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");
98 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK,
99 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
100 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
101 * it can be taken by code in net/if.c also.
102 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
104 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
105 * any need for in_multi itself to be virtualized -- it is bound to an ifp
106 * anyway no matter what happens.
108 struct mtx in_multi_list_mtx;
109 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
111 struct mtx in_multi_free_mtx;
112 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
114 struct sx in_multi_sx;
115 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
120 * Functions with non-static linkage defined in this file should be
121 * declared in in_var.h:
126 * in_joingroup_locked()
128 * in_leavegroup_locked()
134 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
137 static void imf_commit(struct in_mfilter *);
138 static int imf_get_source(struct in_mfilter *imf,
139 const struct sockaddr_in *psin,
140 struct in_msource **);
141 static struct in_msource *
142 imf_graft(struct in_mfilter *, const uint8_t,
143 const struct sockaddr_in *);
144 static void imf_leave(struct in_mfilter *);
145 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
146 static void imf_purge(struct in_mfilter *);
147 static void imf_rollback(struct in_mfilter *);
148 static void imf_reap(struct in_mfilter *);
149 static struct in_mfilter *
150 imo_match_group(const struct ip_moptions *,
151 const struct ifnet *, const struct sockaddr *);
152 static struct in_msource *
153 imo_match_source(struct in_mfilter *, const struct sockaddr *);
154 static void ims_merge(struct ip_msource *ims,
155 const struct in_msource *lims, const int rollback);
156 static int in_getmulti(struct ifnet *, const struct in_addr *,
158 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
159 const int noalloc, struct ip_msource **pims);
161 static int inm_is_ifp_detached(const struct in_multi *);
163 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
164 static void inm_purge(struct in_multi *);
165 static void inm_reap(struct in_multi *);
166 static void inm_release(struct in_multi *);
167 static struct ip_moptions *
168 inp_findmoptions(struct inpcb *);
169 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
170 static int inp_join_group(struct inpcb *, struct sockopt *);
171 static int inp_leave_group(struct inpcb *, struct sockopt *);
172 static struct ifnet *
173 inp_lookup_mcast_ifp(const struct inpcb *,
174 const struct sockaddr_in *, const struct in_addr);
175 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
176 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
177 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
178 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
180 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
183 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
184 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
185 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
186 "Max source filters per group");
188 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
189 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
190 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
191 "Max source filters per socket");
193 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
194 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
195 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
197 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
198 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
199 "Per-interface stack-wide source filters");
203 * Inline function which wraps assertions for a valid ifp.
204 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
208 inm_is_ifp_detached(const struct in_multi *inm)
212 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
213 ifp = inm->inm_ifma->ifma_ifp;
216 * Sanity check that netinet's notion of ifp is the
219 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
222 return (ifp == NULL);
226 static struct grouptask free_gtask;
227 static struct in_multi_head inm_free_list;
228 static void inm_release_task(void *arg __unused);
229 static void inm_init(void)
231 SLIST_INIT(&inm_free_list);
232 taskqgroup_config_gtask_init(NULL, &free_gtask, inm_release_task, "inm release task");
235 #ifdef EARLY_AP_STARTUP
236 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
239 SYSINIT(inm_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_FIRST,
245 inm_release_list_deferred(struct in_multi_head *inmh)
248 if (SLIST_EMPTY(inmh))
250 mtx_lock(&in_multi_free_mtx);
251 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
252 mtx_unlock(&in_multi_free_mtx);
253 GROUPTASK_ENQUEUE(&free_gtask);
257 inm_disconnect(struct in_multi *inm)
260 struct ifmultiaddr *ifma, *ll_ifma;
263 IF_ADDR_WLOCK_ASSERT(ifp);
264 ifma = inm->inm_ifma;
267 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
268 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
269 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
271 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
272 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
273 MPASS(ifma != ll_ifma);
274 ifma->ifma_llifma = NULL;
275 MPASS(ll_ifma->ifma_llifma == NULL);
276 MPASS(ll_ifma->ifma_ifp == ifp);
277 if (--ll_ifma->ifma_refcount == 0) {
278 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
279 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
280 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
282 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
283 if_freemulti(ll_ifma);
290 inm_release_deferred(struct in_multi *inm)
292 struct in_multi_head tmp;
294 IN_MULTI_LIST_LOCK_ASSERT();
295 MPASS(inm->inm_refcount > 0);
296 if (--inm->inm_refcount == 0) {
299 inm->inm_ifma->ifma_protospec = NULL;
300 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
301 inm_release_list_deferred(&tmp);
306 inm_release_task(void *arg __unused)
308 struct in_multi_head inm_free_tmp;
309 struct in_multi *inm, *tinm;
311 SLIST_INIT(&inm_free_tmp);
312 mtx_lock(&in_multi_free_mtx);
313 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
314 mtx_unlock(&in_multi_free_mtx);
316 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
317 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
325 * Initialize an in_mfilter structure to a known state at t0, t1
326 * with an empty source filter list.
329 imf_init(struct in_mfilter *imf, const int st0, const int st1)
331 memset(imf, 0, sizeof(struct in_mfilter));
332 RB_INIT(&imf->imf_sources);
333 imf->imf_st[0] = st0;
334 imf->imf_st[1] = st1;
338 ip_mfilter_alloc(const int mflags, const int st0, const int st1)
340 struct in_mfilter *imf;
342 imf = malloc(sizeof(*imf), M_INMFILTER, mflags);
344 imf_init(imf, st0, st1);
350 ip_mfilter_free(struct in_mfilter *imf)
354 free(imf, M_INMFILTER);
358 * Function for looking up an in_multi record for an IPv4 multicast address
359 * on a given interface. ifp must be valid. If no record found, return NULL.
360 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
363 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
365 struct ifmultiaddr *ifma;
366 struct in_multi *inm;
368 IN_MULTI_LIST_LOCK_ASSERT();
369 IF_ADDR_LOCK_ASSERT(ifp);
372 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
373 if (ifma->ifma_addr->sa_family != AF_INET ||
374 ifma->ifma_protospec == NULL)
376 inm = (struct in_multi *)ifma->ifma_protospec;
377 if (inm->inm_addr.s_addr == ina.s_addr)
385 * Wrapper for inm_lookup_locked().
386 * The IF_ADDR_LOCK will be taken on ifp and released on return.
389 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
391 struct epoch_tracker et;
392 struct in_multi *inm;
394 IN_MULTI_LIST_LOCK_ASSERT();
396 inm = inm_lookup_locked(ifp, ina);
403 * Find an IPv4 multicast group entry for this ip_moptions instance
404 * which matches the specified group, and optionally an interface.
405 * Return its index into the array, or -1 if not found.
407 static struct in_mfilter *
408 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
409 const struct sockaddr *group)
411 const struct sockaddr_in *gsin;
412 struct in_mfilter *imf;
413 struct in_multi *inm;
415 gsin = (const struct sockaddr_in *)group;
417 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
421 if ((ifp == NULL || (inm->inm_ifp == ifp)) &&
422 in_hosteq(inm->inm_addr, gsin->sin_addr)) {
430 * Find an IPv4 multicast source entry for this imo which matches
431 * the given group index for this socket, and source address.
433 * NOTE: This does not check if the entry is in-mode, merely if
434 * it exists, which may not be the desired behaviour.
436 static struct in_msource *
437 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src)
439 struct ip_msource find;
440 struct ip_msource *ims;
441 const sockunion_t *psa;
443 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
445 /* Source trees are keyed in host byte order. */
446 psa = (const sockunion_t *)src;
447 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
448 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
450 return ((struct in_msource *)ims);
454 * Perform filtering for multicast datagrams on a socket by group and source.
456 * Returns 0 if a datagram should be allowed through, or various error codes
457 * if the socket was not a member of the group, or the source was muted, etc.
460 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
461 const struct sockaddr *group, const struct sockaddr *src)
463 struct in_mfilter *imf;
464 struct in_msource *ims;
467 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
469 imf = imo_match_group(imo, ifp, group);
471 return (MCAST_NOTGMEMBER);
474 * Check if the source was included in an (S,G) join.
475 * Allow reception on exclusive memberships by default,
476 * reject reception on inclusive memberships by default.
477 * Exclude source only if an in-mode exclude filter exists.
478 * Include source only if an in-mode include filter exists.
479 * NOTE: We are comparing group state here at IGMP t1 (now)
480 * with socket-layer t0 (since last downcall).
482 mode = imf->imf_st[1];
483 ims = imo_match_source(imf, src);
485 if ((ims == NULL && mode == MCAST_INCLUDE) ||
486 (ims != NULL && ims->imsl_st[0] != mode))
487 return (MCAST_NOTSMEMBER);
493 * Find and return a reference to an in_multi record for (ifp, group),
494 * and bump its reference count.
495 * If one does not exist, try to allocate it, and update link-layer multicast
496 * filters on ifp to listen for group.
497 * Assumes the IN_MULTI lock is held across the call.
498 * Return 0 if successful, otherwise return an appropriate error code.
501 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
502 struct in_multi **pinm)
504 struct sockaddr_in gsin;
505 struct ifmultiaddr *ifma;
506 struct in_ifinfo *ii;
507 struct in_multi *inm;
510 IN_MULTI_LOCK_ASSERT();
512 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
513 IN_MULTI_LIST_LOCK();
514 inm = inm_lookup(ifp, *group);
517 * If we already joined this group, just bump the
518 * refcount and return it.
520 KASSERT(inm->inm_refcount >= 1,
521 ("%s: bad refcount %d", __func__, inm->inm_refcount));
522 inm_acquire_locked(inm);
525 IN_MULTI_LIST_UNLOCK();
529 memset(&gsin, 0, sizeof(gsin));
530 gsin.sin_family = AF_INET;
531 gsin.sin_len = sizeof(struct sockaddr_in);
532 gsin.sin_addr = *group;
535 * Check if a link-layer group is already associated
536 * with this network-layer group on the given ifnet.
538 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
542 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
543 IN_MULTI_LIST_LOCK();
547 * If something other than netinet is occupying the link-layer
548 * group, print a meaningful error message and back out of
550 * Otherwise, bump the refcount on the existing network-layer
551 * group association and return it.
553 if (ifma->ifma_protospec != NULL) {
554 inm = (struct in_multi *)ifma->ifma_protospec;
556 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
558 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
559 ("%s: ifma not AF_INET", __func__));
560 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
561 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
562 !in_hosteq(inm->inm_addr, *group)) {
563 char addrbuf[INET_ADDRSTRLEN];
565 panic("%s: ifma %p is inconsistent with %p (%s)",
566 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
569 inm_acquire_locked(inm);
574 IF_ADDR_WLOCK_ASSERT(ifp);
577 * A new in_multi record is needed; allocate and initialize it.
578 * We DO NOT perform an IGMP join as the in_ layer may need to
579 * push an initial source list down to IGMP to support SSM.
581 * The initial source filter state is INCLUDE, {} as per the RFC.
583 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
585 IF_ADDR_WUNLOCK(ifp);
586 IN_MULTI_LIST_UNLOCK();
587 if_delmulti_ifma(ifma);
590 inm->inm_addr = *group;
592 inm->inm_igi = ii->ii_igmp;
593 inm->inm_ifma = ifma;
594 inm->inm_refcount = 1;
595 inm->inm_state = IGMP_NOT_MEMBER;
596 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
597 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
598 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
599 RB_INIT(&inm->inm_srcs);
601 ifma->ifma_protospec = inm;
605 IF_ADDR_WUNLOCK(ifp);
606 IN_MULTI_LIST_UNLOCK();
611 * Drop a reference to an in_multi record.
613 * If the refcount drops to 0, free the in_multi record and
614 * delete the underlying link-layer membership.
617 inm_release(struct in_multi *inm)
619 struct ifmultiaddr *ifma;
622 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
623 MPASS(inm->inm_refcount == 0);
624 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
626 ifma = inm->inm_ifma;
629 /* XXX this access is not covered by IF_ADDR_LOCK */
630 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
632 CURVNET_SET(ifp->if_vnet);
634 free(inm, M_IPMADDR);
635 if_delmulti_ifma_flags(ifma, 1);
640 free(inm, M_IPMADDR);
641 if_delmulti_ifma_flags(ifma, 1);
646 * Clear recorded source entries for a group.
647 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
648 * FIXME: Should reap.
651 inm_clear_recorded(struct in_multi *inm)
653 struct ip_msource *ims;
655 IN_MULTI_LIST_LOCK_ASSERT();
657 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
660 --inm->inm_st[1].iss_rec;
663 KASSERT(inm->inm_st[1].iss_rec == 0,
664 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
668 * Record a source as pending for a Source-Group IGMPv3 query.
669 * This lives here as it modifies the shared tree.
671 * inm is the group descriptor.
672 * naddr is the address of the source to record in network-byte order.
674 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
675 * lazy-allocate a source node in response to an SG query.
676 * Otherwise, no allocation is performed. This saves some memory
677 * with the trade-off that the source will not be reported to the
678 * router if joined in the window between the query response and
679 * the group actually being joined on the local host.
681 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
682 * This turns off the allocation of a recorded source entry if
683 * the group has not been joined.
685 * Return 0 if the source didn't exist or was already marked as recorded.
686 * Return 1 if the source was marked as recorded by this function.
687 * Return <0 if any error occurred (negated errno code).
690 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
692 struct ip_msource find;
693 struct ip_msource *ims, *nims;
695 IN_MULTI_LIST_LOCK_ASSERT();
697 find.ims_haddr = ntohl(naddr);
698 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
699 if (ims && ims->ims_stp)
702 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
704 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
708 nims->ims_haddr = find.ims_haddr;
709 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
715 * Mark the source as recorded and update the recorded
719 ++inm->inm_st[1].iss_rec;
725 * Return a pointer to an in_msource owned by an in_mfilter,
726 * given its source address.
727 * Lazy-allocate if needed. If this is a new entry its filter state is
730 * imf is the filter set being modified.
731 * haddr is the source address in *host* byte-order.
733 * SMPng: May be called with locks held; malloc must not block.
736 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
737 struct in_msource **plims)
739 struct ip_msource find;
740 struct ip_msource *ims, *nims;
741 struct in_msource *lims;
748 /* key is host byte order */
749 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
750 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
751 lims = (struct in_msource *)ims;
753 if (imf->imf_nsrc == in_mcast_maxsocksrc)
755 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
759 lims = (struct in_msource *)nims;
760 lims->ims_haddr = find.ims_haddr;
761 lims->imsl_st[0] = MCAST_UNDEFINED;
762 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
772 * Graft a source entry into an existing socket-layer filter set,
773 * maintaining any required invariants and checking allocations.
775 * The source is marked as being in the new filter mode at t1.
777 * Return the pointer to the new node, otherwise return NULL.
779 static struct in_msource *
780 imf_graft(struct in_mfilter *imf, const uint8_t st1,
781 const struct sockaddr_in *psin)
783 struct ip_msource *nims;
784 struct in_msource *lims;
786 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
790 lims = (struct in_msource *)nims;
791 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
792 lims->imsl_st[0] = MCAST_UNDEFINED;
793 lims->imsl_st[1] = st1;
794 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
801 * Prune a source entry from an existing socket-layer filter set,
802 * maintaining any required invariants and checking allocations.
804 * The source is marked as being left at t1, it is not freed.
806 * Return 0 if no error occurred, otherwise return an errno value.
809 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
811 struct ip_msource find;
812 struct ip_msource *ims;
813 struct in_msource *lims;
815 /* key is host byte order */
816 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
817 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
820 lims = (struct in_msource *)ims;
821 lims->imsl_st[1] = MCAST_UNDEFINED;
826 * Revert socket-layer filter set deltas at t1 to t0 state.
829 imf_rollback(struct in_mfilter *imf)
831 struct ip_msource *ims, *tims;
832 struct in_msource *lims;
834 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
835 lims = (struct in_msource *)ims;
836 if (lims->imsl_st[0] == lims->imsl_st[1]) {
837 /* no change at t1 */
839 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
840 /* revert change to existing source at t1 */
841 lims->imsl_st[1] = lims->imsl_st[0];
843 /* revert source added t1 */
844 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
845 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
846 free(ims, M_INMFILTER);
850 imf->imf_st[1] = imf->imf_st[0];
854 * Mark socket-layer filter set as INCLUDE {} at t1.
857 imf_leave(struct in_mfilter *imf)
859 struct ip_msource *ims;
860 struct in_msource *lims;
862 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
863 lims = (struct in_msource *)ims;
864 lims->imsl_st[1] = MCAST_UNDEFINED;
866 imf->imf_st[1] = MCAST_INCLUDE;
870 * Mark socket-layer filter set deltas as committed.
873 imf_commit(struct in_mfilter *imf)
875 struct ip_msource *ims;
876 struct in_msource *lims;
878 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
879 lims = (struct in_msource *)ims;
880 lims->imsl_st[0] = lims->imsl_st[1];
882 imf->imf_st[0] = imf->imf_st[1];
886 * Reap unreferenced sources from socket-layer filter set.
889 imf_reap(struct in_mfilter *imf)
891 struct ip_msource *ims, *tims;
892 struct in_msource *lims;
894 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
895 lims = (struct in_msource *)ims;
896 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
897 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
898 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
899 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
900 free(ims, M_INMFILTER);
907 * Purge socket-layer filter set.
910 imf_purge(struct in_mfilter *imf)
912 struct ip_msource *ims, *tims;
914 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
915 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
916 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
917 free(ims, M_INMFILTER);
920 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
921 KASSERT(RB_EMPTY(&imf->imf_sources),
922 ("%s: imf_sources not empty", __func__));
926 * Look up a source filter entry for a multicast group.
928 * inm is the group descriptor to work with.
929 * haddr is the host-byte-order IPv4 address to look up.
930 * noalloc may be non-zero to suppress allocation of sources.
931 * *pims will be set to the address of the retrieved or allocated source.
933 * SMPng: NOTE: may be called with locks held.
934 * Return 0 if successful, otherwise return a non-zero error code.
937 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
938 const int noalloc, struct ip_msource **pims)
940 struct ip_msource find;
941 struct ip_msource *ims, *nims;
943 find.ims_haddr = haddr;
944 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
945 if (ims == NULL && !noalloc) {
946 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
948 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
952 nims->ims_haddr = haddr;
953 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
957 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
967 * Merge socket-layer source into IGMP-layer source.
968 * If rollback is non-zero, perform the inverse of the merge.
971 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
974 int n = rollback ? -1 : 1;
976 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
977 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
978 __func__, n, ims->ims_haddr);
979 ims->ims_st[1].ex -= n;
980 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
981 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
982 __func__, n, ims->ims_haddr);
983 ims->ims_st[1].in -= n;
986 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
987 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
988 __func__, n, ims->ims_haddr);
989 ims->ims_st[1].ex += n;
990 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
991 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
992 __func__, n, ims->ims_haddr);
993 ims->ims_st[1].in += n;
998 * Atomically update the global in_multi state, when a membership's
999 * filter list is being updated in any way.
1001 * imf is the per-inpcb-membership group filter pointer.
1002 * A fake imf may be passed for in-kernel consumers.
1004 * XXX This is a candidate for a set-symmetric-difference style loop
1005 * which would eliminate the repeated lookup from root of ims nodes,
1006 * as they share the same key space.
1008 * If any error occurred this function will back out of refcounts
1009 * and return a non-zero value.
1012 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1014 struct ip_msource *ims, *nims;
1015 struct in_msource *lims;
1016 int schanged, error;
1022 IN_MULTI_LIST_LOCK_ASSERT();
1025 * Update the source filters first, as this may fail.
1026 * Maintain count of in-mode filters at t0, t1. These are
1027 * used to work out if we transition into ASM mode or not.
1028 * Maintain a count of source filters whose state was
1029 * actually modified by this operation.
1031 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1032 lims = (struct in_msource *)ims;
1033 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1034 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1035 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1036 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1040 ims_merge(nims, lims, 0);
1043 struct ip_msource *bims;
1045 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1046 lims = (struct in_msource *)ims;
1047 if (lims->imsl_st[0] == lims->imsl_st[1])
1049 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1052 ims_merge(bims, lims, 1);
1057 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1058 __func__, nsrc0, nsrc1);
1060 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1061 if (imf->imf_st[0] == imf->imf_st[1] &&
1062 imf->imf_st[1] == MCAST_INCLUDE) {
1064 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1065 --inm->inm_st[1].iss_in;
1069 /* Handle filter mode transition on socket. */
1070 if (imf->imf_st[0] != imf->imf_st[1]) {
1071 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1072 __func__, imf->imf_st[0], imf->imf_st[1]);
1074 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1075 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1076 --inm->inm_st[1].iss_ex;
1077 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1078 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1079 --inm->inm_st[1].iss_in;
1082 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1083 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1084 inm->inm_st[1].iss_ex++;
1085 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1086 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1087 inm->inm_st[1].iss_in++;
1092 * Track inm filter state in terms of listener counts.
1093 * If there are any exclusive listeners, stack-wide
1094 * membership is exclusive.
1095 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1096 * If no listeners remain, state is undefined at t1,
1097 * and the IGMP lifecycle for this group should finish.
1099 if (inm->inm_st[1].iss_ex > 0) {
1100 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1101 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1102 } else if (inm->inm_st[1].iss_in > 0) {
1103 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1104 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1106 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1107 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1110 /* Decrement ASM listener count on transition out of ASM mode. */
1111 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1112 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1113 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1114 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1115 --inm->inm_st[1].iss_asm;
1119 /* Increment ASM listener count on transition to ASM mode. */
1120 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1121 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1122 inm->inm_st[1].iss_asm++;
1125 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1130 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1137 * Mark an in_multi's filter set deltas as committed.
1138 * Called by IGMP after a state change has been enqueued.
1141 inm_commit(struct in_multi *inm)
1143 struct ip_msource *ims;
1145 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1146 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1149 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1150 ims->ims_st[0] = ims->ims_st[1];
1152 inm->inm_st[0] = inm->inm_st[1];
1156 * Reap unreferenced nodes from an in_multi's filter set.
1159 inm_reap(struct in_multi *inm)
1161 struct ip_msource *ims, *tims;
1163 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1164 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1165 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1168 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1169 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1170 free(ims, M_IPMSOURCE);
1176 * Purge all source nodes from an in_multi's filter set.
1179 inm_purge(struct in_multi *inm)
1181 struct ip_msource *ims, *tims;
1183 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1184 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1185 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1186 free(ims, M_IPMSOURCE);
1192 * Join a multicast group; unlocked entry point.
1194 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1195 * is not held. Fortunately, ifp is unlikely to have been detached
1196 * at this point, so we assume it's OK to recurse.
1199 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1200 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1205 error = in_joingroup_locked(ifp, gina, imf, pinm);
1212 * Join a multicast group; real entry point.
1214 * Only preserves atomicity at inm level.
1215 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1217 * If the IGMP downcall fails, the group is not joined, and an error
1221 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1222 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1224 struct in_mfilter timf;
1225 struct in_multi *inm;
1228 IN_MULTI_LOCK_ASSERT();
1229 IN_MULTI_LIST_UNLOCK_ASSERT();
1231 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1232 ntohl(gina->s_addr), ifp, ifp->if_xname);
1238 * If no imf was specified (i.e. kernel consumer),
1239 * fake one up and assume it is an ASM join.
1242 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1246 error = in_getmulti(ifp, gina, &inm);
1248 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1251 IN_MULTI_LIST_LOCK();
1252 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1253 error = inm_merge(inm, imf);
1255 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1256 goto out_inm_release;
1259 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1260 error = igmp_change_state(inm);
1262 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1263 goto out_inm_release;
1269 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1270 inm_release_deferred(inm);
1274 IN_MULTI_LIST_UNLOCK();
1280 * Leave a multicast group; unlocked entry point.
1283 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1288 error = in_leavegroup_locked(inm, imf);
1295 * Leave a multicast group; real entry point.
1296 * All source filters will be expunged.
1298 * Only preserves atomicity at inm level.
1300 * Holding the write lock for the INP which contains imf
1301 * is highly advisable. We can't assert for it as imf does not
1302 * contain a back-pointer to the owning inp.
1304 * Note: This is not the same as inm_release(*) as this function also
1305 * makes a state change downcall into IGMP.
1308 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1310 struct in_mfilter timf;
1315 IN_MULTI_LOCK_ASSERT();
1316 IN_MULTI_LIST_UNLOCK_ASSERT();
1318 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1319 inm, ntohl(inm->inm_addr.s_addr),
1320 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1324 * If no imf was specified (i.e. kernel consumer),
1325 * fake one up and assume it is an ASM join.
1328 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1333 * Begin state merge transaction at IGMP layer.
1335 * As this particular invocation should not cause any memory
1336 * to be allocated, and there is no opportunity to roll back
1337 * the transaction, it MUST NOT fail.
1339 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1340 IN_MULTI_LIST_LOCK();
1341 error = inm_merge(inm, imf);
1342 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1344 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1345 CURVNET_SET(inm->inm_ifp->if_vnet);
1346 error = igmp_change_state(inm);
1347 IF_ADDR_WLOCK(inm->inm_ifp);
1348 inm_release_deferred(inm);
1349 IF_ADDR_WUNLOCK(inm->inm_ifp);
1350 IN_MULTI_LIST_UNLOCK();
1353 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1355 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1360 /*#ifndef BURN_BRIDGES*/
1362 * Join an IPv4 multicast group in (*,G) exclusive mode.
1363 * The group must be a 224.0.0.0/24 link-scope group.
1364 * This KPI is for legacy kernel consumers only.
1367 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1369 struct in_multi *pinm;
1372 char addrbuf[INET_ADDRSTRLEN];
1375 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1376 ("%s: %s not in 224.0.0.0/24", __func__,
1377 inet_ntoa_r(*ap, addrbuf)));
1379 error = in_joingroup(ifp, ap, NULL, &pinm);
1387 * Block or unblock an ASM multicast source on an inpcb.
1388 * This implements the delta-based API described in RFC 3678.
1390 * The delta-based API applies only to exclusive-mode memberships.
1391 * An IGMP downcall will be performed.
1393 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1395 * Return 0 if successful, otherwise return an appropriate error code.
1398 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1400 struct group_source_req gsr;
1401 struct rm_priotracker in_ifa_tracker;
1402 sockunion_t *gsa, *ssa;
1404 struct in_mfilter *imf;
1405 struct ip_moptions *imo;
1406 struct in_msource *ims;
1407 struct in_multi *inm;
1415 memset(&gsr, 0, sizeof(struct group_source_req));
1416 gsa = (sockunion_t *)&gsr.gsr_group;
1417 ssa = (sockunion_t *)&gsr.gsr_source;
1419 switch (sopt->sopt_name) {
1420 case IP_BLOCK_SOURCE:
1421 case IP_UNBLOCK_SOURCE: {
1422 struct ip_mreq_source mreqs;
1424 error = sooptcopyin(sopt, &mreqs,
1425 sizeof(struct ip_mreq_source),
1426 sizeof(struct ip_mreq_source));
1430 gsa->sin.sin_family = AF_INET;
1431 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1432 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1434 ssa->sin.sin_family = AF_INET;
1435 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1436 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1438 if (!in_nullhost(mreqs.imr_interface)) {
1439 IN_IFADDR_RLOCK(&in_ifa_tracker);
1440 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1441 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1443 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1446 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1447 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1451 case MCAST_BLOCK_SOURCE:
1452 case MCAST_UNBLOCK_SOURCE:
1453 error = sooptcopyin(sopt, &gsr,
1454 sizeof(struct group_source_req),
1455 sizeof(struct group_source_req));
1459 if (gsa->sin.sin_family != AF_INET ||
1460 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1463 if (ssa->sin.sin_family != AF_INET ||
1464 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1467 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1468 return (EADDRNOTAVAIL);
1470 ifp = ifnet_byindex(gsr.gsr_interface);
1472 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1477 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1478 __func__, sopt->sopt_name);
1479 return (EOPNOTSUPP);
1483 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1489 * Check if we are actually a member of this group.
1491 imo = inp_findmoptions(inp);
1492 imf = imo_match_group(imo, ifp, &gsa->sa);
1494 error = EADDRNOTAVAIL;
1495 goto out_inp_locked;
1500 * Attempting to use the delta-based API on an
1501 * non exclusive-mode membership is an error.
1503 fmode = imf->imf_st[0];
1504 if (fmode != MCAST_EXCLUDE) {
1506 goto out_inp_locked;
1510 * Deal with error cases up-front:
1511 * Asked to block, but already blocked; or
1512 * Asked to unblock, but nothing to unblock.
1513 * If adding a new block entry, allocate it.
1515 ims = imo_match_source(imf, &ssa->sa);
1516 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1517 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1518 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1519 error = EADDRNOTAVAIL;
1520 goto out_inp_locked;
1523 INP_WLOCK_ASSERT(inp);
1526 * Begin state merge transaction at socket layer.
1529 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1530 ims = imf_graft(imf, fmode, &ssa->sin);
1534 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1535 error = imf_prune(imf, &ssa->sin);
1539 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1540 goto out_imf_rollback;
1544 * Begin state merge transaction at IGMP layer.
1546 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1547 IN_MULTI_LIST_LOCK();
1548 error = inm_merge(inm, imf);
1550 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1551 IN_MULTI_LIST_UNLOCK();
1552 goto out_imf_rollback;
1555 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1556 error = igmp_change_state(inm);
1557 IN_MULTI_LIST_UNLOCK();
1559 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1576 * Given an inpcb, return its multicast options structure pointer. Accepts
1577 * an unlocked inpcb pointer, but will return it locked. May sleep.
1579 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1580 * SMPng: NOTE: Returns with the INP write lock held.
1582 static struct ip_moptions *
1583 inp_findmoptions(struct inpcb *inp)
1585 struct ip_moptions *imo;
1588 if (inp->inp_moptions != NULL)
1589 return (inp->inp_moptions);
1593 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1595 imo->imo_multicast_ifp = NULL;
1596 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1597 imo->imo_multicast_vif = -1;
1598 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1599 imo->imo_multicast_loop = in_mcast_loop;
1600 STAILQ_INIT(&imo->imo_head);
1603 if (inp->inp_moptions != NULL) {
1604 free(imo, M_IPMOPTS);
1605 return (inp->inp_moptions);
1607 inp->inp_moptions = imo;
1612 inp_gcmoptions(struct ip_moptions *imo)
1614 struct in_mfilter *imf;
1615 struct in_multi *inm;
1618 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1619 ip_mfilter_remove(&imo->imo_head, imf);
1622 if ((inm = imf->imf_inm) != NULL) {
1623 if ((ifp = inm->inm_ifp) != NULL) {
1624 CURVNET_SET(ifp->if_vnet);
1625 (void)in_leavegroup(inm, imf);
1628 (void)in_leavegroup(inm, imf);
1631 ip_mfilter_free(imf);
1633 free(imo, M_IPMOPTS);
1637 * Discard the IP multicast options (and source filters). To minimize
1638 * the amount of work done while holding locks such as the INP's
1639 * pcbinfo lock (which is used in the receive path), the free
1640 * operation is deferred to the epoch callback task.
1643 inp_freemoptions(struct ip_moptions *imo)
1647 inp_gcmoptions(imo);
1651 * Atomically get source filters on a socket for an IPv4 multicast group.
1652 * Called with INP lock held; returns with lock released.
1655 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1657 struct __msfilterreq msfr;
1660 struct ip_moptions *imo;
1661 struct in_mfilter *imf;
1662 struct ip_msource *ims;
1663 struct in_msource *lims;
1664 struct sockaddr_in *psin;
1665 struct sockaddr_storage *ptss;
1666 struct sockaddr_storage *tss;
1668 size_t nsrcs, ncsrcs;
1670 INP_WLOCK_ASSERT(inp);
1672 imo = inp->inp_moptions;
1673 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1677 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1678 sizeof(struct __msfilterreq));
1682 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1685 ifp = ifnet_byindex(msfr.msfr_ifindex);
1692 * Lookup group on the socket.
1694 gsa = (sockunion_t *)&msfr.msfr_group;
1695 imf = imo_match_group(imo, ifp, &gsa->sa);
1698 return (EADDRNOTAVAIL);
1702 * Ignore memberships which are in limbo.
1704 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1708 msfr.msfr_fmode = imf->imf_st[1];
1711 * If the user specified a buffer, copy out the source filter
1712 * entries to userland gracefully.
1713 * We only copy out the number of entries which userland
1714 * has asked for, but we always tell userland how big the
1715 * buffer really needs to be.
1717 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1718 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1720 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1721 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1722 M_TEMP, M_NOWAIT | M_ZERO);
1730 * Count number of sources in-mode at t0.
1731 * If buffer space exists and remains, copy out source entries.
1733 nsrcs = msfr.msfr_nsrcs;
1736 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1737 lims = (struct in_msource *)ims;
1738 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1739 lims->imsl_st[0] != imf->imf_st[0])
1742 if (tss != NULL && nsrcs > 0) {
1743 psin = (struct sockaddr_in *)ptss;
1744 psin->sin_family = AF_INET;
1745 psin->sin_len = sizeof(struct sockaddr_in);
1746 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1756 error = copyout(tss, msfr.msfr_srcs,
1757 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1763 msfr.msfr_nsrcs = ncsrcs;
1764 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1770 * Return the IP multicast options in response to user getsockopt().
1773 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1775 struct rm_priotracker in_ifa_tracker;
1776 struct ip_mreqn mreqn;
1777 struct ip_moptions *imo;
1779 struct in_ifaddr *ia;
1784 imo = inp->inp_moptions;
1786 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1787 * or is a divert socket, reject it.
1789 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1790 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1791 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1793 return (EOPNOTSUPP);
1797 switch (sopt->sopt_name) {
1798 case IP_MULTICAST_VIF:
1800 optval = imo->imo_multicast_vif;
1804 error = sooptcopyout(sopt, &optval, sizeof(int));
1807 case IP_MULTICAST_IF:
1808 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1810 ifp = imo->imo_multicast_ifp;
1811 if (!in_nullhost(imo->imo_multicast_addr)) {
1812 mreqn.imr_address = imo->imo_multicast_addr;
1813 } else if (ifp != NULL) {
1814 struct epoch_tracker et;
1816 mreqn.imr_ifindex = ifp->if_index;
1817 NET_EPOCH_ENTER(et);
1818 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1821 IA_SIN(ia)->sin_addr;
1826 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1827 error = sooptcopyout(sopt, &mreqn,
1828 sizeof(struct ip_mreqn));
1830 error = sooptcopyout(sopt, &mreqn.imr_address,
1831 sizeof(struct in_addr));
1835 case IP_MULTICAST_TTL:
1837 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1839 optval = coptval = imo->imo_multicast_ttl;
1841 if (sopt->sopt_valsize == sizeof(u_char))
1842 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1844 error = sooptcopyout(sopt, &optval, sizeof(int));
1847 case IP_MULTICAST_LOOP:
1849 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1851 optval = coptval = imo->imo_multicast_loop;
1853 if (sopt->sopt_valsize == sizeof(u_char))
1854 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1856 error = sooptcopyout(sopt, &optval, sizeof(int));
1861 error = EADDRNOTAVAIL;
1864 error = inp_get_source_filters(inp, sopt);
1870 error = ENOPROTOOPT;
1874 INP_UNLOCK_ASSERT(inp);
1880 * Look up the ifnet to use for a multicast group membership,
1881 * given the IPv4 address of an interface, and the IPv4 group address.
1883 * This routine exists to support legacy multicast applications
1884 * which do not understand that multicast memberships are scoped to
1885 * specific physical links in the networking stack, or which need
1886 * to join link-scope groups before IPv4 addresses are configured.
1888 * If inp is non-NULL, use this socket's current FIB number for any
1889 * required FIB lookup.
1890 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1891 * and use its ifp; usually, this points to the default next-hop.
1893 * If the FIB lookup fails, attempt to use the first non-loopback
1894 * interface with multicast capability in the system as a
1895 * last resort. The legacy IPv4 ASM API requires that we do
1896 * this in order to allow groups to be joined when the routing
1897 * table has not yet been populated during boot.
1899 * Returns NULL if no ifp could be found.
1901 * FUTURE: Implement IPv4 source-address selection.
1903 static struct ifnet *
1904 inp_lookup_mcast_ifp(const struct inpcb *inp,
1905 const struct sockaddr_in *gsin, const struct in_addr ina)
1907 struct rm_priotracker in_ifa_tracker;
1909 struct nhop4_basic nh4;
1912 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1913 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1914 ("%s: not multicast", __func__));
1917 if (!in_nullhost(ina)) {
1918 IN_IFADDR_RLOCK(&in_ifa_tracker);
1919 INADDR_TO_IFP(ina, ifp);
1920 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1922 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1923 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1926 struct in_ifaddr *ia;
1930 IN_IFADDR_RLOCK(&in_ifa_tracker);
1931 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1933 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1934 (mifp->if_flags & IFF_MULTICAST)) {
1939 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1947 * Join an IPv4 multicast group, possibly with a source.
1950 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1952 struct group_source_req gsr;
1953 sockunion_t *gsa, *ssa;
1955 struct in_mfilter *imf;
1956 struct ip_moptions *imo;
1957 struct in_multi *inm;
1958 struct in_msource *lims;
1965 memset(&gsr, 0, sizeof(struct group_source_req));
1966 gsa = (sockunion_t *)&gsr.gsr_group;
1967 gsa->ss.ss_family = AF_UNSPEC;
1968 ssa = (sockunion_t *)&gsr.gsr_source;
1969 ssa->ss.ss_family = AF_UNSPEC;
1971 switch (sopt->sopt_name) {
1972 case IP_ADD_MEMBERSHIP: {
1973 struct ip_mreqn mreqn;
1975 if (sopt->sopt_valsize == sizeof(struct ip_mreqn))
1976 error = sooptcopyin(sopt, &mreqn,
1977 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn));
1979 error = sooptcopyin(sopt, &mreqn,
1980 sizeof(struct ip_mreq), sizeof(struct ip_mreq));
1984 gsa->sin.sin_family = AF_INET;
1985 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1986 gsa->sin.sin_addr = mreqn.imr_multiaddr;
1987 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1990 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) &&
1991 mreqn.imr_ifindex != 0)
1992 ifp = ifnet_byindex(mreqn.imr_ifindex);
1994 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1998 case IP_ADD_SOURCE_MEMBERSHIP: {
1999 struct ip_mreq_source mreqs;
2001 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source),
2002 sizeof(struct ip_mreq_source));
2006 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET;
2007 gsa->sin.sin_len = ssa->sin.sin_len =
2008 sizeof(struct sockaddr_in);
2010 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2011 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2014 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2016 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2017 mreqs.imr_interface);
2018 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2019 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2023 case MCAST_JOIN_GROUP:
2024 case MCAST_JOIN_SOURCE_GROUP:
2025 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2026 error = sooptcopyin(sopt, &gsr,
2027 sizeof(struct group_req),
2028 sizeof(struct group_req));
2029 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2030 error = sooptcopyin(sopt, &gsr,
2031 sizeof(struct group_source_req),
2032 sizeof(struct group_source_req));
2037 if (gsa->sin.sin_family != AF_INET ||
2038 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2042 * Overwrite the port field if present, as the sockaddr
2043 * being copied in may be matched with a binary comparison.
2045 gsa->sin.sin_port = 0;
2046 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2047 if (ssa->sin.sin_family != AF_INET ||
2048 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2050 ssa->sin.sin_port = 0;
2053 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2056 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2057 return (EADDRNOTAVAIL);
2058 ifp = ifnet_byindex(gsr.gsr_interface);
2062 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2063 __func__, sopt->sopt_name);
2064 return (EOPNOTSUPP);
2068 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2069 return (EADDRNOTAVAIL);
2074 * Find the membership in the membership list.
2076 imo = inp_findmoptions(inp);
2077 imf = imo_match_group(imo, ifp, &gsa->sa);
2082 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) {
2084 goto out_inp_locked;
2090 if (ssa->ss.ss_family != AF_UNSPEC) {
2092 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2093 * is an error. On an existing inclusive membership,
2094 * it just adds the source to the filter list.
2096 if (imf->imf_st[1] != MCAST_INCLUDE) {
2098 goto out_inp_locked;
2101 * Throw out duplicates.
2103 * XXX FIXME: This makes a naive assumption that
2104 * even if entries exist for *ssa in this imf,
2105 * they will be rejected as dupes, even if they
2106 * are not valid in the current mode (in-mode).
2108 * in_msource is transactioned just as for anything
2109 * else in SSM -- but note naive use of inm_graft()
2110 * below for allocating new filter entries.
2112 * This is only an issue if someone mixes the
2113 * full-state SSM API with the delta-based API,
2114 * which is discouraged in the relevant RFCs.
2116 lims = imo_match_source(imf, &ssa->sa);
2117 if (lims != NULL /*&&
2118 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2119 error = EADDRNOTAVAIL;
2120 goto out_inp_locked;
2124 * MCAST_JOIN_GROUP on an existing exclusive
2125 * membership is an error; return EADDRINUSE
2126 * to preserve 4.4BSD API idempotence, and
2127 * avoid tedious detour to code below.
2128 * NOTE: This is bending RFC 3678 a bit.
2130 * On an existing inclusive membership, this is also
2131 * an error; if you want to change filter mode,
2132 * you must use the userland API setsourcefilter().
2133 * XXX We don't reject this for imf in UNDEFINED
2134 * state at t1, because allocation of a filter
2135 * is atomic with allocation of a membership.
2138 if (imf->imf_st[1] == MCAST_EXCLUDE)
2140 goto out_inp_locked;
2145 * Begin state merge transaction at socket layer.
2147 INP_WLOCK_ASSERT(inp);
2150 * Graft new source into filter list for this inpcb's
2151 * membership of the group. The in_multi may not have
2152 * been allocated yet if this is a new membership, however,
2153 * the in_mfilter slot will be allocated and must be initialized.
2155 * Note: Grafting of exclusive mode filters doesn't happen
2157 * XXX: Should check for non-NULL lims (node exists but may
2158 * not be in-mode) for interop with full-state API.
2160 if (ssa->ss.ss_family != AF_UNSPEC) {
2161 /* Membership starts in IN mode */
2163 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2164 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2167 goto out_inp_locked;
2170 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2172 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2174 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2177 goto out_inp_locked;
2180 /* No address specified; Membership starts in EX mode */
2182 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2183 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2186 goto out_inp_locked;
2192 * Begin state merge transaction at IGMP layer.
2198 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2202 if (in_pcbrele_wlocked(inp)) {
2204 goto out_inp_unlocked;
2207 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2209 goto out_inp_locked;
2211 inm_acquire(imf->imf_inm);
2213 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2214 IN_MULTI_LIST_LOCK();
2215 error = inm_merge(inm, imf);
2217 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2219 IN_MULTI_LIST_UNLOCK();
2222 goto out_inp_locked;
2224 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2225 error = igmp_change_state(inm);
2226 IN_MULTI_LIST_UNLOCK();
2228 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2232 goto out_inp_locked;
2236 ip_mfilter_insert(&imo->imo_head, imf);
2246 if (is_new && imf) {
2247 if (imf->imf_inm != NULL) {
2248 IN_MULTI_LIST_LOCK();
2249 inm_release_deferred(imf->imf_inm);
2250 IN_MULTI_LIST_UNLOCK();
2252 ip_mfilter_free(imf);
2258 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2261 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2263 struct group_source_req gsr;
2264 struct ip_mreq_source mreqs;
2265 struct rm_priotracker in_ifa_tracker;
2266 sockunion_t *gsa, *ssa;
2268 struct in_mfilter *imf;
2269 struct ip_moptions *imo;
2270 struct in_msource *ims;
2271 struct in_multi *inm;
2279 memset(&gsr, 0, sizeof(struct group_source_req));
2280 gsa = (sockunion_t *)&gsr.gsr_group;
2281 gsa->ss.ss_family = AF_UNSPEC;
2282 ssa = (sockunion_t *)&gsr.gsr_source;
2283 ssa->ss.ss_family = AF_UNSPEC;
2285 switch (sopt->sopt_name) {
2286 case IP_DROP_MEMBERSHIP:
2287 case IP_DROP_SOURCE_MEMBERSHIP:
2288 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2289 error = sooptcopyin(sopt, &mreqs,
2290 sizeof(struct ip_mreq),
2291 sizeof(struct ip_mreq));
2293 * Swap interface and sourceaddr arguments,
2294 * as ip_mreq and ip_mreq_source are laid
2297 mreqs.imr_interface = mreqs.imr_sourceaddr;
2298 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2299 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2300 error = sooptcopyin(sopt, &mreqs,
2301 sizeof(struct ip_mreq_source),
2302 sizeof(struct ip_mreq_source));
2307 gsa->sin.sin_family = AF_INET;
2308 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2309 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2311 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2312 ssa->sin.sin_family = AF_INET;
2313 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2314 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2318 * Attempt to look up hinted ifp from interface address.
2319 * Fallthrough with null ifp iff lookup fails, to
2320 * preserve 4.4BSD mcast API idempotence.
2321 * XXX NOTE WELL: The RFC 3678 API is preferred because
2322 * using an IPv4 address as a key is racy.
2324 if (!in_nullhost(mreqs.imr_interface)) {
2325 IN_IFADDR_RLOCK(&in_ifa_tracker);
2326 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2327 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2329 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2330 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2334 case MCAST_LEAVE_GROUP:
2335 case MCAST_LEAVE_SOURCE_GROUP:
2336 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2337 error = sooptcopyin(sopt, &gsr,
2338 sizeof(struct group_req),
2339 sizeof(struct group_req));
2340 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2341 error = sooptcopyin(sopt, &gsr,
2342 sizeof(struct group_source_req),
2343 sizeof(struct group_source_req));
2348 if (gsa->sin.sin_family != AF_INET ||
2349 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2352 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2353 if (ssa->sin.sin_family != AF_INET ||
2354 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2358 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2359 return (EADDRNOTAVAIL);
2361 ifp = ifnet_byindex(gsr.gsr_interface);
2364 return (EADDRNOTAVAIL);
2368 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2369 __func__, sopt->sopt_name);
2370 return (EOPNOTSUPP);
2374 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2380 * Find the membership in the membership list.
2382 imo = inp_findmoptions(inp);
2383 imf = imo_match_group(imo, ifp, &gsa->sa);
2385 error = EADDRNOTAVAIL;
2386 goto out_inp_locked;
2390 if (ssa->ss.ss_family != AF_UNSPEC)
2394 * Begin state merge transaction at socket layer.
2396 INP_WLOCK_ASSERT(inp);
2399 * If we were instructed only to leave a given source, do so.
2400 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2403 ip_mfilter_remove(&imo->imo_head, imf);
2406 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2407 error = EADDRNOTAVAIL;
2408 goto out_inp_locked;
2410 ims = imo_match_source(imf, &ssa->sa);
2412 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2413 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2414 error = EADDRNOTAVAIL;
2415 goto out_inp_locked;
2417 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2418 error = imf_prune(imf, &ssa->sin);
2420 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2422 goto out_inp_locked;
2427 * Begin state merge transaction at IGMP layer.
2430 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2431 IN_MULTI_LIST_LOCK();
2432 error = inm_merge(inm, imf);
2434 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2436 IN_MULTI_LIST_UNLOCK();
2439 goto out_inp_locked;
2442 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2443 error = igmp_change_state(inm);
2444 IN_MULTI_LIST_UNLOCK();
2446 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2450 goto out_inp_locked;
2459 if (is_final && imf) {
2461 * Give up the multicast address record to which
2462 * the membership points.
2464 (void) in_leavegroup_locked(imf->imf_inm, imf);
2465 ip_mfilter_free(imf);
2473 * Select the interface for transmitting IPv4 multicast datagrams.
2475 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2476 * may be passed to this socket option. An address of INADDR_ANY or an
2477 * interface index of 0 is used to remove a previous selection.
2478 * When no interface is selected, one is chosen for every send.
2481 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2483 struct rm_priotracker in_ifa_tracker;
2484 struct in_addr addr;
2485 struct ip_mreqn mreqn;
2487 struct ip_moptions *imo;
2490 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2492 * An interface index was specified using the
2493 * Linux-derived ip_mreqn structure.
2495 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2496 sizeof(struct ip_mreqn));
2500 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2503 if (mreqn.imr_ifindex == 0) {
2506 ifp = ifnet_byindex(mreqn.imr_ifindex);
2508 return (EADDRNOTAVAIL);
2512 * An interface was specified by IPv4 address.
2513 * This is the traditional BSD usage.
2515 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2516 sizeof(struct in_addr));
2519 if (in_nullhost(addr)) {
2522 IN_IFADDR_RLOCK(&in_ifa_tracker);
2523 INADDR_TO_IFP(addr, ifp);
2524 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2526 return (EADDRNOTAVAIL);
2528 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2529 ntohl(addr.s_addr));
2532 /* Reject interfaces which do not support multicast. */
2533 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2534 return (EOPNOTSUPP);
2536 imo = inp_findmoptions(inp);
2537 imo->imo_multicast_ifp = ifp;
2538 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2545 * Atomically set source filters on a socket for an IPv4 multicast group.
2547 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2550 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2552 struct __msfilterreq msfr;
2555 struct in_mfilter *imf;
2556 struct ip_moptions *imo;
2557 struct in_multi *inm;
2560 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2561 sizeof(struct __msfilterreq));
2565 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2568 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2569 msfr.msfr_fmode != MCAST_INCLUDE))
2572 if (msfr.msfr_group.ss_family != AF_INET ||
2573 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2576 gsa = (sockunion_t *)&msfr.msfr_group;
2577 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2580 gsa->sin.sin_port = 0; /* ignore port */
2582 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2583 return (EADDRNOTAVAIL);
2585 ifp = ifnet_byindex(msfr.msfr_ifindex);
2587 return (EADDRNOTAVAIL);
2592 * Take the INP write lock.
2593 * Check if this socket is a member of this group.
2595 imo = inp_findmoptions(inp);
2596 imf = imo_match_group(imo, ifp, &gsa->sa);
2598 error = EADDRNOTAVAIL;
2599 goto out_inp_locked;
2604 * Begin state merge transaction at socket layer.
2606 INP_WLOCK_ASSERT(inp);
2608 imf->imf_st[1] = msfr.msfr_fmode;
2611 * Apply any new source filters, if present.
2612 * Make a copy of the user-space source vector so
2613 * that we may copy them with a single copyin. This
2614 * allows us to deal with page faults up-front.
2616 if (msfr.msfr_nsrcs > 0) {
2617 struct in_msource *lims;
2618 struct sockaddr_in *psin;
2619 struct sockaddr_storage *kss, *pkss;
2624 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2625 __func__, (unsigned long)msfr.msfr_nsrcs);
2626 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2628 error = copyin(msfr.msfr_srcs, kss,
2629 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2638 * Mark all source filters as UNDEFINED at t1.
2639 * Restore new group filter mode, as imf_leave()
2640 * will set it to INCLUDE.
2643 imf->imf_st[1] = msfr.msfr_fmode;
2646 * Update socket layer filters at t1, lazy-allocating
2647 * new entries. This saves a bunch of memory at the
2648 * cost of one RB_FIND() per source entry; duplicate
2649 * entries in the msfr_nsrcs vector are ignored.
2650 * If we encounter an error, rollback transaction.
2652 * XXX This too could be replaced with a set-symmetric
2653 * difference like loop to avoid walking from root
2654 * every time, as the key space is common.
2656 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2657 psin = (struct sockaddr_in *)pkss;
2658 if (psin->sin_family != AF_INET) {
2659 error = EAFNOSUPPORT;
2662 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2666 error = imf_get_source(imf, psin, &lims);
2669 lims->imsl_st[1] = imf->imf_st[1];
2675 goto out_imf_rollback;
2677 INP_WLOCK_ASSERT(inp);
2680 * Begin state merge transaction at IGMP layer.
2682 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2683 IN_MULTI_LIST_LOCK();
2684 error = inm_merge(inm, imf);
2686 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2687 IN_MULTI_LIST_UNLOCK();
2688 goto out_imf_rollback;
2691 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2692 error = igmp_change_state(inm);
2693 IN_MULTI_LIST_UNLOCK();
2695 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2712 * Set the IP multicast options in response to user setsockopt().
2714 * Many of the socket options handled in this function duplicate the
2715 * functionality of socket options in the regular unicast API. However,
2716 * it is not possible to merge the duplicate code, because the idempotence
2717 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2718 * the effects of these options must be treated as separate and distinct.
2720 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2721 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2722 * is refactored to no longer use vifs.
2725 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2727 struct ip_moptions *imo;
2733 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2734 * or is a divert socket, reject it.
2736 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2737 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2738 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2739 return (EOPNOTSUPP);
2741 switch (sopt->sopt_name) {
2742 case IP_MULTICAST_VIF: {
2745 * Select a multicast VIF for transmission.
2746 * Only useful if multicast forwarding is active.
2748 if (legal_vif_num == NULL) {
2752 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2755 if (!legal_vif_num(vifi) && (vifi != -1)) {
2759 imo = inp_findmoptions(inp);
2760 imo->imo_multicast_vif = vifi;
2765 case IP_MULTICAST_IF:
2766 error = inp_set_multicast_if(inp, sopt);
2769 case IP_MULTICAST_TTL: {
2773 * Set the IP time-to-live for outgoing multicast packets.
2774 * The original multicast API required a char argument,
2775 * which is inconsistent with the rest of the socket API.
2776 * We allow either a char or an int.
2778 if (sopt->sopt_valsize == sizeof(u_char)) {
2779 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2786 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2796 imo = inp_findmoptions(inp);
2797 imo->imo_multicast_ttl = ttl;
2802 case IP_MULTICAST_LOOP: {
2806 * Set the loopback flag for outgoing multicast packets.
2807 * Must be zero or one. The original multicast API required a
2808 * char argument, which is inconsistent with the rest
2809 * of the socket API. We allow either a char or an int.
2811 if (sopt->sopt_valsize == sizeof(u_char)) {
2812 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2819 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2823 loop = (u_char)iloop;
2825 imo = inp_findmoptions(inp);
2826 imo->imo_multicast_loop = !!loop;
2831 case IP_ADD_MEMBERSHIP:
2832 case IP_ADD_SOURCE_MEMBERSHIP:
2833 case MCAST_JOIN_GROUP:
2834 case MCAST_JOIN_SOURCE_GROUP:
2835 error = inp_join_group(inp, sopt);
2838 case IP_DROP_MEMBERSHIP:
2839 case IP_DROP_SOURCE_MEMBERSHIP:
2840 case MCAST_LEAVE_GROUP:
2841 case MCAST_LEAVE_SOURCE_GROUP:
2842 error = inp_leave_group(inp, sopt);
2845 case IP_BLOCK_SOURCE:
2846 case IP_UNBLOCK_SOURCE:
2847 case MCAST_BLOCK_SOURCE:
2848 case MCAST_UNBLOCK_SOURCE:
2849 error = inp_block_unblock_source(inp, sopt);
2853 error = inp_set_source_filters(inp, sopt);
2861 INP_UNLOCK_ASSERT(inp);
2867 * Expose IGMP's multicast filter mode and source list(s) to userland,
2868 * keyed by (ifindex, group).
2869 * The filter mode is written out as a uint32_t, followed by
2870 * 0..n of struct in_addr.
2871 * For use by ifmcstat(8).
2872 * SMPng: NOTE: unlocked read of ifindex space.
2875 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2877 struct in_addr src, group;
2878 struct epoch_tracker et;
2880 struct ifmultiaddr *ifma;
2881 struct in_multi *inm;
2882 struct ip_msource *ims;
2886 uint32_t fmode, ifindex;
2891 if (req->newptr != NULL)
2898 if (ifindex <= 0 || ifindex > V_if_index) {
2899 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2904 group.s_addr = name[1];
2905 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2906 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2907 __func__, ntohl(group.s_addr));
2911 ifp = ifnet_byindex(ifindex);
2913 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2918 retval = sysctl_wire_old_buffer(req,
2919 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2923 IN_MULTI_LIST_LOCK();
2925 NET_EPOCH_ENTER(et);
2926 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2927 if (ifma->ifma_addr->sa_family != AF_INET ||
2928 ifma->ifma_protospec == NULL)
2930 inm = (struct in_multi *)ifma->ifma_protospec;
2931 if (!in_hosteq(inm->inm_addr, group))
2933 fmode = inm->inm_st[1].iss_fmode;
2934 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2937 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2938 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2941 * Only copy-out sources which are in-mode.
2943 if (fmode != ims_get_mode(inm, ims, 1)) {
2944 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2948 src.s_addr = htonl(ims->ims_haddr);
2949 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2956 IN_MULTI_LIST_UNLOCK();
2961 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2963 static const char *inm_modestrs[] = {
2964 [MCAST_UNDEFINED] = "un",
2965 [MCAST_INCLUDE] = "in",
2966 [MCAST_EXCLUDE] = "ex",
2968 _Static_assert(MCAST_UNDEFINED == 0 &&
2969 MCAST_EXCLUDE + 1 == nitems(inm_modestrs),
2970 "inm_modestrs: no longer matches #defines");
2973 inm_mode_str(const int mode)
2976 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2977 return (inm_modestrs[mode]);
2981 static const char *inm_statestrs[] = {
2982 [IGMP_NOT_MEMBER] = "not-member",
2983 [IGMP_SILENT_MEMBER] = "silent",
2984 [IGMP_REPORTING_MEMBER] = "reporting",
2985 [IGMP_IDLE_MEMBER] = "idle",
2986 [IGMP_LAZY_MEMBER] = "lazy",
2987 [IGMP_SLEEPING_MEMBER] = "sleeping",
2988 [IGMP_AWAKENING_MEMBER] = "awakening",
2989 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending",
2990 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending",
2991 [IGMP_LEAVING_MEMBER] = "leaving",
2993 _Static_assert(IGMP_NOT_MEMBER == 0 &&
2994 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs),
2995 "inm_statetrs: no longer matches #defines");
2998 inm_state_str(const int state)
3001 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3002 return (inm_statestrs[state]);
3007 * Dump an in_multi structure to the console.
3010 inm_print(const struct in_multi *inm)
3013 char addrbuf[INET_ADDRSTRLEN];
3015 if ((ktr_mask & KTR_IGMPV3) == 0)
3018 printf("%s: --- begin inm %p ---\n", __func__, inm);
3019 printf("addr %s ifp %p(%s) ifma %p\n",
3020 inet_ntoa_r(inm->inm_addr, addrbuf),
3022 inm->inm_ifp->if_xname,
3024 printf("timer %u state %s refcount %u scq.len %u\n",
3026 inm_state_str(inm->inm_state),
3028 inm->inm_scq.mq_len);
3029 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3034 for (t = 0; t < 2; t++) {
3035 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3036 inm_mode_str(inm->inm_st[t].iss_fmode),
3037 inm->inm_st[t].iss_asm,
3038 inm->inm_st[t].iss_ex,
3039 inm->inm_st[t].iss_in,
3040 inm->inm_st[t].iss_rec);
3042 printf("%s: --- end inm %p ---\n", __func__, inm);
3045 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3048 inm_print(const struct in_multi *inm)
3053 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3055 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);