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/rmlock.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/protosw.h>
51 #include <sys/sysctl.h>
53 #include <sys/taskqueue.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/route/nhop.h>
63 #include <net/ethernet.h>
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_fib.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/igmp_var.h>
74 #define KTR_IGMPV3 KTR_INET
77 #ifndef __SOCKUNION_DECLARED
79 struct sockaddr_storage ss;
81 struct sockaddr_dl sdl;
82 struct sockaddr_in sin;
84 typedef union sockunion sockunion_t;
85 #define __SOCKUNION_DECLARED
86 #endif /* __SOCKUNION_DECLARED */
88 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
89 "IPv4 multicast PCB-layer source filter");
90 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
91 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
92 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
93 "IPv4 multicast IGMP-layer source filter");
98 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK,
99 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
100 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
101 * it can be taken by code in net/if.c also.
102 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
104 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
105 * any need for in_multi itself to be virtualized -- it is bound to an ifp
106 * anyway no matter what happens.
108 struct mtx in_multi_list_mtx;
109 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
111 struct mtx in_multi_free_mtx;
112 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
114 struct sx in_multi_sx;
115 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
120 * Functions with non-static linkage defined in this file should be
121 * declared in in_var.h:
124 * in_joingroup_locked()
126 * in_leavegroup_locked()
132 static void imf_commit(struct in_mfilter *);
133 static int imf_get_source(struct in_mfilter *imf,
134 const struct sockaddr_in *psin,
135 struct in_msource **);
136 static struct in_msource *
137 imf_graft(struct in_mfilter *, const uint8_t,
138 const struct sockaddr_in *);
139 static void imf_leave(struct in_mfilter *);
140 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
141 static void imf_purge(struct in_mfilter *);
142 static void imf_rollback(struct in_mfilter *);
143 static void imf_reap(struct in_mfilter *);
144 static struct in_mfilter *
145 imo_match_group(const struct ip_moptions *,
146 const struct ifnet *, const struct sockaddr *);
147 static struct in_msource *
148 imo_match_source(struct in_mfilter *, const struct sockaddr *);
149 static void ims_merge(struct ip_msource *ims,
150 const struct in_msource *lims, const int rollback);
151 static int in_getmulti(struct ifnet *, const struct in_addr *,
153 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
154 const int noalloc, struct ip_msource **pims);
156 static int inm_is_ifp_detached(const struct in_multi *);
158 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
159 static void inm_purge(struct in_multi *);
160 static void inm_reap(struct in_multi *);
161 static void inm_release(struct in_multi *);
162 static struct ip_moptions *
163 inp_findmoptions(struct inpcb *);
164 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
165 static int inp_join_group(struct inpcb *, struct sockopt *);
166 static int inp_leave_group(struct inpcb *, struct sockopt *);
167 static struct ifnet *
168 inp_lookup_mcast_ifp(const struct inpcb *,
169 const struct sockaddr_in *, const struct in_addr);
170 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
171 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
172 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
173 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
175 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast,
176 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
179 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
180 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
181 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
182 "Max source filters per group");
184 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
185 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
186 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
187 "Max source filters per socket");
189 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
190 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
191 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
193 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
194 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
195 "Per-interface stack-wide source filters");
199 * Inline function which wraps assertions for a valid ifp.
200 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
204 inm_is_ifp_detached(const struct in_multi *inm)
208 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
209 ifp = inm->inm_ifma->ifma_ifp;
212 * Sanity check that netinet's notion of ifp is the
215 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
218 return (ifp == NULL);
223 * Interface detach can happen in a taskqueue thread context, so we must use a
224 * dedicated thread to avoid deadlocks when draining inm_release tasks.
226 TASKQUEUE_DEFINE_THREAD(inm_free);
227 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER();
228 static void inm_release_task(void *arg __unused, int pending __unused);
229 static struct task inm_free_task = TASK_INITIALIZER(0, inm_release_task, NULL);
232 inm_release_wait(void *arg __unused)
236 * Make sure all pending multicast addresses are freed before
237 * the VNET or network device is destroyed:
239 taskqueue_drain(taskqueue_inm_free, &inm_free_task);
242 /* XXX-BZ FIXME, see D24914. */
243 VNET_SYSUNINIT(inm_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, inm_release_wait, NULL);
247 inm_release_list_deferred(struct in_multi_head *inmh)
250 if (SLIST_EMPTY(inmh))
252 mtx_lock(&in_multi_free_mtx);
253 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
254 mtx_unlock(&in_multi_free_mtx);
255 taskqueue_enqueue(taskqueue_inm_free, &inm_free_task);
259 inm_disconnect(struct in_multi *inm)
262 struct ifmultiaddr *ifma, *ll_ifma;
265 IF_ADDR_WLOCK_ASSERT(ifp);
266 ifma = inm->inm_ifma;
269 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
270 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
271 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
273 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
274 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
275 MPASS(ifma != ll_ifma);
276 ifma->ifma_llifma = NULL;
277 MPASS(ll_ifma->ifma_llifma == NULL);
278 MPASS(ll_ifma->ifma_ifp == ifp);
279 if (--ll_ifma->ifma_refcount == 0) {
280 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
281 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
282 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
284 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
285 if_freemulti(ll_ifma);
292 inm_release_deferred(struct in_multi *inm)
294 struct in_multi_head tmp;
296 IN_MULTI_LIST_LOCK_ASSERT();
297 MPASS(inm->inm_refcount > 0);
298 if (--inm->inm_refcount == 0) {
301 inm->inm_ifma->ifma_protospec = NULL;
302 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
303 inm_release_list_deferred(&tmp);
308 inm_release_task(void *arg __unused, int pending __unused)
310 struct in_multi_head inm_free_tmp;
311 struct in_multi *inm, *tinm;
313 SLIST_INIT(&inm_free_tmp);
314 mtx_lock(&in_multi_free_mtx);
315 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
316 mtx_unlock(&in_multi_free_mtx);
318 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
319 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
327 * Initialize an in_mfilter structure to a known state at t0, t1
328 * with an empty source filter list.
331 imf_init(struct in_mfilter *imf, const int st0, const int st1)
333 memset(imf, 0, sizeof(struct in_mfilter));
334 RB_INIT(&imf->imf_sources);
335 imf->imf_st[0] = st0;
336 imf->imf_st[1] = st1;
340 ip_mfilter_alloc(const int mflags, const int st0, const int st1)
342 struct in_mfilter *imf;
344 imf = malloc(sizeof(*imf), M_INMFILTER, mflags);
346 imf_init(imf, st0, st1);
352 ip_mfilter_free(struct in_mfilter *imf)
356 free(imf, M_INMFILTER);
360 * Function for looking up an in_multi record for an IPv4 multicast address
361 * on a given interface. ifp must be valid. If no record found, return NULL.
362 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
365 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
367 struct ifmultiaddr *ifma;
368 struct in_multi *inm;
370 IN_MULTI_LIST_LOCK_ASSERT();
371 IF_ADDR_LOCK_ASSERT(ifp);
374 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
375 if (ifma->ifma_addr->sa_family != AF_INET ||
376 ifma->ifma_protospec == NULL)
378 inm = (struct in_multi *)ifma->ifma_protospec;
379 if (inm->inm_addr.s_addr == ina.s_addr)
387 * Wrapper for inm_lookup_locked().
388 * The IF_ADDR_LOCK will be taken on ifp and released on return.
391 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
393 struct epoch_tracker et;
394 struct in_multi *inm;
396 IN_MULTI_LIST_LOCK_ASSERT();
399 inm = inm_lookup_locked(ifp, ina);
406 * Find an IPv4 multicast group entry for this ip_moptions instance
407 * which matches the specified group, and optionally an interface.
408 * Return its index into the array, or -1 if not found.
410 static struct in_mfilter *
411 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
412 const struct sockaddr *group)
414 const struct sockaddr_in *gsin;
415 struct in_mfilter *imf;
416 struct in_multi *inm;
418 gsin = (const struct sockaddr_in *)group;
420 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
424 if ((ifp == NULL || (inm->inm_ifp == ifp)) &&
425 in_hosteq(inm->inm_addr, gsin->sin_addr)) {
433 * Find an IPv4 multicast source entry for this imo which matches
434 * the given group index for this socket, and source address.
436 * NOTE: This does not check if the entry is in-mode, merely if
437 * it exists, which may not be the desired behaviour.
439 static struct in_msource *
440 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src)
442 struct ip_msource find;
443 struct ip_msource *ims;
444 const sockunion_t *psa;
446 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
448 /* Source trees are keyed in host byte order. */
449 psa = (const sockunion_t *)src;
450 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
451 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
453 return ((struct in_msource *)ims);
457 * Perform filtering for multicast datagrams on a socket by group and source.
459 * Returns 0 if a datagram should be allowed through, or various error codes
460 * if the socket was not a member of the group, or the source was muted, etc.
463 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
464 const struct sockaddr *group, const struct sockaddr *src)
466 struct in_mfilter *imf;
467 struct in_msource *ims;
470 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
472 imf = imo_match_group(imo, ifp, group);
474 return (MCAST_NOTGMEMBER);
477 * Check if the source was included in an (S,G) join.
478 * Allow reception on exclusive memberships by default,
479 * reject reception on inclusive memberships by default.
480 * Exclude source only if an in-mode exclude filter exists.
481 * Include source only if an in-mode include filter exists.
482 * NOTE: We are comparing group state here at IGMP t1 (now)
483 * with socket-layer t0 (since last downcall).
485 mode = imf->imf_st[1];
486 ims = imo_match_source(imf, src);
488 if ((ims == NULL && mode == MCAST_INCLUDE) ||
489 (ims != NULL && ims->imsl_st[0] != mode))
490 return (MCAST_NOTSMEMBER);
496 * Find and return a reference to an in_multi record for (ifp, group),
497 * and bump its reference count.
498 * If one does not exist, try to allocate it, and update link-layer multicast
499 * filters on ifp to listen for group.
500 * Assumes the IN_MULTI lock is held across the call.
501 * Return 0 if successful, otherwise return an appropriate error code.
504 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
505 struct in_multi **pinm)
507 struct sockaddr_in gsin;
508 struct ifmultiaddr *ifma;
509 struct in_ifinfo *ii;
510 struct in_multi *inm;
513 IN_MULTI_LOCK_ASSERT();
515 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
516 IN_MULTI_LIST_LOCK();
517 inm = inm_lookup(ifp, *group);
520 * If we already joined this group, just bump the
521 * refcount and return it.
523 KASSERT(inm->inm_refcount >= 1,
524 ("%s: bad refcount %d", __func__, inm->inm_refcount));
525 inm_acquire_locked(inm);
528 IN_MULTI_LIST_UNLOCK();
532 memset(&gsin, 0, sizeof(gsin));
533 gsin.sin_family = AF_INET;
534 gsin.sin_len = sizeof(struct sockaddr_in);
535 gsin.sin_addr = *group;
538 * Check if a link-layer group is already associated
539 * with this network-layer group on the given ifnet.
541 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
545 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
546 IN_MULTI_LIST_LOCK();
550 * If something other than netinet is occupying the link-layer
551 * group, print a meaningful error message and back out of
553 * Otherwise, bump the refcount on the existing network-layer
554 * group association and return it.
556 if (ifma->ifma_protospec != NULL) {
557 inm = (struct in_multi *)ifma->ifma_protospec;
559 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
561 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
562 ("%s: ifma not AF_INET", __func__));
563 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
564 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
565 !in_hosteq(inm->inm_addr, *group)) {
566 char addrbuf[INET_ADDRSTRLEN];
568 panic("%s: ifma %p is inconsistent with %p (%s)",
569 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
572 inm_acquire_locked(inm);
577 IF_ADDR_WLOCK_ASSERT(ifp);
580 * A new in_multi record is needed; allocate and initialize it.
581 * We DO NOT perform an IGMP join as the in_ layer may need to
582 * push an initial source list down to IGMP to support SSM.
584 * The initial source filter state is INCLUDE, {} as per the RFC.
586 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
588 IF_ADDR_WUNLOCK(ifp);
589 IN_MULTI_LIST_UNLOCK();
590 if_delmulti_ifma(ifma);
593 inm->inm_addr = *group;
595 inm->inm_igi = ii->ii_igmp;
596 inm->inm_ifma = ifma;
597 inm->inm_refcount = 1;
598 inm->inm_state = IGMP_NOT_MEMBER;
599 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
600 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
601 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
602 RB_INIT(&inm->inm_srcs);
604 ifma->ifma_protospec = inm;
608 IF_ADDR_WUNLOCK(ifp);
609 IN_MULTI_LIST_UNLOCK();
614 * Drop a reference to an in_multi record.
616 * If the refcount drops to 0, free the in_multi record and
617 * delete the underlying link-layer membership.
620 inm_release(struct in_multi *inm)
622 struct ifmultiaddr *ifma;
625 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
626 MPASS(inm->inm_refcount == 0);
627 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
629 ifma = inm->inm_ifma;
632 /* XXX this access is not covered by IF_ADDR_LOCK */
633 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
635 CURVNET_SET(ifp->if_vnet);
637 free(inm, M_IPMADDR);
638 if_delmulti_ifma_flags(ifma, 1);
643 free(inm, M_IPMADDR);
644 if_delmulti_ifma_flags(ifma, 1);
649 * Clear recorded source entries for a group.
650 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
651 * FIXME: Should reap.
654 inm_clear_recorded(struct in_multi *inm)
656 struct ip_msource *ims;
658 IN_MULTI_LIST_LOCK_ASSERT();
660 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
663 --inm->inm_st[1].iss_rec;
666 KASSERT(inm->inm_st[1].iss_rec == 0,
667 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
671 * Record a source as pending for a Source-Group IGMPv3 query.
672 * This lives here as it modifies the shared tree.
674 * inm is the group descriptor.
675 * naddr is the address of the source to record in network-byte order.
677 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
678 * lazy-allocate a source node in response to an SG query.
679 * Otherwise, no allocation is performed. This saves some memory
680 * with the trade-off that the source will not be reported to the
681 * router if joined in the window between the query response and
682 * the group actually being joined on the local host.
684 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
685 * This turns off the allocation of a recorded source entry if
686 * the group has not been joined.
688 * Return 0 if the source didn't exist or was already marked as recorded.
689 * Return 1 if the source was marked as recorded by this function.
690 * Return <0 if any error occurred (negated errno code).
693 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
695 struct ip_msource find;
696 struct ip_msource *ims, *nims;
698 IN_MULTI_LIST_LOCK_ASSERT();
700 find.ims_haddr = ntohl(naddr);
701 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
702 if (ims && ims->ims_stp)
705 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
707 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
711 nims->ims_haddr = find.ims_haddr;
712 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
718 * Mark the source as recorded and update the recorded
722 ++inm->inm_st[1].iss_rec;
728 * Return a pointer to an in_msource owned by an in_mfilter,
729 * given its source address.
730 * Lazy-allocate if needed. If this is a new entry its filter state is
733 * imf is the filter set being modified.
734 * haddr is the source address in *host* byte-order.
736 * SMPng: May be called with locks held; malloc must not block.
739 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
740 struct in_msource **plims)
742 struct ip_msource find;
743 struct ip_msource *ims, *nims;
744 struct in_msource *lims;
751 /* key is host byte order */
752 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
753 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
754 lims = (struct in_msource *)ims;
756 if (imf->imf_nsrc == in_mcast_maxsocksrc)
758 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
762 lims = (struct in_msource *)nims;
763 lims->ims_haddr = find.ims_haddr;
764 lims->imsl_st[0] = MCAST_UNDEFINED;
765 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
775 * Graft a source entry into an existing socket-layer filter set,
776 * maintaining any required invariants and checking allocations.
778 * The source is marked as being in the new filter mode at t1.
780 * Return the pointer to the new node, otherwise return NULL.
782 static struct in_msource *
783 imf_graft(struct in_mfilter *imf, const uint8_t st1,
784 const struct sockaddr_in *psin)
786 struct ip_msource *nims;
787 struct in_msource *lims;
789 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
793 lims = (struct in_msource *)nims;
794 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
795 lims->imsl_st[0] = MCAST_UNDEFINED;
796 lims->imsl_st[1] = st1;
797 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
804 * Prune a source entry from an existing socket-layer filter set,
805 * maintaining any required invariants and checking allocations.
807 * The source is marked as being left at t1, it is not freed.
809 * Return 0 if no error occurred, otherwise return an errno value.
812 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
814 struct ip_msource find;
815 struct ip_msource *ims;
816 struct in_msource *lims;
818 /* key is host byte order */
819 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
820 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
823 lims = (struct in_msource *)ims;
824 lims->imsl_st[1] = MCAST_UNDEFINED;
829 * Revert socket-layer filter set deltas at t1 to t0 state.
832 imf_rollback(struct in_mfilter *imf)
834 struct ip_msource *ims, *tims;
835 struct in_msource *lims;
837 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
838 lims = (struct in_msource *)ims;
839 if (lims->imsl_st[0] == lims->imsl_st[1]) {
840 /* no change at t1 */
842 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
843 /* revert change to existing source at t1 */
844 lims->imsl_st[1] = lims->imsl_st[0];
846 /* revert source added t1 */
847 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
848 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
849 free(ims, M_INMFILTER);
853 imf->imf_st[1] = imf->imf_st[0];
857 * Mark socket-layer filter set as INCLUDE {} at t1.
860 imf_leave(struct in_mfilter *imf)
862 struct ip_msource *ims;
863 struct in_msource *lims;
865 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
866 lims = (struct in_msource *)ims;
867 lims->imsl_st[1] = MCAST_UNDEFINED;
869 imf->imf_st[1] = MCAST_INCLUDE;
873 * Mark socket-layer filter set deltas as committed.
876 imf_commit(struct in_mfilter *imf)
878 struct ip_msource *ims;
879 struct in_msource *lims;
881 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
882 lims = (struct in_msource *)ims;
883 lims->imsl_st[0] = lims->imsl_st[1];
885 imf->imf_st[0] = imf->imf_st[1];
889 * Reap unreferenced sources from socket-layer filter set.
892 imf_reap(struct in_mfilter *imf)
894 struct ip_msource *ims, *tims;
895 struct in_msource *lims;
897 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
898 lims = (struct in_msource *)ims;
899 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
900 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
901 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
902 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
903 free(ims, M_INMFILTER);
910 * Purge socket-layer filter set.
913 imf_purge(struct in_mfilter *imf)
915 struct ip_msource *ims, *tims;
917 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
918 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
919 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
920 free(ims, M_INMFILTER);
923 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
924 KASSERT(RB_EMPTY(&imf->imf_sources),
925 ("%s: imf_sources not empty", __func__));
929 * Look up a source filter entry for a multicast group.
931 * inm is the group descriptor to work with.
932 * haddr is the host-byte-order IPv4 address to look up.
933 * noalloc may be non-zero to suppress allocation of sources.
934 * *pims will be set to the address of the retrieved or allocated source.
936 * SMPng: NOTE: may be called with locks held.
937 * Return 0 if successful, otherwise return a non-zero error code.
940 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
941 const int noalloc, struct ip_msource **pims)
943 struct ip_msource find;
944 struct ip_msource *ims, *nims;
946 find.ims_haddr = haddr;
947 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
948 if (ims == NULL && !noalloc) {
949 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
951 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
955 nims->ims_haddr = haddr;
956 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
960 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
970 * Merge socket-layer source into IGMP-layer source.
971 * If rollback is non-zero, perform the inverse of the merge.
974 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
977 int n = rollback ? -1 : 1;
979 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
980 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
981 __func__, n, ims->ims_haddr);
982 ims->ims_st[1].ex -= n;
983 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
984 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
985 __func__, n, ims->ims_haddr);
986 ims->ims_st[1].in -= n;
989 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
990 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
991 __func__, n, ims->ims_haddr);
992 ims->ims_st[1].ex += n;
993 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
994 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
995 __func__, n, ims->ims_haddr);
996 ims->ims_st[1].in += n;
1001 * Atomically update the global in_multi state, when a membership's
1002 * filter list is being updated in any way.
1004 * imf is the per-inpcb-membership group filter pointer.
1005 * A fake imf may be passed for in-kernel consumers.
1007 * XXX This is a candidate for a set-symmetric-difference style loop
1008 * which would eliminate the repeated lookup from root of ims nodes,
1009 * as they share the same key space.
1011 * If any error occurred this function will back out of refcounts
1012 * and return a non-zero value.
1015 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1017 struct ip_msource *ims, *nims;
1018 struct in_msource *lims;
1019 int schanged, error;
1025 IN_MULTI_LIST_LOCK_ASSERT();
1028 * Update the source filters first, as this may fail.
1029 * Maintain count of in-mode filters at t0, t1. These are
1030 * used to work out if we transition into ASM mode or not.
1031 * Maintain a count of source filters whose state was
1032 * actually modified by this operation.
1034 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1035 lims = (struct in_msource *)ims;
1036 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1037 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1038 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1039 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1043 ims_merge(nims, lims, 0);
1046 struct ip_msource *bims;
1048 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1049 lims = (struct in_msource *)ims;
1050 if (lims->imsl_st[0] == lims->imsl_st[1])
1052 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1055 ims_merge(bims, lims, 1);
1060 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1061 __func__, nsrc0, nsrc1);
1063 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1064 if (imf->imf_st[0] == imf->imf_st[1] &&
1065 imf->imf_st[1] == MCAST_INCLUDE) {
1067 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1068 --inm->inm_st[1].iss_in;
1072 /* Handle filter mode transition on socket. */
1073 if (imf->imf_st[0] != imf->imf_st[1]) {
1074 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1075 __func__, imf->imf_st[0], imf->imf_st[1]);
1077 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1078 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1079 --inm->inm_st[1].iss_ex;
1080 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1081 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1082 --inm->inm_st[1].iss_in;
1085 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1086 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1087 inm->inm_st[1].iss_ex++;
1088 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1089 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1090 inm->inm_st[1].iss_in++;
1095 * Track inm filter state in terms of listener counts.
1096 * If there are any exclusive listeners, stack-wide
1097 * membership is exclusive.
1098 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1099 * If no listeners remain, state is undefined at t1,
1100 * and the IGMP lifecycle for this group should finish.
1102 if (inm->inm_st[1].iss_ex > 0) {
1103 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1104 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1105 } else if (inm->inm_st[1].iss_in > 0) {
1106 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1107 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1109 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1110 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1113 /* Decrement ASM listener count on transition out of ASM mode. */
1114 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1115 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1116 (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;
1122 /* Increment ASM listener count on transition to ASM mode. */
1123 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1124 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1125 inm->inm_st[1].iss_asm++;
1128 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1133 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1140 * Mark an in_multi's filter set deltas as committed.
1141 * Called by IGMP after a state change has been enqueued.
1144 inm_commit(struct in_multi *inm)
1146 struct ip_msource *ims;
1148 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1149 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1152 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1153 ims->ims_st[0] = ims->ims_st[1];
1155 inm->inm_st[0] = inm->inm_st[1];
1159 * Reap unreferenced nodes from an in_multi's filter set.
1162 inm_reap(struct in_multi *inm)
1164 struct ip_msource *ims, *tims;
1166 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1167 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1168 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1171 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1172 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1173 free(ims, M_IPMSOURCE);
1179 * Purge all source nodes from an in_multi's filter set.
1182 inm_purge(struct in_multi *inm)
1184 struct ip_msource *ims, *tims;
1186 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1187 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1188 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1189 free(ims, M_IPMSOURCE);
1195 * Join a multicast group; unlocked entry point.
1197 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1198 * is not held. Fortunately, ifp is unlikely to have been detached
1199 * at this point, so we assume it's OK to recurse.
1202 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1203 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1208 error = in_joingroup_locked(ifp, gina, imf, pinm);
1215 * Join a multicast group; real entry point.
1217 * Only preserves atomicity at inm level.
1218 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1220 * If the IGMP downcall fails, the group is not joined, and an error
1224 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1225 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1227 struct in_mfilter timf;
1228 struct in_multi *inm;
1231 IN_MULTI_LOCK_ASSERT();
1232 IN_MULTI_LIST_UNLOCK_ASSERT();
1234 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1235 ntohl(gina->s_addr), ifp, ifp->if_xname);
1241 * If no imf was specified (i.e. kernel consumer),
1242 * fake one up and assume it is an ASM join.
1245 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1249 error = in_getmulti(ifp, gina, &inm);
1251 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1254 IN_MULTI_LIST_LOCK();
1255 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1256 error = inm_merge(inm, imf);
1258 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1259 goto out_inm_release;
1262 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1263 error = igmp_change_state(inm);
1265 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1266 goto out_inm_release;
1271 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1273 inm_release_deferred(inm);
1274 IF_ADDR_WUNLOCK(ifp);
1278 IN_MULTI_LIST_UNLOCK();
1284 * Leave a multicast group; unlocked entry point.
1287 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1292 error = in_leavegroup_locked(inm, imf);
1299 * Leave a multicast group; real entry point.
1300 * All source filters will be expunged.
1302 * Only preserves atomicity at inm level.
1304 * Holding the write lock for the INP which contains imf
1305 * is highly advisable. We can't assert for it as imf does not
1306 * contain a back-pointer to the owning inp.
1308 * Note: This is not the same as inm_release(*) as this function also
1309 * makes a state change downcall into IGMP.
1312 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1314 struct in_mfilter timf;
1317 IN_MULTI_LOCK_ASSERT();
1318 IN_MULTI_LIST_UNLOCK_ASSERT();
1322 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1323 inm, ntohl(inm->inm_addr.s_addr),
1324 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1328 * If no imf was specified (i.e. kernel consumer),
1329 * fake one up and assume it is an ASM join.
1332 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1337 * Begin state merge transaction at IGMP layer.
1339 * As this particular invocation should not cause any memory
1340 * to be allocated, and there is no opportunity to roll back
1341 * the transaction, it MUST NOT fail.
1343 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1344 IN_MULTI_LIST_LOCK();
1345 error = inm_merge(inm, imf);
1346 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1348 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1349 CURVNET_SET(inm->inm_ifp->if_vnet);
1350 error = igmp_change_state(inm);
1351 IF_ADDR_WLOCK(inm->inm_ifp);
1352 inm_release_deferred(inm);
1353 IF_ADDR_WUNLOCK(inm->inm_ifp);
1354 IN_MULTI_LIST_UNLOCK();
1357 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1359 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1364 /*#ifndef BURN_BRIDGES*/
1367 * Block or unblock an ASM multicast source on an inpcb.
1368 * This implements the delta-based API described in RFC 3678.
1370 * The delta-based API applies only to exclusive-mode memberships.
1371 * An IGMP downcall will be performed.
1373 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1375 * Return 0 if successful, otherwise return an appropriate error code.
1378 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1380 struct group_source_req gsr;
1381 struct rm_priotracker in_ifa_tracker;
1382 sockunion_t *gsa, *ssa;
1384 struct in_mfilter *imf;
1385 struct ip_moptions *imo;
1386 struct in_msource *ims;
1387 struct in_multi *inm;
1395 memset(&gsr, 0, sizeof(struct group_source_req));
1396 gsa = (sockunion_t *)&gsr.gsr_group;
1397 ssa = (sockunion_t *)&gsr.gsr_source;
1399 switch (sopt->sopt_name) {
1400 case IP_BLOCK_SOURCE:
1401 case IP_UNBLOCK_SOURCE: {
1402 struct ip_mreq_source mreqs;
1404 error = sooptcopyin(sopt, &mreqs,
1405 sizeof(struct ip_mreq_source),
1406 sizeof(struct ip_mreq_source));
1410 gsa->sin.sin_family = AF_INET;
1411 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1412 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1414 ssa->sin.sin_family = AF_INET;
1415 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1416 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1418 if (!in_nullhost(mreqs.imr_interface)) {
1419 IN_IFADDR_RLOCK(&in_ifa_tracker);
1420 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1421 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
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 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1448 return (EADDRNOTAVAIL);
1450 ifp = ifnet_byindex(gsr.gsr_interface);
1452 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1457 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1458 __func__, sopt->sopt_name);
1459 return (EOPNOTSUPP);
1463 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1469 * Check if we are actually a member of this group.
1471 imo = inp_findmoptions(inp);
1472 imf = imo_match_group(imo, ifp, &gsa->sa);
1474 error = EADDRNOTAVAIL;
1475 goto out_inp_locked;
1480 * Attempting to use the delta-based API on an
1481 * non exclusive-mode membership is an error.
1483 fmode = imf->imf_st[0];
1484 if (fmode != MCAST_EXCLUDE) {
1486 goto out_inp_locked;
1490 * Deal with error cases up-front:
1491 * Asked to block, but already blocked; or
1492 * Asked to unblock, but nothing to unblock.
1493 * If adding a new block entry, allocate it.
1495 ims = imo_match_source(imf, &ssa->sa);
1496 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1497 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1498 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1499 error = EADDRNOTAVAIL;
1500 goto out_inp_locked;
1503 INP_WLOCK_ASSERT(inp);
1506 * Begin state merge transaction at socket layer.
1509 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1510 ims = imf_graft(imf, fmode, &ssa->sin);
1514 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1515 error = imf_prune(imf, &ssa->sin);
1519 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1520 goto out_imf_rollback;
1524 * Begin state merge transaction at IGMP layer.
1526 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1527 IN_MULTI_LIST_LOCK();
1528 error = inm_merge(inm, imf);
1530 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1531 IN_MULTI_LIST_UNLOCK();
1532 goto out_imf_rollback;
1535 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1536 error = igmp_change_state(inm);
1537 IN_MULTI_LIST_UNLOCK();
1539 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1556 * Given an inpcb, return its multicast options structure pointer. Accepts
1557 * an unlocked inpcb pointer, but will return it locked. May sleep.
1559 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1560 * SMPng: NOTE: Returns with the INP write lock held.
1562 static struct ip_moptions *
1563 inp_findmoptions(struct inpcb *inp)
1565 struct ip_moptions *imo;
1568 if (inp->inp_moptions != NULL)
1569 return (inp->inp_moptions);
1573 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1575 imo->imo_multicast_ifp = NULL;
1576 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1577 imo->imo_multicast_vif = -1;
1578 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1579 imo->imo_multicast_loop = in_mcast_loop;
1580 STAILQ_INIT(&imo->imo_head);
1583 if (inp->inp_moptions != NULL) {
1584 free(imo, M_IPMOPTS);
1585 return (inp->inp_moptions);
1587 inp->inp_moptions = imo;
1592 inp_gcmoptions(struct ip_moptions *imo)
1594 struct in_mfilter *imf;
1595 struct in_multi *inm;
1598 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1599 ip_mfilter_remove(&imo->imo_head, imf);
1602 if ((inm = imf->imf_inm) != NULL) {
1603 if ((ifp = inm->inm_ifp) != NULL) {
1604 CURVNET_SET(ifp->if_vnet);
1605 (void)in_leavegroup(inm, imf);
1608 (void)in_leavegroup(inm, imf);
1611 ip_mfilter_free(imf);
1613 free(imo, M_IPMOPTS);
1617 * Discard the IP multicast options (and source filters). To minimize
1618 * the amount of work done while holding locks such as the INP's
1619 * pcbinfo lock (which is used in the receive path), the free
1620 * operation is deferred to the epoch callback task.
1623 inp_freemoptions(struct ip_moptions *imo)
1627 inp_gcmoptions(imo);
1631 * Atomically get source filters on a socket for an IPv4 multicast group.
1632 * Called with INP lock held; returns with lock released.
1635 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1637 struct __msfilterreq msfr;
1640 struct ip_moptions *imo;
1641 struct in_mfilter *imf;
1642 struct ip_msource *ims;
1643 struct in_msource *lims;
1644 struct sockaddr_in *psin;
1645 struct sockaddr_storage *ptss;
1646 struct sockaddr_storage *tss;
1648 size_t nsrcs, ncsrcs;
1650 INP_WLOCK_ASSERT(inp);
1652 imo = inp->inp_moptions;
1653 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1657 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1658 sizeof(struct __msfilterreq));
1662 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1665 ifp = ifnet_byindex(msfr.msfr_ifindex);
1672 * Lookup group on the socket.
1674 gsa = (sockunion_t *)&msfr.msfr_group;
1675 imf = imo_match_group(imo, ifp, &gsa->sa);
1678 return (EADDRNOTAVAIL);
1682 * Ignore memberships which are in limbo.
1684 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1688 msfr.msfr_fmode = imf->imf_st[1];
1691 * If the user specified a buffer, copy out the source filter
1692 * entries to userland gracefully.
1693 * We only copy out the number of entries which userland
1694 * has asked for, but we always tell userland how big the
1695 * buffer really needs to be.
1697 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1698 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1700 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1701 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1702 M_TEMP, M_NOWAIT | M_ZERO);
1710 * Count number of sources in-mode at t0.
1711 * If buffer space exists and remains, copy out source entries.
1713 nsrcs = msfr.msfr_nsrcs;
1716 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1717 lims = (struct in_msource *)ims;
1718 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1719 lims->imsl_st[0] != imf->imf_st[0])
1722 if (tss != NULL && nsrcs > 0) {
1723 psin = (struct sockaddr_in *)ptss;
1724 psin->sin_family = AF_INET;
1725 psin->sin_len = sizeof(struct sockaddr_in);
1726 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1736 error = copyout(tss, msfr.msfr_srcs,
1737 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1743 msfr.msfr_nsrcs = ncsrcs;
1744 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1750 * Return the IP multicast options in response to user getsockopt().
1753 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1755 struct rm_priotracker in_ifa_tracker;
1756 struct ip_mreqn mreqn;
1757 struct ip_moptions *imo;
1759 struct in_ifaddr *ia;
1764 imo = inp->inp_moptions;
1766 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1767 * or is a divert socket, reject it.
1769 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1770 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1771 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1773 return (EOPNOTSUPP);
1777 switch (sopt->sopt_name) {
1778 case IP_MULTICAST_VIF:
1780 optval = imo->imo_multicast_vif;
1784 error = sooptcopyout(sopt, &optval, sizeof(int));
1787 case IP_MULTICAST_IF:
1788 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1790 ifp = imo->imo_multicast_ifp;
1791 if (!in_nullhost(imo->imo_multicast_addr)) {
1792 mreqn.imr_address = imo->imo_multicast_addr;
1793 } else if (ifp != NULL) {
1794 struct epoch_tracker et;
1796 mreqn.imr_ifindex = ifp->if_index;
1797 NET_EPOCH_ENTER(et);
1798 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1801 IA_SIN(ia)->sin_addr;
1806 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1807 error = sooptcopyout(sopt, &mreqn,
1808 sizeof(struct ip_mreqn));
1810 error = sooptcopyout(sopt, &mreqn.imr_address,
1811 sizeof(struct in_addr));
1815 case IP_MULTICAST_TTL:
1817 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1819 optval = coptval = imo->imo_multicast_ttl;
1821 if (sopt->sopt_valsize == sizeof(u_char))
1822 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1824 error = sooptcopyout(sopt, &optval, sizeof(int));
1827 case IP_MULTICAST_LOOP:
1829 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1831 optval = coptval = imo->imo_multicast_loop;
1833 if (sopt->sopt_valsize == sizeof(u_char))
1834 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1836 error = sooptcopyout(sopt, &optval, sizeof(int));
1841 error = EADDRNOTAVAIL;
1844 error = inp_get_source_filters(inp, sopt);
1850 error = ENOPROTOOPT;
1854 INP_UNLOCK_ASSERT(inp);
1860 * Look up the ifnet to use for a multicast group membership,
1861 * given the IPv4 address of an interface, and the IPv4 group address.
1863 * This routine exists to support legacy multicast applications
1864 * which do not understand that multicast memberships are scoped to
1865 * specific physical links in the networking stack, or which need
1866 * to join link-scope groups before IPv4 addresses are configured.
1868 * Use this socket's current FIB number for any required FIB lookup.
1869 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1870 * and use its ifp; usually, this points to the default next-hop.
1872 * If the FIB lookup fails, attempt to use the first non-loopback
1873 * interface with multicast capability in the system as a
1874 * last resort. The legacy IPv4 ASM API requires that we do
1875 * this in order to allow groups to be joined when the routing
1876 * table has not yet been populated during boot.
1878 * Returns NULL if no ifp could be found, otherwise return referenced ifp.
1880 * FUTURE: Implement IPv4 source-address selection.
1882 static struct ifnet *
1883 inp_lookup_mcast_ifp(const struct inpcb *inp,
1884 const struct sockaddr_in *gsin, const struct in_addr ina)
1886 struct rm_priotracker in_ifa_tracker;
1888 struct nhop_object *nh;
1890 KASSERT(inp != NULL, ("%s: inp must not be NULL", __func__));
1891 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1892 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1893 ("%s: not multicast", __func__));
1896 if (!in_nullhost(ina)) {
1897 IN_IFADDR_RLOCK(&in_ifa_tracker);
1898 INADDR_TO_IFP(ina, ifp);
1901 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1903 nh = fib4_lookup(inp->inp_inc.inc_fibnum, gsin->sin_addr, 0, NHR_NONE, 0);
1908 struct in_ifaddr *ia;
1912 IN_IFADDR_RLOCK(&in_ifa_tracker);
1913 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1915 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1916 (mifp->if_flags & IFF_MULTICAST)) {
1922 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1930 * Join an IPv4 multicast group, possibly with a source.
1933 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1935 struct group_source_req gsr;
1936 sockunion_t *gsa, *ssa;
1938 struct in_mfilter *imf;
1939 struct ip_moptions *imo;
1940 struct in_multi *inm;
1941 struct in_msource *lims;
1942 struct epoch_tracker et;
1949 memset(&gsr, 0, sizeof(struct group_source_req));
1950 gsa = (sockunion_t *)&gsr.gsr_group;
1951 gsa->ss.ss_family = AF_UNSPEC;
1952 ssa = (sockunion_t *)&gsr.gsr_source;
1953 ssa->ss.ss_family = AF_UNSPEC;
1955 switch (sopt->sopt_name) {
1956 case IP_ADD_MEMBERSHIP: {
1957 struct ip_mreqn mreqn;
1959 if (sopt->sopt_valsize == sizeof(struct ip_mreqn))
1960 error = sooptcopyin(sopt, &mreqn,
1961 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn));
1963 error = sooptcopyin(sopt, &mreqn,
1964 sizeof(struct ip_mreq), sizeof(struct ip_mreq));
1968 gsa->sin.sin_family = AF_INET;
1969 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1970 gsa->sin.sin_addr = mreqn.imr_multiaddr;
1971 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1974 NET_EPOCH_ENTER(et);
1975 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) &&
1976 mreqn.imr_ifindex != 0)
1977 ifp = ifnet_byindex_ref(mreqn.imr_ifindex);
1979 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1984 case IP_ADD_SOURCE_MEMBERSHIP: {
1985 struct ip_mreq_source mreqs;
1987 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source),
1988 sizeof(struct ip_mreq_source));
1992 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET;
1993 gsa->sin.sin_len = ssa->sin.sin_len =
1994 sizeof(struct sockaddr_in);
1996 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1997 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2000 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2002 NET_EPOCH_ENTER(et);
2003 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2004 mreqs.imr_interface);
2006 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2007 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2011 case MCAST_JOIN_GROUP:
2012 case MCAST_JOIN_SOURCE_GROUP:
2013 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2014 error = sooptcopyin(sopt, &gsr,
2015 sizeof(struct group_req),
2016 sizeof(struct group_req));
2017 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2018 error = sooptcopyin(sopt, &gsr,
2019 sizeof(struct group_source_req),
2020 sizeof(struct group_source_req));
2025 if (gsa->sin.sin_family != AF_INET ||
2026 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2030 * Overwrite the port field if present, as the sockaddr
2031 * being copied in may be matched with a binary comparison.
2033 gsa->sin.sin_port = 0;
2034 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2035 if (ssa->sin.sin_family != AF_INET ||
2036 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2038 ssa->sin.sin_port = 0;
2041 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2044 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2045 return (EADDRNOTAVAIL);
2046 NET_EPOCH_ENTER(et);
2047 ifp = ifnet_byindex_ref(gsr.gsr_interface);
2052 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2053 __func__, sopt->sopt_name);
2054 return (EOPNOTSUPP);
2058 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2061 return (EADDRNOTAVAIL);
2067 * Find the membership in the membership list.
2069 imo = inp_findmoptions(inp);
2070 imf = imo_match_group(imo, ifp, &gsa->sa);
2075 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) {
2077 goto out_inp_locked;
2083 if (ssa->ss.ss_family != AF_UNSPEC) {
2085 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2086 * is an error. On an existing inclusive membership,
2087 * it just adds the source to the filter list.
2089 if (imf->imf_st[1] != MCAST_INCLUDE) {
2091 goto out_inp_locked;
2094 * Throw out duplicates.
2096 * XXX FIXME: This makes a naive assumption that
2097 * even if entries exist for *ssa in this imf,
2098 * they will be rejected as dupes, even if they
2099 * are not valid in the current mode (in-mode).
2101 * in_msource is transactioned just as for anything
2102 * else in SSM -- but note naive use of inm_graft()
2103 * below for allocating new filter entries.
2105 * This is only an issue if someone mixes the
2106 * full-state SSM API with the delta-based API,
2107 * which is discouraged in the relevant RFCs.
2109 lims = imo_match_source(imf, &ssa->sa);
2110 if (lims != NULL /*&&
2111 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2112 error = EADDRNOTAVAIL;
2113 goto out_inp_locked;
2117 * MCAST_JOIN_GROUP on an existing exclusive
2118 * membership is an error; return EADDRINUSE
2119 * to preserve 4.4BSD API idempotence, and
2120 * avoid tedious detour to code below.
2121 * NOTE: This is bending RFC 3678 a bit.
2123 * On an existing inclusive membership, this is also
2124 * an error; if you want to change filter mode,
2125 * you must use the userland API setsourcefilter().
2126 * XXX We don't reject this for imf in UNDEFINED
2127 * state at t1, because allocation of a filter
2128 * is atomic with allocation of a membership.
2131 if (imf->imf_st[1] == MCAST_EXCLUDE)
2133 goto out_inp_locked;
2138 * Begin state merge transaction at socket layer.
2140 INP_WLOCK_ASSERT(inp);
2143 * Graft new source into filter list for this inpcb's
2144 * membership of the group. The in_multi may not have
2145 * been allocated yet if this is a new membership, however,
2146 * the in_mfilter slot will be allocated and must be initialized.
2148 * Note: Grafting of exclusive mode filters doesn't happen
2150 * XXX: Should check for non-NULL lims (node exists but may
2151 * not be in-mode) for interop with full-state API.
2153 if (ssa->ss.ss_family != AF_UNSPEC) {
2154 /* Membership starts in IN mode */
2156 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2157 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2160 goto out_inp_locked;
2163 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2165 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2167 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2170 goto out_inp_locked;
2173 /* No address specified; Membership starts in EX mode */
2175 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2176 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2179 goto out_inp_locked;
2185 * Begin state merge transaction at IGMP layer.
2191 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2195 if (in_pcbrele_wlocked(inp)) {
2197 goto out_inp_unlocked;
2200 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2202 goto out_inp_locked;
2205 * NOTE: Refcount from in_joingroup_locked()
2206 * is protecting membership.
2208 ip_mfilter_insert(&imo->imo_head, imf);
2210 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2211 IN_MULTI_LIST_LOCK();
2212 error = inm_merge(inm, imf);
2214 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2216 IN_MULTI_LIST_UNLOCK();
2219 goto out_inp_locked;
2221 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2222 error = igmp_change_state(inm);
2223 IN_MULTI_LIST_UNLOCK();
2225 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2229 goto out_inp_locked;
2241 if (is_new && imf) {
2242 if (imf->imf_inm != NULL) {
2243 IN_MULTI_LIST_LOCK();
2245 inm_release_deferred(imf->imf_inm);
2246 IF_ADDR_WUNLOCK(ifp);
2247 IN_MULTI_LIST_UNLOCK();
2249 ip_mfilter_free(imf);
2256 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2259 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2261 struct group_source_req gsr;
2262 struct ip_mreq_source mreqs;
2263 struct rm_priotracker in_ifa_tracker;
2264 sockunion_t *gsa, *ssa;
2266 struct in_mfilter *imf;
2267 struct ip_moptions *imo;
2268 struct in_msource *ims;
2269 struct in_multi *inm;
2277 memset(&gsr, 0, sizeof(struct group_source_req));
2278 gsa = (sockunion_t *)&gsr.gsr_group;
2279 gsa->ss.ss_family = AF_UNSPEC;
2280 ssa = (sockunion_t *)&gsr.gsr_source;
2281 ssa->ss.ss_family = AF_UNSPEC;
2283 switch (sopt->sopt_name) {
2284 case IP_DROP_MEMBERSHIP:
2285 case IP_DROP_SOURCE_MEMBERSHIP:
2286 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2287 error = sooptcopyin(sopt, &mreqs,
2288 sizeof(struct ip_mreq),
2289 sizeof(struct ip_mreq));
2291 * Swap interface and sourceaddr arguments,
2292 * as ip_mreq and ip_mreq_source are laid
2295 mreqs.imr_interface = mreqs.imr_sourceaddr;
2296 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2297 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2298 error = sooptcopyin(sopt, &mreqs,
2299 sizeof(struct ip_mreq_source),
2300 sizeof(struct ip_mreq_source));
2305 gsa->sin.sin_family = AF_INET;
2306 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2307 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2309 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2310 ssa->sin.sin_family = AF_INET;
2311 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2312 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2316 * Attempt to look up hinted ifp from interface address.
2317 * Fallthrough with null ifp iff lookup fails, to
2318 * preserve 4.4BSD mcast API idempotence.
2319 * XXX NOTE WELL: The RFC 3678 API is preferred because
2320 * using an IPv4 address as a key is racy.
2322 if (!in_nullhost(mreqs.imr_interface)) {
2323 IN_IFADDR_RLOCK(&in_ifa_tracker);
2324 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2325 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2327 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2328 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2332 case MCAST_LEAVE_GROUP:
2333 case MCAST_LEAVE_SOURCE_GROUP:
2334 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2335 error = sooptcopyin(sopt, &gsr,
2336 sizeof(struct group_req),
2337 sizeof(struct group_req));
2338 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2339 error = sooptcopyin(sopt, &gsr,
2340 sizeof(struct group_source_req),
2341 sizeof(struct group_source_req));
2346 if (gsa->sin.sin_family != AF_INET ||
2347 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2350 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2351 if (ssa->sin.sin_family != AF_INET ||
2352 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2356 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2357 return (EADDRNOTAVAIL);
2359 ifp = ifnet_byindex(gsr.gsr_interface);
2362 return (EADDRNOTAVAIL);
2366 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2367 __func__, sopt->sopt_name);
2368 return (EOPNOTSUPP);
2372 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2378 * Find the membership in the membership list.
2380 imo = inp_findmoptions(inp);
2381 imf = imo_match_group(imo, ifp, &gsa->sa);
2383 error = EADDRNOTAVAIL;
2384 goto out_inp_locked;
2388 if (ssa->ss.ss_family != AF_UNSPEC)
2392 * Begin state merge transaction at socket layer.
2394 INP_WLOCK_ASSERT(inp);
2397 * If we were instructed only to leave a given source, do so.
2398 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2401 ip_mfilter_remove(&imo->imo_head, imf);
2405 * Give up the multicast address record to which
2406 * the membership points.
2408 (void) in_leavegroup_locked(imf->imf_inm, imf);
2410 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2411 error = EADDRNOTAVAIL;
2412 goto out_inp_locked;
2414 ims = imo_match_source(imf, &ssa->sa);
2416 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2417 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2418 error = EADDRNOTAVAIL;
2419 goto out_inp_locked;
2421 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2422 error = imf_prune(imf, &ssa->sin);
2424 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2426 goto out_inp_locked;
2431 * Begin state merge transaction at IGMP layer.
2434 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2435 IN_MULTI_LIST_LOCK();
2436 error = inm_merge(inm, imf);
2438 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2440 IN_MULTI_LIST_UNLOCK();
2443 goto out_inp_locked;
2446 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2447 error = igmp_change_state(inm);
2448 IN_MULTI_LIST_UNLOCK();
2450 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2454 goto out_inp_locked;
2463 if (is_final && imf)
2464 ip_mfilter_free(imf);
2471 * Select the interface for transmitting IPv4 multicast datagrams.
2473 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2474 * may be passed to this socket option. An address of INADDR_ANY or an
2475 * interface index of 0 is used to remove a previous selection.
2476 * When no interface is selected, one is chosen for every send.
2479 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2481 struct rm_priotracker in_ifa_tracker;
2482 struct in_addr addr;
2483 struct ip_mreqn mreqn;
2485 struct ip_moptions *imo;
2488 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2490 * An interface index was specified using the
2491 * Linux-derived ip_mreqn structure.
2493 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2494 sizeof(struct ip_mreqn));
2498 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2501 if (mreqn.imr_ifindex == 0) {
2504 ifp = ifnet_byindex(mreqn.imr_ifindex);
2506 return (EADDRNOTAVAIL);
2510 * An interface was specified by IPv4 address.
2511 * This is the traditional BSD usage.
2513 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2514 sizeof(struct in_addr));
2517 if (in_nullhost(addr)) {
2520 IN_IFADDR_RLOCK(&in_ifa_tracker);
2521 INADDR_TO_IFP(addr, ifp);
2522 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2524 return (EADDRNOTAVAIL);
2526 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2527 ntohl(addr.s_addr));
2530 /* Reject interfaces which do not support multicast. */
2531 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2532 return (EOPNOTSUPP);
2534 imo = inp_findmoptions(inp);
2535 imo->imo_multicast_ifp = ifp;
2536 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2543 * Atomically set source filters on a socket for an IPv4 multicast group.
2545 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2548 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2550 struct __msfilterreq msfr;
2553 struct in_mfilter *imf;
2554 struct ip_moptions *imo;
2555 struct in_multi *inm;
2558 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2559 sizeof(struct __msfilterreq));
2563 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2566 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2567 msfr.msfr_fmode != MCAST_INCLUDE))
2570 if (msfr.msfr_group.ss_family != AF_INET ||
2571 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2574 gsa = (sockunion_t *)&msfr.msfr_group;
2575 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2578 gsa->sin.sin_port = 0; /* ignore port */
2580 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2581 return (EADDRNOTAVAIL);
2583 ifp = ifnet_byindex(msfr.msfr_ifindex);
2585 return (EADDRNOTAVAIL);
2590 * Take the INP write lock.
2591 * Check if this socket is a member of this group.
2593 imo = inp_findmoptions(inp);
2594 imf = imo_match_group(imo, ifp, &gsa->sa);
2596 error = EADDRNOTAVAIL;
2597 goto out_inp_locked;
2602 * Begin state merge transaction at socket layer.
2604 INP_WLOCK_ASSERT(inp);
2606 imf->imf_st[1] = msfr.msfr_fmode;
2609 * Apply any new source filters, if present.
2610 * Make a copy of the user-space source vector so
2611 * that we may copy them with a single copyin. This
2612 * allows us to deal with page faults up-front.
2614 if (msfr.msfr_nsrcs > 0) {
2615 struct in_msource *lims;
2616 struct sockaddr_in *psin;
2617 struct sockaddr_storage *kss, *pkss;
2622 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2623 __func__, (unsigned long)msfr.msfr_nsrcs);
2624 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2626 error = copyin(msfr.msfr_srcs, kss,
2627 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2636 * Mark all source filters as UNDEFINED at t1.
2637 * Restore new group filter mode, as imf_leave()
2638 * will set it to INCLUDE.
2641 imf->imf_st[1] = msfr.msfr_fmode;
2644 * Update socket layer filters at t1, lazy-allocating
2645 * new entries. This saves a bunch of memory at the
2646 * cost of one RB_FIND() per source entry; duplicate
2647 * entries in the msfr_nsrcs vector are ignored.
2648 * If we encounter an error, rollback transaction.
2650 * XXX This too could be replaced with a set-symmetric
2651 * difference like loop to avoid walking from root
2652 * every time, as the key space is common.
2654 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2655 psin = (struct sockaddr_in *)pkss;
2656 if (psin->sin_family != AF_INET) {
2657 error = EAFNOSUPPORT;
2660 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2664 error = imf_get_source(imf, psin, &lims);
2667 lims->imsl_st[1] = imf->imf_st[1];
2673 goto out_imf_rollback;
2675 INP_WLOCK_ASSERT(inp);
2678 * Begin state merge transaction at IGMP layer.
2680 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2681 IN_MULTI_LIST_LOCK();
2682 error = inm_merge(inm, imf);
2684 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2685 IN_MULTI_LIST_UNLOCK();
2686 goto out_imf_rollback;
2689 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2690 error = igmp_change_state(inm);
2691 IN_MULTI_LIST_UNLOCK();
2693 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2710 * Set the IP multicast options in response to user setsockopt().
2712 * Many of the socket options handled in this function duplicate the
2713 * functionality of socket options in the regular unicast API. However,
2714 * it is not possible to merge the duplicate code, because the idempotence
2715 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2716 * the effects of these options must be treated as separate and distinct.
2718 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2719 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2720 * is refactored to no longer use vifs.
2723 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2725 struct ip_moptions *imo;
2731 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2732 * or is a divert socket, reject it.
2734 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2735 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2736 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2737 return (EOPNOTSUPP);
2739 switch (sopt->sopt_name) {
2740 case IP_MULTICAST_VIF: {
2743 * Select a multicast VIF for transmission.
2744 * Only useful if multicast forwarding is active.
2746 if (legal_vif_num == NULL) {
2750 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2753 if (!legal_vif_num(vifi) && (vifi != -1)) {
2757 imo = inp_findmoptions(inp);
2758 imo->imo_multicast_vif = vifi;
2763 case IP_MULTICAST_IF:
2764 error = inp_set_multicast_if(inp, sopt);
2767 case IP_MULTICAST_TTL: {
2771 * Set the IP time-to-live for outgoing multicast packets.
2772 * The original multicast API required a char argument,
2773 * which is inconsistent with the rest of the socket API.
2774 * We allow either a char or an int.
2776 if (sopt->sopt_valsize == sizeof(u_char)) {
2777 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2784 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2794 imo = inp_findmoptions(inp);
2795 imo->imo_multicast_ttl = ttl;
2800 case IP_MULTICAST_LOOP: {
2804 * Set the loopback flag for outgoing multicast packets.
2805 * Must be zero or one. The original multicast API required a
2806 * char argument, which is inconsistent with the rest
2807 * of the socket API. We allow either a char or an int.
2809 if (sopt->sopt_valsize == sizeof(u_char)) {
2810 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2817 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2821 loop = (u_char)iloop;
2823 imo = inp_findmoptions(inp);
2824 imo->imo_multicast_loop = !!loop;
2829 case IP_ADD_MEMBERSHIP:
2830 case IP_ADD_SOURCE_MEMBERSHIP:
2831 case MCAST_JOIN_GROUP:
2832 case MCAST_JOIN_SOURCE_GROUP:
2833 error = inp_join_group(inp, sopt);
2836 case IP_DROP_MEMBERSHIP:
2837 case IP_DROP_SOURCE_MEMBERSHIP:
2838 case MCAST_LEAVE_GROUP:
2839 case MCAST_LEAVE_SOURCE_GROUP:
2840 error = inp_leave_group(inp, sopt);
2843 case IP_BLOCK_SOURCE:
2844 case IP_UNBLOCK_SOURCE:
2845 case MCAST_BLOCK_SOURCE:
2846 case MCAST_UNBLOCK_SOURCE:
2847 error = inp_block_unblock_source(inp, sopt);
2851 error = inp_set_source_filters(inp, sopt);
2859 INP_UNLOCK_ASSERT(inp);
2865 * Expose IGMP's multicast filter mode and source list(s) to userland,
2866 * keyed by (ifindex, group).
2867 * The filter mode is written out as a uint32_t, followed by
2868 * 0..n of struct in_addr.
2869 * For use by ifmcstat(8).
2870 * SMPng: NOTE: unlocked read of ifindex space.
2873 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2875 struct in_addr src, group;
2876 struct epoch_tracker et;
2878 struct ifmultiaddr *ifma;
2879 struct in_multi *inm;
2880 struct ip_msource *ims;
2884 uint32_t fmode, ifindex;
2889 if (req->newptr != NULL)
2896 if (ifindex <= 0 || ifindex > V_if_index) {
2897 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2902 group.s_addr = name[1];
2903 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2904 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2905 __func__, ntohl(group.s_addr));
2909 NET_EPOCH_ENTER(et);
2910 ifp = ifnet_byindex(ifindex);
2913 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2918 retval = sysctl_wire_old_buffer(req,
2919 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2925 IN_MULTI_LIST_LOCK();
2927 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2928 if (ifma->ifma_addr->sa_family != AF_INET ||
2929 ifma->ifma_protospec == NULL)
2931 inm = (struct in_multi *)ifma->ifma_protospec;
2932 if (!in_hosteq(inm->inm_addr, group))
2934 fmode = inm->inm_st[1].iss_fmode;
2935 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2938 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2939 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2942 * Only copy-out sources which are in-mode.
2944 if (fmode != ims_get_mode(inm, ims, 1)) {
2945 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2949 src.s_addr = htonl(ims->ims_haddr);
2950 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2956 IN_MULTI_LIST_UNLOCK();
2962 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2964 static const char *inm_modestrs[] = {
2965 [MCAST_UNDEFINED] = "un",
2966 [MCAST_INCLUDE] = "in",
2967 [MCAST_EXCLUDE] = "ex",
2969 _Static_assert(MCAST_UNDEFINED == 0 &&
2970 MCAST_EXCLUDE + 1 == nitems(inm_modestrs),
2971 "inm_modestrs: no longer matches #defines");
2974 inm_mode_str(const int mode)
2977 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2978 return (inm_modestrs[mode]);
2982 static const char *inm_statestrs[] = {
2983 [IGMP_NOT_MEMBER] = "not-member",
2984 [IGMP_SILENT_MEMBER] = "silent",
2985 [IGMP_REPORTING_MEMBER] = "reporting",
2986 [IGMP_IDLE_MEMBER] = "idle",
2987 [IGMP_LAZY_MEMBER] = "lazy",
2988 [IGMP_SLEEPING_MEMBER] = "sleeping",
2989 [IGMP_AWAKENING_MEMBER] = "awakening",
2990 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending",
2991 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending",
2992 [IGMP_LEAVING_MEMBER] = "leaving",
2994 _Static_assert(IGMP_NOT_MEMBER == 0 &&
2995 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs),
2996 "inm_statetrs: no longer matches #defines");
2999 inm_state_str(const int state)
3002 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3003 return (inm_statestrs[state]);
3008 * Dump an in_multi structure to the console.
3011 inm_print(const struct in_multi *inm)
3014 char addrbuf[INET_ADDRSTRLEN];
3016 if ((ktr_mask & KTR_IGMPV3) == 0)
3019 printf("%s: --- begin inm %p ---\n", __func__, inm);
3020 printf("addr %s ifp %p(%s) ifma %p\n",
3021 inet_ntoa_r(inm->inm_addr, addrbuf),
3023 inm->inm_ifp->if_xname,
3025 printf("timer %u state %s refcount %u scq.len %u\n",
3027 inm_state_str(inm->inm_state),
3029 inm->inm_scq.mq_len);
3030 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3035 for (t = 0; t < 2; t++) {
3036 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3037 inm_mode_str(inm->inm_st[t].iss_fmode),
3038 inm->inm_st[t].iss_asm,
3039 inm->inm_st[t].iss_ex,
3040 inm->inm_st[t].iss_in,
3041 inm->inm_st[t].iss_rec);
3043 printf("%s: --- end inm %p ---\n", __func__, inm);
3046 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3049 inm_print(const struct in_multi *inm)
3054 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3056 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);