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);
141 static int inm_is_ifp_detached(const struct in_multi *);
142 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
143 static void inm_purge(struct in_multi *);
144 static void inm_reap(struct in_multi *);
145 static struct ip_moptions *
146 inp_findmoptions(struct inpcb *);
147 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
148 static int inp_join_group(struct inpcb *, struct sockopt *);
149 static int inp_leave_group(struct inpcb *, struct sockopt *);
150 static struct ifnet *
151 inp_lookup_mcast_ifp(const struct inpcb *,
152 const struct sockaddr_in *, const struct in_addr);
153 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
154 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
155 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
156 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
158 SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0, "IPv4 multicast");
160 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
161 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
162 CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxgrpsrc, 0,
163 "Max source filters per group");
164 TUNABLE_ULONG("net.inet.ip.mcast.maxgrpsrc", &in_mcast_maxgrpsrc);
166 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
167 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
168 CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxsocksrc, 0,
169 "Max source filters per socket");
170 TUNABLE_ULONG("net.inet.ip.mcast.maxsocksrc", &in_mcast_maxsocksrc);
172 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
173 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
174 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
175 TUNABLE_INT("net.inet.ip.mcast.loop", &in_mcast_loop);
177 SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
178 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
179 "Per-interface stack-wide source filters");
182 * Inline function which wraps assertions for a valid ifp.
183 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
187 inm_is_ifp_detached(const struct in_multi *inm)
191 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
192 ifp = inm->inm_ifma->ifma_ifp;
195 * Sanity check that netinet's notion of ifp is the
198 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
201 return (ifp == NULL);
205 * Initialize an in_mfilter structure to a known state at t0, t1
206 * with an empty source filter list.
209 imf_init(struct in_mfilter *imf, const int st0, const int st1)
211 memset(imf, 0, sizeof(struct in_mfilter));
212 RB_INIT(&imf->imf_sources);
213 imf->imf_st[0] = st0;
214 imf->imf_st[1] = st1;
218 * Resize the ip_moptions vector to the next power-of-two minus 1.
219 * May be called with locks held; do not sleep.
222 imo_grow(struct ip_moptions *imo)
224 struct in_multi **nmships;
225 struct in_multi **omships;
226 struct in_mfilter *nmfilters;
227 struct in_mfilter *omfilters;
234 omships = imo->imo_membership;
235 omfilters = imo->imo_mfilters;
236 oldmax = imo->imo_max_memberships;
237 newmax = ((oldmax + 1) * 2) - 1;
239 if (newmax <= IP_MAX_MEMBERSHIPS) {
240 nmships = (struct in_multi **)realloc(omships,
241 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
242 nmfilters = (struct in_mfilter *)realloc(omfilters,
243 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
244 if (nmships != NULL && nmfilters != NULL) {
245 /* Initialize newly allocated source filter heads. */
246 for (idx = oldmax; idx < newmax; idx++) {
247 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
250 imo->imo_max_memberships = newmax;
251 imo->imo_membership = nmships;
252 imo->imo_mfilters = nmfilters;
256 if (nmships == NULL || nmfilters == NULL) {
258 free(nmships, M_IPMOPTS);
259 if (nmfilters != NULL)
260 free(nmfilters, M_INMFILTER);
261 return (ETOOMANYREFS);
268 * Find an IPv4 multicast group entry for this ip_moptions instance
269 * which matches the specified group, and optionally an interface.
270 * Return its index into the array, or -1 if not found.
273 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
274 const struct sockaddr *group)
276 const struct sockaddr_in *gsin;
277 struct in_multi **pinm;
281 gsin = (const struct sockaddr_in *)group;
283 /* The imo_membership array may be lazy allocated. */
284 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
287 nmships = imo->imo_num_memberships;
288 pinm = &imo->imo_membership[0];
289 for (idx = 0; idx < nmships; idx++, pinm++) {
292 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
293 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
304 * Find an IPv4 multicast source entry for this imo which matches
305 * the given group index for this socket, and source address.
307 * NOTE: This does not check if the entry is in-mode, merely if
308 * it exists, which may not be the desired behaviour.
310 static struct in_msource *
311 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
312 const struct sockaddr *src)
314 struct ip_msource find;
315 struct in_mfilter *imf;
316 struct ip_msource *ims;
317 const sockunion_t *psa;
319 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
320 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
321 ("%s: invalid index %d\n", __func__, (int)gidx));
323 /* The imo_mfilters array may be lazy allocated. */
324 if (imo->imo_mfilters == NULL)
326 imf = &imo->imo_mfilters[gidx];
328 /* Source trees are keyed in host byte order. */
329 psa = (const sockunion_t *)src;
330 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
331 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
333 return ((struct in_msource *)ims);
337 * Perform filtering for multicast datagrams on a socket by group and source.
339 * Returns 0 if a datagram should be allowed through, or various error codes
340 * if the socket was not a member of the group, or the source was muted, etc.
343 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
344 const struct sockaddr *group, const struct sockaddr *src)
347 struct in_msource *ims;
350 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
352 gidx = imo_match_group(imo, ifp, group);
354 return (MCAST_NOTGMEMBER);
357 * Check if the source was included in an (S,G) join.
358 * Allow reception on exclusive memberships by default,
359 * reject reception on inclusive memberships by default.
360 * Exclude source only if an in-mode exclude filter exists.
361 * Include source only if an in-mode include filter exists.
362 * NOTE: We are comparing group state here at IGMP t1 (now)
363 * with socket-layer t0 (since last downcall).
365 mode = imo->imo_mfilters[gidx].imf_st[1];
366 ims = imo_match_source(imo, gidx, src);
368 if ((ims == NULL && mode == MCAST_INCLUDE) ||
369 (ims != NULL && ims->imsl_st[0] != mode))
370 return (MCAST_NOTSMEMBER);
376 * Find and return a reference to an in_multi record for (ifp, group),
377 * and bump its reference count.
378 * If one does not exist, try to allocate it, and update link-layer multicast
379 * filters on ifp to listen for group.
380 * Assumes the IN_MULTI lock is held across the call.
381 * Return 0 if successful, otherwise return an appropriate error code.
384 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
385 struct in_multi **pinm)
387 struct sockaddr_in gsin;
388 struct ifmultiaddr *ifma;
389 struct in_ifinfo *ii;
390 struct in_multi *inm;
393 IN_MULTI_LOCK_ASSERT();
395 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
397 inm = inm_lookup(ifp, *group);
400 * If we already joined this group, just bump the
401 * refcount and return it.
403 KASSERT(inm->inm_refcount >= 1,
404 ("%s: bad refcount %d", __func__, inm->inm_refcount));
410 memset(&gsin, 0, sizeof(gsin));
411 gsin.sin_family = AF_INET;
412 gsin.sin_len = sizeof(struct sockaddr_in);
413 gsin.sin_addr = *group;
416 * Check if a link-layer group is already associated
417 * with this network-layer group on the given ifnet.
419 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
423 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
427 * If something other than netinet is occupying the link-layer
428 * group, print a meaningful error message and back out of
430 * Otherwise, bump the refcount on the existing network-layer
431 * group association and return it.
433 if (ifma->ifma_protospec != NULL) {
434 inm = (struct in_multi *)ifma->ifma_protospec;
436 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
438 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
439 ("%s: ifma not AF_INET", __func__));
440 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
441 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
442 !in_hosteq(inm->inm_addr, *group))
443 panic("%s: ifma %p is inconsistent with %p (%s)",
444 __func__, ifma, inm, inet_ntoa(*group));
452 IF_ADDR_LOCK_ASSERT(ifp);
455 * A new in_multi record is needed; allocate and initialize it.
456 * We DO NOT perform an IGMP join as the in_ layer may need to
457 * push an initial source list down to IGMP to support SSM.
459 * The initial source filter state is INCLUDE, {} as per the RFC.
461 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
463 if_delmulti_ifma(ifma);
467 inm->inm_addr = *group;
469 inm->inm_igi = ii->ii_igmp;
470 inm->inm_ifma = ifma;
471 inm->inm_refcount = 1;
472 inm->inm_state = IGMP_NOT_MEMBER;
475 * Pending state-changes per group are subject to a bounds check.
477 IFQ_SET_MAXLEN(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
479 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
480 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
481 RB_INIT(&inm->inm_srcs);
483 ifma->ifma_protospec = inm;
492 * Drop a reference to an in_multi record.
494 * If the refcount drops to 0, free the in_multi record and
495 * delete the underlying link-layer membership.
498 inm_release_locked(struct in_multi *inm)
500 struct ifmultiaddr *ifma;
502 IN_MULTI_LOCK_ASSERT();
504 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
506 if (--inm->inm_refcount > 0) {
507 CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__,
512 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
514 ifma = inm->inm_ifma;
516 /* XXX this access is not covered by IF_ADDR_LOCK */
517 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
518 KASSERT(ifma->ifma_protospec == inm,
519 ("%s: ifma_protospec != inm", __func__));
520 ifma->ifma_protospec = NULL;
524 free(inm, M_IPMADDR);
526 if_delmulti_ifma(ifma);
530 * Clear recorded source entries for a group.
531 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
532 * FIXME: Should reap.
535 inm_clear_recorded(struct in_multi *inm)
537 struct ip_msource *ims;
539 IN_MULTI_LOCK_ASSERT();
541 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
544 --inm->inm_st[1].iss_rec;
547 KASSERT(inm->inm_st[1].iss_rec == 0,
548 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
552 * Record a source as pending for a Source-Group IGMPv3 query.
553 * This lives here as it modifies the shared tree.
555 * inm is the group descriptor.
556 * naddr is the address of the source to record in network-byte order.
558 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
559 * lazy-allocate a source node in response to an SG query.
560 * Otherwise, no allocation is performed. This saves some memory
561 * with the trade-off that the source will not be reported to the
562 * router if joined in the window between the query response and
563 * the group actually being joined on the local host.
565 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
566 * This turns off the allocation of a recorded source entry if
567 * the group has not been joined.
569 * Return 0 if the source didn't exist or was already marked as recorded.
570 * Return 1 if the source was marked as recorded by this function.
571 * Return <0 if any error occured (negated errno code).
574 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
576 struct ip_msource find;
577 struct ip_msource *ims, *nims;
579 IN_MULTI_LOCK_ASSERT();
581 find.ims_haddr = ntohl(naddr);
582 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
583 if (ims && ims->ims_stp)
586 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
588 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
592 nims->ims_haddr = find.ims_haddr;
593 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
599 * Mark the source as recorded and update the recorded
603 ++inm->inm_st[1].iss_rec;
609 * Return a pointer to an in_msource owned by an in_mfilter,
610 * given its source address.
611 * Lazy-allocate if needed. If this is a new entry its filter state is
614 * imf is the filter set being modified.
615 * haddr is the source address in *host* byte-order.
617 * SMPng: May be called with locks held; malloc must not block.
620 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
621 struct in_msource **plims)
623 struct ip_msource find;
624 struct ip_msource *ims, *nims;
625 struct in_msource *lims;
632 /* key is host byte order */
633 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
634 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
635 lims = (struct in_msource *)ims;
637 if (imf->imf_nsrc == in_mcast_maxsocksrc)
639 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
643 lims = (struct in_msource *)nims;
644 lims->ims_haddr = find.ims_haddr;
645 lims->imsl_st[0] = MCAST_UNDEFINED;
646 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
656 * Graft a source entry into an existing socket-layer filter set,
657 * maintaining any required invariants and checking allocations.
659 * The source is marked as being in the new filter mode at t1.
661 * Return the pointer to the new node, otherwise return NULL.
663 static struct in_msource *
664 imf_graft(struct in_mfilter *imf, const uint8_t st1,
665 const struct sockaddr_in *psin)
667 struct ip_msource *nims;
668 struct in_msource *lims;
670 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
674 lims = (struct in_msource *)nims;
675 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
676 lims->imsl_st[0] = MCAST_UNDEFINED;
677 lims->imsl_st[1] = st1;
678 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
685 * Prune a source entry from an existing socket-layer filter set,
686 * maintaining any required invariants and checking allocations.
688 * The source is marked as being left at t1, it is not freed.
690 * Return 0 if no error occurred, otherwise return an errno value.
693 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
695 struct ip_msource find;
696 struct ip_msource *ims;
697 struct in_msource *lims;
699 /* key is host byte order */
700 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
701 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
704 lims = (struct in_msource *)ims;
705 lims->imsl_st[1] = MCAST_UNDEFINED;
710 * Revert socket-layer filter set deltas at t1 to t0 state.
713 imf_rollback(struct in_mfilter *imf)
715 struct ip_msource *ims, *tims;
716 struct in_msource *lims;
718 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
719 lims = (struct in_msource *)ims;
720 if (lims->imsl_st[0] == lims->imsl_st[1]) {
721 /* no change at t1 */
723 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
724 /* revert change to existing source at t1 */
725 lims->imsl_st[1] = lims->imsl_st[0];
727 /* revert source added t1 */
728 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
729 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
730 free(ims, M_INMFILTER);
734 imf->imf_st[1] = imf->imf_st[0];
738 * Mark socket-layer filter set as INCLUDE {} at t1.
741 imf_leave(struct in_mfilter *imf)
743 struct ip_msource *ims;
744 struct in_msource *lims;
746 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
747 lims = (struct in_msource *)ims;
748 lims->imsl_st[1] = MCAST_UNDEFINED;
750 imf->imf_st[1] = MCAST_INCLUDE;
754 * Mark socket-layer filter set deltas as committed.
757 imf_commit(struct in_mfilter *imf)
759 struct ip_msource *ims;
760 struct in_msource *lims;
762 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
763 lims = (struct in_msource *)ims;
764 lims->imsl_st[0] = lims->imsl_st[1];
766 imf->imf_st[0] = imf->imf_st[1];
770 * Reap unreferenced sources from socket-layer filter set.
773 imf_reap(struct in_mfilter *imf)
775 struct ip_msource *ims, *tims;
776 struct in_msource *lims;
778 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
779 lims = (struct in_msource *)ims;
780 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
781 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
782 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
783 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
784 free(ims, M_INMFILTER);
791 * Purge socket-layer filter set.
794 imf_purge(struct in_mfilter *imf)
796 struct ip_msource *ims, *tims;
798 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
799 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
800 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
801 free(ims, M_INMFILTER);
804 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
805 KASSERT(RB_EMPTY(&imf->imf_sources),
806 ("%s: imf_sources not empty", __func__));
810 * Look up a source filter entry for a multicast group.
812 * inm is the group descriptor to work with.
813 * haddr is the host-byte-order IPv4 address to look up.
814 * noalloc may be non-zero to suppress allocation of sources.
815 * *pims will be set to the address of the retrieved or allocated source.
817 * SMPng: NOTE: may be called with locks held.
818 * Return 0 if successful, otherwise return a non-zero error code.
821 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
822 const int noalloc, struct ip_msource **pims)
824 struct ip_msource find;
825 struct ip_msource *ims, *nims;
830 find.ims_haddr = haddr;
831 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
832 if (ims == NULL && !noalloc) {
833 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
835 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
839 nims->ims_haddr = haddr;
840 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
844 ia.s_addr = htonl(haddr);
845 CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__,
855 * Merge socket-layer source into IGMP-layer source.
856 * If rollback is non-zero, perform the inverse of the merge.
859 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
862 int n = rollback ? -1 : 1;
866 ia.s_addr = htonl(ims->ims_haddr);
869 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
870 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s",
871 __func__, n, inet_ntoa(ia));
872 ims->ims_st[1].ex -= n;
873 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
874 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s",
875 __func__, n, inet_ntoa(ia));
876 ims->ims_st[1].in -= n;
879 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
880 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s",
881 __func__, n, inet_ntoa(ia));
882 ims->ims_st[1].ex += n;
883 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
884 CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s",
885 __func__, n, inet_ntoa(ia));
886 ims->ims_st[1].in += n;
891 * Atomically update the global in_multi state, when a membership's
892 * filter list is being updated in any way.
894 * imf is the per-inpcb-membership group filter pointer.
895 * A fake imf may be passed for in-kernel consumers.
897 * XXX This is a candidate for a set-symmetric-difference style loop
898 * which would eliminate the repeated lookup from root of ims nodes,
899 * as they share the same key space.
901 * If any error occurred this function will back out of refcounts
902 * and return a non-zero value.
905 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
907 struct ip_msource *ims, *nims;
908 struct in_msource *lims;
917 * Update the source filters first, as this may fail.
918 * Maintain count of in-mode filters at t0, t1. These are
919 * used to work out if we transition into ASM mode or not.
920 * Maintain a count of source filters whose state was
921 * actually modified by this operation.
923 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
924 lims = (struct in_msource *)ims;
925 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
926 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
927 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
928 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
932 ims_merge(nims, lims, 0);
935 struct ip_msource *bims;
937 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
938 lims = (struct in_msource *)ims;
939 if (lims->imsl_st[0] == lims->imsl_st[1])
941 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
944 ims_merge(bims, lims, 1);
949 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
950 __func__, nsrc0, nsrc1);
952 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
953 if (imf->imf_st[0] == imf->imf_st[1] &&
954 imf->imf_st[1] == MCAST_INCLUDE) {
956 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
957 --inm->inm_st[1].iss_in;
961 /* Handle filter mode transition on socket. */
962 if (imf->imf_st[0] != imf->imf_st[1]) {
963 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
964 __func__, imf->imf_st[0], imf->imf_st[1]);
966 if (imf->imf_st[0] == MCAST_EXCLUDE) {
967 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
968 --inm->inm_st[1].iss_ex;
969 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
970 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
971 --inm->inm_st[1].iss_in;
974 if (imf->imf_st[1] == MCAST_EXCLUDE) {
975 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
976 inm->inm_st[1].iss_ex++;
977 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
978 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
979 inm->inm_st[1].iss_in++;
984 * Track inm filter state in terms of listener counts.
985 * If there are any exclusive listeners, stack-wide
986 * membership is exclusive.
987 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
988 * If no listeners remain, state is undefined at t1,
989 * and the IGMP lifecycle for this group should finish.
991 if (inm->inm_st[1].iss_ex > 0) {
992 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
993 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
994 } else if (inm->inm_st[1].iss_in > 0) {
995 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
996 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
998 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
999 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1002 /* Decrement ASM listener count on transition out of ASM mode. */
1003 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1004 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1005 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1006 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1007 --inm->inm_st[1].iss_asm;
1010 /* Increment ASM listener count on transition to ASM mode. */
1011 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1012 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1013 inm->inm_st[1].iss_asm++;
1016 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1021 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1028 * Mark an in_multi's filter set deltas as committed.
1029 * Called by IGMP after a state change has been enqueued.
1032 inm_commit(struct in_multi *inm)
1034 struct ip_msource *ims;
1036 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1037 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1040 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1041 ims->ims_st[0] = ims->ims_st[1];
1043 inm->inm_st[0] = inm->inm_st[1];
1047 * Reap unreferenced nodes from an in_multi's filter set.
1050 inm_reap(struct in_multi *inm)
1052 struct ip_msource *ims, *tims;
1054 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1055 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1056 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1059 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1060 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1061 free(ims, M_IPMSOURCE);
1067 * Purge all source nodes from an in_multi's filter set.
1070 inm_purge(struct in_multi *inm)
1072 struct ip_msource *ims, *tims;
1074 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1075 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1076 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1077 free(ims, M_IPMSOURCE);
1083 * Join a multicast group; unlocked entry point.
1085 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1086 * is not held. Fortunately, ifp is unlikely to have been detached
1087 * at this point, so we assume it's OK to recurse.
1090 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1091 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1096 error = in_joingroup_locked(ifp, gina, imf, pinm);
1103 * Join a multicast group; real entry point.
1105 * Only preserves atomicity at inm level.
1106 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1108 * If the IGMP downcall fails, the group is not joined, and an error
1112 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1113 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1115 struct in_mfilter timf;
1116 struct in_multi *inm;
1119 IN_MULTI_LOCK_ASSERT();
1121 CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__,
1122 inet_ntoa(*gina), ifp, ifp->if_xname);
1128 * If no imf was specified (i.e. kernel consumer),
1129 * fake one up and assume it is an ASM join.
1132 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1136 error = in_getmulti(ifp, gina, &inm);
1138 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1142 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1143 error = inm_merge(inm, imf);
1145 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1146 goto out_inm_release;
1149 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1150 error = igmp_change_state(inm);
1152 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1153 goto out_inm_release;
1158 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1159 inm_release_locked(inm);
1168 * Leave a multicast group; unlocked entry point.
1171 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1179 error = in_leavegroup_locked(inm, imf);
1186 * Leave a multicast group; real entry point.
1187 * All source filters will be expunged.
1189 * Only preserves atomicity at inm level.
1191 * Holding the write lock for the INP which contains imf
1192 * is highly advisable. We can't assert for it as imf does not
1193 * contain a back-pointer to the owning inp.
1195 * Note: This is not the same as inm_release(*) as this function also
1196 * makes a state change downcall into IGMP.
1199 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1201 struct in_mfilter timf;
1206 IN_MULTI_LOCK_ASSERT();
1208 CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__,
1209 inm, inet_ntoa(inm->inm_addr),
1210 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1214 * If no imf was specified (i.e. kernel consumer),
1215 * fake one up and assume it is an ASM join.
1218 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1223 * Begin state merge transaction at IGMP layer.
1225 * As this particular invocation should not cause any memory
1226 * to be allocated, and there is no opportunity to roll back
1227 * the transaction, it MUST NOT fail.
1229 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1230 error = inm_merge(inm, imf);
1231 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1233 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1234 error = igmp_change_state(inm);
1236 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1238 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1239 inm_release_locked(inm);
1244 /*#ifndef BURN_BRIDGES*/
1246 * Join an IPv4 multicast group in (*,G) exclusive mode.
1247 * The group must be a 224.0.0.0/24 link-scope group.
1248 * This KPI is for legacy kernel consumers only.
1251 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1253 struct in_multi *pinm;
1256 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1257 ("%s: %s not in 224.0.0.0/24", __func__, inet_ntoa(*ap)));
1259 error = in_joingroup(ifp, ap, NULL, &pinm);
1267 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1268 * This KPI is for legacy kernel consumers only.
1271 in_delmulti(struct in_multi *inm)
1274 (void)in_leavegroup(inm, NULL);
1279 * Block or unblock an ASM multicast source on an inpcb.
1280 * This implements the delta-based API described in RFC 3678.
1282 * The delta-based API applies only to exclusive-mode memberships.
1283 * An IGMP downcall will be performed.
1285 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1287 * Return 0 if successful, otherwise return an appropriate error code.
1290 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1292 struct group_source_req gsr;
1293 sockunion_t *gsa, *ssa;
1295 struct in_mfilter *imf;
1296 struct ip_moptions *imo;
1297 struct in_msource *ims;
1298 struct in_multi *inm;
1307 memset(&gsr, 0, sizeof(struct group_source_req));
1308 gsa = (sockunion_t *)&gsr.gsr_group;
1309 ssa = (sockunion_t *)&gsr.gsr_source;
1311 switch (sopt->sopt_name) {
1312 case IP_BLOCK_SOURCE:
1313 case IP_UNBLOCK_SOURCE: {
1314 struct ip_mreq_source mreqs;
1316 error = sooptcopyin(sopt, &mreqs,
1317 sizeof(struct ip_mreq_source),
1318 sizeof(struct ip_mreq_source));
1322 gsa->sin.sin_family = AF_INET;
1323 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1324 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1326 ssa->sin.sin_family = AF_INET;
1327 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1328 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1330 if (!in_nullhost(mreqs.imr_interface))
1331 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1333 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1336 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1337 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1341 case MCAST_BLOCK_SOURCE:
1342 case MCAST_UNBLOCK_SOURCE:
1343 error = sooptcopyin(sopt, &gsr,
1344 sizeof(struct group_source_req),
1345 sizeof(struct group_source_req));
1349 if (gsa->sin.sin_family != AF_INET ||
1350 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1353 if (ssa->sin.sin_family != AF_INET ||
1354 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1357 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1358 return (EADDRNOTAVAIL);
1360 ifp = ifnet_byindex(gsr.gsr_interface);
1362 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1367 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1368 __func__, sopt->sopt_name);
1369 return (EOPNOTSUPP);
1373 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1377 * Check if we are actually a member of this group.
1379 imo = inp_findmoptions(inp);
1380 idx = imo_match_group(imo, ifp, &gsa->sa);
1381 if (idx == -1 || imo->imo_mfilters == NULL) {
1382 error = EADDRNOTAVAIL;
1383 goto out_inp_locked;
1386 KASSERT(imo->imo_mfilters != NULL,
1387 ("%s: imo_mfilters not allocated", __func__));
1388 imf = &imo->imo_mfilters[idx];
1389 inm = imo->imo_membership[idx];
1392 * Attempting to use the delta-based API on an
1393 * non exclusive-mode membership is an error.
1395 fmode = imf->imf_st[0];
1396 if (fmode != MCAST_EXCLUDE) {
1398 goto out_inp_locked;
1402 * Deal with error cases up-front:
1403 * Asked to block, but already blocked; or
1404 * Asked to unblock, but nothing to unblock.
1405 * If adding a new block entry, allocate it.
1407 ims = imo_match_source(imo, idx, &ssa->sa);
1408 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1409 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
1410 inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not ");
1411 error = EADDRNOTAVAIL;
1412 goto out_inp_locked;
1415 INP_WLOCK_ASSERT(inp);
1418 * Begin state merge transaction at socket layer.
1421 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1422 ims = imf_graft(imf, fmode, &ssa->sin);
1426 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1427 error = imf_prune(imf, &ssa->sin);
1431 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1432 goto out_imf_rollback;
1436 * Begin state merge transaction at IGMP layer.
1440 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1441 error = inm_merge(inm, imf);
1443 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1444 goto out_imf_rollback;
1447 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1448 error = igmp_change_state(inm);
1450 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1468 * Given an inpcb, return its multicast options structure pointer. Accepts
1469 * an unlocked inpcb pointer, but will return it locked. May sleep.
1471 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1472 * SMPng: NOTE: Returns with the INP write lock held.
1474 static struct ip_moptions *
1475 inp_findmoptions(struct inpcb *inp)
1477 struct ip_moptions *imo;
1478 struct in_multi **immp;
1479 struct in_mfilter *imfp;
1483 if (inp->inp_moptions != NULL)
1484 return (inp->inp_moptions);
1488 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1489 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1491 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1492 M_INMFILTER, M_WAITOK);
1494 imo->imo_multicast_ifp = NULL;
1495 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1496 imo->imo_multicast_vif = -1;
1497 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1498 imo->imo_multicast_loop = in_mcast_loop;
1499 imo->imo_num_memberships = 0;
1500 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1501 imo->imo_membership = immp;
1503 /* Initialize per-group source filters. */
1504 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1505 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1506 imo->imo_mfilters = imfp;
1509 if (inp->inp_moptions != NULL) {
1510 free(imfp, M_INMFILTER);
1511 free(immp, M_IPMOPTS);
1512 free(imo, M_IPMOPTS);
1513 return (inp->inp_moptions);
1515 inp->inp_moptions = imo;
1520 * Discard the IP multicast options (and source filters).
1522 * SMPng: NOTE: assumes INP write lock is held.
1525 inp_freemoptions(struct ip_moptions *imo)
1527 struct in_mfilter *imf;
1528 size_t idx, nmships;
1530 KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1532 nmships = imo->imo_num_memberships;
1533 for (idx = 0; idx < nmships; ++idx) {
1534 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1537 (void)in_leavegroup(imo->imo_membership[idx], imf);
1542 if (imo->imo_mfilters)
1543 free(imo->imo_mfilters, M_INMFILTER);
1544 free(imo->imo_membership, M_IPMOPTS);
1545 free(imo, M_IPMOPTS);
1549 * Atomically get source filters on a socket for an IPv4 multicast group.
1550 * Called with INP lock held; returns with lock released.
1553 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1555 struct __msfilterreq msfr;
1558 struct ip_moptions *imo;
1559 struct in_mfilter *imf;
1560 struct ip_msource *ims;
1561 struct in_msource *lims;
1562 struct sockaddr_in *psin;
1563 struct sockaddr_storage *ptss;
1564 struct sockaddr_storage *tss;
1566 size_t idx, nsrcs, ncsrcs;
1568 INP_WLOCK_ASSERT(inp);
1570 imo = inp->inp_moptions;
1571 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1575 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1576 sizeof(struct __msfilterreq));
1580 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1583 ifp = ifnet_byindex(msfr.msfr_ifindex);
1590 * Lookup group on the socket.
1592 gsa = (sockunion_t *)&msfr.msfr_group;
1593 idx = imo_match_group(imo, ifp, &gsa->sa);
1594 if (idx == -1 || imo->imo_mfilters == NULL) {
1596 return (EADDRNOTAVAIL);
1598 imf = &imo->imo_mfilters[idx];
1601 * Ignore memberships which are in limbo.
1603 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1607 msfr.msfr_fmode = imf->imf_st[1];
1610 * If the user specified a buffer, copy out the source filter
1611 * entries to userland gracefully.
1612 * We only copy out the number of entries which userland
1613 * has asked for, but we always tell userland how big the
1614 * buffer really needs to be.
1617 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1618 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1619 M_TEMP, M_NOWAIT | M_ZERO);
1627 * Count number of sources in-mode at t0.
1628 * If buffer space exists and remains, copy out source entries.
1630 nsrcs = msfr.msfr_nsrcs;
1633 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1634 lims = (struct in_msource *)ims;
1635 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1636 lims->imsl_st[0] != imf->imf_st[0])
1639 if (tss != NULL && nsrcs > 0) {
1640 psin = (struct sockaddr_in *)ptss;
1641 psin->sin_family = AF_INET;
1642 psin->sin_len = sizeof(struct sockaddr_in);
1643 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1653 error = copyout(tss, msfr.msfr_srcs,
1654 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1660 msfr.msfr_nsrcs = ncsrcs;
1661 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1667 * Return the IP multicast options in response to user getsockopt().
1670 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1672 struct ip_mreqn mreqn;
1673 struct ip_moptions *imo;
1675 struct in_ifaddr *ia;
1680 imo = inp->inp_moptions;
1682 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1683 * or is a divert socket, reject it.
1685 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1686 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1687 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1689 return (EOPNOTSUPP);
1693 switch (sopt->sopt_name) {
1694 case IP_MULTICAST_VIF:
1696 optval = imo->imo_multicast_vif;
1700 error = sooptcopyout(sopt, &optval, sizeof(int));
1703 case IP_MULTICAST_IF:
1704 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1706 ifp = imo->imo_multicast_ifp;
1707 if (!in_nullhost(imo->imo_multicast_addr)) {
1708 mreqn.imr_address = imo->imo_multicast_addr;
1709 } else if (ifp != NULL) {
1710 mreqn.imr_ifindex = ifp->if_index;
1714 IA_SIN(ia)->sin_addr;
1715 ifa_free(&ia->ia_ifa);
1720 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1721 error = sooptcopyout(sopt, &mreqn,
1722 sizeof(struct ip_mreqn));
1724 error = sooptcopyout(sopt, &mreqn.imr_address,
1725 sizeof(struct in_addr));
1729 case IP_MULTICAST_TTL:
1731 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1733 optval = coptval = imo->imo_multicast_ttl;
1735 if (sopt->sopt_valsize == sizeof(u_char))
1736 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1738 error = sooptcopyout(sopt, &optval, sizeof(int));
1741 case IP_MULTICAST_LOOP:
1743 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1745 optval = coptval = imo->imo_multicast_loop;
1747 if (sopt->sopt_valsize == sizeof(u_char))
1748 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1750 error = sooptcopyout(sopt, &optval, sizeof(int));
1755 error = EADDRNOTAVAIL;
1758 error = inp_get_source_filters(inp, sopt);
1764 error = ENOPROTOOPT;
1768 INP_UNLOCK_ASSERT(inp);
1774 * Look up the ifnet to use for a multicast group membership,
1775 * given the IPv4 address of an interface, and the IPv4 group address.
1777 * This routine exists to support legacy multicast applications
1778 * which do not understand that multicast memberships are scoped to
1779 * specific physical links in the networking stack, or which need
1780 * to join link-scope groups before IPv4 addresses are configured.
1782 * If inp is non-NULL, use this socket's current FIB number for any
1783 * required FIB lookup.
1784 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1785 * and use its ifp; usually, this points to the default next-hop.
1787 * If the FIB lookup fails, attempt to use the first non-loopback
1788 * interface with multicast capability in the system as a
1789 * last resort. The legacy IPv4 ASM API requires that we do
1790 * this in order to allow groups to be joined when the routing
1791 * table has not yet been populated during boot.
1793 * Returns NULL if no ifp could be found.
1795 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1796 * FUTURE: Implement IPv4 source-address selection.
1798 static struct ifnet *
1799 inp_lookup_mcast_ifp(const struct inpcb *inp,
1800 const struct sockaddr_in *gsin, const struct in_addr ina)
1804 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1805 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1806 ("%s: not multicast", __func__));
1809 if (!in_nullhost(ina)) {
1810 INADDR_TO_IFP(ina, ifp);
1815 memcpy(&ro.ro_dst, gsin, sizeof(struct sockaddr_in));
1816 in_rtalloc_ign(&ro, 0, inp ? inp->inp_inc.inc_fibnum : 0);
1817 if (ro.ro_rt != NULL) {
1818 ifp = ro.ro_rt->rt_ifp;
1819 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1822 struct in_ifaddr *ia;
1827 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1829 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1830 (mifp->if_flags & IFF_MULTICAST)) {
1835 IN_IFADDR_RUNLOCK();
1843 * Join an IPv4 multicast group, possibly with a source.
1846 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1848 struct group_source_req gsr;
1849 sockunion_t *gsa, *ssa;
1851 struct in_mfilter *imf;
1852 struct ip_moptions *imo;
1853 struct in_multi *inm;
1854 struct in_msource *lims;
1863 memset(&gsr, 0, sizeof(struct group_source_req));
1864 gsa = (sockunion_t *)&gsr.gsr_group;
1865 gsa->ss.ss_family = AF_UNSPEC;
1866 ssa = (sockunion_t *)&gsr.gsr_source;
1867 ssa->ss.ss_family = AF_UNSPEC;
1869 switch (sopt->sopt_name) {
1870 case IP_ADD_MEMBERSHIP:
1871 case IP_ADD_SOURCE_MEMBERSHIP: {
1872 struct ip_mreq_source mreqs;
1874 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1875 error = sooptcopyin(sopt, &mreqs,
1876 sizeof(struct ip_mreq),
1877 sizeof(struct ip_mreq));
1879 * Do argument switcharoo from ip_mreq into
1880 * ip_mreq_source to avoid using two instances.
1882 mreqs.imr_interface = mreqs.imr_sourceaddr;
1883 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1884 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1885 error = sooptcopyin(sopt, &mreqs,
1886 sizeof(struct ip_mreq_source),
1887 sizeof(struct ip_mreq_source));
1892 gsa->sin.sin_family = AF_INET;
1893 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1894 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1896 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1897 ssa->sin.sin_family = AF_INET;
1898 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1899 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1902 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1905 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1906 mreqs.imr_interface);
1907 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1908 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1912 case MCAST_JOIN_GROUP:
1913 case MCAST_JOIN_SOURCE_GROUP:
1914 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1915 error = sooptcopyin(sopt, &gsr,
1916 sizeof(struct group_req),
1917 sizeof(struct group_req));
1918 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1919 error = sooptcopyin(sopt, &gsr,
1920 sizeof(struct group_source_req),
1921 sizeof(struct group_source_req));
1926 if (gsa->sin.sin_family != AF_INET ||
1927 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1931 * Overwrite the port field if present, as the sockaddr
1932 * being copied in may be matched with a binary comparison.
1934 gsa->sin.sin_port = 0;
1935 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1936 if (ssa->sin.sin_family != AF_INET ||
1937 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1939 ssa->sin.sin_port = 0;
1942 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1945 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1946 return (EADDRNOTAVAIL);
1947 ifp = ifnet_byindex(gsr.gsr_interface);
1951 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1952 __func__, sopt->sopt_name);
1953 return (EOPNOTSUPP);
1957 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1958 return (EADDRNOTAVAIL);
1960 imo = inp_findmoptions(inp);
1961 idx = imo_match_group(imo, ifp, &gsa->sa);
1965 inm = imo->imo_membership[idx];
1966 imf = &imo->imo_mfilters[idx];
1967 if (ssa->ss.ss_family != AF_UNSPEC) {
1969 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1970 * is an error. On an existing inclusive membership,
1971 * it just adds the source to the filter list.
1973 if (imf->imf_st[1] != MCAST_INCLUDE) {
1975 goto out_inp_locked;
1977 /* Throw out duplicates. */
1978 lims = imo_match_source(imo, idx, &ssa->sa);
1980 error = EADDRNOTAVAIL;
1981 goto out_inp_locked;
1985 * MCAST_JOIN_GROUP on an existing inclusive
1986 * membership is an error; if you want to change
1987 * filter mode, you must use the userland API
1988 * setsourcefilter().
1990 if (imf->imf_st[1] == MCAST_INCLUDE) {
1992 goto out_inp_locked;
1995 * MCAST_JOIN_GROUP on an existing exclusive
1996 * membership is an error; return EADDRINUSE
1997 * to preserve 4.4BSD API idempotence, and
1998 * avoid tedious detour to code below.
1999 * NOTE: This is bending RFC 3678 a bit.
2001 if (imf->imf_st[1] == MCAST_EXCLUDE) {
2003 goto out_inp_locked;
2009 * Begin state merge transaction at socket layer.
2011 INP_WLOCK_ASSERT(inp);
2014 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2015 error = imo_grow(imo);
2017 goto out_inp_locked;
2020 * Allocate the new slot upfront so we can deal with
2021 * grafting the new source filter in same code path
2022 * as for join-source on existing membership.
2024 idx = imo->imo_num_memberships;
2025 imo->imo_membership[idx] = NULL;
2026 imo->imo_num_memberships++;
2027 KASSERT(imo->imo_mfilters != NULL,
2028 ("%s: imf_mfilters vector was not allocated", __func__));
2029 imf = &imo->imo_mfilters[idx];
2030 KASSERT(RB_EMPTY(&imf->imf_sources),
2031 ("%s: imf_sources not empty", __func__));
2035 * Graft new source into filter list for this inpcb's
2036 * membership of the group. The in_multi may not have
2037 * been allocated yet if this is a new membership, however,
2038 * the in_mfilter slot will be allocated and must be initialized.
2040 if (ssa->ss.ss_family != AF_UNSPEC) {
2041 /* Membership starts in IN mode */
2043 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2044 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2046 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2048 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2050 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2056 /* No address specified; Membership starts in EX mode */
2058 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2059 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2064 * Begin state merge transaction at IGMP layer.
2069 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2073 imo->imo_membership[idx] = inm;
2075 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2076 error = inm_merge(inm, imf);
2078 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2080 goto out_imf_rollback;
2082 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2083 error = igmp_change_state(inm);
2085 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2087 goto out_imf_rollback;
2094 INP_WLOCK_ASSERT(inp);
2106 if (error && is_new) {
2107 imo->imo_membership[idx] = NULL;
2108 --imo->imo_num_memberships;
2117 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2120 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2122 struct group_source_req gsr;
2123 struct ip_mreq_source mreqs;
2124 sockunion_t *gsa, *ssa;
2126 struct in_mfilter *imf;
2127 struct ip_moptions *imo;
2128 struct in_msource *ims;
2129 struct in_multi *inm;
2131 int error, is_final;
2137 memset(&gsr, 0, sizeof(struct group_source_req));
2138 gsa = (sockunion_t *)&gsr.gsr_group;
2139 gsa->ss.ss_family = AF_UNSPEC;
2140 ssa = (sockunion_t *)&gsr.gsr_source;
2141 ssa->ss.ss_family = AF_UNSPEC;
2143 switch (sopt->sopt_name) {
2144 case IP_DROP_MEMBERSHIP:
2145 case IP_DROP_SOURCE_MEMBERSHIP:
2146 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2147 error = sooptcopyin(sopt, &mreqs,
2148 sizeof(struct ip_mreq),
2149 sizeof(struct ip_mreq));
2151 * Swap interface and sourceaddr arguments,
2152 * as ip_mreq and ip_mreq_source are laid
2155 mreqs.imr_interface = mreqs.imr_sourceaddr;
2156 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2157 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2158 error = sooptcopyin(sopt, &mreqs,
2159 sizeof(struct ip_mreq_source),
2160 sizeof(struct ip_mreq_source));
2165 gsa->sin.sin_family = AF_INET;
2166 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2167 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2169 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2170 ssa->sin.sin_family = AF_INET;
2171 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2172 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2176 * Attempt to look up hinted ifp from interface address.
2177 * Fallthrough with null ifp iff lookup fails, to
2178 * preserve 4.4BSD mcast API idempotence.
2179 * XXX NOTE WELL: The RFC 3678 API is preferred because
2180 * using an IPv4 address as a key is racy.
2182 if (!in_nullhost(mreqs.imr_interface))
2183 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2185 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
2186 __func__, inet_ntoa(mreqs.imr_interface), ifp);
2190 case MCAST_LEAVE_GROUP:
2191 case MCAST_LEAVE_SOURCE_GROUP:
2192 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2193 error = sooptcopyin(sopt, &gsr,
2194 sizeof(struct group_req),
2195 sizeof(struct group_req));
2196 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2197 error = sooptcopyin(sopt, &gsr,
2198 sizeof(struct group_source_req),
2199 sizeof(struct group_source_req));
2204 if (gsa->sin.sin_family != AF_INET ||
2205 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2208 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2209 if (ssa->sin.sin_family != AF_INET ||
2210 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2214 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2215 return (EADDRNOTAVAIL);
2217 ifp = ifnet_byindex(gsr.gsr_interface);
2220 return (EADDRNOTAVAIL);
2224 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2225 __func__, sopt->sopt_name);
2226 return (EOPNOTSUPP);
2230 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2234 * Find the membership in the membership array.
2236 imo = inp_findmoptions(inp);
2237 idx = imo_match_group(imo, ifp, &gsa->sa);
2239 error = EADDRNOTAVAIL;
2240 goto out_inp_locked;
2242 inm = imo->imo_membership[idx];
2243 imf = &imo->imo_mfilters[idx];
2245 if (ssa->ss.ss_family != AF_UNSPEC)
2249 * Begin state merge transaction at socket layer.
2251 INP_WLOCK_ASSERT(inp);
2254 * If we were instructed only to leave a given source, do so.
2255 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2260 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2261 error = EADDRNOTAVAIL;
2262 goto out_inp_locked;
2264 ims = imo_match_source(imo, idx, &ssa->sa);
2266 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
2267 inet_ntoa(ssa->sin.sin_addr), "not ");
2268 error = EADDRNOTAVAIL;
2269 goto out_inp_locked;
2271 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2272 error = imf_prune(imf, &ssa->sin);
2274 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2276 goto out_inp_locked;
2281 * Begin state merge transaction at IGMP layer.
2287 * Give up the multicast address record to which
2288 * the membership points.
2290 (void)in_leavegroup_locked(inm, imf);
2292 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2293 error = inm_merge(inm, imf);
2295 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2297 goto out_imf_rollback;
2300 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2301 error = igmp_change_state(inm);
2303 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2319 /* Remove the gap in the membership and filter array. */
2320 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2321 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2322 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2324 imo->imo_num_memberships--;
2333 * Select the interface for transmitting IPv4 multicast datagrams.
2335 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2336 * may be passed to this socket option. An address of INADDR_ANY or an
2337 * interface index of 0 is used to remove a previous selection.
2338 * When no interface is selected, one is chosen for every send.
2341 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2343 struct in_addr addr;
2344 struct ip_mreqn mreqn;
2346 struct ip_moptions *imo;
2349 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2351 * An interface index was specified using the
2352 * Linux-derived ip_mreqn structure.
2354 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2355 sizeof(struct ip_mreqn));
2359 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2362 if (mreqn.imr_ifindex == 0) {
2365 ifp = ifnet_byindex(mreqn.imr_ifindex);
2367 return (EADDRNOTAVAIL);
2371 * An interface was specified by IPv4 address.
2372 * This is the traditional BSD usage.
2374 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2375 sizeof(struct in_addr));
2378 if (in_nullhost(addr)) {
2381 INADDR_TO_IFP(addr, ifp);
2383 return (EADDRNOTAVAIL);
2385 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp,
2389 /* Reject interfaces which do not support multicast. */
2390 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2391 return (EOPNOTSUPP);
2393 imo = inp_findmoptions(inp);
2394 imo->imo_multicast_ifp = ifp;
2395 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2402 * Atomically set source filters on a socket for an IPv4 multicast group.
2404 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2407 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2409 struct __msfilterreq msfr;
2412 struct in_mfilter *imf;
2413 struct ip_moptions *imo;
2414 struct in_multi *inm;
2418 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2419 sizeof(struct __msfilterreq));
2423 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc ||
2424 (msfr.msfr_fmode != MCAST_EXCLUDE &&
2425 msfr.msfr_fmode != MCAST_INCLUDE))
2428 if (msfr.msfr_group.ss_family != AF_INET ||
2429 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2432 gsa = (sockunion_t *)&msfr.msfr_group;
2433 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2436 gsa->sin.sin_port = 0; /* ignore port */
2438 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2439 return (EADDRNOTAVAIL);
2441 ifp = ifnet_byindex(msfr.msfr_ifindex);
2443 return (EADDRNOTAVAIL);
2446 * Take the INP write lock.
2447 * Check if this socket is a member of this group.
2449 imo = inp_findmoptions(inp);
2450 idx = imo_match_group(imo, ifp, &gsa->sa);
2451 if (idx == -1 || imo->imo_mfilters == NULL) {
2452 error = EADDRNOTAVAIL;
2453 goto out_inp_locked;
2455 inm = imo->imo_membership[idx];
2456 imf = &imo->imo_mfilters[idx];
2459 * Begin state merge transaction at socket layer.
2461 INP_WLOCK_ASSERT(inp);
2463 imf->imf_st[1] = msfr.msfr_fmode;
2466 * Apply any new source filters, if present.
2467 * Make a copy of the user-space source vector so
2468 * that we may copy them with a single copyin. This
2469 * allows us to deal with page faults up-front.
2471 if (msfr.msfr_nsrcs > 0) {
2472 struct in_msource *lims;
2473 struct sockaddr_in *psin;
2474 struct sockaddr_storage *kss, *pkss;
2479 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2480 __func__, (unsigned long)msfr.msfr_nsrcs);
2481 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2483 error = copyin(msfr.msfr_srcs, kss,
2484 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2493 * Mark all source filters as UNDEFINED at t1.
2494 * Restore new group filter mode, as imf_leave()
2495 * will set it to INCLUDE.
2498 imf->imf_st[1] = msfr.msfr_fmode;
2501 * Update socket layer filters at t1, lazy-allocating
2502 * new entries. This saves a bunch of memory at the
2503 * cost of one RB_FIND() per source entry; duplicate
2504 * entries in the msfr_nsrcs vector are ignored.
2505 * If we encounter an error, rollback transaction.
2507 * XXX This too could be replaced with a set-symmetric
2508 * difference like loop to avoid walking from root
2509 * every time, as the key space is common.
2511 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2512 psin = (struct sockaddr_in *)pkss;
2513 if (psin->sin_family != AF_INET) {
2514 error = EAFNOSUPPORT;
2517 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2521 error = imf_get_source(imf, psin, &lims);
2524 lims->imsl_st[1] = imf->imf_st[1];
2530 goto out_imf_rollback;
2532 INP_WLOCK_ASSERT(inp);
2536 * Begin state merge transaction at IGMP layer.
2538 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2539 error = inm_merge(inm, imf);
2541 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2542 goto out_imf_rollback;
2545 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2546 error = igmp_change_state(inm);
2548 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2566 * Set the IP multicast options in response to user setsockopt().
2568 * Many of the socket options handled in this function duplicate the
2569 * functionality of socket options in the regular unicast API. However,
2570 * it is not possible to merge the duplicate code, because the idempotence
2571 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2572 * the effects of these options must be treated as separate and distinct.
2574 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2575 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2576 * is refactored to no longer use vifs.
2579 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2581 struct ip_moptions *imo;
2587 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2588 * or is a divert socket, reject it.
2590 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2591 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2592 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2593 return (EOPNOTSUPP);
2595 switch (sopt->sopt_name) {
2596 case IP_MULTICAST_VIF: {
2599 * Select a multicast VIF for transmission.
2600 * Only useful if multicast forwarding is active.
2602 if (legal_vif_num == NULL) {
2606 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2609 if (!legal_vif_num(vifi) && (vifi != -1)) {
2613 imo = inp_findmoptions(inp);
2614 imo->imo_multicast_vif = vifi;
2619 case IP_MULTICAST_IF:
2620 error = inp_set_multicast_if(inp, sopt);
2623 case IP_MULTICAST_TTL: {
2627 * Set the IP time-to-live for outgoing multicast packets.
2628 * The original multicast API required a char argument,
2629 * which is inconsistent with the rest of the socket API.
2630 * We allow either a char or an int.
2632 if (sopt->sopt_valsize == sizeof(u_char)) {
2633 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2640 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2650 imo = inp_findmoptions(inp);
2651 imo->imo_multicast_ttl = ttl;
2656 case IP_MULTICAST_LOOP: {
2660 * Set the loopback flag for outgoing multicast packets.
2661 * Must be zero or one. The original multicast API required a
2662 * char argument, which is inconsistent with the rest
2663 * of the socket API. We allow either a char or an int.
2665 if (sopt->sopt_valsize == sizeof(u_char)) {
2666 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2673 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2677 loop = (u_char)iloop;
2679 imo = inp_findmoptions(inp);
2680 imo->imo_multicast_loop = !!loop;
2685 case IP_ADD_MEMBERSHIP:
2686 case IP_ADD_SOURCE_MEMBERSHIP:
2687 case MCAST_JOIN_GROUP:
2688 case MCAST_JOIN_SOURCE_GROUP:
2689 error = inp_join_group(inp, sopt);
2692 case IP_DROP_MEMBERSHIP:
2693 case IP_DROP_SOURCE_MEMBERSHIP:
2694 case MCAST_LEAVE_GROUP:
2695 case MCAST_LEAVE_SOURCE_GROUP:
2696 error = inp_leave_group(inp, sopt);
2699 case IP_BLOCK_SOURCE:
2700 case IP_UNBLOCK_SOURCE:
2701 case MCAST_BLOCK_SOURCE:
2702 case MCAST_UNBLOCK_SOURCE:
2703 error = inp_block_unblock_source(inp, sopt);
2707 error = inp_set_source_filters(inp, sopt);
2715 INP_UNLOCK_ASSERT(inp);
2721 * Expose IGMP's multicast filter mode and source list(s) to userland,
2722 * keyed by (ifindex, group).
2723 * The filter mode is written out as a uint32_t, followed by
2724 * 0..n of struct in_addr.
2725 * For use by ifmcstat(8).
2726 * SMPng: NOTE: unlocked read of ifindex space.
2729 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2731 struct in_addr src, group;
2733 struct ifmultiaddr *ifma;
2734 struct in_multi *inm;
2735 struct ip_msource *ims;
2739 uint32_t fmode, ifindex;
2744 if (req->newptr != NULL)
2751 if (ifindex <= 0 || ifindex > V_if_index) {
2752 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2757 group.s_addr = name[1];
2758 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2759 CTR2(KTR_IGMPV3, "%s: group %s is not multicast",
2760 __func__, inet_ntoa(group));
2764 ifp = ifnet_byindex(ifindex);
2766 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2771 retval = sysctl_wire_old_buffer(req,
2772 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2779 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2780 if (ifma->ifma_addr->sa_family != AF_INET ||
2781 ifma->ifma_protospec == NULL)
2783 inm = (struct in_multi *)ifma->ifma_protospec;
2784 if (!in_hosteq(inm->inm_addr, group))
2786 fmode = inm->inm_st[1].iss_fmode;
2787 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2790 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2793 ina.s_addr = htonl(ims->ims_haddr);
2794 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2798 * Only copy-out sources which are in-mode.
2800 if (fmode != ims_get_mode(inm, ims, 1)) {
2801 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2805 src.s_addr = htonl(ims->ims_haddr);
2806 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2811 IF_ADDR_UNLOCK(ifp);
2820 static const char *inm_modestrs[] = { "un", "in", "ex" };
2823 inm_mode_str(const int mode)
2826 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2827 return (inm_modestrs[mode]);
2831 static const char *inm_statestrs[] = {
2844 inm_state_str(const int state)
2847 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2848 return (inm_statestrs[state]);
2853 * Dump an in_multi structure to the console.
2856 inm_print(const struct in_multi *inm)
2860 if ((ktr_mask & KTR_IGMPV3) == 0)
2863 printf("%s: --- begin inm %p ---\n", __func__, inm);
2864 printf("addr %s ifp %p(%s) ifma %p\n",
2865 inet_ntoa(inm->inm_addr),
2867 inm->inm_ifp->if_xname,
2869 printf("timer %u state %s refcount %u scq.len %u\n",
2871 inm_state_str(inm->inm_state),
2873 inm->inm_scq.ifq_len);
2874 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2879 for (t = 0; t < 2; t++) {
2880 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2881 inm_mode_str(inm->inm_st[t].iss_fmode),
2882 inm->inm_st[t].iss_asm,
2883 inm->inm_st[t].iss_ex,
2884 inm->inm_st[t].iss_in,
2885 inm->inm_st[t].iss_rec);
2887 printf("%s: --- end inm %p ---\n", __func__, inm);
2893 inm_print(const struct in_multi *inm)
2900 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);