2 * Copyright (c) 2007-2009 Bruce Simpson.
3 * Copyright (c) 2005 Robert N. M. Watson.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote
15 * products derived from this software without specific prior written
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * IPv4 multicast socket, group, and socket option processing module.
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
42 #include <sys/malloc.h>
44 #include <sys/protosw.h>
45 #include <sys/rmlock.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/protosw.h>
49 #include <sys/sysctl.h>
51 #include <sys/taskqueue.h>
55 #include <net/if_var.h>
56 #include <net/if_dl.h>
57 #include <net/route.h>
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip_var.h>
65 #include <netinet/igmp_var.h>
68 #define KTR_IGMPV3 KTR_INET
71 #ifndef __SOCKUNION_DECLARED
73 struct sockaddr_storage ss;
75 struct sockaddr_dl sdl;
76 struct sockaddr_in sin;
78 typedef union sockunion sockunion_t;
79 #define __SOCKUNION_DECLARED
80 #endif /* __SOCKUNION_DECLARED */
82 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
83 "IPv4 multicast PCB-layer source filter");
84 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
85 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
86 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
87 "IPv4 multicast IGMP-layer source filter");
91 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
92 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
93 * it can be taken by code in net/if.c also.
94 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
96 * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly
97 * any need for in_multi itself to be virtualized -- it is bound to an ifp
98 * anyway no matter what happens.
100 struct mtx in_multi_mtx;
101 MTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF);
104 * Functions with non-static linkage defined in this file should be
105 * declared in in_var.h:
110 * in_joingroup_locked()
112 * in_leavegroup_locked()
118 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
121 static void imf_commit(struct in_mfilter *);
122 static int imf_get_source(struct in_mfilter *imf,
123 const struct sockaddr_in *psin,
124 struct in_msource **);
125 static struct in_msource *
126 imf_graft(struct in_mfilter *, const uint8_t,
127 const struct sockaddr_in *);
128 static void imf_leave(struct in_mfilter *);
129 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
130 static void imf_purge(struct in_mfilter *);
131 static void imf_rollback(struct in_mfilter *);
132 static void imf_reap(struct in_mfilter *);
133 static int imo_grow(struct ip_moptions *);
134 static size_t imo_match_group(const struct ip_moptions *,
135 const struct ifnet *, const struct sockaddr *);
136 static struct in_msource *
137 imo_match_source(const struct ip_moptions *, const size_t,
138 const struct sockaddr *);
139 static void ims_merge(struct ip_msource *ims,
140 const struct in_msource *lims, const int rollback);
141 static int in_getmulti(struct ifnet *, const struct in_addr *,
143 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
144 const int noalloc, struct ip_msource **pims);
146 static int inm_is_ifp_detached(const struct in_multi *);
148 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
149 static void inm_purge(struct in_multi *);
150 static void inm_reap(struct in_multi *);
151 static struct ip_moptions *
152 inp_findmoptions(struct inpcb *);
153 static void inp_freemoptions_internal(struct ip_moptions *);
154 static void inp_gcmoptions(void *, int);
155 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
156 static int inp_join_group(struct inpcb *, struct sockopt *);
157 static int inp_leave_group(struct inpcb *, struct sockopt *);
158 static struct ifnet *
159 inp_lookup_mcast_ifp(const struct inpcb *,
160 const struct sockaddr_in *, const struct in_addr);
161 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
162 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
163 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
164 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
166 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
169 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
170 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
171 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
172 "Max source filters per group");
174 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
175 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
176 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
177 "Max source filters per socket");
179 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
180 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
181 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
183 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
184 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
185 "Per-interface stack-wide source filters");
187 static STAILQ_HEAD(, ip_moptions) imo_gc_list =
188 STAILQ_HEAD_INITIALIZER(imo_gc_list);
189 static struct task imo_gc_task = TASK_INITIALIZER(0, inp_gcmoptions, NULL);
193 * Inline function which wraps assertions for a valid ifp.
194 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
198 inm_is_ifp_detached(const struct in_multi *inm)
202 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
203 ifp = inm->inm_ifma->ifma_ifp;
206 * Sanity check that netinet's notion of ifp is the
209 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
212 return (ifp == NULL);
217 * Initialize an in_mfilter structure to a known state at t0, t1
218 * with an empty source filter list.
221 imf_init(struct in_mfilter *imf, const int st0, const int st1)
223 memset(imf, 0, sizeof(struct in_mfilter));
224 RB_INIT(&imf->imf_sources);
225 imf->imf_st[0] = st0;
226 imf->imf_st[1] = st1;
230 * Function for looking up an in_multi record for an IPv4 multicast address
231 * on a given interface. ifp must be valid. If no record found, return NULL.
232 * The IN_MULTI_LOCK and IF_ADDR_LOCK on ifp must be held.
235 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
237 struct ifmultiaddr *ifma;
238 struct in_multi *inm;
240 IN_MULTI_LOCK_ASSERT();
241 IF_ADDR_LOCK_ASSERT(ifp);
244 TAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
245 if (ifma->ifma_addr->sa_family == AF_INET) {
246 inm = (struct in_multi *)ifma->ifma_protospec;
247 if (inm->inm_addr.s_addr == ina.s_addr)
256 * Wrapper for inm_lookup_locked().
257 * The IF_ADDR_LOCK will be taken on ifp and released on return.
260 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
262 struct in_multi *inm;
264 IN_MULTI_LOCK_ASSERT();
266 inm = inm_lookup_locked(ifp, ina);
267 IF_ADDR_RUNLOCK(ifp);
273 * Resize the ip_moptions vector to the next power-of-two minus 1.
274 * May be called with locks held; do not sleep.
277 imo_grow(struct ip_moptions *imo)
279 struct in_multi **nmships;
280 struct in_multi **omships;
281 struct in_mfilter *nmfilters;
282 struct in_mfilter *omfilters;
289 omships = imo->imo_membership;
290 omfilters = imo->imo_mfilters;
291 oldmax = imo->imo_max_memberships;
292 newmax = ((oldmax + 1) * 2) - 1;
294 if (newmax <= IP_MAX_MEMBERSHIPS) {
295 nmships = (struct in_multi **)realloc(omships,
296 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
297 nmfilters = (struct in_mfilter *)realloc(omfilters,
298 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
299 if (nmships != NULL && nmfilters != NULL) {
300 /* Initialize newly allocated source filter heads. */
301 for (idx = oldmax; idx < newmax; idx++) {
302 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
305 imo->imo_max_memberships = newmax;
306 imo->imo_membership = nmships;
307 imo->imo_mfilters = nmfilters;
311 if (nmships == NULL || nmfilters == NULL) {
313 free(nmships, M_IPMOPTS);
314 if (nmfilters != NULL)
315 free(nmfilters, M_INMFILTER);
316 return (ETOOMANYREFS);
323 * Find an IPv4 multicast group entry for this ip_moptions instance
324 * which matches the specified group, and optionally an interface.
325 * Return its index into the array, or -1 if not found.
328 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
329 const struct sockaddr *group)
331 const struct sockaddr_in *gsin;
332 struct in_multi **pinm;
336 gsin = (const struct sockaddr_in *)group;
338 /* The imo_membership array may be lazy allocated. */
339 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
342 nmships = imo->imo_num_memberships;
343 pinm = &imo->imo_membership[0];
344 for (idx = 0; idx < nmships; idx++, pinm++) {
347 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
348 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
359 * Find an IPv4 multicast source entry for this imo which matches
360 * the given group index for this socket, and source address.
362 * NOTE: This does not check if the entry is in-mode, merely if
363 * it exists, which may not be the desired behaviour.
365 static struct in_msource *
366 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
367 const struct sockaddr *src)
369 struct ip_msource find;
370 struct in_mfilter *imf;
371 struct ip_msource *ims;
372 const sockunion_t *psa;
374 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
375 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
376 ("%s: invalid index %d\n", __func__, (int)gidx));
378 /* The imo_mfilters array may be lazy allocated. */
379 if (imo->imo_mfilters == NULL)
381 imf = &imo->imo_mfilters[gidx];
383 /* Source trees are keyed in host byte order. */
384 psa = (const sockunion_t *)src;
385 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
386 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
388 return ((struct in_msource *)ims);
392 * Perform filtering for multicast datagrams on a socket by group and source.
394 * Returns 0 if a datagram should be allowed through, or various error codes
395 * if the socket was not a member of the group, or the source was muted, etc.
398 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
399 const struct sockaddr *group, const struct sockaddr *src)
402 struct in_msource *ims;
405 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
407 gidx = imo_match_group(imo, ifp, group);
409 return (MCAST_NOTGMEMBER);
412 * Check if the source was included in an (S,G) join.
413 * Allow reception on exclusive memberships by default,
414 * reject reception on inclusive memberships by default.
415 * Exclude source only if an in-mode exclude filter exists.
416 * Include source only if an in-mode include filter exists.
417 * NOTE: We are comparing group state here at IGMP t1 (now)
418 * with socket-layer t0 (since last downcall).
420 mode = imo->imo_mfilters[gidx].imf_st[1];
421 ims = imo_match_source(imo, gidx, src);
423 if ((ims == NULL && mode == MCAST_INCLUDE) ||
424 (ims != NULL && ims->imsl_st[0] != mode))
425 return (MCAST_NOTSMEMBER);
431 * Find and return a reference to an in_multi record for (ifp, group),
432 * and bump its reference count.
433 * If one does not exist, try to allocate it, and update link-layer multicast
434 * filters on ifp to listen for group.
435 * Assumes the IN_MULTI lock is held across the call.
436 * Return 0 if successful, otherwise return an appropriate error code.
439 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
440 struct in_multi **pinm)
442 struct sockaddr_in gsin;
443 struct ifmultiaddr *ifma;
444 struct in_ifinfo *ii;
445 struct in_multi *inm;
448 IN_MULTI_LOCK_ASSERT();
450 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
452 inm = inm_lookup(ifp, *group);
455 * If we already joined this group, just bump the
456 * refcount and return it.
458 KASSERT(inm->inm_refcount >= 1,
459 ("%s: bad refcount %d", __func__, inm->inm_refcount));
465 memset(&gsin, 0, sizeof(gsin));
466 gsin.sin_family = AF_INET;
467 gsin.sin_len = sizeof(struct sockaddr_in);
468 gsin.sin_addr = *group;
471 * Check if a link-layer group is already associated
472 * with this network-layer group on the given ifnet.
474 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
478 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
482 * If something other than netinet is occupying the link-layer
483 * group, print a meaningful error message and back out of
485 * Otherwise, bump the refcount on the existing network-layer
486 * group association and return it.
488 if (ifma->ifma_protospec != NULL) {
489 inm = (struct in_multi *)ifma->ifma_protospec;
491 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
493 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
494 ("%s: ifma not AF_INET", __func__));
495 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
496 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
497 !in_hosteq(inm->inm_addr, *group))
498 panic("%s: ifma %p is inconsistent with %p (%s)",
499 __func__, ifma, inm, inet_ntoa(*group));
503 IF_ADDR_WUNLOCK(ifp);
507 IF_ADDR_WLOCK_ASSERT(ifp);
510 * A new in_multi record is needed; allocate and initialize it.
511 * We DO NOT perform an IGMP join as the in_ layer may need to
512 * push an initial source list down to IGMP to support SSM.
514 * The initial source filter state is INCLUDE, {} as per the RFC.
516 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
518 if_delmulti_ifma(ifma);
519 IF_ADDR_WUNLOCK(ifp);
522 inm->inm_addr = *group;
524 inm->inm_igi = ii->ii_igmp;
525 inm->inm_ifma = ifma;
526 inm->inm_refcount = 1;
527 inm->inm_state = IGMP_NOT_MEMBER;
528 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
529 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
530 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
531 RB_INIT(&inm->inm_srcs);
533 ifma->ifma_protospec = inm;
537 IF_ADDR_WUNLOCK(ifp);
542 * Drop a reference to an in_multi record.
544 * If the refcount drops to 0, free the in_multi record and
545 * delete the underlying link-layer membership.
548 inm_release_locked(struct in_multi *inm)
550 struct ifmultiaddr *ifma;
552 IN_MULTI_LOCK_ASSERT();
554 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
556 if (--inm->inm_refcount > 0) {
557 CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__,
562 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
564 ifma = inm->inm_ifma;
566 /* XXX this access is not covered by IF_ADDR_LOCK */
567 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
568 KASSERT(ifma->ifma_protospec == inm,
569 ("%s: ifma_protospec != inm", __func__));
570 ifma->ifma_protospec = NULL;
574 free(inm, M_IPMADDR);
576 if_delmulti_ifma(ifma);
580 * Clear recorded source entries for a group.
581 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
582 * FIXME: Should reap.
585 inm_clear_recorded(struct in_multi *inm)
587 struct ip_msource *ims;
589 IN_MULTI_LOCK_ASSERT();
591 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
594 --inm->inm_st[1].iss_rec;
597 KASSERT(inm->inm_st[1].iss_rec == 0,
598 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
602 * Record a source as pending for a Source-Group IGMPv3 query.
603 * This lives here as it modifies the shared tree.
605 * inm is the group descriptor.
606 * naddr is the address of the source to record in network-byte order.
608 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
609 * lazy-allocate a source node in response to an SG query.
610 * Otherwise, no allocation is performed. This saves some memory
611 * with the trade-off that the source will not be reported to the
612 * router if joined in the window between the query response and
613 * the group actually being joined on the local host.
615 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
616 * This turns off the allocation of a recorded source entry if
617 * the group has not been joined.
619 * Return 0 if the source didn't exist or was already marked as recorded.
620 * Return 1 if the source was marked as recorded by this function.
621 * Return <0 if any error occured (negated errno code).
624 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
626 struct ip_msource find;
627 struct ip_msource *ims, *nims;
629 IN_MULTI_LOCK_ASSERT();
631 find.ims_haddr = ntohl(naddr);
632 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
633 if (ims && ims->ims_stp)
636 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
638 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
642 nims->ims_haddr = find.ims_haddr;
643 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
649 * Mark the source as recorded and update the recorded
653 ++inm->inm_st[1].iss_rec;
659 * Return a pointer to an in_msource owned by an in_mfilter,
660 * given its source address.
661 * Lazy-allocate if needed. If this is a new entry its filter state is
664 * imf is the filter set being modified.
665 * haddr is the source address in *host* byte-order.
667 * SMPng: May be called with locks held; malloc must not block.
670 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
671 struct in_msource **plims)
673 struct ip_msource find;
674 struct ip_msource *ims, *nims;
675 struct in_msource *lims;
682 /* key is host byte order */
683 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
684 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
685 lims = (struct in_msource *)ims;
687 if (imf->imf_nsrc == in_mcast_maxsocksrc)
689 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
693 lims = (struct in_msource *)nims;
694 lims->ims_haddr = find.ims_haddr;
695 lims->imsl_st[0] = MCAST_UNDEFINED;
696 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
706 * Graft a source entry into an existing socket-layer filter set,
707 * maintaining any required invariants and checking allocations.
709 * The source is marked as being in the new filter mode at t1.
711 * Return the pointer to the new node, otherwise return NULL.
713 static struct in_msource *
714 imf_graft(struct in_mfilter *imf, const uint8_t st1,
715 const struct sockaddr_in *psin)
717 struct ip_msource *nims;
718 struct in_msource *lims;
720 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
724 lims = (struct in_msource *)nims;
725 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
726 lims->imsl_st[0] = MCAST_UNDEFINED;
727 lims->imsl_st[1] = st1;
728 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
735 * Prune a source entry from an existing socket-layer filter set,
736 * maintaining any required invariants and checking allocations.
738 * The source is marked as being left at t1, it is not freed.
740 * Return 0 if no error occurred, otherwise return an errno value.
743 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
745 struct ip_msource find;
746 struct ip_msource *ims;
747 struct in_msource *lims;
749 /* key is host byte order */
750 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
751 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
754 lims = (struct in_msource *)ims;
755 lims->imsl_st[1] = MCAST_UNDEFINED;
760 * Revert socket-layer filter set deltas at t1 to t0 state.
763 imf_rollback(struct in_mfilter *imf)
765 struct ip_msource *ims, *tims;
766 struct in_msource *lims;
768 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
769 lims = (struct in_msource *)ims;
770 if (lims->imsl_st[0] == lims->imsl_st[1]) {
771 /* no change at t1 */
773 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
774 /* revert change to existing source at t1 */
775 lims->imsl_st[1] = lims->imsl_st[0];
777 /* revert source added t1 */
778 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
779 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
780 free(ims, M_INMFILTER);
784 imf->imf_st[1] = imf->imf_st[0];
788 * Mark socket-layer filter set as INCLUDE {} at t1.
791 imf_leave(struct in_mfilter *imf)
793 struct ip_msource *ims;
794 struct in_msource *lims;
796 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
797 lims = (struct in_msource *)ims;
798 lims->imsl_st[1] = MCAST_UNDEFINED;
800 imf->imf_st[1] = MCAST_INCLUDE;
804 * Mark socket-layer filter set deltas as committed.
807 imf_commit(struct in_mfilter *imf)
809 struct ip_msource *ims;
810 struct in_msource *lims;
812 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
813 lims = (struct in_msource *)ims;
814 lims->imsl_st[0] = lims->imsl_st[1];
816 imf->imf_st[0] = imf->imf_st[1];
820 * Reap unreferenced sources from socket-layer filter set.
823 imf_reap(struct in_mfilter *imf)
825 struct ip_msource *ims, *tims;
826 struct in_msource *lims;
828 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
829 lims = (struct in_msource *)ims;
830 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
831 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
832 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
833 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
834 free(ims, M_INMFILTER);
841 * Purge socket-layer filter set.
844 imf_purge(struct in_mfilter *imf)
846 struct ip_msource *ims, *tims;
848 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
849 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
850 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
851 free(ims, M_INMFILTER);
854 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
855 KASSERT(RB_EMPTY(&imf->imf_sources),
856 ("%s: imf_sources not empty", __func__));
860 * Look up a source filter entry for a multicast group.
862 * inm is the group descriptor to work with.
863 * haddr is the host-byte-order IPv4 address to look up.
864 * noalloc may be non-zero to suppress allocation of sources.
865 * *pims will be set to the address of the retrieved or allocated source.
867 * SMPng: NOTE: may be called with locks held.
868 * Return 0 if successful, otherwise return a non-zero error code.
871 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
872 const int noalloc, struct ip_msource **pims)
874 struct ip_msource find;
875 struct ip_msource *ims, *nims;
880 find.ims_haddr = haddr;
881 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
882 if (ims == NULL && !noalloc) {
883 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
885 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
889 nims->ims_haddr = haddr;
890 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
894 ia.s_addr = htonl(haddr);
895 CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__,
905 * Merge socket-layer source into IGMP-layer source.
906 * If rollback is non-zero, perform the inverse of the merge.
909 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
912 int n = rollback ? -1 : 1;
916 ia.s_addr = htonl(ims->ims_haddr);
919 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
920 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s",
921 __func__, n, inet_ntoa(ia));
922 ims->ims_st[1].ex -= n;
923 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
924 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s",
925 __func__, n, inet_ntoa(ia));
926 ims->ims_st[1].in -= n;
929 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
930 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s",
931 __func__, n, inet_ntoa(ia));
932 ims->ims_st[1].ex += n;
933 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
934 CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s",
935 __func__, n, inet_ntoa(ia));
936 ims->ims_st[1].in += n;
941 * Atomically update the global in_multi state, when a membership's
942 * filter list is being updated in any way.
944 * imf is the per-inpcb-membership group filter pointer.
945 * A fake imf may be passed for in-kernel consumers.
947 * XXX This is a candidate for a set-symmetric-difference style loop
948 * which would eliminate the repeated lookup from root of ims nodes,
949 * as they share the same key space.
951 * If any error occurred this function will back out of refcounts
952 * and return a non-zero value.
955 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
957 struct ip_msource *ims, *nims;
958 struct in_msource *lims;
967 * Update the source filters first, as this may fail.
968 * Maintain count of in-mode filters at t0, t1. These are
969 * used to work out if we transition into ASM mode or not.
970 * Maintain a count of source filters whose state was
971 * actually modified by this operation.
973 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
974 lims = (struct in_msource *)ims;
975 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
976 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
977 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
978 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
982 ims_merge(nims, lims, 0);
985 struct ip_msource *bims;
987 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
988 lims = (struct in_msource *)ims;
989 if (lims->imsl_st[0] == lims->imsl_st[1])
991 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
994 ims_merge(bims, lims, 1);
999 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1000 __func__, nsrc0, nsrc1);
1002 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1003 if (imf->imf_st[0] == imf->imf_st[1] &&
1004 imf->imf_st[1] == MCAST_INCLUDE) {
1006 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1007 --inm->inm_st[1].iss_in;
1011 /* Handle filter mode transition on socket. */
1012 if (imf->imf_st[0] != imf->imf_st[1]) {
1013 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1014 __func__, imf->imf_st[0], imf->imf_st[1]);
1016 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1017 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1018 --inm->inm_st[1].iss_ex;
1019 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1020 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1021 --inm->inm_st[1].iss_in;
1024 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1025 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1026 inm->inm_st[1].iss_ex++;
1027 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1028 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1029 inm->inm_st[1].iss_in++;
1034 * Track inm filter state in terms of listener counts.
1035 * If there are any exclusive listeners, stack-wide
1036 * membership is exclusive.
1037 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1038 * If no listeners remain, state is undefined at t1,
1039 * and the IGMP lifecycle for this group should finish.
1041 if (inm->inm_st[1].iss_ex > 0) {
1042 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1043 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1044 } else if (inm->inm_st[1].iss_in > 0) {
1045 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1046 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1048 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1049 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1052 /* Decrement ASM listener count on transition out of ASM mode. */
1053 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1054 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1055 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1056 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1057 --inm->inm_st[1].iss_asm;
1060 /* Increment ASM listener count on transition to ASM mode. */
1061 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1062 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1063 inm->inm_st[1].iss_asm++;
1066 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1071 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1078 * Mark an in_multi's filter set deltas as committed.
1079 * Called by IGMP after a state change has been enqueued.
1082 inm_commit(struct in_multi *inm)
1084 struct ip_msource *ims;
1086 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1087 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1090 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1091 ims->ims_st[0] = ims->ims_st[1];
1093 inm->inm_st[0] = inm->inm_st[1];
1097 * Reap unreferenced nodes from an in_multi's filter set.
1100 inm_reap(struct in_multi *inm)
1102 struct ip_msource *ims, *tims;
1104 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1105 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1106 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1109 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1110 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1111 free(ims, M_IPMSOURCE);
1117 * Purge all source nodes from an in_multi's filter set.
1120 inm_purge(struct in_multi *inm)
1122 struct ip_msource *ims, *tims;
1124 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1125 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1126 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1127 free(ims, M_IPMSOURCE);
1133 * Join a multicast group; unlocked entry point.
1135 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1136 * is not held. Fortunately, ifp is unlikely to have been detached
1137 * at this point, so we assume it's OK to recurse.
1140 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1141 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1146 error = in_joingroup_locked(ifp, gina, imf, pinm);
1153 * Join a multicast group; real entry point.
1155 * Only preserves atomicity at inm level.
1156 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1158 * If the IGMP downcall fails, the group is not joined, and an error
1162 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1163 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1165 struct in_mfilter timf;
1166 struct in_multi *inm;
1169 IN_MULTI_LOCK_ASSERT();
1171 CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__,
1172 inet_ntoa(*gina), ifp, ifp->if_xname);
1178 * If no imf was specified (i.e. kernel consumer),
1179 * fake one up and assume it is an ASM join.
1182 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1186 error = in_getmulti(ifp, gina, &inm);
1188 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1192 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1193 error = inm_merge(inm, imf);
1195 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1196 goto out_inm_release;
1199 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1200 error = igmp_change_state(inm);
1202 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1203 goto out_inm_release;
1208 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1209 inm_release_locked(inm);
1218 * Leave a multicast group; unlocked entry point.
1221 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1226 error = in_leavegroup_locked(inm, imf);
1233 * Leave a multicast group; real entry point.
1234 * All source filters will be expunged.
1236 * Only preserves atomicity at inm level.
1238 * Holding the write lock for the INP which contains imf
1239 * is highly advisable. We can't assert for it as imf does not
1240 * contain a back-pointer to the owning inp.
1242 * Note: This is not the same as inm_release(*) as this function also
1243 * makes a state change downcall into IGMP.
1246 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1248 struct in_mfilter timf;
1253 IN_MULTI_LOCK_ASSERT();
1255 CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__,
1256 inm, inet_ntoa(inm->inm_addr),
1257 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1261 * If no imf was specified (i.e. kernel consumer),
1262 * fake one up and assume it is an ASM join.
1265 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1270 * Begin state merge transaction at IGMP layer.
1272 * As this particular invocation should not cause any memory
1273 * to be allocated, and there is no opportunity to roll back
1274 * the transaction, it MUST NOT fail.
1276 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1277 error = inm_merge(inm, imf);
1278 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1280 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1281 CURVNET_SET(inm->inm_ifp->if_vnet);
1282 error = igmp_change_state(inm);
1285 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1287 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1288 inm_release_locked(inm);
1293 /*#ifndef BURN_BRIDGES*/
1295 * Join an IPv4 multicast group in (*,G) exclusive mode.
1296 * The group must be a 224.0.0.0/24 link-scope group.
1297 * This KPI is for legacy kernel consumers only.
1300 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1302 struct in_multi *pinm;
1305 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1306 ("%s: %s not in 224.0.0.0/24", __func__, inet_ntoa(*ap)));
1308 error = in_joingroup(ifp, ap, NULL, &pinm);
1316 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1317 * This KPI is for legacy kernel consumers only.
1320 in_delmulti(struct in_multi *inm)
1323 (void)in_leavegroup(inm, NULL);
1328 * Block or unblock an ASM multicast source on an inpcb.
1329 * This implements the delta-based API described in RFC 3678.
1331 * The delta-based API applies only to exclusive-mode memberships.
1332 * An IGMP downcall will be performed.
1334 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1336 * Return 0 if successful, otherwise return an appropriate error code.
1339 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1341 struct group_source_req gsr;
1342 sockunion_t *gsa, *ssa;
1344 struct in_mfilter *imf;
1345 struct ip_moptions *imo;
1346 struct in_msource *ims;
1347 struct in_multi *inm;
1356 memset(&gsr, 0, sizeof(struct group_source_req));
1357 gsa = (sockunion_t *)&gsr.gsr_group;
1358 ssa = (sockunion_t *)&gsr.gsr_source;
1360 switch (sopt->sopt_name) {
1361 case IP_BLOCK_SOURCE:
1362 case IP_UNBLOCK_SOURCE: {
1363 struct ip_mreq_source mreqs;
1365 error = sooptcopyin(sopt, &mreqs,
1366 sizeof(struct ip_mreq_source),
1367 sizeof(struct ip_mreq_source));
1371 gsa->sin.sin_family = AF_INET;
1372 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1373 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1375 ssa->sin.sin_family = AF_INET;
1376 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1377 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1379 if (!in_nullhost(mreqs.imr_interface))
1380 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1382 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1385 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1386 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1390 case MCAST_BLOCK_SOURCE:
1391 case MCAST_UNBLOCK_SOURCE:
1392 error = sooptcopyin(sopt, &gsr,
1393 sizeof(struct group_source_req),
1394 sizeof(struct group_source_req));
1398 if (gsa->sin.sin_family != AF_INET ||
1399 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1402 if (ssa->sin.sin_family != AF_INET ||
1403 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1406 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1407 return (EADDRNOTAVAIL);
1409 ifp = ifnet_byindex(gsr.gsr_interface);
1411 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1416 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1417 __func__, sopt->sopt_name);
1418 return (EOPNOTSUPP);
1422 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1426 * Check if we are actually a member of this group.
1428 imo = inp_findmoptions(inp);
1429 idx = imo_match_group(imo, ifp, &gsa->sa);
1430 if (idx == -1 || imo->imo_mfilters == NULL) {
1431 error = EADDRNOTAVAIL;
1432 goto out_inp_locked;
1435 KASSERT(imo->imo_mfilters != NULL,
1436 ("%s: imo_mfilters not allocated", __func__));
1437 imf = &imo->imo_mfilters[idx];
1438 inm = imo->imo_membership[idx];
1441 * Attempting to use the delta-based API on an
1442 * non exclusive-mode membership is an error.
1444 fmode = imf->imf_st[0];
1445 if (fmode != MCAST_EXCLUDE) {
1447 goto out_inp_locked;
1451 * Deal with error cases up-front:
1452 * Asked to block, but already blocked; or
1453 * Asked to unblock, but nothing to unblock.
1454 * If adding a new block entry, allocate it.
1456 ims = imo_match_source(imo, idx, &ssa->sa);
1457 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1458 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
1459 inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not ");
1460 error = EADDRNOTAVAIL;
1461 goto out_inp_locked;
1464 INP_WLOCK_ASSERT(inp);
1467 * Begin state merge transaction at socket layer.
1470 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1471 ims = imf_graft(imf, fmode, &ssa->sin);
1475 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1476 error = imf_prune(imf, &ssa->sin);
1480 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1481 goto out_imf_rollback;
1485 * Begin state merge transaction at IGMP layer.
1489 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1490 error = inm_merge(inm, imf);
1492 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1493 goto out_in_multi_locked;
1496 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1497 error = igmp_change_state(inm);
1499 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1501 out_in_multi_locked:
1519 * Given an inpcb, return its multicast options structure pointer. Accepts
1520 * an unlocked inpcb pointer, but will return it locked. May sleep.
1522 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1523 * SMPng: NOTE: Returns with the INP write lock held.
1525 static struct ip_moptions *
1526 inp_findmoptions(struct inpcb *inp)
1528 struct ip_moptions *imo;
1529 struct in_multi **immp;
1530 struct in_mfilter *imfp;
1534 if (inp->inp_moptions != NULL)
1535 return (inp->inp_moptions);
1539 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1540 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1542 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1543 M_INMFILTER, M_WAITOK);
1545 imo->imo_multicast_ifp = NULL;
1546 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1547 imo->imo_multicast_vif = -1;
1548 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1549 imo->imo_multicast_loop = in_mcast_loop;
1550 imo->imo_num_memberships = 0;
1551 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1552 imo->imo_membership = immp;
1554 /* Initialize per-group source filters. */
1555 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1556 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1557 imo->imo_mfilters = imfp;
1560 if (inp->inp_moptions != NULL) {
1561 free(imfp, M_INMFILTER);
1562 free(immp, M_IPMOPTS);
1563 free(imo, M_IPMOPTS);
1564 return (inp->inp_moptions);
1566 inp->inp_moptions = imo;
1571 * Discard the IP multicast options (and source filters). To minimize
1572 * the amount of work done while holding locks such as the INP's
1573 * pcbinfo lock (which is used in the receive path), the free
1574 * operation is performed asynchronously in a separate task.
1576 * SMPng: NOTE: assumes INP write lock is held.
1579 inp_freemoptions(struct ip_moptions *imo)
1582 KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1584 STAILQ_INSERT_TAIL(&imo_gc_list, imo, imo_link);
1586 taskqueue_enqueue(taskqueue_thread, &imo_gc_task);
1590 inp_freemoptions_internal(struct ip_moptions *imo)
1592 struct in_mfilter *imf;
1593 size_t idx, nmships;
1595 nmships = imo->imo_num_memberships;
1596 for (idx = 0; idx < nmships; ++idx) {
1597 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1600 (void)in_leavegroup(imo->imo_membership[idx], imf);
1605 if (imo->imo_mfilters)
1606 free(imo->imo_mfilters, M_INMFILTER);
1607 free(imo->imo_membership, M_IPMOPTS);
1608 free(imo, M_IPMOPTS);
1612 inp_gcmoptions(void *context, int pending)
1614 struct ip_moptions *imo;
1617 while (!STAILQ_EMPTY(&imo_gc_list)) {
1618 imo = STAILQ_FIRST(&imo_gc_list);
1619 STAILQ_REMOVE_HEAD(&imo_gc_list, imo_link);
1621 inp_freemoptions_internal(imo);
1628 * Atomically get source filters on a socket for an IPv4 multicast group.
1629 * Called with INP lock held; returns with lock released.
1632 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1634 struct __msfilterreq msfr;
1637 struct ip_moptions *imo;
1638 struct in_mfilter *imf;
1639 struct ip_msource *ims;
1640 struct in_msource *lims;
1641 struct sockaddr_in *psin;
1642 struct sockaddr_storage *ptss;
1643 struct sockaddr_storage *tss;
1645 size_t idx, nsrcs, ncsrcs;
1647 INP_WLOCK_ASSERT(inp);
1649 imo = inp->inp_moptions;
1650 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1654 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1655 sizeof(struct __msfilterreq));
1659 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1662 ifp = ifnet_byindex(msfr.msfr_ifindex);
1669 * Lookup group on the socket.
1671 gsa = (sockunion_t *)&msfr.msfr_group;
1672 idx = imo_match_group(imo, ifp, &gsa->sa);
1673 if (idx == -1 || imo->imo_mfilters == NULL) {
1675 return (EADDRNOTAVAIL);
1677 imf = &imo->imo_mfilters[idx];
1680 * Ignore memberships which are in limbo.
1682 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1686 msfr.msfr_fmode = imf->imf_st[1];
1689 * If the user specified a buffer, copy out the source filter
1690 * entries to userland gracefully.
1691 * We only copy out the number of entries which userland
1692 * has asked for, but we always tell userland how big the
1693 * buffer really needs to be.
1695 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1696 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1698 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1699 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1700 M_TEMP, M_NOWAIT | M_ZERO);
1708 * Count number of sources in-mode at t0.
1709 * If buffer space exists and remains, copy out source entries.
1711 nsrcs = msfr.msfr_nsrcs;
1714 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1715 lims = (struct in_msource *)ims;
1716 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1717 lims->imsl_st[0] != imf->imf_st[0])
1720 if (tss != NULL && nsrcs > 0) {
1721 psin = (struct sockaddr_in *)ptss;
1722 psin->sin_family = AF_INET;
1723 psin->sin_len = sizeof(struct sockaddr_in);
1724 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1734 error = copyout(tss, msfr.msfr_srcs,
1735 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1741 msfr.msfr_nsrcs = ncsrcs;
1742 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1748 * Return the IP multicast options in response to user getsockopt().
1751 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1753 struct rm_priotracker in_ifa_tracker;
1754 struct ip_mreqn mreqn;
1755 struct ip_moptions *imo;
1757 struct in_ifaddr *ia;
1762 imo = inp->inp_moptions;
1764 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1765 * or is a divert socket, reject it.
1767 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1768 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1769 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1771 return (EOPNOTSUPP);
1775 switch (sopt->sopt_name) {
1776 case IP_MULTICAST_VIF:
1778 optval = imo->imo_multicast_vif;
1782 error = sooptcopyout(sopt, &optval, sizeof(int));
1785 case IP_MULTICAST_IF:
1786 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1788 ifp = imo->imo_multicast_ifp;
1789 if (!in_nullhost(imo->imo_multicast_addr)) {
1790 mreqn.imr_address = imo->imo_multicast_addr;
1791 } else if (ifp != NULL) {
1792 mreqn.imr_ifindex = ifp->if_index;
1793 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1796 IA_SIN(ia)->sin_addr;
1797 ifa_free(&ia->ia_ifa);
1802 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1803 error = sooptcopyout(sopt, &mreqn,
1804 sizeof(struct ip_mreqn));
1806 error = sooptcopyout(sopt, &mreqn.imr_address,
1807 sizeof(struct in_addr));
1811 case IP_MULTICAST_TTL:
1813 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1815 optval = coptval = imo->imo_multicast_ttl;
1817 if (sopt->sopt_valsize == sizeof(u_char))
1818 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1820 error = sooptcopyout(sopt, &optval, sizeof(int));
1823 case IP_MULTICAST_LOOP:
1825 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1827 optval = coptval = imo->imo_multicast_loop;
1829 if (sopt->sopt_valsize == sizeof(u_char))
1830 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1832 error = sooptcopyout(sopt, &optval, sizeof(int));
1837 error = EADDRNOTAVAIL;
1840 error = inp_get_source_filters(inp, sopt);
1846 error = ENOPROTOOPT;
1850 INP_UNLOCK_ASSERT(inp);
1856 * Look up the ifnet to use for a multicast group membership,
1857 * given the IPv4 address of an interface, and the IPv4 group address.
1859 * This routine exists to support legacy multicast applications
1860 * which do not understand that multicast memberships are scoped to
1861 * specific physical links in the networking stack, or which need
1862 * to join link-scope groups before IPv4 addresses are configured.
1864 * If inp is non-NULL, use this socket's current FIB number for any
1865 * required FIB lookup.
1866 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1867 * and use its ifp; usually, this points to the default next-hop.
1869 * If the FIB lookup fails, attempt to use the first non-loopback
1870 * interface with multicast capability in the system as a
1871 * last resort. The legacy IPv4 ASM API requires that we do
1872 * this in order to allow groups to be joined when the routing
1873 * table has not yet been populated during boot.
1875 * Returns NULL if no ifp could be found.
1877 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1878 * FUTURE: Implement IPv4 source-address selection.
1880 static struct ifnet *
1881 inp_lookup_mcast_ifp(const struct inpcb *inp,
1882 const struct sockaddr_in *gsin, const struct in_addr ina)
1884 struct rm_priotracker in_ifa_tracker;
1887 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1888 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1889 ("%s: not multicast", __func__));
1892 if (!in_nullhost(ina)) {
1893 INADDR_TO_IFP(ina, ifp);
1898 memcpy(&ro.ro_dst, gsin, sizeof(struct sockaddr_in));
1899 in_rtalloc_ign(&ro, 0, inp ? inp->inp_inc.inc_fibnum : 0);
1900 if (ro.ro_rt != NULL) {
1901 ifp = ro.ro_rt->rt_ifp;
1902 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1905 struct in_ifaddr *ia;
1909 IN_IFADDR_RLOCK(&in_ifa_tracker);
1910 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1912 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1913 (mifp->if_flags & IFF_MULTICAST)) {
1918 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1926 * Join an IPv4 multicast group, possibly with a source.
1929 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1931 struct group_source_req gsr;
1932 sockunion_t *gsa, *ssa;
1934 struct in_mfilter *imf;
1935 struct ip_moptions *imo;
1936 struct in_multi *inm;
1937 struct in_msource *lims;
1947 memset(&gsr, 0, sizeof(struct group_source_req));
1948 gsa = (sockunion_t *)&gsr.gsr_group;
1949 gsa->ss.ss_family = AF_UNSPEC;
1950 ssa = (sockunion_t *)&gsr.gsr_source;
1951 ssa->ss.ss_family = AF_UNSPEC;
1953 switch (sopt->sopt_name) {
1954 case IP_ADD_MEMBERSHIP:
1955 case IP_ADD_SOURCE_MEMBERSHIP: {
1956 struct ip_mreq_source mreqs;
1958 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1959 error = sooptcopyin(sopt, &mreqs,
1960 sizeof(struct ip_mreq),
1961 sizeof(struct ip_mreq));
1963 * Do argument switcharoo from ip_mreq into
1964 * ip_mreq_source to avoid using two instances.
1966 mreqs.imr_interface = mreqs.imr_sourceaddr;
1967 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1968 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1969 error = sooptcopyin(sopt, &mreqs,
1970 sizeof(struct ip_mreq_source),
1971 sizeof(struct ip_mreq_source));
1976 gsa->sin.sin_family = AF_INET;
1977 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1978 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1980 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1981 ssa->sin.sin_family = AF_INET;
1982 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1983 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1986 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1989 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1990 mreqs.imr_interface);
1991 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1992 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1996 case MCAST_JOIN_GROUP:
1997 case MCAST_JOIN_SOURCE_GROUP:
1998 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1999 error = sooptcopyin(sopt, &gsr,
2000 sizeof(struct group_req),
2001 sizeof(struct group_req));
2002 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2003 error = sooptcopyin(sopt, &gsr,
2004 sizeof(struct group_source_req),
2005 sizeof(struct group_source_req));
2010 if (gsa->sin.sin_family != AF_INET ||
2011 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2015 * Overwrite the port field if present, as the sockaddr
2016 * being copied in may be matched with a binary comparison.
2018 gsa->sin.sin_port = 0;
2019 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2020 if (ssa->sin.sin_family != AF_INET ||
2021 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2023 ssa->sin.sin_port = 0;
2026 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2029 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2030 return (EADDRNOTAVAIL);
2031 ifp = ifnet_byindex(gsr.gsr_interface);
2035 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2036 __func__, sopt->sopt_name);
2037 return (EOPNOTSUPP);
2041 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2042 return (EADDRNOTAVAIL);
2044 imo = inp_findmoptions(inp);
2045 idx = imo_match_group(imo, ifp, &gsa->sa);
2049 inm = imo->imo_membership[idx];
2050 imf = &imo->imo_mfilters[idx];
2051 if (ssa->ss.ss_family != AF_UNSPEC) {
2053 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2054 * is an error. On an existing inclusive membership,
2055 * it just adds the source to the filter list.
2057 if (imf->imf_st[1] != MCAST_INCLUDE) {
2059 goto out_inp_locked;
2062 * Throw out duplicates.
2064 * XXX FIXME: This makes a naive assumption that
2065 * even if entries exist for *ssa in this imf,
2066 * they will be rejected as dupes, even if they
2067 * are not valid in the current mode (in-mode).
2069 * in_msource is transactioned just as for anything
2070 * else in SSM -- but note naive use of inm_graft()
2071 * below for allocating new filter entries.
2073 * This is only an issue if someone mixes the
2074 * full-state SSM API with the delta-based API,
2075 * which is discouraged in the relevant RFCs.
2077 lims = imo_match_source(imo, idx, &ssa->sa);
2078 if (lims != NULL /*&&
2079 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2080 error = EADDRNOTAVAIL;
2081 goto out_inp_locked;
2085 * MCAST_JOIN_GROUP on an existing exclusive
2086 * membership is an error; return EADDRINUSE
2087 * to preserve 4.4BSD API idempotence, and
2088 * avoid tedious detour to code below.
2089 * NOTE: This is bending RFC 3678 a bit.
2091 * On an existing inclusive membership, this is also
2092 * an error; if you want to change filter mode,
2093 * you must use the userland API setsourcefilter().
2094 * XXX We don't reject this for imf in UNDEFINED
2095 * state at t1, because allocation of a filter
2096 * is atomic with allocation of a membership.
2099 if (imf->imf_st[1] == MCAST_EXCLUDE)
2101 goto out_inp_locked;
2106 * Begin state merge transaction at socket layer.
2108 INP_WLOCK_ASSERT(inp);
2111 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2112 error = imo_grow(imo);
2114 goto out_inp_locked;
2117 * Allocate the new slot upfront so we can deal with
2118 * grafting the new source filter in same code path
2119 * as for join-source on existing membership.
2121 idx = imo->imo_num_memberships;
2122 imo->imo_membership[idx] = NULL;
2123 imo->imo_num_memberships++;
2124 KASSERT(imo->imo_mfilters != NULL,
2125 ("%s: imf_mfilters vector was not allocated", __func__));
2126 imf = &imo->imo_mfilters[idx];
2127 KASSERT(RB_EMPTY(&imf->imf_sources),
2128 ("%s: imf_sources not empty", __func__));
2132 * Graft new source into filter list for this inpcb's
2133 * membership of the group. The in_multi may not have
2134 * been allocated yet if this is a new membership, however,
2135 * the in_mfilter slot will be allocated and must be initialized.
2137 * Note: Grafting of exclusive mode filters doesn't happen
2139 * XXX: Should check for non-NULL lims (node exists but may
2140 * not be in-mode) for interop with full-state API.
2142 if (ssa->ss.ss_family != AF_UNSPEC) {
2143 /* Membership starts in IN mode */
2145 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2146 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2148 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2150 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2152 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2158 /* No address specified; Membership starts in EX mode */
2160 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2161 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2166 * Begin state merge transaction at IGMP layer.
2171 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2174 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2179 imo->imo_membership[idx] = inm;
2181 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2182 error = inm_merge(inm, imf);
2184 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2186 goto out_in_multi_locked;
2188 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2189 error = igmp_change_state(inm);
2191 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2193 goto out_in_multi_locked;
2197 out_in_multi_locked:
2201 INP_WLOCK_ASSERT(inp);
2213 if (error && is_new) {
2214 imo->imo_membership[idx] = NULL;
2215 --imo->imo_num_memberships;
2224 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2227 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2229 struct group_source_req gsr;
2230 struct ip_mreq_source mreqs;
2231 sockunion_t *gsa, *ssa;
2233 struct in_mfilter *imf;
2234 struct ip_moptions *imo;
2235 struct in_msource *ims;
2236 struct in_multi *inm;
2238 int error, is_final;
2244 memset(&gsr, 0, sizeof(struct group_source_req));
2245 gsa = (sockunion_t *)&gsr.gsr_group;
2246 gsa->ss.ss_family = AF_UNSPEC;
2247 ssa = (sockunion_t *)&gsr.gsr_source;
2248 ssa->ss.ss_family = AF_UNSPEC;
2250 switch (sopt->sopt_name) {
2251 case IP_DROP_MEMBERSHIP:
2252 case IP_DROP_SOURCE_MEMBERSHIP:
2253 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2254 error = sooptcopyin(sopt, &mreqs,
2255 sizeof(struct ip_mreq),
2256 sizeof(struct ip_mreq));
2258 * Swap interface and sourceaddr arguments,
2259 * as ip_mreq and ip_mreq_source are laid
2262 mreqs.imr_interface = mreqs.imr_sourceaddr;
2263 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2264 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2265 error = sooptcopyin(sopt, &mreqs,
2266 sizeof(struct ip_mreq_source),
2267 sizeof(struct ip_mreq_source));
2272 gsa->sin.sin_family = AF_INET;
2273 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2274 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2276 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2277 ssa->sin.sin_family = AF_INET;
2278 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2279 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2283 * Attempt to look up hinted ifp from interface address.
2284 * Fallthrough with null ifp iff lookup fails, to
2285 * preserve 4.4BSD mcast API idempotence.
2286 * XXX NOTE WELL: The RFC 3678 API is preferred because
2287 * using an IPv4 address as a key is racy.
2289 if (!in_nullhost(mreqs.imr_interface))
2290 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2292 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
2293 __func__, inet_ntoa(mreqs.imr_interface), ifp);
2297 case MCAST_LEAVE_GROUP:
2298 case MCAST_LEAVE_SOURCE_GROUP:
2299 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2300 error = sooptcopyin(sopt, &gsr,
2301 sizeof(struct group_req),
2302 sizeof(struct group_req));
2303 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2304 error = sooptcopyin(sopt, &gsr,
2305 sizeof(struct group_source_req),
2306 sizeof(struct group_source_req));
2311 if (gsa->sin.sin_family != AF_INET ||
2312 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2315 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2316 if (ssa->sin.sin_family != AF_INET ||
2317 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2321 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2322 return (EADDRNOTAVAIL);
2324 ifp = ifnet_byindex(gsr.gsr_interface);
2327 return (EADDRNOTAVAIL);
2331 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2332 __func__, sopt->sopt_name);
2333 return (EOPNOTSUPP);
2337 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2341 * Find the membership in the membership array.
2343 imo = inp_findmoptions(inp);
2344 idx = imo_match_group(imo, ifp, &gsa->sa);
2346 error = EADDRNOTAVAIL;
2347 goto out_inp_locked;
2349 inm = imo->imo_membership[idx];
2350 imf = &imo->imo_mfilters[idx];
2352 if (ssa->ss.ss_family != AF_UNSPEC)
2356 * Begin state merge transaction at socket layer.
2358 INP_WLOCK_ASSERT(inp);
2361 * If we were instructed only to leave a given source, do so.
2362 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2367 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2368 error = EADDRNOTAVAIL;
2369 goto out_inp_locked;
2371 ims = imo_match_source(imo, idx, &ssa->sa);
2373 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
2374 inet_ntoa(ssa->sin.sin_addr), "not ");
2375 error = EADDRNOTAVAIL;
2376 goto out_inp_locked;
2378 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2379 error = imf_prune(imf, &ssa->sin);
2381 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2383 goto out_inp_locked;
2388 * Begin state merge transaction at IGMP layer.
2394 * Give up the multicast address record to which
2395 * the membership points.
2397 (void)in_leavegroup_locked(inm, imf);
2399 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2400 error = inm_merge(inm, imf);
2402 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2404 goto out_in_multi_locked;
2407 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2408 error = igmp_change_state(inm);
2410 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2415 out_in_multi_locked:
2427 /* Remove the gap in the membership and filter array. */
2428 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2429 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2430 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2432 imo->imo_num_memberships--;
2441 * Select the interface for transmitting IPv4 multicast datagrams.
2443 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2444 * may be passed to this socket option. An address of INADDR_ANY or an
2445 * interface index of 0 is used to remove a previous selection.
2446 * When no interface is selected, one is chosen for every send.
2449 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2451 struct in_addr addr;
2452 struct ip_mreqn mreqn;
2454 struct ip_moptions *imo;
2457 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2459 * An interface index was specified using the
2460 * Linux-derived ip_mreqn structure.
2462 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2463 sizeof(struct ip_mreqn));
2467 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2470 if (mreqn.imr_ifindex == 0) {
2473 ifp = ifnet_byindex(mreqn.imr_ifindex);
2475 return (EADDRNOTAVAIL);
2479 * An interface was specified by IPv4 address.
2480 * This is the traditional BSD usage.
2482 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2483 sizeof(struct in_addr));
2486 if (in_nullhost(addr)) {
2489 INADDR_TO_IFP(addr, ifp);
2491 return (EADDRNOTAVAIL);
2493 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp,
2497 /* Reject interfaces which do not support multicast. */
2498 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2499 return (EOPNOTSUPP);
2501 imo = inp_findmoptions(inp);
2502 imo->imo_multicast_ifp = ifp;
2503 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2510 * Atomically set source filters on a socket for an IPv4 multicast group.
2512 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2515 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2517 struct __msfilterreq msfr;
2520 struct in_mfilter *imf;
2521 struct ip_moptions *imo;
2522 struct in_multi *inm;
2526 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2527 sizeof(struct __msfilterreq));
2531 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2534 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2535 msfr.msfr_fmode != MCAST_INCLUDE))
2538 if (msfr.msfr_group.ss_family != AF_INET ||
2539 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2542 gsa = (sockunion_t *)&msfr.msfr_group;
2543 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2546 gsa->sin.sin_port = 0; /* ignore port */
2548 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2549 return (EADDRNOTAVAIL);
2551 ifp = ifnet_byindex(msfr.msfr_ifindex);
2553 return (EADDRNOTAVAIL);
2556 * Take the INP write lock.
2557 * Check if this socket is a member of this group.
2559 imo = inp_findmoptions(inp);
2560 idx = imo_match_group(imo, ifp, &gsa->sa);
2561 if (idx == -1 || imo->imo_mfilters == NULL) {
2562 error = EADDRNOTAVAIL;
2563 goto out_inp_locked;
2565 inm = imo->imo_membership[idx];
2566 imf = &imo->imo_mfilters[idx];
2569 * Begin state merge transaction at socket layer.
2571 INP_WLOCK_ASSERT(inp);
2573 imf->imf_st[1] = msfr.msfr_fmode;
2576 * Apply any new source filters, if present.
2577 * Make a copy of the user-space source vector so
2578 * that we may copy them with a single copyin. This
2579 * allows us to deal with page faults up-front.
2581 if (msfr.msfr_nsrcs > 0) {
2582 struct in_msource *lims;
2583 struct sockaddr_in *psin;
2584 struct sockaddr_storage *kss, *pkss;
2589 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2590 __func__, (unsigned long)msfr.msfr_nsrcs);
2591 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2593 error = copyin(msfr.msfr_srcs, kss,
2594 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2603 * Mark all source filters as UNDEFINED at t1.
2604 * Restore new group filter mode, as imf_leave()
2605 * will set it to INCLUDE.
2608 imf->imf_st[1] = msfr.msfr_fmode;
2611 * Update socket layer filters at t1, lazy-allocating
2612 * new entries. This saves a bunch of memory at the
2613 * cost of one RB_FIND() per source entry; duplicate
2614 * entries in the msfr_nsrcs vector are ignored.
2615 * If we encounter an error, rollback transaction.
2617 * XXX This too could be replaced with a set-symmetric
2618 * difference like loop to avoid walking from root
2619 * every time, as the key space is common.
2621 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2622 psin = (struct sockaddr_in *)pkss;
2623 if (psin->sin_family != AF_INET) {
2624 error = EAFNOSUPPORT;
2627 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2631 error = imf_get_source(imf, psin, &lims);
2634 lims->imsl_st[1] = imf->imf_st[1];
2640 goto out_imf_rollback;
2642 INP_WLOCK_ASSERT(inp);
2646 * Begin state merge transaction at IGMP layer.
2648 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2649 error = inm_merge(inm, imf);
2651 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2652 goto out_in_multi_locked;
2655 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2656 error = igmp_change_state(inm);
2658 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2660 out_in_multi_locked:
2678 * Set the IP multicast options in response to user setsockopt().
2680 * Many of the socket options handled in this function duplicate the
2681 * functionality of socket options in the regular unicast API. However,
2682 * it is not possible to merge the duplicate code, because the idempotence
2683 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2684 * the effects of these options must be treated as separate and distinct.
2686 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2687 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2688 * is refactored to no longer use vifs.
2691 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2693 struct ip_moptions *imo;
2699 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2700 * or is a divert socket, reject it.
2702 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2703 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2704 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2705 return (EOPNOTSUPP);
2707 switch (sopt->sopt_name) {
2708 case IP_MULTICAST_VIF: {
2711 * Select a multicast VIF for transmission.
2712 * Only useful if multicast forwarding is active.
2714 if (legal_vif_num == NULL) {
2718 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2721 if (!legal_vif_num(vifi) && (vifi != -1)) {
2725 imo = inp_findmoptions(inp);
2726 imo->imo_multicast_vif = vifi;
2731 case IP_MULTICAST_IF:
2732 error = inp_set_multicast_if(inp, sopt);
2735 case IP_MULTICAST_TTL: {
2739 * Set the IP time-to-live for outgoing multicast packets.
2740 * The original multicast API required a char argument,
2741 * which is inconsistent with the rest of the socket API.
2742 * We allow either a char or an int.
2744 if (sopt->sopt_valsize == sizeof(u_char)) {
2745 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2752 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2762 imo = inp_findmoptions(inp);
2763 imo->imo_multicast_ttl = ttl;
2768 case IP_MULTICAST_LOOP: {
2772 * Set the loopback flag for outgoing multicast packets.
2773 * Must be zero or one. The original multicast API required a
2774 * char argument, which is inconsistent with the rest
2775 * of the socket API. We allow either a char or an int.
2777 if (sopt->sopt_valsize == sizeof(u_char)) {
2778 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2785 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2789 loop = (u_char)iloop;
2791 imo = inp_findmoptions(inp);
2792 imo->imo_multicast_loop = !!loop;
2797 case IP_ADD_MEMBERSHIP:
2798 case IP_ADD_SOURCE_MEMBERSHIP:
2799 case MCAST_JOIN_GROUP:
2800 case MCAST_JOIN_SOURCE_GROUP:
2801 error = inp_join_group(inp, sopt);
2804 case IP_DROP_MEMBERSHIP:
2805 case IP_DROP_SOURCE_MEMBERSHIP:
2806 case MCAST_LEAVE_GROUP:
2807 case MCAST_LEAVE_SOURCE_GROUP:
2808 error = inp_leave_group(inp, sopt);
2811 case IP_BLOCK_SOURCE:
2812 case IP_UNBLOCK_SOURCE:
2813 case MCAST_BLOCK_SOURCE:
2814 case MCAST_UNBLOCK_SOURCE:
2815 error = inp_block_unblock_source(inp, sopt);
2819 error = inp_set_source_filters(inp, sopt);
2827 INP_UNLOCK_ASSERT(inp);
2833 * Expose IGMP's multicast filter mode and source list(s) to userland,
2834 * keyed by (ifindex, group).
2835 * The filter mode is written out as a uint32_t, followed by
2836 * 0..n of struct in_addr.
2837 * For use by ifmcstat(8).
2838 * SMPng: NOTE: unlocked read of ifindex space.
2841 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2843 struct in_addr src, group;
2845 struct ifmultiaddr *ifma;
2846 struct in_multi *inm;
2847 struct ip_msource *ims;
2851 uint32_t fmode, ifindex;
2856 if (req->newptr != NULL)
2863 if (ifindex <= 0 || ifindex > V_if_index) {
2864 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2869 group.s_addr = name[1];
2870 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2871 CTR2(KTR_IGMPV3, "%s: group %s is not multicast",
2872 __func__, inet_ntoa(group));
2876 ifp = ifnet_byindex(ifindex);
2878 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2883 retval = sysctl_wire_old_buffer(req,
2884 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2891 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2892 if (ifma->ifma_addr->sa_family != AF_INET ||
2893 ifma->ifma_protospec == NULL)
2895 inm = (struct in_multi *)ifma->ifma_protospec;
2896 if (!in_hosteq(inm->inm_addr, group))
2898 fmode = inm->inm_st[1].iss_fmode;
2899 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2902 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2905 ina.s_addr = htonl(ims->ims_haddr);
2906 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2910 * Only copy-out sources which are in-mode.
2912 if (fmode != ims_get_mode(inm, ims, 1)) {
2913 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2917 src.s_addr = htonl(ims->ims_haddr);
2918 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2923 IF_ADDR_RUNLOCK(ifp);
2930 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2932 static const char *inm_modestrs[] = { "un", "in", "ex" };
2935 inm_mode_str(const int mode)
2938 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2939 return (inm_modestrs[mode]);
2943 static const char *inm_statestrs[] = {
2956 inm_state_str(const int state)
2959 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2960 return (inm_statestrs[state]);
2965 * Dump an in_multi structure to the console.
2968 inm_print(const struct in_multi *inm)
2972 if ((ktr_mask & KTR_IGMPV3) == 0)
2975 printf("%s: --- begin inm %p ---\n", __func__, inm);
2976 printf("addr %s ifp %p(%s) ifma %p\n",
2977 inet_ntoa(inm->inm_addr),
2979 inm->inm_ifp->if_xname,
2981 printf("timer %u state %s refcount %u scq.len %u\n",
2983 inm_state_str(inm->inm_state),
2985 inm->inm_scq.mq_len);
2986 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2991 for (t = 0; t < 2; t++) {
2992 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2993 inm_mode_str(inm->inm_st[t].iss_fmode),
2994 inm->inm_st[t].iss_asm,
2995 inm->inm_st[t].iss_ex,
2996 inm->inm_st[t].iss_in,
2997 inm->inm_st[t].iss_rec);
2999 printf("%s: --- end inm %p ---\n", __func__, inm);
3002 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3005 inm_print(const struct in_multi *inm)
3010 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3012 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);