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);
670 if_delmulti_ifma_flags(ifma, 1);
675 free(inm, M_IPMADDR);
676 if_delmulti_ifma_flags(ifma, 1);
681 * Clear recorded source entries for a group.
682 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
683 * FIXME: Should reap.
686 inm_clear_recorded(struct in_multi *inm)
688 struct ip_msource *ims;
690 IN_MULTI_LIST_LOCK_ASSERT();
692 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
695 --inm->inm_st[1].iss_rec;
698 KASSERT(inm->inm_st[1].iss_rec == 0,
699 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
703 * Record a source as pending for a Source-Group IGMPv3 query.
704 * This lives here as it modifies the shared tree.
706 * inm is the group descriptor.
707 * naddr is the address of the source to record in network-byte order.
709 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
710 * lazy-allocate a source node in response to an SG query.
711 * Otherwise, no allocation is performed. This saves some memory
712 * with the trade-off that the source will not be reported to the
713 * router if joined in the window between the query response and
714 * the group actually being joined on the local host.
716 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
717 * This turns off the allocation of a recorded source entry if
718 * the group has not been joined.
720 * Return 0 if the source didn't exist or was already marked as recorded.
721 * Return 1 if the source was marked as recorded by this function.
722 * Return <0 if any error occurred (negated errno code).
725 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
727 struct ip_msource find;
728 struct ip_msource *ims, *nims;
730 IN_MULTI_LIST_LOCK_ASSERT();
732 find.ims_haddr = ntohl(naddr);
733 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
734 if (ims && ims->ims_stp)
737 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
739 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
743 nims->ims_haddr = find.ims_haddr;
744 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
750 * Mark the source as recorded and update the recorded
754 ++inm->inm_st[1].iss_rec;
760 * Return a pointer to an in_msource owned by an in_mfilter,
761 * given its source address.
762 * Lazy-allocate if needed. If this is a new entry its filter state is
765 * imf is the filter set being modified.
766 * haddr is the source address in *host* byte-order.
768 * SMPng: May be called with locks held; malloc must not block.
771 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
772 struct in_msource **plims)
774 struct ip_msource find;
775 struct ip_msource *ims, *nims;
776 struct in_msource *lims;
783 /* key is host byte order */
784 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
785 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
786 lims = (struct in_msource *)ims;
788 if (imf->imf_nsrc == in_mcast_maxsocksrc)
790 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
794 lims = (struct in_msource *)nims;
795 lims->ims_haddr = find.ims_haddr;
796 lims->imsl_st[0] = MCAST_UNDEFINED;
797 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
807 * Graft a source entry into an existing socket-layer filter set,
808 * maintaining any required invariants and checking allocations.
810 * The source is marked as being in the new filter mode at t1.
812 * Return the pointer to the new node, otherwise return NULL.
814 static struct in_msource *
815 imf_graft(struct in_mfilter *imf, const uint8_t st1,
816 const struct sockaddr_in *psin)
818 struct ip_msource *nims;
819 struct in_msource *lims;
821 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
825 lims = (struct in_msource *)nims;
826 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
827 lims->imsl_st[0] = MCAST_UNDEFINED;
828 lims->imsl_st[1] = st1;
829 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
836 * Prune a source entry from an existing socket-layer filter set,
837 * maintaining any required invariants and checking allocations.
839 * The source is marked as being left at t1, it is not freed.
841 * Return 0 if no error occurred, otherwise return an errno value.
844 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
846 struct ip_msource find;
847 struct ip_msource *ims;
848 struct in_msource *lims;
850 /* key is host byte order */
851 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
852 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
855 lims = (struct in_msource *)ims;
856 lims->imsl_st[1] = MCAST_UNDEFINED;
861 * Revert socket-layer filter set deltas at t1 to t0 state.
864 imf_rollback(struct in_mfilter *imf)
866 struct ip_msource *ims, *tims;
867 struct in_msource *lims;
869 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
870 lims = (struct in_msource *)ims;
871 if (lims->imsl_st[0] == lims->imsl_st[1]) {
872 /* no change at t1 */
874 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
875 /* revert change to existing source at t1 */
876 lims->imsl_st[1] = lims->imsl_st[0];
878 /* revert source added t1 */
879 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
880 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
881 free(ims, M_INMFILTER);
885 imf->imf_st[1] = imf->imf_st[0];
889 * Mark socket-layer filter set as INCLUDE {} at t1.
892 imf_leave(struct in_mfilter *imf)
894 struct ip_msource *ims;
895 struct in_msource *lims;
897 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
898 lims = (struct in_msource *)ims;
899 lims->imsl_st[1] = MCAST_UNDEFINED;
901 imf->imf_st[1] = MCAST_INCLUDE;
905 * Mark socket-layer filter set deltas as committed.
908 imf_commit(struct in_mfilter *imf)
910 struct ip_msource *ims;
911 struct in_msource *lims;
913 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
914 lims = (struct in_msource *)ims;
915 lims->imsl_st[0] = lims->imsl_st[1];
917 imf->imf_st[0] = imf->imf_st[1];
921 * Reap unreferenced sources from socket-layer filter set.
924 imf_reap(struct in_mfilter *imf)
926 struct ip_msource *ims, *tims;
927 struct in_msource *lims;
929 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
930 lims = (struct in_msource *)ims;
931 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
932 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
933 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
934 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
935 free(ims, M_INMFILTER);
942 * Purge socket-layer filter set.
945 imf_purge(struct in_mfilter *imf)
947 struct ip_msource *ims, *tims;
949 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
950 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
951 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
952 free(ims, M_INMFILTER);
955 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
956 KASSERT(RB_EMPTY(&imf->imf_sources),
957 ("%s: imf_sources not empty", __func__));
961 * Look up a source filter entry for a multicast group.
963 * inm is the group descriptor to work with.
964 * haddr is the host-byte-order IPv4 address to look up.
965 * noalloc may be non-zero to suppress allocation of sources.
966 * *pims will be set to the address of the retrieved or allocated source.
968 * SMPng: NOTE: may be called with locks held.
969 * Return 0 if successful, otherwise return a non-zero error code.
972 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
973 const int noalloc, struct ip_msource **pims)
975 struct ip_msource find;
976 struct ip_msource *ims, *nims;
978 find.ims_haddr = haddr;
979 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
980 if (ims == NULL && !noalloc) {
981 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
983 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
987 nims->ims_haddr = haddr;
988 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
992 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
1002 * Merge socket-layer source into IGMP-layer source.
1003 * If rollback is non-zero, perform the inverse of the merge.
1006 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
1009 int n = rollback ? -1 : 1;
1011 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
1012 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
1013 __func__, n, ims->ims_haddr);
1014 ims->ims_st[1].ex -= n;
1015 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
1016 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
1017 __func__, n, ims->ims_haddr);
1018 ims->ims_st[1].in -= n;
1021 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
1022 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
1023 __func__, n, ims->ims_haddr);
1024 ims->ims_st[1].ex += n;
1025 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
1026 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
1027 __func__, n, ims->ims_haddr);
1028 ims->ims_st[1].in += n;
1033 * Atomically update the global in_multi state, when a membership's
1034 * filter list is being updated in any way.
1036 * imf is the per-inpcb-membership group filter pointer.
1037 * A fake imf may be passed for in-kernel consumers.
1039 * XXX This is a candidate for a set-symmetric-difference style loop
1040 * which would eliminate the repeated lookup from root of ims nodes,
1041 * as they share the same key space.
1043 * If any error occurred this function will back out of refcounts
1044 * and return a non-zero value.
1047 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1049 struct ip_msource *ims, *nims;
1050 struct in_msource *lims;
1051 int schanged, error;
1057 IN_MULTI_LIST_LOCK_ASSERT();
1060 * Update the source filters first, as this may fail.
1061 * Maintain count of in-mode filters at t0, t1. These are
1062 * used to work out if we transition into ASM mode or not.
1063 * Maintain a count of source filters whose state was
1064 * actually modified by this operation.
1066 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1067 lims = (struct in_msource *)ims;
1068 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1069 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1070 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1071 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1075 ims_merge(nims, lims, 0);
1078 struct ip_msource *bims;
1080 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1081 lims = (struct in_msource *)ims;
1082 if (lims->imsl_st[0] == lims->imsl_st[1])
1084 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1087 ims_merge(bims, lims, 1);
1092 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1093 __func__, nsrc0, nsrc1);
1095 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1096 if (imf->imf_st[0] == imf->imf_st[1] &&
1097 imf->imf_st[1] == MCAST_INCLUDE) {
1099 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1100 --inm->inm_st[1].iss_in;
1104 /* Handle filter mode transition on socket. */
1105 if (imf->imf_st[0] != imf->imf_st[1]) {
1106 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1107 __func__, imf->imf_st[0], imf->imf_st[1]);
1109 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1110 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1111 --inm->inm_st[1].iss_ex;
1112 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1113 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1114 --inm->inm_st[1].iss_in;
1117 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1118 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1119 inm->inm_st[1].iss_ex++;
1120 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1121 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1122 inm->inm_st[1].iss_in++;
1127 * Track inm filter state in terms of listener counts.
1128 * If there are any exclusive listeners, stack-wide
1129 * membership is exclusive.
1130 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1131 * If no listeners remain, state is undefined at t1,
1132 * and the IGMP lifecycle for this group should finish.
1134 if (inm->inm_st[1].iss_ex > 0) {
1135 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1136 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1137 } else if (inm->inm_st[1].iss_in > 0) {
1138 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1139 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1141 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1142 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1145 /* Decrement ASM listener count on transition out of ASM mode. */
1146 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1147 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1148 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1149 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1150 --inm->inm_st[1].iss_asm;
1154 /* Increment ASM listener count on transition to ASM mode. */
1155 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1156 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1157 inm->inm_st[1].iss_asm++;
1160 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1165 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1172 * Mark an in_multi's filter set deltas as committed.
1173 * Called by IGMP after a state change has been enqueued.
1176 inm_commit(struct in_multi *inm)
1178 struct ip_msource *ims;
1180 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1181 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1184 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1185 ims->ims_st[0] = ims->ims_st[1];
1187 inm->inm_st[0] = inm->inm_st[1];
1191 * Reap unreferenced nodes from an in_multi's filter set.
1194 inm_reap(struct in_multi *inm)
1196 struct ip_msource *ims, *tims;
1198 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1199 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1200 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1203 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1204 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1205 free(ims, M_IPMSOURCE);
1211 * Purge all source nodes from an in_multi's filter set.
1214 inm_purge(struct in_multi *inm)
1216 struct ip_msource *ims, *tims;
1218 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1219 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1220 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1221 free(ims, M_IPMSOURCE);
1227 * Join a multicast group; unlocked entry point.
1229 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1230 * is not held. Fortunately, ifp is unlikely to have been detached
1231 * at this point, so we assume it's OK to recurse.
1234 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1235 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1240 error = in_joingroup_locked(ifp, gina, imf, pinm);
1247 * Join a multicast group; real entry point.
1249 * Only preserves atomicity at inm level.
1250 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1252 * If the IGMP downcall fails, the group is not joined, and an error
1256 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1257 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1259 struct in_mfilter timf;
1260 struct in_multi *inm;
1263 IN_MULTI_LOCK_ASSERT();
1264 IN_MULTI_LIST_UNLOCK_ASSERT();
1266 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1267 ntohl(gina->s_addr), ifp, ifp->if_xname);
1273 * If no imf was specified (i.e. kernel consumer),
1274 * fake one up and assume it is an ASM join.
1277 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1281 error = in_getmulti(ifp, gina, &inm);
1283 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1286 IN_MULTI_LIST_LOCK();
1287 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1288 error = inm_merge(inm, imf);
1290 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1291 goto out_inm_release;
1294 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1295 error = igmp_change_state(inm);
1297 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1298 goto out_inm_release;
1304 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1305 inm_release_deferred(inm);
1309 IN_MULTI_LIST_UNLOCK();
1315 * Leave a multicast group; unlocked entry point.
1318 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1323 error = in_leavegroup_locked(inm, imf);
1330 * Leave a multicast group; real entry point.
1331 * All source filters will be expunged.
1333 * Only preserves atomicity at inm level.
1335 * Holding the write lock for the INP which contains imf
1336 * is highly advisable. We can't assert for it as imf does not
1337 * contain a back-pointer to the owning inp.
1339 * Note: This is not the same as inm_release(*) as this function also
1340 * makes a state change downcall into IGMP.
1343 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1345 struct in_mfilter timf;
1350 IN_MULTI_LOCK_ASSERT();
1351 IN_MULTI_LIST_UNLOCK_ASSERT();
1353 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1354 inm, ntohl(inm->inm_addr.s_addr),
1355 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1359 * If no imf was specified (i.e. kernel consumer),
1360 * fake one up and assume it is an ASM join.
1363 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1368 * Begin state merge transaction at IGMP layer.
1370 * As this particular invocation should not cause any memory
1371 * to be allocated, and there is no opportunity to roll back
1372 * the transaction, it MUST NOT fail.
1374 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1375 IN_MULTI_LIST_LOCK();
1376 error = inm_merge(inm, imf);
1377 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1379 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1380 CURVNET_SET(inm->inm_ifp->if_vnet);
1381 error = igmp_change_state(inm);
1382 IF_ADDR_WLOCK(inm->inm_ifp);
1383 inm_release_deferred(inm);
1384 IF_ADDR_WUNLOCK(inm->inm_ifp);
1385 IN_MULTI_LIST_UNLOCK();
1388 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1390 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1395 /*#ifndef BURN_BRIDGES*/
1397 * Join an IPv4 multicast group in (*,G) exclusive mode.
1398 * The group must be a 224.0.0.0/24 link-scope group.
1399 * This KPI is for legacy kernel consumers only.
1402 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1404 struct in_multi *pinm;
1407 char addrbuf[INET_ADDRSTRLEN];
1410 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1411 ("%s: %s not in 224.0.0.0/24", __func__,
1412 inet_ntoa_r(*ap, addrbuf)));
1414 error = in_joingroup(ifp, ap, NULL, &pinm);
1422 * Block or unblock an ASM multicast source on an inpcb.
1423 * This implements the delta-based API described in RFC 3678.
1425 * The delta-based API applies only to exclusive-mode memberships.
1426 * An IGMP downcall will be performed.
1428 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1430 * Return 0 if successful, otherwise return an appropriate error code.
1433 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1435 struct group_source_req gsr;
1436 sockunion_t *gsa, *ssa;
1438 struct in_mfilter *imf;
1439 struct ip_moptions *imo;
1440 struct in_msource *ims;
1441 struct in_multi *inm;
1450 memset(&gsr, 0, sizeof(struct group_source_req));
1451 gsa = (sockunion_t *)&gsr.gsr_group;
1452 ssa = (sockunion_t *)&gsr.gsr_source;
1454 switch (sopt->sopt_name) {
1455 case IP_BLOCK_SOURCE:
1456 case IP_UNBLOCK_SOURCE: {
1457 struct ip_mreq_source mreqs;
1459 error = sooptcopyin(sopt, &mreqs,
1460 sizeof(struct ip_mreq_source),
1461 sizeof(struct ip_mreq_source));
1465 gsa->sin.sin_family = AF_INET;
1466 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1467 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1469 ssa->sin.sin_family = AF_INET;
1470 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1471 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1473 if (!in_nullhost(mreqs.imr_interface))
1474 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1476 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1479 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1480 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1484 case MCAST_BLOCK_SOURCE:
1485 case MCAST_UNBLOCK_SOURCE:
1486 error = sooptcopyin(sopt, &gsr,
1487 sizeof(struct group_source_req),
1488 sizeof(struct group_source_req));
1492 if (gsa->sin.sin_family != AF_INET ||
1493 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1496 if (ssa->sin.sin_family != AF_INET ||
1497 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1500 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1501 return (EADDRNOTAVAIL);
1503 ifp = ifnet_byindex(gsr.gsr_interface);
1505 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1510 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1511 __func__, sopt->sopt_name);
1512 return (EOPNOTSUPP);
1516 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1520 * Check if we are actually a member of this group.
1522 imo = inp_findmoptions(inp);
1523 idx = imo_match_group(imo, ifp, &gsa->sa);
1524 if (idx == -1 || imo->imo_mfilters == NULL) {
1525 error = EADDRNOTAVAIL;
1526 goto out_inp_locked;
1529 KASSERT(imo->imo_mfilters != NULL,
1530 ("%s: imo_mfilters not allocated", __func__));
1531 imf = &imo->imo_mfilters[idx];
1532 inm = imo->imo_membership[idx];
1535 * Attempting to use the delta-based API on an
1536 * non exclusive-mode membership is an error.
1538 fmode = imf->imf_st[0];
1539 if (fmode != MCAST_EXCLUDE) {
1541 goto out_inp_locked;
1545 * Deal with error cases up-front:
1546 * Asked to block, but already blocked; or
1547 * Asked to unblock, but nothing to unblock.
1548 * If adding a new block entry, allocate it.
1550 ims = imo_match_source(imo, idx, &ssa->sa);
1551 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1552 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1553 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1554 error = EADDRNOTAVAIL;
1555 goto out_inp_locked;
1558 INP_WLOCK_ASSERT(inp);
1561 * Begin state merge transaction at socket layer.
1564 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1565 ims = imf_graft(imf, fmode, &ssa->sin);
1569 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1570 error = imf_prune(imf, &ssa->sin);
1574 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1575 goto out_imf_rollback;
1579 * Begin state merge transaction at IGMP layer.
1582 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1583 IN_MULTI_LIST_LOCK();
1584 error = inm_merge(inm, imf);
1586 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1587 IN_MULTI_LIST_UNLOCK();
1588 goto out_in_multi_locked;
1591 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1592 error = igmp_change_state(inm);
1593 IN_MULTI_LIST_UNLOCK();
1595 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1597 out_in_multi_locked:
1614 * Given an inpcb, return its multicast options structure pointer. Accepts
1615 * an unlocked inpcb pointer, but will return it locked. May sleep.
1617 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1618 * SMPng: NOTE: Returns with the INP write lock held.
1620 static struct ip_moptions *
1621 inp_findmoptions(struct inpcb *inp)
1623 struct ip_moptions *imo;
1624 struct in_multi **immp;
1625 struct in_mfilter *imfp;
1629 if (inp->inp_moptions != NULL)
1630 return (inp->inp_moptions);
1634 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1635 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1637 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1638 M_INMFILTER, M_WAITOK);
1640 imo->imo_multicast_ifp = NULL;
1641 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1642 imo->imo_multicast_vif = -1;
1643 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1644 imo->imo_multicast_loop = in_mcast_loop;
1645 imo->imo_num_memberships = 0;
1646 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1647 imo->imo_membership = immp;
1649 /* Initialize per-group source filters. */
1650 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1651 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1652 imo->imo_mfilters = imfp;
1655 if (inp->inp_moptions != NULL) {
1656 free(imfp, M_INMFILTER);
1657 free(immp, M_IPMOPTS);
1658 free(imo, M_IPMOPTS);
1659 return (inp->inp_moptions);
1661 inp->inp_moptions = imo;
1666 inp_gcmoptions(epoch_context_t ctx)
1668 struct ip_moptions *imo;
1669 struct in_mfilter *imf;
1670 struct in_multi *inm;
1672 size_t idx, nmships;
1674 imo = __containerof(ctx, struct ip_moptions, imo_epoch_ctx);
1676 nmships = imo->imo_num_memberships;
1677 for (idx = 0; idx < nmships; ++idx) {
1678 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1681 inm = imo->imo_membership[idx];
1684 CURVNET_SET(ifp->if_vnet);
1685 (void)in_leavegroup(inm, imf);
1688 (void)in_leavegroup(inm, imf);
1694 if (imo->imo_mfilters)
1695 free(imo->imo_mfilters, M_INMFILTER);
1696 free(imo->imo_membership, M_IPMOPTS);
1697 free(imo, M_IPMOPTS);
1701 * Discard the IP multicast options (and source filters). To minimize
1702 * the amount of work done while holding locks such as the INP's
1703 * pcbinfo lock (which is used in the receive path), the free
1704 * operation is deferred to the epoch callback task.
1707 inp_freemoptions(struct ip_moptions *imo)
1711 epoch_call(net_epoch_preempt, &imo->imo_epoch_ctx, inp_gcmoptions);
1715 * Atomically get source filters on a socket for an IPv4 multicast group.
1716 * Called with INP lock held; returns with lock released.
1719 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1721 struct __msfilterreq msfr;
1724 struct ip_moptions *imo;
1725 struct in_mfilter *imf;
1726 struct ip_msource *ims;
1727 struct in_msource *lims;
1728 struct sockaddr_in *psin;
1729 struct sockaddr_storage *ptss;
1730 struct sockaddr_storage *tss;
1732 size_t idx, nsrcs, ncsrcs;
1734 INP_WLOCK_ASSERT(inp);
1736 imo = inp->inp_moptions;
1737 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1741 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1742 sizeof(struct __msfilterreq));
1746 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1749 ifp = ifnet_byindex(msfr.msfr_ifindex);
1756 * Lookup group on the socket.
1758 gsa = (sockunion_t *)&msfr.msfr_group;
1759 idx = imo_match_group(imo, ifp, &gsa->sa);
1760 if (idx == -1 || imo->imo_mfilters == NULL) {
1762 return (EADDRNOTAVAIL);
1764 imf = &imo->imo_mfilters[idx];
1767 * Ignore memberships which are in limbo.
1769 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1773 msfr.msfr_fmode = imf->imf_st[1];
1776 * If the user specified a buffer, copy out the source filter
1777 * entries to userland gracefully.
1778 * We only copy out the number of entries which userland
1779 * has asked for, but we always tell userland how big the
1780 * buffer really needs to be.
1782 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1783 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1785 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1786 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1787 M_TEMP, M_NOWAIT | M_ZERO);
1795 * Count number of sources in-mode at t0.
1796 * If buffer space exists and remains, copy out source entries.
1798 nsrcs = msfr.msfr_nsrcs;
1801 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1802 lims = (struct in_msource *)ims;
1803 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1804 lims->imsl_st[0] != imf->imf_st[0])
1807 if (tss != NULL && nsrcs > 0) {
1808 psin = (struct sockaddr_in *)ptss;
1809 psin->sin_family = AF_INET;
1810 psin->sin_len = sizeof(struct sockaddr_in);
1811 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1821 error = copyout(tss, msfr.msfr_srcs,
1822 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1828 msfr.msfr_nsrcs = ncsrcs;
1829 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1835 * Return the IP multicast options in response to user getsockopt().
1838 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1840 struct rm_priotracker in_ifa_tracker;
1841 struct ip_mreqn mreqn;
1842 struct ip_moptions *imo;
1844 struct in_ifaddr *ia;
1849 imo = inp->inp_moptions;
1851 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1852 * or is a divert socket, reject it.
1854 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1855 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1856 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1858 return (EOPNOTSUPP);
1862 switch (sopt->sopt_name) {
1863 case IP_MULTICAST_VIF:
1865 optval = imo->imo_multicast_vif;
1869 error = sooptcopyout(sopt, &optval, sizeof(int));
1872 case IP_MULTICAST_IF:
1873 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1875 ifp = imo->imo_multicast_ifp;
1876 if (!in_nullhost(imo->imo_multicast_addr)) {
1877 mreqn.imr_address = imo->imo_multicast_addr;
1878 } else if (ifp != NULL) {
1879 mreqn.imr_ifindex = ifp->if_index;
1881 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1884 IA_SIN(ia)->sin_addr;
1889 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1890 error = sooptcopyout(sopt, &mreqn,
1891 sizeof(struct ip_mreqn));
1893 error = sooptcopyout(sopt, &mreqn.imr_address,
1894 sizeof(struct in_addr));
1898 case IP_MULTICAST_TTL:
1900 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1902 optval = coptval = imo->imo_multicast_ttl;
1904 if (sopt->sopt_valsize == sizeof(u_char))
1905 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1907 error = sooptcopyout(sopt, &optval, sizeof(int));
1910 case IP_MULTICAST_LOOP:
1912 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1914 optval = coptval = imo->imo_multicast_loop;
1916 if (sopt->sopt_valsize == sizeof(u_char))
1917 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1919 error = sooptcopyout(sopt, &optval, sizeof(int));
1924 error = EADDRNOTAVAIL;
1927 error = inp_get_source_filters(inp, sopt);
1933 error = ENOPROTOOPT;
1937 INP_UNLOCK_ASSERT(inp);
1943 * Look up the ifnet to use for a multicast group membership,
1944 * given the IPv4 address of an interface, and the IPv4 group address.
1946 * This routine exists to support legacy multicast applications
1947 * which do not understand that multicast memberships are scoped to
1948 * specific physical links in the networking stack, or which need
1949 * to join link-scope groups before IPv4 addresses are configured.
1951 * If inp is non-NULL, use this socket's current FIB number for any
1952 * required FIB lookup.
1953 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1954 * and use its ifp; usually, this points to the default next-hop.
1956 * If the FIB lookup fails, attempt to use the first non-loopback
1957 * interface with multicast capability in the system as a
1958 * last resort. The legacy IPv4 ASM API requires that we do
1959 * this in order to allow groups to be joined when the routing
1960 * table has not yet been populated during boot.
1962 * Returns NULL if no ifp could be found.
1964 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1965 * FUTURE: Implement IPv4 source-address selection.
1967 static struct ifnet *
1968 inp_lookup_mcast_ifp(const struct inpcb *inp,
1969 const struct sockaddr_in *gsin, const struct in_addr ina)
1971 struct rm_priotracker in_ifa_tracker;
1973 struct nhop4_basic nh4;
1976 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1977 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1978 ("%s: not multicast", __func__));
1981 if (!in_nullhost(ina)) {
1982 INADDR_TO_IFP(ina, ifp);
1984 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1985 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1988 struct in_ifaddr *ia;
1992 IN_IFADDR_RLOCK(&in_ifa_tracker);
1993 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1995 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1996 (mifp->if_flags & IFF_MULTICAST)) {
2001 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2009 * Join an IPv4 multicast group, possibly with a source.
2012 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
2014 struct group_source_req gsr;
2015 sockunion_t *gsa, *ssa;
2017 struct in_mfilter *imf;
2018 struct ip_moptions *imo;
2019 struct in_multi *inm;
2020 struct in_msource *lims;
2030 memset(&gsr, 0, sizeof(struct group_source_req));
2031 gsa = (sockunion_t *)&gsr.gsr_group;
2032 gsa->ss.ss_family = AF_UNSPEC;
2033 ssa = (sockunion_t *)&gsr.gsr_source;
2034 ssa->ss.ss_family = AF_UNSPEC;
2036 switch (sopt->sopt_name) {
2037 case IP_ADD_MEMBERSHIP:
2038 case IP_ADD_SOURCE_MEMBERSHIP: {
2039 struct ip_mreq_source mreqs;
2041 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
2042 error = sooptcopyin(sopt, &mreqs,
2043 sizeof(struct ip_mreq),
2044 sizeof(struct ip_mreq));
2046 * Do argument switcharoo from ip_mreq into
2047 * ip_mreq_source to avoid using two instances.
2049 mreqs.imr_interface = mreqs.imr_sourceaddr;
2050 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2051 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2052 error = sooptcopyin(sopt, &mreqs,
2053 sizeof(struct ip_mreq_source),
2054 sizeof(struct ip_mreq_source));
2059 gsa->sin.sin_family = AF_INET;
2060 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2061 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2063 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2064 ssa->sin.sin_family = AF_INET;
2065 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2066 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2069 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2072 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2073 mreqs.imr_interface);
2074 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2075 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2079 case MCAST_JOIN_GROUP:
2080 case MCAST_JOIN_SOURCE_GROUP:
2081 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2082 error = sooptcopyin(sopt, &gsr,
2083 sizeof(struct group_req),
2084 sizeof(struct group_req));
2085 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2086 error = sooptcopyin(sopt, &gsr,
2087 sizeof(struct group_source_req),
2088 sizeof(struct group_source_req));
2093 if (gsa->sin.sin_family != AF_INET ||
2094 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2098 * Overwrite the port field if present, as the sockaddr
2099 * being copied in may be matched with a binary comparison.
2101 gsa->sin.sin_port = 0;
2102 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2103 if (ssa->sin.sin_family != AF_INET ||
2104 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2106 ssa->sin.sin_port = 0;
2109 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2112 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2113 return (EADDRNOTAVAIL);
2114 ifp = ifnet_byindex(gsr.gsr_interface);
2118 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2119 __func__, sopt->sopt_name);
2120 return (EOPNOTSUPP);
2124 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2125 return (EADDRNOTAVAIL);
2127 imo = inp_findmoptions(inp);
2128 idx = imo_match_group(imo, ifp, &gsa->sa);
2132 inm = imo->imo_membership[idx];
2133 imf = &imo->imo_mfilters[idx];
2134 if (ssa->ss.ss_family != AF_UNSPEC) {
2136 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2137 * is an error. On an existing inclusive membership,
2138 * it just adds the source to the filter list.
2140 if (imf->imf_st[1] != MCAST_INCLUDE) {
2142 goto out_inp_locked;
2145 * Throw out duplicates.
2147 * XXX FIXME: This makes a naive assumption that
2148 * even if entries exist for *ssa in this imf,
2149 * they will be rejected as dupes, even if they
2150 * are not valid in the current mode (in-mode).
2152 * in_msource is transactioned just as for anything
2153 * else in SSM -- but note naive use of inm_graft()
2154 * below for allocating new filter entries.
2156 * This is only an issue if someone mixes the
2157 * full-state SSM API with the delta-based API,
2158 * which is discouraged in the relevant RFCs.
2160 lims = imo_match_source(imo, idx, &ssa->sa);
2161 if (lims != NULL /*&&
2162 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2163 error = EADDRNOTAVAIL;
2164 goto out_inp_locked;
2168 * MCAST_JOIN_GROUP on an existing exclusive
2169 * membership is an error; return EADDRINUSE
2170 * to preserve 4.4BSD API idempotence, and
2171 * avoid tedious detour to code below.
2172 * NOTE: This is bending RFC 3678 a bit.
2174 * On an existing inclusive membership, this is also
2175 * an error; if you want to change filter mode,
2176 * you must use the userland API setsourcefilter().
2177 * XXX We don't reject this for imf in UNDEFINED
2178 * state at t1, because allocation of a filter
2179 * is atomic with allocation of a membership.
2182 if (imf->imf_st[1] == MCAST_EXCLUDE)
2184 goto out_inp_locked;
2189 * Begin state merge transaction at socket layer.
2191 INP_WLOCK_ASSERT(inp);
2194 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2195 error = imo_grow(imo);
2197 goto out_inp_locked;
2200 * Allocate the new slot upfront so we can deal with
2201 * grafting the new source filter in same code path
2202 * as for join-source on existing membership.
2204 idx = imo->imo_num_memberships;
2205 imo->imo_membership[idx] = NULL;
2206 imo->imo_num_memberships++;
2207 KASSERT(imo->imo_mfilters != NULL,
2208 ("%s: imf_mfilters vector was not allocated", __func__));
2209 imf = &imo->imo_mfilters[idx];
2210 KASSERT(RB_EMPTY(&imf->imf_sources),
2211 ("%s: imf_sources not empty", __func__));
2215 * Graft new source into filter list for this inpcb's
2216 * membership of the group. The in_multi may not have
2217 * been allocated yet if this is a new membership, however,
2218 * the in_mfilter slot will be allocated and must be initialized.
2220 * Note: Grafting of exclusive mode filters doesn't happen
2222 * XXX: Should check for non-NULL lims (node exists but may
2223 * not be in-mode) for interop with full-state API.
2225 if (ssa->ss.ss_family != AF_UNSPEC) {
2226 /* Membership starts in IN mode */
2228 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2229 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2231 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2233 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2235 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2241 /* No address specified; Membership starts in EX mode */
2243 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2244 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2249 * Begin state merge transaction at IGMP layer.
2256 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2259 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2261 IN_MULTI_LIST_UNLOCK();
2264 imo->imo_membership[idx] = inm;
2266 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2267 IN_MULTI_LIST_LOCK();
2268 error = inm_merge(inm, imf);
2270 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2272 IN_MULTI_LIST_UNLOCK();
2273 goto out_in_multi_locked;
2275 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2276 error = igmp_change_state(inm);
2277 IN_MULTI_LIST_UNLOCK();
2279 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2281 goto out_in_multi_locked;
2285 out_in_multi_locked:
2289 if (in_pcbrele_wlocked(inp))
2302 if (error && is_new) {
2303 imo->imo_membership[idx] = NULL;
2304 --imo->imo_num_memberships;
2313 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2316 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2318 struct group_source_req gsr;
2319 struct ip_mreq_source mreqs;
2320 sockunion_t *gsa, *ssa;
2322 struct in_mfilter *imf;
2323 struct ip_moptions *imo;
2324 struct in_msource *ims;
2325 struct in_multi *inm;
2327 int error, is_final;
2333 memset(&gsr, 0, sizeof(struct group_source_req));
2334 gsa = (sockunion_t *)&gsr.gsr_group;
2335 gsa->ss.ss_family = AF_UNSPEC;
2336 ssa = (sockunion_t *)&gsr.gsr_source;
2337 ssa->ss.ss_family = AF_UNSPEC;
2339 switch (sopt->sopt_name) {
2340 case IP_DROP_MEMBERSHIP:
2341 case IP_DROP_SOURCE_MEMBERSHIP:
2342 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2343 error = sooptcopyin(sopt, &mreqs,
2344 sizeof(struct ip_mreq),
2345 sizeof(struct ip_mreq));
2347 * Swap interface and sourceaddr arguments,
2348 * as ip_mreq and ip_mreq_source are laid
2351 mreqs.imr_interface = mreqs.imr_sourceaddr;
2352 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2353 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2354 error = sooptcopyin(sopt, &mreqs,
2355 sizeof(struct ip_mreq_source),
2356 sizeof(struct ip_mreq_source));
2361 gsa->sin.sin_family = AF_INET;
2362 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2363 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2365 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2366 ssa->sin.sin_family = AF_INET;
2367 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2368 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2372 * Attempt to look up hinted ifp from interface address.
2373 * Fallthrough with null ifp iff lookup fails, to
2374 * preserve 4.4BSD mcast API idempotence.
2375 * XXX NOTE WELL: The RFC 3678 API is preferred because
2376 * using an IPv4 address as a key is racy.
2378 if (!in_nullhost(mreqs.imr_interface))
2379 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2381 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2382 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2386 case MCAST_LEAVE_GROUP:
2387 case MCAST_LEAVE_SOURCE_GROUP:
2388 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2389 error = sooptcopyin(sopt, &gsr,
2390 sizeof(struct group_req),
2391 sizeof(struct group_req));
2392 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2393 error = sooptcopyin(sopt, &gsr,
2394 sizeof(struct group_source_req),
2395 sizeof(struct group_source_req));
2400 if (gsa->sin.sin_family != AF_INET ||
2401 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2404 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2405 if (ssa->sin.sin_family != AF_INET ||
2406 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2410 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2411 return (EADDRNOTAVAIL);
2413 ifp = ifnet_byindex(gsr.gsr_interface);
2416 return (EADDRNOTAVAIL);
2420 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2421 __func__, sopt->sopt_name);
2422 return (EOPNOTSUPP);
2426 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2430 * Find the membership in the membership array.
2432 imo = inp_findmoptions(inp);
2433 idx = imo_match_group(imo, ifp, &gsa->sa);
2435 error = EADDRNOTAVAIL;
2436 goto out_inp_locked;
2438 inm = imo->imo_membership[idx];
2439 imf = &imo->imo_mfilters[idx];
2441 if (ssa->ss.ss_family != AF_UNSPEC)
2445 * Begin state merge transaction at socket layer.
2447 INP_WLOCK_ASSERT(inp);
2450 * If we were instructed only to leave a given source, do so.
2451 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2456 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2457 error = EADDRNOTAVAIL;
2458 goto out_inp_locked;
2460 ims = imo_match_source(imo, idx, &ssa->sa);
2462 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2463 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2464 error = EADDRNOTAVAIL;
2465 goto out_inp_locked;
2467 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2468 error = imf_prune(imf, &ssa->sin);
2470 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2472 goto out_inp_locked;
2477 * Begin state merge transaction at IGMP layer.
2485 * Give up the multicast address record to which
2486 * the membership points.
2488 (void)in_leavegroup_locked(inm, imf);
2490 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2491 IN_MULTI_LIST_LOCK();
2492 error = inm_merge(inm, imf);
2494 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2496 IN_MULTI_LIST_UNLOCK();
2497 goto out_in_multi_locked;
2500 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2501 error = igmp_change_state(inm);
2502 IN_MULTI_LIST_UNLOCK();
2504 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2509 out_in_multi_locked:
2513 if (in_pcbrele_wlocked(inp))
2524 /* Remove the gap in the membership and filter array. */
2525 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2526 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2527 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2529 imo->imo_num_memberships--;
2538 * Select the interface for transmitting IPv4 multicast datagrams.
2540 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2541 * may be passed to this socket option. An address of INADDR_ANY or an
2542 * interface index of 0 is used to remove a previous selection.
2543 * When no interface is selected, one is chosen for every send.
2546 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2548 struct in_addr addr;
2549 struct ip_mreqn mreqn;
2551 struct ip_moptions *imo;
2554 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2556 * An interface index was specified using the
2557 * Linux-derived ip_mreqn structure.
2559 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2560 sizeof(struct ip_mreqn));
2564 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2567 if (mreqn.imr_ifindex == 0) {
2570 ifp = ifnet_byindex(mreqn.imr_ifindex);
2572 return (EADDRNOTAVAIL);
2576 * An interface was specified by IPv4 address.
2577 * This is the traditional BSD usage.
2579 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2580 sizeof(struct in_addr));
2583 if (in_nullhost(addr)) {
2586 INADDR_TO_IFP(addr, ifp);
2588 return (EADDRNOTAVAIL);
2590 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2591 ntohl(addr.s_addr));
2594 /* Reject interfaces which do not support multicast. */
2595 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2596 return (EOPNOTSUPP);
2598 imo = inp_findmoptions(inp);
2599 imo->imo_multicast_ifp = ifp;
2600 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2607 * Atomically set source filters on a socket for an IPv4 multicast group.
2609 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2612 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2614 struct __msfilterreq msfr;
2617 struct in_mfilter *imf;
2618 struct ip_moptions *imo;
2619 struct in_multi *inm;
2623 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2624 sizeof(struct __msfilterreq));
2628 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2631 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2632 msfr.msfr_fmode != MCAST_INCLUDE))
2635 if (msfr.msfr_group.ss_family != AF_INET ||
2636 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2639 gsa = (sockunion_t *)&msfr.msfr_group;
2640 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2643 gsa->sin.sin_port = 0; /* ignore port */
2645 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2646 return (EADDRNOTAVAIL);
2648 ifp = ifnet_byindex(msfr.msfr_ifindex);
2650 return (EADDRNOTAVAIL);
2653 * Take the INP write lock.
2654 * Check if this socket is a member of this group.
2656 imo = inp_findmoptions(inp);
2657 idx = imo_match_group(imo, ifp, &gsa->sa);
2658 if (idx == -1 || imo->imo_mfilters == NULL) {
2659 error = EADDRNOTAVAIL;
2660 goto out_inp_locked;
2662 inm = imo->imo_membership[idx];
2663 imf = &imo->imo_mfilters[idx];
2666 * Begin state merge transaction at socket layer.
2668 INP_WLOCK_ASSERT(inp);
2670 imf->imf_st[1] = msfr.msfr_fmode;
2673 * Apply any new source filters, if present.
2674 * Make a copy of the user-space source vector so
2675 * that we may copy them with a single copyin. This
2676 * allows us to deal with page faults up-front.
2678 if (msfr.msfr_nsrcs > 0) {
2679 struct in_msource *lims;
2680 struct sockaddr_in *psin;
2681 struct sockaddr_storage *kss, *pkss;
2686 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2687 __func__, (unsigned long)msfr.msfr_nsrcs);
2688 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2690 error = copyin(msfr.msfr_srcs, kss,
2691 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2700 * Mark all source filters as UNDEFINED at t1.
2701 * Restore new group filter mode, as imf_leave()
2702 * will set it to INCLUDE.
2705 imf->imf_st[1] = msfr.msfr_fmode;
2708 * Update socket layer filters at t1, lazy-allocating
2709 * new entries. This saves a bunch of memory at the
2710 * cost of one RB_FIND() per source entry; duplicate
2711 * entries in the msfr_nsrcs vector are ignored.
2712 * If we encounter an error, rollback transaction.
2714 * XXX This too could be replaced with a set-symmetric
2715 * difference like loop to avoid walking from root
2716 * every time, as the key space is common.
2718 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2719 psin = (struct sockaddr_in *)pkss;
2720 if (psin->sin_family != AF_INET) {
2721 error = EAFNOSUPPORT;
2724 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2728 error = imf_get_source(imf, psin, &lims);
2731 lims->imsl_st[1] = imf->imf_st[1];
2737 goto out_imf_rollback;
2739 INP_WLOCK_ASSERT(inp);
2743 * Begin state merge transaction at IGMP layer.
2745 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2746 IN_MULTI_LIST_LOCK();
2747 error = inm_merge(inm, imf);
2749 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2750 IN_MULTI_LIST_UNLOCK();
2751 goto out_in_multi_locked;
2754 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2755 error = igmp_change_state(inm);
2756 IN_MULTI_LIST_UNLOCK();
2758 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2760 out_in_multi_locked:
2778 * Set the IP multicast options in response to user setsockopt().
2780 * Many of the socket options handled in this function duplicate the
2781 * functionality of socket options in the regular unicast API. However,
2782 * it is not possible to merge the duplicate code, because the idempotence
2783 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2784 * the effects of these options must be treated as separate and distinct.
2786 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2787 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2788 * is refactored to no longer use vifs.
2791 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2793 struct ip_moptions *imo;
2799 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2800 * or is a divert socket, reject it.
2802 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2803 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2804 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2805 return (EOPNOTSUPP);
2807 switch (sopt->sopt_name) {
2808 case IP_MULTICAST_VIF: {
2811 * Select a multicast VIF for transmission.
2812 * Only useful if multicast forwarding is active.
2814 if (legal_vif_num == NULL) {
2818 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2821 if (!legal_vif_num(vifi) && (vifi != -1)) {
2825 imo = inp_findmoptions(inp);
2826 imo->imo_multicast_vif = vifi;
2831 case IP_MULTICAST_IF:
2832 error = inp_set_multicast_if(inp, sopt);
2835 case IP_MULTICAST_TTL: {
2839 * Set the IP time-to-live for outgoing multicast packets.
2840 * The original multicast API required a char argument,
2841 * which is inconsistent with the rest of the socket API.
2842 * We allow either a char or an int.
2844 if (sopt->sopt_valsize == sizeof(u_char)) {
2845 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2852 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2862 imo = inp_findmoptions(inp);
2863 imo->imo_multicast_ttl = ttl;
2868 case IP_MULTICAST_LOOP: {
2872 * Set the loopback flag for outgoing multicast packets.
2873 * Must be zero or one. The original multicast API required a
2874 * char argument, which is inconsistent with the rest
2875 * of the socket API. We allow either a char or an int.
2877 if (sopt->sopt_valsize == sizeof(u_char)) {
2878 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2885 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2889 loop = (u_char)iloop;
2891 imo = inp_findmoptions(inp);
2892 imo->imo_multicast_loop = !!loop;
2897 case IP_ADD_MEMBERSHIP:
2898 case IP_ADD_SOURCE_MEMBERSHIP:
2899 case MCAST_JOIN_GROUP:
2900 case MCAST_JOIN_SOURCE_GROUP:
2901 error = inp_join_group(inp, sopt);
2904 case IP_DROP_MEMBERSHIP:
2905 case IP_DROP_SOURCE_MEMBERSHIP:
2906 case MCAST_LEAVE_GROUP:
2907 case MCAST_LEAVE_SOURCE_GROUP:
2908 error = inp_leave_group(inp, sopt);
2911 case IP_BLOCK_SOURCE:
2912 case IP_UNBLOCK_SOURCE:
2913 case MCAST_BLOCK_SOURCE:
2914 case MCAST_UNBLOCK_SOURCE:
2915 error = inp_block_unblock_source(inp, sopt);
2919 error = inp_set_source_filters(inp, sopt);
2927 INP_UNLOCK_ASSERT(inp);
2933 * Expose IGMP's multicast filter mode and source list(s) to userland,
2934 * keyed by (ifindex, group).
2935 * The filter mode is written out as a uint32_t, followed by
2936 * 0..n of struct in_addr.
2937 * For use by ifmcstat(8).
2938 * SMPng: NOTE: unlocked read of ifindex space.
2941 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2943 struct in_addr src, group;
2945 struct ifmultiaddr *ifma;
2946 struct in_multi *inm;
2947 struct ip_msource *ims;
2951 uint32_t fmode, ifindex;
2956 if (req->newptr != NULL)
2963 if (ifindex <= 0 || ifindex > V_if_index) {
2964 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2969 group.s_addr = name[1];
2970 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2971 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2972 __func__, ntohl(group.s_addr));
2976 ifp = ifnet_byindex(ifindex);
2978 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2983 retval = sysctl_wire_old_buffer(req,
2984 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2988 IN_MULTI_LIST_LOCK();
2991 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2992 if (ifma->ifma_addr->sa_family != AF_INET ||
2993 ifma->ifma_protospec == NULL)
2995 inm = (struct in_multi *)ifma->ifma_protospec;
2996 if (!in_hosteq(inm->inm_addr, group))
2998 fmode = inm->inm_st[1].iss_fmode;
2999 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
3002 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
3003 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
3006 * Only copy-out sources which are in-mode.
3008 if (fmode != ims_get_mode(inm, ims, 1)) {
3009 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
3013 src.s_addr = htonl(ims->ims_haddr);
3014 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
3019 IF_ADDR_RUNLOCK(ifp);
3021 IN_MULTI_LIST_UNLOCK();
3026 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
3028 static const char *inm_modestrs[] = { "un", "in", "ex" };
3031 inm_mode_str(const int mode)
3034 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
3035 return (inm_modestrs[mode]);
3039 static const char *inm_statestrs[] = {
3052 inm_state_str(const int state)
3055 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3056 return (inm_statestrs[state]);
3061 * Dump an in_multi structure to the console.
3064 inm_print(const struct in_multi *inm)
3067 char addrbuf[INET_ADDRSTRLEN];
3069 if ((ktr_mask & KTR_IGMPV3) == 0)
3072 printf("%s: --- begin inm %p ---\n", __func__, inm);
3073 printf("addr %s ifp %p(%s) ifma %p\n",
3074 inet_ntoa_r(inm->inm_addr, addrbuf),
3076 inm->inm_ifp->if_xname,
3078 printf("timer %u state %s refcount %u scq.len %u\n",
3080 inm_state_str(inm->inm_state),
3082 inm->inm_scq.mq_len);
3083 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3088 for (t = 0; t < 2; t++) {
3089 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3090 inm_mode_str(inm->inm_st[t].iss_fmode),
3091 inm->inm_st[t].iss_asm,
3092 inm->inm_st[t].iss_ex,
3093 inm->inm_st[t].iss_in,
3094 inm->inm_st[t].iss_rec);
3096 printf("%s: --- end inm %p ---\n", __func__, inm);
3099 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3102 inm_print(const struct in_multi *inm)
3107 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3109 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);