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 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
267 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
268 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
269 MPASS(ifma != ll_ifma);
270 ifma->ifma_llifma = NULL;
271 MPASS(ll_ifma->ifma_llifma == NULL);
272 MPASS(ll_ifma->ifma_ifp == ifp);
273 if (--ll_ifma->ifma_refcount == 0) {
274 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
275 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
276 if_freemulti(ll_ifma);
283 inm_release_deferred(struct in_multi *inm)
285 struct in_multi_head tmp;
287 IN_MULTI_LIST_LOCK_ASSERT();
288 MPASS(inm->inm_refcount > 0);
289 if (--inm->inm_refcount == 0) {
292 inm->inm_ifma->ifma_protospec = NULL;
293 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
294 inm_release_list_deferred(&tmp);
299 inm_release_task(void *arg __unused)
301 struct in_multi_head inm_free_tmp;
302 struct in_multi *inm, *tinm;
304 SLIST_INIT(&inm_free_tmp);
305 mtx_lock(&in_multi_free_mtx);
306 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
307 mtx_unlock(&in_multi_free_mtx);
309 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
310 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
318 * Initialize an in_mfilter structure to a known state at t0, t1
319 * with an empty source filter list.
322 imf_init(struct in_mfilter *imf, const int st0, const int st1)
324 memset(imf, 0, sizeof(struct in_mfilter));
325 RB_INIT(&imf->imf_sources);
326 imf->imf_st[0] = st0;
327 imf->imf_st[1] = st1;
331 * Function for looking up an in_multi record for an IPv4 multicast address
332 * on a given interface. ifp must be valid. If no record found, return NULL.
333 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
336 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
338 struct ifmultiaddr *ifma;
339 struct in_multi *inm;
341 IN_MULTI_LIST_LOCK_ASSERT();
342 IF_ADDR_LOCK_ASSERT(ifp);
345 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
346 if (ifma->ifma_addr->sa_family != AF_INET ||
347 ifma->ifma_protospec == NULL)
349 inm = (struct in_multi *)ifma->ifma_protospec;
350 if (inm->inm_addr.s_addr == ina.s_addr)
358 * Wrapper for inm_lookup_locked().
359 * The IF_ADDR_LOCK will be taken on ifp and released on return.
362 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
364 struct in_multi *inm;
366 IN_MULTI_LIST_LOCK_ASSERT();
368 inm = inm_lookup_locked(ifp, ina);
369 IF_ADDR_RUNLOCK(ifp);
375 * Resize the ip_moptions vector to the next power-of-two minus 1.
376 * May be called with locks held; do not sleep.
379 imo_grow(struct ip_moptions *imo)
381 struct in_multi **nmships;
382 struct in_multi **omships;
383 struct in_mfilter *nmfilters;
384 struct in_mfilter *omfilters;
391 omships = imo->imo_membership;
392 omfilters = imo->imo_mfilters;
393 oldmax = imo->imo_max_memberships;
394 newmax = ((oldmax + 1) * 2) - 1;
396 if (newmax <= IP_MAX_MEMBERSHIPS) {
397 nmships = (struct in_multi **)realloc(omships,
398 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
399 nmfilters = (struct in_mfilter *)realloc(omfilters,
400 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
401 if (nmships != NULL && nmfilters != NULL) {
402 /* Initialize newly allocated source filter heads. */
403 for (idx = oldmax; idx < newmax; idx++) {
404 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
407 imo->imo_max_memberships = newmax;
408 imo->imo_membership = nmships;
409 imo->imo_mfilters = nmfilters;
413 if (nmships == NULL || nmfilters == NULL) {
415 free(nmships, M_IPMOPTS);
416 if (nmfilters != NULL)
417 free(nmfilters, M_INMFILTER);
418 return (ETOOMANYREFS);
425 * Find an IPv4 multicast group entry for this ip_moptions instance
426 * which matches the specified group, and optionally an interface.
427 * Return its index into the array, or -1 if not found.
430 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
431 const struct sockaddr *group)
433 const struct sockaddr_in *gsin;
434 struct in_multi **pinm;
438 gsin = (const struct sockaddr_in *)group;
440 /* The imo_membership array may be lazy allocated. */
441 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
444 nmships = imo->imo_num_memberships;
445 pinm = &imo->imo_membership[0];
446 for (idx = 0; idx < nmships; idx++, pinm++) {
449 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
450 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
461 * Find an IPv4 multicast source entry for this imo which matches
462 * the given group index for this socket, and source address.
464 * NOTE: This does not check if the entry is in-mode, merely if
465 * it exists, which may not be the desired behaviour.
467 static struct in_msource *
468 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
469 const struct sockaddr *src)
471 struct ip_msource find;
472 struct in_mfilter *imf;
473 struct ip_msource *ims;
474 const sockunion_t *psa;
476 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
477 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
478 ("%s: invalid index %d\n", __func__, (int)gidx));
480 /* The imo_mfilters array may be lazy allocated. */
481 if (imo->imo_mfilters == NULL)
483 imf = &imo->imo_mfilters[gidx];
485 /* Source trees are keyed in host byte order. */
486 psa = (const sockunion_t *)src;
487 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
488 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
490 return ((struct in_msource *)ims);
494 * Perform filtering for multicast datagrams on a socket by group and source.
496 * Returns 0 if a datagram should be allowed through, or various error codes
497 * if the socket was not a member of the group, or the source was muted, etc.
500 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
501 const struct sockaddr *group, const struct sockaddr *src)
504 struct in_msource *ims;
507 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
509 gidx = imo_match_group(imo, ifp, group);
511 return (MCAST_NOTGMEMBER);
514 * Check if the source was included in an (S,G) join.
515 * Allow reception on exclusive memberships by default,
516 * reject reception on inclusive memberships by default.
517 * Exclude source only if an in-mode exclude filter exists.
518 * Include source only if an in-mode include filter exists.
519 * NOTE: We are comparing group state here at IGMP t1 (now)
520 * with socket-layer t0 (since last downcall).
522 mode = imo->imo_mfilters[gidx].imf_st[1];
523 ims = imo_match_source(imo, gidx, src);
525 if ((ims == NULL && mode == MCAST_INCLUDE) ||
526 (ims != NULL && ims->imsl_st[0] != mode))
527 return (MCAST_NOTSMEMBER);
533 * Find and return a reference to an in_multi record for (ifp, group),
534 * and bump its reference count.
535 * If one does not exist, try to allocate it, and update link-layer multicast
536 * filters on ifp to listen for group.
537 * Assumes the IN_MULTI lock is held across the call.
538 * Return 0 if successful, otherwise return an appropriate error code.
541 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
542 struct in_multi **pinm)
544 struct sockaddr_in gsin;
545 struct ifmultiaddr *ifma;
546 struct in_ifinfo *ii;
547 struct in_multi *inm;
550 IN_MULTI_LOCK_ASSERT();
552 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
553 IN_MULTI_LIST_LOCK();
554 inm = inm_lookup(ifp, *group);
557 * If we already joined this group, just bump the
558 * refcount and return it.
560 KASSERT(inm->inm_refcount >= 1,
561 ("%s: bad refcount %d", __func__, inm->inm_refcount));
562 inm_acquire_locked(inm);
565 IN_MULTI_LIST_UNLOCK();
569 memset(&gsin, 0, sizeof(gsin));
570 gsin.sin_family = AF_INET;
571 gsin.sin_len = sizeof(struct sockaddr_in);
572 gsin.sin_addr = *group;
575 * Check if a link-layer group is already associated
576 * with this network-layer group on the given ifnet.
578 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
582 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
583 IN_MULTI_LIST_LOCK();
587 * If something other than netinet is occupying the link-layer
588 * group, print a meaningful error message and back out of
590 * Otherwise, bump the refcount on the existing network-layer
591 * group association and return it.
593 if (ifma->ifma_protospec != NULL) {
594 inm = (struct in_multi *)ifma->ifma_protospec;
596 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
598 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
599 ("%s: ifma not AF_INET", __func__));
600 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
601 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
602 !in_hosteq(inm->inm_addr, *group)) {
603 char addrbuf[INET_ADDRSTRLEN];
605 panic("%s: ifma %p is inconsistent with %p (%s)",
606 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
609 inm_acquire_locked(inm);
614 IF_ADDR_WLOCK_ASSERT(ifp);
617 * A new in_multi record is needed; allocate and initialize it.
618 * We DO NOT perform an IGMP join as the in_ layer may need to
619 * push an initial source list down to IGMP to support SSM.
621 * The initial source filter state is INCLUDE, {} as per the RFC.
623 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
625 IF_ADDR_WUNLOCK(ifp);
626 IN_MULTI_LIST_UNLOCK();
627 if_delmulti_ifma(ifma);
630 inm->inm_addr = *group;
632 inm->inm_igi = ii->ii_igmp;
633 inm->inm_ifma = ifma;
634 inm->inm_refcount = 1;
635 inm->inm_state = IGMP_NOT_MEMBER;
636 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
637 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
638 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
639 RB_INIT(&inm->inm_srcs);
641 ifma->ifma_protospec = inm;
645 IF_ADDR_WUNLOCK(ifp);
646 IN_MULTI_LIST_UNLOCK();
651 * Drop a reference to an in_multi record.
653 * If the refcount drops to 0, free the in_multi record and
654 * delete the underlying link-layer membership.
657 inm_release(struct in_multi *inm)
659 struct ifmultiaddr *ifma;
662 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
663 MPASS(inm->inm_refcount == 0);
664 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
666 ifma = inm->inm_ifma;
669 /* XXX this access is not covered by IF_ADDR_LOCK */
670 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
672 CURVNET_SET(ifp->if_vnet);
674 free(inm, M_IPMADDR);
675 if_delmulti_ifma_flags(ifma, 1);
680 free(inm, M_IPMADDR);
681 if_delmulti_ifma_flags(ifma, 1);
686 * Clear recorded source entries for a group.
687 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
688 * FIXME: Should reap.
691 inm_clear_recorded(struct in_multi *inm)
693 struct ip_msource *ims;
695 IN_MULTI_LIST_LOCK_ASSERT();
697 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
700 --inm->inm_st[1].iss_rec;
703 KASSERT(inm->inm_st[1].iss_rec == 0,
704 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
708 * Record a source as pending for a Source-Group IGMPv3 query.
709 * This lives here as it modifies the shared tree.
711 * inm is the group descriptor.
712 * naddr is the address of the source to record in network-byte order.
714 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
715 * lazy-allocate a source node in response to an SG query.
716 * Otherwise, no allocation is performed. This saves some memory
717 * with the trade-off that the source will not be reported to the
718 * router if joined in the window between the query response and
719 * the group actually being joined on the local host.
721 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
722 * This turns off the allocation of a recorded source entry if
723 * the group has not been joined.
725 * Return 0 if the source didn't exist or was already marked as recorded.
726 * Return 1 if the source was marked as recorded by this function.
727 * Return <0 if any error occurred (negated errno code).
730 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
732 struct ip_msource find;
733 struct ip_msource *ims, *nims;
735 IN_MULTI_LIST_LOCK_ASSERT();
737 find.ims_haddr = ntohl(naddr);
738 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
739 if (ims && ims->ims_stp)
742 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
744 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
748 nims->ims_haddr = find.ims_haddr;
749 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
755 * Mark the source as recorded and update the recorded
759 ++inm->inm_st[1].iss_rec;
765 * Return a pointer to an in_msource owned by an in_mfilter,
766 * given its source address.
767 * Lazy-allocate if needed. If this is a new entry its filter state is
770 * imf is the filter set being modified.
771 * haddr is the source address in *host* byte-order.
773 * SMPng: May be called with locks held; malloc must not block.
776 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
777 struct in_msource **plims)
779 struct ip_msource find;
780 struct ip_msource *ims, *nims;
781 struct in_msource *lims;
788 /* key is host byte order */
789 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
790 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
791 lims = (struct in_msource *)ims;
793 if (imf->imf_nsrc == in_mcast_maxsocksrc)
795 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
799 lims = (struct in_msource *)nims;
800 lims->ims_haddr = find.ims_haddr;
801 lims->imsl_st[0] = MCAST_UNDEFINED;
802 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
812 * Graft a source entry into an existing socket-layer filter set,
813 * maintaining any required invariants and checking allocations.
815 * The source is marked as being in the new filter mode at t1.
817 * Return the pointer to the new node, otherwise return NULL.
819 static struct in_msource *
820 imf_graft(struct in_mfilter *imf, const uint8_t st1,
821 const struct sockaddr_in *psin)
823 struct ip_msource *nims;
824 struct in_msource *lims;
826 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
830 lims = (struct in_msource *)nims;
831 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
832 lims->imsl_st[0] = MCAST_UNDEFINED;
833 lims->imsl_st[1] = st1;
834 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
841 * Prune a source entry from an existing socket-layer filter set,
842 * maintaining any required invariants and checking allocations.
844 * The source is marked as being left at t1, it is not freed.
846 * Return 0 if no error occurred, otherwise return an errno value.
849 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
851 struct ip_msource find;
852 struct ip_msource *ims;
853 struct in_msource *lims;
855 /* key is host byte order */
856 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
857 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
860 lims = (struct in_msource *)ims;
861 lims->imsl_st[1] = MCAST_UNDEFINED;
866 * Revert socket-layer filter set deltas at t1 to t0 state.
869 imf_rollback(struct in_mfilter *imf)
871 struct ip_msource *ims, *tims;
872 struct in_msource *lims;
874 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
875 lims = (struct in_msource *)ims;
876 if (lims->imsl_st[0] == lims->imsl_st[1]) {
877 /* no change at t1 */
879 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
880 /* revert change to existing source at t1 */
881 lims->imsl_st[1] = lims->imsl_st[0];
883 /* revert source added t1 */
884 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
885 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
886 free(ims, M_INMFILTER);
890 imf->imf_st[1] = imf->imf_st[0];
894 * Mark socket-layer filter set as INCLUDE {} at t1.
897 imf_leave(struct in_mfilter *imf)
899 struct ip_msource *ims;
900 struct in_msource *lims;
902 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
903 lims = (struct in_msource *)ims;
904 lims->imsl_st[1] = MCAST_UNDEFINED;
906 imf->imf_st[1] = MCAST_INCLUDE;
910 * Mark socket-layer filter set deltas as committed.
913 imf_commit(struct in_mfilter *imf)
915 struct ip_msource *ims;
916 struct in_msource *lims;
918 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
919 lims = (struct in_msource *)ims;
920 lims->imsl_st[0] = lims->imsl_st[1];
922 imf->imf_st[0] = imf->imf_st[1];
926 * Reap unreferenced sources from socket-layer filter set.
929 imf_reap(struct in_mfilter *imf)
931 struct ip_msource *ims, *tims;
932 struct in_msource *lims;
934 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
935 lims = (struct in_msource *)ims;
936 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
937 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
938 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
939 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
940 free(ims, M_INMFILTER);
947 * Purge socket-layer filter set.
950 imf_purge(struct in_mfilter *imf)
952 struct ip_msource *ims, *tims;
954 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
955 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
956 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
957 free(ims, M_INMFILTER);
960 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
961 KASSERT(RB_EMPTY(&imf->imf_sources),
962 ("%s: imf_sources not empty", __func__));
966 * Look up a source filter entry for a multicast group.
968 * inm is the group descriptor to work with.
969 * haddr is the host-byte-order IPv4 address to look up.
970 * noalloc may be non-zero to suppress allocation of sources.
971 * *pims will be set to the address of the retrieved or allocated source.
973 * SMPng: NOTE: may be called with locks held.
974 * Return 0 if successful, otherwise return a non-zero error code.
977 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
978 const int noalloc, struct ip_msource **pims)
980 struct ip_msource find;
981 struct ip_msource *ims, *nims;
983 find.ims_haddr = haddr;
984 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
985 if (ims == NULL && !noalloc) {
986 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
988 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
992 nims->ims_haddr = haddr;
993 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
997 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
1007 * Merge socket-layer source into IGMP-layer source.
1008 * If rollback is non-zero, perform the inverse of the merge.
1011 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
1014 int n = rollback ? -1 : 1;
1016 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
1017 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
1018 __func__, n, ims->ims_haddr);
1019 ims->ims_st[1].ex -= n;
1020 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
1021 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
1022 __func__, n, ims->ims_haddr);
1023 ims->ims_st[1].in -= n;
1026 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
1027 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
1028 __func__, n, ims->ims_haddr);
1029 ims->ims_st[1].ex += n;
1030 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
1031 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
1032 __func__, n, ims->ims_haddr);
1033 ims->ims_st[1].in += n;
1038 * Atomically update the global in_multi state, when a membership's
1039 * filter list is being updated in any way.
1041 * imf is the per-inpcb-membership group filter pointer.
1042 * A fake imf may be passed for in-kernel consumers.
1044 * XXX This is a candidate for a set-symmetric-difference style loop
1045 * which would eliminate the repeated lookup from root of ims nodes,
1046 * as they share the same key space.
1048 * If any error occurred this function will back out of refcounts
1049 * and return a non-zero value.
1052 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1054 struct ip_msource *ims, *nims;
1055 struct in_msource *lims;
1056 int schanged, error;
1062 IN_MULTI_LIST_LOCK_ASSERT();
1065 * Update the source filters first, as this may fail.
1066 * Maintain count of in-mode filters at t0, t1. These are
1067 * used to work out if we transition into ASM mode or not.
1068 * Maintain a count of source filters whose state was
1069 * actually modified by this operation.
1071 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1072 lims = (struct in_msource *)ims;
1073 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1074 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1075 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1076 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1080 ims_merge(nims, lims, 0);
1083 struct ip_msource *bims;
1085 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1086 lims = (struct in_msource *)ims;
1087 if (lims->imsl_st[0] == lims->imsl_st[1])
1089 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1092 ims_merge(bims, lims, 1);
1097 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1098 __func__, nsrc0, nsrc1);
1100 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1101 if (imf->imf_st[0] == imf->imf_st[1] &&
1102 imf->imf_st[1] == MCAST_INCLUDE) {
1104 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1105 --inm->inm_st[1].iss_in;
1109 /* Handle filter mode transition on socket. */
1110 if (imf->imf_st[0] != imf->imf_st[1]) {
1111 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1112 __func__, imf->imf_st[0], imf->imf_st[1]);
1114 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1115 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1116 --inm->inm_st[1].iss_ex;
1117 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1118 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1119 --inm->inm_st[1].iss_in;
1122 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1123 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1124 inm->inm_st[1].iss_ex++;
1125 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1126 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1127 inm->inm_st[1].iss_in++;
1132 * Track inm filter state in terms of listener counts.
1133 * If there are any exclusive listeners, stack-wide
1134 * membership is exclusive.
1135 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1136 * If no listeners remain, state is undefined at t1,
1137 * and the IGMP lifecycle for this group should finish.
1139 if (inm->inm_st[1].iss_ex > 0) {
1140 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1141 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1142 } else if (inm->inm_st[1].iss_in > 0) {
1143 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1144 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1146 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1147 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1150 /* Decrement ASM listener count on transition out of ASM mode. */
1151 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1152 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1153 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1154 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1155 --inm->inm_st[1].iss_asm;
1159 /* Increment ASM listener count on transition to ASM mode. */
1160 if (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++;
1165 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1170 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1177 * Mark an in_multi's filter set deltas as committed.
1178 * Called by IGMP after a state change has been enqueued.
1181 inm_commit(struct in_multi *inm)
1183 struct ip_msource *ims;
1185 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1186 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1189 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1190 ims->ims_st[0] = ims->ims_st[1];
1192 inm->inm_st[0] = inm->inm_st[1];
1196 * Reap unreferenced nodes from an in_multi's filter set.
1199 inm_reap(struct in_multi *inm)
1201 struct ip_msource *ims, *tims;
1203 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1204 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1205 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1208 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1209 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1210 free(ims, M_IPMSOURCE);
1216 * Purge all source nodes from an in_multi's filter set.
1219 inm_purge(struct in_multi *inm)
1221 struct ip_msource *ims, *tims;
1223 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1224 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1225 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1226 free(ims, M_IPMSOURCE);
1232 * Join a multicast group; unlocked entry point.
1234 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1235 * is not held. Fortunately, ifp is unlikely to have been detached
1236 * at this point, so we assume it's OK to recurse.
1239 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1240 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1245 error = in_joingroup_locked(ifp, gina, imf, pinm);
1252 * Join a multicast group; real entry point.
1254 * Only preserves atomicity at inm level.
1255 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1257 * If the IGMP downcall fails, the group is not joined, and an error
1261 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1262 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1264 struct in_mfilter timf;
1265 struct in_multi *inm;
1268 IN_MULTI_LOCK_ASSERT();
1269 IN_MULTI_LIST_UNLOCK_ASSERT();
1271 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1272 ntohl(gina->s_addr), ifp, ifp->if_xname);
1278 * If no imf was specified (i.e. kernel consumer),
1279 * fake one up and assume it is an ASM join.
1282 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1286 error = in_getmulti(ifp, gina, &inm);
1288 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1291 IN_MULTI_LIST_LOCK();
1292 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1293 error = inm_merge(inm, imf);
1295 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1296 goto out_inm_release;
1299 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1300 error = igmp_change_state(inm);
1302 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1303 goto out_inm_release;
1309 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1310 inm_release_deferred(inm);
1314 IN_MULTI_LIST_UNLOCK();
1320 * Leave a multicast group; unlocked entry point.
1323 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1328 error = in_leavegroup_locked(inm, imf);
1335 * Leave a multicast group; real entry point.
1336 * All source filters will be expunged.
1338 * Only preserves atomicity at inm level.
1340 * Holding the write lock for the INP which contains imf
1341 * is highly advisable. We can't assert for it as imf does not
1342 * contain a back-pointer to the owning inp.
1344 * Note: This is not the same as inm_release(*) as this function also
1345 * makes a state change downcall into IGMP.
1348 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1350 struct in_mfilter timf;
1355 IN_MULTI_LOCK_ASSERT();
1356 IN_MULTI_LIST_UNLOCK_ASSERT();
1358 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1359 inm, ntohl(inm->inm_addr.s_addr),
1360 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1364 * If no imf was specified (i.e. kernel consumer),
1365 * fake one up and assume it is an ASM join.
1368 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1373 * Begin state merge transaction at IGMP layer.
1375 * As this particular invocation should not cause any memory
1376 * to be allocated, and there is no opportunity to roll back
1377 * the transaction, it MUST NOT fail.
1379 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1380 IN_MULTI_LIST_LOCK();
1381 error = inm_merge(inm, imf);
1382 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1384 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1385 CURVNET_SET(inm->inm_ifp->if_vnet);
1386 error = igmp_change_state(inm);
1387 IF_ADDR_WLOCK(inm->inm_ifp);
1388 inm_release_deferred(inm);
1389 IF_ADDR_WUNLOCK(inm->inm_ifp);
1390 IN_MULTI_LIST_UNLOCK();
1393 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1395 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1400 /*#ifndef BURN_BRIDGES*/
1402 * Join an IPv4 multicast group in (*,G) exclusive mode.
1403 * The group must be a 224.0.0.0/24 link-scope group.
1404 * This KPI is for legacy kernel consumers only.
1407 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1409 struct in_multi *pinm;
1412 char addrbuf[INET_ADDRSTRLEN];
1415 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1416 ("%s: %s not in 224.0.0.0/24", __func__,
1417 inet_ntoa_r(*ap, addrbuf)));
1419 error = in_joingroup(ifp, ap, NULL, &pinm);
1427 * Block or unblock an ASM multicast source on an inpcb.
1428 * This implements the delta-based API described in RFC 3678.
1430 * The delta-based API applies only to exclusive-mode memberships.
1431 * An IGMP downcall will be performed.
1433 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1435 * Return 0 if successful, otherwise return an appropriate error code.
1438 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1440 struct group_source_req gsr;
1441 sockunion_t *gsa, *ssa;
1443 struct in_mfilter *imf;
1444 struct ip_moptions *imo;
1445 struct in_msource *ims;
1446 struct in_multi *inm;
1455 memset(&gsr, 0, sizeof(struct group_source_req));
1456 gsa = (sockunion_t *)&gsr.gsr_group;
1457 ssa = (sockunion_t *)&gsr.gsr_source;
1459 switch (sopt->sopt_name) {
1460 case IP_BLOCK_SOURCE:
1461 case IP_UNBLOCK_SOURCE: {
1462 struct ip_mreq_source mreqs;
1464 error = sooptcopyin(sopt, &mreqs,
1465 sizeof(struct ip_mreq_source),
1466 sizeof(struct ip_mreq_source));
1470 gsa->sin.sin_family = AF_INET;
1471 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1472 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1474 ssa->sin.sin_family = AF_INET;
1475 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1476 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1478 if (!in_nullhost(mreqs.imr_interface))
1479 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1481 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1484 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1485 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1489 case MCAST_BLOCK_SOURCE:
1490 case MCAST_UNBLOCK_SOURCE:
1491 error = sooptcopyin(sopt, &gsr,
1492 sizeof(struct group_source_req),
1493 sizeof(struct group_source_req));
1497 if (gsa->sin.sin_family != AF_INET ||
1498 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1501 if (ssa->sin.sin_family != AF_INET ||
1502 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1505 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1506 return (EADDRNOTAVAIL);
1508 ifp = ifnet_byindex(gsr.gsr_interface);
1510 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1515 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1516 __func__, sopt->sopt_name);
1517 return (EOPNOTSUPP);
1521 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1525 * Check if we are actually a member of this group.
1527 imo = inp_findmoptions(inp);
1528 idx = imo_match_group(imo, ifp, &gsa->sa);
1529 if (idx == -1 || imo->imo_mfilters == NULL) {
1530 error = EADDRNOTAVAIL;
1531 goto out_inp_locked;
1534 KASSERT(imo->imo_mfilters != NULL,
1535 ("%s: imo_mfilters not allocated", __func__));
1536 imf = &imo->imo_mfilters[idx];
1537 inm = imo->imo_membership[idx];
1540 * Attempting to use the delta-based API on an
1541 * non exclusive-mode membership is an error.
1543 fmode = imf->imf_st[0];
1544 if (fmode != MCAST_EXCLUDE) {
1546 goto out_inp_locked;
1550 * Deal with error cases up-front:
1551 * Asked to block, but already blocked; or
1552 * Asked to unblock, but nothing to unblock.
1553 * If adding a new block entry, allocate it.
1555 ims = imo_match_source(imo, idx, &ssa->sa);
1556 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1557 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1558 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1559 error = EADDRNOTAVAIL;
1560 goto out_inp_locked;
1563 INP_WLOCK_ASSERT(inp);
1566 * Begin state merge transaction at socket layer.
1569 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1570 ims = imf_graft(imf, fmode, &ssa->sin);
1574 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1575 error = imf_prune(imf, &ssa->sin);
1579 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1580 goto out_imf_rollback;
1584 * Begin state merge transaction at IGMP layer.
1587 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1588 IN_MULTI_LIST_LOCK();
1589 error = inm_merge(inm, imf);
1591 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1592 IN_MULTI_LIST_UNLOCK();
1593 goto out_in_multi_locked;
1596 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1597 error = igmp_change_state(inm);
1598 IN_MULTI_LIST_UNLOCK();
1600 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1602 out_in_multi_locked:
1619 * Given an inpcb, return its multicast options structure pointer. Accepts
1620 * an unlocked inpcb pointer, but will return it locked. May sleep.
1622 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1623 * SMPng: NOTE: Returns with the INP write lock held.
1625 static struct ip_moptions *
1626 inp_findmoptions(struct inpcb *inp)
1628 struct ip_moptions *imo;
1629 struct in_multi **immp;
1630 struct in_mfilter *imfp;
1634 if (inp->inp_moptions != NULL)
1635 return (inp->inp_moptions);
1639 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1640 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1642 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1643 M_INMFILTER, M_WAITOK);
1645 imo->imo_multicast_ifp = NULL;
1646 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1647 imo->imo_multicast_vif = -1;
1648 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1649 imo->imo_multicast_loop = in_mcast_loop;
1650 imo->imo_num_memberships = 0;
1651 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1652 imo->imo_membership = immp;
1654 /* Initialize per-group source filters. */
1655 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1656 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1657 imo->imo_mfilters = imfp;
1660 if (inp->inp_moptions != NULL) {
1661 free(imfp, M_INMFILTER);
1662 free(immp, M_IPMOPTS);
1663 free(imo, M_IPMOPTS);
1664 return (inp->inp_moptions);
1666 inp->inp_moptions = imo;
1671 inp_gcmoptions(epoch_context_t ctx)
1673 struct ip_moptions *imo;
1674 struct in_mfilter *imf;
1675 struct in_multi *inm;
1677 size_t idx, nmships;
1679 imo = __containerof(ctx, struct ip_moptions, imo_epoch_ctx);
1681 nmships = imo->imo_num_memberships;
1682 for (idx = 0; idx < nmships; ++idx) {
1683 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1686 inm = imo->imo_membership[idx];
1689 CURVNET_SET(ifp->if_vnet);
1690 (void)in_leavegroup(inm, imf);
1693 (void)in_leavegroup(inm, imf);
1699 if (imo->imo_mfilters)
1700 free(imo->imo_mfilters, M_INMFILTER);
1701 free(imo->imo_membership, M_IPMOPTS);
1702 free(imo, M_IPMOPTS);
1706 * Discard the IP multicast options (and source filters). To minimize
1707 * the amount of work done while holding locks such as the INP's
1708 * pcbinfo lock (which is used in the receive path), the free
1709 * operation is deferred to the epoch callback task.
1712 inp_freemoptions(struct ip_moptions *imo)
1716 epoch_call(net_epoch_preempt, &imo->imo_epoch_ctx, inp_gcmoptions);
1720 * Atomically get source filters on a socket for an IPv4 multicast group.
1721 * Called with INP lock held; returns with lock released.
1724 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1726 struct __msfilterreq msfr;
1729 struct ip_moptions *imo;
1730 struct in_mfilter *imf;
1731 struct ip_msource *ims;
1732 struct in_msource *lims;
1733 struct sockaddr_in *psin;
1734 struct sockaddr_storage *ptss;
1735 struct sockaddr_storage *tss;
1737 size_t idx, nsrcs, ncsrcs;
1739 INP_WLOCK_ASSERT(inp);
1741 imo = inp->inp_moptions;
1742 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1746 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1747 sizeof(struct __msfilterreq));
1751 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1754 ifp = ifnet_byindex(msfr.msfr_ifindex);
1761 * Lookup group on the socket.
1763 gsa = (sockunion_t *)&msfr.msfr_group;
1764 idx = imo_match_group(imo, ifp, &gsa->sa);
1765 if (idx == -1 || imo->imo_mfilters == NULL) {
1767 return (EADDRNOTAVAIL);
1769 imf = &imo->imo_mfilters[idx];
1772 * Ignore memberships which are in limbo.
1774 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1778 msfr.msfr_fmode = imf->imf_st[1];
1781 * If the user specified a buffer, copy out the source filter
1782 * entries to userland gracefully.
1783 * We only copy out the number of entries which userland
1784 * has asked for, but we always tell userland how big the
1785 * buffer really needs to be.
1787 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1788 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1790 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1791 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1792 M_TEMP, M_NOWAIT | M_ZERO);
1800 * Count number of sources in-mode at t0.
1801 * If buffer space exists and remains, copy out source entries.
1803 nsrcs = msfr.msfr_nsrcs;
1806 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1807 lims = (struct in_msource *)ims;
1808 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1809 lims->imsl_st[0] != imf->imf_st[0])
1812 if (tss != NULL && nsrcs > 0) {
1813 psin = (struct sockaddr_in *)ptss;
1814 psin->sin_family = AF_INET;
1815 psin->sin_len = sizeof(struct sockaddr_in);
1816 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1826 error = copyout(tss, msfr.msfr_srcs,
1827 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1833 msfr.msfr_nsrcs = ncsrcs;
1834 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1840 * Return the IP multicast options in response to user getsockopt().
1843 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1845 struct rm_priotracker in_ifa_tracker;
1846 struct ip_mreqn mreqn;
1847 struct ip_moptions *imo;
1849 struct in_ifaddr *ia;
1854 imo = inp->inp_moptions;
1856 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1857 * or is a divert socket, reject it.
1859 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1860 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1861 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1863 return (EOPNOTSUPP);
1867 switch (sopt->sopt_name) {
1868 case IP_MULTICAST_VIF:
1870 optval = imo->imo_multicast_vif;
1874 error = sooptcopyout(sopt, &optval, sizeof(int));
1877 case IP_MULTICAST_IF:
1878 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1880 ifp = imo->imo_multicast_ifp;
1881 if (!in_nullhost(imo->imo_multicast_addr)) {
1882 mreqn.imr_address = imo->imo_multicast_addr;
1883 } else if (ifp != NULL) {
1884 mreqn.imr_ifindex = ifp->if_index;
1886 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1889 IA_SIN(ia)->sin_addr;
1894 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1895 error = sooptcopyout(sopt, &mreqn,
1896 sizeof(struct ip_mreqn));
1898 error = sooptcopyout(sopt, &mreqn.imr_address,
1899 sizeof(struct in_addr));
1903 case IP_MULTICAST_TTL:
1905 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1907 optval = coptval = imo->imo_multicast_ttl;
1909 if (sopt->sopt_valsize == sizeof(u_char))
1910 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1912 error = sooptcopyout(sopt, &optval, sizeof(int));
1915 case IP_MULTICAST_LOOP:
1917 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1919 optval = coptval = imo->imo_multicast_loop;
1921 if (sopt->sopt_valsize == sizeof(u_char))
1922 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1924 error = sooptcopyout(sopt, &optval, sizeof(int));
1929 error = EADDRNOTAVAIL;
1932 error = inp_get_source_filters(inp, sopt);
1938 error = ENOPROTOOPT;
1942 INP_UNLOCK_ASSERT(inp);
1948 * Look up the ifnet to use for a multicast group membership,
1949 * given the IPv4 address of an interface, and the IPv4 group address.
1951 * This routine exists to support legacy multicast applications
1952 * which do not understand that multicast memberships are scoped to
1953 * specific physical links in the networking stack, or which need
1954 * to join link-scope groups before IPv4 addresses are configured.
1956 * If inp is non-NULL, use this socket's current FIB number for any
1957 * required FIB lookup.
1958 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1959 * and use its ifp; usually, this points to the default next-hop.
1961 * If the FIB lookup fails, attempt to use the first non-loopback
1962 * interface with multicast capability in the system as a
1963 * last resort. The legacy IPv4 ASM API requires that we do
1964 * this in order to allow groups to be joined when the routing
1965 * table has not yet been populated during boot.
1967 * Returns NULL if no ifp could be found.
1969 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1970 * FUTURE: Implement IPv4 source-address selection.
1972 static struct ifnet *
1973 inp_lookup_mcast_ifp(const struct inpcb *inp,
1974 const struct sockaddr_in *gsin, const struct in_addr ina)
1976 struct rm_priotracker in_ifa_tracker;
1978 struct nhop4_basic nh4;
1981 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1982 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1983 ("%s: not multicast", __func__));
1986 if (!in_nullhost(ina)) {
1987 INADDR_TO_IFP(ina, ifp);
1989 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1990 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1993 struct in_ifaddr *ia;
1997 IN_IFADDR_RLOCK(&in_ifa_tracker);
1998 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
2000 if (!(mifp->if_flags & IFF_LOOPBACK) &&
2001 (mifp->if_flags & IFF_MULTICAST)) {
2006 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2014 * Join an IPv4 multicast group, possibly with a source.
2017 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
2019 struct group_source_req gsr;
2020 sockunion_t *gsa, *ssa;
2022 struct in_mfilter *imf;
2023 struct ip_moptions *imo;
2024 struct in_multi *inm;
2025 struct in_msource *lims;
2035 memset(&gsr, 0, sizeof(struct group_source_req));
2036 gsa = (sockunion_t *)&gsr.gsr_group;
2037 gsa->ss.ss_family = AF_UNSPEC;
2038 ssa = (sockunion_t *)&gsr.gsr_source;
2039 ssa->ss.ss_family = AF_UNSPEC;
2041 switch (sopt->sopt_name) {
2042 case IP_ADD_MEMBERSHIP:
2043 case IP_ADD_SOURCE_MEMBERSHIP: {
2044 struct ip_mreq_source mreqs;
2046 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
2047 error = sooptcopyin(sopt, &mreqs,
2048 sizeof(struct ip_mreq),
2049 sizeof(struct ip_mreq));
2051 * Do argument switcharoo from ip_mreq into
2052 * ip_mreq_source to avoid using two instances.
2054 mreqs.imr_interface = mreqs.imr_sourceaddr;
2055 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2056 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2057 error = sooptcopyin(sopt, &mreqs,
2058 sizeof(struct ip_mreq_source),
2059 sizeof(struct ip_mreq_source));
2064 gsa->sin.sin_family = AF_INET;
2065 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2066 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2068 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2069 ssa->sin.sin_family = AF_INET;
2070 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2071 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2074 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2077 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2078 mreqs.imr_interface);
2079 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2080 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2084 case MCAST_JOIN_GROUP:
2085 case MCAST_JOIN_SOURCE_GROUP:
2086 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2087 error = sooptcopyin(sopt, &gsr,
2088 sizeof(struct group_req),
2089 sizeof(struct group_req));
2090 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2091 error = sooptcopyin(sopt, &gsr,
2092 sizeof(struct group_source_req),
2093 sizeof(struct group_source_req));
2098 if (gsa->sin.sin_family != AF_INET ||
2099 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2103 * Overwrite the port field if present, as the sockaddr
2104 * being copied in may be matched with a binary comparison.
2106 gsa->sin.sin_port = 0;
2107 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2108 if (ssa->sin.sin_family != AF_INET ||
2109 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2111 ssa->sin.sin_port = 0;
2114 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2117 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2118 return (EADDRNOTAVAIL);
2119 ifp = ifnet_byindex(gsr.gsr_interface);
2123 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2124 __func__, sopt->sopt_name);
2125 return (EOPNOTSUPP);
2129 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2130 return (EADDRNOTAVAIL);
2132 imo = inp_findmoptions(inp);
2133 idx = imo_match_group(imo, ifp, &gsa->sa);
2137 inm = imo->imo_membership[idx];
2138 imf = &imo->imo_mfilters[idx];
2139 if (ssa->ss.ss_family != AF_UNSPEC) {
2141 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2142 * is an error. On an existing inclusive membership,
2143 * it just adds the source to the filter list.
2145 if (imf->imf_st[1] != MCAST_INCLUDE) {
2147 goto out_inp_locked;
2150 * Throw out duplicates.
2152 * XXX FIXME: This makes a naive assumption that
2153 * even if entries exist for *ssa in this imf,
2154 * they will be rejected as dupes, even if they
2155 * are not valid in the current mode (in-mode).
2157 * in_msource is transactioned just as for anything
2158 * else in SSM -- but note naive use of inm_graft()
2159 * below for allocating new filter entries.
2161 * This is only an issue if someone mixes the
2162 * full-state SSM API with the delta-based API,
2163 * which is discouraged in the relevant RFCs.
2165 lims = imo_match_source(imo, idx, &ssa->sa);
2166 if (lims != NULL /*&&
2167 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2168 error = EADDRNOTAVAIL;
2169 goto out_inp_locked;
2173 * MCAST_JOIN_GROUP on an existing exclusive
2174 * membership is an error; return EADDRINUSE
2175 * to preserve 4.4BSD API idempotence, and
2176 * avoid tedious detour to code below.
2177 * NOTE: This is bending RFC 3678 a bit.
2179 * On an existing inclusive membership, this is also
2180 * an error; if you want to change filter mode,
2181 * you must use the userland API setsourcefilter().
2182 * XXX We don't reject this for imf in UNDEFINED
2183 * state at t1, because allocation of a filter
2184 * is atomic with allocation of a membership.
2187 if (imf->imf_st[1] == MCAST_EXCLUDE)
2189 goto out_inp_locked;
2194 * Begin state merge transaction at socket layer.
2196 INP_WLOCK_ASSERT(inp);
2199 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2200 error = imo_grow(imo);
2202 goto out_inp_locked;
2205 * Allocate the new slot upfront so we can deal with
2206 * grafting the new source filter in same code path
2207 * as for join-source on existing membership.
2209 idx = imo->imo_num_memberships;
2210 imo->imo_membership[idx] = NULL;
2211 imo->imo_num_memberships++;
2212 KASSERT(imo->imo_mfilters != NULL,
2213 ("%s: imf_mfilters vector was not allocated", __func__));
2214 imf = &imo->imo_mfilters[idx];
2215 KASSERT(RB_EMPTY(&imf->imf_sources),
2216 ("%s: imf_sources not empty", __func__));
2220 * Graft new source into filter list for this inpcb's
2221 * membership of the group. The in_multi may not have
2222 * been allocated yet if this is a new membership, however,
2223 * the in_mfilter slot will be allocated and must be initialized.
2225 * Note: Grafting of exclusive mode filters doesn't happen
2227 * XXX: Should check for non-NULL lims (node exists but may
2228 * not be in-mode) for interop with full-state API.
2230 if (ssa->ss.ss_family != AF_UNSPEC) {
2231 /* Membership starts in IN mode */
2233 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2234 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2236 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2238 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2240 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2246 /* No address specified; Membership starts in EX mode */
2248 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2249 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2254 * Begin state merge transaction at IGMP layer.
2261 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2264 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2266 IN_MULTI_LIST_UNLOCK();
2269 imo->imo_membership[idx] = inm;
2271 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2272 IN_MULTI_LIST_LOCK();
2273 error = inm_merge(inm, imf);
2275 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2277 IN_MULTI_LIST_UNLOCK();
2278 goto out_in_multi_locked;
2280 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2281 error = igmp_change_state(inm);
2282 IN_MULTI_LIST_UNLOCK();
2284 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2286 goto out_in_multi_locked;
2290 out_in_multi_locked:
2294 if (in_pcbrele_wlocked(inp))
2307 if (error && is_new) {
2308 imo->imo_membership[idx] = NULL;
2309 --imo->imo_num_memberships;
2318 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2321 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2323 struct group_source_req gsr;
2324 struct ip_mreq_source mreqs;
2325 sockunion_t *gsa, *ssa;
2327 struct in_mfilter *imf;
2328 struct ip_moptions *imo;
2329 struct in_msource *ims;
2330 struct in_multi *inm;
2332 int error, is_final;
2338 memset(&gsr, 0, sizeof(struct group_source_req));
2339 gsa = (sockunion_t *)&gsr.gsr_group;
2340 gsa->ss.ss_family = AF_UNSPEC;
2341 ssa = (sockunion_t *)&gsr.gsr_source;
2342 ssa->ss.ss_family = AF_UNSPEC;
2344 switch (sopt->sopt_name) {
2345 case IP_DROP_MEMBERSHIP:
2346 case IP_DROP_SOURCE_MEMBERSHIP:
2347 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2348 error = sooptcopyin(sopt, &mreqs,
2349 sizeof(struct ip_mreq),
2350 sizeof(struct ip_mreq));
2352 * Swap interface and sourceaddr arguments,
2353 * as ip_mreq and ip_mreq_source are laid
2356 mreqs.imr_interface = mreqs.imr_sourceaddr;
2357 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2358 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2359 error = sooptcopyin(sopt, &mreqs,
2360 sizeof(struct ip_mreq_source),
2361 sizeof(struct ip_mreq_source));
2366 gsa->sin.sin_family = AF_INET;
2367 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2368 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2370 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2371 ssa->sin.sin_family = AF_INET;
2372 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2373 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2377 * Attempt to look up hinted ifp from interface address.
2378 * Fallthrough with null ifp iff lookup fails, to
2379 * preserve 4.4BSD mcast API idempotence.
2380 * XXX NOTE WELL: The RFC 3678 API is preferred because
2381 * using an IPv4 address as a key is racy.
2383 if (!in_nullhost(mreqs.imr_interface))
2384 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2386 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2387 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2391 case MCAST_LEAVE_GROUP:
2392 case MCAST_LEAVE_SOURCE_GROUP:
2393 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2394 error = sooptcopyin(sopt, &gsr,
2395 sizeof(struct group_req),
2396 sizeof(struct group_req));
2397 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2398 error = sooptcopyin(sopt, &gsr,
2399 sizeof(struct group_source_req),
2400 sizeof(struct group_source_req));
2405 if (gsa->sin.sin_family != AF_INET ||
2406 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2409 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2410 if (ssa->sin.sin_family != AF_INET ||
2411 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2415 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2416 return (EADDRNOTAVAIL);
2418 ifp = ifnet_byindex(gsr.gsr_interface);
2421 return (EADDRNOTAVAIL);
2425 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2426 __func__, sopt->sopt_name);
2427 return (EOPNOTSUPP);
2431 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2435 * Find the membership in the membership array.
2437 imo = inp_findmoptions(inp);
2438 idx = imo_match_group(imo, ifp, &gsa->sa);
2440 error = EADDRNOTAVAIL;
2441 goto out_inp_locked;
2443 inm = imo->imo_membership[idx];
2444 imf = &imo->imo_mfilters[idx];
2446 if (ssa->ss.ss_family != AF_UNSPEC)
2450 * Begin state merge transaction at socket layer.
2452 INP_WLOCK_ASSERT(inp);
2455 * If we were instructed only to leave a given source, do so.
2456 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2461 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2462 error = EADDRNOTAVAIL;
2463 goto out_inp_locked;
2465 ims = imo_match_source(imo, idx, &ssa->sa);
2467 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2468 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2469 error = EADDRNOTAVAIL;
2470 goto out_inp_locked;
2472 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2473 error = imf_prune(imf, &ssa->sin);
2475 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2477 goto out_inp_locked;
2482 * Begin state merge transaction at IGMP layer.
2490 * Give up the multicast address record to which
2491 * the membership points.
2493 (void)in_leavegroup_locked(inm, imf);
2495 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2496 IN_MULTI_LIST_LOCK();
2497 error = inm_merge(inm, imf);
2499 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2501 IN_MULTI_LIST_UNLOCK();
2502 goto out_in_multi_locked;
2505 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2506 error = igmp_change_state(inm);
2507 IN_MULTI_LIST_UNLOCK();
2509 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2514 out_in_multi_locked:
2518 if (in_pcbrele_wlocked(inp))
2529 /* Remove the gap in the membership and filter array. */
2530 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2531 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2532 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2534 imo->imo_num_memberships--;
2543 * Select the interface for transmitting IPv4 multicast datagrams.
2545 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2546 * may be passed to this socket option. An address of INADDR_ANY or an
2547 * interface index of 0 is used to remove a previous selection.
2548 * When no interface is selected, one is chosen for every send.
2551 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2553 struct in_addr addr;
2554 struct ip_mreqn mreqn;
2556 struct ip_moptions *imo;
2559 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2561 * An interface index was specified using the
2562 * Linux-derived ip_mreqn structure.
2564 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2565 sizeof(struct ip_mreqn));
2569 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2572 if (mreqn.imr_ifindex == 0) {
2575 ifp = ifnet_byindex(mreqn.imr_ifindex);
2577 return (EADDRNOTAVAIL);
2581 * An interface was specified by IPv4 address.
2582 * This is the traditional BSD usage.
2584 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2585 sizeof(struct in_addr));
2588 if (in_nullhost(addr)) {
2591 INADDR_TO_IFP(addr, ifp);
2593 return (EADDRNOTAVAIL);
2595 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2596 ntohl(addr.s_addr));
2599 /* Reject interfaces which do not support multicast. */
2600 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2601 return (EOPNOTSUPP);
2603 imo = inp_findmoptions(inp);
2604 imo->imo_multicast_ifp = ifp;
2605 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2612 * Atomically set source filters on a socket for an IPv4 multicast group.
2614 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2617 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2619 struct __msfilterreq msfr;
2622 struct in_mfilter *imf;
2623 struct ip_moptions *imo;
2624 struct in_multi *inm;
2628 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2629 sizeof(struct __msfilterreq));
2633 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2636 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2637 msfr.msfr_fmode != MCAST_INCLUDE))
2640 if (msfr.msfr_group.ss_family != AF_INET ||
2641 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2644 gsa = (sockunion_t *)&msfr.msfr_group;
2645 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2648 gsa->sin.sin_port = 0; /* ignore port */
2650 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2651 return (EADDRNOTAVAIL);
2653 ifp = ifnet_byindex(msfr.msfr_ifindex);
2655 return (EADDRNOTAVAIL);
2658 * Take the INP write lock.
2659 * Check if this socket is a member of this group.
2661 imo = inp_findmoptions(inp);
2662 idx = imo_match_group(imo, ifp, &gsa->sa);
2663 if (idx == -1 || imo->imo_mfilters == NULL) {
2664 error = EADDRNOTAVAIL;
2665 goto out_inp_locked;
2667 inm = imo->imo_membership[idx];
2668 imf = &imo->imo_mfilters[idx];
2671 * Begin state merge transaction at socket layer.
2673 INP_WLOCK_ASSERT(inp);
2675 imf->imf_st[1] = msfr.msfr_fmode;
2678 * Apply any new source filters, if present.
2679 * Make a copy of the user-space source vector so
2680 * that we may copy them with a single copyin. This
2681 * allows us to deal with page faults up-front.
2683 if (msfr.msfr_nsrcs > 0) {
2684 struct in_msource *lims;
2685 struct sockaddr_in *psin;
2686 struct sockaddr_storage *kss, *pkss;
2691 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2692 __func__, (unsigned long)msfr.msfr_nsrcs);
2693 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2695 error = copyin(msfr.msfr_srcs, kss,
2696 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2705 * Mark all source filters as UNDEFINED at t1.
2706 * Restore new group filter mode, as imf_leave()
2707 * will set it to INCLUDE.
2710 imf->imf_st[1] = msfr.msfr_fmode;
2713 * Update socket layer filters at t1, lazy-allocating
2714 * new entries. This saves a bunch of memory at the
2715 * cost of one RB_FIND() per source entry; duplicate
2716 * entries in the msfr_nsrcs vector are ignored.
2717 * If we encounter an error, rollback transaction.
2719 * XXX This too could be replaced with a set-symmetric
2720 * difference like loop to avoid walking from root
2721 * every time, as the key space is common.
2723 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2724 psin = (struct sockaddr_in *)pkss;
2725 if (psin->sin_family != AF_INET) {
2726 error = EAFNOSUPPORT;
2729 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2733 error = imf_get_source(imf, psin, &lims);
2736 lims->imsl_st[1] = imf->imf_st[1];
2742 goto out_imf_rollback;
2744 INP_WLOCK_ASSERT(inp);
2748 * Begin state merge transaction at IGMP layer.
2750 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2751 IN_MULTI_LIST_LOCK();
2752 error = inm_merge(inm, imf);
2754 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2755 IN_MULTI_LIST_UNLOCK();
2756 goto out_in_multi_locked;
2759 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2760 error = igmp_change_state(inm);
2761 IN_MULTI_LIST_UNLOCK();
2763 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2765 out_in_multi_locked:
2783 * Set the IP multicast options in response to user setsockopt().
2785 * Many of the socket options handled in this function duplicate the
2786 * functionality of socket options in the regular unicast API. However,
2787 * it is not possible to merge the duplicate code, because the idempotence
2788 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2789 * the effects of these options must be treated as separate and distinct.
2791 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2792 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2793 * is refactored to no longer use vifs.
2796 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2798 struct ip_moptions *imo;
2804 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2805 * or is a divert socket, reject it.
2807 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2808 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2809 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2810 return (EOPNOTSUPP);
2812 switch (sopt->sopt_name) {
2813 case IP_MULTICAST_VIF: {
2816 * Select a multicast VIF for transmission.
2817 * Only useful if multicast forwarding is active.
2819 if (legal_vif_num == NULL) {
2823 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2826 if (!legal_vif_num(vifi) && (vifi != -1)) {
2830 imo = inp_findmoptions(inp);
2831 imo->imo_multicast_vif = vifi;
2836 case IP_MULTICAST_IF:
2837 error = inp_set_multicast_if(inp, sopt);
2840 case IP_MULTICAST_TTL: {
2844 * Set the IP time-to-live for outgoing multicast packets.
2845 * The original multicast API required a char argument,
2846 * which is inconsistent with the rest of the socket API.
2847 * We allow either a char or an int.
2849 if (sopt->sopt_valsize == sizeof(u_char)) {
2850 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2857 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2867 imo = inp_findmoptions(inp);
2868 imo->imo_multicast_ttl = ttl;
2873 case IP_MULTICAST_LOOP: {
2877 * Set the loopback flag for outgoing multicast packets.
2878 * Must be zero or one. The original multicast API required a
2879 * char argument, which is inconsistent with the rest
2880 * of the socket API. We allow either a char or an int.
2882 if (sopt->sopt_valsize == sizeof(u_char)) {
2883 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2890 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2894 loop = (u_char)iloop;
2896 imo = inp_findmoptions(inp);
2897 imo->imo_multicast_loop = !!loop;
2902 case IP_ADD_MEMBERSHIP:
2903 case IP_ADD_SOURCE_MEMBERSHIP:
2904 case MCAST_JOIN_GROUP:
2905 case MCAST_JOIN_SOURCE_GROUP:
2906 error = inp_join_group(inp, sopt);
2909 case IP_DROP_MEMBERSHIP:
2910 case IP_DROP_SOURCE_MEMBERSHIP:
2911 case MCAST_LEAVE_GROUP:
2912 case MCAST_LEAVE_SOURCE_GROUP:
2913 error = inp_leave_group(inp, sopt);
2916 case IP_BLOCK_SOURCE:
2917 case IP_UNBLOCK_SOURCE:
2918 case MCAST_BLOCK_SOURCE:
2919 case MCAST_UNBLOCK_SOURCE:
2920 error = inp_block_unblock_source(inp, sopt);
2924 error = inp_set_source_filters(inp, sopt);
2932 INP_UNLOCK_ASSERT(inp);
2938 * Expose IGMP's multicast filter mode and source list(s) to userland,
2939 * keyed by (ifindex, group).
2940 * The filter mode is written out as a uint32_t, followed by
2941 * 0..n of struct in_addr.
2942 * For use by ifmcstat(8).
2943 * SMPng: NOTE: unlocked read of ifindex space.
2946 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2948 struct in_addr src, group;
2950 struct ifmultiaddr *ifma;
2951 struct in_multi *inm;
2952 struct ip_msource *ims;
2956 uint32_t fmode, ifindex;
2961 if (req->newptr != NULL)
2968 if (ifindex <= 0 || ifindex > V_if_index) {
2969 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2974 group.s_addr = name[1];
2975 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2976 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2977 __func__, ntohl(group.s_addr));
2981 ifp = ifnet_byindex(ifindex);
2983 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2988 retval = sysctl_wire_old_buffer(req,
2989 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2993 IN_MULTI_LIST_LOCK();
2996 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2997 if (ifma->ifma_addr->sa_family != AF_INET ||
2998 ifma->ifma_protospec == NULL)
3000 inm = (struct in_multi *)ifma->ifma_protospec;
3001 if (!in_hosteq(inm->inm_addr, group))
3003 fmode = inm->inm_st[1].iss_fmode;
3004 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
3007 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
3008 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
3011 * Only copy-out sources which are in-mode.
3013 if (fmode != ims_get_mode(inm, ims, 1)) {
3014 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
3018 src.s_addr = htonl(ims->ims_haddr);
3019 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
3024 IF_ADDR_RUNLOCK(ifp);
3026 IN_MULTI_LIST_UNLOCK();
3031 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
3033 static const char *inm_modestrs[] = { "un", "in", "ex" };
3036 inm_mode_str(const int mode)
3039 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
3040 return (inm_modestrs[mode]);
3044 static const char *inm_statestrs[] = {
3057 inm_state_str(const int state)
3060 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3061 return (inm_statestrs[state]);
3066 * Dump an in_multi structure to the console.
3069 inm_print(const struct in_multi *inm)
3072 char addrbuf[INET_ADDRSTRLEN];
3074 if ((ktr_mask & KTR_IGMPV3) == 0)
3077 printf("%s: --- begin inm %p ---\n", __func__, inm);
3078 printf("addr %s ifp %p(%s) ifma %p\n",
3079 inet_ntoa_r(inm->inm_addr, addrbuf),
3081 inm->inm_ifp->if_xname,
3083 printf("timer %u state %s refcount %u scq.len %u\n",
3085 inm_state_str(inm->inm_state),
3087 inm->inm_scq.mq_len);
3088 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3093 for (t = 0; t < 2; t++) {
3094 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3095 inm_mode_str(inm->inm_st[t].iss_fmode),
3096 inm->inm_st[t].iss_asm,
3097 inm->inm_st[t].iss_ex,
3098 inm->inm_st[t].iss_in,
3099 inm->inm_st[t].iss_rec);
3101 printf("%s: --- end inm %p ---\n", __func__, inm);
3104 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3107 inm_print(const struct in_multi *inm)
3112 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3114 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);