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/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
41 #include <sys/malloc.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/protosw.h>
47 #include <sys/sysctl.h>
49 #include <sys/taskqueue.h>
53 #include <net/if_var.h>
54 #include <net/if_dl.h>
55 #include <net/route.h>
56 #include <net/route/nhop.h>
59 #include <net/ethernet.h>
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_fib.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_var.h>
66 #include <net/if_private.h>
67 #include <netinet/ip_var.h>
68 #include <netinet/igmp_var.h>
71 #define KTR_IGMPV3 KTR_INET
74 #ifndef __SOCKUNION_DECLARED
76 struct sockaddr_storage ss;
78 struct sockaddr_dl sdl;
79 struct sockaddr_in sin;
81 typedef union sockunion sockunion_t;
82 #define __SOCKUNION_DECLARED
83 #endif /* __SOCKUNION_DECLARED */
85 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
86 "IPv4 multicast PCB-layer source filter");
87 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
88 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
89 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
90 "IPv4 multicast IGMP-layer source filter");
95 * - Lock order is: IN_MULTI_LOCK, INP_WLOCK, IN_MULTI_LIST_LOCK, IGMP_LOCK,
97 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
98 * it can be taken by code in net/if.c also.
99 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
101 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
102 * any need for in_multi itself to be virtualized -- it is bound to an ifp
103 * anyway no matter what happens.
105 struct mtx in_multi_list_mtx;
106 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
108 struct mtx in_multi_free_mtx;
109 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
111 struct sx in_multi_sx;
112 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
115 * Functions with non-static linkage defined in this file should be
116 * declared in in_var.h:
119 * in_joingroup_locked()
121 * in_leavegroup_locked()
127 static void imf_commit(struct in_mfilter *);
128 static int imf_get_source(struct in_mfilter *imf,
129 const struct sockaddr_in *psin,
130 struct in_msource **);
131 static struct in_msource *
132 imf_graft(struct in_mfilter *, const uint8_t,
133 const struct sockaddr_in *);
134 static void imf_leave(struct in_mfilter *);
135 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
136 static void imf_purge(struct in_mfilter *);
137 static void imf_rollback(struct in_mfilter *);
138 static void imf_reap(struct in_mfilter *);
139 static struct in_mfilter *
140 imo_match_group(const struct ip_moptions *,
141 const struct ifnet *, const struct sockaddr *);
142 static struct in_msource *
143 imo_match_source(struct in_mfilter *, const struct sockaddr *);
144 static void ims_merge(struct ip_msource *ims,
145 const struct in_msource *lims, const int rollback);
146 static int in_getmulti(struct ifnet *, const struct in_addr *,
148 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
149 const int noalloc, struct ip_msource **pims);
151 static int inm_is_ifp_detached(const struct in_multi *);
153 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
154 static void inm_purge(struct in_multi *);
155 static void inm_reap(struct in_multi *);
156 static void inm_release(struct in_multi *);
157 static struct ip_moptions *
158 inp_findmoptions(struct inpcb *);
159 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
160 static int inp_join_group(struct inpcb *, struct sockopt *);
161 static int inp_leave_group(struct inpcb *, struct sockopt *);
162 static struct ifnet *
163 inp_lookup_mcast_ifp(const struct inpcb *,
164 const struct sockaddr_in *, const struct in_addr);
165 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
166 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
167 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
168 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
170 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast,
171 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
174 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
175 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
176 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
177 "Max source filters per group");
179 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
180 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
181 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
182 "Max source filters per socket");
184 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
185 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
186 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
188 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
189 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
190 "Per-interface stack-wide source filters");
194 * Inline function which wraps assertions for a valid ifp.
195 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
199 inm_is_ifp_detached(const struct in_multi *inm)
203 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
204 ifp = inm->inm_ifma->ifma_ifp;
207 * Sanity check that netinet's notion of ifp is the
210 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
213 return (ifp == NULL);
218 * Interface detach can happen in a taskqueue thread context, so we must use a
219 * dedicated thread to avoid deadlocks when draining inm_release tasks.
221 TASKQUEUE_DEFINE_THREAD(inm_free);
222 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER();
223 static void inm_release_task(void *arg __unused, int pending __unused);
224 static struct task inm_free_task = TASK_INITIALIZER(0, inm_release_task, NULL);
227 inm_release_wait(void *arg __unused)
231 * Make sure all pending multicast addresses are freed before
232 * the VNET or network device is destroyed:
234 taskqueue_drain(taskqueue_inm_free, &inm_free_task);
237 /* XXX-BZ FIXME, see D24914. */
238 VNET_SYSUNINIT(inm_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, inm_release_wait, NULL);
242 inm_release_list_deferred(struct in_multi_head *inmh)
245 if (SLIST_EMPTY(inmh))
247 mtx_lock(&in_multi_free_mtx);
248 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
249 mtx_unlock(&in_multi_free_mtx);
250 taskqueue_enqueue(taskqueue_inm_free, &inm_free_task);
254 inm_disconnect(struct in_multi *inm)
257 struct ifmultiaddr *ifma, *ll_ifma;
260 IF_ADDR_WLOCK_ASSERT(ifp);
261 ifma = inm->inm_ifma;
264 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
265 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
266 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
268 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
269 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
270 MPASS(ifma != ll_ifma);
271 ifma->ifma_llifma = NULL;
272 MPASS(ll_ifma->ifma_llifma == NULL);
273 MPASS(ll_ifma->ifma_ifp == ifp);
274 if (--ll_ifma->ifma_refcount == 0) {
275 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
276 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
277 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
279 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
280 if_freemulti(ll_ifma);
286 inm_release_deferred(struct in_multi *inm)
288 struct in_multi_head tmp;
290 IN_MULTI_LIST_LOCK_ASSERT();
291 MPASS(inm->inm_refcount > 0);
292 if (--inm->inm_refcount == 0) {
295 inm->inm_ifma->ifma_protospec = NULL;
296 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
297 inm_release_list_deferred(&tmp);
302 inm_release_task(void *arg __unused, int pending __unused)
304 struct in_multi_head inm_free_tmp;
305 struct in_multi *inm, *tinm;
307 SLIST_INIT(&inm_free_tmp);
308 mtx_lock(&in_multi_free_mtx);
309 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
310 mtx_unlock(&in_multi_free_mtx);
312 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
313 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
321 * Initialize an in_mfilter structure to a known state at t0, t1
322 * with an empty source filter list.
325 imf_init(struct in_mfilter *imf, const int st0, const int st1)
327 memset(imf, 0, sizeof(struct in_mfilter));
328 RB_INIT(&imf->imf_sources);
329 imf->imf_st[0] = st0;
330 imf->imf_st[1] = st1;
334 ip_mfilter_alloc(const int mflags, const int st0, const int st1)
336 struct in_mfilter *imf;
338 imf = malloc(sizeof(*imf), M_INMFILTER, mflags);
340 imf_init(imf, st0, st1);
346 ip_mfilter_free(struct in_mfilter *imf)
350 free(imf, M_INMFILTER);
354 * Function for looking up an in_multi record for an IPv4 multicast address
355 * on a given interface. ifp must be valid. If no record found, return NULL.
356 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
359 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
361 struct ifmultiaddr *ifma;
362 struct in_multi *inm;
364 IN_MULTI_LIST_LOCK_ASSERT();
365 IF_ADDR_LOCK_ASSERT(ifp);
367 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
368 inm = inm_ifmultiaddr_get_inm(ifma);
371 if (inm->inm_addr.s_addr == ina.s_addr)
378 * Wrapper for inm_lookup_locked().
379 * The IF_ADDR_LOCK will be taken on ifp and released on return.
382 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
384 struct epoch_tracker et;
385 struct in_multi *inm;
387 IN_MULTI_LIST_LOCK_ASSERT();
390 inm = inm_lookup_locked(ifp, ina);
397 * Find an IPv4 multicast group entry for this ip_moptions instance
398 * which matches the specified group, and optionally an interface.
399 * Return its index into the array, or -1 if not found.
401 static struct in_mfilter *
402 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
403 const struct sockaddr *group)
405 const struct sockaddr_in *gsin;
406 struct in_mfilter *imf;
407 struct in_multi *inm;
409 gsin = (const struct sockaddr_in *)group;
411 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
415 if ((ifp == NULL || (inm->inm_ifp == ifp)) &&
416 in_hosteq(inm->inm_addr, gsin->sin_addr)) {
424 * Find an IPv4 multicast source entry for this imo which matches
425 * the given group index for this socket, and source address.
427 * NOTE: This does not check if the entry is in-mode, merely if
428 * it exists, which may not be the desired behaviour.
430 static struct in_msource *
431 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src)
433 struct ip_msource find;
434 struct ip_msource *ims;
435 const sockunion_t *psa;
437 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
439 /* Source trees are keyed in host byte order. */
440 psa = (const sockunion_t *)src;
441 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
442 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
444 return ((struct in_msource *)ims);
448 * Perform filtering for multicast datagrams on a socket by group and source.
450 * Returns 0 if a datagram should be allowed through, or various error codes
451 * if the socket was not a member of the group, or the source was muted, etc.
454 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
455 const struct sockaddr *group, const struct sockaddr *src)
457 struct in_mfilter *imf;
458 struct in_msource *ims;
461 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
463 imf = imo_match_group(imo, ifp, group);
465 return (MCAST_NOTGMEMBER);
468 * Check if the source was included in an (S,G) join.
469 * Allow reception on exclusive memberships by default,
470 * reject reception on inclusive memberships by default.
471 * Exclude source only if an in-mode exclude filter exists.
472 * Include source only if an in-mode include filter exists.
473 * NOTE: We are comparing group state here at IGMP t1 (now)
474 * with socket-layer t0 (since last downcall).
476 mode = imf->imf_st[1];
477 ims = imo_match_source(imf, src);
479 if ((ims == NULL && mode == MCAST_INCLUDE) ||
480 (ims != NULL && ims->imsl_st[0] == MCAST_EXCLUDE))
481 return (MCAST_NOTSMEMBER);
487 * Find and return a reference to an in_multi record for (ifp, group),
488 * and bump its reference count.
489 * If one does not exist, try to allocate it, and update link-layer multicast
490 * filters on ifp to listen for group.
491 * Assumes the IN_MULTI lock is held across the call.
492 * Return 0 if successful, otherwise return an appropriate error code.
495 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
496 struct in_multi **pinm)
498 struct sockaddr_in gsin;
499 struct ifmultiaddr *ifma;
500 struct in_ifinfo *ii;
501 struct in_multi *inm;
504 IN_MULTI_LOCK_ASSERT();
506 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
507 IN_MULTI_LIST_LOCK();
508 inm = inm_lookup(ifp, *group);
511 * If we already joined this group, just bump the
512 * refcount and return it.
514 KASSERT(inm->inm_refcount >= 1,
515 ("%s: bad refcount %d", __func__, inm->inm_refcount));
516 inm_acquire_locked(inm);
519 IN_MULTI_LIST_UNLOCK();
523 memset(&gsin, 0, sizeof(gsin));
524 gsin.sin_family = AF_INET;
525 gsin.sin_len = sizeof(struct sockaddr_in);
526 gsin.sin_addr = *group;
529 * Check if a link-layer group is already associated
530 * with this network-layer group on the given ifnet.
532 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
536 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
537 IN_MULTI_LIST_LOCK();
541 * If something other than netinet is occupying the link-layer
542 * group, print a meaningful error message and back out of
544 * Otherwise, bump the refcount on the existing network-layer
545 * group association and return it.
547 if (ifma->ifma_protospec != NULL) {
548 inm = (struct in_multi *)ifma->ifma_protospec;
550 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
552 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
553 ("%s: ifma not AF_INET", __func__));
554 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
555 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
556 !in_hosteq(inm->inm_addr, *group)) {
557 char addrbuf[INET_ADDRSTRLEN];
559 panic("%s: ifma %p is inconsistent with %p (%s)",
560 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
563 inm_acquire_locked(inm);
568 IF_ADDR_WLOCK_ASSERT(ifp);
571 * A new in_multi record is needed; allocate and initialize it.
572 * We DO NOT perform an IGMP join as the in_ layer may need to
573 * push an initial source list down to IGMP to support SSM.
575 * The initial source filter state is INCLUDE, {} as per the RFC.
577 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
579 IF_ADDR_WUNLOCK(ifp);
580 IN_MULTI_LIST_UNLOCK();
581 if_delmulti_ifma(ifma);
584 inm->inm_addr = *group;
586 inm->inm_igi = ii->ii_igmp;
587 inm->inm_ifma = ifma;
588 inm->inm_refcount = 1;
589 inm->inm_state = IGMP_NOT_MEMBER;
590 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
591 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
592 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
593 RB_INIT(&inm->inm_srcs);
595 ifma->ifma_protospec = inm;
599 IF_ADDR_WUNLOCK(ifp);
600 IN_MULTI_LIST_UNLOCK();
605 * Drop a reference to an in_multi record.
607 * If the refcount drops to 0, free the in_multi record and
608 * delete the underlying link-layer membership.
611 inm_release(struct in_multi *inm)
613 struct ifmultiaddr *ifma;
616 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
617 MPASS(inm->inm_refcount == 0);
618 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
620 ifma = inm->inm_ifma;
623 /* XXX this access is not covered by IF_ADDR_LOCK */
624 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
626 CURVNET_SET(ifp->if_vnet);
628 free(inm, M_IPMADDR);
629 if_delmulti_ifma_flags(ifma, 1);
634 free(inm, M_IPMADDR);
635 if_delmulti_ifma_flags(ifma, 1);
640 * Clear recorded source entries for a group.
641 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
642 * FIXME: Should reap.
645 inm_clear_recorded(struct in_multi *inm)
647 struct ip_msource *ims;
649 IN_MULTI_LIST_LOCK_ASSERT();
651 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
654 --inm->inm_st[1].iss_rec;
657 KASSERT(inm->inm_st[1].iss_rec == 0,
658 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
662 * Record a source as pending for a Source-Group IGMPv3 query.
663 * This lives here as it modifies the shared tree.
665 * inm is the group descriptor.
666 * naddr is the address of the source to record in network-byte order.
668 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
669 * lazy-allocate a source node in response to an SG query.
670 * Otherwise, no allocation is performed. This saves some memory
671 * with the trade-off that the source will not be reported to the
672 * router if joined in the window between the query response and
673 * the group actually being joined on the local host.
675 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
676 * This turns off the allocation of a recorded source entry if
677 * the group has not been joined.
679 * Return 0 if the source didn't exist or was already marked as recorded.
680 * Return 1 if the source was marked as recorded by this function.
681 * Return <0 if any error occurred (negated errno code).
684 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
686 struct ip_msource find;
687 struct ip_msource *ims, *nims;
689 IN_MULTI_LIST_LOCK_ASSERT();
691 find.ims_haddr = ntohl(naddr);
692 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
693 if (ims && ims->ims_stp)
696 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
698 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
702 nims->ims_haddr = find.ims_haddr;
703 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
709 * Mark the source as recorded and update the recorded
713 ++inm->inm_st[1].iss_rec;
719 * Return a pointer to an in_msource owned by an in_mfilter,
720 * given its source address.
721 * Lazy-allocate if needed. If this is a new entry its filter state is
724 * imf is the filter set being modified.
725 * haddr is the source address in *host* byte-order.
727 * SMPng: May be called with locks held; malloc must not block.
730 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
731 struct in_msource **plims)
733 struct ip_msource find;
734 struct ip_msource *ims, *nims;
735 struct in_msource *lims;
742 /* key is host byte order */
743 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
744 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
745 lims = (struct in_msource *)ims;
747 if (imf->imf_nsrc == in_mcast_maxsocksrc)
749 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
753 lims = (struct in_msource *)nims;
754 lims->ims_haddr = find.ims_haddr;
755 lims->imsl_st[0] = MCAST_UNDEFINED;
756 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
766 * Graft a source entry into an existing socket-layer filter set,
767 * maintaining any required invariants and checking allocations.
769 * The source is marked as being in the new filter mode at t1.
771 * Return the pointer to the new node, otherwise return NULL.
773 static struct in_msource *
774 imf_graft(struct in_mfilter *imf, const uint8_t st1,
775 const struct sockaddr_in *psin)
777 struct ip_msource *nims;
778 struct in_msource *lims;
780 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
784 lims = (struct in_msource *)nims;
785 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
786 lims->imsl_st[0] = MCAST_UNDEFINED;
787 lims->imsl_st[1] = st1;
788 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
795 * Prune a source entry from an existing socket-layer filter set,
796 * maintaining any required invariants and checking allocations.
798 * The source is marked as being left at t1, it is not freed.
800 * Return 0 if no error occurred, otherwise return an errno value.
803 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
805 struct ip_msource find;
806 struct ip_msource *ims;
807 struct in_msource *lims;
809 /* key is host byte order */
810 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
811 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
814 lims = (struct in_msource *)ims;
815 lims->imsl_st[1] = MCAST_UNDEFINED;
820 * Revert socket-layer filter set deltas at t1 to t0 state.
823 imf_rollback(struct in_mfilter *imf)
825 struct ip_msource *ims, *tims;
826 struct in_msource *lims;
828 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
829 lims = (struct in_msource *)ims;
830 if (lims->imsl_st[0] == lims->imsl_st[1]) {
831 /* no change at t1 */
833 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
834 /* revert change to existing source at t1 */
835 lims->imsl_st[1] = lims->imsl_st[0];
837 /* revert source added t1 */
838 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
839 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
840 free(ims, M_INMFILTER);
844 imf->imf_st[1] = imf->imf_st[0];
848 * Mark socket-layer filter set as INCLUDE {} at t1.
851 imf_leave(struct in_mfilter *imf)
853 struct ip_msource *ims;
854 struct in_msource *lims;
856 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
857 lims = (struct in_msource *)ims;
858 lims->imsl_st[1] = MCAST_UNDEFINED;
860 imf->imf_st[1] = MCAST_INCLUDE;
864 * Mark socket-layer filter set deltas as committed.
867 imf_commit(struct in_mfilter *imf)
869 struct ip_msource *ims;
870 struct in_msource *lims;
872 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
873 lims = (struct in_msource *)ims;
874 lims->imsl_st[0] = lims->imsl_st[1];
876 imf->imf_st[0] = imf->imf_st[1];
880 * Reap unreferenced sources from socket-layer filter set.
883 imf_reap(struct in_mfilter *imf)
885 struct ip_msource *ims, *tims;
886 struct in_msource *lims;
888 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
889 lims = (struct in_msource *)ims;
890 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
891 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
892 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
893 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
894 free(ims, M_INMFILTER);
901 * Purge socket-layer filter set.
904 imf_purge(struct in_mfilter *imf)
906 struct ip_msource *ims, *tims;
908 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
909 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
910 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
911 free(ims, M_INMFILTER);
914 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
915 KASSERT(RB_EMPTY(&imf->imf_sources),
916 ("%s: imf_sources not empty", __func__));
920 * Look up a source filter entry for a multicast group.
922 * inm is the group descriptor to work with.
923 * haddr is the host-byte-order IPv4 address to look up.
924 * noalloc may be non-zero to suppress allocation of sources.
925 * *pims will be set to the address of the retrieved or allocated source.
927 * SMPng: NOTE: may be called with locks held.
928 * Return 0 if successful, otherwise return a non-zero error code.
931 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
932 const int noalloc, struct ip_msource **pims)
934 struct ip_msource find;
935 struct ip_msource *ims, *nims;
937 find.ims_haddr = haddr;
938 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
939 if (ims == NULL && !noalloc) {
940 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
942 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
946 nims->ims_haddr = haddr;
947 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
951 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
961 * Merge socket-layer source into IGMP-layer source.
962 * If rollback is non-zero, perform the inverse of the merge.
965 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
968 int n = rollback ? -1 : 1;
970 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
971 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
972 __func__, n, ims->ims_haddr);
973 ims->ims_st[1].ex -= n;
974 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
975 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
976 __func__, n, ims->ims_haddr);
977 ims->ims_st[1].in -= n;
980 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
981 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
982 __func__, n, ims->ims_haddr);
983 ims->ims_st[1].ex += n;
984 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
985 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
986 __func__, n, ims->ims_haddr);
987 ims->ims_st[1].in += n;
992 * Atomically update the global in_multi state, when a membership's
993 * filter list is being updated in any way.
995 * imf is the per-inpcb-membership group filter pointer.
996 * A fake imf may be passed for in-kernel consumers.
998 * XXX This is a candidate for a set-symmetric-difference style loop
999 * which would eliminate the repeated lookup from root of ims nodes,
1000 * as they share the same key space.
1002 * If any error occurred this function will back out of refcounts
1003 * and return a non-zero value.
1006 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1008 struct ip_msource *ims, *nims;
1009 struct in_msource *lims;
1010 int schanged, error;
1016 IN_MULTI_LIST_LOCK_ASSERT();
1019 * Update the source filters first, as this may fail.
1020 * Maintain count of in-mode filters at t0, t1. These are
1021 * used to work out if we transition into ASM mode or not.
1022 * Maintain a count of source filters whose state was
1023 * actually modified by this operation.
1025 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1026 lims = (struct in_msource *)ims;
1027 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1028 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1029 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1030 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1034 ims_merge(nims, lims, 0);
1037 struct ip_msource *bims;
1039 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1040 lims = (struct in_msource *)ims;
1041 if (lims->imsl_st[0] == lims->imsl_st[1])
1043 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1046 ims_merge(bims, lims, 1);
1051 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1052 __func__, nsrc0, nsrc1);
1054 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1055 if (imf->imf_st[0] == imf->imf_st[1] &&
1056 imf->imf_st[1] == MCAST_INCLUDE) {
1058 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1059 --inm->inm_st[1].iss_in;
1063 /* Handle filter mode transition on socket. */
1064 if (imf->imf_st[0] != imf->imf_st[1]) {
1065 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1066 __func__, imf->imf_st[0], imf->imf_st[1]);
1068 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1069 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1070 --inm->inm_st[1].iss_ex;
1071 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1072 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1073 --inm->inm_st[1].iss_in;
1076 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1077 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1078 inm->inm_st[1].iss_ex++;
1079 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1080 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1081 inm->inm_st[1].iss_in++;
1086 * Track inm filter state in terms of listener counts.
1087 * If there are any exclusive listeners, stack-wide
1088 * membership is exclusive.
1089 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1090 * If no listeners remain, state is undefined at t1,
1091 * and the IGMP lifecycle for this group should finish.
1093 if (inm->inm_st[1].iss_ex > 0) {
1094 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1095 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1096 } else if (inm->inm_st[1].iss_in > 0) {
1097 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1098 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1100 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1101 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1104 /* Decrement ASM listener count on transition out of ASM mode. */
1105 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1106 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1107 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1108 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1109 --inm->inm_st[1].iss_asm;
1113 /* Increment ASM listener count on transition to ASM mode. */
1114 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1115 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1116 inm->inm_st[1].iss_asm++;
1119 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1124 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1131 * Mark an in_multi's filter set deltas as committed.
1132 * Called by IGMP after a state change has been enqueued.
1135 inm_commit(struct in_multi *inm)
1137 struct ip_msource *ims;
1139 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1140 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1143 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1144 ims->ims_st[0] = ims->ims_st[1];
1146 inm->inm_st[0] = inm->inm_st[1];
1150 * Reap unreferenced nodes from an in_multi's filter set.
1153 inm_reap(struct in_multi *inm)
1155 struct ip_msource *ims, *tims;
1157 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1158 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1159 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1162 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1163 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1164 free(ims, M_IPMSOURCE);
1170 * Purge all source nodes from an in_multi's filter set.
1173 inm_purge(struct in_multi *inm)
1175 struct ip_msource *ims, *tims;
1177 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1178 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1179 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1180 free(ims, M_IPMSOURCE);
1183 mbufq_drain(&inm->inm_scq);
1187 * Join a multicast group; unlocked entry point.
1189 * SMPng: XXX: in_joingroup() is called from in_control(). Fortunately,
1190 * ifp is unlikely to have been detached at this point, so we assume
1191 * it's OK to recurse.
1194 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1195 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1200 error = in_joingroup_locked(ifp, gina, imf, pinm);
1207 * Join a multicast group; real entry point.
1209 * Only preserves atomicity at inm level.
1210 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1212 * If the IGMP downcall fails, the group is not joined, and an error
1216 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1217 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1219 struct in_mfilter timf;
1220 struct in_multi *inm;
1223 IN_MULTI_LOCK_ASSERT();
1224 IN_MULTI_LIST_UNLOCK_ASSERT();
1226 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1227 ntohl(gina->s_addr), ifp, ifp->if_xname);
1233 * If no imf was specified (i.e. kernel consumer),
1234 * fake one up and assume it is an ASM join.
1237 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1241 error = in_getmulti(ifp, gina, &inm);
1243 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1246 IN_MULTI_LIST_LOCK();
1247 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1248 error = inm_merge(inm, imf);
1250 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1251 goto out_inm_release;
1254 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1255 error = igmp_change_state(inm);
1257 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1258 goto out_inm_release;
1263 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1265 inm_release_deferred(inm);
1266 IF_ADDR_WUNLOCK(ifp);
1270 IN_MULTI_LIST_UNLOCK();
1276 * Leave a multicast group; unlocked entry point.
1279 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1284 error = in_leavegroup_locked(inm, imf);
1291 * Leave a multicast group; real entry point.
1292 * All source filters will be expunged.
1294 * Only preserves atomicity at inm level.
1296 * Holding the write lock for the INP which contains imf
1297 * is highly advisable. We can't assert for it as imf does not
1298 * contain a back-pointer to the owning inp.
1300 * Note: This is not the same as inm_release(*) as this function also
1301 * makes a state change downcall into IGMP.
1304 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1306 struct in_mfilter timf;
1309 IN_MULTI_LOCK_ASSERT();
1310 IN_MULTI_LIST_UNLOCK_ASSERT();
1314 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1315 inm, ntohl(inm->inm_addr.s_addr),
1316 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1320 * If no imf was specified (i.e. kernel consumer),
1321 * fake one up and assume it is an ASM join.
1324 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1329 * Begin state merge transaction at IGMP layer.
1331 * As this particular invocation should not cause any memory
1332 * to be allocated, and there is no opportunity to roll back
1333 * the transaction, it MUST NOT fail.
1335 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1336 IN_MULTI_LIST_LOCK();
1337 error = inm_merge(inm, imf);
1338 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1340 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1341 CURVNET_SET(inm->inm_ifp->if_vnet);
1342 error = igmp_change_state(inm);
1343 IF_ADDR_WLOCK(inm->inm_ifp);
1344 inm_release_deferred(inm);
1345 IF_ADDR_WUNLOCK(inm->inm_ifp);
1346 IN_MULTI_LIST_UNLOCK();
1349 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1351 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1356 /*#ifndef BURN_BRIDGES*/
1359 * Block or unblock an ASM multicast source on an inpcb.
1360 * This implements the delta-based API described in RFC 3678.
1362 * The delta-based API applies only to exclusive-mode memberships.
1363 * An IGMP downcall will be performed.
1365 * Return 0 if successful, otherwise return an appropriate error code.
1368 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1370 struct epoch_tracker et;
1371 struct group_source_req gsr;
1372 sockunion_t *gsa, *ssa;
1374 struct in_mfilter *imf;
1375 struct ip_moptions *imo;
1376 struct in_msource *ims;
1377 struct in_multi *inm;
1385 memset(&gsr, 0, sizeof(struct group_source_req));
1386 gsa = (sockunion_t *)&gsr.gsr_group;
1387 ssa = (sockunion_t *)&gsr.gsr_source;
1389 switch (sopt->sopt_name) {
1390 case IP_BLOCK_SOURCE:
1391 case IP_UNBLOCK_SOURCE: {
1392 struct ip_mreq_source mreqs;
1394 error = sooptcopyin(sopt, &mreqs,
1395 sizeof(struct ip_mreq_source),
1396 sizeof(struct ip_mreq_source));
1400 gsa->sin.sin_family = AF_INET;
1401 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1402 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1404 ssa->sin.sin_family = AF_INET;
1405 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1406 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1408 if (!in_nullhost(mreqs.imr_interface)) {
1409 NET_EPOCH_ENTER(et);
1410 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1414 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1417 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1418 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1422 case MCAST_BLOCK_SOURCE:
1423 case MCAST_UNBLOCK_SOURCE:
1424 error = sooptcopyin(sopt, &gsr,
1425 sizeof(struct group_source_req),
1426 sizeof(struct group_source_req));
1430 if (gsa->sin.sin_family != AF_INET ||
1431 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1434 if (ssa->sin.sin_family != AF_INET ||
1435 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1438 NET_EPOCH_ENTER(et);
1439 ifp = ifnet_byindex(gsr.gsr_interface);
1442 return (EADDRNOTAVAIL);
1444 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1449 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1450 __func__, sopt->sopt_name);
1451 return (EOPNOTSUPP);
1455 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1461 * Check if we are actually a member of this group.
1463 imo = inp_findmoptions(inp);
1464 imf = imo_match_group(imo, ifp, &gsa->sa);
1466 error = EADDRNOTAVAIL;
1467 goto out_inp_locked;
1472 * Attempting to use the delta-based API on an
1473 * non exclusive-mode membership is an error.
1475 fmode = imf->imf_st[0];
1476 if (fmode != MCAST_EXCLUDE) {
1478 goto out_inp_locked;
1482 * Deal with error cases up-front:
1483 * Asked to block, but already blocked; or
1484 * Asked to unblock, but nothing to unblock.
1485 * If adding a new block entry, allocate it.
1487 ims = imo_match_source(imf, &ssa->sa);
1488 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1489 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1490 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1491 error = EADDRNOTAVAIL;
1492 goto out_inp_locked;
1495 INP_WLOCK_ASSERT(inp);
1498 * Begin state merge transaction at socket layer.
1501 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1502 ims = imf_graft(imf, fmode, &ssa->sin);
1506 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1507 error = imf_prune(imf, &ssa->sin);
1511 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1512 goto out_imf_rollback;
1516 * Begin state merge transaction at IGMP layer.
1518 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1519 IN_MULTI_LIST_LOCK();
1520 error = inm_merge(inm, imf);
1522 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1523 IN_MULTI_LIST_UNLOCK();
1524 goto out_imf_rollback;
1527 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1528 error = igmp_change_state(inm);
1529 IN_MULTI_LIST_UNLOCK();
1531 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1548 * Given an inpcb, return its multicast options structure pointer. Accepts
1549 * an unlocked inpcb pointer, but will return it locked. May sleep.
1551 * SMPng: NOTE: Returns with the INP write lock held.
1553 static struct ip_moptions *
1554 inp_findmoptions(struct inpcb *inp)
1556 struct ip_moptions *imo;
1559 if (inp->inp_moptions != NULL)
1560 return (inp->inp_moptions);
1564 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1566 imo->imo_multicast_ifp = NULL;
1567 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1568 imo->imo_multicast_vif = -1;
1569 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1570 imo->imo_multicast_loop = in_mcast_loop;
1571 STAILQ_INIT(&imo->imo_head);
1574 if (inp->inp_moptions != NULL) {
1575 free(imo, M_IPMOPTS);
1576 return (inp->inp_moptions);
1578 inp->inp_moptions = imo;
1583 inp_freemoptions(struct ip_moptions *imo)
1585 struct in_mfilter *imf;
1586 struct in_multi *inm;
1592 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1593 ip_mfilter_remove(&imo->imo_head, imf);
1596 if ((inm = imf->imf_inm) != NULL) {
1597 if ((ifp = inm->inm_ifp) != NULL) {
1598 CURVNET_SET(ifp->if_vnet);
1599 (void)in_leavegroup(inm, imf);
1602 (void)in_leavegroup(inm, imf);
1605 ip_mfilter_free(imf);
1607 free(imo, M_IPMOPTS);
1611 * Atomically get source filters on a socket for an IPv4 multicast group.
1612 * Called with INP lock held; returns with lock released.
1615 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1617 struct epoch_tracker et;
1618 struct __msfilterreq msfr;
1621 struct ip_moptions *imo;
1622 struct in_mfilter *imf;
1623 struct ip_msource *ims;
1624 struct in_msource *lims;
1625 struct sockaddr_in *psin;
1626 struct sockaddr_storage *ptss;
1627 struct sockaddr_storage *tss;
1629 size_t nsrcs, ncsrcs;
1631 INP_WLOCK_ASSERT(inp);
1633 imo = inp->inp_moptions;
1634 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1638 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1639 sizeof(struct __msfilterreq));
1643 NET_EPOCH_ENTER(et);
1644 ifp = ifnet_byindex(msfr.msfr_ifindex);
1645 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifnet pointer left */
1652 * Lookup group on the socket.
1654 gsa = (sockunion_t *)&msfr.msfr_group;
1655 imf = imo_match_group(imo, ifp, &gsa->sa);
1658 return (EADDRNOTAVAIL);
1662 * Ignore memberships which are in limbo.
1664 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1668 msfr.msfr_fmode = imf->imf_st[1];
1671 * If the user specified a buffer, copy out the source filter
1672 * entries to userland gracefully.
1673 * We only copy out the number of entries which userland
1674 * has asked for, but we always tell userland how big the
1675 * buffer really needs to be.
1677 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1678 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1680 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1681 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1682 M_TEMP, M_NOWAIT | M_ZERO);
1690 * Count number of sources in-mode at t0.
1691 * If buffer space exists and remains, copy out source entries.
1693 nsrcs = msfr.msfr_nsrcs;
1696 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1697 lims = (struct in_msource *)ims;
1698 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1699 lims->imsl_st[0] != imf->imf_st[0])
1702 if (tss != NULL && nsrcs > 0) {
1703 psin = (struct sockaddr_in *)ptss;
1704 psin->sin_family = AF_INET;
1705 psin->sin_len = sizeof(struct sockaddr_in);
1706 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1716 error = copyout(tss, msfr.msfr_srcs,
1717 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1723 msfr.msfr_nsrcs = ncsrcs;
1724 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1730 * Return the IP multicast options in response to user getsockopt().
1733 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1735 struct ip_mreqn mreqn;
1736 struct ip_moptions *imo;
1738 struct in_ifaddr *ia;
1743 imo = inp->inp_moptions;
1744 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM reject it. */
1745 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1746 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM) {
1748 return (EOPNOTSUPP);
1752 switch (sopt->sopt_name) {
1753 case IP_MULTICAST_VIF:
1755 optval = imo->imo_multicast_vif;
1759 error = sooptcopyout(sopt, &optval, sizeof(int));
1762 case IP_MULTICAST_IF:
1763 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1765 ifp = imo->imo_multicast_ifp;
1766 if (!in_nullhost(imo->imo_multicast_addr)) {
1767 mreqn.imr_address = imo->imo_multicast_addr;
1768 } else if (ifp != NULL) {
1769 struct epoch_tracker et;
1771 mreqn.imr_ifindex = ifp->if_index;
1772 NET_EPOCH_ENTER(et);
1776 IA_SIN(ia)->sin_addr;
1781 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1782 error = sooptcopyout(sopt, &mreqn,
1783 sizeof(struct ip_mreqn));
1785 error = sooptcopyout(sopt, &mreqn.imr_address,
1786 sizeof(struct in_addr));
1790 case IP_MULTICAST_TTL:
1792 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1794 optval = coptval = imo->imo_multicast_ttl;
1796 if (sopt->sopt_valsize == sizeof(u_char))
1797 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1799 error = sooptcopyout(sopt, &optval, sizeof(int));
1802 case IP_MULTICAST_LOOP:
1804 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1806 optval = coptval = imo->imo_multicast_loop;
1808 if (sopt->sopt_valsize == sizeof(u_char))
1809 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1811 error = sooptcopyout(sopt, &optval, sizeof(int));
1816 error = EADDRNOTAVAIL;
1819 error = inp_get_source_filters(inp, sopt);
1825 error = ENOPROTOOPT;
1829 INP_UNLOCK_ASSERT(inp);
1835 * Look up the ifnet to use for a multicast group membership,
1836 * given the IPv4 address of an interface, and the IPv4 group address.
1838 * This routine exists to support legacy multicast applications
1839 * which do not understand that multicast memberships are scoped to
1840 * specific physical links in the networking stack, or which need
1841 * to join link-scope groups before IPv4 addresses are configured.
1843 * Use this socket's current FIB number for any required FIB lookup.
1844 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1845 * and use its ifp; usually, this points to the default next-hop.
1847 * If the FIB lookup fails, attempt to use the first non-loopback
1848 * interface with multicast capability in the system as a
1849 * last resort. The legacy IPv4 ASM API requires that we do
1850 * this in order to allow groups to be joined when the routing
1851 * table has not yet been populated during boot.
1853 * Returns NULL if no ifp could be found, otherwise return referenced ifp.
1855 * FUTURE: Implement IPv4 source-address selection.
1857 static struct ifnet *
1858 inp_lookup_mcast_ifp(const struct inpcb *inp,
1859 const struct sockaddr_in *gsin, const struct in_addr ina)
1862 struct nhop_object *nh;
1865 KASSERT(inp != NULL, ("%s: inp must not be NULL", __func__));
1866 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1867 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1868 ("%s: not multicast", __func__));
1871 if (!in_nullhost(ina)) {
1872 INADDR_TO_IFP(ina, ifp);
1876 nh = fib4_lookup(inp->inp_inc.inc_fibnum, gsin->sin_addr, 0, NHR_NONE, 0);
1881 struct in_ifaddr *ia;
1885 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1887 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1888 (mifp->if_flags & IFF_MULTICAST)) {
1901 * Join an IPv4 multicast group, possibly with a source.
1904 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1906 struct group_source_req gsr;
1907 sockunion_t *gsa, *ssa;
1909 struct in_mfilter *imf;
1910 struct ip_moptions *imo;
1911 struct in_multi *inm;
1912 struct in_msource *lims;
1913 struct epoch_tracker et;
1920 memset(&gsr, 0, sizeof(struct group_source_req));
1921 gsa = (sockunion_t *)&gsr.gsr_group;
1922 gsa->ss.ss_family = AF_UNSPEC;
1923 ssa = (sockunion_t *)&gsr.gsr_source;
1924 ssa->ss.ss_family = AF_UNSPEC;
1926 switch (sopt->sopt_name) {
1927 case IP_ADD_MEMBERSHIP: {
1928 struct ip_mreqn mreqn;
1930 if (sopt->sopt_valsize == sizeof(struct ip_mreqn))
1931 error = sooptcopyin(sopt, &mreqn,
1932 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn));
1934 error = sooptcopyin(sopt, &mreqn,
1935 sizeof(struct ip_mreq), sizeof(struct ip_mreq));
1939 gsa->sin.sin_family = AF_INET;
1940 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1941 gsa->sin.sin_addr = mreqn.imr_multiaddr;
1942 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1945 NET_EPOCH_ENTER(et);
1946 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) &&
1947 mreqn.imr_ifindex != 0)
1948 ifp = ifnet_byindex_ref(mreqn.imr_ifindex);
1950 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1955 case IP_ADD_SOURCE_MEMBERSHIP: {
1956 struct ip_mreq_source mreqs;
1958 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source),
1959 sizeof(struct ip_mreq_source));
1963 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET;
1964 gsa->sin.sin_len = ssa->sin.sin_len =
1965 sizeof(struct sockaddr_in);
1967 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1968 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1971 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1973 NET_EPOCH_ENTER(et);
1974 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1975 mreqs.imr_interface);
1977 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1978 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1982 case MCAST_JOIN_GROUP:
1983 case MCAST_JOIN_SOURCE_GROUP:
1984 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1985 error = sooptcopyin(sopt, &gsr,
1986 sizeof(struct group_req),
1987 sizeof(struct group_req));
1988 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1989 error = sooptcopyin(sopt, &gsr,
1990 sizeof(struct group_source_req),
1991 sizeof(struct group_source_req));
1996 if (gsa->sin.sin_family != AF_INET ||
1997 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2001 * Overwrite the port field if present, as the sockaddr
2002 * being copied in may be matched with a binary comparison.
2004 gsa->sin.sin_port = 0;
2005 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2006 if (ssa->sin.sin_family != AF_INET ||
2007 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2009 ssa->sin.sin_port = 0;
2012 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2015 NET_EPOCH_ENTER(et);
2016 ifp = ifnet_byindex_ref(gsr.gsr_interface);
2019 return (EADDRNOTAVAIL);
2023 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2024 __func__, sopt->sopt_name);
2025 return (EOPNOTSUPP);
2029 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2032 return (EADDRNOTAVAIL);
2038 * Find the membership in the membership list.
2040 imo = inp_findmoptions(inp);
2041 imf = imo_match_group(imo, ifp, &gsa->sa);
2046 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) {
2048 goto out_inp_locked;
2054 if (ssa->ss.ss_family != AF_UNSPEC) {
2056 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2057 * is an error. On an existing inclusive membership,
2058 * it just adds the source to the filter list.
2060 if (imf->imf_st[1] != MCAST_INCLUDE) {
2062 goto out_inp_locked;
2065 * Throw out duplicates.
2067 * XXX FIXME: This makes a naive assumption that
2068 * even if entries exist for *ssa in this imf,
2069 * they will be rejected as dupes, even if they
2070 * are not valid in the current mode (in-mode).
2072 * in_msource is transactioned just as for anything
2073 * else in SSM -- but note naive use of inm_graft()
2074 * below for allocating new filter entries.
2076 * This is only an issue if someone mixes the
2077 * full-state SSM API with the delta-based API,
2078 * which is discouraged in the relevant RFCs.
2080 lims = imo_match_source(imf, &ssa->sa);
2081 if (lims != NULL /*&&
2082 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2083 error = EADDRNOTAVAIL;
2084 goto out_inp_locked;
2088 * MCAST_JOIN_GROUP on an existing exclusive
2089 * membership is an error; return EADDRINUSE
2090 * to preserve 4.4BSD API idempotence, and
2091 * avoid tedious detour to code below.
2092 * NOTE: This is bending RFC 3678 a bit.
2094 * On an existing inclusive membership, this is also
2095 * an error; if you want to change filter mode,
2096 * you must use the userland API setsourcefilter().
2097 * XXX We don't reject this for imf in UNDEFINED
2098 * state at t1, because allocation of a filter
2099 * is atomic with allocation of a membership.
2102 if (imf->imf_st[1] == MCAST_EXCLUDE)
2104 goto out_inp_locked;
2109 * Begin state merge transaction at socket layer.
2111 INP_WLOCK_ASSERT(inp);
2114 * Graft new source into filter list for this inpcb's
2115 * membership of the group. The in_multi may not have
2116 * been allocated yet if this is a new membership, however,
2117 * the in_mfilter slot will be allocated and must be initialized.
2119 * Note: Grafting of exclusive mode filters doesn't happen
2121 * XXX: Should check for non-NULL lims (node exists but may
2122 * not be in-mode) for interop with full-state API.
2124 if (ssa->ss.ss_family != AF_UNSPEC) {
2125 /* Membership starts in IN mode */
2127 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2128 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2131 goto out_inp_locked;
2134 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2136 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2138 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2141 goto out_inp_locked;
2144 /* No address specified; Membership starts in EX mode */
2146 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2147 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2150 goto out_inp_locked;
2156 * Begin state merge transaction at IGMP layer.
2162 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2166 if (in_pcbrele_wlocked(inp)) {
2168 goto out_inp_unlocked;
2171 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2173 goto out_inp_locked;
2176 * NOTE: Refcount from in_joingroup_locked()
2177 * is protecting membership.
2179 ip_mfilter_insert(&imo->imo_head, imf);
2181 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2182 IN_MULTI_LIST_LOCK();
2183 error = inm_merge(inm, imf);
2185 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2187 IN_MULTI_LIST_UNLOCK();
2190 goto out_inp_locked;
2192 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2193 error = igmp_change_state(inm);
2194 IN_MULTI_LIST_UNLOCK();
2196 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2200 goto out_inp_locked;
2212 if (is_new && imf) {
2213 if (imf->imf_inm != NULL) {
2214 IN_MULTI_LIST_LOCK();
2216 inm_release_deferred(imf->imf_inm);
2217 IF_ADDR_WUNLOCK(ifp);
2218 IN_MULTI_LIST_UNLOCK();
2220 ip_mfilter_free(imf);
2227 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2230 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2232 struct epoch_tracker et;
2233 struct group_source_req gsr;
2234 struct ip_mreq_source mreqs;
2235 sockunion_t *gsa, *ssa;
2237 struct in_mfilter *imf;
2238 struct ip_moptions *imo;
2239 struct in_msource *ims;
2240 struct in_multi *inm;
2248 memset(&gsr, 0, sizeof(struct group_source_req));
2249 gsa = (sockunion_t *)&gsr.gsr_group;
2250 gsa->ss.ss_family = AF_UNSPEC;
2251 ssa = (sockunion_t *)&gsr.gsr_source;
2252 ssa->ss.ss_family = AF_UNSPEC;
2254 switch (sopt->sopt_name) {
2255 case IP_DROP_MEMBERSHIP:
2256 case IP_DROP_SOURCE_MEMBERSHIP:
2257 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2258 error = sooptcopyin(sopt, &mreqs,
2259 sizeof(struct ip_mreq),
2260 sizeof(struct ip_mreq));
2262 * Swap interface and sourceaddr arguments,
2263 * as ip_mreq and ip_mreq_source are laid
2266 mreqs.imr_interface = mreqs.imr_sourceaddr;
2267 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2268 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2269 error = sooptcopyin(sopt, &mreqs,
2270 sizeof(struct ip_mreq_source),
2271 sizeof(struct ip_mreq_source));
2276 gsa->sin.sin_family = AF_INET;
2277 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2278 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2280 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2281 ssa->sin.sin_family = AF_INET;
2282 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2283 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2287 * Attempt to look up hinted ifp from interface address.
2288 * Fallthrough with null ifp iff lookup fails, to
2289 * preserve 4.4BSD mcast API idempotence.
2290 * XXX NOTE WELL: The RFC 3678 API is preferred because
2291 * using an IPv4 address as a key is racy.
2293 if (!in_nullhost(mreqs.imr_interface)) {
2294 NET_EPOCH_ENTER(et);
2295 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2299 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2300 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2304 case MCAST_LEAVE_GROUP:
2305 case MCAST_LEAVE_SOURCE_GROUP:
2306 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2307 error = sooptcopyin(sopt, &gsr,
2308 sizeof(struct group_req),
2309 sizeof(struct group_req));
2310 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2311 error = sooptcopyin(sopt, &gsr,
2312 sizeof(struct group_source_req),
2313 sizeof(struct group_source_req));
2318 if (gsa->sin.sin_family != AF_INET ||
2319 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2322 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2323 if (ssa->sin.sin_family != AF_INET ||
2324 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2328 NET_EPOCH_ENTER(et);
2329 ifp = ifnet_byindex(gsr.gsr_interface);
2330 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2332 return (EADDRNOTAVAIL);
2336 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2337 __func__, sopt->sopt_name);
2338 return (EOPNOTSUPP);
2342 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2348 * Find the membership in the membership list.
2350 imo = inp_findmoptions(inp);
2351 imf = imo_match_group(imo, ifp, &gsa->sa);
2353 error = EADDRNOTAVAIL;
2354 goto out_inp_locked;
2358 if (ssa->ss.ss_family != AF_UNSPEC)
2362 * Begin state merge transaction at socket layer.
2364 INP_WLOCK_ASSERT(inp);
2367 * If we were instructed only to leave a given source, do so.
2368 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2371 ip_mfilter_remove(&imo->imo_head, imf);
2375 * Give up the multicast address record to which
2376 * the membership points.
2378 (void) in_leavegroup_locked(imf->imf_inm, imf);
2380 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2381 error = EADDRNOTAVAIL;
2382 goto out_inp_locked;
2384 ims = imo_match_source(imf, &ssa->sa);
2386 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2387 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2388 error = EADDRNOTAVAIL;
2389 goto out_inp_locked;
2391 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2392 error = imf_prune(imf, &ssa->sin);
2394 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2396 goto out_inp_locked;
2401 * Begin state merge transaction at IGMP layer.
2404 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2405 IN_MULTI_LIST_LOCK();
2406 error = inm_merge(inm, imf);
2408 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2410 IN_MULTI_LIST_UNLOCK();
2413 goto out_inp_locked;
2416 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2417 error = igmp_change_state(inm);
2418 IN_MULTI_LIST_UNLOCK();
2420 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2424 goto out_inp_locked;
2433 if (is_final && imf)
2434 ip_mfilter_free(imf);
2441 * Select the interface for transmitting IPv4 multicast datagrams.
2443 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2444 * may be passed to this socket option. An address of INADDR_ANY or an
2445 * interface index of 0 is used to remove a previous selection.
2446 * When no interface is selected, one is chosen for every send.
2449 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2451 struct in_addr addr;
2452 struct ip_mreqn mreqn;
2454 struct ip_moptions *imo;
2457 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2459 * An interface index was specified using the
2460 * Linux-derived ip_mreqn structure.
2462 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2463 sizeof(struct ip_mreqn));
2467 if (mreqn.imr_ifindex < 0)
2470 if (mreqn.imr_ifindex == 0) {
2473 struct epoch_tracker et;
2475 NET_EPOCH_ENTER(et);
2476 ifp = ifnet_byindex(mreqn.imr_ifindex);
2477 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2479 return (EADDRNOTAVAIL);
2483 * An interface was specified by IPv4 address.
2484 * This is the traditional BSD usage.
2486 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2487 sizeof(struct in_addr));
2490 if (in_nullhost(addr)) {
2493 struct epoch_tracker et;
2495 NET_EPOCH_ENTER(et);
2496 INADDR_TO_IFP(addr, ifp);
2500 return (EADDRNOTAVAIL);
2502 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2503 ntohl(addr.s_addr));
2506 /* Reject interfaces which do not support multicast. */
2507 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2508 return (EOPNOTSUPP);
2510 imo = inp_findmoptions(inp);
2511 imo->imo_multicast_ifp = ifp;
2512 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2519 * Atomically set source filters on a socket for an IPv4 multicast group.
2522 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2524 struct epoch_tracker et;
2525 struct __msfilterreq msfr;
2528 struct in_mfilter *imf;
2529 struct ip_moptions *imo;
2530 struct in_multi *inm;
2533 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2534 sizeof(struct __msfilterreq));
2538 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2541 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2542 msfr.msfr_fmode != MCAST_INCLUDE))
2545 if (msfr.msfr_group.ss_family != AF_INET ||
2546 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2549 gsa = (sockunion_t *)&msfr.msfr_group;
2550 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2553 gsa->sin.sin_port = 0; /* ignore port */
2555 NET_EPOCH_ENTER(et);
2556 ifp = ifnet_byindex(msfr.msfr_ifindex);
2557 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2559 return (EADDRNOTAVAIL);
2564 * Take the INP write lock.
2565 * Check if this socket is a member of this group.
2567 imo = inp_findmoptions(inp);
2568 imf = imo_match_group(imo, ifp, &gsa->sa);
2570 error = EADDRNOTAVAIL;
2571 goto out_inp_locked;
2576 * Begin state merge transaction at socket layer.
2578 INP_WLOCK_ASSERT(inp);
2580 imf->imf_st[1] = msfr.msfr_fmode;
2583 * Apply any new source filters, if present.
2584 * Make a copy of the user-space source vector so
2585 * that we may copy them with a single copyin. This
2586 * allows us to deal with page faults up-front.
2588 if (msfr.msfr_nsrcs > 0) {
2589 struct in_msource *lims;
2590 struct sockaddr_in *psin;
2591 struct sockaddr_storage *kss, *pkss;
2596 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2597 __func__, (unsigned long)msfr.msfr_nsrcs);
2598 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2600 error = copyin(msfr.msfr_srcs, kss,
2601 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2610 * Mark all source filters as UNDEFINED at t1.
2611 * Restore new group filter mode, as imf_leave()
2612 * will set it to INCLUDE.
2615 imf->imf_st[1] = msfr.msfr_fmode;
2618 * Update socket layer filters at t1, lazy-allocating
2619 * new entries. This saves a bunch of memory at the
2620 * cost of one RB_FIND() per source entry; duplicate
2621 * entries in the msfr_nsrcs vector are ignored.
2622 * If we encounter an error, rollback transaction.
2624 * XXX This too could be replaced with a set-symmetric
2625 * difference like loop to avoid walking from root
2626 * every time, as the key space is common.
2628 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2629 psin = (struct sockaddr_in *)pkss;
2630 if (psin->sin_family != AF_INET) {
2631 error = EAFNOSUPPORT;
2634 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2638 error = imf_get_source(imf, psin, &lims);
2641 lims->imsl_st[1] = imf->imf_st[1];
2647 goto out_imf_rollback;
2649 INP_WLOCK_ASSERT(inp);
2652 * Begin state merge transaction at IGMP layer.
2654 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2655 IN_MULTI_LIST_LOCK();
2656 error = inm_merge(inm, imf);
2658 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2659 IN_MULTI_LIST_UNLOCK();
2660 goto out_imf_rollback;
2663 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2664 error = igmp_change_state(inm);
2665 IN_MULTI_LIST_UNLOCK();
2667 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2684 * Set the IP multicast options in response to user setsockopt().
2686 * Many of the socket options handled in this function duplicate the
2687 * functionality of socket options in the regular unicast API. However,
2688 * it is not possible to merge the duplicate code, because the idempotence
2689 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2690 * the effects of these options must be treated as separate and distinct.
2692 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2693 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2694 * is refactored to no longer use vifs.
2697 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2699 struct ip_moptions *imo;
2704 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
2705 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2706 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)
2707 return (EOPNOTSUPP);
2709 switch (sopt->sopt_name) {
2710 case IP_MULTICAST_VIF: {
2713 * Select a multicast VIF for transmission.
2714 * Only useful if multicast forwarding is active.
2716 if (legal_vif_num == NULL) {
2720 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2723 if (!legal_vif_num(vifi) && (vifi != -1)) {
2727 imo = inp_findmoptions(inp);
2728 imo->imo_multicast_vif = vifi;
2733 case IP_MULTICAST_IF:
2734 error = inp_set_multicast_if(inp, sopt);
2737 case IP_MULTICAST_TTL: {
2741 * Set the IP time-to-live for outgoing multicast packets.
2742 * The original multicast API required a char argument,
2743 * which is inconsistent with the rest of the socket API.
2744 * We allow either a char or an int.
2746 if (sopt->sopt_valsize == sizeof(u_char)) {
2747 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2754 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2764 imo = inp_findmoptions(inp);
2765 imo->imo_multicast_ttl = ttl;
2770 case IP_MULTICAST_LOOP: {
2774 * Set the loopback flag for outgoing multicast packets.
2775 * Must be zero or one. The original multicast API required a
2776 * char argument, which is inconsistent with the rest
2777 * of the socket API. We allow either a char or an int.
2779 if (sopt->sopt_valsize == sizeof(u_char)) {
2780 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2787 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2791 loop = (u_char)iloop;
2793 imo = inp_findmoptions(inp);
2794 imo->imo_multicast_loop = !!loop;
2799 case IP_ADD_MEMBERSHIP:
2800 case IP_ADD_SOURCE_MEMBERSHIP:
2801 case MCAST_JOIN_GROUP:
2802 case MCAST_JOIN_SOURCE_GROUP:
2803 error = inp_join_group(inp, sopt);
2806 case IP_DROP_MEMBERSHIP:
2807 case IP_DROP_SOURCE_MEMBERSHIP:
2808 case MCAST_LEAVE_GROUP:
2809 case MCAST_LEAVE_SOURCE_GROUP:
2810 error = inp_leave_group(inp, sopt);
2813 case IP_BLOCK_SOURCE:
2814 case IP_UNBLOCK_SOURCE:
2815 case MCAST_BLOCK_SOURCE:
2816 case MCAST_UNBLOCK_SOURCE:
2817 error = inp_block_unblock_source(inp, sopt);
2821 error = inp_set_source_filters(inp, sopt);
2829 INP_UNLOCK_ASSERT(inp);
2835 * Expose IGMP's multicast filter mode and source list(s) to userland,
2836 * keyed by (ifindex, group).
2837 * The filter mode is written out as a uint32_t, followed by
2838 * 0..n of struct in_addr.
2839 * For use by ifmcstat(8).
2840 * SMPng: NOTE: unlocked read of ifindex space.
2843 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2845 struct in_addr src, group;
2846 struct epoch_tracker et;
2848 struct ifmultiaddr *ifma;
2849 struct in_multi *inm;
2850 struct ip_msource *ims;
2854 uint32_t fmode, ifindex;
2859 if (req->newptr != NULL)
2865 group.s_addr = name[1];
2866 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2867 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2868 __func__, ntohl(group.s_addr));
2873 NET_EPOCH_ENTER(et);
2874 ifp = ifnet_byindex(ifindex);
2877 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2882 retval = sysctl_wire_old_buffer(req,
2883 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2889 IN_MULTI_LIST_LOCK();
2891 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2892 inm = inm_ifmultiaddr_get_inm(ifma);
2895 if (!in_hosteq(inm->inm_addr, group))
2897 fmode = inm->inm_st[1].iss_fmode;
2898 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2901 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2902 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2905 * Only copy-out sources which are in-mode.
2907 if (fmode != ims_get_mode(inm, ims, 1)) {
2908 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2912 src.s_addr = htonl(ims->ims_haddr);
2913 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2919 IN_MULTI_LIST_UNLOCK();
2925 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2927 static const char *inm_modestrs[] = {
2928 [MCAST_UNDEFINED] = "un",
2929 [MCAST_INCLUDE] = "in",
2930 [MCAST_EXCLUDE] = "ex",
2932 _Static_assert(MCAST_UNDEFINED == 0 &&
2933 MCAST_EXCLUDE + 1 == nitems(inm_modestrs),
2934 "inm_modestrs: no longer matches #defines");
2937 inm_mode_str(const int mode)
2940 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2941 return (inm_modestrs[mode]);
2945 static const char *inm_statestrs[] = {
2946 [IGMP_NOT_MEMBER] = "not-member",
2947 [IGMP_SILENT_MEMBER] = "silent",
2948 [IGMP_REPORTING_MEMBER] = "reporting",
2949 [IGMP_IDLE_MEMBER] = "idle",
2950 [IGMP_LAZY_MEMBER] = "lazy",
2951 [IGMP_SLEEPING_MEMBER] = "sleeping",
2952 [IGMP_AWAKENING_MEMBER] = "awakening",
2953 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending",
2954 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending",
2955 [IGMP_LEAVING_MEMBER] = "leaving",
2957 _Static_assert(IGMP_NOT_MEMBER == 0 &&
2958 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs),
2959 "inm_statetrs: no longer matches #defines");
2962 inm_state_str(const int state)
2965 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2966 return (inm_statestrs[state]);
2971 * Dump an in_multi structure to the console.
2974 inm_print(const struct in_multi *inm)
2977 char addrbuf[INET_ADDRSTRLEN];
2979 if ((ktr_mask & KTR_IGMPV3) == 0)
2982 printf("%s: --- begin inm %p ---\n", __func__, inm);
2983 printf("addr %s ifp %p(%s) ifma %p\n",
2984 inet_ntoa_r(inm->inm_addr, addrbuf),
2986 inm->inm_ifp->if_xname,
2988 printf("timer %u state %s refcount %u scq.len %u\n",
2990 inm_state_str(inm->inm_state),
2992 inm->inm_scq.mq_len);
2993 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2998 for (t = 0; t < 2; t++) {
2999 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3000 inm_mode_str(inm->inm_st[t].iss_fmode),
3001 inm->inm_st[t].iss_asm,
3002 inm->inm_st[t].iss_ex,
3003 inm->inm_st[t].iss_in,
3004 inm->inm_st[t].iss_rec);
3006 printf("%s: --- end inm %p ---\n", __func__, inm);
3009 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3012 inm_print(const struct in_multi *inm)
3017 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3019 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);