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 <netinet/ip_var.h>
70 #include <netinet/igmp_var.h>
73 #define KTR_IGMPV3 KTR_INET
76 #ifndef __SOCKUNION_DECLARED
78 struct sockaddr_storage ss;
80 struct sockaddr_dl sdl;
81 struct sockaddr_in sin;
83 typedef union sockunion sockunion_t;
84 #define __SOCKUNION_DECLARED
85 #endif /* __SOCKUNION_DECLARED */
87 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
88 "IPv4 multicast PCB-layer source filter");
89 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
90 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
91 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
92 "IPv4 multicast IGMP-layer source filter");
97 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK,
98 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
99 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
100 * it can be taken by code in net/if.c also.
101 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
103 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
104 * any need for in_multi itself to be virtualized -- it is bound to an ifp
105 * anyway no matter what happens.
107 struct mtx in_multi_list_mtx;
108 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
110 struct mtx in_multi_free_mtx;
111 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
113 struct sx in_multi_sx;
114 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
117 * Functions with non-static linkage defined in this file should be
118 * declared in in_var.h:
121 * in_joingroup_locked()
123 * in_leavegroup_locked()
129 static void imf_commit(struct in_mfilter *);
130 static int imf_get_source(struct in_mfilter *imf,
131 const struct sockaddr_in *psin,
132 struct in_msource **);
133 static struct in_msource *
134 imf_graft(struct in_mfilter *, const uint8_t,
135 const struct sockaddr_in *);
136 static void imf_leave(struct in_mfilter *);
137 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
138 static void imf_purge(struct in_mfilter *);
139 static void imf_rollback(struct in_mfilter *);
140 static void imf_reap(struct in_mfilter *);
141 static struct in_mfilter *
142 imo_match_group(const struct ip_moptions *,
143 const struct ifnet *, const struct sockaddr *);
144 static struct in_msource *
145 imo_match_source(struct in_mfilter *, const struct sockaddr *);
146 static void ims_merge(struct ip_msource *ims,
147 const struct in_msource *lims, const int rollback);
148 static int in_getmulti(struct ifnet *, const struct in_addr *,
150 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
151 const int noalloc, struct ip_msource **pims);
153 static int inm_is_ifp_detached(const struct in_multi *);
155 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
156 static void inm_purge(struct in_multi *);
157 static void inm_reap(struct in_multi *);
158 static void inm_release(struct in_multi *);
159 static struct ip_moptions *
160 inp_findmoptions(struct inpcb *);
161 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
162 static int inp_join_group(struct inpcb *, struct sockopt *);
163 static int inp_leave_group(struct inpcb *, struct sockopt *);
164 static struct ifnet *
165 inp_lookup_mcast_ifp(const struct inpcb *,
166 const struct sockaddr_in *, const struct in_addr);
167 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
168 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
169 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
170 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
172 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast,
173 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
176 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
177 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
178 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
179 "Max source filters per group");
181 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
182 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
183 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
184 "Max source filters per socket");
186 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
187 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
188 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
190 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
191 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
192 "Per-interface stack-wide source filters");
196 * Inline function which wraps assertions for a valid ifp.
197 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
201 inm_is_ifp_detached(const struct in_multi *inm)
205 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
206 ifp = inm->inm_ifma->ifma_ifp;
209 * Sanity check that netinet's notion of ifp is the
212 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
215 return (ifp == NULL);
220 * Interface detach can happen in a taskqueue thread context, so we must use a
221 * dedicated thread to avoid deadlocks when draining inm_release tasks.
223 TASKQUEUE_DEFINE_THREAD(inm_free);
224 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER();
225 static void inm_release_task(void *arg __unused, int pending __unused);
226 static struct task inm_free_task = TASK_INITIALIZER(0, inm_release_task, NULL);
229 inm_release_wait(void *arg __unused)
233 * Make sure all pending multicast addresses are freed before
234 * the VNET or network device is destroyed:
236 taskqueue_drain(taskqueue_inm_free, &inm_free_task);
239 /* XXX-BZ FIXME, see D24914. */
240 VNET_SYSUNINIT(inm_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, inm_release_wait, NULL);
244 inm_release_list_deferred(struct in_multi_head *inmh)
247 if (SLIST_EMPTY(inmh))
249 mtx_lock(&in_multi_free_mtx);
250 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
251 mtx_unlock(&in_multi_free_mtx);
252 taskqueue_enqueue(taskqueue_inm_free, &inm_free_task);
256 inm_disconnect(struct in_multi *inm)
259 struct ifmultiaddr *ifma, *ll_ifma;
262 IF_ADDR_WLOCK_ASSERT(ifp);
263 ifma = inm->inm_ifma;
266 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
267 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
268 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
270 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
271 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
272 MPASS(ifma != ll_ifma);
273 ifma->ifma_llifma = NULL;
274 MPASS(ll_ifma->ifma_llifma == NULL);
275 MPASS(ll_ifma->ifma_ifp == ifp);
276 if (--ll_ifma->ifma_refcount == 0) {
277 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
278 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
279 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
281 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
282 if_freemulti(ll_ifma);
288 inm_release_deferred(struct in_multi *inm)
290 struct in_multi_head tmp;
292 IN_MULTI_LIST_LOCK_ASSERT();
293 MPASS(inm->inm_refcount > 0);
294 if (--inm->inm_refcount == 0) {
297 inm->inm_ifma->ifma_protospec = NULL;
298 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
299 inm_release_list_deferred(&tmp);
304 inm_release_task(void *arg __unused, int pending __unused)
306 struct in_multi_head inm_free_tmp;
307 struct in_multi *inm, *tinm;
309 SLIST_INIT(&inm_free_tmp);
310 mtx_lock(&in_multi_free_mtx);
311 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
312 mtx_unlock(&in_multi_free_mtx);
314 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
315 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
323 * Initialize an in_mfilter structure to a known state at t0, t1
324 * with an empty source filter list.
327 imf_init(struct in_mfilter *imf, const int st0, const int st1)
329 memset(imf, 0, sizeof(struct in_mfilter));
330 RB_INIT(&imf->imf_sources);
331 imf->imf_st[0] = st0;
332 imf->imf_st[1] = st1;
336 ip_mfilter_alloc(const int mflags, const int st0, const int st1)
338 struct in_mfilter *imf;
340 imf = malloc(sizeof(*imf), M_INMFILTER, mflags);
342 imf_init(imf, st0, st1);
348 ip_mfilter_free(struct in_mfilter *imf)
352 free(imf, M_INMFILTER);
356 * Function for looking up an in_multi record for an IPv4 multicast address
357 * on a given interface. ifp must be valid. If no record found, return NULL.
358 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
361 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
363 struct ifmultiaddr *ifma;
364 struct in_multi *inm;
366 IN_MULTI_LIST_LOCK_ASSERT();
367 IF_ADDR_LOCK_ASSERT(ifp);
369 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
370 inm = inm_ifmultiaddr_get_inm(ifma);
373 if (inm->inm_addr.s_addr == ina.s_addr)
380 * Wrapper for inm_lookup_locked().
381 * The IF_ADDR_LOCK will be taken on ifp and released on return.
384 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
386 struct epoch_tracker et;
387 struct in_multi *inm;
389 IN_MULTI_LIST_LOCK_ASSERT();
392 inm = inm_lookup_locked(ifp, ina);
399 * Find an IPv4 multicast group entry for this ip_moptions instance
400 * which matches the specified group, and optionally an interface.
401 * Return its index into the array, or -1 if not found.
403 static struct in_mfilter *
404 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
405 const struct sockaddr *group)
407 const struct sockaddr_in *gsin;
408 struct in_mfilter *imf;
409 struct in_multi *inm;
411 gsin = (const struct sockaddr_in *)group;
413 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
417 if ((ifp == NULL || (inm->inm_ifp == ifp)) &&
418 in_hosteq(inm->inm_addr, gsin->sin_addr)) {
426 * Find an IPv4 multicast source entry for this imo which matches
427 * the given group index for this socket, and source address.
429 * NOTE: This does not check if the entry is in-mode, merely if
430 * it exists, which may not be the desired behaviour.
432 static struct in_msource *
433 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src)
435 struct ip_msource find;
436 struct ip_msource *ims;
437 const sockunion_t *psa;
439 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
441 /* Source trees are keyed in host byte order. */
442 psa = (const sockunion_t *)src;
443 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
444 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
446 return ((struct in_msource *)ims);
450 * Perform filtering for multicast datagrams on a socket by group and source.
452 * Returns 0 if a datagram should be allowed through, or various error codes
453 * if the socket was not a member of the group, or the source was muted, etc.
456 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
457 const struct sockaddr *group, const struct sockaddr *src)
459 struct in_mfilter *imf;
460 struct in_msource *ims;
463 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
465 imf = imo_match_group(imo, ifp, group);
467 return (MCAST_NOTGMEMBER);
470 * Check if the source was included in an (S,G) join.
471 * Allow reception on exclusive memberships by default,
472 * reject reception on inclusive memberships by default.
473 * Exclude source only if an in-mode exclude filter exists.
474 * Include source only if an in-mode include filter exists.
475 * NOTE: We are comparing group state here at IGMP t1 (now)
476 * with socket-layer t0 (since last downcall).
478 mode = imf->imf_st[1];
479 ims = imo_match_source(imf, src);
481 if ((ims == NULL && mode == MCAST_INCLUDE) ||
482 (ims != NULL && ims->imsl_st[0] != mode))
483 return (MCAST_NOTSMEMBER);
489 * Find and return a reference to an in_multi record for (ifp, group),
490 * and bump its reference count.
491 * If one does not exist, try to allocate it, and update link-layer multicast
492 * filters on ifp to listen for group.
493 * Assumes the IN_MULTI lock is held across the call.
494 * Return 0 if successful, otherwise return an appropriate error code.
497 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
498 struct in_multi **pinm)
500 struct sockaddr_in gsin;
501 struct ifmultiaddr *ifma;
502 struct in_ifinfo *ii;
503 struct in_multi *inm;
506 IN_MULTI_LOCK_ASSERT();
508 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
509 IN_MULTI_LIST_LOCK();
510 inm = inm_lookup(ifp, *group);
513 * If we already joined this group, just bump the
514 * refcount and return it.
516 KASSERT(inm->inm_refcount >= 1,
517 ("%s: bad refcount %d", __func__, inm->inm_refcount));
518 inm_acquire_locked(inm);
521 IN_MULTI_LIST_UNLOCK();
525 memset(&gsin, 0, sizeof(gsin));
526 gsin.sin_family = AF_INET;
527 gsin.sin_len = sizeof(struct sockaddr_in);
528 gsin.sin_addr = *group;
531 * Check if a link-layer group is already associated
532 * with this network-layer group on the given ifnet.
534 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
538 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
539 IN_MULTI_LIST_LOCK();
543 * If something other than netinet is occupying the link-layer
544 * group, print a meaningful error message and back out of
546 * Otherwise, bump the refcount on the existing network-layer
547 * group association and return it.
549 if (ifma->ifma_protospec != NULL) {
550 inm = (struct in_multi *)ifma->ifma_protospec;
552 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
554 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
555 ("%s: ifma not AF_INET", __func__));
556 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
557 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
558 !in_hosteq(inm->inm_addr, *group)) {
559 char addrbuf[INET_ADDRSTRLEN];
561 panic("%s: ifma %p is inconsistent with %p (%s)",
562 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
565 inm_acquire_locked(inm);
570 IF_ADDR_WLOCK_ASSERT(ifp);
573 * A new in_multi record is needed; allocate and initialize it.
574 * We DO NOT perform an IGMP join as the in_ layer may need to
575 * push an initial source list down to IGMP to support SSM.
577 * The initial source filter state is INCLUDE, {} as per the RFC.
579 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
581 IF_ADDR_WUNLOCK(ifp);
582 IN_MULTI_LIST_UNLOCK();
583 if_delmulti_ifma(ifma);
586 inm->inm_addr = *group;
588 inm->inm_igi = ii->ii_igmp;
589 inm->inm_ifma = ifma;
590 inm->inm_refcount = 1;
591 inm->inm_state = IGMP_NOT_MEMBER;
592 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
593 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
594 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
595 RB_INIT(&inm->inm_srcs);
597 ifma->ifma_protospec = inm;
601 IF_ADDR_WUNLOCK(ifp);
602 IN_MULTI_LIST_UNLOCK();
607 * Drop a reference to an in_multi record.
609 * If the refcount drops to 0, free the in_multi record and
610 * delete the underlying link-layer membership.
613 inm_release(struct in_multi *inm)
615 struct ifmultiaddr *ifma;
618 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
619 MPASS(inm->inm_refcount == 0);
620 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
622 ifma = inm->inm_ifma;
625 /* XXX this access is not covered by IF_ADDR_LOCK */
626 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
628 CURVNET_SET(ifp->if_vnet);
630 free(inm, M_IPMADDR);
631 if_delmulti_ifma_flags(ifma, 1);
636 free(inm, M_IPMADDR);
637 if_delmulti_ifma_flags(ifma, 1);
642 * Clear recorded source entries for a group.
643 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
644 * FIXME: Should reap.
647 inm_clear_recorded(struct in_multi *inm)
649 struct ip_msource *ims;
651 IN_MULTI_LIST_LOCK_ASSERT();
653 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
656 --inm->inm_st[1].iss_rec;
659 KASSERT(inm->inm_st[1].iss_rec == 0,
660 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
664 * Record a source as pending for a Source-Group IGMPv3 query.
665 * This lives here as it modifies the shared tree.
667 * inm is the group descriptor.
668 * naddr is the address of the source to record in network-byte order.
670 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
671 * lazy-allocate a source node in response to an SG query.
672 * Otherwise, no allocation is performed. This saves some memory
673 * with the trade-off that the source will not be reported to the
674 * router if joined in the window between the query response and
675 * the group actually being joined on the local host.
677 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
678 * This turns off the allocation of a recorded source entry if
679 * the group has not been joined.
681 * Return 0 if the source didn't exist or was already marked as recorded.
682 * Return 1 if the source was marked as recorded by this function.
683 * Return <0 if any error occurred (negated errno code).
686 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
688 struct ip_msource find;
689 struct ip_msource *ims, *nims;
691 IN_MULTI_LIST_LOCK_ASSERT();
693 find.ims_haddr = ntohl(naddr);
694 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
695 if (ims && ims->ims_stp)
698 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
700 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
704 nims->ims_haddr = find.ims_haddr;
705 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
711 * Mark the source as recorded and update the recorded
715 ++inm->inm_st[1].iss_rec;
721 * Return a pointer to an in_msource owned by an in_mfilter,
722 * given its source address.
723 * Lazy-allocate if needed. If this is a new entry its filter state is
726 * imf is the filter set being modified.
727 * haddr is the source address in *host* byte-order.
729 * SMPng: May be called with locks held; malloc must not block.
732 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
733 struct in_msource **plims)
735 struct ip_msource find;
736 struct ip_msource *ims, *nims;
737 struct in_msource *lims;
744 /* key is host byte order */
745 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
746 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
747 lims = (struct in_msource *)ims;
749 if (imf->imf_nsrc == in_mcast_maxsocksrc)
751 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
755 lims = (struct in_msource *)nims;
756 lims->ims_haddr = find.ims_haddr;
757 lims->imsl_st[0] = MCAST_UNDEFINED;
758 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
768 * Graft a source entry into an existing socket-layer filter set,
769 * maintaining any required invariants and checking allocations.
771 * The source is marked as being in the new filter mode at t1.
773 * Return the pointer to the new node, otherwise return NULL.
775 static struct in_msource *
776 imf_graft(struct in_mfilter *imf, const uint8_t st1,
777 const struct sockaddr_in *psin)
779 struct ip_msource *nims;
780 struct in_msource *lims;
782 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
786 lims = (struct in_msource *)nims;
787 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
788 lims->imsl_st[0] = MCAST_UNDEFINED;
789 lims->imsl_st[1] = st1;
790 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
797 * Prune a source entry from an existing socket-layer filter set,
798 * maintaining any required invariants and checking allocations.
800 * The source is marked as being left at t1, it is not freed.
802 * Return 0 if no error occurred, otherwise return an errno value.
805 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
807 struct ip_msource find;
808 struct ip_msource *ims;
809 struct in_msource *lims;
811 /* key is host byte order */
812 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
813 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
816 lims = (struct in_msource *)ims;
817 lims->imsl_st[1] = MCAST_UNDEFINED;
822 * Revert socket-layer filter set deltas at t1 to t0 state.
825 imf_rollback(struct in_mfilter *imf)
827 struct ip_msource *ims, *tims;
828 struct in_msource *lims;
830 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
831 lims = (struct in_msource *)ims;
832 if (lims->imsl_st[0] == lims->imsl_st[1]) {
833 /* no change at t1 */
835 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
836 /* revert change to existing source at t1 */
837 lims->imsl_st[1] = lims->imsl_st[0];
839 /* revert source added t1 */
840 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
841 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
842 free(ims, M_INMFILTER);
846 imf->imf_st[1] = imf->imf_st[0];
850 * Mark socket-layer filter set as INCLUDE {} at t1.
853 imf_leave(struct in_mfilter *imf)
855 struct ip_msource *ims;
856 struct in_msource *lims;
858 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
859 lims = (struct in_msource *)ims;
860 lims->imsl_st[1] = MCAST_UNDEFINED;
862 imf->imf_st[1] = MCAST_INCLUDE;
866 * Mark socket-layer filter set deltas as committed.
869 imf_commit(struct in_mfilter *imf)
871 struct ip_msource *ims;
872 struct in_msource *lims;
874 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
875 lims = (struct in_msource *)ims;
876 lims->imsl_st[0] = lims->imsl_st[1];
878 imf->imf_st[0] = imf->imf_st[1];
882 * Reap unreferenced sources from socket-layer filter set.
885 imf_reap(struct in_mfilter *imf)
887 struct ip_msource *ims, *tims;
888 struct in_msource *lims;
890 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
891 lims = (struct in_msource *)ims;
892 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
893 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
894 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
895 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
896 free(ims, M_INMFILTER);
903 * Purge socket-layer filter set.
906 imf_purge(struct in_mfilter *imf)
908 struct ip_msource *ims, *tims;
910 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
911 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
912 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
913 free(ims, M_INMFILTER);
916 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
917 KASSERT(RB_EMPTY(&imf->imf_sources),
918 ("%s: imf_sources not empty", __func__));
922 * Look up a source filter entry for a multicast group.
924 * inm is the group descriptor to work with.
925 * haddr is the host-byte-order IPv4 address to look up.
926 * noalloc may be non-zero to suppress allocation of sources.
927 * *pims will be set to the address of the retrieved or allocated source.
929 * SMPng: NOTE: may be called with locks held.
930 * Return 0 if successful, otherwise return a non-zero error code.
933 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
934 const int noalloc, struct ip_msource **pims)
936 struct ip_msource find;
937 struct ip_msource *ims, *nims;
939 find.ims_haddr = haddr;
940 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
941 if (ims == NULL && !noalloc) {
942 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
944 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
948 nims->ims_haddr = haddr;
949 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
953 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
963 * Merge socket-layer source into IGMP-layer source.
964 * If rollback is non-zero, perform the inverse of the merge.
967 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
970 int n = rollback ? -1 : 1;
972 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
973 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
974 __func__, n, ims->ims_haddr);
975 ims->ims_st[1].ex -= n;
976 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
977 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
978 __func__, n, ims->ims_haddr);
979 ims->ims_st[1].in -= n;
982 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
983 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
984 __func__, n, ims->ims_haddr);
985 ims->ims_st[1].ex += n;
986 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
987 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
988 __func__, n, ims->ims_haddr);
989 ims->ims_st[1].in += n;
994 * Atomically update the global in_multi state, when a membership's
995 * filter list is being updated in any way.
997 * imf is the per-inpcb-membership group filter pointer.
998 * A fake imf may be passed for in-kernel consumers.
1000 * XXX This is a candidate for a set-symmetric-difference style loop
1001 * which would eliminate the repeated lookup from root of ims nodes,
1002 * as they share the same key space.
1004 * If any error occurred this function will back out of refcounts
1005 * and return a non-zero value.
1008 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1010 struct ip_msource *ims, *nims;
1011 struct in_msource *lims;
1012 int schanged, error;
1018 IN_MULTI_LIST_LOCK_ASSERT();
1021 * Update the source filters first, as this may fail.
1022 * Maintain count of in-mode filters at t0, t1. These are
1023 * used to work out if we transition into ASM mode or not.
1024 * Maintain a count of source filters whose state was
1025 * actually modified by this operation.
1027 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1028 lims = (struct in_msource *)ims;
1029 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1030 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1031 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1032 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1036 ims_merge(nims, lims, 0);
1039 struct ip_msource *bims;
1041 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1042 lims = (struct in_msource *)ims;
1043 if (lims->imsl_st[0] == lims->imsl_st[1])
1045 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1048 ims_merge(bims, lims, 1);
1053 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1054 __func__, nsrc0, nsrc1);
1056 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1057 if (imf->imf_st[0] == imf->imf_st[1] &&
1058 imf->imf_st[1] == MCAST_INCLUDE) {
1060 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1061 --inm->inm_st[1].iss_in;
1065 /* Handle filter mode transition on socket. */
1066 if (imf->imf_st[0] != imf->imf_st[1]) {
1067 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1068 __func__, imf->imf_st[0], imf->imf_st[1]);
1070 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1071 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1072 --inm->inm_st[1].iss_ex;
1073 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1074 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1075 --inm->inm_st[1].iss_in;
1078 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1079 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1080 inm->inm_st[1].iss_ex++;
1081 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1082 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1083 inm->inm_st[1].iss_in++;
1088 * Track inm filter state in terms of listener counts.
1089 * If there are any exclusive listeners, stack-wide
1090 * membership is exclusive.
1091 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1092 * If no listeners remain, state is undefined at t1,
1093 * and the IGMP lifecycle for this group should finish.
1095 if (inm->inm_st[1].iss_ex > 0) {
1096 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1097 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1098 } else if (inm->inm_st[1].iss_in > 0) {
1099 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1100 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1102 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1103 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1106 /* Decrement ASM listener count on transition out of ASM mode. */
1107 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1108 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1109 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1110 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1111 --inm->inm_st[1].iss_asm;
1115 /* Increment ASM listener count on transition to ASM mode. */
1116 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1117 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1118 inm->inm_st[1].iss_asm++;
1121 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1126 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1133 * Mark an in_multi's filter set deltas as committed.
1134 * Called by IGMP after a state change has been enqueued.
1137 inm_commit(struct in_multi *inm)
1139 struct ip_msource *ims;
1141 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1142 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1145 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1146 ims->ims_st[0] = ims->ims_st[1];
1148 inm->inm_st[0] = inm->inm_st[1];
1152 * Reap unreferenced nodes from an in_multi's filter set.
1155 inm_reap(struct in_multi *inm)
1157 struct ip_msource *ims, *tims;
1159 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1160 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1161 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1164 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1165 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1166 free(ims, M_IPMSOURCE);
1172 * Purge all source nodes from an in_multi's filter set.
1175 inm_purge(struct in_multi *inm)
1177 struct ip_msource *ims, *tims;
1179 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1180 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1181 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1182 free(ims, M_IPMSOURCE);
1188 * Join a multicast group; unlocked entry point.
1190 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1191 * is not held. Fortunately, ifp is unlikely to have been detached
1192 * at this point, so we assume it's OK to recurse.
1195 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1196 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1201 error = in_joingroup_locked(ifp, gina, imf, pinm);
1208 * Join a multicast group; real entry point.
1210 * Only preserves atomicity at inm level.
1211 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1213 * If the IGMP downcall fails, the group is not joined, and an error
1217 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1218 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1220 struct in_mfilter timf;
1221 struct in_multi *inm;
1224 IN_MULTI_LOCK_ASSERT();
1225 IN_MULTI_LIST_UNLOCK_ASSERT();
1227 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1228 ntohl(gina->s_addr), ifp, ifp->if_xname);
1234 * If no imf was specified (i.e. kernel consumer),
1235 * fake one up and assume it is an ASM join.
1238 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1242 error = in_getmulti(ifp, gina, &inm);
1244 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1247 IN_MULTI_LIST_LOCK();
1248 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1249 error = inm_merge(inm, imf);
1251 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1252 goto out_inm_release;
1255 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1256 error = igmp_change_state(inm);
1258 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1259 goto out_inm_release;
1264 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1266 inm_release_deferred(inm);
1267 IF_ADDR_WUNLOCK(ifp);
1271 IN_MULTI_LIST_UNLOCK();
1277 * Leave a multicast group; unlocked entry point.
1280 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1285 error = in_leavegroup_locked(inm, imf);
1292 * Leave a multicast group; real entry point.
1293 * All source filters will be expunged.
1295 * Only preserves atomicity at inm level.
1297 * Holding the write lock for the INP which contains imf
1298 * is highly advisable. We can't assert for it as imf does not
1299 * contain a back-pointer to the owning inp.
1301 * Note: This is not the same as inm_release(*) as this function also
1302 * makes a state change downcall into IGMP.
1305 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1307 struct in_mfilter timf;
1310 IN_MULTI_LOCK_ASSERT();
1311 IN_MULTI_LIST_UNLOCK_ASSERT();
1315 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1316 inm, ntohl(inm->inm_addr.s_addr),
1317 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1321 * If no imf was specified (i.e. kernel consumer),
1322 * fake one up and assume it is an ASM join.
1325 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1330 * Begin state merge transaction at IGMP layer.
1332 * As this particular invocation should not cause any memory
1333 * to be allocated, and there is no opportunity to roll back
1334 * the transaction, it MUST NOT fail.
1336 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1337 IN_MULTI_LIST_LOCK();
1338 error = inm_merge(inm, imf);
1339 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1341 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1342 CURVNET_SET(inm->inm_ifp->if_vnet);
1343 error = igmp_change_state(inm);
1344 IF_ADDR_WLOCK(inm->inm_ifp);
1345 inm_release_deferred(inm);
1346 IF_ADDR_WUNLOCK(inm->inm_ifp);
1347 IN_MULTI_LIST_UNLOCK();
1350 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1352 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1357 /*#ifndef BURN_BRIDGES*/
1360 * Block or unblock an ASM multicast source on an inpcb.
1361 * This implements the delta-based API described in RFC 3678.
1363 * The delta-based API applies only to exclusive-mode memberships.
1364 * An IGMP downcall will be performed.
1366 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1368 * Return 0 if successful, otherwise return an appropriate error code.
1371 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1373 struct epoch_tracker et;
1374 struct group_source_req gsr;
1375 sockunion_t *gsa, *ssa;
1377 struct in_mfilter *imf;
1378 struct ip_moptions *imo;
1379 struct in_msource *ims;
1380 struct in_multi *inm;
1388 memset(&gsr, 0, sizeof(struct group_source_req));
1389 gsa = (sockunion_t *)&gsr.gsr_group;
1390 ssa = (sockunion_t *)&gsr.gsr_source;
1392 switch (sopt->sopt_name) {
1393 case IP_BLOCK_SOURCE:
1394 case IP_UNBLOCK_SOURCE: {
1395 struct ip_mreq_source mreqs;
1397 error = sooptcopyin(sopt, &mreqs,
1398 sizeof(struct ip_mreq_source),
1399 sizeof(struct ip_mreq_source));
1403 gsa->sin.sin_family = AF_INET;
1404 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1405 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1407 ssa->sin.sin_family = AF_INET;
1408 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1409 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1411 if (!in_nullhost(mreqs.imr_interface)) {
1412 NET_EPOCH_ENTER(et);
1413 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1417 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1420 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1421 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1425 case MCAST_BLOCK_SOURCE:
1426 case MCAST_UNBLOCK_SOURCE:
1427 error = sooptcopyin(sopt, &gsr,
1428 sizeof(struct group_source_req),
1429 sizeof(struct group_source_req));
1433 if (gsa->sin.sin_family != AF_INET ||
1434 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1437 if (ssa->sin.sin_family != AF_INET ||
1438 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1441 NET_EPOCH_ENTER(et);
1442 ifp = ifnet_byindex(gsr.gsr_interface);
1445 return (EADDRNOTAVAIL);
1447 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1452 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1453 __func__, sopt->sopt_name);
1454 return (EOPNOTSUPP);
1458 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1464 * Check if we are actually a member of this group.
1466 imo = inp_findmoptions(inp);
1467 imf = imo_match_group(imo, ifp, &gsa->sa);
1469 error = EADDRNOTAVAIL;
1470 goto out_inp_locked;
1475 * Attempting to use the delta-based API on an
1476 * non exclusive-mode membership is an error.
1478 fmode = imf->imf_st[0];
1479 if (fmode != MCAST_EXCLUDE) {
1481 goto out_inp_locked;
1485 * Deal with error cases up-front:
1486 * Asked to block, but already blocked; or
1487 * Asked to unblock, but nothing to unblock.
1488 * If adding a new block entry, allocate it.
1490 ims = imo_match_source(imf, &ssa->sa);
1491 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1492 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1493 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1494 error = EADDRNOTAVAIL;
1495 goto out_inp_locked;
1498 INP_WLOCK_ASSERT(inp);
1501 * Begin state merge transaction at socket layer.
1504 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1505 ims = imf_graft(imf, fmode, &ssa->sin);
1509 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1510 error = imf_prune(imf, &ssa->sin);
1514 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1515 goto out_imf_rollback;
1519 * Begin state merge transaction at IGMP layer.
1521 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1522 IN_MULTI_LIST_LOCK();
1523 error = inm_merge(inm, imf);
1525 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1526 IN_MULTI_LIST_UNLOCK();
1527 goto out_imf_rollback;
1530 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1531 error = igmp_change_state(inm);
1532 IN_MULTI_LIST_UNLOCK();
1534 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1551 * Given an inpcb, return its multicast options structure pointer. Accepts
1552 * an unlocked inpcb pointer, but will return it locked. May sleep.
1554 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1555 * SMPng: NOTE: Returns with the INP write lock held.
1557 static struct ip_moptions *
1558 inp_findmoptions(struct inpcb *inp)
1560 struct ip_moptions *imo;
1563 if (inp->inp_moptions != NULL)
1564 return (inp->inp_moptions);
1568 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1570 imo->imo_multicast_ifp = NULL;
1571 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1572 imo->imo_multicast_vif = -1;
1573 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1574 imo->imo_multicast_loop = in_mcast_loop;
1575 STAILQ_INIT(&imo->imo_head);
1578 if (inp->inp_moptions != NULL) {
1579 free(imo, M_IPMOPTS);
1580 return (inp->inp_moptions);
1582 inp->inp_moptions = imo;
1587 inp_freemoptions(struct ip_moptions *imo)
1589 struct in_mfilter *imf;
1590 struct in_multi *inm;
1596 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1597 ip_mfilter_remove(&imo->imo_head, imf);
1600 if ((inm = imf->imf_inm) != NULL) {
1601 if ((ifp = inm->inm_ifp) != NULL) {
1602 CURVNET_SET(ifp->if_vnet);
1603 (void)in_leavegroup(inm, imf);
1606 (void)in_leavegroup(inm, imf);
1609 ip_mfilter_free(imf);
1611 free(imo, M_IPMOPTS);
1615 * Atomically get source filters on a socket for an IPv4 multicast group.
1616 * Called with INP lock held; returns with lock released.
1619 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1621 struct epoch_tracker et;
1622 struct __msfilterreq msfr;
1625 struct ip_moptions *imo;
1626 struct in_mfilter *imf;
1627 struct ip_msource *ims;
1628 struct in_msource *lims;
1629 struct sockaddr_in *psin;
1630 struct sockaddr_storage *ptss;
1631 struct sockaddr_storage *tss;
1633 size_t nsrcs, ncsrcs;
1635 INP_WLOCK_ASSERT(inp);
1637 imo = inp->inp_moptions;
1638 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1642 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1643 sizeof(struct __msfilterreq));
1647 NET_EPOCH_ENTER(et);
1648 ifp = ifnet_byindex(msfr.msfr_ifindex);
1649 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifnet pointer left */
1656 * Lookup group on the socket.
1658 gsa = (sockunion_t *)&msfr.msfr_group;
1659 imf = imo_match_group(imo, ifp, &gsa->sa);
1662 return (EADDRNOTAVAIL);
1666 * Ignore memberships which are in limbo.
1668 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1672 msfr.msfr_fmode = imf->imf_st[1];
1675 * If the user specified a buffer, copy out the source filter
1676 * entries to userland gracefully.
1677 * We only copy out the number of entries which userland
1678 * has asked for, but we always tell userland how big the
1679 * buffer really needs to be.
1681 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1682 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1684 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1685 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1686 M_TEMP, M_NOWAIT | M_ZERO);
1694 * Count number of sources in-mode at t0.
1695 * If buffer space exists and remains, copy out source entries.
1697 nsrcs = msfr.msfr_nsrcs;
1700 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1701 lims = (struct in_msource *)ims;
1702 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1703 lims->imsl_st[0] != imf->imf_st[0])
1706 if (tss != NULL && nsrcs > 0) {
1707 psin = (struct sockaddr_in *)ptss;
1708 psin->sin_family = AF_INET;
1709 psin->sin_len = sizeof(struct sockaddr_in);
1710 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1720 error = copyout(tss, msfr.msfr_srcs,
1721 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1727 msfr.msfr_nsrcs = ncsrcs;
1728 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1734 * Return the IP multicast options in response to user getsockopt().
1737 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1739 struct ip_mreqn mreqn;
1740 struct ip_moptions *imo;
1742 struct in_ifaddr *ia;
1747 imo = inp->inp_moptions;
1748 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM reject it. */
1749 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1750 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM) {
1752 return (EOPNOTSUPP);
1756 switch (sopt->sopt_name) {
1757 case IP_MULTICAST_VIF:
1759 optval = imo->imo_multicast_vif;
1763 error = sooptcopyout(sopt, &optval, sizeof(int));
1766 case IP_MULTICAST_IF:
1767 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1769 ifp = imo->imo_multicast_ifp;
1770 if (!in_nullhost(imo->imo_multicast_addr)) {
1771 mreqn.imr_address = imo->imo_multicast_addr;
1772 } else if (ifp != NULL) {
1773 struct epoch_tracker et;
1775 mreqn.imr_ifindex = ifp->if_index;
1776 NET_EPOCH_ENTER(et);
1780 IA_SIN(ia)->sin_addr;
1785 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1786 error = sooptcopyout(sopt, &mreqn,
1787 sizeof(struct ip_mreqn));
1789 error = sooptcopyout(sopt, &mreqn.imr_address,
1790 sizeof(struct in_addr));
1794 case IP_MULTICAST_TTL:
1796 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1798 optval = coptval = imo->imo_multicast_ttl;
1800 if (sopt->sopt_valsize == sizeof(u_char))
1801 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1803 error = sooptcopyout(sopt, &optval, sizeof(int));
1806 case IP_MULTICAST_LOOP:
1808 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1810 optval = coptval = imo->imo_multicast_loop;
1812 if (sopt->sopt_valsize == sizeof(u_char))
1813 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1815 error = sooptcopyout(sopt, &optval, sizeof(int));
1820 error = EADDRNOTAVAIL;
1823 error = inp_get_source_filters(inp, sopt);
1829 error = ENOPROTOOPT;
1833 INP_UNLOCK_ASSERT(inp);
1839 * Look up the ifnet to use for a multicast group membership,
1840 * given the IPv4 address of an interface, and the IPv4 group address.
1842 * This routine exists to support legacy multicast applications
1843 * which do not understand that multicast memberships are scoped to
1844 * specific physical links in the networking stack, or which need
1845 * to join link-scope groups before IPv4 addresses are configured.
1847 * Use this socket's current FIB number for any required FIB lookup.
1848 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1849 * and use its ifp; usually, this points to the default next-hop.
1851 * If the FIB lookup fails, attempt to use the first non-loopback
1852 * interface with multicast capability in the system as a
1853 * last resort. The legacy IPv4 ASM API requires that we do
1854 * this in order to allow groups to be joined when the routing
1855 * table has not yet been populated during boot.
1857 * Returns NULL if no ifp could be found, otherwise return referenced ifp.
1859 * FUTURE: Implement IPv4 source-address selection.
1861 static struct ifnet *
1862 inp_lookup_mcast_ifp(const struct inpcb *inp,
1863 const struct sockaddr_in *gsin, const struct in_addr ina)
1866 struct nhop_object *nh;
1869 KASSERT(inp != NULL, ("%s: inp must not be NULL", __func__));
1870 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1871 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1872 ("%s: not multicast", __func__));
1875 if (!in_nullhost(ina)) {
1876 INADDR_TO_IFP(ina, ifp);
1880 nh = fib4_lookup(inp->inp_inc.inc_fibnum, gsin->sin_addr, 0, NHR_NONE, 0);
1885 struct in_ifaddr *ia;
1889 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1891 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1892 (mifp->if_flags & IFF_MULTICAST)) {
1905 * Join an IPv4 multicast group, possibly with a source.
1908 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1910 struct group_source_req gsr;
1911 sockunion_t *gsa, *ssa;
1913 struct in_mfilter *imf;
1914 struct ip_moptions *imo;
1915 struct in_multi *inm;
1916 struct in_msource *lims;
1917 struct epoch_tracker et;
1924 memset(&gsr, 0, sizeof(struct group_source_req));
1925 gsa = (sockunion_t *)&gsr.gsr_group;
1926 gsa->ss.ss_family = AF_UNSPEC;
1927 ssa = (sockunion_t *)&gsr.gsr_source;
1928 ssa->ss.ss_family = AF_UNSPEC;
1930 switch (sopt->sopt_name) {
1931 case IP_ADD_MEMBERSHIP: {
1932 struct ip_mreqn mreqn;
1934 if (sopt->sopt_valsize == sizeof(struct ip_mreqn))
1935 error = sooptcopyin(sopt, &mreqn,
1936 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn));
1938 error = sooptcopyin(sopt, &mreqn,
1939 sizeof(struct ip_mreq), sizeof(struct ip_mreq));
1943 gsa->sin.sin_family = AF_INET;
1944 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1945 gsa->sin.sin_addr = mreqn.imr_multiaddr;
1946 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1949 NET_EPOCH_ENTER(et);
1950 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) &&
1951 mreqn.imr_ifindex != 0)
1952 ifp = ifnet_byindex_ref(mreqn.imr_ifindex);
1954 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1959 case IP_ADD_SOURCE_MEMBERSHIP: {
1960 struct ip_mreq_source mreqs;
1962 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source),
1963 sizeof(struct ip_mreq_source));
1967 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET;
1968 gsa->sin.sin_len = ssa->sin.sin_len =
1969 sizeof(struct sockaddr_in);
1971 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1972 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1975 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1977 NET_EPOCH_ENTER(et);
1978 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1979 mreqs.imr_interface);
1981 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1982 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1986 case MCAST_JOIN_GROUP:
1987 case MCAST_JOIN_SOURCE_GROUP:
1988 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1989 error = sooptcopyin(sopt, &gsr,
1990 sizeof(struct group_req),
1991 sizeof(struct group_req));
1992 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1993 error = sooptcopyin(sopt, &gsr,
1994 sizeof(struct group_source_req),
1995 sizeof(struct group_source_req));
2000 if (gsa->sin.sin_family != AF_INET ||
2001 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2005 * Overwrite the port field if present, as the sockaddr
2006 * being copied in may be matched with a binary comparison.
2008 gsa->sin.sin_port = 0;
2009 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2010 if (ssa->sin.sin_family != AF_INET ||
2011 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2013 ssa->sin.sin_port = 0;
2016 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2019 NET_EPOCH_ENTER(et);
2020 ifp = ifnet_byindex_ref(gsr.gsr_interface);
2023 return (EADDRNOTAVAIL);
2027 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2028 __func__, sopt->sopt_name);
2029 return (EOPNOTSUPP);
2033 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2036 return (EADDRNOTAVAIL);
2042 * Find the membership in the membership list.
2044 imo = inp_findmoptions(inp);
2045 imf = imo_match_group(imo, ifp, &gsa->sa);
2050 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) {
2052 goto out_inp_locked;
2058 if (ssa->ss.ss_family != AF_UNSPEC) {
2060 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2061 * is an error. On an existing inclusive membership,
2062 * it just adds the source to the filter list.
2064 if (imf->imf_st[1] != MCAST_INCLUDE) {
2066 goto out_inp_locked;
2069 * Throw out duplicates.
2071 * XXX FIXME: This makes a naive assumption that
2072 * even if entries exist for *ssa in this imf,
2073 * they will be rejected as dupes, even if they
2074 * are not valid in the current mode (in-mode).
2076 * in_msource is transactioned just as for anything
2077 * else in SSM -- but note naive use of inm_graft()
2078 * below for allocating new filter entries.
2080 * This is only an issue if someone mixes the
2081 * full-state SSM API with the delta-based API,
2082 * which is discouraged in the relevant RFCs.
2084 lims = imo_match_source(imf, &ssa->sa);
2085 if (lims != NULL /*&&
2086 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2087 error = EADDRNOTAVAIL;
2088 goto out_inp_locked;
2092 * MCAST_JOIN_GROUP on an existing exclusive
2093 * membership is an error; return EADDRINUSE
2094 * to preserve 4.4BSD API idempotence, and
2095 * avoid tedious detour to code below.
2096 * NOTE: This is bending RFC 3678 a bit.
2098 * On an existing inclusive membership, this is also
2099 * an error; if you want to change filter mode,
2100 * you must use the userland API setsourcefilter().
2101 * XXX We don't reject this for imf in UNDEFINED
2102 * state at t1, because allocation of a filter
2103 * is atomic with allocation of a membership.
2106 if (imf->imf_st[1] == MCAST_EXCLUDE)
2108 goto out_inp_locked;
2113 * Begin state merge transaction at socket layer.
2115 INP_WLOCK_ASSERT(inp);
2118 * Graft new source into filter list for this inpcb's
2119 * membership of the group. The in_multi may not have
2120 * been allocated yet if this is a new membership, however,
2121 * the in_mfilter slot will be allocated and must be initialized.
2123 * Note: Grafting of exclusive mode filters doesn't happen
2125 * XXX: Should check for non-NULL lims (node exists but may
2126 * not be in-mode) for interop with full-state API.
2128 if (ssa->ss.ss_family != AF_UNSPEC) {
2129 /* Membership starts in IN mode */
2131 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2132 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2135 goto out_inp_locked;
2138 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2140 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2142 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2145 goto out_inp_locked;
2148 /* No address specified; Membership starts in EX mode */
2150 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2151 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2154 goto out_inp_locked;
2160 * Begin state merge transaction at IGMP layer.
2166 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2170 if (in_pcbrele_wlocked(inp)) {
2172 goto out_inp_unlocked;
2175 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2177 goto out_inp_locked;
2180 * NOTE: Refcount from in_joingroup_locked()
2181 * is protecting membership.
2183 ip_mfilter_insert(&imo->imo_head, imf);
2185 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2186 IN_MULTI_LIST_LOCK();
2187 error = inm_merge(inm, imf);
2189 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2191 IN_MULTI_LIST_UNLOCK();
2194 goto out_inp_locked;
2196 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2197 error = igmp_change_state(inm);
2198 IN_MULTI_LIST_UNLOCK();
2200 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2204 goto out_inp_locked;
2216 if (is_new && imf) {
2217 if (imf->imf_inm != NULL) {
2218 IN_MULTI_LIST_LOCK();
2220 inm_release_deferred(imf->imf_inm);
2221 IF_ADDR_WUNLOCK(ifp);
2222 IN_MULTI_LIST_UNLOCK();
2224 ip_mfilter_free(imf);
2231 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2234 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2236 struct epoch_tracker et;
2237 struct group_source_req gsr;
2238 struct ip_mreq_source mreqs;
2239 sockunion_t *gsa, *ssa;
2241 struct in_mfilter *imf;
2242 struct ip_moptions *imo;
2243 struct in_msource *ims;
2244 struct in_multi *inm;
2252 memset(&gsr, 0, sizeof(struct group_source_req));
2253 gsa = (sockunion_t *)&gsr.gsr_group;
2254 gsa->ss.ss_family = AF_UNSPEC;
2255 ssa = (sockunion_t *)&gsr.gsr_source;
2256 ssa->ss.ss_family = AF_UNSPEC;
2258 switch (sopt->sopt_name) {
2259 case IP_DROP_MEMBERSHIP:
2260 case IP_DROP_SOURCE_MEMBERSHIP:
2261 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2262 error = sooptcopyin(sopt, &mreqs,
2263 sizeof(struct ip_mreq),
2264 sizeof(struct ip_mreq));
2266 * Swap interface and sourceaddr arguments,
2267 * as ip_mreq and ip_mreq_source are laid
2270 mreqs.imr_interface = mreqs.imr_sourceaddr;
2271 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2272 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2273 error = sooptcopyin(sopt, &mreqs,
2274 sizeof(struct ip_mreq_source),
2275 sizeof(struct ip_mreq_source));
2280 gsa->sin.sin_family = AF_INET;
2281 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2282 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2284 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2285 ssa->sin.sin_family = AF_INET;
2286 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2287 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2291 * Attempt to look up hinted ifp from interface address.
2292 * Fallthrough with null ifp iff lookup fails, to
2293 * preserve 4.4BSD mcast API idempotence.
2294 * XXX NOTE WELL: The RFC 3678 API is preferred because
2295 * using an IPv4 address as a key is racy.
2297 if (!in_nullhost(mreqs.imr_interface)) {
2298 NET_EPOCH_ENTER(et);
2299 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2303 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2304 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2308 case MCAST_LEAVE_GROUP:
2309 case MCAST_LEAVE_SOURCE_GROUP:
2310 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2311 error = sooptcopyin(sopt, &gsr,
2312 sizeof(struct group_req),
2313 sizeof(struct group_req));
2314 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2315 error = sooptcopyin(sopt, &gsr,
2316 sizeof(struct group_source_req),
2317 sizeof(struct group_source_req));
2322 if (gsa->sin.sin_family != AF_INET ||
2323 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2326 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2327 if (ssa->sin.sin_family != AF_INET ||
2328 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2332 NET_EPOCH_ENTER(et);
2333 ifp = ifnet_byindex(gsr.gsr_interface);
2334 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2336 return (EADDRNOTAVAIL);
2340 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2341 __func__, sopt->sopt_name);
2342 return (EOPNOTSUPP);
2346 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2352 * Find the membership in the membership list.
2354 imo = inp_findmoptions(inp);
2355 imf = imo_match_group(imo, ifp, &gsa->sa);
2357 error = EADDRNOTAVAIL;
2358 goto out_inp_locked;
2362 if (ssa->ss.ss_family != AF_UNSPEC)
2366 * Begin state merge transaction at socket layer.
2368 INP_WLOCK_ASSERT(inp);
2371 * If we were instructed only to leave a given source, do so.
2372 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2375 ip_mfilter_remove(&imo->imo_head, imf);
2379 * Give up the multicast address record to which
2380 * the membership points.
2382 (void) in_leavegroup_locked(imf->imf_inm, imf);
2384 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2385 error = EADDRNOTAVAIL;
2386 goto out_inp_locked;
2388 ims = imo_match_source(imf, &ssa->sa);
2390 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2391 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2392 error = EADDRNOTAVAIL;
2393 goto out_inp_locked;
2395 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2396 error = imf_prune(imf, &ssa->sin);
2398 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2400 goto out_inp_locked;
2405 * Begin state merge transaction at IGMP layer.
2408 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2409 IN_MULTI_LIST_LOCK();
2410 error = inm_merge(inm, imf);
2412 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2414 IN_MULTI_LIST_UNLOCK();
2417 goto out_inp_locked;
2420 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2421 error = igmp_change_state(inm);
2422 IN_MULTI_LIST_UNLOCK();
2424 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2428 goto out_inp_locked;
2437 if (is_final && imf)
2438 ip_mfilter_free(imf);
2445 * Select the interface for transmitting IPv4 multicast datagrams.
2447 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2448 * may be passed to this socket option. An address of INADDR_ANY or an
2449 * interface index of 0 is used to remove a previous selection.
2450 * When no interface is selected, one is chosen for every send.
2453 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2455 struct in_addr addr;
2456 struct ip_mreqn mreqn;
2458 struct ip_moptions *imo;
2461 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2463 * An interface index was specified using the
2464 * Linux-derived ip_mreqn structure.
2466 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2467 sizeof(struct ip_mreqn));
2471 if (mreqn.imr_ifindex < 0)
2474 if (mreqn.imr_ifindex == 0) {
2477 struct epoch_tracker et;
2479 NET_EPOCH_ENTER(et);
2480 ifp = ifnet_byindex(mreqn.imr_ifindex);
2481 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2483 return (EADDRNOTAVAIL);
2487 * An interface was specified by IPv4 address.
2488 * This is the traditional BSD usage.
2490 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2491 sizeof(struct in_addr));
2494 if (in_nullhost(addr)) {
2497 struct epoch_tracker et;
2499 NET_EPOCH_ENTER(et);
2500 INADDR_TO_IFP(addr, ifp);
2504 return (EADDRNOTAVAIL);
2506 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2507 ntohl(addr.s_addr));
2510 /* Reject interfaces which do not support multicast. */
2511 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2512 return (EOPNOTSUPP);
2514 imo = inp_findmoptions(inp);
2515 imo->imo_multicast_ifp = ifp;
2516 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2523 * Atomically set source filters on a socket for an IPv4 multicast group.
2525 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2528 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2530 struct epoch_tracker et;
2531 struct __msfilterreq msfr;
2534 struct in_mfilter *imf;
2535 struct ip_moptions *imo;
2536 struct in_multi *inm;
2539 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2540 sizeof(struct __msfilterreq));
2544 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2547 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2548 msfr.msfr_fmode != MCAST_INCLUDE))
2551 if (msfr.msfr_group.ss_family != AF_INET ||
2552 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2555 gsa = (sockunion_t *)&msfr.msfr_group;
2556 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2559 gsa->sin.sin_port = 0; /* ignore port */
2561 NET_EPOCH_ENTER(et);
2562 ifp = ifnet_byindex(msfr.msfr_ifindex);
2563 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2565 return (EADDRNOTAVAIL);
2570 * Take the INP write lock.
2571 * Check if this socket is a member of this group.
2573 imo = inp_findmoptions(inp);
2574 imf = imo_match_group(imo, ifp, &gsa->sa);
2576 error = EADDRNOTAVAIL;
2577 goto out_inp_locked;
2582 * Begin state merge transaction at socket layer.
2584 INP_WLOCK_ASSERT(inp);
2586 imf->imf_st[1] = msfr.msfr_fmode;
2589 * Apply any new source filters, if present.
2590 * Make a copy of the user-space source vector so
2591 * that we may copy them with a single copyin. This
2592 * allows us to deal with page faults up-front.
2594 if (msfr.msfr_nsrcs > 0) {
2595 struct in_msource *lims;
2596 struct sockaddr_in *psin;
2597 struct sockaddr_storage *kss, *pkss;
2602 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2603 __func__, (unsigned long)msfr.msfr_nsrcs);
2604 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2606 error = copyin(msfr.msfr_srcs, kss,
2607 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2616 * Mark all source filters as UNDEFINED at t1.
2617 * Restore new group filter mode, as imf_leave()
2618 * will set it to INCLUDE.
2621 imf->imf_st[1] = msfr.msfr_fmode;
2624 * Update socket layer filters at t1, lazy-allocating
2625 * new entries. This saves a bunch of memory at the
2626 * cost of one RB_FIND() per source entry; duplicate
2627 * entries in the msfr_nsrcs vector are ignored.
2628 * If we encounter an error, rollback transaction.
2630 * XXX This too could be replaced with a set-symmetric
2631 * difference like loop to avoid walking from root
2632 * every time, as the key space is common.
2634 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2635 psin = (struct sockaddr_in *)pkss;
2636 if (psin->sin_family != AF_INET) {
2637 error = EAFNOSUPPORT;
2640 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2644 error = imf_get_source(imf, psin, &lims);
2647 lims->imsl_st[1] = imf->imf_st[1];
2653 goto out_imf_rollback;
2655 INP_WLOCK_ASSERT(inp);
2658 * Begin state merge transaction at IGMP layer.
2660 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2661 IN_MULTI_LIST_LOCK();
2662 error = inm_merge(inm, imf);
2664 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2665 IN_MULTI_LIST_UNLOCK();
2666 goto out_imf_rollback;
2669 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2670 error = igmp_change_state(inm);
2671 IN_MULTI_LIST_UNLOCK();
2673 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2690 * Set the IP multicast options in response to user setsockopt().
2692 * Many of the socket options handled in this function duplicate the
2693 * functionality of socket options in the regular unicast API. However,
2694 * it is not possible to merge the duplicate code, because the idempotence
2695 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2696 * the effects of these options must be treated as separate and distinct.
2698 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2699 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2700 * is refactored to no longer use vifs.
2703 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2705 struct ip_moptions *imo;
2710 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
2711 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2712 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)
2713 return (EOPNOTSUPP);
2715 switch (sopt->sopt_name) {
2716 case IP_MULTICAST_VIF: {
2719 * Select a multicast VIF for transmission.
2720 * Only useful if multicast forwarding is active.
2722 if (legal_vif_num == NULL) {
2726 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2729 if (!legal_vif_num(vifi) && (vifi != -1)) {
2733 imo = inp_findmoptions(inp);
2734 imo->imo_multicast_vif = vifi;
2739 case IP_MULTICAST_IF:
2740 error = inp_set_multicast_if(inp, sopt);
2743 case IP_MULTICAST_TTL: {
2747 * Set the IP time-to-live for outgoing multicast packets.
2748 * The original multicast API required a char argument,
2749 * which is inconsistent with the rest of the socket API.
2750 * We allow either a char or an int.
2752 if (sopt->sopt_valsize == sizeof(u_char)) {
2753 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2760 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2770 imo = inp_findmoptions(inp);
2771 imo->imo_multicast_ttl = ttl;
2776 case IP_MULTICAST_LOOP: {
2780 * Set the loopback flag for outgoing multicast packets.
2781 * Must be zero or one. The original multicast API required a
2782 * char argument, which is inconsistent with the rest
2783 * of the socket API. We allow either a char or an int.
2785 if (sopt->sopt_valsize == sizeof(u_char)) {
2786 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2793 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2797 loop = (u_char)iloop;
2799 imo = inp_findmoptions(inp);
2800 imo->imo_multicast_loop = !!loop;
2805 case IP_ADD_MEMBERSHIP:
2806 case IP_ADD_SOURCE_MEMBERSHIP:
2807 case MCAST_JOIN_GROUP:
2808 case MCAST_JOIN_SOURCE_GROUP:
2809 error = inp_join_group(inp, sopt);
2812 case IP_DROP_MEMBERSHIP:
2813 case IP_DROP_SOURCE_MEMBERSHIP:
2814 case MCAST_LEAVE_GROUP:
2815 case MCAST_LEAVE_SOURCE_GROUP:
2816 error = inp_leave_group(inp, sopt);
2819 case IP_BLOCK_SOURCE:
2820 case IP_UNBLOCK_SOURCE:
2821 case MCAST_BLOCK_SOURCE:
2822 case MCAST_UNBLOCK_SOURCE:
2823 error = inp_block_unblock_source(inp, sopt);
2827 error = inp_set_source_filters(inp, sopt);
2835 INP_UNLOCK_ASSERT(inp);
2841 * Expose IGMP's multicast filter mode and source list(s) to userland,
2842 * keyed by (ifindex, group).
2843 * The filter mode is written out as a uint32_t, followed by
2844 * 0..n of struct in_addr.
2845 * For use by ifmcstat(8).
2846 * SMPng: NOTE: unlocked read of ifindex space.
2849 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2851 struct in_addr src, group;
2852 struct epoch_tracker et;
2854 struct ifmultiaddr *ifma;
2855 struct in_multi *inm;
2856 struct ip_msource *ims;
2860 uint32_t fmode, ifindex;
2865 if (req->newptr != NULL)
2871 group.s_addr = name[1];
2872 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2873 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2874 __func__, ntohl(group.s_addr));
2879 NET_EPOCH_ENTER(et);
2880 ifp = ifnet_byindex(ifindex);
2883 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2888 retval = sysctl_wire_old_buffer(req,
2889 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2895 IN_MULTI_LIST_LOCK();
2897 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2898 inm = inm_ifmultiaddr_get_inm(ifma);
2901 if (!in_hosteq(inm->inm_addr, group))
2903 fmode = inm->inm_st[1].iss_fmode;
2904 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2907 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2908 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2911 * Only copy-out sources which are in-mode.
2913 if (fmode != ims_get_mode(inm, ims, 1)) {
2914 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2918 src.s_addr = htonl(ims->ims_haddr);
2919 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2925 IN_MULTI_LIST_UNLOCK();
2931 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2933 static const char *inm_modestrs[] = {
2934 [MCAST_UNDEFINED] = "un",
2935 [MCAST_INCLUDE] = "in",
2936 [MCAST_EXCLUDE] = "ex",
2938 _Static_assert(MCAST_UNDEFINED == 0 &&
2939 MCAST_EXCLUDE + 1 == nitems(inm_modestrs),
2940 "inm_modestrs: no longer matches #defines");
2943 inm_mode_str(const int mode)
2946 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2947 return (inm_modestrs[mode]);
2951 static const char *inm_statestrs[] = {
2952 [IGMP_NOT_MEMBER] = "not-member",
2953 [IGMP_SILENT_MEMBER] = "silent",
2954 [IGMP_REPORTING_MEMBER] = "reporting",
2955 [IGMP_IDLE_MEMBER] = "idle",
2956 [IGMP_LAZY_MEMBER] = "lazy",
2957 [IGMP_SLEEPING_MEMBER] = "sleeping",
2958 [IGMP_AWAKENING_MEMBER] = "awakening",
2959 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending",
2960 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending",
2961 [IGMP_LEAVING_MEMBER] = "leaving",
2963 _Static_assert(IGMP_NOT_MEMBER == 0 &&
2964 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs),
2965 "inm_statetrs: no longer matches #defines");
2968 inm_state_str(const int state)
2971 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2972 return (inm_statestrs[state]);
2977 * Dump an in_multi structure to the console.
2980 inm_print(const struct in_multi *inm)
2983 char addrbuf[INET_ADDRSTRLEN];
2985 if ((ktr_mask & KTR_IGMPV3) == 0)
2988 printf("%s: --- begin inm %p ---\n", __func__, inm);
2989 printf("addr %s ifp %p(%s) ifma %p\n",
2990 inet_ntoa_r(inm->inm_addr, addrbuf),
2992 inm->inm_ifp->if_xname,
2994 printf("timer %u state %s refcount %u scq.len %u\n",
2996 inm_state_str(inm->inm_state),
2998 inm->inm_scq.mq_len);
2999 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3004 for (t = 0; t < 2; t++) {
3005 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3006 inm_mode_str(inm->inm_st[t].iss_fmode),
3007 inm->inm_st[t].iss_asm,
3008 inm->inm_st[t].iss_ex,
3009 inm->inm_st[t].iss_in,
3010 inm->inm_st[t].iss_rec);
3012 printf("%s: --- end inm %p ---\n", __func__, inm);
3015 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3018 inm_print(const struct in_multi *inm)
3023 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3025 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);