2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2007-2009 Bruce Simpson.
5 * Copyright (c) 2005 Robert N. M. Watson.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote
17 * products derived from this software without specific prior written
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * IPv4 multicast socket, group, and socket option processing module.
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/protosw.h>
47 #include <sys/rmlock.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/protosw.h>
51 #include <sys/sysctl.h>
53 #include <sys/taskqueue.h>
54 #include <sys/gtaskqueue.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
63 #include <net/ethernet.h>
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_fib.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/igmp_var.h>
74 #define KTR_IGMPV3 KTR_INET
77 #ifndef __SOCKUNION_DECLARED
79 struct sockaddr_storage ss;
81 struct sockaddr_dl sdl;
82 struct sockaddr_in sin;
84 typedef union sockunion sockunion_t;
85 #define __SOCKUNION_DECLARED
86 #endif /* __SOCKUNION_DECLARED */
88 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
89 "IPv4 multicast PCB-layer source filter");
90 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
91 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
92 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
93 "IPv4 multicast IGMP-layer source filter");
97 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
98 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
99 * it can be taken by code in net/if.c also.
100 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
102 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
103 * any need for in_multi itself to be virtualized -- it is bound to an ifp
104 * anyway no matter what happens.
106 struct mtx in_multi_list_mtx;
107 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
109 struct mtx in_multi_free_mtx;
110 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
112 struct sx in_multi_sx;
113 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
118 * Functions with non-static linkage defined in this file should be
119 * declared in in_var.h:
124 * in_joingroup_locked()
126 * in_leavegroup_locked()
132 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
135 static void imf_commit(struct in_mfilter *);
136 static int imf_get_source(struct in_mfilter *imf,
137 const struct sockaddr_in *psin,
138 struct in_msource **);
139 static struct in_msource *
140 imf_graft(struct in_mfilter *, const uint8_t,
141 const struct sockaddr_in *);
142 static void imf_leave(struct in_mfilter *);
143 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
144 static void imf_purge(struct in_mfilter *);
145 static void imf_rollback(struct in_mfilter *);
146 static void imf_reap(struct in_mfilter *);
147 static int imo_grow(struct ip_moptions *);
148 static size_t imo_match_group(const struct ip_moptions *,
149 const struct ifnet *, const struct sockaddr *);
150 static struct in_msource *
151 imo_match_source(const struct ip_moptions *, const size_t,
152 const struct sockaddr *);
153 static void ims_merge(struct ip_msource *ims,
154 const struct in_msource *lims, const int rollback);
155 static int in_getmulti(struct ifnet *, const struct in_addr *,
157 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
158 const int noalloc, struct ip_msource **pims);
160 static int inm_is_ifp_detached(const struct in_multi *);
162 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
163 static void inm_purge(struct in_multi *);
164 static void inm_reap(struct in_multi *);
165 static void inm_release(struct in_multi *);
166 static struct ip_moptions *
167 inp_findmoptions(struct inpcb *);
168 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
169 static int inp_join_group(struct inpcb *, struct sockopt *);
170 static int inp_leave_group(struct inpcb *, struct sockopt *);
171 static struct ifnet *
172 inp_lookup_mcast_ifp(const struct inpcb *,
173 const struct sockaddr_in *, const struct in_addr);
174 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
175 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
176 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
177 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
179 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
182 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
183 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
184 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
185 "Max source filters per group");
187 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
188 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
189 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
190 "Max source filters per socket");
192 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
193 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
194 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
196 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
197 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
198 "Per-interface stack-wide source filters");
202 * Inline function which wraps assertions for a valid ifp.
203 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
207 inm_is_ifp_detached(const struct in_multi *inm)
211 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
212 ifp = inm->inm_ifma->ifma_ifp;
215 * Sanity check that netinet's notion of ifp is the
218 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
221 return (ifp == NULL);
225 static struct grouptask free_gtask;
226 static struct in_multi_head inm_free_list;
227 static void inm_release_task(void *arg __unused);
228 static void inm_init(void)
230 SLIST_INIT(&inm_free_list);
231 taskqgroup_config_gtask_init(NULL, &free_gtask, inm_release_task, "inm release task");
234 #ifdef EARLY_AP_STARTUP
235 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
238 SYSINIT(inm_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_FIRST,
244 inm_release_list_deferred(struct in_multi_head *inmh)
247 if (SLIST_EMPTY(inmh))
249 mtx_lock(&in_multi_free_mtx);
250 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
251 mtx_unlock(&in_multi_free_mtx);
252 GROUPTASK_ENQUEUE(&free_gtask);
256 inm_disconnect(struct in_multi *inm)
259 struct ifmultiaddr *ifma, *ll_ifma;
262 IF_ADDR_WLOCK_ASSERT(ifp);
263 ifma = inm->inm_ifma;
266 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
267 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
268 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
270 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
271 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
272 MPASS(ifma != ll_ifma);
273 ifma->ifma_llifma = NULL;
274 MPASS(ll_ifma->ifma_llifma == NULL);
275 MPASS(ll_ifma->ifma_ifp == ifp);
276 if (--ll_ifma->ifma_refcount == 0) {
277 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
278 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
279 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
281 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
282 if_freemulti(ll_ifma);
289 inm_release_deferred(struct in_multi *inm)
291 struct in_multi_head tmp;
293 IN_MULTI_LIST_LOCK_ASSERT();
294 MPASS(inm->inm_refcount > 0);
295 if (--inm->inm_refcount == 0) {
298 inm->inm_ifma->ifma_protospec = NULL;
299 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
300 inm_release_list_deferred(&tmp);
305 inm_release_task(void *arg __unused)
307 struct in_multi_head inm_free_tmp;
308 struct in_multi *inm, *tinm;
310 SLIST_INIT(&inm_free_tmp);
311 mtx_lock(&in_multi_free_mtx);
312 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
313 mtx_unlock(&in_multi_free_mtx);
315 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
316 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
324 * Initialize an in_mfilter structure to a known state at t0, t1
325 * with an empty source filter list.
328 imf_init(struct in_mfilter *imf, const int st0, const int st1)
330 memset(imf, 0, sizeof(struct in_mfilter));
331 RB_INIT(&imf->imf_sources);
332 imf->imf_st[0] = st0;
333 imf->imf_st[1] = st1;
337 * Function for looking up an in_multi record for an IPv4 multicast address
338 * on a given interface. ifp must be valid. If no record found, return NULL.
339 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
342 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
344 struct ifmultiaddr *ifma;
345 struct in_multi *inm;
347 IN_MULTI_LIST_LOCK_ASSERT();
348 IF_ADDR_LOCK_ASSERT(ifp);
351 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
352 if (ifma->ifma_addr->sa_family != AF_INET ||
353 ifma->ifma_protospec == NULL)
355 inm = (struct in_multi *)ifma->ifma_protospec;
356 if (inm->inm_addr.s_addr == ina.s_addr)
364 * Wrapper for inm_lookup_locked().
365 * The IF_ADDR_LOCK will be taken on ifp and released on return.
368 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
370 struct in_multi *inm;
372 IN_MULTI_LIST_LOCK_ASSERT();
374 inm = inm_lookup_locked(ifp, ina);
375 IF_ADDR_RUNLOCK(ifp);
381 * Resize the ip_moptions vector to the next power-of-two minus 1.
382 * May be called with locks held; do not sleep.
385 imo_grow(struct ip_moptions *imo)
387 struct in_multi **nmships;
388 struct in_multi **omships;
389 struct in_mfilter *nmfilters;
390 struct in_mfilter *omfilters;
397 omships = imo->imo_membership;
398 omfilters = imo->imo_mfilters;
399 oldmax = imo->imo_max_memberships;
400 newmax = ((oldmax + 1) * 2) - 1;
402 if (newmax <= IP_MAX_MEMBERSHIPS) {
403 nmships = (struct in_multi **)realloc(omships,
404 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
405 nmfilters = (struct in_mfilter *)realloc(omfilters,
406 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
407 if (nmships != NULL && nmfilters != NULL) {
408 /* Initialize newly allocated source filter heads. */
409 for (idx = oldmax; idx < newmax; idx++) {
410 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
413 imo->imo_max_memberships = newmax;
414 imo->imo_membership = nmships;
415 imo->imo_mfilters = nmfilters;
419 if (nmships == NULL || nmfilters == NULL) {
421 free(nmships, M_IPMOPTS);
422 if (nmfilters != NULL)
423 free(nmfilters, M_INMFILTER);
424 return (ETOOMANYREFS);
431 * Find an IPv4 multicast group entry for this ip_moptions instance
432 * which matches the specified group, and optionally an interface.
433 * Return its index into the array, or -1 if not found.
436 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
437 const struct sockaddr *group)
439 const struct sockaddr_in *gsin;
440 struct in_multi **pinm;
444 gsin = (const struct sockaddr_in *)group;
446 /* The imo_membership array may be lazy allocated. */
447 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
450 nmships = imo->imo_num_memberships;
451 pinm = &imo->imo_membership[0];
452 for (idx = 0; idx < nmships; idx++, pinm++) {
455 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
456 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
467 * Find an IPv4 multicast source entry for this imo which matches
468 * the given group index for this socket, and source address.
470 * NOTE: This does not check if the entry is in-mode, merely if
471 * it exists, which may not be the desired behaviour.
473 static struct in_msource *
474 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
475 const struct sockaddr *src)
477 struct ip_msource find;
478 struct in_mfilter *imf;
479 struct ip_msource *ims;
480 const sockunion_t *psa;
482 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
483 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
484 ("%s: invalid index %d\n", __func__, (int)gidx));
486 /* The imo_mfilters array may be lazy allocated. */
487 if (imo->imo_mfilters == NULL)
489 imf = &imo->imo_mfilters[gidx];
491 /* Source trees are keyed in host byte order. */
492 psa = (const sockunion_t *)src;
493 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
494 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
496 return ((struct in_msource *)ims);
500 * Perform filtering for multicast datagrams on a socket by group and source.
502 * Returns 0 if a datagram should be allowed through, or various error codes
503 * if the socket was not a member of the group, or the source was muted, etc.
506 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
507 const struct sockaddr *group, const struct sockaddr *src)
510 struct in_msource *ims;
513 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
515 gidx = imo_match_group(imo, ifp, group);
517 return (MCAST_NOTGMEMBER);
520 * Check if the source was included in an (S,G) join.
521 * Allow reception on exclusive memberships by default,
522 * reject reception on inclusive memberships by default.
523 * Exclude source only if an in-mode exclude filter exists.
524 * Include source only if an in-mode include filter exists.
525 * NOTE: We are comparing group state here at IGMP t1 (now)
526 * with socket-layer t0 (since last downcall).
528 mode = imo->imo_mfilters[gidx].imf_st[1];
529 ims = imo_match_source(imo, gidx, src);
531 if ((ims == NULL && mode == MCAST_INCLUDE) ||
532 (ims != NULL && ims->imsl_st[0] != mode))
533 return (MCAST_NOTSMEMBER);
539 * Find and return a reference to an in_multi record for (ifp, group),
540 * and bump its reference count.
541 * If one does not exist, try to allocate it, and update link-layer multicast
542 * filters on ifp to listen for group.
543 * Assumes the IN_MULTI lock is held across the call.
544 * Return 0 if successful, otherwise return an appropriate error code.
547 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
548 struct in_multi **pinm)
550 struct sockaddr_in gsin;
551 struct ifmultiaddr *ifma;
552 struct in_ifinfo *ii;
553 struct in_multi *inm;
556 IN_MULTI_LOCK_ASSERT();
558 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
559 IN_MULTI_LIST_LOCK();
560 inm = inm_lookup(ifp, *group);
563 * If we already joined this group, just bump the
564 * refcount and return it.
566 KASSERT(inm->inm_refcount >= 1,
567 ("%s: bad refcount %d", __func__, inm->inm_refcount));
568 inm_acquire_locked(inm);
571 IN_MULTI_LIST_UNLOCK();
575 memset(&gsin, 0, sizeof(gsin));
576 gsin.sin_family = AF_INET;
577 gsin.sin_len = sizeof(struct sockaddr_in);
578 gsin.sin_addr = *group;
581 * Check if a link-layer group is already associated
582 * with this network-layer group on the given ifnet.
584 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
588 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
589 IN_MULTI_LIST_LOCK();
593 * If something other than netinet is occupying the link-layer
594 * group, print a meaningful error message and back out of
596 * Otherwise, bump the refcount on the existing network-layer
597 * group association and return it.
599 if (ifma->ifma_protospec != NULL) {
600 inm = (struct in_multi *)ifma->ifma_protospec;
602 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
604 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
605 ("%s: ifma not AF_INET", __func__));
606 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
607 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
608 !in_hosteq(inm->inm_addr, *group)) {
609 char addrbuf[INET_ADDRSTRLEN];
611 panic("%s: ifma %p is inconsistent with %p (%s)",
612 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
615 inm_acquire_locked(inm);
620 IF_ADDR_WLOCK_ASSERT(ifp);
623 * A new in_multi record is needed; allocate and initialize it.
624 * We DO NOT perform an IGMP join as the in_ layer may need to
625 * push an initial source list down to IGMP to support SSM.
627 * The initial source filter state is INCLUDE, {} as per the RFC.
629 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
631 IF_ADDR_WUNLOCK(ifp);
632 IN_MULTI_LIST_UNLOCK();
633 if_delmulti_ifma(ifma);
636 inm->inm_addr = *group;
638 inm->inm_igi = ii->ii_igmp;
639 inm->inm_ifma = ifma;
640 inm->inm_refcount = 1;
641 inm->inm_state = IGMP_NOT_MEMBER;
642 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
643 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
644 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
645 RB_INIT(&inm->inm_srcs);
647 ifma->ifma_protospec = inm;
651 IF_ADDR_WUNLOCK(ifp);
652 IN_MULTI_LIST_UNLOCK();
657 * Drop a reference to an in_multi record.
659 * If the refcount drops to 0, free the in_multi record and
660 * delete the underlying link-layer membership.
663 inm_release(struct in_multi *inm)
665 struct ifmultiaddr *ifma;
668 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
669 MPASS(inm->inm_refcount == 0);
670 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
672 ifma = inm->inm_ifma;
675 /* XXX this access is not covered by IF_ADDR_LOCK */
676 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
678 CURVNET_SET(ifp->if_vnet);
680 free(inm, M_IPMADDR);
681 if_delmulti_ifma_flags(ifma, 1);
686 free(inm, M_IPMADDR);
687 if_delmulti_ifma_flags(ifma, 1);
692 * Clear recorded source entries for a group.
693 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
694 * FIXME: Should reap.
697 inm_clear_recorded(struct in_multi *inm)
699 struct ip_msource *ims;
701 IN_MULTI_LIST_LOCK_ASSERT();
703 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
706 --inm->inm_st[1].iss_rec;
709 KASSERT(inm->inm_st[1].iss_rec == 0,
710 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
714 * Record a source as pending for a Source-Group IGMPv3 query.
715 * This lives here as it modifies the shared tree.
717 * inm is the group descriptor.
718 * naddr is the address of the source to record in network-byte order.
720 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
721 * lazy-allocate a source node in response to an SG query.
722 * Otherwise, no allocation is performed. This saves some memory
723 * with the trade-off that the source will not be reported to the
724 * router if joined in the window between the query response and
725 * the group actually being joined on the local host.
727 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
728 * This turns off the allocation of a recorded source entry if
729 * the group has not been joined.
731 * Return 0 if the source didn't exist or was already marked as recorded.
732 * Return 1 if the source was marked as recorded by this function.
733 * Return <0 if any error occurred (negated errno code).
736 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
738 struct ip_msource find;
739 struct ip_msource *ims, *nims;
741 IN_MULTI_LIST_LOCK_ASSERT();
743 find.ims_haddr = ntohl(naddr);
744 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
745 if (ims && ims->ims_stp)
748 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
750 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
754 nims->ims_haddr = find.ims_haddr;
755 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
761 * Mark the source as recorded and update the recorded
765 ++inm->inm_st[1].iss_rec;
771 * Return a pointer to an in_msource owned by an in_mfilter,
772 * given its source address.
773 * Lazy-allocate if needed. If this is a new entry its filter state is
776 * imf is the filter set being modified.
777 * haddr is the source address in *host* byte-order.
779 * SMPng: May be called with locks held; malloc must not block.
782 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
783 struct in_msource **plims)
785 struct ip_msource find;
786 struct ip_msource *ims, *nims;
787 struct in_msource *lims;
794 /* key is host byte order */
795 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
796 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
797 lims = (struct in_msource *)ims;
799 if (imf->imf_nsrc == in_mcast_maxsocksrc)
801 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
805 lims = (struct in_msource *)nims;
806 lims->ims_haddr = find.ims_haddr;
807 lims->imsl_st[0] = MCAST_UNDEFINED;
808 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
818 * Graft a source entry into an existing socket-layer filter set,
819 * maintaining any required invariants and checking allocations.
821 * The source is marked as being in the new filter mode at t1.
823 * Return the pointer to the new node, otherwise return NULL.
825 static struct in_msource *
826 imf_graft(struct in_mfilter *imf, const uint8_t st1,
827 const struct sockaddr_in *psin)
829 struct ip_msource *nims;
830 struct in_msource *lims;
832 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
836 lims = (struct in_msource *)nims;
837 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
838 lims->imsl_st[0] = MCAST_UNDEFINED;
839 lims->imsl_st[1] = st1;
840 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
847 * Prune a source entry from an existing socket-layer filter set,
848 * maintaining any required invariants and checking allocations.
850 * The source is marked as being left at t1, it is not freed.
852 * Return 0 if no error occurred, otherwise return an errno value.
855 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
857 struct ip_msource find;
858 struct ip_msource *ims;
859 struct in_msource *lims;
861 /* key is host byte order */
862 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
863 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
866 lims = (struct in_msource *)ims;
867 lims->imsl_st[1] = MCAST_UNDEFINED;
872 * Revert socket-layer filter set deltas at t1 to t0 state.
875 imf_rollback(struct in_mfilter *imf)
877 struct ip_msource *ims, *tims;
878 struct in_msource *lims;
880 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
881 lims = (struct in_msource *)ims;
882 if (lims->imsl_st[0] == lims->imsl_st[1]) {
883 /* no change at t1 */
885 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
886 /* revert change to existing source at t1 */
887 lims->imsl_st[1] = lims->imsl_st[0];
889 /* revert source added t1 */
890 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
891 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
892 free(ims, M_INMFILTER);
896 imf->imf_st[1] = imf->imf_st[0];
900 * Mark socket-layer filter set as INCLUDE {} at t1.
903 imf_leave(struct in_mfilter *imf)
905 struct ip_msource *ims;
906 struct in_msource *lims;
908 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
909 lims = (struct in_msource *)ims;
910 lims->imsl_st[1] = MCAST_UNDEFINED;
912 imf->imf_st[1] = MCAST_INCLUDE;
916 * Mark socket-layer filter set deltas as committed.
919 imf_commit(struct in_mfilter *imf)
921 struct ip_msource *ims;
922 struct in_msource *lims;
924 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
925 lims = (struct in_msource *)ims;
926 lims->imsl_st[0] = lims->imsl_st[1];
928 imf->imf_st[0] = imf->imf_st[1];
932 * Reap unreferenced sources from socket-layer filter set.
935 imf_reap(struct in_mfilter *imf)
937 struct ip_msource *ims, *tims;
938 struct in_msource *lims;
940 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
941 lims = (struct in_msource *)ims;
942 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
943 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
944 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
945 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
946 free(ims, M_INMFILTER);
953 * Purge socket-layer filter set.
956 imf_purge(struct in_mfilter *imf)
958 struct ip_msource *ims, *tims;
960 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
961 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
962 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
963 free(ims, M_INMFILTER);
966 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
967 KASSERT(RB_EMPTY(&imf->imf_sources),
968 ("%s: imf_sources not empty", __func__));
972 * Look up a source filter entry for a multicast group.
974 * inm is the group descriptor to work with.
975 * haddr is the host-byte-order IPv4 address to look up.
976 * noalloc may be non-zero to suppress allocation of sources.
977 * *pims will be set to the address of the retrieved or allocated source.
979 * SMPng: NOTE: may be called with locks held.
980 * Return 0 if successful, otherwise return a non-zero error code.
983 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
984 const int noalloc, struct ip_msource **pims)
986 struct ip_msource find;
987 struct ip_msource *ims, *nims;
989 find.ims_haddr = haddr;
990 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
991 if (ims == NULL && !noalloc) {
992 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
994 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
998 nims->ims_haddr = haddr;
999 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
1003 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
1013 * Merge socket-layer source into IGMP-layer source.
1014 * If rollback is non-zero, perform the inverse of the merge.
1017 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
1020 int n = rollback ? -1 : 1;
1022 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
1023 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
1024 __func__, n, ims->ims_haddr);
1025 ims->ims_st[1].ex -= n;
1026 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
1027 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
1028 __func__, n, ims->ims_haddr);
1029 ims->ims_st[1].in -= n;
1032 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
1033 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
1034 __func__, n, ims->ims_haddr);
1035 ims->ims_st[1].ex += n;
1036 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
1037 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
1038 __func__, n, ims->ims_haddr);
1039 ims->ims_st[1].in += n;
1044 * Atomically update the global in_multi state, when a membership's
1045 * filter list is being updated in any way.
1047 * imf is the per-inpcb-membership group filter pointer.
1048 * A fake imf may be passed for in-kernel consumers.
1050 * XXX This is a candidate for a set-symmetric-difference style loop
1051 * which would eliminate the repeated lookup from root of ims nodes,
1052 * as they share the same key space.
1054 * If any error occurred this function will back out of refcounts
1055 * and return a non-zero value.
1058 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1060 struct ip_msource *ims, *nims;
1061 struct in_msource *lims;
1062 int schanged, error;
1068 IN_MULTI_LIST_LOCK_ASSERT();
1071 * Update the source filters first, as this may fail.
1072 * Maintain count of in-mode filters at t0, t1. These are
1073 * used to work out if we transition into ASM mode or not.
1074 * Maintain a count of source filters whose state was
1075 * actually modified by this operation.
1077 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1078 lims = (struct in_msource *)ims;
1079 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1080 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1081 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1082 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1086 ims_merge(nims, lims, 0);
1089 struct ip_msource *bims;
1091 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1092 lims = (struct in_msource *)ims;
1093 if (lims->imsl_st[0] == lims->imsl_st[1])
1095 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1098 ims_merge(bims, lims, 1);
1103 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1104 __func__, nsrc0, nsrc1);
1106 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1107 if (imf->imf_st[0] == imf->imf_st[1] &&
1108 imf->imf_st[1] == MCAST_INCLUDE) {
1110 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1111 --inm->inm_st[1].iss_in;
1115 /* Handle filter mode transition on socket. */
1116 if (imf->imf_st[0] != imf->imf_st[1]) {
1117 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1118 __func__, imf->imf_st[0], imf->imf_st[1]);
1120 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1121 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1122 --inm->inm_st[1].iss_ex;
1123 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1124 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1125 --inm->inm_st[1].iss_in;
1128 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1129 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1130 inm->inm_st[1].iss_ex++;
1131 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1132 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1133 inm->inm_st[1].iss_in++;
1138 * Track inm filter state in terms of listener counts.
1139 * If there are any exclusive listeners, stack-wide
1140 * membership is exclusive.
1141 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1142 * If no listeners remain, state is undefined at t1,
1143 * and the IGMP lifecycle for this group should finish.
1145 if (inm->inm_st[1].iss_ex > 0) {
1146 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1147 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1148 } else if (inm->inm_st[1].iss_in > 0) {
1149 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1150 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1152 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1153 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1156 /* Decrement ASM listener count on transition out of ASM mode. */
1157 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1158 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1159 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1160 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1161 --inm->inm_st[1].iss_asm;
1165 /* Increment ASM listener count on transition to ASM mode. */
1166 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1167 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1168 inm->inm_st[1].iss_asm++;
1171 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1176 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1183 * Mark an in_multi's filter set deltas as committed.
1184 * Called by IGMP after a state change has been enqueued.
1187 inm_commit(struct in_multi *inm)
1189 struct ip_msource *ims;
1191 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1192 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1195 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1196 ims->ims_st[0] = ims->ims_st[1];
1198 inm->inm_st[0] = inm->inm_st[1];
1202 * Reap unreferenced nodes from an in_multi's filter set.
1205 inm_reap(struct in_multi *inm)
1207 struct ip_msource *ims, *tims;
1209 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1210 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1211 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1214 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1215 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1216 free(ims, M_IPMSOURCE);
1222 * Purge all source nodes from an in_multi's filter set.
1225 inm_purge(struct in_multi *inm)
1227 struct ip_msource *ims, *tims;
1229 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1230 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1231 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1232 free(ims, M_IPMSOURCE);
1238 * Join a multicast group; unlocked entry point.
1240 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1241 * is not held. Fortunately, ifp is unlikely to have been detached
1242 * at this point, so we assume it's OK to recurse.
1245 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1246 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1251 error = in_joingroup_locked(ifp, gina, imf, pinm);
1258 * Join a multicast group; real entry point.
1260 * Only preserves atomicity at inm level.
1261 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1263 * If the IGMP downcall fails, the group is not joined, and an error
1267 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1268 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1270 struct in_mfilter timf;
1271 struct in_multi *inm;
1274 IN_MULTI_LOCK_ASSERT();
1275 IN_MULTI_LIST_UNLOCK_ASSERT();
1277 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1278 ntohl(gina->s_addr), ifp, ifp->if_xname);
1284 * If no imf was specified (i.e. kernel consumer),
1285 * fake one up and assume it is an ASM join.
1288 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1292 error = in_getmulti(ifp, gina, &inm);
1294 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1297 IN_MULTI_LIST_LOCK();
1298 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1299 error = inm_merge(inm, imf);
1301 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1302 goto out_inm_release;
1305 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1306 error = igmp_change_state(inm);
1308 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1309 goto out_inm_release;
1315 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1316 inm_release_deferred(inm);
1320 IN_MULTI_LIST_UNLOCK();
1326 * Leave a multicast group; unlocked entry point.
1329 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1334 error = in_leavegroup_locked(inm, imf);
1341 * Leave a multicast group; real entry point.
1342 * All source filters will be expunged.
1344 * Only preserves atomicity at inm level.
1346 * Holding the write lock for the INP which contains imf
1347 * is highly advisable. We can't assert for it as imf does not
1348 * contain a back-pointer to the owning inp.
1350 * Note: This is not the same as inm_release(*) as this function also
1351 * makes a state change downcall into IGMP.
1354 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1356 struct in_mfilter timf;
1361 IN_MULTI_LOCK_ASSERT();
1362 IN_MULTI_LIST_UNLOCK_ASSERT();
1364 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1365 inm, ntohl(inm->inm_addr.s_addr),
1366 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1370 * If no imf was specified (i.e. kernel consumer),
1371 * fake one up and assume it is an ASM join.
1374 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1379 * Begin state merge transaction at IGMP layer.
1381 * As this particular invocation should not cause any memory
1382 * to be allocated, and there is no opportunity to roll back
1383 * the transaction, it MUST NOT fail.
1385 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1386 IN_MULTI_LIST_LOCK();
1387 error = inm_merge(inm, imf);
1388 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1390 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1391 CURVNET_SET(inm->inm_ifp->if_vnet);
1392 error = igmp_change_state(inm);
1393 IF_ADDR_WLOCK(inm->inm_ifp);
1394 inm_release_deferred(inm);
1395 IF_ADDR_WUNLOCK(inm->inm_ifp);
1396 IN_MULTI_LIST_UNLOCK();
1399 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1401 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1406 /*#ifndef BURN_BRIDGES*/
1408 * Join an IPv4 multicast group in (*,G) exclusive mode.
1409 * The group must be a 224.0.0.0/24 link-scope group.
1410 * This KPI is for legacy kernel consumers only.
1413 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1415 struct in_multi *pinm;
1418 char addrbuf[INET_ADDRSTRLEN];
1421 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1422 ("%s: %s not in 224.0.0.0/24", __func__,
1423 inet_ntoa_r(*ap, addrbuf)));
1425 error = in_joingroup(ifp, ap, NULL, &pinm);
1433 * Block or unblock an ASM multicast source on an inpcb.
1434 * This implements the delta-based API described in RFC 3678.
1436 * The delta-based API applies only to exclusive-mode memberships.
1437 * An IGMP downcall will be performed.
1439 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1441 * Return 0 if successful, otherwise return an appropriate error code.
1444 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1446 struct group_source_req gsr;
1447 sockunion_t *gsa, *ssa;
1449 struct in_mfilter *imf;
1450 struct ip_moptions *imo;
1451 struct in_msource *ims;
1452 struct in_multi *inm;
1461 memset(&gsr, 0, sizeof(struct group_source_req));
1462 gsa = (sockunion_t *)&gsr.gsr_group;
1463 ssa = (sockunion_t *)&gsr.gsr_source;
1465 switch (sopt->sopt_name) {
1466 case IP_BLOCK_SOURCE:
1467 case IP_UNBLOCK_SOURCE: {
1468 struct ip_mreq_source mreqs;
1470 error = sooptcopyin(sopt, &mreqs,
1471 sizeof(struct ip_mreq_source),
1472 sizeof(struct ip_mreq_source));
1476 gsa->sin.sin_family = AF_INET;
1477 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1478 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1480 ssa->sin.sin_family = AF_INET;
1481 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1482 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1484 if (!in_nullhost(mreqs.imr_interface))
1485 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1487 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1490 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1491 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1495 case MCAST_BLOCK_SOURCE:
1496 case MCAST_UNBLOCK_SOURCE:
1497 error = sooptcopyin(sopt, &gsr,
1498 sizeof(struct group_source_req),
1499 sizeof(struct group_source_req));
1503 if (gsa->sin.sin_family != AF_INET ||
1504 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1507 if (ssa->sin.sin_family != AF_INET ||
1508 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1511 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1512 return (EADDRNOTAVAIL);
1514 ifp = ifnet_byindex(gsr.gsr_interface);
1516 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1521 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1522 __func__, sopt->sopt_name);
1523 return (EOPNOTSUPP);
1527 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1531 * Check if we are actually a member of this group.
1533 imo = inp_findmoptions(inp);
1534 idx = imo_match_group(imo, ifp, &gsa->sa);
1535 if (idx == -1 || imo->imo_mfilters == NULL) {
1536 error = EADDRNOTAVAIL;
1537 goto out_inp_locked;
1540 KASSERT(imo->imo_mfilters != NULL,
1541 ("%s: imo_mfilters not allocated", __func__));
1542 imf = &imo->imo_mfilters[idx];
1543 inm = imo->imo_membership[idx];
1546 * Attempting to use the delta-based API on an
1547 * non exclusive-mode membership is an error.
1549 fmode = imf->imf_st[0];
1550 if (fmode != MCAST_EXCLUDE) {
1552 goto out_inp_locked;
1556 * Deal with error cases up-front:
1557 * Asked to block, but already blocked; or
1558 * Asked to unblock, but nothing to unblock.
1559 * If adding a new block entry, allocate it.
1561 ims = imo_match_source(imo, idx, &ssa->sa);
1562 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1563 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1564 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1565 error = EADDRNOTAVAIL;
1566 goto out_inp_locked;
1569 INP_WLOCK_ASSERT(inp);
1572 * Begin state merge transaction at socket layer.
1575 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1576 ims = imf_graft(imf, fmode, &ssa->sin);
1580 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1581 error = imf_prune(imf, &ssa->sin);
1585 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1586 goto out_imf_rollback;
1590 * Begin state merge transaction at IGMP layer.
1593 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1594 IN_MULTI_LIST_LOCK();
1595 error = inm_merge(inm, imf);
1597 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1598 IN_MULTI_LIST_UNLOCK();
1599 goto out_in_multi_locked;
1602 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1603 error = igmp_change_state(inm);
1604 IN_MULTI_LIST_UNLOCK();
1606 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1608 out_in_multi_locked:
1625 * Given an inpcb, return its multicast options structure pointer. Accepts
1626 * an unlocked inpcb pointer, but will return it locked. May sleep.
1628 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1629 * SMPng: NOTE: Returns with the INP write lock held.
1631 static struct ip_moptions *
1632 inp_findmoptions(struct inpcb *inp)
1634 struct ip_moptions *imo;
1635 struct in_multi **immp;
1636 struct in_mfilter *imfp;
1640 if (inp->inp_moptions != NULL)
1641 return (inp->inp_moptions);
1645 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1646 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1648 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1649 M_INMFILTER, M_WAITOK);
1651 imo->imo_multicast_ifp = NULL;
1652 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1653 imo->imo_multicast_vif = -1;
1654 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1655 imo->imo_multicast_loop = in_mcast_loop;
1656 imo->imo_num_memberships = 0;
1657 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1658 imo->imo_membership = immp;
1660 /* Initialize per-group source filters. */
1661 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1662 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1663 imo->imo_mfilters = imfp;
1666 if (inp->inp_moptions != NULL) {
1667 free(imfp, M_INMFILTER);
1668 free(immp, M_IPMOPTS);
1669 free(imo, M_IPMOPTS);
1670 return (inp->inp_moptions);
1672 inp->inp_moptions = imo;
1677 inp_gcmoptions(struct ip_moptions *imo)
1679 struct in_mfilter *imf;
1680 struct in_multi *inm;
1682 size_t idx, nmships;
1684 nmships = imo->imo_num_memberships;
1685 for (idx = 0; idx < nmships; ++idx) {
1686 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1689 inm = imo->imo_membership[idx];
1692 CURVNET_SET(ifp->if_vnet);
1693 (void)in_leavegroup(inm, imf);
1696 (void)in_leavegroup(inm, imf);
1702 if (imo->imo_mfilters)
1703 free(imo->imo_mfilters, M_INMFILTER);
1704 free(imo->imo_membership, M_IPMOPTS);
1705 free(imo, M_IPMOPTS);
1709 * Discard the IP multicast options (and source filters). To minimize
1710 * the amount of work done while holding locks such as the INP's
1711 * pcbinfo lock (which is used in the receive path), the free
1712 * operation is deferred to the epoch callback task.
1715 inp_freemoptions(struct ip_moptions *imo)
1719 inp_gcmoptions(imo);
1723 * Atomically get source filters on a socket for an IPv4 multicast group.
1724 * Called with INP lock held; returns with lock released.
1727 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1729 struct __msfilterreq msfr;
1732 struct ip_moptions *imo;
1733 struct in_mfilter *imf;
1734 struct ip_msource *ims;
1735 struct in_msource *lims;
1736 struct sockaddr_in *psin;
1737 struct sockaddr_storage *ptss;
1738 struct sockaddr_storage *tss;
1740 size_t idx, nsrcs, ncsrcs;
1742 INP_WLOCK_ASSERT(inp);
1744 imo = inp->inp_moptions;
1745 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1749 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1750 sizeof(struct __msfilterreq));
1754 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1757 ifp = ifnet_byindex(msfr.msfr_ifindex);
1764 * Lookup group on the socket.
1766 gsa = (sockunion_t *)&msfr.msfr_group;
1767 idx = imo_match_group(imo, ifp, &gsa->sa);
1768 if (idx == -1 || imo->imo_mfilters == NULL) {
1770 return (EADDRNOTAVAIL);
1772 imf = &imo->imo_mfilters[idx];
1775 * Ignore memberships which are in limbo.
1777 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1781 msfr.msfr_fmode = imf->imf_st[1];
1784 * If the user specified a buffer, copy out the source filter
1785 * entries to userland gracefully.
1786 * We only copy out the number of entries which userland
1787 * has asked for, but we always tell userland how big the
1788 * buffer really needs to be.
1790 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1791 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1793 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1794 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1795 M_TEMP, M_NOWAIT | M_ZERO);
1803 * Count number of sources in-mode at t0.
1804 * If buffer space exists and remains, copy out source entries.
1806 nsrcs = msfr.msfr_nsrcs;
1809 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1810 lims = (struct in_msource *)ims;
1811 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1812 lims->imsl_st[0] != imf->imf_st[0])
1815 if (tss != NULL && nsrcs > 0) {
1816 psin = (struct sockaddr_in *)ptss;
1817 psin->sin_family = AF_INET;
1818 psin->sin_len = sizeof(struct sockaddr_in);
1819 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1829 error = copyout(tss, msfr.msfr_srcs,
1830 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1836 msfr.msfr_nsrcs = ncsrcs;
1837 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1843 * Return the IP multicast options in response to user getsockopt().
1846 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1848 struct rm_priotracker in_ifa_tracker;
1849 struct ip_mreqn mreqn;
1850 struct ip_moptions *imo;
1852 struct in_ifaddr *ia;
1857 imo = inp->inp_moptions;
1859 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1860 * or is a divert socket, reject it.
1862 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1863 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1864 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1866 return (EOPNOTSUPP);
1870 switch (sopt->sopt_name) {
1871 case IP_MULTICAST_VIF:
1873 optval = imo->imo_multicast_vif;
1877 error = sooptcopyout(sopt, &optval, sizeof(int));
1880 case IP_MULTICAST_IF:
1881 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1883 ifp = imo->imo_multicast_ifp;
1884 if (!in_nullhost(imo->imo_multicast_addr)) {
1885 mreqn.imr_address = imo->imo_multicast_addr;
1886 } else if (ifp != NULL) {
1887 mreqn.imr_ifindex = ifp->if_index;
1889 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1892 IA_SIN(ia)->sin_addr;
1897 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1898 error = sooptcopyout(sopt, &mreqn,
1899 sizeof(struct ip_mreqn));
1901 error = sooptcopyout(sopt, &mreqn.imr_address,
1902 sizeof(struct in_addr));
1906 case IP_MULTICAST_TTL:
1908 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1910 optval = coptval = imo->imo_multicast_ttl;
1912 if (sopt->sopt_valsize == sizeof(u_char))
1913 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1915 error = sooptcopyout(sopt, &optval, sizeof(int));
1918 case IP_MULTICAST_LOOP:
1920 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1922 optval = coptval = imo->imo_multicast_loop;
1924 if (sopt->sopt_valsize == sizeof(u_char))
1925 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1927 error = sooptcopyout(sopt, &optval, sizeof(int));
1932 error = EADDRNOTAVAIL;
1935 error = inp_get_source_filters(inp, sopt);
1941 error = ENOPROTOOPT;
1945 INP_UNLOCK_ASSERT(inp);
1951 * Look up the ifnet to use for a multicast group membership,
1952 * given the IPv4 address of an interface, and the IPv4 group address.
1954 * This routine exists to support legacy multicast applications
1955 * which do not understand that multicast memberships are scoped to
1956 * specific physical links in the networking stack, or which need
1957 * to join link-scope groups before IPv4 addresses are configured.
1959 * If inp is non-NULL, use this socket's current FIB number for any
1960 * required FIB lookup.
1961 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1962 * and use its ifp; usually, this points to the default next-hop.
1964 * If the FIB lookup fails, attempt to use the first non-loopback
1965 * interface with multicast capability in the system as a
1966 * last resort. The legacy IPv4 ASM API requires that we do
1967 * this in order to allow groups to be joined when the routing
1968 * table has not yet been populated during boot.
1970 * Returns NULL if no ifp could be found.
1972 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1973 * FUTURE: Implement IPv4 source-address selection.
1975 static struct ifnet *
1976 inp_lookup_mcast_ifp(const struct inpcb *inp,
1977 const struct sockaddr_in *gsin, const struct in_addr ina)
1979 struct rm_priotracker in_ifa_tracker;
1981 struct nhop4_basic nh4;
1984 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1985 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1986 ("%s: not multicast", __func__));
1989 if (!in_nullhost(ina)) {
1990 INADDR_TO_IFP(ina, ifp);
1992 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1993 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1996 struct in_ifaddr *ia;
2000 IN_IFADDR_RLOCK(&in_ifa_tracker);
2001 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
2003 if (!(mifp->if_flags & IFF_LOOPBACK) &&
2004 (mifp->if_flags & IFF_MULTICAST)) {
2009 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2017 * Join an IPv4 multicast group, possibly with a source.
2020 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
2022 struct group_source_req gsr;
2023 sockunion_t *gsa, *ssa;
2025 struct in_mfilter *imf;
2026 struct ip_moptions *imo;
2027 struct in_multi *inm;
2028 struct in_msource *lims;
2038 memset(&gsr, 0, sizeof(struct group_source_req));
2039 gsa = (sockunion_t *)&gsr.gsr_group;
2040 gsa->ss.ss_family = AF_UNSPEC;
2041 ssa = (sockunion_t *)&gsr.gsr_source;
2042 ssa->ss.ss_family = AF_UNSPEC;
2044 switch (sopt->sopt_name) {
2045 case IP_ADD_MEMBERSHIP:
2046 case IP_ADD_SOURCE_MEMBERSHIP: {
2047 struct ip_mreq_source mreqs;
2049 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
2050 error = sooptcopyin(sopt, &mreqs,
2051 sizeof(struct ip_mreq),
2052 sizeof(struct ip_mreq));
2054 * Do argument switcharoo from ip_mreq into
2055 * ip_mreq_source to avoid using two instances.
2057 mreqs.imr_interface = mreqs.imr_sourceaddr;
2058 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2059 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2060 error = sooptcopyin(sopt, &mreqs,
2061 sizeof(struct ip_mreq_source),
2062 sizeof(struct ip_mreq_source));
2067 gsa->sin.sin_family = AF_INET;
2068 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2069 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2071 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2072 ssa->sin.sin_family = AF_INET;
2073 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2074 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2077 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2080 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2081 mreqs.imr_interface);
2082 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2083 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2087 case MCAST_JOIN_GROUP:
2088 case MCAST_JOIN_SOURCE_GROUP:
2089 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2090 error = sooptcopyin(sopt, &gsr,
2091 sizeof(struct group_req),
2092 sizeof(struct group_req));
2093 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2094 error = sooptcopyin(sopt, &gsr,
2095 sizeof(struct group_source_req),
2096 sizeof(struct group_source_req));
2101 if (gsa->sin.sin_family != AF_INET ||
2102 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2106 * Overwrite the port field if present, as the sockaddr
2107 * being copied in may be matched with a binary comparison.
2109 gsa->sin.sin_port = 0;
2110 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2111 if (ssa->sin.sin_family != AF_INET ||
2112 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2114 ssa->sin.sin_port = 0;
2117 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2120 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2121 return (EADDRNOTAVAIL);
2122 ifp = ifnet_byindex(gsr.gsr_interface);
2126 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2127 __func__, sopt->sopt_name);
2128 return (EOPNOTSUPP);
2132 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2133 return (EADDRNOTAVAIL);
2135 imo = inp_findmoptions(inp);
2136 idx = imo_match_group(imo, ifp, &gsa->sa);
2140 inm = imo->imo_membership[idx];
2141 imf = &imo->imo_mfilters[idx];
2142 if (ssa->ss.ss_family != AF_UNSPEC) {
2144 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2145 * is an error. On an existing inclusive membership,
2146 * it just adds the source to the filter list.
2148 if (imf->imf_st[1] != MCAST_INCLUDE) {
2150 goto out_inp_locked;
2153 * Throw out duplicates.
2155 * XXX FIXME: This makes a naive assumption that
2156 * even if entries exist for *ssa in this imf,
2157 * they will be rejected as dupes, even if they
2158 * are not valid in the current mode (in-mode).
2160 * in_msource is transactioned just as for anything
2161 * else in SSM -- but note naive use of inm_graft()
2162 * below for allocating new filter entries.
2164 * This is only an issue if someone mixes the
2165 * full-state SSM API with the delta-based API,
2166 * which is discouraged in the relevant RFCs.
2168 lims = imo_match_source(imo, idx, &ssa->sa);
2169 if (lims != NULL /*&&
2170 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2171 error = EADDRNOTAVAIL;
2172 goto out_inp_locked;
2176 * MCAST_JOIN_GROUP on an existing exclusive
2177 * membership is an error; return EADDRINUSE
2178 * to preserve 4.4BSD API idempotence, and
2179 * avoid tedious detour to code below.
2180 * NOTE: This is bending RFC 3678 a bit.
2182 * On an existing inclusive membership, this is also
2183 * an error; if you want to change filter mode,
2184 * you must use the userland API setsourcefilter().
2185 * XXX We don't reject this for imf in UNDEFINED
2186 * state at t1, because allocation of a filter
2187 * is atomic with allocation of a membership.
2190 if (imf->imf_st[1] == MCAST_EXCLUDE)
2192 goto out_inp_locked;
2197 * Begin state merge transaction at socket layer.
2199 INP_WLOCK_ASSERT(inp);
2202 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2203 error = imo_grow(imo);
2205 goto out_inp_locked;
2208 * Allocate the new slot upfront so we can deal with
2209 * grafting the new source filter in same code path
2210 * as for join-source on existing membership.
2212 idx = imo->imo_num_memberships;
2213 imo->imo_membership[idx] = NULL;
2214 imo->imo_num_memberships++;
2215 KASSERT(imo->imo_mfilters != NULL,
2216 ("%s: imf_mfilters vector was not allocated", __func__));
2217 imf = &imo->imo_mfilters[idx];
2218 KASSERT(RB_EMPTY(&imf->imf_sources),
2219 ("%s: imf_sources not empty", __func__));
2223 * Graft new source into filter list for this inpcb's
2224 * membership of the group. The in_multi may not have
2225 * been allocated yet if this is a new membership, however,
2226 * the in_mfilter slot will be allocated and must be initialized.
2228 * Note: Grafting of exclusive mode filters doesn't happen
2230 * XXX: Should check for non-NULL lims (node exists but may
2231 * not be in-mode) for interop with full-state API.
2233 if (ssa->ss.ss_family != AF_UNSPEC) {
2234 /* Membership starts in IN mode */
2236 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2237 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2239 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2241 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2243 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2249 /* No address specified; Membership starts in EX mode */
2251 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2252 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2257 * Begin state merge transaction at IGMP layer.
2264 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2267 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2269 IN_MULTI_LIST_UNLOCK();
2273 imo->imo_membership[idx] = inm;
2275 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2276 IN_MULTI_LIST_LOCK();
2277 error = inm_merge(inm, imf);
2279 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2281 IN_MULTI_LIST_UNLOCK();
2282 goto out_in_multi_locked;
2284 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2285 error = igmp_change_state(inm);
2286 IN_MULTI_LIST_UNLOCK();
2288 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2290 goto out_in_multi_locked;
2294 out_in_multi_locked:
2298 if (in_pcbrele_wlocked(inp))
2311 if (error && is_new) {
2312 inm = imo->imo_membership[idx];
2314 IN_MULTI_LIST_LOCK();
2315 inm_release_deferred(inm);
2316 IN_MULTI_LIST_UNLOCK();
2318 imo->imo_membership[idx] = NULL;
2319 --imo->imo_num_memberships;
2328 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2331 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2333 struct group_source_req gsr;
2334 struct ip_mreq_source mreqs;
2335 sockunion_t *gsa, *ssa;
2337 struct in_mfilter *imf;
2338 struct ip_moptions *imo;
2339 struct in_msource *ims;
2340 struct in_multi *inm;
2342 int error, is_final;
2348 memset(&gsr, 0, sizeof(struct group_source_req));
2349 gsa = (sockunion_t *)&gsr.gsr_group;
2350 gsa->ss.ss_family = AF_UNSPEC;
2351 ssa = (sockunion_t *)&gsr.gsr_source;
2352 ssa->ss.ss_family = AF_UNSPEC;
2354 switch (sopt->sopt_name) {
2355 case IP_DROP_MEMBERSHIP:
2356 case IP_DROP_SOURCE_MEMBERSHIP:
2357 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2358 error = sooptcopyin(sopt, &mreqs,
2359 sizeof(struct ip_mreq),
2360 sizeof(struct ip_mreq));
2362 * Swap interface and sourceaddr arguments,
2363 * as ip_mreq and ip_mreq_source are laid
2366 mreqs.imr_interface = mreqs.imr_sourceaddr;
2367 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2368 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2369 error = sooptcopyin(sopt, &mreqs,
2370 sizeof(struct ip_mreq_source),
2371 sizeof(struct ip_mreq_source));
2376 gsa->sin.sin_family = AF_INET;
2377 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2378 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2380 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2381 ssa->sin.sin_family = AF_INET;
2382 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2383 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2387 * Attempt to look up hinted ifp from interface address.
2388 * Fallthrough with null ifp iff lookup fails, to
2389 * preserve 4.4BSD mcast API idempotence.
2390 * XXX NOTE WELL: The RFC 3678 API is preferred because
2391 * using an IPv4 address as a key is racy.
2393 if (!in_nullhost(mreqs.imr_interface))
2394 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2396 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2397 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2401 case MCAST_LEAVE_GROUP:
2402 case MCAST_LEAVE_SOURCE_GROUP:
2403 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2404 error = sooptcopyin(sopt, &gsr,
2405 sizeof(struct group_req),
2406 sizeof(struct group_req));
2407 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2408 error = sooptcopyin(sopt, &gsr,
2409 sizeof(struct group_source_req),
2410 sizeof(struct group_source_req));
2415 if (gsa->sin.sin_family != AF_INET ||
2416 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2419 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2420 if (ssa->sin.sin_family != AF_INET ||
2421 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2425 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2426 return (EADDRNOTAVAIL);
2428 ifp = ifnet_byindex(gsr.gsr_interface);
2431 return (EADDRNOTAVAIL);
2435 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2436 __func__, sopt->sopt_name);
2437 return (EOPNOTSUPP);
2441 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2445 * Find the membership in the membership array.
2447 imo = inp_findmoptions(inp);
2448 idx = imo_match_group(imo, ifp, &gsa->sa);
2450 error = EADDRNOTAVAIL;
2451 goto out_inp_locked;
2453 inm = imo->imo_membership[idx];
2454 imf = &imo->imo_mfilters[idx];
2456 if (ssa->ss.ss_family != AF_UNSPEC)
2460 * Begin state merge transaction at socket layer.
2462 INP_WLOCK_ASSERT(inp);
2465 * If we were instructed only to leave a given source, do so.
2466 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2471 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2472 error = EADDRNOTAVAIL;
2473 goto out_inp_locked;
2475 ims = imo_match_source(imo, idx, &ssa->sa);
2477 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2478 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2479 error = EADDRNOTAVAIL;
2480 goto out_inp_locked;
2482 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2483 error = imf_prune(imf, &ssa->sin);
2485 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2487 goto out_inp_locked;
2492 * Begin state merge transaction at IGMP layer.
2500 * Give up the multicast address record to which
2501 * the membership points.
2503 (void)in_leavegroup_locked(inm, imf);
2505 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2506 IN_MULTI_LIST_LOCK();
2507 error = inm_merge(inm, imf);
2509 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2511 IN_MULTI_LIST_UNLOCK();
2512 goto out_in_multi_locked;
2515 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2516 error = igmp_change_state(inm);
2517 IN_MULTI_LIST_UNLOCK();
2519 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2524 out_in_multi_locked:
2528 if (in_pcbrele_wlocked(inp))
2539 /* Remove the gap in the membership and filter array. */
2540 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2541 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2542 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2544 imo->imo_num_memberships--;
2553 * Select the interface for transmitting IPv4 multicast datagrams.
2555 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2556 * may be passed to this socket option. An address of INADDR_ANY or an
2557 * interface index of 0 is used to remove a previous selection.
2558 * When no interface is selected, one is chosen for every send.
2561 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2563 struct in_addr addr;
2564 struct ip_mreqn mreqn;
2566 struct ip_moptions *imo;
2569 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2571 * An interface index was specified using the
2572 * Linux-derived ip_mreqn structure.
2574 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2575 sizeof(struct ip_mreqn));
2579 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2582 if (mreqn.imr_ifindex == 0) {
2585 ifp = ifnet_byindex(mreqn.imr_ifindex);
2587 return (EADDRNOTAVAIL);
2591 * An interface was specified by IPv4 address.
2592 * This is the traditional BSD usage.
2594 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2595 sizeof(struct in_addr));
2598 if (in_nullhost(addr)) {
2601 INADDR_TO_IFP(addr, ifp);
2603 return (EADDRNOTAVAIL);
2605 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2606 ntohl(addr.s_addr));
2609 /* Reject interfaces which do not support multicast. */
2610 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2611 return (EOPNOTSUPP);
2613 imo = inp_findmoptions(inp);
2614 imo->imo_multicast_ifp = ifp;
2615 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2622 * Atomically set source filters on a socket for an IPv4 multicast group.
2624 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2627 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2629 struct __msfilterreq msfr;
2632 struct in_mfilter *imf;
2633 struct ip_moptions *imo;
2634 struct in_multi *inm;
2638 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2639 sizeof(struct __msfilterreq));
2643 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2646 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2647 msfr.msfr_fmode != MCAST_INCLUDE))
2650 if (msfr.msfr_group.ss_family != AF_INET ||
2651 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2654 gsa = (sockunion_t *)&msfr.msfr_group;
2655 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2658 gsa->sin.sin_port = 0; /* ignore port */
2660 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2661 return (EADDRNOTAVAIL);
2663 ifp = ifnet_byindex(msfr.msfr_ifindex);
2665 return (EADDRNOTAVAIL);
2668 * Take the INP write lock.
2669 * Check if this socket is a member of this group.
2671 imo = inp_findmoptions(inp);
2672 idx = imo_match_group(imo, ifp, &gsa->sa);
2673 if (idx == -1 || imo->imo_mfilters == NULL) {
2674 error = EADDRNOTAVAIL;
2675 goto out_inp_locked;
2677 inm = imo->imo_membership[idx];
2678 imf = &imo->imo_mfilters[idx];
2681 * Begin state merge transaction at socket layer.
2683 INP_WLOCK_ASSERT(inp);
2685 imf->imf_st[1] = msfr.msfr_fmode;
2688 * Apply any new source filters, if present.
2689 * Make a copy of the user-space source vector so
2690 * that we may copy them with a single copyin. This
2691 * allows us to deal with page faults up-front.
2693 if (msfr.msfr_nsrcs > 0) {
2694 struct in_msource *lims;
2695 struct sockaddr_in *psin;
2696 struct sockaddr_storage *kss, *pkss;
2701 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2702 __func__, (unsigned long)msfr.msfr_nsrcs);
2703 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2705 error = copyin(msfr.msfr_srcs, kss,
2706 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2715 * Mark all source filters as UNDEFINED at t1.
2716 * Restore new group filter mode, as imf_leave()
2717 * will set it to INCLUDE.
2720 imf->imf_st[1] = msfr.msfr_fmode;
2723 * Update socket layer filters at t1, lazy-allocating
2724 * new entries. This saves a bunch of memory at the
2725 * cost of one RB_FIND() per source entry; duplicate
2726 * entries in the msfr_nsrcs vector are ignored.
2727 * If we encounter an error, rollback transaction.
2729 * XXX This too could be replaced with a set-symmetric
2730 * difference like loop to avoid walking from root
2731 * every time, as the key space is common.
2733 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2734 psin = (struct sockaddr_in *)pkss;
2735 if (psin->sin_family != AF_INET) {
2736 error = EAFNOSUPPORT;
2739 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2743 error = imf_get_source(imf, psin, &lims);
2746 lims->imsl_st[1] = imf->imf_st[1];
2752 goto out_imf_rollback;
2754 INP_WLOCK_ASSERT(inp);
2758 * Begin state merge transaction at IGMP layer.
2760 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2761 IN_MULTI_LIST_LOCK();
2762 error = inm_merge(inm, imf);
2764 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2765 IN_MULTI_LIST_UNLOCK();
2766 goto out_in_multi_locked;
2769 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2770 error = igmp_change_state(inm);
2771 IN_MULTI_LIST_UNLOCK();
2773 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2775 out_in_multi_locked:
2793 * Set the IP multicast options in response to user setsockopt().
2795 * Many of the socket options handled in this function duplicate the
2796 * functionality of socket options in the regular unicast API. However,
2797 * it is not possible to merge the duplicate code, because the idempotence
2798 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2799 * the effects of these options must be treated as separate and distinct.
2801 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2802 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2803 * is refactored to no longer use vifs.
2806 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2808 struct ip_moptions *imo;
2814 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2815 * or is a divert socket, reject it.
2817 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2818 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2819 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2820 return (EOPNOTSUPP);
2822 switch (sopt->sopt_name) {
2823 case IP_MULTICAST_VIF: {
2826 * Select a multicast VIF for transmission.
2827 * Only useful if multicast forwarding is active.
2829 if (legal_vif_num == NULL) {
2833 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2836 if (!legal_vif_num(vifi) && (vifi != -1)) {
2840 imo = inp_findmoptions(inp);
2841 imo->imo_multicast_vif = vifi;
2846 case IP_MULTICAST_IF:
2847 error = inp_set_multicast_if(inp, sopt);
2850 case IP_MULTICAST_TTL: {
2854 * Set the IP time-to-live for outgoing multicast packets.
2855 * The original multicast API required a char argument,
2856 * which is inconsistent with the rest of the socket API.
2857 * We allow either a char or an int.
2859 if (sopt->sopt_valsize == sizeof(u_char)) {
2860 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2867 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2877 imo = inp_findmoptions(inp);
2878 imo->imo_multicast_ttl = ttl;
2883 case IP_MULTICAST_LOOP: {
2887 * Set the loopback flag for outgoing multicast packets.
2888 * Must be zero or one. The original multicast API required a
2889 * char argument, which is inconsistent with the rest
2890 * of the socket API. We allow either a char or an int.
2892 if (sopt->sopt_valsize == sizeof(u_char)) {
2893 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2900 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2904 loop = (u_char)iloop;
2906 imo = inp_findmoptions(inp);
2907 imo->imo_multicast_loop = !!loop;
2912 case IP_ADD_MEMBERSHIP:
2913 case IP_ADD_SOURCE_MEMBERSHIP:
2914 case MCAST_JOIN_GROUP:
2915 case MCAST_JOIN_SOURCE_GROUP:
2916 error = inp_join_group(inp, sopt);
2919 case IP_DROP_MEMBERSHIP:
2920 case IP_DROP_SOURCE_MEMBERSHIP:
2921 case MCAST_LEAVE_GROUP:
2922 case MCAST_LEAVE_SOURCE_GROUP:
2923 error = inp_leave_group(inp, sopt);
2926 case IP_BLOCK_SOURCE:
2927 case IP_UNBLOCK_SOURCE:
2928 case MCAST_BLOCK_SOURCE:
2929 case MCAST_UNBLOCK_SOURCE:
2930 error = inp_block_unblock_source(inp, sopt);
2934 error = inp_set_source_filters(inp, sopt);
2942 INP_UNLOCK_ASSERT(inp);
2948 * Expose IGMP's multicast filter mode and source list(s) to userland,
2949 * keyed by (ifindex, group).
2950 * The filter mode is written out as a uint32_t, followed by
2951 * 0..n of struct in_addr.
2952 * For use by ifmcstat(8).
2953 * SMPng: NOTE: unlocked read of ifindex space.
2956 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2958 struct in_addr src, group;
2960 struct ifmultiaddr *ifma;
2961 struct in_multi *inm;
2962 struct ip_msource *ims;
2966 uint32_t fmode, ifindex;
2971 if (req->newptr != NULL)
2978 if (ifindex <= 0 || ifindex > V_if_index) {
2979 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2984 group.s_addr = name[1];
2985 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2986 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2987 __func__, ntohl(group.s_addr));
2991 ifp = ifnet_byindex(ifindex);
2993 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2998 retval = sysctl_wire_old_buffer(req,
2999 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
3003 IN_MULTI_LIST_LOCK();
3006 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3007 if (ifma->ifma_addr->sa_family != AF_INET ||
3008 ifma->ifma_protospec == NULL)
3010 inm = (struct in_multi *)ifma->ifma_protospec;
3011 if (!in_hosteq(inm->inm_addr, group))
3013 fmode = inm->inm_st[1].iss_fmode;
3014 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
3017 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
3018 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
3021 * Only copy-out sources which are in-mode.
3023 if (fmode != ims_get_mode(inm, ims, 1)) {
3024 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
3028 src.s_addr = htonl(ims->ims_haddr);
3029 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
3034 IF_ADDR_RUNLOCK(ifp);
3036 IN_MULTI_LIST_UNLOCK();
3041 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
3043 static const char *inm_modestrs[] = { "un", "in", "ex" };
3046 inm_mode_str(const int mode)
3049 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
3050 return (inm_modestrs[mode]);
3054 static const char *inm_statestrs[] = {
3067 inm_state_str(const int state)
3070 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3071 return (inm_statestrs[state]);
3076 * Dump an in_multi structure to the console.
3079 inm_print(const struct in_multi *inm)
3082 char addrbuf[INET_ADDRSTRLEN];
3084 if ((ktr_mask & KTR_IGMPV3) == 0)
3087 printf("%s: --- begin inm %p ---\n", __func__, inm);
3088 printf("addr %s ifp %p(%s) ifma %p\n",
3089 inet_ntoa_r(inm->inm_addr, addrbuf),
3091 inm->inm_ifp->if_xname,
3093 printf("timer %u state %s refcount %u scq.len %u\n",
3095 inm_state_str(inm->inm_state),
3097 inm->inm_scq.mq_len);
3098 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3103 for (t = 0; t < 2; t++) {
3104 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3105 inm_mode_str(inm->inm_st[t].iss_fmode),
3106 inm->inm_st[t].iss_asm,
3107 inm->inm_st[t].iss_ex,
3108 inm->inm_st[t].iss_in,
3109 inm->inm_st[t].iss_rec);
3111 printf("%s: --- end inm %p ---\n", __func__, inm);
3114 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3117 inm_print(const struct in_multi *inm)
3122 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3124 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);