2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2007-2009 Bruce Simpson.
5 * Copyright (c) 2005 Robert N. M. Watson.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote
17 * products derived from this software without specific prior written
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * IPv4 multicast socket, group, and socket option processing module.
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
44 #include <sys/malloc.h>
46 #include <sys/protosw.h>
47 #include <sys/rmlock.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/protosw.h>
51 #include <sys/sysctl.h>
53 #include <sys/taskqueue.h>
54 #include <sys/gtaskqueue.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
63 #include <net/ethernet.h>
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_fib.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/igmp_var.h>
74 #define KTR_IGMPV3 KTR_INET
77 #ifndef __SOCKUNION_DECLARED
79 struct sockaddr_storage ss;
81 struct sockaddr_dl sdl;
82 struct sockaddr_in sin;
84 typedef union sockunion sockunion_t;
85 #define __SOCKUNION_DECLARED
86 #endif /* __SOCKUNION_DECLARED */
88 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
89 "IPv4 multicast PCB-layer source filter");
90 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
91 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
92 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
93 "IPv4 multicast IGMP-layer source filter");
97 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
98 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
99 * it can be taken by code in net/if.c also.
100 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
102 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
103 * any need for in_multi itself to be virtualized -- it is bound to an ifp
104 * anyway no matter what happens.
106 struct mtx in_multi_list_mtx;
107 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
109 struct mtx in_multi_free_mtx;
110 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
112 struct sx in_multi_sx;
113 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
118 * Functions with non-static linkage defined in this file should be
119 * declared in in_var.h:
124 * in_joingroup_locked()
126 * in_leavegroup_locked()
132 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
135 static void imf_commit(struct in_mfilter *);
136 static int imf_get_source(struct in_mfilter *imf,
137 const struct sockaddr_in *psin,
138 struct in_msource **);
139 static struct in_msource *
140 imf_graft(struct in_mfilter *, const uint8_t,
141 const struct sockaddr_in *);
142 static void imf_leave(struct in_mfilter *);
143 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
144 static void imf_purge(struct in_mfilter *);
145 static void imf_rollback(struct in_mfilter *);
146 static void imf_reap(struct in_mfilter *);
147 static int imo_grow(struct ip_moptions *);
148 static size_t imo_match_group(const struct ip_moptions *,
149 const struct ifnet *, const struct sockaddr *);
150 static struct in_msource *
151 imo_match_source(const struct ip_moptions *, const size_t,
152 const struct sockaddr *);
153 static void ims_merge(struct ip_msource *ims,
154 const struct in_msource *lims, const int rollback);
155 static int in_getmulti(struct ifnet *, const struct in_addr *,
157 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
158 const int noalloc, struct ip_msource **pims);
160 static int inm_is_ifp_detached(const struct in_multi *);
162 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
163 static void inm_purge(struct in_multi *);
164 static void inm_reap(struct in_multi *);
165 static void inm_release(struct in_multi *);
166 static struct ip_moptions *
167 inp_findmoptions(struct inpcb *);
168 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
169 static int inp_join_group(struct inpcb *, struct sockopt *);
170 static int inp_leave_group(struct inpcb *, struct sockopt *);
171 static struct ifnet *
172 inp_lookup_mcast_ifp(const struct inpcb *,
173 const struct sockaddr_in *, const struct in_addr);
174 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
175 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
176 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
177 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
179 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
182 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
183 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
184 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
185 "Max source filters per group");
187 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
188 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
189 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
190 "Max source filters per socket");
192 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
193 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
194 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
196 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
197 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
198 "Per-interface stack-wide source filters");
202 * Inline function which wraps assertions for a valid ifp.
203 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
207 inm_is_ifp_detached(const struct in_multi *inm)
211 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
212 ifp = inm->inm_ifma->ifma_ifp;
215 * Sanity check that netinet's notion of ifp is the
218 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
221 return (ifp == NULL);
225 static struct grouptask free_gtask;
226 static struct in_multi_head inm_free_list;
227 static void inm_release_task(void *arg __unused);
228 static void inm_init(void)
230 SLIST_INIT(&inm_free_list);
231 taskqgroup_config_gtask_init(NULL, &free_gtask, inm_release_task, "inm release task");
234 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
239 inm_release_list_deferred(struct in_multi_head *inmh)
242 if (SLIST_EMPTY(inmh))
244 mtx_lock(&in_multi_free_mtx);
245 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
246 mtx_unlock(&in_multi_free_mtx);
247 GROUPTASK_ENQUEUE(&free_gtask);
251 inm_disconnect(struct in_multi *inm)
254 struct ifmultiaddr *ifma, *ll_ifma;
257 IF_ADDR_WLOCK_ASSERT(ifp);
258 ifma = inm->inm_ifma;
261 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
262 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
263 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
264 MPASS(ifma != ll_ifma);
265 ifma->ifma_llifma = NULL;
266 MPASS(ll_ifma->ifma_llifma == NULL);
267 MPASS(ll_ifma->ifma_ifp == ifp);
268 if (--ll_ifma->ifma_refcount == 0) {
269 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
270 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
271 if_freemulti(ll_ifma);
278 inm_release_deferred(struct in_multi *inm)
280 struct in_multi_head tmp;
282 IN_MULTI_LIST_LOCK_ASSERT();
283 MPASS(inm->inm_refcount > 0);
284 if (--inm->inm_refcount == 0) {
287 inm->inm_ifma->ifma_protospec = NULL;
288 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
289 inm_release_list_deferred(&tmp);
294 inm_release_task(void *arg __unused)
296 struct in_multi_head inm_free_tmp;
297 struct in_multi *inm, *tinm;
299 SLIST_INIT(&inm_free_tmp);
300 mtx_lock(&in_multi_free_mtx);
301 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
302 mtx_unlock(&in_multi_free_mtx);
304 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
305 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
313 * Initialize an in_mfilter structure to a known state at t0, t1
314 * with an empty source filter list.
317 imf_init(struct in_mfilter *imf, const int st0, const int st1)
319 memset(imf, 0, sizeof(struct in_mfilter));
320 RB_INIT(&imf->imf_sources);
321 imf->imf_st[0] = st0;
322 imf->imf_st[1] = st1;
326 * Function for looking up an in_multi record for an IPv4 multicast address
327 * on a given interface. ifp must be valid. If no record found, return NULL.
328 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
331 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
333 struct ifmultiaddr *ifma;
334 struct in_multi *inm;
336 IN_MULTI_LIST_LOCK_ASSERT();
337 IF_ADDR_LOCK_ASSERT(ifp);
340 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
341 if (ifma->ifma_addr->sa_family != AF_INET ||
342 ifma->ifma_protospec == NULL)
344 inm = (struct in_multi *)ifma->ifma_protospec;
345 if (inm->inm_addr.s_addr == ina.s_addr)
353 * Wrapper for inm_lookup_locked().
354 * The IF_ADDR_LOCK will be taken on ifp and released on return.
357 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
359 struct in_multi *inm;
361 IN_MULTI_LIST_LOCK_ASSERT();
363 inm = inm_lookup_locked(ifp, ina);
364 IF_ADDR_RUNLOCK(ifp);
370 * Resize the ip_moptions vector to the next power-of-two minus 1.
371 * May be called with locks held; do not sleep.
374 imo_grow(struct ip_moptions *imo)
376 struct in_multi **nmships;
377 struct in_multi **omships;
378 struct in_mfilter *nmfilters;
379 struct in_mfilter *omfilters;
386 omships = imo->imo_membership;
387 omfilters = imo->imo_mfilters;
388 oldmax = imo->imo_max_memberships;
389 newmax = ((oldmax + 1) * 2) - 1;
391 if (newmax <= IP_MAX_MEMBERSHIPS) {
392 nmships = (struct in_multi **)realloc(omships,
393 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
394 nmfilters = (struct in_mfilter *)realloc(omfilters,
395 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
396 if (nmships != NULL && nmfilters != NULL) {
397 /* Initialize newly allocated source filter heads. */
398 for (idx = oldmax; idx < newmax; idx++) {
399 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
402 imo->imo_max_memberships = newmax;
403 imo->imo_membership = nmships;
404 imo->imo_mfilters = nmfilters;
408 if (nmships == NULL || nmfilters == NULL) {
410 free(nmships, M_IPMOPTS);
411 if (nmfilters != NULL)
412 free(nmfilters, M_INMFILTER);
413 return (ETOOMANYREFS);
420 * Find an IPv4 multicast group entry for this ip_moptions instance
421 * which matches the specified group, and optionally an interface.
422 * Return its index into the array, or -1 if not found.
425 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
426 const struct sockaddr *group)
428 const struct sockaddr_in *gsin;
429 struct in_multi **pinm;
433 gsin = (const struct sockaddr_in *)group;
435 /* The imo_membership array may be lazy allocated. */
436 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
439 nmships = imo->imo_num_memberships;
440 pinm = &imo->imo_membership[0];
441 for (idx = 0; idx < nmships; idx++, pinm++) {
444 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
445 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
456 * Find an IPv4 multicast source entry for this imo which matches
457 * the given group index for this socket, and source address.
459 * NOTE: This does not check if the entry is in-mode, merely if
460 * it exists, which may not be the desired behaviour.
462 static struct in_msource *
463 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
464 const struct sockaddr *src)
466 struct ip_msource find;
467 struct in_mfilter *imf;
468 struct ip_msource *ims;
469 const sockunion_t *psa;
471 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
472 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
473 ("%s: invalid index %d\n", __func__, (int)gidx));
475 /* The imo_mfilters array may be lazy allocated. */
476 if (imo->imo_mfilters == NULL)
478 imf = &imo->imo_mfilters[gidx];
480 /* Source trees are keyed in host byte order. */
481 psa = (const sockunion_t *)src;
482 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
483 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
485 return ((struct in_msource *)ims);
489 * Perform filtering for multicast datagrams on a socket by group and source.
491 * Returns 0 if a datagram should be allowed through, or various error codes
492 * if the socket was not a member of the group, or the source was muted, etc.
495 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
496 const struct sockaddr *group, const struct sockaddr *src)
499 struct in_msource *ims;
502 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
504 gidx = imo_match_group(imo, ifp, group);
506 return (MCAST_NOTGMEMBER);
509 * Check if the source was included in an (S,G) join.
510 * Allow reception on exclusive memberships by default,
511 * reject reception on inclusive memberships by default.
512 * Exclude source only if an in-mode exclude filter exists.
513 * Include source only if an in-mode include filter exists.
514 * NOTE: We are comparing group state here at IGMP t1 (now)
515 * with socket-layer t0 (since last downcall).
517 mode = imo->imo_mfilters[gidx].imf_st[1];
518 ims = imo_match_source(imo, gidx, src);
520 if ((ims == NULL && mode == MCAST_INCLUDE) ||
521 (ims != NULL && ims->imsl_st[0] != mode))
522 return (MCAST_NOTSMEMBER);
528 * Find and return a reference to an in_multi record for (ifp, group),
529 * and bump its reference count.
530 * If one does not exist, try to allocate it, and update link-layer multicast
531 * filters on ifp to listen for group.
532 * Assumes the IN_MULTI lock is held across the call.
533 * Return 0 if successful, otherwise return an appropriate error code.
536 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
537 struct in_multi **pinm)
539 struct sockaddr_in gsin;
540 struct ifmultiaddr *ifma;
541 struct in_ifinfo *ii;
542 struct in_multi *inm;
545 IN_MULTI_LOCK_ASSERT();
547 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
548 IN_MULTI_LIST_LOCK();
549 inm = inm_lookup(ifp, *group);
552 * If we already joined this group, just bump the
553 * refcount and return it.
555 KASSERT(inm->inm_refcount >= 1,
556 ("%s: bad refcount %d", __func__, inm->inm_refcount));
557 inm_acquire_locked(inm);
560 IN_MULTI_LIST_UNLOCK();
564 memset(&gsin, 0, sizeof(gsin));
565 gsin.sin_family = AF_INET;
566 gsin.sin_len = sizeof(struct sockaddr_in);
567 gsin.sin_addr = *group;
570 * Check if a link-layer group is already associated
571 * with this network-layer group on the given ifnet.
573 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
577 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
578 IN_MULTI_LIST_LOCK();
582 * If something other than netinet is occupying the link-layer
583 * group, print a meaningful error message and back out of
585 * Otherwise, bump the refcount on the existing network-layer
586 * group association and return it.
588 if (ifma->ifma_protospec != NULL) {
589 inm = (struct in_multi *)ifma->ifma_protospec;
591 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
593 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
594 ("%s: ifma not AF_INET", __func__));
595 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
596 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
597 !in_hosteq(inm->inm_addr, *group)) {
598 char addrbuf[INET_ADDRSTRLEN];
600 panic("%s: ifma %p is inconsistent with %p (%s)",
601 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
604 inm_acquire_locked(inm);
609 IF_ADDR_WLOCK_ASSERT(ifp);
612 * A new in_multi record is needed; allocate and initialize it.
613 * We DO NOT perform an IGMP join as the in_ layer may need to
614 * push an initial source list down to IGMP to support SSM.
616 * The initial source filter state is INCLUDE, {} as per the RFC.
618 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
620 IF_ADDR_WUNLOCK(ifp);
621 IN_MULTI_LIST_UNLOCK();
622 if_delmulti_ifma(ifma);
625 inm->inm_addr = *group;
627 inm->inm_igi = ii->ii_igmp;
628 inm->inm_ifma = ifma;
629 inm->inm_refcount = 1;
630 inm->inm_state = IGMP_NOT_MEMBER;
631 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
632 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
633 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
634 RB_INIT(&inm->inm_srcs);
636 ifma->ifma_protospec = inm;
640 IF_ADDR_WUNLOCK(ifp);
641 IN_MULTI_LIST_UNLOCK();
646 * Drop a reference to an in_multi record.
648 * If the refcount drops to 0, free the in_multi record and
649 * delete the underlying link-layer membership.
652 inm_release(struct in_multi *inm)
654 struct ifmultiaddr *ifma;
657 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
658 MPASS(inm->inm_refcount == 0);
659 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
661 ifma = inm->inm_ifma;
664 /* XXX this access is not covered by IF_ADDR_LOCK */
665 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
667 CURVNET_SET(ifp->if_vnet);
669 free(inm, M_IPMADDR);
671 if_delmulti_ifma_flags(ifma, 1);
679 * Clear recorded source entries for a group.
680 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
681 * FIXME: Should reap.
684 inm_clear_recorded(struct in_multi *inm)
686 struct ip_msource *ims;
688 IN_MULTI_LIST_LOCK_ASSERT();
690 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
693 --inm->inm_st[1].iss_rec;
696 KASSERT(inm->inm_st[1].iss_rec == 0,
697 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
701 * Record a source as pending for a Source-Group IGMPv3 query.
702 * This lives here as it modifies the shared tree.
704 * inm is the group descriptor.
705 * naddr is the address of the source to record in network-byte order.
707 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
708 * lazy-allocate a source node in response to an SG query.
709 * Otherwise, no allocation is performed. This saves some memory
710 * with the trade-off that the source will not be reported to the
711 * router if joined in the window between the query response and
712 * the group actually being joined on the local host.
714 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
715 * This turns off the allocation of a recorded source entry if
716 * the group has not been joined.
718 * Return 0 if the source didn't exist or was already marked as recorded.
719 * Return 1 if the source was marked as recorded by this function.
720 * Return <0 if any error occurred (negated errno code).
723 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
725 struct ip_msource find;
726 struct ip_msource *ims, *nims;
728 IN_MULTI_LIST_LOCK_ASSERT();
730 find.ims_haddr = ntohl(naddr);
731 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
732 if (ims && ims->ims_stp)
735 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
737 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
741 nims->ims_haddr = find.ims_haddr;
742 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
748 * Mark the source as recorded and update the recorded
752 ++inm->inm_st[1].iss_rec;
758 * Return a pointer to an in_msource owned by an in_mfilter,
759 * given its source address.
760 * Lazy-allocate if needed. If this is a new entry its filter state is
763 * imf is the filter set being modified.
764 * haddr is the source address in *host* byte-order.
766 * SMPng: May be called with locks held; malloc must not block.
769 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
770 struct in_msource **plims)
772 struct ip_msource find;
773 struct ip_msource *ims, *nims;
774 struct in_msource *lims;
781 /* key is host byte order */
782 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
783 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
784 lims = (struct in_msource *)ims;
786 if (imf->imf_nsrc == in_mcast_maxsocksrc)
788 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
792 lims = (struct in_msource *)nims;
793 lims->ims_haddr = find.ims_haddr;
794 lims->imsl_st[0] = MCAST_UNDEFINED;
795 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
805 * Graft a source entry into an existing socket-layer filter set,
806 * maintaining any required invariants and checking allocations.
808 * The source is marked as being in the new filter mode at t1.
810 * Return the pointer to the new node, otherwise return NULL.
812 static struct in_msource *
813 imf_graft(struct in_mfilter *imf, const uint8_t st1,
814 const struct sockaddr_in *psin)
816 struct ip_msource *nims;
817 struct in_msource *lims;
819 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
823 lims = (struct in_msource *)nims;
824 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
825 lims->imsl_st[0] = MCAST_UNDEFINED;
826 lims->imsl_st[1] = st1;
827 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
834 * Prune a source entry from an existing socket-layer filter set,
835 * maintaining any required invariants and checking allocations.
837 * The source is marked as being left at t1, it is not freed.
839 * Return 0 if no error occurred, otherwise return an errno value.
842 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
844 struct ip_msource find;
845 struct ip_msource *ims;
846 struct in_msource *lims;
848 /* key is host byte order */
849 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
850 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
853 lims = (struct in_msource *)ims;
854 lims->imsl_st[1] = MCAST_UNDEFINED;
859 * Revert socket-layer filter set deltas at t1 to t0 state.
862 imf_rollback(struct in_mfilter *imf)
864 struct ip_msource *ims, *tims;
865 struct in_msource *lims;
867 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
868 lims = (struct in_msource *)ims;
869 if (lims->imsl_st[0] == lims->imsl_st[1]) {
870 /* no change at t1 */
872 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
873 /* revert change to existing source at t1 */
874 lims->imsl_st[1] = lims->imsl_st[0];
876 /* revert source added t1 */
877 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
878 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
879 free(ims, M_INMFILTER);
883 imf->imf_st[1] = imf->imf_st[0];
887 * Mark socket-layer filter set as INCLUDE {} at t1.
890 imf_leave(struct in_mfilter *imf)
892 struct ip_msource *ims;
893 struct in_msource *lims;
895 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
896 lims = (struct in_msource *)ims;
897 lims->imsl_st[1] = MCAST_UNDEFINED;
899 imf->imf_st[1] = MCAST_INCLUDE;
903 * Mark socket-layer filter set deltas as committed.
906 imf_commit(struct in_mfilter *imf)
908 struct ip_msource *ims;
909 struct in_msource *lims;
911 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
912 lims = (struct in_msource *)ims;
913 lims->imsl_st[0] = lims->imsl_st[1];
915 imf->imf_st[0] = imf->imf_st[1];
919 * Reap unreferenced sources from socket-layer filter set.
922 imf_reap(struct in_mfilter *imf)
924 struct ip_msource *ims, *tims;
925 struct in_msource *lims;
927 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
928 lims = (struct in_msource *)ims;
929 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
930 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
931 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
932 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
933 free(ims, M_INMFILTER);
940 * Purge socket-layer filter set.
943 imf_purge(struct in_mfilter *imf)
945 struct ip_msource *ims, *tims;
947 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
948 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
949 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
950 free(ims, M_INMFILTER);
953 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
954 KASSERT(RB_EMPTY(&imf->imf_sources),
955 ("%s: imf_sources not empty", __func__));
959 * Look up a source filter entry for a multicast group.
961 * inm is the group descriptor to work with.
962 * haddr is the host-byte-order IPv4 address to look up.
963 * noalloc may be non-zero to suppress allocation of sources.
964 * *pims will be set to the address of the retrieved or allocated source.
966 * SMPng: NOTE: may be called with locks held.
967 * Return 0 if successful, otherwise return a non-zero error code.
970 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
971 const int noalloc, struct ip_msource **pims)
973 struct ip_msource find;
974 struct ip_msource *ims, *nims;
976 find.ims_haddr = haddr;
977 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
978 if (ims == NULL && !noalloc) {
979 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
981 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
985 nims->ims_haddr = haddr;
986 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
990 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
1000 * Merge socket-layer source into IGMP-layer source.
1001 * If rollback is non-zero, perform the inverse of the merge.
1004 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
1007 int n = rollback ? -1 : 1;
1009 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
1010 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
1011 __func__, n, ims->ims_haddr);
1012 ims->ims_st[1].ex -= n;
1013 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
1014 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
1015 __func__, n, ims->ims_haddr);
1016 ims->ims_st[1].in -= n;
1019 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
1020 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
1021 __func__, n, ims->ims_haddr);
1022 ims->ims_st[1].ex += n;
1023 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
1024 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
1025 __func__, n, ims->ims_haddr);
1026 ims->ims_st[1].in += n;
1031 * Atomically update the global in_multi state, when a membership's
1032 * filter list is being updated in any way.
1034 * imf is the per-inpcb-membership group filter pointer.
1035 * A fake imf may be passed for in-kernel consumers.
1037 * XXX This is a candidate for a set-symmetric-difference style loop
1038 * which would eliminate the repeated lookup from root of ims nodes,
1039 * as they share the same key space.
1041 * If any error occurred this function will back out of refcounts
1042 * and return a non-zero value.
1045 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1047 struct ip_msource *ims, *nims;
1048 struct in_msource *lims;
1049 int schanged, error;
1055 IN_MULTI_LIST_LOCK_ASSERT();
1058 * Update the source filters first, as this may fail.
1059 * Maintain count of in-mode filters at t0, t1. These are
1060 * used to work out if we transition into ASM mode or not.
1061 * Maintain a count of source filters whose state was
1062 * actually modified by this operation.
1064 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1065 lims = (struct in_msource *)ims;
1066 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1067 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1068 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1069 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1073 ims_merge(nims, lims, 0);
1076 struct ip_msource *bims;
1078 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1079 lims = (struct in_msource *)ims;
1080 if (lims->imsl_st[0] == lims->imsl_st[1])
1082 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1085 ims_merge(bims, lims, 1);
1090 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1091 __func__, nsrc0, nsrc1);
1093 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1094 if (imf->imf_st[0] == imf->imf_st[1] &&
1095 imf->imf_st[1] == MCAST_INCLUDE) {
1097 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1098 --inm->inm_st[1].iss_in;
1102 /* Handle filter mode transition on socket. */
1103 if (imf->imf_st[0] != imf->imf_st[1]) {
1104 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1105 __func__, imf->imf_st[0], imf->imf_st[1]);
1107 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1108 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1109 --inm->inm_st[1].iss_ex;
1110 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1111 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1112 --inm->inm_st[1].iss_in;
1115 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1116 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1117 inm->inm_st[1].iss_ex++;
1118 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1119 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1120 inm->inm_st[1].iss_in++;
1125 * Track inm filter state in terms of listener counts.
1126 * If there are any exclusive listeners, stack-wide
1127 * membership is exclusive.
1128 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1129 * If no listeners remain, state is undefined at t1,
1130 * and the IGMP lifecycle for this group should finish.
1132 if (inm->inm_st[1].iss_ex > 0) {
1133 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1134 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1135 } else if (inm->inm_st[1].iss_in > 0) {
1136 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1137 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1139 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1140 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1143 /* Decrement ASM listener count on transition out of ASM mode. */
1144 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1145 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1146 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1147 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1148 --inm->inm_st[1].iss_asm;
1152 /* Increment ASM listener count on transition to ASM mode. */
1153 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1154 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1155 inm->inm_st[1].iss_asm++;
1158 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1163 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1170 * Mark an in_multi's filter set deltas as committed.
1171 * Called by IGMP after a state change has been enqueued.
1174 inm_commit(struct in_multi *inm)
1176 struct ip_msource *ims;
1178 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1179 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1182 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1183 ims->ims_st[0] = ims->ims_st[1];
1185 inm->inm_st[0] = inm->inm_st[1];
1189 * Reap unreferenced nodes from an in_multi's filter set.
1192 inm_reap(struct in_multi *inm)
1194 struct ip_msource *ims, *tims;
1196 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1197 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1198 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1201 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1202 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1203 free(ims, M_IPMSOURCE);
1209 * Purge all source nodes from an in_multi's filter set.
1212 inm_purge(struct in_multi *inm)
1214 struct ip_msource *ims, *tims;
1216 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1217 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1218 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1219 free(ims, M_IPMSOURCE);
1225 * Join a multicast group; unlocked entry point.
1227 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1228 * is not held. Fortunately, ifp is unlikely to have been detached
1229 * at this point, so we assume it's OK to recurse.
1232 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1233 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1238 error = in_joingroup_locked(ifp, gina, imf, pinm);
1245 * Join a multicast group; real entry point.
1247 * Only preserves atomicity at inm level.
1248 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1250 * If the IGMP downcall fails, the group is not joined, and an error
1254 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1255 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1257 struct in_mfilter timf;
1258 struct in_multi *inm;
1261 IN_MULTI_LOCK_ASSERT();
1262 IN_MULTI_LIST_UNLOCK_ASSERT();
1264 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1265 ntohl(gina->s_addr), ifp, ifp->if_xname);
1271 * If no imf was specified (i.e. kernel consumer),
1272 * fake one up and assume it is an ASM join.
1275 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1279 error = in_getmulti(ifp, gina, &inm);
1281 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1284 IN_MULTI_LIST_LOCK();
1285 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1286 error = inm_merge(inm, imf);
1288 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1289 goto out_inm_release;
1292 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1293 error = igmp_change_state(inm);
1295 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1296 goto out_inm_release;
1302 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1303 inm_release_deferred(inm);
1307 IN_MULTI_LIST_UNLOCK();
1313 * Leave a multicast group; unlocked entry point.
1316 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1321 error = in_leavegroup_locked(inm, imf);
1328 * Leave a multicast group; real entry point.
1329 * All source filters will be expunged.
1331 * Only preserves atomicity at inm level.
1333 * Holding the write lock for the INP which contains imf
1334 * is highly advisable. We can't assert for it as imf does not
1335 * contain a back-pointer to the owning inp.
1337 * Note: This is not the same as inm_release(*) as this function also
1338 * makes a state change downcall into IGMP.
1341 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1343 struct in_mfilter timf;
1348 IN_MULTI_LOCK_ASSERT();
1349 IN_MULTI_LIST_UNLOCK_ASSERT();
1351 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1352 inm, ntohl(inm->inm_addr.s_addr),
1353 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1357 * If no imf was specified (i.e. kernel consumer),
1358 * fake one up and assume it is an ASM join.
1361 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1366 * Begin state merge transaction at IGMP layer.
1368 * As this particular invocation should not cause any memory
1369 * to be allocated, and there is no opportunity to roll back
1370 * the transaction, it MUST NOT fail.
1372 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1373 IN_MULTI_LIST_LOCK();
1374 error = inm_merge(inm, imf);
1375 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1377 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1378 CURVNET_SET(inm->inm_ifp->if_vnet);
1379 error = igmp_change_state(inm);
1380 IF_ADDR_WLOCK(inm->inm_ifp);
1381 inm_release_deferred(inm);
1382 IF_ADDR_WUNLOCK(inm->inm_ifp);
1383 IN_MULTI_LIST_UNLOCK();
1386 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1388 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1393 /*#ifndef BURN_BRIDGES*/
1395 * Join an IPv4 multicast group in (*,G) exclusive mode.
1396 * The group must be a 224.0.0.0/24 link-scope group.
1397 * This KPI is for legacy kernel consumers only.
1400 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1402 struct in_multi *pinm;
1405 char addrbuf[INET_ADDRSTRLEN];
1408 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1409 ("%s: %s not in 224.0.0.0/24", __func__,
1410 inet_ntoa_r(*ap, addrbuf)));
1412 error = in_joingroup(ifp, ap, NULL, &pinm);
1420 * Block or unblock an ASM multicast source on an inpcb.
1421 * This implements the delta-based API described in RFC 3678.
1423 * The delta-based API applies only to exclusive-mode memberships.
1424 * An IGMP downcall will be performed.
1426 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1428 * Return 0 if successful, otherwise return an appropriate error code.
1431 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1433 struct group_source_req gsr;
1434 sockunion_t *gsa, *ssa;
1436 struct in_mfilter *imf;
1437 struct ip_moptions *imo;
1438 struct in_msource *ims;
1439 struct in_multi *inm;
1448 memset(&gsr, 0, sizeof(struct group_source_req));
1449 gsa = (sockunion_t *)&gsr.gsr_group;
1450 ssa = (sockunion_t *)&gsr.gsr_source;
1452 switch (sopt->sopt_name) {
1453 case IP_BLOCK_SOURCE:
1454 case IP_UNBLOCK_SOURCE: {
1455 struct ip_mreq_source mreqs;
1457 error = sooptcopyin(sopt, &mreqs,
1458 sizeof(struct ip_mreq_source),
1459 sizeof(struct ip_mreq_source));
1463 gsa->sin.sin_family = AF_INET;
1464 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1465 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1467 ssa->sin.sin_family = AF_INET;
1468 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1469 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1471 if (!in_nullhost(mreqs.imr_interface))
1472 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1474 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1477 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1478 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1482 case MCAST_BLOCK_SOURCE:
1483 case MCAST_UNBLOCK_SOURCE:
1484 error = sooptcopyin(sopt, &gsr,
1485 sizeof(struct group_source_req),
1486 sizeof(struct group_source_req));
1490 if (gsa->sin.sin_family != AF_INET ||
1491 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1494 if (ssa->sin.sin_family != AF_INET ||
1495 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1498 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1499 return (EADDRNOTAVAIL);
1501 ifp = ifnet_byindex(gsr.gsr_interface);
1503 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1508 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1509 __func__, sopt->sopt_name);
1510 return (EOPNOTSUPP);
1514 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1518 * Check if we are actually a member of this group.
1520 imo = inp_findmoptions(inp);
1521 idx = imo_match_group(imo, ifp, &gsa->sa);
1522 if (idx == -1 || imo->imo_mfilters == NULL) {
1523 error = EADDRNOTAVAIL;
1524 goto out_inp_locked;
1527 KASSERT(imo->imo_mfilters != NULL,
1528 ("%s: imo_mfilters not allocated", __func__));
1529 imf = &imo->imo_mfilters[idx];
1530 inm = imo->imo_membership[idx];
1533 * Attempting to use the delta-based API on an
1534 * non exclusive-mode membership is an error.
1536 fmode = imf->imf_st[0];
1537 if (fmode != MCAST_EXCLUDE) {
1539 goto out_inp_locked;
1543 * Deal with error cases up-front:
1544 * Asked to block, but already blocked; or
1545 * Asked to unblock, but nothing to unblock.
1546 * If adding a new block entry, allocate it.
1548 ims = imo_match_source(imo, idx, &ssa->sa);
1549 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1550 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1551 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1552 error = EADDRNOTAVAIL;
1553 goto out_inp_locked;
1556 INP_WLOCK_ASSERT(inp);
1559 * Begin state merge transaction at socket layer.
1562 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1563 ims = imf_graft(imf, fmode, &ssa->sin);
1567 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1568 error = imf_prune(imf, &ssa->sin);
1572 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1573 goto out_imf_rollback;
1577 * Begin state merge transaction at IGMP layer.
1580 IN_MULTI_LIST_LOCK();
1581 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1582 error = inm_merge(inm, imf);
1584 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1585 goto out_in_multi_locked;
1588 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1589 error = igmp_change_state(inm);
1591 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1593 out_in_multi_locked:
1611 * Given an inpcb, return its multicast options structure pointer. Accepts
1612 * an unlocked inpcb pointer, but will return it locked. May sleep.
1614 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1615 * SMPng: NOTE: Returns with the INP write lock held.
1617 static struct ip_moptions *
1618 inp_findmoptions(struct inpcb *inp)
1620 struct ip_moptions *imo;
1621 struct in_multi **immp;
1622 struct in_mfilter *imfp;
1626 if (inp->inp_moptions != NULL)
1627 return (inp->inp_moptions);
1631 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1632 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1634 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1635 M_INMFILTER, M_WAITOK);
1637 imo->imo_multicast_ifp = NULL;
1638 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1639 imo->imo_multicast_vif = -1;
1640 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1641 imo->imo_multicast_loop = in_mcast_loop;
1642 imo->imo_num_memberships = 0;
1643 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1644 imo->imo_membership = immp;
1646 /* Initialize per-group source filters. */
1647 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1648 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1649 imo->imo_mfilters = imfp;
1652 if (inp->inp_moptions != NULL) {
1653 free(imfp, M_INMFILTER);
1654 free(immp, M_IPMOPTS);
1655 free(imo, M_IPMOPTS);
1656 return (inp->inp_moptions);
1658 inp->inp_moptions = imo;
1663 inp_gcmoptions(epoch_context_t ctx)
1665 struct ip_moptions *imo;
1666 struct in_mfilter *imf;
1667 struct in_multi *inm;
1669 size_t idx, nmships;
1671 imo = __containerof(ctx, struct ip_moptions, imo_epoch_ctx);
1673 nmships = imo->imo_num_memberships;
1674 for (idx = 0; idx < nmships; ++idx) {
1675 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1678 inm = imo->imo_membership[idx];
1681 CURVNET_SET(ifp->if_vnet);
1682 (void)in_leavegroup(inm, imf);
1689 if (imo->imo_mfilters)
1690 free(imo->imo_mfilters, M_INMFILTER);
1691 free(imo->imo_membership, M_IPMOPTS);
1692 free(imo, M_IPMOPTS);
1696 * Discard the IP multicast options (and source filters). To minimize
1697 * the amount of work done while holding locks such as the INP's
1698 * pcbinfo lock (which is used in the receive path), the free
1699 * operation is deferred to the epoch callback task.
1702 inp_freemoptions(struct ip_moptions *imo)
1706 epoch_call(net_epoch_preempt, &imo->imo_epoch_ctx, inp_gcmoptions);
1710 * Atomically get source filters on a socket for an IPv4 multicast group.
1711 * Called with INP lock held; returns with lock released.
1714 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1716 struct __msfilterreq msfr;
1719 struct ip_moptions *imo;
1720 struct in_mfilter *imf;
1721 struct ip_msource *ims;
1722 struct in_msource *lims;
1723 struct sockaddr_in *psin;
1724 struct sockaddr_storage *ptss;
1725 struct sockaddr_storage *tss;
1727 size_t idx, nsrcs, ncsrcs;
1729 INP_WLOCK_ASSERT(inp);
1731 imo = inp->inp_moptions;
1732 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1736 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1737 sizeof(struct __msfilterreq));
1741 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1744 ifp = ifnet_byindex(msfr.msfr_ifindex);
1751 * Lookup group on the socket.
1753 gsa = (sockunion_t *)&msfr.msfr_group;
1754 idx = imo_match_group(imo, ifp, &gsa->sa);
1755 if (idx == -1 || imo->imo_mfilters == NULL) {
1757 return (EADDRNOTAVAIL);
1759 imf = &imo->imo_mfilters[idx];
1762 * Ignore memberships which are in limbo.
1764 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1768 msfr.msfr_fmode = imf->imf_st[1];
1771 * If the user specified a buffer, copy out the source filter
1772 * entries to userland gracefully.
1773 * We only copy out the number of entries which userland
1774 * has asked for, but we always tell userland how big the
1775 * buffer really needs to be.
1777 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1778 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1780 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1781 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1782 M_TEMP, M_NOWAIT | M_ZERO);
1790 * Count number of sources in-mode at t0.
1791 * If buffer space exists and remains, copy out source entries.
1793 nsrcs = msfr.msfr_nsrcs;
1796 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1797 lims = (struct in_msource *)ims;
1798 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1799 lims->imsl_st[0] != imf->imf_st[0])
1802 if (tss != NULL && nsrcs > 0) {
1803 psin = (struct sockaddr_in *)ptss;
1804 psin->sin_family = AF_INET;
1805 psin->sin_len = sizeof(struct sockaddr_in);
1806 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1816 error = copyout(tss, msfr.msfr_srcs,
1817 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1823 msfr.msfr_nsrcs = ncsrcs;
1824 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1830 * Return the IP multicast options in response to user getsockopt().
1833 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1835 struct rm_priotracker in_ifa_tracker;
1836 struct ip_mreqn mreqn;
1837 struct ip_moptions *imo;
1839 struct in_ifaddr *ia;
1844 imo = inp->inp_moptions;
1846 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1847 * or is a divert socket, reject it.
1849 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1850 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1851 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1853 return (EOPNOTSUPP);
1857 switch (sopt->sopt_name) {
1858 case IP_MULTICAST_VIF:
1860 optval = imo->imo_multicast_vif;
1864 error = sooptcopyout(sopt, &optval, sizeof(int));
1867 case IP_MULTICAST_IF:
1868 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1870 ifp = imo->imo_multicast_ifp;
1871 if (!in_nullhost(imo->imo_multicast_addr)) {
1872 mreqn.imr_address = imo->imo_multicast_addr;
1873 } else if (ifp != NULL) {
1874 mreqn.imr_ifindex = ifp->if_index;
1875 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1878 IA_SIN(ia)->sin_addr;
1879 ifa_free(&ia->ia_ifa);
1884 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1885 error = sooptcopyout(sopt, &mreqn,
1886 sizeof(struct ip_mreqn));
1888 error = sooptcopyout(sopt, &mreqn.imr_address,
1889 sizeof(struct in_addr));
1893 case IP_MULTICAST_TTL:
1895 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1897 optval = coptval = imo->imo_multicast_ttl;
1899 if (sopt->sopt_valsize == sizeof(u_char))
1900 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1902 error = sooptcopyout(sopt, &optval, sizeof(int));
1905 case IP_MULTICAST_LOOP:
1907 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1909 optval = coptval = imo->imo_multicast_loop;
1911 if (sopt->sopt_valsize == sizeof(u_char))
1912 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1914 error = sooptcopyout(sopt, &optval, sizeof(int));
1919 error = EADDRNOTAVAIL;
1922 error = inp_get_source_filters(inp, sopt);
1928 error = ENOPROTOOPT;
1932 INP_UNLOCK_ASSERT(inp);
1938 * Look up the ifnet to use for a multicast group membership,
1939 * given the IPv4 address of an interface, and the IPv4 group address.
1941 * This routine exists to support legacy multicast applications
1942 * which do not understand that multicast memberships are scoped to
1943 * specific physical links in the networking stack, or which need
1944 * to join link-scope groups before IPv4 addresses are configured.
1946 * If inp is non-NULL, use this socket's current FIB number for any
1947 * required FIB lookup.
1948 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1949 * and use its ifp; usually, this points to the default next-hop.
1951 * If the FIB lookup fails, attempt to use the first non-loopback
1952 * interface with multicast capability in the system as a
1953 * last resort. The legacy IPv4 ASM API requires that we do
1954 * this in order to allow groups to be joined when the routing
1955 * table has not yet been populated during boot.
1957 * Returns NULL if no ifp could be found.
1959 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1960 * FUTURE: Implement IPv4 source-address selection.
1962 static struct ifnet *
1963 inp_lookup_mcast_ifp(const struct inpcb *inp,
1964 const struct sockaddr_in *gsin, const struct in_addr ina)
1966 struct rm_priotracker in_ifa_tracker;
1968 struct nhop4_basic nh4;
1971 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1972 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1973 ("%s: not multicast", __func__));
1976 if (!in_nullhost(ina)) {
1977 INADDR_TO_IFP(ina, ifp);
1979 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1980 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1983 struct in_ifaddr *ia;
1987 IN_IFADDR_RLOCK(&in_ifa_tracker);
1988 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1990 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1991 (mifp->if_flags & IFF_MULTICAST)) {
1996 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2004 * Join an IPv4 multicast group, possibly with a source.
2007 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
2009 struct group_source_req gsr;
2010 sockunion_t *gsa, *ssa;
2012 struct in_mfilter *imf;
2013 struct ip_moptions *imo;
2014 struct in_multi *inm;
2015 struct in_msource *lims;
2025 memset(&gsr, 0, sizeof(struct group_source_req));
2026 gsa = (sockunion_t *)&gsr.gsr_group;
2027 gsa->ss.ss_family = AF_UNSPEC;
2028 ssa = (sockunion_t *)&gsr.gsr_source;
2029 ssa->ss.ss_family = AF_UNSPEC;
2031 switch (sopt->sopt_name) {
2032 case IP_ADD_MEMBERSHIP:
2033 case IP_ADD_SOURCE_MEMBERSHIP: {
2034 struct ip_mreq_source mreqs;
2036 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
2037 error = sooptcopyin(sopt, &mreqs,
2038 sizeof(struct ip_mreq),
2039 sizeof(struct ip_mreq));
2041 * Do argument switcharoo from ip_mreq into
2042 * ip_mreq_source to avoid using two instances.
2044 mreqs.imr_interface = mreqs.imr_sourceaddr;
2045 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2046 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2047 error = sooptcopyin(sopt, &mreqs,
2048 sizeof(struct ip_mreq_source),
2049 sizeof(struct ip_mreq_source));
2054 gsa->sin.sin_family = AF_INET;
2055 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2056 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2058 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2059 ssa->sin.sin_family = AF_INET;
2060 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2061 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2064 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2067 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2068 mreqs.imr_interface);
2069 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2070 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2074 case MCAST_JOIN_GROUP:
2075 case MCAST_JOIN_SOURCE_GROUP:
2076 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2077 error = sooptcopyin(sopt, &gsr,
2078 sizeof(struct group_req),
2079 sizeof(struct group_req));
2080 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2081 error = sooptcopyin(sopt, &gsr,
2082 sizeof(struct group_source_req),
2083 sizeof(struct group_source_req));
2088 if (gsa->sin.sin_family != AF_INET ||
2089 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2093 * Overwrite the port field if present, as the sockaddr
2094 * being copied in may be matched with a binary comparison.
2096 gsa->sin.sin_port = 0;
2097 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2098 if (ssa->sin.sin_family != AF_INET ||
2099 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2101 ssa->sin.sin_port = 0;
2104 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2107 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2108 return (EADDRNOTAVAIL);
2109 ifp = ifnet_byindex(gsr.gsr_interface);
2113 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2114 __func__, sopt->sopt_name);
2115 return (EOPNOTSUPP);
2119 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2120 return (EADDRNOTAVAIL);
2122 imo = inp_findmoptions(inp);
2123 idx = imo_match_group(imo, ifp, &gsa->sa);
2127 inm = imo->imo_membership[idx];
2128 imf = &imo->imo_mfilters[idx];
2129 if (ssa->ss.ss_family != AF_UNSPEC) {
2131 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2132 * is an error. On an existing inclusive membership,
2133 * it just adds the source to the filter list.
2135 if (imf->imf_st[1] != MCAST_INCLUDE) {
2137 goto out_inp_locked;
2140 * Throw out duplicates.
2142 * XXX FIXME: This makes a naive assumption that
2143 * even if entries exist for *ssa in this imf,
2144 * they will be rejected as dupes, even if they
2145 * are not valid in the current mode (in-mode).
2147 * in_msource is transactioned just as for anything
2148 * else in SSM -- but note naive use of inm_graft()
2149 * below for allocating new filter entries.
2151 * This is only an issue if someone mixes the
2152 * full-state SSM API with the delta-based API,
2153 * which is discouraged in the relevant RFCs.
2155 lims = imo_match_source(imo, idx, &ssa->sa);
2156 if (lims != NULL /*&&
2157 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2158 error = EADDRNOTAVAIL;
2159 goto out_inp_locked;
2163 * MCAST_JOIN_GROUP on an existing exclusive
2164 * membership is an error; return EADDRINUSE
2165 * to preserve 4.4BSD API idempotence, and
2166 * avoid tedious detour to code below.
2167 * NOTE: This is bending RFC 3678 a bit.
2169 * On an existing inclusive membership, this is also
2170 * an error; if you want to change filter mode,
2171 * you must use the userland API setsourcefilter().
2172 * XXX We don't reject this for imf in UNDEFINED
2173 * state at t1, because allocation of a filter
2174 * is atomic with allocation of a membership.
2177 if (imf->imf_st[1] == MCAST_EXCLUDE)
2179 goto out_inp_locked;
2184 * Begin state merge transaction at socket layer.
2186 INP_WLOCK_ASSERT(inp);
2189 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2190 error = imo_grow(imo);
2192 goto out_inp_locked;
2195 * Allocate the new slot upfront so we can deal with
2196 * grafting the new source filter in same code path
2197 * as for join-source on existing membership.
2199 idx = imo->imo_num_memberships;
2200 imo->imo_membership[idx] = NULL;
2201 imo->imo_num_memberships++;
2202 KASSERT(imo->imo_mfilters != NULL,
2203 ("%s: imf_mfilters vector was not allocated", __func__));
2204 imf = &imo->imo_mfilters[idx];
2205 KASSERT(RB_EMPTY(&imf->imf_sources),
2206 ("%s: imf_sources not empty", __func__));
2210 * Graft new source into filter list for this inpcb's
2211 * membership of the group. The in_multi may not have
2212 * been allocated yet if this is a new membership, however,
2213 * the in_mfilter slot will be allocated and must be initialized.
2215 * Note: Grafting of exclusive mode filters doesn't happen
2217 * XXX: Should check for non-NULL lims (node exists but may
2218 * not be in-mode) for interop with full-state API.
2220 if (ssa->ss.ss_family != AF_UNSPEC) {
2221 /* Membership starts in IN mode */
2223 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2224 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2226 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2228 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2230 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2236 /* No address specified; Membership starts in EX mode */
2238 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2239 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2244 * Begin state merge transaction at IGMP layer.
2251 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2254 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2256 IN_MULTI_LIST_UNLOCK();
2259 imo->imo_membership[idx] = inm;
2261 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2262 IN_MULTI_LIST_LOCK();
2263 error = inm_merge(inm, imf);
2265 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2267 IN_MULTI_LIST_UNLOCK();
2268 goto out_in_multi_locked;
2270 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2271 error = igmp_change_state(inm);
2272 IN_MULTI_LIST_UNLOCK();
2274 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2276 goto out_in_multi_locked;
2280 out_in_multi_locked:
2284 if (in_pcbrele_wlocked(inp))
2297 if (error && is_new) {
2298 imo->imo_membership[idx] = NULL;
2299 --imo->imo_num_memberships;
2308 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2311 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2313 struct group_source_req gsr;
2314 struct ip_mreq_source mreqs;
2315 sockunion_t *gsa, *ssa;
2317 struct in_mfilter *imf;
2318 struct ip_moptions *imo;
2319 struct in_msource *ims;
2320 struct in_multi *inm;
2322 int error, is_final;
2328 memset(&gsr, 0, sizeof(struct group_source_req));
2329 gsa = (sockunion_t *)&gsr.gsr_group;
2330 gsa->ss.ss_family = AF_UNSPEC;
2331 ssa = (sockunion_t *)&gsr.gsr_source;
2332 ssa->ss.ss_family = AF_UNSPEC;
2334 switch (sopt->sopt_name) {
2335 case IP_DROP_MEMBERSHIP:
2336 case IP_DROP_SOURCE_MEMBERSHIP:
2337 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2338 error = sooptcopyin(sopt, &mreqs,
2339 sizeof(struct ip_mreq),
2340 sizeof(struct ip_mreq));
2342 * Swap interface and sourceaddr arguments,
2343 * as ip_mreq and ip_mreq_source are laid
2346 mreqs.imr_interface = mreqs.imr_sourceaddr;
2347 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2348 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2349 error = sooptcopyin(sopt, &mreqs,
2350 sizeof(struct ip_mreq_source),
2351 sizeof(struct ip_mreq_source));
2356 gsa->sin.sin_family = AF_INET;
2357 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2358 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2360 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2361 ssa->sin.sin_family = AF_INET;
2362 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2363 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2367 * Attempt to look up hinted ifp from interface address.
2368 * Fallthrough with null ifp iff lookup fails, to
2369 * preserve 4.4BSD mcast API idempotence.
2370 * XXX NOTE WELL: The RFC 3678 API is preferred because
2371 * using an IPv4 address as a key is racy.
2373 if (!in_nullhost(mreqs.imr_interface))
2374 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2376 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2377 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2381 case MCAST_LEAVE_GROUP:
2382 case MCAST_LEAVE_SOURCE_GROUP:
2383 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2384 error = sooptcopyin(sopt, &gsr,
2385 sizeof(struct group_req),
2386 sizeof(struct group_req));
2387 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2388 error = sooptcopyin(sopt, &gsr,
2389 sizeof(struct group_source_req),
2390 sizeof(struct group_source_req));
2395 if (gsa->sin.sin_family != AF_INET ||
2396 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2399 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2400 if (ssa->sin.sin_family != AF_INET ||
2401 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2405 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2406 return (EADDRNOTAVAIL);
2408 ifp = ifnet_byindex(gsr.gsr_interface);
2411 return (EADDRNOTAVAIL);
2415 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2416 __func__, sopt->sopt_name);
2417 return (EOPNOTSUPP);
2421 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2425 * Find the membership in the membership array.
2427 imo = inp_findmoptions(inp);
2428 idx = imo_match_group(imo, ifp, &gsa->sa);
2430 error = EADDRNOTAVAIL;
2431 goto out_inp_locked;
2433 inm = imo->imo_membership[idx];
2434 imf = &imo->imo_mfilters[idx];
2436 if (ssa->ss.ss_family != AF_UNSPEC)
2440 * Begin state merge transaction at socket layer.
2442 INP_WLOCK_ASSERT(inp);
2445 * If we were instructed only to leave a given source, do so.
2446 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2451 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2452 error = EADDRNOTAVAIL;
2453 goto out_inp_locked;
2455 ims = imo_match_source(imo, idx, &ssa->sa);
2457 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2458 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2459 error = EADDRNOTAVAIL;
2460 goto out_inp_locked;
2462 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2463 error = imf_prune(imf, &ssa->sin);
2465 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2467 goto out_inp_locked;
2472 * Begin state merge transaction at IGMP layer.
2480 * Give up the multicast address record to which
2481 * the membership points.
2483 (void)in_leavegroup_locked(inm, imf);
2485 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2486 IN_MULTI_LIST_LOCK();
2487 error = inm_merge(inm, imf);
2489 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2491 goto out_in_multi_locked;
2494 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2495 error = igmp_change_state(inm);
2496 IN_MULTI_LIST_UNLOCK();
2498 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2503 out_in_multi_locked:
2507 if (in_pcbrele_wlocked(inp))
2518 /* Remove the gap in the membership and filter array. */
2519 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2520 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2521 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2523 imo->imo_num_memberships--;
2532 * Select the interface for transmitting IPv4 multicast datagrams.
2534 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2535 * may be passed to this socket option. An address of INADDR_ANY or an
2536 * interface index of 0 is used to remove a previous selection.
2537 * When no interface is selected, one is chosen for every send.
2540 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2542 struct in_addr addr;
2543 struct ip_mreqn mreqn;
2545 struct ip_moptions *imo;
2548 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2550 * An interface index was specified using the
2551 * Linux-derived ip_mreqn structure.
2553 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2554 sizeof(struct ip_mreqn));
2558 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2561 if (mreqn.imr_ifindex == 0) {
2564 ifp = ifnet_byindex(mreqn.imr_ifindex);
2566 return (EADDRNOTAVAIL);
2570 * An interface was specified by IPv4 address.
2571 * This is the traditional BSD usage.
2573 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2574 sizeof(struct in_addr));
2577 if (in_nullhost(addr)) {
2580 INADDR_TO_IFP(addr, ifp);
2582 return (EADDRNOTAVAIL);
2584 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2585 ntohl(addr.s_addr));
2588 /* Reject interfaces which do not support multicast. */
2589 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2590 return (EOPNOTSUPP);
2592 imo = inp_findmoptions(inp);
2593 imo->imo_multicast_ifp = ifp;
2594 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2601 * Atomically set source filters on a socket for an IPv4 multicast group.
2603 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2606 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2608 struct __msfilterreq msfr;
2611 struct in_mfilter *imf;
2612 struct ip_moptions *imo;
2613 struct in_multi *inm;
2617 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2618 sizeof(struct __msfilterreq));
2622 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2625 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2626 msfr.msfr_fmode != MCAST_INCLUDE))
2629 if (msfr.msfr_group.ss_family != AF_INET ||
2630 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2633 gsa = (sockunion_t *)&msfr.msfr_group;
2634 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2637 gsa->sin.sin_port = 0; /* ignore port */
2639 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2640 return (EADDRNOTAVAIL);
2642 ifp = ifnet_byindex(msfr.msfr_ifindex);
2644 return (EADDRNOTAVAIL);
2647 * Take the INP write lock.
2648 * Check if this socket is a member of this group.
2650 imo = inp_findmoptions(inp);
2651 idx = imo_match_group(imo, ifp, &gsa->sa);
2652 if (idx == -1 || imo->imo_mfilters == NULL) {
2653 error = EADDRNOTAVAIL;
2654 goto out_inp_locked;
2656 inm = imo->imo_membership[idx];
2657 imf = &imo->imo_mfilters[idx];
2660 * Begin state merge transaction at socket layer.
2662 INP_WLOCK_ASSERT(inp);
2664 imf->imf_st[1] = msfr.msfr_fmode;
2667 * Apply any new source filters, if present.
2668 * Make a copy of the user-space source vector so
2669 * that we may copy them with a single copyin. This
2670 * allows us to deal with page faults up-front.
2672 if (msfr.msfr_nsrcs > 0) {
2673 struct in_msource *lims;
2674 struct sockaddr_in *psin;
2675 struct sockaddr_storage *kss, *pkss;
2680 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2681 __func__, (unsigned long)msfr.msfr_nsrcs);
2682 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2684 error = copyin(msfr.msfr_srcs, kss,
2685 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2694 * Mark all source filters as UNDEFINED at t1.
2695 * Restore new group filter mode, as imf_leave()
2696 * will set it to INCLUDE.
2699 imf->imf_st[1] = msfr.msfr_fmode;
2702 * Update socket layer filters at t1, lazy-allocating
2703 * new entries. This saves a bunch of memory at the
2704 * cost of one RB_FIND() per source entry; duplicate
2705 * entries in the msfr_nsrcs vector are ignored.
2706 * If we encounter an error, rollback transaction.
2708 * XXX This too could be replaced with a set-symmetric
2709 * difference like loop to avoid walking from root
2710 * every time, as the key space is common.
2712 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2713 psin = (struct sockaddr_in *)pkss;
2714 if (psin->sin_family != AF_INET) {
2715 error = EAFNOSUPPORT;
2718 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2722 error = imf_get_source(imf, psin, &lims);
2725 lims->imsl_st[1] = imf->imf_st[1];
2731 goto out_imf_rollback;
2733 INP_WLOCK_ASSERT(inp);
2735 IN_MULTI_LIST_LOCK();
2738 * Begin state merge transaction at IGMP layer.
2740 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2741 error = inm_merge(inm, imf);
2743 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2744 IN_MULTI_LIST_UNLOCK();
2745 goto out_in_multi_locked;
2748 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2749 error = igmp_change_state(inm);
2750 IN_MULTI_LIST_UNLOCK();
2752 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2754 out_in_multi_locked:
2772 * Set the IP multicast options in response to user setsockopt().
2774 * Many of the socket options handled in this function duplicate the
2775 * functionality of socket options in the regular unicast API. However,
2776 * it is not possible to merge the duplicate code, because the idempotence
2777 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2778 * the effects of these options must be treated as separate and distinct.
2780 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2781 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2782 * is refactored to no longer use vifs.
2785 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2787 struct ip_moptions *imo;
2793 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2794 * or is a divert socket, reject it.
2796 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2797 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2798 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2799 return (EOPNOTSUPP);
2801 switch (sopt->sopt_name) {
2802 case IP_MULTICAST_VIF: {
2805 * Select a multicast VIF for transmission.
2806 * Only useful if multicast forwarding is active.
2808 if (legal_vif_num == NULL) {
2812 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2815 if (!legal_vif_num(vifi) && (vifi != -1)) {
2819 imo = inp_findmoptions(inp);
2820 imo->imo_multicast_vif = vifi;
2825 case IP_MULTICAST_IF:
2826 error = inp_set_multicast_if(inp, sopt);
2829 case IP_MULTICAST_TTL: {
2833 * Set the IP time-to-live for outgoing multicast packets.
2834 * The original multicast API required a char argument,
2835 * which is inconsistent with the rest of the socket API.
2836 * We allow either a char or an int.
2838 if (sopt->sopt_valsize == sizeof(u_char)) {
2839 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2846 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2856 imo = inp_findmoptions(inp);
2857 imo->imo_multicast_ttl = ttl;
2862 case IP_MULTICAST_LOOP: {
2866 * Set the loopback flag for outgoing multicast packets.
2867 * Must be zero or one. The original multicast API required a
2868 * char argument, which is inconsistent with the rest
2869 * of the socket API. We allow either a char or an int.
2871 if (sopt->sopt_valsize == sizeof(u_char)) {
2872 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2879 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2883 loop = (u_char)iloop;
2885 imo = inp_findmoptions(inp);
2886 imo->imo_multicast_loop = !!loop;
2891 case IP_ADD_MEMBERSHIP:
2892 case IP_ADD_SOURCE_MEMBERSHIP:
2893 case MCAST_JOIN_GROUP:
2894 case MCAST_JOIN_SOURCE_GROUP:
2895 error = inp_join_group(inp, sopt);
2898 case IP_DROP_MEMBERSHIP:
2899 case IP_DROP_SOURCE_MEMBERSHIP:
2900 case MCAST_LEAVE_GROUP:
2901 case MCAST_LEAVE_SOURCE_GROUP:
2902 error = inp_leave_group(inp, sopt);
2905 case IP_BLOCK_SOURCE:
2906 case IP_UNBLOCK_SOURCE:
2907 case MCAST_BLOCK_SOURCE:
2908 case MCAST_UNBLOCK_SOURCE:
2909 error = inp_block_unblock_source(inp, sopt);
2913 error = inp_set_source_filters(inp, sopt);
2921 INP_UNLOCK_ASSERT(inp);
2927 * Expose IGMP's multicast filter mode and source list(s) to userland,
2928 * keyed by (ifindex, group).
2929 * The filter mode is written out as a uint32_t, followed by
2930 * 0..n of struct in_addr.
2931 * For use by ifmcstat(8).
2932 * SMPng: NOTE: unlocked read of ifindex space.
2935 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2937 struct in_addr src, group;
2939 struct ifmultiaddr *ifma;
2940 struct in_multi *inm;
2941 struct ip_msource *ims;
2945 uint32_t fmode, ifindex;
2950 if (req->newptr != NULL)
2957 if (ifindex <= 0 || ifindex > V_if_index) {
2958 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2963 group.s_addr = name[1];
2964 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2965 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2966 __func__, ntohl(group.s_addr));
2970 ifp = ifnet_byindex(ifindex);
2972 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2977 retval = sysctl_wire_old_buffer(req,
2978 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2982 IN_MULTI_LIST_LOCK();
2985 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2986 if (ifma->ifma_addr->sa_family != AF_INET ||
2987 ifma->ifma_protospec == NULL)
2989 inm = (struct in_multi *)ifma->ifma_protospec;
2990 if (!in_hosteq(inm->inm_addr, group))
2992 fmode = inm->inm_st[1].iss_fmode;
2993 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2996 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2997 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
3000 * Only copy-out sources which are in-mode.
3002 if (fmode != ims_get_mode(inm, ims, 1)) {
3003 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
3007 src.s_addr = htonl(ims->ims_haddr);
3008 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
3013 IF_ADDR_RUNLOCK(ifp);
3015 IN_MULTI_LIST_UNLOCK();
3020 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
3022 static const char *inm_modestrs[] = { "un", "in", "ex" };
3025 inm_mode_str(const int mode)
3028 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
3029 return (inm_modestrs[mode]);
3033 static const char *inm_statestrs[] = {
3046 inm_state_str(const int state)
3049 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3050 return (inm_statestrs[state]);
3055 * Dump an in_multi structure to the console.
3058 inm_print(const struct in_multi *inm)
3061 char addrbuf[INET_ADDRSTRLEN];
3063 if ((ktr_mask & KTR_IGMPV3) == 0)
3066 printf("%s: --- begin inm %p ---\n", __func__, inm);
3067 printf("addr %s ifp %p(%s) ifma %p\n",
3068 inet_ntoa_r(inm->inm_addr, addrbuf),
3070 inm->inm_ifp->if_xname,
3072 printf("timer %u state %s refcount %u scq.len %u\n",
3074 inm_state_str(inm->inm_state),
3076 inm->inm_scq.mq_len);
3077 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3082 for (t = 0; t < 2; t++) {
3083 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3084 inm_mode_str(inm->inm_st[t].iss_fmode),
3085 inm->inm_st[t].iss_asm,
3086 inm->inm_st[t].iss_ex,
3087 inm->inm_st[t].iss_in,
3088 inm->inm_st[t].iss_rec);
3090 printf("%s: --- end inm %p ---\n", __func__, inm);
3093 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3096 inm_print(const struct in_multi *inm)
3101 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3103 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);