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>
49 #include <sys/taskqueue.h>
53 #include <net/if_dl.h>
54 #include <net/route.h>
57 #include <netinet/in.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/in_pcb.h>
60 #include <netinet/in_var.h>
61 #include <netinet/ip_var.h>
62 #include <netinet/igmp_var.h>
65 #define KTR_IGMPV3 KTR_INET
68 #ifndef __SOCKUNION_DECLARED
70 struct sockaddr_storage ss;
72 struct sockaddr_dl sdl;
73 struct sockaddr_in sin;
75 typedef union sockunion sockunion_t;
76 #define __SOCKUNION_DECLARED
77 #endif /* __SOCKUNION_DECLARED */
79 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
80 "IPv4 multicast PCB-layer source filter");
81 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
82 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
83 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
84 "IPv4 multicast IGMP-layer source filter");
88 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
89 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
90 * it can be taken by code in net/if.c also.
91 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
93 * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly
94 * any need for in_multi itself to be virtualized -- it is bound to an ifp
95 * anyway no matter what happens.
97 struct mtx in_multi_mtx;
98 MTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF);
101 * Functions with non-static linkage defined in this file should be
102 * declared in in_var.h:
107 * in_joingroup_locked()
109 * in_leavegroup_locked()
115 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
118 static void imf_commit(struct in_mfilter *);
119 static int imf_get_source(struct in_mfilter *imf,
120 const struct sockaddr_in *psin,
121 struct in_msource **);
122 static struct in_msource *
123 imf_graft(struct in_mfilter *, const uint8_t,
124 const struct sockaddr_in *);
125 static void imf_leave(struct in_mfilter *);
126 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
127 static void imf_purge(struct in_mfilter *);
128 static void imf_rollback(struct in_mfilter *);
129 static void imf_reap(struct in_mfilter *);
130 static int imo_grow(struct ip_moptions *);
131 static size_t imo_match_group(const struct ip_moptions *,
132 const struct ifnet *, const struct sockaddr *);
133 static struct in_msource *
134 imo_match_source(const struct ip_moptions *, const size_t,
135 const struct sockaddr *);
136 static void ims_merge(struct ip_msource *ims,
137 const struct in_msource *lims, const int rollback);
138 static int in_getmulti(struct ifnet *, const struct in_addr *,
140 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
141 const int noalloc, struct ip_msource **pims);
142 static int inm_is_ifp_detached(const struct in_multi *);
143 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
144 static void inm_purge(struct in_multi *);
145 static void inm_reap(struct in_multi *);
146 static struct ip_moptions *
147 inp_findmoptions(struct inpcb *);
148 static void inp_freemoptions_internal(struct ip_moptions *);
149 static void inp_gcmoptions(void *, int);
150 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
151 static int inp_join_group(struct inpcb *, struct sockopt *);
152 static int inp_leave_group(struct inpcb *, struct sockopt *);
153 static struct ifnet *
154 inp_lookup_mcast_ifp(const struct inpcb *,
155 const struct sockaddr_in *, const struct in_addr);
156 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
157 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
158 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
159 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
161 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
164 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
165 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
166 CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxgrpsrc, 0,
167 "Max source filters per group");
168 TUNABLE_ULONG("net.inet.ip.mcast.maxgrpsrc", &in_mcast_maxgrpsrc);
170 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
171 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
172 CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxsocksrc, 0,
173 "Max source filters per socket");
174 TUNABLE_ULONG("net.inet.ip.mcast.maxsocksrc", &in_mcast_maxsocksrc);
176 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
177 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
178 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
179 TUNABLE_INT("net.inet.ip.mcast.loop", &in_mcast_loop);
181 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
182 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
183 "Per-interface stack-wide source filters");
185 static STAILQ_HEAD(, ip_moptions) imo_gc_list =
186 STAILQ_HEAD_INITIALIZER(imo_gc_list);
187 static struct task imo_gc_task = TASK_INITIALIZER(0, inp_gcmoptions, NULL);
190 * Inline function which wraps assertions for a valid ifp.
191 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
195 inm_is_ifp_detached(const struct in_multi *inm)
199 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
200 ifp = inm->inm_ifma->ifma_ifp;
203 * Sanity check that netinet's notion of ifp is the
206 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
209 return (ifp == NULL);
213 * Initialize an in_mfilter structure to a known state at t0, t1
214 * with an empty source filter list.
217 imf_init(struct in_mfilter *imf, const int st0, const int st1)
219 memset(imf, 0, sizeof(struct in_mfilter));
220 RB_INIT(&imf->imf_sources);
221 imf->imf_st[0] = st0;
222 imf->imf_st[1] = st1;
226 * Resize the ip_moptions vector to the next power-of-two minus 1.
227 * May be called with locks held; do not sleep.
230 imo_grow(struct ip_moptions *imo)
232 struct in_multi **nmships;
233 struct in_multi **omships;
234 struct in_mfilter *nmfilters;
235 struct in_mfilter *omfilters;
242 omships = imo->imo_membership;
243 omfilters = imo->imo_mfilters;
244 oldmax = imo->imo_max_memberships;
245 newmax = ((oldmax + 1) * 2) - 1;
247 if (newmax <= IP_MAX_MEMBERSHIPS) {
248 nmships = (struct in_multi **)realloc(omships,
249 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
250 nmfilters = (struct in_mfilter *)realloc(omfilters,
251 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
252 if (nmships != NULL && nmfilters != NULL) {
253 /* Initialize newly allocated source filter heads. */
254 for (idx = oldmax; idx < newmax; idx++) {
255 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
258 imo->imo_max_memberships = newmax;
259 imo->imo_membership = nmships;
260 imo->imo_mfilters = nmfilters;
264 if (nmships == NULL || nmfilters == NULL) {
266 free(nmships, M_IPMOPTS);
267 if (nmfilters != NULL)
268 free(nmfilters, M_INMFILTER);
269 return (ETOOMANYREFS);
276 * Find an IPv4 multicast group entry for this ip_moptions instance
277 * which matches the specified group, and optionally an interface.
278 * Return its index into the array, or -1 if not found.
281 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
282 const struct sockaddr *group)
284 const struct sockaddr_in *gsin;
285 struct in_multi **pinm;
289 gsin = (const struct sockaddr_in *)group;
291 /* The imo_membership array may be lazy allocated. */
292 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
295 nmships = imo->imo_num_memberships;
296 pinm = &imo->imo_membership[0];
297 for (idx = 0; idx < nmships; idx++, pinm++) {
300 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
301 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
312 * Find an IPv4 multicast source entry for this imo which matches
313 * the given group index for this socket, and source address.
315 * NOTE: This does not check if the entry is in-mode, merely if
316 * it exists, which may not be the desired behaviour.
318 static struct in_msource *
319 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
320 const struct sockaddr *src)
322 struct ip_msource find;
323 struct in_mfilter *imf;
324 struct ip_msource *ims;
325 const sockunion_t *psa;
327 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
328 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
329 ("%s: invalid index %d\n", __func__, (int)gidx));
331 /* The imo_mfilters array may be lazy allocated. */
332 if (imo->imo_mfilters == NULL)
334 imf = &imo->imo_mfilters[gidx];
336 /* Source trees are keyed in host byte order. */
337 psa = (const sockunion_t *)src;
338 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
339 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
341 return ((struct in_msource *)ims);
345 * Perform filtering for multicast datagrams on a socket by group and source.
347 * Returns 0 if a datagram should be allowed through, or various error codes
348 * if the socket was not a member of the group, or the source was muted, etc.
351 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
352 const struct sockaddr *group, const struct sockaddr *src)
355 struct in_msource *ims;
358 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
360 gidx = imo_match_group(imo, ifp, group);
362 return (MCAST_NOTGMEMBER);
365 * Check if the source was included in an (S,G) join.
366 * Allow reception on exclusive memberships by default,
367 * reject reception on inclusive memberships by default.
368 * Exclude source only if an in-mode exclude filter exists.
369 * Include source only if an in-mode include filter exists.
370 * NOTE: We are comparing group state here at IGMP t1 (now)
371 * with socket-layer t0 (since last downcall).
373 mode = imo->imo_mfilters[gidx].imf_st[1];
374 ims = imo_match_source(imo, gidx, src);
376 if ((ims == NULL && mode == MCAST_INCLUDE) ||
377 (ims != NULL && ims->imsl_st[0] != mode))
378 return (MCAST_NOTSMEMBER);
384 * Find and return a reference to an in_multi record for (ifp, group),
385 * and bump its reference count.
386 * If one does not exist, try to allocate it, and update link-layer multicast
387 * filters on ifp to listen for group.
388 * Assumes the IN_MULTI lock is held across the call.
389 * Return 0 if successful, otherwise return an appropriate error code.
392 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
393 struct in_multi **pinm)
395 struct sockaddr_in gsin;
396 struct ifmultiaddr *ifma;
397 struct in_ifinfo *ii;
398 struct in_multi *inm;
401 IN_MULTI_LOCK_ASSERT();
403 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
405 inm = inm_lookup(ifp, *group);
408 * If we already joined this group, just bump the
409 * refcount and return it.
411 KASSERT(inm->inm_refcount >= 1,
412 ("%s: bad refcount %d", __func__, inm->inm_refcount));
418 memset(&gsin, 0, sizeof(gsin));
419 gsin.sin_family = AF_INET;
420 gsin.sin_len = sizeof(struct sockaddr_in);
421 gsin.sin_addr = *group;
424 * Check if a link-layer group is already associated
425 * with this network-layer group on the given ifnet.
427 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
431 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
435 * If something other than netinet is occupying the link-layer
436 * group, print a meaningful error message and back out of
438 * Otherwise, bump the refcount on the existing network-layer
439 * group association and return it.
441 if (ifma->ifma_protospec != NULL) {
442 inm = (struct in_multi *)ifma->ifma_protospec;
444 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
446 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
447 ("%s: ifma not AF_INET", __func__));
448 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
449 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
450 !in_hosteq(inm->inm_addr, *group))
451 panic("%s: ifma %p is inconsistent with %p (%s)",
452 __func__, ifma, inm, inet_ntoa(*group));
460 IF_ADDR_LOCK_ASSERT(ifp);
463 * A new in_multi record is needed; allocate and initialize it.
464 * We DO NOT perform an IGMP join as the in_ layer may need to
465 * push an initial source list down to IGMP to support SSM.
467 * The initial source filter state is INCLUDE, {} as per the RFC.
469 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
471 if_delmulti_ifma(ifma);
475 inm->inm_addr = *group;
477 inm->inm_igi = ii->ii_igmp;
478 inm->inm_ifma = ifma;
479 inm->inm_refcount = 1;
480 inm->inm_state = IGMP_NOT_MEMBER;
483 * Pending state-changes per group are subject to a bounds check.
485 IFQ_SET_MAXLEN(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
487 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
488 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
489 RB_INIT(&inm->inm_srcs);
491 ifma->ifma_protospec = inm;
500 * Drop a reference to an in_multi record.
502 * If the refcount drops to 0, free the in_multi record and
503 * delete the underlying link-layer membership.
506 inm_release_locked(struct in_multi *inm)
508 struct ifmultiaddr *ifma;
510 IN_MULTI_LOCK_ASSERT();
512 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
514 if (--inm->inm_refcount > 0) {
515 CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__,
520 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
522 ifma = inm->inm_ifma;
524 /* XXX this access is not covered by IF_ADDR_LOCK */
525 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
526 KASSERT(ifma->ifma_protospec == inm,
527 ("%s: ifma_protospec != inm", __func__));
528 ifma->ifma_protospec = NULL;
532 free(inm, M_IPMADDR);
534 if_delmulti_ifma(ifma);
538 * Clear recorded source entries for a group.
539 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
540 * FIXME: Should reap.
543 inm_clear_recorded(struct in_multi *inm)
545 struct ip_msource *ims;
547 IN_MULTI_LOCK_ASSERT();
549 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
552 --inm->inm_st[1].iss_rec;
555 KASSERT(inm->inm_st[1].iss_rec == 0,
556 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
560 * Record a source as pending for a Source-Group IGMPv3 query.
561 * This lives here as it modifies the shared tree.
563 * inm is the group descriptor.
564 * naddr is the address of the source to record in network-byte order.
566 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
567 * lazy-allocate a source node in response to an SG query.
568 * Otherwise, no allocation is performed. This saves some memory
569 * with the trade-off that the source will not be reported to the
570 * router if joined in the window between the query response and
571 * the group actually being joined on the local host.
573 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
574 * This turns off the allocation of a recorded source entry if
575 * the group has not been joined.
577 * Return 0 if the source didn't exist or was already marked as recorded.
578 * Return 1 if the source was marked as recorded by this function.
579 * Return <0 if any error occured (negated errno code).
582 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
584 struct ip_msource find;
585 struct ip_msource *ims, *nims;
587 IN_MULTI_LOCK_ASSERT();
589 find.ims_haddr = ntohl(naddr);
590 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
591 if (ims && ims->ims_stp)
594 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
596 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
600 nims->ims_haddr = find.ims_haddr;
601 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
607 * Mark the source as recorded and update the recorded
611 ++inm->inm_st[1].iss_rec;
617 * Return a pointer to an in_msource owned by an in_mfilter,
618 * given its source address.
619 * Lazy-allocate if needed. If this is a new entry its filter state is
622 * imf is the filter set being modified.
623 * haddr is the source address in *host* byte-order.
625 * SMPng: May be called with locks held; malloc must not block.
628 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
629 struct in_msource **plims)
631 struct ip_msource find;
632 struct ip_msource *ims, *nims;
633 struct in_msource *lims;
640 /* key is host byte order */
641 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
642 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
643 lims = (struct in_msource *)ims;
645 if (imf->imf_nsrc == in_mcast_maxsocksrc)
647 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
651 lims = (struct in_msource *)nims;
652 lims->ims_haddr = find.ims_haddr;
653 lims->imsl_st[0] = MCAST_UNDEFINED;
654 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
664 * Graft a source entry into an existing socket-layer filter set,
665 * maintaining any required invariants and checking allocations.
667 * The source is marked as being in the new filter mode at t1.
669 * Return the pointer to the new node, otherwise return NULL.
671 static struct in_msource *
672 imf_graft(struct in_mfilter *imf, const uint8_t st1,
673 const struct sockaddr_in *psin)
675 struct ip_msource *nims;
676 struct in_msource *lims;
678 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
682 lims = (struct in_msource *)nims;
683 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
684 lims->imsl_st[0] = MCAST_UNDEFINED;
685 lims->imsl_st[1] = st1;
686 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
693 * Prune a source entry from an existing socket-layer filter set,
694 * maintaining any required invariants and checking allocations.
696 * The source is marked as being left at t1, it is not freed.
698 * Return 0 if no error occurred, otherwise return an errno value.
701 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
703 struct ip_msource find;
704 struct ip_msource *ims;
705 struct in_msource *lims;
707 /* key is host byte order */
708 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
709 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
712 lims = (struct in_msource *)ims;
713 lims->imsl_st[1] = MCAST_UNDEFINED;
718 * Revert socket-layer filter set deltas at t1 to t0 state.
721 imf_rollback(struct in_mfilter *imf)
723 struct ip_msource *ims, *tims;
724 struct in_msource *lims;
726 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
727 lims = (struct in_msource *)ims;
728 if (lims->imsl_st[0] == lims->imsl_st[1]) {
729 /* no change at t1 */
731 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
732 /* revert change to existing source at t1 */
733 lims->imsl_st[1] = lims->imsl_st[0];
735 /* revert source added t1 */
736 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
737 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
738 free(ims, M_INMFILTER);
742 imf->imf_st[1] = imf->imf_st[0];
746 * Mark socket-layer filter set as INCLUDE {} at t1.
749 imf_leave(struct in_mfilter *imf)
751 struct ip_msource *ims;
752 struct in_msource *lims;
754 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
755 lims = (struct in_msource *)ims;
756 lims->imsl_st[1] = MCAST_UNDEFINED;
758 imf->imf_st[1] = MCAST_INCLUDE;
762 * Mark socket-layer filter set deltas as committed.
765 imf_commit(struct in_mfilter *imf)
767 struct ip_msource *ims;
768 struct in_msource *lims;
770 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
771 lims = (struct in_msource *)ims;
772 lims->imsl_st[0] = lims->imsl_st[1];
774 imf->imf_st[0] = imf->imf_st[1];
778 * Reap unreferenced sources from socket-layer filter set.
781 imf_reap(struct in_mfilter *imf)
783 struct ip_msource *ims, *tims;
784 struct in_msource *lims;
786 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
787 lims = (struct in_msource *)ims;
788 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
789 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
790 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
791 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
792 free(ims, M_INMFILTER);
799 * Purge socket-layer filter set.
802 imf_purge(struct in_mfilter *imf)
804 struct ip_msource *ims, *tims;
806 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
807 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
808 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
809 free(ims, M_INMFILTER);
812 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
813 KASSERT(RB_EMPTY(&imf->imf_sources),
814 ("%s: imf_sources not empty", __func__));
818 * Look up a source filter entry for a multicast group.
820 * inm is the group descriptor to work with.
821 * haddr is the host-byte-order IPv4 address to look up.
822 * noalloc may be non-zero to suppress allocation of sources.
823 * *pims will be set to the address of the retrieved or allocated source.
825 * SMPng: NOTE: may be called with locks held.
826 * Return 0 if successful, otherwise return a non-zero error code.
829 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
830 const int noalloc, struct ip_msource **pims)
832 struct ip_msource find;
833 struct ip_msource *ims, *nims;
838 find.ims_haddr = haddr;
839 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
840 if (ims == NULL && !noalloc) {
841 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
843 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
847 nims->ims_haddr = haddr;
848 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
852 ia.s_addr = htonl(haddr);
853 CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__,
863 * Merge socket-layer source into IGMP-layer source.
864 * If rollback is non-zero, perform the inverse of the merge.
867 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
870 int n = rollback ? -1 : 1;
874 ia.s_addr = htonl(ims->ims_haddr);
877 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
878 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s",
879 __func__, n, inet_ntoa(ia));
880 ims->ims_st[1].ex -= n;
881 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
882 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s",
883 __func__, n, inet_ntoa(ia));
884 ims->ims_st[1].in -= n;
887 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
888 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s",
889 __func__, n, inet_ntoa(ia));
890 ims->ims_st[1].ex += n;
891 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
892 CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s",
893 __func__, n, inet_ntoa(ia));
894 ims->ims_st[1].in += n;
899 * Atomically update the global in_multi state, when a membership's
900 * filter list is being updated in any way.
902 * imf is the per-inpcb-membership group filter pointer.
903 * A fake imf may be passed for in-kernel consumers.
905 * XXX This is a candidate for a set-symmetric-difference style loop
906 * which would eliminate the repeated lookup from root of ims nodes,
907 * as they share the same key space.
909 * If any error occurred this function will back out of refcounts
910 * and return a non-zero value.
913 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
915 struct ip_msource *ims, *nims;
916 struct in_msource *lims;
925 * Update the source filters first, as this may fail.
926 * Maintain count of in-mode filters at t0, t1. These are
927 * used to work out if we transition into ASM mode or not.
928 * Maintain a count of source filters whose state was
929 * actually modified by this operation.
931 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
932 lims = (struct in_msource *)ims;
933 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
934 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
935 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
936 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
940 ims_merge(nims, lims, 0);
943 struct ip_msource *bims;
945 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
946 lims = (struct in_msource *)ims;
947 if (lims->imsl_st[0] == lims->imsl_st[1])
949 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
952 ims_merge(bims, lims, 1);
957 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
958 __func__, nsrc0, nsrc1);
960 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
961 if (imf->imf_st[0] == imf->imf_st[1] &&
962 imf->imf_st[1] == MCAST_INCLUDE) {
964 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
965 --inm->inm_st[1].iss_in;
969 /* Handle filter mode transition on socket. */
970 if (imf->imf_st[0] != imf->imf_st[1]) {
971 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
972 __func__, imf->imf_st[0], imf->imf_st[1]);
974 if (imf->imf_st[0] == 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[0] == MCAST_INCLUDE) {
978 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
979 --inm->inm_st[1].iss_in;
982 if (imf->imf_st[1] == MCAST_EXCLUDE) {
983 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
984 inm->inm_st[1].iss_ex++;
985 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
986 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
987 inm->inm_st[1].iss_in++;
992 * Track inm filter state in terms of listener counts.
993 * If there are any exclusive listeners, stack-wide
994 * membership is exclusive.
995 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
996 * If no listeners remain, state is undefined at t1,
997 * and the IGMP lifecycle for this group should finish.
999 if (inm->inm_st[1].iss_ex > 0) {
1000 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1001 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1002 } else if (inm->inm_st[1].iss_in > 0) {
1003 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1004 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1006 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1007 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1010 /* Decrement ASM listener count on transition out of ASM mode. */
1011 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1012 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1013 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1014 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1015 --inm->inm_st[1].iss_asm;
1018 /* Increment ASM listener count on transition to ASM mode. */
1019 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1020 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1021 inm->inm_st[1].iss_asm++;
1024 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1029 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1036 * Mark an in_multi's filter set deltas as committed.
1037 * Called by IGMP after a state change has been enqueued.
1040 inm_commit(struct in_multi *inm)
1042 struct ip_msource *ims;
1044 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1045 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1048 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1049 ims->ims_st[0] = ims->ims_st[1];
1051 inm->inm_st[0] = inm->inm_st[1];
1055 * Reap unreferenced nodes from an in_multi's filter set.
1058 inm_reap(struct in_multi *inm)
1060 struct ip_msource *ims, *tims;
1062 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1063 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1064 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1067 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1068 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1069 free(ims, M_IPMSOURCE);
1075 * Purge all source nodes from an in_multi's filter set.
1078 inm_purge(struct in_multi *inm)
1080 struct ip_msource *ims, *tims;
1082 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1083 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1084 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1085 free(ims, M_IPMSOURCE);
1091 * Join a multicast group; unlocked entry point.
1093 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1094 * is not held. Fortunately, ifp is unlikely to have been detached
1095 * at this point, so we assume it's OK to recurse.
1098 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1099 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1104 error = in_joingroup_locked(ifp, gina, imf, pinm);
1111 * Join a multicast group; real entry point.
1113 * Only preserves atomicity at inm level.
1114 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1116 * If the IGMP downcall fails, the group is not joined, and an error
1120 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1121 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1123 struct in_mfilter timf;
1124 struct in_multi *inm;
1127 IN_MULTI_LOCK_ASSERT();
1129 CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__,
1130 inet_ntoa(*gina), ifp, ifp->if_xname);
1136 * If no imf was specified (i.e. kernel consumer),
1137 * fake one up and assume it is an ASM join.
1140 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1144 error = in_getmulti(ifp, gina, &inm);
1146 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1150 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1151 error = inm_merge(inm, imf);
1153 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1154 goto out_inm_release;
1157 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1158 error = igmp_change_state(inm);
1160 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1161 goto out_inm_release;
1166 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1167 inm_release_locked(inm);
1176 * Leave a multicast group; unlocked entry point.
1179 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1187 error = in_leavegroup_locked(inm, imf);
1194 * Leave a multicast group; real entry point.
1195 * All source filters will be expunged.
1197 * Only preserves atomicity at inm level.
1199 * Holding the write lock for the INP which contains imf
1200 * is highly advisable. We can't assert for it as imf does not
1201 * contain a back-pointer to the owning inp.
1203 * Note: This is not the same as inm_release(*) as this function also
1204 * makes a state change downcall into IGMP.
1207 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1209 struct in_mfilter timf;
1214 IN_MULTI_LOCK_ASSERT();
1216 CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__,
1217 inm, inet_ntoa(inm->inm_addr),
1218 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1222 * If no imf was specified (i.e. kernel consumer),
1223 * fake one up and assume it is an ASM join.
1226 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1231 * Begin state merge transaction at IGMP layer.
1233 * As this particular invocation should not cause any memory
1234 * to be allocated, and there is no opportunity to roll back
1235 * the transaction, it MUST NOT fail.
1237 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1238 error = inm_merge(inm, imf);
1239 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1241 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1242 error = igmp_change_state(inm);
1244 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1246 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1247 inm_release_locked(inm);
1252 /*#ifndef BURN_BRIDGES*/
1254 * Join an IPv4 multicast group in (*,G) exclusive mode.
1255 * The group must be a 224.0.0.0/24 link-scope group.
1256 * This KPI is for legacy kernel consumers only.
1259 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1261 struct in_multi *pinm;
1264 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1265 ("%s: %s not in 224.0.0.0/24", __func__, inet_ntoa(*ap)));
1267 error = in_joingroup(ifp, ap, NULL, &pinm);
1275 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1276 * This KPI is for legacy kernel consumers only.
1279 in_delmulti(struct in_multi *inm)
1282 (void)in_leavegroup(inm, NULL);
1287 * Block or unblock an ASM multicast source on an inpcb.
1288 * This implements the delta-based API described in RFC 3678.
1290 * The delta-based API applies only to exclusive-mode memberships.
1291 * An IGMP downcall will be performed.
1293 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1295 * Return 0 if successful, otherwise return an appropriate error code.
1298 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1300 struct group_source_req gsr;
1301 sockunion_t *gsa, *ssa;
1303 struct in_mfilter *imf;
1304 struct ip_moptions *imo;
1305 struct in_msource *ims;
1306 struct in_multi *inm;
1315 memset(&gsr, 0, sizeof(struct group_source_req));
1316 gsa = (sockunion_t *)&gsr.gsr_group;
1317 ssa = (sockunion_t *)&gsr.gsr_source;
1319 switch (sopt->sopt_name) {
1320 case IP_BLOCK_SOURCE:
1321 case IP_UNBLOCK_SOURCE: {
1322 struct ip_mreq_source mreqs;
1324 error = sooptcopyin(sopt, &mreqs,
1325 sizeof(struct ip_mreq_source),
1326 sizeof(struct ip_mreq_source));
1330 gsa->sin.sin_family = AF_INET;
1331 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1332 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1334 ssa->sin.sin_family = AF_INET;
1335 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1336 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1338 if (!in_nullhost(mreqs.imr_interface))
1339 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1341 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1344 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1345 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1349 case MCAST_BLOCK_SOURCE:
1350 case MCAST_UNBLOCK_SOURCE:
1351 error = sooptcopyin(sopt, &gsr,
1352 sizeof(struct group_source_req),
1353 sizeof(struct group_source_req));
1357 if (gsa->sin.sin_family != AF_INET ||
1358 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1361 if (ssa->sin.sin_family != AF_INET ||
1362 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1365 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1366 return (EADDRNOTAVAIL);
1368 ifp = ifnet_byindex(gsr.gsr_interface);
1370 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1375 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1376 __func__, sopt->sopt_name);
1377 return (EOPNOTSUPP);
1381 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1385 * Check if we are actually a member of this group.
1387 imo = inp_findmoptions(inp);
1388 idx = imo_match_group(imo, ifp, &gsa->sa);
1389 if (idx == -1 || imo->imo_mfilters == NULL) {
1390 error = EADDRNOTAVAIL;
1391 goto out_inp_locked;
1394 KASSERT(imo->imo_mfilters != NULL,
1395 ("%s: imo_mfilters not allocated", __func__));
1396 imf = &imo->imo_mfilters[idx];
1397 inm = imo->imo_membership[idx];
1400 * Attempting to use the delta-based API on an
1401 * non exclusive-mode membership is an error.
1403 fmode = imf->imf_st[0];
1404 if (fmode != MCAST_EXCLUDE) {
1406 goto out_inp_locked;
1410 * Deal with error cases up-front:
1411 * Asked to block, but already blocked; or
1412 * Asked to unblock, but nothing to unblock.
1413 * If adding a new block entry, allocate it.
1415 ims = imo_match_source(imo, idx, &ssa->sa);
1416 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1417 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
1418 inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not ");
1419 error = EADDRNOTAVAIL;
1420 goto out_inp_locked;
1423 INP_WLOCK_ASSERT(inp);
1426 * Begin state merge transaction at socket layer.
1429 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1430 ims = imf_graft(imf, fmode, &ssa->sin);
1434 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1435 error = imf_prune(imf, &ssa->sin);
1439 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1440 goto out_imf_rollback;
1444 * Begin state merge transaction at IGMP layer.
1448 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1449 error = inm_merge(inm, imf);
1451 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1452 goto out_imf_rollback;
1455 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1456 error = igmp_change_state(inm);
1458 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1476 * Given an inpcb, return its multicast options structure pointer. Accepts
1477 * an unlocked inpcb pointer, but will return it locked. May sleep.
1479 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1480 * SMPng: NOTE: Returns with the INP write lock held.
1482 static struct ip_moptions *
1483 inp_findmoptions(struct inpcb *inp)
1485 struct ip_moptions *imo;
1486 struct in_multi **immp;
1487 struct in_mfilter *imfp;
1491 if (inp->inp_moptions != NULL)
1492 return (inp->inp_moptions);
1496 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1497 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1499 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1500 M_INMFILTER, M_WAITOK);
1502 imo->imo_multicast_ifp = NULL;
1503 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1504 imo->imo_multicast_vif = -1;
1505 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1506 imo->imo_multicast_loop = in_mcast_loop;
1507 imo->imo_num_memberships = 0;
1508 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1509 imo->imo_membership = immp;
1511 /* Initialize per-group source filters. */
1512 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1513 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1514 imo->imo_mfilters = imfp;
1517 if (inp->inp_moptions != NULL) {
1518 free(imfp, M_INMFILTER);
1519 free(immp, M_IPMOPTS);
1520 free(imo, M_IPMOPTS);
1521 return (inp->inp_moptions);
1523 inp->inp_moptions = imo;
1528 * Discard the IP multicast options (and source filters). To minimize
1529 * the amount of work done while holding locks such as the INP's
1530 * pcbinfo lock (which is used in the receive path), the free
1531 * operation is performed asynchronously in a separate task.
1533 * SMPng: NOTE: assumes INP write lock is held.
1536 inp_freemoptions(struct ip_moptions *imo)
1539 KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1541 STAILQ_INSERT_TAIL(&imo_gc_list, imo, imo_link);
1543 taskqueue_enqueue(taskqueue_thread, &imo_gc_task);
1547 inp_freemoptions_internal(struct ip_moptions *imo)
1549 struct in_mfilter *imf;
1550 size_t idx, nmships;
1552 nmships = imo->imo_num_memberships;
1553 for (idx = 0; idx < nmships; ++idx) {
1554 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1557 (void)in_leavegroup(imo->imo_membership[idx], imf);
1562 if (imo->imo_mfilters)
1563 free(imo->imo_mfilters, M_INMFILTER);
1564 free(imo->imo_membership, M_IPMOPTS);
1565 free(imo, M_IPMOPTS);
1569 inp_gcmoptions(void *context, int pending)
1571 struct ip_moptions *imo;
1574 while (!STAILQ_EMPTY(&imo_gc_list)) {
1575 imo = STAILQ_FIRST(&imo_gc_list);
1576 STAILQ_REMOVE_HEAD(&imo_gc_list, imo_link);
1578 inp_freemoptions_internal(imo);
1585 * Atomically get source filters on a socket for an IPv4 multicast group.
1586 * Called with INP lock held; returns with lock released.
1589 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1591 struct __msfilterreq msfr;
1594 struct ip_moptions *imo;
1595 struct in_mfilter *imf;
1596 struct ip_msource *ims;
1597 struct in_msource *lims;
1598 struct sockaddr_in *psin;
1599 struct sockaddr_storage *ptss;
1600 struct sockaddr_storage *tss;
1602 size_t idx, nsrcs, ncsrcs;
1604 INP_WLOCK_ASSERT(inp);
1606 imo = inp->inp_moptions;
1607 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1611 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1612 sizeof(struct __msfilterreq));
1616 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1619 ifp = ifnet_byindex(msfr.msfr_ifindex);
1626 * Lookup group on the socket.
1628 gsa = (sockunion_t *)&msfr.msfr_group;
1629 idx = imo_match_group(imo, ifp, &gsa->sa);
1630 if (idx == -1 || imo->imo_mfilters == NULL) {
1632 return (EADDRNOTAVAIL);
1634 imf = &imo->imo_mfilters[idx];
1637 * Ignore memberships which are in limbo.
1639 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1643 msfr.msfr_fmode = imf->imf_st[1];
1646 * If the user specified a buffer, copy out the source filter
1647 * entries to userland gracefully.
1648 * We only copy out the number of entries which userland
1649 * has asked for, but we always tell userland how big the
1650 * buffer really needs to be.
1653 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1654 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1655 M_TEMP, M_NOWAIT | M_ZERO);
1663 * Count number of sources in-mode at t0.
1664 * If buffer space exists and remains, copy out source entries.
1666 nsrcs = msfr.msfr_nsrcs;
1669 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1670 lims = (struct in_msource *)ims;
1671 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1672 lims->imsl_st[0] != imf->imf_st[0])
1675 if (tss != NULL && nsrcs > 0) {
1676 psin = (struct sockaddr_in *)ptss;
1677 psin->sin_family = AF_INET;
1678 psin->sin_len = sizeof(struct sockaddr_in);
1679 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1689 error = copyout(tss, msfr.msfr_srcs,
1690 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1696 msfr.msfr_nsrcs = ncsrcs;
1697 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1703 * Return the IP multicast options in response to user getsockopt().
1706 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1708 struct ip_mreqn mreqn;
1709 struct ip_moptions *imo;
1711 struct in_ifaddr *ia;
1716 imo = inp->inp_moptions;
1718 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1719 * or is a divert socket, reject it.
1721 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1722 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1723 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1725 return (EOPNOTSUPP);
1729 switch (sopt->sopt_name) {
1730 case IP_MULTICAST_VIF:
1732 optval = imo->imo_multicast_vif;
1736 error = sooptcopyout(sopt, &optval, sizeof(int));
1739 case IP_MULTICAST_IF:
1740 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1742 ifp = imo->imo_multicast_ifp;
1743 if (!in_nullhost(imo->imo_multicast_addr)) {
1744 mreqn.imr_address = imo->imo_multicast_addr;
1745 } else if (ifp != NULL) {
1746 mreqn.imr_ifindex = ifp->if_index;
1750 IA_SIN(ia)->sin_addr;
1751 ifa_free(&ia->ia_ifa);
1756 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1757 error = sooptcopyout(sopt, &mreqn,
1758 sizeof(struct ip_mreqn));
1760 error = sooptcopyout(sopt, &mreqn.imr_address,
1761 sizeof(struct in_addr));
1765 case IP_MULTICAST_TTL:
1767 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1769 optval = coptval = imo->imo_multicast_ttl;
1771 if (sopt->sopt_valsize == sizeof(u_char))
1772 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1774 error = sooptcopyout(sopt, &optval, sizeof(int));
1777 case IP_MULTICAST_LOOP:
1779 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1781 optval = coptval = imo->imo_multicast_loop;
1783 if (sopt->sopt_valsize == sizeof(u_char))
1784 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1786 error = sooptcopyout(sopt, &optval, sizeof(int));
1791 error = EADDRNOTAVAIL;
1794 error = inp_get_source_filters(inp, sopt);
1800 error = ENOPROTOOPT;
1804 INP_UNLOCK_ASSERT(inp);
1810 * Look up the ifnet to use for a multicast group membership,
1811 * given the IPv4 address of an interface, and the IPv4 group address.
1813 * This routine exists to support legacy multicast applications
1814 * which do not understand that multicast memberships are scoped to
1815 * specific physical links in the networking stack, or which need
1816 * to join link-scope groups before IPv4 addresses are configured.
1818 * If inp is non-NULL, use this socket's current FIB number for any
1819 * required FIB lookup.
1820 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1821 * and use its ifp; usually, this points to the default next-hop.
1823 * If the FIB lookup fails, attempt to use the first non-loopback
1824 * interface with multicast capability in the system as a
1825 * last resort. The legacy IPv4 ASM API requires that we do
1826 * this in order to allow groups to be joined when the routing
1827 * table has not yet been populated during boot.
1829 * Returns NULL if no ifp could be found.
1831 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1832 * FUTURE: Implement IPv4 source-address selection.
1834 static struct ifnet *
1835 inp_lookup_mcast_ifp(const struct inpcb *inp,
1836 const struct sockaddr_in *gsin, const struct in_addr ina)
1840 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1841 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1842 ("%s: not multicast", __func__));
1845 if (!in_nullhost(ina)) {
1846 INADDR_TO_IFP(ina, ifp);
1851 memcpy(&ro.ro_dst, gsin, sizeof(struct sockaddr_in));
1852 in_rtalloc_ign(&ro, 0, inp ? inp->inp_inc.inc_fibnum : 0);
1853 if (ro.ro_rt != NULL) {
1854 ifp = ro.ro_rt->rt_ifp;
1855 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1858 struct in_ifaddr *ia;
1863 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1865 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1866 (mifp->if_flags & IFF_MULTICAST)) {
1871 IN_IFADDR_RUNLOCK();
1879 * Join an IPv4 multicast group, possibly with a source.
1882 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1884 struct group_source_req gsr;
1885 sockunion_t *gsa, *ssa;
1887 struct in_mfilter *imf;
1888 struct ip_moptions *imo;
1889 struct in_multi *inm;
1890 struct in_msource *lims;
1900 memset(&gsr, 0, sizeof(struct group_source_req));
1901 gsa = (sockunion_t *)&gsr.gsr_group;
1902 gsa->ss.ss_family = AF_UNSPEC;
1903 ssa = (sockunion_t *)&gsr.gsr_source;
1904 ssa->ss.ss_family = AF_UNSPEC;
1906 switch (sopt->sopt_name) {
1907 case IP_ADD_MEMBERSHIP:
1908 case IP_ADD_SOURCE_MEMBERSHIP: {
1909 struct ip_mreq_source mreqs;
1911 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1912 error = sooptcopyin(sopt, &mreqs,
1913 sizeof(struct ip_mreq),
1914 sizeof(struct ip_mreq));
1916 * Do argument switcharoo from ip_mreq into
1917 * ip_mreq_source to avoid using two instances.
1919 mreqs.imr_interface = mreqs.imr_sourceaddr;
1920 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1921 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1922 error = sooptcopyin(sopt, &mreqs,
1923 sizeof(struct ip_mreq_source),
1924 sizeof(struct ip_mreq_source));
1929 gsa->sin.sin_family = AF_INET;
1930 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1931 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1933 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1934 ssa->sin.sin_family = AF_INET;
1935 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1936 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1939 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1942 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1943 mreqs.imr_interface);
1944 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1945 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1949 case MCAST_JOIN_GROUP:
1950 case MCAST_JOIN_SOURCE_GROUP:
1951 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1952 error = sooptcopyin(sopt, &gsr,
1953 sizeof(struct group_req),
1954 sizeof(struct group_req));
1955 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1956 error = sooptcopyin(sopt, &gsr,
1957 sizeof(struct group_source_req),
1958 sizeof(struct group_source_req));
1963 if (gsa->sin.sin_family != AF_INET ||
1964 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1968 * Overwrite the port field if present, as the sockaddr
1969 * being copied in may be matched with a binary comparison.
1971 gsa->sin.sin_port = 0;
1972 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1973 if (ssa->sin.sin_family != AF_INET ||
1974 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1976 ssa->sin.sin_port = 0;
1979 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1982 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1983 return (EADDRNOTAVAIL);
1984 ifp = ifnet_byindex(gsr.gsr_interface);
1988 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1989 __func__, sopt->sopt_name);
1990 return (EOPNOTSUPP);
1994 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1995 return (EADDRNOTAVAIL);
1997 imo = inp_findmoptions(inp);
1998 idx = imo_match_group(imo, ifp, &gsa->sa);
2002 inm = imo->imo_membership[idx];
2003 imf = &imo->imo_mfilters[idx];
2004 if (ssa->ss.ss_family != AF_UNSPEC) {
2006 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2007 * is an error. On an existing inclusive membership,
2008 * it just adds the source to the filter list.
2010 if (imf->imf_st[1] != MCAST_INCLUDE) {
2012 goto out_inp_locked;
2015 * Throw out duplicates.
2017 * XXX FIXME: This makes a naive assumption that
2018 * even if entries exist for *ssa in this imf,
2019 * they will be rejected as dupes, even if they
2020 * are not valid in the current mode (in-mode).
2022 * in_msource is transactioned just as for anything
2023 * else in SSM -- but note naive use of inm_graft()
2024 * below for allocating new filter entries.
2026 * This is only an issue if someone mixes the
2027 * full-state SSM API with the delta-based API,
2028 * which is discouraged in the relevant RFCs.
2030 lims = imo_match_source(imo, idx, &ssa->sa);
2031 if (lims != NULL /*&&
2032 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2033 error = EADDRNOTAVAIL;
2034 goto out_inp_locked;
2038 * MCAST_JOIN_GROUP on an existing exclusive
2039 * membership is an error; return EADDRINUSE
2040 * to preserve 4.4BSD API idempotence, and
2041 * avoid tedious detour to code below.
2042 * NOTE: This is bending RFC 3678 a bit.
2044 * On an existing inclusive membership, this is also
2045 * an error; if you want to change filter mode,
2046 * you must use the userland API setsourcefilter().
2047 * XXX We don't reject this for imf in UNDEFINED
2048 * state at t1, because allocation of a filter
2049 * is atomic with allocation of a membership.
2052 if (imf->imf_st[1] == MCAST_EXCLUDE)
2054 goto out_inp_locked;
2059 * Begin state merge transaction at socket layer.
2061 INP_WLOCK_ASSERT(inp);
2064 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2065 error = imo_grow(imo);
2067 goto out_inp_locked;
2070 * Allocate the new slot upfront so we can deal with
2071 * grafting the new source filter in same code path
2072 * as for join-source on existing membership.
2074 idx = imo->imo_num_memberships;
2075 imo->imo_membership[idx] = NULL;
2076 imo->imo_num_memberships++;
2077 KASSERT(imo->imo_mfilters != NULL,
2078 ("%s: imf_mfilters vector was not allocated", __func__));
2079 imf = &imo->imo_mfilters[idx];
2080 KASSERT(RB_EMPTY(&imf->imf_sources),
2081 ("%s: imf_sources not empty", __func__));
2085 * Graft new source into filter list for this inpcb's
2086 * membership of the group. The in_multi may not have
2087 * been allocated yet if this is a new membership, however,
2088 * the in_mfilter slot will be allocated and must be initialized.
2090 * Note: Grafting of exclusive mode filters doesn't happen
2092 * XXX: Should check for non-NULL lims (node exists but may
2093 * not be in-mode) for interop with full-state API.
2095 if (ssa->ss.ss_family != AF_UNSPEC) {
2096 /* Membership starts in IN mode */
2098 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2099 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2101 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2103 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2105 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2111 /* No address specified; Membership starts in EX mode */
2113 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2114 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2119 * Begin state merge transaction at IGMP layer.
2124 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2128 imo->imo_membership[idx] = inm;
2130 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2131 error = inm_merge(inm, imf);
2133 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2135 goto out_imf_rollback;
2137 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2138 error = igmp_change_state(inm);
2140 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2142 goto out_imf_rollback;
2149 INP_WLOCK_ASSERT(inp);
2161 if (error && is_new) {
2162 imo->imo_membership[idx] = NULL;
2163 --imo->imo_num_memberships;
2172 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2175 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2177 struct group_source_req gsr;
2178 struct ip_mreq_source mreqs;
2179 sockunion_t *gsa, *ssa;
2181 struct in_mfilter *imf;
2182 struct ip_moptions *imo;
2183 struct in_msource *ims;
2184 struct in_multi *inm;
2186 int error, is_final;
2192 memset(&gsr, 0, sizeof(struct group_source_req));
2193 gsa = (sockunion_t *)&gsr.gsr_group;
2194 gsa->ss.ss_family = AF_UNSPEC;
2195 ssa = (sockunion_t *)&gsr.gsr_source;
2196 ssa->ss.ss_family = AF_UNSPEC;
2198 switch (sopt->sopt_name) {
2199 case IP_DROP_MEMBERSHIP:
2200 case IP_DROP_SOURCE_MEMBERSHIP:
2201 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2202 error = sooptcopyin(sopt, &mreqs,
2203 sizeof(struct ip_mreq),
2204 sizeof(struct ip_mreq));
2206 * Swap interface and sourceaddr arguments,
2207 * as ip_mreq and ip_mreq_source are laid
2210 mreqs.imr_interface = mreqs.imr_sourceaddr;
2211 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2212 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2213 error = sooptcopyin(sopt, &mreqs,
2214 sizeof(struct ip_mreq_source),
2215 sizeof(struct ip_mreq_source));
2220 gsa->sin.sin_family = AF_INET;
2221 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2222 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2224 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2225 ssa->sin.sin_family = AF_INET;
2226 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2227 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2231 * Attempt to look up hinted ifp from interface address.
2232 * Fallthrough with null ifp iff lookup fails, to
2233 * preserve 4.4BSD mcast API idempotence.
2234 * XXX NOTE WELL: The RFC 3678 API is preferred because
2235 * using an IPv4 address as a key is racy.
2237 if (!in_nullhost(mreqs.imr_interface))
2238 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2240 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
2241 __func__, inet_ntoa(mreqs.imr_interface), ifp);
2245 case MCAST_LEAVE_GROUP:
2246 case MCAST_LEAVE_SOURCE_GROUP:
2247 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2248 error = sooptcopyin(sopt, &gsr,
2249 sizeof(struct group_req),
2250 sizeof(struct group_req));
2251 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2252 error = sooptcopyin(sopt, &gsr,
2253 sizeof(struct group_source_req),
2254 sizeof(struct group_source_req));
2259 if (gsa->sin.sin_family != AF_INET ||
2260 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2263 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2264 if (ssa->sin.sin_family != AF_INET ||
2265 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2269 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2270 return (EADDRNOTAVAIL);
2272 ifp = ifnet_byindex(gsr.gsr_interface);
2275 return (EADDRNOTAVAIL);
2279 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2280 __func__, sopt->sopt_name);
2281 return (EOPNOTSUPP);
2285 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2289 * Find the membership in the membership array.
2291 imo = inp_findmoptions(inp);
2292 idx = imo_match_group(imo, ifp, &gsa->sa);
2294 error = EADDRNOTAVAIL;
2295 goto out_inp_locked;
2297 inm = imo->imo_membership[idx];
2298 imf = &imo->imo_mfilters[idx];
2300 if (ssa->ss.ss_family != AF_UNSPEC)
2304 * Begin state merge transaction at socket layer.
2306 INP_WLOCK_ASSERT(inp);
2309 * If we were instructed only to leave a given source, do so.
2310 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2315 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2316 error = EADDRNOTAVAIL;
2317 goto out_inp_locked;
2319 ims = imo_match_source(imo, idx, &ssa->sa);
2321 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
2322 inet_ntoa(ssa->sin.sin_addr), "not ");
2323 error = EADDRNOTAVAIL;
2324 goto out_inp_locked;
2326 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2327 error = imf_prune(imf, &ssa->sin);
2329 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2331 goto out_inp_locked;
2336 * Begin state merge transaction at IGMP layer.
2342 * Give up the multicast address record to which
2343 * the membership points.
2345 (void)in_leavegroup_locked(inm, imf);
2347 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2348 error = inm_merge(inm, imf);
2350 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2352 goto out_imf_rollback;
2355 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2356 error = igmp_change_state(inm);
2358 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2374 /* Remove the gap in the membership and filter array. */
2375 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2376 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2377 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2379 imo->imo_num_memberships--;
2388 * Select the interface for transmitting IPv4 multicast datagrams.
2390 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2391 * may be passed to this socket option. An address of INADDR_ANY or an
2392 * interface index of 0 is used to remove a previous selection.
2393 * When no interface is selected, one is chosen for every send.
2396 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2398 struct in_addr addr;
2399 struct ip_mreqn mreqn;
2401 struct ip_moptions *imo;
2404 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2406 * An interface index was specified using the
2407 * Linux-derived ip_mreqn structure.
2409 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2410 sizeof(struct ip_mreqn));
2414 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2417 if (mreqn.imr_ifindex == 0) {
2420 ifp = ifnet_byindex(mreqn.imr_ifindex);
2422 return (EADDRNOTAVAIL);
2426 * An interface was specified by IPv4 address.
2427 * This is the traditional BSD usage.
2429 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2430 sizeof(struct in_addr));
2433 if (in_nullhost(addr)) {
2436 INADDR_TO_IFP(addr, ifp);
2438 return (EADDRNOTAVAIL);
2440 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp,
2444 /* Reject interfaces which do not support multicast. */
2445 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2446 return (EOPNOTSUPP);
2448 imo = inp_findmoptions(inp);
2449 imo->imo_multicast_ifp = ifp;
2450 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2457 * Atomically set source filters on a socket for an IPv4 multicast group.
2459 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2462 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2464 struct __msfilterreq msfr;
2467 struct in_mfilter *imf;
2468 struct ip_moptions *imo;
2469 struct in_multi *inm;
2473 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2474 sizeof(struct __msfilterreq));
2478 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2481 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2482 msfr.msfr_fmode != MCAST_INCLUDE))
2485 if (msfr.msfr_group.ss_family != AF_INET ||
2486 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2489 gsa = (sockunion_t *)&msfr.msfr_group;
2490 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2493 gsa->sin.sin_port = 0; /* ignore port */
2495 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2496 return (EADDRNOTAVAIL);
2498 ifp = ifnet_byindex(msfr.msfr_ifindex);
2500 return (EADDRNOTAVAIL);
2503 * Take the INP write lock.
2504 * Check if this socket is a member of this group.
2506 imo = inp_findmoptions(inp);
2507 idx = imo_match_group(imo, ifp, &gsa->sa);
2508 if (idx == -1 || imo->imo_mfilters == NULL) {
2509 error = EADDRNOTAVAIL;
2510 goto out_inp_locked;
2512 inm = imo->imo_membership[idx];
2513 imf = &imo->imo_mfilters[idx];
2516 * Begin state merge transaction at socket layer.
2518 INP_WLOCK_ASSERT(inp);
2520 imf->imf_st[1] = msfr.msfr_fmode;
2523 * Apply any new source filters, if present.
2524 * Make a copy of the user-space source vector so
2525 * that we may copy them with a single copyin. This
2526 * allows us to deal with page faults up-front.
2528 if (msfr.msfr_nsrcs > 0) {
2529 struct in_msource *lims;
2530 struct sockaddr_in *psin;
2531 struct sockaddr_storage *kss, *pkss;
2536 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2537 __func__, (unsigned long)msfr.msfr_nsrcs);
2538 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2540 error = copyin(msfr.msfr_srcs, kss,
2541 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2550 * Mark all source filters as UNDEFINED at t1.
2551 * Restore new group filter mode, as imf_leave()
2552 * will set it to INCLUDE.
2555 imf->imf_st[1] = msfr.msfr_fmode;
2558 * Update socket layer filters at t1, lazy-allocating
2559 * new entries. This saves a bunch of memory at the
2560 * cost of one RB_FIND() per source entry; duplicate
2561 * entries in the msfr_nsrcs vector are ignored.
2562 * If we encounter an error, rollback transaction.
2564 * XXX This too could be replaced with a set-symmetric
2565 * difference like loop to avoid walking from root
2566 * every time, as the key space is common.
2568 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2569 psin = (struct sockaddr_in *)pkss;
2570 if (psin->sin_family != AF_INET) {
2571 error = EAFNOSUPPORT;
2574 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2578 error = imf_get_source(imf, psin, &lims);
2581 lims->imsl_st[1] = imf->imf_st[1];
2587 goto out_imf_rollback;
2589 INP_WLOCK_ASSERT(inp);
2593 * Begin state merge transaction at IGMP layer.
2595 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2596 error = inm_merge(inm, imf);
2598 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2599 goto out_imf_rollback;
2602 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2603 error = igmp_change_state(inm);
2605 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2623 * Set the IP multicast options in response to user setsockopt().
2625 * Many of the socket options handled in this function duplicate the
2626 * functionality of socket options in the regular unicast API. However,
2627 * it is not possible to merge the duplicate code, because the idempotence
2628 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2629 * the effects of these options must be treated as separate and distinct.
2631 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2632 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2633 * is refactored to no longer use vifs.
2636 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2638 struct ip_moptions *imo;
2644 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2645 * or is a divert socket, reject it.
2647 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2648 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2649 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2650 return (EOPNOTSUPP);
2652 switch (sopt->sopt_name) {
2653 case IP_MULTICAST_VIF: {
2656 * Select a multicast VIF for transmission.
2657 * Only useful if multicast forwarding is active.
2659 if (legal_vif_num == NULL) {
2663 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2666 if (!legal_vif_num(vifi) && (vifi != -1)) {
2670 imo = inp_findmoptions(inp);
2671 imo->imo_multicast_vif = vifi;
2676 case IP_MULTICAST_IF:
2677 error = inp_set_multicast_if(inp, sopt);
2680 case IP_MULTICAST_TTL: {
2684 * Set the IP time-to-live for outgoing multicast packets.
2685 * The original multicast API required a char argument,
2686 * which is inconsistent with the rest of the socket API.
2687 * We allow either a char or an int.
2689 if (sopt->sopt_valsize == sizeof(u_char)) {
2690 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2697 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2707 imo = inp_findmoptions(inp);
2708 imo->imo_multicast_ttl = ttl;
2713 case IP_MULTICAST_LOOP: {
2717 * Set the loopback flag for outgoing multicast packets.
2718 * Must be zero or one. The original multicast API required a
2719 * char argument, which is inconsistent with the rest
2720 * of the socket API. We allow either a char or an int.
2722 if (sopt->sopt_valsize == sizeof(u_char)) {
2723 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2730 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2734 loop = (u_char)iloop;
2736 imo = inp_findmoptions(inp);
2737 imo->imo_multicast_loop = !!loop;
2742 case IP_ADD_MEMBERSHIP:
2743 case IP_ADD_SOURCE_MEMBERSHIP:
2744 case MCAST_JOIN_GROUP:
2745 case MCAST_JOIN_SOURCE_GROUP:
2746 error = inp_join_group(inp, sopt);
2749 case IP_DROP_MEMBERSHIP:
2750 case IP_DROP_SOURCE_MEMBERSHIP:
2751 case MCAST_LEAVE_GROUP:
2752 case MCAST_LEAVE_SOURCE_GROUP:
2753 error = inp_leave_group(inp, sopt);
2756 case IP_BLOCK_SOURCE:
2757 case IP_UNBLOCK_SOURCE:
2758 case MCAST_BLOCK_SOURCE:
2759 case MCAST_UNBLOCK_SOURCE:
2760 error = inp_block_unblock_source(inp, sopt);
2764 error = inp_set_source_filters(inp, sopt);
2772 INP_UNLOCK_ASSERT(inp);
2778 * Expose IGMP's multicast filter mode and source list(s) to userland,
2779 * keyed by (ifindex, group).
2780 * The filter mode is written out as a uint32_t, followed by
2781 * 0..n of struct in_addr.
2782 * For use by ifmcstat(8).
2783 * SMPng: NOTE: unlocked read of ifindex space.
2786 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2788 struct in_addr src, group;
2790 struct ifmultiaddr *ifma;
2791 struct in_multi *inm;
2792 struct ip_msource *ims;
2796 uint32_t fmode, ifindex;
2801 if (req->newptr != NULL)
2808 if (ifindex <= 0 || ifindex > V_if_index) {
2809 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2814 group.s_addr = name[1];
2815 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2816 CTR2(KTR_IGMPV3, "%s: group %s is not multicast",
2817 __func__, inet_ntoa(group));
2821 ifp = ifnet_byindex(ifindex);
2823 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2828 retval = sysctl_wire_old_buffer(req,
2829 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2836 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2837 if (ifma->ifma_addr->sa_family != AF_INET ||
2838 ifma->ifma_protospec == NULL)
2840 inm = (struct in_multi *)ifma->ifma_protospec;
2841 if (!in_hosteq(inm->inm_addr, group))
2843 fmode = inm->inm_st[1].iss_fmode;
2844 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2847 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2850 ina.s_addr = htonl(ims->ims_haddr);
2851 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2855 * Only copy-out sources which are in-mode.
2857 if (fmode != ims_get_mode(inm, ims, 1)) {
2858 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2862 src.s_addr = htonl(ims->ims_haddr);
2863 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2868 IF_ADDR_UNLOCK(ifp);
2877 static const char *inm_modestrs[] = { "un", "in", "ex" };
2880 inm_mode_str(const int mode)
2883 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2884 return (inm_modestrs[mode]);
2888 static const char *inm_statestrs[] = {
2901 inm_state_str(const int state)
2904 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2905 return (inm_statestrs[state]);
2910 * Dump an in_multi structure to the console.
2913 inm_print(const struct in_multi *inm)
2917 if ((ktr_mask & KTR_IGMPV3) == 0)
2920 printf("%s: --- begin inm %p ---\n", __func__, inm);
2921 printf("addr %s ifp %p(%s) ifma %p\n",
2922 inet_ntoa(inm->inm_addr),
2924 inm->inm_ifp->if_xname,
2926 printf("timer %u state %s refcount %u scq.len %u\n",
2928 inm_state_str(inm->inm_state),
2930 inm->inm_scq.ifq_len);
2931 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2936 for (t = 0; t < 2; t++) {
2937 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2938 inm_mode_str(inm->inm_st[t].iss_fmode),
2939 inm->inm_st[t].iss_asm,
2940 inm->inm_st[t].iss_ex,
2941 inm->inm_st[t].iss_in,
2942 inm->inm_st[t].iss_rec);
2944 printf("%s: --- end inm %p ---\n", __func__, inm);
2950 inm_print(const struct in_multi *inm)
2957 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);