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 void inp_freemoptions_internal(struct ip_moptions *);
169 static void inp_gcmoptions(void *, int);
170 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
171 static int inp_join_group(struct inpcb *, struct sockopt *);
172 static int inp_leave_group(struct inpcb *, struct sockopt *);
173 static struct ifnet *
174 inp_lookup_mcast_ifp(const struct inpcb *,
175 const struct sockaddr_in *, const struct in_addr);
176 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
177 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
178 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
179 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
181 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
184 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
185 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
186 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
187 "Max source filters per group");
189 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
190 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
191 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
192 "Max source filters per socket");
194 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
195 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
196 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
198 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
199 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
200 "Per-interface stack-wide source filters");
202 static STAILQ_HEAD(, ip_moptions) imo_gc_list =
203 STAILQ_HEAD_INITIALIZER(imo_gc_list);
204 static struct task imo_gc_task = TASK_INITIALIZER(0, inp_gcmoptions, NULL);
208 * Inline function which wraps assertions for a valid ifp.
209 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
213 inm_is_ifp_detached(const struct in_multi *inm)
217 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
218 ifp = inm->inm_ifma->ifma_ifp;
221 * Sanity check that netinet's notion of ifp is the
224 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
227 return (ifp == NULL);
231 static struct grouptask free_gtask;
232 static struct in_multi_head inm_free_list;
233 static void inm_release_task(void *arg __unused);
234 static void inm_init(void)
236 SLIST_INIT(&inm_free_list);
237 taskqgroup_config_gtask_init(NULL, &free_gtask, inm_release_task, "inm release task");
240 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
245 inm_release_list_deferred(struct in_multi_head *inmh)
248 if (SLIST_EMPTY(inmh))
250 mtx_lock(&in_multi_free_mtx);
251 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
252 mtx_unlock(&in_multi_free_mtx);
253 GROUPTASK_ENQUEUE(&free_gtask);
257 inm_disconnect(struct in_multi *inm)
260 struct ifmultiaddr *ifma, *ll_ifma;
263 IF_ADDR_WLOCK_ASSERT(ifp);
264 ifma = inm->inm_ifma;
267 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
268 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
269 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
270 MPASS(ifma != ll_ifma);
271 ifma->ifma_llifma = NULL;
272 MPASS(ll_ifma->ifma_llifma == NULL);
273 MPASS(ll_ifma->ifma_ifp == ifp);
274 if (--ll_ifma->ifma_refcount == 0) {
275 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
276 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
277 if_freemulti(ll_ifma);
284 inm_release_deferred(struct in_multi *inm)
286 struct in_multi_head tmp;
288 IN_MULTI_LIST_LOCK_ASSERT();
289 MPASS(inm->inm_refcount > 0);
290 if (--inm->inm_refcount == 0) {
293 inm->inm_ifma->ifma_protospec = NULL;
294 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
295 inm_release_list_deferred(&tmp);
300 inm_release_task(void *arg __unused)
302 struct in_multi_head inm_free_tmp;
303 struct in_multi *inm, *tinm;
305 SLIST_INIT(&inm_free_tmp);
306 mtx_lock(&in_multi_free_mtx);
307 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
308 mtx_unlock(&in_multi_free_mtx);
310 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
311 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
319 * Initialize an in_mfilter structure to a known state at t0, t1
320 * with an empty source filter list.
323 imf_init(struct in_mfilter *imf, const int st0, const int st1)
325 memset(imf, 0, sizeof(struct in_mfilter));
326 RB_INIT(&imf->imf_sources);
327 imf->imf_st[0] = st0;
328 imf->imf_st[1] = st1;
332 * Function for looking up an in_multi record for an IPv4 multicast address
333 * on a given interface. ifp must be valid. If no record found, return NULL.
334 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
337 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
339 struct ifmultiaddr *ifma;
340 struct in_multi *inm;
342 IN_MULTI_LIST_LOCK_ASSERT();
343 IF_ADDR_LOCK_ASSERT(ifp);
346 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
347 if (ifma->ifma_addr->sa_family != AF_INET ||
348 ifma->ifma_protospec == NULL)
350 inm = (struct in_multi *)ifma->ifma_protospec;
351 if (inm->inm_addr.s_addr == ina.s_addr)
359 * Wrapper for inm_lookup_locked().
360 * The IF_ADDR_LOCK will be taken on ifp and released on return.
363 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
365 struct in_multi *inm;
367 IN_MULTI_LIST_LOCK_ASSERT();
369 inm = inm_lookup_locked(ifp, ina);
370 IF_ADDR_RUNLOCK(ifp);
376 * Resize the ip_moptions vector to the next power-of-two minus 1.
377 * May be called with locks held; do not sleep.
380 imo_grow(struct ip_moptions *imo)
382 struct in_multi **nmships;
383 struct in_multi **omships;
384 struct in_mfilter *nmfilters;
385 struct in_mfilter *omfilters;
392 omships = imo->imo_membership;
393 omfilters = imo->imo_mfilters;
394 oldmax = imo->imo_max_memberships;
395 newmax = ((oldmax + 1) * 2) - 1;
397 if (newmax <= IP_MAX_MEMBERSHIPS) {
398 nmships = (struct in_multi **)realloc(omships,
399 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
400 nmfilters = (struct in_mfilter *)realloc(omfilters,
401 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
402 if (nmships != NULL && nmfilters != NULL) {
403 /* Initialize newly allocated source filter heads. */
404 for (idx = oldmax; idx < newmax; idx++) {
405 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
408 imo->imo_max_memberships = newmax;
409 imo->imo_membership = nmships;
410 imo->imo_mfilters = nmfilters;
414 if (nmships == NULL || nmfilters == NULL) {
416 free(nmships, M_IPMOPTS);
417 if (nmfilters != NULL)
418 free(nmfilters, M_INMFILTER);
419 return (ETOOMANYREFS);
426 * Find an IPv4 multicast group entry for this ip_moptions instance
427 * which matches the specified group, and optionally an interface.
428 * Return its index into the array, or -1 if not found.
431 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
432 const struct sockaddr *group)
434 const struct sockaddr_in *gsin;
435 struct in_multi **pinm;
439 gsin = (const struct sockaddr_in *)group;
441 /* The imo_membership array may be lazy allocated. */
442 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
445 nmships = imo->imo_num_memberships;
446 pinm = &imo->imo_membership[0];
447 for (idx = 0; idx < nmships; idx++, pinm++) {
450 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
451 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
462 * Find an IPv4 multicast source entry for this imo which matches
463 * the given group index for this socket, and source address.
465 * NOTE: This does not check if the entry is in-mode, merely if
466 * it exists, which may not be the desired behaviour.
468 static struct in_msource *
469 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
470 const struct sockaddr *src)
472 struct ip_msource find;
473 struct in_mfilter *imf;
474 struct ip_msource *ims;
475 const sockunion_t *psa;
477 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
478 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
479 ("%s: invalid index %d\n", __func__, (int)gidx));
481 /* The imo_mfilters array may be lazy allocated. */
482 if (imo->imo_mfilters == NULL)
484 imf = &imo->imo_mfilters[gidx];
486 /* Source trees are keyed in host byte order. */
487 psa = (const sockunion_t *)src;
488 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
489 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
491 return ((struct in_msource *)ims);
495 * Perform filtering for multicast datagrams on a socket by group and source.
497 * Returns 0 if a datagram should be allowed through, or various error codes
498 * if the socket was not a member of the group, or the source was muted, etc.
501 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
502 const struct sockaddr *group, const struct sockaddr *src)
505 struct in_msource *ims;
508 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
510 gidx = imo_match_group(imo, ifp, group);
512 return (MCAST_NOTGMEMBER);
515 * Check if the source was included in an (S,G) join.
516 * Allow reception on exclusive memberships by default,
517 * reject reception on inclusive memberships by default.
518 * Exclude source only if an in-mode exclude filter exists.
519 * Include source only if an in-mode include filter exists.
520 * NOTE: We are comparing group state here at IGMP t1 (now)
521 * with socket-layer t0 (since last downcall).
523 mode = imo->imo_mfilters[gidx].imf_st[1];
524 ims = imo_match_source(imo, gidx, src);
526 if ((ims == NULL && mode == MCAST_INCLUDE) ||
527 (ims != NULL && ims->imsl_st[0] != mode))
528 return (MCAST_NOTSMEMBER);
534 * Find and return a reference to an in_multi record for (ifp, group),
535 * and bump its reference count.
536 * If one does not exist, try to allocate it, and update link-layer multicast
537 * filters on ifp to listen for group.
538 * Assumes the IN_MULTI lock is held across the call.
539 * Return 0 if successful, otherwise return an appropriate error code.
542 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
543 struct in_multi **pinm)
545 struct sockaddr_in gsin;
546 struct ifmultiaddr *ifma;
547 struct in_ifinfo *ii;
548 struct in_multi *inm;
551 IN_MULTI_LOCK_ASSERT();
553 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
554 IN_MULTI_LIST_LOCK();
555 inm = inm_lookup(ifp, *group);
558 * If we already joined this group, just bump the
559 * refcount and return it.
561 KASSERT(inm->inm_refcount >= 1,
562 ("%s: bad refcount %d", __func__, inm->inm_refcount));
563 inm_acquire_locked(inm);
566 IN_MULTI_LIST_UNLOCK();
570 memset(&gsin, 0, sizeof(gsin));
571 gsin.sin_family = AF_INET;
572 gsin.sin_len = sizeof(struct sockaddr_in);
573 gsin.sin_addr = *group;
576 * Check if a link-layer group is already associated
577 * with this network-layer group on the given ifnet.
579 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
583 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
584 IN_MULTI_LIST_LOCK();
588 * If something other than netinet is occupying the link-layer
589 * group, print a meaningful error message and back out of
591 * Otherwise, bump the refcount on the existing network-layer
592 * group association and return it.
594 if (ifma->ifma_protospec != NULL) {
595 inm = (struct in_multi *)ifma->ifma_protospec;
597 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
599 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
600 ("%s: ifma not AF_INET", __func__));
601 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
602 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
603 !in_hosteq(inm->inm_addr, *group)) {
604 char addrbuf[INET_ADDRSTRLEN];
606 panic("%s: ifma %p is inconsistent with %p (%s)",
607 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
610 inm_acquire_locked(inm);
615 IF_ADDR_WLOCK_ASSERT(ifp);
618 * A new in_multi record is needed; allocate and initialize it.
619 * We DO NOT perform an IGMP join as the in_ layer may need to
620 * push an initial source list down to IGMP to support SSM.
622 * The initial source filter state is INCLUDE, {} as per the RFC.
624 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
626 IF_ADDR_WUNLOCK(ifp);
627 IN_MULTI_LIST_UNLOCK();
628 if_delmulti_ifma(ifma);
631 inm->inm_addr = *group;
633 inm->inm_igi = ii->ii_igmp;
634 inm->inm_ifma = ifma;
635 inm->inm_refcount = 1;
636 inm->inm_state = IGMP_NOT_MEMBER;
637 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
638 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
639 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
640 RB_INIT(&inm->inm_srcs);
642 ifma->ifma_protospec = inm;
646 IF_ADDR_WUNLOCK(ifp);
647 IN_MULTI_LIST_UNLOCK();
652 * Drop a reference to an in_multi record.
654 * If the refcount drops to 0, free the in_multi record and
655 * delete the underlying link-layer membership.
658 inm_release(struct in_multi *inm)
660 struct ifmultiaddr *ifma;
663 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
664 MPASS(inm->inm_refcount == 0);
665 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
667 ifma = inm->inm_ifma;
670 /* XXX this access is not covered by IF_ADDR_LOCK */
671 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
673 CURVNET_SET(ifp->if_vnet);
675 free(inm, M_IPMADDR);
677 if_delmulti_ifma_flags(ifma, 1);
685 * Clear recorded source entries for a group.
686 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
687 * FIXME: Should reap.
690 inm_clear_recorded(struct in_multi *inm)
692 struct ip_msource *ims;
694 IN_MULTI_LIST_LOCK_ASSERT();
696 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
699 --inm->inm_st[1].iss_rec;
702 KASSERT(inm->inm_st[1].iss_rec == 0,
703 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
707 * Record a source as pending for a Source-Group IGMPv3 query.
708 * This lives here as it modifies the shared tree.
710 * inm is the group descriptor.
711 * naddr is the address of the source to record in network-byte order.
713 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
714 * lazy-allocate a source node in response to an SG query.
715 * Otherwise, no allocation is performed. This saves some memory
716 * with the trade-off that the source will not be reported to the
717 * router if joined in the window between the query response and
718 * the group actually being joined on the local host.
720 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
721 * This turns off the allocation of a recorded source entry if
722 * the group has not been joined.
724 * Return 0 if the source didn't exist or was already marked as recorded.
725 * Return 1 if the source was marked as recorded by this function.
726 * Return <0 if any error occurred (negated errno code).
729 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
731 struct ip_msource find;
732 struct ip_msource *ims, *nims;
734 IN_MULTI_LIST_LOCK_ASSERT();
736 find.ims_haddr = ntohl(naddr);
737 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
738 if (ims && ims->ims_stp)
741 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
743 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
747 nims->ims_haddr = find.ims_haddr;
748 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
754 * Mark the source as recorded and update the recorded
758 ++inm->inm_st[1].iss_rec;
764 * Return a pointer to an in_msource owned by an in_mfilter,
765 * given its source address.
766 * Lazy-allocate if needed. If this is a new entry its filter state is
769 * imf is the filter set being modified.
770 * haddr is the source address in *host* byte-order.
772 * SMPng: May be called with locks held; malloc must not block.
775 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
776 struct in_msource **plims)
778 struct ip_msource find;
779 struct ip_msource *ims, *nims;
780 struct in_msource *lims;
787 /* key is host byte order */
788 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
789 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
790 lims = (struct in_msource *)ims;
792 if (imf->imf_nsrc == in_mcast_maxsocksrc)
794 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
798 lims = (struct in_msource *)nims;
799 lims->ims_haddr = find.ims_haddr;
800 lims->imsl_st[0] = MCAST_UNDEFINED;
801 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
811 * Graft a source entry into an existing socket-layer filter set,
812 * maintaining any required invariants and checking allocations.
814 * The source is marked as being in the new filter mode at t1.
816 * Return the pointer to the new node, otherwise return NULL.
818 static struct in_msource *
819 imf_graft(struct in_mfilter *imf, const uint8_t st1,
820 const struct sockaddr_in *psin)
822 struct ip_msource *nims;
823 struct in_msource *lims;
825 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
829 lims = (struct in_msource *)nims;
830 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
831 lims->imsl_st[0] = MCAST_UNDEFINED;
832 lims->imsl_st[1] = st1;
833 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
840 * Prune a source entry from an existing socket-layer filter set,
841 * maintaining any required invariants and checking allocations.
843 * The source is marked as being left at t1, it is not freed.
845 * Return 0 if no error occurred, otherwise return an errno value.
848 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
850 struct ip_msource find;
851 struct ip_msource *ims;
852 struct in_msource *lims;
854 /* key is host byte order */
855 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
856 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
859 lims = (struct in_msource *)ims;
860 lims->imsl_st[1] = MCAST_UNDEFINED;
865 * Revert socket-layer filter set deltas at t1 to t0 state.
868 imf_rollback(struct in_mfilter *imf)
870 struct ip_msource *ims, *tims;
871 struct in_msource *lims;
873 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
874 lims = (struct in_msource *)ims;
875 if (lims->imsl_st[0] == lims->imsl_st[1]) {
876 /* no change at t1 */
878 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
879 /* revert change to existing source at t1 */
880 lims->imsl_st[1] = lims->imsl_st[0];
882 /* revert source added t1 */
883 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
884 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
885 free(ims, M_INMFILTER);
889 imf->imf_st[1] = imf->imf_st[0];
893 * Mark socket-layer filter set as INCLUDE {} at t1.
896 imf_leave(struct in_mfilter *imf)
898 struct ip_msource *ims;
899 struct in_msource *lims;
901 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
902 lims = (struct in_msource *)ims;
903 lims->imsl_st[1] = MCAST_UNDEFINED;
905 imf->imf_st[1] = MCAST_INCLUDE;
909 * Mark socket-layer filter set deltas as committed.
912 imf_commit(struct in_mfilter *imf)
914 struct ip_msource *ims;
915 struct in_msource *lims;
917 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
918 lims = (struct in_msource *)ims;
919 lims->imsl_st[0] = lims->imsl_st[1];
921 imf->imf_st[0] = imf->imf_st[1];
925 * Reap unreferenced sources from socket-layer filter set.
928 imf_reap(struct in_mfilter *imf)
930 struct ip_msource *ims, *tims;
931 struct in_msource *lims;
933 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
934 lims = (struct in_msource *)ims;
935 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
936 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
937 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
938 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
939 free(ims, M_INMFILTER);
946 * Purge socket-layer filter set.
949 imf_purge(struct in_mfilter *imf)
951 struct ip_msource *ims, *tims;
953 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
954 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
955 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
956 free(ims, M_INMFILTER);
959 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
960 KASSERT(RB_EMPTY(&imf->imf_sources),
961 ("%s: imf_sources not empty", __func__));
965 * Look up a source filter entry for a multicast group.
967 * inm is the group descriptor to work with.
968 * haddr is the host-byte-order IPv4 address to look up.
969 * noalloc may be non-zero to suppress allocation of sources.
970 * *pims will be set to the address of the retrieved or allocated source.
972 * SMPng: NOTE: may be called with locks held.
973 * Return 0 if successful, otherwise return a non-zero error code.
976 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
977 const int noalloc, struct ip_msource **pims)
979 struct ip_msource find;
980 struct ip_msource *ims, *nims;
982 find.ims_haddr = haddr;
983 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
984 if (ims == NULL && !noalloc) {
985 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
987 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
991 nims->ims_haddr = haddr;
992 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
996 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
1006 * Merge socket-layer source into IGMP-layer source.
1007 * If rollback is non-zero, perform the inverse of the merge.
1010 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
1013 int n = rollback ? -1 : 1;
1015 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
1016 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
1017 __func__, n, ims->ims_haddr);
1018 ims->ims_st[1].ex -= n;
1019 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
1020 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
1021 __func__, n, ims->ims_haddr);
1022 ims->ims_st[1].in -= n;
1025 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
1026 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
1027 __func__, n, ims->ims_haddr);
1028 ims->ims_st[1].ex += n;
1029 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
1030 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
1031 __func__, n, ims->ims_haddr);
1032 ims->ims_st[1].in += n;
1037 * Atomically update the global in_multi state, when a membership's
1038 * filter list is being updated in any way.
1040 * imf is the per-inpcb-membership group filter pointer.
1041 * A fake imf may be passed for in-kernel consumers.
1043 * XXX This is a candidate for a set-symmetric-difference style loop
1044 * which would eliminate the repeated lookup from root of ims nodes,
1045 * as they share the same key space.
1047 * If any error occurred this function will back out of refcounts
1048 * and return a non-zero value.
1051 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1053 struct ip_msource *ims, *nims;
1054 struct in_msource *lims;
1055 int schanged, error;
1061 IN_MULTI_LIST_LOCK_ASSERT();
1064 * Update the source filters first, as this may fail.
1065 * Maintain count of in-mode filters at t0, t1. These are
1066 * used to work out if we transition into ASM mode or not.
1067 * Maintain a count of source filters whose state was
1068 * actually modified by this operation.
1070 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1071 lims = (struct in_msource *)ims;
1072 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1073 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1074 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1075 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1079 ims_merge(nims, lims, 0);
1082 struct ip_msource *bims;
1084 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1085 lims = (struct in_msource *)ims;
1086 if (lims->imsl_st[0] == lims->imsl_st[1])
1088 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1091 ims_merge(bims, lims, 1);
1096 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1097 __func__, nsrc0, nsrc1);
1099 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1100 if (imf->imf_st[0] == imf->imf_st[1] &&
1101 imf->imf_st[1] == MCAST_INCLUDE) {
1103 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1104 --inm->inm_st[1].iss_in;
1108 /* Handle filter mode transition on socket. */
1109 if (imf->imf_st[0] != imf->imf_st[1]) {
1110 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1111 __func__, imf->imf_st[0], imf->imf_st[1]);
1113 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1114 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1115 --inm->inm_st[1].iss_ex;
1116 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1117 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1118 --inm->inm_st[1].iss_in;
1121 if (imf->imf_st[1] == 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[1] == MCAST_INCLUDE && nsrc1 > 0) {
1125 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1126 inm->inm_st[1].iss_in++;
1131 * Track inm filter state in terms of listener counts.
1132 * If there are any exclusive listeners, stack-wide
1133 * membership is exclusive.
1134 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1135 * If no listeners remain, state is undefined at t1,
1136 * and the IGMP lifecycle for this group should finish.
1138 if (inm->inm_st[1].iss_ex > 0) {
1139 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1140 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1141 } else if (inm->inm_st[1].iss_in > 0) {
1142 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1143 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1145 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1146 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1149 /* Decrement ASM listener count on transition out of ASM mode. */
1150 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1151 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1152 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1153 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1154 --inm->inm_st[1].iss_asm;
1158 /* Increment ASM listener count on transition to ASM mode. */
1159 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1160 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1161 inm->inm_st[1].iss_asm++;
1164 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1169 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1176 * Mark an in_multi's filter set deltas as committed.
1177 * Called by IGMP after a state change has been enqueued.
1180 inm_commit(struct in_multi *inm)
1182 struct ip_msource *ims;
1184 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1185 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1188 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1189 ims->ims_st[0] = ims->ims_st[1];
1191 inm->inm_st[0] = inm->inm_st[1];
1195 * Reap unreferenced nodes from an in_multi's filter set.
1198 inm_reap(struct in_multi *inm)
1200 struct ip_msource *ims, *tims;
1202 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1203 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1204 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1207 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1208 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1209 free(ims, M_IPMSOURCE);
1215 * Purge all source nodes from an in_multi's filter set.
1218 inm_purge(struct in_multi *inm)
1220 struct ip_msource *ims, *tims;
1222 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1223 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1224 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1225 free(ims, M_IPMSOURCE);
1231 * Join a multicast group; unlocked entry point.
1233 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1234 * is not held. Fortunately, ifp is unlikely to have been detached
1235 * at this point, so we assume it's OK to recurse.
1238 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1239 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1244 error = in_joingroup_locked(ifp, gina, imf, pinm);
1251 * Join a multicast group; real entry point.
1253 * Only preserves atomicity at inm level.
1254 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1256 * If the IGMP downcall fails, the group is not joined, and an error
1260 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1261 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1263 struct in_mfilter timf;
1264 struct in_multi *inm;
1267 IN_MULTI_LOCK_ASSERT();
1268 IN_MULTI_LIST_UNLOCK_ASSERT();
1270 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1271 ntohl(gina->s_addr), ifp, ifp->if_xname);
1277 * If no imf was specified (i.e. kernel consumer),
1278 * fake one up and assume it is an ASM join.
1281 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1285 error = in_getmulti(ifp, gina, &inm);
1287 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1290 IN_MULTI_LIST_LOCK();
1291 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1292 error = inm_merge(inm, imf);
1294 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1295 goto out_inm_release;
1298 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1299 error = igmp_change_state(inm);
1301 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1302 goto out_inm_release;
1308 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1309 inm_release_deferred(inm);
1313 IN_MULTI_LIST_UNLOCK();
1319 * Leave a multicast group; unlocked entry point.
1322 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1327 error = in_leavegroup_locked(inm, imf);
1334 * Leave a multicast group; real entry point.
1335 * All source filters will be expunged.
1337 * Only preserves atomicity at inm level.
1339 * Holding the write lock for the INP which contains imf
1340 * is highly advisable. We can't assert for it as imf does not
1341 * contain a back-pointer to the owning inp.
1343 * Note: This is not the same as inm_release(*) as this function also
1344 * makes a state change downcall into IGMP.
1347 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1349 struct in_mfilter timf;
1354 IN_MULTI_LOCK_ASSERT();
1355 IN_MULTI_LIST_UNLOCK_ASSERT();
1357 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1358 inm, ntohl(inm->inm_addr.s_addr),
1359 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1363 * If no imf was specified (i.e. kernel consumer),
1364 * fake one up and assume it is an ASM join.
1367 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1372 * Begin state merge transaction at IGMP layer.
1374 * As this particular invocation should not cause any memory
1375 * to be allocated, and there is no opportunity to roll back
1376 * the transaction, it MUST NOT fail.
1378 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1379 IN_MULTI_LIST_LOCK();
1380 error = inm_merge(inm, imf);
1381 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1383 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1384 CURVNET_SET(inm->inm_ifp->if_vnet);
1385 error = igmp_change_state(inm);
1386 IF_ADDR_WLOCK(inm->inm_ifp);
1387 inm_release_deferred(inm);
1388 IF_ADDR_WUNLOCK(inm->inm_ifp);
1389 IN_MULTI_LIST_UNLOCK();
1392 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1394 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1399 /*#ifndef BURN_BRIDGES*/
1401 * Join an IPv4 multicast group in (*,G) exclusive mode.
1402 * The group must be a 224.0.0.0/24 link-scope group.
1403 * This KPI is for legacy kernel consumers only.
1406 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1408 struct in_multi *pinm;
1411 char addrbuf[INET_ADDRSTRLEN];
1414 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1415 ("%s: %s not in 224.0.0.0/24", __func__,
1416 inet_ntoa_r(*ap, addrbuf)));
1418 error = in_joingroup(ifp, ap, NULL, &pinm);
1426 * Block or unblock an ASM multicast source on an inpcb.
1427 * This implements the delta-based API described in RFC 3678.
1429 * The delta-based API applies only to exclusive-mode memberships.
1430 * An IGMP downcall will be performed.
1432 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1434 * Return 0 if successful, otherwise return an appropriate error code.
1437 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1439 struct group_source_req gsr;
1440 sockunion_t *gsa, *ssa;
1442 struct in_mfilter *imf;
1443 struct ip_moptions *imo;
1444 struct in_msource *ims;
1445 struct in_multi *inm;
1454 memset(&gsr, 0, sizeof(struct group_source_req));
1455 gsa = (sockunion_t *)&gsr.gsr_group;
1456 ssa = (sockunion_t *)&gsr.gsr_source;
1458 switch (sopt->sopt_name) {
1459 case IP_BLOCK_SOURCE:
1460 case IP_UNBLOCK_SOURCE: {
1461 struct ip_mreq_source mreqs;
1463 error = sooptcopyin(sopt, &mreqs,
1464 sizeof(struct ip_mreq_source),
1465 sizeof(struct ip_mreq_source));
1469 gsa->sin.sin_family = AF_INET;
1470 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1471 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1473 ssa->sin.sin_family = AF_INET;
1474 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1475 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1477 if (!in_nullhost(mreqs.imr_interface))
1478 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1480 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1483 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1484 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1488 case MCAST_BLOCK_SOURCE:
1489 case MCAST_UNBLOCK_SOURCE:
1490 error = sooptcopyin(sopt, &gsr,
1491 sizeof(struct group_source_req),
1492 sizeof(struct group_source_req));
1496 if (gsa->sin.sin_family != AF_INET ||
1497 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1500 if (ssa->sin.sin_family != AF_INET ||
1501 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1504 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1505 return (EADDRNOTAVAIL);
1507 ifp = ifnet_byindex(gsr.gsr_interface);
1509 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1514 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1515 __func__, sopt->sopt_name);
1516 return (EOPNOTSUPP);
1520 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1524 * Check if we are actually a member of this group.
1526 imo = inp_findmoptions(inp);
1527 idx = imo_match_group(imo, ifp, &gsa->sa);
1528 if (idx == -1 || imo->imo_mfilters == NULL) {
1529 error = EADDRNOTAVAIL;
1530 goto out_inp_locked;
1533 KASSERT(imo->imo_mfilters != NULL,
1534 ("%s: imo_mfilters not allocated", __func__));
1535 imf = &imo->imo_mfilters[idx];
1536 inm = imo->imo_membership[idx];
1539 * Attempting to use the delta-based API on an
1540 * non exclusive-mode membership is an error.
1542 fmode = imf->imf_st[0];
1543 if (fmode != MCAST_EXCLUDE) {
1545 goto out_inp_locked;
1549 * Deal with error cases up-front:
1550 * Asked to block, but already blocked; or
1551 * Asked to unblock, but nothing to unblock.
1552 * If adding a new block entry, allocate it.
1554 ims = imo_match_source(imo, idx, &ssa->sa);
1555 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1556 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1557 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1558 error = EADDRNOTAVAIL;
1559 goto out_inp_locked;
1562 INP_WLOCK_ASSERT(inp);
1565 * Begin state merge transaction at socket layer.
1568 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1569 ims = imf_graft(imf, fmode, &ssa->sin);
1573 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1574 error = imf_prune(imf, &ssa->sin);
1578 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1579 goto out_imf_rollback;
1583 * Begin state merge transaction at IGMP layer.
1586 IN_MULTI_LIST_LOCK();
1587 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1588 error = inm_merge(inm, imf);
1590 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1591 goto out_in_multi_locked;
1594 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1595 error = igmp_change_state(inm);
1597 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1599 out_in_multi_locked:
1617 * Given an inpcb, return its multicast options structure pointer. Accepts
1618 * an unlocked inpcb pointer, but will return it locked. May sleep.
1620 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1621 * SMPng: NOTE: Returns with the INP write lock held.
1623 static struct ip_moptions *
1624 inp_findmoptions(struct inpcb *inp)
1626 struct ip_moptions *imo;
1627 struct in_multi **immp;
1628 struct in_mfilter *imfp;
1632 if (inp->inp_moptions != NULL)
1633 return (inp->inp_moptions);
1637 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1638 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1640 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1641 M_INMFILTER, M_WAITOK);
1643 imo->imo_multicast_ifp = NULL;
1644 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1645 imo->imo_multicast_vif = -1;
1646 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1647 imo->imo_multicast_loop = in_mcast_loop;
1648 imo->imo_num_memberships = 0;
1649 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1650 imo->imo_membership = immp;
1652 /* Initialize per-group source filters. */
1653 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1654 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1655 imo->imo_mfilters = imfp;
1658 if (inp->inp_moptions != NULL) {
1659 free(imfp, M_INMFILTER);
1660 free(immp, M_IPMOPTS);
1661 free(imo, M_IPMOPTS);
1662 return (inp->inp_moptions);
1664 inp->inp_moptions = imo;
1669 * Discard the IP multicast options (and source filters). To minimize
1670 * the amount of work done while holding locks such as the INP's
1671 * pcbinfo lock (which is used in the receive path), the free
1672 * operation is performed asynchronously in a separate task.
1674 * SMPng: NOTE: assumes INP write lock is held.
1677 inp_freemoptions(struct ip_moptions *imo, struct inpcbinfo *pcbinfo)
1684 INP_INFO_LOCK_ASSERT(pcbinfo);
1685 wlock = INP_INFO_WLOCKED(pcbinfo);
1687 INP_INFO_WUNLOCK(pcbinfo);
1689 INP_INFO_RUNLOCK(pcbinfo);
1691 KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1692 IN_MULTI_LIST_LOCK();
1693 STAILQ_INSERT_TAIL(&imo_gc_list, imo, imo_link);
1694 IN_MULTI_LIST_UNLOCK();
1695 taskqueue_enqueue(taskqueue_thread, &imo_gc_task);
1697 INP_INFO_WLOCK(pcbinfo);
1699 INP_INFO_RLOCK(pcbinfo);
1703 inp_freemoptions_internal(struct ip_moptions *imo)
1705 struct in_mfilter *imf;
1706 size_t idx, nmships;
1708 nmships = imo->imo_num_memberships;
1709 for (idx = 0; idx < nmships; ++idx) {
1710 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1713 (void)in_leavegroup(imo->imo_membership[idx], imf);
1718 if (imo->imo_mfilters)
1719 free(imo->imo_mfilters, M_INMFILTER);
1720 free(imo->imo_membership, M_IPMOPTS);
1721 free(imo, M_IPMOPTS);
1725 inp_gcmoptions(void *context, int pending)
1727 struct ip_moptions *imo;
1729 IN_MULTI_LIST_LOCK();
1730 while (!STAILQ_EMPTY(&imo_gc_list)) {
1731 imo = STAILQ_FIRST(&imo_gc_list);
1732 STAILQ_REMOVE_HEAD(&imo_gc_list, imo_link);
1733 IN_MULTI_LIST_UNLOCK();
1734 inp_freemoptions_internal(imo);
1735 IN_MULTI_LIST_LOCK();
1737 IN_MULTI_LIST_UNLOCK();
1741 * Atomically get source filters on a socket for an IPv4 multicast group.
1742 * Called with INP lock held; returns with lock released.
1745 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1747 struct __msfilterreq msfr;
1750 struct ip_moptions *imo;
1751 struct in_mfilter *imf;
1752 struct ip_msource *ims;
1753 struct in_msource *lims;
1754 struct sockaddr_in *psin;
1755 struct sockaddr_storage *ptss;
1756 struct sockaddr_storage *tss;
1758 size_t idx, nsrcs, ncsrcs;
1760 INP_WLOCK_ASSERT(inp);
1762 imo = inp->inp_moptions;
1763 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1767 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1768 sizeof(struct __msfilterreq));
1772 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1775 ifp = ifnet_byindex(msfr.msfr_ifindex);
1782 * Lookup group on the socket.
1784 gsa = (sockunion_t *)&msfr.msfr_group;
1785 idx = imo_match_group(imo, ifp, &gsa->sa);
1786 if (idx == -1 || imo->imo_mfilters == NULL) {
1788 return (EADDRNOTAVAIL);
1790 imf = &imo->imo_mfilters[idx];
1793 * Ignore memberships which are in limbo.
1795 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1799 msfr.msfr_fmode = imf->imf_st[1];
1802 * If the user specified a buffer, copy out the source filter
1803 * entries to userland gracefully.
1804 * We only copy out the number of entries which userland
1805 * has asked for, but we always tell userland how big the
1806 * buffer really needs to be.
1808 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1809 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1811 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1812 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1813 M_TEMP, M_NOWAIT | M_ZERO);
1821 * Count number of sources in-mode at t0.
1822 * If buffer space exists and remains, copy out source entries.
1824 nsrcs = msfr.msfr_nsrcs;
1827 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1828 lims = (struct in_msource *)ims;
1829 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1830 lims->imsl_st[0] != imf->imf_st[0])
1833 if (tss != NULL && nsrcs > 0) {
1834 psin = (struct sockaddr_in *)ptss;
1835 psin->sin_family = AF_INET;
1836 psin->sin_len = sizeof(struct sockaddr_in);
1837 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1847 error = copyout(tss, msfr.msfr_srcs,
1848 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1854 msfr.msfr_nsrcs = ncsrcs;
1855 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1861 * Return the IP multicast options in response to user getsockopt().
1864 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1866 struct rm_priotracker in_ifa_tracker;
1867 struct ip_mreqn mreqn;
1868 struct ip_moptions *imo;
1870 struct in_ifaddr *ia;
1875 imo = inp->inp_moptions;
1877 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1878 * or is a divert socket, reject it.
1880 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1881 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1882 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1884 return (EOPNOTSUPP);
1888 switch (sopt->sopt_name) {
1889 case IP_MULTICAST_VIF:
1891 optval = imo->imo_multicast_vif;
1895 error = sooptcopyout(sopt, &optval, sizeof(int));
1898 case IP_MULTICAST_IF:
1899 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1901 ifp = imo->imo_multicast_ifp;
1902 if (!in_nullhost(imo->imo_multicast_addr)) {
1903 mreqn.imr_address = imo->imo_multicast_addr;
1904 } else if (ifp != NULL) {
1905 mreqn.imr_ifindex = ifp->if_index;
1906 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1909 IA_SIN(ia)->sin_addr;
1910 ifa_free(&ia->ia_ifa);
1915 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1916 error = sooptcopyout(sopt, &mreqn,
1917 sizeof(struct ip_mreqn));
1919 error = sooptcopyout(sopt, &mreqn.imr_address,
1920 sizeof(struct in_addr));
1924 case IP_MULTICAST_TTL:
1926 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1928 optval = coptval = imo->imo_multicast_ttl;
1930 if (sopt->sopt_valsize == sizeof(u_char))
1931 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1933 error = sooptcopyout(sopt, &optval, sizeof(int));
1936 case IP_MULTICAST_LOOP:
1938 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1940 optval = coptval = imo->imo_multicast_loop;
1942 if (sopt->sopt_valsize == sizeof(u_char))
1943 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1945 error = sooptcopyout(sopt, &optval, sizeof(int));
1950 error = EADDRNOTAVAIL;
1953 error = inp_get_source_filters(inp, sopt);
1959 error = ENOPROTOOPT;
1963 INP_UNLOCK_ASSERT(inp);
1969 * Look up the ifnet to use for a multicast group membership,
1970 * given the IPv4 address of an interface, and the IPv4 group address.
1972 * This routine exists to support legacy multicast applications
1973 * which do not understand that multicast memberships are scoped to
1974 * specific physical links in the networking stack, or which need
1975 * to join link-scope groups before IPv4 addresses are configured.
1977 * If inp is non-NULL, use this socket's current FIB number for any
1978 * required FIB lookup.
1979 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1980 * and use its ifp; usually, this points to the default next-hop.
1982 * If the FIB lookup fails, attempt to use the first non-loopback
1983 * interface with multicast capability in the system as a
1984 * last resort. The legacy IPv4 ASM API requires that we do
1985 * this in order to allow groups to be joined when the routing
1986 * table has not yet been populated during boot.
1988 * Returns NULL if no ifp could be found.
1990 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1991 * FUTURE: Implement IPv4 source-address selection.
1993 static struct ifnet *
1994 inp_lookup_mcast_ifp(const struct inpcb *inp,
1995 const struct sockaddr_in *gsin, const struct in_addr ina)
1997 struct rm_priotracker in_ifa_tracker;
1999 struct nhop4_basic nh4;
2002 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
2003 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
2004 ("%s: not multicast", __func__));
2007 if (!in_nullhost(ina)) {
2008 INADDR_TO_IFP(ina, ifp);
2010 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
2011 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
2014 struct in_ifaddr *ia;
2018 IN_IFADDR_RLOCK(&in_ifa_tracker);
2019 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
2021 if (!(mifp->if_flags & IFF_LOOPBACK) &&
2022 (mifp->if_flags & IFF_MULTICAST)) {
2027 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2035 * Join an IPv4 multicast group, possibly with a source.
2038 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
2040 struct group_source_req gsr;
2041 sockunion_t *gsa, *ssa;
2043 struct in_mfilter *imf;
2044 struct ip_moptions *imo;
2045 struct in_multi *inm;
2046 struct in_msource *lims;
2056 memset(&gsr, 0, sizeof(struct group_source_req));
2057 gsa = (sockunion_t *)&gsr.gsr_group;
2058 gsa->ss.ss_family = AF_UNSPEC;
2059 ssa = (sockunion_t *)&gsr.gsr_source;
2060 ssa->ss.ss_family = AF_UNSPEC;
2062 switch (sopt->sopt_name) {
2063 case IP_ADD_MEMBERSHIP:
2064 case IP_ADD_SOURCE_MEMBERSHIP: {
2065 struct ip_mreq_source mreqs;
2067 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
2068 error = sooptcopyin(sopt, &mreqs,
2069 sizeof(struct ip_mreq),
2070 sizeof(struct ip_mreq));
2072 * Do argument switcharoo from ip_mreq into
2073 * ip_mreq_source to avoid using two instances.
2075 mreqs.imr_interface = mreqs.imr_sourceaddr;
2076 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2077 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2078 error = sooptcopyin(sopt, &mreqs,
2079 sizeof(struct ip_mreq_source),
2080 sizeof(struct ip_mreq_source));
2085 gsa->sin.sin_family = AF_INET;
2086 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2087 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2089 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2090 ssa->sin.sin_family = AF_INET;
2091 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2092 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2095 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2098 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2099 mreqs.imr_interface);
2100 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2101 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2105 case MCAST_JOIN_GROUP:
2106 case MCAST_JOIN_SOURCE_GROUP:
2107 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2108 error = sooptcopyin(sopt, &gsr,
2109 sizeof(struct group_req),
2110 sizeof(struct group_req));
2111 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2112 error = sooptcopyin(sopt, &gsr,
2113 sizeof(struct group_source_req),
2114 sizeof(struct group_source_req));
2119 if (gsa->sin.sin_family != AF_INET ||
2120 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2124 * Overwrite the port field if present, as the sockaddr
2125 * being copied in may be matched with a binary comparison.
2127 gsa->sin.sin_port = 0;
2128 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2129 if (ssa->sin.sin_family != AF_INET ||
2130 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2132 ssa->sin.sin_port = 0;
2135 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2138 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2139 return (EADDRNOTAVAIL);
2140 ifp = ifnet_byindex(gsr.gsr_interface);
2144 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2145 __func__, sopt->sopt_name);
2146 return (EOPNOTSUPP);
2150 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2151 return (EADDRNOTAVAIL);
2153 imo = inp_findmoptions(inp);
2154 idx = imo_match_group(imo, ifp, &gsa->sa);
2158 inm = imo->imo_membership[idx];
2159 imf = &imo->imo_mfilters[idx];
2160 if (ssa->ss.ss_family != AF_UNSPEC) {
2162 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2163 * is an error. On an existing inclusive membership,
2164 * it just adds the source to the filter list.
2166 if (imf->imf_st[1] != MCAST_INCLUDE) {
2168 goto out_inp_locked;
2171 * Throw out duplicates.
2173 * XXX FIXME: This makes a naive assumption that
2174 * even if entries exist for *ssa in this imf,
2175 * they will be rejected as dupes, even if they
2176 * are not valid in the current mode (in-mode).
2178 * in_msource is transactioned just as for anything
2179 * else in SSM -- but note naive use of inm_graft()
2180 * below for allocating new filter entries.
2182 * This is only an issue if someone mixes the
2183 * full-state SSM API with the delta-based API,
2184 * which is discouraged in the relevant RFCs.
2186 lims = imo_match_source(imo, idx, &ssa->sa);
2187 if (lims != NULL /*&&
2188 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2189 error = EADDRNOTAVAIL;
2190 goto out_inp_locked;
2194 * MCAST_JOIN_GROUP on an existing exclusive
2195 * membership is an error; return EADDRINUSE
2196 * to preserve 4.4BSD API idempotence, and
2197 * avoid tedious detour to code below.
2198 * NOTE: This is bending RFC 3678 a bit.
2200 * On an existing inclusive membership, this is also
2201 * an error; if you want to change filter mode,
2202 * you must use the userland API setsourcefilter().
2203 * XXX We don't reject this for imf in UNDEFINED
2204 * state at t1, because allocation of a filter
2205 * is atomic with allocation of a membership.
2208 if (imf->imf_st[1] == MCAST_EXCLUDE)
2210 goto out_inp_locked;
2215 * Begin state merge transaction at socket layer.
2217 INP_WLOCK_ASSERT(inp);
2220 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2221 error = imo_grow(imo);
2223 goto out_inp_locked;
2226 * Allocate the new slot upfront so we can deal with
2227 * grafting the new source filter in same code path
2228 * as for join-source on existing membership.
2230 idx = imo->imo_num_memberships;
2231 imo->imo_membership[idx] = NULL;
2232 imo->imo_num_memberships++;
2233 KASSERT(imo->imo_mfilters != NULL,
2234 ("%s: imf_mfilters vector was not allocated", __func__));
2235 imf = &imo->imo_mfilters[idx];
2236 KASSERT(RB_EMPTY(&imf->imf_sources),
2237 ("%s: imf_sources not empty", __func__));
2241 * Graft new source into filter list for this inpcb's
2242 * membership of the group. The in_multi may not have
2243 * been allocated yet if this is a new membership, however,
2244 * the in_mfilter slot will be allocated and must be initialized.
2246 * Note: Grafting of exclusive mode filters doesn't happen
2248 * XXX: Should check for non-NULL lims (node exists but may
2249 * not be in-mode) for interop with full-state API.
2251 if (ssa->ss.ss_family != AF_UNSPEC) {
2252 /* Membership starts in IN mode */
2254 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2255 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2257 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2259 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2261 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2267 /* No address specified; Membership starts in EX mode */
2269 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2270 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2275 * Begin state merge transaction at IGMP layer.
2282 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2285 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2287 IN_MULTI_LIST_UNLOCK();
2290 imo->imo_membership[idx] = inm;
2292 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2293 IN_MULTI_LIST_LOCK();
2294 error = inm_merge(inm, imf);
2296 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2298 IN_MULTI_LIST_UNLOCK();
2299 goto out_in_multi_locked;
2301 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2302 error = igmp_change_state(inm);
2303 IN_MULTI_LIST_UNLOCK();
2305 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2307 goto out_in_multi_locked;
2311 out_in_multi_locked:
2315 if (in_pcbrele_wlocked(inp))
2328 if (error && is_new) {
2329 imo->imo_membership[idx] = NULL;
2330 --imo->imo_num_memberships;
2339 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2342 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2344 struct group_source_req gsr;
2345 struct ip_mreq_source mreqs;
2346 sockunion_t *gsa, *ssa;
2348 struct in_mfilter *imf;
2349 struct ip_moptions *imo;
2350 struct in_msource *ims;
2351 struct in_multi *inm;
2353 int error, is_final;
2359 memset(&gsr, 0, sizeof(struct group_source_req));
2360 gsa = (sockunion_t *)&gsr.gsr_group;
2361 gsa->ss.ss_family = AF_UNSPEC;
2362 ssa = (sockunion_t *)&gsr.gsr_source;
2363 ssa->ss.ss_family = AF_UNSPEC;
2365 switch (sopt->sopt_name) {
2366 case IP_DROP_MEMBERSHIP:
2367 case IP_DROP_SOURCE_MEMBERSHIP:
2368 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2369 error = sooptcopyin(sopt, &mreqs,
2370 sizeof(struct ip_mreq),
2371 sizeof(struct ip_mreq));
2373 * Swap interface and sourceaddr arguments,
2374 * as ip_mreq and ip_mreq_source are laid
2377 mreqs.imr_interface = mreqs.imr_sourceaddr;
2378 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2379 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2380 error = sooptcopyin(sopt, &mreqs,
2381 sizeof(struct ip_mreq_source),
2382 sizeof(struct ip_mreq_source));
2387 gsa->sin.sin_family = AF_INET;
2388 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2389 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2391 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2392 ssa->sin.sin_family = AF_INET;
2393 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2394 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2398 * Attempt to look up hinted ifp from interface address.
2399 * Fallthrough with null ifp iff lookup fails, to
2400 * preserve 4.4BSD mcast API idempotence.
2401 * XXX NOTE WELL: The RFC 3678 API is preferred because
2402 * using an IPv4 address as a key is racy.
2404 if (!in_nullhost(mreqs.imr_interface))
2405 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2407 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2408 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2412 case MCAST_LEAVE_GROUP:
2413 case MCAST_LEAVE_SOURCE_GROUP:
2414 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2415 error = sooptcopyin(sopt, &gsr,
2416 sizeof(struct group_req),
2417 sizeof(struct group_req));
2418 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2419 error = sooptcopyin(sopt, &gsr,
2420 sizeof(struct group_source_req),
2421 sizeof(struct group_source_req));
2426 if (gsa->sin.sin_family != AF_INET ||
2427 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2430 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2431 if (ssa->sin.sin_family != AF_INET ||
2432 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2436 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2437 return (EADDRNOTAVAIL);
2439 ifp = ifnet_byindex(gsr.gsr_interface);
2442 return (EADDRNOTAVAIL);
2446 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2447 __func__, sopt->sopt_name);
2448 return (EOPNOTSUPP);
2452 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2456 * Find the membership in the membership array.
2458 imo = inp_findmoptions(inp);
2459 idx = imo_match_group(imo, ifp, &gsa->sa);
2461 error = EADDRNOTAVAIL;
2462 goto out_inp_locked;
2464 inm = imo->imo_membership[idx];
2465 imf = &imo->imo_mfilters[idx];
2467 if (ssa->ss.ss_family != AF_UNSPEC)
2471 * Begin state merge transaction at socket layer.
2473 INP_WLOCK_ASSERT(inp);
2476 * If we were instructed only to leave a given source, do so.
2477 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2482 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2483 error = EADDRNOTAVAIL;
2484 goto out_inp_locked;
2486 ims = imo_match_source(imo, idx, &ssa->sa);
2488 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2489 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2490 error = EADDRNOTAVAIL;
2491 goto out_inp_locked;
2493 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2494 error = imf_prune(imf, &ssa->sin);
2496 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2498 goto out_inp_locked;
2503 * Begin state merge transaction at IGMP layer.
2511 * Give up the multicast address record to which
2512 * the membership points.
2514 (void)in_leavegroup_locked(inm, imf);
2516 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2517 IN_MULTI_LIST_LOCK();
2518 error = inm_merge(inm, imf);
2520 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
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 in_addr addr;
2574 struct ip_mreqn mreqn;
2576 struct ip_moptions *imo;
2579 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2581 * An interface index was specified using the
2582 * Linux-derived ip_mreqn structure.
2584 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2585 sizeof(struct ip_mreqn));
2589 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2592 if (mreqn.imr_ifindex == 0) {
2595 ifp = ifnet_byindex(mreqn.imr_ifindex);
2597 return (EADDRNOTAVAIL);
2601 * An interface was specified by IPv4 address.
2602 * This is the traditional BSD usage.
2604 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2605 sizeof(struct in_addr));
2608 if (in_nullhost(addr)) {
2611 INADDR_TO_IFP(addr, ifp);
2613 return (EADDRNOTAVAIL);
2615 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2616 ntohl(addr.s_addr));
2619 /* Reject interfaces which do not support multicast. */
2620 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2621 return (EOPNOTSUPP);
2623 imo = inp_findmoptions(inp);
2624 imo->imo_multicast_ifp = ifp;
2625 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2632 * Atomically set source filters on a socket for an IPv4 multicast group.
2634 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2637 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2639 struct __msfilterreq msfr;
2642 struct in_mfilter *imf;
2643 struct ip_moptions *imo;
2644 struct in_multi *inm;
2648 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2649 sizeof(struct __msfilterreq));
2653 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2656 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2657 msfr.msfr_fmode != MCAST_INCLUDE))
2660 if (msfr.msfr_group.ss_family != AF_INET ||
2661 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2664 gsa = (sockunion_t *)&msfr.msfr_group;
2665 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2668 gsa->sin.sin_port = 0; /* ignore port */
2670 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2671 return (EADDRNOTAVAIL);
2673 ifp = ifnet_byindex(msfr.msfr_ifindex);
2675 return (EADDRNOTAVAIL);
2678 * Take the INP write lock.
2679 * Check if this socket is a member of this group.
2681 imo = inp_findmoptions(inp);
2682 idx = imo_match_group(imo, ifp, &gsa->sa);
2683 if (idx == -1 || imo->imo_mfilters == NULL) {
2684 error = EADDRNOTAVAIL;
2685 goto out_inp_locked;
2687 inm = imo->imo_membership[idx];
2688 imf = &imo->imo_mfilters[idx];
2691 * Begin state merge transaction at socket layer.
2693 INP_WLOCK_ASSERT(inp);
2695 imf->imf_st[1] = msfr.msfr_fmode;
2698 * Apply any new source filters, if present.
2699 * Make a copy of the user-space source vector so
2700 * that we may copy them with a single copyin. This
2701 * allows us to deal with page faults up-front.
2703 if (msfr.msfr_nsrcs > 0) {
2704 struct in_msource *lims;
2705 struct sockaddr_in *psin;
2706 struct sockaddr_storage *kss, *pkss;
2711 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2712 __func__, (unsigned long)msfr.msfr_nsrcs);
2713 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2715 error = copyin(msfr.msfr_srcs, kss,
2716 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2725 * Mark all source filters as UNDEFINED at t1.
2726 * Restore new group filter mode, as imf_leave()
2727 * will set it to INCLUDE.
2730 imf->imf_st[1] = msfr.msfr_fmode;
2733 * Update socket layer filters at t1, lazy-allocating
2734 * new entries. This saves a bunch of memory at the
2735 * cost of one RB_FIND() per source entry; duplicate
2736 * entries in the msfr_nsrcs vector are ignored.
2737 * If we encounter an error, rollback transaction.
2739 * XXX This too could be replaced with a set-symmetric
2740 * difference like loop to avoid walking from root
2741 * every time, as the key space is common.
2743 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2744 psin = (struct sockaddr_in *)pkss;
2745 if (psin->sin_family != AF_INET) {
2746 error = EAFNOSUPPORT;
2749 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2753 error = imf_get_source(imf, psin, &lims);
2756 lims->imsl_st[1] = imf->imf_st[1];
2762 goto out_imf_rollback;
2764 INP_WLOCK_ASSERT(inp);
2766 IN_MULTI_LIST_LOCK();
2769 * Begin state merge transaction at IGMP layer.
2771 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2772 error = inm_merge(inm, imf);
2774 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2775 IN_MULTI_LIST_UNLOCK();
2776 goto out_in_multi_locked;
2779 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2780 error = igmp_change_state(inm);
2781 IN_MULTI_LIST_UNLOCK();
2783 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2785 out_in_multi_locked:
2803 * Set the IP multicast options in response to user setsockopt().
2805 * Many of the socket options handled in this function duplicate the
2806 * functionality of socket options in the regular unicast API. However,
2807 * it is not possible to merge the duplicate code, because the idempotence
2808 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2809 * the effects of these options must be treated as separate and distinct.
2811 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2812 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2813 * is refactored to no longer use vifs.
2816 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2818 struct ip_moptions *imo;
2824 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2825 * or is a divert socket, reject it.
2827 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2828 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2829 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2830 return (EOPNOTSUPP);
2832 switch (sopt->sopt_name) {
2833 case IP_MULTICAST_VIF: {
2836 * Select a multicast VIF for transmission.
2837 * Only useful if multicast forwarding is active.
2839 if (legal_vif_num == NULL) {
2843 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2846 if (!legal_vif_num(vifi) && (vifi != -1)) {
2850 imo = inp_findmoptions(inp);
2851 imo->imo_multicast_vif = vifi;
2856 case IP_MULTICAST_IF:
2857 error = inp_set_multicast_if(inp, sopt);
2860 case IP_MULTICAST_TTL: {
2864 * Set the IP time-to-live for outgoing multicast packets.
2865 * The original multicast API required a char argument,
2866 * which is inconsistent with the rest of the socket API.
2867 * We allow either a char or an int.
2869 if (sopt->sopt_valsize == sizeof(u_char)) {
2870 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2877 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2887 imo = inp_findmoptions(inp);
2888 imo->imo_multicast_ttl = ttl;
2893 case IP_MULTICAST_LOOP: {
2897 * Set the loopback flag for outgoing multicast packets.
2898 * Must be zero or one. The original multicast API required a
2899 * char argument, which is inconsistent with the rest
2900 * of the socket API. We allow either a char or an int.
2902 if (sopt->sopt_valsize == sizeof(u_char)) {
2903 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2910 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2914 loop = (u_char)iloop;
2916 imo = inp_findmoptions(inp);
2917 imo->imo_multicast_loop = !!loop;
2922 case IP_ADD_MEMBERSHIP:
2923 case IP_ADD_SOURCE_MEMBERSHIP:
2924 case MCAST_JOIN_GROUP:
2925 case MCAST_JOIN_SOURCE_GROUP:
2926 error = inp_join_group(inp, sopt);
2929 case IP_DROP_MEMBERSHIP:
2930 case IP_DROP_SOURCE_MEMBERSHIP:
2931 case MCAST_LEAVE_GROUP:
2932 case MCAST_LEAVE_SOURCE_GROUP:
2933 error = inp_leave_group(inp, sopt);
2936 case IP_BLOCK_SOURCE:
2937 case IP_UNBLOCK_SOURCE:
2938 case MCAST_BLOCK_SOURCE:
2939 case MCAST_UNBLOCK_SOURCE:
2940 error = inp_block_unblock_source(inp, sopt);
2944 error = inp_set_source_filters(inp, sopt);
2952 INP_UNLOCK_ASSERT(inp);
2958 * Expose IGMP's multicast filter mode and source list(s) to userland,
2959 * keyed by (ifindex, group).
2960 * The filter mode is written out as a uint32_t, followed by
2961 * 0..n of struct in_addr.
2962 * For use by ifmcstat(8).
2963 * SMPng: NOTE: unlocked read of ifindex space.
2966 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2968 struct in_addr src, group;
2970 struct ifmultiaddr *ifma;
2971 struct in_multi *inm;
2972 struct ip_msource *ims;
2976 uint32_t fmode, ifindex;
2981 if (req->newptr != NULL)
2988 if (ifindex <= 0 || ifindex > V_if_index) {
2989 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2994 group.s_addr = name[1];
2995 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2996 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2997 __func__, ntohl(group.s_addr));
3001 ifp = ifnet_byindex(ifindex);
3003 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
3008 retval = sysctl_wire_old_buffer(req,
3009 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
3013 IN_MULTI_LIST_LOCK();
3016 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3017 if (ifma->ifma_addr->sa_family != AF_INET ||
3018 ifma->ifma_protospec == NULL)
3020 inm = (struct in_multi *)ifma->ifma_protospec;
3021 if (!in_hosteq(inm->inm_addr, group))
3023 fmode = inm->inm_st[1].iss_fmode;
3024 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
3027 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
3028 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
3031 * Only copy-out sources which are in-mode.
3033 if (fmode != ims_get_mode(inm, ims, 1)) {
3034 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
3038 src.s_addr = htonl(ims->ims_haddr);
3039 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
3044 IF_ADDR_RUNLOCK(ifp);
3046 IN_MULTI_LIST_UNLOCK();
3051 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
3053 static const char *inm_modestrs[] = { "un", "in", "ex" };
3056 inm_mode_str(const int mode)
3059 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
3060 return (inm_modestrs[mode]);
3064 static const char *inm_statestrs[] = {
3077 inm_state_str(const int state)
3080 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3081 return (inm_statestrs[state]);
3086 * Dump an in_multi structure to the console.
3089 inm_print(const struct in_multi *inm)
3092 char addrbuf[INET_ADDRSTRLEN];
3094 if ((ktr_mask & KTR_IGMPV3) == 0)
3097 printf("%s: --- begin inm %p ---\n", __func__, inm);
3098 printf("addr %s ifp %p(%s) ifma %p\n",
3099 inet_ntoa_r(inm->inm_addr, addrbuf),
3101 inm->inm_ifp->if_xname,
3103 printf("timer %u state %s refcount %u scq.len %u\n",
3105 inm_state_str(inm->inm_state),
3107 inm->inm_scq.mq_len);
3108 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3113 for (t = 0; t < 2; t++) {
3114 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3115 inm_mode_str(inm->inm_st[t].iss_fmode),
3116 inm->inm_st[t].iss_asm,
3117 inm->inm_st[t].iss_ex,
3118 inm->inm_st[t].iss_in,
3119 inm->inm_st[t].iss_rec);
3121 printf("%s: --- end inm %p ---\n", __func__, inm);
3124 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3127 inm_print(const struct in_multi *inm)
3132 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3134 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);