in_mcast.c revision 196932
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 196932 2009-09-07 16:00:33Z syrinx $"); 37170613Sbms 38170613Sbms#include <sys/param.h> 39170613Sbms#include <sys/systm.h> 40170613Sbms#include <sys/kernel.h> 41170613Sbms#include <sys/malloc.h> 42170613Sbms#include <sys/mbuf.h> 43171746Scsjp#include <sys/protosw.h> 44170613Sbms#include <sys/socket.h> 45170613Sbms#include <sys/socketvar.h> 46189592Sbms#include <sys/protosw.h> 47170613Sbms#include <sys/sysctl.h> 48189592Sbms#include <sys/ktr.h> 49189592Sbms#include <sys/tree.h> 50170613Sbms 51170613Sbms#include <net/if.h> 52170613Sbms#include <net/if_dl.h> 53170613Sbms#include <net/route.h> 54185571Sbz#include <net/vnet.h> 55170613Sbms 56170613Sbms#include <netinet/in.h> 57170613Sbms#include <netinet/in_systm.h> 58170613Sbms#include <netinet/in_pcb.h> 59170613Sbms#include <netinet/in_var.h> 60170613Sbms#include <netinet/ip_var.h> 61170613Sbms#include <netinet/igmp_var.h> 62170613Sbms 63189592Sbms#ifndef KTR_IGMPV3 64191659Sbms#define KTR_IGMPV3 KTR_INET 65189592Sbms#endif 66189592Sbms 67170613Sbms#ifndef __SOCKUNION_DECLARED 68170613Sbmsunion sockunion { 69170613Sbms struct sockaddr_storage ss; 70170613Sbms struct sockaddr sa; 71170613Sbms struct sockaddr_dl sdl; 72170613Sbms struct sockaddr_in sin; 73170613Sbms}; 74170613Sbmstypedef union sockunion sockunion_t; 75170613Sbms#define __SOCKUNION_DECLARED 76170613Sbms#endif /* __SOCKUNION_DECLARED */ 77170613Sbms 78189592Sbmsstatic MALLOC_DEFINE(M_INMFILTER, "in_mfilter", 79189592Sbms "IPv4 multicast PCB-layer source filter"); 80170613Sbmsstatic MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group"); 81170613Sbmsstatic MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options"); 82189592Sbmsstatic MALLOC_DEFINE(M_IPMSOURCE, "ip_msource", 83189592Sbms "IPv4 multicast IGMP-layer source filter"); 84170613Sbms 85170613Sbms/* 86189592Sbms * Locking: 87189592Sbms * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK. 88189592Sbms * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however 89189592Sbms * it can be taken by code in net/if.c also. 90189592Sbms * - ip_moptions and in_mfilter are covered by the INP_WLOCK. 91189592Sbms * 92189592Sbms * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly 93189592Sbms * any need for in_multi itself to be virtualized -- it is bound to an ifp 94189592Sbms * anyway no matter what happens. 95170613Sbms */ 96170613Sbmsstruct mtx in_multi_mtx; 97189592SbmsMTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF); 98170613Sbms 99170613Sbms/* 100170613Sbms * Functions with non-static linkage defined in this file should be 101170613Sbms * declared in in_var.h: 102189592Sbms * imo_multi_filter() 103170613Sbms * in_addmulti() 104170613Sbms * in_delmulti() 105189592Sbms * in_joingroup() 106189592Sbms * in_joingroup_locked() 107189592Sbms * in_leavegroup() 108189592Sbms * in_leavegroup_locked() 109170613Sbms * and ip_var.h: 110170613Sbms * inp_freemoptions() 111170613Sbms * inp_getmoptions() 112170613Sbms * inp_setmoptions() 113189592Sbms * 114189592Sbms * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti() 115189592Sbms * and in_delmulti(). 116170613Sbms */ 117189592Sbmsstatic void imf_commit(struct in_mfilter *); 118189592Sbmsstatic int imf_get_source(struct in_mfilter *imf, 119189592Sbms const struct sockaddr_in *psin, 120189592Sbms struct in_msource **); 121189592Sbmsstatic struct in_msource * 122189592Sbms imf_graft(struct in_mfilter *, const uint8_t, 123189592Sbms const struct sockaddr_in *); 124189592Sbmsstatic void imf_leave(struct in_mfilter *); 125189592Sbmsstatic int imf_prune(struct in_mfilter *, const struct sockaddr_in *); 126189592Sbmsstatic void imf_purge(struct in_mfilter *); 127189592Sbmsstatic void imf_rollback(struct in_mfilter *); 128189592Sbmsstatic void imf_reap(struct in_mfilter *); 129170613Sbmsstatic int imo_grow(struct ip_moptions *); 130189592Sbmsstatic size_t imo_match_group(const struct ip_moptions *, 131189592Sbms const struct ifnet *, const struct sockaddr *); 132189592Sbmsstatic struct in_msource * 133189592Sbms imo_match_source(const struct ip_moptions *, const size_t, 134189592Sbms const struct sockaddr *); 135189592Sbmsstatic void ims_merge(struct ip_msource *ims, 136189592Sbms const struct in_msource *lims, const int rollback); 137189592Sbmsstatic int in_getmulti(struct ifnet *, const struct in_addr *, 138189592Sbms struct in_multi **); 139189592Sbmsstatic int inm_get_source(struct in_multi *inm, const in_addr_t haddr, 140189592Sbms const int noalloc, struct ip_msource **pims); 141189592Sbmsstatic int inm_is_ifp_detached(const struct in_multi *); 142189592Sbmsstatic int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *); 143189592Sbmsstatic void inm_purge(struct in_multi *); 144189592Sbmsstatic void inm_reap(struct in_multi *); 145170613Sbmsstatic struct ip_moptions * 146170613Sbms inp_findmoptions(struct inpcb *); 147170613Sbmsstatic int inp_get_source_filters(struct inpcb *, struct sockopt *); 148170613Sbmsstatic int inp_join_group(struct inpcb *, struct sockopt *); 149170613Sbmsstatic int inp_leave_group(struct inpcb *, struct sockopt *); 150189592Sbmsstatic struct ifnet * 151189592Sbms inp_lookup_mcast_ifp(const struct inpcb *, 152189592Sbms const struct sockaddr_in *, const struct in_addr); 153189592Sbmsstatic int inp_block_unblock_source(struct inpcb *, struct sockopt *); 154170613Sbmsstatic int inp_set_multicast_if(struct inpcb *, struct sockopt *); 155170613Sbmsstatic int inp_set_source_filters(struct inpcb *, struct sockopt *); 156189592Sbmsstatic int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS); 157170613Sbms 158189357SbmsSYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0, "IPv4 multicast"); 159189357Sbms 160189592Sbmsstatic u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER; 161189592SbmsSYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc, 162189592Sbms CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxgrpsrc, 0, 163189592Sbms "Max source filters per group"); 164189592SbmsTUNABLE_ULONG("net.inet.ip.mcast.maxgrpsrc", &in_mcast_maxgrpsrc); 165189592Sbms 166189592Sbmsstatic u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER; 167189592SbmsSYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc, 168189592Sbms CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxsocksrc, 0, 169189592Sbms "Max source filters per socket"); 170189592SbmsTUNABLE_ULONG("net.inet.ip.mcast.maxsocksrc", &in_mcast_maxsocksrc); 171189592Sbms 172189357Sbmsint in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP; 173189357SbmsSYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN, 174189357Sbms &in_mcast_loop, 0, "Loopback multicast datagrams by default"); 175189357SbmsTUNABLE_INT("net.inet.ip.mcast.loop", &in_mcast_loop); 176189357Sbms 177189592SbmsSYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters, 178189592Sbms CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters, 179189592Sbms "Per-interface stack-wide source filters"); 180189592Sbms 181170613Sbms/* 182189592Sbms * Inline function which wraps assertions for a valid ifp. 183189592Sbms * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp 184189592Sbms * is detached. 185189592Sbms */ 186189592Sbmsstatic int __inline 187189592Sbmsinm_is_ifp_detached(const struct in_multi *inm) 188189592Sbms{ 189189592Sbms struct ifnet *ifp; 190189592Sbms 191189592Sbms KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); 192189592Sbms ifp = inm->inm_ifma->ifma_ifp; 193189592Sbms if (ifp != NULL) { 194189592Sbms /* 195189592Sbms * Sanity check that netinet's notion of ifp is the 196189592Sbms * same as net's. 197189592Sbms */ 198189592Sbms KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); 199189592Sbms } 200189592Sbms 201189592Sbms return (ifp == NULL); 202189592Sbms} 203189592Sbms 204189592Sbms/* 205189592Sbms * Initialize an in_mfilter structure to a known state at t0, t1 206189592Sbms * with an empty source filter list. 207189592Sbms */ 208189592Sbmsstatic __inline void 209189592Sbmsimf_init(struct in_mfilter *imf, const int st0, const int st1) 210189592Sbms{ 211189592Sbms memset(imf, 0, sizeof(struct in_mfilter)); 212189592Sbms RB_INIT(&imf->imf_sources); 213189592Sbms imf->imf_st[0] = st0; 214189592Sbms imf->imf_st[1] = st1; 215189592Sbms} 216189592Sbms 217189592Sbms/* 218170613Sbms * Resize the ip_moptions vector to the next power-of-two minus 1. 219170613Sbms * May be called with locks held; do not sleep. 220170613Sbms */ 221170613Sbmsstatic int 222170613Sbmsimo_grow(struct ip_moptions *imo) 223170613Sbms{ 224170613Sbms struct in_multi **nmships; 225170613Sbms struct in_multi **omships; 226170613Sbms struct in_mfilter *nmfilters; 227170613Sbms struct in_mfilter *omfilters; 228170613Sbms size_t idx; 229170613Sbms size_t newmax; 230170613Sbms size_t oldmax; 231170613Sbms 232170613Sbms nmships = NULL; 233170613Sbms nmfilters = NULL; 234170613Sbms omships = imo->imo_membership; 235170613Sbms omfilters = imo->imo_mfilters; 236170613Sbms oldmax = imo->imo_max_memberships; 237170613Sbms newmax = ((oldmax + 1) * 2) - 1; 238170613Sbms 239170613Sbms if (newmax <= IP_MAX_MEMBERSHIPS) { 240170613Sbms nmships = (struct in_multi **)realloc(omships, 241170613Sbms sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT); 242170613Sbms nmfilters = (struct in_mfilter *)realloc(omfilters, 243189592Sbms sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT); 244170613Sbms if (nmships != NULL && nmfilters != NULL) { 245170613Sbms /* Initialize newly allocated source filter heads. */ 246170613Sbms for (idx = oldmax; idx < newmax; idx++) { 247189592Sbms imf_init(&nmfilters[idx], MCAST_UNDEFINED, 248189592Sbms MCAST_EXCLUDE); 249170613Sbms } 250170613Sbms imo->imo_max_memberships = newmax; 251170613Sbms imo->imo_membership = nmships; 252170613Sbms imo->imo_mfilters = nmfilters; 253170613Sbms } 254170613Sbms } 255170613Sbms 256170613Sbms if (nmships == NULL || nmfilters == NULL) { 257170613Sbms if (nmships != NULL) 258170613Sbms free(nmships, M_IPMOPTS); 259170613Sbms if (nmfilters != NULL) 260189592Sbms free(nmfilters, M_INMFILTER); 261170613Sbms return (ETOOMANYREFS); 262170613Sbms } 263170613Sbms 264170613Sbms return (0); 265170613Sbms} 266170613Sbms 267170613Sbms/* 268170613Sbms * Find an IPv4 multicast group entry for this ip_moptions instance 269170613Sbms * which matches the specified group, and optionally an interface. 270170613Sbms * Return its index into the array, or -1 if not found. 271170613Sbms */ 272189592Sbmsstatic size_t 273189592Sbmsimo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp, 274189592Sbms const struct sockaddr *group) 275170613Sbms{ 276189592Sbms const struct sockaddr_in *gsin; 277170613Sbms struct in_multi **pinm; 278170613Sbms int idx; 279170613Sbms int nmships; 280170613Sbms 281189592Sbms gsin = (const struct sockaddr_in *)group; 282170613Sbms 283170613Sbms /* The imo_membership array may be lazy allocated. */ 284170613Sbms if (imo->imo_membership == NULL || imo->imo_num_memberships == 0) 285170613Sbms return (-1); 286170613Sbms 287170613Sbms nmships = imo->imo_num_memberships; 288170613Sbms pinm = &imo->imo_membership[0]; 289170613Sbms for (idx = 0; idx < nmships; idx++, pinm++) { 290170613Sbms if (*pinm == NULL) 291170613Sbms continue; 292170613Sbms if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) && 293189592Sbms in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) { 294170613Sbms break; 295170613Sbms } 296170613Sbms } 297170613Sbms if (idx >= nmships) 298170613Sbms idx = -1; 299170613Sbms 300170613Sbms return (idx); 301170613Sbms} 302170613Sbms 303170613Sbms/* 304189592Sbms * Find an IPv4 multicast source entry for this imo which matches 305170613Sbms * the given group index for this socket, and source address. 306189592Sbms * 307189592Sbms * NOTE: This does not check if the entry is in-mode, merely if 308189592Sbms * it exists, which may not be the desired behaviour. 309170613Sbms */ 310189592Sbmsstatic struct in_msource * 311189592Sbmsimo_match_source(const struct ip_moptions *imo, const size_t gidx, 312189592Sbms const struct sockaddr *src) 313170613Sbms{ 314189592Sbms struct ip_msource find; 315170613Sbms struct in_mfilter *imf; 316189592Sbms struct ip_msource *ims; 317189592Sbms const sockunion_t *psa; 318170613Sbms 319170613Sbms KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__)); 320170613Sbms KASSERT(gidx != -1 && gidx < imo->imo_num_memberships, 321170613Sbms ("%s: invalid index %d\n", __func__, (int)gidx)); 322170613Sbms 323170613Sbms /* The imo_mfilters array may be lazy allocated. */ 324170613Sbms if (imo->imo_mfilters == NULL) 325170613Sbms return (NULL); 326170613Sbms imf = &imo->imo_mfilters[gidx]; 327170613Sbms 328189592Sbms /* Source trees are keyed in host byte order. */ 329189592Sbms psa = (const sockunion_t *)src; 330189592Sbms find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr); 331189592Sbms ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 332189592Sbms 333189592Sbms return ((struct in_msource *)ims); 334170613Sbms} 335170613Sbms 336170613Sbms/* 337189592Sbms * Perform filtering for multicast datagrams on a socket by group and source. 338189592Sbms * 339189592Sbms * Returns 0 if a datagram should be allowed through, or various error codes 340189592Sbms * if the socket was not a member of the group, or the source was muted, etc. 341170613Sbms */ 342189592Sbmsint 343189592Sbmsimo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp, 344189592Sbms const struct sockaddr *group, const struct sockaddr *src) 345170613Sbms{ 346189592Sbms size_t gidx; 347189592Sbms struct in_msource *ims; 348189592Sbms int mode; 349189592Sbms 350189592Sbms KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 351189592Sbms 352189592Sbms gidx = imo_match_group(imo, ifp, group); 353189592Sbms if (gidx == -1) 354189592Sbms return (MCAST_NOTGMEMBER); 355189592Sbms 356189592Sbms /* 357189592Sbms * Check if the source was included in an (S,G) join. 358189592Sbms * Allow reception on exclusive memberships by default, 359189592Sbms * reject reception on inclusive memberships by default. 360189592Sbms * Exclude source only if an in-mode exclude filter exists. 361189592Sbms * Include source only if an in-mode include filter exists. 362189592Sbms * NOTE: We are comparing group state here at IGMP t1 (now) 363189592Sbms * with socket-layer t0 (since last downcall). 364189592Sbms */ 365189592Sbms mode = imo->imo_mfilters[gidx].imf_st[1]; 366189592Sbms ims = imo_match_source(imo, gidx, src); 367189592Sbms 368189592Sbms if ((ims == NULL && mode == MCAST_INCLUDE) || 369189592Sbms (ims != NULL && ims->imsl_st[0] != mode)) 370189592Sbms return (MCAST_NOTSMEMBER); 371189592Sbms 372189592Sbms return (MCAST_PASS); 373189592Sbms} 374189592Sbms 375189592Sbms/* 376189592Sbms * Find and return a reference to an in_multi record for (ifp, group), 377189592Sbms * and bump its reference count. 378189592Sbms * If one does not exist, try to allocate it, and update link-layer multicast 379189592Sbms * filters on ifp to listen for group. 380189592Sbms * Assumes the IN_MULTI lock is held across the call. 381189592Sbms * Return 0 if successful, otherwise return an appropriate error code. 382189592Sbms */ 383189592Sbmsstatic int 384189592Sbmsin_getmulti(struct ifnet *ifp, const struct in_addr *group, 385189592Sbms struct in_multi **pinm) 386189592Sbms{ 387189592Sbms struct sockaddr_in gsin; 388189592Sbms struct ifmultiaddr *ifma; 389189592Sbms struct in_ifinfo *ii; 390189592Sbms struct in_multi *inm; 391189592Sbms int error; 392170613Sbms 393189592Sbms IN_MULTI_LOCK_ASSERT(); 394170613Sbms 395189592Sbms ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET]; 396170613Sbms 397189592Sbms inm = inm_lookup(ifp, *group); 398170613Sbms if (inm != NULL) { 399170613Sbms /* 400170613Sbms * If we already joined this group, just bump the 401170613Sbms * refcount and return it. 402170613Sbms */ 403170613Sbms KASSERT(inm->inm_refcount >= 1, 404170613Sbms ("%s: bad refcount %d", __func__, inm->inm_refcount)); 405170613Sbms ++inm->inm_refcount; 406189592Sbms *pinm = inm; 407189592Sbms return (0); 408189592Sbms } 409170613Sbms 410189592Sbms memset(&gsin, 0, sizeof(gsin)); 411189592Sbms gsin.sin_family = AF_INET; 412189592Sbms gsin.sin_len = sizeof(struct sockaddr_in); 413189592Sbms gsin.sin_addr = *group; 414170613Sbms 415189592Sbms /* 416189592Sbms * Check if a link-layer group is already associated 417189592Sbms * with this network-layer group on the given ifnet. 418189592Sbms */ 419189592Sbms error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma); 420189592Sbms if (error != 0) 421189592Sbms return (error); 422189592Sbms 423189931Sbms /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */ 424189931Sbms IF_ADDR_LOCK(ifp); 425189931Sbms 426189592Sbms /* 427189592Sbms * If something other than netinet is occupying the link-layer 428189592Sbms * group, print a meaningful error message and back out of 429189592Sbms * the allocation. 430189592Sbms * Otherwise, bump the refcount on the existing network-layer 431189592Sbms * group association and return it. 432189592Sbms */ 433189592Sbms if (ifma->ifma_protospec != NULL) { 434189592Sbms inm = (struct in_multi *)ifma->ifma_protospec; 435170613Sbms#ifdef INVARIANTS 436189592Sbms KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr", 437189592Sbms __func__)); 438189592Sbms KASSERT(ifma->ifma_addr->sa_family == AF_INET, 439189592Sbms ("%s: ifma not AF_INET", __func__)); 440189592Sbms KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__)); 441189592Sbms if (inm->inm_ifma != ifma || inm->inm_ifp != ifp || 442189592Sbms !in_hosteq(inm->inm_addr, *group)) 443189592Sbms panic("%s: ifma %p is inconsistent with %p (%s)", 444189592Sbms __func__, ifma, inm, inet_ntoa(*group)); 445170613Sbms#endif 446189592Sbms ++inm->inm_refcount; 447189592Sbms *pinm = inm; 448189931Sbms IF_ADDR_UNLOCK(ifp); 449189592Sbms return (0); 450189592Sbms } 451189592Sbms 452189931Sbms IF_ADDR_LOCK_ASSERT(ifp); 453189931Sbms 454189592Sbms /* 455189592Sbms * A new in_multi record is needed; allocate and initialize it. 456189592Sbms * We DO NOT perform an IGMP join as the in_ layer may need to 457189592Sbms * push an initial source list down to IGMP to support SSM. 458189592Sbms * 459189592Sbms * The initial source filter state is INCLUDE, {} as per the RFC. 460189592Sbms */ 461189592Sbms inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO); 462189592Sbms if (inm == NULL) { 463189592Sbms if_delmulti_ifma(ifma); 464189931Sbms IF_ADDR_UNLOCK(ifp); 465189592Sbms return (ENOMEM); 466189592Sbms } 467189592Sbms inm->inm_addr = *group; 468189592Sbms inm->inm_ifp = ifp; 469189592Sbms inm->inm_igi = ii->ii_igmp; 470189592Sbms inm->inm_ifma = ifma; 471189592Sbms inm->inm_refcount = 1; 472189592Sbms inm->inm_state = IGMP_NOT_MEMBER; 473189592Sbms 474189592Sbms /* 475189592Sbms * Pending state-changes per group are subject to a bounds check. 476189592Sbms */ 477189592Sbms IFQ_SET_MAXLEN(&inm->inm_scq, IGMP_MAX_STATE_CHANGES); 478189592Sbms 479189592Sbms inm->inm_st[0].iss_fmode = MCAST_UNDEFINED; 480189592Sbms inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 481189592Sbms RB_INIT(&inm->inm_srcs); 482189592Sbms 483189592Sbms ifma->ifma_protospec = inm; 484189592Sbms 485189592Sbms *pinm = inm; 486189592Sbms 487189931Sbms IF_ADDR_UNLOCK(ifp); 488189592Sbms return (0); 489189592Sbms} 490189592Sbms 491189592Sbms/* 492189592Sbms * Drop a reference to an in_multi record. 493189592Sbms * 494189592Sbms * If the refcount drops to 0, free the in_multi record and 495189592Sbms * delete the underlying link-layer membership. 496189592Sbms */ 497189592Sbmsvoid 498189592Sbmsinm_release_locked(struct in_multi *inm) 499189592Sbms{ 500189592Sbms struct ifmultiaddr *ifma; 501189592Sbms 502189592Sbms IN_MULTI_LOCK_ASSERT(); 503189592Sbms 504189592Sbms CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount); 505189592Sbms 506189592Sbms if (--inm->inm_refcount > 0) { 507189592Sbms CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__, 508189592Sbms inm->inm_refcount); 509189592Sbms return; 510189592Sbms } 511189592Sbms 512189592Sbms CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm); 513189592Sbms 514189592Sbms ifma = inm->inm_ifma; 515189592Sbms 516189931Sbms /* XXX this access is not covered by IF_ADDR_LOCK */ 517189592Sbms CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma); 518189592Sbms KASSERT(ifma->ifma_protospec == inm, 519189592Sbms ("%s: ifma_protospec != inm", __func__)); 520189592Sbms ifma->ifma_protospec = NULL; 521189592Sbms 522189592Sbms inm_purge(inm); 523189592Sbms 524189592Sbms free(inm, M_IPMADDR); 525189592Sbms 526189592Sbms if_delmulti_ifma(ifma); 527189592Sbms} 528189592Sbms 529189592Sbms/* 530189592Sbms * Clear recorded source entries for a group. 531189592Sbms * Used by the IGMP code. Caller must hold the IN_MULTI lock. 532189592Sbms * FIXME: Should reap. 533189592Sbms */ 534189592Sbmsvoid 535189592Sbmsinm_clear_recorded(struct in_multi *inm) 536189592Sbms{ 537189592Sbms struct ip_msource *ims; 538189592Sbms 539189592Sbms IN_MULTI_LOCK_ASSERT(); 540189592Sbms 541189592Sbms RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 542189592Sbms if (ims->ims_stp) { 543189592Sbms ims->ims_stp = 0; 544189592Sbms --inm->inm_st[1].iss_rec; 545170613Sbms } 546189592Sbms } 547189592Sbms KASSERT(inm->inm_st[1].iss_rec == 0, 548189592Sbms ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec)); 549189592Sbms} 550170613Sbms 551189592Sbms/* 552189592Sbms * Record a source as pending for a Source-Group IGMPv3 query. 553189592Sbms * This lives here as it modifies the shared tree. 554189592Sbms * 555189592Sbms * inm is the group descriptor. 556189592Sbms * naddr is the address of the source to record in network-byte order. 557189592Sbms * 558189592Sbms * If the net.inet.igmp.sgalloc sysctl is non-zero, we will 559189592Sbms * lazy-allocate a source node in response to an SG query. 560189592Sbms * Otherwise, no allocation is performed. This saves some memory 561189592Sbms * with the trade-off that the source will not be reported to the 562189592Sbms * router if joined in the window between the query response and 563189592Sbms * the group actually being joined on the local host. 564189592Sbms * 565189592Sbms * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed. 566189592Sbms * This turns off the allocation of a recorded source entry if 567189592Sbms * the group has not been joined. 568189592Sbms * 569189592Sbms * Return 0 if the source didn't exist or was already marked as recorded. 570189592Sbms * Return 1 if the source was marked as recorded by this function. 571189592Sbms * Return <0 if any error occured (negated errno code). 572189592Sbms */ 573189592Sbmsint 574189592Sbmsinm_record_source(struct in_multi *inm, const in_addr_t naddr) 575189592Sbms{ 576189592Sbms struct ip_msource find; 577189592Sbms struct ip_msource *ims, *nims; 578189592Sbms 579189592Sbms IN_MULTI_LOCK_ASSERT(); 580189592Sbms 581189592Sbms find.ims_haddr = ntohl(naddr); 582189592Sbms ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 583189592Sbms if (ims && ims->ims_stp) 584189592Sbms return (0); 585189592Sbms if (ims == NULL) { 586189592Sbms if (inm->inm_nsrc == in_mcast_maxgrpsrc) 587189592Sbms return (-ENOSPC); 588189592Sbms nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 589189592Sbms M_NOWAIT | M_ZERO); 590189592Sbms if (nims == NULL) 591189592Sbms return (-ENOMEM); 592189592Sbms nims->ims_haddr = find.ims_haddr; 593189592Sbms RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 594189592Sbms ++inm->inm_nsrc; 595189592Sbms ims = nims; 596189592Sbms } 597189592Sbms 598189592Sbms /* 599189592Sbms * Mark the source as recorded and update the recorded 600189592Sbms * source count. 601189592Sbms */ 602189592Sbms ++ims->ims_stp; 603189592Sbms ++inm->inm_st[1].iss_rec; 604189592Sbms 605189592Sbms return (1); 606189592Sbms} 607189592Sbms 608189592Sbms/* 609189592Sbms * Return a pointer to an in_msource owned by an in_mfilter, 610189592Sbms * given its source address. 611189592Sbms * Lazy-allocate if needed. If this is a new entry its filter state is 612189592Sbms * undefined at t0. 613189592Sbms * 614189592Sbms * imf is the filter set being modified. 615189592Sbms * haddr is the source address in *host* byte-order. 616189592Sbms * 617189592Sbms * SMPng: May be called with locks held; malloc must not block. 618189592Sbms */ 619189592Sbmsstatic int 620189592Sbmsimf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin, 621189592Sbms struct in_msource **plims) 622189592Sbms{ 623189592Sbms struct ip_msource find; 624189592Sbms struct ip_msource *ims, *nims; 625189592Sbms struct in_msource *lims; 626189592Sbms int error; 627189592Sbms 628189592Sbms error = 0; 629189592Sbms ims = NULL; 630189592Sbms lims = NULL; 631189592Sbms 632189592Sbms /* key is host byte order */ 633189592Sbms find.ims_haddr = ntohl(psin->sin_addr.s_addr); 634189592Sbms ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 635189592Sbms lims = (struct in_msource *)ims; 636189592Sbms if (lims == NULL) { 637189592Sbms if (imf->imf_nsrc == in_mcast_maxsocksrc) 638189592Sbms return (ENOSPC); 639189592Sbms nims = malloc(sizeof(struct in_msource), M_INMFILTER, 640189592Sbms M_NOWAIT | M_ZERO); 641189592Sbms if (nims == NULL) 642189592Sbms return (ENOMEM); 643189592Sbms lims = (struct in_msource *)nims; 644189592Sbms lims->ims_haddr = find.ims_haddr; 645189592Sbms lims->imsl_st[0] = MCAST_UNDEFINED; 646189592Sbms RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 647189592Sbms ++imf->imf_nsrc; 648189592Sbms } 649189592Sbms 650189592Sbms *plims = lims; 651189592Sbms 652189592Sbms return (error); 653189592Sbms} 654189592Sbms 655189592Sbms/* 656189592Sbms * Graft a source entry into an existing socket-layer filter set, 657189592Sbms * maintaining any required invariants and checking allocations. 658189592Sbms * 659189592Sbms * The source is marked as being in the new filter mode at t1. 660189592Sbms * 661189592Sbms * Return the pointer to the new node, otherwise return NULL. 662189592Sbms */ 663189592Sbmsstatic struct in_msource * 664189592Sbmsimf_graft(struct in_mfilter *imf, const uint8_t st1, 665189592Sbms const struct sockaddr_in *psin) 666189592Sbms{ 667189592Sbms struct ip_msource *nims; 668189592Sbms struct in_msource *lims; 669189592Sbms 670189592Sbms nims = malloc(sizeof(struct in_msource), M_INMFILTER, 671189592Sbms M_NOWAIT | M_ZERO); 672189592Sbms if (nims == NULL) 673189592Sbms return (NULL); 674189592Sbms lims = (struct in_msource *)nims; 675189592Sbms lims->ims_haddr = ntohl(psin->sin_addr.s_addr); 676189592Sbms lims->imsl_st[0] = MCAST_UNDEFINED; 677189592Sbms lims->imsl_st[1] = st1; 678189592Sbms RB_INSERT(ip_msource_tree, &imf->imf_sources, nims); 679189592Sbms ++imf->imf_nsrc; 680189592Sbms 681189592Sbms return (lims); 682189592Sbms} 683189592Sbms 684189592Sbms/* 685189592Sbms * Prune a source entry from an existing socket-layer filter set, 686189592Sbms * maintaining any required invariants and checking allocations. 687189592Sbms * 688189592Sbms * The source is marked as being left at t1, it is not freed. 689189592Sbms * 690189592Sbms * Return 0 if no error occurred, otherwise return an errno value. 691189592Sbms */ 692189592Sbmsstatic int 693189592Sbmsimf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin) 694189592Sbms{ 695189592Sbms struct ip_msource find; 696189592Sbms struct ip_msource *ims; 697189592Sbms struct in_msource *lims; 698189592Sbms 699189592Sbms /* key is host byte order */ 700189592Sbms find.ims_haddr = ntohl(psin->sin_addr.s_addr); 701189592Sbms ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find); 702189592Sbms if (ims == NULL) 703189592Sbms return (ENOENT); 704189592Sbms lims = (struct in_msource *)ims; 705189592Sbms lims->imsl_st[1] = MCAST_UNDEFINED; 706189592Sbms return (0); 707189592Sbms} 708189592Sbms 709189592Sbms/* 710189592Sbms * Revert socket-layer filter set deltas at t1 to t0 state. 711189592Sbms */ 712189592Sbmsstatic void 713189592Sbmsimf_rollback(struct in_mfilter *imf) 714189592Sbms{ 715189592Sbms struct ip_msource *ims, *tims; 716189592Sbms struct in_msource *lims; 717189592Sbms 718189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 719189592Sbms lims = (struct in_msource *)ims; 720189592Sbms if (lims->imsl_st[0] == lims->imsl_st[1]) { 721189592Sbms /* no change at t1 */ 722189592Sbms continue; 723189592Sbms } else if (lims->imsl_st[0] != MCAST_UNDEFINED) { 724189592Sbms /* revert change to existing source at t1 */ 725189592Sbms lims->imsl_st[1] = lims->imsl_st[0]; 726189592Sbms } else { 727189592Sbms /* revert source added t1 */ 728189592Sbms CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 729189592Sbms RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 730189592Sbms free(ims, M_INMFILTER); 731189592Sbms imf->imf_nsrc--; 732189592Sbms } 733189592Sbms } 734189592Sbms imf->imf_st[1] = imf->imf_st[0]; 735189592Sbms} 736189592Sbms 737189592Sbms/* 738189592Sbms * Mark socket-layer filter set as INCLUDE {} at t1. 739189592Sbms */ 740189592Sbmsstatic void 741189592Sbmsimf_leave(struct in_mfilter *imf) 742189592Sbms{ 743189592Sbms struct ip_msource *ims; 744189592Sbms struct in_msource *lims; 745189592Sbms 746189592Sbms RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 747189592Sbms lims = (struct in_msource *)ims; 748189592Sbms lims->imsl_st[1] = MCAST_UNDEFINED; 749189592Sbms } 750189592Sbms imf->imf_st[1] = MCAST_INCLUDE; 751189592Sbms} 752189592Sbms 753189592Sbms/* 754189592Sbms * Mark socket-layer filter set deltas as committed. 755189592Sbms */ 756189592Sbmsstatic void 757189592Sbmsimf_commit(struct in_mfilter *imf) 758189592Sbms{ 759189592Sbms struct ip_msource *ims; 760189592Sbms struct in_msource *lims; 761189592Sbms 762189592Sbms RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 763189592Sbms lims = (struct in_msource *)ims; 764189592Sbms lims->imsl_st[0] = lims->imsl_st[1]; 765189592Sbms } 766189592Sbms imf->imf_st[0] = imf->imf_st[1]; 767189592Sbms} 768189592Sbms 769189592Sbms/* 770189592Sbms * Reap unreferenced sources from socket-layer filter set. 771189592Sbms */ 772189592Sbmsstatic void 773189592Sbmsimf_reap(struct in_mfilter *imf) 774189592Sbms{ 775189592Sbms struct ip_msource *ims, *tims; 776189592Sbms struct in_msource *lims; 777189592Sbms 778189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 779189592Sbms lims = (struct in_msource *)ims; 780189592Sbms if ((lims->imsl_st[0] == MCAST_UNDEFINED) && 781189592Sbms (lims->imsl_st[1] == MCAST_UNDEFINED)) { 782189592Sbms CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims); 783189592Sbms RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 784189592Sbms free(ims, M_INMFILTER); 785189592Sbms imf->imf_nsrc--; 786189592Sbms } 787189592Sbms } 788189592Sbms} 789189592Sbms 790189592Sbms/* 791189592Sbms * Purge socket-layer filter set. 792189592Sbms */ 793189592Sbmsstatic void 794189592Sbmsimf_purge(struct in_mfilter *imf) 795189592Sbms{ 796189592Sbms struct ip_msource *ims, *tims; 797189592Sbms 798189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) { 799189592Sbms CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 800189592Sbms RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims); 801189592Sbms free(ims, M_INMFILTER); 802189592Sbms imf->imf_nsrc--; 803189592Sbms } 804189592Sbms imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED; 805189592Sbms KASSERT(RB_EMPTY(&imf->imf_sources), 806189592Sbms ("%s: imf_sources not empty", __func__)); 807189592Sbms} 808189592Sbms 809189592Sbms/* 810189592Sbms * Look up a source filter entry for a multicast group. 811189592Sbms * 812189592Sbms * inm is the group descriptor to work with. 813189592Sbms * haddr is the host-byte-order IPv4 address to look up. 814189592Sbms * noalloc may be non-zero to suppress allocation of sources. 815189592Sbms * *pims will be set to the address of the retrieved or allocated source. 816189592Sbms * 817189592Sbms * SMPng: NOTE: may be called with locks held. 818189592Sbms * Return 0 if successful, otherwise return a non-zero error code. 819189592Sbms */ 820189592Sbmsstatic int 821189592Sbmsinm_get_source(struct in_multi *inm, const in_addr_t haddr, 822189592Sbms const int noalloc, struct ip_msource **pims) 823189592Sbms{ 824189592Sbms struct ip_msource find; 825189592Sbms struct ip_msource *ims, *nims; 826189592Sbms#ifdef KTR 827189592Sbms struct in_addr ia; 828189592Sbms#endif 829189592Sbms 830189592Sbms find.ims_haddr = haddr; 831189592Sbms ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find); 832189592Sbms if (ims == NULL && !noalloc) { 833189592Sbms if (inm->inm_nsrc == in_mcast_maxgrpsrc) 834189592Sbms return (ENOSPC); 835189592Sbms nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE, 836189592Sbms M_NOWAIT | M_ZERO); 837189592Sbms if (nims == NULL) 838189592Sbms return (ENOMEM); 839189592Sbms nims->ims_haddr = haddr; 840189592Sbms RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims); 841189592Sbms ++inm->inm_nsrc; 842189592Sbms ims = nims; 843189592Sbms#ifdef KTR 844189592Sbms ia.s_addr = htonl(haddr); 845189592Sbms CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__, 846189592Sbms inet_ntoa(ia), ims); 847189592Sbms#endif 848189592Sbms } 849189592Sbms 850189592Sbms *pims = ims; 851189592Sbms return (0); 852189592Sbms} 853189592Sbms 854189592Sbms/* 855189592Sbms * Merge socket-layer source into IGMP-layer source. 856189592Sbms * If rollback is non-zero, perform the inverse of the merge. 857189592Sbms */ 858189592Sbmsstatic void 859189592Sbmsims_merge(struct ip_msource *ims, const struct in_msource *lims, 860189592Sbms const int rollback) 861189592Sbms{ 862189592Sbms int n = rollback ? -1 : 1; 863189592Sbms#ifdef KTR 864189592Sbms struct in_addr ia; 865189592Sbms 866189592Sbms ia.s_addr = htonl(ims->ims_haddr); 867189592Sbms#endif 868189592Sbms 869189592Sbms if (lims->imsl_st[0] == MCAST_EXCLUDE) { 870189592Sbms CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s", 871189592Sbms __func__, n, inet_ntoa(ia)); 872189592Sbms ims->ims_st[1].ex -= n; 873189592Sbms } else if (lims->imsl_st[0] == MCAST_INCLUDE) { 874189592Sbms CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s", 875189592Sbms __func__, n, inet_ntoa(ia)); 876189592Sbms ims->ims_st[1].in -= n; 877189592Sbms } 878189592Sbms 879189592Sbms if (lims->imsl_st[1] == MCAST_EXCLUDE) { 880189592Sbms CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s", 881189592Sbms __func__, n, inet_ntoa(ia)); 882189592Sbms ims->ims_st[1].ex += n; 883189592Sbms } else if (lims->imsl_st[1] == MCAST_INCLUDE) { 884189592Sbms CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s", 885189592Sbms __func__, n, inet_ntoa(ia)); 886189592Sbms ims->ims_st[1].in += n; 887189592Sbms } 888189592Sbms} 889189592Sbms 890189592Sbms/* 891189592Sbms * Atomically update the global in_multi state, when a membership's 892189592Sbms * filter list is being updated in any way. 893189592Sbms * 894189592Sbms * imf is the per-inpcb-membership group filter pointer. 895189592Sbms * A fake imf may be passed for in-kernel consumers. 896189592Sbms * 897189592Sbms * XXX This is a candidate for a set-symmetric-difference style loop 898189592Sbms * which would eliminate the repeated lookup from root of ims nodes, 899189592Sbms * as they share the same key space. 900189592Sbms * 901189592Sbms * If any error occurred this function will back out of refcounts 902189592Sbms * and return a non-zero value. 903189592Sbms */ 904189592Sbmsstatic int 905189592Sbmsinm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 906189592Sbms{ 907189592Sbms struct ip_msource *ims, *nims; 908189592Sbms struct in_msource *lims; 909189592Sbms int schanged, error; 910189592Sbms int nsrc0, nsrc1; 911189592Sbms 912189592Sbms schanged = 0; 913189592Sbms error = 0; 914189592Sbms nsrc1 = nsrc0 = 0; 915189592Sbms 916189592Sbms /* 917189592Sbms * Update the source filters first, as this may fail. 918189592Sbms * Maintain count of in-mode filters at t0, t1. These are 919189592Sbms * used to work out if we transition into ASM mode or not. 920189592Sbms * Maintain a count of source filters whose state was 921189592Sbms * actually modified by this operation. 922189592Sbms */ 923189592Sbms RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 924189592Sbms lims = (struct in_msource *)ims; 925189592Sbms if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++; 926189592Sbms if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++; 927189592Sbms if (lims->imsl_st[0] == lims->imsl_st[1]) continue; 928189592Sbms error = inm_get_source(inm, lims->ims_haddr, 0, &nims); 929189592Sbms ++schanged; 930189592Sbms if (error) 931170613Sbms break; 932189592Sbms ims_merge(nims, lims, 0); 933189592Sbms } 934189592Sbms if (error) { 935189592Sbms struct ip_msource *bims; 936189592Sbms 937189592Sbms RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) { 938189592Sbms lims = (struct in_msource *)ims; 939189592Sbms if (lims->imsl_st[0] == lims->imsl_st[1]) 940189592Sbms continue; 941189592Sbms (void)inm_get_source(inm, lims->ims_haddr, 1, &bims); 942189592Sbms if (bims == NULL) 943189592Sbms continue; 944189592Sbms ims_merge(bims, lims, 1); 945170613Sbms } 946189592Sbms goto out_reap; 947189592Sbms } 948170613Sbms 949189592Sbms CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1", 950189592Sbms __func__, nsrc0, nsrc1); 951170613Sbms 952189592Sbms /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */ 953189592Sbms if (imf->imf_st[0] == imf->imf_st[1] && 954189592Sbms imf->imf_st[1] == MCAST_INCLUDE) { 955189592Sbms if (nsrc1 == 0) { 956189592Sbms CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 957189592Sbms --inm->inm_st[1].iss_in; 958189592Sbms } 959189592Sbms } 960170613Sbms 961189592Sbms /* Handle filter mode transition on socket. */ 962189592Sbms if (imf->imf_st[0] != imf->imf_st[1]) { 963189592Sbms CTR3(KTR_IGMPV3, "%s: imf transition %d to %d", 964189592Sbms __func__, imf->imf_st[0], imf->imf_st[1]); 965189592Sbms 966189592Sbms if (imf->imf_st[0] == MCAST_EXCLUDE) { 967189592Sbms CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__); 968189592Sbms --inm->inm_st[1].iss_ex; 969189592Sbms } else if (imf->imf_st[0] == MCAST_INCLUDE) { 970189592Sbms CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__); 971189592Sbms --inm->inm_st[1].iss_in; 972189592Sbms } 973189592Sbms 974189592Sbms if (imf->imf_st[1] == 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[1] == MCAST_INCLUDE && nsrc1 > 0) { 978189592Sbms CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__); 979189592Sbms inm->inm_st[1].iss_in++; 980189592Sbms } 981189592Sbms } 982189592Sbms 983189592Sbms /* 984189592Sbms * Track inm filter state in terms of listener counts. 985189592Sbms * If there are any exclusive listeners, stack-wide 986189592Sbms * membership is exclusive. 987189592Sbms * Otherwise, if only inclusive listeners, stack-wide is inclusive. 988189592Sbms * If no listeners remain, state is undefined at t1, 989189592Sbms * and the IGMP lifecycle for this group should finish. 990189592Sbms */ 991189592Sbms if (inm->inm_st[1].iss_ex > 0) { 992189592Sbms CTR1(KTR_IGMPV3, "%s: transition to EX", __func__); 993189592Sbms inm->inm_st[1].iss_fmode = MCAST_EXCLUDE; 994189592Sbms } else if (inm->inm_st[1].iss_in > 0) { 995189592Sbms CTR1(KTR_IGMPV3, "%s: transition to IN", __func__); 996189592Sbms inm->inm_st[1].iss_fmode = MCAST_INCLUDE; 997189592Sbms } else { 998189592Sbms CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__); 999189592Sbms inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; 1000189592Sbms } 1001189592Sbms 1002189592Sbms /* Decrement ASM listener count on transition out of ASM mode. */ 1003189592Sbms if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) { 1004189592Sbms if ((imf->imf_st[1] != MCAST_EXCLUDE) || 1005189592Sbms (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) 1006189592Sbms CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__); 1007189592Sbms --inm->inm_st[1].iss_asm; 1008189592Sbms } 1009189592Sbms 1010189592Sbms /* Increment ASM listener count on transition to ASM mode. */ 1011189592Sbms if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) { 1012189592Sbms CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__); 1013189592Sbms inm->inm_st[1].iss_asm++; 1014189592Sbms } 1015189592Sbms 1016189592Sbms CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm); 1017189592Sbms inm_print(inm); 1018189592Sbms 1019189592Sbmsout_reap: 1020189592Sbms if (schanged > 0) { 1021189592Sbms CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__); 1022189592Sbms inm_reap(inm); 1023189592Sbms } 1024189592Sbms return (error); 1025189592Sbms} 1026189592Sbms 1027189592Sbms/* 1028189592Sbms * Mark an in_multi's filter set deltas as committed. 1029189592Sbms * Called by IGMP after a state change has been enqueued. 1030189592Sbms */ 1031189592Sbmsvoid 1032189592Sbmsinm_commit(struct in_multi *inm) 1033189592Sbms{ 1034189592Sbms struct ip_msource *ims; 1035189592Sbms 1036189592Sbms CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm); 1037189592Sbms CTR1(KTR_IGMPV3, "%s: pre commit:", __func__); 1038189592Sbms inm_print(inm); 1039189592Sbms 1040189592Sbms RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 1041189592Sbms ims->ims_st[0] = ims->ims_st[1]; 1042189592Sbms } 1043189592Sbms inm->inm_st[0] = inm->inm_st[1]; 1044189592Sbms} 1045189592Sbms 1046189592Sbms/* 1047189592Sbms * Reap unreferenced nodes from an in_multi's filter set. 1048189592Sbms */ 1049189592Sbmsstatic void 1050189592Sbmsinm_reap(struct in_multi *inm) 1051189592Sbms{ 1052189592Sbms struct ip_msource *ims, *tims; 1053189592Sbms 1054189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1055189592Sbms if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 || 1056189592Sbms ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 || 1057189592Sbms ims->ims_stp != 0) 1058189592Sbms continue; 1059189592Sbms CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1060189592Sbms RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1061189592Sbms free(ims, M_IPMSOURCE); 1062189592Sbms inm->inm_nsrc--; 1063189592Sbms } 1064189592Sbms} 1065189592Sbms 1066189592Sbms/* 1067189592Sbms * Purge all source nodes from an in_multi's filter set. 1068189592Sbms */ 1069189592Sbmsstatic void 1070189592Sbmsinm_purge(struct in_multi *inm) 1071189592Sbms{ 1072189592Sbms struct ip_msource *ims, *tims; 1073189592Sbms 1074189592Sbms RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) { 1075189592Sbms CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims); 1076189592Sbms RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims); 1077189592Sbms free(ims, M_IPMSOURCE); 1078189592Sbms inm->inm_nsrc--; 1079189592Sbms } 1080189592Sbms} 1081189592Sbms 1082189592Sbms/* 1083189592Sbms * Join a multicast group; unlocked entry point. 1084189592Sbms * 1085189592Sbms * SMPng: XXX: in_joingroup() is called from in_control() when Giant 1086189592Sbms * is not held. Fortunately, ifp is unlikely to have been detached 1087189592Sbms * at this point, so we assume it's OK to recurse. 1088189592Sbms */ 1089189592Sbmsint 1090189592Sbmsin_joingroup(struct ifnet *ifp, const struct in_addr *gina, 1091189592Sbms /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1092189592Sbms{ 1093189592Sbms int error; 1094189592Sbms 1095189592Sbms IN_MULTI_LOCK(); 1096189592Sbms error = in_joingroup_locked(ifp, gina, imf, pinm); 1097170613Sbms IN_MULTI_UNLOCK(); 1098170613Sbms 1099189592Sbms return (error); 1100170613Sbms} 1101170613Sbms 1102170613Sbms/* 1103189592Sbms * Join a multicast group; real entry point. 1104170613Sbms * 1105189592Sbms * Only preserves atomicity at inm level. 1106189592Sbms * NOTE: imf argument cannot be const due to sys/tree.h limitations. 1107170613Sbms * 1108189592Sbms * If the IGMP downcall fails, the group is not joined, and an error 1109189592Sbms * code is returned. 1110170613Sbms */ 1111189592Sbmsint 1112189592Sbmsin_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina, 1113189592Sbms /*const*/ struct in_mfilter *imf, struct in_multi **pinm) 1114170613Sbms{ 1115189592Sbms struct in_mfilter timf; 1116189592Sbms struct in_multi *inm; 1117189592Sbms int error; 1118170613Sbms 1119189592Sbms IN_MULTI_LOCK_ASSERT(); 1120170613Sbms 1121189592Sbms CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__, 1122189592Sbms inet_ntoa(*gina), ifp, ifp->if_xname); 1123189592Sbms 1124189592Sbms error = 0; 1125189592Sbms inm = NULL; 1126189592Sbms 1127189592Sbms /* 1128189592Sbms * If no imf was specified (i.e. kernel consumer), 1129189592Sbms * fake one up and assume it is an ASM join. 1130189592Sbms */ 1131189592Sbms if (imf == NULL) { 1132189592Sbms imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE); 1133189592Sbms imf = &timf; 1134170613Sbms } 1135170613Sbms 1136189592Sbms error = in_getmulti(ifp, gina, &inm); 1137189592Sbms if (error) { 1138189592Sbms CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__); 1139189592Sbms return (error); 1140189592Sbms } 1141189592Sbms 1142189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1143189592Sbms error = inm_merge(inm, imf); 1144189592Sbms if (error) { 1145189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1146189592Sbms goto out_inm_release; 1147189592Sbms } 1148189592Sbms 1149189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1150189592Sbms error = igmp_change_state(inm); 1151189592Sbms if (error) { 1152189592Sbms CTR1(KTR_IGMPV3, "%s: failed to update source", __func__); 1153189592Sbms goto out_inm_release; 1154189592Sbms } 1155189592Sbms 1156189592Sbmsout_inm_release: 1157189592Sbms if (error) { 1158189592Sbms CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1159189592Sbms inm_release_locked(inm); 1160189592Sbms } else { 1161189592Sbms *pinm = inm; 1162189592Sbms } 1163189592Sbms 1164189592Sbms return (error); 1165189592Sbms} 1166189592Sbms 1167189592Sbms/* 1168189592Sbms * Leave a multicast group; unlocked entry point. 1169189592Sbms */ 1170189592Sbmsint 1171189592Sbmsin_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1172189592Sbms{ 1173189592Sbms struct ifnet *ifp; 1174189851Srwatson int error; 1175189592Sbms 1176189592Sbms ifp = inm->inm_ifp; 1177189592Sbms 1178170613Sbms IN_MULTI_LOCK(); 1179189592Sbms error = in_leavegroup_locked(inm, imf); 1180170613Sbms IN_MULTI_UNLOCK(); 1181170613Sbms 1182189592Sbms return (error); 1183170613Sbms} 1184170613Sbms 1185170613Sbms/* 1186189592Sbms * Leave a multicast group; real entry point. 1187189592Sbms * All source filters will be expunged. 1188170613Sbms * 1189189592Sbms * Only preserves atomicity at inm level. 1190189592Sbms * 1191189592Sbms * Holding the write lock for the INP which contains imf 1192189592Sbms * is highly advisable. We can't assert for it as imf does not 1193189592Sbms * contain a back-pointer to the owning inp. 1194189592Sbms * 1195189592Sbms * Note: This is not the same as inm_release(*) as this function also 1196189592Sbms * makes a state change downcall into IGMP. 1197170613Sbms */ 1198189592Sbmsint 1199189592Sbmsin_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf) 1200170613Sbms{ 1201189592Sbms struct in_mfilter timf; 1202189592Sbms int error; 1203170613Sbms 1204189592Sbms error = 0; 1205189592Sbms 1206170613Sbms IN_MULTI_LOCK_ASSERT(); 1207170613Sbms 1208189592Sbms CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__, 1209189592Sbms inm, inet_ntoa(inm->inm_addr), 1210189592Sbms (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname), 1211189592Sbms imf); 1212170613Sbms 1213189592Sbms /* 1214189592Sbms * If no imf was specified (i.e. kernel consumer), 1215189592Sbms * fake one up and assume it is an ASM join. 1216189592Sbms */ 1217189592Sbms if (imf == NULL) { 1218189592Sbms imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED); 1219189592Sbms imf = &timf; 1220189592Sbms } 1221170613Sbms 1222189592Sbms /* 1223189592Sbms * Begin state merge transaction at IGMP layer. 1224189592Sbms * 1225189592Sbms * As this particular invocation should not cause any memory 1226189592Sbms * to be allocated, and there is no opportunity to roll back 1227189592Sbms * the transaction, it MUST NOT fail. 1228189592Sbms */ 1229189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1230189592Sbms error = inm_merge(inm, imf); 1231189592Sbms KASSERT(error == 0, ("%s: failed to merge inm state", __func__)); 1232170613Sbms 1233189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1234189592Sbms error = igmp_change_state(inm); 1235189592Sbms if (error) 1236189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1237189592Sbms 1238189592Sbms CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm); 1239189592Sbms inm_release_locked(inm); 1240189592Sbms 1241189592Sbms return (error); 1242170613Sbms} 1243170613Sbms 1244189592Sbms/*#ifndef BURN_BRIDGES*/ 1245170613Sbms/* 1246189592Sbms * Join an IPv4 multicast group in (*,G) exclusive mode. 1247189592Sbms * The group must be a 224.0.0.0/24 link-scope group. 1248189592Sbms * This KPI is for legacy kernel consumers only. 1249170613Sbms */ 1250189592Sbmsstruct in_multi * 1251189592Sbmsin_addmulti(struct in_addr *ap, struct ifnet *ifp) 1252189592Sbms{ 1253189592Sbms struct in_multi *pinm; 1254189592Sbms int error; 1255189592Sbms 1256189592Sbms KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)), 1257189592Sbms ("%s: %s not in 224.0.0.0/24", __func__, inet_ntoa(*ap))); 1258189592Sbms 1259189592Sbms error = in_joingroup(ifp, ap, NULL, &pinm); 1260189592Sbms if (error != 0) 1261189592Sbms pinm = NULL; 1262189592Sbms 1263189592Sbms return (pinm); 1264189592Sbms} 1265189592Sbms 1266189592Sbms/* 1267189592Sbms * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode. 1268189592Sbms * This KPI is for legacy kernel consumers only. 1269189592Sbms */ 1270189592Sbmsvoid 1271189592Sbmsin_delmulti(struct in_multi *inm) 1272189592Sbms{ 1273189592Sbms 1274189592Sbms (void)in_leavegroup(inm, NULL); 1275189592Sbms} 1276189592Sbms/*#endif*/ 1277189592Sbms 1278189592Sbms/* 1279189592Sbms * Block or unblock an ASM multicast source on an inpcb. 1280189592Sbms * This implements the delta-based API described in RFC 3678. 1281189592Sbms * 1282189592Sbms * The delta-based API applies only to exclusive-mode memberships. 1283189592Sbms * An IGMP downcall will be performed. 1284189592Sbms * 1285189592Sbms * SMPng: NOTE: Must take Giant as a join may create a new ifma. 1286189592Sbms * 1287189592Sbms * Return 0 if successful, otherwise return an appropriate error code. 1288189592Sbms */ 1289170613Sbmsstatic int 1290189592Sbmsinp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt) 1291170613Sbms{ 1292170613Sbms struct group_source_req gsr; 1293170613Sbms sockunion_t *gsa, *ssa; 1294170613Sbms struct ifnet *ifp; 1295170613Sbms struct in_mfilter *imf; 1296170613Sbms struct ip_moptions *imo; 1297170613Sbms struct in_msource *ims; 1298189592Sbms struct in_multi *inm; 1299170613Sbms size_t idx; 1300189592Sbms uint16_t fmode; 1301189592Sbms int error, doblock; 1302170613Sbms 1303170613Sbms ifp = NULL; 1304170613Sbms error = 0; 1305189592Sbms doblock = 0; 1306170613Sbms 1307170613Sbms memset(&gsr, 0, sizeof(struct group_source_req)); 1308170613Sbms gsa = (sockunion_t *)&gsr.gsr_group; 1309170613Sbms ssa = (sockunion_t *)&gsr.gsr_source; 1310170613Sbms 1311170613Sbms switch (sopt->sopt_name) { 1312170613Sbms case IP_BLOCK_SOURCE: 1313170613Sbms case IP_UNBLOCK_SOURCE: { 1314170613Sbms struct ip_mreq_source mreqs; 1315170613Sbms 1316170613Sbms error = sooptcopyin(sopt, &mreqs, 1317170613Sbms sizeof(struct ip_mreq_source), 1318170613Sbms sizeof(struct ip_mreq_source)); 1319170613Sbms if (error) 1320170613Sbms return (error); 1321170613Sbms 1322170613Sbms gsa->sin.sin_family = AF_INET; 1323170613Sbms gsa->sin.sin_len = sizeof(struct sockaddr_in); 1324170613Sbms gsa->sin.sin_addr = mreqs.imr_multiaddr; 1325170613Sbms 1326170613Sbms ssa->sin.sin_family = AF_INET; 1327170613Sbms ssa->sin.sin_len = sizeof(struct sockaddr_in); 1328170613Sbms ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1329170613Sbms 1330189592Sbms if (!in_nullhost(mreqs.imr_interface)) 1331170613Sbms INADDR_TO_IFP(mreqs.imr_interface, ifp); 1332170613Sbms 1333170613Sbms if (sopt->sopt_name == IP_BLOCK_SOURCE) 1334189592Sbms doblock = 1; 1335170613Sbms 1336189592Sbms CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p", 1337189592Sbms __func__, inet_ntoa(mreqs.imr_interface), ifp); 1338170613Sbms break; 1339170613Sbms } 1340170613Sbms 1341170613Sbms case MCAST_BLOCK_SOURCE: 1342170613Sbms case MCAST_UNBLOCK_SOURCE: 1343170613Sbms error = sooptcopyin(sopt, &gsr, 1344170613Sbms sizeof(struct group_source_req), 1345170613Sbms sizeof(struct group_source_req)); 1346170613Sbms if (error) 1347170613Sbms return (error); 1348170613Sbms 1349170613Sbms if (gsa->sin.sin_family != AF_INET || 1350170613Sbms gsa->sin.sin_len != sizeof(struct sockaddr_in)) 1351170613Sbms return (EINVAL); 1352170613Sbms 1353170613Sbms if (ssa->sin.sin_family != AF_INET || 1354170613Sbms ssa->sin.sin_len != sizeof(struct sockaddr_in)) 1355170613Sbms return (EINVAL); 1356170613Sbms 1357181803Sbz if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1358170613Sbms return (EADDRNOTAVAIL); 1359170613Sbms 1360170613Sbms ifp = ifnet_byindex(gsr.gsr_interface); 1361170613Sbms 1362170613Sbms if (sopt->sopt_name == MCAST_BLOCK_SOURCE) 1363189592Sbms doblock = 1; 1364170613Sbms break; 1365170613Sbms 1366170613Sbms default: 1367189592Sbms CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1368189592Sbms __func__, sopt->sopt_name); 1369170613Sbms return (EOPNOTSUPP); 1370170613Sbms break; 1371170613Sbms } 1372170613Sbms 1373170613Sbms if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1374170613Sbms return (EINVAL); 1375170613Sbms 1376170613Sbms /* 1377170613Sbms * Check if we are actually a member of this group. 1378170613Sbms */ 1379170613Sbms imo = inp_findmoptions(inp); 1380170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 1381170613Sbms if (idx == -1 || imo->imo_mfilters == NULL) { 1382170613Sbms error = EADDRNOTAVAIL; 1383189592Sbms goto out_inp_locked; 1384170613Sbms } 1385170613Sbms 1386170613Sbms KASSERT(imo->imo_mfilters != NULL, 1387170613Sbms ("%s: imo_mfilters not allocated", __func__)); 1388170613Sbms imf = &imo->imo_mfilters[idx]; 1389189592Sbms inm = imo->imo_membership[idx]; 1390170613Sbms 1391170613Sbms /* 1392189592Sbms * Attempting to use the delta-based API on an 1393189592Sbms * non exclusive-mode membership is an error. 1394170613Sbms */ 1395189592Sbms fmode = imf->imf_st[0]; 1396189592Sbms if (fmode != MCAST_EXCLUDE) { 1397189592Sbms error = EINVAL; 1398189592Sbms goto out_inp_locked; 1399170613Sbms } 1400189592Sbms 1401189592Sbms /* 1402189592Sbms * Deal with error cases up-front: 1403189592Sbms * Asked to block, but already blocked; or 1404189592Sbms * Asked to unblock, but nothing to unblock. 1405189592Sbms * If adding a new block entry, allocate it. 1406189592Sbms */ 1407170613Sbms ims = imo_match_source(imo, idx, &ssa->sa); 1408189592Sbms if ((ims != NULL && doblock) || (ims == NULL && !doblock)) { 1409189592Sbms CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__, 1410189592Sbms inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not "); 1411189592Sbms error = EADDRNOTAVAIL; 1412189592Sbms goto out_inp_locked; 1413189592Sbms } 1414189592Sbms 1415189592Sbms INP_WLOCK_ASSERT(inp); 1416189592Sbms 1417189592Sbms /* 1418189592Sbms * Begin state merge transaction at socket layer. 1419189592Sbms */ 1420189592Sbms if (doblock) { 1421189592Sbms CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 1422189592Sbms ims = imf_graft(imf, fmode, &ssa->sin); 1423189592Sbms if (ims == NULL) 1424189592Sbms error = ENOMEM; 1425170613Sbms } else { 1426189592Sbms CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 1427189592Sbms error = imf_prune(imf, &ssa->sin); 1428170613Sbms } 1429170613Sbms 1430189592Sbms if (error) { 1431189592Sbms CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__); 1432189592Sbms goto out_imf_rollback; 1433189592Sbms } 1434189592Sbms 1435189592Sbms /* 1436189592Sbms * Begin state merge transaction at IGMP layer. 1437189592Sbms */ 1438189592Sbms IN_MULTI_LOCK(); 1439189592Sbms 1440189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 1441189592Sbms error = inm_merge(inm, imf); 1442189592Sbms if (error) { 1443189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 1444189592Sbms goto out_imf_rollback; 1445189592Sbms } 1446189592Sbms 1447189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 1448189592Sbms error = igmp_change_state(inm); 1449189592Sbms if (error) 1450189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 1451189592Sbms 1452189592Sbms IN_MULTI_UNLOCK(); 1453189592Sbms 1454189592Sbmsout_imf_rollback: 1455189592Sbms if (error) 1456189592Sbms imf_rollback(imf); 1457189592Sbms else 1458189592Sbms imf_commit(imf); 1459189592Sbms 1460189592Sbms imf_reap(imf); 1461189592Sbms 1462189592Sbmsout_inp_locked: 1463178285Srwatson INP_WUNLOCK(inp); 1464170613Sbms return (error); 1465170613Sbms} 1466170613Sbms 1467170613Sbms/* 1468170613Sbms * Given an inpcb, return its multicast options structure pointer. Accepts 1469170613Sbms * an unlocked inpcb pointer, but will return it locked. May sleep. 1470189592Sbms * 1471189592Sbms * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 1472189592Sbms * SMPng: NOTE: Returns with the INP write lock held. 1473170613Sbms */ 1474170613Sbmsstatic struct ip_moptions * 1475170613Sbmsinp_findmoptions(struct inpcb *inp) 1476170613Sbms{ 1477170613Sbms struct ip_moptions *imo; 1478170613Sbms struct in_multi **immp; 1479170613Sbms struct in_mfilter *imfp; 1480170613Sbms size_t idx; 1481170613Sbms 1482178285Srwatson INP_WLOCK(inp); 1483170613Sbms if (inp->inp_moptions != NULL) 1484170613Sbms return (inp->inp_moptions); 1485170613Sbms 1486178285Srwatson INP_WUNLOCK(inp); 1487170613Sbms 1488189592Sbms imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK); 1489189592Sbms immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS, 1490189592Sbms M_WAITOK | M_ZERO); 1491189592Sbms imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS, 1492189592Sbms M_INMFILTER, M_WAITOK); 1493170613Sbms 1494170613Sbms imo->imo_multicast_ifp = NULL; 1495170613Sbms imo->imo_multicast_addr.s_addr = INADDR_ANY; 1496170613Sbms imo->imo_multicast_vif = -1; 1497170613Sbms imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL; 1498189357Sbms imo->imo_multicast_loop = in_mcast_loop; 1499170613Sbms imo->imo_num_memberships = 0; 1500170613Sbms imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; 1501170613Sbms imo->imo_membership = immp; 1502170613Sbms 1503170613Sbms /* Initialize per-group source filters. */ 1504189592Sbms for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++) 1505189592Sbms imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE); 1506170613Sbms imo->imo_mfilters = imfp; 1507170613Sbms 1508178285Srwatson INP_WLOCK(inp); 1509170613Sbms if (inp->inp_moptions != NULL) { 1510189592Sbms free(imfp, M_INMFILTER); 1511170613Sbms free(immp, M_IPMOPTS); 1512170613Sbms free(imo, M_IPMOPTS); 1513170613Sbms return (inp->inp_moptions); 1514170613Sbms } 1515170613Sbms inp->inp_moptions = imo; 1516170613Sbms return (imo); 1517170613Sbms} 1518170613Sbms 1519170613Sbms/* 1520170613Sbms * Discard the IP multicast options (and source filters). 1521189592Sbms * 1522189592Sbms * SMPng: NOTE: assumes INP write lock is held. 1523170613Sbms */ 1524170613Sbmsvoid 1525170613Sbmsinp_freemoptions(struct ip_moptions *imo) 1526170613Sbms{ 1527170613Sbms struct in_mfilter *imf; 1528170613Sbms size_t idx, nmships; 1529170613Sbms 1530170613Sbms KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__)); 1531170613Sbms 1532170613Sbms nmships = imo->imo_num_memberships; 1533170613Sbms for (idx = 0; idx < nmships; ++idx) { 1534189592Sbms imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL; 1535189592Sbms if (imf) 1536189592Sbms imf_leave(imf); 1537189592Sbms (void)in_leavegroup(imo->imo_membership[idx], imf); 1538189592Sbms if (imf) 1539189592Sbms imf_purge(imf); 1540170613Sbms } 1541170613Sbms 1542189592Sbms if (imo->imo_mfilters) 1543189592Sbms free(imo->imo_mfilters, M_INMFILTER); 1544170613Sbms free(imo->imo_membership, M_IPMOPTS); 1545170613Sbms free(imo, M_IPMOPTS); 1546170613Sbms} 1547170613Sbms 1548170613Sbms/* 1549170613Sbms * Atomically get source filters on a socket for an IPv4 multicast group. 1550170613Sbms * Called with INP lock held; returns with lock released. 1551170613Sbms */ 1552170613Sbmsstatic int 1553170613Sbmsinp_get_source_filters(struct inpcb *inp, struct sockopt *sopt) 1554170613Sbms{ 1555170613Sbms struct __msfilterreq msfr; 1556170613Sbms sockunion_t *gsa; 1557170613Sbms struct ifnet *ifp; 1558170613Sbms struct ip_moptions *imo; 1559170613Sbms struct in_mfilter *imf; 1560189592Sbms struct ip_msource *ims; 1561189592Sbms struct in_msource *lims; 1562189592Sbms struct sockaddr_in *psin; 1563170613Sbms struct sockaddr_storage *ptss; 1564170613Sbms struct sockaddr_storage *tss; 1565170613Sbms int error; 1566189592Sbms size_t idx, nsrcs, ncsrcs; 1567170613Sbms 1568178285Srwatson INP_WLOCK_ASSERT(inp); 1569170613Sbms 1570170613Sbms imo = inp->inp_moptions; 1571170613Sbms KASSERT(imo != NULL, ("%s: null ip_moptions", __func__)); 1572170613Sbms 1573178285Srwatson INP_WUNLOCK(inp); 1574170613Sbms 1575170613Sbms error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 1576170613Sbms sizeof(struct __msfilterreq)); 1577170613Sbms if (error) 1578170613Sbms return (error); 1579170613Sbms 1580181803Sbz if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 1581170613Sbms return (EINVAL); 1582170613Sbms 1583170613Sbms ifp = ifnet_byindex(msfr.msfr_ifindex); 1584170613Sbms if (ifp == NULL) 1585170613Sbms return (EINVAL); 1586170613Sbms 1587178285Srwatson INP_WLOCK(inp); 1588170613Sbms 1589170613Sbms /* 1590170613Sbms * Lookup group on the socket. 1591170613Sbms */ 1592170613Sbms gsa = (sockunion_t *)&msfr.msfr_group; 1593170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 1594170613Sbms if (idx == -1 || imo->imo_mfilters == NULL) { 1595178285Srwatson INP_WUNLOCK(inp); 1596170613Sbms return (EADDRNOTAVAIL); 1597170613Sbms } 1598170613Sbms imf = &imo->imo_mfilters[idx]; 1599170613Sbms 1600170613Sbms /* 1601189592Sbms * Ignore memberships which are in limbo. 1602189592Sbms */ 1603189592Sbms if (imf->imf_st[1] == MCAST_UNDEFINED) { 1604189592Sbms INP_WUNLOCK(inp); 1605189592Sbms return (EAGAIN); 1606189592Sbms } 1607189592Sbms msfr.msfr_fmode = imf->imf_st[1]; 1608189592Sbms 1609189592Sbms /* 1610170613Sbms * If the user specified a buffer, copy out the source filter 1611170613Sbms * entries to userland gracefully. 1612189592Sbms * We only copy out the number of entries which userland 1613189592Sbms * has asked for, but we always tell userland how big the 1614189592Sbms * buffer really needs to be. 1615170613Sbms */ 1616170613Sbms tss = NULL; 1617170613Sbms if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) { 1618184214Sdes tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 1619189592Sbms M_TEMP, M_NOWAIT | M_ZERO); 1620170613Sbms if (tss == NULL) { 1621189592Sbms INP_WUNLOCK(inp); 1622189592Sbms return (ENOBUFS); 1623170613Sbms } 1624170613Sbms } 1625170613Sbms 1626189592Sbms /* 1627189592Sbms * Count number of sources in-mode at t0. 1628189592Sbms * If buffer space exists and remains, copy out source entries. 1629189592Sbms */ 1630189592Sbms nsrcs = msfr.msfr_nsrcs; 1631189592Sbms ncsrcs = 0; 1632189592Sbms ptss = tss; 1633189592Sbms RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) { 1634189592Sbms lims = (struct in_msource *)ims; 1635189592Sbms if (lims->imsl_st[0] == MCAST_UNDEFINED || 1636189592Sbms lims->imsl_st[0] != imf->imf_st[0]) 1637189592Sbms continue; 1638189592Sbms ++ncsrcs; 1639191659Sbms if (tss != NULL && nsrcs > 0) { 1640191659Sbms psin = (struct sockaddr_in *)ptss; 1641189592Sbms psin->sin_family = AF_INET; 1642189592Sbms psin->sin_len = sizeof(struct sockaddr_in); 1643189592Sbms psin->sin_addr.s_addr = htonl(lims->ims_haddr); 1644191659Sbms psin->sin_port = 0; 1645191659Sbms ++ptss; 1646191659Sbms --nsrcs; 1647189592Sbms } 1648189592Sbms } 1649189592Sbms 1650178285Srwatson INP_WUNLOCK(inp); 1651170613Sbms 1652170613Sbms if (tss != NULL) { 1653170613Sbms error = copyout(tss, msfr.msfr_srcs, 1654170613Sbms sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 1655184205Sdes free(tss, M_TEMP); 1656189592Sbms if (error) 1657189592Sbms return (error); 1658170613Sbms } 1659170613Sbms 1660189592Sbms msfr.msfr_nsrcs = ncsrcs; 1661170613Sbms error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq)); 1662170613Sbms 1663170613Sbms return (error); 1664170613Sbms} 1665170613Sbms 1666170613Sbms/* 1667170613Sbms * Return the IP multicast options in response to user getsockopt(). 1668170613Sbms */ 1669170613Sbmsint 1670170613Sbmsinp_getmoptions(struct inpcb *inp, struct sockopt *sopt) 1671170613Sbms{ 1672170613Sbms struct ip_mreqn mreqn; 1673170613Sbms struct ip_moptions *imo; 1674170613Sbms struct ifnet *ifp; 1675170613Sbms struct in_ifaddr *ia; 1676170613Sbms int error, optval; 1677170613Sbms u_char coptval; 1678170613Sbms 1679178285Srwatson INP_WLOCK(inp); 1680170613Sbms imo = inp->inp_moptions; 1681171746Scsjp /* 1682171746Scsjp * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 1683171746Scsjp * or is a divert socket, reject it. 1684171746Scsjp */ 1685171746Scsjp if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 1686171746Scsjp (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 1687171746Scsjp inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) { 1688178285Srwatson INP_WUNLOCK(inp); 1689171746Scsjp return (EOPNOTSUPP); 1690171746Scsjp } 1691170613Sbms 1692170613Sbms error = 0; 1693170613Sbms switch (sopt->sopt_name) { 1694170613Sbms case IP_MULTICAST_VIF: 1695170613Sbms if (imo != NULL) 1696170613Sbms optval = imo->imo_multicast_vif; 1697170613Sbms else 1698170613Sbms optval = -1; 1699178285Srwatson INP_WUNLOCK(inp); 1700170613Sbms error = sooptcopyout(sopt, &optval, sizeof(int)); 1701170613Sbms break; 1702170613Sbms 1703170613Sbms case IP_MULTICAST_IF: 1704170613Sbms memset(&mreqn, 0, sizeof(struct ip_mreqn)); 1705170613Sbms if (imo != NULL) { 1706170613Sbms ifp = imo->imo_multicast_ifp; 1707189592Sbms if (!in_nullhost(imo->imo_multicast_addr)) { 1708170613Sbms mreqn.imr_address = imo->imo_multicast_addr; 1709170613Sbms } else if (ifp != NULL) { 1710170613Sbms mreqn.imr_ifindex = ifp->if_index; 1711170613Sbms IFP_TO_IA(ifp, ia); 1712170613Sbms if (ia != NULL) { 1713170613Sbms mreqn.imr_address = 1714170613Sbms IA_SIN(ia)->sin_addr; 1715194760Srwatson ifa_free(&ia->ia_ifa); 1716170613Sbms } 1717170613Sbms } 1718170613Sbms } 1719178285Srwatson INP_WUNLOCK(inp); 1720170613Sbms if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 1721170613Sbms error = sooptcopyout(sopt, &mreqn, 1722170613Sbms sizeof(struct ip_mreqn)); 1723170613Sbms } else { 1724170613Sbms error = sooptcopyout(sopt, &mreqn.imr_address, 1725170613Sbms sizeof(struct in_addr)); 1726170613Sbms } 1727170613Sbms break; 1728170613Sbms 1729170613Sbms case IP_MULTICAST_TTL: 1730170613Sbms if (imo == 0) 1731170613Sbms optval = coptval = IP_DEFAULT_MULTICAST_TTL; 1732170613Sbms else 1733170613Sbms optval = coptval = imo->imo_multicast_ttl; 1734178285Srwatson INP_WUNLOCK(inp); 1735170613Sbms if (sopt->sopt_valsize == sizeof(u_char)) 1736170613Sbms error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1737170613Sbms else 1738170613Sbms error = sooptcopyout(sopt, &optval, sizeof(int)); 1739170613Sbms break; 1740170613Sbms 1741170613Sbms case IP_MULTICAST_LOOP: 1742170613Sbms if (imo == 0) 1743170613Sbms optval = coptval = IP_DEFAULT_MULTICAST_LOOP; 1744170613Sbms else 1745170613Sbms optval = coptval = imo->imo_multicast_loop; 1746178285Srwatson INP_WUNLOCK(inp); 1747170613Sbms if (sopt->sopt_valsize == sizeof(u_char)) 1748170613Sbms error = sooptcopyout(sopt, &coptval, sizeof(u_char)); 1749170613Sbms else 1750170613Sbms error = sooptcopyout(sopt, &optval, sizeof(int)); 1751170613Sbms break; 1752170613Sbms 1753170613Sbms case IP_MSFILTER: 1754170613Sbms if (imo == NULL) { 1755170613Sbms error = EADDRNOTAVAIL; 1756178285Srwatson INP_WUNLOCK(inp); 1757170613Sbms } else { 1758170613Sbms error = inp_get_source_filters(inp, sopt); 1759170613Sbms } 1760170613Sbms break; 1761170613Sbms 1762170613Sbms default: 1763178285Srwatson INP_WUNLOCK(inp); 1764170613Sbms error = ENOPROTOOPT; 1765170613Sbms break; 1766170613Sbms } 1767170613Sbms 1768170613Sbms INP_UNLOCK_ASSERT(inp); 1769170613Sbms 1770170613Sbms return (error); 1771170613Sbms} 1772170613Sbms 1773170613Sbms/* 1774189592Sbms * Look up the ifnet to use for a multicast group membership, 1775189592Sbms * given the IPv4 address of an interface, and the IPv4 group address. 1776189592Sbms * 1777189592Sbms * This routine exists to support legacy multicast applications 1778189592Sbms * which do not understand that multicast memberships are scoped to 1779189592Sbms * specific physical links in the networking stack, or which need 1780189592Sbms * to join link-scope groups before IPv4 addresses are configured. 1781189592Sbms * 1782189592Sbms * If inp is non-NULL, use this socket's current FIB number for any 1783189592Sbms * required FIB lookup. 1784189592Sbms * If ina is INADDR_ANY, look up the group address in the unicast FIB, 1785189592Sbms * and use its ifp; usually, this points to the default next-hop. 1786189592Sbms * 1787189592Sbms * If the FIB lookup fails, attempt to use the first non-loopback 1788189592Sbms * interface with multicast capability in the system as a 1789189592Sbms * last resort. The legacy IPv4 ASM API requires that we do 1790189592Sbms * this in order to allow groups to be joined when the routing 1791189592Sbms * table has not yet been populated during boot. 1792189592Sbms * 1793189592Sbms * Returns NULL if no ifp could be found. 1794189592Sbms * 1795189592Sbms * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP. 1796189592Sbms * FUTURE: Implement IPv4 source-address selection. 1797189592Sbms */ 1798189592Sbmsstatic struct ifnet * 1799189592Sbmsinp_lookup_mcast_ifp(const struct inpcb *inp, 1800189592Sbms const struct sockaddr_in *gsin, const struct in_addr ina) 1801189592Sbms{ 1802189592Sbms struct ifnet *ifp; 1803189592Sbms 1804189592Sbms KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__)); 1805189592Sbms KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)), 1806189592Sbms ("%s: not multicast", __func__)); 1807189592Sbms 1808189592Sbms ifp = NULL; 1809189592Sbms if (!in_nullhost(ina)) { 1810189592Sbms INADDR_TO_IFP(ina, ifp); 1811189592Sbms } else { 1812189592Sbms struct route ro; 1813189592Sbms 1814189592Sbms ro.ro_rt = NULL; 1815189592Sbms memcpy(&ro.ro_dst, gsin, sizeof(struct sockaddr_in)); 1816189592Sbms in_rtalloc_ign(&ro, 0, inp ? inp->inp_inc.inc_fibnum : 0); 1817189592Sbms if (ro.ro_rt != NULL) { 1818189592Sbms ifp = ro.ro_rt->rt_ifp; 1819189592Sbms KASSERT(ifp != NULL, ("%s: null ifp", __func__)); 1820189592Sbms RTFREE(ro.ro_rt); 1821189592Sbms } else { 1822189592Sbms struct in_ifaddr *ia; 1823189592Sbms struct ifnet *mifp; 1824189592Sbms 1825189592Sbms mifp = NULL; 1826194951Srwatson IN_IFADDR_RLOCK(); 1827189592Sbms TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1828189592Sbms mifp = ia->ia_ifp; 1829189592Sbms if (!(mifp->if_flags & IFF_LOOPBACK) && 1830189592Sbms (mifp->if_flags & IFF_MULTICAST)) { 1831189592Sbms ifp = mifp; 1832189592Sbms break; 1833189592Sbms } 1834189592Sbms } 1835194951Srwatson IN_IFADDR_RUNLOCK(); 1836189592Sbms } 1837189592Sbms } 1838189592Sbms 1839189592Sbms return (ifp); 1840189592Sbms} 1841189592Sbms 1842189592Sbms/* 1843170613Sbms * Join an IPv4 multicast group, possibly with a source. 1844170613Sbms */ 1845170613Sbmsstatic int 1846170613Sbmsinp_join_group(struct inpcb *inp, struct sockopt *sopt) 1847170613Sbms{ 1848170613Sbms struct group_source_req gsr; 1849170613Sbms sockunion_t *gsa, *ssa; 1850170613Sbms struct ifnet *ifp; 1851170613Sbms struct in_mfilter *imf; 1852170613Sbms struct ip_moptions *imo; 1853170613Sbms struct in_multi *inm; 1854189592Sbms struct in_msource *lims; 1855170613Sbms size_t idx; 1856189592Sbms int error, is_new; 1857170613Sbms 1858170613Sbms ifp = NULL; 1859189592Sbms imf = NULL; 1860170613Sbms error = 0; 1861189592Sbms is_new = 0; 1862170613Sbms 1863170613Sbms memset(&gsr, 0, sizeof(struct group_source_req)); 1864170613Sbms gsa = (sockunion_t *)&gsr.gsr_group; 1865170613Sbms gsa->ss.ss_family = AF_UNSPEC; 1866170613Sbms ssa = (sockunion_t *)&gsr.gsr_source; 1867170613Sbms ssa->ss.ss_family = AF_UNSPEC; 1868170613Sbms 1869170613Sbms switch (sopt->sopt_name) { 1870170613Sbms case IP_ADD_MEMBERSHIP: 1871170613Sbms case IP_ADD_SOURCE_MEMBERSHIP: { 1872170613Sbms struct ip_mreq_source mreqs; 1873170613Sbms 1874170613Sbms if (sopt->sopt_name == IP_ADD_MEMBERSHIP) { 1875170613Sbms error = sooptcopyin(sopt, &mreqs, 1876170613Sbms sizeof(struct ip_mreq), 1877170613Sbms sizeof(struct ip_mreq)); 1878170613Sbms /* 1879170613Sbms * Do argument switcharoo from ip_mreq into 1880170613Sbms * ip_mreq_source to avoid using two instances. 1881170613Sbms */ 1882170613Sbms mreqs.imr_interface = mreqs.imr_sourceaddr; 1883170613Sbms mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 1884170613Sbms } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) { 1885170613Sbms error = sooptcopyin(sopt, &mreqs, 1886170613Sbms sizeof(struct ip_mreq_source), 1887170613Sbms sizeof(struct ip_mreq_source)); 1888170613Sbms } 1889170613Sbms if (error) 1890170613Sbms return (error); 1891170613Sbms 1892170613Sbms gsa->sin.sin_family = AF_INET; 1893170613Sbms gsa->sin.sin_len = sizeof(struct sockaddr_in); 1894170613Sbms gsa->sin.sin_addr = mreqs.imr_multiaddr; 1895170613Sbms 1896170613Sbms if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) { 1897170613Sbms ssa->sin.sin_family = AF_INET; 1898170613Sbms ssa->sin.sin_len = sizeof(struct sockaddr_in); 1899170613Sbms ssa->sin.sin_addr = mreqs.imr_sourceaddr; 1900170613Sbms } 1901170613Sbms 1902196932Ssyrinx if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1903196932Ssyrinx return (EINVAL); 1904196932Ssyrinx 1905189592Sbms ifp = inp_lookup_mcast_ifp(inp, &gsa->sin, 1906189592Sbms mreqs.imr_interface); 1907189592Sbms CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p", 1908189592Sbms __func__, inet_ntoa(mreqs.imr_interface), ifp); 1909170613Sbms break; 1910170613Sbms } 1911170613Sbms 1912170613Sbms case MCAST_JOIN_GROUP: 1913170613Sbms case MCAST_JOIN_SOURCE_GROUP: 1914170613Sbms if (sopt->sopt_name == MCAST_JOIN_GROUP) { 1915170613Sbms error = sooptcopyin(sopt, &gsr, 1916170613Sbms sizeof(struct group_req), 1917170613Sbms sizeof(struct group_req)); 1918170613Sbms } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 1919170613Sbms error = sooptcopyin(sopt, &gsr, 1920170613Sbms sizeof(struct group_source_req), 1921170613Sbms sizeof(struct group_source_req)); 1922170613Sbms } 1923170613Sbms if (error) 1924170613Sbms return (error); 1925170613Sbms 1926170613Sbms if (gsa->sin.sin_family != AF_INET || 1927170613Sbms gsa->sin.sin_len != sizeof(struct sockaddr_in)) 1928170613Sbms return (EINVAL); 1929170613Sbms 1930170613Sbms /* 1931170613Sbms * Overwrite the port field if present, as the sockaddr 1932170613Sbms * being copied in may be matched with a binary comparison. 1933170613Sbms */ 1934170613Sbms gsa->sin.sin_port = 0; 1935170613Sbms if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) { 1936170613Sbms if (ssa->sin.sin_family != AF_INET || 1937170613Sbms ssa->sin.sin_len != sizeof(struct sockaddr_in)) 1938170613Sbms return (EINVAL); 1939170613Sbms ssa->sin.sin_port = 0; 1940170613Sbms } 1941170613Sbms 1942196932Ssyrinx if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 1943196932Ssyrinx return (EINVAL); 1944196932Ssyrinx 1945181803Sbz if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 1946170613Sbms return (EADDRNOTAVAIL); 1947170613Sbms ifp = ifnet_byindex(gsr.gsr_interface); 1948170613Sbms break; 1949170613Sbms 1950170613Sbms default: 1951189592Sbms CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 1952189592Sbms __func__, sopt->sopt_name); 1953170613Sbms return (EOPNOTSUPP); 1954170613Sbms break; 1955170613Sbms } 1956170613Sbms 1957170613Sbms if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) 1958170613Sbms return (EADDRNOTAVAIL); 1959170613Sbms 1960170613Sbms /* 1961189592Sbms * MCAST_JOIN_SOURCE on an exclusive membership is an error. 1962189592Sbms * On an existing inclusive membership, it just adds the 1963189592Sbms * source to the filter list. 1964170613Sbms */ 1965170613Sbms imo = inp_findmoptions(inp); 1966170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 1967189592Sbms if (idx == -1) { 1968189592Sbms is_new = 1; 1969189592Sbms } else { 1970189592Sbms inm = imo->imo_membership[idx]; 1971189592Sbms imf = &imo->imo_mfilters[idx]; 1972189592Sbms if (ssa->ss.ss_family != AF_UNSPEC && 1973189592Sbms imf->imf_st[1] != MCAST_INCLUDE) { 1974189592Sbms error = EINVAL; 1975189592Sbms goto out_inp_locked; 1976189592Sbms } 1977189592Sbms lims = imo_match_source(imo, idx, &ssa->sa); 1978189592Sbms if (lims != NULL) { 1979170613Sbms error = EADDRNOTAVAIL; 1980189592Sbms goto out_inp_locked; 1981170613Sbms } 1982170613Sbms } 1983170613Sbms 1984170613Sbms /* 1985189592Sbms * Begin state merge transaction at socket layer. 1986170613Sbms */ 1987189592Sbms INP_WLOCK_ASSERT(inp); 1988189592Sbms 1989189592Sbms if (is_new) { 1990189592Sbms if (imo->imo_num_memberships == imo->imo_max_memberships) { 1991189592Sbms error = imo_grow(imo); 1992189592Sbms if (error) 1993189592Sbms goto out_inp_locked; 1994189592Sbms } 1995189592Sbms /* 1996189592Sbms * Allocate the new slot upfront so we can deal with 1997189592Sbms * grafting the new source filter in same code path 1998189592Sbms * as for join-source on existing membership. 1999189592Sbms */ 2000189592Sbms idx = imo->imo_num_memberships; 2001189592Sbms imo->imo_membership[idx] = NULL; 2002189592Sbms imo->imo_num_memberships++; 2003189592Sbms KASSERT(imo->imo_mfilters != NULL, 2004189592Sbms ("%s: imf_mfilters vector was not allocated", __func__)); 2005189592Sbms imf = &imo->imo_mfilters[idx]; 2006189592Sbms KASSERT(RB_EMPTY(&imf->imf_sources), 2007189592Sbms ("%s: imf_sources not empty", __func__)); 2008170613Sbms } 2009170613Sbms 2010170613Sbms /* 2011189592Sbms * Graft new source into filter list for this inpcb's 2012189592Sbms * membership of the group. The in_multi may not have 2013189592Sbms * been allocated yet if this is a new membership. 2014170613Sbms */ 2015189592Sbms if (ssa->ss.ss_family != AF_UNSPEC) { 2016189592Sbms /* Membership starts in IN mode */ 2017189592Sbms if (is_new) { 2018189592Sbms CTR1(KTR_IGMPV3, "%s: new join w/source", __func__); 2019189592Sbms imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE); 2020189592Sbms } else { 2021189592Sbms CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow"); 2022189592Sbms } 2023189592Sbms lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin); 2024189592Sbms if (lims == NULL) { 2025189592Sbms CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2026189592Sbms __func__); 2027189592Sbms error = ENOMEM; 2028189592Sbms goto out_imo_free; 2029189592Sbms } 2030170613Sbms } 2031170613Sbms 2032170613Sbms /* 2033189592Sbms * Begin state merge transaction at IGMP layer. 2034170613Sbms */ 2035189592Sbms IN_MULTI_LOCK(); 2036170613Sbms 2037189592Sbms if (is_new) { 2038189592Sbms error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf, 2039189592Sbms &inm); 2040189592Sbms if (error) 2041189592Sbms goto out_imo_free; 2042189592Sbms imo->imo_membership[idx] = inm; 2043189592Sbms } else { 2044189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2045189592Sbms error = inm_merge(inm, imf); 2046170613Sbms if (error) { 2047189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2048189592Sbms __func__); 2049189592Sbms goto out_imf_rollback; 2050170613Sbms } 2051189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2052189592Sbms error = igmp_change_state(inm); 2053189592Sbms if (error) { 2054189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2055189592Sbms __func__); 2056189592Sbms goto out_imf_rollback; 2057189592Sbms } 2058170613Sbms } 2059170613Sbms 2060189592Sbms IN_MULTI_UNLOCK(); 2061189592Sbms 2062189592Sbmsout_imf_rollback: 2063189592Sbms INP_WLOCK_ASSERT(inp); 2064189592Sbms if (error) { 2065189592Sbms imf_rollback(imf); 2066189592Sbms if (is_new) 2067189592Sbms imf_purge(imf); 2068189592Sbms else 2069189592Sbms imf_reap(imf); 2070189592Sbms } else { 2071189592Sbms imf_commit(imf); 2072189592Sbms } 2073189592Sbms 2074189592Sbmsout_imo_free: 2075189592Sbms if (error && is_new) { 2076189592Sbms imo->imo_membership[idx] = NULL; 2077189592Sbms --imo->imo_num_memberships; 2078189592Sbms } 2079189592Sbms 2080189592Sbmsout_inp_locked: 2081178285Srwatson INP_WUNLOCK(inp); 2082170613Sbms return (error); 2083170613Sbms} 2084170613Sbms 2085170613Sbms/* 2086170613Sbms * Leave an IPv4 multicast group on an inpcb, possibly with a source. 2087170613Sbms */ 2088170613Sbmsstatic int 2089170613Sbmsinp_leave_group(struct inpcb *inp, struct sockopt *sopt) 2090170613Sbms{ 2091170613Sbms struct group_source_req gsr; 2092170613Sbms struct ip_mreq_source mreqs; 2093170613Sbms sockunion_t *gsa, *ssa; 2094170613Sbms struct ifnet *ifp; 2095170613Sbms struct in_mfilter *imf; 2096170613Sbms struct ip_moptions *imo; 2097189592Sbms struct in_msource *ims; 2098170613Sbms struct in_multi *inm; 2099170613Sbms size_t idx; 2100189592Sbms int error, is_final; 2101170613Sbms 2102170613Sbms ifp = NULL; 2103170613Sbms error = 0; 2104189592Sbms is_final = 1; 2105170613Sbms 2106170613Sbms memset(&gsr, 0, sizeof(struct group_source_req)); 2107170613Sbms gsa = (sockunion_t *)&gsr.gsr_group; 2108170613Sbms gsa->ss.ss_family = AF_UNSPEC; 2109170613Sbms ssa = (sockunion_t *)&gsr.gsr_source; 2110170613Sbms ssa->ss.ss_family = AF_UNSPEC; 2111170613Sbms 2112170613Sbms switch (sopt->sopt_name) { 2113170613Sbms case IP_DROP_MEMBERSHIP: 2114170613Sbms case IP_DROP_SOURCE_MEMBERSHIP: 2115170613Sbms if (sopt->sopt_name == IP_DROP_MEMBERSHIP) { 2116170613Sbms error = sooptcopyin(sopt, &mreqs, 2117170613Sbms sizeof(struct ip_mreq), 2118170613Sbms sizeof(struct ip_mreq)); 2119170613Sbms /* 2120170613Sbms * Swap interface and sourceaddr arguments, 2121170613Sbms * as ip_mreq and ip_mreq_source are laid 2122170613Sbms * out differently. 2123170613Sbms */ 2124170613Sbms mreqs.imr_interface = mreqs.imr_sourceaddr; 2125170613Sbms mreqs.imr_sourceaddr.s_addr = INADDR_ANY; 2126170613Sbms } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2127170613Sbms error = sooptcopyin(sopt, &mreqs, 2128170613Sbms sizeof(struct ip_mreq_source), 2129170613Sbms sizeof(struct ip_mreq_source)); 2130170613Sbms } 2131170613Sbms if (error) 2132170613Sbms return (error); 2133170613Sbms 2134170613Sbms gsa->sin.sin_family = AF_INET; 2135170613Sbms gsa->sin.sin_len = sizeof(struct sockaddr_in); 2136170613Sbms gsa->sin.sin_addr = mreqs.imr_multiaddr; 2137170613Sbms 2138170613Sbms if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) { 2139170613Sbms ssa->sin.sin_family = AF_INET; 2140170613Sbms ssa->sin.sin_len = sizeof(struct sockaddr_in); 2141170613Sbms ssa->sin.sin_addr = mreqs.imr_sourceaddr; 2142170613Sbms } 2143170613Sbms 2144189592Sbms if (!in_nullhost(gsa->sin.sin_addr)) 2145170613Sbms INADDR_TO_IFP(mreqs.imr_interface, ifp); 2146170613Sbms 2147189592Sbms CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p", 2148189592Sbms __func__, inet_ntoa(mreqs.imr_interface), ifp); 2149189592Sbms 2150170613Sbms break; 2151170613Sbms 2152170613Sbms case MCAST_LEAVE_GROUP: 2153170613Sbms case MCAST_LEAVE_SOURCE_GROUP: 2154170613Sbms if (sopt->sopt_name == MCAST_LEAVE_GROUP) { 2155170613Sbms error = sooptcopyin(sopt, &gsr, 2156170613Sbms sizeof(struct group_req), 2157170613Sbms sizeof(struct group_req)); 2158170613Sbms } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2159170613Sbms error = sooptcopyin(sopt, &gsr, 2160170613Sbms sizeof(struct group_source_req), 2161170613Sbms sizeof(struct group_source_req)); 2162170613Sbms } 2163170613Sbms if (error) 2164170613Sbms return (error); 2165170613Sbms 2166170613Sbms if (gsa->sin.sin_family != AF_INET || 2167170613Sbms gsa->sin.sin_len != sizeof(struct sockaddr_in)) 2168170613Sbms return (EINVAL); 2169170613Sbms 2170170613Sbms if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) { 2171170613Sbms if (ssa->sin.sin_family != AF_INET || 2172170613Sbms ssa->sin.sin_len != sizeof(struct sockaddr_in)) 2173170613Sbms return (EINVAL); 2174170613Sbms } 2175170613Sbms 2176181803Sbz if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface) 2177170613Sbms return (EADDRNOTAVAIL); 2178170613Sbms 2179170613Sbms ifp = ifnet_byindex(gsr.gsr_interface); 2180170613Sbms break; 2181170613Sbms 2182170613Sbms default: 2183189592Sbms CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d", 2184189592Sbms __func__, sopt->sopt_name); 2185170613Sbms return (EOPNOTSUPP); 2186170613Sbms break; 2187170613Sbms } 2188170613Sbms 2189170613Sbms if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2190170613Sbms return (EINVAL); 2191170613Sbms 2192170613Sbms /* 2193170613Sbms * Find the membership in the membership array. 2194170613Sbms */ 2195170613Sbms imo = inp_findmoptions(inp); 2196170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 2197170613Sbms if (idx == -1) { 2198170613Sbms error = EADDRNOTAVAIL; 2199189592Sbms goto out_inp_locked; 2200170613Sbms } 2201189592Sbms inm = imo->imo_membership[idx]; 2202170613Sbms imf = &imo->imo_mfilters[idx]; 2203170613Sbms 2204189592Sbms if (ssa->ss.ss_family != AF_UNSPEC) 2205189592Sbms is_final = 0; 2206189592Sbms 2207170613Sbms /* 2208189592Sbms * Begin state merge transaction at socket layer. 2209189592Sbms */ 2210189592Sbms INP_WLOCK_ASSERT(inp); 2211189592Sbms 2212189592Sbms /* 2213170613Sbms * If we were instructed only to leave a given source, do so. 2214189592Sbms * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships. 2215170613Sbms */ 2216189592Sbms if (is_final) { 2217189592Sbms imf_leave(imf); 2218189592Sbms } else { 2219189592Sbms if (imf->imf_st[0] == MCAST_EXCLUDE) { 2220189592Sbms error = EADDRNOTAVAIL; 2221189592Sbms goto out_inp_locked; 2222170613Sbms } 2223189592Sbms ims = imo_match_source(imo, idx, &ssa->sa); 2224189592Sbms if (ims == NULL) { 2225189592Sbms CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__, 2226189592Sbms inet_ntoa(ssa->sin.sin_addr), "not "); 2227189592Sbms error = EADDRNOTAVAIL; 2228189592Sbms goto out_inp_locked; 2229189592Sbms } 2230189592Sbms CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block"); 2231189592Sbms error = imf_prune(imf, &ssa->sin); 2232189592Sbms if (error) { 2233189592Sbms CTR1(KTR_IGMPV3, "%s: merge imf state failed", 2234189592Sbms __func__); 2235189592Sbms goto out_inp_locked; 2236189592Sbms } 2237170613Sbms } 2238170613Sbms 2239170613Sbms /* 2240189592Sbms * Begin state merge transaction at IGMP layer. 2241170613Sbms */ 2242189592Sbms IN_MULTI_LOCK(); 2243170613Sbms 2244189592Sbms if (is_final) { 2245189592Sbms /* 2246189592Sbms * Give up the multicast address record to which 2247189592Sbms * the membership points. 2248189592Sbms */ 2249189592Sbms (void)in_leavegroup_locked(inm, imf); 2250189592Sbms } else { 2251189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2252189592Sbms error = inm_merge(inm, imf); 2253189592Sbms if (error) { 2254189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", 2255189592Sbms __func__); 2256189592Sbms goto out_imf_rollback; 2257170613Sbms } 2258189592Sbms 2259189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2260189592Sbms error = igmp_change_state(inm); 2261189592Sbms if (error) { 2262189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", 2263189592Sbms __func__); 2264189592Sbms } 2265170613Sbms } 2266170613Sbms 2267189592Sbms IN_MULTI_UNLOCK(); 2268170613Sbms 2269189592Sbmsout_imf_rollback: 2270189592Sbms if (error) 2271189592Sbms imf_rollback(imf); 2272189592Sbms else 2273189592Sbms imf_commit(imf); 2274189592Sbms 2275189592Sbms imf_reap(imf); 2276189592Sbms 2277189592Sbms if (is_final) { 2278189592Sbms /* Remove the gap in the membership array. */ 2279189592Sbms for (++idx; idx < imo->imo_num_memberships; ++idx) 2280189592Sbms imo->imo_membership[idx-1] = imo->imo_membership[idx]; 2281189592Sbms imo->imo_num_memberships--; 2282189592Sbms } 2283189592Sbms 2284189592Sbmsout_inp_locked: 2285178285Srwatson INP_WUNLOCK(inp); 2286170613Sbms return (error); 2287170613Sbms} 2288170613Sbms 2289170613Sbms/* 2290170613Sbms * Select the interface for transmitting IPv4 multicast datagrams. 2291170613Sbms * 2292170613Sbms * Either an instance of struct in_addr or an instance of struct ip_mreqn 2293170613Sbms * may be passed to this socket option. An address of INADDR_ANY or an 2294170613Sbms * interface index of 0 is used to remove a previous selection. 2295170613Sbms * When no interface is selected, one is chosen for every send. 2296170613Sbms */ 2297170613Sbmsstatic int 2298170613Sbmsinp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt) 2299170613Sbms{ 2300170613Sbms struct in_addr addr; 2301170613Sbms struct ip_mreqn mreqn; 2302170613Sbms struct ifnet *ifp; 2303170613Sbms struct ip_moptions *imo; 2304170613Sbms int error; 2305170613Sbms 2306170613Sbms if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) { 2307170613Sbms /* 2308170613Sbms * An interface index was specified using the 2309170613Sbms * Linux-derived ip_mreqn structure. 2310170613Sbms */ 2311170613Sbms error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn), 2312170613Sbms sizeof(struct ip_mreqn)); 2313170613Sbms if (error) 2314170613Sbms return (error); 2315170613Sbms 2316181803Sbz if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex) 2317170613Sbms return (EINVAL); 2318170613Sbms 2319170613Sbms if (mreqn.imr_ifindex == 0) { 2320170613Sbms ifp = NULL; 2321170613Sbms } else { 2322170613Sbms ifp = ifnet_byindex(mreqn.imr_ifindex); 2323170613Sbms if (ifp == NULL) 2324170613Sbms return (EADDRNOTAVAIL); 2325170613Sbms } 2326170613Sbms } else { 2327170613Sbms /* 2328170613Sbms * An interface was specified by IPv4 address. 2329170613Sbms * This is the traditional BSD usage. 2330170613Sbms */ 2331170613Sbms error = sooptcopyin(sopt, &addr, sizeof(struct in_addr), 2332170613Sbms sizeof(struct in_addr)); 2333170613Sbms if (error) 2334170613Sbms return (error); 2335189592Sbms if (in_nullhost(addr)) { 2336170613Sbms ifp = NULL; 2337170613Sbms } else { 2338170613Sbms INADDR_TO_IFP(addr, ifp); 2339170613Sbms if (ifp == NULL) 2340170613Sbms return (EADDRNOTAVAIL); 2341170613Sbms } 2342189592Sbms CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp, 2343189592Sbms inet_ntoa(addr)); 2344170613Sbms } 2345170613Sbms 2346170613Sbms /* Reject interfaces which do not support multicast. */ 2347170613Sbms if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0) 2348170613Sbms return (EOPNOTSUPP); 2349170613Sbms 2350170613Sbms imo = inp_findmoptions(inp); 2351170613Sbms imo->imo_multicast_ifp = ifp; 2352170613Sbms imo->imo_multicast_addr.s_addr = INADDR_ANY; 2353178285Srwatson INP_WUNLOCK(inp); 2354170613Sbms 2355170613Sbms return (0); 2356170613Sbms} 2357170613Sbms 2358170613Sbms/* 2359170613Sbms * Atomically set source filters on a socket for an IPv4 multicast group. 2360189592Sbms * 2361189592Sbms * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held. 2362170613Sbms */ 2363170613Sbmsstatic int 2364170613Sbmsinp_set_source_filters(struct inpcb *inp, struct sockopt *sopt) 2365170613Sbms{ 2366170613Sbms struct __msfilterreq msfr; 2367170613Sbms sockunion_t *gsa; 2368170613Sbms struct ifnet *ifp; 2369170613Sbms struct in_mfilter *imf; 2370170613Sbms struct ip_moptions *imo; 2371189592Sbms struct in_multi *inm; 2372170613Sbms size_t idx; 2373170613Sbms int error; 2374170613Sbms 2375170613Sbms error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq), 2376170613Sbms sizeof(struct __msfilterreq)); 2377170613Sbms if (error) 2378170613Sbms return (error); 2379170613Sbms 2380189592Sbms if (msfr.msfr_nsrcs > in_mcast_maxsocksrc || 2381170613Sbms (msfr.msfr_fmode != MCAST_EXCLUDE && 2382170613Sbms msfr.msfr_fmode != MCAST_INCLUDE)) 2383170613Sbms return (EINVAL); 2384170613Sbms 2385170613Sbms if (msfr.msfr_group.ss_family != AF_INET || 2386170613Sbms msfr.msfr_group.ss_len != sizeof(struct sockaddr_in)) 2387170613Sbms return (EINVAL); 2388170613Sbms 2389170613Sbms gsa = (sockunion_t *)&msfr.msfr_group; 2390170613Sbms if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr))) 2391170613Sbms return (EINVAL); 2392170613Sbms 2393170613Sbms gsa->sin.sin_port = 0; /* ignore port */ 2394170613Sbms 2395181803Sbz if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex) 2396170613Sbms return (EADDRNOTAVAIL); 2397170613Sbms 2398170613Sbms ifp = ifnet_byindex(msfr.msfr_ifindex); 2399170613Sbms if (ifp == NULL) 2400170613Sbms return (EADDRNOTAVAIL); 2401170613Sbms 2402170613Sbms /* 2403189592Sbms * Take the INP write lock. 2404170613Sbms * Check if this socket is a member of this group. 2405170613Sbms */ 2406170613Sbms imo = inp_findmoptions(inp); 2407170613Sbms idx = imo_match_group(imo, ifp, &gsa->sa); 2408170613Sbms if (idx == -1 || imo->imo_mfilters == NULL) { 2409170613Sbms error = EADDRNOTAVAIL; 2410189592Sbms goto out_inp_locked; 2411170613Sbms } 2412189592Sbms inm = imo->imo_membership[idx]; 2413170613Sbms imf = &imo->imo_mfilters[idx]; 2414170613Sbms 2415170613Sbms /* 2416189592Sbms * Begin state merge transaction at socket layer. 2417170613Sbms */ 2418189592Sbms INP_WLOCK_ASSERT(inp); 2419170613Sbms 2420189592Sbms imf->imf_st[1] = msfr.msfr_fmode; 2421189592Sbms 2422170613Sbms /* 2423170613Sbms * Apply any new source filters, if present. 2424189592Sbms * Make a copy of the user-space source vector so 2425189592Sbms * that we may copy them with a single copyin. This 2426189592Sbms * allows us to deal with page faults up-front. 2427170613Sbms */ 2428170613Sbms if (msfr.msfr_nsrcs > 0) { 2429189592Sbms struct in_msource *lims; 2430189592Sbms struct sockaddr_in *psin; 2431189592Sbms struct sockaddr_storage *kss, *pkss; 2432189592Sbms int i; 2433170613Sbms 2434178285Srwatson INP_WUNLOCK(inp); 2435189592Sbms 2436189592Sbms CTR2(KTR_IGMPV3, "%s: loading %lu source list entries", 2437189592Sbms __func__, (unsigned long)msfr.msfr_nsrcs); 2438184214Sdes kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs, 2439170613Sbms M_TEMP, M_WAITOK); 2440170613Sbms error = copyin(msfr.msfr_srcs, kss, 2441170613Sbms sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs); 2442170613Sbms if (error) { 2443184205Sdes free(kss, M_TEMP); 2444170613Sbms return (error); 2445170613Sbms } 2446170613Sbms 2447189592Sbms INP_WLOCK(inp); 2448189592Sbms 2449170613Sbms /* 2450189592Sbms * Mark all source filters as UNDEFINED at t1. 2451189592Sbms * Restore new group filter mode, as imf_leave() 2452189592Sbms * will set it to INCLUDE. 2453170613Sbms */ 2454189592Sbms imf_leave(imf); 2455189592Sbms imf->imf_st[1] = msfr.msfr_fmode; 2456189592Sbms 2457189592Sbms /* 2458189592Sbms * Update socket layer filters at t1, lazy-allocating 2459189592Sbms * new entries. This saves a bunch of memory at the 2460189592Sbms * cost of one RB_FIND() per source entry; duplicate 2461189592Sbms * entries in the msfr_nsrcs vector are ignored. 2462189592Sbms * If we encounter an error, rollback transaction. 2463189592Sbms * 2464189592Sbms * XXX This too could be replaced with a set-symmetric 2465189592Sbms * difference like loop to avoid walking from root 2466189592Sbms * every time, as the key space is common. 2467189592Sbms */ 2468189592Sbms for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) { 2469189592Sbms psin = (struct sockaddr_in *)pkss; 2470189592Sbms if (psin->sin_family != AF_INET) { 2471170613Sbms error = EAFNOSUPPORT; 2472170613Sbms break; 2473170613Sbms } 2474189592Sbms if (psin->sin_len != sizeof(struct sockaddr_in)) { 2475189592Sbms error = EINVAL; 2476189592Sbms break; 2477189592Sbms } 2478189592Sbms error = imf_get_source(imf, psin, &lims); 2479170613Sbms if (error) 2480170613Sbms break; 2481189592Sbms lims->imsl_st[1] = imf->imf_st[1]; 2482170613Sbms } 2483189592Sbms free(kss, M_TEMP); 2484189592Sbms } 2485170613Sbms 2486189592Sbms if (error) 2487189592Sbms goto out_imf_rollback; 2488170613Sbms 2489189592Sbms INP_WLOCK_ASSERT(inp); 2490189592Sbms IN_MULTI_LOCK(); 2491170613Sbms 2492170613Sbms /* 2493189592Sbms * Begin state merge transaction at IGMP layer. 2494170613Sbms */ 2495189592Sbms CTR1(KTR_IGMPV3, "%s: merge inm state", __func__); 2496189592Sbms error = inm_merge(inm, imf); 2497189592Sbms if (error) { 2498189592Sbms CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__); 2499189592Sbms goto out_imf_rollback; 2500189592Sbms } 2501170613Sbms 2502189592Sbms CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__); 2503189592Sbms error = igmp_change_state(inm); 2504189592Sbms if (error) 2505189592Sbms CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__); 2506189592Sbms 2507189592Sbms IN_MULTI_UNLOCK(); 2508189592Sbms 2509189592Sbmsout_imf_rollback: 2510189592Sbms if (error) 2511189592Sbms imf_rollback(imf); 2512189592Sbms else 2513189592Sbms imf_commit(imf); 2514189592Sbms 2515189592Sbms imf_reap(imf); 2516189592Sbms 2517189592Sbmsout_inp_locked: 2518178285Srwatson INP_WUNLOCK(inp); 2519170613Sbms return (error); 2520170613Sbms} 2521170613Sbms 2522170613Sbms/* 2523170613Sbms * Set the IP multicast options in response to user setsockopt(). 2524170613Sbms * 2525170613Sbms * Many of the socket options handled in this function duplicate the 2526170613Sbms * functionality of socket options in the regular unicast API. However, 2527170613Sbms * it is not possible to merge the duplicate code, because the idempotence 2528170613Sbms * of the IPv4 multicast part of the BSD Sockets API must be preserved; 2529170613Sbms * the effects of these options must be treated as separate and distinct. 2530189592Sbms * 2531189592Sbms * SMPng: XXX: Unlocked read of inp_socket believed OK. 2532189592Sbms * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING 2533189592Sbms * is refactored to no longer use vifs. 2534170613Sbms */ 2535170613Sbmsint 2536170613Sbmsinp_setmoptions(struct inpcb *inp, struct sockopt *sopt) 2537170613Sbms{ 2538170613Sbms struct ip_moptions *imo; 2539170613Sbms int error; 2540170613Sbms 2541170613Sbms error = 0; 2542170613Sbms 2543171746Scsjp /* 2544171746Scsjp * If socket is neither of type SOCK_RAW or SOCK_DGRAM, 2545171746Scsjp * or is a divert socket, reject it. 2546171746Scsjp */ 2547171746Scsjp if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT || 2548171746Scsjp (inp->inp_socket->so_proto->pr_type != SOCK_RAW && 2549189592Sbms inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) 2550171746Scsjp return (EOPNOTSUPP); 2551171746Scsjp 2552170613Sbms switch (sopt->sopt_name) { 2553170613Sbms case IP_MULTICAST_VIF: { 2554170613Sbms int vifi; 2555170613Sbms /* 2556170613Sbms * Select a multicast VIF for transmission. 2557170613Sbms * Only useful if multicast forwarding is active. 2558170613Sbms */ 2559170613Sbms if (legal_vif_num == NULL) { 2560170613Sbms error = EOPNOTSUPP; 2561170613Sbms break; 2562170613Sbms } 2563170613Sbms error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int)); 2564170613Sbms if (error) 2565170613Sbms break; 2566170613Sbms if (!legal_vif_num(vifi) && (vifi != -1)) { 2567170613Sbms error = EINVAL; 2568170613Sbms break; 2569170613Sbms } 2570170613Sbms imo = inp_findmoptions(inp); 2571170613Sbms imo->imo_multicast_vif = vifi; 2572178285Srwatson INP_WUNLOCK(inp); 2573170613Sbms break; 2574170613Sbms } 2575170613Sbms 2576170613Sbms case IP_MULTICAST_IF: 2577170613Sbms error = inp_set_multicast_if(inp, sopt); 2578170613Sbms break; 2579170613Sbms 2580170613Sbms case IP_MULTICAST_TTL: { 2581170613Sbms u_char ttl; 2582170613Sbms 2583170613Sbms /* 2584170613Sbms * Set the IP time-to-live for outgoing multicast packets. 2585170613Sbms * The original multicast API required a char argument, 2586170613Sbms * which is inconsistent with the rest of the socket API. 2587170613Sbms * We allow either a char or an int. 2588170613Sbms */ 2589170613Sbms if (sopt->sopt_valsize == sizeof(u_char)) { 2590170613Sbms error = sooptcopyin(sopt, &ttl, sizeof(u_char), 2591170613Sbms sizeof(u_char)); 2592170613Sbms if (error) 2593170613Sbms break; 2594170613Sbms } else { 2595170613Sbms u_int ittl; 2596170613Sbms 2597170613Sbms error = sooptcopyin(sopt, &ittl, sizeof(u_int), 2598170613Sbms sizeof(u_int)); 2599170613Sbms if (error) 2600170613Sbms break; 2601170613Sbms if (ittl > 255) { 2602170613Sbms error = EINVAL; 2603170613Sbms break; 2604170613Sbms } 2605170613Sbms ttl = (u_char)ittl; 2606170613Sbms } 2607170613Sbms imo = inp_findmoptions(inp); 2608170613Sbms imo->imo_multicast_ttl = ttl; 2609178285Srwatson INP_WUNLOCK(inp); 2610170613Sbms break; 2611170613Sbms } 2612170613Sbms 2613170613Sbms case IP_MULTICAST_LOOP: { 2614170613Sbms u_char loop; 2615170613Sbms 2616170613Sbms /* 2617170613Sbms * Set the loopback flag for outgoing multicast packets. 2618170613Sbms * Must be zero or one. The original multicast API required a 2619170613Sbms * char argument, which is inconsistent with the rest 2620170613Sbms * of the socket API. We allow either a char or an int. 2621170613Sbms */ 2622170613Sbms if (sopt->sopt_valsize == sizeof(u_char)) { 2623170613Sbms error = sooptcopyin(sopt, &loop, sizeof(u_char), 2624170613Sbms sizeof(u_char)); 2625170613Sbms if (error) 2626170613Sbms break; 2627170613Sbms } else { 2628170613Sbms u_int iloop; 2629170613Sbms 2630170613Sbms error = sooptcopyin(sopt, &iloop, sizeof(u_int), 2631170613Sbms sizeof(u_int)); 2632170613Sbms if (error) 2633170613Sbms break; 2634170613Sbms loop = (u_char)iloop; 2635170613Sbms } 2636170613Sbms imo = inp_findmoptions(inp); 2637170613Sbms imo->imo_multicast_loop = !!loop; 2638178285Srwatson INP_WUNLOCK(inp); 2639170613Sbms break; 2640170613Sbms } 2641170613Sbms 2642170613Sbms case IP_ADD_MEMBERSHIP: 2643170613Sbms case IP_ADD_SOURCE_MEMBERSHIP: 2644170613Sbms case MCAST_JOIN_GROUP: 2645170613Sbms case MCAST_JOIN_SOURCE_GROUP: 2646170613Sbms error = inp_join_group(inp, sopt); 2647170613Sbms break; 2648170613Sbms 2649170613Sbms case IP_DROP_MEMBERSHIP: 2650170613Sbms case IP_DROP_SOURCE_MEMBERSHIP: 2651170613Sbms case MCAST_LEAVE_GROUP: 2652170613Sbms case MCAST_LEAVE_SOURCE_GROUP: 2653170613Sbms error = inp_leave_group(inp, sopt); 2654170613Sbms break; 2655170613Sbms 2656170613Sbms case IP_BLOCK_SOURCE: 2657170613Sbms case IP_UNBLOCK_SOURCE: 2658170613Sbms case MCAST_BLOCK_SOURCE: 2659170613Sbms case MCAST_UNBLOCK_SOURCE: 2660189592Sbms error = inp_block_unblock_source(inp, sopt); 2661170613Sbms break; 2662170613Sbms 2663170613Sbms case IP_MSFILTER: 2664170613Sbms error = inp_set_source_filters(inp, sopt); 2665170613Sbms break; 2666170613Sbms 2667170613Sbms default: 2668170613Sbms error = EOPNOTSUPP; 2669170613Sbms break; 2670170613Sbms } 2671170613Sbms 2672170613Sbms INP_UNLOCK_ASSERT(inp); 2673170613Sbms 2674170613Sbms return (error); 2675170613Sbms} 2676189592Sbms 2677189592Sbms/* 2678189592Sbms * Expose IGMP's multicast filter mode and source list(s) to userland, 2679189592Sbms * keyed by (ifindex, group). 2680189592Sbms * The filter mode is written out as a uint32_t, followed by 2681189592Sbms * 0..n of struct in_addr. 2682189592Sbms * For use by ifmcstat(8). 2683189592Sbms * SMPng: NOTE: unlocked read of ifindex space. 2684189592Sbms */ 2685189592Sbmsstatic int 2686189592Sbmssysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS) 2687189592Sbms{ 2688189592Sbms struct in_addr src, group; 2689189592Sbms struct ifnet *ifp; 2690189592Sbms struct ifmultiaddr *ifma; 2691189592Sbms struct in_multi *inm; 2692189592Sbms struct ip_msource *ims; 2693189592Sbms int *name; 2694189592Sbms int retval; 2695189592Sbms u_int namelen; 2696189592Sbms uint32_t fmode, ifindex; 2697189592Sbms 2698189592Sbms name = (int *)arg1; 2699189592Sbms namelen = arg2; 2700189592Sbms 2701189592Sbms if (req->newptr != NULL) 2702189592Sbms return (EPERM); 2703189592Sbms 2704189592Sbms if (namelen != 2) 2705189592Sbms return (EINVAL); 2706189592Sbms 2707189592Sbms ifindex = name[0]; 2708189592Sbms if (ifindex <= 0 || ifindex > V_if_index) { 2709189592Sbms CTR2(KTR_IGMPV3, "%s: ifindex %u out of range", 2710189592Sbms __func__, ifindex); 2711189592Sbms return (ENOENT); 2712189592Sbms } 2713189592Sbms 2714189592Sbms group.s_addr = name[1]; 2715189592Sbms if (!IN_MULTICAST(ntohl(group.s_addr))) { 2716189592Sbms CTR2(KTR_IGMPV3, "%s: group %s is not multicast", 2717189592Sbms __func__, inet_ntoa(group)); 2718189592Sbms return (EINVAL); 2719189592Sbms } 2720189592Sbms 2721189592Sbms ifp = ifnet_byindex(ifindex); 2722189592Sbms if (ifp == NULL) { 2723189592Sbms CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u", 2724189592Sbms __func__, ifindex); 2725189592Sbms return (ENOENT); 2726189592Sbms } 2727189592Sbms 2728189592Sbms retval = sysctl_wire_old_buffer(req, 2729189592Sbms sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr))); 2730189592Sbms if (retval) 2731189592Sbms return (retval); 2732189592Sbms 2733189592Sbms IN_MULTI_LOCK(); 2734189592Sbms 2735189592Sbms IF_ADDR_LOCK(ifp); 2736189592Sbms TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 2737189592Sbms if (ifma->ifma_addr->sa_family != AF_INET || 2738189592Sbms ifma->ifma_protospec == NULL) 2739189592Sbms continue; 2740189592Sbms inm = (struct in_multi *)ifma->ifma_protospec; 2741189592Sbms if (!in_hosteq(inm->inm_addr, group)) 2742189592Sbms continue; 2743189592Sbms fmode = inm->inm_st[1].iss_fmode; 2744189592Sbms retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t)); 2745189592Sbms if (retval != 0) 2746189592Sbms break; 2747189592Sbms RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) { 2748189592Sbms#ifdef KTR 2749189592Sbms struct in_addr ina; 2750189592Sbms ina.s_addr = htonl(ims->ims_haddr); 2751189592Sbms CTR2(KTR_IGMPV3, "%s: visit node %s", __func__, 2752189592Sbms inet_ntoa(ina)); 2753189592Sbms#endif 2754189592Sbms /* 2755189592Sbms * Only copy-out sources which are in-mode. 2756189592Sbms */ 2757189592Sbms if (fmode != ims_get_mode(inm, ims, 1)) { 2758189592Sbms CTR1(KTR_IGMPV3, "%s: skip non-in-mode", 2759189592Sbms __func__); 2760189592Sbms continue; 2761189592Sbms } 2762189592Sbms src.s_addr = htonl(ims->ims_haddr); 2763189592Sbms retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr)); 2764189592Sbms if (retval != 0) 2765189592Sbms break; 2766189592Sbms } 2767189592Sbms } 2768189592Sbms IF_ADDR_UNLOCK(ifp); 2769189592Sbms 2770189592Sbms IN_MULTI_UNLOCK(); 2771189592Sbms 2772189592Sbms return (retval); 2773189592Sbms} 2774189592Sbms 2775189592Sbms#ifdef KTR 2776189592Sbms 2777189592Sbmsstatic const char *inm_modestrs[] = { "un", "in", "ex" }; 2778189592Sbms 2779189592Sbmsstatic const char * 2780189592Sbmsinm_mode_str(const int mode) 2781189592Sbms{ 2782189592Sbms 2783189592Sbms if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE) 2784189592Sbms return (inm_modestrs[mode]); 2785189592Sbms return ("??"); 2786189592Sbms} 2787189592Sbms 2788189592Sbmsstatic const char *inm_statestrs[] = { 2789189592Sbms "not-member", 2790189592Sbms "silent", 2791189592Sbms "idle", 2792189592Sbms "lazy", 2793189592Sbms "sleeping", 2794189592Sbms "awakening", 2795189592Sbms "query-pending", 2796189592Sbms "sg-query-pending", 2797189592Sbms "leaving" 2798189592Sbms}; 2799189592Sbms 2800189592Sbmsstatic const char * 2801189592Sbmsinm_state_str(const int state) 2802189592Sbms{ 2803189592Sbms 2804189592Sbms if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER) 2805189592Sbms return (inm_statestrs[state]); 2806189592Sbms return ("??"); 2807189592Sbms} 2808189592Sbms 2809189592Sbms/* 2810189592Sbms * Dump an in_multi structure to the console. 2811189592Sbms */ 2812189592Sbmsvoid 2813189592Sbmsinm_print(const struct in_multi *inm) 2814189592Sbms{ 2815189592Sbms int t; 2816189592Sbms 2817190753Skan if ((ktr_mask & KTR_IGMPV3) == 0) 2818189635Sbms return; 2819189635Sbms 2820189592Sbms printf("%s: --- begin inm %p ---\n", __func__, inm); 2821189592Sbms printf("addr %s ifp %p(%s) ifma %p\n", 2822189592Sbms inet_ntoa(inm->inm_addr), 2823189592Sbms inm->inm_ifp, 2824189592Sbms inm->inm_ifp->if_xname, 2825189592Sbms inm->inm_ifma); 2826189592Sbms printf("timer %u state %s refcount %u scq.len %u\n", 2827189592Sbms inm->inm_timer, 2828189592Sbms inm_state_str(inm->inm_state), 2829189592Sbms inm->inm_refcount, 2830189592Sbms inm->inm_scq.ifq_len); 2831189592Sbms printf("igi %p nsrc %lu sctimer %u scrv %u\n", 2832189592Sbms inm->inm_igi, 2833189592Sbms inm->inm_nsrc, 2834189592Sbms inm->inm_sctimer, 2835189592Sbms inm->inm_scrv); 2836189592Sbms for (t = 0; t < 2; t++) { 2837189592Sbms printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t, 2838189592Sbms inm_mode_str(inm->inm_st[t].iss_fmode), 2839189592Sbms inm->inm_st[t].iss_asm, 2840189592Sbms inm->inm_st[t].iss_ex, 2841189592Sbms inm->inm_st[t].iss_in, 2842189592Sbms inm->inm_st[t].iss_rec); 2843189592Sbms } 2844189592Sbms printf("%s: --- end inm %p ---\n", __func__, inm); 2845189592Sbms} 2846189592Sbms 2847189592Sbms#else /* !KTR */ 2848189592Sbms 2849189592Sbmsvoid 2850189592Sbmsinm_print(const struct in_multi *inm) 2851189592Sbms{ 2852189592Sbms 2853189592Sbms} 2854189592Sbms 2855189592Sbms#endif /* KTR */ 2856189592Sbms 2857189592SbmsRB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp); 2858