in_mcast.c revision 189635
1170613Sbms/*- 2189592Sbms * Copyright (c) 2007-2009 Bruce Simpson. 3170613Sbms * Copyright (c) 2005 Robert N. M. Watson. 4170613Sbms * All rights reserved. 5170613Sbms * 6170613Sbms * Redistribution and use in source and binary forms, with or without 7170613Sbms * modification, are permitted provided that the following conditions 8170613Sbms * are met: 9170613Sbms * 1. Redistributions of source code must retain the above copyright 10170613Sbms * notice, this list of conditions and the following disclaimer. 11170613Sbms * 2. Redistributions in binary form must reproduce the above copyright 12170613Sbms * notice, this list of conditions and the following disclaimer in the 13170613Sbms * documentation and/or other materials provided with the distribution. 14170613Sbms * 3. The name of the author may not be used to endorse or promote 15170613Sbms * products derived from this software without specific prior written 16170613Sbms * permission. 17170613Sbms * 18170613Sbms * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 19170613Sbms * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 20170613Sbms * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 21170613Sbms * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 22170613Sbms * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 23170613Sbms * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 24170613Sbms * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 25170613Sbms * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 26170613Sbms * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 27170613Sbms * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 28170613Sbms * SUCH DAMAGE. 29170613Sbms */ 30170613Sbms 31170613Sbms/* 32170613Sbms * IPv4 multicast socket, group, and socket option processing module. 33170613Sbms */ 34170613Sbms 35170613Sbms#include <sys/cdefs.h> 36170613Sbms__FBSDID("$FreeBSD: head/sys/netinet/in_mcast.c 189635 2009-03-10 17:48:49Z bms $"); 37170613Sbms 38189106Sbz#include "opt_route.h" 39189106Sbz 40170613Sbms#include <sys/param.h> 41170613Sbms#include <sys/systm.h> 42170613Sbms#include <sys/kernel.h> 43170613Sbms#include <sys/malloc.h> 44170613Sbms#include <sys/mbuf.h> 45171746Scsjp#include <sys/protosw.h> 46170613Sbms#include <sys/socket.h> 47170613Sbms#include <sys/socketvar.h> 48189592Sbms#include <sys/protosw.h> 49170613Sbms#include <sys/sysctl.h> 50181803Sbz#include <sys/vimage.h> 51189592Sbms#include <sys/ktr.h> 52189592Sbms#include <sys/tree.h> 53170613Sbms 54170613Sbms#include <net/if.h> 55170613Sbms#include <net/if_dl.h> 56170613Sbms#include <net/route.h> 57185571Sbz#include <net/vnet.h> 58170613Sbms 59170613Sbms#include <netinet/in.h> 60170613Sbms#include <netinet/in_systm.h> 61170613Sbms#include <netinet/in_pcb.h> 62170613Sbms#include <netinet/in_var.h> 63170613Sbms#include <netinet/ip_var.h> 64170613Sbms#include <netinet/igmp_var.h> 65185571Sbz#include <netinet/vinet.h> 66170613Sbms 67189592Sbms#ifndef KTR_IGMPV3 68189592Sbms#define KTR_IGMPV3 KTR_SUBSYS 69189592Sbms#endif 70189592Sbms 71170613Sbms#ifndef __SOCKUNION_DECLARED 72170613Sbmsunion sockunion { 73170613Sbms struct sockaddr_storage ss; 74170613Sbms struct sockaddr sa; 75170613Sbms struct sockaddr_dl sdl; 76170613Sbms struct sockaddr_in sin; 77170613Sbms}; 78170613Sbmstypedef union sockunion sockunion_t; 79170613Sbms#define __SOCKUNION_DECLARED 80170613Sbms#endif /* __SOCKUNION_DECLARED */ 81170613Sbms 82189592Sbmsstatic MALLOC_DEFINE(M_INMFILTER, "in_mfilter", 83189592Sbms "IPv4 multicast PCB-layer source filter"); 84170613Sbmsstatic MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group"); 85170613Sbmsstatic MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options"); 86189592Sbmsstatic MALLOC_DEFINE(M_IPMSOURCE, "ip_msource", 87189592Sbms "IPv4 multicast IGMP-layer source filter"); 88170613Sbms 89189592Sbms#ifdef VIMAGE_GLOBALS 90189592Sbmsstruct in_multihead in_multihead; /* XXX now unused; retain for ABI */ 91189592Sbms#endif 92189592Sbms 93170613Sbms/* 94189592Sbms * Locking: 95189592Sbms * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 96189592Sbms * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however 97189592Sbms * it can be taken by code in net/if.c also. 98189592Sbms * - ip_moptions and in_mfilter are covered by the INP_WLOCK. 99189592Sbms * 100189592Sbms * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly 101189592Sbms * any need for in_multi itself to be virtualized -- it is bound to an ifp 102189592Sbms * anyway no matter what happens. 103170613Sbms */ 104170613Sbmsstruct mtx in_multi_mtx; 105189592SbmsMTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF); 106170613Sbms 107170613Sbms/* 108170613Sbms * Functions with non-static linkage defined in this file should be 109170613Sbms * declared in in_var.h: 110189592Sbms * imo_multi_filter() 111170613Sbms * in_addmulti() 112170613Sbms * in_delmulti() 113189592Sbms * in_joingroup() 114189592Sbms * in_joingroup_locked() 115189592Sbms * in_leavegroup() 116189592Sbms * in_leavegroup_locked() 117170613Sbms * and ip_var.h: 118170613Sbms * inp_freemoptions() 119170613Sbms * inp_getmoptions() 120170613Sbms * inp_setmoptions() 121189592Sbms * 122189592Sbms * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti() 123189592Sbms * and in_delmulti(). 124170613Sbms */ 125189592Sbmsstatic void imf_commit(struct in_mfilter *); 126189592Sbmsstatic int imf_get_source(struct in_mfilter *imf, 127189592Sbms const struct sockaddr_in *psin, 128189592Sbms struct in_msource **); 129189592Sbmsstatic struct in_msource * 130189592Sbms imf_graft(struct in_mfilter *, const uint8_t, 131189592Sbms const struct sockaddr_in *); 132189592Sbmsstatic void imf_leave(struct in_mfilter *); 133189592Sbmsstatic int imf_prune(struct in_mfilter *, const struct sockaddr_in *); 134189592Sbmsstatic void imf_purge(struct in_mfilter *); 135189592Sbmsstatic void imf_rollback(struct in_mfilter *); 136189592Sbmsstatic void imf_reap(struct in_mfilter *); 137170613Sbmsstatic int imo_grow(struct ip_moptions *); 138189592Sbmsstatic size_t imo_match_group(const struct ip_moptions *, 139189592Sbms const struct ifnet *, const struct sockaddr *); 140189592Sbmsstatic struct in_msource * 141189592Sbms imo_match_source(const struct ip_moptions *, const size_t, 142189592Sbms const struct sockaddr *); 143189592Sbmsstatic void ims_merge(struct ip_msource *ims, 144189592Sbms const struct in_msource *lims, const int rollback); 145189592Sbmsstatic int in_getmulti(struct ifnet *, const struct in_addr *, 146189592Sbms struct in_multi **); 147189592Sbmsstatic int inm_get_source(struct in_multi *inm, const in_addr_t haddr, 148189592Sbms const int noalloc, struct ip_msource **pims); 149189592Sbmsstatic int inm_is_ifp_detached(const struct in_multi *); 150189592Sbmsstatic int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *); 151189592Sbmsstatic void inm_purge(struct in_multi *); 152189592Sbmsstatic void inm_reap(struct in_multi *); 153170613Sbmsstatic struct ip_moptions * 154170613Sbms inp_findmoptions(struct inpcb *); 155170613Sbmsstatic int inp_get_source_filters(struct inpcb *, struct sockopt *); 156170613Sbmsstatic int inp_join_group(struct inpcb *, struct sockopt *); 157170613Sbmsstatic int inp_leave_group(struct inpcb *, struct sockopt *); 158189592Sbmsstatic struct ifnet * 159189592Sbms inp_lookup_mcast_ifp(const struct inpcb *, 160189592Sbms const struct sockaddr_in *, const struct in_addr); 161189592Sbmsstatic int inp_block_unblock_source(struct inpcb *, struct sockopt *); 162170613Sbmsstatic int inp_set_multicast_if(struct inpcb *, struct sockopt *); 163170613Sbmsstatic int inp_set_source_filters(struct inpcb *, struct sockopt *); 164189592Sbmsstatic int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS); 165170613Sbms 166189357SbmsSYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0, "IPv4 multicast"); 167189357Sbms 168189592Sbmsstatic u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER; 169189592SbmsSYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc, 170189592Sbms CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxgrpsrc, 0, 171189592Sbms "Max source filters per group"); 172189592SbmsTUNABLE_ULONG("net.inet.ip.mcast.maxgrpsrc", &in_mcast_maxgrpsrc); 173189592Sbms 174189592Sbmsstatic u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER; 175189592SbmsSYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc, 176189592Sbms CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxsocksrc, 0, 177189592Sbms "Max source filters per socket"); 178189592SbmsTUNABLE_ULONG("net.inet.ip.mcast.maxsocksrc", &in_mcast_maxsocksrc); 179189592Sbms 180189357Sbmsint in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP; 181189357SbmsSYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN, 182189357Sbms &in_mcast_loop, 0, "Loopback multicast datagrams by default"); 183189357SbmsTUNABLE_INT("net.inet.ip.mcast.loop", &in_mcast_loop); 184189357Sbms 185189592SbmsSYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters, 186189592Sbms CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters, 187189592Sbms "Per-interface stack-wide source filters"); 188189592Sbms 189170613Sbms/* 190189592Sbms * Inline function which wraps assertions for a valid ifp. 191189592Sbms * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 192189592Sbms * is detached. 193189592Sbms */ 194189592Sbmsstatic int __inline 195189592Sbmsinm_is_ifp_detached(const struct in_multi *inm) 196189592Sbms{ 197189592Sbms struct ifnet *ifp; 198189592Sbms 199189592Sbms KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 200189592Sbms ifp = inm->inm_ifma->ifma_ifp; 201189592Sbms if (ifp != NULL) { 202189592Sbms /* 203189592Sbms * Sanity check that netinet's notion of ifp is the 204189592Sbms * same as net's. 205189592Sbms */ 206189592Sbms KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 207189592Sbms } 208189592Sbms 209189592Sbms return (ifp == NULL); 210189592Sbms} 211189592Sbms 212189592Sbms/* 213189592Sbms * Initialize an in_mfilter structure to a known state at t0, t1 214189592Sbms * with an empty source filter list. 215189592Sbms */ 216189592Sbmsstatic __inline void 217189592Sbmsimf_init(struct in_mfilter *imf, const int st0, const int st1) 218189592Sbms{ 219189592Sbms memset(imf, 0, sizeof(struct in_mfilter)); 220189592Sbms RB_INIT(&imf->imf_sources); 221189592Sbms imf->imf_st[0] = st0; 222189592Sbms imf->imf_st[1] = st1; 223189592Sbms} 224189592Sbms 225189592Sbms/* 226170613Sbms * Resize the ip_moptions vector to the next power-of-two minus 1. 227170613Sbms * May be called with locks held; do not sleep. 228170613Sbms */ 229170613Sbmsstatic int 230170613Sbmsimo_grow(struct ip_moptions *imo) 231170613Sbms{ 232170613Sbms struct in_multi **nmships; 233170613Sbms struct in_multi **omships; 234170613Sbms struct in_mfilter *nmfilters; 235170613Sbms struct in_mfilter *omfilters; 236170613Sbms size_t idx; 237170613Sbms size_t newmax; 238170613Sbms size_t oldmax; 239170613Sbms 240170613Sbms nmships = NULL; 241170613Sbms nmfilters = NULL; 242170613Sbms omships = imo->imo_membership; 243170613Sbms omfilters = imo->imo_mfilters; 244170613Sbms oldmax = imo->imo_max_memberships; 245170613Sbms newmax = ((oldmax + 1) * 2) - 1; 246170613Sbms 247170613Sbms if (newmax <= IP_MAX_MEMBERSHIPS) { 248170613Sbms nmships = (struct in_multi **)realloc(omships, 249170613Sbms sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT); 250170613Sbms nmfilters = (struct in_mfilter *)realloc(omfilters, 251189592Sbms sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT); 252170613Sbms if (nmships != NULL && nmfilters != NULL) { 253170613Sbms /* Initialize newly allocated source filter heads. */ 254170613Sbms for (idx = oldmax; idx < newmax; idx++) { 255189592Sbms imf_init(&nmfilters[idx], MCAST_UNDEFINED, 256189592Sbms MCAST_EXCLUDE); 257170613Sbms } 258170613Sbms imo->imo_max_memberships = newmax; 259170613Sbms imo->imo_membership = nmships; 260170613Sbms imo->imo_mfilters = nmfilters; 261170613Sbms } 262170613Sbms } 263170613Sbms 264170613Sbms if (nmships == NULL || nmfilters == NULL) { 265170613Sbms if (nmships != NULL) 266170613Sbms free(nmships, M_IPMOPTS); 267170613Sbms if (nmfilters != NULL) 268189592Sbms free(nmfilters, M_INMFILTER); 269170613Sbms return (ETOOMANYREFS); 270170613Sbms } 271170613Sbms 272170613Sbms return (0); 273170613Sbms} 274170613Sbms 275170613Sbms/* 276170613Sbms * Find an IPv4 multicast group entry for this ip_moptions instance 277170613Sbms * which matches the specified group, and optionally an interface. 278170613Sbms * Return its index into the array, or -1 if not found. 279170613Sbms */ 280189592Sbmsstatic size_t 281189592Sbmsimo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp, 282189592Sbms const struct sockaddr *group) 283170613Sbms{ 284189592Sbms const struct sockaddr_in *gsin; 285170613Sbms struct in_multi **pinm; 286170613Sbms int idx; 287170613Sbms int nmships; 288170613Sbms 289189592Sbms gsin = (const struct sockaddr_in *)group; 290170613Sbms 291170613Sbms /* The imo_membership array may be lazy allocated. */ 292170613Sbms if (imo->imo_membership == NULL || imo->imo_num_memberships == 0) 293170613Sbms return (-1); 294170613Sbms 295170613Sbms nmships = imo->imo_num_memberships; 296170613Sbms pinm = &imo->imo_membership[0]; 297170613Sbms for (idx = 0; idx < nmships; idx++, pinm++) { 298170613Sbms if (*pinm == NULL) 299170613Sbms continue; 300170613Sbms if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) && 301189592Sbms in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) { 302170613Sbms break; 303170613Sbms } 304170613Sbms } 305170613Sbms if (idx >= nmships) 306170613Sbms idx = -1; 307170613Sbms 308170613Sbms return (idx); 309170613Sbms} 310170613Sbms 311170613Sbms/* 312189592Sbms * Find an IPv4 multicast source entry for this imo which matches 313170613Sbms * the given group index for this socket, and source address. 314189592Sbms * 315189592Sbms * NOTE: This does not check if the entry is in-mode, merely if 316189592Sbms * it exists, which may not be the desired behaviour. 317170613Sbms */ 318189592Sbmsstatic struct in_msource * 319189592Sbmsimo_match_source(const struct ip_moptions *imo, const size_t gidx, 320189592Sbms const struct sockaddr *src) 321170613Sbms{ 322189592Sbms struct ip_msource find; 323170613Sbms struct in_mfilter *imf; 324189592Sbms struct ip_msource *ims; 325189592Sbms const sockunion_t *psa; 326170613Sbms 327170613Sbms KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__)); 328170613Sbms KASSERT(gidx != -1 && gidx < imo->imo_num_memberships, 329170613Sbms ("%s: invalid index %d\n", __func__, (int)gidx)); 330170613Sbms 331170613Sbms /* The imo_mfilters array may be lazy allocated. */ 332170613Sbms if (imo->imo_mfilters == NULL) 333170613Sbms return (NULL); 334170613Sbms imf = &imo->imo_mfilters[gidx]; 335170613Sbms 336189592Sbms /* Source trees are keyed in host byte order. */ 337189592Sbms psa = (const sockunion_t *)src; 338189592Sbms find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr); 339189592Sbms ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 340189592Sbms 341189592Sbms return ((struct in_msource *)ims); 342170613Sbms} 343170613Sbms 344170613Sbms/* 345189592Sbms * Perform filtering for multicast datagrams on a socket by group and source. 346189592Sbms * 347189592Sbms * Returns 0 if a datagram should be allowed through, or various error codes 348189592Sbms * if the socket was not a member of the group, or the source was muted, etc. 349170613Sbms */ 350189592Sbmsint 351189592Sbmsimo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp, 352189592Sbms const struct sockaddr *group, const struct sockaddr *src) 353170613Sbms{ 354189592Sbms size_t gidx; 355189592Sbms struct in_msource *ims; 356189592Sbms int mode; 357189592Sbms 358189592Sbms KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 359189592Sbms 360189592Sbms gidx = imo_match_group(imo, ifp, group); 361189592Sbms if (gidx == -1) 362189592Sbms return (MCAST_NOTGMEMBER); 363189592Sbms 364189592Sbms /* 365189592Sbms * Check if the source was included in an (S,G) join. 366189592Sbms * Allow reception on exclusive memberships by default, 367189592Sbms * reject reception on inclusive memberships by default. 368189592Sbms * Exclude source only if an in-mode exclude filter exists. 369189592Sbms * Include source only if an in-mode include filter exists. 370189592Sbms * NOTE: We are comparing group state here at IGMP t1 (now) 371189592Sbms * with socket-layer t0 (since last downcall). 372189592Sbms */ 373189592Sbms mode = imo->imo_mfilters[gidx].imf_st[1]; 374189592Sbms ims = imo_match_source(imo, gidx, src); 375189592Sbms 376189592Sbms if ((ims == NULL && mode == MCAST_INCLUDE) || 377189592Sbms (ims != NULL && ims->imsl_st[0] != mode)) 378189592Sbms return (MCAST_NOTSMEMBER); 379189592Sbms 380189592Sbms return (MCAST_PASS); 381189592Sbms} 382189592Sbms 383189592Sbms/* 384189592Sbms * Find and return a reference to an in_multi record for (ifp, group), 385189592Sbms * and bump its reference count. 386189592Sbms * If one does not exist, try to allocate it, and update link-layer multicast 387189592Sbms * filters on ifp to listen for group. 388189592Sbms * Assumes the IN_MULTI lock is held across the call. 389189592Sbms * Return 0 if successful, otherwise return an appropriate error code. 390189592Sbms */ 391189592Sbmsstatic int 392189592Sbmsin_getmulti(struct ifnet *ifp, const struct in_addr *group, 393189592Sbms struct in_multi **pinm) 394189592Sbms{ 395183550Szec INIT_VNET_INET(ifp->if_vnet); 396189592Sbms struct sockaddr_in gsin; 397189592Sbms struct ifmultiaddr *ifma; 398189592Sbms struct in_ifinfo *ii; 399189592Sbms struct in_multi *inm; 400189592Sbms int error; 401170613Sbms 402189592Sbms#if defined(INVARIANTS) && defined(IFF_ASSERTGIANT) 403189592Sbms IFF_ASSERTGIANT(ifp); 404189592Sbms#endif 405189592Sbms IN_MULTI_LOCK_ASSERT(); 406170613Sbms 407189592Sbms ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET]; 408170613Sbms 409189592Sbms inm = inm_lookup(ifp, *group); 410170613Sbms if (inm != NULL) { 411170613Sbms /* 412170613Sbms * If we already joined this group, just bump the 413170613Sbms * refcount and return it. 414170613Sbms */ 415170613Sbms KASSERT(inm->inm_refcount >= 1, 416170613Sbms ("%s: bad refcount %d", __func__, inm->inm_refcount)); 417170613Sbms ++inm->inm_refcount; 418189592Sbms *pinm = inm; 419189592Sbms return (0); 420189592Sbms } 421170613Sbms 422189592Sbms memset(&gsin, 0, sizeof(gsin)); 423189592Sbms gsin.sin_family = AF_INET; 424189592Sbms gsin.sin_len = sizeof(struct sockaddr_in); 425189592Sbms gsin.sin_addr = *group; 426170613Sbms 427189592Sbms /* 428189592Sbms * Check if a link-layer group is already associated 429189592Sbms * with this network-layer group on the given ifnet. 430189592Sbms */ 431189592Sbms error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma); 432189592Sbms if (error != 0) 433189592Sbms return (error); 434189592Sbms 435189592Sbms /* 436189592Sbms * If something other than netinet is occupying the link-layer 437189592Sbms * group, print a meaningful error message and back out of 438189592Sbms * the allocation. 439189592Sbms * Otherwise, bump the refcount on the existing network-layer 440189592Sbms * group association and return it. 441189592Sbms */ 442189592Sbms if (ifma->ifma_protospec != NULL) { 443189592Sbms inm = (struct in_multi *)ifma->ifma_protospec; 444170613Sbms#ifdef INVARIANTS 445189592Sbms KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 446189592Sbms __func__)); 447189592Sbms KASSERT(ifma->ifma_addr->sa_family == AF_INET, 448189592Sbms ("%s: ifma not AF_INET", __func__)); 449189592Sbms KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 450189592Sbms if (inm->inm_ifma != ifma || inm->inm_ifp != ifp || 451189592Sbms !in_hosteq(inm->inm_addr, *group)) 452189592Sbms panic("%s: ifma %p is inconsistent with %p (%s)", 453189592Sbms __func__, ifma, inm, inet_ntoa(*group)); 454170613Sbms#endif 455189592Sbms ++inm->inm_refcount; 456189592Sbms *pinm = inm; 457189592Sbms return (0); 458189592Sbms } 459189592Sbms 460189592Sbms /* 461189592Sbms * A new in_multi record is needed; allocate and initialize it. 462189592Sbms * We DO NOT perform an IGMP join as the in_ layer may need to 463189592Sbms * push an initial source list down to IGMP to support SSM. 464189592Sbms * 465189592Sbms * The initial source filter state is INCLUDE, {} as per the RFC. 466189592Sbms */ 467189592Sbms inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO); 468189592Sbms if (inm == NULL) { 469189592Sbms if_delmulti_ifma(ifma); 470189592Sbms return (ENOMEM); 471189592Sbms } 472189592Sbms inm->inm_addr = *group; 473189592Sbms inm->inm_ifp = ifp; 474189592Sbms inm->inm_igi = ii->ii_igmp; 475189592Sbms inm->inm_ifma = ifma; 476189592Sbms inm->inm_refcount = 1; 477189592Sbms inm->inm_state = IGMP_NOT_MEMBER; 478189592Sbms 479189592Sbms /* 480189592Sbms * Pending state-changes per group are subject to a bounds check. 481189592Sbms */ 482189592Sbms IFQ_SET_MAXLEN(&inm->inm_scq, IGMP_MAX_STATE_CHANGES); 483189592Sbms 484189592Sbms inm->inm_st[0].iss_fmode = MCAST_UNDEFINED; 485189592Sbms inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 486189592Sbms RB_INIT(&inm->inm_srcs); 487189592Sbms 488189592Sbms ifma->ifma_protospec = inm; 489189592Sbms 490189592Sbms *pinm = inm; 491189592Sbms 492189592Sbms return (0); 493189592Sbms} 494189592Sbms 495189592Sbms/* 496189592Sbms * Drop a reference to an in_multi record. 497189592Sbms * 498189592Sbms * If the refcount drops to 0, free the in_multi record and 499189592Sbms * delete the underlying link-layer membership. 500189592Sbms */ 501189592Sbmsvoid 502189592Sbmsinm_release_locked(struct in_multi *inm) 503189592Sbms{ 504189592Sbms struct ifmultiaddr *ifma; 505189592Sbms 506189592Sbms#if defined(INVARIANTS) && defined(IFF_ASSERTGIANT) 507189592Sbms if (!inm_is_ifp_detached(inm)) 508189592Sbms IFF_ASSERTGIANT(ifp); 509189592Sbms#endif 510189592Sbms 511189592Sbms IN_MULTI_LOCK_ASSERT(); 512189592Sbms 513189592Sbms CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount); 514189592Sbms 515189592Sbms if (--inm->inm_refcount > 0) { 516189592Sbms CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__, 517189592Sbms inm->inm_refcount); 518189592Sbms return; 519189592Sbms } 520189592Sbms 521189592Sbms CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm); 522189592Sbms 523189592Sbms ifma = inm->inm_ifma; 524189592Sbms 525189592Sbms CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma); 526189592Sbms KASSERT(ifma->ifma_protospec == inm, 527189592Sbms ("%s: ifma_protospec != inm", __func__)); 528189592Sbms ifma->ifma_protospec = NULL; 529189592Sbms 530189592Sbms inm_purge(inm); 531189592Sbms 532189592Sbms free(inm, M_IPMADDR); 533189592Sbms 534189592Sbms if_delmulti_ifma(ifma); 535189592Sbms} 536189592Sbms 537189592Sbms/* 538189592Sbms * Clear recorded source entries for a group. 539189592Sbms * Used by the IGMP code. Caller must hold the IN_MULTI lock. 540189592Sbms * FIXME: Should reap. 541189592Sbms */ 542189592Sbmsvoid 543189592Sbmsinm_clear_recorded(struct in_multi *inm) 544189592Sbms{ 545189592Sbms struct ip_msource *ims; 546189592Sbms 547189592Sbms IN_MULTI_LOCK_ASSERT(); 548189592Sbms 549189592Sbms RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 550189592Sbms if (ims->ims_stp) { 551189592Sbms ims->ims_stp = 0; 552189592Sbms --inm->inm_st[1].iss_rec; 553170613Sbms } 554189592Sbms } 555189592Sbms KASSERT(inm->inm_st[1].iss_rec == 0, 556189592Sbms ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec)); 557189592Sbms} 558170613Sbms 559189592Sbms/* 560189592Sbms * Record a source as pending for a Source-Group IGMPv3 query. 561189592Sbms * This lives here as it modifies the shared tree. 562189592Sbms * 563189592Sbms * inm is the group descriptor. 564189592Sbms * naddr is the address of the source to record in network-byte order. 565189592Sbms * 566189592Sbms * If the net.inet.igmp.sgalloc sysctl is non-zero, we will 567189592Sbms * lazy-allocate a source node in response to an SG query. 568189592Sbms * Otherwise, no allocation is performed. This saves some memory 569189592Sbms * with the trade-off that the source will not be reported to the 570189592Sbms * router if joined in the window between the query response and 571189592Sbms * the group actually being joined on the local host. 572189592Sbms * 573189592Sbms * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed. 574189592Sbms * This turns off the allocation of a recorded source entry if 575189592Sbms * the group has not been joined. 576189592Sbms * 577189592Sbms * Return 0 if the source didn't exist or was already marked as recorded. 578189592Sbms * Return 1 if the source was marked as recorded by this function. 579189592Sbms * Return <0 if any error occured (negated errno code). 580189592Sbms */ 581189592Sbmsint 582189592Sbmsinm_record_source(struct in_multi *inm, const in_addr_t naddr) 583189592Sbms{ 584189592Sbms struct ip_msource find; 585189592Sbms struct ip_msource *ims, *nims; 586189592Sbms 587189592Sbms IN_MULTI_LOCK_ASSERT(); 588189592Sbms 589189592Sbms find.ims_haddr = ntohl(naddr); 590189592Sbms ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 591189592Sbms if (ims && ims->ims_stp) 592189592Sbms return (0); 593189592Sbms if (ims == NULL) { 594189592Sbms if (inm->inm_nsrc == in_mcast_maxgrpsrc) 595189592Sbms return (-ENOSPC); 596189592Sbms nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 597189592Sbms M_NOWAIT | M_ZERO); 598189592Sbms if (nims == NULL) 599189592Sbms return (-ENOMEM); 600189592Sbms nims->ims_haddr = find.ims_haddr; 601189592Sbms RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 602189592Sbms ++inm->inm_nsrc; 603189592Sbms ims = nims; 604189592Sbms } 605189592Sbms 606189592Sbms /* 607189592Sbms * Mark the source as recorded and update the recorded 608189592Sbms * source count. 609189592Sbms */ 610189592Sbms ++ims->ims_stp; 611189592Sbms ++inm->inm_st[1].iss_rec; 612189592Sbms 613189592Sbms return (1); 614189592Sbms} 615189592Sbms 616189592Sbms/* 617189592Sbms * Return a pointer to an in_msource owned by an in_mfilter, 618189592Sbms * given its source address. 619189592Sbms * Lazy-allocate if needed. If this is a new entry its filter state is 620189592Sbms * undefined at t0. 621189592Sbms * 622189592Sbms * imf is the filter set being modified. 623189592Sbms * haddr is the source address in *host* byte-order. 624189592Sbms * 625189592Sbms * SMPng: May be called with locks held; malloc must not block. 626189592Sbms */ 627189592Sbmsstatic int 628189592Sbmsimf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin, 629189592Sbms struct in_msource **plims) 630189592Sbms{ 631189592Sbms struct ip_msource find; 632189592Sbms struct ip_msource *ims, *nims; 633189592Sbms struct in_msource *lims; 634189592Sbms int error; 635189592Sbms 636189592Sbms error = 0; 637189592Sbms ims = NULL; 638189592Sbms lims = NULL; 639189592Sbms 640189592Sbms /* key is host byte order */ 641189592Sbms find.ims_haddr = ntohl(psin->sin_addr.s_addr); 642189592Sbms ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 643189592Sbms lims = (struct in_msource *)ims; 644189592Sbms if (lims == NULL) { 645189592Sbms if (imf->imf_nsrc == in_mcast_maxsocksrc) 646189592Sbms return (ENOSPC); 647189592Sbms nims = malloc(sizeof(struct in_msource), M_INMFILTER, 648189592Sbms M_NOWAIT | M_ZERO); 649189592Sbms if (nims == NULL) 650189592Sbms return (ENOMEM); 651189592Sbms lims = (struct in_msource *)nims; 652189592Sbms lims->ims_haddr = find.ims_haddr; 653189592Sbms lims->imsl_st[0] = MCAST_UNDEFINED; 654189592Sbms RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 655189592Sbms ++imf->imf_nsrc; 656189592Sbms } 657189592Sbms 658189592Sbms *plims = lims; 659189592Sbms 660189592Sbms return (error); 661189592Sbms} 662189592Sbms 663189592Sbms/* 664189592Sbms * Graft a source entry into an existing socket-layer filter set, 665189592Sbms * maintaining any required invariants and checking allocations. 666189592Sbms * 667189592Sbms * The source is marked as being in the new filter mode at t1. 668189592Sbms * 669189592Sbms * Return the pointer to the new node, otherwise return NULL. 670189592Sbms */ 671189592Sbmsstatic struct in_msource * 672189592Sbmsimf_graft(struct in_mfilter *imf, const uint8_t st1, 673189592Sbms const struct sockaddr_in *psin) 674189592Sbms{ 675189592Sbms struct ip_msource *nims; 676189592Sbms struct in_msource *lims; 677189592Sbms 678189592Sbms nims = malloc(sizeof(struct in_msource), M_INMFILTER, 679189592Sbms M_NOWAIT | M_ZERO); 680189592Sbms if (nims == NULL) 681189592Sbms return (NULL); 682189592Sbms lims = (struct in_msource *)nims; 683189592Sbms lims->ims_haddr = ntohl(psin->sin_addr.s_addr); 684189592Sbms lims->imsl_st[0] = MCAST_UNDEFINED; 685189592Sbms lims->imsl_st[1] = st1; 686189592Sbms RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 687189592Sbms ++imf->imf_nsrc; 688189592Sbms 689189592Sbms return (lims); 690189592Sbms} 691189592Sbms 692189592Sbms/* 693189592Sbms * Prune a source entry from an existing socket-layer filter set, 694189592Sbms * maintaining any required invariants and checking allocations. 695189592Sbms * 696189592Sbms * The source is marked as being left at t1, it is not freed. 697189592Sbms * 698189592Sbms * Return 0 if no error occurred, otherwise return an errno value. 699189592Sbms */ 700189592Sbmsstatic int 701189592Sbmsimf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin) 702189592Sbms{ 703189592Sbms struct ip_msource find; 704189592Sbms struct ip_msource *ims; 705189592Sbms struct in_msource *lims; 706189592Sbms 707189592Sbms /* key is host byte order */ 708189592Sbms find.ims_haddr = ntohl(psin->sin_addr.s_addr); 709189592Sbms ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 710189592Sbms if (ims == NULL) 711189592Sbms return (ENOENT); 712189592Sbms lims = (struct in_msource *)ims; 713189592Sbms lims->imsl_st[1] = MCAST_UNDEFINED; 714189592Sbms return (0); 715189592Sbms} 716189592Sbms 717189592Sbms/* 718189592Sbms * Revert socket-layer filter set deltas at t1 to t0 state. 719189592Sbms */ 720189592Sbmsstatic void 721189592Sbmsimf_rollback(struct in_mfilter *imf) 722189592Sbms{ 723189592Sbms struct ip_msource *ims, *tims; 724189592Sbms struct in_msource *lims; 725189592Sbms 726189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 727189592Sbms lims = (struct in_msource *)ims; 728189592Sbms if (lims->imsl_st[0] == lims->imsl_st[1]) { 729189592Sbms /* no change at t1 */ 730189592Sbms continue; 731189592Sbms } else if (lims->imsl_st[0] != MCAST_UNDEFINED) { 732189592Sbms /* revert change to existing source at t1 */ 733189592Sbms lims->imsl_st[1] = lims->imsl_st[0]; 734189592Sbms } else { 735189592Sbms /* revert source added t1 */ 736189592Sbms CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 737189592Sbms RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 738189592Sbms free(ims, M_INMFILTER); 739189592Sbms imf->imf_nsrc--; 740189592Sbms } 741189592Sbms } 742189592Sbms imf->imf_st[1] = imf->imf_st[0]; 743189592Sbms} 744189592Sbms 745189592Sbms/* 746189592Sbms * Mark socket-layer filter set as INCLUDE {} at t1. 747189592Sbms */ 748189592Sbmsstatic void 749189592Sbmsimf_leave(struct in_mfilter *imf) 750189592Sbms{ 751189592Sbms struct ip_msource *ims; 752189592Sbms struct in_msource *lims; 753189592Sbms 754189592Sbms RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 755189592Sbms lims = (struct in_msource *)ims; 756189592Sbms lims->imsl_st[1] = MCAST_UNDEFINED; 757189592Sbms } 758189592Sbms imf->imf_st[1] = MCAST_INCLUDE; 759189592Sbms} 760189592Sbms 761189592Sbms/* 762189592Sbms * Mark socket-layer filter set deltas as committed. 763189592Sbms */ 764189592Sbmsstatic void 765189592Sbmsimf_commit(struct in_mfilter *imf) 766189592Sbms{ 767189592Sbms struct ip_msource *ims; 768189592Sbms struct in_msource *lims; 769189592Sbms 770189592Sbms RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 771189592Sbms lims = (struct in_msource *)ims; 772189592Sbms lims->imsl_st[0] = lims->imsl_st[1]; 773189592Sbms } 774189592Sbms imf->imf_st[0] = imf->imf_st[1]; 775189592Sbms} 776189592Sbms 777189592Sbms/* 778189592Sbms * Reap unreferenced sources from socket-layer filter set. 779189592Sbms */ 780189592Sbmsstatic void 781189592Sbmsimf_reap(struct in_mfilter *imf) 782189592Sbms{ 783189592Sbms struct ip_msource *ims, *tims; 784189592Sbms struct in_msource *lims; 785189592Sbms 786189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 787189592Sbms lims = (struct in_msource *)ims; 788189592Sbms if ((lims->imsl_st[0] == MCAST_UNDEFINED) && 789189592Sbms (lims->imsl_st[1] == MCAST_UNDEFINED)) { 790189592Sbms CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims); 791189592Sbms RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 792189592Sbms free(ims, M_INMFILTER); 793189592Sbms imf->imf_nsrc--; 794189592Sbms } 795189592Sbms } 796189592Sbms} 797189592Sbms 798189592Sbms/* 799189592Sbms * Purge socket-layer filter set. 800189592Sbms */ 801189592Sbmsstatic void 802189592Sbmsimf_purge(struct in_mfilter *imf) 803189592Sbms{ 804189592Sbms struct ip_msource *ims, *tims; 805189592Sbms 806189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 807189592Sbms CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 808189592Sbms RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 809189592Sbms free(ims, M_INMFILTER); 810189592Sbms imf->imf_nsrc--; 811189592Sbms } 812189592Sbms imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED; 813189592Sbms KASSERT(RB_EMPTY(&imf->imf_sources), 814189592Sbms ("%s: imf_sources not empty", __func__)); 815189592Sbms} 816189592Sbms 817189592Sbms/* 818189592Sbms * Look up a source filter entry for a multicast group. 819189592Sbms * 820189592Sbms * inm is the group descriptor to work with. 821189592Sbms * haddr is the host-byte-order IPv4 address to look up. 822189592Sbms * noalloc may be non-zero to suppress allocation of sources. 823189592Sbms * *pims will be set to the address of the retrieved or allocated source. 824189592Sbms * 825189592Sbms * SMPng: NOTE: may be called with locks held. 826189592Sbms * Return 0 if successful, otherwise return a non-zero error code. 827189592Sbms */ 828189592Sbmsstatic int 829189592Sbmsinm_get_source(struct in_multi *inm, const in_addr_t haddr, 830189592Sbms const int noalloc, struct ip_msource **pims) 831189592Sbms{ 832189592Sbms struct ip_msource find; 833189592Sbms struct ip_msource *ims, *nims; 834189592Sbms#ifdef KTR 835189592Sbms struct in_addr ia; 836189592Sbms#endif 837189592Sbms 838189592Sbms find.ims_haddr = haddr; 839189592Sbms ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 840189592Sbms if (ims == NULL && !noalloc) { 841189592Sbms if (inm->inm_nsrc == in_mcast_maxgrpsrc) 842189592Sbms return (ENOSPC); 843189592Sbms nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 844189592Sbms M_NOWAIT | M_ZERO); 845189592Sbms if (nims == NULL) 846189592Sbms return (ENOMEM); 847189592Sbms nims->ims_haddr = haddr; 848189592Sbms RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 849189592Sbms ++inm->inm_nsrc; 850189592Sbms ims = nims; 851189592Sbms#ifdef KTR 852189592Sbms ia.s_addr = htonl(haddr); 853189592Sbms CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__, 854189592Sbms inet_ntoa(ia), ims); 855189592Sbms#endif 856189592Sbms } 857189592Sbms 858189592Sbms *pims = ims; 859189592Sbms return (0); 860189592Sbms} 861189592Sbms 862189592Sbms/* 863189592Sbms * Merge socket-layer source into IGMP-layer source. 864189592Sbms * If rollback is non-zero, perform the inverse of the merge. 865189592Sbms */ 866189592Sbmsstatic void 867189592Sbmsims_merge(struct ip_msource *ims, const struct in_msource *lims, 868189592Sbms const int rollback) 869189592Sbms{ 870189592Sbms int n = rollback ? -1 : 1; 871189592Sbms#ifdef KTR 872189592Sbms struct in_addr ia; 873189592Sbms 874189592Sbms ia.s_addr = htonl(ims->ims_haddr); 875189592Sbms#endif 876189592Sbms 877189592Sbms if (lims->imsl_st[0] == MCAST_EXCLUDE) { 878189592Sbms CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s", 879189592Sbms __func__, n, inet_ntoa(ia)); 880189592Sbms ims->ims_st[1].ex -= n; 881189592Sbms } else if (lims->imsl_st[0] == MCAST_INCLUDE) { 882189592Sbms CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s", 883189592Sbms __func__, n, inet_ntoa(ia)); 884189592Sbms ims->ims_st[1].in -= n; 885189592Sbms } 886189592Sbms 887189592Sbms if (lims->imsl_st[1] == MCAST_EXCLUDE) { 888189592Sbms CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s", 889189592Sbms __func__, n, inet_ntoa(ia)); 890189592Sbms ims->ims_st[1].ex += n; 891189592Sbms } else if (lims->imsl_st[1] == MCAST_INCLUDE) { 892189592Sbms CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s", 893189592Sbms __func__, n, inet_ntoa(ia)); 894189592Sbms ims->ims_st[1].in += n; 895189592Sbms } 896189592Sbms} 897189592Sbms 898189592Sbms/* 899189592Sbms * Atomically update the global in_multi state, when a membership's 900189592Sbms * filter list is being updated in any way. 901189592Sbms * 902189592Sbms * imf is the per-inpcb-membership group filter pointer. 903189592Sbms * A fake imf may be passed for in-kernel consumers. 904189592Sbms * 905189592Sbms * XXX This is a candidate for a set-symmetric-difference style loop 906189592Sbms * which would eliminate the repeated lookup from root of ims nodes, 907189592Sbms * as they share the same key space. 908189592Sbms * 909189592Sbms * If any error occurred this function will back out of refcounts 910189592Sbms * and return a non-zero value. 911189592Sbms */ 912189592Sbmsstatic int 913189592Sbmsinm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 914189592Sbms{ 915189592Sbms struct ip_msource *ims, *nims; 916189592Sbms struct in_msource *lims; 917189592Sbms int schanged, error; 918189592Sbms int nsrc0, nsrc1; 919189592Sbms 920189592Sbms schanged = 0; 921189592Sbms error = 0; 922189592Sbms nsrc1 = nsrc0 = 0; 923189592Sbms 924189592Sbms /* 925189592Sbms * Update the source filters first, as this may fail. 926189592Sbms * Maintain count of in-mode filters at t0, t1. These are 927189592Sbms * used to work out if we transition into ASM mode or not. 928189592Sbms * Maintain a count of source filters whose state was 929189592Sbms * actually modified by this operation. 930189592Sbms */ 931189592Sbms RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 932189592Sbms lims = (struct in_msource *)ims; 933189592Sbms if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++; 934189592Sbms if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++; 935189592Sbms if (lims->imsl_st[0] == lims->imsl_st[1]) continue; 936189592Sbms error = inm_get_source(inm, lims->ims_haddr, 0, &nims); 937189592Sbms ++schanged; 938189592Sbms if (error) 939170613Sbms break; 940189592Sbms ims_merge(nims, lims, 0); 941189592Sbms } 942189592Sbms if (error) { 943189592Sbms struct ip_msource *bims; 944189592Sbms 945189592Sbms RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) { 946189592Sbms lims = (struct in_msource *)ims; 947189592Sbms if (lims->imsl_st[0] == lims->imsl_st[1]) 948189592Sbms continue; 949189592Sbms (void)inm_get_source(inm, lims->ims_haddr, 1, &bims); 950189592Sbms if (bims == NULL) 951189592Sbms continue; 952189592Sbms ims_merge(bims, lims, 1); 953170613Sbms } 954189592Sbms goto out_reap; 955189592Sbms } 956170613Sbms 957189592Sbms CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1", 958189592Sbms __func__, nsrc0, nsrc1); 959170613Sbms 960189592Sbms /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 961189592Sbms if (imf->imf_st[0] == imf->imf_st[1] && 962189592Sbms imf->imf_st[1] == MCAST_INCLUDE) { 963189592Sbms if (nsrc1 == 0) { 964189592Sbms CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 965189592Sbms --inm->inm_st[1].iss_in; 966189592Sbms } 967189592Sbms } 968170613Sbms 969189592Sbms /* Handle filter mode transition on socket. */ 970189592Sbms if (imf->imf_st[0] != imf->imf_st[1]) { 971189592Sbms CTR3(KTR_IGMPV3, "%s: imf transition %d to %d", 972189592Sbms __func__, imf->imf_st[0], imf->imf_st[1]); 973189592Sbms 974189592Sbms if (imf->imf_st[0] == MCAST_EXCLUDE) { 975189592Sbms CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__); 976189592Sbms --inm->inm_st[1].iss_ex; 977189592Sbms } else if (imf->imf_st[0] == MCAST_INCLUDE) { 978189592Sbms CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 979189592Sbms --inm->inm_st[1].iss_in; 980189592Sbms } 981189592Sbms 982189592Sbms if (imf->imf_st[1] == MCAST_EXCLUDE) { 983189592Sbms CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__); 984189592Sbms inm->inm_st[1].iss_ex++; 985189592Sbms } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) { 986189592Sbms CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__); 987189592Sbms inm->inm_st[1].iss_in++; 988189592Sbms } 989189592Sbms } 990189592Sbms 991189592Sbms /* 992189592Sbms * Track inm filter state in terms of listener counts. 993189592Sbms * If there are any exclusive listeners, stack-wide 994189592Sbms * membership is exclusive. 995189592Sbms * Otherwise, if only inclusive listeners, stack-wide is inclusive. 996189592Sbms * If no listeners remain, state is undefined at t1, 997189592Sbms * and the IGMP lifecycle for this group should finish. 998189592Sbms */ 999189592Sbms if (inm->inm_st[1].iss_ex > 0) { 1000189592Sbms CTR1(KTR_IGMPV3, "%s: transition to EX", __func__); 1001189592Sbms inm->inm_st[1].iss_fmode = MCAST_EXCLUDE; 1002189592Sbms } else if (inm->inm_st[1].iss_in > 0) { 1003189592Sbms CTR1(KTR_IGMPV3, "%s: transition to IN", __func__); 1004189592Sbms inm->inm_st[1].iss_fmode = MCAST_INCLUDE; 1005189592Sbms } else { 1006189592Sbms CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__); 1007189592Sbms inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 1008189592Sbms } 1009189592Sbms 1010189592Sbms /* Decrement ASM listener count on transition out of ASM mode. */ 1011189592Sbms if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1012189592Sbms if ((imf->imf_st[1] != MCAST_EXCLUDE) || 1013189592Sbms (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) 1014189592Sbms CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__); 1015189592Sbms --inm->inm_st[1].iss_asm; 1016189592Sbms } 1017189592Sbms 1018189592Sbms /* Increment ASM listener count on transition to ASM mode. */ 1019189592Sbms if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1020189592Sbms CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__); 1021189592Sbms inm->inm_st[1].iss_asm++; 1022189592Sbms } 1023189592Sbms 1024189592Sbms CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm); 1025189592Sbms inm_print(inm); 1026189592Sbms 1027189592Sbmsout_reap: 1028189592Sbms if (schanged > 0) { 1029189592Sbms CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__); 1030189592Sbms inm_reap(inm); 1031189592Sbms } 1032189592Sbms return (error); 1033189592Sbms} 1034189592Sbms 1035189592Sbms/* 1036189592Sbms * Mark an in_multi's filter set deltas as committed. 1037189592Sbms * Called by IGMP after a state change has been enqueued. 1038189592Sbms */ 1039189592Sbmsvoid 1040189592Sbmsinm_commit(struct in_multi *inm) 1041189592Sbms{ 1042189592Sbms struct ip_msource *ims; 1043189592Sbms 1044189592Sbms CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm); 1045189592Sbms CTR1(KTR_IGMPV3, "%s: pre commit:", __func__); 1046189592Sbms inm_print(inm); 1047189592Sbms 1048189592Sbms RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 1049189592Sbms ims->ims_st[0] = ims->ims_st[1]; 1050189592Sbms } 1051189592Sbms inm->inm_st[0] = inm->inm_st[1]; 1052189592Sbms} 1053189592Sbms 1054189592Sbms/* 1055189592Sbms * Reap unreferenced nodes from an in_multi's filter set. 1056189592Sbms */ 1057189592Sbmsstatic void 1058189592Sbmsinm_reap(struct in_multi *inm) 1059189592Sbms{ 1060189592Sbms struct ip_msource *ims, *tims; 1061189592Sbms 1062189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1063189592Sbms if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 || 1064189592Sbms ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 || 1065189592Sbms ims->ims_stp != 0) 1066189592Sbms continue; 1067189592Sbms CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1068189592Sbms RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1069189592Sbms free(ims, M_IPMSOURCE); 1070189592Sbms inm->inm_nsrc--; 1071189592Sbms } 1072189592Sbms} 1073189592Sbms 1074189592Sbms/* 1075189592Sbms * Purge all source nodes from an in_multi's filter set. 1076189592Sbms */ 1077189592Sbmsstatic void 1078189592Sbmsinm_purge(struct in_multi *inm) 1079189592Sbms{ 1080189592Sbms struct ip_msource *ims, *tims; 1081189592Sbms 1082189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1083189592Sbms CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1084189592Sbms RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1085189592Sbms free(ims, M_IPMSOURCE); 1086189592Sbms inm->inm_nsrc--; 1087189592Sbms } 1088189592Sbms} 1089189592Sbms 1090189592Sbms/* 1091189592Sbms * Join a multicast group; unlocked entry point. 1092189592Sbms * 1093189592Sbms * SMPng: XXX: in_joingroup() is called from in_control() when Giant 1094189592Sbms * is not held. Fortunately, ifp is unlikely to have been detached 1095189592Sbms * at this point, so we assume it's OK to recurse. 1096189592Sbms */ 1097189592Sbmsint 1098189592Sbmsin_joingroup(struct ifnet *ifp, const struct in_addr *gina, 1099189592Sbms /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1100189592Sbms{ 1101189592Sbms int error; 1102189592Sbms 1103189592Sbms IFF_LOCKGIANT(ifp); 1104189592Sbms IN_MULTI_LOCK(); 1105189592Sbms error = in_joingroup_locked(ifp, gina, imf, pinm); 1106170613Sbms IN_MULTI_UNLOCK(); 1107170613Sbms IFF_UNLOCKGIANT(ifp); 1108170613Sbms 1109189592Sbms return (error); 1110170613Sbms} 1111170613Sbms 1112170613Sbms/* 1113189592Sbms * Join a multicast group; real entry point. 1114170613Sbms * 1115189592Sbms * Only preserves atomicity at inm level. 1116189592Sbms * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1117170613Sbms * 1118189592Sbms * If the IGMP downcall fails, the group is not joined, and an error 1119189592Sbms * code is returned. 1120170613Sbms */ 1121189592Sbmsint 1122189592Sbmsin_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina, 1123189592Sbms /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1124170613Sbms{ 1125189592Sbms struct in_mfilter timf; 1126189592Sbms struct in_multi *inm; 1127189592Sbms int error; 1128170613Sbms 1129189592Sbms IN_MULTI_LOCK_ASSERT(); 1130170613Sbms 1131189592Sbms CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__, 1132189592Sbms inet_ntoa(*gina), ifp, ifp->if_xname); 1133189592Sbms 1134189592Sbms error = 0; 1135189592Sbms inm = NULL; 1136189592Sbms 1137189592Sbms /* 1138189592Sbms * If no imf was specified (i.e. kernel consumer), 1139189592Sbms * fake one up and assume it is an ASM join. 1140189592Sbms */ 1141189592Sbms if (imf == NULL) { 1142189592Sbms imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1143189592Sbms imf = &timf; 1144170613Sbms } 1145170613Sbms 1146189592Sbms error = in_getmulti(ifp, gina, &inm); 1147189592Sbms if (error) { 1148189592Sbms CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__); 1149189592Sbms return (error); 1150189592Sbms } 1151189592Sbms 1152189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1153189592Sbms error = inm_merge(inm, imf); 1154189592Sbms if (error) { 1155189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1156189592Sbms goto out_inm_release; 1157189592Sbms } 1158189592Sbms 1159189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1160189592Sbms error = igmp_change_state(inm); 1161189592Sbms if (error) { 1162189592Sbms CTR1(KTR_IGMPV3, "%s: failed to update source", __func__); 1163189592Sbms goto out_inm_release; 1164189592Sbms } 1165189592Sbms 1166189592Sbmsout_inm_release: 1167189592Sbms if (error) { 1168189592Sbms CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1169189592Sbms inm_release_locked(inm); 1170189592Sbms } else { 1171189592Sbms *pinm = inm; 1172189592Sbms } 1173189592Sbms 1174189592Sbms return (error); 1175189592Sbms} 1176189592Sbms 1177189592Sbms/* 1178189592Sbms * Leave a multicast group; unlocked entry point. 1179189592Sbms */ 1180189592Sbmsint 1181189592Sbmsin_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1182189592Sbms{ 1183189592Sbms struct ifnet *ifp; 1184189592Sbms int detached, error; 1185189592Sbms 1186189592Sbms detached = inm_is_ifp_detached(inm); 1187189592Sbms ifp = inm->inm_ifp; 1188189592Sbms if (!detached) 1189189592Sbms IFF_LOCKGIANT(ifp); 1190189592Sbms 1191170613Sbms IN_MULTI_LOCK(); 1192189592Sbms error = in_leavegroup_locked(inm, imf); 1193170613Sbms IN_MULTI_UNLOCK(); 1194170613Sbms 1195189592Sbms if (!detached) 1196170613Sbms IFF_UNLOCKGIANT(ifp); 1197189592Sbms 1198189592Sbms return (error); 1199170613Sbms} 1200170613Sbms 1201170613Sbms/* 1202189592Sbms * Leave a multicast group; real entry point. 1203189592Sbms * All source filters will be expunged. 1204170613Sbms * 1205189592Sbms * Only preserves atomicity at inm level. 1206189592Sbms * 1207189592Sbms * Holding the write lock for the INP which contains imf 1208189592Sbms * is highly advisable. We can't assert for it as imf does not 1209189592Sbms * contain a back-pointer to the owning inp. 1210189592Sbms * 1211189592Sbms * Note: This is not the same as inm_release(*) as this function also 1212189592Sbms * makes a state change downcall into IGMP. 1213170613Sbms */ 1214189592Sbmsint 1215189592Sbmsin_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1216170613Sbms{ 1217189592Sbms struct in_mfilter timf; 1218189592Sbms int error; 1219170613Sbms 1220189592Sbms error = 0; 1221189592Sbms 1222189592Sbms#if defined(INVARIANTS) && defined(IFF_ASSERTGIANT) 1223189592Sbms if (!inm_is_ifp_detached(inm)) 1224189592Sbms IFF_ASSERTGIANT(inm->inm_ifp); 1225189592Sbms#endif 1226189592Sbms 1227170613Sbms IN_MULTI_LOCK_ASSERT(); 1228170613Sbms 1229189592Sbms CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__, 1230189592Sbms inm, inet_ntoa(inm->inm_addr), 1231189592Sbms (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname), 1232189592Sbms imf); 1233170613Sbms 1234189592Sbms /* 1235189592Sbms * If no imf was specified (i.e. kernel consumer), 1236189592Sbms * fake one up and assume it is an ASM join. 1237189592Sbms */ 1238189592Sbms if (imf == NULL) { 1239189592Sbms imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1240189592Sbms imf = &timf; 1241189592Sbms } 1242170613Sbms 1243189592Sbms /* 1244189592Sbms * Begin state merge transaction at IGMP layer. 1245189592Sbms * 1246189592Sbms * As this particular invocation should not cause any memory 1247189592Sbms * to be allocated, and there is no opportunity to roll back 1248189592Sbms * the transaction, it MUST NOT fail. 1249189592Sbms */ 1250189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1251189592Sbms error = inm_merge(inm, imf); 1252189592Sbms KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1253170613Sbms 1254189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1255189592Sbms error = igmp_change_state(inm); 1256189592Sbms if (error) 1257189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1258189592Sbms 1259189592Sbms CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1260189592Sbms inm_release_locked(inm); 1261189592Sbms 1262189592Sbms return (error); 1263170613Sbms} 1264170613Sbms 1265189592Sbms/*#ifndef BURN_BRIDGES*/ 1266170613Sbms/* 1267189592Sbms * Join an IPv4 multicast group in (*,G) exclusive mode. 1268189592Sbms * The group must be a 224.0.0.0/24 link-scope group. 1269189592Sbms * This KPI is for legacy kernel consumers only. 1270170613Sbms */ 1271189592Sbmsstruct in_multi * 1272189592Sbmsin_addmulti(struct in_addr *ap, struct ifnet *ifp) 1273189592Sbms{ 1274189592Sbms struct in_multi *pinm; 1275189592Sbms int error; 1276189592Sbms 1277189592Sbms KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)), 1278189592Sbms ("%s: %s not in 224.0.0.0/24", __func__, inet_ntoa(*ap))); 1279189592Sbms 1280189592Sbms error = in_joingroup(ifp, ap, NULL, &pinm); 1281189592Sbms if (error != 0) 1282189592Sbms pinm = NULL; 1283189592Sbms 1284189592Sbms return (pinm); 1285189592Sbms} 1286189592Sbms 1287189592Sbms/* 1288189592Sbms * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode. 1289189592Sbms * This KPI is for legacy kernel consumers only. 1290189592Sbms */ 1291189592Sbmsvoid 1292189592Sbmsin_delmulti(struct in_multi *inm) 1293189592Sbms{ 1294189592Sbms 1295189592Sbms (void)in_leavegroup(inm, NULL); 1296189592Sbms} 1297189592Sbms/*#endif*/ 1298189592Sbms 1299189592Sbms/* 1300189592Sbms * Block or unblock an ASM multicast source on an inpcb. 1301189592Sbms * This implements the delta-based API described in RFC 3678. 1302189592Sbms * 1303189592Sbms * The delta-based API applies only to exclusive-mode memberships. 1304189592Sbms * An IGMP downcall will be performed. 1305189592Sbms * 1306189592Sbms * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1307189592Sbms * 1308189592Sbms * Return 0 if successful, otherwise return an appropriate error code. 1309189592Sbms */ 1310170613Sbmsstatic int 1311189592Sbmsinp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1312170613Sbms{ 1313183550Szec INIT_VNET_NET(curvnet); 1314183550Szec INIT_VNET_INET(curvnet); 1315170613Sbms struct group_source_req gsr; 1316170613Sbms sockunion_t *gsa, *ssa; 1317170613Sbms struct ifnet *ifp; 1318170613Sbms struct in_mfilter *imf; 1319170613Sbms struct ip_moptions *imo; 1320170613Sbms struct in_msource *ims; 1321189592Sbms struct in_multi *inm; 1322170613Sbms size_t idx; 1323189592Sbms uint16_t fmode; 1324189592Sbms int error, doblock; 1325170613Sbms 1326170613Sbms ifp = NULL; 1327170613Sbms error = 0; 1328189592Sbms doblock = 0; 1329170613Sbms 1330170613Sbms memset(&gsr, 0, sizeof(struct group_source_req)); 1331170613Sbms gsa = (sockunion_t *)&gsr.gsr_group; 1332170613Sbms ssa = (sockunion_t *)&gsr.gsr_source; 1333170613Sbms 1334170613Sbms switch (sopt->sopt_name) { 1335170613Sbms case IP_BLOCK_SOURCE: 1336170613Sbms case IP_UNBLOCK_SOURCE: { 1337170613Sbms struct ip_mreq_source mreqs; 1338170613Sbms 1339170613Sbms error = sooptcopyin(sopt, &mreqs, 1340170613Sbms sizeof(struct ip_mreq_source), 1341170613Sbms sizeof(struct ip_mreq_source)); 1342170613Sbms if (error) 1343170613Sbms return (error); 1344170613Sbms 1345170613Sbms gsa->sin.sin_family = AF_INET; 1346170613Sbms gsa->sin.sin_len = sizeof(struct sockaddr_in); 1347170613Sbms gsa->sin.sin_addr = mreqs.imr_multiaddr; 1348170613Sbms 1349170613Sbms ssa->sin.sin_family = AF_INET; 1350170613Sbms ssa->sin.sin_len = sizeof(struct sockaddr_in); 1351170613Sbms ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1352170613Sbms 1353189592Sbms if (!in_nullhost(mreqs.imr_interface)) 1354170613Sbms INADDR_TO_IFP(mreqs.imr_interface, ifp); 1355170613Sbms 1356170613Sbms if (sopt->sopt_name == IP_BLOCK_SOURCE) 1357189592Sbms doblock = 1; 1358170613Sbms 1359189592Sbms CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p", 1360189592Sbms __func__, inet_ntoa(mreqs.imr_interface), ifp); 1361170613Sbms break; 1362170613Sbms } 1363170613Sbms 1364170613Sbms case MCAST_BLOCK_SOURCE: 1365170613Sbms case MCAST_UNBLOCK_SOURCE: 1366170613Sbms error = sooptcopyin(sopt, &gsr, 1367170613Sbms sizeof(struct group_source_req), 1368170613Sbms sizeof(struct group_source_req)); 1369170613Sbms if (error) 1370170613Sbms return (error); 1371170613Sbms 1372170613Sbms if (gsa->sin.sin_family != AF_INET || 1373170613Sbms gsa->sin.sin_len != sizeof(struct sockaddr_in)) 1374170613Sbms return (EINVAL); 1375170613Sbms 1376170613Sbms if (ssa->sin.sin_family != AF_INET || 1377170613Sbms ssa->sin.sin_len != sizeof(struct sockaddr_in)) 1378170613Sbms return (EINVAL); 1379170613Sbms 1380181803Sbz if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1381170613Sbms return (EADDRNOTAVAIL); 1382170613Sbms 1383170613Sbms ifp = ifnet_byindex(gsr.gsr_interface); 1384170613Sbms 1385170613Sbms if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1386189592Sbms doblock = 1; 1387170613Sbms break; 1388170613Sbms 1389170613Sbms default: 1390189592Sbms CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1391189592Sbms __func__, sopt->sopt_name); 1392170613Sbms return (EOPNOTSUPP); 1393170613Sbms break; 1394170613Sbms } 1395170613Sbms 1396170613Sbms if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1397170613Sbms return (EINVAL); 1398170613Sbms 1399189592Sbms IFF_LOCKGIANT(ifp); 1400189592Sbms 1401170613Sbms /* 1402170613Sbms * Check if we are actually a member of this group. 1403170613Sbms */ 1404170613Sbms imo = inp_findmoptions(inp); 1405170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 1406170613Sbms if (idx == -1 || imo->imo_mfilters == NULL) { 1407170613Sbms error = EADDRNOTAVAIL; 1408189592Sbms goto out_inp_locked; 1409170613Sbms } 1410170613Sbms 1411170613Sbms KASSERT(imo->imo_mfilters != NULL, 1412170613Sbms ("%s: imo_mfilters not allocated", __func__)); 1413170613Sbms imf = &imo->imo_mfilters[idx]; 1414189592Sbms inm = imo->imo_membership[idx]; 1415170613Sbms 1416170613Sbms /* 1417189592Sbms * Attempting to use the delta-based API on an 1418189592Sbms * non exclusive-mode membership is an error. 1419170613Sbms */ 1420189592Sbms fmode = imf->imf_st[0]; 1421189592Sbms if (fmode != MCAST_EXCLUDE) { 1422189592Sbms error = EINVAL; 1423189592Sbms goto out_inp_locked; 1424170613Sbms } 1425189592Sbms 1426189592Sbms /* 1427189592Sbms * Deal with error cases up-front: 1428189592Sbms * Asked to block, but already blocked; or 1429189592Sbms * Asked to unblock, but nothing to unblock. 1430189592Sbms * If adding a new block entry, allocate it. 1431189592Sbms */ 1432170613Sbms ims = imo_match_source(imo, idx, &ssa->sa); 1433189592Sbms if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1434189592Sbms CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__, 1435189592Sbms inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not "); 1436189592Sbms error = EADDRNOTAVAIL; 1437189592Sbms goto out_inp_locked; 1438189592Sbms } 1439189592Sbms 1440189592Sbms INP_WLOCK_ASSERT(inp); 1441189592Sbms 1442189592Sbms /* 1443189592Sbms * Begin state merge transaction at socket layer. 1444189592Sbms */ 1445189592Sbms if (doblock) { 1446189592Sbms CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 1447189592Sbms ims = imf_graft(imf, fmode, &ssa->sin); 1448189592Sbms if (ims == NULL) 1449189592Sbms error = ENOMEM; 1450170613Sbms } else { 1451189592Sbms CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 1452189592Sbms error = imf_prune(imf, &ssa->sin); 1453170613Sbms } 1454170613Sbms 1455189592Sbms if (error) { 1456189592Sbms CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__); 1457189592Sbms goto out_imf_rollback; 1458189592Sbms } 1459189592Sbms 1460189592Sbms /* 1461189592Sbms * Begin state merge transaction at IGMP layer. 1462189592Sbms */ 1463189592Sbms IN_MULTI_LOCK(); 1464189592Sbms 1465189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1466189592Sbms error = inm_merge(inm, imf); 1467189592Sbms if (error) { 1468189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1469189592Sbms goto out_imf_rollback; 1470189592Sbms } 1471189592Sbms 1472189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1473189592Sbms error = igmp_change_state(inm); 1474189592Sbms if (error) 1475189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1476189592Sbms 1477189592Sbms IN_MULTI_UNLOCK(); 1478189592Sbms 1479189592Sbmsout_imf_rollback: 1480189592Sbms if (error) 1481189592Sbms imf_rollback(imf); 1482189592Sbms else 1483189592Sbms imf_commit(imf); 1484189592Sbms 1485189592Sbms imf_reap(imf); 1486189592Sbms 1487189592Sbmsout_inp_locked: 1488178285Srwatson INP_WUNLOCK(inp); 1489189592Sbms IFF_UNLOCKGIANT(ifp); 1490170613Sbms return (error); 1491170613Sbms} 1492170613Sbms 1493170613Sbms/* 1494170613Sbms * Given an inpcb, return its multicast options structure pointer. Accepts 1495170613Sbms * an unlocked inpcb pointer, but will return it locked. May sleep. 1496189592Sbms * 1497189592Sbms * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1498189592Sbms * SMPng: NOTE: Returns with the INP write lock held. 1499170613Sbms */ 1500170613Sbmsstatic struct ip_moptions * 1501170613Sbmsinp_findmoptions(struct inpcb *inp) 1502170613Sbms{ 1503170613Sbms struct ip_moptions *imo; 1504170613Sbms struct in_multi **immp; 1505170613Sbms struct in_mfilter *imfp; 1506170613Sbms size_t idx; 1507170613Sbms 1508178285Srwatson INP_WLOCK(inp); 1509170613Sbms if (inp->inp_moptions != NULL) 1510170613Sbms return (inp->inp_moptions); 1511170613Sbms 1512178285Srwatson INP_WUNLOCK(inp); 1513170613Sbms 1514189592Sbms imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK); 1515189592Sbms immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS, 1516189592Sbms M_WAITOK | M_ZERO); 1517189592Sbms imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS, 1518189592Sbms M_INMFILTER, M_WAITOK); 1519170613Sbms 1520170613Sbms imo->imo_multicast_ifp = NULL; 1521170613Sbms imo->imo_multicast_addr.s_addr = INADDR_ANY; 1522170613Sbms imo->imo_multicast_vif = -1; 1523170613Sbms imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL; 1524189357Sbms imo->imo_multicast_loop = in_mcast_loop; 1525170613Sbms imo->imo_num_memberships = 0; 1526170613Sbms imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; 1527170613Sbms imo->imo_membership = immp; 1528170613Sbms 1529170613Sbms /* Initialize per-group source filters. */ 1530189592Sbms for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++) 1531189592Sbms imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE); 1532170613Sbms imo->imo_mfilters = imfp; 1533170613Sbms 1534178285Srwatson INP_WLOCK(inp); 1535170613Sbms if (inp->inp_moptions != NULL) { 1536189592Sbms free(imfp, M_INMFILTER); 1537170613Sbms free(immp, M_IPMOPTS); 1538170613Sbms free(imo, M_IPMOPTS); 1539170613Sbms return (inp->inp_moptions); 1540170613Sbms } 1541170613Sbms inp->inp_moptions = imo; 1542170613Sbms return (imo); 1543170613Sbms} 1544170613Sbms 1545170613Sbms/* 1546170613Sbms * Discard the IP multicast options (and source filters). 1547189592Sbms * 1548189592Sbms * SMPng: NOTE: assumes INP write lock is held. 1549170613Sbms */ 1550170613Sbmsvoid 1551170613Sbmsinp_freemoptions(struct ip_moptions *imo) 1552170613Sbms{ 1553170613Sbms struct in_mfilter *imf; 1554170613Sbms size_t idx, nmships; 1555170613Sbms 1556170613Sbms KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__)); 1557170613Sbms 1558170613Sbms nmships = imo->imo_num_memberships; 1559170613Sbms for (idx = 0; idx < nmships; ++idx) { 1560189592Sbms imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL; 1561189592Sbms if (imf) 1562189592Sbms imf_leave(imf); 1563189592Sbms (void)in_leavegroup(imo->imo_membership[idx], imf); 1564189592Sbms if (imf) 1565189592Sbms imf_purge(imf); 1566170613Sbms } 1567170613Sbms 1568189592Sbms if (imo->imo_mfilters) 1569189592Sbms free(imo->imo_mfilters, M_INMFILTER); 1570170613Sbms free(imo->imo_membership, M_IPMOPTS); 1571170613Sbms free(imo, M_IPMOPTS); 1572170613Sbms} 1573170613Sbms 1574170613Sbms/* 1575170613Sbms * Atomically get source filters on a socket for an IPv4 multicast group. 1576170613Sbms * Called with INP lock held; returns with lock released. 1577170613Sbms */ 1578170613Sbmsstatic int 1579170613Sbmsinp_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1580170613Sbms{ 1581183550Szec INIT_VNET_NET(curvnet); 1582170613Sbms struct __msfilterreq msfr; 1583170613Sbms sockunion_t *gsa; 1584170613Sbms struct ifnet *ifp; 1585170613Sbms struct ip_moptions *imo; 1586170613Sbms struct in_mfilter *imf; 1587189592Sbms struct ip_msource *ims; 1588189592Sbms struct in_msource *lims; 1589189592Sbms struct sockaddr_in *psin; 1590170613Sbms struct sockaddr_storage *ptss; 1591170613Sbms struct sockaddr_storage *tss; 1592170613Sbms int error; 1593189592Sbms size_t idx, nsrcs, ncsrcs; 1594170613Sbms 1595178285Srwatson INP_WLOCK_ASSERT(inp); 1596170613Sbms 1597170613Sbms imo = inp->inp_moptions; 1598170613Sbms KASSERT(imo != NULL, ("%s: null ip_moptions", __func__)); 1599170613Sbms 1600178285Srwatson INP_WUNLOCK(inp); 1601170613Sbms 1602170613Sbms error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1603170613Sbms sizeof(struct __msfilterreq)); 1604170613Sbms if (error) 1605170613Sbms return (error); 1606170613Sbms 1607181803Sbz if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 1608170613Sbms return (EINVAL); 1609170613Sbms 1610170613Sbms ifp = ifnet_byindex(msfr.msfr_ifindex); 1611170613Sbms if (ifp == NULL) 1612170613Sbms return (EINVAL); 1613170613Sbms 1614178285Srwatson INP_WLOCK(inp); 1615170613Sbms 1616170613Sbms /* 1617170613Sbms * Lookup group on the socket. 1618170613Sbms */ 1619170613Sbms gsa = (sockunion_t *)&msfr.msfr_group; 1620170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 1621170613Sbms if (idx == -1 || imo->imo_mfilters == NULL) { 1622178285Srwatson INP_WUNLOCK(inp); 1623170613Sbms return (EADDRNOTAVAIL); 1624170613Sbms } 1625170613Sbms imf = &imo->imo_mfilters[idx]; 1626170613Sbms 1627170613Sbms /* 1628189592Sbms * Ignore memberships which are in limbo. 1629189592Sbms */ 1630189592Sbms if (imf->imf_st[1] == MCAST_UNDEFINED) { 1631189592Sbms INP_WUNLOCK(inp); 1632189592Sbms return (EAGAIN); 1633189592Sbms } 1634189592Sbms msfr.msfr_fmode = imf->imf_st[1]; 1635189592Sbms 1636189592Sbms /* 1637170613Sbms * If the user specified a buffer, copy out the source filter 1638170613Sbms * entries to userland gracefully. 1639189592Sbms * We only copy out the number of entries which userland 1640189592Sbms * has asked for, but we always tell userland how big the 1641189592Sbms * buffer really needs to be. 1642170613Sbms */ 1643170613Sbms tss = NULL; 1644170613Sbms if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1645184214Sdes tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1646189592Sbms M_TEMP, M_NOWAIT | M_ZERO); 1647170613Sbms if (tss == NULL) { 1648189592Sbms INP_WUNLOCK(inp); 1649189592Sbms return (ENOBUFS); 1650170613Sbms } 1651170613Sbms } 1652170613Sbms 1653189592Sbms /* 1654189592Sbms * Count number of sources in-mode at t0. 1655189592Sbms * If buffer space exists and remains, copy out source entries. 1656189592Sbms */ 1657189592Sbms nsrcs = msfr.msfr_nsrcs; 1658189592Sbms ncsrcs = 0; 1659189592Sbms ptss = tss; 1660189592Sbms RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1661189592Sbms lims = (struct in_msource *)ims; 1662189592Sbms if (lims->imsl_st[0] == MCAST_UNDEFINED || 1663189592Sbms lims->imsl_st[0] != imf->imf_st[0]) 1664189592Sbms continue; 1665189592Sbms ++ncsrcs; 1666189592Sbms if (tss != NULL && nsrcs-- > 0) { 1667189592Sbms psin = (struct sockaddr_in *)ptss++; 1668189592Sbms psin->sin_family = AF_INET; 1669189592Sbms psin->sin_len = sizeof(struct sockaddr_in); 1670189592Sbms psin->sin_addr.s_addr = htonl(lims->ims_haddr); 1671189592Sbms } 1672189592Sbms } 1673189592Sbms 1674178285Srwatson INP_WUNLOCK(inp); 1675170613Sbms 1676170613Sbms if (tss != NULL) { 1677170613Sbms error = copyout(tss, msfr.msfr_srcs, 1678170613Sbms sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1679184205Sdes free(tss, M_TEMP); 1680189592Sbms if (error) 1681189592Sbms return (error); 1682170613Sbms } 1683170613Sbms 1684189592Sbms msfr.msfr_nsrcs = ncsrcs; 1685170613Sbms error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1686170613Sbms 1687170613Sbms return (error); 1688170613Sbms} 1689170613Sbms 1690170613Sbms/* 1691170613Sbms * Return the IP multicast options in response to user getsockopt(). 1692170613Sbms */ 1693170613Sbmsint 1694170613Sbmsinp_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1695170613Sbms{ 1696183550Szec INIT_VNET_INET(curvnet); 1697170613Sbms struct ip_mreqn mreqn; 1698170613Sbms struct ip_moptions *imo; 1699170613Sbms struct ifnet *ifp; 1700170613Sbms struct in_ifaddr *ia; 1701170613Sbms int error, optval; 1702170613Sbms u_char coptval; 1703170613Sbms 1704178285Srwatson INP_WLOCK(inp); 1705170613Sbms imo = inp->inp_moptions; 1706171746Scsjp /* 1707171746Scsjp * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 1708171746Scsjp * or is a divert socket, reject it. 1709171746Scsjp */ 1710171746Scsjp if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 1711171746Scsjp (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1712171746Scsjp inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) { 1713178285Srwatson INP_WUNLOCK(inp); 1714171746Scsjp return (EOPNOTSUPP); 1715171746Scsjp } 1716170613Sbms 1717170613Sbms error = 0; 1718170613Sbms switch (sopt->sopt_name) { 1719170613Sbms case IP_MULTICAST_VIF: 1720170613Sbms if (imo != NULL) 1721170613Sbms optval = imo->imo_multicast_vif; 1722170613Sbms else 1723170613Sbms optval = -1; 1724178285Srwatson INP_WUNLOCK(inp); 1725170613Sbms error = sooptcopyout(sopt, &optval, sizeof(int)); 1726170613Sbms break; 1727170613Sbms 1728170613Sbms case IP_MULTICAST_IF: 1729170613Sbms memset(&mreqn, 0, sizeof(struct ip_mreqn)); 1730170613Sbms if (imo != NULL) { 1731170613Sbms ifp = imo->imo_multicast_ifp; 1732189592Sbms if (!in_nullhost(imo->imo_multicast_addr)) { 1733170613Sbms mreqn.imr_address = imo->imo_multicast_addr; 1734170613Sbms } else if (ifp != NULL) { 1735170613Sbms mreqn.imr_ifindex = ifp->if_index; 1736170613Sbms IFP_TO_IA(ifp, ia); 1737170613Sbms if (ia != NULL) { 1738170613Sbms mreqn.imr_address = 1739170613Sbms IA_SIN(ia)->sin_addr; 1740170613Sbms } 1741170613Sbms } 1742170613Sbms } 1743178285Srwatson INP_WUNLOCK(inp); 1744170613Sbms if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 1745170613Sbms error = sooptcopyout(sopt, &mreqn, 1746170613Sbms sizeof(struct ip_mreqn)); 1747170613Sbms } else { 1748170613Sbms error = sooptcopyout(sopt, &mreqn.imr_address, 1749170613Sbms sizeof(struct in_addr)); 1750170613Sbms } 1751170613Sbms break; 1752170613Sbms 1753170613Sbms case IP_MULTICAST_TTL: 1754170613Sbms if (imo == 0) 1755170613Sbms optval = coptval = IP_DEFAULT_MULTICAST_TTL; 1756170613Sbms else 1757170613Sbms optval = coptval = imo->imo_multicast_ttl; 1758178285Srwatson INP_WUNLOCK(inp); 1759170613Sbms if (sopt->sopt_valsize == sizeof(u_char)) 1760170613Sbms error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1761170613Sbms else 1762170613Sbms error = sooptcopyout(sopt, &optval, sizeof(int)); 1763170613Sbms break; 1764170613Sbms 1765170613Sbms case IP_MULTICAST_LOOP: 1766170613Sbms if (imo == 0) 1767170613Sbms optval = coptval = IP_DEFAULT_MULTICAST_LOOP; 1768170613Sbms else 1769170613Sbms optval = coptval = imo->imo_multicast_loop; 1770178285Srwatson INP_WUNLOCK(inp); 1771170613Sbms if (sopt->sopt_valsize == sizeof(u_char)) 1772170613Sbms error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1773170613Sbms else 1774170613Sbms error = sooptcopyout(sopt, &optval, sizeof(int)); 1775170613Sbms break; 1776170613Sbms 1777170613Sbms case IP_MSFILTER: 1778170613Sbms if (imo == NULL) { 1779170613Sbms error = EADDRNOTAVAIL; 1780178285Srwatson INP_WUNLOCK(inp); 1781170613Sbms } else { 1782170613Sbms error = inp_get_source_filters(inp, sopt); 1783170613Sbms } 1784170613Sbms break; 1785170613Sbms 1786170613Sbms default: 1787178285Srwatson INP_WUNLOCK(inp); 1788170613Sbms error = ENOPROTOOPT; 1789170613Sbms break; 1790170613Sbms } 1791170613Sbms 1792170613Sbms INP_UNLOCK_ASSERT(inp); 1793170613Sbms 1794170613Sbms return (error); 1795170613Sbms} 1796170613Sbms 1797170613Sbms/* 1798189592Sbms * Look up the ifnet to use for a multicast group membership, 1799189592Sbms * given the IPv4 address of an interface, and the IPv4 group address. 1800189592Sbms * 1801189592Sbms * This routine exists to support legacy multicast applications 1802189592Sbms * which do not understand that multicast memberships are scoped to 1803189592Sbms * specific physical links in the networking stack, or which need 1804189592Sbms * to join link-scope groups before IPv4 addresses are configured. 1805189592Sbms * 1806189592Sbms * If inp is non-NULL, use this socket's current FIB number for any 1807189592Sbms * required FIB lookup. 1808189592Sbms * If ina is INADDR_ANY, look up the group address in the unicast FIB, 1809189592Sbms * and use its ifp; usually, this points to the default next-hop. 1810189592Sbms * 1811189592Sbms * If the FIB lookup fails, attempt to use the first non-loopback 1812189592Sbms * interface with multicast capability in the system as a 1813189592Sbms * last resort. The legacy IPv4 ASM API requires that we do 1814189592Sbms * this in order to allow groups to be joined when the routing 1815189592Sbms * table has not yet been populated during boot. 1816189592Sbms * 1817189592Sbms * Returns NULL if no ifp could be found. 1818189592Sbms * 1819189592Sbms * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP. 1820189592Sbms * FUTURE: Implement IPv4 source-address selection. 1821189592Sbms */ 1822189592Sbmsstatic struct ifnet * 1823189592Sbmsinp_lookup_mcast_ifp(const struct inpcb *inp, 1824189592Sbms const struct sockaddr_in *gsin, const struct in_addr ina) 1825189592Sbms{ 1826189592Sbms struct ifnet *ifp; 1827189592Sbms 1828189592Sbms KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__)); 1829189592Sbms KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)), 1830189592Sbms ("%s: not multicast", __func__)); 1831189592Sbms 1832189592Sbms ifp = NULL; 1833189592Sbms if (!in_nullhost(ina)) { 1834189592Sbms INADDR_TO_IFP(ina, ifp); 1835189592Sbms } else { 1836189592Sbms struct route ro; 1837189592Sbms 1838189592Sbms ro.ro_rt = NULL; 1839189592Sbms memcpy(&ro.ro_dst, gsin, sizeof(struct sockaddr_in)); 1840189592Sbms in_rtalloc_ign(&ro, 0, inp ? inp->inp_inc.inc_fibnum : 0); 1841189592Sbms if (ro.ro_rt != NULL) { 1842189592Sbms ifp = ro.ro_rt->rt_ifp; 1843189592Sbms KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 1844189592Sbms RTFREE(ro.ro_rt); 1845189592Sbms } else { 1846189592Sbms struct in_ifaddr *ia; 1847189592Sbms struct ifnet *mifp; 1848189592Sbms 1849189592Sbms mifp = NULL; 1850189592Sbms TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1851189592Sbms mifp = ia->ia_ifp; 1852189592Sbms if (!(mifp->if_flags & IFF_LOOPBACK) && 1853189592Sbms (mifp->if_flags & IFF_MULTICAST)) { 1854189592Sbms ifp = mifp; 1855189592Sbms break; 1856189592Sbms } 1857189592Sbms } 1858189592Sbms } 1859189592Sbms } 1860189592Sbms 1861189592Sbms return (ifp); 1862189592Sbms} 1863189592Sbms 1864189592Sbms/* 1865170613Sbms * Join an IPv4 multicast group, possibly with a source. 1866170613Sbms */ 1867170613Sbmsstatic int 1868170613Sbmsinp_join_group(struct inpcb *inp, struct sockopt *sopt) 1869170613Sbms{ 1870183550Szec INIT_VNET_NET(curvnet); 1871183550Szec INIT_VNET_INET(curvnet); 1872170613Sbms struct group_source_req gsr; 1873170613Sbms sockunion_t *gsa, *ssa; 1874170613Sbms struct ifnet *ifp; 1875170613Sbms struct in_mfilter *imf; 1876170613Sbms struct ip_moptions *imo; 1877170613Sbms struct in_multi *inm; 1878189592Sbms struct in_msource *lims; 1879170613Sbms size_t idx; 1880189592Sbms int error, is_new; 1881170613Sbms 1882170613Sbms ifp = NULL; 1883189592Sbms imf = NULL; 1884170613Sbms error = 0; 1885189592Sbms is_new = 0; 1886170613Sbms 1887170613Sbms memset(&gsr, 0, sizeof(struct group_source_req)); 1888170613Sbms gsa = (sockunion_t *)&gsr.gsr_group; 1889170613Sbms gsa->ss.ss_family = AF_UNSPEC; 1890170613Sbms ssa = (sockunion_t *)&gsr.gsr_source; 1891170613Sbms ssa->ss.ss_family = AF_UNSPEC; 1892170613Sbms 1893170613Sbms switch (sopt->sopt_name) { 1894170613Sbms case IP_ADD_MEMBERSHIP: 1895170613Sbms case IP_ADD_SOURCE_MEMBERSHIP: { 1896170613Sbms struct ip_mreq_source mreqs; 1897170613Sbms 1898170613Sbms if (sopt->sopt_name == IP_ADD_MEMBERSHIP) { 1899170613Sbms error = sooptcopyin(sopt, &mreqs, 1900170613Sbms sizeof(struct ip_mreq), 1901170613Sbms sizeof(struct ip_mreq)); 1902170613Sbms /* 1903170613Sbms * Do argument switcharoo from ip_mreq into 1904170613Sbms * ip_mreq_source to avoid using two instances. 1905170613Sbms */ 1906170613Sbms mreqs.imr_interface = mreqs.imr_sourceaddr; 1907170613Sbms mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 1908170613Sbms } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) { 1909170613Sbms error = sooptcopyin(sopt, &mreqs, 1910170613Sbms sizeof(struct ip_mreq_source), 1911170613Sbms sizeof(struct ip_mreq_source)); 1912170613Sbms } 1913170613Sbms if (error) 1914170613Sbms return (error); 1915170613Sbms 1916170613Sbms gsa->sin.sin_family = AF_INET; 1917170613Sbms gsa->sin.sin_len = sizeof(struct sockaddr_in); 1918170613Sbms gsa->sin.sin_addr = mreqs.imr_multiaddr; 1919170613Sbms 1920170613Sbms if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) { 1921170613Sbms ssa->sin.sin_family = AF_INET; 1922170613Sbms ssa->sin.sin_len = sizeof(struct sockaddr_in); 1923170613Sbms ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1924170613Sbms } 1925170613Sbms 1926189592Sbms ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 1927189592Sbms mreqs.imr_interface); 1928189592Sbms CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p", 1929189592Sbms __func__, inet_ntoa(mreqs.imr_interface), ifp); 1930170613Sbms break; 1931170613Sbms } 1932170613Sbms 1933170613Sbms case MCAST_JOIN_GROUP: 1934170613Sbms case MCAST_JOIN_SOURCE_GROUP: 1935170613Sbms if (sopt->sopt_name == MCAST_JOIN_GROUP) { 1936170613Sbms error = sooptcopyin(sopt, &gsr, 1937170613Sbms sizeof(struct group_req), 1938170613Sbms sizeof(struct group_req)); 1939170613Sbms } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 1940170613Sbms error = sooptcopyin(sopt, &gsr, 1941170613Sbms sizeof(struct group_source_req), 1942170613Sbms sizeof(struct group_source_req)); 1943170613Sbms } 1944170613Sbms if (error) 1945170613Sbms return (error); 1946170613Sbms 1947170613Sbms if (gsa->sin.sin_family != AF_INET || 1948170613Sbms gsa->sin.sin_len != sizeof(struct sockaddr_in)) 1949170613Sbms return (EINVAL); 1950170613Sbms 1951170613Sbms /* 1952170613Sbms * Overwrite the port field if present, as the sockaddr 1953170613Sbms * being copied in may be matched with a binary comparison. 1954170613Sbms */ 1955170613Sbms gsa->sin.sin_port = 0; 1956170613Sbms if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 1957170613Sbms if (ssa->sin.sin_family != AF_INET || 1958170613Sbms ssa->sin.sin_len != sizeof(struct sockaddr_in)) 1959170613Sbms return (EINVAL); 1960170613Sbms ssa->sin.sin_port = 0; 1961170613Sbms } 1962170613Sbms 1963181803Sbz if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1964170613Sbms return (EADDRNOTAVAIL); 1965170613Sbms ifp = ifnet_byindex(gsr.gsr_interface); 1966170613Sbms break; 1967170613Sbms 1968170613Sbms default: 1969189592Sbms CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1970189592Sbms __func__, sopt->sopt_name); 1971170613Sbms return (EOPNOTSUPP); 1972170613Sbms break; 1973170613Sbms } 1974170613Sbms 1975170613Sbms if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1976170613Sbms return (EINVAL); 1977170613Sbms 1978170613Sbms if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) 1979170613Sbms return (EADDRNOTAVAIL); 1980170613Sbms 1981189592Sbms IFF_LOCKGIANT(ifp); 1982189592Sbms 1983170613Sbms /* 1984189592Sbms * MCAST_JOIN_SOURCE on an exclusive membership is an error. 1985189592Sbms * On an existing inclusive membership, it just adds the 1986189592Sbms * source to the filter list. 1987170613Sbms */ 1988170613Sbms imo = inp_findmoptions(inp); 1989170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 1990189592Sbms if (idx == -1) { 1991189592Sbms is_new = 1; 1992189592Sbms } else { 1993189592Sbms inm = imo->imo_membership[idx]; 1994189592Sbms imf = &imo->imo_mfilters[idx]; 1995189592Sbms if (ssa->ss.ss_family != AF_UNSPEC && 1996189592Sbms imf->imf_st[1] != MCAST_INCLUDE) { 1997189592Sbms error = EINVAL; 1998189592Sbms goto out_inp_locked; 1999189592Sbms } 2000189592Sbms lims = imo_match_source(imo, idx, &ssa->sa); 2001189592Sbms if (lims != NULL) { 2002170613Sbms error = EADDRNOTAVAIL; 2003189592Sbms goto out_inp_locked; 2004170613Sbms } 2005170613Sbms } 2006170613Sbms 2007170613Sbms /* 2008189592Sbms * Begin state merge transaction at socket layer. 2009170613Sbms */ 2010189592Sbms INP_WLOCK_ASSERT(inp); 2011189592Sbms 2012189592Sbms if (is_new) { 2013189592Sbms if (imo->imo_num_memberships == imo->imo_max_memberships) { 2014189592Sbms error = imo_grow(imo); 2015189592Sbms if (error) 2016189592Sbms goto out_inp_locked; 2017189592Sbms } 2018189592Sbms /* 2019189592Sbms * Allocate the new slot upfront so we can deal with 2020189592Sbms * grafting the new source filter in same code path 2021189592Sbms * as for join-source on existing membership. 2022189592Sbms */ 2023189592Sbms idx = imo->imo_num_memberships; 2024189592Sbms imo->imo_membership[idx] = NULL; 2025189592Sbms imo->imo_num_memberships++; 2026189592Sbms KASSERT(imo->imo_mfilters != NULL, 2027189592Sbms ("%s: imf_mfilters vector was not allocated", __func__)); 2028189592Sbms imf = &imo->imo_mfilters[idx]; 2029189592Sbms KASSERT(RB_EMPTY(&imf->imf_sources), 2030189592Sbms ("%s: imf_sources not empty", __func__)); 2031170613Sbms } 2032170613Sbms 2033170613Sbms /* 2034189592Sbms * Graft new source into filter list for this inpcb's 2035189592Sbms * membership of the group. The in_multi may not have 2036189592Sbms * been allocated yet if this is a new membership. 2037170613Sbms */ 2038189592Sbms if (ssa->ss.ss_family != AF_UNSPEC) { 2039189592Sbms /* Membership starts in IN mode */ 2040189592Sbms if (is_new) { 2041189592Sbms CTR1(KTR_IGMPV3, "%s: new join w/source", __func__); 2042189592Sbms imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE); 2043189592Sbms } else { 2044189592Sbms CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 2045189592Sbms } 2046189592Sbms lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin); 2047189592Sbms if (lims == NULL) { 2048189592Sbms CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2049189592Sbms __func__); 2050189592Sbms error = ENOMEM; 2051189592Sbms goto out_imo_free; 2052189592Sbms } 2053170613Sbms } 2054170613Sbms 2055170613Sbms /* 2056189592Sbms * Begin state merge transaction at IGMP layer. 2057170613Sbms */ 2058189592Sbms IN_MULTI_LOCK(); 2059170613Sbms 2060189592Sbms if (is_new) { 2061189592Sbms error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf, 2062189592Sbms &inm); 2063189592Sbms if (error) 2064189592Sbms goto out_imo_free; 2065189592Sbms imo->imo_membership[idx] = inm; 2066189592Sbms } else { 2067189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2068189592Sbms error = inm_merge(inm, imf); 2069170613Sbms if (error) { 2070189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2071189592Sbms __func__); 2072189592Sbms goto out_imf_rollback; 2073170613Sbms } 2074189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2075189592Sbms error = igmp_change_state(inm); 2076189592Sbms if (error) { 2077189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2078189592Sbms __func__); 2079189592Sbms goto out_imf_rollback; 2080189592Sbms } 2081170613Sbms } 2082170613Sbms 2083189592Sbms IN_MULTI_UNLOCK(); 2084189592Sbms 2085189592Sbmsout_imf_rollback: 2086189592Sbms INP_WLOCK_ASSERT(inp); 2087189592Sbms if (error) { 2088189592Sbms imf_rollback(imf); 2089189592Sbms if (is_new) 2090189592Sbms imf_purge(imf); 2091189592Sbms else 2092189592Sbms imf_reap(imf); 2093189592Sbms } else { 2094189592Sbms imf_commit(imf); 2095189592Sbms } 2096189592Sbms 2097189592Sbmsout_imo_free: 2098189592Sbms if (error && is_new) { 2099189592Sbms imo->imo_membership[idx] = NULL; 2100189592Sbms --imo->imo_num_memberships; 2101189592Sbms } 2102189592Sbms 2103189592Sbmsout_inp_locked: 2104178285Srwatson INP_WUNLOCK(inp); 2105189592Sbms IFF_UNLOCKGIANT(ifp); 2106170613Sbms return (error); 2107170613Sbms} 2108170613Sbms 2109170613Sbms/* 2110170613Sbms * Leave an IPv4 multicast group on an inpcb, possibly with a source. 2111170613Sbms */ 2112170613Sbmsstatic int 2113170613Sbmsinp_leave_group(struct inpcb *inp, struct sockopt *sopt) 2114170613Sbms{ 2115183550Szec INIT_VNET_NET(curvnet); 2116183550Szec INIT_VNET_INET(curvnet); 2117170613Sbms struct group_source_req gsr; 2118170613Sbms struct ip_mreq_source mreqs; 2119170613Sbms sockunion_t *gsa, *ssa; 2120170613Sbms struct ifnet *ifp; 2121170613Sbms struct in_mfilter *imf; 2122170613Sbms struct ip_moptions *imo; 2123189592Sbms struct in_msource *ims; 2124170613Sbms struct in_multi *inm; 2125170613Sbms size_t idx; 2126189592Sbms int error, is_final; 2127170613Sbms 2128170613Sbms ifp = NULL; 2129170613Sbms error = 0; 2130189592Sbms is_final = 1; 2131170613Sbms 2132170613Sbms memset(&gsr, 0, sizeof(struct group_source_req)); 2133170613Sbms gsa = (sockunion_t *)&gsr.gsr_group; 2134170613Sbms gsa->ss.ss_family = AF_UNSPEC; 2135170613Sbms ssa = (sockunion_t *)&gsr.gsr_source; 2136170613Sbms ssa->ss.ss_family = AF_UNSPEC; 2137170613Sbms 2138170613Sbms switch (sopt->sopt_name) { 2139170613Sbms case IP_DROP_MEMBERSHIP: 2140170613Sbms case IP_DROP_SOURCE_MEMBERSHIP: 2141170613Sbms if (sopt->sopt_name == IP_DROP_MEMBERSHIP) { 2142170613Sbms error = sooptcopyin(sopt, &mreqs, 2143170613Sbms sizeof(struct ip_mreq), 2144170613Sbms sizeof(struct ip_mreq)); 2145170613Sbms /* 2146170613Sbms * Swap interface and sourceaddr arguments, 2147170613Sbms * as ip_mreq and ip_mreq_source are laid 2148170613Sbms * out differently. 2149170613Sbms */ 2150170613Sbms mreqs.imr_interface = mreqs.imr_sourceaddr; 2151170613Sbms mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 2152170613Sbms } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2153170613Sbms error = sooptcopyin(sopt, &mreqs, 2154170613Sbms sizeof(struct ip_mreq_source), 2155170613Sbms sizeof(struct ip_mreq_source)); 2156170613Sbms } 2157170613Sbms if (error) 2158170613Sbms return (error); 2159170613Sbms 2160170613Sbms gsa->sin.sin_family = AF_INET; 2161170613Sbms gsa->sin.sin_len = sizeof(struct sockaddr_in); 2162170613Sbms gsa->sin.sin_addr = mreqs.imr_multiaddr; 2163170613Sbms 2164170613Sbms if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2165170613Sbms ssa->sin.sin_family = AF_INET; 2166170613Sbms ssa->sin.sin_len = sizeof(struct sockaddr_in); 2167170613Sbms ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2168170613Sbms } 2169170613Sbms 2170189592Sbms if (!in_nullhost(gsa->sin.sin_addr)) 2171170613Sbms INADDR_TO_IFP(mreqs.imr_interface, ifp); 2172170613Sbms 2173189592Sbms CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p", 2174189592Sbms __func__, inet_ntoa(mreqs.imr_interface), ifp); 2175189592Sbms 2176170613Sbms break; 2177170613Sbms 2178170613Sbms case MCAST_LEAVE_GROUP: 2179170613Sbms case MCAST_LEAVE_SOURCE_GROUP: 2180170613Sbms if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2181170613Sbms error = sooptcopyin(sopt, &gsr, 2182170613Sbms sizeof(struct group_req), 2183170613Sbms sizeof(struct group_req)); 2184170613Sbms } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2185170613Sbms error = sooptcopyin(sopt, &gsr, 2186170613Sbms sizeof(struct group_source_req), 2187170613Sbms sizeof(struct group_source_req)); 2188170613Sbms } 2189170613Sbms if (error) 2190170613Sbms return (error); 2191170613Sbms 2192170613Sbms if (gsa->sin.sin_family != AF_INET || 2193170613Sbms gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2194170613Sbms return (EINVAL); 2195170613Sbms 2196170613Sbms if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2197170613Sbms if (ssa->sin.sin_family != AF_INET || 2198170613Sbms ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2199170613Sbms return (EINVAL); 2200170613Sbms } 2201170613Sbms 2202181803Sbz if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2203170613Sbms return (EADDRNOTAVAIL); 2204170613Sbms 2205170613Sbms ifp = ifnet_byindex(gsr.gsr_interface); 2206170613Sbms break; 2207170613Sbms 2208170613Sbms default: 2209189592Sbms CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2210189592Sbms __func__, sopt->sopt_name); 2211170613Sbms return (EOPNOTSUPP); 2212170613Sbms break; 2213170613Sbms } 2214170613Sbms 2215170613Sbms if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2216170613Sbms return (EINVAL); 2217170613Sbms 2218189592Sbms if (ifp) 2219189592Sbms IFF_LOCKGIANT(ifp); 2220189592Sbms 2221170613Sbms /* 2222170613Sbms * Find the membership in the membership array. 2223170613Sbms */ 2224170613Sbms imo = inp_findmoptions(inp); 2225170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 2226170613Sbms if (idx == -1) { 2227170613Sbms error = EADDRNOTAVAIL; 2228189592Sbms goto out_inp_locked; 2229170613Sbms } 2230189592Sbms inm = imo->imo_membership[idx]; 2231170613Sbms imf = &imo->imo_mfilters[idx]; 2232170613Sbms 2233189592Sbms if (ssa->ss.ss_family != AF_UNSPEC) 2234189592Sbms is_final = 0; 2235189592Sbms 2236170613Sbms /* 2237189592Sbms * Begin state merge transaction at socket layer. 2238189592Sbms */ 2239189592Sbms INP_WLOCK_ASSERT(inp); 2240189592Sbms 2241189592Sbms /* 2242170613Sbms * If we were instructed only to leave a given source, do so. 2243189592Sbms * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2244170613Sbms */ 2245189592Sbms if (is_final) { 2246189592Sbms imf_leave(imf); 2247189592Sbms } else { 2248189592Sbms if (imf->imf_st[0] == MCAST_EXCLUDE) { 2249189592Sbms error = EADDRNOTAVAIL; 2250189592Sbms goto out_inp_locked; 2251170613Sbms } 2252189592Sbms ims = imo_match_source(imo, idx, &ssa->sa); 2253189592Sbms if (ims == NULL) { 2254189592Sbms CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__, 2255189592Sbms inet_ntoa(ssa->sin.sin_addr), "not "); 2256189592Sbms error = EADDRNOTAVAIL; 2257189592Sbms goto out_inp_locked; 2258189592Sbms } 2259189592Sbms CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 2260189592Sbms error = imf_prune(imf, &ssa->sin); 2261189592Sbms if (error) { 2262189592Sbms CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2263189592Sbms __func__); 2264189592Sbms goto out_inp_locked; 2265189592Sbms } 2266170613Sbms } 2267170613Sbms 2268170613Sbms /* 2269189592Sbms * Begin state merge transaction at IGMP layer. 2270170613Sbms */ 2271189592Sbms IN_MULTI_LOCK(); 2272170613Sbms 2273189592Sbms if (is_final) { 2274189592Sbms /* 2275189592Sbms * Give up the multicast address record to which 2276189592Sbms * the membership points. 2277189592Sbms */ 2278189592Sbms (void)in_leavegroup_locked(inm, imf); 2279189592Sbms } else { 2280189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2281189592Sbms error = inm_merge(inm, imf); 2282189592Sbms if (error) { 2283189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2284189592Sbms __func__); 2285189592Sbms goto out_imf_rollback; 2286170613Sbms } 2287189592Sbms 2288189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2289189592Sbms error = igmp_change_state(inm); 2290189592Sbms if (error) { 2291189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2292189592Sbms __func__); 2293189592Sbms } 2294170613Sbms } 2295170613Sbms 2296189592Sbms IN_MULTI_UNLOCK(); 2297170613Sbms 2298189592Sbmsout_imf_rollback: 2299189592Sbms if (error) 2300189592Sbms imf_rollback(imf); 2301189592Sbms else 2302189592Sbms imf_commit(imf); 2303189592Sbms 2304189592Sbms imf_reap(imf); 2305189592Sbms 2306189592Sbms if (is_final) { 2307189592Sbms /* Remove the gap in the membership array. */ 2308189592Sbms for (++idx; idx < imo->imo_num_memberships; ++idx) 2309189592Sbms imo->imo_membership[idx-1] = imo->imo_membership[idx]; 2310189592Sbms imo->imo_num_memberships--; 2311189592Sbms } 2312189592Sbms 2313189592Sbmsout_inp_locked: 2314178285Srwatson INP_WUNLOCK(inp); 2315189592Sbms if (ifp) 2316189592Sbms IFF_UNLOCKGIANT(ifp); 2317170613Sbms return (error); 2318170613Sbms} 2319170613Sbms 2320170613Sbms/* 2321170613Sbms * Select the interface for transmitting IPv4 multicast datagrams. 2322170613Sbms * 2323170613Sbms * Either an instance of struct in_addr or an instance of struct ip_mreqn 2324170613Sbms * may be passed to this socket option. An address of INADDR_ANY or an 2325170613Sbms * interface index of 0 is used to remove a previous selection. 2326170613Sbms * When no interface is selected, one is chosen for every send. 2327170613Sbms */ 2328170613Sbmsstatic int 2329170613Sbmsinp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2330170613Sbms{ 2331183550Szec INIT_VNET_NET(curvnet); 2332170613Sbms struct in_addr addr; 2333170613Sbms struct ip_mreqn mreqn; 2334170613Sbms struct ifnet *ifp; 2335170613Sbms struct ip_moptions *imo; 2336170613Sbms int error; 2337170613Sbms 2338170613Sbms if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 2339170613Sbms /* 2340170613Sbms * An interface index was specified using the 2341170613Sbms * Linux-derived ip_mreqn structure. 2342170613Sbms */ 2343170613Sbms error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn), 2344170613Sbms sizeof(struct ip_mreqn)); 2345170613Sbms if (error) 2346170613Sbms return (error); 2347170613Sbms 2348181803Sbz if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex) 2349170613Sbms return (EINVAL); 2350170613Sbms 2351170613Sbms if (mreqn.imr_ifindex == 0) { 2352170613Sbms ifp = NULL; 2353170613Sbms } else { 2354170613Sbms ifp = ifnet_byindex(mreqn.imr_ifindex); 2355170613Sbms if (ifp == NULL) 2356170613Sbms return (EADDRNOTAVAIL); 2357170613Sbms } 2358170613Sbms } else { 2359170613Sbms /* 2360170613Sbms * An interface was specified by IPv4 address. 2361170613Sbms * This is the traditional BSD usage. 2362170613Sbms */ 2363170613Sbms error = sooptcopyin(sopt, &addr, sizeof(struct in_addr), 2364170613Sbms sizeof(struct in_addr)); 2365170613Sbms if (error) 2366170613Sbms return (error); 2367189592Sbms if (in_nullhost(addr)) { 2368170613Sbms ifp = NULL; 2369170613Sbms } else { 2370170613Sbms INADDR_TO_IFP(addr, ifp); 2371170613Sbms if (ifp == NULL) 2372170613Sbms return (EADDRNOTAVAIL); 2373170613Sbms } 2374189592Sbms CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp, 2375189592Sbms inet_ntoa(addr)); 2376170613Sbms } 2377170613Sbms 2378170613Sbms /* Reject interfaces which do not support multicast. */ 2379170613Sbms if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0) 2380170613Sbms return (EOPNOTSUPP); 2381170613Sbms 2382170613Sbms imo = inp_findmoptions(inp); 2383170613Sbms imo->imo_multicast_ifp = ifp; 2384170613Sbms imo->imo_multicast_addr.s_addr = INADDR_ANY; 2385178285Srwatson INP_WUNLOCK(inp); 2386170613Sbms 2387170613Sbms return (0); 2388170613Sbms} 2389170613Sbms 2390170613Sbms/* 2391170613Sbms * Atomically set source filters on a socket for an IPv4 multicast group. 2392189592Sbms * 2393189592Sbms * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2394170613Sbms */ 2395170613Sbmsstatic int 2396170613Sbmsinp_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2397170613Sbms{ 2398183550Szec INIT_VNET_NET(curvnet); 2399170613Sbms struct __msfilterreq msfr; 2400170613Sbms sockunion_t *gsa; 2401170613Sbms struct ifnet *ifp; 2402170613Sbms struct in_mfilter *imf; 2403170613Sbms struct ip_moptions *imo; 2404189592Sbms struct in_multi *inm; 2405170613Sbms size_t idx; 2406170613Sbms int error; 2407170613Sbms 2408170613Sbms error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2409170613Sbms sizeof(struct __msfilterreq)); 2410170613Sbms if (error) 2411170613Sbms return (error); 2412170613Sbms 2413189592Sbms if (msfr.msfr_nsrcs > in_mcast_maxsocksrc || 2414170613Sbms (msfr.msfr_fmode != MCAST_EXCLUDE && 2415170613Sbms msfr.msfr_fmode != MCAST_INCLUDE)) 2416170613Sbms return (EINVAL); 2417170613Sbms 2418170613Sbms if (msfr.msfr_group.ss_family != AF_INET || 2419170613Sbms msfr.msfr_group.ss_len != sizeof(struct sockaddr_in)) 2420170613Sbms return (EINVAL); 2421170613Sbms 2422170613Sbms gsa = (sockunion_t *)&msfr.msfr_group; 2423170613Sbms if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2424170613Sbms return (EINVAL); 2425170613Sbms 2426170613Sbms gsa->sin.sin_port = 0; /* ignore port */ 2427170613Sbms 2428181803Sbz if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 2429170613Sbms return (EADDRNOTAVAIL); 2430170613Sbms 2431170613Sbms ifp = ifnet_byindex(msfr.msfr_ifindex); 2432170613Sbms if (ifp == NULL) 2433170613Sbms return (EADDRNOTAVAIL); 2434170613Sbms 2435189592Sbms IFF_LOCKGIANT(ifp); 2436189592Sbms 2437170613Sbms /* 2438189592Sbms * Take the INP write lock. 2439170613Sbms * Check if this socket is a member of this group. 2440170613Sbms */ 2441170613Sbms imo = inp_findmoptions(inp); 2442170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 2443170613Sbms if (idx == -1 || imo->imo_mfilters == NULL) { 2444170613Sbms error = EADDRNOTAVAIL; 2445189592Sbms goto out_inp_locked; 2446170613Sbms } 2447189592Sbms inm = imo->imo_membership[idx]; 2448170613Sbms imf = &imo->imo_mfilters[idx]; 2449170613Sbms 2450170613Sbms /* 2451189592Sbms * Begin state merge transaction at socket layer. 2452170613Sbms */ 2453189592Sbms INP_WLOCK_ASSERT(inp); 2454170613Sbms 2455189592Sbms imf->imf_st[1] = msfr.msfr_fmode; 2456189592Sbms 2457170613Sbms /* 2458170613Sbms * Apply any new source filters, if present. 2459189592Sbms * Make a copy of the user-space source vector so 2460189592Sbms * that we may copy them with a single copyin. This 2461189592Sbms * allows us to deal with page faults up-front. 2462170613Sbms */ 2463170613Sbms if (msfr.msfr_nsrcs > 0) { 2464189592Sbms struct in_msource *lims; 2465189592Sbms struct sockaddr_in *psin; 2466189592Sbms struct sockaddr_storage *kss, *pkss; 2467189592Sbms int i; 2468170613Sbms 2469178285Srwatson INP_WUNLOCK(inp); 2470189592Sbms 2471189592Sbms CTR2(KTR_IGMPV3, "%s: loading %lu source list entries", 2472189592Sbms __func__, (unsigned long)msfr.msfr_nsrcs); 2473184214Sdes kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2474170613Sbms M_TEMP, M_WAITOK); 2475170613Sbms error = copyin(msfr.msfr_srcs, kss, 2476170613Sbms sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2477170613Sbms if (error) { 2478184205Sdes free(kss, M_TEMP); 2479170613Sbms return (error); 2480170613Sbms } 2481170613Sbms 2482189592Sbms INP_WLOCK(inp); 2483189592Sbms 2484170613Sbms /* 2485189592Sbms * Mark all source filters as UNDEFINED at t1. 2486189592Sbms * Restore new group filter mode, as imf_leave() 2487189592Sbms * will set it to INCLUDE. 2488170613Sbms */ 2489189592Sbms imf_leave(imf); 2490189592Sbms imf->imf_st[1] = msfr.msfr_fmode; 2491189592Sbms 2492189592Sbms /* 2493189592Sbms * Update socket layer filters at t1, lazy-allocating 2494189592Sbms * new entries. This saves a bunch of memory at the 2495189592Sbms * cost of one RB_FIND() per source entry; duplicate 2496189592Sbms * entries in the msfr_nsrcs vector are ignored. 2497189592Sbms * If we encounter an error, rollback transaction. 2498189592Sbms * 2499189592Sbms * XXX This too could be replaced with a set-symmetric 2500189592Sbms * difference like loop to avoid walking from root 2501189592Sbms * every time, as the key space is common. 2502189592Sbms */ 2503189592Sbms for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2504189592Sbms psin = (struct sockaddr_in *)pkss; 2505189592Sbms if (psin->sin_family != AF_INET) { 2506170613Sbms error = EAFNOSUPPORT; 2507170613Sbms break; 2508170613Sbms } 2509189592Sbms if (psin->sin_len != sizeof(struct sockaddr_in)) { 2510189592Sbms error = EINVAL; 2511189592Sbms break; 2512189592Sbms } 2513189592Sbms error = imf_get_source(imf, psin, &lims); 2514170613Sbms if (error) 2515170613Sbms break; 2516189592Sbms lims->imsl_st[1] = imf->imf_st[1]; 2517170613Sbms } 2518189592Sbms free(kss, M_TEMP); 2519189592Sbms } 2520170613Sbms 2521189592Sbms if (error) 2522189592Sbms goto out_imf_rollback; 2523170613Sbms 2524189592Sbms INP_WLOCK_ASSERT(inp); 2525189592Sbms IN_MULTI_LOCK(); 2526170613Sbms 2527170613Sbms /* 2528189592Sbms * Begin state merge transaction at IGMP layer. 2529170613Sbms */ 2530189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2531189592Sbms error = inm_merge(inm, imf); 2532189592Sbms if (error) { 2533189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 2534189592Sbms goto out_imf_rollback; 2535189592Sbms } 2536170613Sbms 2537189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2538189592Sbms error = igmp_change_state(inm); 2539189592Sbms if (error) 2540189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 2541189592Sbms 2542189592Sbms IN_MULTI_UNLOCK(); 2543189592Sbms 2544189592Sbmsout_imf_rollback: 2545189592Sbms if (error) 2546189592Sbms imf_rollback(imf); 2547189592Sbms else 2548189592Sbms imf_commit(imf); 2549189592Sbms 2550189592Sbms imf_reap(imf); 2551189592Sbms 2552189592Sbmsout_inp_locked: 2553178285Srwatson INP_WUNLOCK(inp); 2554189592Sbms IFF_UNLOCKGIANT(ifp); 2555170613Sbms return (error); 2556170613Sbms} 2557170613Sbms 2558170613Sbms/* 2559170613Sbms * Set the IP multicast options in response to user setsockopt(). 2560170613Sbms * 2561170613Sbms * Many of the socket options handled in this function duplicate the 2562170613Sbms * functionality of socket options in the regular unicast API. However, 2563170613Sbms * it is not possible to merge the duplicate code, because the idempotence 2564170613Sbms * of the IPv4 multicast part of the BSD Sockets API must be preserved; 2565170613Sbms * the effects of these options must be treated as separate and distinct. 2566189592Sbms * 2567189592Sbms * SMPng: XXX: Unlocked read of inp_socket believed OK. 2568189592Sbms * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING 2569189592Sbms * is refactored to no longer use vifs. 2570170613Sbms */ 2571170613Sbmsint 2572170613Sbmsinp_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2573170613Sbms{ 2574170613Sbms struct ip_moptions *imo; 2575170613Sbms int error; 2576170613Sbms 2577170613Sbms error = 0; 2578170613Sbms 2579171746Scsjp /* 2580171746Scsjp * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 2581171746Scsjp * or is a divert socket, reject it. 2582171746Scsjp */ 2583171746Scsjp if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 2584171746Scsjp (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2585189592Sbms inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) 2586171746Scsjp return (EOPNOTSUPP); 2587171746Scsjp 2588170613Sbms switch (sopt->sopt_name) { 2589170613Sbms case IP_MULTICAST_VIF: { 2590170613Sbms int vifi; 2591170613Sbms /* 2592170613Sbms * Select a multicast VIF for transmission. 2593170613Sbms * Only useful if multicast forwarding is active. 2594170613Sbms */ 2595170613Sbms if (legal_vif_num == NULL) { 2596170613Sbms error = EOPNOTSUPP; 2597170613Sbms break; 2598170613Sbms } 2599170613Sbms error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int)); 2600170613Sbms if (error) 2601170613Sbms break; 2602170613Sbms if (!legal_vif_num(vifi) && (vifi != -1)) { 2603170613Sbms error = EINVAL; 2604170613Sbms break; 2605170613Sbms } 2606170613Sbms imo = inp_findmoptions(inp); 2607170613Sbms imo->imo_multicast_vif = vifi; 2608178285Srwatson INP_WUNLOCK(inp); 2609170613Sbms break; 2610170613Sbms } 2611170613Sbms 2612170613Sbms case IP_MULTICAST_IF: 2613170613Sbms error = inp_set_multicast_if(inp, sopt); 2614170613Sbms break; 2615170613Sbms 2616170613Sbms case IP_MULTICAST_TTL: { 2617170613Sbms u_char ttl; 2618170613Sbms 2619170613Sbms /* 2620170613Sbms * Set the IP time-to-live for outgoing multicast packets. 2621170613Sbms * The original multicast API required a char argument, 2622170613Sbms * which is inconsistent with the rest of the socket API. 2623170613Sbms * We allow either a char or an int. 2624170613Sbms */ 2625170613Sbms if (sopt->sopt_valsize == sizeof(u_char)) { 2626170613Sbms error = sooptcopyin(sopt, &ttl, sizeof(u_char), 2627170613Sbms sizeof(u_char)); 2628170613Sbms if (error) 2629170613Sbms break; 2630170613Sbms } else { 2631170613Sbms u_int ittl; 2632170613Sbms 2633170613Sbms error = sooptcopyin(sopt, &ittl, sizeof(u_int), 2634170613Sbms sizeof(u_int)); 2635170613Sbms if (error) 2636170613Sbms break; 2637170613Sbms if (ittl > 255) { 2638170613Sbms error = EINVAL; 2639170613Sbms break; 2640170613Sbms } 2641170613Sbms ttl = (u_char)ittl; 2642170613Sbms } 2643170613Sbms imo = inp_findmoptions(inp); 2644170613Sbms imo->imo_multicast_ttl = ttl; 2645178285Srwatson INP_WUNLOCK(inp); 2646170613Sbms break; 2647170613Sbms } 2648170613Sbms 2649170613Sbms case IP_MULTICAST_LOOP: { 2650170613Sbms u_char loop; 2651170613Sbms 2652170613Sbms /* 2653170613Sbms * Set the loopback flag for outgoing multicast packets. 2654170613Sbms * Must be zero or one. The original multicast API required a 2655170613Sbms * char argument, which is inconsistent with the rest 2656170613Sbms * of the socket API. We allow either a char or an int. 2657170613Sbms */ 2658170613Sbms if (sopt->sopt_valsize == sizeof(u_char)) { 2659170613Sbms error = sooptcopyin(sopt, &loop, sizeof(u_char), 2660170613Sbms sizeof(u_char)); 2661170613Sbms if (error) 2662170613Sbms break; 2663170613Sbms } else { 2664170613Sbms u_int iloop; 2665170613Sbms 2666170613Sbms error = sooptcopyin(sopt, &iloop, sizeof(u_int), 2667170613Sbms sizeof(u_int)); 2668170613Sbms if (error) 2669170613Sbms break; 2670170613Sbms loop = (u_char)iloop; 2671170613Sbms } 2672170613Sbms imo = inp_findmoptions(inp); 2673170613Sbms imo->imo_multicast_loop = !!loop; 2674178285Srwatson INP_WUNLOCK(inp); 2675170613Sbms break; 2676170613Sbms } 2677170613Sbms 2678170613Sbms case IP_ADD_MEMBERSHIP: 2679170613Sbms case IP_ADD_SOURCE_MEMBERSHIP: 2680170613Sbms case MCAST_JOIN_GROUP: 2681170613Sbms case MCAST_JOIN_SOURCE_GROUP: 2682170613Sbms error = inp_join_group(inp, sopt); 2683170613Sbms break; 2684170613Sbms 2685170613Sbms case IP_DROP_MEMBERSHIP: 2686170613Sbms case IP_DROP_SOURCE_MEMBERSHIP: 2687170613Sbms case MCAST_LEAVE_GROUP: 2688170613Sbms case MCAST_LEAVE_SOURCE_GROUP: 2689170613Sbms error = inp_leave_group(inp, sopt); 2690170613Sbms break; 2691170613Sbms 2692170613Sbms case IP_BLOCK_SOURCE: 2693170613Sbms case IP_UNBLOCK_SOURCE: 2694170613Sbms case MCAST_BLOCK_SOURCE: 2695170613Sbms case MCAST_UNBLOCK_SOURCE: 2696189592Sbms error = inp_block_unblock_source(inp, sopt); 2697170613Sbms break; 2698170613Sbms 2699170613Sbms case IP_MSFILTER: 2700170613Sbms error = inp_set_source_filters(inp, sopt); 2701170613Sbms break; 2702170613Sbms 2703170613Sbms default: 2704170613Sbms error = EOPNOTSUPP; 2705170613Sbms break; 2706170613Sbms } 2707170613Sbms 2708170613Sbms INP_UNLOCK_ASSERT(inp); 2709170613Sbms 2710170613Sbms return (error); 2711170613Sbms} 2712189592Sbms 2713189592Sbms/* 2714189592Sbms * Expose IGMP's multicast filter mode and source list(s) to userland, 2715189592Sbms * keyed by (ifindex, group). 2716189592Sbms * The filter mode is written out as a uint32_t, followed by 2717189592Sbms * 0..n of struct in_addr. 2718189592Sbms * For use by ifmcstat(8). 2719189592Sbms * SMPng: NOTE: unlocked read of ifindex space. 2720189592Sbms */ 2721189592Sbmsstatic int 2722189592Sbmssysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS) 2723189592Sbms{ 2724189592Sbms INIT_VNET_NET(curvnet); 2725189592Sbms struct in_addr src, group; 2726189592Sbms struct ifnet *ifp; 2727189592Sbms struct ifmultiaddr *ifma; 2728189592Sbms struct in_multi *inm; 2729189592Sbms struct ip_msource *ims; 2730189592Sbms int *name; 2731189592Sbms int retval; 2732189592Sbms u_int namelen; 2733189592Sbms uint32_t fmode, ifindex; 2734189592Sbms 2735189592Sbms name = (int *)arg1; 2736189592Sbms namelen = arg2; 2737189592Sbms 2738189592Sbms if (req->newptr != NULL) 2739189592Sbms return (EPERM); 2740189592Sbms 2741189592Sbms if (namelen != 2) 2742189592Sbms return (EINVAL); 2743189592Sbms 2744189592Sbms ifindex = name[0]; 2745189592Sbms if (ifindex <= 0 || ifindex > V_if_index) { 2746189592Sbms CTR2(KTR_IGMPV3, "%s: ifindex %u out of range", 2747189592Sbms __func__, ifindex); 2748189592Sbms return (ENOENT); 2749189592Sbms } 2750189592Sbms 2751189592Sbms group.s_addr = name[1]; 2752189592Sbms if (!IN_MULTICAST(ntohl(group.s_addr))) { 2753189592Sbms CTR2(KTR_IGMPV3, "%s: group %s is not multicast", 2754189592Sbms __func__, inet_ntoa(group)); 2755189592Sbms return (EINVAL); 2756189592Sbms } 2757189592Sbms 2758189592Sbms ifp = ifnet_byindex(ifindex); 2759189592Sbms if (ifp == NULL) { 2760189592Sbms CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u", 2761189592Sbms __func__, ifindex); 2762189592Sbms return (ENOENT); 2763189592Sbms } 2764189592Sbms 2765189592Sbms retval = sysctl_wire_old_buffer(req, 2766189592Sbms sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr))); 2767189592Sbms if (retval) 2768189592Sbms return (retval); 2769189592Sbms 2770189592Sbms IN_MULTI_LOCK(); 2771189592Sbms 2772189592Sbms IF_ADDR_LOCK(ifp); 2773189592Sbms TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2774189592Sbms if (ifma->ifma_addr->sa_family != AF_INET || 2775189592Sbms ifma->ifma_protospec == NULL) 2776189592Sbms continue; 2777189592Sbms inm = (struct in_multi *)ifma->ifma_protospec; 2778189592Sbms if (!in_hosteq(inm->inm_addr, group)) 2779189592Sbms continue; 2780189592Sbms fmode = inm->inm_st[1].iss_fmode; 2781189592Sbms retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 2782189592Sbms if (retval != 0) 2783189592Sbms break; 2784189592Sbms RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 2785189592Sbms#ifdef KTR 2786189592Sbms struct in_addr ina; 2787189592Sbms ina.s_addr = htonl(ims->ims_haddr); 2788189592Sbms CTR2(KTR_IGMPV3, "%s: visit node %s", __func__, 2789189592Sbms inet_ntoa(ina)); 2790189592Sbms#endif 2791189592Sbms /* 2792189592Sbms * Only copy-out sources which are in-mode. 2793189592Sbms */ 2794189592Sbms if (fmode != ims_get_mode(inm, ims, 1)) { 2795189592Sbms CTR1(KTR_IGMPV3, "%s: skip non-in-mode", 2796189592Sbms __func__); 2797189592Sbms continue; 2798189592Sbms } 2799189592Sbms src.s_addr = htonl(ims->ims_haddr); 2800189592Sbms retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr)); 2801189592Sbms if (retval != 0) 2802189592Sbms break; 2803189592Sbms } 2804189592Sbms } 2805189592Sbms IF_ADDR_UNLOCK(ifp); 2806189592Sbms 2807189592Sbms IN_MULTI_UNLOCK(); 2808189592Sbms 2809189592Sbms return (retval); 2810189592Sbms} 2811189592Sbms 2812189592Sbms#ifdef KTR 2813189592Sbms 2814189592Sbmsstatic const char *inm_modestrs[] = { "un", "in", "ex" }; 2815189592Sbms 2816189592Sbmsstatic const char * 2817189592Sbmsinm_mode_str(const int mode) 2818189592Sbms{ 2819189592Sbms 2820189592Sbms if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 2821189592Sbms return (inm_modestrs[mode]); 2822189592Sbms return ("??"); 2823189592Sbms} 2824189592Sbms 2825189592Sbmsstatic const char *inm_statestrs[] = { 2826189592Sbms "not-member", 2827189592Sbms "silent", 2828189592Sbms "idle", 2829189592Sbms "lazy", 2830189592Sbms "sleeping", 2831189592Sbms "awakening", 2832189592Sbms "query-pending", 2833189592Sbms "sg-query-pending", 2834189592Sbms "leaving" 2835189592Sbms}; 2836189592Sbms 2837189592Sbmsstatic const char * 2838189592Sbmsinm_state_str(const int state) 2839189592Sbms{ 2840189592Sbms 2841189592Sbms if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER) 2842189592Sbms return (inm_statestrs[state]); 2843189592Sbms return ("??"); 2844189592Sbms} 2845189592Sbms 2846189592Sbms/* 2847189592Sbms * Dump an in_multi structure to the console. 2848189592Sbms */ 2849189592Sbmsvoid 2850189592Sbmsinm_print(const struct in_multi *inm) 2851189592Sbms{ 2852189592Sbms int t; 2853189592Sbms 2854189635Sbms if ((KTR_COMPILE & KTR_IGMPV3) == 0) 2855189635Sbms return; 2856189635Sbms 2857189592Sbms printf("%s: --- begin inm %p ---\n", __func__, inm); 2858189592Sbms printf("addr %s ifp %p(%s) ifma %p\n", 2859189592Sbms inet_ntoa(inm->inm_addr), 2860189592Sbms inm->inm_ifp, 2861189592Sbms inm->inm_ifp->if_xname, 2862189592Sbms inm->inm_ifma); 2863189592Sbms printf("timer %u state %s refcount %u scq.len %u\n", 2864189592Sbms inm->inm_timer, 2865189592Sbms inm_state_str(inm->inm_state), 2866189592Sbms inm->inm_refcount, 2867189592Sbms inm->inm_scq.ifq_len); 2868189592Sbms printf("igi %p nsrc %lu sctimer %u scrv %u\n", 2869189592Sbms inm->inm_igi, 2870189592Sbms inm->inm_nsrc, 2871189592Sbms inm->inm_sctimer, 2872189592Sbms inm->inm_scrv); 2873189592Sbms for (t = 0; t < 2; t++) { 2874189592Sbms printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 2875189592Sbms inm_mode_str(inm->inm_st[t].iss_fmode), 2876189592Sbms inm->inm_st[t].iss_asm, 2877189592Sbms inm->inm_st[t].iss_ex, 2878189592Sbms inm->inm_st[t].iss_in, 2879189592Sbms inm->inm_st[t].iss_rec); 2880189592Sbms } 2881189592Sbms printf("%s: --- end inm %p ---\n", __func__, inm); 2882189592Sbms} 2883189592Sbms 2884189592Sbms#else /* !KTR */ 2885189592Sbms 2886189592Sbmsvoid 2887189592Sbmsinm_print(const struct in_multi *inm) 2888189592Sbms{ 2889189592Sbms 2890189592Sbms} 2891189592Sbms 2892189592Sbms#endif /* KTR */ 2893189592Sbms 2894189592SbmsRB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp); 2895