2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2007-2009 Bruce Simpson.
5 * Copyright (c) 2005 Robert N. M. Watson.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote
17 * products derived from this software without specific prior written
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * IPv4 multicast socket, group, and socket option processing module.
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/protosw.h>
50 #include <sys/sysctl.h>
52 #include <sys/taskqueue.h>
56 #include <net/if_var.h>
57 #include <net/if_dl.h>
58 #include <net/route.h>
59 #include <net/route/nhop.h>
62 #include <net/ethernet.h>
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/in_fib.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/in_var.h>
69 #include <net/if_private.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/igmp_var.h>
74 #define KTR_IGMPV3 KTR_INET
77 #ifndef __SOCKUNION_DECLARED
79 struct sockaddr_storage ss;
81 struct sockaddr_dl sdl;
82 struct sockaddr_in sin;
84 typedef union sockunion sockunion_t;
85 #define __SOCKUNION_DECLARED
86 #endif /* __SOCKUNION_DECLARED */
88 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
89 "IPv4 multicast PCB-layer source filter");
90 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
91 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
92 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
93 "IPv4 multicast IGMP-layer source filter");
98 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK,
99 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
100 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
101 * it can be taken by code in net/if.c also.
102 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
104 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
105 * any need for in_multi itself to be virtualized -- it is bound to an ifp
106 * anyway no matter what happens.
108 struct mtx in_multi_list_mtx;
109 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
111 struct mtx in_multi_free_mtx;
112 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
114 struct sx in_multi_sx;
115 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
118 * Functions with non-static linkage defined in this file should be
119 * declared in in_var.h:
122 * in_joingroup_locked()
124 * in_leavegroup_locked()
130 static void imf_commit(struct in_mfilter *);
131 static int imf_get_source(struct in_mfilter *imf,
132 const struct sockaddr_in *psin,
133 struct in_msource **);
134 static struct in_msource *
135 imf_graft(struct in_mfilter *, const uint8_t,
136 const struct sockaddr_in *);
137 static void imf_leave(struct in_mfilter *);
138 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
139 static void imf_purge(struct in_mfilter *);
140 static void imf_rollback(struct in_mfilter *);
141 static void imf_reap(struct in_mfilter *);
142 static struct in_mfilter *
143 imo_match_group(const struct ip_moptions *,
144 const struct ifnet *, const struct sockaddr *);
145 static struct in_msource *
146 imo_match_source(struct in_mfilter *, const struct sockaddr *);
147 static void ims_merge(struct ip_msource *ims,
148 const struct in_msource *lims, const int rollback);
149 static int in_getmulti(struct ifnet *, const struct in_addr *,
151 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
152 const int noalloc, struct ip_msource **pims);
154 static int inm_is_ifp_detached(const struct in_multi *);
156 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
157 static void inm_purge(struct in_multi *);
158 static void inm_reap(struct in_multi *);
159 static void inm_release(struct in_multi *);
160 static struct ip_moptions *
161 inp_findmoptions(struct inpcb *);
162 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
163 static int inp_join_group(struct inpcb *, struct sockopt *);
164 static int inp_leave_group(struct inpcb *, struct sockopt *);
165 static struct ifnet *
166 inp_lookup_mcast_ifp(const struct inpcb *,
167 const struct sockaddr_in *, const struct in_addr);
168 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
169 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
170 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
171 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
173 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast,
174 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
177 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
178 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
179 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
180 "Max source filters per group");
182 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
183 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
184 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
185 "Max source filters per socket");
187 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
188 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
189 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
191 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
192 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
193 "Per-interface stack-wide source filters");
197 * Inline function which wraps assertions for a valid ifp.
198 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
202 inm_is_ifp_detached(const struct in_multi *inm)
206 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
207 ifp = inm->inm_ifma->ifma_ifp;
210 * Sanity check that netinet's notion of ifp is the
213 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
216 return (ifp == NULL);
221 * Interface detach can happen in a taskqueue thread context, so we must use a
222 * dedicated thread to avoid deadlocks when draining inm_release tasks.
224 TASKQUEUE_DEFINE_THREAD(inm_free);
225 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER();
226 static void inm_release_task(void *arg __unused, int pending __unused);
227 static struct task inm_free_task = TASK_INITIALIZER(0, inm_release_task, NULL);
230 inm_release_wait(void *arg __unused)
234 * Make sure all pending multicast addresses are freed before
235 * the VNET or network device is destroyed:
237 taskqueue_drain(taskqueue_inm_free, &inm_free_task);
240 /* XXX-BZ FIXME, see D24914. */
241 VNET_SYSUNINIT(inm_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, inm_release_wait, NULL);
245 inm_release_list_deferred(struct in_multi_head *inmh)
248 if (SLIST_EMPTY(inmh))
250 mtx_lock(&in_multi_free_mtx);
251 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
252 mtx_unlock(&in_multi_free_mtx);
253 taskqueue_enqueue(taskqueue_inm_free, &inm_free_task);
257 inm_disconnect(struct in_multi *inm)
260 struct ifmultiaddr *ifma, *ll_ifma;
263 IF_ADDR_WLOCK_ASSERT(ifp);
264 ifma = inm->inm_ifma;
267 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
268 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
269 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
271 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
272 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
273 MPASS(ifma != ll_ifma);
274 ifma->ifma_llifma = NULL;
275 MPASS(ll_ifma->ifma_llifma == NULL);
276 MPASS(ll_ifma->ifma_ifp == ifp);
277 if (--ll_ifma->ifma_refcount == 0) {
278 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
279 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
280 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
282 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
283 if_freemulti(ll_ifma);
289 inm_release_deferred(struct in_multi *inm)
291 struct in_multi_head tmp;
293 IN_MULTI_LIST_LOCK_ASSERT();
294 MPASS(inm->inm_refcount > 0);
295 if (--inm->inm_refcount == 0) {
298 inm->inm_ifma->ifma_protospec = NULL;
299 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
300 inm_release_list_deferred(&tmp);
305 inm_release_task(void *arg __unused, int pending __unused)
307 struct in_multi_head inm_free_tmp;
308 struct in_multi *inm, *tinm;
310 SLIST_INIT(&inm_free_tmp);
311 mtx_lock(&in_multi_free_mtx);
312 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
313 mtx_unlock(&in_multi_free_mtx);
315 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
316 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
324 * Initialize an in_mfilter structure to a known state at t0, t1
325 * with an empty source filter list.
328 imf_init(struct in_mfilter *imf, const int st0, const int st1)
330 memset(imf, 0, sizeof(struct in_mfilter));
331 RB_INIT(&imf->imf_sources);
332 imf->imf_st[0] = st0;
333 imf->imf_st[1] = st1;
337 ip_mfilter_alloc(const int mflags, const int st0, const int st1)
339 struct in_mfilter *imf;
341 imf = malloc(sizeof(*imf), M_INMFILTER, mflags);
343 imf_init(imf, st0, st1);
349 ip_mfilter_free(struct in_mfilter *imf)
353 free(imf, M_INMFILTER);
357 * Function for looking up an in_multi record for an IPv4 multicast address
358 * on a given interface. ifp must be valid. If no record found, return NULL.
359 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
362 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
364 struct ifmultiaddr *ifma;
365 struct in_multi *inm;
367 IN_MULTI_LIST_LOCK_ASSERT();
368 IF_ADDR_LOCK_ASSERT(ifp);
370 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
371 inm = inm_ifmultiaddr_get_inm(ifma);
374 if (inm->inm_addr.s_addr == ina.s_addr)
381 * Wrapper for inm_lookup_locked().
382 * The IF_ADDR_LOCK will be taken on ifp and released on return.
385 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
387 struct epoch_tracker et;
388 struct in_multi *inm;
390 IN_MULTI_LIST_LOCK_ASSERT();
393 inm = inm_lookup_locked(ifp, ina);
400 * Find an IPv4 multicast group entry for this ip_moptions instance
401 * which matches the specified group, and optionally an interface.
402 * Return its index into the array, or -1 if not found.
404 static struct in_mfilter *
405 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
406 const struct sockaddr *group)
408 const struct sockaddr_in *gsin;
409 struct in_mfilter *imf;
410 struct in_multi *inm;
412 gsin = (const struct sockaddr_in *)group;
414 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
418 if ((ifp == NULL || (inm->inm_ifp == ifp)) &&
419 in_hosteq(inm->inm_addr, gsin->sin_addr)) {
427 * Find an IPv4 multicast source entry for this imo which matches
428 * the given group index for this socket, and source address.
430 * NOTE: This does not check if the entry is in-mode, merely if
431 * it exists, which may not be the desired behaviour.
433 static struct in_msource *
434 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src)
436 struct ip_msource find;
437 struct ip_msource *ims;
438 const sockunion_t *psa;
440 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
442 /* Source trees are keyed in host byte order. */
443 psa = (const sockunion_t *)src;
444 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
445 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
447 return ((struct in_msource *)ims);
451 * Perform filtering for multicast datagrams on a socket by group and source.
453 * Returns 0 if a datagram should be allowed through, or various error codes
454 * if the socket was not a member of the group, or the source was muted, etc.
457 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
458 const struct sockaddr *group, const struct sockaddr *src)
460 struct in_mfilter *imf;
461 struct in_msource *ims;
464 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
466 imf = imo_match_group(imo, ifp, group);
468 return (MCAST_NOTGMEMBER);
471 * Check if the source was included in an (S,G) join.
472 * Allow reception on exclusive memberships by default,
473 * reject reception on inclusive memberships by default.
474 * Exclude source only if an in-mode exclude filter exists.
475 * Include source only if an in-mode include filter exists.
476 * NOTE: We are comparing group state here at IGMP t1 (now)
477 * with socket-layer t0 (since last downcall).
479 mode = imf->imf_st[1];
480 ims = imo_match_source(imf, src);
482 if ((ims == NULL && mode == MCAST_INCLUDE) ||
483 (ims != NULL && ims->imsl_st[0] == MCAST_EXCLUDE))
484 return (MCAST_NOTSMEMBER);
490 * Find and return a reference to an in_multi record for (ifp, group),
491 * and bump its reference count.
492 * If one does not exist, try to allocate it, and update link-layer multicast
493 * filters on ifp to listen for group.
494 * Assumes the IN_MULTI lock is held across the call.
495 * Return 0 if successful, otherwise return an appropriate error code.
498 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
499 struct in_multi **pinm)
501 struct sockaddr_in gsin;
502 struct ifmultiaddr *ifma;
503 struct in_ifinfo *ii;
504 struct in_multi *inm;
507 IN_MULTI_LOCK_ASSERT();
509 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
510 IN_MULTI_LIST_LOCK();
511 inm = inm_lookup(ifp, *group);
514 * If we already joined this group, just bump the
515 * refcount and return it.
517 KASSERT(inm->inm_refcount >= 1,
518 ("%s: bad refcount %d", __func__, inm->inm_refcount));
519 inm_acquire_locked(inm);
522 IN_MULTI_LIST_UNLOCK();
526 memset(&gsin, 0, sizeof(gsin));
527 gsin.sin_family = AF_INET;
528 gsin.sin_len = sizeof(struct sockaddr_in);
529 gsin.sin_addr = *group;
532 * Check if a link-layer group is already associated
533 * with this network-layer group on the given ifnet.
535 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
539 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
540 IN_MULTI_LIST_LOCK();
544 * If something other than netinet is occupying the link-layer
545 * group, print a meaningful error message and back out of
547 * Otherwise, bump the refcount on the existing network-layer
548 * group association and return it.
550 if (ifma->ifma_protospec != NULL) {
551 inm = (struct in_multi *)ifma->ifma_protospec;
553 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
555 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
556 ("%s: ifma not AF_INET", __func__));
557 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
558 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
559 !in_hosteq(inm->inm_addr, *group)) {
560 char addrbuf[INET_ADDRSTRLEN];
562 panic("%s: ifma %p is inconsistent with %p (%s)",
563 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
566 inm_acquire_locked(inm);
571 IF_ADDR_WLOCK_ASSERT(ifp);
574 * A new in_multi record is needed; allocate and initialize it.
575 * We DO NOT perform an IGMP join as the in_ layer may need to
576 * push an initial source list down to IGMP to support SSM.
578 * The initial source filter state is INCLUDE, {} as per the RFC.
580 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
582 IF_ADDR_WUNLOCK(ifp);
583 IN_MULTI_LIST_UNLOCK();
584 if_delmulti_ifma(ifma);
587 inm->inm_addr = *group;
589 inm->inm_igi = ii->ii_igmp;
590 inm->inm_ifma = ifma;
591 inm->inm_refcount = 1;
592 inm->inm_state = IGMP_NOT_MEMBER;
593 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
594 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
595 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
596 RB_INIT(&inm->inm_srcs);
598 ifma->ifma_protospec = inm;
602 IF_ADDR_WUNLOCK(ifp);
603 IN_MULTI_LIST_UNLOCK();
608 * Drop a reference to an in_multi record.
610 * If the refcount drops to 0, free the in_multi record and
611 * delete the underlying link-layer membership.
614 inm_release(struct in_multi *inm)
616 struct ifmultiaddr *ifma;
619 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
620 MPASS(inm->inm_refcount == 0);
621 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
623 ifma = inm->inm_ifma;
626 /* XXX this access is not covered by IF_ADDR_LOCK */
627 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
629 CURVNET_SET(ifp->if_vnet);
631 free(inm, M_IPMADDR);
632 if_delmulti_ifma_flags(ifma, 1);
637 free(inm, M_IPMADDR);
638 if_delmulti_ifma_flags(ifma, 1);
643 * Clear recorded source entries for a group.
644 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
645 * FIXME: Should reap.
648 inm_clear_recorded(struct in_multi *inm)
650 struct ip_msource *ims;
652 IN_MULTI_LIST_LOCK_ASSERT();
654 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
657 --inm->inm_st[1].iss_rec;
660 KASSERT(inm->inm_st[1].iss_rec == 0,
661 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
665 * Record a source as pending for a Source-Group IGMPv3 query.
666 * This lives here as it modifies the shared tree.
668 * inm is the group descriptor.
669 * naddr is the address of the source to record in network-byte order.
671 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
672 * lazy-allocate a source node in response to an SG query.
673 * Otherwise, no allocation is performed. This saves some memory
674 * with the trade-off that the source will not be reported to the
675 * router if joined in the window between the query response and
676 * the group actually being joined on the local host.
678 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
679 * This turns off the allocation of a recorded source entry if
680 * the group has not been joined.
682 * Return 0 if the source didn't exist or was already marked as recorded.
683 * Return 1 if the source was marked as recorded by this function.
684 * Return <0 if any error occurred (negated errno code).
687 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
689 struct ip_msource find;
690 struct ip_msource *ims, *nims;
692 IN_MULTI_LIST_LOCK_ASSERT();
694 find.ims_haddr = ntohl(naddr);
695 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
696 if (ims && ims->ims_stp)
699 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
701 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
705 nims->ims_haddr = find.ims_haddr;
706 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
712 * Mark the source as recorded and update the recorded
716 ++inm->inm_st[1].iss_rec;
722 * Return a pointer to an in_msource owned by an in_mfilter,
723 * given its source address.
724 * Lazy-allocate if needed. If this is a new entry its filter state is
727 * imf is the filter set being modified.
728 * haddr is the source address in *host* byte-order.
730 * SMPng: May be called with locks held; malloc must not block.
733 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
734 struct in_msource **plims)
736 struct ip_msource find;
737 struct ip_msource *ims, *nims;
738 struct in_msource *lims;
745 /* key is host byte order */
746 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
747 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
748 lims = (struct in_msource *)ims;
750 if (imf->imf_nsrc == in_mcast_maxsocksrc)
752 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
756 lims = (struct in_msource *)nims;
757 lims->ims_haddr = find.ims_haddr;
758 lims->imsl_st[0] = MCAST_UNDEFINED;
759 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
769 * Graft a source entry into an existing socket-layer filter set,
770 * maintaining any required invariants and checking allocations.
772 * The source is marked as being in the new filter mode at t1.
774 * Return the pointer to the new node, otherwise return NULL.
776 static struct in_msource *
777 imf_graft(struct in_mfilter *imf, const uint8_t st1,
778 const struct sockaddr_in *psin)
780 struct ip_msource *nims;
781 struct in_msource *lims;
783 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
787 lims = (struct in_msource *)nims;
788 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
789 lims->imsl_st[0] = MCAST_UNDEFINED;
790 lims->imsl_st[1] = st1;
791 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
798 * Prune a source entry from an existing socket-layer filter set,
799 * maintaining any required invariants and checking allocations.
801 * The source is marked as being left at t1, it is not freed.
803 * Return 0 if no error occurred, otherwise return an errno value.
806 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
808 struct ip_msource find;
809 struct ip_msource *ims;
810 struct in_msource *lims;
812 /* key is host byte order */
813 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
814 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
817 lims = (struct in_msource *)ims;
818 lims->imsl_st[1] = MCAST_UNDEFINED;
823 * Revert socket-layer filter set deltas at t1 to t0 state.
826 imf_rollback(struct in_mfilter *imf)
828 struct ip_msource *ims, *tims;
829 struct in_msource *lims;
831 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
832 lims = (struct in_msource *)ims;
833 if (lims->imsl_st[0] == lims->imsl_st[1]) {
834 /* no change at t1 */
836 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
837 /* revert change to existing source at t1 */
838 lims->imsl_st[1] = lims->imsl_st[0];
840 /* revert source added t1 */
841 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
842 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
843 free(ims, M_INMFILTER);
847 imf->imf_st[1] = imf->imf_st[0];
851 * Mark socket-layer filter set as INCLUDE {} at t1.
854 imf_leave(struct in_mfilter *imf)
856 struct ip_msource *ims;
857 struct in_msource *lims;
859 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
860 lims = (struct in_msource *)ims;
861 lims->imsl_st[1] = MCAST_UNDEFINED;
863 imf->imf_st[1] = MCAST_INCLUDE;
867 * Mark socket-layer filter set deltas as committed.
870 imf_commit(struct in_mfilter *imf)
872 struct ip_msource *ims;
873 struct in_msource *lims;
875 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
876 lims = (struct in_msource *)ims;
877 lims->imsl_st[0] = lims->imsl_st[1];
879 imf->imf_st[0] = imf->imf_st[1];
883 * Reap unreferenced sources from socket-layer filter set.
886 imf_reap(struct in_mfilter *imf)
888 struct ip_msource *ims, *tims;
889 struct in_msource *lims;
891 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
892 lims = (struct in_msource *)ims;
893 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
894 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
895 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
896 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
897 free(ims, M_INMFILTER);
904 * Purge socket-layer filter set.
907 imf_purge(struct in_mfilter *imf)
909 struct ip_msource *ims, *tims;
911 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
912 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
913 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
914 free(ims, M_INMFILTER);
917 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
918 KASSERT(RB_EMPTY(&imf->imf_sources),
919 ("%s: imf_sources not empty", __func__));
923 * Look up a source filter entry for a multicast group.
925 * inm is the group descriptor to work with.
926 * haddr is the host-byte-order IPv4 address to look up.
927 * noalloc may be non-zero to suppress allocation of sources.
928 * *pims will be set to the address of the retrieved or allocated source.
930 * SMPng: NOTE: may be called with locks held.
931 * Return 0 if successful, otherwise return a non-zero error code.
934 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
935 const int noalloc, struct ip_msource **pims)
937 struct ip_msource find;
938 struct ip_msource *ims, *nims;
940 find.ims_haddr = haddr;
941 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
942 if (ims == NULL && !noalloc) {
943 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
945 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
949 nims->ims_haddr = haddr;
950 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
954 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
964 * Merge socket-layer source into IGMP-layer source.
965 * If rollback is non-zero, perform the inverse of the merge.
968 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
971 int n = rollback ? -1 : 1;
973 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
974 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
975 __func__, n, ims->ims_haddr);
976 ims->ims_st[1].ex -= n;
977 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
978 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
979 __func__, n, ims->ims_haddr);
980 ims->ims_st[1].in -= n;
983 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
984 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
985 __func__, n, ims->ims_haddr);
986 ims->ims_st[1].ex += n;
987 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
988 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
989 __func__, n, ims->ims_haddr);
990 ims->ims_st[1].in += n;
995 * Atomically update the global in_multi state, when a membership's
996 * filter list is being updated in any way.
998 * imf is the per-inpcb-membership group filter pointer.
999 * A fake imf may be passed for in-kernel consumers.
1001 * XXX This is a candidate for a set-symmetric-difference style loop
1002 * which would eliminate the repeated lookup from root of ims nodes,
1003 * as they share the same key space.
1005 * If any error occurred this function will back out of refcounts
1006 * and return a non-zero value.
1009 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1011 struct ip_msource *ims, *nims;
1012 struct in_msource *lims;
1013 int schanged, error;
1019 IN_MULTI_LIST_LOCK_ASSERT();
1022 * Update the source filters first, as this may fail.
1023 * Maintain count of in-mode filters at t0, t1. These are
1024 * used to work out if we transition into ASM mode or not.
1025 * Maintain a count of source filters whose state was
1026 * actually modified by this operation.
1028 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1029 lims = (struct in_msource *)ims;
1030 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1031 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1032 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1033 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1037 ims_merge(nims, lims, 0);
1040 struct ip_msource *bims;
1042 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1043 lims = (struct in_msource *)ims;
1044 if (lims->imsl_st[0] == lims->imsl_st[1])
1046 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1049 ims_merge(bims, lims, 1);
1054 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1055 __func__, nsrc0, nsrc1);
1057 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1058 if (imf->imf_st[0] == imf->imf_st[1] &&
1059 imf->imf_st[1] == MCAST_INCLUDE) {
1061 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1062 --inm->inm_st[1].iss_in;
1066 /* Handle filter mode transition on socket. */
1067 if (imf->imf_st[0] != imf->imf_st[1]) {
1068 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1069 __func__, imf->imf_st[0], imf->imf_st[1]);
1071 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1072 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1073 --inm->inm_st[1].iss_ex;
1074 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1075 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1076 --inm->inm_st[1].iss_in;
1079 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1080 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1081 inm->inm_st[1].iss_ex++;
1082 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1083 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1084 inm->inm_st[1].iss_in++;
1089 * Track inm filter state in terms of listener counts.
1090 * If there are any exclusive listeners, stack-wide
1091 * membership is exclusive.
1092 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1093 * If no listeners remain, state is undefined at t1,
1094 * and the IGMP lifecycle for this group should finish.
1096 if (inm->inm_st[1].iss_ex > 0) {
1097 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1098 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1099 } else if (inm->inm_st[1].iss_in > 0) {
1100 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1101 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1103 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1104 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1107 /* Decrement ASM listener count on transition out of ASM mode. */
1108 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1109 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1110 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1111 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1112 --inm->inm_st[1].iss_asm;
1116 /* Increment ASM listener count on transition to ASM mode. */
1117 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1118 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1119 inm->inm_st[1].iss_asm++;
1122 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1127 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1134 * Mark an in_multi's filter set deltas as committed.
1135 * Called by IGMP after a state change has been enqueued.
1138 inm_commit(struct in_multi *inm)
1140 struct ip_msource *ims;
1142 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1143 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1146 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1147 ims->ims_st[0] = ims->ims_st[1];
1149 inm->inm_st[0] = inm->inm_st[1];
1153 * Reap unreferenced nodes from an in_multi's filter set.
1156 inm_reap(struct in_multi *inm)
1158 struct ip_msource *ims, *tims;
1160 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1161 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1162 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1165 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1166 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1167 free(ims, M_IPMSOURCE);
1173 * Purge all source nodes from an in_multi's filter set.
1176 inm_purge(struct in_multi *inm)
1178 struct ip_msource *ims, *tims;
1180 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1181 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1182 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1183 free(ims, M_IPMSOURCE);
1189 * Join a multicast group; unlocked entry point.
1191 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1192 * is not held. Fortunately, ifp is unlikely to have been detached
1193 * at this point, so we assume it's OK to recurse.
1196 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1197 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1202 error = in_joingroup_locked(ifp, gina, imf, pinm);
1209 * Join a multicast group; real entry point.
1211 * Only preserves atomicity at inm level.
1212 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1214 * If the IGMP downcall fails, the group is not joined, and an error
1218 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1219 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1221 struct in_mfilter timf;
1222 struct in_multi *inm;
1225 IN_MULTI_LOCK_ASSERT();
1226 IN_MULTI_LIST_UNLOCK_ASSERT();
1228 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1229 ntohl(gina->s_addr), ifp, ifp->if_xname);
1235 * If no imf was specified (i.e. kernel consumer),
1236 * fake one up and assume it is an ASM join.
1239 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1243 error = in_getmulti(ifp, gina, &inm);
1245 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1248 IN_MULTI_LIST_LOCK();
1249 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1250 error = inm_merge(inm, imf);
1252 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1253 goto out_inm_release;
1256 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1257 error = igmp_change_state(inm);
1259 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1260 goto out_inm_release;
1265 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1267 inm_release_deferred(inm);
1268 IF_ADDR_WUNLOCK(ifp);
1272 IN_MULTI_LIST_UNLOCK();
1278 * Leave a multicast group; unlocked entry point.
1281 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1286 error = in_leavegroup_locked(inm, imf);
1293 * Leave a multicast group; real entry point.
1294 * All source filters will be expunged.
1296 * Only preserves atomicity at inm level.
1298 * Holding the write lock for the INP which contains imf
1299 * is highly advisable. We can't assert for it as imf does not
1300 * contain a back-pointer to the owning inp.
1302 * Note: This is not the same as inm_release(*) as this function also
1303 * makes a state change downcall into IGMP.
1306 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1308 struct in_mfilter timf;
1311 IN_MULTI_LOCK_ASSERT();
1312 IN_MULTI_LIST_UNLOCK_ASSERT();
1316 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1317 inm, ntohl(inm->inm_addr.s_addr),
1318 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1322 * If no imf was specified (i.e. kernel consumer),
1323 * fake one up and assume it is an ASM join.
1326 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1331 * Begin state merge transaction at IGMP layer.
1333 * As this particular invocation should not cause any memory
1334 * to be allocated, and there is no opportunity to roll back
1335 * the transaction, it MUST NOT fail.
1337 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1338 IN_MULTI_LIST_LOCK();
1339 error = inm_merge(inm, imf);
1340 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1342 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1343 CURVNET_SET(inm->inm_ifp->if_vnet);
1344 error = igmp_change_state(inm);
1345 IF_ADDR_WLOCK(inm->inm_ifp);
1346 inm_release_deferred(inm);
1347 IF_ADDR_WUNLOCK(inm->inm_ifp);
1348 IN_MULTI_LIST_UNLOCK();
1351 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1353 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1358 /*#ifndef BURN_BRIDGES*/
1361 * Block or unblock an ASM multicast source on an inpcb.
1362 * This implements the delta-based API described in RFC 3678.
1364 * The delta-based API applies only to exclusive-mode memberships.
1365 * An IGMP downcall will be performed.
1367 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1369 * Return 0 if successful, otherwise return an appropriate error code.
1372 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1374 struct epoch_tracker et;
1375 struct group_source_req gsr;
1376 sockunion_t *gsa, *ssa;
1378 struct in_mfilter *imf;
1379 struct ip_moptions *imo;
1380 struct in_msource *ims;
1381 struct in_multi *inm;
1389 memset(&gsr, 0, sizeof(struct group_source_req));
1390 gsa = (sockunion_t *)&gsr.gsr_group;
1391 ssa = (sockunion_t *)&gsr.gsr_source;
1393 switch (sopt->sopt_name) {
1394 case IP_BLOCK_SOURCE:
1395 case IP_UNBLOCK_SOURCE: {
1396 struct ip_mreq_source mreqs;
1398 error = sooptcopyin(sopt, &mreqs,
1399 sizeof(struct ip_mreq_source),
1400 sizeof(struct ip_mreq_source));
1404 gsa->sin.sin_family = AF_INET;
1405 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1406 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1408 ssa->sin.sin_family = AF_INET;
1409 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1410 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1412 if (!in_nullhost(mreqs.imr_interface)) {
1413 NET_EPOCH_ENTER(et);
1414 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1418 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1421 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1422 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1426 case MCAST_BLOCK_SOURCE:
1427 case MCAST_UNBLOCK_SOURCE:
1428 error = sooptcopyin(sopt, &gsr,
1429 sizeof(struct group_source_req),
1430 sizeof(struct group_source_req));
1434 if (gsa->sin.sin_family != AF_INET ||
1435 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1438 if (ssa->sin.sin_family != AF_INET ||
1439 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1442 NET_EPOCH_ENTER(et);
1443 ifp = ifnet_byindex(gsr.gsr_interface);
1446 return (EADDRNOTAVAIL);
1448 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1453 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1454 __func__, sopt->sopt_name);
1455 return (EOPNOTSUPP);
1459 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1465 * Check if we are actually a member of this group.
1467 imo = inp_findmoptions(inp);
1468 imf = imo_match_group(imo, ifp, &gsa->sa);
1470 error = EADDRNOTAVAIL;
1471 goto out_inp_locked;
1476 * Attempting to use the delta-based API on an
1477 * non exclusive-mode membership is an error.
1479 fmode = imf->imf_st[0];
1480 if (fmode != MCAST_EXCLUDE) {
1482 goto out_inp_locked;
1486 * Deal with error cases up-front:
1487 * Asked to block, but already blocked; or
1488 * Asked to unblock, but nothing to unblock.
1489 * If adding a new block entry, allocate it.
1491 ims = imo_match_source(imf, &ssa->sa);
1492 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1493 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1494 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1495 error = EADDRNOTAVAIL;
1496 goto out_inp_locked;
1499 INP_WLOCK_ASSERT(inp);
1502 * Begin state merge transaction at socket layer.
1505 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1506 ims = imf_graft(imf, fmode, &ssa->sin);
1510 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1511 error = imf_prune(imf, &ssa->sin);
1515 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1516 goto out_imf_rollback;
1520 * Begin state merge transaction at IGMP layer.
1522 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1523 IN_MULTI_LIST_LOCK();
1524 error = inm_merge(inm, imf);
1526 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1527 IN_MULTI_LIST_UNLOCK();
1528 goto out_imf_rollback;
1531 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1532 error = igmp_change_state(inm);
1533 IN_MULTI_LIST_UNLOCK();
1535 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1552 * Given an inpcb, return its multicast options structure pointer. Accepts
1553 * an unlocked inpcb pointer, but will return it locked. May sleep.
1555 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1556 * SMPng: NOTE: Returns with the INP write lock held.
1558 static struct ip_moptions *
1559 inp_findmoptions(struct inpcb *inp)
1561 struct ip_moptions *imo;
1564 if (inp->inp_moptions != NULL)
1565 return (inp->inp_moptions);
1569 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1571 imo->imo_multicast_ifp = NULL;
1572 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1573 imo->imo_multicast_vif = -1;
1574 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1575 imo->imo_multicast_loop = in_mcast_loop;
1576 STAILQ_INIT(&imo->imo_head);
1579 if (inp->inp_moptions != NULL) {
1580 free(imo, M_IPMOPTS);
1581 return (inp->inp_moptions);
1583 inp->inp_moptions = imo;
1588 inp_freemoptions(struct ip_moptions *imo)
1590 struct in_mfilter *imf;
1591 struct in_multi *inm;
1597 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1598 ip_mfilter_remove(&imo->imo_head, imf);
1601 if ((inm = imf->imf_inm) != NULL) {
1602 if ((ifp = inm->inm_ifp) != NULL) {
1603 CURVNET_SET(ifp->if_vnet);
1604 (void)in_leavegroup(inm, imf);
1607 (void)in_leavegroup(inm, imf);
1610 ip_mfilter_free(imf);
1612 free(imo, M_IPMOPTS);
1616 * Atomically get source filters on a socket for an IPv4 multicast group.
1617 * Called with INP lock held; returns with lock released.
1620 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1622 struct epoch_tracker et;
1623 struct __msfilterreq msfr;
1626 struct ip_moptions *imo;
1627 struct in_mfilter *imf;
1628 struct ip_msource *ims;
1629 struct in_msource *lims;
1630 struct sockaddr_in *psin;
1631 struct sockaddr_storage *ptss;
1632 struct sockaddr_storage *tss;
1634 size_t nsrcs, ncsrcs;
1636 INP_WLOCK_ASSERT(inp);
1638 imo = inp->inp_moptions;
1639 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1643 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1644 sizeof(struct __msfilterreq));
1648 NET_EPOCH_ENTER(et);
1649 ifp = ifnet_byindex(msfr.msfr_ifindex);
1650 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifnet pointer left */
1657 * Lookup group on the socket.
1659 gsa = (sockunion_t *)&msfr.msfr_group;
1660 imf = imo_match_group(imo, ifp, &gsa->sa);
1663 return (EADDRNOTAVAIL);
1667 * Ignore memberships which are in limbo.
1669 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1673 msfr.msfr_fmode = imf->imf_st[1];
1676 * If the user specified a buffer, copy out the source filter
1677 * entries to userland gracefully.
1678 * We only copy out the number of entries which userland
1679 * has asked for, but we always tell userland how big the
1680 * buffer really needs to be.
1682 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1683 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1685 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1686 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1687 M_TEMP, M_NOWAIT | M_ZERO);
1695 * Count number of sources in-mode at t0.
1696 * If buffer space exists and remains, copy out source entries.
1698 nsrcs = msfr.msfr_nsrcs;
1701 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1702 lims = (struct in_msource *)ims;
1703 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1704 lims->imsl_st[0] != imf->imf_st[0])
1707 if (tss != NULL && nsrcs > 0) {
1708 psin = (struct sockaddr_in *)ptss;
1709 psin->sin_family = AF_INET;
1710 psin->sin_len = sizeof(struct sockaddr_in);
1711 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1721 error = copyout(tss, msfr.msfr_srcs,
1722 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1728 msfr.msfr_nsrcs = ncsrcs;
1729 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1735 * Return the IP multicast options in response to user getsockopt().
1738 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1740 struct ip_mreqn mreqn;
1741 struct ip_moptions *imo;
1743 struct in_ifaddr *ia;
1748 imo = inp->inp_moptions;
1749 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM reject it. */
1750 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1751 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM) {
1753 return (EOPNOTSUPP);
1757 switch (sopt->sopt_name) {
1758 case IP_MULTICAST_VIF:
1760 optval = imo->imo_multicast_vif;
1764 error = sooptcopyout(sopt, &optval, sizeof(int));
1767 case IP_MULTICAST_IF:
1768 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1770 ifp = imo->imo_multicast_ifp;
1771 if (!in_nullhost(imo->imo_multicast_addr)) {
1772 mreqn.imr_address = imo->imo_multicast_addr;
1773 } else if (ifp != NULL) {
1774 struct epoch_tracker et;
1776 mreqn.imr_ifindex = ifp->if_index;
1777 NET_EPOCH_ENTER(et);
1781 IA_SIN(ia)->sin_addr;
1786 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1787 error = sooptcopyout(sopt, &mreqn,
1788 sizeof(struct ip_mreqn));
1790 error = sooptcopyout(sopt, &mreqn.imr_address,
1791 sizeof(struct in_addr));
1795 case IP_MULTICAST_TTL:
1797 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1799 optval = coptval = imo->imo_multicast_ttl;
1801 if (sopt->sopt_valsize == sizeof(u_char))
1802 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1804 error = sooptcopyout(sopt, &optval, sizeof(int));
1807 case IP_MULTICAST_LOOP:
1809 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1811 optval = coptval = imo->imo_multicast_loop;
1813 if (sopt->sopt_valsize == sizeof(u_char))
1814 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1816 error = sooptcopyout(sopt, &optval, sizeof(int));
1821 error = EADDRNOTAVAIL;
1824 error = inp_get_source_filters(inp, sopt);
1830 error = ENOPROTOOPT;
1834 INP_UNLOCK_ASSERT(inp);
1840 * Look up the ifnet to use for a multicast group membership,
1841 * given the IPv4 address of an interface, and the IPv4 group address.
1843 * This routine exists to support legacy multicast applications
1844 * which do not understand that multicast memberships are scoped to
1845 * specific physical links in the networking stack, or which need
1846 * to join link-scope groups before IPv4 addresses are configured.
1848 * Use this socket's current FIB number for any required FIB lookup.
1849 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1850 * and use its ifp; usually, this points to the default next-hop.
1852 * If the FIB lookup fails, attempt to use the first non-loopback
1853 * interface with multicast capability in the system as a
1854 * last resort. The legacy IPv4 ASM API requires that we do
1855 * this in order to allow groups to be joined when the routing
1856 * table has not yet been populated during boot.
1858 * Returns NULL if no ifp could be found, otherwise return referenced ifp.
1860 * FUTURE: Implement IPv4 source-address selection.
1862 static struct ifnet *
1863 inp_lookup_mcast_ifp(const struct inpcb *inp,
1864 const struct sockaddr_in *gsin, const struct in_addr ina)
1867 struct nhop_object *nh;
1870 KASSERT(inp != NULL, ("%s: inp must not be NULL", __func__));
1871 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1872 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1873 ("%s: not multicast", __func__));
1876 if (!in_nullhost(ina)) {
1877 INADDR_TO_IFP(ina, ifp);
1881 nh = fib4_lookup(inp->inp_inc.inc_fibnum, gsin->sin_addr, 0, NHR_NONE, 0);
1886 struct in_ifaddr *ia;
1890 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1892 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1893 (mifp->if_flags & IFF_MULTICAST)) {
1906 * Join an IPv4 multicast group, possibly with a source.
1909 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1911 struct group_source_req gsr;
1912 sockunion_t *gsa, *ssa;
1914 struct in_mfilter *imf;
1915 struct ip_moptions *imo;
1916 struct in_multi *inm;
1917 struct in_msource *lims;
1918 struct epoch_tracker et;
1925 memset(&gsr, 0, sizeof(struct group_source_req));
1926 gsa = (sockunion_t *)&gsr.gsr_group;
1927 gsa->ss.ss_family = AF_UNSPEC;
1928 ssa = (sockunion_t *)&gsr.gsr_source;
1929 ssa->ss.ss_family = AF_UNSPEC;
1931 switch (sopt->sopt_name) {
1932 case IP_ADD_MEMBERSHIP: {
1933 struct ip_mreqn mreqn;
1935 if (sopt->sopt_valsize == sizeof(struct ip_mreqn))
1936 error = sooptcopyin(sopt, &mreqn,
1937 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn));
1939 error = sooptcopyin(sopt, &mreqn,
1940 sizeof(struct ip_mreq), sizeof(struct ip_mreq));
1944 gsa->sin.sin_family = AF_INET;
1945 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1946 gsa->sin.sin_addr = mreqn.imr_multiaddr;
1947 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1950 NET_EPOCH_ENTER(et);
1951 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) &&
1952 mreqn.imr_ifindex != 0)
1953 ifp = ifnet_byindex_ref(mreqn.imr_ifindex);
1955 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1960 case IP_ADD_SOURCE_MEMBERSHIP: {
1961 struct ip_mreq_source mreqs;
1963 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source),
1964 sizeof(struct ip_mreq_source));
1968 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET;
1969 gsa->sin.sin_len = ssa->sin.sin_len =
1970 sizeof(struct sockaddr_in);
1972 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1973 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1976 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1978 NET_EPOCH_ENTER(et);
1979 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1980 mreqs.imr_interface);
1982 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1983 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1987 case MCAST_JOIN_GROUP:
1988 case MCAST_JOIN_SOURCE_GROUP:
1989 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1990 error = sooptcopyin(sopt, &gsr,
1991 sizeof(struct group_req),
1992 sizeof(struct group_req));
1993 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1994 error = sooptcopyin(sopt, &gsr,
1995 sizeof(struct group_source_req),
1996 sizeof(struct group_source_req));
2001 if (gsa->sin.sin_family != AF_INET ||
2002 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2006 * Overwrite the port field if present, as the sockaddr
2007 * being copied in may be matched with a binary comparison.
2009 gsa->sin.sin_port = 0;
2010 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2011 if (ssa->sin.sin_family != AF_INET ||
2012 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2014 ssa->sin.sin_port = 0;
2017 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2020 NET_EPOCH_ENTER(et);
2021 ifp = ifnet_byindex_ref(gsr.gsr_interface);
2024 return (EADDRNOTAVAIL);
2028 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2029 __func__, sopt->sopt_name);
2030 return (EOPNOTSUPP);
2034 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2037 return (EADDRNOTAVAIL);
2043 * Find the membership in the membership list.
2045 imo = inp_findmoptions(inp);
2046 imf = imo_match_group(imo, ifp, &gsa->sa);
2051 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) {
2053 goto out_inp_locked;
2059 if (ssa->ss.ss_family != AF_UNSPEC) {
2061 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2062 * is an error. On an existing inclusive membership,
2063 * it just adds the source to the filter list.
2065 if (imf->imf_st[1] != MCAST_INCLUDE) {
2067 goto out_inp_locked;
2070 * Throw out duplicates.
2072 * XXX FIXME: This makes a naive assumption that
2073 * even if entries exist for *ssa in this imf,
2074 * they will be rejected as dupes, even if they
2075 * are not valid in the current mode (in-mode).
2077 * in_msource is transactioned just as for anything
2078 * else in SSM -- but note naive use of inm_graft()
2079 * below for allocating new filter entries.
2081 * This is only an issue if someone mixes the
2082 * full-state SSM API with the delta-based API,
2083 * which is discouraged in the relevant RFCs.
2085 lims = imo_match_source(imf, &ssa->sa);
2086 if (lims != NULL /*&&
2087 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2088 error = EADDRNOTAVAIL;
2089 goto out_inp_locked;
2093 * MCAST_JOIN_GROUP on an existing exclusive
2094 * membership is an error; return EADDRINUSE
2095 * to preserve 4.4BSD API idempotence, and
2096 * avoid tedious detour to code below.
2097 * NOTE: This is bending RFC 3678 a bit.
2099 * On an existing inclusive membership, this is also
2100 * an error; if you want to change filter mode,
2101 * you must use the userland API setsourcefilter().
2102 * XXX We don't reject this for imf in UNDEFINED
2103 * state at t1, because allocation of a filter
2104 * is atomic with allocation of a membership.
2107 if (imf->imf_st[1] == MCAST_EXCLUDE)
2109 goto out_inp_locked;
2114 * Begin state merge transaction at socket layer.
2116 INP_WLOCK_ASSERT(inp);
2119 * Graft new source into filter list for this inpcb's
2120 * membership of the group. The in_multi may not have
2121 * been allocated yet if this is a new membership, however,
2122 * the in_mfilter slot will be allocated and must be initialized.
2124 * Note: Grafting of exclusive mode filters doesn't happen
2126 * XXX: Should check for non-NULL lims (node exists but may
2127 * not be in-mode) for interop with full-state API.
2129 if (ssa->ss.ss_family != AF_UNSPEC) {
2130 /* Membership starts in IN mode */
2132 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2133 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2136 goto out_inp_locked;
2139 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2141 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2143 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2146 goto out_inp_locked;
2149 /* No address specified; Membership starts in EX mode */
2151 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2152 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2155 goto out_inp_locked;
2161 * Begin state merge transaction at IGMP layer.
2167 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2171 if (in_pcbrele_wlocked(inp)) {
2173 goto out_inp_unlocked;
2176 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2178 goto out_inp_locked;
2181 * NOTE: Refcount from in_joingroup_locked()
2182 * is protecting membership.
2184 ip_mfilter_insert(&imo->imo_head, imf);
2186 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2187 IN_MULTI_LIST_LOCK();
2188 error = inm_merge(inm, imf);
2190 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2192 IN_MULTI_LIST_UNLOCK();
2195 goto out_inp_locked;
2197 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2198 error = igmp_change_state(inm);
2199 IN_MULTI_LIST_UNLOCK();
2201 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2205 goto out_inp_locked;
2217 if (is_new && imf) {
2218 if (imf->imf_inm != NULL) {
2219 IN_MULTI_LIST_LOCK();
2221 inm_release_deferred(imf->imf_inm);
2222 IF_ADDR_WUNLOCK(ifp);
2223 IN_MULTI_LIST_UNLOCK();
2225 ip_mfilter_free(imf);
2232 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2235 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2237 struct epoch_tracker et;
2238 struct group_source_req gsr;
2239 struct ip_mreq_source mreqs;
2240 sockunion_t *gsa, *ssa;
2242 struct in_mfilter *imf;
2243 struct ip_moptions *imo;
2244 struct in_msource *ims;
2245 struct in_multi *inm;
2253 memset(&gsr, 0, sizeof(struct group_source_req));
2254 gsa = (sockunion_t *)&gsr.gsr_group;
2255 gsa->ss.ss_family = AF_UNSPEC;
2256 ssa = (sockunion_t *)&gsr.gsr_source;
2257 ssa->ss.ss_family = AF_UNSPEC;
2259 switch (sopt->sopt_name) {
2260 case IP_DROP_MEMBERSHIP:
2261 case IP_DROP_SOURCE_MEMBERSHIP:
2262 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2263 error = sooptcopyin(sopt, &mreqs,
2264 sizeof(struct ip_mreq),
2265 sizeof(struct ip_mreq));
2267 * Swap interface and sourceaddr arguments,
2268 * as ip_mreq and ip_mreq_source are laid
2271 mreqs.imr_interface = mreqs.imr_sourceaddr;
2272 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2273 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2274 error = sooptcopyin(sopt, &mreqs,
2275 sizeof(struct ip_mreq_source),
2276 sizeof(struct ip_mreq_source));
2281 gsa->sin.sin_family = AF_INET;
2282 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2283 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2285 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2286 ssa->sin.sin_family = AF_INET;
2287 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2288 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2292 * Attempt to look up hinted ifp from interface address.
2293 * Fallthrough with null ifp iff lookup fails, to
2294 * preserve 4.4BSD mcast API idempotence.
2295 * XXX NOTE WELL: The RFC 3678 API is preferred because
2296 * using an IPv4 address as a key is racy.
2298 if (!in_nullhost(mreqs.imr_interface)) {
2299 NET_EPOCH_ENTER(et);
2300 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2304 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2305 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2309 case MCAST_LEAVE_GROUP:
2310 case MCAST_LEAVE_SOURCE_GROUP:
2311 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2312 error = sooptcopyin(sopt, &gsr,
2313 sizeof(struct group_req),
2314 sizeof(struct group_req));
2315 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2316 error = sooptcopyin(sopt, &gsr,
2317 sizeof(struct group_source_req),
2318 sizeof(struct group_source_req));
2323 if (gsa->sin.sin_family != AF_INET ||
2324 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2327 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2328 if (ssa->sin.sin_family != AF_INET ||
2329 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2333 NET_EPOCH_ENTER(et);
2334 ifp = ifnet_byindex(gsr.gsr_interface);
2335 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2337 return (EADDRNOTAVAIL);
2341 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2342 __func__, sopt->sopt_name);
2343 return (EOPNOTSUPP);
2347 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2353 * Find the membership in the membership list.
2355 imo = inp_findmoptions(inp);
2356 imf = imo_match_group(imo, ifp, &gsa->sa);
2358 error = EADDRNOTAVAIL;
2359 goto out_inp_locked;
2363 if (ssa->ss.ss_family != AF_UNSPEC)
2367 * Begin state merge transaction at socket layer.
2369 INP_WLOCK_ASSERT(inp);
2372 * If we were instructed only to leave a given source, do so.
2373 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2376 ip_mfilter_remove(&imo->imo_head, imf);
2380 * Give up the multicast address record to which
2381 * the membership points.
2383 (void) in_leavegroup_locked(imf->imf_inm, imf);
2385 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2386 error = EADDRNOTAVAIL;
2387 goto out_inp_locked;
2389 ims = imo_match_source(imf, &ssa->sa);
2391 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2392 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2393 error = EADDRNOTAVAIL;
2394 goto out_inp_locked;
2396 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2397 error = imf_prune(imf, &ssa->sin);
2399 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2401 goto out_inp_locked;
2406 * Begin state merge transaction at IGMP layer.
2409 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2410 IN_MULTI_LIST_LOCK();
2411 error = inm_merge(inm, imf);
2413 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2415 IN_MULTI_LIST_UNLOCK();
2418 goto out_inp_locked;
2421 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2422 error = igmp_change_state(inm);
2423 IN_MULTI_LIST_UNLOCK();
2425 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2429 goto out_inp_locked;
2438 if (is_final && imf)
2439 ip_mfilter_free(imf);
2446 * Select the interface for transmitting IPv4 multicast datagrams.
2448 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2449 * may be passed to this socket option. An address of INADDR_ANY or an
2450 * interface index of 0 is used to remove a previous selection.
2451 * When no interface is selected, one is chosen for every send.
2454 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2456 struct in_addr addr;
2457 struct ip_mreqn mreqn;
2459 struct ip_moptions *imo;
2462 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2464 * An interface index was specified using the
2465 * Linux-derived ip_mreqn structure.
2467 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2468 sizeof(struct ip_mreqn));
2472 if (mreqn.imr_ifindex < 0)
2475 if (mreqn.imr_ifindex == 0) {
2478 struct epoch_tracker et;
2480 NET_EPOCH_ENTER(et);
2481 ifp = ifnet_byindex(mreqn.imr_ifindex);
2482 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2484 return (EADDRNOTAVAIL);
2488 * An interface was specified by IPv4 address.
2489 * This is the traditional BSD usage.
2491 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2492 sizeof(struct in_addr));
2495 if (in_nullhost(addr)) {
2498 struct epoch_tracker et;
2500 NET_EPOCH_ENTER(et);
2501 INADDR_TO_IFP(addr, ifp);
2505 return (EADDRNOTAVAIL);
2507 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2508 ntohl(addr.s_addr));
2511 /* Reject interfaces which do not support multicast. */
2512 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2513 return (EOPNOTSUPP);
2515 imo = inp_findmoptions(inp);
2516 imo->imo_multicast_ifp = ifp;
2517 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2524 * Atomically set source filters on a socket for an IPv4 multicast group.
2526 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2529 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2531 struct epoch_tracker et;
2532 struct __msfilterreq msfr;
2535 struct in_mfilter *imf;
2536 struct ip_moptions *imo;
2537 struct in_multi *inm;
2540 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2541 sizeof(struct __msfilterreq));
2545 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2548 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2549 msfr.msfr_fmode != MCAST_INCLUDE))
2552 if (msfr.msfr_group.ss_family != AF_INET ||
2553 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2556 gsa = (sockunion_t *)&msfr.msfr_group;
2557 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2560 gsa->sin.sin_port = 0; /* ignore port */
2562 NET_EPOCH_ENTER(et);
2563 ifp = ifnet_byindex(msfr.msfr_ifindex);
2564 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2566 return (EADDRNOTAVAIL);
2571 * Take the INP write lock.
2572 * Check if this socket is a member of this group.
2574 imo = inp_findmoptions(inp);
2575 imf = imo_match_group(imo, ifp, &gsa->sa);
2577 error = EADDRNOTAVAIL;
2578 goto out_inp_locked;
2583 * Begin state merge transaction at socket layer.
2585 INP_WLOCK_ASSERT(inp);
2587 imf->imf_st[1] = msfr.msfr_fmode;
2590 * Apply any new source filters, if present.
2591 * Make a copy of the user-space source vector so
2592 * that we may copy them with a single copyin. This
2593 * allows us to deal with page faults up-front.
2595 if (msfr.msfr_nsrcs > 0) {
2596 struct in_msource *lims;
2597 struct sockaddr_in *psin;
2598 struct sockaddr_storage *kss, *pkss;
2603 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2604 __func__, (unsigned long)msfr.msfr_nsrcs);
2605 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2607 error = copyin(msfr.msfr_srcs, kss,
2608 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2617 * Mark all source filters as UNDEFINED at t1.
2618 * Restore new group filter mode, as imf_leave()
2619 * will set it to INCLUDE.
2622 imf->imf_st[1] = msfr.msfr_fmode;
2625 * Update socket layer filters at t1, lazy-allocating
2626 * new entries. This saves a bunch of memory at the
2627 * cost of one RB_FIND() per source entry; duplicate
2628 * entries in the msfr_nsrcs vector are ignored.
2629 * If we encounter an error, rollback transaction.
2631 * XXX This too could be replaced with a set-symmetric
2632 * difference like loop to avoid walking from root
2633 * every time, as the key space is common.
2635 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2636 psin = (struct sockaddr_in *)pkss;
2637 if (psin->sin_family != AF_INET) {
2638 error = EAFNOSUPPORT;
2641 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2645 error = imf_get_source(imf, psin, &lims);
2648 lims->imsl_st[1] = imf->imf_st[1];
2654 goto out_imf_rollback;
2656 INP_WLOCK_ASSERT(inp);
2659 * Begin state merge transaction at IGMP layer.
2661 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2662 IN_MULTI_LIST_LOCK();
2663 error = inm_merge(inm, imf);
2665 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2666 IN_MULTI_LIST_UNLOCK();
2667 goto out_imf_rollback;
2670 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2671 error = igmp_change_state(inm);
2672 IN_MULTI_LIST_UNLOCK();
2674 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2691 * Set the IP multicast options in response to user setsockopt().
2693 * Many of the socket options handled in this function duplicate the
2694 * functionality of socket options in the regular unicast API. However,
2695 * it is not possible to merge the duplicate code, because the idempotence
2696 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2697 * the effects of these options must be treated as separate and distinct.
2699 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2700 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2701 * is refactored to no longer use vifs.
2704 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2706 struct ip_moptions *imo;
2711 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
2712 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2713 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)
2714 return (EOPNOTSUPP);
2716 switch (sopt->sopt_name) {
2717 case IP_MULTICAST_VIF: {
2720 * Select a multicast VIF for transmission.
2721 * Only useful if multicast forwarding is active.
2723 if (legal_vif_num == NULL) {
2727 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2730 if (!legal_vif_num(vifi) && (vifi != -1)) {
2734 imo = inp_findmoptions(inp);
2735 imo->imo_multicast_vif = vifi;
2740 case IP_MULTICAST_IF:
2741 error = inp_set_multicast_if(inp, sopt);
2744 case IP_MULTICAST_TTL: {
2748 * Set the IP time-to-live for outgoing multicast packets.
2749 * The original multicast API required a char argument,
2750 * which is inconsistent with the rest of the socket API.
2751 * We allow either a char or an int.
2753 if (sopt->sopt_valsize == sizeof(u_char)) {
2754 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2761 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2771 imo = inp_findmoptions(inp);
2772 imo->imo_multicast_ttl = ttl;
2777 case IP_MULTICAST_LOOP: {
2781 * Set the loopback flag for outgoing multicast packets.
2782 * Must be zero or one. The original multicast API required a
2783 * char argument, which is inconsistent with the rest
2784 * of the socket API. We allow either a char or an int.
2786 if (sopt->sopt_valsize == sizeof(u_char)) {
2787 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2794 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2798 loop = (u_char)iloop;
2800 imo = inp_findmoptions(inp);
2801 imo->imo_multicast_loop = !!loop;
2806 case IP_ADD_MEMBERSHIP:
2807 case IP_ADD_SOURCE_MEMBERSHIP:
2808 case MCAST_JOIN_GROUP:
2809 case MCAST_JOIN_SOURCE_GROUP:
2810 error = inp_join_group(inp, sopt);
2813 case IP_DROP_MEMBERSHIP:
2814 case IP_DROP_SOURCE_MEMBERSHIP:
2815 case MCAST_LEAVE_GROUP:
2816 case MCAST_LEAVE_SOURCE_GROUP:
2817 error = inp_leave_group(inp, sopt);
2820 case IP_BLOCK_SOURCE:
2821 case IP_UNBLOCK_SOURCE:
2822 case MCAST_BLOCK_SOURCE:
2823 case MCAST_UNBLOCK_SOURCE:
2824 error = inp_block_unblock_source(inp, sopt);
2828 error = inp_set_source_filters(inp, sopt);
2836 INP_UNLOCK_ASSERT(inp);
2842 * Expose IGMP's multicast filter mode and source list(s) to userland,
2843 * keyed by (ifindex, group).
2844 * The filter mode is written out as a uint32_t, followed by
2845 * 0..n of struct in_addr.
2846 * For use by ifmcstat(8).
2847 * SMPng: NOTE: unlocked read of ifindex space.
2850 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2852 struct in_addr src, group;
2853 struct epoch_tracker et;
2855 struct ifmultiaddr *ifma;
2856 struct in_multi *inm;
2857 struct ip_msource *ims;
2861 uint32_t fmode, ifindex;
2866 if (req->newptr != NULL)
2872 group.s_addr = name[1];
2873 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2874 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2875 __func__, ntohl(group.s_addr));
2880 NET_EPOCH_ENTER(et);
2881 ifp = ifnet_byindex(ifindex);
2884 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2889 retval = sysctl_wire_old_buffer(req,
2890 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2896 IN_MULTI_LIST_LOCK();
2898 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2899 inm = inm_ifmultiaddr_get_inm(ifma);
2902 if (!in_hosteq(inm->inm_addr, group))
2904 fmode = inm->inm_st[1].iss_fmode;
2905 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2908 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2909 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2912 * Only copy-out sources which are in-mode.
2914 if (fmode != ims_get_mode(inm, ims, 1)) {
2915 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2919 src.s_addr = htonl(ims->ims_haddr);
2920 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2926 IN_MULTI_LIST_UNLOCK();
2932 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2934 static const char *inm_modestrs[] = {
2935 [MCAST_UNDEFINED] = "un",
2936 [MCAST_INCLUDE] = "in",
2937 [MCAST_EXCLUDE] = "ex",
2939 _Static_assert(MCAST_UNDEFINED == 0 &&
2940 MCAST_EXCLUDE + 1 == nitems(inm_modestrs),
2941 "inm_modestrs: no longer matches #defines");
2944 inm_mode_str(const int mode)
2947 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2948 return (inm_modestrs[mode]);
2952 static const char *inm_statestrs[] = {
2953 [IGMP_NOT_MEMBER] = "not-member",
2954 [IGMP_SILENT_MEMBER] = "silent",
2955 [IGMP_REPORTING_MEMBER] = "reporting",
2956 [IGMP_IDLE_MEMBER] = "idle",
2957 [IGMP_LAZY_MEMBER] = "lazy",
2958 [IGMP_SLEEPING_MEMBER] = "sleeping",
2959 [IGMP_AWAKENING_MEMBER] = "awakening",
2960 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending",
2961 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending",
2962 [IGMP_LEAVING_MEMBER] = "leaving",
2964 _Static_assert(IGMP_NOT_MEMBER == 0 &&
2965 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs),
2966 "inm_statetrs: no longer matches #defines");
2969 inm_state_str(const int state)
2972 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2973 return (inm_statestrs[state]);
2978 * Dump an in_multi structure to the console.
2981 inm_print(const struct in_multi *inm)
2984 char addrbuf[INET_ADDRSTRLEN];
2986 if ((ktr_mask & KTR_IGMPV3) == 0)
2989 printf("%s: --- begin inm %p ---\n", __func__, inm);
2990 printf("addr %s ifp %p(%s) ifma %p\n",
2991 inet_ntoa_r(inm->inm_addr, addrbuf),
2993 inm->inm_ifp->if_xname,
2995 printf("timer %u state %s refcount %u scq.len %u\n",
2997 inm_state_str(inm->inm_state),
2999 inm->inm_scq.mq_len);
3000 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3005 for (t = 0; t < 2; t++) {
3006 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3007 inm_mode_str(inm->inm_st[t].iss_fmode),
3008 inm->inm_st[t].iss_asm,
3009 inm->inm_st[t].iss_ex,
3010 inm->inm_st[t].iss_in,
3011 inm->inm_st[t].iss_rec);
3013 printf("%s: --- end inm %p ---\n", __func__, inm);
3016 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3019 inm_print(const struct in_multi *inm)
3024 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3026 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);