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>
41 #include <sys/malloc.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/protosw.h>
47 #include <sys/sysctl.h>
52 #include <net/if_dl.h>
53 #include <net/route.h>
56 #include <netinet/in.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/in_pcb.h>
59 #include <netinet/in_var.h>
60 #include <netinet/ip_var.h>
61 #include <netinet/igmp_var.h>
64 #define KTR_IGMPV3 KTR_INET
67 #ifndef __SOCKUNION_DECLARED
69 struct sockaddr_storage ss;
71 struct sockaddr_dl sdl;
72 struct sockaddr_in sin;
74 typedef union sockunion sockunion_t;
75 #define __SOCKUNION_DECLARED
76 #endif /* __SOCKUNION_DECLARED */
78 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
79 "IPv4 multicast PCB-layer source filter");
80 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
81 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
82 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
83 "IPv4 multicast IGMP-layer source filter");
87 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
88 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
89 * it can be taken by code in net/if.c also.
90 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
92 * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly
93 * any need for in_multi itself to be virtualized -- it is bound to an ifp
94 * anyway no matter what happens.
96 struct mtx in_multi_mtx;
97 MTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF);
100 * Functions with non-static linkage defined in this file should be
101 * declared in in_var.h:
106 * in_joingroup_locked()
108 * in_leavegroup_locked()
114 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
117 static void imf_commit(struct in_mfilter *);
118 static int imf_get_source(struct in_mfilter *imf,
119 const struct sockaddr_in *psin,
120 struct in_msource **);
121 static struct in_msource *
122 imf_graft(struct in_mfilter *, const uint8_t,
123 const struct sockaddr_in *);
124 static void imf_leave(struct in_mfilter *);
125 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
126 static void imf_purge(struct in_mfilter *);
127 static void imf_rollback(struct in_mfilter *);
128 static void imf_reap(struct in_mfilter *);
129 static int imo_grow(struct ip_moptions *);
130 static size_t imo_match_group(const struct ip_moptions *,
131 const struct ifnet *, const struct sockaddr *);
132 static struct in_msource *
133 imo_match_source(const struct ip_moptions *, const size_t,
134 const struct sockaddr *);
135 static void ims_merge(struct ip_msource *ims,
136 const struct in_msource *lims, const int rollback);
137 static int in_getmulti(struct ifnet *, const struct in_addr *,
139 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
140 const int noalloc, struct ip_msource **pims);
142 static int inm_is_ifp_detached(const struct in_multi *);
144 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
145 static void inm_purge(struct in_multi *);
146 static void inm_reap(struct in_multi *);
147 static struct ip_moptions *
148 inp_findmoptions(struct inpcb *);
149 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
150 static int inp_join_group(struct inpcb *, struct sockopt *);
151 static int inp_leave_group(struct inpcb *, struct sockopt *);
152 static struct ifnet *
153 inp_lookup_mcast_ifp(const struct inpcb *,
154 const struct sockaddr_in *, const struct in_addr);
155 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
156 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
157 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
158 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
160 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
163 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
164 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
165 CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxgrpsrc, 0,
166 "Max source filters per group");
167 TUNABLE_ULONG("net.inet.ip.mcast.maxgrpsrc", &in_mcast_maxgrpsrc);
169 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
170 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
171 CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxsocksrc, 0,
172 "Max source filters per socket");
173 TUNABLE_ULONG("net.inet.ip.mcast.maxsocksrc", &in_mcast_maxsocksrc);
175 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
176 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
177 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
178 TUNABLE_INT("net.inet.ip.mcast.loop", &in_mcast_loop);
180 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
181 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
182 "Per-interface stack-wide source filters");
186 * Inline function which wraps assertions for a valid ifp.
187 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
191 inm_is_ifp_detached(const struct in_multi *inm)
195 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
196 ifp = inm->inm_ifma->ifma_ifp;
199 * Sanity check that netinet's notion of ifp is the
202 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
205 return (ifp == NULL);
210 * Initialize an in_mfilter structure to a known state at t0, t1
211 * with an empty source filter list.
214 imf_init(struct in_mfilter *imf, const int st0, const int st1)
216 memset(imf, 0, sizeof(struct in_mfilter));
217 RB_INIT(&imf->imf_sources);
218 imf->imf_st[0] = st0;
219 imf->imf_st[1] = st1;
223 * Resize the ip_moptions vector to the next power-of-two minus 1.
224 * May be called with locks held; do not sleep.
227 imo_grow(struct ip_moptions *imo)
229 struct in_multi **nmships;
230 struct in_multi **omships;
231 struct in_mfilter *nmfilters;
232 struct in_mfilter *omfilters;
239 omships = imo->imo_membership;
240 omfilters = imo->imo_mfilters;
241 oldmax = imo->imo_max_memberships;
242 newmax = ((oldmax + 1) * 2) - 1;
244 if (newmax <= IP_MAX_MEMBERSHIPS) {
245 nmships = (struct in_multi **)realloc(omships,
246 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
247 nmfilters = (struct in_mfilter *)realloc(omfilters,
248 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
249 if (nmships != NULL && nmfilters != NULL) {
250 /* Initialize newly allocated source filter heads. */
251 for (idx = oldmax; idx < newmax; idx++) {
252 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
255 imo->imo_max_memberships = newmax;
256 imo->imo_membership = nmships;
257 imo->imo_mfilters = nmfilters;
261 if (nmships == NULL || nmfilters == NULL) {
263 free(nmships, M_IPMOPTS);
264 if (nmfilters != NULL)
265 free(nmfilters, M_INMFILTER);
266 return (ETOOMANYREFS);
273 * Find an IPv4 multicast group entry for this ip_moptions instance
274 * which matches the specified group, and optionally an interface.
275 * Return its index into the array, or -1 if not found.
278 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
279 const struct sockaddr *group)
281 const struct sockaddr_in *gsin;
282 struct in_multi **pinm;
286 gsin = (const struct sockaddr_in *)group;
288 /* The imo_membership array may be lazy allocated. */
289 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
292 nmships = imo->imo_num_memberships;
293 pinm = &imo->imo_membership[0];
294 for (idx = 0; idx < nmships; idx++, pinm++) {
297 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
298 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
309 * Find an IPv4 multicast source entry for this imo which matches
310 * the given group index for this socket, and source address.
312 * NOTE: This does not check if the entry is in-mode, merely if
313 * it exists, which may not be the desired behaviour.
315 static struct in_msource *
316 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
317 const struct sockaddr *src)
319 struct ip_msource find;
320 struct in_mfilter *imf;
321 struct ip_msource *ims;
322 const sockunion_t *psa;
324 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
325 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
326 ("%s: invalid index %d\n", __func__, (int)gidx));
328 /* The imo_mfilters array may be lazy allocated. */
329 if (imo->imo_mfilters == NULL)
331 imf = &imo->imo_mfilters[gidx];
333 /* Source trees are keyed in host byte order. */
334 psa = (const sockunion_t *)src;
335 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
336 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
338 return ((struct in_msource *)ims);
342 * Perform filtering for multicast datagrams on a socket by group and source.
344 * Returns 0 if a datagram should be allowed through, or various error codes
345 * if the socket was not a member of the group, or the source was muted, etc.
348 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
349 const struct sockaddr *group, const struct sockaddr *src)
352 struct in_msource *ims;
355 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
357 gidx = imo_match_group(imo, ifp, group);
359 return (MCAST_NOTGMEMBER);
362 * Check if the source was included in an (S,G) join.
363 * Allow reception on exclusive memberships by default,
364 * reject reception on inclusive memberships by default.
365 * Exclude source only if an in-mode exclude filter exists.
366 * Include source only if an in-mode include filter exists.
367 * NOTE: We are comparing group state here at IGMP t1 (now)
368 * with socket-layer t0 (since last downcall).
370 mode = imo->imo_mfilters[gidx].imf_st[1];
371 ims = imo_match_source(imo, gidx, src);
373 if ((ims == NULL && mode == MCAST_INCLUDE) ||
374 (ims != NULL && ims->imsl_st[0] != mode))
375 return (MCAST_NOTSMEMBER);
381 * Find and return a reference to an in_multi record for (ifp, group),
382 * and bump its reference count.
383 * If one does not exist, try to allocate it, and update link-layer multicast
384 * filters on ifp to listen for group.
385 * Assumes the IN_MULTI lock is held across the call.
386 * Return 0 if successful, otherwise return an appropriate error code.
389 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
390 struct in_multi **pinm)
392 struct sockaddr_in gsin;
393 struct ifmultiaddr *ifma;
394 struct in_ifinfo *ii;
395 struct in_multi *inm;
398 IN_MULTI_LOCK_ASSERT();
400 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
402 inm = inm_lookup(ifp, *group);
405 * If we already joined this group, just bump the
406 * refcount and return it.
408 KASSERT(inm->inm_refcount >= 1,
409 ("%s: bad refcount %d", __func__, inm->inm_refcount));
415 memset(&gsin, 0, sizeof(gsin));
416 gsin.sin_family = AF_INET;
417 gsin.sin_len = sizeof(struct sockaddr_in);
418 gsin.sin_addr = *group;
421 * Check if a link-layer group is already associated
422 * with this network-layer group on the given ifnet.
424 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
428 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
432 * If something other than netinet is occupying the link-layer
433 * group, print a meaningful error message and back out of
435 * Otherwise, bump the refcount on the existing network-layer
436 * group association and return it.
438 if (ifma->ifma_protospec != NULL) {
439 inm = (struct in_multi *)ifma->ifma_protospec;
441 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
443 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
444 ("%s: ifma not AF_INET", __func__));
445 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
446 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
447 !in_hosteq(inm->inm_addr, *group))
448 panic("%s: ifma %p is inconsistent with %p (%s)",
449 __func__, ifma, inm, inet_ntoa(*group));
453 IF_ADDR_WUNLOCK(ifp);
457 IF_ADDR_WLOCK_ASSERT(ifp);
460 * A new in_multi record is needed; allocate and initialize it.
461 * We DO NOT perform an IGMP join as the in_ layer may need to
462 * push an initial source list down to IGMP to support SSM.
464 * The initial source filter state is INCLUDE, {} as per the RFC.
466 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
468 if_delmulti_ifma(ifma);
469 IF_ADDR_WUNLOCK(ifp);
472 inm->inm_addr = *group;
474 inm->inm_igi = ii->ii_igmp;
475 inm->inm_ifma = ifma;
476 inm->inm_refcount = 1;
477 inm->inm_state = IGMP_NOT_MEMBER;
480 * Pending state-changes per group are subject to a bounds check.
482 IFQ_SET_MAXLEN(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
484 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
485 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
486 RB_INIT(&inm->inm_srcs);
488 ifma->ifma_protospec = inm;
492 IF_ADDR_WUNLOCK(ifp);
497 * Drop a reference to an in_multi record.
499 * If the refcount drops to 0, free the in_multi record and
500 * delete the underlying link-layer membership.
503 inm_release_locked(struct in_multi *inm)
505 struct ifmultiaddr *ifma;
507 IN_MULTI_LOCK_ASSERT();
509 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
511 if (--inm->inm_refcount > 0) {
512 CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__,
517 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
519 ifma = inm->inm_ifma;
521 /* XXX this access is not covered by IF_ADDR_LOCK */
522 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
523 KASSERT(ifma->ifma_protospec == inm,
524 ("%s: ifma_protospec != inm", __func__));
525 ifma->ifma_protospec = NULL;
529 free(inm, M_IPMADDR);
531 if_delmulti_ifma(ifma);
535 * Clear recorded source entries for a group.
536 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
537 * FIXME: Should reap.
540 inm_clear_recorded(struct in_multi *inm)
542 struct ip_msource *ims;
544 IN_MULTI_LOCK_ASSERT();
546 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
549 --inm->inm_st[1].iss_rec;
552 KASSERT(inm->inm_st[1].iss_rec == 0,
553 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
557 * Record a source as pending for a Source-Group IGMPv3 query.
558 * This lives here as it modifies the shared tree.
560 * inm is the group descriptor.
561 * naddr is the address of the source to record in network-byte order.
563 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
564 * lazy-allocate a source node in response to an SG query.
565 * Otherwise, no allocation is performed. This saves some memory
566 * with the trade-off that the source will not be reported to the
567 * router if joined in the window between the query response and
568 * the group actually being joined on the local host.
570 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
571 * This turns off the allocation of a recorded source entry if
572 * the group has not been joined.
574 * Return 0 if the source didn't exist or was already marked as recorded.
575 * Return 1 if the source was marked as recorded by this function.
576 * Return <0 if any error occured (negated errno code).
579 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
581 struct ip_msource find;
582 struct ip_msource *ims, *nims;
584 IN_MULTI_LOCK_ASSERT();
586 find.ims_haddr = ntohl(naddr);
587 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
588 if (ims && ims->ims_stp)
591 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
593 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
597 nims->ims_haddr = find.ims_haddr;
598 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
604 * Mark the source as recorded and update the recorded
608 ++inm->inm_st[1].iss_rec;
614 * Return a pointer to an in_msource owned by an in_mfilter,
615 * given its source address.
616 * Lazy-allocate if needed. If this is a new entry its filter state is
619 * imf is the filter set being modified.
620 * haddr is the source address in *host* byte-order.
622 * SMPng: May be called with locks held; malloc must not block.
625 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
626 struct in_msource **plims)
628 struct ip_msource find;
629 struct ip_msource *ims, *nims;
630 struct in_msource *lims;
637 /* key is host byte order */
638 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
639 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
640 lims = (struct in_msource *)ims;
642 if (imf->imf_nsrc == in_mcast_maxsocksrc)
644 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
648 lims = (struct in_msource *)nims;
649 lims->ims_haddr = find.ims_haddr;
650 lims->imsl_st[0] = MCAST_UNDEFINED;
651 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
661 * Graft a source entry into an existing socket-layer filter set,
662 * maintaining any required invariants and checking allocations.
664 * The source is marked as being in the new filter mode at t1.
666 * Return the pointer to the new node, otherwise return NULL.
668 static struct in_msource *
669 imf_graft(struct in_mfilter *imf, const uint8_t st1,
670 const struct sockaddr_in *psin)
672 struct ip_msource *nims;
673 struct in_msource *lims;
675 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
679 lims = (struct in_msource *)nims;
680 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
681 lims->imsl_st[0] = MCAST_UNDEFINED;
682 lims->imsl_st[1] = st1;
683 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
690 * Prune a source entry from an existing socket-layer filter set,
691 * maintaining any required invariants and checking allocations.
693 * The source is marked as being left at t1, it is not freed.
695 * Return 0 if no error occurred, otherwise return an errno value.
698 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
700 struct ip_msource find;
701 struct ip_msource *ims;
702 struct in_msource *lims;
704 /* key is host byte order */
705 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
706 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
709 lims = (struct in_msource *)ims;
710 lims->imsl_st[1] = MCAST_UNDEFINED;
715 * Revert socket-layer filter set deltas at t1 to t0 state.
718 imf_rollback(struct in_mfilter *imf)
720 struct ip_msource *ims, *tims;
721 struct in_msource *lims;
723 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
724 lims = (struct in_msource *)ims;
725 if (lims->imsl_st[0] == lims->imsl_st[1]) {
726 /* no change at t1 */
728 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
729 /* revert change to existing source at t1 */
730 lims->imsl_st[1] = lims->imsl_st[0];
732 /* revert source added t1 */
733 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
734 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
735 free(ims, M_INMFILTER);
739 imf->imf_st[1] = imf->imf_st[0];
743 * Mark socket-layer filter set as INCLUDE {} at t1.
746 imf_leave(struct in_mfilter *imf)
748 struct ip_msource *ims;
749 struct in_msource *lims;
751 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
752 lims = (struct in_msource *)ims;
753 lims->imsl_st[1] = MCAST_UNDEFINED;
755 imf->imf_st[1] = MCAST_INCLUDE;
759 * Mark socket-layer filter set deltas as committed.
762 imf_commit(struct in_mfilter *imf)
764 struct ip_msource *ims;
765 struct in_msource *lims;
767 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
768 lims = (struct in_msource *)ims;
769 lims->imsl_st[0] = lims->imsl_st[1];
771 imf->imf_st[0] = imf->imf_st[1];
775 * Reap unreferenced sources from socket-layer filter set.
778 imf_reap(struct in_mfilter *imf)
780 struct ip_msource *ims, *tims;
781 struct in_msource *lims;
783 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
784 lims = (struct in_msource *)ims;
785 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
786 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
787 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
788 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
789 free(ims, M_INMFILTER);
796 * Purge socket-layer filter set.
799 imf_purge(struct in_mfilter *imf)
801 struct ip_msource *ims, *tims;
803 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
804 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
805 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
806 free(ims, M_INMFILTER);
809 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
810 KASSERT(RB_EMPTY(&imf->imf_sources),
811 ("%s: imf_sources not empty", __func__));
815 * Look up a source filter entry for a multicast group.
817 * inm is the group descriptor to work with.
818 * haddr is the host-byte-order IPv4 address to look up.
819 * noalloc may be non-zero to suppress allocation of sources.
820 * *pims will be set to the address of the retrieved or allocated source.
822 * SMPng: NOTE: may be called with locks held.
823 * Return 0 if successful, otherwise return a non-zero error code.
826 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
827 const int noalloc, struct ip_msource **pims)
829 struct ip_msource find;
830 struct ip_msource *ims, *nims;
835 find.ims_haddr = haddr;
836 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
837 if (ims == NULL && !noalloc) {
838 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
840 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
844 nims->ims_haddr = haddr;
845 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
849 ia.s_addr = htonl(haddr);
850 CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__,
860 * Merge socket-layer source into IGMP-layer source.
861 * If rollback is non-zero, perform the inverse of the merge.
864 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
867 int n = rollback ? -1 : 1;
871 ia.s_addr = htonl(ims->ims_haddr);
874 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
875 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s",
876 __func__, n, inet_ntoa(ia));
877 ims->ims_st[1].ex -= n;
878 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
879 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s",
880 __func__, n, inet_ntoa(ia));
881 ims->ims_st[1].in -= n;
884 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
885 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s",
886 __func__, n, inet_ntoa(ia));
887 ims->ims_st[1].ex += n;
888 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
889 CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s",
890 __func__, n, inet_ntoa(ia));
891 ims->ims_st[1].in += n;
896 * Atomically update the global in_multi state, when a membership's
897 * filter list is being updated in any way.
899 * imf is the per-inpcb-membership group filter pointer.
900 * A fake imf may be passed for in-kernel consumers.
902 * XXX This is a candidate for a set-symmetric-difference style loop
903 * which would eliminate the repeated lookup from root of ims nodes,
904 * as they share the same key space.
906 * If any error occurred this function will back out of refcounts
907 * and return a non-zero value.
910 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
912 struct ip_msource *ims, *nims;
913 struct in_msource *lims;
922 * Update the source filters first, as this may fail.
923 * Maintain count of in-mode filters at t0, t1. These are
924 * used to work out if we transition into ASM mode or not.
925 * Maintain a count of source filters whose state was
926 * actually modified by this operation.
928 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
929 lims = (struct in_msource *)ims;
930 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
931 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
932 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
933 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
937 ims_merge(nims, lims, 0);
940 struct ip_msource *bims;
942 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
943 lims = (struct in_msource *)ims;
944 if (lims->imsl_st[0] == lims->imsl_st[1])
946 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
949 ims_merge(bims, lims, 1);
954 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
955 __func__, nsrc0, nsrc1);
957 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
958 if (imf->imf_st[0] == imf->imf_st[1] &&
959 imf->imf_st[1] == MCAST_INCLUDE) {
961 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
962 --inm->inm_st[1].iss_in;
966 /* Handle filter mode transition on socket. */
967 if (imf->imf_st[0] != imf->imf_st[1]) {
968 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
969 __func__, imf->imf_st[0], imf->imf_st[1]);
971 if (imf->imf_st[0] == MCAST_EXCLUDE) {
972 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
973 --inm->inm_st[1].iss_ex;
974 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
975 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
976 --inm->inm_st[1].iss_in;
979 if (imf->imf_st[1] == MCAST_EXCLUDE) {
980 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
981 inm->inm_st[1].iss_ex++;
982 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
983 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
984 inm->inm_st[1].iss_in++;
989 * Track inm filter state in terms of listener counts.
990 * If there are any exclusive listeners, stack-wide
991 * membership is exclusive.
992 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
993 * If no listeners remain, state is undefined at t1,
994 * and the IGMP lifecycle for this group should finish.
996 if (inm->inm_st[1].iss_ex > 0) {
997 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
998 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
999 } else if (inm->inm_st[1].iss_in > 0) {
1000 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1001 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1003 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1004 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1007 /* Decrement ASM listener count on transition out of ASM mode. */
1008 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1009 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1010 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1011 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1012 --inm->inm_st[1].iss_asm;
1015 /* Increment ASM listener count on transition to ASM mode. */
1016 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1017 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1018 inm->inm_st[1].iss_asm++;
1021 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1026 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1033 * Mark an in_multi's filter set deltas as committed.
1034 * Called by IGMP after a state change has been enqueued.
1037 inm_commit(struct in_multi *inm)
1039 struct ip_msource *ims;
1041 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1042 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1045 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1046 ims->ims_st[0] = ims->ims_st[1];
1048 inm->inm_st[0] = inm->inm_st[1];
1052 * Reap unreferenced nodes from an in_multi's filter set.
1055 inm_reap(struct in_multi *inm)
1057 struct ip_msource *ims, *tims;
1059 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1060 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1061 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1064 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1065 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1066 free(ims, M_IPMSOURCE);
1072 * Purge all source nodes from an in_multi's filter set.
1075 inm_purge(struct in_multi *inm)
1077 struct ip_msource *ims, *tims;
1079 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1080 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1081 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1082 free(ims, M_IPMSOURCE);
1088 * Join a multicast group; unlocked entry point.
1090 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1091 * is not held. Fortunately, ifp is unlikely to have been detached
1092 * at this point, so we assume it's OK to recurse.
1095 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1096 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1101 error = in_joingroup_locked(ifp, gina, imf, pinm);
1108 * Join a multicast group; real entry point.
1110 * Only preserves atomicity at inm level.
1111 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1113 * If the IGMP downcall fails, the group is not joined, and an error
1117 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1118 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1120 struct in_mfilter timf;
1121 struct in_multi *inm;
1124 IN_MULTI_LOCK_ASSERT();
1126 CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__,
1127 inet_ntoa(*gina), ifp, ifp->if_xname);
1133 * If no imf was specified (i.e. kernel consumer),
1134 * fake one up and assume it is an ASM join.
1137 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1141 error = in_getmulti(ifp, gina, &inm);
1143 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1147 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1148 error = inm_merge(inm, imf);
1150 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1151 goto out_inm_release;
1154 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1155 error = igmp_change_state(inm);
1157 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1158 goto out_inm_release;
1163 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1164 inm_release_locked(inm);
1173 * Leave a multicast group; unlocked entry point.
1176 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1184 error = in_leavegroup_locked(inm, imf);
1191 * Leave a multicast group; real entry point.
1192 * All source filters will be expunged.
1194 * Only preserves atomicity at inm level.
1196 * Holding the write lock for the INP which contains imf
1197 * is highly advisable. We can't assert for it as imf does not
1198 * contain a back-pointer to the owning inp.
1200 * Note: This is not the same as inm_release(*) as this function also
1201 * makes a state change downcall into IGMP.
1204 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1206 struct in_mfilter timf;
1211 IN_MULTI_LOCK_ASSERT();
1213 CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__,
1214 inm, inet_ntoa(inm->inm_addr),
1215 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1219 * If no imf was specified (i.e. kernel consumer),
1220 * fake one up and assume it is an ASM join.
1223 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1228 * Begin state merge transaction at IGMP layer.
1230 * As this particular invocation should not cause any memory
1231 * to be allocated, and there is no opportunity to roll back
1232 * the transaction, it MUST NOT fail.
1234 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1235 error = inm_merge(inm, imf);
1236 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1238 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1239 error = igmp_change_state(inm);
1241 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1243 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1244 inm_release_locked(inm);
1249 /*#ifndef BURN_BRIDGES*/
1251 * Join an IPv4 multicast group in (*,G) exclusive mode.
1252 * The group must be a 224.0.0.0/24 link-scope group.
1253 * This KPI is for legacy kernel consumers only.
1256 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1258 struct in_multi *pinm;
1261 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1262 ("%s: %s not in 224.0.0.0/24", __func__, inet_ntoa(*ap)));
1264 error = in_joingroup(ifp, ap, NULL, &pinm);
1272 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1273 * This KPI is for legacy kernel consumers only.
1276 in_delmulti(struct in_multi *inm)
1279 (void)in_leavegroup(inm, NULL);
1284 * Block or unblock an ASM multicast source on an inpcb.
1285 * This implements the delta-based API described in RFC 3678.
1287 * The delta-based API applies only to exclusive-mode memberships.
1288 * An IGMP downcall will be performed.
1290 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1292 * Return 0 if successful, otherwise return an appropriate error code.
1295 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1297 struct group_source_req gsr;
1298 sockunion_t *gsa, *ssa;
1300 struct in_mfilter *imf;
1301 struct ip_moptions *imo;
1302 struct in_msource *ims;
1303 struct in_multi *inm;
1312 memset(&gsr, 0, sizeof(struct group_source_req));
1313 gsa = (sockunion_t *)&gsr.gsr_group;
1314 ssa = (sockunion_t *)&gsr.gsr_source;
1316 switch (sopt->sopt_name) {
1317 case IP_BLOCK_SOURCE:
1318 case IP_UNBLOCK_SOURCE: {
1319 struct ip_mreq_source mreqs;
1321 error = sooptcopyin(sopt, &mreqs,
1322 sizeof(struct ip_mreq_source),
1323 sizeof(struct ip_mreq_source));
1327 gsa->sin.sin_family = AF_INET;
1328 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1329 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1331 ssa->sin.sin_family = AF_INET;
1332 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1333 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1335 if (!in_nullhost(mreqs.imr_interface))
1336 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1338 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1341 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1342 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1346 case MCAST_BLOCK_SOURCE:
1347 case MCAST_UNBLOCK_SOURCE:
1348 error = sooptcopyin(sopt, &gsr,
1349 sizeof(struct group_source_req),
1350 sizeof(struct group_source_req));
1354 if (gsa->sin.sin_family != AF_INET ||
1355 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1358 if (ssa->sin.sin_family != AF_INET ||
1359 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1362 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1363 return (EADDRNOTAVAIL);
1365 ifp = ifnet_byindex(gsr.gsr_interface);
1367 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1372 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1373 __func__, sopt->sopt_name);
1374 return (EOPNOTSUPP);
1378 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1382 * Check if we are actually a member of this group.
1384 imo = inp_findmoptions(inp);
1385 idx = imo_match_group(imo, ifp, &gsa->sa);
1386 if (idx == -1 || imo->imo_mfilters == NULL) {
1387 error = EADDRNOTAVAIL;
1388 goto out_inp_locked;
1391 KASSERT(imo->imo_mfilters != NULL,
1392 ("%s: imo_mfilters not allocated", __func__));
1393 imf = &imo->imo_mfilters[idx];
1394 inm = imo->imo_membership[idx];
1397 * Attempting to use the delta-based API on an
1398 * non exclusive-mode membership is an error.
1400 fmode = imf->imf_st[0];
1401 if (fmode != MCAST_EXCLUDE) {
1403 goto out_inp_locked;
1407 * Deal with error cases up-front:
1408 * Asked to block, but already blocked; or
1409 * Asked to unblock, but nothing to unblock.
1410 * If adding a new block entry, allocate it.
1412 ims = imo_match_source(imo, idx, &ssa->sa);
1413 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1414 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
1415 inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not ");
1416 error = EADDRNOTAVAIL;
1417 goto out_inp_locked;
1420 INP_WLOCK_ASSERT(inp);
1423 * Begin state merge transaction at socket layer.
1426 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1427 ims = imf_graft(imf, fmode, &ssa->sin);
1431 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1432 error = imf_prune(imf, &ssa->sin);
1436 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1437 goto out_imf_rollback;
1441 * Begin state merge transaction at IGMP layer.
1445 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1446 error = inm_merge(inm, imf);
1448 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1449 goto out_imf_rollback;
1452 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1453 error = igmp_change_state(inm);
1455 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1473 * Given an inpcb, return its multicast options structure pointer. Accepts
1474 * an unlocked inpcb pointer, but will return it locked. May sleep.
1476 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1477 * SMPng: NOTE: Returns with the INP write lock held.
1479 static struct ip_moptions *
1480 inp_findmoptions(struct inpcb *inp)
1482 struct ip_moptions *imo;
1483 struct in_multi **immp;
1484 struct in_mfilter *imfp;
1488 if (inp->inp_moptions != NULL)
1489 return (inp->inp_moptions);
1493 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1494 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1496 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1497 M_INMFILTER, M_WAITOK);
1499 imo->imo_multicast_ifp = NULL;
1500 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1501 imo->imo_multicast_vif = -1;
1502 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1503 imo->imo_multicast_loop = in_mcast_loop;
1504 imo->imo_num_memberships = 0;
1505 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1506 imo->imo_membership = immp;
1508 /* Initialize per-group source filters. */
1509 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1510 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1511 imo->imo_mfilters = imfp;
1514 if (inp->inp_moptions != NULL) {
1515 free(imfp, M_INMFILTER);
1516 free(immp, M_IPMOPTS);
1517 free(imo, M_IPMOPTS);
1518 return (inp->inp_moptions);
1520 inp->inp_moptions = imo;
1525 * Discard the IP multicast options (and source filters).
1527 * SMPng: NOTE: assumes INP write lock is held.
1530 inp_freemoptions(struct ip_moptions *imo)
1532 struct in_mfilter *imf;
1533 size_t idx, nmships;
1535 KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1537 nmships = imo->imo_num_memberships;
1538 for (idx = 0; idx < nmships; ++idx) {
1539 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1542 (void)in_leavegroup(imo->imo_membership[idx], imf);
1547 if (imo->imo_mfilters)
1548 free(imo->imo_mfilters, M_INMFILTER);
1549 free(imo->imo_membership, M_IPMOPTS);
1550 free(imo, M_IPMOPTS);
1554 * Atomically get source filters on a socket for an IPv4 multicast group.
1555 * Called with INP lock held; returns with lock released.
1558 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1560 struct __msfilterreq msfr;
1563 struct ip_moptions *imo;
1564 struct in_mfilter *imf;
1565 struct ip_msource *ims;
1566 struct in_msource *lims;
1567 struct sockaddr_in *psin;
1568 struct sockaddr_storage *ptss;
1569 struct sockaddr_storage *tss;
1571 size_t idx, nsrcs, ncsrcs;
1573 INP_WLOCK_ASSERT(inp);
1575 imo = inp->inp_moptions;
1576 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1580 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1581 sizeof(struct __msfilterreq));
1585 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1588 ifp = ifnet_byindex(msfr.msfr_ifindex);
1595 * Lookup group on the socket.
1597 gsa = (sockunion_t *)&msfr.msfr_group;
1598 idx = imo_match_group(imo, ifp, &gsa->sa);
1599 if (idx == -1 || imo->imo_mfilters == NULL) {
1601 return (EADDRNOTAVAIL);
1603 imf = &imo->imo_mfilters[idx];
1606 * Ignore memberships which are in limbo.
1608 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1612 msfr.msfr_fmode = imf->imf_st[1];
1615 * If the user specified a buffer, copy out the source filter
1616 * entries to userland gracefully.
1617 * We only copy out the number of entries which userland
1618 * has asked for, but we always tell userland how big the
1619 * buffer really needs to be.
1621 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1622 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1624 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1625 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1626 M_TEMP, M_NOWAIT | M_ZERO);
1634 * Count number of sources in-mode at t0.
1635 * If buffer space exists and remains, copy out source entries.
1637 nsrcs = msfr.msfr_nsrcs;
1640 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1641 lims = (struct in_msource *)ims;
1642 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1643 lims->imsl_st[0] != imf->imf_st[0])
1646 if (tss != NULL && nsrcs > 0) {
1647 psin = (struct sockaddr_in *)ptss;
1648 psin->sin_family = AF_INET;
1649 psin->sin_len = sizeof(struct sockaddr_in);
1650 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1660 error = copyout(tss, msfr.msfr_srcs,
1661 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1667 msfr.msfr_nsrcs = ncsrcs;
1668 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1674 * Return the IP multicast options in response to user getsockopt().
1677 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1679 struct ip_mreqn mreqn;
1680 struct ip_moptions *imo;
1682 struct in_ifaddr *ia;
1687 imo = inp->inp_moptions;
1689 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1690 * or is a divert socket, reject it.
1692 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1693 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1694 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1696 return (EOPNOTSUPP);
1700 switch (sopt->sopt_name) {
1701 case IP_MULTICAST_VIF:
1703 optval = imo->imo_multicast_vif;
1707 error = sooptcopyout(sopt, &optval, sizeof(int));
1710 case IP_MULTICAST_IF:
1711 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1713 ifp = imo->imo_multicast_ifp;
1714 if (!in_nullhost(imo->imo_multicast_addr)) {
1715 mreqn.imr_address = imo->imo_multicast_addr;
1716 } else if (ifp != NULL) {
1717 mreqn.imr_ifindex = ifp->if_index;
1721 IA_SIN(ia)->sin_addr;
1722 ifa_free(&ia->ia_ifa);
1727 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1728 error = sooptcopyout(sopt, &mreqn,
1729 sizeof(struct ip_mreqn));
1731 error = sooptcopyout(sopt, &mreqn.imr_address,
1732 sizeof(struct in_addr));
1736 case IP_MULTICAST_TTL:
1738 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1740 optval = coptval = imo->imo_multicast_ttl;
1742 if (sopt->sopt_valsize == sizeof(u_char))
1743 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1745 error = sooptcopyout(sopt, &optval, sizeof(int));
1748 case IP_MULTICAST_LOOP:
1750 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1752 optval = coptval = imo->imo_multicast_loop;
1754 if (sopt->sopt_valsize == sizeof(u_char))
1755 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1757 error = sooptcopyout(sopt, &optval, sizeof(int));
1762 error = EADDRNOTAVAIL;
1765 error = inp_get_source_filters(inp, sopt);
1771 error = ENOPROTOOPT;
1775 INP_UNLOCK_ASSERT(inp);
1781 * Look up the ifnet to use for a multicast group membership,
1782 * given the IPv4 address of an interface, and the IPv4 group address.
1784 * This routine exists to support legacy multicast applications
1785 * which do not understand that multicast memberships are scoped to
1786 * specific physical links in the networking stack, or which need
1787 * to join link-scope groups before IPv4 addresses are configured.
1789 * If inp is non-NULL, use this socket's current FIB number for any
1790 * required FIB lookup.
1791 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1792 * and use its ifp; usually, this points to the default next-hop.
1794 * If the FIB lookup fails, attempt to use the first non-loopback
1795 * interface with multicast capability in the system as a
1796 * last resort. The legacy IPv4 ASM API requires that we do
1797 * this in order to allow groups to be joined when the routing
1798 * table has not yet been populated during boot.
1800 * Returns NULL if no ifp could be found.
1802 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1803 * FUTURE: Implement IPv4 source-address selection.
1805 static struct ifnet *
1806 inp_lookup_mcast_ifp(const struct inpcb *inp,
1807 const struct sockaddr_in *gsin, const struct in_addr ina)
1811 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1812 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1813 ("%s: not multicast", __func__));
1816 if (!in_nullhost(ina)) {
1817 INADDR_TO_IFP(ina, ifp);
1822 memcpy(&ro.ro_dst, gsin, sizeof(struct sockaddr_in));
1823 in_rtalloc_ign(&ro, 0, inp ? inp->inp_inc.inc_fibnum : 0);
1824 if (ro.ro_rt != NULL) {
1825 ifp = ro.ro_rt->rt_ifp;
1826 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1829 struct in_ifaddr *ia;
1834 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1836 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1837 (mifp->if_flags & IFF_MULTICAST)) {
1842 IN_IFADDR_RUNLOCK();
1850 * Join an IPv4 multicast group, possibly with a source.
1853 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1855 struct group_source_req gsr;
1856 sockunion_t *gsa, *ssa;
1858 struct in_mfilter *imf;
1859 struct ip_moptions *imo;
1860 struct in_multi *inm;
1861 struct in_msource *lims;
1871 memset(&gsr, 0, sizeof(struct group_source_req));
1872 gsa = (sockunion_t *)&gsr.gsr_group;
1873 gsa->ss.ss_family = AF_UNSPEC;
1874 ssa = (sockunion_t *)&gsr.gsr_source;
1875 ssa->ss.ss_family = AF_UNSPEC;
1877 switch (sopt->sopt_name) {
1878 case IP_ADD_MEMBERSHIP:
1879 case IP_ADD_SOURCE_MEMBERSHIP: {
1880 struct ip_mreq_source mreqs;
1882 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1883 error = sooptcopyin(sopt, &mreqs,
1884 sizeof(struct ip_mreq),
1885 sizeof(struct ip_mreq));
1887 * Do argument switcharoo from ip_mreq into
1888 * ip_mreq_source to avoid using two instances.
1890 mreqs.imr_interface = mreqs.imr_sourceaddr;
1891 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1892 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1893 error = sooptcopyin(sopt, &mreqs,
1894 sizeof(struct ip_mreq_source),
1895 sizeof(struct ip_mreq_source));
1900 gsa->sin.sin_family = AF_INET;
1901 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1902 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1904 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1905 ssa->sin.sin_family = AF_INET;
1906 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1907 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1910 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1913 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1914 mreqs.imr_interface);
1915 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1916 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1920 case MCAST_JOIN_GROUP:
1921 case MCAST_JOIN_SOURCE_GROUP:
1922 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1923 error = sooptcopyin(sopt, &gsr,
1924 sizeof(struct group_req),
1925 sizeof(struct group_req));
1926 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1927 error = sooptcopyin(sopt, &gsr,
1928 sizeof(struct group_source_req),
1929 sizeof(struct group_source_req));
1934 if (gsa->sin.sin_family != AF_INET ||
1935 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1939 * Overwrite the port field if present, as the sockaddr
1940 * being copied in may be matched with a binary comparison.
1942 gsa->sin.sin_port = 0;
1943 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1944 if (ssa->sin.sin_family != AF_INET ||
1945 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1947 ssa->sin.sin_port = 0;
1950 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1953 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1954 return (EADDRNOTAVAIL);
1955 ifp = ifnet_byindex(gsr.gsr_interface);
1959 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1960 __func__, sopt->sopt_name);
1961 return (EOPNOTSUPP);
1965 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1966 return (EADDRNOTAVAIL);
1968 imo = inp_findmoptions(inp);
1969 idx = imo_match_group(imo, ifp, &gsa->sa);
1973 inm = imo->imo_membership[idx];
1974 imf = &imo->imo_mfilters[idx];
1975 if (ssa->ss.ss_family != AF_UNSPEC) {
1977 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1978 * is an error. On an existing inclusive membership,
1979 * it just adds the source to the filter list.
1981 if (imf->imf_st[1] != MCAST_INCLUDE) {
1983 goto out_inp_locked;
1986 * Throw out duplicates.
1988 * XXX FIXME: This makes a naive assumption that
1989 * even if entries exist for *ssa in this imf,
1990 * they will be rejected as dupes, even if they
1991 * are not valid in the current mode (in-mode).
1993 * in_msource is transactioned just as for anything
1994 * else in SSM -- but note naive use of inm_graft()
1995 * below for allocating new filter entries.
1997 * This is only an issue if someone mixes the
1998 * full-state SSM API with the delta-based API,
1999 * which is discouraged in the relevant RFCs.
2001 lims = imo_match_source(imo, idx, &ssa->sa);
2002 if (lims != NULL /*&&
2003 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2004 error = EADDRNOTAVAIL;
2005 goto out_inp_locked;
2009 * MCAST_JOIN_GROUP on an existing exclusive
2010 * membership is an error; return EADDRINUSE
2011 * to preserve 4.4BSD API idempotence, and
2012 * avoid tedious detour to code below.
2013 * NOTE: This is bending RFC 3678 a bit.
2015 * On an existing inclusive membership, this is also
2016 * an error; if you want to change filter mode,
2017 * you must use the userland API setsourcefilter().
2018 * XXX We don't reject this for imf in UNDEFINED
2019 * state at t1, because allocation of a filter
2020 * is atomic with allocation of a membership.
2023 if (imf->imf_st[1] == MCAST_EXCLUDE)
2025 goto out_inp_locked;
2030 * Begin state merge transaction at socket layer.
2032 INP_WLOCK_ASSERT(inp);
2035 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2036 error = imo_grow(imo);
2038 goto out_inp_locked;
2041 * Allocate the new slot upfront so we can deal with
2042 * grafting the new source filter in same code path
2043 * as for join-source on existing membership.
2045 idx = imo->imo_num_memberships;
2046 imo->imo_membership[idx] = NULL;
2047 imo->imo_num_memberships++;
2048 KASSERT(imo->imo_mfilters != NULL,
2049 ("%s: imf_mfilters vector was not allocated", __func__));
2050 imf = &imo->imo_mfilters[idx];
2051 KASSERT(RB_EMPTY(&imf->imf_sources),
2052 ("%s: imf_sources not empty", __func__));
2056 * Graft new source into filter list for this inpcb's
2057 * membership of the group. The in_multi may not have
2058 * been allocated yet if this is a new membership, however,
2059 * the in_mfilter slot will be allocated and must be initialized.
2061 * Note: Grafting of exclusive mode filters doesn't happen
2063 * XXX: Should check for non-NULL lims (node exists but may
2064 * not be in-mode) for interop with full-state API.
2066 if (ssa->ss.ss_family != AF_UNSPEC) {
2067 /* Membership starts in IN mode */
2069 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2070 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2072 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2074 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2076 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2082 /* No address specified; Membership starts in EX mode */
2084 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2085 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2090 * Begin state merge transaction at IGMP layer.
2095 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2099 imo->imo_membership[idx] = inm;
2101 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2102 error = inm_merge(inm, imf);
2104 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2106 goto out_imf_rollback;
2108 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2109 error = igmp_change_state(inm);
2111 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2113 goto out_imf_rollback;
2120 INP_WLOCK_ASSERT(inp);
2132 if (error && is_new) {
2133 imo->imo_membership[idx] = NULL;
2134 --imo->imo_num_memberships;
2143 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2146 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2148 struct group_source_req gsr;
2149 struct ip_mreq_source mreqs;
2150 sockunion_t *gsa, *ssa;
2152 struct in_mfilter *imf;
2153 struct ip_moptions *imo;
2154 struct in_msource *ims;
2155 struct in_multi *inm;
2157 int error, is_final;
2163 memset(&gsr, 0, sizeof(struct group_source_req));
2164 gsa = (sockunion_t *)&gsr.gsr_group;
2165 gsa->ss.ss_family = AF_UNSPEC;
2166 ssa = (sockunion_t *)&gsr.gsr_source;
2167 ssa->ss.ss_family = AF_UNSPEC;
2169 switch (sopt->sopt_name) {
2170 case IP_DROP_MEMBERSHIP:
2171 case IP_DROP_SOURCE_MEMBERSHIP:
2172 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2173 error = sooptcopyin(sopt, &mreqs,
2174 sizeof(struct ip_mreq),
2175 sizeof(struct ip_mreq));
2177 * Swap interface and sourceaddr arguments,
2178 * as ip_mreq and ip_mreq_source are laid
2181 mreqs.imr_interface = mreqs.imr_sourceaddr;
2182 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2183 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2184 error = sooptcopyin(sopt, &mreqs,
2185 sizeof(struct ip_mreq_source),
2186 sizeof(struct ip_mreq_source));
2191 gsa->sin.sin_family = AF_INET;
2192 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2193 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2195 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2196 ssa->sin.sin_family = AF_INET;
2197 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2198 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2202 * Attempt to look up hinted ifp from interface address.
2203 * Fallthrough with null ifp iff lookup fails, to
2204 * preserve 4.4BSD mcast API idempotence.
2205 * XXX NOTE WELL: The RFC 3678 API is preferred because
2206 * using an IPv4 address as a key is racy.
2208 if (!in_nullhost(mreqs.imr_interface))
2209 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2211 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
2212 __func__, inet_ntoa(mreqs.imr_interface), ifp);
2216 case MCAST_LEAVE_GROUP:
2217 case MCAST_LEAVE_SOURCE_GROUP:
2218 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2219 error = sooptcopyin(sopt, &gsr,
2220 sizeof(struct group_req),
2221 sizeof(struct group_req));
2222 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2223 error = sooptcopyin(sopt, &gsr,
2224 sizeof(struct group_source_req),
2225 sizeof(struct group_source_req));
2230 if (gsa->sin.sin_family != AF_INET ||
2231 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2234 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2235 if (ssa->sin.sin_family != AF_INET ||
2236 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2240 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2241 return (EADDRNOTAVAIL);
2243 ifp = ifnet_byindex(gsr.gsr_interface);
2246 return (EADDRNOTAVAIL);
2250 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2251 __func__, sopt->sopt_name);
2252 return (EOPNOTSUPP);
2256 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2260 * Find the membership in the membership array.
2262 imo = inp_findmoptions(inp);
2263 idx = imo_match_group(imo, ifp, &gsa->sa);
2265 error = EADDRNOTAVAIL;
2266 goto out_inp_locked;
2268 inm = imo->imo_membership[idx];
2269 imf = &imo->imo_mfilters[idx];
2271 if (ssa->ss.ss_family != AF_UNSPEC)
2275 * Begin state merge transaction at socket layer.
2277 INP_WLOCK_ASSERT(inp);
2280 * If we were instructed only to leave a given source, do so.
2281 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2286 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2287 error = EADDRNOTAVAIL;
2288 goto out_inp_locked;
2290 ims = imo_match_source(imo, idx, &ssa->sa);
2292 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
2293 inet_ntoa(ssa->sin.sin_addr), "not ");
2294 error = EADDRNOTAVAIL;
2295 goto out_inp_locked;
2297 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2298 error = imf_prune(imf, &ssa->sin);
2300 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2302 goto out_inp_locked;
2307 * Begin state merge transaction at IGMP layer.
2313 * Give up the multicast address record to which
2314 * the membership points.
2316 (void)in_leavegroup_locked(inm, imf);
2318 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2319 error = inm_merge(inm, imf);
2321 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2323 goto out_imf_rollback;
2326 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2327 error = igmp_change_state(inm);
2329 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2345 /* Remove the gap in the membership and filter array. */
2346 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2347 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2348 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2350 imo->imo_num_memberships--;
2359 * Select the interface for transmitting IPv4 multicast datagrams.
2361 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2362 * may be passed to this socket option. An address of INADDR_ANY or an
2363 * interface index of 0 is used to remove a previous selection.
2364 * When no interface is selected, one is chosen for every send.
2367 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2369 struct in_addr addr;
2370 struct ip_mreqn mreqn;
2372 struct ip_moptions *imo;
2375 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2377 * An interface index was specified using the
2378 * Linux-derived ip_mreqn structure.
2380 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2381 sizeof(struct ip_mreqn));
2385 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2388 if (mreqn.imr_ifindex == 0) {
2391 ifp = ifnet_byindex(mreqn.imr_ifindex);
2393 return (EADDRNOTAVAIL);
2397 * An interface was specified by IPv4 address.
2398 * This is the traditional BSD usage.
2400 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2401 sizeof(struct in_addr));
2404 if (in_nullhost(addr)) {
2407 INADDR_TO_IFP(addr, ifp);
2409 return (EADDRNOTAVAIL);
2411 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp,
2415 /* Reject interfaces which do not support multicast. */
2416 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2417 return (EOPNOTSUPP);
2419 imo = inp_findmoptions(inp);
2420 imo->imo_multicast_ifp = ifp;
2421 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2428 * Atomically set source filters on a socket for an IPv4 multicast group.
2430 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2433 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2435 struct __msfilterreq msfr;
2438 struct in_mfilter *imf;
2439 struct ip_moptions *imo;
2440 struct in_multi *inm;
2444 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2445 sizeof(struct __msfilterreq));
2449 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2452 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2453 msfr.msfr_fmode != MCAST_INCLUDE))
2456 if (msfr.msfr_group.ss_family != AF_INET ||
2457 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2460 gsa = (sockunion_t *)&msfr.msfr_group;
2461 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2464 gsa->sin.sin_port = 0; /* ignore port */
2466 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2467 return (EADDRNOTAVAIL);
2469 ifp = ifnet_byindex(msfr.msfr_ifindex);
2471 return (EADDRNOTAVAIL);
2474 * Take the INP write lock.
2475 * Check if this socket is a member of this group.
2477 imo = inp_findmoptions(inp);
2478 idx = imo_match_group(imo, ifp, &gsa->sa);
2479 if (idx == -1 || imo->imo_mfilters == NULL) {
2480 error = EADDRNOTAVAIL;
2481 goto out_inp_locked;
2483 inm = imo->imo_membership[idx];
2484 imf = &imo->imo_mfilters[idx];
2487 * Begin state merge transaction at socket layer.
2489 INP_WLOCK_ASSERT(inp);
2491 imf->imf_st[1] = msfr.msfr_fmode;
2494 * Apply any new source filters, if present.
2495 * Make a copy of the user-space source vector so
2496 * that we may copy them with a single copyin. This
2497 * allows us to deal with page faults up-front.
2499 if (msfr.msfr_nsrcs > 0) {
2500 struct in_msource *lims;
2501 struct sockaddr_in *psin;
2502 struct sockaddr_storage *kss, *pkss;
2507 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2508 __func__, (unsigned long)msfr.msfr_nsrcs);
2509 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2511 error = copyin(msfr.msfr_srcs, kss,
2512 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2521 * Mark all source filters as UNDEFINED at t1.
2522 * Restore new group filter mode, as imf_leave()
2523 * will set it to INCLUDE.
2526 imf->imf_st[1] = msfr.msfr_fmode;
2529 * Update socket layer filters at t1, lazy-allocating
2530 * new entries. This saves a bunch of memory at the
2531 * cost of one RB_FIND() per source entry; duplicate
2532 * entries in the msfr_nsrcs vector are ignored.
2533 * If we encounter an error, rollback transaction.
2535 * XXX This too could be replaced with a set-symmetric
2536 * difference like loop to avoid walking from root
2537 * every time, as the key space is common.
2539 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2540 psin = (struct sockaddr_in *)pkss;
2541 if (psin->sin_family != AF_INET) {
2542 error = EAFNOSUPPORT;
2545 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2549 error = imf_get_source(imf, psin, &lims);
2552 lims->imsl_st[1] = imf->imf_st[1];
2558 goto out_imf_rollback;
2560 INP_WLOCK_ASSERT(inp);
2564 * Begin state merge transaction at IGMP layer.
2566 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2567 error = inm_merge(inm, imf);
2569 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2570 goto out_imf_rollback;
2573 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2574 error = igmp_change_state(inm);
2576 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2594 * Set the IP multicast options in response to user setsockopt().
2596 * Many of the socket options handled in this function duplicate the
2597 * functionality of socket options in the regular unicast API. However,
2598 * it is not possible to merge the duplicate code, because the idempotence
2599 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2600 * the effects of these options must be treated as separate and distinct.
2602 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2603 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2604 * is refactored to no longer use vifs.
2607 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2609 struct ip_moptions *imo;
2615 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2616 * or is a divert socket, reject it.
2618 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2619 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2620 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2621 return (EOPNOTSUPP);
2623 switch (sopt->sopt_name) {
2624 case IP_MULTICAST_VIF: {
2627 * Select a multicast VIF for transmission.
2628 * Only useful if multicast forwarding is active.
2630 if (legal_vif_num == NULL) {
2634 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2637 if (!legal_vif_num(vifi) && (vifi != -1)) {
2641 imo = inp_findmoptions(inp);
2642 imo->imo_multicast_vif = vifi;
2647 case IP_MULTICAST_IF:
2648 error = inp_set_multicast_if(inp, sopt);
2651 case IP_MULTICAST_TTL: {
2655 * Set the IP time-to-live for outgoing multicast packets.
2656 * The original multicast API required a char argument,
2657 * which is inconsistent with the rest of the socket API.
2658 * We allow either a char or an int.
2660 if (sopt->sopt_valsize == sizeof(u_char)) {
2661 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2668 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2678 imo = inp_findmoptions(inp);
2679 imo->imo_multicast_ttl = ttl;
2684 case IP_MULTICAST_LOOP: {
2688 * Set the loopback flag for outgoing multicast packets.
2689 * Must be zero or one. The original multicast API required a
2690 * char argument, which is inconsistent with the rest
2691 * of the socket API. We allow either a char or an int.
2693 if (sopt->sopt_valsize == sizeof(u_char)) {
2694 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2701 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2705 loop = (u_char)iloop;
2707 imo = inp_findmoptions(inp);
2708 imo->imo_multicast_loop = !!loop;
2713 case IP_ADD_MEMBERSHIP:
2714 case IP_ADD_SOURCE_MEMBERSHIP:
2715 case MCAST_JOIN_GROUP:
2716 case MCAST_JOIN_SOURCE_GROUP:
2717 error = inp_join_group(inp, sopt);
2720 case IP_DROP_MEMBERSHIP:
2721 case IP_DROP_SOURCE_MEMBERSHIP:
2722 case MCAST_LEAVE_GROUP:
2723 case MCAST_LEAVE_SOURCE_GROUP:
2724 error = inp_leave_group(inp, sopt);
2727 case IP_BLOCK_SOURCE:
2728 case IP_UNBLOCK_SOURCE:
2729 case MCAST_BLOCK_SOURCE:
2730 case MCAST_UNBLOCK_SOURCE:
2731 error = inp_block_unblock_source(inp, sopt);
2735 error = inp_set_source_filters(inp, sopt);
2743 INP_UNLOCK_ASSERT(inp);
2749 * Expose IGMP's multicast filter mode and source list(s) to userland,
2750 * keyed by (ifindex, group).
2751 * The filter mode is written out as a uint32_t, followed by
2752 * 0..n of struct in_addr.
2753 * For use by ifmcstat(8).
2754 * SMPng: NOTE: unlocked read of ifindex space.
2757 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2759 struct in_addr src, group;
2761 struct ifmultiaddr *ifma;
2762 struct in_multi *inm;
2763 struct ip_msource *ims;
2767 uint32_t fmode, ifindex;
2772 if (req->newptr != NULL)
2779 if (ifindex <= 0 || ifindex > V_if_index) {
2780 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2785 group.s_addr = name[1];
2786 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2787 CTR2(KTR_IGMPV3, "%s: group %s is not multicast",
2788 __func__, inet_ntoa(group));
2792 ifp = ifnet_byindex(ifindex);
2794 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2799 retval = sysctl_wire_old_buffer(req,
2800 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2807 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2808 if (ifma->ifma_addr->sa_family != AF_INET ||
2809 ifma->ifma_protospec == NULL)
2811 inm = (struct in_multi *)ifma->ifma_protospec;
2812 if (!in_hosteq(inm->inm_addr, group))
2814 fmode = inm->inm_st[1].iss_fmode;
2815 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2818 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2821 ina.s_addr = htonl(ims->ims_haddr);
2822 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2826 * Only copy-out sources which are in-mode.
2828 if (fmode != ims_get_mode(inm, ims, 1)) {
2829 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2833 src.s_addr = htonl(ims->ims_haddr);
2834 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2839 IF_ADDR_RUNLOCK(ifp);
2848 static const char *inm_modestrs[] = { "un", "in", "ex" };
2851 inm_mode_str(const int mode)
2854 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2855 return (inm_modestrs[mode]);
2859 static const char *inm_statestrs[] = {
2872 inm_state_str(const int state)
2875 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2876 return (inm_statestrs[state]);
2881 * Dump an in_multi structure to the console.
2884 inm_print(const struct in_multi *inm)
2888 if ((ktr_mask & KTR_IGMPV3) == 0)
2891 printf("%s: --- begin inm %p ---\n", __func__, inm);
2892 printf("addr %s ifp %p(%s) ifma %p\n",
2893 inet_ntoa(inm->inm_addr),
2895 inm->inm_ifp->if_xname,
2897 printf("timer %u state %s refcount %u scq.len %u\n",
2899 inm_state_str(inm->inm_state),
2901 inm->inm_scq.ifq_len);
2902 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2907 for (t = 0; t < 2; t++) {
2908 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2909 inm_mode_str(inm->inm_st[t].iss_fmode),
2910 inm->inm_st[t].iss_asm,
2911 inm->inm_st[t].iss_ex,
2912 inm->inm_st[t].iss_in,
2913 inm->inm_st[t].iss_rec);
2915 printf("%s: --- end inm %p ---\n", __func__, inm);
2921 inm_print(const struct in_multi *inm)
2928 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);