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 in_multi *inm;
393 IN_MULTI_LIST_LOCK_ASSERT();
395 inm = inm_lookup_locked(ifp, ina);
396 IF_ADDR_RUNLOCK(ifp);
402 * Find an IPv4 multicast group entry for this ip_moptions instance
403 * which matches the specified group, and optionally an interface.
404 * Return its index into the array, or -1 if not found.
406 static struct in_mfilter *
407 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
408 const struct sockaddr *group)
410 const struct sockaddr_in *gsin;
411 struct in_mfilter *imf;
412 struct in_multi *inm;
414 gsin = (const struct sockaddr_in *)group;
416 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
420 if ((ifp == NULL || (inm->inm_ifp == ifp)) &&
421 in_hosteq(inm->inm_addr, gsin->sin_addr)) {
429 * Find an IPv4 multicast source entry for this imo which matches
430 * the given group index for this socket, and source address.
432 * NOTE: This does not check if the entry is in-mode, merely if
433 * it exists, which may not be the desired behaviour.
435 static struct in_msource *
436 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src)
438 struct ip_msource find;
439 struct ip_msource *ims;
440 const sockunion_t *psa;
442 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
444 /* Source trees are keyed in host byte order. */
445 psa = (const sockunion_t *)src;
446 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
447 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
449 return ((struct in_msource *)ims);
453 * Perform filtering for multicast datagrams on a socket by group and source.
455 * Returns 0 if a datagram should be allowed through, or various error codes
456 * if the socket was not a member of the group, or the source was muted, etc.
459 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
460 const struct sockaddr *group, const struct sockaddr *src)
462 struct in_mfilter *imf;
463 struct in_msource *ims;
466 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
468 imf = imo_match_group(imo, ifp, group);
470 return (MCAST_NOTGMEMBER);
473 * Check if the source was included in an (S,G) join.
474 * Allow reception on exclusive memberships by default,
475 * reject reception on inclusive memberships by default.
476 * Exclude source only if an in-mode exclude filter exists.
477 * Include source only if an in-mode include filter exists.
478 * NOTE: We are comparing group state here at IGMP t1 (now)
479 * with socket-layer t0 (since last downcall).
481 mode = imf->imf_st[1];
482 ims = imo_match_source(imf, src);
484 if ((ims == NULL && mode == MCAST_INCLUDE) ||
485 (ims != NULL && ims->imsl_st[0] != mode))
486 return (MCAST_NOTSMEMBER);
492 * Find and return a reference to an in_multi record for (ifp, group),
493 * and bump its reference count.
494 * If one does not exist, try to allocate it, and update link-layer multicast
495 * filters on ifp to listen for group.
496 * Assumes the IN_MULTI lock is held across the call.
497 * Return 0 if successful, otherwise return an appropriate error code.
500 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
501 struct in_multi **pinm)
503 struct sockaddr_in gsin;
504 struct ifmultiaddr *ifma;
505 struct in_ifinfo *ii;
506 struct in_multi *inm;
509 IN_MULTI_LOCK_ASSERT();
511 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
512 IN_MULTI_LIST_LOCK();
513 inm = inm_lookup(ifp, *group);
516 * If we already joined this group, just bump the
517 * refcount and return it.
519 KASSERT(inm->inm_refcount >= 1,
520 ("%s: bad refcount %d", __func__, inm->inm_refcount));
521 inm_acquire_locked(inm);
524 IN_MULTI_LIST_UNLOCK();
528 memset(&gsin, 0, sizeof(gsin));
529 gsin.sin_family = AF_INET;
530 gsin.sin_len = sizeof(struct sockaddr_in);
531 gsin.sin_addr = *group;
534 * Check if a link-layer group is already associated
535 * with this network-layer group on the given ifnet.
537 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
541 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
542 IN_MULTI_LIST_LOCK();
546 * If something other than netinet is occupying the link-layer
547 * group, print a meaningful error message and back out of
549 * Otherwise, bump the refcount on the existing network-layer
550 * group association and return it.
552 if (ifma->ifma_protospec != NULL) {
553 inm = (struct in_multi *)ifma->ifma_protospec;
555 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
557 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
558 ("%s: ifma not AF_INET", __func__));
559 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
560 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
561 !in_hosteq(inm->inm_addr, *group)) {
562 char addrbuf[INET_ADDRSTRLEN];
564 panic("%s: ifma %p is inconsistent with %p (%s)",
565 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
568 inm_acquire_locked(inm);
573 IF_ADDR_WLOCK_ASSERT(ifp);
576 * A new in_multi record is needed; allocate and initialize it.
577 * We DO NOT perform an IGMP join as the in_ layer may need to
578 * push an initial source list down to IGMP to support SSM.
580 * The initial source filter state is INCLUDE, {} as per the RFC.
582 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
584 IF_ADDR_WUNLOCK(ifp);
585 IN_MULTI_LIST_UNLOCK();
586 if_delmulti_ifma(ifma);
589 inm->inm_addr = *group;
591 inm->inm_igi = ii->ii_igmp;
592 inm->inm_ifma = ifma;
593 inm->inm_refcount = 1;
594 inm->inm_state = IGMP_NOT_MEMBER;
595 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
596 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
597 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
598 RB_INIT(&inm->inm_srcs);
600 ifma->ifma_protospec = inm;
604 IF_ADDR_WUNLOCK(ifp);
605 IN_MULTI_LIST_UNLOCK();
610 * Drop a reference to an in_multi record.
612 * If the refcount drops to 0, free the in_multi record and
613 * delete the underlying link-layer membership.
616 inm_release(struct in_multi *inm)
618 struct ifmultiaddr *ifma;
621 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
622 MPASS(inm->inm_refcount == 0);
623 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
625 ifma = inm->inm_ifma;
628 /* XXX this access is not covered by IF_ADDR_LOCK */
629 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
631 CURVNET_SET(ifp->if_vnet);
633 free(inm, M_IPMADDR);
634 if_delmulti_ifma_flags(ifma, 1);
639 free(inm, M_IPMADDR);
640 if_delmulti_ifma_flags(ifma, 1);
645 * Clear recorded source entries for a group.
646 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
647 * FIXME: Should reap.
650 inm_clear_recorded(struct in_multi *inm)
652 struct ip_msource *ims;
654 IN_MULTI_LIST_LOCK_ASSERT();
656 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
659 --inm->inm_st[1].iss_rec;
662 KASSERT(inm->inm_st[1].iss_rec == 0,
663 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
667 * Record a source as pending for a Source-Group IGMPv3 query.
668 * This lives here as it modifies the shared tree.
670 * inm is the group descriptor.
671 * naddr is the address of the source to record in network-byte order.
673 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
674 * lazy-allocate a source node in response to an SG query.
675 * Otherwise, no allocation is performed. This saves some memory
676 * with the trade-off that the source will not be reported to the
677 * router if joined in the window between the query response and
678 * the group actually being joined on the local host.
680 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
681 * This turns off the allocation of a recorded source entry if
682 * the group has not been joined.
684 * Return 0 if the source didn't exist or was already marked as recorded.
685 * Return 1 if the source was marked as recorded by this function.
686 * Return <0 if any error occurred (negated errno code).
689 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
691 struct ip_msource find;
692 struct ip_msource *ims, *nims;
694 IN_MULTI_LIST_LOCK_ASSERT();
696 find.ims_haddr = ntohl(naddr);
697 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
698 if (ims && ims->ims_stp)
701 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
703 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
707 nims->ims_haddr = find.ims_haddr;
708 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
714 * Mark the source as recorded and update the recorded
718 ++inm->inm_st[1].iss_rec;
724 * Return a pointer to an in_msource owned by an in_mfilter,
725 * given its source address.
726 * Lazy-allocate if needed. If this is a new entry its filter state is
729 * imf is the filter set being modified.
730 * haddr is the source address in *host* byte-order.
732 * SMPng: May be called with locks held; malloc must not block.
735 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
736 struct in_msource **plims)
738 struct ip_msource find;
739 struct ip_msource *ims, *nims;
740 struct in_msource *lims;
747 /* key is host byte order */
748 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
749 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
750 lims = (struct in_msource *)ims;
752 if (imf->imf_nsrc == in_mcast_maxsocksrc)
754 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
758 lims = (struct in_msource *)nims;
759 lims->ims_haddr = find.ims_haddr;
760 lims->imsl_st[0] = MCAST_UNDEFINED;
761 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
771 * Graft a source entry into an existing socket-layer filter set,
772 * maintaining any required invariants and checking allocations.
774 * The source is marked as being in the new filter mode at t1.
776 * Return the pointer to the new node, otherwise return NULL.
778 static struct in_msource *
779 imf_graft(struct in_mfilter *imf, const uint8_t st1,
780 const struct sockaddr_in *psin)
782 struct ip_msource *nims;
783 struct in_msource *lims;
785 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
789 lims = (struct in_msource *)nims;
790 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
791 lims->imsl_st[0] = MCAST_UNDEFINED;
792 lims->imsl_st[1] = st1;
793 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
800 * Prune a source entry from an existing socket-layer filter set,
801 * maintaining any required invariants and checking allocations.
803 * The source is marked as being left at t1, it is not freed.
805 * Return 0 if no error occurred, otherwise return an errno value.
808 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
810 struct ip_msource find;
811 struct ip_msource *ims;
812 struct in_msource *lims;
814 /* key is host byte order */
815 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
816 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
819 lims = (struct in_msource *)ims;
820 lims->imsl_st[1] = MCAST_UNDEFINED;
825 * Revert socket-layer filter set deltas at t1 to t0 state.
828 imf_rollback(struct in_mfilter *imf)
830 struct ip_msource *ims, *tims;
831 struct in_msource *lims;
833 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
834 lims = (struct in_msource *)ims;
835 if (lims->imsl_st[0] == lims->imsl_st[1]) {
836 /* no change at t1 */
838 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
839 /* revert change to existing source at t1 */
840 lims->imsl_st[1] = lims->imsl_st[0];
842 /* revert source added t1 */
843 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
844 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
845 free(ims, M_INMFILTER);
849 imf->imf_st[1] = imf->imf_st[0];
853 * Mark socket-layer filter set as INCLUDE {} at t1.
856 imf_leave(struct in_mfilter *imf)
858 struct ip_msource *ims;
859 struct in_msource *lims;
861 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
862 lims = (struct in_msource *)ims;
863 lims->imsl_st[1] = MCAST_UNDEFINED;
865 imf->imf_st[1] = MCAST_INCLUDE;
869 * Mark socket-layer filter set deltas as committed.
872 imf_commit(struct in_mfilter *imf)
874 struct ip_msource *ims;
875 struct in_msource *lims;
877 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
878 lims = (struct in_msource *)ims;
879 lims->imsl_st[0] = lims->imsl_st[1];
881 imf->imf_st[0] = imf->imf_st[1];
885 * Reap unreferenced sources from socket-layer filter set.
888 imf_reap(struct in_mfilter *imf)
890 struct ip_msource *ims, *tims;
891 struct in_msource *lims;
893 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
894 lims = (struct in_msource *)ims;
895 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
896 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
897 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
898 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
899 free(ims, M_INMFILTER);
906 * Purge socket-layer filter set.
909 imf_purge(struct in_mfilter *imf)
911 struct ip_msource *ims, *tims;
913 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
914 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
915 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
916 free(ims, M_INMFILTER);
919 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
920 KASSERT(RB_EMPTY(&imf->imf_sources),
921 ("%s: imf_sources not empty", __func__));
925 * Look up a source filter entry for a multicast group.
927 * inm is the group descriptor to work with.
928 * haddr is the host-byte-order IPv4 address to look up.
929 * noalloc may be non-zero to suppress allocation of sources.
930 * *pims will be set to the address of the retrieved or allocated source.
932 * SMPng: NOTE: may be called with locks held.
933 * Return 0 if successful, otherwise return a non-zero error code.
936 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
937 const int noalloc, struct ip_msource **pims)
939 struct ip_msource find;
940 struct ip_msource *ims, *nims;
942 find.ims_haddr = haddr;
943 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
944 if (ims == NULL && !noalloc) {
945 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
947 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
951 nims->ims_haddr = haddr;
952 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
956 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
966 * Merge socket-layer source into IGMP-layer source.
967 * If rollback is non-zero, perform the inverse of the merge.
970 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
973 int n = rollback ? -1 : 1;
975 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
976 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
977 __func__, n, ims->ims_haddr);
978 ims->ims_st[1].ex -= n;
979 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
980 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
981 __func__, n, ims->ims_haddr);
982 ims->ims_st[1].in -= n;
985 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
986 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
987 __func__, n, ims->ims_haddr);
988 ims->ims_st[1].ex += n;
989 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
990 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
991 __func__, n, ims->ims_haddr);
992 ims->ims_st[1].in += n;
997 * Atomically update the global in_multi state, when a membership's
998 * filter list is being updated in any way.
1000 * imf is the per-inpcb-membership group filter pointer.
1001 * A fake imf may be passed for in-kernel consumers.
1003 * XXX This is a candidate for a set-symmetric-difference style loop
1004 * which would eliminate the repeated lookup from root of ims nodes,
1005 * as they share the same key space.
1007 * If any error occurred this function will back out of refcounts
1008 * and return a non-zero value.
1011 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1013 struct ip_msource *ims, *nims;
1014 struct in_msource *lims;
1015 int schanged, error;
1021 IN_MULTI_LIST_LOCK_ASSERT();
1024 * Update the source filters first, as this may fail.
1025 * Maintain count of in-mode filters at t0, t1. These are
1026 * used to work out if we transition into ASM mode or not.
1027 * Maintain a count of source filters whose state was
1028 * actually modified by this operation.
1030 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1031 lims = (struct in_msource *)ims;
1032 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1033 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1034 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1035 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1039 ims_merge(nims, lims, 0);
1042 struct ip_msource *bims;
1044 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1045 lims = (struct in_msource *)ims;
1046 if (lims->imsl_st[0] == lims->imsl_st[1])
1048 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1051 ims_merge(bims, lims, 1);
1056 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1057 __func__, nsrc0, nsrc1);
1059 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1060 if (imf->imf_st[0] == imf->imf_st[1] &&
1061 imf->imf_st[1] == MCAST_INCLUDE) {
1063 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1064 --inm->inm_st[1].iss_in;
1068 /* Handle filter mode transition on socket. */
1069 if (imf->imf_st[0] != imf->imf_st[1]) {
1070 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1071 __func__, imf->imf_st[0], imf->imf_st[1]);
1073 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1074 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1075 --inm->inm_st[1].iss_ex;
1076 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1077 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1078 --inm->inm_st[1].iss_in;
1081 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1082 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1083 inm->inm_st[1].iss_ex++;
1084 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1085 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1086 inm->inm_st[1].iss_in++;
1091 * Track inm filter state in terms of listener counts.
1092 * If there are any exclusive listeners, stack-wide
1093 * membership is exclusive.
1094 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1095 * If no listeners remain, state is undefined at t1,
1096 * and the IGMP lifecycle for this group should finish.
1098 if (inm->inm_st[1].iss_ex > 0) {
1099 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1100 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1101 } else if (inm->inm_st[1].iss_in > 0) {
1102 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1103 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1105 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1106 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1109 /* Decrement ASM listener count on transition out of ASM mode. */
1110 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1111 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1112 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1113 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1114 --inm->inm_st[1].iss_asm;
1118 /* Increment ASM listener count on transition to ASM mode. */
1119 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1120 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1121 inm->inm_st[1].iss_asm++;
1124 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1129 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1136 * Mark an in_multi's filter set deltas as committed.
1137 * Called by IGMP after a state change has been enqueued.
1140 inm_commit(struct in_multi *inm)
1142 struct ip_msource *ims;
1144 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1145 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1148 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1149 ims->ims_st[0] = ims->ims_st[1];
1151 inm->inm_st[0] = inm->inm_st[1];
1155 * Reap unreferenced nodes from an in_multi's filter set.
1158 inm_reap(struct in_multi *inm)
1160 struct ip_msource *ims, *tims;
1162 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1163 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1164 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1167 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1168 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1169 free(ims, M_IPMSOURCE);
1175 * Purge all source nodes from an in_multi's filter set.
1178 inm_purge(struct in_multi *inm)
1180 struct ip_msource *ims, *tims;
1182 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1183 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1184 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1185 free(ims, M_IPMSOURCE);
1191 * Join a multicast group; unlocked entry point.
1193 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1194 * is not held. Fortunately, ifp is unlikely to have been detached
1195 * at this point, so we assume it's OK to recurse.
1198 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1199 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1204 error = in_joingroup_locked(ifp, gina, imf, pinm);
1211 * Join a multicast group; real entry point.
1213 * Only preserves atomicity at inm level.
1214 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1216 * If the IGMP downcall fails, the group is not joined, and an error
1220 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1221 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1223 struct in_mfilter timf;
1224 struct in_multi *inm;
1227 IN_MULTI_LOCK_ASSERT();
1228 IN_MULTI_LIST_UNLOCK_ASSERT();
1230 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1231 ntohl(gina->s_addr), ifp, ifp->if_xname);
1237 * If no imf was specified (i.e. kernel consumer),
1238 * fake one up and assume it is an ASM join.
1241 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1245 error = in_getmulti(ifp, gina, &inm);
1247 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1250 IN_MULTI_LIST_LOCK();
1251 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1252 error = inm_merge(inm, imf);
1254 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1255 goto out_inm_release;
1258 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1259 error = igmp_change_state(inm);
1261 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1262 goto out_inm_release;
1268 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1269 inm_release_deferred(inm);
1273 IN_MULTI_LIST_UNLOCK();
1279 * Leave a multicast group; unlocked entry point.
1282 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1287 error = in_leavegroup_locked(inm, imf);
1294 * Leave a multicast group; real entry point.
1295 * All source filters will be expunged.
1297 * Only preserves atomicity at inm level.
1299 * Holding the write lock for the INP which contains imf
1300 * is highly advisable. We can't assert for it as imf does not
1301 * contain a back-pointer to the owning inp.
1303 * Note: This is not the same as inm_release(*) as this function also
1304 * makes a state change downcall into IGMP.
1307 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1309 struct in_mfilter timf;
1314 IN_MULTI_LOCK_ASSERT();
1315 IN_MULTI_LIST_UNLOCK_ASSERT();
1317 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1318 inm, ntohl(inm->inm_addr.s_addr),
1319 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1323 * If no imf was specified (i.e. kernel consumer),
1324 * fake one up and assume it is an ASM join.
1327 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1332 * Begin state merge transaction at IGMP layer.
1334 * As this particular invocation should not cause any memory
1335 * to be allocated, and there is no opportunity to roll back
1336 * the transaction, it MUST NOT fail.
1338 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1339 IN_MULTI_LIST_LOCK();
1340 error = inm_merge(inm, imf);
1341 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1343 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1344 CURVNET_SET(inm->inm_ifp->if_vnet);
1345 error = igmp_change_state(inm);
1346 IF_ADDR_WLOCK(inm->inm_ifp);
1347 inm_release_deferred(inm);
1348 IF_ADDR_WUNLOCK(inm->inm_ifp);
1349 IN_MULTI_LIST_UNLOCK();
1352 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1354 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1359 /*#ifndef BURN_BRIDGES*/
1361 * Join an IPv4 multicast group in (*,G) exclusive mode.
1362 * The group must be a 224.0.0.0/24 link-scope group.
1363 * This KPI is for legacy kernel consumers only.
1366 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1368 struct in_multi *pinm;
1371 char addrbuf[INET_ADDRSTRLEN];
1374 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1375 ("%s: %s not in 224.0.0.0/24", __func__,
1376 inet_ntoa_r(*ap, addrbuf)));
1378 error = in_joingroup(ifp, ap, NULL, &pinm);
1386 * Block or unblock an ASM multicast source on an inpcb.
1387 * This implements the delta-based API described in RFC 3678.
1389 * The delta-based API applies only to exclusive-mode memberships.
1390 * An IGMP downcall will be performed.
1392 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1394 * Return 0 if successful, otherwise return an appropriate error code.
1397 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1399 struct group_source_req gsr;
1400 struct rm_priotracker in_ifa_tracker;
1401 sockunion_t *gsa, *ssa;
1403 struct in_mfilter *imf;
1404 struct ip_moptions *imo;
1405 struct in_msource *ims;
1406 struct in_multi *inm;
1414 memset(&gsr, 0, sizeof(struct group_source_req));
1415 gsa = (sockunion_t *)&gsr.gsr_group;
1416 ssa = (sockunion_t *)&gsr.gsr_source;
1418 switch (sopt->sopt_name) {
1419 case IP_BLOCK_SOURCE:
1420 case IP_UNBLOCK_SOURCE: {
1421 struct ip_mreq_source mreqs;
1423 error = sooptcopyin(sopt, &mreqs,
1424 sizeof(struct ip_mreq_source),
1425 sizeof(struct ip_mreq_source));
1429 gsa->sin.sin_family = AF_INET;
1430 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1431 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1433 ssa->sin.sin_family = AF_INET;
1434 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1435 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1437 if (!in_nullhost(mreqs.imr_interface)) {
1438 IN_IFADDR_RLOCK(&in_ifa_tracker);
1439 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1440 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1442 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1445 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1446 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1450 case MCAST_BLOCK_SOURCE:
1451 case MCAST_UNBLOCK_SOURCE:
1452 error = sooptcopyin(sopt, &gsr,
1453 sizeof(struct group_source_req),
1454 sizeof(struct group_source_req));
1458 if (gsa->sin.sin_family != AF_INET ||
1459 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1462 if (ssa->sin.sin_family != AF_INET ||
1463 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1466 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1467 return (EADDRNOTAVAIL);
1469 ifp = ifnet_byindex(gsr.gsr_interface);
1471 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1476 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1477 __func__, sopt->sopt_name);
1478 return (EOPNOTSUPP);
1482 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1488 * Check if we are actually a member of this group.
1490 imo = inp_findmoptions(inp);
1491 imf = imo_match_group(imo, ifp, &gsa->sa);
1493 error = EADDRNOTAVAIL;
1494 goto out_inp_locked;
1499 * Attempting to use the delta-based API on an
1500 * non exclusive-mode membership is an error.
1502 fmode = imf->imf_st[0];
1503 if (fmode != MCAST_EXCLUDE) {
1505 goto out_inp_locked;
1509 * Deal with error cases up-front:
1510 * Asked to block, but already blocked; or
1511 * Asked to unblock, but nothing to unblock.
1512 * If adding a new block entry, allocate it.
1514 ims = imo_match_source(imf, &ssa->sa);
1515 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1516 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1517 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1518 error = EADDRNOTAVAIL;
1519 goto out_inp_locked;
1522 INP_WLOCK_ASSERT(inp);
1525 * Begin state merge transaction at socket layer.
1528 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1529 ims = imf_graft(imf, fmode, &ssa->sin);
1533 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1534 error = imf_prune(imf, &ssa->sin);
1538 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1539 goto out_imf_rollback;
1543 * Begin state merge transaction at IGMP layer.
1545 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1546 IN_MULTI_LIST_LOCK();
1547 error = inm_merge(inm, imf);
1549 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1550 IN_MULTI_LIST_UNLOCK();
1551 goto out_imf_rollback;
1554 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1555 error = igmp_change_state(inm);
1556 IN_MULTI_LIST_UNLOCK();
1558 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1575 * Given an inpcb, return its multicast options structure pointer. Accepts
1576 * an unlocked inpcb pointer, but will return it locked. May sleep.
1578 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1579 * SMPng: NOTE: Returns with the INP write lock held.
1581 static struct ip_moptions *
1582 inp_findmoptions(struct inpcb *inp)
1584 struct ip_moptions *imo;
1587 if (inp->inp_moptions != NULL)
1588 return (inp->inp_moptions);
1592 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1594 imo->imo_multicast_ifp = NULL;
1595 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1596 imo->imo_multicast_vif = -1;
1597 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1598 imo->imo_multicast_loop = in_mcast_loop;
1599 STAILQ_INIT(&imo->imo_head);
1602 if (inp->inp_moptions != NULL) {
1603 free(imo, M_IPMOPTS);
1604 return (inp->inp_moptions);
1606 inp->inp_moptions = imo;
1611 inp_gcmoptions(struct ip_moptions *imo)
1613 struct in_mfilter *imf;
1614 struct in_multi *inm;
1617 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1618 ip_mfilter_remove(&imo->imo_head, imf);
1621 if ((inm = imf->imf_inm) != NULL) {
1622 if ((ifp = inm->inm_ifp) != NULL) {
1623 CURVNET_SET(ifp->if_vnet);
1624 (void)in_leavegroup(inm, imf);
1627 (void)in_leavegroup(inm, imf);
1630 ip_mfilter_free(imf);
1632 free(imo, M_IPMOPTS);
1636 * Discard the IP multicast options (and source filters). To minimize
1637 * the amount of work done while holding locks such as the INP's
1638 * pcbinfo lock (which is used in the receive path), the free
1639 * operation is deferred to the epoch callback task.
1642 inp_freemoptions(struct ip_moptions *imo)
1646 inp_gcmoptions(imo);
1650 * Atomically get source filters on a socket for an IPv4 multicast group.
1651 * Called with INP lock held; returns with lock released.
1654 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1656 struct __msfilterreq msfr;
1659 struct ip_moptions *imo;
1660 struct in_mfilter *imf;
1661 struct ip_msource *ims;
1662 struct in_msource *lims;
1663 struct sockaddr_in *psin;
1664 struct sockaddr_storage *ptss;
1665 struct sockaddr_storage *tss;
1667 size_t nsrcs, ncsrcs;
1669 INP_WLOCK_ASSERT(inp);
1671 imo = inp->inp_moptions;
1672 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1676 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1677 sizeof(struct __msfilterreq));
1681 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1684 ifp = ifnet_byindex(msfr.msfr_ifindex);
1691 * Lookup group on the socket.
1693 gsa = (sockunion_t *)&msfr.msfr_group;
1694 imf = imo_match_group(imo, ifp, &gsa->sa);
1697 return (EADDRNOTAVAIL);
1701 * Ignore memberships which are in limbo.
1703 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1707 msfr.msfr_fmode = imf->imf_st[1];
1710 * If the user specified a buffer, copy out the source filter
1711 * entries to userland gracefully.
1712 * We only copy out the number of entries which userland
1713 * has asked for, but we always tell userland how big the
1714 * buffer really needs to be.
1716 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1717 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1719 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1720 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1721 M_TEMP, M_NOWAIT | M_ZERO);
1729 * Count number of sources in-mode at t0.
1730 * If buffer space exists and remains, copy out source entries.
1732 nsrcs = msfr.msfr_nsrcs;
1735 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1736 lims = (struct in_msource *)ims;
1737 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1738 lims->imsl_st[0] != imf->imf_st[0])
1741 if (tss != NULL && nsrcs > 0) {
1742 psin = (struct sockaddr_in *)ptss;
1743 psin->sin_family = AF_INET;
1744 psin->sin_len = sizeof(struct sockaddr_in);
1745 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1755 error = copyout(tss, msfr.msfr_srcs,
1756 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1762 msfr.msfr_nsrcs = ncsrcs;
1763 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1769 * Return the IP multicast options in response to user getsockopt().
1772 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1774 struct rm_priotracker in_ifa_tracker;
1775 struct ip_mreqn mreqn;
1776 struct ip_moptions *imo;
1778 struct in_ifaddr *ia;
1783 imo = inp->inp_moptions;
1785 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1786 * or is a divert socket, reject it.
1788 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1789 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1790 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1792 return (EOPNOTSUPP);
1796 switch (sopt->sopt_name) {
1797 case IP_MULTICAST_VIF:
1799 optval = imo->imo_multicast_vif;
1803 error = sooptcopyout(sopt, &optval, sizeof(int));
1806 case IP_MULTICAST_IF:
1807 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1809 ifp = imo->imo_multicast_ifp;
1810 if (!in_nullhost(imo->imo_multicast_addr)) {
1811 mreqn.imr_address = imo->imo_multicast_addr;
1812 } else if (ifp != NULL) {
1813 mreqn.imr_ifindex = ifp->if_index;
1815 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1818 IA_SIN(ia)->sin_addr;
1823 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1824 error = sooptcopyout(sopt, &mreqn,
1825 sizeof(struct ip_mreqn));
1827 error = sooptcopyout(sopt, &mreqn.imr_address,
1828 sizeof(struct in_addr));
1832 case IP_MULTICAST_TTL:
1834 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1836 optval = coptval = imo->imo_multicast_ttl;
1838 if (sopt->sopt_valsize == sizeof(u_char))
1839 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1841 error = sooptcopyout(sopt, &optval, sizeof(int));
1844 case IP_MULTICAST_LOOP:
1846 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1848 optval = coptval = imo->imo_multicast_loop;
1850 if (sopt->sopt_valsize == sizeof(u_char))
1851 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1853 error = sooptcopyout(sopt, &optval, sizeof(int));
1858 error = EADDRNOTAVAIL;
1861 error = inp_get_source_filters(inp, sopt);
1867 error = ENOPROTOOPT;
1871 INP_UNLOCK_ASSERT(inp);
1877 * Look up the ifnet to use for a multicast group membership,
1878 * given the IPv4 address of an interface, and the IPv4 group address.
1880 * This routine exists to support legacy multicast applications
1881 * which do not understand that multicast memberships are scoped to
1882 * specific physical links in the networking stack, or which need
1883 * to join link-scope groups before IPv4 addresses are configured.
1885 * If inp is non-NULL, use this socket's current FIB number for any
1886 * required FIB lookup.
1887 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1888 * and use its ifp; usually, this points to the default next-hop.
1890 * If the FIB lookup fails, attempt to use the first non-loopback
1891 * interface with multicast capability in the system as a
1892 * last resort. The legacy IPv4 ASM API requires that we do
1893 * this in order to allow groups to be joined when the routing
1894 * table has not yet been populated during boot.
1896 * Returns NULL if no ifp could be found.
1898 * FUTURE: Implement IPv4 source-address selection.
1900 static struct ifnet *
1901 inp_lookup_mcast_ifp(const struct inpcb *inp,
1902 const struct sockaddr_in *gsin, const struct in_addr ina)
1904 struct rm_priotracker in_ifa_tracker;
1906 struct nhop4_basic nh4;
1909 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1910 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1911 ("%s: not multicast", __func__));
1914 if (!in_nullhost(ina)) {
1915 IN_IFADDR_RLOCK(&in_ifa_tracker);
1916 INADDR_TO_IFP(ina, ifp);
1917 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1919 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1920 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1923 struct in_ifaddr *ia;
1927 IN_IFADDR_RLOCK(&in_ifa_tracker);
1928 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1930 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1931 (mifp->if_flags & IFF_MULTICAST)) {
1936 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1944 * Join an IPv4 multicast group, possibly with a source.
1947 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1949 struct group_source_req gsr;
1950 sockunion_t *gsa, *ssa;
1952 struct in_mfilter *imf;
1953 struct ip_moptions *imo;
1954 struct in_multi *inm;
1955 struct in_msource *lims;
1962 memset(&gsr, 0, sizeof(struct group_source_req));
1963 gsa = (sockunion_t *)&gsr.gsr_group;
1964 gsa->ss.ss_family = AF_UNSPEC;
1965 ssa = (sockunion_t *)&gsr.gsr_source;
1966 ssa->ss.ss_family = AF_UNSPEC;
1968 switch (sopt->sopt_name) {
1969 case IP_ADD_MEMBERSHIP: {
1970 struct ip_mreqn mreqn;
1972 if (sopt->sopt_valsize == sizeof(struct ip_mreqn))
1973 error = sooptcopyin(sopt, &mreqn,
1974 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn));
1976 error = sooptcopyin(sopt, &mreqn,
1977 sizeof(struct ip_mreq), sizeof(struct ip_mreq));
1981 gsa->sin.sin_family = AF_INET;
1982 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1983 gsa->sin.sin_addr = mreqn.imr_multiaddr;
1984 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1987 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) &&
1988 mreqn.imr_ifindex != 0)
1989 ifp = ifnet_byindex(mreqn.imr_ifindex);
1991 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1995 case IP_ADD_SOURCE_MEMBERSHIP: {
1996 struct ip_mreq_source mreqs;
1998 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source),
1999 sizeof(struct ip_mreq_source));
2003 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET;
2004 gsa->sin.sin_len = ssa->sin.sin_len =
2005 sizeof(struct sockaddr_in);
2007 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2008 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2011 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2013 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2014 mreqs.imr_interface);
2015 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2016 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2020 case MCAST_JOIN_GROUP:
2021 case MCAST_JOIN_SOURCE_GROUP:
2022 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2023 error = sooptcopyin(sopt, &gsr,
2024 sizeof(struct group_req),
2025 sizeof(struct group_req));
2026 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2027 error = sooptcopyin(sopt, &gsr,
2028 sizeof(struct group_source_req),
2029 sizeof(struct group_source_req));
2034 if (gsa->sin.sin_family != AF_INET ||
2035 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2039 * Overwrite the port field if present, as the sockaddr
2040 * being copied in may be matched with a binary comparison.
2042 gsa->sin.sin_port = 0;
2043 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2044 if (ssa->sin.sin_family != AF_INET ||
2045 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2047 ssa->sin.sin_port = 0;
2050 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2053 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2054 return (EADDRNOTAVAIL);
2055 ifp = ifnet_byindex(gsr.gsr_interface);
2059 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2060 __func__, sopt->sopt_name);
2061 return (EOPNOTSUPP);
2065 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2066 return (EADDRNOTAVAIL);
2071 * Find the membership in the membership list.
2073 imo = inp_findmoptions(inp);
2074 imf = imo_match_group(imo, ifp, &gsa->sa);
2079 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) {
2081 goto out_inp_locked;
2087 if (ssa->ss.ss_family != AF_UNSPEC) {
2089 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2090 * is an error. On an existing inclusive membership,
2091 * it just adds the source to the filter list.
2093 if (imf->imf_st[1] != MCAST_INCLUDE) {
2095 goto out_inp_locked;
2098 * Throw out duplicates.
2100 * XXX FIXME: This makes a naive assumption that
2101 * even if entries exist for *ssa in this imf,
2102 * they will be rejected as dupes, even if they
2103 * are not valid in the current mode (in-mode).
2105 * in_msource is transactioned just as for anything
2106 * else in SSM -- but note naive use of inm_graft()
2107 * below for allocating new filter entries.
2109 * This is only an issue if someone mixes the
2110 * full-state SSM API with the delta-based API,
2111 * which is discouraged in the relevant RFCs.
2113 lims = imo_match_source(imf, &ssa->sa);
2114 if (lims != NULL /*&&
2115 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2116 error = EADDRNOTAVAIL;
2117 goto out_inp_locked;
2121 * MCAST_JOIN_GROUP on an existing exclusive
2122 * membership is an error; return EADDRINUSE
2123 * to preserve 4.4BSD API idempotence, and
2124 * avoid tedious detour to code below.
2125 * NOTE: This is bending RFC 3678 a bit.
2127 * On an existing inclusive membership, this is also
2128 * an error; if you want to change filter mode,
2129 * you must use the userland API setsourcefilter().
2130 * XXX We don't reject this for imf in UNDEFINED
2131 * state at t1, because allocation of a filter
2132 * is atomic with allocation of a membership.
2135 if (imf->imf_st[1] == MCAST_EXCLUDE)
2137 goto out_inp_locked;
2142 * Begin state merge transaction at socket layer.
2144 INP_WLOCK_ASSERT(inp);
2147 * Graft new source into filter list for this inpcb's
2148 * membership of the group. The in_multi may not have
2149 * been allocated yet if this is a new membership, however,
2150 * the in_mfilter slot will be allocated and must be initialized.
2152 * Note: Grafting of exclusive mode filters doesn't happen
2154 * XXX: Should check for non-NULL lims (node exists but may
2155 * not be in-mode) for interop with full-state API.
2157 if (ssa->ss.ss_family != AF_UNSPEC) {
2158 /* Membership starts in IN mode */
2160 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2161 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2164 goto out_inp_locked;
2167 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2169 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2171 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2174 goto out_inp_locked;
2177 /* No address specified; Membership starts in EX mode */
2179 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2180 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2183 goto out_inp_locked;
2189 * Begin state merge transaction at IGMP layer.
2195 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2199 if (in_pcbrele_wlocked(inp)) {
2201 goto out_inp_unlocked;
2204 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2206 goto out_inp_locked;
2209 * NOTE: Refcount from in_joingroup_locked()
2210 * is protecting membership.
2212 ip_mfilter_insert(&imo->imo_head, imf);
2214 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2215 IN_MULTI_LIST_LOCK();
2216 error = inm_merge(inm, imf);
2218 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2220 IN_MULTI_LIST_UNLOCK();
2223 goto out_inp_locked;
2225 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2226 error = igmp_change_state(inm);
2227 IN_MULTI_LIST_UNLOCK();
2229 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2233 goto out_inp_locked;
2245 if (is_new && imf) {
2246 if (imf->imf_inm != NULL) {
2247 IN_MULTI_LIST_LOCK();
2248 inm_release_deferred(imf->imf_inm);
2249 IN_MULTI_LIST_UNLOCK();
2251 ip_mfilter_free(imf);
2257 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2260 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2262 struct group_source_req gsr;
2263 struct ip_mreq_source mreqs;
2264 struct rm_priotracker in_ifa_tracker;
2265 sockunion_t *gsa, *ssa;
2267 struct in_mfilter *imf;
2268 struct ip_moptions *imo;
2269 struct in_msource *ims;
2270 struct in_multi *inm;
2278 memset(&gsr, 0, sizeof(struct group_source_req));
2279 gsa = (sockunion_t *)&gsr.gsr_group;
2280 gsa->ss.ss_family = AF_UNSPEC;
2281 ssa = (sockunion_t *)&gsr.gsr_source;
2282 ssa->ss.ss_family = AF_UNSPEC;
2284 switch (sopt->sopt_name) {
2285 case IP_DROP_MEMBERSHIP:
2286 case IP_DROP_SOURCE_MEMBERSHIP:
2287 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2288 error = sooptcopyin(sopt, &mreqs,
2289 sizeof(struct ip_mreq),
2290 sizeof(struct ip_mreq));
2292 * Swap interface and sourceaddr arguments,
2293 * as ip_mreq and ip_mreq_source are laid
2296 mreqs.imr_interface = mreqs.imr_sourceaddr;
2297 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2298 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2299 error = sooptcopyin(sopt, &mreqs,
2300 sizeof(struct ip_mreq_source),
2301 sizeof(struct ip_mreq_source));
2306 gsa->sin.sin_family = AF_INET;
2307 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2308 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2310 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2311 ssa->sin.sin_family = AF_INET;
2312 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2313 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2317 * Attempt to look up hinted ifp from interface address.
2318 * Fallthrough with null ifp iff lookup fails, to
2319 * preserve 4.4BSD mcast API idempotence.
2320 * XXX NOTE WELL: The RFC 3678 API is preferred because
2321 * using an IPv4 address as a key is racy.
2323 if (!in_nullhost(mreqs.imr_interface)) {
2324 IN_IFADDR_RLOCK(&in_ifa_tracker);
2325 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2326 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2328 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2329 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2333 case MCAST_LEAVE_GROUP:
2334 case MCAST_LEAVE_SOURCE_GROUP:
2335 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2336 error = sooptcopyin(sopt, &gsr,
2337 sizeof(struct group_req),
2338 sizeof(struct group_req));
2339 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2340 error = sooptcopyin(sopt, &gsr,
2341 sizeof(struct group_source_req),
2342 sizeof(struct group_source_req));
2347 if (gsa->sin.sin_family != AF_INET ||
2348 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2351 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2352 if (ssa->sin.sin_family != AF_INET ||
2353 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2357 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2358 return (EADDRNOTAVAIL);
2360 ifp = ifnet_byindex(gsr.gsr_interface);
2363 return (EADDRNOTAVAIL);
2367 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2368 __func__, sopt->sopt_name);
2369 return (EOPNOTSUPP);
2373 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2379 * Find the membership in the membership list.
2381 imo = inp_findmoptions(inp);
2382 imf = imo_match_group(imo, ifp, &gsa->sa);
2384 error = EADDRNOTAVAIL;
2385 goto out_inp_locked;
2389 if (ssa->ss.ss_family != AF_UNSPEC)
2393 * Begin state merge transaction at socket layer.
2395 INP_WLOCK_ASSERT(inp);
2398 * If we were instructed only to leave a given source, do so.
2399 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2402 ip_mfilter_remove(&imo->imo_head, imf);
2406 * Give up the multicast address record to which
2407 * the membership points.
2409 (void) in_leavegroup_locked(imf->imf_inm, imf);
2411 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2412 error = EADDRNOTAVAIL;
2413 goto out_inp_locked;
2415 ims = imo_match_source(imf, &ssa->sa);
2417 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2418 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2419 error = EADDRNOTAVAIL;
2420 goto out_inp_locked;
2422 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2423 error = imf_prune(imf, &ssa->sin);
2425 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2427 goto out_inp_locked;
2432 * Begin state merge transaction at IGMP layer.
2435 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2436 IN_MULTI_LIST_LOCK();
2437 error = inm_merge(inm, imf);
2439 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2441 IN_MULTI_LIST_UNLOCK();
2444 goto out_inp_locked;
2447 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2448 error = igmp_change_state(inm);
2449 IN_MULTI_LIST_UNLOCK();
2451 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2455 goto out_inp_locked;
2464 if (is_final && imf)
2465 ip_mfilter_free(imf);
2472 * Select the interface for transmitting IPv4 multicast datagrams.
2474 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2475 * may be passed to this socket option. An address of INADDR_ANY or an
2476 * interface index of 0 is used to remove a previous selection.
2477 * When no interface is selected, one is chosen for every send.
2480 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2482 struct rm_priotracker in_ifa_tracker;
2483 struct in_addr addr;
2484 struct ip_mreqn mreqn;
2486 struct ip_moptions *imo;
2489 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2491 * An interface index was specified using the
2492 * Linux-derived ip_mreqn structure.
2494 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2495 sizeof(struct ip_mreqn));
2499 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2502 if (mreqn.imr_ifindex == 0) {
2505 ifp = ifnet_byindex(mreqn.imr_ifindex);
2507 return (EADDRNOTAVAIL);
2511 * An interface was specified by IPv4 address.
2512 * This is the traditional BSD usage.
2514 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2515 sizeof(struct in_addr));
2518 if (in_nullhost(addr)) {
2521 IN_IFADDR_RLOCK(&in_ifa_tracker);
2522 INADDR_TO_IFP(addr, ifp);
2523 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2525 return (EADDRNOTAVAIL);
2527 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2528 ntohl(addr.s_addr));
2531 /* Reject interfaces which do not support multicast. */
2532 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2533 return (EOPNOTSUPP);
2535 imo = inp_findmoptions(inp);
2536 imo->imo_multicast_ifp = ifp;
2537 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2544 * Atomically set source filters on a socket for an IPv4 multicast group.
2546 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2549 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2551 struct __msfilterreq msfr;
2554 struct in_mfilter *imf;
2555 struct ip_moptions *imo;
2556 struct in_multi *inm;
2559 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2560 sizeof(struct __msfilterreq));
2564 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2567 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2568 msfr.msfr_fmode != MCAST_INCLUDE))
2571 if (msfr.msfr_group.ss_family != AF_INET ||
2572 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2575 gsa = (sockunion_t *)&msfr.msfr_group;
2576 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2579 gsa->sin.sin_port = 0; /* ignore port */
2581 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2582 return (EADDRNOTAVAIL);
2584 ifp = ifnet_byindex(msfr.msfr_ifindex);
2586 return (EADDRNOTAVAIL);
2591 * Take the INP write lock.
2592 * Check if this socket is a member of this group.
2594 imo = inp_findmoptions(inp);
2595 imf = imo_match_group(imo, ifp, &gsa->sa);
2597 error = EADDRNOTAVAIL;
2598 goto out_inp_locked;
2603 * Begin state merge transaction at socket layer.
2605 INP_WLOCK_ASSERT(inp);
2607 imf->imf_st[1] = msfr.msfr_fmode;
2610 * Apply any new source filters, if present.
2611 * Make a copy of the user-space source vector so
2612 * that we may copy them with a single copyin. This
2613 * allows us to deal with page faults up-front.
2615 if (msfr.msfr_nsrcs > 0) {
2616 struct in_msource *lims;
2617 struct sockaddr_in *psin;
2618 struct sockaddr_storage *kss, *pkss;
2623 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2624 __func__, (unsigned long)msfr.msfr_nsrcs);
2625 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2627 error = copyin(msfr.msfr_srcs, kss,
2628 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2637 * Mark all source filters as UNDEFINED at t1.
2638 * Restore new group filter mode, as imf_leave()
2639 * will set it to INCLUDE.
2642 imf->imf_st[1] = msfr.msfr_fmode;
2645 * Update socket layer filters at t1, lazy-allocating
2646 * new entries. This saves a bunch of memory at the
2647 * cost of one RB_FIND() per source entry; duplicate
2648 * entries in the msfr_nsrcs vector are ignored.
2649 * If we encounter an error, rollback transaction.
2651 * XXX This too could be replaced with a set-symmetric
2652 * difference like loop to avoid walking from root
2653 * every time, as the key space is common.
2655 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2656 psin = (struct sockaddr_in *)pkss;
2657 if (psin->sin_family != AF_INET) {
2658 error = EAFNOSUPPORT;
2661 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2665 error = imf_get_source(imf, psin, &lims);
2668 lims->imsl_st[1] = imf->imf_st[1];
2674 goto out_imf_rollback;
2676 INP_WLOCK_ASSERT(inp);
2679 * Begin state merge transaction at IGMP layer.
2681 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2682 IN_MULTI_LIST_LOCK();
2683 error = inm_merge(inm, imf);
2685 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2686 IN_MULTI_LIST_UNLOCK();
2687 goto out_imf_rollback;
2690 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2691 error = igmp_change_state(inm);
2692 IN_MULTI_LIST_UNLOCK();
2694 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2711 * Set the IP multicast options in response to user setsockopt().
2713 * Many of the socket options handled in this function duplicate the
2714 * functionality of socket options in the regular unicast API. However,
2715 * it is not possible to merge the duplicate code, because the idempotence
2716 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2717 * the effects of these options must be treated as separate and distinct.
2719 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2720 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2721 * is refactored to no longer use vifs.
2724 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2726 struct ip_moptions *imo;
2732 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2733 * or is a divert socket, reject it.
2735 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2736 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2737 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2738 return (EOPNOTSUPP);
2740 switch (sopt->sopt_name) {
2741 case IP_MULTICAST_VIF: {
2744 * Select a multicast VIF for transmission.
2745 * Only useful if multicast forwarding is active.
2747 if (legal_vif_num == NULL) {
2751 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2754 if (!legal_vif_num(vifi) && (vifi != -1)) {
2758 imo = inp_findmoptions(inp);
2759 imo->imo_multicast_vif = vifi;
2764 case IP_MULTICAST_IF:
2765 error = inp_set_multicast_if(inp, sopt);
2768 case IP_MULTICAST_TTL: {
2772 * Set the IP time-to-live for outgoing multicast packets.
2773 * The original multicast API required a char argument,
2774 * which is inconsistent with the rest of the socket API.
2775 * We allow either a char or an int.
2777 if (sopt->sopt_valsize == sizeof(u_char)) {
2778 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2785 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2795 imo = inp_findmoptions(inp);
2796 imo->imo_multicast_ttl = ttl;
2801 case IP_MULTICAST_LOOP: {
2805 * Set the loopback flag for outgoing multicast packets.
2806 * Must be zero or one. The original multicast API required a
2807 * char argument, which is inconsistent with the rest
2808 * of the socket API. We allow either a char or an int.
2810 if (sopt->sopt_valsize == sizeof(u_char)) {
2811 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2818 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2822 loop = (u_char)iloop;
2824 imo = inp_findmoptions(inp);
2825 imo->imo_multicast_loop = !!loop;
2830 case IP_ADD_MEMBERSHIP:
2831 case IP_ADD_SOURCE_MEMBERSHIP:
2832 case MCAST_JOIN_GROUP:
2833 case MCAST_JOIN_SOURCE_GROUP:
2834 error = inp_join_group(inp, sopt);
2837 case IP_DROP_MEMBERSHIP:
2838 case IP_DROP_SOURCE_MEMBERSHIP:
2839 case MCAST_LEAVE_GROUP:
2840 case MCAST_LEAVE_SOURCE_GROUP:
2841 error = inp_leave_group(inp, sopt);
2844 case IP_BLOCK_SOURCE:
2845 case IP_UNBLOCK_SOURCE:
2846 case MCAST_BLOCK_SOURCE:
2847 case MCAST_UNBLOCK_SOURCE:
2848 error = inp_block_unblock_source(inp, sopt);
2852 error = inp_set_source_filters(inp, sopt);
2860 INP_UNLOCK_ASSERT(inp);
2866 * Expose IGMP's multicast filter mode and source list(s) to userland,
2867 * keyed by (ifindex, group).
2868 * The filter mode is written out as a uint32_t, followed by
2869 * 0..n of struct in_addr.
2870 * For use by ifmcstat(8).
2871 * SMPng: NOTE: unlocked read of ifindex space.
2874 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2876 struct in_addr src, group;
2878 struct ifmultiaddr *ifma;
2879 struct in_multi *inm;
2880 struct ip_msource *ims;
2884 uint32_t fmode, ifindex;
2889 if (req->newptr != NULL)
2896 if (ifindex <= 0 || ifindex > V_if_index) {
2897 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2902 group.s_addr = name[1];
2903 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2904 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2905 __func__, ntohl(group.s_addr));
2909 ifp = ifnet_byindex(ifindex);
2911 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2916 retval = sysctl_wire_old_buffer(req,
2917 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2921 IN_MULTI_LIST_LOCK();
2924 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2925 if (ifma->ifma_addr->sa_family != AF_INET ||
2926 ifma->ifma_protospec == NULL)
2928 inm = (struct in_multi *)ifma->ifma_protospec;
2929 if (!in_hosteq(inm->inm_addr, group))
2931 fmode = inm->inm_st[1].iss_fmode;
2932 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2935 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2936 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2939 * Only copy-out sources which are in-mode.
2941 if (fmode != ims_get_mode(inm, ims, 1)) {
2942 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2946 src.s_addr = htonl(ims->ims_haddr);
2947 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2952 IF_ADDR_RUNLOCK(ifp);
2954 IN_MULTI_LIST_UNLOCK();
2959 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2961 static const char *inm_modestrs[] = {
2962 [MCAST_UNDEFINED] = "un",
2963 [MCAST_INCLUDE] = "in",
2964 [MCAST_EXCLUDE] = "ex",
2966 _Static_assert(MCAST_UNDEFINED == 0 &&
2967 MCAST_EXCLUDE + 1 == nitems(inm_modestrs),
2968 "inm_modestrs: no longer matches #defines");
2971 inm_mode_str(const int mode)
2974 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2975 return (inm_modestrs[mode]);
2979 static const char *inm_statestrs[] = {
2980 [IGMP_NOT_MEMBER] = "not-member",
2981 [IGMP_SILENT_MEMBER] = "silent",
2982 [IGMP_REPORTING_MEMBER] = "reporting",
2983 [IGMP_IDLE_MEMBER] = "idle",
2984 [IGMP_LAZY_MEMBER] = "lazy",
2985 [IGMP_SLEEPING_MEMBER] = "sleeping",
2986 [IGMP_AWAKENING_MEMBER] = "awakening",
2987 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending",
2988 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending",
2989 [IGMP_LEAVING_MEMBER] = "leaving",
2991 _Static_assert(IGMP_NOT_MEMBER == 0 &&
2992 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs),
2993 "inm_statetrs: no longer matches #defines");
2996 inm_state_str(const int state)
2999 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3000 return (inm_statestrs[state]);
3005 * Dump an in_multi structure to the console.
3008 inm_print(const struct in_multi *inm)
3011 char addrbuf[INET_ADDRSTRLEN];
3013 if ((ktr_mask & KTR_IGMPV3) == 0)
3016 printf("%s: --- begin inm %p ---\n", __func__, inm);
3017 printf("addr %s ifp %p(%s) ifma %p\n",
3018 inet_ntoa_r(inm->inm_addr, addrbuf),
3020 inm->inm_ifp->if_xname,
3022 printf("timer %u state %s refcount %u scq.len %u\n",
3024 inm_state_str(inm->inm_state),
3026 inm->inm_scq.mq_len);
3027 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3032 for (t = 0; t < 2; t++) {
3033 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3034 inm_mode_str(inm->inm_st[t].iss_fmode),
3035 inm->inm_st[t].iss_asm,
3036 inm->inm_st[t].iss_ex,
3037 inm->inm_st[t].iss_in,
3038 inm->inm_st[t].iss_rec);
3040 printf("%s: --- end inm %p ---\n", __func__, inm);
3043 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3046 inm_print(const struct in_multi *inm)
3051 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3053 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);