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>
54 #include <sys/gtaskqueue.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.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");
97 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
98 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
99 * it can be taken by code in net/if.c also.
100 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
102 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
103 * any need for in_multi itself to be virtualized -- it is bound to an ifp
104 * anyway no matter what happens.
106 struct mtx in_multi_list_mtx;
107 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
109 struct mtx in_multi_free_mtx;
110 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
112 struct sx in_multi_sx;
113 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
118 * Functions with non-static linkage defined in this file should be
119 * declared in in_var.h:
124 * in_joingroup_locked()
126 * in_leavegroup_locked()
132 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
135 static void imf_commit(struct in_mfilter *);
136 static int imf_get_source(struct in_mfilter *imf,
137 const struct sockaddr_in *psin,
138 struct in_msource **);
139 static struct in_msource *
140 imf_graft(struct in_mfilter *, const uint8_t,
141 const struct sockaddr_in *);
142 static void imf_leave(struct in_mfilter *);
143 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
144 static void imf_purge(struct in_mfilter *);
145 static void imf_rollback(struct in_mfilter *);
146 static void imf_reap(struct in_mfilter *);
147 static int imo_grow(struct ip_moptions *);
148 static size_t imo_match_group(const struct ip_moptions *,
149 const struct ifnet *, const struct sockaddr *);
150 static struct in_msource *
151 imo_match_source(const struct ip_moptions *, const size_t,
152 const struct sockaddr *);
153 static void ims_merge(struct ip_msource *ims,
154 const struct in_msource *lims, const int rollback);
155 static int in_getmulti(struct ifnet *, const struct in_addr *,
157 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
158 const int noalloc, struct ip_msource **pims);
160 static int inm_is_ifp_detached(const struct in_multi *);
162 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
163 static void inm_purge(struct in_multi *);
164 static void inm_reap(struct in_multi *);
165 static void inm_release(struct in_multi *);
166 static struct ip_moptions *
167 inp_findmoptions(struct inpcb *);
168 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
169 static int inp_join_group(struct inpcb *, struct sockopt *);
170 static int inp_leave_group(struct inpcb *, struct sockopt *);
171 static struct ifnet *
172 inp_lookup_mcast_ifp(const struct inpcb *,
173 const struct sockaddr_in *, const struct in_addr);
174 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
175 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
176 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
177 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
179 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
182 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
183 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
184 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
185 "Max source filters per group");
187 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
188 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
189 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
190 "Max source filters per socket");
192 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
193 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
194 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
196 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
197 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
198 "Per-interface stack-wide source filters");
202 * Inline function which wraps assertions for a valid ifp.
203 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
207 inm_is_ifp_detached(const struct in_multi *inm)
211 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
212 ifp = inm->inm_ifma->ifma_ifp;
215 * Sanity check that netinet's notion of ifp is the
218 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
221 return (ifp == NULL);
225 static struct grouptask free_gtask;
226 static struct in_multi_head inm_free_list;
227 static void inm_release_task(void *arg __unused);
228 static void inm_init(void)
230 SLIST_INIT(&inm_free_list);
231 taskqgroup_config_gtask_init(NULL, &free_gtask, inm_release_task, "inm release task");
234 #ifdef EARLY_AP_STARTUP
235 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
238 SYSINIT(inm_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_FIRST,
244 inm_release_list_deferred(struct in_multi_head *inmh)
247 if (SLIST_EMPTY(inmh))
249 mtx_lock(&in_multi_free_mtx);
250 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
251 mtx_unlock(&in_multi_free_mtx);
252 GROUPTASK_ENQUEUE(&free_gtask);
256 inm_disconnect(struct in_multi *inm)
259 struct ifmultiaddr *ifma, *ll_ifma;
262 IF_ADDR_WLOCK_ASSERT(ifp);
263 ifma = inm->inm_ifma;
266 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
267 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
268 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
270 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
271 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
272 MPASS(ifma != ll_ifma);
273 ifma->ifma_llifma = NULL;
274 MPASS(ll_ifma->ifma_llifma == NULL);
275 MPASS(ll_ifma->ifma_ifp == ifp);
276 if (--ll_ifma->ifma_refcount == 0) {
277 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
278 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
279 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
281 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
282 if_freemulti(ll_ifma);
289 inm_release_deferred(struct in_multi *inm)
291 struct in_multi_head tmp;
293 IN_MULTI_LIST_LOCK_ASSERT();
294 MPASS(inm->inm_refcount > 0);
295 if (--inm->inm_refcount == 0) {
298 inm->inm_ifma->ifma_protospec = NULL;
299 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
300 inm_release_list_deferred(&tmp);
305 inm_release_task(void *arg __unused)
307 struct in_multi_head inm_free_tmp;
308 struct in_multi *inm, *tinm;
310 SLIST_INIT(&inm_free_tmp);
311 mtx_lock(&in_multi_free_mtx);
312 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
313 mtx_unlock(&in_multi_free_mtx);
315 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
316 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
324 * Initialize an in_mfilter structure to a known state at t0, t1
325 * with an empty source filter list.
328 imf_init(struct in_mfilter *imf, const int st0, const int st1)
330 memset(imf, 0, sizeof(struct in_mfilter));
331 RB_INIT(&imf->imf_sources);
332 imf->imf_st[0] = st0;
333 imf->imf_st[1] = st1;
337 * Function for looking up an in_multi record for an IPv4 multicast address
338 * on a given interface. ifp must be valid. If no record found, return NULL.
339 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
342 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
344 struct ifmultiaddr *ifma;
345 struct in_multi *inm;
347 IN_MULTI_LIST_LOCK_ASSERT();
348 IF_ADDR_LOCK_ASSERT(ifp);
351 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
352 if (ifma->ifma_addr->sa_family != AF_INET ||
353 ifma->ifma_protospec == NULL)
355 inm = (struct in_multi *)ifma->ifma_protospec;
356 if (inm->inm_addr.s_addr == ina.s_addr)
364 * Wrapper for inm_lookup_locked().
365 * The IF_ADDR_LOCK will be taken on ifp and released on return.
368 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
370 struct epoch_tracker et;
371 struct in_multi *inm;
373 IN_MULTI_LIST_LOCK_ASSERT();
375 inm = inm_lookup_locked(ifp, ina);
382 * Resize the ip_moptions vector to the next power-of-two minus 1.
383 * May be called with locks held; do not sleep.
386 imo_grow(struct ip_moptions *imo)
388 struct in_multi **nmships;
389 struct in_multi **omships;
390 struct in_mfilter *nmfilters;
391 struct in_mfilter *omfilters;
398 omships = imo->imo_membership;
399 omfilters = imo->imo_mfilters;
400 oldmax = imo->imo_max_memberships;
401 newmax = ((oldmax + 1) * 2) - 1;
403 if (newmax <= IP_MAX_MEMBERSHIPS) {
404 nmships = (struct in_multi **)realloc(omships,
405 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
406 nmfilters = (struct in_mfilter *)realloc(omfilters,
407 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
408 if (nmships != NULL && nmfilters != NULL) {
409 /* Initialize newly allocated source filter heads. */
410 for (idx = oldmax; idx < newmax; idx++) {
411 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
414 imo->imo_max_memberships = newmax;
415 imo->imo_membership = nmships;
416 imo->imo_mfilters = nmfilters;
420 if (nmships == NULL || nmfilters == NULL) {
422 free(nmships, M_IPMOPTS);
423 if (nmfilters != NULL)
424 free(nmfilters, M_INMFILTER);
425 return (ETOOMANYREFS);
432 * Find an IPv4 multicast group entry for this ip_moptions instance
433 * which matches the specified group, and optionally an interface.
434 * Return its index into the array, or -1 if not found.
437 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
438 const struct sockaddr *group)
440 const struct sockaddr_in *gsin;
441 struct in_multi **pinm;
445 gsin = (const struct sockaddr_in *)group;
447 /* The imo_membership array may be lazy allocated. */
448 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
451 nmships = imo->imo_num_memberships;
452 pinm = &imo->imo_membership[0];
453 for (idx = 0; idx < nmships; idx++, pinm++) {
456 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
457 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
468 * Find an IPv4 multicast source entry for this imo which matches
469 * the given group index for this socket, and source address.
471 * NOTE: This does not check if the entry is in-mode, merely if
472 * it exists, which may not be the desired behaviour.
474 static struct in_msource *
475 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
476 const struct sockaddr *src)
478 struct ip_msource find;
479 struct in_mfilter *imf;
480 struct ip_msource *ims;
481 const sockunion_t *psa;
483 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
484 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
485 ("%s: invalid index %d\n", __func__, (int)gidx));
487 /* The imo_mfilters array may be lazy allocated. */
488 if (imo->imo_mfilters == NULL)
490 imf = &imo->imo_mfilters[gidx];
492 /* Source trees are keyed in host byte order. */
493 psa = (const sockunion_t *)src;
494 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
495 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
497 return ((struct in_msource *)ims);
501 * Perform filtering for multicast datagrams on a socket by group and source.
503 * Returns 0 if a datagram should be allowed through, or various error codes
504 * if the socket was not a member of the group, or the source was muted, etc.
507 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
508 const struct sockaddr *group, const struct sockaddr *src)
511 struct in_msource *ims;
514 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
516 gidx = imo_match_group(imo, ifp, group);
518 return (MCAST_NOTGMEMBER);
521 * Check if the source was included in an (S,G) join.
522 * Allow reception on exclusive memberships by default,
523 * reject reception on inclusive memberships by default.
524 * Exclude source only if an in-mode exclude filter exists.
525 * Include source only if an in-mode include filter exists.
526 * NOTE: We are comparing group state here at IGMP t1 (now)
527 * with socket-layer t0 (since last downcall).
529 mode = imo->imo_mfilters[gidx].imf_st[1];
530 ims = imo_match_source(imo, gidx, src);
532 if ((ims == NULL && mode == MCAST_INCLUDE) ||
533 (ims != NULL && ims->imsl_st[0] != mode))
534 return (MCAST_NOTSMEMBER);
540 * Find and return a reference to an in_multi record for (ifp, group),
541 * and bump its reference count.
542 * If one does not exist, try to allocate it, and update link-layer multicast
543 * filters on ifp to listen for group.
544 * Assumes the IN_MULTI lock is held across the call.
545 * Return 0 if successful, otherwise return an appropriate error code.
548 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
549 struct in_multi **pinm)
551 struct sockaddr_in gsin;
552 struct ifmultiaddr *ifma;
553 struct in_ifinfo *ii;
554 struct in_multi *inm;
557 IN_MULTI_LOCK_ASSERT();
559 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
560 IN_MULTI_LIST_LOCK();
561 inm = inm_lookup(ifp, *group);
564 * If we already joined this group, just bump the
565 * refcount and return it.
567 KASSERT(inm->inm_refcount >= 1,
568 ("%s: bad refcount %d", __func__, inm->inm_refcount));
569 inm_acquire_locked(inm);
572 IN_MULTI_LIST_UNLOCK();
576 memset(&gsin, 0, sizeof(gsin));
577 gsin.sin_family = AF_INET;
578 gsin.sin_len = sizeof(struct sockaddr_in);
579 gsin.sin_addr = *group;
582 * Check if a link-layer group is already associated
583 * with this network-layer group on the given ifnet.
585 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
589 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
590 IN_MULTI_LIST_LOCK();
594 * If something other than netinet is occupying the link-layer
595 * group, print a meaningful error message and back out of
597 * Otherwise, bump the refcount on the existing network-layer
598 * group association and return it.
600 if (ifma->ifma_protospec != NULL) {
601 inm = (struct in_multi *)ifma->ifma_protospec;
603 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
605 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
606 ("%s: ifma not AF_INET", __func__));
607 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
608 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
609 !in_hosteq(inm->inm_addr, *group)) {
610 char addrbuf[INET_ADDRSTRLEN];
612 panic("%s: ifma %p is inconsistent with %p (%s)",
613 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
616 inm_acquire_locked(inm);
621 IF_ADDR_WLOCK_ASSERT(ifp);
624 * A new in_multi record is needed; allocate and initialize it.
625 * We DO NOT perform an IGMP join as the in_ layer may need to
626 * push an initial source list down to IGMP to support SSM.
628 * The initial source filter state is INCLUDE, {} as per the RFC.
630 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
632 IF_ADDR_WUNLOCK(ifp);
633 IN_MULTI_LIST_UNLOCK();
634 if_delmulti_ifma(ifma);
637 inm->inm_addr = *group;
639 inm->inm_igi = ii->ii_igmp;
640 inm->inm_ifma = ifma;
641 inm->inm_refcount = 1;
642 inm->inm_state = IGMP_NOT_MEMBER;
643 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
644 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
645 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
646 RB_INIT(&inm->inm_srcs);
648 ifma->ifma_protospec = inm;
652 IF_ADDR_WUNLOCK(ifp);
653 IN_MULTI_LIST_UNLOCK();
658 * Drop a reference to an in_multi record.
660 * If the refcount drops to 0, free the in_multi record and
661 * delete the underlying link-layer membership.
664 inm_release(struct in_multi *inm)
666 struct ifmultiaddr *ifma;
669 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
670 MPASS(inm->inm_refcount == 0);
671 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
673 ifma = inm->inm_ifma;
676 /* XXX this access is not covered by IF_ADDR_LOCK */
677 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
679 CURVNET_SET(ifp->if_vnet);
681 free(inm, M_IPMADDR);
682 if_delmulti_ifma_flags(ifma, 1);
687 free(inm, M_IPMADDR);
688 if_delmulti_ifma_flags(ifma, 1);
693 * Clear recorded source entries for a group.
694 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
695 * FIXME: Should reap.
698 inm_clear_recorded(struct in_multi *inm)
700 struct ip_msource *ims;
702 IN_MULTI_LIST_LOCK_ASSERT();
704 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
707 --inm->inm_st[1].iss_rec;
710 KASSERT(inm->inm_st[1].iss_rec == 0,
711 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
715 * Record a source as pending for a Source-Group IGMPv3 query.
716 * This lives here as it modifies the shared tree.
718 * inm is the group descriptor.
719 * naddr is the address of the source to record in network-byte order.
721 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
722 * lazy-allocate a source node in response to an SG query.
723 * Otherwise, no allocation is performed. This saves some memory
724 * with the trade-off that the source will not be reported to the
725 * router if joined in the window between the query response and
726 * the group actually being joined on the local host.
728 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
729 * This turns off the allocation of a recorded source entry if
730 * the group has not been joined.
732 * Return 0 if the source didn't exist or was already marked as recorded.
733 * Return 1 if the source was marked as recorded by this function.
734 * Return <0 if any error occurred (negated errno code).
737 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
739 struct ip_msource find;
740 struct ip_msource *ims, *nims;
742 IN_MULTI_LIST_LOCK_ASSERT();
744 find.ims_haddr = ntohl(naddr);
745 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
746 if (ims && ims->ims_stp)
749 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
751 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
755 nims->ims_haddr = find.ims_haddr;
756 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
762 * Mark the source as recorded and update the recorded
766 ++inm->inm_st[1].iss_rec;
772 * Return a pointer to an in_msource owned by an in_mfilter,
773 * given its source address.
774 * Lazy-allocate if needed. If this is a new entry its filter state is
777 * imf is the filter set being modified.
778 * haddr is the source address in *host* byte-order.
780 * SMPng: May be called with locks held; malloc must not block.
783 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
784 struct in_msource **plims)
786 struct ip_msource find;
787 struct ip_msource *ims, *nims;
788 struct in_msource *lims;
795 /* key is host byte order */
796 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
797 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
798 lims = (struct in_msource *)ims;
800 if (imf->imf_nsrc == in_mcast_maxsocksrc)
802 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
806 lims = (struct in_msource *)nims;
807 lims->ims_haddr = find.ims_haddr;
808 lims->imsl_st[0] = MCAST_UNDEFINED;
809 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
819 * Graft a source entry into an existing socket-layer filter set,
820 * maintaining any required invariants and checking allocations.
822 * The source is marked as being in the new filter mode at t1.
824 * Return the pointer to the new node, otherwise return NULL.
826 static struct in_msource *
827 imf_graft(struct in_mfilter *imf, const uint8_t st1,
828 const struct sockaddr_in *psin)
830 struct ip_msource *nims;
831 struct in_msource *lims;
833 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
837 lims = (struct in_msource *)nims;
838 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
839 lims->imsl_st[0] = MCAST_UNDEFINED;
840 lims->imsl_st[1] = st1;
841 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
848 * Prune a source entry from an existing socket-layer filter set,
849 * maintaining any required invariants and checking allocations.
851 * The source is marked as being left at t1, it is not freed.
853 * Return 0 if no error occurred, otherwise return an errno value.
856 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
858 struct ip_msource find;
859 struct ip_msource *ims;
860 struct in_msource *lims;
862 /* key is host byte order */
863 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
864 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
867 lims = (struct in_msource *)ims;
868 lims->imsl_st[1] = MCAST_UNDEFINED;
873 * Revert socket-layer filter set deltas at t1 to t0 state.
876 imf_rollback(struct in_mfilter *imf)
878 struct ip_msource *ims, *tims;
879 struct in_msource *lims;
881 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
882 lims = (struct in_msource *)ims;
883 if (lims->imsl_st[0] == lims->imsl_st[1]) {
884 /* no change at t1 */
886 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
887 /* revert change to existing source at t1 */
888 lims->imsl_st[1] = lims->imsl_st[0];
890 /* revert source added t1 */
891 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
892 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
893 free(ims, M_INMFILTER);
897 imf->imf_st[1] = imf->imf_st[0];
901 * Mark socket-layer filter set as INCLUDE {} at t1.
904 imf_leave(struct in_mfilter *imf)
906 struct ip_msource *ims;
907 struct in_msource *lims;
909 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
910 lims = (struct in_msource *)ims;
911 lims->imsl_st[1] = MCAST_UNDEFINED;
913 imf->imf_st[1] = MCAST_INCLUDE;
917 * Mark socket-layer filter set deltas as committed.
920 imf_commit(struct in_mfilter *imf)
922 struct ip_msource *ims;
923 struct in_msource *lims;
925 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
926 lims = (struct in_msource *)ims;
927 lims->imsl_st[0] = lims->imsl_st[1];
929 imf->imf_st[0] = imf->imf_st[1];
933 * Reap unreferenced sources from socket-layer filter set.
936 imf_reap(struct in_mfilter *imf)
938 struct ip_msource *ims, *tims;
939 struct in_msource *lims;
941 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
942 lims = (struct in_msource *)ims;
943 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
944 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
945 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
946 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
947 free(ims, M_INMFILTER);
954 * Purge socket-layer filter set.
957 imf_purge(struct in_mfilter *imf)
959 struct ip_msource *ims, *tims;
961 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
962 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
963 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
964 free(ims, M_INMFILTER);
967 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
968 KASSERT(RB_EMPTY(&imf->imf_sources),
969 ("%s: imf_sources not empty", __func__));
973 * Look up a source filter entry for a multicast group.
975 * inm is the group descriptor to work with.
976 * haddr is the host-byte-order IPv4 address to look up.
977 * noalloc may be non-zero to suppress allocation of sources.
978 * *pims will be set to the address of the retrieved or allocated source.
980 * SMPng: NOTE: may be called with locks held.
981 * Return 0 if successful, otherwise return a non-zero error code.
984 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
985 const int noalloc, struct ip_msource **pims)
987 struct ip_msource find;
988 struct ip_msource *ims, *nims;
990 find.ims_haddr = haddr;
991 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
992 if (ims == NULL && !noalloc) {
993 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
995 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
999 nims->ims_haddr = haddr;
1000 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
1004 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
1014 * Merge socket-layer source into IGMP-layer source.
1015 * If rollback is non-zero, perform the inverse of the merge.
1018 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
1021 int n = rollback ? -1 : 1;
1023 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
1024 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
1025 __func__, n, ims->ims_haddr);
1026 ims->ims_st[1].ex -= n;
1027 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
1028 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
1029 __func__, n, ims->ims_haddr);
1030 ims->ims_st[1].in -= n;
1033 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
1034 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
1035 __func__, n, ims->ims_haddr);
1036 ims->ims_st[1].ex += n;
1037 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
1038 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
1039 __func__, n, ims->ims_haddr);
1040 ims->ims_st[1].in += n;
1045 * Atomically update the global in_multi state, when a membership's
1046 * filter list is being updated in any way.
1048 * imf is the per-inpcb-membership group filter pointer.
1049 * A fake imf may be passed for in-kernel consumers.
1051 * XXX This is a candidate for a set-symmetric-difference style loop
1052 * which would eliminate the repeated lookup from root of ims nodes,
1053 * as they share the same key space.
1055 * If any error occurred this function will back out of refcounts
1056 * and return a non-zero value.
1059 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1061 struct ip_msource *ims, *nims;
1062 struct in_msource *lims;
1063 int schanged, error;
1069 IN_MULTI_LIST_LOCK_ASSERT();
1072 * Update the source filters first, as this may fail.
1073 * Maintain count of in-mode filters at t0, t1. These are
1074 * used to work out if we transition into ASM mode or not.
1075 * Maintain a count of source filters whose state was
1076 * actually modified by this operation.
1078 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1079 lims = (struct in_msource *)ims;
1080 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1081 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1082 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1083 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1087 ims_merge(nims, lims, 0);
1090 struct ip_msource *bims;
1092 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1093 lims = (struct in_msource *)ims;
1094 if (lims->imsl_st[0] == lims->imsl_st[1])
1096 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1099 ims_merge(bims, lims, 1);
1104 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1105 __func__, nsrc0, nsrc1);
1107 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1108 if (imf->imf_st[0] == imf->imf_st[1] &&
1109 imf->imf_st[1] == MCAST_INCLUDE) {
1111 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1112 --inm->inm_st[1].iss_in;
1116 /* Handle filter mode transition on socket. */
1117 if (imf->imf_st[0] != imf->imf_st[1]) {
1118 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1119 __func__, imf->imf_st[0], imf->imf_st[1]);
1121 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1122 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1123 --inm->inm_st[1].iss_ex;
1124 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1125 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1126 --inm->inm_st[1].iss_in;
1129 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1130 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1131 inm->inm_st[1].iss_ex++;
1132 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1133 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1134 inm->inm_st[1].iss_in++;
1139 * Track inm filter state in terms of listener counts.
1140 * If there are any exclusive listeners, stack-wide
1141 * membership is exclusive.
1142 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1143 * If no listeners remain, state is undefined at t1,
1144 * and the IGMP lifecycle for this group should finish.
1146 if (inm->inm_st[1].iss_ex > 0) {
1147 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1148 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1149 } else if (inm->inm_st[1].iss_in > 0) {
1150 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1151 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1153 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1154 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1157 /* Decrement ASM listener count on transition out of ASM mode. */
1158 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1159 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1160 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1161 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1162 --inm->inm_st[1].iss_asm;
1166 /* Increment ASM listener count on transition to ASM mode. */
1167 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1168 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1169 inm->inm_st[1].iss_asm++;
1172 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1177 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1184 * Mark an in_multi's filter set deltas as committed.
1185 * Called by IGMP after a state change has been enqueued.
1188 inm_commit(struct in_multi *inm)
1190 struct ip_msource *ims;
1192 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1193 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1196 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1197 ims->ims_st[0] = ims->ims_st[1];
1199 inm->inm_st[0] = inm->inm_st[1];
1203 * Reap unreferenced nodes from an in_multi's filter set.
1206 inm_reap(struct in_multi *inm)
1208 struct ip_msource *ims, *tims;
1210 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1211 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1212 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1215 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1216 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1217 free(ims, M_IPMSOURCE);
1223 * Purge all source nodes from an in_multi's filter set.
1226 inm_purge(struct in_multi *inm)
1228 struct ip_msource *ims, *tims;
1230 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1231 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1232 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1233 free(ims, M_IPMSOURCE);
1239 * Join a multicast group; unlocked entry point.
1241 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1242 * is not held. Fortunately, ifp is unlikely to have been detached
1243 * at this point, so we assume it's OK to recurse.
1246 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1247 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1252 error = in_joingroup_locked(ifp, gina, imf, pinm);
1259 * Join a multicast group; real entry point.
1261 * Only preserves atomicity at inm level.
1262 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1264 * If the IGMP downcall fails, the group is not joined, and an error
1268 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1269 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1271 struct in_mfilter timf;
1272 struct in_multi *inm;
1275 IN_MULTI_LOCK_ASSERT();
1276 IN_MULTI_LIST_UNLOCK_ASSERT();
1278 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1279 ntohl(gina->s_addr), ifp, ifp->if_xname);
1285 * If no imf was specified (i.e. kernel consumer),
1286 * fake one up and assume it is an ASM join.
1289 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1293 error = in_getmulti(ifp, gina, &inm);
1295 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1298 IN_MULTI_LIST_LOCK();
1299 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1300 error = inm_merge(inm, imf);
1302 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1303 goto out_inm_release;
1306 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1307 error = igmp_change_state(inm);
1309 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1310 goto out_inm_release;
1316 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1317 inm_release_deferred(inm);
1321 IN_MULTI_LIST_UNLOCK();
1327 * Leave a multicast group; unlocked entry point.
1330 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1335 error = in_leavegroup_locked(inm, imf);
1342 * Leave a multicast group; real entry point.
1343 * All source filters will be expunged.
1345 * Only preserves atomicity at inm level.
1347 * Holding the write lock for the INP which contains imf
1348 * is highly advisable. We can't assert for it as imf does not
1349 * contain a back-pointer to the owning inp.
1351 * Note: This is not the same as inm_release(*) as this function also
1352 * makes a state change downcall into IGMP.
1355 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1357 struct in_mfilter timf;
1362 IN_MULTI_LOCK_ASSERT();
1363 IN_MULTI_LIST_UNLOCK_ASSERT();
1365 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1366 inm, ntohl(inm->inm_addr.s_addr),
1367 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1371 * If no imf was specified (i.e. kernel consumer),
1372 * fake one up and assume it is an ASM join.
1375 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1380 * Begin state merge transaction at IGMP layer.
1382 * As this particular invocation should not cause any memory
1383 * to be allocated, and there is no opportunity to roll back
1384 * the transaction, it MUST NOT fail.
1386 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1387 IN_MULTI_LIST_LOCK();
1388 error = inm_merge(inm, imf);
1389 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1391 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1392 CURVNET_SET(inm->inm_ifp->if_vnet);
1393 error = igmp_change_state(inm);
1394 IF_ADDR_WLOCK(inm->inm_ifp);
1395 inm_release_deferred(inm);
1396 IF_ADDR_WUNLOCK(inm->inm_ifp);
1397 IN_MULTI_LIST_UNLOCK();
1400 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1402 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1407 /*#ifndef BURN_BRIDGES*/
1409 * Join an IPv4 multicast group in (*,G) exclusive mode.
1410 * The group must be a 224.0.0.0/24 link-scope group.
1411 * This KPI is for legacy kernel consumers only.
1414 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1416 struct in_multi *pinm;
1419 char addrbuf[INET_ADDRSTRLEN];
1422 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1423 ("%s: %s not in 224.0.0.0/24", __func__,
1424 inet_ntoa_r(*ap, addrbuf)));
1426 error = in_joingroup(ifp, ap, NULL, &pinm);
1434 * Block or unblock an ASM multicast source on an inpcb.
1435 * This implements the delta-based API described in RFC 3678.
1437 * The delta-based API applies only to exclusive-mode memberships.
1438 * An IGMP downcall will be performed.
1440 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1442 * Return 0 if successful, otherwise return an appropriate error code.
1445 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1447 struct group_source_req gsr;
1448 struct rm_priotracker in_ifa_tracker;
1449 sockunion_t *gsa, *ssa;
1451 struct in_mfilter *imf;
1452 struct ip_moptions *imo;
1453 struct in_msource *ims;
1454 struct in_multi *inm;
1463 memset(&gsr, 0, sizeof(struct group_source_req));
1464 gsa = (sockunion_t *)&gsr.gsr_group;
1465 ssa = (sockunion_t *)&gsr.gsr_source;
1467 switch (sopt->sopt_name) {
1468 case IP_BLOCK_SOURCE:
1469 case IP_UNBLOCK_SOURCE: {
1470 struct ip_mreq_source mreqs;
1472 error = sooptcopyin(sopt, &mreqs,
1473 sizeof(struct ip_mreq_source),
1474 sizeof(struct ip_mreq_source));
1478 gsa->sin.sin_family = AF_INET;
1479 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1480 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1482 ssa->sin.sin_family = AF_INET;
1483 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1484 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1486 if (!in_nullhost(mreqs.imr_interface)) {
1487 IN_IFADDR_RLOCK(&in_ifa_tracker);
1488 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1489 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1491 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1494 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1495 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1499 case MCAST_BLOCK_SOURCE:
1500 case MCAST_UNBLOCK_SOURCE:
1501 error = sooptcopyin(sopt, &gsr,
1502 sizeof(struct group_source_req),
1503 sizeof(struct group_source_req));
1507 if (gsa->sin.sin_family != AF_INET ||
1508 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1511 if (ssa->sin.sin_family != AF_INET ||
1512 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1515 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1516 return (EADDRNOTAVAIL);
1518 ifp = ifnet_byindex(gsr.gsr_interface);
1520 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1525 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1526 __func__, sopt->sopt_name);
1527 return (EOPNOTSUPP);
1531 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1535 * Check if we are actually a member of this group.
1537 imo = inp_findmoptions(inp);
1538 idx = imo_match_group(imo, ifp, &gsa->sa);
1539 if (idx == -1 || imo->imo_mfilters == NULL) {
1540 error = EADDRNOTAVAIL;
1541 goto out_inp_locked;
1544 KASSERT(imo->imo_mfilters != NULL,
1545 ("%s: imo_mfilters not allocated", __func__));
1546 imf = &imo->imo_mfilters[idx];
1547 inm = imo->imo_membership[idx];
1550 * Attempting to use the delta-based API on an
1551 * non exclusive-mode membership is an error.
1553 fmode = imf->imf_st[0];
1554 if (fmode != MCAST_EXCLUDE) {
1556 goto out_inp_locked;
1560 * Deal with error cases up-front:
1561 * Asked to block, but already blocked; or
1562 * Asked to unblock, but nothing to unblock.
1563 * If adding a new block entry, allocate it.
1565 ims = imo_match_source(imo, idx, &ssa->sa);
1566 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1567 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1568 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1569 error = EADDRNOTAVAIL;
1570 goto out_inp_locked;
1573 INP_WLOCK_ASSERT(inp);
1576 * Begin state merge transaction at socket layer.
1579 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1580 ims = imf_graft(imf, fmode, &ssa->sin);
1584 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1585 error = imf_prune(imf, &ssa->sin);
1589 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1590 goto out_imf_rollback;
1594 * Begin state merge transaction at IGMP layer.
1597 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1598 IN_MULTI_LIST_LOCK();
1599 error = inm_merge(inm, imf);
1601 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1602 IN_MULTI_LIST_UNLOCK();
1603 goto out_in_multi_locked;
1606 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1607 error = igmp_change_state(inm);
1608 IN_MULTI_LIST_UNLOCK();
1610 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1612 out_in_multi_locked:
1629 * Given an inpcb, return its multicast options structure pointer. Accepts
1630 * an unlocked inpcb pointer, but will return it locked. May sleep.
1632 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1633 * SMPng: NOTE: Returns with the INP write lock held.
1635 static struct ip_moptions *
1636 inp_findmoptions(struct inpcb *inp)
1638 struct ip_moptions *imo;
1639 struct in_multi **immp;
1640 struct in_mfilter *imfp;
1644 if (inp->inp_moptions != NULL)
1645 return (inp->inp_moptions);
1649 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1650 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1652 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1653 M_INMFILTER, M_WAITOK);
1655 imo->imo_multicast_ifp = NULL;
1656 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1657 imo->imo_multicast_vif = -1;
1658 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1659 imo->imo_multicast_loop = in_mcast_loop;
1660 imo->imo_num_memberships = 0;
1661 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1662 imo->imo_membership = immp;
1664 /* Initialize per-group source filters. */
1665 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1666 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1667 imo->imo_mfilters = imfp;
1670 if (inp->inp_moptions != NULL) {
1671 free(imfp, M_INMFILTER);
1672 free(immp, M_IPMOPTS);
1673 free(imo, M_IPMOPTS);
1674 return (inp->inp_moptions);
1676 inp->inp_moptions = imo;
1681 inp_gcmoptions(struct ip_moptions *imo)
1683 struct in_mfilter *imf;
1684 struct in_multi *inm;
1686 size_t idx, nmships;
1688 nmships = imo->imo_num_memberships;
1689 for (idx = 0; idx < nmships; ++idx) {
1690 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1693 inm = imo->imo_membership[idx];
1696 CURVNET_SET(ifp->if_vnet);
1697 (void)in_leavegroup(inm, imf);
1700 (void)in_leavegroup(inm, imf);
1706 if (imo->imo_mfilters)
1707 free(imo->imo_mfilters, M_INMFILTER);
1708 free(imo->imo_membership, M_IPMOPTS);
1709 free(imo, M_IPMOPTS);
1713 * Discard the IP multicast options (and source filters). To minimize
1714 * the amount of work done while holding locks such as the INP's
1715 * pcbinfo lock (which is used in the receive path), the free
1716 * operation is deferred to the epoch callback task.
1719 inp_freemoptions(struct ip_moptions *imo)
1723 inp_gcmoptions(imo);
1727 * Atomically get source filters on a socket for an IPv4 multicast group.
1728 * Called with INP lock held; returns with lock released.
1731 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1733 struct __msfilterreq msfr;
1736 struct ip_moptions *imo;
1737 struct in_mfilter *imf;
1738 struct ip_msource *ims;
1739 struct in_msource *lims;
1740 struct sockaddr_in *psin;
1741 struct sockaddr_storage *ptss;
1742 struct sockaddr_storage *tss;
1744 size_t idx, nsrcs, ncsrcs;
1746 INP_WLOCK_ASSERT(inp);
1748 imo = inp->inp_moptions;
1749 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1753 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1754 sizeof(struct __msfilterreq));
1758 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1761 ifp = ifnet_byindex(msfr.msfr_ifindex);
1768 * Lookup group on the socket.
1770 gsa = (sockunion_t *)&msfr.msfr_group;
1771 idx = imo_match_group(imo, ifp, &gsa->sa);
1772 if (idx == -1 || imo->imo_mfilters == NULL) {
1774 return (EADDRNOTAVAIL);
1776 imf = &imo->imo_mfilters[idx];
1779 * Ignore memberships which are in limbo.
1781 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1785 msfr.msfr_fmode = imf->imf_st[1];
1788 * If the user specified a buffer, copy out the source filter
1789 * entries to userland gracefully.
1790 * We only copy out the number of entries which userland
1791 * has asked for, but we always tell userland how big the
1792 * buffer really needs to be.
1794 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1795 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1797 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1798 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1799 M_TEMP, M_NOWAIT | M_ZERO);
1807 * Count number of sources in-mode at t0.
1808 * If buffer space exists and remains, copy out source entries.
1810 nsrcs = msfr.msfr_nsrcs;
1813 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1814 lims = (struct in_msource *)ims;
1815 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1816 lims->imsl_st[0] != imf->imf_st[0])
1819 if (tss != NULL && nsrcs > 0) {
1820 psin = (struct sockaddr_in *)ptss;
1821 psin->sin_family = AF_INET;
1822 psin->sin_len = sizeof(struct sockaddr_in);
1823 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1833 error = copyout(tss, msfr.msfr_srcs,
1834 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1840 msfr.msfr_nsrcs = ncsrcs;
1841 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1847 * Return the IP multicast options in response to user getsockopt().
1850 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1852 struct rm_priotracker in_ifa_tracker;
1853 struct ip_mreqn mreqn;
1854 struct ip_moptions *imo;
1856 struct in_ifaddr *ia;
1861 imo = inp->inp_moptions;
1863 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1864 * or is a divert socket, reject it.
1866 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1867 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1868 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1870 return (EOPNOTSUPP);
1874 switch (sopt->sopt_name) {
1875 case IP_MULTICAST_VIF:
1877 optval = imo->imo_multicast_vif;
1881 error = sooptcopyout(sopt, &optval, sizeof(int));
1884 case IP_MULTICAST_IF:
1885 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1887 ifp = imo->imo_multicast_ifp;
1888 if (!in_nullhost(imo->imo_multicast_addr)) {
1889 mreqn.imr_address = imo->imo_multicast_addr;
1890 } else if (ifp != NULL) {
1891 struct epoch_tracker et;
1893 mreqn.imr_ifindex = ifp->if_index;
1894 NET_EPOCH_ENTER(et);
1895 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1898 IA_SIN(ia)->sin_addr;
1903 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1904 error = sooptcopyout(sopt, &mreqn,
1905 sizeof(struct ip_mreqn));
1907 error = sooptcopyout(sopt, &mreqn.imr_address,
1908 sizeof(struct in_addr));
1912 case IP_MULTICAST_TTL:
1914 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1916 optval = coptval = imo->imo_multicast_ttl;
1918 if (sopt->sopt_valsize == sizeof(u_char))
1919 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1921 error = sooptcopyout(sopt, &optval, sizeof(int));
1924 case IP_MULTICAST_LOOP:
1926 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1928 optval = coptval = imo->imo_multicast_loop;
1930 if (sopt->sopt_valsize == sizeof(u_char))
1931 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1933 error = sooptcopyout(sopt, &optval, sizeof(int));
1938 error = EADDRNOTAVAIL;
1941 error = inp_get_source_filters(inp, sopt);
1947 error = ENOPROTOOPT;
1951 INP_UNLOCK_ASSERT(inp);
1957 * Look up the ifnet to use for a multicast group membership,
1958 * given the IPv4 address of an interface, and the IPv4 group address.
1960 * This routine exists to support legacy multicast applications
1961 * which do not understand that multicast memberships are scoped to
1962 * specific physical links in the networking stack, or which need
1963 * to join link-scope groups before IPv4 addresses are configured.
1965 * If inp is non-NULL, use this socket's current FIB number for any
1966 * required FIB lookup.
1967 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1968 * and use its ifp; usually, this points to the default next-hop.
1970 * If the FIB lookup fails, attempt to use the first non-loopback
1971 * interface with multicast capability in the system as a
1972 * last resort. The legacy IPv4 ASM API requires that we do
1973 * this in order to allow groups to be joined when the routing
1974 * table has not yet been populated during boot.
1976 * Returns NULL if no ifp could be found.
1978 * FUTURE: Implement IPv4 source-address selection.
1980 static struct ifnet *
1981 inp_lookup_mcast_ifp(const struct inpcb *inp,
1982 const struct sockaddr_in *gsin, const struct in_addr ina)
1984 struct rm_priotracker in_ifa_tracker;
1986 struct nhop4_basic nh4;
1989 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1990 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1991 ("%s: not multicast", __func__));
1994 if (!in_nullhost(ina)) {
1995 IN_IFADDR_RLOCK(&in_ifa_tracker);
1996 INADDR_TO_IFP(ina, ifp);
1997 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1999 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
2000 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
2003 struct in_ifaddr *ia;
2007 IN_IFADDR_RLOCK(&in_ifa_tracker);
2008 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
2010 if (!(mifp->if_flags & IFF_LOOPBACK) &&
2011 (mifp->if_flags & IFF_MULTICAST)) {
2016 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2024 * Join an IPv4 multicast group, possibly with a source.
2027 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
2029 struct group_source_req gsr;
2030 sockunion_t *gsa, *ssa;
2032 struct in_mfilter *imf;
2033 struct ip_moptions *imo;
2034 struct in_multi *inm;
2035 struct in_msource *lims;
2045 memset(&gsr, 0, sizeof(struct group_source_req));
2046 gsa = (sockunion_t *)&gsr.gsr_group;
2047 gsa->ss.ss_family = AF_UNSPEC;
2048 ssa = (sockunion_t *)&gsr.gsr_source;
2049 ssa->ss.ss_family = AF_UNSPEC;
2051 switch (sopt->sopt_name) {
2052 case IP_ADD_MEMBERSHIP:
2053 case IP_ADD_SOURCE_MEMBERSHIP: {
2054 struct ip_mreq_source mreqs;
2056 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
2057 error = sooptcopyin(sopt, &mreqs,
2058 sizeof(struct ip_mreq),
2059 sizeof(struct ip_mreq));
2061 * Do argument switcharoo from ip_mreq into
2062 * ip_mreq_source to avoid using two instances.
2064 mreqs.imr_interface = mreqs.imr_sourceaddr;
2065 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2066 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2067 error = sooptcopyin(sopt, &mreqs,
2068 sizeof(struct ip_mreq_source),
2069 sizeof(struct ip_mreq_source));
2074 gsa->sin.sin_family = AF_INET;
2075 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2076 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2078 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2079 ssa->sin.sin_family = AF_INET;
2080 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2081 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2084 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2087 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2088 mreqs.imr_interface);
2089 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2090 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2094 case MCAST_JOIN_GROUP:
2095 case MCAST_JOIN_SOURCE_GROUP:
2096 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2097 error = sooptcopyin(sopt, &gsr,
2098 sizeof(struct group_req),
2099 sizeof(struct group_req));
2100 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2101 error = sooptcopyin(sopt, &gsr,
2102 sizeof(struct group_source_req),
2103 sizeof(struct group_source_req));
2108 if (gsa->sin.sin_family != AF_INET ||
2109 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2113 * Overwrite the port field if present, as the sockaddr
2114 * being copied in may be matched with a binary comparison.
2116 gsa->sin.sin_port = 0;
2117 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2118 if (ssa->sin.sin_family != AF_INET ||
2119 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2121 ssa->sin.sin_port = 0;
2124 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2127 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2128 return (EADDRNOTAVAIL);
2129 ifp = ifnet_byindex(gsr.gsr_interface);
2133 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2134 __func__, sopt->sopt_name);
2135 return (EOPNOTSUPP);
2139 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2140 return (EADDRNOTAVAIL);
2142 imo = inp_findmoptions(inp);
2143 idx = imo_match_group(imo, ifp, &gsa->sa);
2147 inm = imo->imo_membership[idx];
2148 imf = &imo->imo_mfilters[idx];
2149 if (ssa->ss.ss_family != AF_UNSPEC) {
2151 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2152 * is an error. On an existing inclusive membership,
2153 * it just adds the source to the filter list.
2155 if (imf->imf_st[1] != MCAST_INCLUDE) {
2157 goto out_inp_locked;
2160 * Throw out duplicates.
2162 * XXX FIXME: This makes a naive assumption that
2163 * even if entries exist for *ssa in this imf,
2164 * they will be rejected as dupes, even if they
2165 * are not valid in the current mode (in-mode).
2167 * in_msource is transactioned just as for anything
2168 * else in SSM -- but note naive use of inm_graft()
2169 * below for allocating new filter entries.
2171 * This is only an issue if someone mixes the
2172 * full-state SSM API with the delta-based API,
2173 * which is discouraged in the relevant RFCs.
2175 lims = imo_match_source(imo, idx, &ssa->sa);
2176 if (lims != NULL /*&&
2177 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2178 error = EADDRNOTAVAIL;
2179 goto out_inp_locked;
2183 * MCAST_JOIN_GROUP on an existing exclusive
2184 * membership is an error; return EADDRINUSE
2185 * to preserve 4.4BSD API idempotence, and
2186 * avoid tedious detour to code below.
2187 * NOTE: This is bending RFC 3678 a bit.
2189 * On an existing inclusive membership, this is also
2190 * an error; if you want to change filter mode,
2191 * you must use the userland API setsourcefilter().
2192 * XXX We don't reject this for imf in UNDEFINED
2193 * state at t1, because allocation of a filter
2194 * is atomic with allocation of a membership.
2197 if (imf->imf_st[1] == MCAST_EXCLUDE)
2199 goto out_inp_locked;
2204 * Begin state merge transaction at socket layer.
2206 INP_WLOCK_ASSERT(inp);
2209 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2210 error = imo_grow(imo);
2212 goto out_inp_locked;
2215 * Allocate the new slot upfront so we can deal with
2216 * grafting the new source filter in same code path
2217 * as for join-source on existing membership.
2219 idx = imo->imo_num_memberships;
2220 imo->imo_membership[idx] = NULL;
2221 imo->imo_num_memberships++;
2222 KASSERT(imo->imo_mfilters != NULL,
2223 ("%s: imf_mfilters vector was not allocated", __func__));
2224 imf = &imo->imo_mfilters[idx];
2225 KASSERT(RB_EMPTY(&imf->imf_sources),
2226 ("%s: imf_sources not empty", __func__));
2230 * Graft new source into filter list for this inpcb's
2231 * membership of the group. The in_multi may not have
2232 * been allocated yet if this is a new membership, however,
2233 * the in_mfilter slot will be allocated and must be initialized.
2235 * Note: Grafting of exclusive mode filters doesn't happen
2237 * XXX: Should check for non-NULL lims (node exists but may
2238 * not be in-mode) for interop with full-state API.
2240 if (ssa->ss.ss_family != AF_UNSPEC) {
2241 /* Membership starts in IN mode */
2243 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2244 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2246 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2248 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2250 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2256 /* No address specified; Membership starts in EX mode */
2258 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2259 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2264 * Begin state merge transaction at IGMP layer.
2271 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2274 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2276 IN_MULTI_LIST_UNLOCK();
2280 imo->imo_membership[idx] = inm;
2282 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2283 IN_MULTI_LIST_LOCK();
2284 error = inm_merge(inm, imf);
2286 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2288 IN_MULTI_LIST_UNLOCK();
2289 goto out_in_multi_locked;
2291 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2292 error = igmp_change_state(inm);
2293 IN_MULTI_LIST_UNLOCK();
2295 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2297 goto out_in_multi_locked;
2301 out_in_multi_locked:
2305 if (in_pcbrele_wlocked(inp))
2318 if (error && is_new) {
2319 inm = imo->imo_membership[idx];
2321 IN_MULTI_LIST_LOCK();
2322 inm_release_deferred(inm);
2323 IN_MULTI_LIST_UNLOCK();
2325 imo->imo_membership[idx] = NULL;
2326 --imo->imo_num_memberships;
2335 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2338 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2340 struct group_source_req gsr;
2341 struct ip_mreq_source mreqs;
2342 struct rm_priotracker in_ifa_tracker;
2343 sockunion_t *gsa, *ssa;
2345 struct in_mfilter *imf;
2346 struct ip_moptions *imo;
2347 struct in_msource *ims;
2348 struct in_multi *inm;
2350 int error, is_final;
2356 memset(&gsr, 0, sizeof(struct group_source_req));
2357 gsa = (sockunion_t *)&gsr.gsr_group;
2358 gsa->ss.ss_family = AF_UNSPEC;
2359 ssa = (sockunion_t *)&gsr.gsr_source;
2360 ssa->ss.ss_family = AF_UNSPEC;
2362 switch (sopt->sopt_name) {
2363 case IP_DROP_MEMBERSHIP:
2364 case IP_DROP_SOURCE_MEMBERSHIP:
2365 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2366 error = sooptcopyin(sopt, &mreqs,
2367 sizeof(struct ip_mreq),
2368 sizeof(struct ip_mreq));
2370 * Swap interface and sourceaddr arguments,
2371 * as ip_mreq and ip_mreq_source are laid
2374 mreqs.imr_interface = mreqs.imr_sourceaddr;
2375 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2376 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2377 error = sooptcopyin(sopt, &mreqs,
2378 sizeof(struct ip_mreq_source),
2379 sizeof(struct ip_mreq_source));
2384 gsa->sin.sin_family = AF_INET;
2385 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2386 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2388 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2389 ssa->sin.sin_family = AF_INET;
2390 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2391 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2395 * Attempt to look up hinted ifp from interface address.
2396 * Fallthrough with null ifp iff lookup fails, to
2397 * preserve 4.4BSD mcast API idempotence.
2398 * XXX NOTE WELL: The RFC 3678 API is preferred because
2399 * using an IPv4 address as a key is racy.
2401 if (!in_nullhost(mreqs.imr_interface)) {
2402 IN_IFADDR_RLOCK(&in_ifa_tracker);
2403 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2404 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2406 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2407 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2411 case MCAST_LEAVE_GROUP:
2412 case MCAST_LEAVE_SOURCE_GROUP:
2413 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2414 error = sooptcopyin(sopt, &gsr,
2415 sizeof(struct group_req),
2416 sizeof(struct group_req));
2417 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2418 error = sooptcopyin(sopt, &gsr,
2419 sizeof(struct group_source_req),
2420 sizeof(struct group_source_req));
2425 if (gsa->sin.sin_family != AF_INET ||
2426 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2429 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2430 if (ssa->sin.sin_family != AF_INET ||
2431 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2435 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2436 return (EADDRNOTAVAIL);
2438 ifp = ifnet_byindex(gsr.gsr_interface);
2441 return (EADDRNOTAVAIL);
2445 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2446 __func__, sopt->sopt_name);
2447 return (EOPNOTSUPP);
2451 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2455 * Find the membership in the membership array.
2457 imo = inp_findmoptions(inp);
2458 idx = imo_match_group(imo, ifp, &gsa->sa);
2460 error = EADDRNOTAVAIL;
2461 goto out_inp_locked;
2463 inm = imo->imo_membership[idx];
2464 imf = &imo->imo_mfilters[idx];
2466 if (ssa->ss.ss_family != AF_UNSPEC)
2470 * Begin state merge transaction at socket layer.
2472 INP_WLOCK_ASSERT(inp);
2475 * If we were instructed only to leave a given source, do so.
2476 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2481 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2482 error = EADDRNOTAVAIL;
2483 goto out_inp_locked;
2485 ims = imo_match_source(imo, idx, &ssa->sa);
2487 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2488 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2489 error = EADDRNOTAVAIL;
2490 goto out_inp_locked;
2492 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2493 error = imf_prune(imf, &ssa->sin);
2495 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2497 goto out_inp_locked;
2502 * Begin state merge transaction at IGMP layer.
2510 * Give up the multicast address record to which
2511 * the membership points.
2513 (void)in_leavegroup_locked(inm, imf);
2515 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2516 IN_MULTI_LIST_LOCK();
2517 error = inm_merge(inm, imf);
2519 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2521 IN_MULTI_LIST_UNLOCK();
2522 goto out_in_multi_locked;
2525 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2526 error = igmp_change_state(inm);
2527 IN_MULTI_LIST_UNLOCK();
2529 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2534 out_in_multi_locked:
2538 if (in_pcbrele_wlocked(inp))
2549 /* Remove the gap in the membership and filter array. */
2550 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2551 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2552 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2554 imo->imo_num_memberships--;
2563 * Select the interface for transmitting IPv4 multicast datagrams.
2565 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2566 * may be passed to this socket option. An address of INADDR_ANY or an
2567 * interface index of 0 is used to remove a previous selection.
2568 * When no interface is selected, one is chosen for every send.
2571 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2573 struct rm_priotracker in_ifa_tracker;
2574 struct in_addr addr;
2575 struct ip_mreqn mreqn;
2577 struct ip_moptions *imo;
2580 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2582 * An interface index was specified using the
2583 * Linux-derived ip_mreqn structure.
2585 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2586 sizeof(struct ip_mreqn));
2590 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2593 if (mreqn.imr_ifindex == 0) {
2596 ifp = ifnet_byindex(mreqn.imr_ifindex);
2598 return (EADDRNOTAVAIL);
2602 * An interface was specified by IPv4 address.
2603 * This is the traditional BSD usage.
2605 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2606 sizeof(struct in_addr));
2609 if (in_nullhost(addr)) {
2612 IN_IFADDR_RLOCK(&in_ifa_tracker);
2613 INADDR_TO_IFP(addr, ifp);
2614 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2616 return (EADDRNOTAVAIL);
2618 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2619 ntohl(addr.s_addr));
2622 /* Reject interfaces which do not support multicast. */
2623 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2624 return (EOPNOTSUPP);
2626 imo = inp_findmoptions(inp);
2627 imo->imo_multicast_ifp = ifp;
2628 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2635 * Atomically set source filters on a socket for an IPv4 multicast group.
2637 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2640 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2642 struct __msfilterreq msfr;
2645 struct in_mfilter *imf;
2646 struct ip_moptions *imo;
2647 struct in_multi *inm;
2651 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2652 sizeof(struct __msfilterreq));
2656 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2659 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2660 msfr.msfr_fmode != MCAST_INCLUDE))
2663 if (msfr.msfr_group.ss_family != AF_INET ||
2664 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2667 gsa = (sockunion_t *)&msfr.msfr_group;
2668 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2671 gsa->sin.sin_port = 0; /* ignore port */
2673 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2674 return (EADDRNOTAVAIL);
2676 ifp = ifnet_byindex(msfr.msfr_ifindex);
2678 return (EADDRNOTAVAIL);
2681 * Take the INP write lock.
2682 * Check if this socket is a member of this group.
2684 imo = inp_findmoptions(inp);
2685 idx = imo_match_group(imo, ifp, &gsa->sa);
2686 if (idx == -1 || imo->imo_mfilters == NULL) {
2687 error = EADDRNOTAVAIL;
2688 goto out_inp_locked;
2690 inm = imo->imo_membership[idx];
2691 imf = &imo->imo_mfilters[idx];
2694 * Begin state merge transaction at socket layer.
2696 INP_WLOCK_ASSERT(inp);
2698 imf->imf_st[1] = msfr.msfr_fmode;
2701 * Apply any new source filters, if present.
2702 * Make a copy of the user-space source vector so
2703 * that we may copy them with a single copyin. This
2704 * allows us to deal with page faults up-front.
2706 if (msfr.msfr_nsrcs > 0) {
2707 struct in_msource *lims;
2708 struct sockaddr_in *psin;
2709 struct sockaddr_storage *kss, *pkss;
2714 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2715 __func__, (unsigned long)msfr.msfr_nsrcs);
2716 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2718 error = copyin(msfr.msfr_srcs, kss,
2719 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2728 * Mark all source filters as UNDEFINED at t1.
2729 * Restore new group filter mode, as imf_leave()
2730 * will set it to INCLUDE.
2733 imf->imf_st[1] = msfr.msfr_fmode;
2736 * Update socket layer filters at t1, lazy-allocating
2737 * new entries. This saves a bunch of memory at the
2738 * cost of one RB_FIND() per source entry; duplicate
2739 * entries in the msfr_nsrcs vector are ignored.
2740 * If we encounter an error, rollback transaction.
2742 * XXX This too could be replaced with a set-symmetric
2743 * difference like loop to avoid walking from root
2744 * every time, as the key space is common.
2746 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2747 psin = (struct sockaddr_in *)pkss;
2748 if (psin->sin_family != AF_INET) {
2749 error = EAFNOSUPPORT;
2752 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2756 error = imf_get_source(imf, psin, &lims);
2759 lims->imsl_st[1] = imf->imf_st[1];
2765 goto out_imf_rollback;
2767 INP_WLOCK_ASSERT(inp);
2771 * Begin state merge transaction at IGMP layer.
2773 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2774 IN_MULTI_LIST_LOCK();
2775 error = inm_merge(inm, imf);
2777 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2778 IN_MULTI_LIST_UNLOCK();
2779 goto out_in_multi_locked;
2782 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2783 error = igmp_change_state(inm);
2784 IN_MULTI_LIST_UNLOCK();
2786 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2788 out_in_multi_locked:
2806 * Set the IP multicast options in response to user setsockopt().
2808 * Many of the socket options handled in this function duplicate the
2809 * functionality of socket options in the regular unicast API. However,
2810 * it is not possible to merge the duplicate code, because the idempotence
2811 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2812 * the effects of these options must be treated as separate and distinct.
2814 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2815 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2816 * is refactored to no longer use vifs.
2819 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2821 struct ip_moptions *imo;
2827 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2828 * or is a divert socket, reject it.
2830 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2831 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2832 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2833 return (EOPNOTSUPP);
2835 switch (sopt->sopt_name) {
2836 case IP_MULTICAST_VIF: {
2839 * Select a multicast VIF for transmission.
2840 * Only useful if multicast forwarding is active.
2842 if (legal_vif_num == NULL) {
2846 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2849 if (!legal_vif_num(vifi) && (vifi != -1)) {
2853 imo = inp_findmoptions(inp);
2854 imo->imo_multicast_vif = vifi;
2859 case IP_MULTICAST_IF:
2860 error = inp_set_multicast_if(inp, sopt);
2863 case IP_MULTICAST_TTL: {
2867 * Set the IP time-to-live for outgoing multicast packets.
2868 * The original multicast API required a char argument,
2869 * which is inconsistent with the rest of the socket API.
2870 * We allow either a char or an int.
2872 if (sopt->sopt_valsize == sizeof(u_char)) {
2873 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2880 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2890 imo = inp_findmoptions(inp);
2891 imo->imo_multicast_ttl = ttl;
2896 case IP_MULTICAST_LOOP: {
2900 * Set the loopback flag for outgoing multicast packets.
2901 * Must be zero or one. The original multicast API required a
2902 * char argument, which is inconsistent with the rest
2903 * of the socket API. We allow either a char or an int.
2905 if (sopt->sopt_valsize == sizeof(u_char)) {
2906 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2913 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2917 loop = (u_char)iloop;
2919 imo = inp_findmoptions(inp);
2920 imo->imo_multicast_loop = !!loop;
2925 case IP_ADD_MEMBERSHIP:
2926 case IP_ADD_SOURCE_MEMBERSHIP:
2927 case MCAST_JOIN_GROUP:
2928 case MCAST_JOIN_SOURCE_GROUP:
2929 error = inp_join_group(inp, sopt);
2932 case IP_DROP_MEMBERSHIP:
2933 case IP_DROP_SOURCE_MEMBERSHIP:
2934 case MCAST_LEAVE_GROUP:
2935 case MCAST_LEAVE_SOURCE_GROUP:
2936 error = inp_leave_group(inp, sopt);
2939 case IP_BLOCK_SOURCE:
2940 case IP_UNBLOCK_SOURCE:
2941 case MCAST_BLOCK_SOURCE:
2942 case MCAST_UNBLOCK_SOURCE:
2943 error = inp_block_unblock_source(inp, sopt);
2947 error = inp_set_source_filters(inp, sopt);
2955 INP_UNLOCK_ASSERT(inp);
2961 * Expose IGMP's multicast filter mode and source list(s) to userland,
2962 * keyed by (ifindex, group).
2963 * The filter mode is written out as a uint32_t, followed by
2964 * 0..n of struct in_addr.
2965 * For use by ifmcstat(8).
2966 * SMPng: NOTE: unlocked read of ifindex space.
2969 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2971 struct in_addr src, group;
2972 struct epoch_tracker et;
2974 struct ifmultiaddr *ifma;
2975 struct in_multi *inm;
2976 struct ip_msource *ims;
2980 uint32_t fmode, ifindex;
2985 if (req->newptr != NULL)
2992 if (ifindex <= 0 || ifindex > V_if_index) {
2993 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2998 group.s_addr = name[1];
2999 if (!IN_MULTICAST(ntohl(group.s_addr))) {
3000 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
3001 __func__, ntohl(group.s_addr));
3005 ifp = ifnet_byindex(ifindex);
3007 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
3012 retval = sysctl_wire_old_buffer(req,
3013 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
3017 IN_MULTI_LIST_LOCK();
3019 NET_EPOCH_ENTER(et);
3020 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3021 if (ifma->ifma_addr->sa_family != AF_INET ||
3022 ifma->ifma_protospec == NULL)
3024 inm = (struct in_multi *)ifma->ifma_protospec;
3025 if (!in_hosteq(inm->inm_addr, group))
3027 fmode = inm->inm_st[1].iss_fmode;
3028 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
3031 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
3032 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
3035 * Only copy-out sources which are in-mode.
3037 if (fmode != ims_get_mode(inm, ims, 1)) {
3038 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
3042 src.s_addr = htonl(ims->ims_haddr);
3043 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
3050 IN_MULTI_LIST_UNLOCK();
3055 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
3057 static const char *inm_modestrs[] = { "un", "in", "ex" };
3060 inm_mode_str(const int mode)
3063 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
3064 return (inm_modestrs[mode]);
3068 static const char *inm_statestrs[] = {
3081 inm_state_str(const int state)
3084 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3085 return (inm_statestrs[state]);
3090 * Dump an in_multi structure to the console.
3093 inm_print(const struct in_multi *inm)
3096 char addrbuf[INET_ADDRSTRLEN];
3098 if ((ktr_mask & KTR_IGMPV3) == 0)
3101 printf("%s: --- begin inm %p ---\n", __func__, inm);
3102 printf("addr %s ifp %p(%s) ifma %p\n",
3103 inet_ntoa_r(inm->inm_addr, addrbuf),
3105 inm->inm_ifp->if_xname,
3107 printf("timer %u state %s refcount %u scq.len %u\n",
3109 inm_state_str(inm->inm_state),
3111 inm->inm_scq.mq_len);
3112 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3117 for (t = 0; t < 2; t++) {
3118 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3119 inm_mode_str(inm->inm_st[t].iss_fmode),
3120 inm->inm_st[t].iss_asm,
3121 inm->inm_st[t].iss_ex,
3122 inm->inm_st[t].iss_in,
3123 inm->inm_st[t].iss_rec);
3125 printf("%s: --- end inm %p ---\n", __func__, inm);
3128 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3131 inm_print(const struct in_multi *inm)
3136 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3138 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);