2 * Copyright (c) 2007-2009 Bruce Simpson.
3 * Copyright (c) 2005 Robert N. M. Watson.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote
15 * products derived from this software without specific prior written
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * IPv4 multicast socket, group, and socket option processing module.
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
42 #include <sys/malloc.h>
44 #include <sys/protosw.h>
45 #include <sys/rmlock.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/protosw.h>
49 #include <sys/sysctl.h>
51 #include <sys/taskqueue.h>
55 #include <net/if_var.h>
56 #include <net/if_dl.h>
57 #include <net/route.h>
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_fib.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip_var.h>
66 #include <netinet/igmp_var.h>
69 #define KTR_IGMPV3 KTR_INET
72 #ifndef __SOCKUNION_DECLARED
74 struct sockaddr_storage ss;
76 struct sockaddr_dl sdl;
77 struct sockaddr_in sin;
79 typedef union sockunion sockunion_t;
80 #define __SOCKUNION_DECLARED
81 #endif /* __SOCKUNION_DECLARED */
83 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
84 "IPv4 multicast PCB-layer source filter");
85 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
86 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
87 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
88 "IPv4 multicast IGMP-layer source filter");
92 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
93 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
94 * it can be taken by code in net/if.c also.
95 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
97 * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly
98 * any need for in_multi itself to be virtualized -- it is bound to an ifp
99 * anyway no matter what happens.
101 struct mtx in_multi_mtx;
102 MTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF);
105 * Functions with non-static linkage defined in this file should be
106 * declared in in_var.h:
111 * in_joingroup_locked()
113 * in_leavegroup_locked()
119 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
122 static void imf_commit(struct in_mfilter *);
123 static int imf_get_source(struct in_mfilter *imf,
124 const struct sockaddr_in *psin,
125 struct in_msource **);
126 static struct in_msource *
127 imf_graft(struct in_mfilter *, const uint8_t,
128 const struct sockaddr_in *);
129 static void imf_leave(struct in_mfilter *);
130 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
131 static void imf_purge(struct in_mfilter *);
132 static void imf_rollback(struct in_mfilter *);
133 static void imf_reap(struct in_mfilter *);
134 static int imo_grow(struct ip_moptions *);
135 static size_t imo_match_group(const struct ip_moptions *,
136 const struct ifnet *, const struct sockaddr *);
137 static struct in_msource *
138 imo_match_source(const struct ip_moptions *, const size_t,
139 const struct sockaddr *);
140 static void ims_merge(struct ip_msource *ims,
141 const struct in_msource *lims, const int rollback);
142 static int in_getmulti(struct ifnet *, const struct in_addr *,
144 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
145 const int noalloc, struct ip_msource **pims);
147 static int inm_is_ifp_detached(const struct in_multi *);
149 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
150 static void inm_purge(struct in_multi *);
151 static void inm_reap(struct in_multi *);
152 static struct ip_moptions *
153 inp_findmoptions(struct inpcb *);
154 static void inp_freemoptions_internal(struct ip_moptions *);
155 static void inp_gcmoptions(void *, int);
156 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
157 static int inp_join_group(struct inpcb *, struct sockopt *);
158 static int inp_leave_group(struct inpcb *, struct sockopt *);
159 static struct ifnet *
160 inp_lookup_mcast_ifp(const struct inpcb *,
161 const struct sockaddr_in *, const struct in_addr);
162 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
163 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
164 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
165 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
167 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
170 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
171 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
172 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
173 "Max source filters per group");
175 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
176 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
177 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
178 "Max source filters per socket");
180 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
181 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
182 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
184 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
185 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
186 "Per-interface stack-wide source filters");
188 static STAILQ_HEAD(, ip_moptions) imo_gc_list =
189 STAILQ_HEAD_INITIALIZER(imo_gc_list);
190 static struct task imo_gc_task = TASK_INITIALIZER(0, inp_gcmoptions, NULL);
194 * Inline function which wraps assertions for a valid ifp.
195 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
199 inm_is_ifp_detached(const struct in_multi *inm)
203 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
204 ifp = inm->inm_ifma->ifma_ifp;
207 * Sanity check that netinet's notion of ifp is the
210 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
213 return (ifp == NULL);
218 * Initialize an in_mfilter structure to a known state at t0, t1
219 * with an empty source filter list.
222 imf_init(struct in_mfilter *imf, const int st0, const int st1)
224 memset(imf, 0, sizeof(struct in_mfilter));
225 RB_INIT(&imf->imf_sources);
226 imf->imf_st[0] = st0;
227 imf->imf_st[1] = st1;
231 * Function for looking up an in_multi record for an IPv4 multicast address
232 * on a given interface. ifp must be valid. If no record found, return NULL.
233 * The IN_MULTI_LOCK and IF_ADDR_LOCK on ifp must be held.
236 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
238 struct ifmultiaddr *ifma;
239 struct in_multi *inm;
241 IN_MULTI_LOCK_ASSERT();
242 IF_ADDR_LOCK_ASSERT(ifp);
245 TAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
246 if (ifma->ifma_addr->sa_family == AF_INET) {
247 inm = (struct in_multi *)ifma->ifma_protospec;
248 if (inm->inm_addr.s_addr == ina.s_addr)
257 * Wrapper for inm_lookup_locked().
258 * The IF_ADDR_LOCK will be taken on ifp and released on return.
261 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
263 struct in_multi *inm;
265 IN_MULTI_LOCK_ASSERT();
267 inm = inm_lookup_locked(ifp, ina);
268 IF_ADDR_RUNLOCK(ifp);
274 * Resize the ip_moptions vector to the next power-of-two minus 1.
275 * May be called with locks held; do not sleep.
278 imo_grow(struct ip_moptions *imo)
280 struct in_multi **nmships;
281 struct in_multi **omships;
282 struct in_mfilter *nmfilters;
283 struct in_mfilter *omfilters;
290 omships = imo->imo_membership;
291 omfilters = imo->imo_mfilters;
292 oldmax = imo->imo_max_memberships;
293 newmax = ((oldmax + 1) * 2) - 1;
295 if (newmax <= IP_MAX_MEMBERSHIPS) {
296 nmships = (struct in_multi **)realloc(omships,
297 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
298 nmfilters = (struct in_mfilter *)realloc(omfilters,
299 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
300 if (nmships != NULL && nmfilters != NULL) {
301 /* Initialize newly allocated source filter heads. */
302 for (idx = oldmax; idx < newmax; idx++) {
303 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
306 imo->imo_max_memberships = newmax;
307 imo->imo_membership = nmships;
308 imo->imo_mfilters = nmfilters;
312 if (nmships == NULL || nmfilters == NULL) {
314 free(nmships, M_IPMOPTS);
315 if (nmfilters != NULL)
316 free(nmfilters, M_INMFILTER);
317 return (ETOOMANYREFS);
324 * Find an IPv4 multicast group entry for this ip_moptions instance
325 * which matches the specified group, and optionally an interface.
326 * Return its index into the array, or -1 if not found.
329 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
330 const struct sockaddr *group)
332 const struct sockaddr_in *gsin;
333 struct in_multi **pinm;
337 gsin = (const struct sockaddr_in *)group;
339 /* The imo_membership array may be lazy allocated. */
340 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
343 nmships = imo->imo_num_memberships;
344 pinm = &imo->imo_membership[0];
345 for (idx = 0; idx < nmships; idx++, pinm++) {
348 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
349 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
360 * Find an IPv4 multicast source entry for this imo which matches
361 * the given group index for this socket, and source address.
363 * NOTE: This does not check if the entry is in-mode, merely if
364 * it exists, which may not be the desired behaviour.
366 static struct in_msource *
367 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
368 const struct sockaddr *src)
370 struct ip_msource find;
371 struct in_mfilter *imf;
372 struct ip_msource *ims;
373 const sockunion_t *psa;
375 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
376 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
377 ("%s: invalid index %d\n", __func__, (int)gidx));
379 /* The imo_mfilters array may be lazy allocated. */
380 if (imo->imo_mfilters == NULL)
382 imf = &imo->imo_mfilters[gidx];
384 /* Source trees are keyed in host byte order. */
385 psa = (const sockunion_t *)src;
386 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
387 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
389 return ((struct in_msource *)ims);
393 * Perform filtering for multicast datagrams on a socket by group and source.
395 * Returns 0 if a datagram should be allowed through, or various error codes
396 * if the socket was not a member of the group, or the source was muted, etc.
399 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
400 const struct sockaddr *group, const struct sockaddr *src)
403 struct in_msource *ims;
406 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
408 gidx = imo_match_group(imo, ifp, group);
410 return (MCAST_NOTGMEMBER);
413 * Check if the source was included in an (S,G) join.
414 * Allow reception on exclusive memberships by default,
415 * reject reception on inclusive memberships by default.
416 * Exclude source only if an in-mode exclude filter exists.
417 * Include source only if an in-mode include filter exists.
418 * NOTE: We are comparing group state here at IGMP t1 (now)
419 * with socket-layer t0 (since last downcall).
421 mode = imo->imo_mfilters[gidx].imf_st[1];
422 ims = imo_match_source(imo, gidx, src);
424 if ((ims == NULL && mode == MCAST_INCLUDE) ||
425 (ims != NULL && ims->imsl_st[0] != mode))
426 return (MCAST_NOTSMEMBER);
432 * Find and return a reference to an in_multi record for (ifp, group),
433 * and bump its reference count.
434 * If one does not exist, try to allocate it, and update link-layer multicast
435 * filters on ifp to listen for group.
436 * Assumes the IN_MULTI lock is held across the call.
437 * Return 0 if successful, otherwise return an appropriate error code.
440 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
441 struct in_multi **pinm)
443 struct sockaddr_in gsin;
444 struct ifmultiaddr *ifma;
445 struct in_ifinfo *ii;
446 struct in_multi *inm;
449 IN_MULTI_LOCK_ASSERT();
451 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
453 inm = inm_lookup(ifp, *group);
456 * If we already joined this group, just bump the
457 * refcount and return it.
459 KASSERT(inm->inm_refcount >= 1,
460 ("%s: bad refcount %d", __func__, inm->inm_refcount));
466 memset(&gsin, 0, sizeof(gsin));
467 gsin.sin_family = AF_INET;
468 gsin.sin_len = sizeof(struct sockaddr_in);
469 gsin.sin_addr = *group;
472 * Check if a link-layer group is already associated
473 * with this network-layer group on the given ifnet.
475 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
479 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
483 * If something other than netinet is occupying the link-layer
484 * group, print a meaningful error message and back out of
486 * Otherwise, bump the refcount on the existing network-layer
487 * group association and return it.
489 if (ifma->ifma_protospec != NULL) {
490 inm = (struct in_multi *)ifma->ifma_protospec;
492 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
494 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
495 ("%s: ifma not AF_INET", __func__));
496 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
497 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
498 !in_hosteq(inm->inm_addr, *group)) {
499 char addrbuf[INET_ADDRSTRLEN];
501 panic("%s: ifma %p is inconsistent with %p (%s)",
502 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
507 IF_ADDR_WUNLOCK(ifp);
511 IF_ADDR_WLOCK_ASSERT(ifp);
514 * A new in_multi record is needed; allocate and initialize it.
515 * We DO NOT perform an IGMP join as the in_ layer may need to
516 * push an initial source list down to IGMP to support SSM.
518 * The initial source filter state is INCLUDE, {} as per the RFC.
520 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
522 IF_ADDR_WUNLOCK(ifp);
523 if_delmulti_ifma(ifma);
526 inm->inm_addr = *group;
528 inm->inm_igi = ii->ii_igmp;
529 inm->inm_ifma = ifma;
530 inm->inm_refcount = 1;
531 inm->inm_state = IGMP_NOT_MEMBER;
532 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
533 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
534 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
535 RB_INIT(&inm->inm_srcs);
537 ifma->ifma_protospec = inm;
541 IF_ADDR_WUNLOCK(ifp);
546 * Drop a reference to an in_multi record.
548 * If the refcount drops to 0, free the in_multi record and
549 * delete the underlying link-layer membership.
552 inm_release_locked(struct in_multi *inm)
554 struct ifmultiaddr *ifma;
556 IN_MULTI_LOCK_ASSERT();
558 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
560 if (--inm->inm_refcount > 0) {
561 CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__,
566 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
568 ifma = inm->inm_ifma;
570 /* XXX this access is not covered by IF_ADDR_LOCK */
571 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
572 KASSERT(ifma->ifma_protospec == inm,
573 ("%s: ifma_protospec != inm", __func__));
574 ifma->ifma_protospec = NULL;
578 free(inm, M_IPMADDR);
580 if_delmulti_ifma(ifma);
584 * Clear recorded source entries for a group.
585 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
586 * FIXME: Should reap.
589 inm_clear_recorded(struct in_multi *inm)
591 struct ip_msource *ims;
593 IN_MULTI_LOCK_ASSERT();
595 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
598 --inm->inm_st[1].iss_rec;
601 KASSERT(inm->inm_st[1].iss_rec == 0,
602 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
606 * Record a source as pending for a Source-Group IGMPv3 query.
607 * This lives here as it modifies the shared tree.
609 * inm is the group descriptor.
610 * naddr is the address of the source to record in network-byte order.
612 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
613 * lazy-allocate a source node in response to an SG query.
614 * Otherwise, no allocation is performed. This saves some memory
615 * with the trade-off that the source will not be reported to the
616 * router if joined in the window between the query response and
617 * the group actually being joined on the local host.
619 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
620 * This turns off the allocation of a recorded source entry if
621 * the group has not been joined.
623 * Return 0 if the source didn't exist or was already marked as recorded.
624 * Return 1 if the source was marked as recorded by this function.
625 * Return <0 if any error occurred (negated errno code).
628 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
630 struct ip_msource find;
631 struct ip_msource *ims, *nims;
633 IN_MULTI_LOCK_ASSERT();
635 find.ims_haddr = ntohl(naddr);
636 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
637 if (ims && ims->ims_stp)
640 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
642 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
646 nims->ims_haddr = find.ims_haddr;
647 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
653 * Mark the source as recorded and update the recorded
657 ++inm->inm_st[1].iss_rec;
663 * Return a pointer to an in_msource owned by an in_mfilter,
664 * given its source address.
665 * Lazy-allocate if needed. If this is a new entry its filter state is
668 * imf is the filter set being modified.
669 * haddr is the source address in *host* byte-order.
671 * SMPng: May be called with locks held; malloc must not block.
674 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
675 struct in_msource **plims)
677 struct ip_msource find;
678 struct ip_msource *ims, *nims;
679 struct in_msource *lims;
686 /* key is host byte order */
687 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
688 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
689 lims = (struct in_msource *)ims;
691 if (imf->imf_nsrc == in_mcast_maxsocksrc)
693 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
697 lims = (struct in_msource *)nims;
698 lims->ims_haddr = find.ims_haddr;
699 lims->imsl_st[0] = MCAST_UNDEFINED;
700 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
710 * Graft a source entry into an existing socket-layer filter set,
711 * maintaining any required invariants and checking allocations.
713 * The source is marked as being in the new filter mode at t1.
715 * Return the pointer to the new node, otherwise return NULL.
717 static struct in_msource *
718 imf_graft(struct in_mfilter *imf, const uint8_t st1,
719 const struct sockaddr_in *psin)
721 struct ip_msource *nims;
722 struct in_msource *lims;
724 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
728 lims = (struct in_msource *)nims;
729 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
730 lims->imsl_st[0] = MCAST_UNDEFINED;
731 lims->imsl_st[1] = st1;
732 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
739 * Prune a source entry from an existing socket-layer filter set,
740 * maintaining any required invariants and checking allocations.
742 * The source is marked as being left at t1, it is not freed.
744 * Return 0 if no error occurred, otherwise return an errno value.
747 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
749 struct ip_msource find;
750 struct ip_msource *ims;
751 struct in_msource *lims;
753 /* key is host byte order */
754 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
755 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
758 lims = (struct in_msource *)ims;
759 lims->imsl_st[1] = MCAST_UNDEFINED;
764 * Revert socket-layer filter set deltas at t1 to t0 state.
767 imf_rollback(struct in_mfilter *imf)
769 struct ip_msource *ims, *tims;
770 struct in_msource *lims;
772 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
773 lims = (struct in_msource *)ims;
774 if (lims->imsl_st[0] == lims->imsl_st[1]) {
775 /* no change at t1 */
777 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
778 /* revert change to existing source at t1 */
779 lims->imsl_st[1] = lims->imsl_st[0];
781 /* revert source added t1 */
782 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
783 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
784 free(ims, M_INMFILTER);
788 imf->imf_st[1] = imf->imf_st[0];
792 * Mark socket-layer filter set as INCLUDE {} at t1.
795 imf_leave(struct in_mfilter *imf)
797 struct ip_msource *ims;
798 struct in_msource *lims;
800 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
801 lims = (struct in_msource *)ims;
802 lims->imsl_st[1] = MCAST_UNDEFINED;
804 imf->imf_st[1] = MCAST_INCLUDE;
808 * Mark socket-layer filter set deltas as committed.
811 imf_commit(struct in_mfilter *imf)
813 struct ip_msource *ims;
814 struct in_msource *lims;
816 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
817 lims = (struct in_msource *)ims;
818 lims->imsl_st[0] = lims->imsl_st[1];
820 imf->imf_st[0] = imf->imf_st[1];
824 * Reap unreferenced sources from socket-layer filter set.
827 imf_reap(struct in_mfilter *imf)
829 struct ip_msource *ims, *tims;
830 struct in_msource *lims;
832 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
833 lims = (struct in_msource *)ims;
834 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
835 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
836 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
837 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
838 free(ims, M_INMFILTER);
845 * Purge socket-layer filter set.
848 imf_purge(struct in_mfilter *imf)
850 struct ip_msource *ims, *tims;
852 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
853 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
854 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
855 free(ims, M_INMFILTER);
858 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
859 KASSERT(RB_EMPTY(&imf->imf_sources),
860 ("%s: imf_sources not empty", __func__));
864 * Look up a source filter entry for a multicast group.
866 * inm is the group descriptor to work with.
867 * haddr is the host-byte-order IPv4 address to look up.
868 * noalloc may be non-zero to suppress allocation of sources.
869 * *pims will be set to the address of the retrieved or allocated source.
871 * SMPng: NOTE: may be called with locks held.
872 * Return 0 if successful, otherwise return a non-zero error code.
875 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
876 const int noalloc, struct ip_msource **pims)
878 struct ip_msource find;
879 struct ip_msource *ims, *nims;
882 char addrbuf[INET_ADDRSTRLEN];
885 find.ims_haddr = haddr;
886 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
887 if (ims == NULL && !noalloc) {
888 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
890 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
894 nims->ims_haddr = haddr;
895 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
899 ia.s_addr = htonl(haddr);
900 CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__,
901 inet_ntoa_r(ia, addrbuf), ims);
910 * Merge socket-layer source into IGMP-layer source.
911 * If rollback is non-zero, perform the inverse of the merge.
914 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
917 int n = rollback ? -1 : 1;
919 char addrbuf[INET_ADDRSTRLEN];
922 ia.s_addr = htonl(ims->ims_haddr);
925 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
926 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s",
927 __func__, n, inet_ntoa_r(ia, addrbuf));
928 ims->ims_st[1].ex -= n;
929 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
930 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s",
931 __func__, n, inet_ntoa_r(ia, addrbuf));
932 ims->ims_st[1].in -= n;
935 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
936 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s",
937 __func__, n, inet_ntoa_r(ia, addrbuf));
938 ims->ims_st[1].ex += n;
939 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
940 CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s",
941 __func__, n, inet_ntoa_r(ia, addrbuf));
942 ims->ims_st[1].in += n;
947 * Atomically update the global in_multi state, when a membership's
948 * filter list is being updated in any way.
950 * imf is the per-inpcb-membership group filter pointer.
951 * A fake imf may be passed for in-kernel consumers.
953 * XXX This is a candidate for a set-symmetric-difference style loop
954 * which would eliminate the repeated lookup from root of ims nodes,
955 * as they share the same key space.
957 * If any error occurred this function will back out of refcounts
958 * and return a non-zero value.
961 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
963 struct ip_msource *ims, *nims;
964 struct in_msource *lims;
973 * Update the source filters first, as this may fail.
974 * Maintain count of in-mode filters at t0, t1. These are
975 * used to work out if we transition into ASM mode or not.
976 * Maintain a count of source filters whose state was
977 * actually modified by this operation.
979 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
980 lims = (struct in_msource *)ims;
981 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
982 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
983 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
984 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
988 ims_merge(nims, lims, 0);
991 struct ip_msource *bims;
993 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
994 lims = (struct in_msource *)ims;
995 if (lims->imsl_st[0] == lims->imsl_st[1])
997 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1000 ims_merge(bims, lims, 1);
1005 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1006 __func__, nsrc0, nsrc1);
1008 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1009 if (imf->imf_st[0] == imf->imf_st[1] &&
1010 imf->imf_st[1] == MCAST_INCLUDE) {
1012 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1013 --inm->inm_st[1].iss_in;
1017 /* Handle filter mode transition on socket. */
1018 if (imf->imf_st[0] != imf->imf_st[1]) {
1019 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1020 __func__, imf->imf_st[0], imf->imf_st[1]);
1022 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1023 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1024 --inm->inm_st[1].iss_ex;
1025 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1026 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1027 --inm->inm_st[1].iss_in;
1030 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1031 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1032 inm->inm_st[1].iss_ex++;
1033 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1034 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1035 inm->inm_st[1].iss_in++;
1040 * Track inm filter state in terms of listener counts.
1041 * If there are any exclusive listeners, stack-wide
1042 * membership is exclusive.
1043 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1044 * If no listeners remain, state is undefined at t1,
1045 * and the IGMP lifecycle for this group should finish.
1047 if (inm->inm_st[1].iss_ex > 0) {
1048 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1049 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1050 } else if (inm->inm_st[1].iss_in > 0) {
1051 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1052 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1054 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1055 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1058 /* Decrement ASM listener count on transition out of ASM mode. */
1059 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1060 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1061 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1062 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1063 --inm->inm_st[1].iss_asm;
1066 /* Increment ASM listener count on transition to ASM mode. */
1067 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1068 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1069 inm->inm_st[1].iss_asm++;
1072 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1077 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1084 * Mark an in_multi's filter set deltas as committed.
1085 * Called by IGMP after a state change has been enqueued.
1088 inm_commit(struct in_multi *inm)
1090 struct ip_msource *ims;
1092 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1093 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1096 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1097 ims->ims_st[0] = ims->ims_st[1];
1099 inm->inm_st[0] = inm->inm_st[1];
1103 * Reap unreferenced nodes from an in_multi's filter set.
1106 inm_reap(struct in_multi *inm)
1108 struct ip_msource *ims, *tims;
1110 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1111 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1112 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1115 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1116 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1117 free(ims, M_IPMSOURCE);
1123 * Purge all source nodes from an in_multi's filter set.
1126 inm_purge(struct in_multi *inm)
1128 struct ip_msource *ims, *tims;
1130 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1131 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1132 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1133 free(ims, M_IPMSOURCE);
1139 * Join a multicast group; unlocked entry point.
1141 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1142 * is not held. Fortunately, ifp is unlikely to have been detached
1143 * at this point, so we assume it's OK to recurse.
1146 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1147 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1152 error = in_joingroup_locked(ifp, gina, imf, pinm);
1159 * Join a multicast group; real entry point.
1161 * Only preserves atomicity at inm level.
1162 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1164 * If the IGMP downcall fails, the group is not joined, and an error
1168 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1169 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1171 struct in_mfilter timf;
1172 struct in_multi *inm;
1175 char addrbuf[INET_ADDRSTRLEN];
1178 IN_MULTI_LOCK_ASSERT();
1180 CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__,
1181 inet_ntoa_r(*gina, addrbuf), ifp, ifp->if_xname);
1187 * If no imf was specified (i.e. kernel consumer),
1188 * fake one up and assume it is an ASM join.
1191 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1195 error = in_getmulti(ifp, gina, &inm);
1197 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1201 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1202 error = inm_merge(inm, imf);
1204 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1205 goto out_inm_release;
1208 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1209 error = igmp_change_state(inm);
1211 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1212 goto out_inm_release;
1217 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1218 inm_release_locked(inm);
1227 * Leave a multicast group; unlocked entry point.
1230 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1235 error = in_leavegroup_locked(inm, imf);
1242 * Leave a multicast group; real entry point.
1243 * All source filters will be expunged.
1245 * Only preserves atomicity at inm level.
1247 * Holding the write lock for the INP which contains imf
1248 * is highly advisable. We can't assert for it as imf does not
1249 * contain a back-pointer to the owning inp.
1251 * Note: This is not the same as inm_release(*) as this function also
1252 * makes a state change downcall into IGMP.
1255 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1257 struct in_mfilter timf;
1260 char addrbuf[INET_ADDRSTRLEN];
1265 IN_MULTI_LOCK_ASSERT();
1267 CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__,
1268 inm, inet_ntoa_r(inm->inm_addr, addrbuf),
1269 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1273 * If no imf was specified (i.e. kernel consumer),
1274 * fake one up and assume it is an ASM join.
1277 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1282 * Begin state merge transaction at IGMP layer.
1284 * As this particular invocation should not cause any memory
1285 * to be allocated, and there is no opportunity to roll back
1286 * the transaction, it MUST NOT fail.
1288 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1289 error = inm_merge(inm, imf);
1290 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1292 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1293 CURVNET_SET(inm->inm_ifp->if_vnet);
1294 error = igmp_change_state(inm);
1297 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1299 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1300 inm_release_locked(inm);
1305 /*#ifndef BURN_BRIDGES*/
1307 * Join an IPv4 multicast group in (*,G) exclusive mode.
1308 * The group must be a 224.0.0.0/24 link-scope group.
1309 * This KPI is for legacy kernel consumers only.
1312 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1314 struct in_multi *pinm;
1317 char addrbuf[INET_ADDRSTRLEN];
1320 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1321 ("%s: %s not in 224.0.0.0/24", __func__,
1322 inet_ntoa_r(*ap, addrbuf)));
1324 error = in_joingroup(ifp, ap, NULL, &pinm);
1332 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1333 * This KPI is for legacy kernel consumers only.
1336 in_delmulti(struct in_multi *inm)
1339 (void)in_leavegroup(inm, NULL);
1344 * Block or unblock an ASM multicast source on an inpcb.
1345 * This implements the delta-based API described in RFC 3678.
1347 * The delta-based API applies only to exclusive-mode memberships.
1348 * An IGMP downcall will be performed.
1350 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1352 * Return 0 if successful, otherwise return an appropriate error code.
1355 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1357 struct group_source_req gsr;
1358 sockunion_t *gsa, *ssa;
1360 struct in_mfilter *imf;
1361 struct ip_moptions *imo;
1362 struct in_msource *ims;
1363 struct in_multi *inm;
1368 char addrbuf[INET_ADDRSTRLEN];
1375 memset(&gsr, 0, sizeof(struct group_source_req));
1376 gsa = (sockunion_t *)&gsr.gsr_group;
1377 ssa = (sockunion_t *)&gsr.gsr_source;
1379 switch (sopt->sopt_name) {
1380 case IP_BLOCK_SOURCE:
1381 case IP_UNBLOCK_SOURCE: {
1382 struct ip_mreq_source mreqs;
1384 error = sooptcopyin(sopt, &mreqs,
1385 sizeof(struct ip_mreq_source),
1386 sizeof(struct ip_mreq_source));
1390 gsa->sin.sin_family = AF_INET;
1391 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1392 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1394 ssa->sin.sin_family = AF_INET;
1395 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1396 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1398 if (!in_nullhost(mreqs.imr_interface))
1399 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1401 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1404 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1405 __func__, inet_ntoa_r(mreqs.imr_interface, addrbuf), ifp);
1409 case MCAST_BLOCK_SOURCE:
1410 case MCAST_UNBLOCK_SOURCE:
1411 error = sooptcopyin(sopt, &gsr,
1412 sizeof(struct group_source_req),
1413 sizeof(struct group_source_req));
1417 if (gsa->sin.sin_family != AF_INET ||
1418 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1421 if (ssa->sin.sin_family != AF_INET ||
1422 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1425 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1426 return (EADDRNOTAVAIL);
1428 ifp = ifnet_byindex(gsr.gsr_interface);
1430 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1435 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1436 __func__, sopt->sopt_name);
1437 return (EOPNOTSUPP);
1441 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1445 * Check if we are actually a member of this group.
1447 imo = inp_findmoptions(inp);
1448 idx = imo_match_group(imo, ifp, &gsa->sa);
1449 if (idx == -1 || imo->imo_mfilters == NULL) {
1450 error = EADDRNOTAVAIL;
1451 goto out_inp_locked;
1454 KASSERT(imo->imo_mfilters != NULL,
1455 ("%s: imo_mfilters not allocated", __func__));
1456 imf = &imo->imo_mfilters[idx];
1457 inm = imo->imo_membership[idx];
1460 * Attempting to use the delta-based API on an
1461 * non exclusive-mode membership is an error.
1463 fmode = imf->imf_st[0];
1464 if (fmode != MCAST_EXCLUDE) {
1466 goto out_inp_locked;
1470 * Deal with error cases up-front:
1471 * Asked to block, but already blocked; or
1472 * Asked to unblock, but nothing to unblock.
1473 * If adding a new block entry, allocate it.
1475 ims = imo_match_source(imo, idx, &ssa->sa);
1476 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1477 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
1478 inet_ntoa_r(ssa->sin.sin_addr, addrbuf),
1479 doblock ? "" : "not ");
1480 error = EADDRNOTAVAIL;
1481 goto out_inp_locked;
1484 INP_WLOCK_ASSERT(inp);
1487 * Begin state merge transaction at socket layer.
1490 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1491 ims = imf_graft(imf, fmode, &ssa->sin);
1495 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1496 error = imf_prune(imf, &ssa->sin);
1500 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1501 goto out_imf_rollback;
1505 * Begin state merge transaction at IGMP layer.
1509 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1510 error = inm_merge(inm, imf);
1512 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1513 goto out_in_multi_locked;
1516 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1517 error = igmp_change_state(inm);
1519 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1521 out_in_multi_locked:
1539 * Given an inpcb, return its multicast options structure pointer. Accepts
1540 * an unlocked inpcb pointer, but will return it locked. May sleep.
1542 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1543 * SMPng: NOTE: Returns with the INP write lock held.
1545 static struct ip_moptions *
1546 inp_findmoptions(struct inpcb *inp)
1548 struct ip_moptions *imo;
1549 struct in_multi **immp;
1550 struct in_mfilter *imfp;
1554 if (inp->inp_moptions != NULL)
1555 return (inp->inp_moptions);
1559 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1560 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1562 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1563 M_INMFILTER, M_WAITOK);
1565 imo->imo_multicast_ifp = NULL;
1566 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1567 imo->imo_multicast_vif = -1;
1568 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1569 imo->imo_multicast_loop = in_mcast_loop;
1570 imo->imo_num_memberships = 0;
1571 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1572 imo->imo_membership = immp;
1574 /* Initialize per-group source filters. */
1575 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1576 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1577 imo->imo_mfilters = imfp;
1580 if (inp->inp_moptions != NULL) {
1581 free(imfp, M_INMFILTER);
1582 free(immp, M_IPMOPTS);
1583 free(imo, M_IPMOPTS);
1584 return (inp->inp_moptions);
1586 inp->inp_moptions = imo;
1591 * Discard the IP multicast options (and source filters). To minimize
1592 * the amount of work done while holding locks such as the INP's
1593 * pcbinfo lock (which is used in the receive path), the free
1594 * operation is performed asynchronously in a separate task.
1596 * SMPng: NOTE: assumes INP write lock is held.
1599 inp_freemoptions(struct ip_moptions *imo)
1602 KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1604 STAILQ_INSERT_TAIL(&imo_gc_list, imo, imo_link);
1606 taskqueue_enqueue(taskqueue_thread, &imo_gc_task);
1610 inp_freemoptions_internal(struct ip_moptions *imo)
1612 struct in_mfilter *imf;
1613 size_t idx, nmships;
1615 nmships = imo->imo_num_memberships;
1616 for (idx = 0; idx < nmships; ++idx) {
1617 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1620 (void)in_leavegroup(imo->imo_membership[idx], imf);
1625 if (imo->imo_mfilters)
1626 free(imo->imo_mfilters, M_INMFILTER);
1627 free(imo->imo_membership, M_IPMOPTS);
1628 free(imo, M_IPMOPTS);
1632 inp_gcmoptions(void *context, int pending)
1634 struct ip_moptions *imo;
1637 while (!STAILQ_EMPTY(&imo_gc_list)) {
1638 imo = STAILQ_FIRST(&imo_gc_list);
1639 STAILQ_REMOVE_HEAD(&imo_gc_list, imo_link);
1641 inp_freemoptions_internal(imo);
1648 * Atomically get source filters on a socket for an IPv4 multicast group.
1649 * Called with INP lock held; returns with lock released.
1652 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1654 struct __msfilterreq msfr;
1657 struct ip_moptions *imo;
1658 struct in_mfilter *imf;
1659 struct ip_msource *ims;
1660 struct in_msource *lims;
1661 struct sockaddr_in *psin;
1662 struct sockaddr_storage *ptss;
1663 struct sockaddr_storage *tss;
1665 size_t idx, nsrcs, ncsrcs;
1667 INP_WLOCK_ASSERT(inp);
1669 imo = inp->inp_moptions;
1670 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1674 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1675 sizeof(struct __msfilterreq));
1679 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1682 ifp = ifnet_byindex(msfr.msfr_ifindex);
1689 * Lookup group on the socket.
1691 gsa = (sockunion_t *)&msfr.msfr_group;
1692 idx = imo_match_group(imo, ifp, &gsa->sa);
1693 if (idx == -1 || imo->imo_mfilters == NULL) {
1695 return (EADDRNOTAVAIL);
1697 imf = &imo->imo_mfilters[idx];
1700 * Ignore memberships which are in limbo.
1702 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1706 msfr.msfr_fmode = imf->imf_st[1];
1709 * If the user specified a buffer, copy out the source filter
1710 * entries to userland gracefully.
1711 * We only copy out the number of entries which userland
1712 * has asked for, but we always tell userland how big the
1713 * buffer really needs to be.
1715 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1716 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1718 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1719 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1720 M_TEMP, M_NOWAIT | M_ZERO);
1728 * Count number of sources in-mode at t0.
1729 * If buffer space exists and remains, copy out source entries.
1731 nsrcs = msfr.msfr_nsrcs;
1734 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1735 lims = (struct in_msource *)ims;
1736 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1737 lims->imsl_st[0] != imf->imf_st[0])
1740 if (tss != NULL && nsrcs > 0) {
1741 psin = (struct sockaddr_in *)ptss;
1742 psin->sin_family = AF_INET;
1743 psin->sin_len = sizeof(struct sockaddr_in);
1744 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1754 error = copyout(tss, msfr.msfr_srcs,
1755 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1761 msfr.msfr_nsrcs = ncsrcs;
1762 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1768 * Return the IP multicast options in response to user getsockopt().
1771 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1773 struct rm_priotracker in_ifa_tracker;
1774 struct ip_mreqn mreqn;
1775 struct ip_moptions *imo;
1777 struct in_ifaddr *ia;
1782 imo = inp->inp_moptions;
1784 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1785 * or is a divert socket, reject it.
1787 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1788 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1789 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1791 return (EOPNOTSUPP);
1795 switch (sopt->sopt_name) {
1796 case IP_MULTICAST_VIF:
1798 optval = imo->imo_multicast_vif;
1802 error = sooptcopyout(sopt, &optval, sizeof(int));
1805 case IP_MULTICAST_IF:
1806 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1808 ifp = imo->imo_multicast_ifp;
1809 if (!in_nullhost(imo->imo_multicast_addr)) {
1810 mreqn.imr_address = imo->imo_multicast_addr;
1811 } else if (ifp != NULL) {
1812 mreqn.imr_ifindex = ifp->if_index;
1813 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1816 IA_SIN(ia)->sin_addr;
1817 ifa_free(&ia->ia_ifa);
1822 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1823 error = sooptcopyout(sopt, &mreqn,
1824 sizeof(struct ip_mreqn));
1826 error = sooptcopyout(sopt, &mreqn.imr_address,
1827 sizeof(struct in_addr));
1831 case IP_MULTICAST_TTL:
1833 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1835 optval = coptval = imo->imo_multicast_ttl;
1837 if (sopt->sopt_valsize == sizeof(u_char))
1838 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1840 error = sooptcopyout(sopt, &optval, sizeof(int));
1843 case IP_MULTICAST_LOOP:
1845 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1847 optval = coptval = imo->imo_multicast_loop;
1849 if (sopt->sopt_valsize == sizeof(u_char))
1850 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1852 error = sooptcopyout(sopt, &optval, sizeof(int));
1857 error = EADDRNOTAVAIL;
1860 error = inp_get_source_filters(inp, sopt);
1866 error = ENOPROTOOPT;
1870 INP_UNLOCK_ASSERT(inp);
1876 * Look up the ifnet to use for a multicast group membership,
1877 * given the IPv4 address of an interface, and the IPv4 group address.
1879 * This routine exists to support legacy multicast applications
1880 * which do not understand that multicast memberships are scoped to
1881 * specific physical links in the networking stack, or which need
1882 * to join link-scope groups before IPv4 addresses are configured.
1884 * If inp is non-NULL, use this socket's current FIB number for any
1885 * required FIB lookup.
1886 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1887 * and use its ifp; usually, this points to the default next-hop.
1889 * If the FIB lookup fails, attempt to use the first non-loopback
1890 * interface with multicast capability in the system as a
1891 * last resort. The legacy IPv4 ASM API requires that we do
1892 * this in order to allow groups to be joined when the routing
1893 * table has not yet been populated during boot.
1895 * Returns NULL if no ifp could be found.
1897 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1898 * FUTURE: Implement IPv4 source-address selection.
1900 static struct ifnet *
1901 inp_lookup_mcast_ifp(const struct inpcb *inp,
1902 const struct sockaddr_in *gsin, const struct in_addr ina)
1904 struct rm_priotracker in_ifa_tracker;
1906 struct nhop4_basic nh4;
1909 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1910 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1911 ("%s: not multicast", __func__));
1914 if (!in_nullhost(ina)) {
1915 INADDR_TO_IFP(ina, ifp);
1917 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1918 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1921 struct in_ifaddr *ia;
1925 IN_IFADDR_RLOCK(&in_ifa_tracker);
1926 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1928 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1929 (mifp->if_flags & IFF_MULTICAST)) {
1934 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1942 * Join an IPv4 multicast group, possibly with a source.
1945 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1947 struct group_source_req gsr;
1948 sockunion_t *gsa, *ssa;
1950 struct in_mfilter *imf;
1951 struct ip_moptions *imo;
1952 struct in_multi *inm;
1953 struct in_msource *lims;
1957 char addrbuf[INET_ADDRSTRLEN];
1966 memset(&gsr, 0, sizeof(struct group_source_req));
1967 gsa = (sockunion_t *)&gsr.gsr_group;
1968 gsa->ss.ss_family = AF_UNSPEC;
1969 ssa = (sockunion_t *)&gsr.gsr_source;
1970 ssa->ss.ss_family = AF_UNSPEC;
1972 switch (sopt->sopt_name) {
1973 case IP_ADD_MEMBERSHIP:
1974 case IP_ADD_SOURCE_MEMBERSHIP: {
1975 struct ip_mreq_source mreqs;
1977 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1978 error = sooptcopyin(sopt, &mreqs,
1979 sizeof(struct ip_mreq),
1980 sizeof(struct ip_mreq));
1982 * Do argument switcharoo from ip_mreq into
1983 * ip_mreq_source to avoid using two instances.
1985 mreqs.imr_interface = mreqs.imr_sourceaddr;
1986 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1987 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1988 error = sooptcopyin(sopt, &mreqs,
1989 sizeof(struct ip_mreq_source),
1990 sizeof(struct ip_mreq_source));
1995 gsa->sin.sin_family = AF_INET;
1996 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1997 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1999 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2000 ssa->sin.sin_family = AF_INET;
2001 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2002 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2005 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2008 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2009 mreqs.imr_interface);
2010 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
2011 __func__, inet_ntoa_r(mreqs.imr_interface, addrbuf), ifp);
2015 case MCAST_JOIN_GROUP:
2016 case MCAST_JOIN_SOURCE_GROUP:
2017 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2018 error = sooptcopyin(sopt, &gsr,
2019 sizeof(struct group_req),
2020 sizeof(struct group_req));
2021 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2022 error = sooptcopyin(sopt, &gsr,
2023 sizeof(struct group_source_req),
2024 sizeof(struct group_source_req));
2029 if (gsa->sin.sin_family != AF_INET ||
2030 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2034 * Overwrite the port field if present, as the sockaddr
2035 * being copied in may be matched with a binary comparison.
2037 gsa->sin.sin_port = 0;
2038 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2039 if (ssa->sin.sin_family != AF_INET ||
2040 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2042 ssa->sin.sin_port = 0;
2045 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2048 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2049 return (EADDRNOTAVAIL);
2050 ifp = ifnet_byindex(gsr.gsr_interface);
2054 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2055 __func__, sopt->sopt_name);
2056 return (EOPNOTSUPP);
2060 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2061 return (EADDRNOTAVAIL);
2063 imo = inp_findmoptions(inp);
2064 idx = imo_match_group(imo, ifp, &gsa->sa);
2068 inm = imo->imo_membership[idx];
2069 imf = &imo->imo_mfilters[idx];
2070 if (ssa->ss.ss_family != AF_UNSPEC) {
2072 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2073 * is an error. On an existing inclusive membership,
2074 * it just adds the source to the filter list.
2076 if (imf->imf_st[1] != MCAST_INCLUDE) {
2078 goto out_inp_locked;
2081 * Throw out duplicates.
2083 * XXX FIXME: This makes a naive assumption that
2084 * even if entries exist for *ssa in this imf,
2085 * they will be rejected as dupes, even if they
2086 * are not valid in the current mode (in-mode).
2088 * in_msource is transactioned just as for anything
2089 * else in SSM -- but note naive use of inm_graft()
2090 * below for allocating new filter entries.
2092 * This is only an issue if someone mixes the
2093 * full-state SSM API with the delta-based API,
2094 * which is discouraged in the relevant RFCs.
2096 lims = imo_match_source(imo, idx, &ssa->sa);
2097 if (lims != NULL /*&&
2098 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2099 error = EADDRNOTAVAIL;
2100 goto out_inp_locked;
2104 * MCAST_JOIN_GROUP on an existing exclusive
2105 * membership is an error; return EADDRINUSE
2106 * to preserve 4.4BSD API idempotence, and
2107 * avoid tedious detour to code below.
2108 * NOTE: This is bending RFC 3678 a bit.
2110 * On an existing inclusive membership, this is also
2111 * an error; if you want to change filter mode,
2112 * you must use the userland API setsourcefilter().
2113 * XXX We don't reject this for imf in UNDEFINED
2114 * state at t1, because allocation of a filter
2115 * is atomic with allocation of a membership.
2118 if (imf->imf_st[1] == MCAST_EXCLUDE)
2120 goto out_inp_locked;
2125 * Begin state merge transaction at socket layer.
2127 INP_WLOCK_ASSERT(inp);
2130 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2131 error = imo_grow(imo);
2133 goto out_inp_locked;
2136 * Allocate the new slot upfront so we can deal with
2137 * grafting the new source filter in same code path
2138 * as for join-source on existing membership.
2140 idx = imo->imo_num_memberships;
2141 imo->imo_membership[idx] = NULL;
2142 imo->imo_num_memberships++;
2143 KASSERT(imo->imo_mfilters != NULL,
2144 ("%s: imf_mfilters vector was not allocated", __func__));
2145 imf = &imo->imo_mfilters[idx];
2146 KASSERT(RB_EMPTY(&imf->imf_sources),
2147 ("%s: imf_sources not empty", __func__));
2151 * Graft new source into filter list for this inpcb's
2152 * membership of the group. The in_multi may not have
2153 * been allocated yet if this is a new membership, however,
2154 * the in_mfilter slot will be allocated and must be initialized.
2156 * Note: Grafting of exclusive mode filters doesn't happen
2158 * XXX: Should check for non-NULL lims (node exists but may
2159 * not be in-mode) for interop with full-state API.
2161 if (ssa->ss.ss_family != AF_UNSPEC) {
2162 /* Membership starts in IN mode */
2164 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2165 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2167 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2169 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2171 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2177 /* No address specified; Membership starts in EX mode */
2179 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2180 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2185 * Begin state merge transaction at IGMP layer.
2190 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2193 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2198 imo->imo_membership[idx] = inm;
2200 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2201 error = inm_merge(inm, imf);
2203 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2205 goto out_in_multi_locked;
2207 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2208 error = igmp_change_state(inm);
2210 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2212 goto out_in_multi_locked;
2216 out_in_multi_locked:
2220 INP_WLOCK_ASSERT(inp);
2232 if (error && is_new) {
2233 imo->imo_membership[idx] = NULL;
2234 --imo->imo_num_memberships;
2243 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2246 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2248 struct group_source_req gsr;
2249 struct ip_mreq_source mreqs;
2250 sockunion_t *gsa, *ssa;
2252 struct in_mfilter *imf;
2253 struct ip_moptions *imo;
2254 struct in_msource *ims;
2255 struct in_multi *inm;
2257 int error, is_final;
2259 char addrbuf[INET_ADDRSTRLEN];
2266 memset(&gsr, 0, sizeof(struct group_source_req));
2267 gsa = (sockunion_t *)&gsr.gsr_group;
2268 gsa->ss.ss_family = AF_UNSPEC;
2269 ssa = (sockunion_t *)&gsr.gsr_source;
2270 ssa->ss.ss_family = AF_UNSPEC;
2272 switch (sopt->sopt_name) {
2273 case IP_DROP_MEMBERSHIP:
2274 case IP_DROP_SOURCE_MEMBERSHIP:
2275 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2276 error = sooptcopyin(sopt, &mreqs,
2277 sizeof(struct ip_mreq),
2278 sizeof(struct ip_mreq));
2280 * Swap interface and sourceaddr arguments,
2281 * as ip_mreq and ip_mreq_source are laid
2284 mreqs.imr_interface = mreqs.imr_sourceaddr;
2285 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2286 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2287 error = sooptcopyin(sopt, &mreqs,
2288 sizeof(struct ip_mreq_source),
2289 sizeof(struct ip_mreq_source));
2294 gsa->sin.sin_family = AF_INET;
2295 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2296 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2298 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2299 ssa->sin.sin_family = AF_INET;
2300 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2301 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2305 * Attempt to look up hinted ifp from interface address.
2306 * Fallthrough with null ifp iff lookup fails, to
2307 * preserve 4.4BSD mcast API idempotence.
2308 * XXX NOTE WELL: The RFC 3678 API is preferred because
2309 * using an IPv4 address as a key is racy.
2311 if (!in_nullhost(mreqs.imr_interface))
2312 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2314 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
2315 __func__, inet_ntoa_r(mreqs.imr_interface, addrbuf), ifp);
2319 case MCAST_LEAVE_GROUP:
2320 case MCAST_LEAVE_SOURCE_GROUP:
2321 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2322 error = sooptcopyin(sopt, &gsr,
2323 sizeof(struct group_req),
2324 sizeof(struct group_req));
2325 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2326 error = sooptcopyin(sopt, &gsr,
2327 sizeof(struct group_source_req),
2328 sizeof(struct group_source_req));
2333 if (gsa->sin.sin_family != AF_INET ||
2334 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2337 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2338 if (ssa->sin.sin_family != AF_INET ||
2339 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2343 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2344 return (EADDRNOTAVAIL);
2346 ifp = ifnet_byindex(gsr.gsr_interface);
2349 return (EADDRNOTAVAIL);
2353 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2354 __func__, sopt->sopt_name);
2355 return (EOPNOTSUPP);
2359 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2363 * Find the membership in the membership array.
2365 imo = inp_findmoptions(inp);
2366 idx = imo_match_group(imo, ifp, &gsa->sa);
2368 error = EADDRNOTAVAIL;
2369 goto out_inp_locked;
2371 inm = imo->imo_membership[idx];
2372 imf = &imo->imo_mfilters[idx];
2374 if (ssa->ss.ss_family != AF_UNSPEC)
2378 * Begin state merge transaction at socket layer.
2380 INP_WLOCK_ASSERT(inp);
2383 * If we were instructed only to leave a given source, do so.
2384 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2389 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2390 error = EADDRNOTAVAIL;
2391 goto out_inp_locked;
2393 ims = imo_match_source(imo, idx, &ssa->sa);
2395 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
2396 inet_ntoa_r(ssa->sin.sin_addr, addrbuf), "not ");
2397 error = EADDRNOTAVAIL;
2398 goto out_inp_locked;
2400 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2401 error = imf_prune(imf, &ssa->sin);
2403 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2405 goto out_inp_locked;
2410 * Begin state merge transaction at IGMP layer.
2416 * Give up the multicast address record to which
2417 * the membership points.
2419 (void)in_leavegroup_locked(inm, imf);
2421 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2422 error = inm_merge(inm, imf);
2424 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2426 goto out_in_multi_locked;
2429 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2430 error = igmp_change_state(inm);
2432 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2437 out_in_multi_locked:
2449 /* Remove the gap in the membership and filter array. */
2450 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2451 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2452 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2454 imo->imo_num_memberships--;
2463 * Select the interface for transmitting IPv4 multicast datagrams.
2465 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2466 * may be passed to this socket option. An address of INADDR_ANY or an
2467 * interface index of 0 is used to remove a previous selection.
2468 * When no interface is selected, one is chosen for every send.
2471 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2473 struct in_addr addr;
2474 struct ip_mreqn mreqn;
2476 struct ip_moptions *imo;
2479 char addrbuf[INET_ADDRSTRLEN];
2482 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2484 * An interface index was specified using the
2485 * Linux-derived ip_mreqn structure.
2487 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2488 sizeof(struct ip_mreqn));
2492 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2495 if (mreqn.imr_ifindex == 0) {
2498 ifp = ifnet_byindex(mreqn.imr_ifindex);
2500 return (EADDRNOTAVAIL);
2504 * An interface was specified by IPv4 address.
2505 * This is the traditional BSD usage.
2507 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2508 sizeof(struct in_addr));
2511 if (in_nullhost(addr)) {
2514 INADDR_TO_IFP(addr, ifp);
2516 return (EADDRNOTAVAIL);
2518 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp,
2519 inet_ntoa_r(addr, addrbuf));
2522 /* Reject interfaces which do not support multicast. */
2523 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2524 return (EOPNOTSUPP);
2526 imo = inp_findmoptions(inp);
2527 imo->imo_multicast_ifp = ifp;
2528 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2535 * Atomically set source filters on a socket for an IPv4 multicast group.
2537 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2540 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2542 struct __msfilterreq msfr;
2545 struct in_mfilter *imf;
2546 struct ip_moptions *imo;
2547 struct in_multi *inm;
2551 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2552 sizeof(struct __msfilterreq));
2556 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2559 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2560 msfr.msfr_fmode != MCAST_INCLUDE))
2563 if (msfr.msfr_group.ss_family != AF_INET ||
2564 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2567 gsa = (sockunion_t *)&msfr.msfr_group;
2568 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2571 gsa->sin.sin_port = 0; /* ignore port */
2573 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2574 return (EADDRNOTAVAIL);
2576 ifp = ifnet_byindex(msfr.msfr_ifindex);
2578 return (EADDRNOTAVAIL);
2581 * Take the INP write lock.
2582 * Check if this socket is a member of this group.
2584 imo = inp_findmoptions(inp);
2585 idx = imo_match_group(imo, ifp, &gsa->sa);
2586 if (idx == -1 || imo->imo_mfilters == NULL) {
2587 error = EADDRNOTAVAIL;
2588 goto out_inp_locked;
2590 inm = imo->imo_membership[idx];
2591 imf = &imo->imo_mfilters[idx];
2594 * Begin state merge transaction at socket layer.
2596 INP_WLOCK_ASSERT(inp);
2598 imf->imf_st[1] = msfr.msfr_fmode;
2601 * Apply any new source filters, if present.
2602 * Make a copy of the user-space source vector so
2603 * that we may copy them with a single copyin. This
2604 * allows us to deal with page faults up-front.
2606 if (msfr.msfr_nsrcs > 0) {
2607 struct in_msource *lims;
2608 struct sockaddr_in *psin;
2609 struct sockaddr_storage *kss, *pkss;
2614 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2615 __func__, (unsigned long)msfr.msfr_nsrcs);
2616 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2618 error = copyin(msfr.msfr_srcs, kss,
2619 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2628 * Mark all source filters as UNDEFINED at t1.
2629 * Restore new group filter mode, as imf_leave()
2630 * will set it to INCLUDE.
2633 imf->imf_st[1] = msfr.msfr_fmode;
2636 * Update socket layer filters at t1, lazy-allocating
2637 * new entries. This saves a bunch of memory at the
2638 * cost of one RB_FIND() per source entry; duplicate
2639 * entries in the msfr_nsrcs vector are ignored.
2640 * If we encounter an error, rollback transaction.
2642 * XXX This too could be replaced with a set-symmetric
2643 * difference like loop to avoid walking from root
2644 * every time, as the key space is common.
2646 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2647 psin = (struct sockaddr_in *)pkss;
2648 if (psin->sin_family != AF_INET) {
2649 error = EAFNOSUPPORT;
2652 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2656 error = imf_get_source(imf, psin, &lims);
2659 lims->imsl_st[1] = imf->imf_st[1];
2665 goto out_imf_rollback;
2667 INP_WLOCK_ASSERT(inp);
2671 * Begin state merge transaction at IGMP layer.
2673 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2674 error = inm_merge(inm, imf);
2676 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2677 goto out_in_multi_locked;
2680 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2681 error = igmp_change_state(inm);
2683 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2685 out_in_multi_locked:
2703 * Set the IP multicast options in response to user setsockopt().
2705 * Many of the socket options handled in this function duplicate the
2706 * functionality of socket options in the regular unicast API. However,
2707 * it is not possible to merge the duplicate code, because the idempotence
2708 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2709 * the effects of these options must be treated as separate and distinct.
2711 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2712 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2713 * is refactored to no longer use vifs.
2716 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2718 struct ip_moptions *imo;
2724 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2725 * or is a divert socket, reject it.
2727 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2728 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2729 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2730 return (EOPNOTSUPP);
2732 switch (sopt->sopt_name) {
2733 case IP_MULTICAST_VIF: {
2736 * Select a multicast VIF for transmission.
2737 * Only useful if multicast forwarding is active.
2739 if (legal_vif_num == NULL) {
2743 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2746 if (!legal_vif_num(vifi) && (vifi != -1)) {
2750 imo = inp_findmoptions(inp);
2751 imo->imo_multicast_vif = vifi;
2756 case IP_MULTICAST_IF:
2757 error = inp_set_multicast_if(inp, sopt);
2760 case IP_MULTICAST_TTL: {
2764 * Set the IP time-to-live for outgoing multicast packets.
2765 * The original multicast API required a char argument,
2766 * which is inconsistent with the rest of the socket API.
2767 * We allow either a char or an int.
2769 if (sopt->sopt_valsize == sizeof(u_char)) {
2770 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2777 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2787 imo = inp_findmoptions(inp);
2788 imo->imo_multicast_ttl = ttl;
2793 case IP_MULTICAST_LOOP: {
2797 * Set the loopback flag for outgoing multicast packets.
2798 * Must be zero or one. The original multicast API required a
2799 * char argument, which is inconsistent with the rest
2800 * of the socket API. We allow either a char or an int.
2802 if (sopt->sopt_valsize == sizeof(u_char)) {
2803 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2810 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2814 loop = (u_char)iloop;
2816 imo = inp_findmoptions(inp);
2817 imo->imo_multicast_loop = !!loop;
2822 case IP_ADD_MEMBERSHIP:
2823 case IP_ADD_SOURCE_MEMBERSHIP:
2824 case MCAST_JOIN_GROUP:
2825 case MCAST_JOIN_SOURCE_GROUP:
2826 error = inp_join_group(inp, sopt);
2829 case IP_DROP_MEMBERSHIP:
2830 case IP_DROP_SOURCE_MEMBERSHIP:
2831 case MCAST_LEAVE_GROUP:
2832 case MCAST_LEAVE_SOURCE_GROUP:
2833 error = inp_leave_group(inp, sopt);
2836 case IP_BLOCK_SOURCE:
2837 case IP_UNBLOCK_SOURCE:
2838 case MCAST_BLOCK_SOURCE:
2839 case MCAST_UNBLOCK_SOURCE:
2840 error = inp_block_unblock_source(inp, sopt);
2844 error = inp_set_source_filters(inp, sopt);
2852 INP_UNLOCK_ASSERT(inp);
2858 * Expose IGMP's multicast filter mode and source list(s) to userland,
2859 * keyed by (ifindex, group).
2860 * The filter mode is written out as a uint32_t, followed by
2861 * 0..n of struct in_addr.
2862 * For use by ifmcstat(8).
2863 * SMPng: NOTE: unlocked read of ifindex space.
2866 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2868 struct in_addr src, group;
2870 struct ifmultiaddr *ifma;
2871 struct in_multi *inm;
2872 struct ip_msource *ims;
2876 uint32_t fmode, ifindex;
2878 char addrbuf[INET_ADDRSTRLEN];
2884 if (req->newptr != NULL)
2891 if (ifindex <= 0 || ifindex > V_if_index) {
2892 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2897 group.s_addr = name[1];
2898 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2899 CTR2(KTR_IGMPV3, "%s: group %s is not multicast",
2900 __func__, inet_ntoa_r(group, addrbuf));
2904 ifp = ifnet_byindex(ifindex);
2906 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2911 retval = sysctl_wire_old_buffer(req,
2912 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2919 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2920 if (ifma->ifma_addr->sa_family != AF_INET ||
2921 ifma->ifma_protospec == NULL)
2923 inm = (struct in_multi *)ifma->ifma_protospec;
2924 if (!in_hosteq(inm->inm_addr, group))
2926 fmode = inm->inm_st[1].iss_fmode;
2927 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2930 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2933 ina.s_addr = htonl(ims->ims_haddr);
2934 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2935 inet_ntoa_r(ina, addrbuf));
2938 * Only copy-out sources which are in-mode.
2940 if (fmode != ims_get_mode(inm, ims, 1)) {
2941 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2945 src.s_addr = htonl(ims->ims_haddr);
2946 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2951 IF_ADDR_RUNLOCK(ifp);
2958 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2960 static const char *inm_modestrs[] = { "un", "in", "ex" };
2963 inm_mode_str(const int mode)
2966 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2967 return (inm_modestrs[mode]);
2971 static const char *inm_statestrs[] = {
2984 inm_state_str(const int state)
2987 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2988 return (inm_statestrs[state]);
2993 * Dump an in_multi structure to the console.
2996 inm_print(const struct in_multi *inm)
2999 char addrbuf[INET_ADDRSTRLEN];
3001 if ((ktr_mask & KTR_IGMPV3) == 0)
3004 printf("%s: --- begin inm %p ---\n", __func__, inm);
3005 printf("addr %s ifp %p(%s) ifma %p\n",
3006 inet_ntoa_r(inm->inm_addr, addrbuf),
3008 inm->inm_ifp->if_xname,
3010 printf("timer %u state %s refcount %u scq.len %u\n",
3012 inm_state_str(inm->inm_state),
3014 inm->inm_scq.mq_len);
3015 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3020 for (t = 0; t < 2; t++) {
3021 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3022 inm_mode_str(inm->inm_st[t].iss_fmode),
3023 inm->inm_st[t].iss_asm,
3024 inm->inm_st[t].iss_ex,
3025 inm->inm_st[t].iss_in,
3026 inm->inm_st[t].iss_rec);
3028 printf("%s: --- end inm %p ---\n", __func__, inm);
3031 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3034 inm_print(const struct in_multi *inm)
3039 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3041 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);