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 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
239 inm_release_list_deferred(struct in_multi_head *inmh)
242 if (SLIST_EMPTY(inmh))
244 mtx_lock(&in_multi_free_mtx);
245 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
246 mtx_unlock(&in_multi_free_mtx);
247 GROUPTASK_ENQUEUE(&free_gtask);
251 inm_disconnect(struct in_multi *inm)
254 struct ifmultiaddr *ifma, *ll_ifma;
257 IF_ADDR_WLOCK_ASSERT(ifp);
258 ifma = inm->inm_ifma;
261 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
262 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
263 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
264 MPASS(ifma != ll_ifma);
265 ifma->ifma_llifma = NULL;
266 MPASS(ll_ifma->ifma_llifma == NULL);
267 MPASS(ll_ifma->ifma_ifp == ifp);
268 if (--ll_ifma->ifma_refcount == 0) {
269 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
270 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
271 if_freemulti(ll_ifma);
278 inm_release_deferred(struct in_multi *inm)
280 struct in_multi_head tmp;
282 IN_MULTI_LIST_LOCK_ASSERT();
283 MPASS(inm->inm_refcount > 0);
284 if (--inm->inm_refcount == 0) {
287 inm->inm_ifma->ifma_protospec = NULL;
288 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
289 inm_release_list_deferred(&tmp);
294 inm_release_task(void *arg __unused)
296 struct in_multi_head inm_free_tmp;
297 struct in_multi *inm, *tinm;
299 SLIST_INIT(&inm_free_tmp);
300 mtx_lock(&in_multi_free_mtx);
301 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
302 mtx_unlock(&in_multi_free_mtx);
304 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
305 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
313 * Initialize an in_mfilter structure to a known state at t0, t1
314 * with an empty source filter list.
317 imf_init(struct in_mfilter *imf, const int st0, const int st1)
319 memset(imf, 0, sizeof(struct in_mfilter));
320 RB_INIT(&imf->imf_sources);
321 imf->imf_st[0] = st0;
322 imf->imf_st[1] = st1;
326 * Function for looking up an in_multi record for an IPv4 multicast address
327 * on a given interface. ifp must be valid. If no record found, return NULL.
328 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
331 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
333 struct ifmultiaddr *ifma;
334 struct in_multi *inm;
336 IN_MULTI_LIST_LOCK_ASSERT();
337 IF_ADDR_LOCK_ASSERT(ifp);
340 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
341 if (ifma->ifma_addr->sa_family != AF_INET ||
342 ifma->ifma_protospec == NULL)
344 inm = (struct in_multi *)ifma->ifma_protospec;
345 if (inm->inm_addr.s_addr == ina.s_addr)
353 * Wrapper for inm_lookup_locked().
354 * The IF_ADDR_LOCK will be taken on ifp and released on return.
357 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
359 struct in_multi *inm;
361 IN_MULTI_LIST_LOCK_ASSERT();
363 inm = inm_lookup_locked(ifp, ina);
364 IF_ADDR_RUNLOCK(ifp);
370 * Resize the ip_moptions vector to the next power-of-two minus 1.
371 * May be called with locks held; do not sleep.
374 imo_grow(struct ip_moptions *imo)
376 struct in_multi **nmships;
377 struct in_multi **omships;
378 struct in_mfilter *nmfilters;
379 struct in_mfilter *omfilters;
386 omships = imo->imo_membership;
387 omfilters = imo->imo_mfilters;
388 oldmax = imo->imo_max_memberships;
389 newmax = ((oldmax + 1) * 2) - 1;
391 if (newmax <= IP_MAX_MEMBERSHIPS) {
392 nmships = (struct in_multi **)realloc(omships,
393 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
394 nmfilters = (struct in_mfilter *)realloc(omfilters,
395 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
396 if (nmships != NULL && nmfilters != NULL) {
397 /* Initialize newly allocated source filter heads. */
398 for (idx = oldmax; idx < newmax; idx++) {
399 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
402 imo->imo_max_memberships = newmax;
403 imo->imo_membership = nmships;
404 imo->imo_mfilters = nmfilters;
408 if (nmships == NULL || nmfilters == NULL) {
410 free(nmships, M_IPMOPTS);
411 if (nmfilters != NULL)
412 free(nmfilters, M_INMFILTER);
413 return (ETOOMANYREFS);
420 * Find an IPv4 multicast group entry for this ip_moptions instance
421 * which matches the specified group, and optionally an interface.
422 * Return its index into the array, or -1 if not found.
425 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
426 const struct sockaddr *group)
428 const struct sockaddr_in *gsin;
429 struct in_multi **pinm;
433 gsin = (const struct sockaddr_in *)group;
435 /* The imo_membership array may be lazy allocated. */
436 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
439 nmships = imo->imo_num_memberships;
440 pinm = &imo->imo_membership[0];
441 for (idx = 0; idx < nmships; idx++, pinm++) {
444 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
445 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
456 * Find an IPv4 multicast source entry for this imo which matches
457 * the given group index for this socket, and source address.
459 * NOTE: This does not check if the entry is in-mode, merely if
460 * it exists, which may not be the desired behaviour.
462 static struct in_msource *
463 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
464 const struct sockaddr *src)
466 struct ip_msource find;
467 struct in_mfilter *imf;
468 struct ip_msource *ims;
469 const sockunion_t *psa;
471 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
472 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
473 ("%s: invalid index %d\n", __func__, (int)gidx));
475 /* The imo_mfilters array may be lazy allocated. */
476 if (imo->imo_mfilters == NULL)
478 imf = &imo->imo_mfilters[gidx];
480 /* Source trees are keyed in host byte order. */
481 psa = (const sockunion_t *)src;
482 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
483 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
485 return ((struct in_msource *)ims);
489 * Perform filtering for multicast datagrams on a socket by group and source.
491 * Returns 0 if a datagram should be allowed through, or various error codes
492 * if the socket was not a member of the group, or the source was muted, etc.
495 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
496 const struct sockaddr *group, const struct sockaddr *src)
499 struct in_msource *ims;
502 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
504 gidx = imo_match_group(imo, ifp, group);
506 return (MCAST_NOTGMEMBER);
509 * Check if the source was included in an (S,G) join.
510 * Allow reception on exclusive memberships by default,
511 * reject reception on inclusive memberships by default.
512 * Exclude source only if an in-mode exclude filter exists.
513 * Include source only if an in-mode include filter exists.
514 * NOTE: We are comparing group state here at IGMP t1 (now)
515 * with socket-layer t0 (since last downcall).
517 mode = imo->imo_mfilters[gidx].imf_st[1];
518 ims = imo_match_source(imo, gidx, src);
520 if ((ims == NULL && mode == MCAST_INCLUDE) ||
521 (ims != NULL && ims->imsl_st[0] != mode))
522 return (MCAST_NOTSMEMBER);
528 * Find and return a reference to an in_multi record for (ifp, group),
529 * and bump its reference count.
530 * If one does not exist, try to allocate it, and update link-layer multicast
531 * filters on ifp to listen for group.
532 * Assumes the IN_MULTI lock is held across the call.
533 * Return 0 if successful, otherwise return an appropriate error code.
536 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
537 struct in_multi **pinm)
539 struct sockaddr_in gsin;
540 struct ifmultiaddr *ifma;
541 struct in_ifinfo *ii;
542 struct in_multi *inm;
545 IN_MULTI_LOCK_ASSERT();
547 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
548 IN_MULTI_LIST_LOCK();
549 inm = inm_lookup(ifp, *group);
552 * If we already joined this group, just bump the
553 * refcount and return it.
555 KASSERT(inm->inm_refcount >= 1,
556 ("%s: bad refcount %d", __func__, inm->inm_refcount));
557 inm_acquire_locked(inm);
560 IN_MULTI_LIST_UNLOCK();
564 memset(&gsin, 0, sizeof(gsin));
565 gsin.sin_family = AF_INET;
566 gsin.sin_len = sizeof(struct sockaddr_in);
567 gsin.sin_addr = *group;
570 * Check if a link-layer group is already associated
571 * with this network-layer group on the given ifnet.
573 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
577 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
578 IN_MULTI_LIST_LOCK();
582 * If something other than netinet is occupying the link-layer
583 * group, print a meaningful error message and back out of
585 * Otherwise, bump the refcount on the existing network-layer
586 * group association and return it.
588 if (ifma->ifma_protospec != NULL) {
589 inm = (struct in_multi *)ifma->ifma_protospec;
591 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
593 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
594 ("%s: ifma not AF_INET", __func__));
595 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
596 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
597 !in_hosteq(inm->inm_addr, *group)) {
598 char addrbuf[INET_ADDRSTRLEN];
600 panic("%s: ifma %p is inconsistent with %p (%s)",
601 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
604 inm_acquire_locked(inm);
609 IF_ADDR_WLOCK_ASSERT(ifp);
612 * A new in_multi record is needed; allocate and initialize it.
613 * We DO NOT perform an IGMP join as the in_ layer may need to
614 * push an initial source list down to IGMP to support SSM.
616 * The initial source filter state is INCLUDE, {} as per the RFC.
618 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
620 IF_ADDR_WUNLOCK(ifp);
621 IN_MULTI_LIST_UNLOCK();
622 if_delmulti_ifma(ifma);
625 inm->inm_addr = *group;
627 inm->inm_igi = ii->ii_igmp;
628 inm->inm_ifma = ifma;
629 inm->inm_refcount = 1;
630 inm->inm_state = IGMP_NOT_MEMBER;
631 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
632 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
633 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
634 RB_INIT(&inm->inm_srcs);
636 ifma->ifma_protospec = inm;
640 IF_ADDR_WUNLOCK(ifp);
641 IN_MULTI_LIST_UNLOCK();
646 * Drop a reference to an in_multi record.
648 * If the refcount drops to 0, free the in_multi record and
649 * delete the underlying link-layer membership.
652 inm_release(struct in_multi *inm)
654 struct ifmultiaddr *ifma;
656 struct vnet *saved_vnet;
658 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
659 MPASS(inm->inm_refcount == 0);
660 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
662 ifma = inm->inm_ifma;
665 /* XXX this access is not covered by IF_ADDR_LOCK */
666 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
668 saved_vnet = curvnet;
669 curvnet = ifp->if_vnet;
672 free(inm, M_IPMADDR);
674 if_delmulti_ifma_flags(ifma, 1);
676 curvnet = saved_vnet;
682 * Clear recorded source entries for a group.
683 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
684 * FIXME: Should reap.
687 inm_clear_recorded(struct in_multi *inm)
689 struct ip_msource *ims;
691 IN_MULTI_LIST_LOCK_ASSERT();
693 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
696 --inm->inm_st[1].iss_rec;
699 KASSERT(inm->inm_st[1].iss_rec == 0,
700 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
704 * Record a source as pending for a Source-Group IGMPv3 query.
705 * This lives here as it modifies the shared tree.
707 * inm is the group descriptor.
708 * naddr is the address of the source to record in network-byte order.
710 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
711 * lazy-allocate a source node in response to an SG query.
712 * Otherwise, no allocation is performed. This saves some memory
713 * with the trade-off that the source will not be reported to the
714 * router if joined in the window between the query response and
715 * the group actually being joined on the local host.
717 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
718 * This turns off the allocation of a recorded source entry if
719 * the group has not been joined.
721 * Return 0 if the source didn't exist or was already marked as recorded.
722 * Return 1 if the source was marked as recorded by this function.
723 * Return <0 if any error occurred (negated errno code).
726 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
728 struct ip_msource find;
729 struct ip_msource *ims, *nims;
731 IN_MULTI_LIST_LOCK_ASSERT();
733 find.ims_haddr = ntohl(naddr);
734 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
735 if (ims && ims->ims_stp)
738 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
740 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
744 nims->ims_haddr = find.ims_haddr;
745 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
751 * Mark the source as recorded and update the recorded
755 ++inm->inm_st[1].iss_rec;
761 * Return a pointer to an in_msource owned by an in_mfilter,
762 * given its source address.
763 * Lazy-allocate if needed. If this is a new entry its filter state is
766 * imf is the filter set being modified.
767 * haddr is the source address in *host* byte-order.
769 * SMPng: May be called with locks held; malloc must not block.
772 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
773 struct in_msource **plims)
775 struct ip_msource find;
776 struct ip_msource *ims, *nims;
777 struct in_msource *lims;
784 /* key is host byte order */
785 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
786 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
787 lims = (struct in_msource *)ims;
789 if (imf->imf_nsrc == in_mcast_maxsocksrc)
791 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
795 lims = (struct in_msource *)nims;
796 lims->ims_haddr = find.ims_haddr;
797 lims->imsl_st[0] = MCAST_UNDEFINED;
798 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
808 * Graft a source entry into an existing socket-layer filter set,
809 * maintaining any required invariants and checking allocations.
811 * The source is marked as being in the new filter mode at t1.
813 * Return the pointer to the new node, otherwise return NULL.
815 static struct in_msource *
816 imf_graft(struct in_mfilter *imf, const uint8_t st1,
817 const struct sockaddr_in *psin)
819 struct ip_msource *nims;
820 struct in_msource *lims;
822 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
826 lims = (struct in_msource *)nims;
827 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
828 lims->imsl_st[0] = MCAST_UNDEFINED;
829 lims->imsl_st[1] = st1;
830 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
837 * Prune a source entry from an existing socket-layer filter set,
838 * maintaining any required invariants and checking allocations.
840 * The source is marked as being left at t1, it is not freed.
842 * Return 0 if no error occurred, otherwise return an errno value.
845 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
847 struct ip_msource find;
848 struct ip_msource *ims;
849 struct in_msource *lims;
851 /* key is host byte order */
852 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
853 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
856 lims = (struct in_msource *)ims;
857 lims->imsl_st[1] = MCAST_UNDEFINED;
862 * Revert socket-layer filter set deltas at t1 to t0 state.
865 imf_rollback(struct in_mfilter *imf)
867 struct ip_msource *ims, *tims;
868 struct in_msource *lims;
870 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
871 lims = (struct in_msource *)ims;
872 if (lims->imsl_st[0] == lims->imsl_st[1]) {
873 /* no change at t1 */
875 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
876 /* revert change to existing source at t1 */
877 lims->imsl_st[1] = lims->imsl_st[0];
879 /* revert source added t1 */
880 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
881 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
882 free(ims, M_INMFILTER);
886 imf->imf_st[1] = imf->imf_st[0];
890 * Mark socket-layer filter set as INCLUDE {} at t1.
893 imf_leave(struct in_mfilter *imf)
895 struct ip_msource *ims;
896 struct in_msource *lims;
898 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
899 lims = (struct in_msource *)ims;
900 lims->imsl_st[1] = MCAST_UNDEFINED;
902 imf->imf_st[1] = MCAST_INCLUDE;
906 * Mark socket-layer filter set deltas as committed.
909 imf_commit(struct in_mfilter *imf)
911 struct ip_msource *ims;
912 struct in_msource *lims;
914 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
915 lims = (struct in_msource *)ims;
916 lims->imsl_st[0] = lims->imsl_st[1];
918 imf->imf_st[0] = imf->imf_st[1];
922 * Reap unreferenced sources from socket-layer filter set.
925 imf_reap(struct in_mfilter *imf)
927 struct ip_msource *ims, *tims;
928 struct in_msource *lims;
930 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
931 lims = (struct in_msource *)ims;
932 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
933 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
934 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
935 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
936 free(ims, M_INMFILTER);
943 * Purge socket-layer filter set.
946 imf_purge(struct in_mfilter *imf)
948 struct ip_msource *ims, *tims;
950 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
951 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
952 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
953 free(ims, M_INMFILTER);
956 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
957 KASSERT(RB_EMPTY(&imf->imf_sources),
958 ("%s: imf_sources not empty", __func__));
962 * Look up a source filter entry for a multicast group.
964 * inm is the group descriptor to work with.
965 * haddr is the host-byte-order IPv4 address to look up.
966 * noalloc may be non-zero to suppress allocation of sources.
967 * *pims will be set to the address of the retrieved or allocated source.
969 * SMPng: NOTE: may be called with locks held.
970 * Return 0 if successful, otherwise return a non-zero error code.
973 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
974 const int noalloc, struct ip_msource **pims)
976 struct ip_msource find;
977 struct ip_msource *ims, *nims;
979 find.ims_haddr = haddr;
980 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
981 if (ims == NULL && !noalloc) {
982 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
984 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
988 nims->ims_haddr = haddr;
989 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
993 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
1003 * Merge socket-layer source into IGMP-layer source.
1004 * If rollback is non-zero, perform the inverse of the merge.
1007 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
1010 int n = rollback ? -1 : 1;
1012 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
1013 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
1014 __func__, n, ims->ims_haddr);
1015 ims->ims_st[1].ex -= n;
1016 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
1017 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
1018 __func__, n, ims->ims_haddr);
1019 ims->ims_st[1].in -= n;
1022 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
1023 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
1024 __func__, n, ims->ims_haddr);
1025 ims->ims_st[1].ex += n;
1026 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
1027 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
1028 __func__, n, ims->ims_haddr);
1029 ims->ims_st[1].in += n;
1034 * Atomically update the global in_multi state, when a membership's
1035 * filter list is being updated in any way.
1037 * imf is the per-inpcb-membership group filter pointer.
1038 * A fake imf may be passed for in-kernel consumers.
1040 * XXX This is a candidate for a set-symmetric-difference style loop
1041 * which would eliminate the repeated lookup from root of ims nodes,
1042 * as they share the same key space.
1044 * If any error occurred this function will back out of refcounts
1045 * and return a non-zero value.
1048 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1050 struct ip_msource *ims, *nims;
1051 struct in_msource *lims;
1052 int schanged, error;
1058 IN_MULTI_LIST_LOCK_ASSERT();
1061 * Update the source filters first, as this may fail.
1062 * Maintain count of in-mode filters at t0, t1. These are
1063 * used to work out if we transition into ASM mode or not.
1064 * Maintain a count of source filters whose state was
1065 * actually modified by this operation.
1067 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1068 lims = (struct in_msource *)ims;
1069 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1070 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1071 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1072 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1076 ims_merge(nims, lims, 0);
1079 struct ip_msource *bims;
1081 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1082 lims = (struct in_msource *)ims;
1083 if (lims->imsl_st[0] == lims->imsl_st[1])
1085 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1088 ims_merge(bims, lims, 1);
1093 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1094 __func__, nsrc0, nsrc1);
1096 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1097 if (imf->imf_st[0] == imf->imf_st[1] &&
1098 imf->imf_st[1] == MCAST_INCLUDE) {
1100 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1101 --inm->inm_st[1].iss_in;
1105 /* Handle filter mode transition on socket. */
1106 if (imf->imf_st[0] != imf->imf_st[1]) {
1107 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1108 __func__, imf->imf_st[0], imf->imf_st[1]);
1110 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1111 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1112 --inm->inm_st[1].iss_ex;
1113 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1114 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1115 --inm->inm_st[1].iss_in;
1118 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1119 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1120 inm->inm_st[1].iss_ex++;
1121 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1122 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1123 inm->inm_st[1].iss_in++;
1128 * Track inm filter state in terms of listener counts.
1129 * If there are any exclusive listeners, stack-wide
1130 * membership is exclusive.
1131 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1132 * If no listeners remain, state is undefined at t1,
1133 * and the IGMP lifecycle for this group should finish.
1135 if (inm->inm_st[1].iss_ex > 0) {
1136 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1137 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1138 } else if (inm->inm_st[1].iss_in > 0) {
1139 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1140 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1142 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1143 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1146 /* Decrement ASM listener count on transition out of ASM mode. */
1147 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1148 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1149 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1150 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1151 --inm->inm_st[1].iss_asm;
1155 /* Increment ASM listener count on transition to ASM mode. */
1156 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1157 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1158 inm->inm_st[1].iss_asm++;
1161 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1166 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1173 * Mark an in_multi's filter set deltas as committed.
1174 * Called by IGMP after a state change has been enqueued.
1177 inm_commit(struct in_multi *inm)
1179 struct ip_msource *ims;
1181 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1182 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1185 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1186 ims->ims_st[0] = ims->ims_st[1];
1188 inm->inm_st[0] = inm->inm_st[1];
1192 * Reap unreferenced nodes from an in_multi's filter set.
1195 inm_reap(struct in_multi *inm)
1197 struct ip_msource *ims, *tims;
1199 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1200 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1201 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1204 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1205 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1206 free(ims, M_IPMSOURCE);
1212 * Purge all source nodes from an in_multi's filter set.
1215 inm_purge(struct in_multi *inm)
1217 struct ip_msource *ims, *tims;
1219 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1220 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1221 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1222 free(ims, M_IPMSOURCE);
1228 * Join a multicast group; unlocked entry point.
1230 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1231 * is not held. Fortunately, ifp is unlikely to have been detached
1232 * at this point, so we assume it's OK to recurse.
1235 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1236 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1241 error = in_joingroup_locked(ifp, gina, imf, pinm);
1248 * Join a multicast group; real entry point.
1250 * Only preserves atomicity at inm level.
1251 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1253 * If the IGMP downcall fails, the group is not joined, and an error
1257 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1258 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1260 struct in_mfilter timf;
1261 struct in_multi *inm;
1264 IN_MULTI_LOCK_ASSERT();
1265 IN_MULTI_LIST_UNLOCK_ASSERT();
1267 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1268 ntohl(gina->s_addr), ifp, ifp->if_xname);
1274 * If no imf was specified (i.e. kernel consumer),
1275 * fake one up and assume it is an ASM join.
1278 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1282 error = in_getmulti(ifp, gina, &inm);
1284 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1287 IN_MULTI_LIST_LOCK();
1288 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1289 error = inm_merge(inm, imf);
1291 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1292 goto out_inm_release;
1295 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1296 error = igmp_change_state(inm);
1298 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1299 goto out_inm_release;
1305 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1306 inm_release_deferred(inm);
1310 IN_MULTI_LIST_UNLOCK();
1316 * Leave a multicast group; unlocked entry point.
1319 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1324 error = in_leavegroup_locked(inm, imf);
1331 * Leave a multicast group; real entry point.
1332 * All source filters will be expunged.
1334 * Only preserves atomicity at inm level.
1336 * Holding the write lock for the INP which contains imf
1337 * is highly advisable. We can't assert for it as imf does not
1338 * contain a back-pointer to the owning inp.
1340 * Note: This is not the same as inm_release(*) as this function also
1341 * makes a state change downcall into IGMP.
1344 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1346 struct in_mfilter timf;
1351 IN_MULTI_LOCK_ASSERT();
1352 IN_MULTI_LIST_UNLOCK_ASSERT();
1354 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1355 inm, ntohl(inm->inm_addr.s_addr),
1356 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1360 * If no imf was specified (i.e. kernel consumer),
1361 * fake one up and assume it is an ASM join.
1364 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1369 * Begin state merge transaction at IGMP layer.
1371 * As this particular invocation should not cause any memory
1372 * to be allocated, and there is no opportunity to roll back
1373 * the transaction, it MUST NOT fail.
1375 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1376 IN_MULTI_LIST_LOCK();
1377 error = inm_merge(inm, imf);
1378 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1380 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1381 CURVNET_SET(inm->inm_ifp->if_vnet);
1382 error = igmp_change_state(inm);
1383 IF_ADDR_WLOCK(inm->inm_ifp);
1384 inm_release_deferred(inm);
1385 IF_ADDR_WUNLOCK(inm->inm_ifp);
1386 IN_MULTI_LIST_UNLOCK();
1389 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1391 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1396 /*#ifndef BURN_BRIDGES*/
1398 * Join an IPv4 multicast group in (*,G) exclusive mode.
1399 * The group must be a 224.0.0.0/24 link-scope group.
1400 * This KPI is for legacy kernel consumers only.
1403 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1405 struct in_multi *pinm;
1408 char addrbuf[INET_ADDRSTRLEN];
1411 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1412 ("%s: %s not in 224.0.0.0/24", __func__,
1413 inet_ntoa_r(*ap, addrbuf)));
1415 error = in_joingroup(ifp, ap, NULL, &pinm);
1423 * Block or unblock an ASM multicast source on an inpcb.
1424 * This implements the delta-based API described in RFC 3678.
1426 * The delta-based API applies only to exclusive-mode memberships.
1427 * An IGMP downcall will be performed.
1429 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1431 * Return 0 if successful, otherwise return an appropriate error code.
1434 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1436 struct group_source_req gsr;
1437 sockunion_t *gsa, *ssa;
1439 struct in_mfilter *imf;
1440 struct ip_moptions *imo;
1441 struct in_msource *ims;
1442 struct in_multi *inm;
1451 memset(&gsr, 0, sizeof(struct group_source_req));
1452 gsa = (sockunion_t *)&gsr.gsr_group;
1453 ssa = (sockunion_t *)&gsr.gsr_source;
1455 switch (sopt->sopt_name) {
1456 case IP_BLOCK_SOURCE:
1457 case IP_UNBLOCK_SOURCE: {
1458 struct ip_mreq_source mreqs;
1460 error = sooptcopyin(sopt, &mreqs,
1461 sizeof(struct ip_mreq_source),
1462 sizeof(struct ip_mreq_source));
1466 gsa->sin.sin_family = AF_INET;
1467 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1468 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1470 ssa->sin.sin_family = AF_INET;
1471 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1472 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1474 if (!in_nullhost(mreqs.imr_interface))
1475 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1477 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1480 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1481 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1485 case MCAST_BLOCK_SOURCE:
1486 case MCAST_UNBLOCK_SOURCE:
1487 error = sooptcopyin(sopt, &gsr,
1488 sizeof(struct group_source_req),
1489 sizeof(struct group_source_req));
1493 if (gsa->sin.sin_family != AF_INET ||
1494 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1497 if (ssa->sin.sin_family != AF_INET ||
1498 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1501 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1502 return (EADDRNOTAVAIL);
1504 ifp = ifnet_byindex(gsr.gsr_interface);
1506 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1511 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1512 __func__, sopt->sopt_name);
1513 return (EOPNOTSUPP);
1517 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1521 * Check if we are actually a member of this group.
1523 imo = inp_findmoptions(inp);
1524 idx = imo_match_group(imo, ifp, &gsa->sa);
1525 if (idx == -1 || imo->imo_mfilters == NULL) {
1526 error = EADDRNOTAVAIL;
1527 goto out_inp_locked;
1530 KASSERT(imo->imo_mfilters != NULL,
1531 ("%s: imo_mfilters not allocated", __func__));
1532 imf = &imo->imo_mfilters[idx];
1533 inm = imo->imo_membership[idx];
1536 * Attempting to use the delta-based API on an
1537 * non exclusive-mode membership is an error.
1539 fmode = imf->imf_st[0];
1540 if (fmode != MCAST_EXCLUDE) {
1542 goto out_inp_locked;
1546 * Deal with error cases up-front:
1547 * Asked to block, but already blocked; or
1548 * Asked to unblock, but nothing to unblock.
1549 * If adding a new block entry, allocate it.
1551 ims = imo_match_source(imo, idx, &ssa->sa);
1552 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1553 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1554 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1555 error = EADDRNOTAVAIL;
1556 goto out_inp_locked;
1559 INP_WLOCK_ASSERT(inp);
1562 * Begin state merge transaction at socket layer.
1565 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1566 ims = imf_graft(imf, fmode, &ssa->sin);
1570 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1571 error = imf_prune(imf, &ssa->sin);
1575 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1576 goto out_imf_rollback;
1580 * Begin state merge transaction at IGMP layer.
1583 IN_MULTI_LIST_LOCK();
1584 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1585 error = inm_merge(inm, imf);
1587 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1588 goto out_in_multi_locked;
1591 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1592 error = igmp_change_state(inm);
1594 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1596 out_in_multi_locked:
1614 * Given an inpcb, return its multicast options structure pointer. Accepts
1615 * an unlocked inpcb pointer, but will return it locked. May sleep.
1617 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1618 * SMPng: NOTE: Returns with the INP write lock held.
1620 static struct ip_moptions *
1621 inp_findmoptions(struct inpcb *inp)
1623 struct ip_moptions *imo;
1624 struct in_multi **immp;
1625 struct in_mfilter *imfp;
1629 if (inp->inp_moptions != NULL)
1630 return (inp->inp_moptions);
1634 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1635 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1637 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1638 M_INMFILTER, M_WAITOK);
1640 imo->imo_multicast_ifp = NULL;
1641 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1642 imo->imo_multicast_vif = -1;
1643 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1644 imo->imo_multicast_loop = in_mcast_loop;
1645 imo->imo_num_memberships = 0;
1646 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1647 imo->imo_membership = immp;
1649 /* Initialize per-group source filters. */
1650 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1651 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1652 imo->imo_mfilters = imfp;
1655 if (inp->inp_moptions != NULL) {
1656 free(imfp, M_INMFILTER);
1657 free(immp, M_IPMOPTS);
1658 free(imo, M_IPMOPTS);
1659 return (inp->inp_moptions);
1661 inp->inp_moptions = imo;
1666 inp_gcmoptions(epoch_context_t ctx)
1668 struct ip_moptions *imo;
1669 struct in_mfilter *imf;
1670 struct in_multi *inm;
1672 struct vnet *saved_vnet;
1673 size_t idx, nmships;
1675 imo = __containerof(ctx, struct ip_moptions, imo_epoch_ctx);
1677 nmships = imo->imo_num_memberships;
1678 for (idx = 0; idx < nmships; ++idx) {
1679 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1682 inm = imo->imo_membership[idx];
1685 saved_vnet = curvnet;
1686 curvnet = ifp->if_vnet;
1688 (void)in_leavegroup(inm, imf);
1690 curvnet = saved_vnet;
1695 if (imo->imo_mfilters)
1696 free(imo->imo_mfilters, M_INMFILTER);
1697 free(imo->imo_membership, M_IPMOPTS);
1698 free(imo, M_IPMOPTS);
1702 * Discard the IP multicast options (and source filters). To minimize
1703 * the amount of work done while holding locks such as the INP's
1704 * pcbinfo lock (which is used in the receive path), the free
1705 * operation is deferred to the epoch callback task.
1708 inp_freemoptions(struct ip_moptions *imo)
1712 epoch_call(net_epoch_preempt, &imo->imo_epoch_ctx, inp_gcmoptions);
1716 * Atomically get source filters on a socket for an IPv4 multicast group.
1717 * Called with INP lock held; returns with lock released.
1720 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1722 struct __msfilterreq msfr;
1725 struct ip_moptions *imo;
1726 struct in_mfilter *imf;
1727 struct ip_msource *ims;
1728 struct in_msource *lims;
1729 struct sockaddr_in *psin;
1730 struct sockaddr_storage *ptss;
1731 struct sockaddr_storage *tss;
1733 size_t idx, nsrcs, ncsrcs;
1735 INP_WLOCK_ASSERT(inp);
1737 imo = inp->inp_moptions;
1738 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1742 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1743 sizeof(struct __msfilterreq));
1747 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1750 ifp = ifnet_byindex(msfr.msfr_ifindex);
1757 * Lookup group on the socket.
1759 gsa = (sockunion_t *)&msfr.msfr_group;
1760 idx = imo_match_group(imo, ifp, &gsa->sa);
1761 if (idx == -1 || imo->imo_mfilters == NULL) {
1763 return (EADDRNOTAVAIL);
1765 imf = &imo->imo_mfilters[idx];
1768 * Ignore memberships which are in limbo.
1770 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1774 msfr.msfr_fmode = imf->imf_st[1];
1777 * If the user specified a buffer, copy out the source filter
1778 * entries to userland gracefully.
1779 * We only copy out the number of entries which userland
1780 * has asked for, but we always tell userland how big the
1781 * buffer really needs to be.
1783 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1784 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1786 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1787 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1788 M_TEMP, M_NOWAIT | M_ZERO);
1796 * Count number of sources in-mode at t0.
1797 * If buffer space exists and remains, copy out source entries.
1799 nsrcs = msfr.msfr_nsrcs;
1802 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1803 lims = (struct in_msource *)ims;
1804 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1805 lims->imsl_st[0] != imf->imf_st[0])
1808 if (tss != NULL && nsrcs > 0) {
1809 psin = (struct sockaddr_in *)ptss;
1810 psin->sin_family = AF_INET;
1811 psin->sin_len = sizeof(struct sockaddr_in);
1812 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1822 error = copyout(tss, msfr.msfr_srcs,
1823 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1829 msfr.msfr_nsrcs = ncsrcs;
1830 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1836 * Return the IP multicast options in response to user getsockopt().
1839 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1841 struct rm_priotracker in_ifa_tracker;
1842 struct ip_mreqn mreqn;
1843 struct ip_moptions *imo;
1845 struct in_ifaddr *ia;
1850 imo = inp->inp_moptions;
1852 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1853 * or is a divert socket, reject it.
1855 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1856 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1857 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1859 return (EOPNOTSUPP);
1863 switch (sopt->sopt_name) {
1864 case IP_MULTICAST_VIF:
1866 optval = imo->imo_multicast_vif;
1870 error = sooptcopyout(sopt, &optval, sizeof(int));
1873 case IP_MULTICAST_IF:
1874 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1876 ifp = imo->imo_multicast_ifp;
1877 if (!in_nullhost(imo->imo_multicast_addr)) {
1878 mreqn.imr_address = imo->imo_multicast_addr;
1879 } else if (ifp != NULL) {
1880 mreqn.imr_ifindex = ifp->if_index;
1881 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1884 IA_SIN(ia)->sin_addr;
1885 ifa_free(&ia->ia_ifa);
1890 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1891 error = sooptcopyout(sopt, &mreqn,
1892 sizeof(struct ip_mreqn));
1894 error = sooptcopyout(sopt, &mreqn.imr_address,
1895 sizeof(struct in_addr));
1899 case IP_MULTICAST_TTL:
1901 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1903 optval = coptval = imo->imo_multicast_ttl;
1905 if (sopt->sopt_valsize == sizeof(u_char))
1906 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1908 error = sooptcopyout(sopt, &optval, sizeof(int));
1911 case IP_MULTICAST_LOOP:
1913 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1915 optval = coptval = imo->imo_multicast_loop;
1917 if (sopt->sopt_valsize == sizeof(u_char))
1918 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1920 error = sooptcopyout(sopt, &optval, sizeof(int));
1925 error = EADDRNOTAVAIL;
1928 error = inp_get_source_filters(inp, sopt);
1934 error = ENOPROTOOPT;
1938 INP_UNLOCK_ASSERT(inp);
1944 * Look up the ifnet to use for a multicast group membership,
1945 * given the IPv4 address of an interface, and the IPv4 group address.
1947 * This routine exists to support legacy multicast applications
1948 * which do not understand that multicast memberships are scoped to
1949 * specific physical links in the networking stack, or which need
1950 * to join link-scope groups before IPv4 addresses are configured.
1952 * If inp is non-NULL, use this socket's current FIB number for any
1953 * required FIB lookup.
1954 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1955 * and use its ifp; usually, this points to the default next-hop.
1957 * If the FIB lookup fails, attempt to use the first non-loopback
1958 * interface with multicast capability in the system as a
1959 * last resort. The legacy IPv4 ASM API requires that we do
1960 * this in order to allow groups to be joined when the routing
1961 * table has not yet been populated during boot.
1963 * Returns NULL if no ifp could be found.
1965 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1966 * FUTURE: Implement IPv4 source-address selection.
1968 static struct ifnet *
1969 inp_lookup_mcast_ifp(const struct inpcb *inp,
1970 const struct sockaddr_in *gsin, const struct in_addr ina)
1972 struct rm_priotracker in_ifa_tracker;
1974 struct nhop4_basic nh4;
1977 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1978 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1979 ("%s: not multicast", __func__));
1982 if (!in_nullhost(ina)) {
1983 INADDR_TO_IFP(ina, ifp);
1985 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1986 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1989 struct in_ifaddr *ia;
1993 IN_IFADDR_RLOCK(&in_ifa_tracker);
1994 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1996 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1997 (mifp->if_flags & IFF_MULTICAST)) {
2002 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2010 * Join an IPv4 multicast group, possibly with a source.
2013 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
2015 struct group_source_req gsr;
2016 sockunion_t *gsa, *ssa;
2018 struct in_mfilter *imf;
2019 struct ip_moptions *imo;
2020 struct in_multi *inm;
2021 struct in_msource *lims;
2031 memset(&gsr, 0, sizeof(struct group_source_req));
2032 gsa = (sockunion_t *)&gsr.gsr_group;
2033 gsa->ss.ss_family = AF_UNSPEC;
2034 ssa = (sockunion_t *)&gsr.gsr_source;
2035 ssa->ss.ss_family = AF_UNSPEC;
2037 switch (sopt->sopt_name) {
2038 case IP_ADD_MEMBERSHIP:
2039 case IP_ADD_SOURCE_MEMBERSHIP: {
2040 struct ip_mreq_source mreqs;
2042 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
2043 error = sooptcopyin(sopt, &mreqs,
2044 sizeof(struct ip_mreq),
2045 sizeof(struct ip_mreq));
2047 * Do argument switcharoo from ip_mreq into
2048 * ip_mreq_source to avoid using two instances.
2050 mreqs.imr_interface = mreqs.imr_sourceaddr;
2051 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2052 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2053 error = sooptcopyin(sopt, &mreqs,
2054 sizeof(struct ip_mreq_source),
2055 sizeof(struct ip_mreq_source));
2060 gsa->sin.sin_family = AF_INET;
2061 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2062 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2064 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2065 ssa->sin.sin_family = AF_INET;
2066 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2067 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2070 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2073 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2074 mreqs.imr_interface);
2075 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2076 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2080 case MCAST_JOIN_GROUP:
2081 case MCAST_JOIN_SOURCE_GROUP:
2082 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2083 error = sooptcopyin(sopt, &gsr,
2084 sizeof(struct group_req),
2085 sizeof(struct group_req));
2086 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2087 error = sooptcopyin(sopt, &gsr,
2088 sizeof(struct group_source_req),
2089 sizeof(struct group_source_req));
2094 if (gsa->sin.sin_family != AF_INET ||
2095 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2099 * Overwrite the port field if present, as the sockaddr
2100 * being copied in may be matched with a binary comparison.
2102 gsa->sin.sin_port = 0;
2103 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2104 if (ssa->sin.sin_family != AF_INET ||
2105 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2107 ssa->sin.sin_port = 0;
2110 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2113 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2114 return (EADDRNOTAVAIL);
2115 ifp = ifnet_byindex(gsr.gsr_interface);
2119 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2120 __func__, sopt->sopt_name);
2121 return (EOPNOTSUPP);
2125 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2126 return (EADDRNOTAVAIL);
2128 imo = inp_findmoptions(inp);
2129 idx = imo_match_group(imo, ifp, &gsa->sa);
2133 inm = imo->imo_membership[idx];
2134 imf = &imo->imo_mfilters[idx];
2135 if (ssa->ss.ss_family != AF_UNSPEC) {
2137 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2138 * is an error. On an existing inclusive membership,
2139 * it just adds the source to the filter list.
2141 if (imf->imf_st[1] != MCAST_INCLUDE) {
2143 goto out_inp_locked;
2146 * Throw out duplicates.
2148 * XXX FIXME: This makes a naive assumption that
2149 * even if entries exist for *ssa in this imf,
2150 * they will be rejected as dupes, even if they
2151 * are not valid in the current mode (in-mode).
2153 * in_msource is transactioned just as for anything
2154 * else in SSM -- but note naive use of inm_graft()
2155 * below for allocating new filter entries.
2157 * This is only an issue if someone mixes the
2158 * full-state SSM API with the delta-based API,
2159 * which is discouraged in the relevant RFCs.
2161 lims = imo_match_source(imo, idx, &ssa->sa);
2162 if (lims != NULL /*&&
2163 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2164 error = EADDRNOTAVAIL;
2165 goto out_inp_locked;
2169 * MCAST_JOIN_GROUP on an existing exclusive
2170 * membership is an error; return EADDRINUSE
2171 * to preserve 4.4BSD API idempotence, and
2172 * avoid tedious detour to code below.
2173 * NOTE: This is bending RFC 3678 a bit.
2175 * On an existing inclusive membership, this is also
2176 * an error; if you want to change filter mode,
2177 * you must use the userland API setsourcefilter().
2178 * XXX We don't reject this for imf in UNDEFINED
2179 * state at t1, because allocation of a filter
2180 * is atomic with allocation of a membership.
2183 if (imf->imf_st[1] == MCAST_EXCLUDE)
2185 goto out_inp_locked;
2190 * Begin state merge transaction at socket layer.
2192 INP_WLOCK_ASSERT(inp);
2195 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2196 error = imo_grow(imo);
2198 goto out_inp_locked;
2201 * Allocate the new slot upfront so we can deal with
2202 * grafting the new source filter in same code path
2203 * as for join-source on existing membership.
2205 idx = imo->imo_num_memberships;
2206 imo->imo_membership[idx] = NULL;
2207 imo->imo_num_memberships++;
2208 KASSERT(imo->imo_mfilters != NULL,
2209 ("%s: imf_mfilters vector was not allocated", __func__));
2210 imf = &imo->imo_mfilters[idx];
2211 KASSERT(RB_EMPTY(&imf->imf_sources),
2212 ("%s: imf_sources not empty", __func__));
2216 * Graft new source into filter list for this inpcb's
2217 * membership of the group. The in_multi may not have
2218 * been allocated yet if this is a new membership, however,
2219 * the in_mfilter slot will be allocated and must be initialized.
2221 * Note: Grafting of exclusive mode filters doesn't happen
2223 * XXX: Should check for non-NULL lims (node exists but may
2224 * not be in-mode) for interop with full-state API.
2226 if (ssa->ss.ss_family != AF_UNSPEC) {
2227 /* Membership starts in IN mode */
2229 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2230 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2232 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2234 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2236 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2242 /* No address specified; Membership starts in EX mode */
2244 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2245 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2250 * Begin state merge transaction at IGMP layer.
2257 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2260 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2262 IN_MULTI_LIST_UNLOCK();
2265 imo->imo_membership[idx] = inm;
2267 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2268 IN_MULTI_LIST_LOCK();
2269 error = inm_merge(inm, imf);
2271 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2273 IN_MULTI_LIST_UNLOCK();
2274 goto out_in_multi_locked;
2276 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2277 error = igmp_change_state(inm);
2278 IN_MULTI_LIST_UNLOCK();
2280 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2282 goto out_in_multi_locked;
2286 out_in_multi_locked:
2290 if (in_pcbrele_wlocked(inp))
2303 if (error && is_new) {
2304 imo->imo_membership[idx] = NULL;
2305 --imo->imo_num_memberships;
2314 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2317 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2319 struct group_source_req gsr;
2320 struct ip_mreq_source mreqs;
2321 sockunion_t *gsa, *ssa;
2323 struct in_mfilter *imf;
2324 struct ip_moptions *imo;
2325 struct in_msource *ims;
2326 struct in_multi *inm;
2328 int error, is_final;
2334 memset(&gsr, 0, sizeof(struct group_source_req));
2335 gsa = (sockunion_t *)&gsr.gsr_group;
2336 gsa->ss.ss_family = AF_UNSPEC;
2337 ssa = (sockunion_t *)&gsr.gsr_source;
2338 ssa->ss.ss_family = AF_UNSPEC;
2340 switch (sopt->sopt_name) {
2341 case IP_DROP_MEMBERSHIP:
2342 case IP_DROP_SOURCE_MEMBERSHIP:
2343 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2344 error = sooptcopyin(sopt, &mreqs,
2345 sizeof(struct ip_mreq),
2346 sizeof(struct ip_mreq));
2348 * Swap interface and sourceaddr arguments,
2349 * as ip_mreq and ip_mreq_source are laid
2352 mreqs.imr_interface = mreqs.imr_sourceaddr;
2353 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2354 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2355 error = sooptcopyin(sopt, &mreqs,
2356 sizeof(struct ip_mreq_source),
2357 sizeof(struct ip_mreq_source));
2362 gsa->sin.sin_family = AF_INET;
2363 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2364 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2366 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2367 ssa->sin.sin_family = AF_INET;
2368 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2369 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2373 * Attempt to look up hinted ifp from interface address.
2374 * Fallthrough with null ifp iff lookup fails, to
2375 * preserve 4.4BSD mcast API idempotence.
2376 * XXX NOTE WELL: The RFC 3678 API is preferred because
2377 * using an IPv4 address as a key is racy.
2379 if (!in_nullhost(mreqs.imr_interface))
2380 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2382 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2383 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2387 case MCAST_LEAVE_GROUP:
2388 case MCAST_LEAVE_SOURCE_GROUP:
2389 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2390 error = sooptcopyin(sopt, &gsr,
2391 sizeof(struct group_req),
2392 sizeof(struct group_req));
2393 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2394 error = sooptcopyin(sopt, &gsr,
2395 sizeof(struct group_source_req),
2396 sizeof(struct group_source_req));
2401 if (gsa->sin.sin_family != AF_INET ||
2402 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2405 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2406 if (ssa->sin.sin_family != AF_INET ||
2407 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2411 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2412 return (EADDRNOTAVAIL);
2414 ifp = ifnet_byindex(gsr.gsr_interface);
2417 return (EADDRNOTAVAIL);
2421 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2422 __func__, sopt->sopt_name);
2423 return (EOPNOTSUPP);
2427 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2431 * Find the membership in the membership array.
2433 imo = inp_findmoptions(inp);
2434 idx = imo_match_group(imo, ifp, &gsa->sa);
2436 error = EADDRNOTAVAIL;
2437 goto out_inp_locked;
2439 inm = imo->imo_membership[idx];
2440 imf = &imo->imo_mfilters[idx];
2442 if (ssa->ss.ss_family != AF_UNSPEC)
2446 * Begin state merge transaction at socket layer.
2448 INP_WLOCK_ASSERT(inp);
2451 * If we were instructed only to leave a given source, do so.
2452 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2457 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2458 error = EADDRNOTAVAIL;
2459 goto out_inp_locked;
2461 ims = imo_match_source(imo, idx, &ssa->sa);
2463 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2464 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2465 error = EADDRNOTAVAIL;
2466 goto out_inp_locked;
2468 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2469 error = imf_prune(imf, &ssa->sin);
2471 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2473 goto out_inp_locked;
2478 * Begin state merge transaction at IGMP layer.
2486 * Give up the multicast address record to which
2487 * the membership points.
2489 (void)in_leavegroup_locked(inm, imf);
2491 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2492 IN_MULTI_LIST_LOCK();
2493 error = inm_merge(inm, imf);
2495 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2497 goto out_in_multi_locked;
2500 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2501 error = igmp_change_state(inm);
2502 IN_MULTI_LIST_UNLOCK();
2504 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2509 out_in_multi_locked:
2513 if (in_pcbrele_wlocked(inp))
2524 /* Remove the gap in the membership and filter array. */
2525 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2526 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2527 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2529 imo->imo_num_memberships--;
2538 * Select the interface for transmitting IPv4 multicast datagrams.
2540 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2541 * may be passed to this socket option. An address of INADDR_ANY or an
2542 * interface index of 0 is used to remove a previous selection.
2543 * When no interface is selected, one is chosen for every send.
2546 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2548 struct in_addr addr;
2549 struct ip_mreqn mreqn;
2551 struct ip_moptions *imo;
2554 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2556 * An interface index was specified using the
2557 * Linux-derived ip_mreqn structure.
2559 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2560 sizeof(struct ip_mreqn));
2564 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2567 if (mreqn.imr_ifindex == 0) {
2570 ifp = ifnet_byindex(mreqn.imr_ifindex);
2572 return (EADDRNOTAVAIL);
2576 * An interface was specified by IPv4 address.
2577 * This is the traditional BSD usage.
2579 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2580 sizeof(struct in_addr));
2583 if (in_nullhost(addr)) {
2586 INADDR_TO_IFP(addr, ifp);
2588 return (EADDRNOTAVAIL);
2590 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2591 ntohl(addr.s_addr));
2594 /* Reject interfaces which do not support multicast. */
2595 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2596 return (EOPNOTSUPP);
2598 imo = inp_findmoptions(inp);
2599 imo->imo_multicast_ifp = ifp;
2600 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2607 * Atomically set source filters on a socket for an IPv4 multicast group.
2609 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2612 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2614 struct __msfilterreq msfr;
2617 struct in_mfilter *imf;
2618 struct ip_moptions *imo;
2619 struct in_multi *inm;
2623 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2624 sizeof(struct __msfilterreq));
2628 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2631 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2632 msfr.msfr_fmode != MCAST_INCLUDE))
2635 if (msfr.msfr_group.ss_family != AF_INET ||
2636 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2639 gsa = (sockunion_t *)&msfr.msfr_group;
2640 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2643 gsa->sin.sin_port = 0; /* ignore port */
2645 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2646 return (EADDRNOTAVAIL);
2648 ifp = ifnet_byindex(msfr.msfr_ifindex);
2650 return (EADDRNOTAVAIL);
2653 * Take the INP write lock.
2654 * Check if this socket is a member of this group.
2656 imo = inp_findmoptions(inp);
2657 idx = imo_match_group(imo, ifp, &gsa->sa);
2658 if (idx == -1 || imo->imo_mfilters == NULL) {
2659 error = EADDRNOTAVAIL;
2660 goto out_inp_locked;
2662 inm = imo->imo_membership[idx];
2663 imf = &imo->imo_mfilters[idx];
2666 * Begin state merge transaction at socket layer.
2668 INP_WLOCK_ASSERT(inp);
2670 imf->imf_st[1] = msfr.msfr_fmode;
2673 * Apply any new source filters, if present.
2674 * Make a copy of the user-space source vector so
2675 * that we may copy them with a single copyin. This
2676 * allows us to deal with page faults up-front.
2678 if (msfr.msfr_nsrcs > 0) {
2679 struct in_msource *lims;
2680 struct sockaddr_in *psin;
2681 struct sockaddr_storage *kss, *pkss;
2686 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2687 __func__, (unsigned long)msfr.msfr_nsrcs);
2688 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2690 error = copyin(msfr.msfr_srcs, kss,
2691 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2700 * Mark all source filters as UNDEFINED at t1.
2701 * Restore new group filter mode, as imf_leave()
2702 * will set it to INCLUDE.
2705 imf->imf_st[1] = msfr.msfr_fmode;
2708 * Update socket layer filters at t1, lazy-allocating
2709 * new entries. This saves a bunch of memory at the
2710 * cost of one RB_FIND() per source entry; duplicate
2711 * entries in the msfr_nsrcs vector are ignored.
2712 * If we encounter an error, rollback transaction.
2714 * XXX This too could be replaced with a set-symmetric
2715 * difference like loop to avoid walking from root
2716 * every time, as the key space is common.
2718 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2719 psin = (struct sockaddr_in *)pkss;
2720 if (psin->sin_family != AF_INET) {
2721 error = EAFNOSUPPORT;
2724 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2728 error = imf_get_source(imf, psin, &lims);
2731 lims->imsl_st[1] = imf->imf_st[1];
2737 goto out_imf_rollback;
2739 INP_WLOCK_ASSERT(inp);
2741 IN_MULTI_LIST_LOCK();
2744 * Begin state merge transaction at IGMP layer.
2746 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2747 error = inm_merge(inm, imf);
2749 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2750 IN_MULTI_LIST_UNLOCK();
2751 goto out_in_multi_locked;
2754 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2755 error = igmp_change_state(inm);
2756 IN_MULTI_LIST_UNLOCK();
2758 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2760 out_in_multi_locked:
2778 * Set the IP multicast options in response to user setsockopt().
2780 * Many of the socket options handled in this function duplicate the
2781 * functionality of socket options in the regular unicast API. However,
2782 * it is not possible to merge the duplicate code, because the idempotence
2783 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2784 * the effects of these options must be treated as separate and distinct.
2786 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2787 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2788 * is refactored to no longer use vifs.
2791 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2793 struct ip_moptions *imo;
2799 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2800 * or is a divert socket, reject it.
2802 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2803 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2804 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2805 return (EOPNOTSUPP);
2807 switch (sopt->sopt_name) {
2808 case IP_MULTICAST_VIF: {
2811 * Select a multicast VIF for transmission.
2812 * Only useful if multicast forwarding is active.
2814 if (legal_vif_num == NULL) {
2818 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2821 if (!legal_vif_num(vifi) && (vifi != -1)) {
2825 imo = inp_findmoptions(inp);
2826 imo->imo_multicast_vif = vifi;
2831 case IP_MULTICAST_IF:
2832 error = inp_set_multicast_if(inp, sopt);
2835 case IP_MULTICAST_TTL: {
2839 * Set the IP time-to-live for outgoing multicast packets.
2840 * The original multicast API required a char argument,
2841 * which is inconsistent with the rest of the socket API.
2842 * We allow either a char or an int.
2844 if (sopt->sopt_valsize == sizeof(u_char)) {
2845 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2852 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2862 imo = inp_findmoptions(inp);
2863 imo->imo_multicast_ttl = ttl;
2868 case IP_MULTICAST_LOOP: {
2872 * Set the loopback flag for outgoing multicast packets.
2873 * Must be zero or one. The original multicast API required a
2874 * char argument, which is inconsistent with the rest
2875 * of the socket API. We allow either a char or an int.
2877 if (sopt->sopt_valsize == sizeof(u_char)) {
2878 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2885 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2889 loop = (u_char)iloop;
2891 imo = inp_findmoptions(inp);
2892 imo->imo_multicast_loop = !!loop;
2897 case IP_ADD_MEMBERSHIP:
2898 case IP_ADD_SOURCE_MEMBERSHIP:
2899 case MCAST_JOIN_GROUP:
2900 case MCAST_JOIN_SOURCE_GROUP:
2901 error = inp_join_group(inp, sopt);
2904 case IP_DROP_MEMBERSHIP:
2905 case IP_DROP_SOURCE_MEMBERSHIP:
2906 case MCAST_LEAVE_GROUP:
2907 case MCAST_LEAVE_SOURCE_GROUP:
2908 error = inp_leave_group(inp, sopt);
2911 case IP_BLOCK_SOURCE:
2912 case IP_UNBLOCK_SOURCE:
2913 case MCAST_BLOCK_SOURCE:
2914 case MCAST_UNBLOCK_SOURCE:
2915 error = inp_block_unblock_source(inp, sopt);
2919 error = inp_set_source_filters(inp, sopt);
2927 INP_UNLOCK_ASSERT(inp);
2933 * Expose IGMP's multicast filter mode and source list(s) to userland,
2934 * keyed by (ifindex, group).
2935 * The filter mode is written out as a uint32_t, followed by
2936 * 0..n of struct in_addr.
2937 * For use by ifmcstat(8).
2938 * SMPng: NOTE: unlocked read of ifindex space.
2941 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2943 struct in_addr src, group;
2945 struct ifmultiaddr *ifma;
2946 struct in_multi *inm;
2947 struct ip_msource *ims;
2951 uint32_t fmode, ifindex;
2956 if (req->newptr != NULL)
2963 if (ifindex <= 0 || ifindex > V_if_index) {
2964 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2969 group.s_addr = name[1];
2970 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2971 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2972 __func__, ntohl(group.s_addr));
2976 ifp = ifnet_byindex(ifindex);
2978 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2983 retval = sysctl_wire_old_buffer(req,
2984 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2988 IN_MULTI_LIST_LOCK();
2991 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2992 if (ifma->ifma_addr->sa_family != AF_INET ||
2993 ifma->ifma_protospec == NULL)
2995 inm = (struct in_multi *)ifma->ifma_protospec;
2996 if (!in_hosteq(inm->inm_addr, group))
2998 fmode = inm->inm_st[1].iss_fmode;
2999 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
3002 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
3003 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
3006 * Only copy-out sources which are in-mode.
3008 if (fmode != ims_get_mode(inm, ims, 1)) {
3009 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
3013 src.s_addr = htonl(ims->ims_haddr);
3014 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
3019 IF_ADDR_RUNLOCK(ifp);
3021 IN_MULTI_LIST_UNLOCK();
3026 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
3028 static const char *inm_modestrs[] = { "un", "in", "ex" };
3031 inm_mode_str(const int mode)
3034 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
3035 return (inm_modestrs[mode]);
3039 static const char *inm_statestrs[] = {
3052 inm_state_str(const int state)
3055 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3056 return (inm_statestrs[state]);
3061 * Dump an in_multi structure to the console.
3064 inm_print(const struct in_multi *inm)
3067 char addrbuf[INET_ADDRSTRLEN];
3069 if ((ktr_mask & KTR_IGMPV3) == 0)
3072 printf("%s: --- begin inm %p ---\n", __func__, inm);
3073 printf("addr %s ifp %p(%s) ifma %p\n",
3074 inet_ntoa_r(inm->inm_addr, addrbuf),
3076 inm->inm_ifp->if_xname,
3078 printf("timer %u state %s refcount %u scq.len %u\n",
3080 inm_state_str(inm->inm_state),
3082 inm->inm_scq.mq_len);
3083 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3088 for (t = 0; t < 2; t++) {
3089 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3090 inm_mode_str(inm->inm_st[t].iss_fmode),
3091 inm->inm_st[t].iss_asm,
3092 inm->inm_st[t].iss_ex,
3093 inm->inm_st[t].iss_in,
3094 inm->inm_st[t].iss_rec);
3096 printf("%s: --- end inm %p ---\n", __func__, inm);
3099 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3102 inm_print(const struct in_multi *inm)
3107 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3109 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);