sooptcopyin(sopt, &gsr, 1976 sizeof(struct group_req), 1977 sizeof(struct group_req)); 1978 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 1979 error = sooptcopyin(sopt, &gsr, 1980 sizeof(struct group_source_req), 1981 sizeof(struct group_source_req)); 1982 } 1983 if (error) 1984 return (error); 1985 1986 if (gsa->sin6.sin6_family != AF_INET6 || 1987 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 1988 return (EINVAL); 1989 1990 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 1991 if (ssa->sin6.sin6_family != AF_INET6 || 1992 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 1993 return (EINVAL); 1994 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr)) 1995 return (EINVAL); 1996 /* 1997 * TODO: Validate embedded scope ID in source 1998 * list entry against passed-in ifp, if and only 1999 * if source list filter entry is iface or node local. 2000 */ 2001 in6_clearscope(&ssa->sin6.sin6_addr); 2002 ssa->sin6.sin6_port = 0; 2003 ssa->sin6.sin6_scope_id = 0; 2004 } 2005 2006 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2007 return (EADDRNOTAVAIL); 2008 ifp = ifnet_byindex(gsr.gsr_interface); 2009 break; 2010 2011 default: 2012 CTR2(KTR_MLD, "%s: unknown sopt_name %d", 2013 __func__, sopt->sopt_name); 2014 return (EOPNOTSUPP); 2015 break; 2016 } 2017 2018 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 2019 return (EINVAL); 2020 2021 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) 2022 return (EADDRNOTAVAIL); 2023 2024 gsa->sin6.sin6_port = 0; 2025 gsa->sin6.sin6_scope_id = 0; 2026 2027 /* 2028 * Always set the scope zone ID on memberships created from userland. 2029 * Use the passed-in ifp to do this. 2030 * XXX The in6_setscope() return value is meaningless. 2031 * XXX SCOPE6_LOCK() is taken by in6_setscope(). 2032 */ 2033 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 2034 2035 imo = in6p_findmoptions(inp); 2036 idx = im6o_match_group(imo, ifp, &gsa->sa); 2037 if (idx == -1) { 2038 is_new = 1; 2039 } else { 2040 inm = imo->im6o_membership[idx]; 2041 imf = &imo->im6o_mfilters[idx]; 2042 if (ssa->ss.ss_family != AF_UNSPEC) { 2043 /* 2044 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership 2045 * is an error. On an existing inclusive membership, 2046 * it just adds the source to the filter list. 2047 */ 2048 if (imf->im6f_st[1] != MCAST_INCLUDE) { 2049 error = EINVAL; 2050 goto out_in6p_locked; 2051 } 2052 /* 2053 * Throw out duplicates. 2054 * 2055 * XXX FIXME: This makes a naive assumption that 2056 * even if entries exist for *ssa in this imf, 2057 * they will be rejected as dupes, even if they 2058 * are not valid in the current mode (in-mode). 2059 * 2060 * in6_msource is transactioned just as for anything 2061 * else in SSM -- but note naive use of in6m_graft() 2062 * below for allocating new filter entries. 2063 * 2064 * This is only an issue if someone mixes the 2065 * full-state SSM API with the delta-based API, 2066 * which is discouraged in the relevant RFCs. 2067 */ 2068 lims = im6o_match_source(imo, idx, &ssa->sa); 2069 if (lims != NULL /*&& 2070 lims->im6sl_st[1] == MCAST_INCLUDE*/) { 2071 error = EADDRNOTAVAIL; 2072 goto out_in6p_locked; 2073 } 2074 } else { 2075 /* 2076 * MCAST_JOIN_GROUP alone, on any existing membership, 2077 * is rejected, to stop the same inpcb tying up 2078 * multiple refs to the in_multi. 2079 * On an existing inclusive membership, this is also 2080 * an error; if you want to change filter mode, 2081 * you must use the userland API setsourcefilter(). 2082 * XXX We don't reject this for imf in UNDEFINED 2083 * state at t1, because allocation of a filter 2084 * is atomic with allocation of a membership. 2085 */ 2086 error = EINVAL; 2087 goto out_in6p_locked; 2088 } 2089 } 2090 2091 /* 2092 * Begin state merge transaction at socket layer. 2093 */ 2094 INP_WLOCK_ASSERT(inp); 2095 2096 if (is_new) { 2097 if (imo->im6o_num_memberships == imo->im6o_max_memberships) { 2098 error = im6o_grow(imo); 2099 if (error) 2100 goto out_in6p_locked; 2101 } 2102 /* 2103 * Allocate the new slot upfront so we can deal with 2104 * grafting the new source filter in same code path 2105 * as for join-source on existing membership. 2106 */ 2107 idx = imo->im6o_num_memberships; 2108 imo->im6o_membership[idx] = NULL; 2109 imo->im6o_num_memberships++; 2110 KASSERT(imo->im6o_mfilters != NULL, 2111 ("%s: im6f_mfilters vector was not allocated", __func__)); 2112 imf = &imo->im6o_mfilters[idx]; 2113 KASSERT(RB_EMPTY(&imf->im6f_sources), 2114 ("%s: im6f_sources not empty", __func__)); 2115 } 2116 2117 /* 2118 * Graft new source into filter list for this inpcb's 2119 * membership of the group. The in6_multi may not have 2120 * been allocated yet if this is a new membership, however, 2121 * the in_mfilter slot will be allocated and must be initialized. 2122 * 2123 * Note: Grafting of exclusive mode filters doesn't happen 2124 * in this path. 2125 * XXX: Should check for non-NULL lims (node exists but may 2126 * not be in-mode) for interop with full-state API. 2127 */ 2128 if (ssa->ss.ss_family != AF_UNSPEC) { 2129 /* Membership starts in IN mode */ 2130 if (is_new) { 2131 CTR1(KTR_MLD, "%s: new join w/source", __func__); 2132 im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE); 2133 } else { 2134 CTR2(KTR_MLD, "%s: %s source", __func__, "allow"); 2135 } 2136 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6); 2137 if (lims == NULL) { 2138 CTR1(KTR_MLD, "%s: merge imf state failed", 2139 __func__); 2140 error = ENOMEM; 2141 goto out_im6o_free; 2142 } 2143 } else { 2144 /* No address specified; Membership starts in EX mode */ 2145 if (is_new) { 2146 CTR1(KTR_MLD, "%s: new join w/o source", __func__); 2147 im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE); 2148 } 2149 } 2150 2151 /* 2152 * Begin state merge transaction at MLD layer. 2153 */ 2154 in_pcbref(inp); 2155 INP_WUNLOCK(inp); 2156 IN6_MULTI_LOCK(); 2157 2158 if (is_new) { 2159 error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf, 2160 &inm, 0); 2161 if (error) { 2162 IN6_MULTI_UNLOCK(); 2163 goto out_im6o_free; 2164 } 2165 imo->im6o_membership[idx] = inm; 2166 } else { 2167 CTR1(KTR_MLD, "%s: merge inm state", __func__); 2168 IN6_MULTI_LIST_LOCK(); 2169 error = in6m_merge(inm, imf); 2170 if (error) 2171 CTR1(KTR_MLD, "%s: failed to merge inm state", 2172 __func__); 2173 else { 2174 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 2175 error = mld_change_state(inm, 0); 2176 if (error) 2177 CTR1(KTR_MLD, "%s: failed mld downcall", 2178 __func__); 2179 } 2180 IN6_MULTI_LIST_UNLOCK(); 2181 } 2182 2183 IN6_MULTI_UNLOCK(); 2184 INP_WLOCK(inp); 2185 if (in_pcbrele_wlocked(inp)) 2186 return (ENXIO); 2187 if (error) { 2188 im6f_rollback(imf); 2189 if (is_new) 2190 im6f_purge(imf); 2191 else 2192 im6f_reap(imf); 2193 } else { 2194 im6f_commit(imf); 2195 } 2196 2197 out_im6o_free: 2198 if (error && is_new) { 2199 imo->im6o_membership[idx] = NULL; 2200 --imo->im6o_num_memberships; 2201 } 2202 2203 out_in6p_locked: 2204 INP_WUNLOCK(inp); 2205 return (error); 2206 } 2207 2208 /* 2209 * Leave an IPv6 multicast group on an inpcb, possibly with a source. 2210 */ 2211 static int 2212 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt) 2213 { 2214 struct ipv6_mreq mreq; 2215 struct group_source_req gsr; 2216 sockunion_t *gsa, *ssa; 2217 struct ifnet *ifp; 2218 struct in6_mfilter *imf; 2219 struct ip6_moptions *imo; 2220 struct in6_msource *ims; 2221 struct in6_multi *inm; 2222 uint32_t ifindex; 2223 size_t idx; 2224 int error, is_final; 2225 #ifdef KTR 2226 char ip6tbuf[INET6_ADDRSTRLEN]; 2227 #endif 2228 2229 ifp = NULL; 2230 ifindex = 0; 2231 error = 0; 2232 is_final = 1; 2233 2234 memset(&gsr, 0, sizeof(struct group_source_req)); 2235 gsa = (sockunion_t *)&gsr.gsr_group; 2236 gsa->ss.ss_family = AF_UNSPEC; 2237 ssa = (sockunion_t *)&gsr.gsr_source; 2238 ssa->ss.ss_family = AF_UNSPEC; 2239 2240 /* 2241 * Chew everything passed in up into a struct group_source_req 2242 * as that is easier to process. 2243 * Note: Any embedded scope ID in the multicast group passed 2244 * in by userland is ignored, the interface index is the recommended 2245 * mechanism to specify an interface; see below. 2246 */ 2247 switch (sopt->sopt_name) { 2248 case IPV6_LEAVE_GROUP: 2249 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq), 2250 sizeof(struct ipv6_mreq)); 2251 if (error) 2252 return (error); 2253 gsa->sin6.sin6_family = AF_INET6; 2254 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6); 2255 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr; 2256 gsa->sin6.sin6_port = 0; 2257 gsa->sin6.sin6_scope_id = 0; 2258 ifindex = mreq.ipv6mr_interface; 2259 break; 2260 2261 case MCAST_LEAVE_GROUP: 2262 case MCAST_LEAVE_SOURCE_GROUP: 2263 if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2264 error = sooptcopyin(sopt, &gsr, 2265 sizeof(struct group_req), 2266 sizeof(struct group_req)); 2267 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2268 error = sooptcopyin(sopt, &gsr, 2269 sizeof(struct group_source_req), 2270 sizeof(struct group_source_req)); 2271 } 2272 if (error) 2273 return (error); 2274 2275 if (gsa->sin6.sin6_family != AF_INET6 || 2276 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 2277 return (EINVAL); 2278 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2279 if (ssa->sin6.sin6_family != AF_INET6 || 2280 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6)) 2281 return (EINVAL); 2282 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr)) 2283 return (EINVAL); 2284 /* 2285 * TODO: Validate embedded scope ID in source 2286 * list entry against passed-in ifp, if and only 2287 * if source list filter entry is iface or node local. 2288 */ 2289 in6_clearscope(&ssa->sin6.sin6_addr); 2290 } 2291 gsa->sin6.sin6_port = 0; 2292 gsa->sin6.sin6_scope_id = 0; 2293 ifindex = gsr.gsr_interface; 2294 break; 2295 2296 default: 2297 CTR2(KTR_MLD, "%s: unknown sopt_name %d", 2298 __func__, sopt->sopt_name); 2299 return (EOPNOTSUPP); 2300 break; 2301 } 2302 2303 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 2304 return (EINVAL); 2305 2306 /* 2307 * Validate interface index if provided. If no interface index 2308 * was provided separately, attempt to look the membership up 2309 * from the default scope as a last resort to disambiguate 2310 * the membership we are being asked to leave. 2311 * XXX SCOPE6 lock potentially taken here. 2312 */ 2313 if (ifindex != 0) { 2314 if (V_if_index < ifindex) 2315 return (EADDRNOTAVAIL); 2316 ifp = ifnet_byindex(ifindex); 2317 if (ifp == NULL) 2318 return (EADDRNOTAVAIL); 2319 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 2320 } else { 2321 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone); 2322 if (error) 2323 return (EADDRNOTAVAIL); 2324 /* 2325 * Some badly behaved applications don't pass an ifindex 2326 * or a scope ID, which is an API violation. In this case, 2327 * perform a lookup as per a v6 join. 2328 * 2329 * XXX For now, stomp on zone ID for the corner case. 2330 * This is not the 'KAME way', but we need to see the ifp 2331 * directly until such time as this implementation is 2332 * refactored, assuming the scope IDs are the way to go. 2333 */ 2334 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]); 2335 if (ifindex == 0) { 2336 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up " 2337 "ifp for group %s.", __func__, 2338 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr)); 2339 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6); 2340 } else { 2341 ifp = ifnet_byindex(ifindex); 2342 } 2343 if (ifp == NULL) 2344 return (EADDRNOTAVAIL); 2345 } 2346 2347 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp); 2348 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__)); 2349 2350 /* 2351 * Find the membership in the membership array. 2352 */ 2353 imo = in6p_findmoptions(inp); 2354 idx = im6o_match_group(imo, ifp, &gsa->sa); 2355 if (idx == -1) { 2356 error = EADDRNOTAVAIL; 2357 goto out_in6p_locked; 2358 } 2359 inm = imo->im6o_membership[idx]; 2360 imf = &imo->im6o_mfilters[idx]; 2361 2362 if (ssa->ss.ss_family != AF_UNSPEC) 2363 is_final = 0; 2364 2365 /* 2366 * Begin state merge transaction at socket layer. 2367 */ 2368 INP_WLOCK_ASSERT(inp); 2369 2370 /* 2371 * If we were instructed only to leave a given source, do so. 2372 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2373 */ 2374 if (is_final) { 2375 im6f_leave(imf); 2376 } else { 2377 if (imf->im6f_st[0] == MCAST_EXCLUDE) { 2378 error = EADDRNOTAVAIL; 2379 goto out_in6p_locked; 2380 } 2381 ims = im6o_match_source(imo, idx, &ssa->sa); 2382 if (ims == NULL) { 2383 CTR3(KTR_MLD, "%s: source %p %spresent", __func__, 2384 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr), 2385 "not "); 2386 error = EADDRNOTAVAIL; 2387 goto out_in6p_locked; 2388 } 2389 CTR2(KTR_MLD, "%s: %s source", __func__, "block"); 2390 error = im6f_prune(imf, &ssa->sin6); 2391 if (error) { 2392 CTR1(KTR_MLD, "%s: merge imf state failed", 2393 __func__); 2394 goto out_in6p_locked; 2395 } 2396 } 2397 2398 /* 2399 * Begin state merge transaction at MLD layer. 2400 */ 2401 in_pcbref(inp); 2402 INP_WUNLOCK(inp); 2403 IN6_MULTI_LOCK(); 2404 2405 if (is_final) { 2406 /* 2407 * Give up the multicast address record to which 2408 * the membership points. 2409 */ 2410 (void)in6_leavegroup_locked(inm, imf); 2411 } else { 2412 CTR1(KTR_MLD, "%s: merge inm state", __func__); 2413 IN6_MULTI_LIST_LOCK(); 2414 error = in6m_merge(inm, imf); 2415 if (error) 2416 CTR1(KTR_MLD, "%s: failed to merge inm state", 2417 __func__); 2418 else { 2419 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 2420 error = mld_change_state(inm, 0); 2421 if (error) 2422 CTR1(KTR_MLD, "%s: failed mld downcall", 2423 __func__); 2424 } 2425 IN6_MULTI_LIST_UNLOCK(); 2426 } 2427 2428 IN6_MULTI_UNLOCK(); 2429 INP_WLOCK(inp); 2430 if (in_pcbrele_wlocked(inp)) 2431 return (ENXIO); 2432 2433 if (error) 2434 im6f_rollback(imf); 2435 else 2436 im6f_commit(imf); 2437 2438 im6f_reap(imf); 2439 2440 if (is_final) { 2441 /* Remove the gap in the membership array. */ 2442 for (++idx; idx < imo->im6o_num_memberships; ++idx) { 2443 imo->im6o_membership[idx-1] = imo->im6o_membership[idx]; 2444 imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx]; 2445 } 2446 imo->im6o_num_memberships--; 2447 } 2448 2449 out_in6p_locked: 2450 INP_WUNLOCK(inp); 2451 return (error); 2452 } 2453 2454 /* 2455 * Select the interface for transmitting IPv6 multicast datagrams. 2456 * 2457 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn 2458 * may be passed to this socket option. An address of in6addr_any or an 2459 * interface index of 0 is used to remove a previous selection. 2460 * When no interface is selected, one is chosen for every send. 2461 */ 2462 static int 2463 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2464 { 2465 struct ifnet *ifp; 2466 struct ip6_moptions *imo; 2467 u_int ifindex; 2468 int error; 2469 2470 if (sopt->sopt_valsize != sizeof(u_int)) 2471 return (EINVAL); 2472 2473 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int)); 2474 if (error) 2475 return (error); 2476 if (V_if_index < ifindex) 2477 return (EINVAL); 2478 if (ifindex == 0) 2479 ifp = NULL; 2480 else { 2481 ifp = ifnet_byindex(ifindex); 2482 if (ifp == NULL) 2483 return (EINVAL); 2484 if ((ifp->if_flags & IFF_MULTICAST) == 0) 2485 return (EADDRNOTAVAIL); 2486 } 2487 imo = in6p_findmoptions(inp); 2488 imo->im6o_multicast_ifp = ifp; 2489 INP_WUNLOCK(inp); 2490 2491 return (0); 2492 } 2493 2494 /* 2495 * Atomically set source filters on a socket for an IPv6 multicast group. 2496 * 2497 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2498 */ 2499 static int 2500 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2501 { 2502 struct __msfilterreq msfr; 2503 sockunion_t *gsa; 2504 struct ifnet *ifp; 2505 struct in6_mfilter *imf; 2506 struct ip6_moptions *imo; 2507 struct in6_multi *inm; 2508 size_t idx; 2509 int error; 2510 2511 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2512 sizeof(struct __msfilterreq)); 2513 if (error) 2514 return (error); 2515 2516 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc) 2517 return (ENOBUFS); 2518 2519 if (msfr.msfr_fmode != MCAST_EXCLUDE && 2520 msfr.msfr_fmode != MCAST_INCLUDE) 2521 return (EINVAL); 2522 2523 if (msfr.msfr_group.ss_family != AF_INET6 || 2524 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6)) 2525 return (EINVAL); 2526 2527 gsa = (sockunion_t *)&msfr.msfr_group; 2528 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr)) 2529 return (EINVAL); 2530 2531 gsa->sin6.sin6_port = 0; /* ignore port */ 2532 2533 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 2534 return (EADDRNOTAVAIL); 2535 ifp = ifnet_byindex(msfr.msfr_ifindex); 2536 if (ifp == NULL) 2537 return (EADDRNOTAVAIL); 2538 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL); 2539 2540 /* 2541 * Take the INP write lock. 2542 * Check if this socket is a member of this group. 2543 */ 2544 imo = in6p_findmoptions(inp); 2545 idx = im6o_match_group(imo, ifp, &gsa->sa); 2546 if (idx == -1 || imo->im6o_mfilters == NULL) { 2547 error = EADDRNOTAVAIL; 2548 goto out_in6p_locked; 2549 } 2550 inm = imo->im6o_membership[idx]; 2551 imf = &imo->im6o_mfilters[idx]; 2552 2553 /* 2554 * Begin state merge transaction at socket layer. 2555 */ 2556 INP_WLOCK_ASSERT(inp); 2557 2558 imf->im6f_st[1] = msfr.msfr_fmode; 2559 2560 /* 2561 * Apply any new source filters, if present. 2562 * Make a copy of the user-space source vector so 2563 * that we may copy them with a single copyin. This 2564 * allows us to deal with page faults up-front. 2565 */ 2566 if (msfr.msfr_nsrcs > 0) { 2567 struct in6_msource *lims; 2568 struct sockaddr_in6 *psin; 2569 struct sockaddr_storage *kss, *pkss; 2570 int i; 2571 2572 INP_WUNLOCK(inp); 2573 2574 CTR2(KTR_MLD, "%s: loading %lu source list entries", 2575 __func__, (unsigned long)msfr.msfr_nsrcs); 2576 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2577 M_TEMP, M_WAITOK); 2578 error = copyin(msfr.msfr_srcs, kss, 2579 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2580 if (error) { 2581 free(kss, M_TEMP); 2582 return (error); 2583 } 2584 2585 INP_WLOCK(inp); 2586 2587 /* 2588 * Mark all source filters as UNDEFINED at t1. 2589 * Restore new group filter mode, as im6f_leave() 2590 * will set it to INCLUDE. 2591 */ 2592 im6f_leave(imf); 2593 imf->im6f_st[1] = msfr.msfr_fmode; 2594 2595 /* 2596 * Update socket layer filters at t1, lazy-allocating 2597 * new entries. This saves a bunch of memory at the 2598 * cost of one RB_FIND() per source entry; duplicate 2599 * entries in the msfr_nsrcs vector are ignored. 2600 * If we encounter an error, rollback transaction. 2601 * 2602 * XXX This too could be replaced with a set-symmetric 2603 * difference like loop to avoid walking from root 2604 * every time, as the key space is common. 2605 */ 2606 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2607 psin = (struct sockaddr_in6 *)pkss; 2608 if (psin->sin6_family != AF_INET6) { 2609 error = EAFNOSUPPORT; 2610 break; 2611 } 2612 if (psin->sin6_len != sizeof(struct sockaddr_in6)) { 2613 error = EINVAL; 2614 break; 2615 } 2616 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) { 2617 error = EINVAL; 2618 break; 2619 } 2620 /* 2621 * TODO: Validate embedded scope ID in source 2622 * list entry against passed-in ifp, if and only 2623 * if source list filter entry is iface or node local. 2624 */ 2625 in6_clearscope(&psin->sin6_addr); 2626 error = im6f_get_source(imf, psin, &lims); 2627 if (error) 2628 break; 2629 lims->im6sl_st[1] = imf->im6f_st[1]; 2630 } 2631 free(kss, M_TEMP); 2632 } 2633 2634 if (error) 2635 goto out_im6f_rollback; 2636 2637 INP_WLOCK_ASSERT(inp); 2638 IN6_MULTI_LIST_LOCK(); 2639 2640 /* 2641 * Begin state merge transaction at MLD layer. 2642 */ 2643 CTR1(KTR_MLD, "%s: merge inm state", __func__); 2644 error = in6m_merge(inm, imf); 2645 if (error) 2646 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__); 2647 else { 2648 CTR1(KTR_MLD, "%s: doing mld downcall", __func__); 2649 error = mld_change_state(inm, 0); 2650 if (error) 2651 CTR1(KTR_MLD, "%s: failed mld downcall", __func__); 2652 } 2653 2654 IN6_MULTI_LIST_UNLOCK(); 2655 2656 out_im6f_rollback: 2657 if (error) 2658 im6f_rollback(imf); 2659 else 2660 im6f_commit(imf); 2661 2662 im6f_reap(imf); 2663 2664 out_in6p_locked: 2665 INP_WUNLOCK(inp); 2666 return (error); 2667 } 2668 2669 /* 2670 * Set the IP multicast options in response to user setsockopt(). 2671 * 2672 * Many of the socket options handled in this function duplicate the 2673 * functionality of socket options in the regular unicast API. However, 2674 * it is not possible to merge the duplicate code, because the idempotence 2675 * of the IPv6 multicast part of the BSD Sockets API must be preserved; 2676 * the effects of these options must be treated as separate and distinct. 2677 * 2678 * SMPng: XXX: Unlocked read of inp_socket believed OK. 2679 */ 2680 int 2681 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2682 { 2683 struct ip6_moptions *im6o; 2684 int error; 2685 2686 error = 0; 2687 2688 /* 2689 * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 2690 * or is a divert socket, reject it. 2691 */ 2692 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 2693 (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2694 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) 2695 return (EOPNOTSUPP); 2696 2697 switch (sopt->sopt_name) { 2698 case IPV6_MULTICAST_IF: 2699 error = in6p_set_multicast_if(inp, sopt); 2700 break; 2701 2702 case IPV6_MULTICAST_HOPS: { 2703 int hlim; 2704 2705 if (sopt->sopt_valsize != sizeof(int)) { 2706 error = EINVAL; 2707 break; 2708 } 2709 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int)); 2710 if (error) 2711 break; 2712 if (hlim < -1 || hlim > 255) { 2713 error = EINVAL; 2714 break; 2715 } else if (hlim == -1) { 2716 hlim = V_ip6_defmcasthlim; 2717 } 2718 im6o = in6p_findmoptions(inp); 2719 im6o->im6o_multicast_hlim = hlim; 2720 INP_WUNLOCK(inp); 2721 break; 2722 } 2723 2724 case IPV6_MULTICAST_LOOP: { 2725 u_int loop; 2726 2727 /* 2728 * Set the loopback flag for outgoing multicast packets. 2729 * Must be zero or one. 2730 */ 2731 if (sopt->sopt_valsize != sizeof(u_int)) { 2732 error = EINVAL; 2733 break; 2734 } 2735 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int)); 2736 if (error) 2737 break; 2738 if (loop > 1) { 2739 error = EINVAL; 2740 break; 2741 } 2742 im6o = in6p_findmoptions(inp); 2743 im6o->im6o_multicast_loop = loop; 2744 INP_WUNLOCK(inp); 2745 break; 2746 } 2747 2748 case IPV6_JOIN_GROUP: 2749 case MCAST_JOIN_GROUP: 2750 case MCAST_JOIN_SOURCE_GROUP: 2751 error = in6p_join_group(inp, sopt); 2752 break; 2753 2754 case IPV6_LEAVE_GROUP: 2755 case MCAST_LEAVE_GROUP: 2756 case MCAST_LEAVE_SOURCE_GROUP: 2757 error = in6p_leave_group(inp, sopt); 2758 break; 2759 2760 case MCAST_BLOCK_SOURCE: 2761 case MCAST_UNBLOCK_SOURCE: 2762 error = in6p_block_unblock_source(inp, sopt); 2763 break; 2764 2765 case IPV6_MSFILTER: 2766 error = in6p_set_source_filters(inp, sopt); 2767 break; 2768 2769 default: 2770 error = EOPNOTSUPP; 2771 break; 2772 } 2773 2774 INP_UNLOCK_ASSERT(inp); 2775 2776 return (error); 2777 } 2778 2779 /* 2780 * Expose MLD's multicast filter mode and source list(s) to userland, 2781 * keyed by (ifindex, group). 2782 * The filter mode is written out as a uint32_t, followed by 2783 * 0..n of struct in6_addr. 2784 * For use by ifmcstat(8). 2785 * SMPng: NOTE: unlocked read of ifindex space. 2786 */ 2787 static int 2788 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS) 2789 { 2790 struct in6_addr mcaddr; 2791 struct in6_addr src; 2792 struct ifnet *ifp; 2793 struct ifmultiaddr *ifma; 2794 struct in6_multi *inm; 2795 struct ip6_msource *ims; 2796 int *name; 2797 int retval; 2798 u_int namelen; 2799 uint32_t fmode, ifindex; 2800 #ifdef KTR 2801 char ip6tbuf[INET6_ADDRSTRLEN]; 2802 #endif 2803 2804 name = (int *)arg1; 2805 namelen = arg2; 2806 2807 if (req->newptr != NULL) 2808 return (EPERM); 2809 2810 /* int: ifindex + 4 * 32 bits of IPv6 address */ 2811 if (namelen != 5) 2812 return (EINVAL); 2813 2814 ifindex = name[0]; 2815 if (ifindex <= 0 || ifindex > V_if_index) { 2816 CTR2(KTR_MLD, "%s: ifindex %u out of range", 2817 __func__, ifindex); 2818 return (ENOENT); 2819 } 2820 2821 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr)); 2822 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) { 2823 CTR2(KTR_MLD, "%s: group %s is not multicast", 2824 __func__, ip6_sprintf(ip6tbuf, &mcaddr)); 2825 return (EINVAL); 2826 } 2827 2828 ifp = ifnet_byindex(ifindex); 2829 if (ifp == NULL) { 2830 CTR2(KTR_MLD, "%s: no ifp for ifindex %u", 2831 __func__, ifindex); 2832 return (ENOENT); 2833 } 2834 /* 2835 * Internal MLD lookups require that scope/zone ID is set. 2836 */ 2837 (void)in6_setscope(&mcaddr, ifp, NULL); 2838 2839 retval = sysctl_wire_old_buffer(req, 2840 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr))); 2841 if (retval) 2842 return (retval); 2843 2844 IN6_MULTI_LOCK(); 2845 IN6_MULTI_LIST_LOCK(); 2846 IF_ADDR_RLOCK(ifp); 2847 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2848 if (ifma->ifma_addr->sa_family != AF_INET6 || 2849 ifma->ifma_protospec == NULL) 2850 continue; 2851 inm = (struct in6_multi *)ifma->ifma_protospec; 2852 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr)) 2853 continue; 2854 fmode = inm->in6m_st[1].iss_fmode; 2855 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 2856 if (retval != 0) 2857 break; 2858 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) { 2859 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims); 2860 /* 2861 * Only copy-out sources which are in-mode. 2862 */ 2863 if (fmode != im6s_get_mode(inm, ims, 1)) { 2864 CTR1(KTR_MLD, "%s: skip non-in-mode", 2865 __func__); 2866 continue; 2867 } 2868 src = ims->im6s_addr; 2869 retval = SYSCTL_OUT(req, &src, 2870 sizeof(struct in6_addr)); 2871 if (retval != 0) 2872 break; 2873 } 2874 } 2875 IF_ADDR_RUNLOCK(ifp); 2876 2877 IN6_MULTI_LIST_UNLOCK(); 2878 IN6_MULTI_UNLOCK(); 2879 2880 return (retval); 2881 } 2882 2883 #ifdef KTR 2884 2885 static const char *in6m_modestrs[] = { "un", "in", "ex" }; 2886 2887 static const char * 2888 in6m_mode_str(const int mode) 2889 { 2890 2891 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 2892 return (in6m_modestrs[mode]); 2893 return ("??"); 2894 } 2895 2896 static const char *in6m_statestrs[] = { 2897 "not-member", 2898 "silent", 2899 "idle", 2900 "lazy", 2901 "sleeping", 2902 "awakening", 2903 "query-pending", 2904 "sg-query-pending", 2905 "leaving" 2906 }; 2907 2908 static const char * 2909 in6m_state_str(const int state) 2910 { 2911 2912 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER) 2913 return (in6m_statestrs[state]); 2914 return ("??"); 2915 } 2916 2917 /* 2918 * Dump an in6_multi structure to the console. 2919 */ 2920 void 2921 in6m_print(const struct in6_multi *inm) 2922 { 2923 int t; 2924 char ip6tbuf[INET6_ADDRSTRLEN]; 2925 2926 if ((ktr_mask & KTR_MLD) == 0) 2927 return; 2928 2929 printf("%s: --- begin in6m %p ---\n", __func__, inm); 2930 printf("addr %s ifp %p(%s) ifma %p\n", 2931 ip6_sprintf(ip6tbuf, &inm->in6m_addr), 2932 inm->in6m_ifp, 2933 if_name(inm->in6m_ifp), 2934 inm->in6m_ifma); 2935 printf("timer %u state %s refcount %u scq.len %u\n", 2936 inm->in6m_timer, 2937 in6m_state_str(inm->in6m_state), 2938 inm->in6m_refcount, 2939 mbufq_len(&inm->in6m_scq)); 2940 printf("mli %p nsrc %lu sctimer %u scrv %u\n", 2941 inm->in6m_mli, 2942 inm->in6m_nsrc, 2943 inm->in6m_sctimer, 2944 inm->in6m_scrv); 2945 for (t = 0; t < 2; t++) { 2946 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 2947 in6m_mode_str(inm->in6m_st[t].iss_fmode), 2948 inm->in6m_st[t].iss_asm, 2949 inm->in6m_st[t].iss_ex, 2950 inm->in6m_st[t].iss_in, 2951 inm->in6m_st[t].iss_rec); 2952 } 2953 printf("%s: --- end in6m %p ---\n", __func__, inm); 2954 } 2955 2956 #else /* !KTR */ 2957 2958 void 2959 in6m_print(const struct in6_multi *inm) 2960 { 2961 2962 } 2963 2964 #endif /* KTR */ 2965