in_mcast.c revision 227309
1/*-
2 * Copyright (c) 2007-2009 Bruce Simpson.
3 * Copyright (c) 2005 Robert N. M. Watson.
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote
15 *    products derived from this software without specific prior written
16 *    permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31/*
32 * IPv4 multicast socket, group, and socket option processing module.
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD: head/sys/netinet/in_mcast.c 227309 2011-11-07 15:43:11Z ed $");
37
38#include <sys/param.h>
39#include <sys/systm.h>
40#include <sys/kernel.h>
41#include <sys/malloc.h>
42#include <sys/mbuf.h>
43#include <sys/protosw.h>
44#include <sys/socket.h>
45#include <sys/socketvar.h>
46#include <sys/protosw.h>
47#include <sys/sysctl.h>
48#include <sys/ktr.h>
49#include <sys/tree.h>
50
51#include <net/if.h>
52#include <net/if_dl.h>
53#include <net/route.h>
54#include <net/vnet.h>
55
56#include <netinet/in.h>
57#include <netinet/in_systm.h>
58#include <netinet/in_pcb.h>
59#include <netinet/in_var.h>
60#include <netinet/ip_var.h>
61#include <netinet/igmp_var.h>
62
63#ifndef KTR_IGMPV3
64#define KTR_IGMPV3 KTR_INET
65#endif
66
67#ifndef __SOCKUNION_DECLARED
68union sockunion {
69	struct sockaddr_storage	ss;
70	struct sockaddr		sa;
71	struct sockaddr_dl	sdl;
72	struct sockaddr_in	sin;
73};
74typedef union sockunion sockunion_t;
75#define __SOCKUNION_DECLARED
76#endif /* __SOCKUNION_DECLARED */
77
78static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
79    "IPv4 multicast PCB-layer source filter");
80static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
81static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
82static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
83    "IPv4 multicast IGMP-layer source filter");
84
85/*
86 * Locking:
87 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
88 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
89 *   it can be taken by code in net/if.c also.
90 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
91 *
92 * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly
93 * any need for in_multi itself to be virtualized -- it is bound to an ifp
94 * anyway no matter what happens.
95 */
96struct mtx in_multi_mtx;
97MTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF);
98
99/*
100 * Functions with non-static linkage defined in this file should be
101 * declared in in_var.h:
102 *  imo_multi_filter()
103 *  in_addmulti()
104 *  in_delmulti()
105 *  in_joingroup()
106 *  in_joingroup_locked()
107 *  in_leavegroup()
108 *  in_leavegroup_locked()
109 * and ip_var.h:
110 *  inp_freemoptions()
111 *  inp_getmoptions()
112 *  inp_setmoptions()
113 *
114 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
115 * and in_delmulti().
116 */
117static void	imf_commit(struct in_mfilter *);
118static int	imf_get_source(struct in_mfilter *imf,
119		    const struct sockaddr_in *psin,
120		    struct in_msource **);
121static struct in_msource *
122		imf_graft(struct in_mfilter *, const uint8_t,
123		    const struct sockaddr_in *);
124static void	imf_leave(struct in_mfilter *);
125static int	imf_prune(struct in_mfilter *, const struct sockaddr_in *);
126static void	imf_purge(struct in_mfilter *);
127static void	imf_rollback(struct in_mfilter *);
128static void	imf_reap(struct in_mfilter *);
129static int	imo_grow(struct ip_moptions *);
130static size_t	imo_match_group(const struct ip_moptions *,
131		    const struct ifnet *, const struct sockaddr *);
132static struct in_msource *
133		imo_match_source(const struct ip_moptions *, const size_t,
134		    const struct sockaddr *);
135static void	ims_merge(struct ip_msource *ims,
136		    const struct in_msource *lims, const int rollback);
137static int	in_getmulti(struct ifnet *, const struct in_addr *,
138		    struct in_multi **);
139static int	inm_get_source(struct in_multi *inm, const in_addr_t haddr,
140		    const int noalloc, struct ip_msource **pims);
141static int	inm_is_ifp_detached(const struct in_multi *);
142static int	inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
143static void	inm_purge(struct in_multi *);
144static void	inm_reap(struct in_multi *);
145static struct ip_moptions *
146		inp_findmoptions(struct inpcb *);
147static int	inp_get_source_filters(struct inpcb *, struct sockopt *);
148static int	inp_join_group(struct inpcb *, struct sockopt *);
149static int	inp_leave_group(struct inpcb *, struct sockopt *);
150static struct ifnet *
151		inp_lookup_mcast_ifp(const struct inpcb *,
152		    const struct sockaddr_in *, const struct in_addr);
153static int	inp_block_unblock_source(struct inpcb *, struct sockopt *);
154static int	inp_set_multicast_if(struct inpcb *, struct sockopt *);
155static int	inp_set_source_filters(struct inpcb *, struct sockopt *);
156static int	sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
157
158static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
159    "IPv4 multicast");
160
161static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
162SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
163    CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxgrpsrc, 0,
164    "Max source filters per group");
165TUNABLE_ULONG("net.inet.ip.mcast.maxgrpsrc", &in_mcast_maxgrpsrc);
166
167static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
168SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
169    CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxsocksrc, 0,
170    "Max source filters per socket");
171TUNABLE_ULONG("net.inet.ip.mcast.maxsocksrc", &in_mcast_maxsocksrc);
172
173int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
174SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
175    &in_mcast_loop, 0, "Loopback multicast datagrams by default");
176TUNABLE_INT("net.inet.ip.mcast.loop", &in_mcast_loop);
177
178static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
179    CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
180    "Per-interface stack-wide source filters");
181
182/*
183 * Inline function which wraps assertions for a valid ifp.
184 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
185 * is detached.
186 */
187static int __inline
188inm_is_ifp_detached(const struct in_multi *inm)
189{
190	struct ifnet *ifp;
191
192	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
193	ifp = inm->inm_ifma->ifma_ifp;
194	if (ifp != NULL) {
195		/*
196		 * Sanity check that netinet's notion of ifp is the
197		 * same as net's.
198		 */
199		KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
200	}
201
202	return (ifp == NULL);
203}
204
205/*
206 * Initialize an in_mfilter structure to a known state at t0, t1
207 * with an empty source filter list.
208 */
209static __inline void
210imf_init(struct in_mfilter *imf, const int st0, const int st1)
211{
212	memset(imf, 0, sizeof(struct in_mfilter));
213	RB_INIT(&imf->imf_sources);
214	imf->imf_st[0] = st0;
215	imf->imf_st[1] = st1;
216}
217
218/*
219 * Resize the ip_moptions vector to the next power-of-two minus 1.
220 * May be called with locks held; do not sleep.
221 */
222static int
223imo_grow(struct ip_moptions *imo)
224{
225	struct in_multi		**nmships;
226	struct in_multi		**omships;
227	struct in_mfilter	 *nmfilters;
228	struct in_mfilter	 *omfilters;
229	size_t			  idx;
230	size_t			  newmax;
231	size_t			  oldmax;
232
233	nmships = NULL;
234	nmfilters = NULL;
235	omships = imo->imo_membership;
236	omfilters = imo->imo_mfilters;
237	oldmax = imo->imo_max_memberships;
238	newmax = ((oldmax + 1) * 2) - 1;
239
240	if (newmax <= IP_MAX_MEMBERSHIPS) {
241		nmships = (struct in_multi **)realloc(omships,
242		    sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
243		nmfilters = (struct in_mfilter *)realloc(omfilters,
244		    sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
245		if (nmships != NULL && nmfilters != NULL) {
246			/* Initialize newly allocated source filter heads. */
247			for (idx = oldmax; idx < newmax; idx++) {
248				imf_init(&nmfilters[idx], MCAST_UNDEFINED,
249				    MCAST_EXCLUDE);
250			}
251			imo->imo_max_memberships = newmax;
252			imo->imo_membership = nmships;
253			imo->imo_mfilters = nmfilters;
254		}
255	}
256
257	if (nmships == NULL || nmfilters == NULL) {
258		if (nmships != NULL)
259			free(nmships, M_IPMOPTS);
260		if (nmfilters != NULL)
261			free(nmfilters, M_INMFILTER);
262		return (ETOOMANYREFS);
263	}
264
265	return (0);
266}
267
268/*
269 * Find an IPv4 multicast group entry for this ip_moptions instance
270 * which matches the specified group, and optionally an interface.
271 * Return its index into the array, or -1 if not found.
272 */
273static size_t
274imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
275    const struct sockaddr *group)
276{
277	const struct sockaddr_in *gsin;
278	struct in_multi	**pinm;
279	int		  idx;
280	int		  nmships;
281
282	gsin = (const struct sockaddr_in *)group;
283
284	/* The imo_membership array may be lazy allocated. */
285	if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
286		return (-1);
287
288	nmships = imo->imo_num_memberships;
289	pinm = &imo->imo_membership[0];
290	for (idx = 0; idx < nmships; idx++, pinm++) {
291		if (*pinm == NULL)
292			continue;
293		if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
294		    in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
295			break;
296		}
297	}
298	if (idx >= nmships)
299		idx = -1;
300
301	return (idx);
302}
303
304/*
305 * Find an IPv4 multicast source entry for this imo which matches
306 * the given group index for this socket, and source address.
307 *
308 * NOTE: This does not check if the entry is in-mode, merely if
309 * it exists, which may not be the desired behaviour.
310 */
311static struct in_msource *
312imo_match_source(const struct ip_moptions *imo, const size_t gidx,
313    const struct sockaddr *src)
314{
315	struct ip_msource	 find;
316	struct in_mfilter	*imf;
317	struct ip_msource	*ims;
318	const sockunion_t	*psa;
319
320	KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
321	KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
322	    ("%s: invalid index %d\n", __func__, (int)gidx));
323
324	/* The imo_mfilters array may be lazy allocated. */
325	if (imo->imo_mfilters == NULL)
326		return (NULL);
327	imf = &imo->imo_mfilters[gidx];
328
329	/* Source trees are keyed in host byte order. */
330	psa = (const sockunion_t *)src;
331	find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
332	ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
333
334	return ((struct in_msource *)ims);
335}
336
337/*
338 * Perform filtering for multicast datagrams on a socket by group and source.
339 *
340 * Returns 0 if a datagram should be allowed through, or various error codes
341 * if the socket was not a member of the group, or the source was muted, etc.
342 */
343int
344imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
345    const struct sockaddr *group, const struct sockaddr *src)
346{
347	size_t gidx;
348	struct in_msource *ims;
349	int mode;
350
351	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
352
353	gidx = imo_match_group(imo, ifp, group);
354	if (gidx == -1)
355		return (MCAST_NOTGMEMBER);
356
357	/*
358	 * Check if the source was included in an (S,G) join.
359	 * Allow reception on exclusive memberships by default,
360	 * reject reception on inclusive memberships by default.
361	 * Exclude source only if an in-mode exclude filter exists.
362	 * Include source only if an in-mode include filter exists.
363	 * NOTE: We are comparing group state here at IGMP t1 (now)
364	 * with socket-layer t0 (since last downcall).
365	 */
366	mode = imo->imo_mfilters[gidx].imf_st[1];
367	ims = imo_match_source(imo, gidx, src);
368
369	if ((ims == NULL && mode == MCAST_INCLUDE) ||
370	    (ims != NULL && ims->imsl_st[0] != mode))
371		return (MCAST_NOTSMEMBER);
372
373	return (MCAST_PASS);
374}
375
376/*
377 * Find and return a reference to an in_multi record for (ifp, group),
378 * and bump its reference count.
379 * If one does not exist, try to allocate it, and update link-layer multicast
380 * filters on ifp to listen for group.
381 * Assumes the IN_MULTI lock is held across the call.
382 * Return 0 if successful, otherwise return an appropriate error code.
383 */
384static int
385in_getmulti(struct ifnet *ifp, const struct in_addr *group,
386    struct in_multi **pinm)
387{
388	struct sockaddr_in	 gsin;
389	struct ifmultiaddr	*ifma;
390	struct in_ifinfo	*ii;
391	struct in_multi		*inm;
392	int error;
393
394	IN_MULTI_LOCK_ASSERT();
395
396	ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
397
398	inm = inm_lookup(ifp, *group);
399	if (inm != NULL) {
400		/*
401		 * If we already joined this group, just bump the
402		 * refcount and return it.
403		 */
404		KASSERT(inm->inm_refcount >= 1,
405		    ("%s: bad refcount %d", __func__, inm->inm_refcount));
406		++inm->inm_refcount;
407		*pinm = inm;
408		return (0);
409	}
410
411	memset(&gsin, 0, sizeof(gsin));
412	gsin.sin_family = AF_INET;
413	gsin.sin_len = sizeof(struct sockaddr_in);
414	gsin.sin_addr = *group;
415
416	/*
417	 * Check if a link-layer group is already associated
418	 * with this network-layer group on the given ifnet.
419	 */
420	error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
421	if (error != 0)
422		return (error);
423
424	/* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
425	IF_ADDR_LOCK(ifp);
426
427	/*
428	 * If something other than netinet is occupying the link-layer
429	 * group, print a meaningful error message and back out of
430	 * the allocation.
431	 * Otherwise, bump the refcount on the existing network-layer
432	 * group association and return it.
433	 */
434	if (ifma->ifma_protospec != NULL) {
435		inm = (struct in_multi *)ifma->ifma_protospec;
436#ifdef INVARIANTS
437		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
438		    __func__));
439		KASSERT(ifma->ifma_addr->sa_family == AF_INET,
440		    ("%s: ifma not AF_INET", __func__));
441		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
442		if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
443		    !in_hosteq(inm->inm_addr, *group))
444			panic("%s: ifma %p is inconsistent with %p (%s)",
445			    __func__, ifma, inm, inet_ntoa(*group));
446#endif
447		++inm->inm_refcount;
448		*pinm = inm;
449		IF_ADDR_UNLOCK(ifp);
450		return (0);
451	}
452
453	IF_ADDR_LOCK_ASSERT(ifp);
454
455	/*
456	 * A new in_multi record is needed; allocate and initialize it.
457	 * We DO NOT perform an IGMP join as the in_ layer may need to
458	 * push an initial source list down to IGMP to support SSM.
459	 *
460	 * The initial source filter state is INCLUDE, {} as per the RFC.
461	 */
462	inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
463	if (inm == NULL) {
464		if_delmulti_ifma(ifma);
465		IF_ADDR_UNLOCK(ifp);
466		return (ENOMEM);
467	}
468	inm->inm_addr = *group;
469	inm->inm_ifp = ifp;
470	inm->inm_igi = ii->ii_igmp;
471	inm->inm_ifma = ifma;
472	inm->inm_refcount = 1;
473	inm->inm_state = IGMP_NOT_MEMBER;
474
475	/*
476	 * Pending state-changes per group are subject to a bounds check.
477	 */
478	IFQ_SET_MAXLEN(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
479
480	inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
481	inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
482	RB_INIT(&inm->inm_srcs);
483
484	ifma->ifma_protospec = inm;
485
486	*pinm = inm;
487
488	IF_ADDR_UNLOCK(ifp);
489	return (0);
490}
491
492/*
493 * Drop a reference to an in_multi record.
494 *
495 * If the refcount drops to 0, free the in_multi record and
496 * delete the underlying link-layer membership.
497 */
498void
499inm_release_locked(struct in_multi *inm)
500{
501	struct ifmultiaddr *ifma;
502
503	IN_MULTI_LOCK_ASSERT();
504
505	CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
506
507	if (--inm->inm_refcount > 0) {
508		CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__,
509		    inm->inm_refcount);
510		return;
511	}
512
513	CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
514
515	ifma = inm->inm_ifma;
516
517	/* XXX this access is not covered by IF_ADDR_LOCK */
518	CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
519	KASSERT(ifma->ifma_protospec == inm,
520	    ("%s: ifma_protospec != inm", __func__));
521	ifma->ifma_protospec = NULL;
522
523	inm_purge(inm);
524
525	free(inm, M_IPMADDR);
526
527	if_delmulti_ifma(ifma);
528}
529
530/*
531 * Clear recorded source entries for a group.
532 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
533 * FIXME: Should reap.
534 */
535void
536inm_clear_recorded(struct in_multi *inm)
537{
538	struct ip_msource	*ims;
539
540	IN_MULTI_LOCK_ASSERT();
541
542	RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
543		if (ims->ims_stp) {
544			ims->ims_stp = 0;
545			--inm->inm_st[1].iss_rec;
546		}
547	}
548	KASSERT(inm->inm_st[1].iss_rec == 0,
549	    ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
550}
551
552/*
553 * Record a source as pending for a Source-Group IGMPv3 query.
554 * This lives here as it modifies the shared tree.
555 *
556 * inm is the group descriptor.
557 * naddr is the address of the source to record in network-byte order.
558 *
559 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
560 * lazy-allocate a source node in response to an SG query.
561 * Otherwise, no allocation is performed. This saves some memory
562 * with the trade-off that the source will not be reported to the
563 * router if joined in the window between the query response and
564 * the group actually being joined on the local host.
565 *
566 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
567 * This turns off the allocation of a recorded source entry if
568 * the group has not been joined.
569 *
570 * Return 0 if the source didn't exist or was already marked as recorded.
571 * Return 1 if the source was marked as recorded by this function.
572 * Return <0 if any error occured (negated errno code).
573 */
574int
575inm_record_source(struct in_multi *inm, const in_addr_t naddr)
576{
577	struct ip_msource	 find;
578	struct ip_msource	*ims, *nims;
579
580	IN_MULTI_LOCK_ASSERT();
581
582	find.ims_haddr = ntohl(naddr);
583	ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
584	if (ims && ims->ims_stp)
585		return (0);
586	if (ims == NULL) {
587		if (inm->inm_nsrc == in_mcast_maxgrpsrc)
588			return (-ENOSPC);
589		nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
590		    M_NOWAIT | M_ZERO);
591		if (nims == NULL)
592			return (-ENOMEM);
593		nims->ims_haddr = find.ims_haddr;
594		RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
595		++inm->inm_nsrc;
596		ims = nims;
597	}
598
599	/*
600	 * Mark the source as recorded and update the recorded
601	 * source count.
602	 */
603	++ims->ims_stp;
604	++inm->inm_st[1].iss_rec;
605
606	return (1);
607}
608
609/*
610 * Return a pointer to an in_msource owned by an in_mfilter,
611 * given its source address.
612 * Lazy-allocate if needed. If this is a new entry its filter state is
613 * undefined at t0.
614 *
615 * imf is the filter set being modified.
616 * haddr is the source address in *host* byte-order.
617 *
618 * SMPng: May be called with locks held; malloc must not block.
619 */
620static int
621imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
622    struct in_msource **plims)
623{
624	struct ip_msource	 find;
625	struct ip_msource	*ims, *nims;
626	struct in_msource	*lims;
627	int			 error;
628
629	error = 0;
630	ims = NULL;
631	lims = NULL;
632
633	/* key is host byte order */
634	find.ims_haddr = ntohl(psin->sin_addr.s_addr);
635	ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
636	lims = (struct in_msource *)ims;
637	if (lims == NULL) {
638		if (imf->imf_nsrc == in_mcast_maxsocksrc)
639			return (ENOSPC);
640		nims = malloc(sizeof(struct in_msource), M_INMFILTER,
641		    M_NOWAIT | M_ZERO);
642		if (nims == NULL)
643			return (ENOMEM);
644		lims = (struct in_msource *)nims;
645		lims->ims_haddr = find.ims_haddr;
646		lims->imsl_st[0] = MCAST_UNDEFINED;
647		RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
648		++imf->imf_nsrc;
649	}
650
651	*plims = lims;
652
653	return (error);
654}
655
656/*
657 * Graft a source entry into an existing socket-layer filter set,
658 * maintaining any required invariants and checking allocations.
659 *
660 * The source is marked as being in the new filter mode at t1.
661 *
662 * Return the pointer to the new node, otherwise return NULL.
663 */
664static struct in_msource *
665imf_graft(struct in_mfilter *imf, const uint8_t st1,
666    const struct sockaddr_in *psin)
667{
668	struct ip_msource	*nims;
669	struct in_msource	*lims;
670
671	nims = malloc(sizeof(struct in_msource), M_INMFILTER,
672	    M_NOWAIT | M_ZERO);
673	if (nims == NULL)
674		return (NULL);
675	lims = (struct in_msource *)nims;
676	lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
677	lims->imsl_st[0] = MCAST_UNDEFINED;
678	lims->imsl_st[1] = st1;
679	RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
680	++imf->imf_nsrc;
681
682	return (lims);
683}
684
685/*
686 * Prune a source entry from an existing socket-layer filter set,
687 * maintaining any required invariants and checking allocations.
688 *
689 * The source is marked as being left at t1, it is not freed.
690 *
691 * Return 0 if no error occurred, otherwise return an errno value.
692 */
693static int
694imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
695{
696	struct ip_msource	 find;
697	struct ip_msource	*ims;
698	struct in_msource	*lims;
699
700	/* key is host byte order */
701	find.ims_haddr = ntohl(psin->sin_addr.s_addr);
702	ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
703	if (ims == NULL)
704		return (ENOENT);
705	lims = (struct in_msource *)ims;
706	lims->imsl_st[1] = MCAST_UNDEFINED;
707	return (0);
708}
709
710/*
711 * Revert socket-layer filter set deltas at t1 to t0 state.
712 */
713static void
714imf_rollback(struct in_mfilter *imf)
715{
716	struct ip_msource	*ims, *tims;
717	struct in_msource	*lims;
718
719	RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
720		lims = (struct in_msource *)ims;
721		if (lims->imsl_st[0] == lims->imsl_st[1]) {
722			/* no change at t1 */
723			continue;
724		} else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
725			/* revert change to existing source at t1 */
726			lims->imsl_st[1] = lims->imsl_st[0];
727		} else {
728			/* revert source added t1 */
729			CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
730			RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
731			free(ims, M_INMFILTER);
732			imf->imf_nsrc--;
733		}
734	}
735	imf->imf_st[1] = imf->imf_st[0];
736}
737
738/*
739 * Mark socket-layer filter set as INCLUDE {} at t1.
740 */
741static void
742imf_leave(struct in_mfilter *imf)
743{
744	struct ip_msource	*ims;
745	struct in_msource	*lims;
746
747	RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
748		lims = (struct in_msource *)ims;
749		lims->imsl_st[1] = MCAST_UNDEFINED;
750	}
751	imf->imf_st[1] = MCAST_INCLUDE;
752}
753
754/*
755 * Mark socket-layer filter set deltas as committed.
756 */
757static void
758imf_commit(struct in_mfilter *imf)
759{
760	struct ip_msource	*ims;
761	struct in_msource	*lims;
762
763	RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
764		lims = (struct in_msource *)ims;
765		lims->imsl_st[0] = lims->imsl_st[1];
766	}
767	imf->imf_st[0] = imf->imf_st[1];
768}
769
770/*
771 * Reap unreferenced sources from socket-layer filter set.
772 */
773static void
774imf_reap(struct in_mfilter *imf)
775{
776	struct ip_msource	*ims, *tims;
777	struct in_msource	*lims;
778
779	RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
780		lims = (struct in_msource *)ims;
781		if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
782		    (lims->imsl_st[1] == MCAST_UNDEFINED)) {
783			CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
784			RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
785			free(ims, M_INMFILTER);
786			imf->imf_nsrc--;
787		}
788	}
789}
790
791/*
792 * Purge socket-layer filter set.
793 */
794static void
795imf_purge(struct in_mfilter *imf)
796{
797	struct ip_msource	*ims, *tims;
798
799	RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
800		CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
801		RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
802		free(ims, M_INMFILTER);
803		imf->imf_nsrc--;
804	}
805	imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
806	KASSERT(RB_EMPTY(&imf->imf_sources),
807	    ("%s: imf_sources not empty", __func__));
808}
809
810/*
811 * Look up a source filter entry for a multicast group.
812 *
813 * inm is the group descriptor to work with.
814 * haddr is the host-byte-order IPv4 address to look up.
815 * noalloc may be non-zero to suppress allocation of sources.
816 * *pims will be set to the address of the retrieved or allocated source.
817 *
818 * SMPng: NOTE: may be called with locks held.
819 * Return 0 if successful, otherwise return a non-zero error code.
820 */
821static int
822inm_get_source(struct in_multi *inm, const in_addr_t haddr,
823    const int noalloc, struct ip_msource **pims)
824{
825	struct ip_msource	 find;
826	struct ip_msource	*ims, *nims;
827#ifdef KTR
828	struct in_addr ia;
829#endif
830
831	find.ims_haddr = haddr;
832	ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
833	if (ims == NULL && !noalloc) {
834		if (inm->inm_nsrc == in_mcast_maxgrpsrc)
835			return (ENOSPC);
836		nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
837		    M_NOWAIT | M_ZERO);
838		if (nims == NULL)
839			return (ENOMEM);
840		nims->ims_haddr = haddr;
841		RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
842		++inm->inm_nsrc;
843		ims = nims;
844#ifdef KTR
845		ia.s_addr = htonl(haddr);
846		CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__,
847		    inet_ntoa(ia), ims);
848#endif
849	}
850
851	*pims = ims;
852	return (0);
853}
854
855/*
856 * Merge socket-layer source into IGMP-layer source.
857 * If rollback is non-zero, perform the inverse of the merge.
858 */
859static void
860ims_merge(struct ip_msource *ims, const struct in_msource *lims,
861    const int rollback)
862{
863	int n = rollback ? -1 : 1;
864#ifdef KTR
865	struct in_addr ia;
866
867	ia.s_addr = htonl(ims->ims_haddr);
868#endif
869
870	if (lims->imsl_st[0] == MCAST_EXCLUDE) {
871		CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s",
872		    __func__, n, inet_ntoa(ia));
873		ims->ims_st[1].ex -= n;
874	} else if (lims->imsl_st[0] == MCAST_INCLUDE) {
875		CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s",
876		    __func__, n, inet_ntoa(ia));
877		ims->ims_st[1].in -= n;
878	}
879
880	if (lims->imsl_st[1] == MCAST_EXCLUDE) {
881		CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s",
882		    __func__, n, inet_ntoa(ia));
883		ims->ims_st[1].ex += n;
884	} else if (lims->imsl_st[1] == MCAST_INCLUDE) {
885		CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s",
886		    __func__, n, inet_ntoa(ia));
887		ims->ims_st[1].in += n;
888	}
889}
890
891/*
892 * Atomically update the global in_multi state, when a membership's
893 * filter list is being updated in any way.
894 *
895 * imf is the per-inpcb-membership group filter pointer.
896 * A fake imf may be passed for in-kernel consumers.
897 *
898 * XXX This is a candidate for a set-symmetric-difference style loop
899 * which would eliminate the repeated lookup from root of ims nodes,
900 * as they share the same key space.
901 *
902 * If any error occurred this function will back out of refcounts
903 * and return a non-zero value.
904 */
905static int
906inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
907{
908	struct ip_msource	*ims, *nims;
909	struct in_msource	*lims;
910	int			 schanged, error;
911	int			 nsrc0, nsrc1;
912
913	schanged = 0;
914	error = 0;
915	nsrc1 = nsrc0 = 0;
916
917	/*
918	 * Update the source filters first, as this may fail.
919	 * Maintain count of in-mode filters at t0, t1. These are
920	 * used to work out if we transition into ASM mode or not.
921	 * Maintain a count of source filters whose state was
922	 * actually modified by this operation.
923	 */
924	RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
925		lims = (struct in_msource *)ims;
926		if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
927		if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
928		if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
929		error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
930		++schanged;
931		if (error)
932			break;
933		ims_merge(nims, lims, 0);
934	}
935	if (error) {
936		struct ip_msource *bims;
937
938		RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
939			lims = (struct in_msource *)ims;
940			if (lims->imsl_st[0] == lims->imsl_st[1])
941				continue;
942			(void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
943			if (bims == NULL)
944				continue;
945			ims_merge(bims, lims, 1);
946		}
947		goto out_reap;
948	}
949
950	CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
951	    __func__, nsrc0, nsrc1);
952
953	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
954	if (imf->imf_st[0] == imf->imf_st[1] &&
955	    imf->imf_st[1] == MCAST_INCLUDE) {
956		if (nsrc1 == 0) {
957			CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
958			--inm->inm_st[1].iss_in;
959		}
960	}
961
962	/* Handle filter mode transition on socket. */
963	if (imf->imf_st[0] != imf->imf_st[1]) {
964		CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
965		    __func__, imf->imf_st[0], imf->imf_st[1]);
966
967		if (imf->imf_st[0] == MCAST_EXCLUDE) {
968			CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
969			--inm->inm_st[1].iss_ex;
970		} else if (imf->imf_st[0] == MCAST_INCLUDE) {
971			CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
972			--inm->inm_st[1].iss_in;
973		}
974
975		if (imf->imf_st[1] == MCAST_EXCLUDE) {
976			CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
977			inm->inm_st[1].iss_ex++;
978		} else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
979			CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
980			inm->inm_st[1].iss_in++;
981		}
982	}
983
984	/*
985	 * Track inm filter state in terms of listener counts.
986	 * If there are any exclusive listeners, stack-wide
987	 * membership is exclusive.
988	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
989	 * If no listeners remain, state is undefined at t1,
990	 * and the IGMP lifecycle for this group should finish.
991	 */
992	if (inm->inm_st[1].iss_ex > 0) {
993		CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
994		inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
995	} else if (inm->inm_st[1].iss_in > 0) {
996		CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
997		inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
998	} else {
999		CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1000		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1001	}
1002
1003	/* Decrement ASM listener count on transition out of ASM mode. */
1004	if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1005		if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1006		    (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1007			CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1008			--inm->inm_st[1].iss_asm;
1009	}
1010
1011	/* Increment ASM listener count on transition to ASM mode. */
1012	if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1013		CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1014		inm->inm_st[1].iss_asm++;
1015	}
1016
1017	CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1018	inm_print(inm);
1019
1020out_reap:
1021	if (schanged > 0) {
1022		CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1023		inm_reap(inm);
1024	}
1025	return (error);
1026}
1027
1028/*
1029 * Mark an in_multi's filter set deltas as committed.
1030 * Called by IGMP after a state change has been enqueued.
1031 */
1032void
1033inm_commit(struct in_multi *inm)
1034{
1035	struct ip_msource	*ims;
1036
1037	CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1038	CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1039	inm_print(inm);
1040
1041	RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1042		ims->ims_st[0] = ims->ims_st[1];
1043	}
1044	inm->inm_st[0] = inm->inm_st[1];
1045}
1046
1047/*
1048 * Reap unreferenced nodes from an in_multi's filter set.
1049 */
1050static void
1051inm_reap(struct in_multi *inm)
1052{
1053	struct ip_msource	*ims, *tims;
1054
1055	RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1056		if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1057		    ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1058		    ims->ims_stp != 0)
1059			continue;
1060		CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1061		RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1062		free(ims, M_IPMSOURCE);
1063		inm->inm_nsrc--;
1064	}
1065}
1066
1067/*
1068 * Purge all source nodes from an in_multi's filter set.
1069 */
1070static void
1071inm_purge(struct in_multi *inm)
1072{
1073	struct ip_msource	*ims, *tims;
1074
1075	RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1076		CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1077		RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1078		free(ims, M_IPMSOURCE);
1079		inm->inm_nsrc--;
1080	}
1081}
1082
1083/*
1084 * Join a multicast group; unlocked entry point.
1085 *
1086 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1087 * is not held. Fortunately, ifp is unlikely to have been detached
1088 * at this point, so we assume it's OK to recurse.
1089 */
1090int
1091in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1092    /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1093{
1094	int error;
1095
1096	IN_MULTI_LOCK();
1097	error = in_joingroup_locked(ifp, gina, imf, pinm);
1098	IN_MULTI_UNLOCK();
1099
1100	return (error);
1101}
1102
1103/*
1104 * Join a multicast group; real entry point.
1105 *
1106 * Only preserves atomicity at inm level.
1107 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1108 *
1109 * If the IGMP downcall fails, the group is not joined, and an error
1110 * code is returned.
1111 */
1112int
1113in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1114    /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1115{
1116	struct in_mfilter	 timf;
1117	struct in_multi		*inm;
1118	int			 error;
1119
1120	IN_MULTI_LOCK_ASSERT();
1121
1122	CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__,
1123	    inet_ntoa(*gina), ifp, ifp->if_xname);
1124
1125	error = 0;
1126	inm = NULL;
1127
1128	/*
1129	 * If no imf was specified (i.e. kernel consumer),
1130	 * fake one up and assume it is an ASM join.
1131	 */
1132	if (imf == NULL) {
1133		imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1134		imf = &timf;
1135	}
1136
1137	error = in_getmulti(ifp, gina, &inm);
1138	if (error) {
1139		CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1140		return (error);
1141	}
1142
1143	CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1144	error = inm_merge(inm, imf);
1145	if (error) {
1146		CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1147		goto out_inm_release;
1148	}
1149
1150	CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1151	error = igmp_change_state(inm);
1152	if (error) {
1153		CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1154		goto out_inm_release;
1155	}
1156
1157out_inm_release:
1158	if (error) {
1159		CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1160		inm_release_locked(inm);
1161	} else {
1162		*pinm = inm;
1163	}
1164
1165	return (error);
1166}
1167
1168/*
1169 * Leave a multicast group; unlocked entry point.
1170 */
1171int
1172in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1173{
1174	struct ifnet *ifp;
1175	int error;
1176
1177	ifp = inm->inm_ifp;
1178
1179	IN_MULTI_LOCK();
1180	error = in_leavegroup_locked(inm, imf);
1181	IN_MULTI_UNLOCK();
1182
1183	return (error);
1184}
1185
1186/*
1187 * Leave a multicast group; real entry point.
1188 * All source filters will be expunged.
1189 *
1190 * Only preserves atomicity at inm level.
1191 *
1192 * Holding the write lock for the INP which contains imf
1193 * is highly advisable. We can't assert for it as imf does not
1194 * contain a back-pointer to the owning inp.
1195 *
1196 * Note: This is not the same as inm_release(*) as this function also
1197 * makes a state change downcall into IGMP.
1198 */
1199int
1200in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1201{
1202	struct in_mfilter	 timf;
1203	int			 error;
1204
1205	error = 0;
1206
1207	IN_MULTI_LOCK_ASSERT();
1208
1209	CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__,
1210	    inm, inet_ntoa(inm->inm_addr),
1211	    (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1212	    imf);
1213
1214	/*
1215	 * If no imf was specified (i.e. kernel consumer),
1216	 * fake one up and assume it is an ASM join.
1217	 */
1218	if (imf == NULL) {
1219		imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1220		imf = &timf;
1221	}
1222
1223	/*
1224	 * Begin state merge transaction at IGMP layer.
1225	 *
1226	 * As this particular invocation should not cause any memory
1227	 * to be allocated, and there is no opportunity to roll back
1228	 * the transaction, it MUST NOT fail.
1229	 */
1230	CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1231	error = inm_merge(inm, imf);
1232	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1233
1234	CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1235	error = igmp_change_state(inm);
1236	if (error)
1237		CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1238
1239	CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1240	inm_release_locked(inm);
1241
1242	return (error);
1243}
1244
1245/*#ifndef BURN_BRIDGES*/
1246/*
1247 * Join an IPv4 multicast group in (*,G) exclusive mode.
1248 * The group must be a 224.0.0.0/24 link-scope group.
1249 * This KPI is for legacy kernel consumers only.
1250 */
1251struct in_multi *
1252in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1253{
1254	struct in_multi *pinm;
1255	int error;
1256
1257	KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1258	    ("%s: %s not in 224.0.0.0/24", __func__, inet_ntoa(*ap)));
1259
1260	error = in_joingroup(ifp, ap, NULL, &pinm);
1261	if (error != 0)
1262		pinm = NULL;
1263
1264	return (pinm);
1265}
1266
1267/*
1268 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1269 * This KPI is for legacy kernel consumers only.
1270 */
1271void
1272in_delmulti(struct in_multi *inm)
1273{
1274
1275	(void)in_leavegroup(inm, NULL);
1276}
1277/*#endif*/
1278
1279/*
1280 * Block or unblock an ASM multicast source on an inpcb.
1281 * This implements the delta-based API described in RFC 3678.
1282 *
1283 * The delta-based API applies only to exclusive-mode memberships.
1284 * An IGMP downcall will be performed.
1285 *
1286 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1287 *
1288 * Return 0 if successful, otherwise return an appropriate error code.
1289 */
1290static int
1291inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1292{
1293	struct group_source_req		 gsr;
1294	sockunion_t			*gsa, *ssa;
1295	struct ifnet			*ifp;
1296	struct in_mfilter		*imf;
1297	struct ip_moptions		*imo;
1298	struct in_msource		*ims;
1299	struct in_multi			*inm;
1300	size_t				 idx;
1301	uint16_t			 fmode;
1302	int				 error, doblock;
1303
1304	ifp = NULL;
1305	error = 0;
1306	doblock = 0;
1307
1308	memset(&gsr, 0, sizeof(struct group_source_req));
1309	gsa = (sockunion_t *)&gsr.gsr_group;
1310	ssa = (sockunion_t *)&gsr.gsr_source;
1311
1312	switch (sopt->sopt_name) {
1313	case IP_BLOCK_SOURCE:
1314	case IP_UNBLOCK_SOURCE: {
1315		struct ip_mreq_source	 mreqs;
1316
1317		error = sooptcopyin(sopt, &mreqs,
1318		    sizeof(struct ip_mreq_source),
1319		    sizeof(struct ip_mreq_source));
1320		if (error)
1321			return (error);
1322
1323		gsa->sin.sin_family = AF_INET;
1324		gsa->sin.sin_len = sizeof(struct sockaddr_in);
1325		gsa->sin.sin_addr = mreqs.imr_multiaddr;
1326
1327		ssa->sin.sin_family = AF_INET;
1328		ssa->sin.sin_len = sizeof(struct sockaddr_in);
1329		ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1330
1331		if (!in_nullhost(mreqs.imr_interface))
1332			INADDR_TO_IFP(mreqs.imr_interface, ifp);
1333
1334		if (sopt->sopt_name == IP_BLOCK_SOURCE)
1335			doblock = 1;
1336
1337		CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1338		    __func__, inet_ntoa(mreqs.imr_interface), ifp);
1339		break;
1340	    }
1341
1342	case MCAST_BLOCK_SOURCE:
1343	case MCAST_UNBLOCK_SOURCE:
1344		error = sooptcopyin(sopt, &gsr,
1345		    sizeof(struct group_source_req),
1346		    sizeof(struct group_source_req));
1347		if (error)
1348			return (error);
1349
1350		if (gsa->sin.sin_family != AF_INET ||
1351		    gsa->sin.sin_len != sizeof(struct sockaddr_in))
1352			return (EINVAL);
1353
1354		if (ssa->sin.sin_family != AF_INET ||
1355		    ssa->sin.sin_len != sizeof(struct sockaddr_in))
1356			return (EINVAL);
1357
1358		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1359			return (EADDRNOTAVAIL);
1360
1361		ifp = ifnet_byindex(gsr.gsr_interface);
1362
1363		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1364			doblock = 1;
1365		break;
1366
1367	default:
1368		CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1369		    __func__, sopt->sopt_name);
1370		return (EOPNOTSUPP);
1371		break;
1372	}
1373
1374	if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1375		return (EINVAL);
1376
1377	/*
1378	 * Check if we are actually a member of this group.
1379	 */
1380	imo = inp_findmoptions(inp);
1381	idx = imo_match_group(imo, ifp, &gsa->sa);
1382	if (idx == -1 || imo->imo_mfilters == NULL) {
1383		error = EADDRNOTAVAIL;
1384		goto out_inp_locked;
1385	}
1386
1387	KASSERT(imo->imo_mfilters != NULL,
1388	    ("%s: imo_mfilters not allocated", __func__));
1389	imf = &imo->imo_mfilters[idx];
1390	inm = imo->imo_membership[idx];
1391
1392	/*
1393	 * Attempting to use the delta-based API on an
1394	 * non exclusive-mode membership is an error.
1395	 */
1396	fmode = imf->imf_st[0];
1397	if (fmode != MCAST_EXCLUDE) {
1398		error = EINVAL;
1399		goto out_inp_locked;
1400	}
1401
1402	/*
1403	 * Deal with error cases up-front:
1404	 *  Asked to block, but already blocked; or
1405	 *  Asked to unblock, but nothing to unblock.
1406	 * If adding a new block entry, allocate it.
1407	 */
1408	ims = imo_match_source(imo, idx, &ssa->sa);
1409	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1410		CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
1411		    inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not ");
1412		error = EADDRNOTAVAIL;
1413		goto out_inp_locked;
1414	}
1415
1416	INP_WLOCK_ASSERT(inp);
1417
1418	/*
1419	 * Begin state merge transaction at socket layer.
1420	 */
1421	if (doblock) {
1422		CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1423		ims = imf_graft(imf, fmode, &ssa->sin);
1424		if (ims == NULL)
1425			error = ENOMEM;
1426	} else {
1427		CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1428		error = imf_prune(imf, &ssa->sin);
1429	}
1430
1431	if (error) {
1432		CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1433		goto out_imf_rollback;
1434	}
1435
1436	/*
1437	 * Begin state merge transaction at IGMP layer.
1438	 */
1439	IN_MULTI_LOCK();
1440
1441	CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1442	error = inm_merge(inm, imf);
1443	if (error) {
1444		CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1445		goto out_imf_rollback;
1446	}
1447
1448	CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1449	error = igmp_change_state(inm);
1450	if (error)
1451		CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1452
1453	IN_MULTI_UNLOCK();
1454
1455out_imf_rollback:
1456	if (error)
1457		imf_rollback(imf);
1458	else
1459		imf_commit(imf);
1460
1461	imf_reap(imf);
1462
1463out_inp_locked:
1464	INP_WUNLOCK(inp);
1465	return (error);
1466}
1467
1468/*
1469 * Given an inpcb, return its multicast options structure pointer.  Accepts
1470 * an unlocked inpcb pointer, but will return it locked.  May sleep.
1471 *
1472 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1473 * SMPng: NOTE: Returns with the INP write lock held.
1474 */
1475static struct ip_moptions *
1476inp_findmoptions(struct inpcb *inp)
1477{
1478	struct ip_moptions	 *imo;
1479	struct in_multi		**immp;
1480	struct in_mfilter	 *imfp;
1481	size_t			  idx;
1482
1483	INP_WLOCK(inp);
1484	if (inp->inp_moptions != NULL)
1485		return (inp->inp_moptions);
1486
1487	INP_WUNLOCK(inp);
1488
1489	imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1490	immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1491	    M_WAITOK | M_ZERO);
1492	imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1493	    M_INMFILTER, M_WAITOK);
1494
1495	imo->imo_multicast_ifp = NULL;
1496	imo->imo_multicast_addr.s_addr = INADDR_ANY;
1497	imo->imo_multicast_vif = -1;
1498	imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1499	imo->imo_multicast_loop = in_mcast_loop;
1500	imo->imo_num_memberships = 0;
1501	imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1502	imo->imo_membership = immp;
1503
1504	/* Initialize per-group source filters. */
1505	for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1506		imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1507	imo->imo_mfilters = imfp;
1508
1509	INP_WLOCK(inp);
1510	if (inp->inp_moptions != NULL) {
1511		free(imfp, M_INMFILTER);
1512		free(immp, M_IPMOPTS);
1513		free(imo, M_IPMOPTS);
1514		return (inp->inp_moptions);
1515	}
1516	inp->inp_moptions = imo;
1517	return (imo);
1518}
1519
1520/*
1521 * Discard the IP multicast options (and source filters).
1522 *
1523 * SMPng: NOTE: assumes INP write lock is held.
1524 */
1525void
1526inp_freemoptions(struct ip_moptions *imo)
1527{
1528	struct in_mfilter	*imf;
1529	size_t			 idx, nmships;
1530
1531	KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1532
1533	nmships = imo->imo_num_memberships;
1534	for (idx = 0; idx < nmships; ++idx) {
1535		imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1536		if (imf)
1537			imf_leave(imf);
1538		(void)in_leavegroup(imo->imo_membership[idx], imf);
1539		if (imf)
1540			imf_purge(imf);
1541	}
1542
1543	if (imo->imo_mfilters)
1544		free(imo->imo_mfilters, M_INMFILTER);
1545	free(imo->imo_membership, M_IPMOPTS);
1546	free(imo, M_IPMOPTS);
1547}
1548
1549/*
1550 * Atomically get source filters on a socket for an IPv4 multicast group.
1551 * Called with INP lock held; returns with lock released.
1552 */
1553static int
1554inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1555{
1556	struct __msfilterreq	 msfr;
1557	sockunion_t		*gsa;
1558	struct ifnet		*ifp;
1559	struct ip_moptions	*imo;
1560	struct in_mfilter	*imf;
1561	struct ip_msource	*ims;
1562	struct in_msource	*lims;
1563	struct sockaddr_in	*psin;
1564	struct sockaddr_storage	*ptss;
1565	struct sockaddr_storage	*tss;
1566	int			 error;
1567	size_t			 idx, nsrcs, ncsrcs;
1568
1569	INP_WLOCK_ASSERT(inp);
1570
1571	imo = inp->inp_moptions;
1572	KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1573
1574	INP_WUNLOCK(inp);
1575
1576	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1577	    sizeof(struct __msfilterreq));
1578	if (error)
1579		return (error);
1580
1581	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1582		return (EINVAL);
1583
1584	ifp = ifnet_byindex(msfr.msfr_ifindex);
1585	if (ifp == NULL)
1586		return (EINVAL);
1587
1588	INP_WLOCK(inp);
1589
1590	/*
1591	 * Lookup group on the socket.
1592	 */
1593	gsa = (sockunion_t *)&msfr.msfr_group;
1594	idx = imo_match_group(imo, ifp, &gsa->sa);
1595	if (idx == -1 || imo->imo_mfilters == NULL) {
1596		INP_WUNLOCK(inp);
1597		return (EADDRNOTAVAIL);
1598	}
1599	imf = &imo->imo_mfilters[idx];
1600
1601	/*
1602	 * Ignore memberships which are in limbo.
1603	 */
1604	if (imf->imf_st[1] == MCAST_UNDEFINED) {
1605		INP_WUNLOCK(inp);
1606		return (EAGAIN);
1607	}
1608	msfr.msfr_fmode = imf->imf_st[1];
1609
1610	/*
1611	 * If the user specified a buffer, copy out the source filter
1612	 * entries to userland gracefully.
1613	 * We only copy out the number of entries which userland
1614	 * has asked for, but we always tell userland how big the
1615	 * buffer really needs to be.
1616	 */
1617	tss = NULL;
1618	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1619		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1620		    M_TEMP, M_NOWAIT | M_ZERO);
1621		if (tss == NULL) {
1622			INP_WUNLOCK(inp);
1623			return (ENOBUFS);
1624		}
1625	}
1626
1627	/*
1628	 * Count number of sources in-mode at t0.
1629	 * If buffer space exists and remains, copy out source entries.
1630	 */
1631	nsrcs = msfr.msfr_nsrcs;
1632	ncsrcs = 0;
1633	ptss = tss;
1634	RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1635		lims = (struct in_msource *)ims;
1636		if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1637		    lims->imsl_st[0] != imf->imf_st[0])
1638			continue;
1639		++ncsrcs;
1640		if (tss != NULL && nsrcs > 0) {
1641			psin = (struct sockaddr_in *)ptss;
1642			psin->sin_family = AF_INET;
1643			psin->sin_len = sizeof(struct sockaddr_in);
1644			psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1645			psin->sin_port = 0;
1646			++ptss;
1647			--nsrcs;
1648		}
1649	}
1650
1651	INP_WUNLOCK(inp);
1652
1653	if (tss != NULL) {
1654		error = copyout(tss, msfr.msfr_srcs,
1655		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1656		free(tss, M_TEMP);
1657		if (error)
1658			return (error);
1659	}
1660
1661	msfr.msfr_nsrcs = ncsrcs;
1662	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1663
1664	return (error);
1665}
1666
1667/*
1668 * Return the IP multicast options in response to user getsockopt().
1669 */
1670int
1671inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1672{
1673	struct ip_mreqn		 mreqn;
1674	struct ip_moptions	*imo;
1675	struct ifnet		*ifp;
1676	struct in_ifaddr	*ia;
1677	int			 error, optval;
1678	u_char			 coptval;
1679
1680	INP_WLOCK(inp);
1681	imo = inp->inp_moptions;
1682	/*
1683	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1684	 * or is a divert socket, reject it.
1685	 */
1686	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1687	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1688	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1689		INP_WUNLOCK(inp);
1690		return (EOPNOTSUPP);
1691	}
1692
1693	error = 0;
1694	switch (sopt->sopt_name) {
1695	case IP_MULTICAST_VIF:
1696		if (imo != NULL)
1697			optval = imo->imo_multicast_vif;
1698		else
1699			optval = -1;
1700		INP_WUNLOCK(inp);
1701		error = sooptcopyout(sopt, &optval, sizeof(int));
1702		break;
1703
1704	case IP_MULTICAST_IF:
1705		memset(&mreqn, 0, sizeof(struct ip_mreqn));
1706		if (imo != NULL) {
1707			ifp = imo->imo_multicast_ifp;
1708			if (!in_nullhost(imo->imo_multicast_addr)) {
1709				mreqn.imr_address = imo->imo_multicast_addr;
1710			} else if (ifp != NULL) {
1711				mreqn.imr_ifindex = ifp->if_index;
1712				IFP_TO_IA(ifp, ia);
1713				if (ia != NULL) {
1714					mreqn.imr_address =
1715					    IA_SIN(ia)->sin_addr;
1716					ifa_free(&ia->ia_ifa);
1717				}
1718			}
1719		}
1720		INP_WUNLOCK(inp);
1721		if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1722			error = sooptcopyout(sopt, &mreqn,
1723			    sizeof(struct ip_mreqn));
1724		} else {
1725			error = sooptcopyout(sopt, &mreqn.imr_address,
1726			    sizeof(struct in_addr));
1727		}
1728		break;
1729
1730	case IP_MULTICAST_TTL:
1731		if (imo == 0)
1732			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1733		else
1734			optval = coptval = imo->imo_multicast_ttl;
1735		INP_WUNLOCK(inp);
1736		if (sopt->sopt_valsize == sizeof(u_char))
1737			error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1738		else
1739			error = sooptcopyout(sopt, &optval, sizeof(int));
1740		break;
1741
1742	case IP_MULTICAST_LOOP:
1743		if (imo == 0)
1744			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1745		else
1746			optval = coptval = imo->imo_multicast_loop;
1747		INP_WUNLOCK(inp);
1748		if (sopt->sopt_valsize == sizeof(u_char))
1749			error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1750		else
1751			error = sooptcopyout(sopt, &optval, sizeof(int));
1752		break;
1753
1754	case IP_MSFILTER:
1755		if (imo == NULL) {
1756			error = EADDRNOTAVAIL;
1757			INP_WUNLOCK(inp);
1758		} else {
1759			error = inp_get_source_filters(inp, sopt);
1760		}
1761		break;
1762
1763	default:
1764		INP_WUNLOCK(inp);
1765		error = ENOPROTOOPT;
1766		break;
1767	}
1768
1769	INP_UNLOCK_ASSERT(inp);
1770
1771	return (error);
1772}
1773
1774/*
1775 * Look up the ifnet to use for a multicast group membership,
1776 * given the IPv4 address of an interface, and the IPv4 group address.
1777 *
1778 * This routine exists to support legacy multicast applications
1779 * which do not understand that multicast memberships are scoped to
1780 * specific physical links in the networking stack, or which need
1781 * to join link-scope groups before IPv4 addresses are configured.
1782 *
1783 * If inp is non-NULL, use this socket's current FIB number for any
1784 * required FIB lookup.
1785 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1786 * and use its ifp; usually, this points to the default next-hop.
1787 *
1788 * If the FIB lookup fails, attempt to use the first non-loopback
1789 * interface with multicast capability in the system as a
1790 * last resort. The legacy IPv4 ASM API requires that we do
1791 * this in order to allow groups to be joined when the routing
1792 * table has not yet been populated during boot.
1793 *
1794 * Returns NULL if no ifp could be found.
1795 *
1796 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1797 * FUTURE: Implement IPv4 source-address selection.
1798 */
1799static struct ifnet *
1800inp_lookup_mcast_ifp(const struct inpcb *inp,
1801    const struct sockaddr_in *gsin, const struct in_addr ina)
1802{
1803	struct ifnet *ifp;
1804
1805	KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1806	KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1807	    ("%s: not multicast", __func__));
1808
1809	ifp = NULL;
1810	if (!in_nullhost(ina)) {
1811		INADDR_TO_IFP(ina, ifp);
1812	} else {
1813		struct route ro;
1814
1815		ro.ro_rt = NULL;
1816		memcpy(&ro.ro_dst, gsin, sizeof(struct sockaddr_in));
1817		in_rtalloc_ign(&ro, 0, inp ? inp->inp_inc.inc_fibnum : 0);
1818		if (ro.ro_rt != NULL) {
1819			ifp = ro.ro_rt->rt_ifp;
1820			KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1821			RTFREE(ro.ro_rt);
1822		} else {
1823			struct in_ifaddr *ia;
1824			struct ifnet *mifp;
1825
1826			mifp = NULL;
1827			IN_IFADDR_RLOCK();
1828			TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1829				mifp = ia->ia_ifp;
1830				if (!(mifp->if_flags & IFF_LOOPBACK) &&
1831				     (mifp->if_flags & IFF_MULTICAST)) {
1832					ifp = mifp;
1833					break;
1834				}
1835			}
1836			IN_IFADDR_RUNLOCK();
1837		}
1838	}
1839
1840	return (ifp);
1841}
1842
1843/*
1844 * Join an IPv4 multicast group, possibly with a source.
1845 */
1846static int
1847inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1848{
1849	struct group_source_req		 gsr;
1850	sockunion_t			*gsa, *ssa;
1851	struct ifnet			*ifp;
1852	struct in_mfilter		*imf;
1853	struct ip_moptions		*imo;
1854	struct in_multi			*inm;
1855	struct in_msource		*lims;
1856	size_t				 idx;
1857	int				 error, is_new;
1858
1859	ifp = NULL;
1860	imf = NULL;
1861	lims = NULL;
1862	error = 0;
1863	is_new = 0;
1864
1865	memset(&gsr, 0, sizeof(struct group_source_req));
1866	gsa = (sockunion_t *)&gsr.gsr_group;
1867	gsa->ss.ss_family = AF_UNSPEC;
1868	ssa = (sockunion_t *)&gsr.gsr_source;
1869	ssa->ss.ss_family = AF_UNSPEC;
1870
1871	switch (sopt->sopt_name) {
1872	case IP_ADD_MEMBERSHIP:
1873	case IP_ADD_SOURCE_MEMBERSHIP: {
1874		struct ip_mreq_source	 mreqs;
1875
1876		if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1877			error = sooptcopyin(sopt, &mreqs,
1878			    sizeof(struct ip_mreq),
1879			    sizeof(struct ip_mreq));
1880			/*
1881			 * Do argument switcharoo from ip_mreq into
1882			 * ip_mreq_source to avoid using two instances.
1883			 */
1884			mreqs.imr_interface = mreqs.imr_sourceaddr;
1885			mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1886		} else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1887			error = sooptcopyin(sopt, &mreqs,
1888			    sizeof(struct ip_mreq_source),
1889			    sizeof(struct ip_mreq_source));
1890		}
1891		if (error)
1892			return (error);
1893
1894		gsa->sin.sin_family = AF_INET;
1895		gsa->sin.sin_len = sizeof(struct sockaddr_in);
1896		gsa->sin.sin_addr = mreqs.imr_multiaddr;
1897
1898		if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1899			ssa->sin.sin_family = AF_INET;
1900			ssa->sin.sin_len = sizeof(struct sockaddr_in);
1901			ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1902		}
1903
1904		if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1905			return (EINVAL);
1906
1907		ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1908		    mreqs.imr_interface);
1909		CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1910		    __func__, inet_ntoa(mreqs.imr_interface), ifp);
1911		break;
1912	}
1913
1914	case MCAST_JOIN_GROUP:
1915	case MCAST_JOIN_SOURCE_GROUP:
1916		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1917			error = sooptcopyin(sopt, &gsr,
1918			    sizeof(struct group_req),
1919			    sizeof(struct group_req));
1920		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1921			error = sooptcopyin(sopt, &gsr,
1922			    sizeof(struct group_source_req),
1923			    sizeof(struct group_source_req));
1924		}
1925		if (error)
1926			return (error);
1927
1928		if (gsa->sin.sin_family != AF_INET ||
1929		    gsa->sin.sin_len != sizeof(struct sockaddr_in))
1930			return (EINVAL);
1931
1932		/*
1933		 * Overwrite the port field if present, as the sockaddr
1934		 * being copied in may be matched with a binary comparison.
1935		 */
1936		gsa->sin.sin_port = 0;
1937		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1938			if (ssa->sin.sin_family != AF_INET ||
1939			    ssa->sin.sin_len != sizeof(struct sockaddr_in))
1940				return (EINVAL);
1941			ssa->sin.sin_port = 0;
1942		}
1943
1944		if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1945			return (EINVAL);
1946
1947		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1948			return (EADDRNOTAVAIL);
1949		ifp = ifnet_byindex(gsr.gsr_interface);
1950		break;
1951
1952	default:
1953		CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1954		    __func__, sopt->sopt_name);
1955		return (EOPNOTSUPP);
1956		break;
1957	}
1958
1959	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1960		return (EADDRNOTAVAIL);
1961
1962	imo = inp_findmoptions(inp);
1963	idx = imo_match_group(imo, ifp, &gsa->sa);
1964	if (idx == -1) {
1965		is_new = 1;
1966	} else {
1967		inm = imo->imo_membership[idx];
1968		imf = &imo->imo_mfilters[idx];
1969		if (ssa->ss.ss_family != AF_UNSPEC) {
1970			/*
1971			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1972			 * is an error. On an existing inclusive membership,
1973			 * it just adds the source to the filter list.
1974			 */
1975			if (imf->imf_st[1] != MCAST_INCLUDE) {
1976				error = EINVAL;
1977				goto out_inp_locked;
1978			}
1979			/*
1980			 * Throw out duplicates.
1981			 *
1982			 * XXX FIXME: This makes a naive assumption that
1983			 * even if entries exist for *ssa in this imf,
1984			 * they will be rejected as dupes, even if they
1985			 * are not valid in the current mode (in-mode).
1986			 *
1987			 * in_msource is transactioned just as for anything
1988			 * else in SSM -- but note naive use of inm_graft()
1989			 * below for allocating new filter entries.
1990			 *
1991			 * This is only an issue if someone mixes the
1992			 * full-state SSM API with the delta-based API,
1993			 * which is discouraged in the relevant RFCs.
1994			 */
1995			lims = imo_match_source(imo, idx, &ssa->sa);
1996			if (lims != NULL /*&&
1997			    lims->imsl_st[1] == MCAST_INCLUDE*/) {
1998				error = EADDRNOTAVAIL;
1999				goto out_inp_locked;
2000			}
2001		} else {
2002			/*
2003			 * MCAST_JOIN_GROUP on an existing exclusive
2004			 * membership is an error; return EADDRINUSE
2005			 * to preserve 4.4BSD API idempotence, and
2006			 * avoid tedious detour to code below.
2007			 * NOTE: This is bending RFC 3678 a bit.
2008			 *
2009			 * On an existing inclusive membership, this is also
2010			 * an error; if you want to change filter mode,
2011			 * you must use the userland API setsourcefilter().
2012			 * XXX We don't reject this for imf in UNDEFINED
2013			 * state at t1, because allocation of a filter
2014			 * is atomic with allocation of a membership.
2015			 */
2016			error = EINVAL;
2017			if (imf->imf_st[1] == MCAST_EXCLUDE)
2018				error = EADDRINUSE;
2019			goto out_inp_locked;
2020		}
2021	}
2022
2023	/*
2024	 * Begin state merge transaction at socket layer.
2025	 */
2026	INP_WLOCK_ASSERT(inp);
2027
2028	if (is_new) {
2029		if (imo->imo_num_memberships == imo->imo_max_memberships) {
2030			error = imo_grow(imo);
2031			if (error)
2032				goto out_inp_locked;
2033		}
2034		/*
2035		 * Allocate the new slot upfront so we can deal with
2036		 * grafting the new source filter in same code path
2037		 * as for join-source on existing membership.
2038		 */
2039		idx = imo->imo_num_memberships;
2040		imo->imo_membership[idx] = NULL;
2041		imo->imo_num_memberships++;
2042		KASSERT(imo->imo_mfilters != NULL,
2043		    ("%s: imf_mfilters vector was not allocated", __func__));
2044		imf = &imo->imo_mfilters[idx];
2045		KASSERT(RB_EMPTY(&imf->imf_sources),
2046		    ("%s: imf_sources not empty", __func__));
2047	}
2048
2049	/*
2050	 * Graft new source into filter list for this inpcb's
2051	 * membership of the group. The in_multi may not have
2052	 * been allocated yet if this is a new membership, however,
2053	 * the in_mfilter slot will be allocated and must be initialized.
2054	 *
2055	 * Note: Grafting of exclusive mode filters doesn't happen
2056	 * in this path.
2057	 * XXX: Should check for non-NULL lims (node exists but may
2058	 * not be in-mode) for interop with full-state API.
2059	 */
2060	if (ssa->ss.ss_family != AF_UNSPEC) {
2061		/* Membership starts in IN mode */
2062		if (is_new) {
2063			CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2064			imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2065		} else {
2066			CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2067		}
2068		lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2069		if (lims == NULL) {
2070			CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2071			    __func__);
2072			error = ENOMEM;
2073			goto out_imo_free;
2074		}
2075	} else {
2076		/* No address specified; Membership starts in EX mode */
2077		if (is_new) {
2078			CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2079			imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2080		}
2081	}
2082
2083	/*
2084	 * Begin state merge transaction at IGMP layer.
2085	 */
2086	IN_MULTI_LOCK();
2087
2088	if (is_new) {
2089		error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2090		    &inm);
2091		if (error)
2092			goto out_imo_free;
2093		imo->imo_membership[idx] = inm;
2094	} else {
2095		CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2096		error = inm_merge(inm, imf);
2097		if (error) {
2098			CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2099			    __func__);
2100			goto out_imf_rollback;
2101		}
2102		CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2103		error = igmp_change_state(inm);
2104		if (error) {
2105			CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2106			    __func__);
2107			goto out_imf_rollback;
2108		}
2109	}
2110
2111	IN_MULTI_UNLOCK();
2112
2113out_imf_rollback:
2114	INP_WLOCK_ASSERT(inp);
2115	if (error) {
2116		imf_rollback(imf);
2117		if (is_new)
2118			imf_purge(imf);
2119		else
2120			imf_reap(imf);
2121	} else {
2122		imf_commit(imf);
2123	}
2124
2125out_imo_free:
2126	if (error && is_new) {
2127		imo->imo_membership[idx] = NULL;
2128		--imo->imo_num_memberships;
2129	}
2130
2131out_inp_locked:
2132	INP_WUNLOCK(inp);
2133	return (error);
2134}
2135
2136/*
2137 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2138 */
2139static int
2140inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2141{
2142	struct group_source_req		 gsr;
2143	struct ip_mreq_source		 mreqs;
2144	sockunion_t			*gsa, *ssa;
2145	struct ifnet			*ifp;
2146	struct in_mfilter		*imf;
2147	struct ip_moptions		*imo;
2148	struct in_msource		*ims;
2149	struct in_multi			*inm;
2150	size_t				 idx;
2151	int				 error, is_final;
2152
2153	ifp = NULL;
2154	error = 0;
2155	is_final = 1;
2156
2157	memset(&gsr, 0, sizeof(struct group_source_req));
2158	gsa = (sockunion_t *)&gsr.gsr_group;
2159	gsa->ss.ss_family = AF_UNSPEC;
2160	ssa = (sockunion_t *)&gsr.gsr_source;
2161	ssa->ss.ss_family = AF_UNSPEC;
2162
2163	switch (sopt->sopt_name) {
2164	case IP_DROP_MEMBERSHIP:
2165	case IP_DROP_SOURCE_MEMBERSHIP:
2166		if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2167			error = sooptcopyin(sopt, &mreqs,
2168			    sizeof(struct ip_mreq),
2169			    sizeof(struct ip_mreq));
2170			/*
2171			 * Swap interface and sourceaddr arguments,
2172			 * as ip_mreq and ip_mreq_source are laid
2173			 * out differently.
2174			 */
2175			mreqs.imr_interface = mreqs.imr_sourceaddr;
2176			mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2177		} else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2178			error = sooptcopyin(sopt, &mreqs,
2179			    sizeof(struct ip_mreq_source),
2180			    sizeof(struct ip_mreq_source));
2181		}
2182		if (error)
2183			return (error);
2184
2185		gsa->sin.sin_family = AF_INET;
2186		gsa->sin.sin_len = sizeof(struct sockaddr_in);
2187		gsa->sin.sin_addr = mreqs.imr_multiaddr;
2188
2189		if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2190			ssa->sin.sin_family = AF_INET;
2191			ssa->sin.sin_len = sizeof(struct sockaddr_in);
2192			ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2193		}
2194
2195		/*
2196		 * Attempt to look up hinted ifp from interface address.
2197		 * Fallthrough with null ifp iff lookup fails, to
2198		 * preserve 4.4BSD mcast API idempotence.
2199		 * XXX NOTE WELL: The RFC 3678 API is preferred because
2200		 * using an IPv4 address as a key is racy.
2201		 */
2202		if (!in_nullhost(mreqs.imr_interface))
2203			INADDR_TO_IFP(mreqs.imr_interface, ifp);
2204
2205		CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
2206		    __func__, inet_ntoa(mreqs.imr_interface), ifp);
2207
2208		break;
2209
2210	case MCAST_LEAVE_GROUP:
2211	case MCAST_LEAVE_SOURCE_GROUP:
2212		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2213			error = sooptcopyin(sopt, &gsr,
2214			    sizeof(struct group_req),
2215			    sizeof(struct group_req));
2216		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2217			error = sooptcopyin(sopt, &gsr,
2218			    sizeof(struct group_source_req),
2219			    sizeof(struct group_source_req));
2220		}
2221		if (error)
2222			return (error);
2223
2224		if (gsa->sin.sin_family != AF_INET ||
2225		    gsa->sin.sin_len != sizeof(struct sockaddr_in))
2226			return (EINVAL);
2227
2228		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2229			if (ssa->sin.sin_family != AF_INET ||
2230			    ssa->sin.sin_len != sizeof(struct sockaddr_in))
2231				return (EINVAL);
2232		}
2233
2234		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2235			return (EADDRNOTAVAIL);
2236
2237		ifp = ifnet_byindex(gsr.gsr_interface);
2238
2239		if (ifp == NULL)
2240			return (EADDRNOTAVAIL);
2241		break;
2242
2243	default:
2244		CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2245		    __func__, sopt->sopt_name);
2246		return (EOPNOTSUPP);
2247		break;
2248	}
2249
2250	if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2251		return (EINVAL);
2252
2253	/*
2254	 * Find the membership in the membership array.
2255	 */
2256	imo = inp_findmoptions(inp);
2257	idx = imo_match_group(imo, ifp, &gsa->sa);
2258	if (idx == -1) {
2259		error = EADDRNOTAVAIL;
2260		goto out_inp_locked;
2261	}
2262	inm = imo->imo_membership[idx];
2263	imf = &imo->imo_mfilters[idx];
2264
2265	if (ssa->ss.ss_family != AF_UNSPEC)
2266		is_final = 0;
2267
2268	/*
2269	 * Begin state merge transaction at socket layer.
2270	 */
2271	INP_WLOCK_ASSERT(inp);
2272
2273	/*
2274	 * If we were instructed only to leave a given source, do so.
2275	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2276	 */
2277	if (is_final) {
2278		imf_leave(imf);
2279	} else {
2280		if (imf->imf_st[0] == MCAST_EXCLUDE) {
2281			error = EADDRNOTAVAIL;
2282			goto out_inp_locked;
2283		}
2284		ims = imo_match_source(imo, idx, &ssa->sa);
2285		if (ims == NULL) {
2286			CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
2287			    inet_ntoa(ssa->sin.sin_addr), "not ");
2288			error = EADDRNOTAVAIL;
2289			goto out_inp_locked;
2290		}
2291		CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2292		error = imf_prune(imf, &ssa->sin);
2293		if (error) {
2294			CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2295			    __func__);
2296			goto out_inp_locked;
2297		}
2298	}
2299
2300	/*
2301	 * Begin state merge transaction at IGMP layer.
2302	 */
2303	IN_MULTI_LOCK();
2304
2305	if (is_final) {
2306		/*
2307		 * Give up the multicast address record to which
2308		 * the membership points.
2309		 */
2310		(void)in_leavegroup_locked(inm, imf);
2311	} else {
2312		CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2313		error = inm_merge(inm, imf);
2314		if (error) {
2315			CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2316			    __func__);
2317			goto out_imf_rollback;
2318		}
2319
2320		CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2321		error = igmp_change_state(inm);
2322		if (error) {
2323			CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2324			    __func__);
2325		}
2326	}
2327
2328	IN_MULTI_UNLOCK();
2329
2330out_imf_rollback:
2331	if (error)
2332		imf_rollback(imf);
2333	else
2334		imf_commit(imf);
2335
2336	imf_reap(imf);
2337
2338	if (is_final) {
2339		/* Remove the gap in the membership and filter array. */
2340		for (++idx; idx < imo->imo_num_memberships; ++idx) {
2341			imo->imo_membership[idx-1] = imo->imo_membership[idx];
2342			imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2343		}
2344		imo->imo_num_memberships--;
2345	}
2346
2347out_inp_locked:
2348	INP_WUNLOCK(inp);
2349	return (error);
2350}
2351
2352/*
2353 * Select the interface for transmitting IPv4 multicast datagrams.
2354 *
2355 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2356 * may be passed to this socket option. An address of INADDR_ANY or an
2357 * interface index of 0 is used to remove a previous selection.
2358 * When no interface is selected, one is chosen for every send.
2359 */
2360static int
2361inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2362{
2363	struct in_addr		 addr;
2364	struct ip_mreqn		 mreqn;
2365	struct ifnet		*ifp;
2366	struct ip_moptions	*imo;
2367	int			 error;
2368
2369	if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2370		/*
2371		 * An interface index was specified using the
2372		 * Linux-derived ip_mreqn structure.
2373		 */
2374		error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2375		    sizeof(struct ip_mreqn));
2376		if (error)
2377			return (error);
2378
2379		if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2380			return (EINVAL);
2381
2382		if (mreqn.imr_ifindex == 0) {
2383			ifp = NULL;
2384		} else {
2385			ifp = ifnet_byindex(mreqn.imr_ifindex);
2386			if (ifp == NULL)
2387				return (EADDRNOTAVAIL);
2388		}
2389	} else {
2390		/*
2391		 * An interface was specified by IPv4 address.
2392		 * This is the traditional BSD usage.
2393		 */
2394		error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2395		    sizeof(struct in_addr));
2396		if (error)
2397			return (error);
2398		if (in_nullhost(addr)) {
2399			ifp = NULL;
2400		} else {
2401			INADDR_TO_IFP(addr, ifp);
2402			if (ifp == NULL)
2403				return (EADDRNOTAVAIL);
2404		}
2405		CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp,
2406		    inet_ntoa(addr));
2407	}
2408
2409	/* Reject interfaces which do not support multicast. */
2410	if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2411		return (EOPNOTSUPP);
2412
2413	imo = inp_findmoptions(inp);
2414	imo->imo_multicast_ifp = ifp;
2415	imo->imo_multicast_addr.s_addr = INADDR_ANY;
2416	INP_WUNLOCK(inp);
2417
2418	return (0);
2419}
2420
2421/*
2422 * Atomically set source filters on a socket for an IPv4 multicast group.
2423 *
2424 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2425 */
2426static int
2427inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2428{
2429	struct __msfilterreq	 msfr;
2430	sockunion_t		*gsa;
2431	struct ifnet		*ifp;
2432	struct in_mfilter	*imf;
2433	struct ip_moptions	*imo;
2434	struct in_multi		*inm;
2435	size_t			 idx;
2436	int			 error;
2437
2438	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2439	    sizeof(struct __msfilterreq));
2440	if (error)
2441		return (error);
2442
2443	if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2444		return (ENOBUFS);
2445
2446	if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2447	     msfr.msfr_fmode != MCAST_INCLUDE))
2448		return (EINVAL);
2449
2450	if (msfr.msfr_group.ss_family != AF_INET ||
2451	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2452		return (EINVAL);
2453
2454	gsa = (sockunion_t *)&msfr.msfr_group;
2455	if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2456		return (EINVAL);
2457
2458	gsa->sin.sin_port = 0;	/* ignore port */
2459
2460	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2461		return (EADDRNOTAVAIL);
2462
2463	ifp = ifnet_byindex(msfr.msfr_ifindex);
2464	if (ifp == NULL)
2465		return (EADDRNOTAVAIL);
2466
2467	/*
2468	 * Take the INP write lock.
2469	 * Check if this socket is a member of this group.
2470	 */
2471	imo = inp_findmoptions(inp);
2472	idx = imo_match_group(imo, ifp, &gsa->sa);
2473	if (idx == -1 || imo->imo_mfilters == NULL) {
2474		error = EADDRNOTAVAIL;
2475		goto out_inp_locked;
2476	}
2477	inm = imo->imo_membership[idx];
2478	imf = &imo->imo_mfilters[idx];
2479
2480	/*
2481	 * Begin state merge transaction at socket layer.
2482	 */
2483	INP_WLOCK_ASSERT(inp);
2484
2485	imf->imf_st[1] = msfr.msfr_fmode;
2486
2487	/*
2488	 * Apply any new source filters, if present.
2489	 * Make a copy of the user-space source vector so
2490	 * that we may copy them with a single copyin. This
2491	 * allows us to deal with page faults up-front.
2492	 */
2493	if (msfr.msfr_nsrcs > 0) {
2494		struct in_msource	*lims;
2495		struct sockaddr_in	*psin;
2496		struct sockaddr_storage	*kss, *pkss;
2497		int			 i;
2498
2499		INP_WUNLOCK(inp);
2500
2501		CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2502		    __func__, (unsigned long)msfr.msfr_nsrcs);
2503		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2504		    M_TEMP, M_WAITOK);
2505		error = copyin(msfr.msfr_srcs, kss,
2506		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2507		if (error) {
2508			free(kss, M_TEMP);
2509			return (error);
2510		}
2511
2512		INP_WLOCK(inp);
2513
2514		/*
2515		 * Mark all source filters as UNDEFINED at t1.
2516		 * Restore new group filter mode, as imf_leave()
2517		 * will set it to INCLUDE.
2518		 */
2519		imf_leave(imf);
2520		imf->imf_st[1] = msfr.msfr_fmode;
2521
2522		/*
2523		 * Update socket layer filters at t1, lazy-allocating
2524		 * new entries. This saves a bunch of memory at the
2525		 * cost of one RB_FIND() per source entry; duplicate
2526		 * entries in the msfr_nsrcs vector are ignored.
2527		 * If we encounter an error, rollback transaction.
2528		 *
2529		 * XXX This too could be replaced with a set-symmetric
2530		 * difference like loop to avoid walking from root
2531		 * every time, as the key space is common.
2532		 */
2533		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2534			psin = (struct sockaddr_in *)pkss;
2535			if (psin->sin_family != AF_INET) {
2536				error = EAFNOSUPPORT;
2537				break;
2538			}
2539			if (psin->sin_len != sizeof(struct sockaddr_in)) {
2540				error = EINVAL;
2541				break;
2542			}
2543			error = imf_get_source(imf, psin, &lims);
2544			if (error)
2545				break;
2546			lims->imsl_st[1] = imf->imf_st[1];
2547		}
2548		free(kss, M_TEMP);
2549	}
2550
2551	if (error)
2552		goto out_imf_rollback;
2553
2554	INP_WLOCK_ASSERT(inp);
2555	IN_MULTI_LOCK();
2556
2557	/*
2558	 * Begin state merge transaction at IGMP layer.
2559	 */
2560	CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2561	error = inm_merge(inm, imf);
2562	if (error) {
2563		CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2564		goto out_imf_rollback;
2565	}
2566
2567	CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2568	error = igmp_change_state(inm);
2569	if (error)
2570		CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2571
2572	IN_MULTI_UNLOCK();
2573
2574out_imf_rollback:
2575	if (error)
2576		imf_rollback(imf);
2577	else
2578		imf_commit(imf);
2579
2580	imf_reap(imf);
2581
2582out_inp_locked:
2583	INP_WUNLOCK(inp);
2584	return (error);
2585}
2586
2587/*
2588 * Set the IP multicast options in response to user setsockopt().
2589 *
2590 * Many of the socket options handled in this function duplicate the
2591 * functionality of socket options in the regular unicast API. However,
2592 * it is not possible to merge the duplicate code, because the idempotence
2593 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2594 * the effects of these options must be treated as separate and distinct.
2595 *
2596 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2597 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2598 * is refactored to no longer use vifs.
2599 */
2600int
2601inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2602{
2603	struct ip_moptions	*imo;
2604	int			 error;
2605
2606	error = 0;
2607
2608	/*
2609	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2610	 * or is a divert socket, reject it.
2611	 */
2612	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2613	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2614	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2615		return (EOPNOTSUPP);
2616
2617	switch (sopt->sopt_name) {
2618	case IP_MULTICAST_VIF: {
2619		int vifi;
2620		/*
2621		 * Select a multicast VIF for transmission.
2622		 * Only useful if multicast forwarding is active.
2623		 */
2624		if (legal_vif_num == NULL) {
2625			error = EOPNOTSUPP;
2626			break;
2627		}
2628		error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2629		if (error)
2630			break;
2631		if (!legal_vif_num(vifi) && (vifi != -1)) {
2632			error = EINVAL;
2633			break;
2634		}
2635		imo = inp_findmoptions(inp);
2636		imo->imo_multicast_vif = vifi;
2637		INP_WUNLOCK(inp);
2638		break;
2639	}
2640
2641	case IP_MULTICAST_IF:
2642		error = inp_set_multicast_if(inp, sopt);
2643		break;
2644
2645	case IP_MULTICAST_TTL: {
2646		u_char ttl;
2647
2648		/*
2649		 * Set the IP time-to-live for outgoing multicast packets.
2650		 * The original multicast API required a char argument,
2651		 * which is inconsistent with the rest of the socket API.
2652		 * We allow either a char or an int.
2653		 */
2654		if (sopt->sopt_valsize == sizeof(u_char)) {
2655			error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2656			    sizeof(u_char));
2657			if (error)
2658				break;
2659		} else {
2660			u_int ittl;
2661
2662			error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2663			    sizeof(u_int));
2664			if (error)
2665				break;
2666			if (ittl > 255) {
2667				error = EINVAL;
2668				break;
2669			}
2670			ttl = (u_char)ittl;
2671		}
2672		imo = inp_findmoptions(inp);
2673		imo->imo_multicast_ttl = ttl;
2674		INP_WUNLOCK(inp);
2675		break;
2676	}
2677
2678	case IP_MULTICAST_LOOP: {
2679		u_char loop;
2680
2681		/*
2682		 * Set the loopback flag for outgoing multicast packets.
2683		 * Must be zero or one.  The original multicast API required a
2684		 * char argument, which is inconsistent with the rest
2685		 * of the socket API.  We allow either a char or an int.
2686		 */
2687		if (sopt->sopt_valsize == sizeof(u_char)) {
2688			error = sooptcopyin(sopt, &loop, sizeof(u_char),
2689			    sizeof(u_char));
2690			if (error)
2691				break;
2692		} else {
2693			u_int iloop;
2694
2695			error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2696					    sizeof(u_int));
2697			if (error)
2698				break;
2699			loop = (u_char)iloop;
2700		}
2701		imo = inp_findmoptions(inp);
2702		imo->imo_multicast_loop = !!loop;
2703		INP_WUNLOCK(inp);
2704		break;
2705	}
2706
2707	case IP_ADD_MEMBERSHIP:
2708	case IP_ADD_SOURCE_MEMBERSHIP:
2709	case MCAST_JOIN_GROUP:
2710	case MCAST_JOIN_SOURCE_GROUP:
2711		error = inp_join_group(inp, sopt);
2712		break;
2713
2714	case IP_DROP_MEMBERSHIP:
2715	case IP_DROP_SOURCE_MEMBERSHIP:
2716	case MCAST_LEAVE_GROUP:
2717	case MCAST_LEAVE_SOURCE_GROUP:
2718		error = inp_leave_group(inp, sopt);
2719		break;
2720
2721	case IP_BLOCK_SOURCE:
2722	case IP_UNBLOCK_SOURCE:
2723	case MCAST_BLOCK_SOURCE:
2724	case MCAST_UNBLOCK_SOURCE:
2725		error = inp_block_unblock_source(inp, sopt);
2726		break;
2727
2728	case IP_MSFILTER:
2729		error = inp_set_source_filters(inp, sopt);
2730		break;
2731
2732	default:
2733		error = EOPNOTSUPP;
2734		break;
2735	}
2736
2737	INP_UNLOCK_ASSERT(inp);
2738
2739	return (error);
2740}
2741
2742/*
2743 * Expose IGMP's multicast filter mode and source list(s) to userland,
2744 * keyed by (ifindex, group).
2745 * The filter mode is written out as a uint32_t, followed by
2746 * 0..n of struct in_addr.
2747 * For use by ifmcstat(8).
2748 * SMPng: NOTE: unlocked read of ifindex space.
2749 */
2750static int
2751sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2752{
2753	struct in_addr			 src, group;
2754	struct ifnet			*ifp;
2755	struct ifmultiaddr		*ifma;
2756	struct in_multi			*inm;
2757	struct ip_msource		*ims;
2758	int				*name;
2759	int				 retval;
2760	u_int				 namelen;
2761	uint32_t			 fmode, ifindex;
2762
2763	name = (int *)arg1;
2764	namelen = arg2;
2765
2766	if (req->newptr != NULL)
2767		return (EPERM);
2768
2769	if (namelen != 2)
2770		return (EINVAL);
2771
2772	ifindex = name[0];
2773	if (ifindex <= 0 || ifindex > V_if_index) {
2774		CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2775		    __func__, ifindex);
2776		return (ENOENT);
2777	}
2778
2779	group.s_addr = name[1];
2780	if (!IN_MULTICAST(ntohl(group.s_addr))) {
2781		CTR2(KTR_IGMPV3, "%s: group %s is not multicast",
2782		    __func__, inet_ntoa(group));
2783		return (EINVAL);
2784	}
2785
2786	ifp = ifnet_byindex(ifindex);
2787	if (ifp == NULL) {
2788		CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2789		    __func__, ifindex);
2790		return (ENOENT);
2791	}
2792
2793	retval = sysctl_wire_old_buffer(req,
2794	    sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2795	if (retval)
2796		return (retval);
2797
2798	IN_MULTI_LOCK();
2799
2800	IF_ADDR_LOCK(ifp);
2801	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2802		if (ifma->ifma_addr->sa_family != AF_INET ||
2803		    ifma->ifma_protospec == NULL)
2804			continue;
2805		inm = (struct in_multi *)ifma->ifma_protospec;
2806		if (!in_hosteq(inm->inm_addr, group))
2807			continue;
2808		fmode = inm->inm_st[1].iss_fmode;
2809		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2810		if (retval != 0)
2811			break;
2812		RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2813#ifdef KTR
2814			struct in_addr ina;
2815			ina.s_addr = htonl(ims->ims_haddr);
2816			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2817			    inet_ntoa(ina));
2818#endif
2819			/*
2820			 * Only copy-out sources which are in-mode.
2821			 */
2822			if (fmode != ims_get_mode(inm, ims, 1)) {
2823				CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2824				    __func__);
2825				continue;
2826			}
2827			src.s_addr = htonl(ims->ims_haddr);
2828			retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2829			if (retval != 0)
2830				break;
2831		}
2832	}
2833	IF_ADDR_UNLOCK(ifp);
2834
2835	IN_MULTI_UNLOCK();
2836
2837	return (retval);
2838}
2839
2840#ifdef KTR
2841
2842static const char *inm_modestrs[] = { "un", "in", "ex" };
2843
2844static const char *
2845inm_mode_str(const int mode)
2846{
2847
2848	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2849		return (inm_modestrs[mode]);
2850	return ("??");
2851}
2852
2853static const char *inm_statestrs[] = {
2854	"not-member",
2855	"silent",
2856	"idle",
2857	"lazy",
2858	"sleeping",
2859	"awakening",
2860	"query-pending",
2861	"sg-query-pending",
2862	"leaving"
2863};
2864
2865static const char *
2866inm_state_str(const int state)
2867{
2868
2869	if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2870		return (inm_statestrs[state]);
2871	return ("??");
2872}
2873
2874/*
2875 * Dump an in_multi structure to the console.
2876 */
2877void
2878inm_print(const struct in_multi *inm)
2879{
2880	int t;
2881
2882	if ((ktr_mask & KTR_IGMPV3) == 0)
2883		return;
2884
2885	printf("%s: --- begin inm %p ---\n", __func__, inm);
2886	printf("addr %s ifp %p(%s) ifma %p\n",
2887	    inet_ntoa(inm->inm_addr),
2888	    inm->inm_ifp,
2889	    inm->inm_ifp->if_xname,
2890	    inm->inm_ifma);
2891	printf("timer %u state %s refcount %u scq.len %u\n",
2892	    inm->inm_timer,
2893	    inm_state_str(inm->inm_state),
2894	    inm->inm_refcount,
2895	    inm->inm_scq.ifq_len);
2896	printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2897	    inm->inm_igi,
2898	    inm->inm_nsrc,
2899	    inm->inm_sctimer,
2900	    inm->inm_scrv);
2901	for (t = 0; t < 2; t++) {
2902		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2903		    inm_mode_str(inm->inm_st[t].iss_fmode),
2904		    inm->inm_st[t].iss_asm,
2905		    inm->inm_st[t].iss_ex,
2906		    inm->inm_st[t].iss_in,
2907		    inm->inm_st[t].iss_rec);
2908	}
2909	printf("%s: --- end inm %p ---\n", __func__, inm);
2910}
2911
2912#else /* !KTR */
2913
2914void
2915inm_print(const struct in_multi *inm)
2916{
2917
2918}
2919
2920#endif /* KTR */
2921
2922RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);
2923