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");
119 * Functions with non-static linkage defined in this file should be
120 * declared in in_var.h:
123 * in_joingroup_locked()
125 * in_leavegroup_locked()
131 static void imf_commit(struct in_mfilter *);
132 static int imf_get_source(struct in_mfilter *imf,
133 const struct sockaddr_in *psin,
134 struct in_msource **);
135 static struct in_msource *
136 imf_graft(struct in_mfilter *, const uint8_t,
137 const struct sockaddr_in *);
138 static void imf_leave(struct in_mfilter *);
139 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
140 static void imf_purge(struct in_mfilter *);
141 static void imf_rollback(struct in_mfilter *);
142 static void imf_reap(struct in_mfilter *);
143 static struct in_mfilter *
144 imo_match_group(const struct ip_moptions *,
145 const struct ifnet *, const struct sockaddr *);
146 static struct in_msource *
147 imo_match_source(struct in_mfilter *, const struct sockaddr *);
148 static void ims_merge(struct ip_msource *ims,
149 const struct in_msource *lims, const int rollback);
150 static int in_getmulti(struct ifnet *, const struct in_addr *,
152 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
153 const int noalloc, struct ip_msource **pims);
155 static int inm_is_ifp_detached(const struct in_multi *);
157 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
158 static void inm_purge(struct in_multi *);
159 static void inm_reap(struct in_multi *);
160 static void inm_release(struct in_multi *);
161 static struct ip_moptions *
162 inp_findmoptions(struct inpcb *);
163 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
164 static int inp_join_group(struct inpcb *, struct sockopt *);
165 static int inp_leave_group(struct inpcb *, struct sockopt *);
166 static struct ifnet *
167 inp_lookup_mcast_ifp(const struct inpcb *,
168 const struct sockaddr_in *, const struct in_addr);
169 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
170 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
171 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
172 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
174 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast,
175 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
178 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
179 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
180 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
181 "Max source filters per group");
183 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
184 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
185 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
186 "Max source filters per socket");
188 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
189 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
190 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
192 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
193 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
194 "Per-interface stack-wide source filters");
198 * Inline function which wraps assertions for a valid ifp.
199 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
203 inm_is_ifp_detached(const struct in_multi *inm)
207 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
208 ifp = inm->inm_ifma->ifma_ifp;
211 * Sanity check that netinet's notion of ifp is the
214 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
217 return (ifp == NULL);
222 * Interface detach can happen in a taskqueue thread context, so we must use a
223 * dedicated thread to avoid deadlocks when draining inm_release tasks.
225 TASKQUEUE_DEFINE_THREAD(inm_free);
226 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER();
227 static void inm_release_task(void *arg __unused, int pending __unused);
228 static struct task inm_free_task = TASK_INITIALIZER(0, inm_release_task, NULL);
231 inm_release_wait(void *arg __unused)
235 * Make sure all pending multicast addresses are freed before
236 * the VNET or network device is destroyed:
238 taskqueue_drain(taskqueue_inm_free, &inm_free_task);
241 /* XXX-BZ FIXME, see D24914. */
242 VNET_SYSUNINIT(inm_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, inm_release_wait, NULL);
246 inm_release_list_deferred(struct in_multi_head *inmh)
249 if (SLIST_EMPTY(inmh))
251 mtx_lock(&in_multi_free_mtx);
252 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
253 mtx_unlock(&in_multi_free_mtx);
254 taskqueue_enqueue(taskqueue_inm_free, &inm_free_task);
258 inm_disconnect(struct in_multi *inm)
261 struct ifmultiaddr *ifma, *ll_ifma;
264 IF_ADDR_WLOCK_ASSERT(ifp);
265 ifma = inm->inm_ifma;
268 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
269 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
270 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
272 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
273 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
274 MPASS(ifma != ll_ifma);
275 ifma->ifma_llifma = NULL;
276 MPASS(ll_ifma->ifma_llifma == NULL);
277 MPASS(ll_ifma->ifma_ifp == ifp);
278 if (--ll_ifma->ifma_refcount == 0) {
279 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
280 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
281 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
283 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
284 if_freemulti(ll_ifma);
291 inm_release_deferred(struct in_multi *inm)
293 struct in_multi_head tmp;
295 IN_MULTI_LIST_LOCK_ASSERT();
296 MPASS(inm->inm_refcount > 0);
297 if (--inm->inm_refcount == 0) {
300 inm->inm_ifma->ifma_protospec = NULL;
301 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
302 inm_release_list_deferred(&tmp);
307 inm_release_task(void *arg __unused, int pending __unused)
309 struct in_multi_head inm_free_tmp;
310 struct in_multi *inm, *tinm;
312 SLIST_INIT(&inm_free_tmp);
313 mtx_lock(&in_multi_free_mtx);
314 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
315 mtx_unlock(&in_multi_free_mtx);
317 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
318 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
326 * Initialize an in_mfilter structure to a known state at t0, t1
327 * with an empty source filter list.
330 imf_init(struct in_mfilter *imf, const int st0, const int st1)
332 memset(imf, 0, sizeof(struct in_mfilter));
333 RB_INIT(&imf->imf_sources);
334 imf->imf_st[0] = st0;
335 imf->imf_st[1] = st1;
339 ip_mfilter_alloc(const int mflags, const int st0, const int st1)
341 struct in_mfilter *imf;
343 imf = malloc(sizeof(*imf), M_INMFILTER, mflags);
345 imf_init(imf, st0, st1);
351 ip_mfilter_free(struct in_mfilter *imf)
355 free(imf, M_INMFILTER);
359 * Function for looking up an in_multi record for an IPv4 multicast address
360 * on a given interface. ifp must be valid. If no record found, return NULL.
361 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
364 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
366 struct ifmultiaddr *ifma;
367 struct in_multi *inm;
369 IN_MULTI_LIST_LOCK_ASSERT();
370 IF_ADDR_LOCK_ASSERT(ifp);
373 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
374 if (ifma->ifma_addr->sa_family != AF_INET ||
375 ifma->ifma_protospec == NULL)
377 inm = (struct in_multi *)ifma->ifma_protospec;
378 if (inm->inm_addr.s_addr == ina.s_addr)
386 * Wrapper for inm_lookup_locked().
387 * The IF_ADDR_LOCK will be taken on ifp and released on return.
390 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
392 struct epoch_tracker et;
393 struct in_multi *inm;
395 IN_MULTI_LIST_LOCK_ASSERT();
398 inm = inm_lookup_locked(ifp, ina);
405 * Find an IPv4 multicast group entry for this ip_moptions instance
406 * which matches the specified group, and optionally an interface.
407 * Return its index into the array, or -1 if not found.
409 static struct in_mfilter *
410 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
411 const struct sockaddr *group)
413 const struct sockaddr_in *gsin;
414 struct in_mfilter *imf;
415 struct in_multi *inm;
417 gsin = (const struct sockaddr_in *)group;
419 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
423 if ((ifp == NULL || (inm->inm_ifp == ifp)) &&
424 in_hosteq(inm->inm_addr, gsin->sin_addr)) {
432 * Find an IPv4 multicast source entry for this imo which matches
433 * the given group index for this socket, and source address.
435 * NOTE: This does not check if the entry is in-mode, merely if
436 * it exists, which may not be the desired behaviour.
438 static struct in_msource *
439 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src)
441 struct ip_msource find;
442 struct ip_msource *ims;
443 const sockunion_t *psa;
445 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
447 /* Source trees are keyed in host byte order. */
448 psa = (const sockunion_t *)src;
449 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
450 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
452 return ((struct in_msource *)ims);
456 * Perform filtering for multicast datagrams on a socket by group and source.
458 * Returns 0 if a datagram should be allowed through, or various error codes
459 * if the socket was not a member of the group, or the source was muted, etc.
462 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
463 const struct sockaddr *group, const struct sockaddr *src)
465 struct in_mfilter *imf;
466 struct in_msource *ims;
469 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
471 imf = imo_match_group(imo, ifp, group);
473 return (MCAST_NOTGMEMBER);
476 * Check if the source was included in an (S,G) join.
477 * Allow reception on exclusive memberships by default,
478 * reject reception on inclusive memberships by default.
479 * Exclude source only if an in-mode exclude filter exists.
480 * Include source only if an in-mode include filter exists.
481 * NOTE: We are comparing group state here at IGMP t1 (now)
482 * with socket-layer t0 (since last downcall).
484 mode = imf->imf_st[1];
485 ims = imo_match_source(imf, src);
487 if ((ims == NULL && mode == MCAST_INCLUDE) ||
488 (ims != NULL && ims->imsl_st[0] != mode))
489 return (MCAST_NOTSMEMBER);
495 * Find and return a reference to an in_multi record for (ifp, group),
496 * and bump its reference count.
497 * If one does not exist, try to allocate it, and update link-layer multicast
498 * filters on ifp to listen for group.
499 * Assumes the IN_MULTI lock is held across the call.
500 * Return 0 if successful, otherwise return an appropriate error code.
503 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
504 struct in_multi **pinm)
506 struct sockaddr_in gsin;
507 struct ifmultiaddr *ifma;
508 struct in_ifinfo *ii;
509 struct in_multi *inm;
512 IN_MULTI_LOCK_ASSERT();
514 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
515 IN_MULTI_LIST_LOCK();
516 inm = inm_lookup(ifp, *group);
519 * If we already joined this group, just bump the
520 * refcount and return it.
522 KASSERT(inm->inm_refcount >= 1,
523 ("%s: bad refcount %d", __func__, inm->inm_refcount));
524 inm_acquire_locked(inm);
527 IN_MULTI_LIST_UNLOCK();
531 memset(&gsin, 0, sizeof(gsin));
532 gsin.sin_family = AF_INET;
533 gsin.sin_len = sizeof(struct sockaddr_in);
534 gsin.sin_addr = *group;
537 * Check if a link-layer group is already associated
538 * with this network-layer group on the given ifnet.
540 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
544 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
545 IN_MULTI_LIST_LOCK();
549 * If something other than netinet is occupying the link-layer
550 * group, print a meaningful error message and back out of
552 * Otherwise, bump the refcount on the existing network-layer
553 * group association and return it.
555 if (ifma->ifma_protospec != NULL) {
556 inm = (struct in_multi *)ifma->ifma_protospec;
558 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
560 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
561 ("%s: ifma not AF_INET", __func__));
562 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
563 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
564 !in_hosteq(inm->inm_addr, *group)) {
565 char addrbuf[INET_ADDRSTRLEN];
567 panic("%s: ifma %p is inconsistent with %p (%s)",
568 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
571 inm_acquire_locked(inm);
576 IF_ADDR_WLOCK_ASSERT(ifp);
579 * A new in_multi record is needed; allocate and initialize it.
580 * We DO NOT perform an IGMP join as the in_ layer may need to
581 * push an initial source list down to IGMP to support SSM.
583 * The initial source filter state is INCLUDE, {} as per the RFC.
585 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
587 IF_ADDR_WUNLOCK(ifp);
588 IN_MULTI_LIST_UNLOCK();
589 if_delmulti_ifma(ifma);
592 inm->inm_addr = *group;
594 inm->inm_igi = ii->ii_igmp;
595 inm->inm_ifma = ifma;
596 inm->inm_refcount = 1;
597 inm->inm_state = IGMP_NOT_MEMBER;
598 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
599 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
600 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
601 RB_INIT(&inm->inm_srcs);
603 ifma->ifma_protospec = inm;
607 IF_ADDR_WUNLOCK(ifp);
608 IN_MULTI_LIST_UNLOCK();
613 * Drop a reference to an in_multi record.
615 * If the refcount drops to 0, free the in_multi record and
616 * delete the underlying link-layer membership.
619 inm_release(struct in_multi *inm)
621 struct ifmultiaddr *ifma;
624 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
625 MPASS(inm->inm_refcount == 0);
626 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
628 ifma = inm->inm_ifma;
631 /* XXX this access is not covered by IF_ADDR_LOCK */
632 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
634 CURVNET_SET(ifp->if_vnet);
636 free(inm, M_IPMADDR);
637 if_delmulti_ifma_flags(ifma, 1);
642 free(inm, M_IPMADDR);
643 if_delmulti_ifma_flags(ifma, 1);
648 * Clear recorded source entries for a group.
649 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
650 * FIXME: Should reap.
653 inm_clear_recorded(struct in_multi *inm)
655 struct ip_msource *ims;
657 IN_MULTI_LIST_LOCK_ASSERT();
659 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
662 --inm->inm_st[1].iss_rec;
665 KASSERT(inm->inm_st[1].iss_rec == 0,
666 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
670 * Record a source as pending for a Source-Group IGMPv3 query.
671 * This lives here as it modifies the shared tree.
673 * inm is the group descriptor.
674 * naddr is the address of the source to record in network-byte order.
676 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
677 * lazy-allocate a source node in response to an SG query.
678 * Otherwise, no allocation is performed. This saves some memory
679 * with the trade-off that the source will not be reported to the
680 * router if joined in the window between the query response and
681 * the group actually being joined on the local host.
683 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
684 * This turns off the allocation of a recorded source entry if
685 * the group has not been joined.
687 * Return 0 if the source didn't exist or was already marked as recorded.
688 * Return 1 if the source was marked as recorded by this function.
689 * Return <0 if any error occurred (negated errno code).
692 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
694 struct ip_msource find;
695 struct ip_msource *ims, *nims;
697 IN_MULTI_LIST_LOCK_ASSERT();
699 find.ims_haddr = ntohl(naddr);
700 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
701 if (ims && ims->ims_stp)
704 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
706 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
710 nims->ims_haddr = find.ims_haddr;
711 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
717 * Mark the source as recorded and update the recorded
721 ++inm->inm_st[1].iss_rec;
727 * Return a pointer to an in_msource owned by an in_mfilter,
728 * given its source address.
729 * Lazy-allocate if needed. If this is a new entry its filter state is
732 * imf is the filter set being modified.
733 * haddr is the source address in *host* byte-order.
735 * SMPng: May be called with locks held; malloc must not block.
738 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
739 struct in_msource **plims)
741 struct ip_msource find;
742 struct ip_msource *ims, *nims;
743 struct in_msource *lims;
750 /* key is host byte order */
751 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
752 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
753 lims = (struct in_msource *)ims;
755 if (imf->imf_nsrc == in_mcast_maxsocksrc)
757 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
761 lims = (struct in_msource *)nims;
762 lims->ims_haddr = find.ims_haddr;
763 lims->imsl_st[0] = MCAST_UNDEFINED;
764 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
774 * Graft a source entry into an existing socket-layer filter set,
775 * maintaining any required invariants and checking allocations.
777 * The source is marked as being in the new filter mode at t1.
779 * Return the pointer to the new node, otherwise return NULL.
781 static struct in_msource *
782 imf_graft(struct in_mfilter *imf, const uint8_t st1,
783 const struct sockaddr_in *psin)
785 struct ip_msource *nims;
786 struct in_msource *lims;
788 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
792 lims = (struct in_msource *)nims;
793 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
794 lims->imsl_st[0] = MCAST_UNDEFINED;
795 lims->imsl_st[1] = st1;
796 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
803 * Prune a source entry from an existing socket-layer filter set,
804 * maintaining any required invariants and checking allocations.
806 * The source is marked as being left at t1, it is not freed.
808 * Return 0 if no error occurred, otherwise return an errno value.
811 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
813 struct ip_msource find;
814 struct ip_msource *ims;
815 struct in_msource *lims;
817 /* key is host byte order */
818 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
819 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
822 lims = (struct in_msource *)ims;
823 lims->imsl_st[1] = MCAST_UNDEFINED;
828 * Revert socket-layer filter set deltas at t1 to t0 state.
831 imf_rollback(struct in_mfilter *imf)
833 struct ip_msource *ims, *tims;
834 struct in_msource *lims;
836 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
837 lims = (struct in_msource *)ims;
838 if (lims->imsl_st[0] == lims->imsl_st[1]) {
839 /* no change at t1 */
841 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
842 /* revert change to existing source at t1 */
843 lims->imsl_st[1] = lims->imsl_st[0];
845 /* revert source added t1 */
846 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
847 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
848 free(ims, M_INMFILTER);
852 imf->imf_st[1] = imf->imf_st[0];
856 * Mark socket-layer filter set as INCLUDE {} at t1.
859 imf_leave(struct in_mfilter *imf)
861 struct ip_msource *ims;
862 struct in_msource *lims;
864 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
865 lims = (struct in_msource *)ims;
866 lims->imsl_st[1] = MCAST_UNDEFINED;
868 imf->imf_st[1] = MCAST_INCLUDE;
872 * Mark socket-layer filter set deltas as committed.
875 imf_commit(struct in_mfilter *imf)
877 struct ip_msource *ims;
878 struct in_msource *lims;
880 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
881 lims = (struct in_msource *)ims;
882 lims->imsl_st[0] = lims->imsl_st[1];
884 imf->imf_st[0] = imf->imf_st[1];
888 * Reap unreferenced sources from socket-layer filter set.
891 imf_reap(struct in_mfilter *imf)
893 struct ip_msource *ims, *tims;
894 struct in_msource *lims;
896 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
897 lims = (struct in_msource *)ims;
898 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
899 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
900 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
901 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
902 free(ims, M_INMFILTER);
909 * Purge socket-layer filter set.
912 imf_purge(struct in_mfilter *imf)
914 struct ip_msource *ims, *tims;
916 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
917 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
918 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
919 free(ims, M_INMFILTER);
922 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
923 KASSERT(RB_EMPTY(&imf->imf_sources),
924 ("%s: imf_sources not empty", __func__));
928 * Look up a source filter entry for a multicast group.
930 * inm is the group descriptor to work with.
931 * haddr is the host-byte-order IPv4 address to look up.
932 * noalloc may be non-zero to suppress allocation of sources.
933 * *pims will be set to the address of the retrieved or allocated source.
935 * SMPng: NOTE: may be called with locks held.
936 * Return 0 if successful, otherwise return a non-zero error code.
939 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
940 const int noalloc, struct ip_msource **pims)
942 struct ip_msource find;
943 struct ip_msource *ims, *nims;
945 find.ims_haddr = haddr;
946 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
947 if (ims == NULL && !noalloc) {
948 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
950 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
954 nims->ims_haddr = haddr;
955 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
959 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
969 * Merge socket-layer source into IGMP-layer source.
970 * If rollback is non-zero, perform the inverse of the merge.
973 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
976 int n = rollback ? -1 : 1;
978 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
979 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
980 __func__, n, ims->ims_haddr);
981 ims->ims_st[1].ex -= n;
982 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
983 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
984 __func__, n, ims->ims_haddr);
985 ims->ims_st[1].in -= n;
988 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
989 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
990 __func__, n, ims->ims_haddr);
991 ims->ims_st[1].ex += n;
992 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
993 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
994 __func__, n, ims->ims_haddr);
995 ims->ims_st[1].in += n;
1000 * Atomically update the global in_multi state, when a membership's
1001 * filter list is being updated in any way.
1003 * imf is the per-inpcb-membership group filter pointer.
1004 * A fake imf may be passed for in-kernel consumers.
1006 * XXX This is a candidate for a set-symmetric-difference style loop
1007 * which would eliminate the repeated lookup from root of ims nodes,
1008 * as they share the same key space.
1010 * If any error occurred this function will back out of refcounts
1011 * and return a non-zero value.
1014 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1016 struct ip_msource *ims, *nims;
1017 struct in_msource *lims;
1018 int schanged, error;
1024 IN_MULTI_LIST_LOCK_ASSERT();
1027 * Update the source filters first, as this may fail.
1028 * Maintain count of in-mode filters at t0, t1. These are
1029 * used to work out if we transition into ASM mode or not.
1030 * Maintain a count of source filters whose state was
1031 * actually modified by this operation.
1033 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1034 lims = (struct in_msource *)ims;
1035 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1036 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1037 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1038 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1042 ims_merge(nims, lims, 0);
1045 struct ip_msource *bims;
1047 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1048 lims = (struct in_msource *)ims;
1049 if (lims->imsl_st[0] == lims->imsl_st[1])
1051 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1054 ims_merge(bims, lims, 1);
1059 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1060 __func__, nsrc0, nsrc1);
1062 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1063 if (imf->imf_st[0] == imf->imf_st[1] &&
1064 imf->imf_st[1] == MCAST_INCLUDE) {
1066 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1067 --inm->inm_st[1].iss_in;
1071 /* Handle filter mode transition on socket. */
1072 if (imf->imf_st[0] != imf->imf_st[1]) {
1073 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1074 __func__, imf->imf_st[0], imf->imf_st[1]);
1076 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1077 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1078 --inm->inm_st[1].iss_ex;
1079 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1080 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1081 --inm->inm_st[1].iss_in;
1084 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1085 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1086 inm->inm_st[1].iss_ex++;
1087 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1088 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1089 inm->inm_st[1].iss_in++;
1094 * Track inm filter state in terms of listener counts.
1095 * If there are any exclusive listeners, stack-wide
1096 * membership is exclusive.
1097 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1098 * If no listeners remain, state is undefined at t1,
1099 * and the IGMP lifecycle for this group should finish.
1101 if (inm->inm_st[1].iss_ex > 0) {
1102 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1103 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1104 } else if (inm->inm_st[1].iss_in > 0) {
1105 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1106 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1108 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1109 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1112 /* Decrement ASM listener count on transition out of ASM mode. */
1113 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1114 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1115 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1116 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1117 --inm->inm_st[1].iss_asm;
1121 /* Increment ASM listener count on transition to ASM mode. */
1122 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1123 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1124 inm->inm_st[1].iss_asm++;
1127 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1132 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1139 * Mark an in_multi's filter set deltas as committed.
1140 * Called by IGMP after a state change has been enqueued.
1143 inm_commit(struct in_multi *inm)
1145 struct ip_msource *ims;
1147 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1148 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1151 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1152 ims->ims_st[0] = ims->ims_st[1];
1154 inm->inm_st[0] = inm->inm_st[1];
1158 * Reap unreferenced nodes from an in_multi's filter set.
1161 inm_reap(struct in_multi *inm)
1163 struct ip_msource *ims, *tims;
1165 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1166 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1167 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1170 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1171 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1172 free(ims, M_IPMSOURCE);
1178 * Purge all source nodes from an in_multi's filter set.
1181 inm_purge(struct in_multi *inm)
1183 struct ip_msource *ims, *tims;
1185 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1186 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1187 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1188 free(ims, M_IPMSOURCE);
1194 * Join a multicast group; unlocked entry point.
1196 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1197 * is not held. Fortunately, ifp is unlikely to have been detached
1198 * at this point, so we assume it's OK to recurse.
1201 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1202 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1207 error = in_joingroup_locked(ifp, gina, imf, pinm);
1214 * Join a multicast group; real entry point.
1216 * Only preserves atomicity at inm level.
1217 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1219 * If the IGMP downcall fails, the group is not joined, and an error
1223 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1224 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1226 struct in_mfilter timf;
1227 struct in_multi *inm;
1230 IN_MULTI_LOCK_ASSERT();
1231 IN_MULTI_LIST_UNLOCK_ASSERT();
1233 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1234 ntohl(gina->s_addr), ifp, ifp->if_xname);
1240 * If no imf was specified (i.e. kernel consumer),
1241 * fake one up and assume it is an ASM join.
1244 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1248 error = in_getmulti(ifp, gina, &inm);
1250 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1253 IN_MULTI_LIST_LOCK();
1254 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1255 error = inm_merge(inm, imf);
1257 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1258 goto out_inm_release;
1261 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1262 error = igmp_change_state(inm);
1264 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1265 goto out_inm_release;
1270 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1272 inm_release_deferred(inm);
1273 IF_ADDR_WUNLOCK(ifp);
1277 IN_MULTI_LIST_UNLOCK();
1283 * Leave a multicast group; unlocked entry point.
1286 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1291 error = in_leavegroup_locked(inm, imf);
1298 * Leave a multicast group; real entry point.
1299 * All source filters will be expunged.
1301 * Only preserves atomicity at inm level.
1303 * Holding the write lock for the INP which contains imf
1304 * is highly advisable. We can't assert for it as imf does not
1305 * contain a back-pointer to the owning inp.
1307 * Note: This is not the same as inm_release(*) as this function also
1308 * makes a state change downcall into IGMP.
1311 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1313 struct in_mfilter timf;
1316 IN_MULTI_LOCK_ASSERT();
1317 IN_MULTI_LIST_UNLOCK_ASSERT();
1321 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1322 inm, ntohl(inm->inm_addr.s_addr),
1323 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1327 * If no imf was specified (i.e. kernel consumer),
1328 * fake one up and assume it is an ASM join.
1331 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1336 * Begin state merge transaction at IGMP layer.
1338 * As this particular invocation should not cause any memory
1339 * to be allocated, and there is no opportunity to roll back
1340 * the transaction, it MUST NOT fail.
1342 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1343 IN_MULTI_LIST_LOCK();
1344 error = inm_merge(inm, imf);
1345 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1347 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1348 CURVNET_SET(inm->inm_ifp->if_vnet);
1349 error = igmp_change_state(inm);
1350 IF_ADDR_WLOCK(inm->inm_ifp);
1351 inm_release_deferred(inm);
1352 IF_ADDR_WUNLOCK(inm->inm_ifp);
1353 IN_MULTI_LIST_UNLOCK();
1356 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1358 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1363 /*#ifndef BURN_BRIDGES*/
1366 * Block or unblock an ASM multicast source on an inpcb.
1367 * This implements the delta-based API described in RFC 3678.
1369 * The delta-based API applies only to exclusive-mode memberships.
1370 * An IGMP downcall will be performed.
1372 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1374 * Return 0 if successful, otherwise return an appropriate error code.
1377 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1379 struct epoch_tracker et;
1380 struct group_source_req gsr;
1381 sockunion_t *gsa, *ssa;
1383 struct in_mfilter *imf;
1384 struct ip_moptions *imo;
1385 struct in_msource *ims;
1386 struct in_multi *inm;
1394 memset(&gsr, 0, sizeof(struct group_source_req));
1395 gsa = (sockunion_t *)&gsr.gsr_group;
1396 ssa = (sockunion_t *)&gsr.gsr_source;
1398 switch (sopt->sopt_name) {
1399 case IP_BLOCK_SOURCE:
1400 case IP_UNBLOCK_SOURCE: {
1401 struct ip_mreq_source mreqs;
1403 error = sooptcopyin(sopt, &mreqs,
1404 sizeof(struct ip_mreq_source),
1405 sizeof(struct ip_mreq_source));
1409 gsa->sin.sin_family = AF_INET;
1410 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1411 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1413 ssa->sin.sin_family = AF_INET;
1414 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1415 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1417 if (!in_nullhost(mreqs.imr_interface)) {
1418 NET_EPOCH_ENTER(et);
1419 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1423 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1426 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1427 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1431 case MCAST_BLOCK_SOURCE:
1432 case MCAST_UNBLOCK_SOURCE:
1433 error = sooptcopyin(sopt, &gsr,
1434 sizeof(struct group_source_req),
1435 sizeof(struct group_source_req));
1439 if (gsa->sin.sin_family != AF_INET ||
1440 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1443 if (ssa->sin.sin_family != AF_INET ||
1444 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1447 NET_EPOCH_ENTER(et);
1448 ifp = ifnet_byindex(gsr.gsr_interface);
1451 return (EADDRNOTAVAIL);
1453 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1458 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1459 __func__, sopt->sopt_name);
1460 return (EOPNOTSUPP);
1464 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1470 * Check if we are actually a member of this group.
1472 imo = inp_findmoptions(inp);
1473 imf = imo_match_group(imo, ifp, &gsa->sa);
1475 error = EADDRNOTAVAIL;
1476 goto out_inp_locked;
1481 * Attempting to use the delta-based API on an
1482 * non exclusive-mode membership is an error.
1484 fmode = imf->imf_st[0];
1485 if (fmode != MCAST_EXCLUDE) {
1487 goto out_inp_locked;
1491 * Deal with error cases up-front:
1492 * Asked to block, but already blocked; or
1493 * Asked to unblock, but nothing to unblock.
1494 * If adding a new block entry, allocate it.
1496 ims = imo_match_source(imf, &ssa->sa);
1497 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1498 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1499 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1500 error = EADDRNOTAVAIL;
1501 goto out_inp_locked;
1504 INP_WLOCK_ASSERT(inp);
1507 * Begin state merge transaction at socket layer.
1510 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1511 ims = imf_graft(imf, fmode, &ssa->sin);
1515 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1516 error = imf_prune(imf, &ssa->sin);
1520 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1521 goto out_imf_rollback;
1525 * Begin state merge transaction at IGMP layer.
1527 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1528 IN_MULTI_LIST_LOCK();
1529 error = inm_merge(inm, imf);
1531 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1532 IN_MULTI_LIST_UNLOCK();
1533 goto out_imf_rollback;
1536 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1537 error = igmp_change_state(inm);
1538 IN_MULTI_LIST_UNLOCK();
1540 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1557 * Given an inpcb, return its multicast options structure pointer. Accepts
1558 * an unlocked inpcb pointer, but will return it locked. May sleep.
1560 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1561 * SMPng: NOTE: Returns with the INP write lock held.
1563 static struct ip_moptions *
1564 inp_findmoptions(struct inpcb *inp)
1566 struct ip_moptions *imo;
1569 if (inp->inp_moptions != NULL)
1570 return (inp->inp_moptions);
1574 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1576 imo->imo_multicast_ifp = NULL;
1577 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1578 imo->imo_multicast_vif = -1;
1579 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1580 imo->imo_multicast_loop = in_mcast_loop;
1581 STAILQ_INIT(&imo->imo_head);
1584 if (inp->inp_moptions != NULL) {
1585 free(imo, M_IPMOPTS);
1586 return (inp->inp_moptions);
1588 inp->inp_moptions = imo;
1593 inp_freemoptions(struct ip_moptions *imo)
1595 struct in_mfilter *imf;
1596 struct in_multi *inm;
1602 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1603 ip_mfilter_remove(&imo->imo_head, imf);
1606 if ((inm = imf->imf_inm) != NULL) {
1607 if ((ifp = inm->inm_ifp) != NULL) {
1608 CURVNET_SET(ifp->if_vnet);
1609 (void)in_leavegroup(inm, imf);
1612 (void)in_leavegroup(inm, imf);
1615 ip_mfilter_free(imf);
1617 free(imo, M_IPMOPTS);
1621 * Atomically get source filters on a socket for an IPv4 multicast group.
1622 * Called with INP lock held; returns with lock released.
1625 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1627 struct epoch_tracker et;
1628 struct __msfilterreq msfr;
1631 struct ip_moptions *imo;
1632 struct in_mfilter *imf;
1633 struct ip_msource *ims;
1634 struct in_msource *lims;
1635 struct sockaddr_in *psin;
1636 struct sockaddr_storage *ptss;
1637 struct sockaddr_storage *tss;
1639 size_t nsrcs, ncsrcs;
1641 INP_WLOCK_ASSERT(inp);
1643 imo = inp->inp_moptions;
1644 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1648 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1649 sizeof(struct __msfilterreq));
1653 NET_EPOCH_ENTER(et);
1654 ifp = ifnet_byindex(msfr.msfr_ifindex);
1655 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifnet pointer left */
1662 * Lookup group on the socket.
1664 gsa = (sockunion_t *)&msfr.msfr_group;
1665 imf = imo_match_group(imo, ifp, &gsa->sa);
1668 return (EADDRNOTAVAIL);
1672 * Ignore memberships which are in limbo.
1674 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1678 msfr.msfr_fmode = imf->imf_st[1];
1681 * If the user specified a buffer, copy out the source filter
1682 * entries to userland gracefully.
1683 * We only copy out the number of entries which userland
1684 * has asked for, but we always tell userland how big the
1685 * buffer really needs to be.
1687 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1688 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1690 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1691 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1692 M_TEMP, M_NOWAIT | M_ZERO);
1700 * Count number of sources in-mode at t0.
1701 * If buffer space exists and remains, copy out source entries.
1703 nsrcs = msfr.msfr_nsrcs;
1706 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1707 lims = (struct in_msource *)ims;
1708 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1709 lims->imsl_st[0] != imf->imf_st[0])
1712 if (tss != NULL && nsrcs > 0) {
1713 psin = (struct sockaddr_in *)ptss;
1714 psin->sin_family = AF_INET;
1715 psin->sin_len = sizeof(struct sockaddr_in);
1716 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1726 error = copyout(tss, msfr.msfr_srcs,
1727 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1733 msfr.msfr_nsrcs = ncsrcs;
1734 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1740 * Return the IP multicast options in response to user getsockopt().
1743 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1745 struct ip_mreqn mreqn;
1746 struct ip_moptions *imo;
1748 struct in_ifaddr *ia;
1753 imo = inp->inp_moptions;
1755 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1756 * or is a divert socket, reject it.
1758 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1759 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1760 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1762 return (EOPNOTSUPP);
1766 switch (sopt->sopt_name) {
1767 case IP_MULTICAST_VIF:
1769 optval = imo->imo_multicast_vif;
1773 error = sooptcopyout(sopt, &optval, sizeof(int));
1776 case IP_MULTICAST_IF:
1777 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1779 ifp = imo->imo_multicast_ifp;
1780 if (!in_nullhost(imo->imo_multicast_addr)) {
1781 mreqn.imr_address = imo->imo_multicast_addr;
1782 } else if (ifp != NULL) {
1783 struct epoch_tracker et;
1785 mreqn.imr_ifindex = ifp->if_index;
1786 NET_EPOCH_ENTER(et);
1790 IA_SIN(ia)->sin_addr;
1795 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1796 error = sooptcopyout(sopt, &mreqn,
1797 sizeof(struct ip_mreqn));
1799 error = sooptcopyout(sopt, &mreqn.imr_address,
1800 sizeof(struct in_addr));
1804 case IP_MULTICAST_TTL:
1806 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1808 optval = coptval = imo->imo_multicast_ttl;
1810 if (sopt->sopt_valsize == sizeof(u_char))
1811 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1813 error = sooptcopyout(sopt, &optval, sizeof(int));
1816 case IP_MULTICAST_LOOP:
1818 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1820 optval = coptval = imo->imo_multicast_loop;
1822 if (sopt->sopt_valsize == sizeof(u_char))
1823 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1825 error = sooptcopyout(sopt, &optval, sizeof(int));
1830 error = EADDRNOTAVAIL;
1833 error = inp_get_source_filters(inp, sopt);
1839 error = ENOPROTOOPT;
1843 INP_UNLOCK_ASSERT(inp);
1849 * Look up the ifnet to use for a multicast group membership,
1850 * given the IPv4 address of an interface, and the IPv4 group address.
1852 * This routine exists to support legacy multicast applications
1853 * which do not understand that multicast memberships are scoped to
1854 * specific physical links in the networking stack, or which need
1855 * to join link-scope groups before IPv4 addresses are configured.
1857 * Use this socket's current FIB number for any required FIB lookup.
1858 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1859 * and use its ifp; usually, this points to the default next-hop.
1861 * If the FIB lookup fails, attempt to use the first non-loopback
1862 * interface with multicast capability in the system as a
1863 * last resort. The legacy IPv4 ASM API requires that we do
1864 * this in order to allow groups to be joined when the routing
1865 * table has not yet been populated during boot.
1867 * Returns NULL if no ifp could be found, otherwise return referenced ifp.
1869 * FUTURE: Implement IPv4 source-address selection.
1871 static struct ifnet *
1872 inp_lookup_mcast_ifp(const struct inpcb *inp,
1873 const struct sockaddr_in *gsin, const struct in_addr ina)
1876 struct nhop_object *nh;
1879 KASSERT(inp != NULL, ("%s: inp must not be NULL", __func__));
1880 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1881 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1882 ("%s: not multicast", __func__));
1885 if (!in_nullhost(ina)) {
1886 INADDR_TO_IFP(ina, ifp);
1890 nh = fib4_lookup(inp->inp_inc.inc_fibnum, gsin->sin_addr, 0, NHR_NONE, 0);
1895 struct in_ifaddr *ia;
1899 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1901 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1902 (mifp->if_flags & IFF_MULTICAST)) {
1915 * Join an IPv4 multicast group, possibly with a source.
1918 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1920 struct group_source_req gsr;
1921 sockunion_t *gsa, *ssa;
1923 struct in_mfilter *imf;
1924 struct ip_moptions *imo;
1925 struct in_multi *inm;
1926 struct in_msource *lims;
1927 struct epoch_tracker et;
1934 memset(&gsr, 0, sizeof(struct group_source_req));
1935 gsa = (sockunion_t *)&gsr.gsr_group;
1936 gsa->ss.ss_family = AF_UNSPEC;
1937 ssa = (sockunion_t *)&gsr.gsr_source;
1938 ssa->ss.ss_family = AF_UNSPEC;
1940 switch (sopt->sopt_name) {
1941 case IP_ADD_MEMBERSHIP: {
1942 struct ip_mreqn mreqn;
1944 if (sopt->sopt_valsize == sizeof(struct ip_mreqn))
1945 error = sooptcopyin(sopt, &mreqn,
1946 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn));
1948 error = sooptcopyin(sopt, &mreqn,
1949 sizeof(struct ip_mreq), sizeof(struct ip_mreq));
1953 gsa->sin.sin_family = AF_INET;
1954 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1955 gsa->sin.sin_addr = mreqn.imr_multiaddr;
1956 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1959 NET_EPOCH_ENTER(et);
1960 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) &&
1961 mreqn.imr_ifindex != 0)
1962 ifp = ifnet_byindex_ref(mreqn.imr_ifindex);
1964 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1969 case IP_ADD_SOURCE_MEMBERSHIP: {
1970 struct ip_mreq_source mreqs;
1972 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source),
1973 sizeof(struct ip_mreq_source));
1977 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET;
1978 gsa->sin.sin_len = ssa->sin.sin_len =
1979 sizeof(struct sockaddr_in);
1981 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1982 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1985 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1987 NET_EPOCH_ENTER(et);
1988 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1989 mreqs.imr_interface);
1991 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1992 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1996 case MCAST_JOIN_GROUP:
1997 case MCAST_JOIN_SOURCE_GROUP:
1998 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1999 error = sooptcopyin(sopt, &gsr,
2000 sizeof(struct group_req),
2001 sizeof(struct group_req));
2002 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2003 error = sooptcopyin(sopt, &gsr,
2004 sizeof(struct group_source_req),
2005 sizeof(struct group_source_req));
2010 if (gsa->sin.sin_family != AF_INET ||
2011 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2015 * Overwrite the port field if present, as the sockaddr
2016 * being copied in may be matched with a binary comparison.
2018 gsa->sin.sin_port = 0;
2019 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2020 if (ssa->sin.sin_family != AF_INET ||
2021 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2023 ssa->sin.sin_port = 0;
2026 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2029 NET_EPOCH_ENTER(et);
2030 ifp = ifnet_byindex_ref(gsr.gsr_interface);
2033 return (EADDRNOTAVAIL);
2037 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2038 __func__, sopt->sopt_name);
2039 return (EOPNOTSUPP);
2043 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2046 return (EADDRNOTAVAIL);
2052 * Find the membership in the membership list.
2054 imo = inp_findmoptions(inp);
2055 imf = imo_match_group(imo, ifp, &gsa->sa);
2060 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) {
2062 goto out_inp_locked;
2068 if (ssa->ss.ss_family != AF_UNSPEC) {
2070 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2071 * is an error. On an existing inclusive membership,
2072 * it just adds the source to the filter list.
2074 if (imf->imf_st[1] != MCAST_INCLUDE) {
2076 goto out_inp_locked;
2079 * Throw out duplicates.
2081 * XXX FIXME: This makes a naive assumption that
2082 * even if entries exist for *ssa in this imf,
2083 * they will be rejected as dupes, even if they
2084 * are not valid in the current mode (in-mode).
2086 * in_msource is transactioned just as for anything
2087 * else in SSM -- but note naive use of inm_graft()
2088 * below for allocating new filter entries.
2090 * This is only an issue if someone mixes the
2091 * full-state SSM API with the delta-based API,
2092 * which is discouraged in the relevant RFCs.
2094 lims = imo_match_source(imf, &ssa->sa);
2095 if (lims != NULL /*&&
2096 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2097 error = EADDRNOTAVAIL;
2098 goto out_inp_locked;
2102 * MCAST_JOIN_GROUP on an existing exclusive
2103 * membership is an error; return EADDRINUSE
2104 * to preserve 4.4BSD API idempotence, and
2105 * avoid tedious detour to code below.
2106 * NOTE: This is bending RFC 3678 a bit.
2108 * On an existing inclusive membership, this is also
2109 * an error; if you want to change filter mode,
2110 * you must use the userland API setsourcefilter().
2111 * XXX We don't reject this for imf in UNDEFINED
2112 * state at t1, because allocation of a filter
2113 * is atomic with allocation of a membership.
2116 if (imf->imf_st[1] == MCAST_EXCLUDE)
2118 goto out_inp_locked;
2123 * Begin state merge transaction at socket layer.
2125 INP_WLOCK_ASSERT(inp);
2128 * Graft new source into filter list for this inpcb's
2129 * membership of the group. The in_multi may not have
2130 * been allocated yet if this is a new membership, however,
2131 * the in_mfilter slot will be allocated and must be initialized.
2133 * Note: Grafting of exclusive mode filters doesn't happen
2135 * XXX: Should check for non-NULL lims (node exists but may
2136 * not be in-mode) for interop with full-state API.
2138 if (ssa->ss.ss_family != AF_UNSPEC) {
2139 /* Membership starts in IN mode */
2141 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2142 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2145 goto out_inp_locked;
2148 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2150 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2152 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2155 goto out_inp_locked;
2158 /* No address specified; Membership starts in EX mode */
2160 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2161 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2164 goto out_inp_locked;
2170 * Begin state merge transaction at IGMP layer.
2176 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2180 if (in_pcbrele_wlocked(inp)) {
2182 goto out_inp_unlocked;
2185 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2187 goto out_inp_locked;
2190 * NOTE: Refcount from in_joingroup_locked()
2191 * is protecting membership.
2193 ip_mfilter_insert(&imo->imo_head, imf);
2195 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2196 IN_MULTI_LIST_LOCK();
2197 error = inm_merge(inm, imf);
2199 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2201 IN_MULTI_LIST_UNLOCK();
2204 goto out_inp_locked;
2206 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2207 error = igmp_change_state(inm);
2208 IN_MULTI_LIST_UNLOCK();
2210 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2214 goto out_inp_locked;
2226 if (is_new && imf) {
2227 if (imf->imf_inm != NULL) {
2228 IN_MULTI_LIST_LOCK();
2230 inm_release_deferred(imf->imf_inm);
2231 IF_ADDR_WUNLOCK(ifp);
2232 IN_MULTI_LIST_UNLOCK();
2234 ip_mfilter_free(imf);
2241 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2244 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2246 struct epoch_tracker et;
2247 struct group_source_req gsr;
2248 struct ip_mreq_source mreqs;
2249 sockunion_t *gsa, *ssa;
2251 struct in_mfilter *imf;
2252 struct ip_moptions *imo;
2253 struct in_msource *ims;
2254 struct in_multi *inm;
2262 memset(&gsr, 0, sizeof(struct group_source_req));
2263 gsa = (sockunion_t *)&gsr.gsr_group;
2264 gsa->ss.ss_family = AF_UNSPEC;
2265 ssa = (sockunion_t *)&gsr.gsr_source;
2266 ssa->ss.ss_family = AF_UNSPEC;
2268 switch (sopt->sopt_name) {
2269 case IP_DROP_MEMBERSHIP:
2270 case IP_DROP_SOURCE_MEMBERSHIP:
2271 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2272 error = sooptcopyin(sopt, &mreqs,
2273 sizeof(struct ip_mreq),
2274 sizeof(struct ip_mreq));
2276 * Swap interface and sourceaddr arguments,
2277 * as ip_mreq and ip_mreq_source are laid
2280 mreqs.imr_interface = mreqs.imr_sourceaddr;
2281 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2282 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2283 error = sooptcopyin(sopt, &mreqs,
2284 sizeof(struct ip_mreq_source),
2285 sizeof(struct ip_mreq_source));
2290 gsa->sin.sin_family = AF_INET;
2291 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2292 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2294 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2295 ssa->sin.sin_family = AF_INET;
2296 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2297 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2301 * Attempt to look up hinted ifp from interface address.
2302 * Fallthrough with null ifp iff lookup fails, to
2303 * preserve 4.4BSD mcast API idempotence.
2304 * XXX NOTE WELL: The RFC 3678 API is preferred because
2305 * using an IPv4 address as a key is racy.
2307 if (!in_nullhost(mreqs.imr_interface)) {
2308 NET_EPOCH_ENTER(et);
2309 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2313 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2314 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2318 case MCAST_LEAVE_GROUP:
2319 case MCAST_LEAVE_SOURCE_GROUP:
2320 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2321 error = sooptcopyin(sopt, &gsr,
2322 sizeof(struct group_req),
2323 sizeof(struct group_req));
2324 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2325 error = sooptcopyin(sopt, &gsr,
2326 sizeof(struct group_source_req),
2327 sizeof(struct group_source_req));
2332 if (gsa->sin.sin_family != AF_INET ||
2333 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2336 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2337 if (ssa->sin.sin_family != AF_INET ||
2338 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2342 NET_EPOCH_ENTER(et);
2343 ifp = ifnet_byindex(gsr.gsr_interface);
2344 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2346 return (EADDRNOTAVAIL);
2350 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2351 __func__, sopt->sopt_name);
2352 return (EOPNOTSUPP);
2356 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2362 * Find the membership in the membership list.
2364 imo = inp_findmoptions(inp);
2365 imf = imo_match_group(imo, ifp, &gsa->sa);
2367 error = EADDRNOTAVAIL;
2368 goto out_inp_locked;
2372 if (ssa->ss.ss_family != AF_UNSPEC)
2376 * Begin state merge transaction at socket layer.
2378 INP_WLOCK_ASSERT(inp);
2381 * If we were instructed only to leave a given source, do so.
2382 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2385 ip_mfilter_remove(&imo->imo_head, imf);
2389 * Give up the multicast address record to which
2390 * the membership points.
2392 (void) in_leavegroup_locked(imf->imf_inm, imf);
2394 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2395 error = EADDRNOTAVAIL;
2396 goto out_inp_locked;
2398 ims = imo_match_source(imf, &ssa->sa);
2400 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2401 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2402 error = EADDRNOTAVAIL;
2403 goto out_inp_locked;
2405 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2406 error = imf_prune(imf, &ssa->sin);
2408 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2410 goto out_inp_locked;
2415 * Begin state merge transaction at IGMP layer.
2418 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2419 IN_MULTI_LIST_LOCK();
2420 error = inm_merge(inm, imf);
2422 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2424 IN_MULTI_LIST_UNLOCK();
2427 goto out_inp_locked;
2430 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2431 error = igmp_change_state(inm);
2432 IN_MULTI_LIST_UNLOCK();
2434 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2438 goto out_inp_locked;
2447 if (is_final && imf)
2448 ip_mfilter_free(imf);
2455 * Select the interface for transmitting IPv4 multicast datagrams.
2457 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2458 * may be passed to this socket option. An address of INADDR_ANY or an
2459 * interface index of 0 is used to remove a previous selection.
2460 * When no interface is selected, one is chosen for every send.
2463 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2465 struct in_addr addr;
2466 struct ip_mreqn mreqn;
2468 struct ip_moptions *imo;
2471 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2473 * An interface index was specified using the
2474 * Linux-derived ip_mreqn structure.
2476 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2477 sizeof(struct ip_mreqn));
2481 if (mreqn.imr_ifindex < 0)
2484 if (mreqn.imr_ifindex == 0) {
2487 struct epoch_tracker et;
2489 NET_EPOCH_ENTER(et);
2490 ifp = ifnet_byindex(mreqn.imr_ifindex);
2491 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2493 return (EADDRNOTAVAIL);
2497 * An interface was specified by IPv4 address.
2498 * This is the traditional BSD usage.
2500 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2501 sizeof(struct in_addr));
2504 if (in_nullhost(addr)) {
2507 struct epoch_tracker et;
2509 NET_EPOCH_ENTER(et);
2510 INADDR_TO_IFP(addr, ifp);
2514 return (EADDRNOTAVAIL);
2516 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2517 ntohl(addr.s_addr));
2520 /* Reject interfaces which do not support multicast. */
2521 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2522 return (EOPNOTSUPP);
2524 imo = inp_findmoptions(inp);
2525 imo->imo_multicast_ifp = ifp;
2526 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2533 * Atomically set source filters on a socket for an IPv4 multicast group.
2535 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2538 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2540 struct epoch_tracker et;
2541 struct __msfilterreq msfr;
2544 struct in_mfilter *imf;
2545 struct ip_moptions *imo;
2546 struct in_multi *inm;
2549 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2550 sizeof(struct __msfilterreq));
2554 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2557 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2558 msfr.msfr_fmode != MCAST_INCLUDE))
2561 if (msfr.msfr_group.ss_family != AF_INET ||
2562 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2565 gsa = (sockunion_t *)&msfr.msfr_group;
2566 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2569 gsa->sin.sin_port = 0; /* ignore port */
2571 NET_EPOCH_ENTER(et);
2572 ifp = ifnet_byindex(msfr.msfr_ifindex);
2573 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2575 return (EADDRNOTAVAIL);
2580 * Take the INP write lock.
2581 * Check if this socket is a member of this group.
2583 imo = inp_findmoptions(inp);
2584 imf = imo_match_group(imo, ifp, &gsa->sa);
2586 error = EADDRNOTAVAIL;
2587 goto out_inp_locked;
2592 * Begin state merge transaction at socket layer.
2594 INP_WLOCK_ASSERT(inp);
2596 imf->imf_st[1] = msfr.msfr_fmode;
2599 * Apply any new source filters, if present.
2600 * Make a copy of the user-space source vector so
2601 * that we may copy them with a single copyin. This
2602 * allows us to deal with page faults up-front.
2604 if (msfr.msfr_nsrcs > 0) {
2605 struct in_msource *lims;
2606 struct sockaddr_in *psin;
2607 struct sockaddr_storage *kss, *pkss;
2612 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2613 __func__, (unsigned long)msfr.msfr_nsrcs);
2614 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2616 error = copyin(msfr.msfr_srcs, kss,
2617 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2626 * Mark all source filters as UNDEFINED at t1.
2627 * Restore new group filter mode, as imf_leave()
2628 * will set it to INCLUDE.
2631 imf->imf_st[1] = msfr.msfr_fmode;
2634 * Update socket layer filters at t1, lazy-allocating
2635 * new entries. This saves a bunch of memory at the
2636 * cost of one RB_FIND() per source entry; duplicate
2637 * entries in the msfr_nsrcs vector are ignored.
2638 * If we encounter an error, rollback transaction.
2640 * XXX This too could be replaced with a set-symmetric
2641 * difference like loop to avoid walking from root
2642 * every time, as the key space is common.
2644 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2645 psin = (struct sockaddr_in *)pkss;
2646 if (psin->sin_family != AF_INET) {
2647 error = EAFNOSUPPORT;
2650 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2654 error = imf_get_source(imf, psin, &lims);
2657 lims->imsl_st[1] = imf->imf_st[1];
2663 goto out_imf_rollback;
2665 INP_WLOCK_ASSERT(inp);
2668 * Begin state merge transaction at IGMP layer.
2670 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2671 IN_MULTI_LIST_LOCK();
2672 error = inm_merge(inm, imf);
2674 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2675 IN_MULTI_LIST_UNLOCK();
2676 goto out_imf_rollback;
2679 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2680 error = igmp_change_state(inm);
2681 IN_MULTI_LIST_UNLOCK();
2683 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2700 * Set the IP multicast options in response to user setsockopt().
2702 * Many of the socket options handled in this function duplicate the
2703 * functionality of socket options in the regular unicast API. However,
2704 * it is not possible to merge the duplicate code, because the idempotence
2705 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2706 * the effects of these options must be treated as separate and distinct.
2708 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2709 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2710 * is refactored to no longer use vifs.
2713 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2715 struct ip_moptions *imo;
2721 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2722 * or is a divert socket, reject it.
2724 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2725 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2726 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2727 return (EOPNOTSUPP);
2729 switch (sopt->sopt_name) {
2730 case IP_MULTICAST_VIF: {
2733 * Select a multicast VIF for transmission.
2734 * Only useful if multicast forwarding is active.
2736 if (legal_vif_num == NULL) {
2740 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2743 if (!legal_vif_num(vifi) && (vifi != -1)) {
2747 imo = inp_findmoptions(inp);
2748 imo->imo_multicast_vif = vifi;
2753 case IP_MULTICAST_IF:
2754 error = inp_set_multicast_if(inp, sopt);
2757 case IP_MULTICAST_TTL: {
2761 * Set the IP time-to-live for outgoing multicast packets.
2762 * The original multicast API required a char argument,
2763 * which is inconsistent with the rest of the socket API.
2764 * We allow either a char or an int.
2766 if (sopt->sopt_valsize == sizeof(u_char)) {
2767 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2774 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2784 imo = inp_findmoptions(inp);
2785 imo->imo_multicast_ttl = ttl;
2790 case IP_MULTICAST_LOOP: {
2794 * Set the loopback flag for outgoing multicast packets.
2795 * Must be zero or one. The original multicast API required a
2796 * char argument, which is inconsistent with the rest
2797 * of the socket API. We allow either a char or an int.
2799 if (sopt->sopt_valsize == sizeof(u_char)) {
2800 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2807 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2811 loop = (u_char)iloop;
2813 imo = inp_findmoptions(inp);
2814 imo->imo_multicast_loop = !!loop;
2819 case IP_ADD_MEMBERSHIP:
2820 case IP_ADD_SOURCE_MEMBERSHIP:
2821 case MCAST_JOIN_GROUP:
2822 case MCAST_JOIN_SOURCE_GROUP:
2823 error = inp_join_group(inp, sopt);
2826 case IP_DROP_MEMBERSHIP:
2827 case IP_DROP_SOURCE_MEMBERSHIP:
2828 case MCAST_LEAVE_GROUP:
2829 case MCAST_LEAVE_SOURCE_GROUP:
2830 error = inp_leave_group(inp, sopt);
2833 case IP_BLOCK_SOURCE:
2834 case IP_UNBLOCK_SOURCE:
2835 case MCAST_BLOCK_SOURCE:
2836 case MCAST_UNBLOCK_SOURCE:
2837 error = inp_block_unblock_source(inp, sopt);
2841 error = inp_set_source_filters(inp, sopt);
2849 INP_UNLOCK_ASSERT(inp);
2855 * Expose IGMP's multicast filter mode and source list(s) to userland,
2856 * keyed by (ifindex, group).
2857 * The filter mode is written out as a uint32_t, followed by
2858 * 0..n of struct in_addr.
2859 * For use by ifmcstat(8).
2860 * SMPng: NOTE: unlocked read of ifindex space.
2863 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2865 struct in_addr src, group;
2866 struct epoch_tracker et;
2868 struct ifmultiaddr *ifma;
2869 struct in_multi *inm;
2870 struct ip_msource *ims;
2874 uint32_t fmode, ifindex;
2879 if (req->newptr != NULL)
2885 group.s_addr = name[1];
2886 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2887 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2888 __func__, ntohl(group.s_addr));
2893 NET_EPOCH_ENTER(et);
2894 ifp = ifnet_byindex(ifindex);
2897 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2902 retval = sysctl_wire_old_buffer(req,
2903 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2909 IN_MULTI_LIST_LOCK();
2911 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2912 if (ifma->ifma_addr->sa_family != AF_INET ||
2913 ifma->ifma_protospec == NULL)
2915 inm = (struct in_multi *)ifma->ifma_protospec;
2916 if (!in_hosteq(inm->inm_addr, group))
2918 fmode = inm->inm_st[1].iss_fmode;
2919 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2922 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2923 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2926 * Only copy-out sources which are in-mode.
2928 if (fmode != ims_get_mode(inm, ims, 1)) {
2929 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2933 src.s_addr = htonl(ims->ims_haddr);
2934 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2940 IN_MULTI_LIST_UNLOCK();
2946 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2948 static const char *inm_modestrs[] = {
2949 [MCAST_UNDEFINED] = "un",
2950 [MCAST_INCLUDE] = "in",
2951 [MCAST_EXCLUDE] = "ex",
2953 _Static_assert(MCAST_UNDEFINED == 0 &&
2954 MCAST_EXCLUDE + 1 == nitems(inm_modestrs),
2955 "inm_modestrs: no longer matches #defines");
2958 inm_mode_str(const int mode)
2961 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2962 return (inm_modestrs[mode]);
2966 static const char *inm_statestrs[] = {
2967 [IGMP_NOT_MEMBER] = "not-member",
2968 [IGMP_SILENT_MEMBER] = "silent",
2969 [IGMP_REPORTING_MEMBER] = "reporting",
2970 [IGMP_IDLE_MEMBER] = "idle",
2971 [IGMP_LAZY_MEMBER] = "lazy",
2972 [IGMP_SLEEPING_MEMBER] = "sleeping",
2973 [IGMP_AWAKENING_MEMBER] = "awakening",
2974 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending",
2975 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending",
2976 [IGMP_LEAVING_MEMBER] = "leaving",
2978 _Static_assert(IGMP_NOT_MEMBER == 0 &&
2979 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs),
2980 "inm_statetrs: no longer matches #defines");
2983 inm_state_str(const int state)
2986 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2987 return (inm_statestrs[state]);
2992 * Dump an in_multi structure to the console.
2995 inm_print(const struct in_multi *inm)
2998 char addrbuf[INET_ADDRSTRLEN];
3000 if ((ktr_mask & KTR_IGMPV3) == 0)
3003 printf("%s: --- begin inm %p ---\n", __func__, inm);
3004 printf("addr %s ifp %p(%s) ifma %p\n",
3005 inet_ntoa_r(inm->inm_addr, addrbuf),
3007 inm->inm_ifp->if_xname,
3009 printf("timer %u state %s refcount %u scq.len %u\n",
3011 inm_state_str(inm->inm_state),
3013 inm->inm_scq.mq_len);
3014 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3019 for (t = 0; t < 2; t++) {
3020 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3021 inm_mode_str(inm->inm_st[t].iss_fmode),
3022 inm->inm_st[t].iss_asm,
3023 inm->inm_st[t].iss_ex,
3024 inm->inm_st[t].iss_in,
3025 inm->inm_st[t].iss_rec);
3027 printf("%s: --- end inm %p ---\n", __func__, inm);
3030 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3033 inm_print(const struct in_multi *inm)
3038 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3040 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);