xref: /freebsd/sys/netinet/in_mcast.c (revision 273c26a3c3bea87a241d6879abd4f991db180bf0)
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("$FreeBSD$");
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/protosw.h>
45 #include <sys/rmlock.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/protosw.h>
49 #include <sys/sysctl.h>
50 #include <sys/ktr.h>
51 #include <sys/taskqueue.h>
52 #include <sys/tree.h>
53 
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/if_dl.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_fib.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip_var.h>
66 #include <netinet/igmp_var.h>
67 
68 #ifndef KTR_IGMPV3
69 #define KTR_IGMPV3 KTR_INET
70 #endif
71 
72 #ifndef __SOCKUNION_DECLARED
73 union sockunion {
74 	struct sockaddr_storage	ss;
75 	struct sockaddr		sa;
76 	struct sockaddr_dl	sdl;
77 	struct sockaddr_in	sin;
78 };
79 typedef union sockunion sockunion_t;
80 #define __SOCKUNION_DECLARED
81 #endif /* __SOCKUNION_DECLARED */
82 
83 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
84     "IPv4 multicast PCB-layer source filter");
85 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
86 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
87 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
88     "IPv4 multicast IGMP-layer source filter");
89 
90 /*
91  * Locking:
92  * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
93  * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
94  *   it can be taken by code in net/if.c also.
95  * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
96  *
97  * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly
98  * any need for in_multi itself to be virtualized -- it is bound to an ifp
99  * anyway no matter what happens.
100  */
101 struct mtx in_multi_mtx;
102 MTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF);
103 
104 /*
105  * Functions with non-static linkage defined in this file should be
106  * declared in in_var.h:
107  *  imo_multi_filter()
108  *  in_addmulti()
109  *  in_delmulti()
110  *  in_joingroup()
111  *  in_joingroup_locked()
112  *  in_leavegroup()
113  *  in_leavegroup_locked()
114  * and ip_var.h:
115  *  inp_freemoptions()
116  *  inp_getmoptions()
117  *  inp_setmoptions()
118  *
119  * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
120  * and in_delmulti().
121  */
122 static void	imf_commit(struct in_mfilter *);
123 static int	imf_get_source(struct in_mfilter *imf,
124 		    const struct sockaddr_in *psin,
125 		    struct in_msource **);
126 static struct in_msource *
127 		imf_graft(struct in_mfilter *, const uint8_t,
128 		    const struct sockaddr_in *);
129 static void	imf_leave(struct in_mfilter *);
130 static int	imf_prune(struct in_mfilter *, const struct sockaddr_in *);
131 static void	imf_purge(struct in_mfilter *);
132 static void	imf_rollback(struct in_mfilter *);
133 static void	imf_reap(struct in_mfilter *);
134 static int	imo_grow(struct ip_moptions *);
135 static size_t	imo_match_group(const struct ip_moptions *,
136 		    const struct ifnet *, const struct sockaddr *);
137 static struct in_msource *
138 		imo_match_source(const struct ip_moptions *, const size_t,
139 		    const struct sockaddr *);
140 static void	ims_merge(struct ip_msource *ims,
141 		    const struct in_msource *lims, const int rollback);
142 static int	in_getmulti(struct ifnet *, const struct in_addr *,
143 		    struct in_multi **);
144 static int	inm_get_source(struct in_multi *inm, const in_addr_t haddr,
145 		    const int noalloc, struct ip_msource **pims);
146 #ifdef KTR
147 static int	inm_is_ifp_detached(const struct in_multi *);
148 #endif
149 static int	inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
150 static void	inm_purge(struct in_multi *);
151 static void	inm_reap(struct in_multi *);
152 static struct ip_moptions *
153 		inp_findmoptions(struct inpcb *);
154 static void	inp_freemoptions_internal(struct ip_moptions *);
155 static void	inp_gcmoptions(void *, int);
156 static int	inp_get_source_filters(struct inpcb *, struct sockopt *);
157 static int	inp_join_group(struct inpcb *, struct sockopt *);
158 static int	inp_leave_group(struct inpcb *, struct sockopt *);
159 static struct ifnet *
160 		inp_lookup_mcast_ifp(const struct inpcb *,
161 		    const struct sockaddr_in *, const struct in_addr);
162 static int	inp_block_unblock_source(struct inpcb *, struct sockopt *);
163 static int	inp_set_multicast_if(struct inpcb *, struct sockopt *);
164 static int	inp_set_source_filters(struct inpcb *, struct sockopt *);
165 static int	sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
166 
167 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
168     "IPv4 multicast");
169 
170 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
171 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
172     CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
173     "Max source filters per group");
174 
175 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
176 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
177     CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
178     "Max source filters per socket");
179 
180 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
181 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
182     &in_mcast_loop, 0, "Loopback multicast datagrams by default");
183 
184 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
185     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
186     "Per-interface stack-wide source filters");
187 
188 static STAILQ_HEAD(, ip_moptions) imo_gc_list =
189     STAILQ_HEAD_INITIALIZER(imo_gc_list);
190 static struct task imo_gc_task = TASK_INITIALIZER(0, inp_gcmoptions, NULL);
191 
192 #ifdef KTR
193 /*
194  * Inline function which wraps assertions for a valid ifp.
195  * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
196  * is detached.
197  */
198 static int __inline
199 inm_is_ifp_detached(const struct in_multi *inm)
200 {
201 	struct ifnet *ifp;
202 
203 	KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
204 	ifp = inm->inm_ifma->ifma_ifp;
205 	if (ifp != NULL) {
206 		/*
207 		 * Sanity check that netinet's notion of ifp is the
208 		 * same as net's.
209 		 */
210 		KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
211 	}
212 
213 	return (ifp == NULL);
214 }
215 #endif
216 
217 /*
218  * Initialize an in_mfilter structure to a known state at t0, t1
219  * with an empty source filter list.
220  */
221 static __inline void
222 imf_init(struct in_mfilter *imf, const int st0, const int st1)
223 {
224 	memset(imf, 0, sizeof(struct in_mfilter));
225 	RB_INIT(&imf->imf_sources);
226 	imf->imf_st[0] = st0;
227 	imf->imf_st[1] = st1;
228 }
229 
230 /*
231  * Function for looking up an in_multi record for an IPv4 multicast address
232  * on a given interface. ifp must be valid. If no record found, return NULL.
233  * The IN_MULTI_LOCK and IF_ADDR_LOCK on ifp must be held.
234  */
235 struct in_multi *
236 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
237 {
238 	struct ifmultiaddr *ifma;
239 	struct in_multi *inm;
240 
241 	IN_MULTI_LOCK_ASSERT();
242 	IF_ADDR_LOCK_ASSERT(ifp);
243 
244 	inm = NULL;
245 	TAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
246 		if (ifma->ifma_addr->sa_family == AF_INET) {
247 			inm = (struct in_multi *)ifma->ifma_protospec;
248 			if (inm->inm_addr.s_addr == ina.s_addr)
249 				break;
250 			inm = NULL;
251 		}
252 	}
253 	return (inm);
254 }
255 
256 /*
257  * Wrapper for inm_lookup_locked().
258  * The IF_ADDR_LOCK will be taken on ifp and released on return.
259  */
260 struct in_multi *
261 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
262 {
263 	struct in_multi *inm;
264 
265 	IN_MULTI_LOCK_ASSERT();
266 	IF_ADDR_RLOCK(ifp);
267 	inm = inm_lookup_locked(ifp, ina);
268 	IF_ADDR_RUNLOCK(ifp);
269 
270 	return (inm);
271 }
272 
273 /*
274  * Resize the ip_moptions vector to the next power-of-two minus 1.
275  * May be called with locks held; do not sleep.
276  */
277 static int
278 imo_grow(struct ip_moptions *imo)
279 {
280 	struct in_multi		**nmships;
281 	struct in_multi		**omships;
282 	struct in_mfilter	 *nmfilters;
283 	struct in_mfilter	 *omfilters;
284 	size_t			  idx;
285 	size_t			  newmax;
286 	size_t			  oldmax;
287 
288 	nmships = NULL;
289 	nmfilters = NULL;
290 	omships = imo->imo_membership;
291 	omfilters = imo->imo_mfilters;
292 	oldmax = imo->imo_max_memberships;
293 	newmax = ((oldmax + 1) * 2) - 1;
294 
295 	if (newmax <= IP_MAX_MEMBERSHIPS) {
296 		nmships = (struct in_multi **)realloc(omships,
297 		    sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
298 		nmfilters = (struct in_mfilter *)realloc(omfilters,
299 		    sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
300 		if (nmships != NULL && nmfilters != NULL) {
301 			/* Initialize newly allocated source filter heads. */
302 			for (idx = oldmax; idx < newmax; idx++) {
303 				imf_init(&nmfilters[idx], MCAST_UNDEFINED,
304 				    MCAST_EXCLUDE);
305 			}
306 			imo->imo_max_memberships = newmax;
307 			imo->imo_membership = nmships;
308 			imo->imo_mfilters = nmfilters;
309 		}
310 	}
311 
312 	if (nmships == NULL || nmfilters == NULL) {
313 		if (nmships != NULL)
314 			free(nmships, M_IPMOPTS);
315 		if (nmfilters != NULL)
316 			free(nmfilters, M_INMFILTER);
317 		return (ETOOMANYREFS);
318 	}
319 
320 	return (0);
321 }
322 
323 /*
324  * Find an IPv4 multicast group entry for this ip_moptions instance
325  * which matches the specified group, and optionally an interface.
326  * Return its index into the array, or -1 if not found.
327  */
328 static size_t
329 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
330     const struct sockaddr *group)
331 {
332 	const struct sockaddr_in *gsin;
333 	struct in_multi	**pinm;
334 	int		  idx;
335 	int		  nmships;
336 
337 	gsin = (const struct sockaddr_in *)group;
338 
339 	/* The imo_membership array may be lazy allocated. */
340 	if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
341 		return (-1);
342 
343 	nmships = imo->imo_num_memberships;
344 	pinm = &imo->imo_membership[0];
345 	for (idx = 0; idx < nmships; idx++, pinm++) {
346 		if (*pinm == NULL)
347 			continue;
348 		if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
349 		    in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
350 			break;
351 		}
352 	}
353 	if (idx >= nmships)
354 		idx = -1;
355 
356 	return (idx);
357 }
358 
359 /*
360  * Find an IPv4 multicast source entry for this imo which matches
361  * the given group index for this socket, and source address.
362  *
363  * NOTE: This does not check if the entry is in-mode, merely if
364  * it exists, which may not be the desired behaviour.
365  */
366 static struct in_msource *
367 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
368     const struct sockaddr *src)
369 {
370 	struct ip_msource	 find;
371 	struct in_mfilter	*imf;
372 	struct ip_msource	*ims;
373 	const sockunion_t	*psa;
374 
375 	KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
376 	KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
377 	    ("%s: invalid index %d\n", __func__, (int)gidx));
378 
379 	/* The imo_mfilters array may be lazy allocated. */
380 	if (imo->imo_mfilters == NULL)
381 		return (NULL);
382 	imf = &imo->imo_mfilters[gidx];
383 
384 	/* Source trees are keyed in host byte order. */
385 	psa = (const sockunion_t *)src;
386 	find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
387 	ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
388 
389 	return ((struct in_msource *)ims);
390 }
391 
392 /*
393  * Perform filtering for multicast datagrams on a socket by group and source.
394  *
395  * Returns 0 if a datagram should be allowed through, or various error codes
396  * if the socket was not a member of the group, or the source was muted, etc.
397  */
398 int
399 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
400     const struct sockaddr *group, const struct sockaddr *src)
401 {
402 	size_t gidx;
403 	struct in_msource *ims;
404 	int mode;
405 
406 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
407 
408 	gidx = imo_match_group(imo, ifp, group);
409 	if (gidx == -1)
410 		return (MCAST_NOTGMEMBER);
411 
412 	/*
413 	 * Check if the source was included in an (S,G) join.
414 	 * Allow reception on exclusive memberships by default,
415 	 * reject reception on inclusive memberships by default.
416 	 * Exclude source only if an in-mode exclude filter exists.
417 	 * Include source only if an in-mode include filter exists.
418 	 * NOTE: We are comparing group state here at IGMP t1 (now)
419 	 * with socket-layer t0 (since last downcall).
420 	 */
421 	mode = imo->imo_mfilters[gidx].imf_st[1];
422 	ims = imo_match_source(imo, gidx, src);
423 
424 	if ((ims == NULL && mode == MCAST_INCLUDE) ||
425 	    (ims != NULL && ims->imsl_st[0] != mode))
426 		return (MCAST_NOTSMEMBER);
427 
428 	return (MCAST_PASS);
429 }
430 
431 /*
432  * Find and return a reference to an in_multi record for (ifp, group),
433  * and bump its reference count.
434  * If one does not exist, try to allocate it, and update link-layer multicast
435  * filters on ifp to listen for group.
436  * Assumes the IN_MULTI lock is held across the call.
437  * Return 0 if successful, otherwise return an appropriate error code.
438  */
439 static int
440 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
441     struct in_multi **pinm)
442 {
443 	struct sockaddr_in	 gsin;
444 	struct ifmultiaddr	*ifma;
445 	struct in_ifinfo	*ii;
446 	struct in_multi		*inm;
447 	int error;
448 
449 	IN_MULTI_LOCK_ASSERT();
450 
451 	ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
452 
453 	inm = inm_lookup(ifp, *group);
454 	if (inm != NULL) {
455 		/*
456 		 * If we already joined this group, just bump the
457 		 * refcount and return it.
458 		 */
459 		KASSERT(inm->inm_refcount >= 1,
460 		    ("%s: bad refcount %d", __func__, inm->inm_refcount));
461 		++inm->inm_refcount;
462 		*pinm = inm;
463 		return (0);
464 	}
465 
466 	memset(&gsin, 0, sizeof(gsin));
467 	gsin.sin_family = AF_INET;
468 	gsin.sin_len = sizeof(struct sockaddr_in);
469 	gsin.sin_addr = *group;
470 
471 	/*
472 	 * Check if a link-layer group is already associated
473 	 * with this network-layer group on the given ifnet.
474 	 */
475 	error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
476 	if (error != 0)
477 		return (error);
478 
479 	/* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
480 	IF_ADDR_WLOCK(ifp);
481 
482 	/*
483 	 * If something other than netinet is occupying the link-layer
484 	 * group, print a meaningful error message and back out of
485 	 * the allocation.
486 	 * Otherwise, bump the refcount on the existing network-layer
487 	 * group association and return it.
488 	 */
489 	if (ifma->ifma_protospec != NULL) {
490 		inm = (struct in_multi *)ifma->ifma_protospec;
491 #ifdef INVARIANTS
492 		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
493 		    __func__));
494 		KASSERT(ifma->ifma_addr->sa_family == AF_INET,
495 		    ("%s: ifma not AF_INET", __func__));
496 		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
497 		if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
498 		    !in_hosteq(inm->inm_addr, *group))
499 			panic("%s: ifma %p is inconsistent with %p (%s)",
500 			    __func__, ifma, inm, inet_ntoa(*group));
501 #endif
502 		++inm->inm_refcount;
503 		*pinm = inm;
504 		IF_ADDR_WUNLOCK(ifp);
505 		return (0);
506 	}
507 
508 	IF_ADDR_WLOCK_ASSERT(ifp);
509 
510 	/*
511 	 * A new in_multi record is needed; allocate and initialize it.
512 	 * We DO NOT perform an IGMP join as the in_ layer may need to
513 	 * push an initial source list down to IGMP to support SSM.
514 	 *
515 	 * The initial source filter state is INCLUDE, {} as per the RFC.
516 	 */
517 	inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
518 	if (inm == NULL) {
519 		IF_ADDR_WUNLOCK(ifp);
520 		if_delmulti_ifma(ifma);
521 		return (ENOMEM);
522 	}
523 	inm->inm_addr = *group;
524 	inm->inm_ifp = ifp;
525 	inm->inm_igi = ii->ii_igmp;
526 	inm->inm_ifma = ifma;
527 	inm->inm_refcount = 1;
528 	inm->inm_state = IGMP_NOT_MEMBER;
529 	mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
530 	inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
531 	inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
532 	RB_INIT(&inm->inm_srcs);
533 
534 	ifma->ifma_protospec = inm;
535 
536 	*pinm = inm;
537 
538 	IF_ADDR_WUNLOCK(ifp);
539 	return (0);
540 }
541 
542 /*
543  * Drop a reference to an in_multi record.
544  *
545  * If the refcount drops to 0, free the in_multi record and
546  * delete the underlying link-layer membership.
547  */
548 void
549 inm_release_locked(struct in_multi *inm)
550 {
551 	struct ifmultiaddr *ifma;
552 
553 	IN_MULTI_LOCK_ASSERT();
554 
555 	CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
556 
557 	if (--inm->inm_refcount > 0) {
558 		CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__,
559 		    inm->inm_refcount);
560 		return;
561 	}
562 
563 	CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
564 
565 	ifma = inm->inm_ifma;
566 
567 	/* XXX this access is not covered by IF_ADDR_LOCK */
568 	CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
569 	KASSERT(ifma->ifma_protospec == inm,
570 	    ("%s: ifma_protospec != inm", __func__));
571 	ifma->ifma_protospec = NULL;
572 
573 	inm_purge(inm);
574 
575 	free(inm, M_IPMADDR);
576 
577 	if_delmulti_ifma(ifma);
578 }
579 
580 /*
581  * Clear recorded source entries for a group.
582  * Used by the IGMP code. Caller must hold the IN_MULTI lock.
583  * FIXME: Should reap.
584  */
585 void
586 inm_clear_recorded(struct in_multi *inm)
587 {
588 	struct ip_msource	*ims;
589 
590 	IN_MULTI_LOCK_ASSERT();
591 
592 	RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
593 		if (ims->ims_stp) {
594 			ims->ims_stp = 0;
595 			--inm->inm_st[1].iss_rec;
596 		}
597 	}
598 	KASSERT(inm->inm_st[1].iss_rec == 0,
599 	    ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
600 }
601 
602 /*
603  * Record a source as pending for a Source-Group IGMPv3 query.
604  * This lives here as it modifies the shared tree.
605  *
606  * inm is the group descriptor.
607  * naddr is the address of the source to record in network-byte order.
608  *
609  * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
610  * lazy-allocate a source node in response to an SG query.
611  * Otherwise, no allocation is performed. This saves some memory
612  * with the trade-off that the source will not be reported to the
613  * router if joined in the window between the query response and
614  * the group actually being joined on the local host.
615  *
616  * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
617  * This turns off the allocation of a recorded source entry if
618  * the group has not been joined.
619  *
620  * Return 0 if the source didn't exist or was already marked as recorded.
621  * Return 1 if the source was marked as recorded by this function.
622  * Return <0 if any error occurred (negated errno code).
623  */
624 int
625 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
626 {
627 	struct ip_msource	 find;
628 	struct ip_msource	*ims, *nims;
629 
630 	IN_MULTI_LOCK_ASSERT();
631 
632 	find.ims_haddr = ntohl(naddr);
633 	ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
634 	if (ims && ims->ims_stp)
635 		return (0);
636 	if (ims == NULL) {
637 		if (inm->inm_nsrc == in_mcast_maxgrpsrc)
638 			return (-ENOSPC);
639 		nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
640 		    M_NOWAIT | M_ZERO);
641 		if (nims == NULL)
642 			return (-ENOMEM);
643 		nims->ims_haddr = find.ims_haddr;
644 		RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
645 		++inm->inm_nsrc;
646 		ims = nims;
647 	}
648 
649 	/*
650 	 * Mark the source as recorded and update the recorded
651 	 * source count.
652 	 */
653 	++ims->ims_stp;
654 	++inm->inm_st[1].iss_rec;
655 
656 	return (1);
657 }
658 
659 /*
660  * Return a pointer to an in_msource owned by an in_mfilter,
661  * given its source address.
662  * Lazy-allocate if needed. If this is a new entry its filter state is
663  * undefined at t0.
664  *
665  * imf is the filter set being modified.
666  * haddr is the source address in *host* byte-order.
667  *
668  * SMPng: May be called with locks held; malloc must not block.
669  */
670 static int
671 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
672     struct in_msource **plims)
673 {
674 	struct ip_msource	 find;
675 	struct ip_msource	*ims, *nims;
676 	struct in_msource	*lims;
677 	int			 error;
678 
679 	error = 0;
680 	ims = NULL;
681 	lims = NULL;
682 
683 	/* key is host byte order */
684 	find.ims_haddr = ntohl(psin->sin_addr.s_addr);
685 	ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
686 	lims = (struct in_msource *)ims;
687 	if (lims == NULL) {
688 		if (imf->imf_nsrc == in_mcast_maxsocksrc)
689 			return (ENOSPC);
690 		nims = malloc(sizeof(struct in_msource), M_INMFILTER,
691 		    M_NOWAIT | M_ZERO);
692 		if (nims == NULL)
693 			return (ENOMEM);
694 		lims = (struct in_msource *)nims;
695 		lims->ims_haddr = find.ims_haddr;
696 		lims->imsl_st[0] = MCAST_UNDEFINED;
697 		RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
698 		++imf->imf_nsrc;
699 	}
700 
701 	*plims = lims;
702 
703 	return (error);
704 }
705 
706 /*
707  * Graft a source entry into an existing socket-layer filter set,
708  * maintaining any required invariants and checking allocations.
709  *
710  * The source is marked as being in the new filter mode at t1.
711  *
712  * Return the pointer to the new node, otherwise return NULL.
713  */
714 static struct in_msource *
715 imf_graft(struct in_mfilter *imf, const uint8_t st1,
716     const struct sockaddr_in *psin)
717 {
718 	struct ip_msource	*nims;
719 	struct in_msource	*lims;
720 
721 	nims = malloc(sizeof(struct in_msource), M_INMFILTER,
722 	    M_NOWAIT | M_ZERO);
723 	if (nims == NULL)
724 		return (NULL);
725 	lims = (struct in_msource *)nims;
726 	lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
727 	lims->imsl_st[0] = MCAST_UNDEFINED;
728 	lims->imsl_st[1] = st1;
729 	RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
730 	++imf->imf_nsrc;
731 
732 	return (lims);
733 }
734 
735 /*
736  * Prune a source entry from an existing socket-layer filter set,
737  * maintaining any required invariants and checking allocations.
738  *
739  * The source is marked as being left at t1, it is not freed.
740  *
741  * Return 0 if no error occurred, otherwise return an errno value.
742  */
743 static int
744 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
745 {
746 	struct ip_msource	 find;
747 	struct ip_msource	*ims;
748 	struct in_msource	*lims;
749 
750 	/* key is host byte order */
751 	find.ims_haddr = ntohl(psin->sin_addr.s_addr);
752 	ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
753 	if (ims == NULL)
754 		return (ENOENT);
755 	lims = (struct in_msource *)ims;
756 	lims->imsl_st[1] = MCAST_UNDEFINED;
757 	return (0);
758 }
759 
760 /*
761  * Revert socket-layer filter set deltas at t1 to t0 state.
762  */
763 static void
764 imf_rollback(struct in_mfilter *imf)
765 {
766 	struct ip_msource	*ims, *tims;
767 	struct in_msource	*lims;
768 
769 	RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
770 		lims = (struct in_msource *)ims;
771 		if (lims->imsl_st[0] == lims->imsl_st[1]) {
772 			/* no change at t1 */
773 			continue;
774 		} else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
775 			/* revert change to existing source at t1 */
776 			lims->imsl_st[1] = lims->imsl_st[0];
777 		} else {
778 			/* revert source added t1 */
779 			CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
780 			RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
781 			free(ims, M_INMFILTER);
782 			imf->imf_nsrc--;
783 		}
784 	}
785 	imf->imf_st[1] = imf->imf_st[0];
786 }
787 
788 /*
789  * Mark socket-layer filter set as INCLUDE {} at t1.
790  */
791 static void
792 imf_leave(struct in_mfilter *imf)
793 {
794 	struct ip_msource	*ims;
795 	struct in_msource	*lims;
796 
797 	RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
798 		lims = (struct in_msource *)ims;
799 		lims->imsl_st[1] = MCAST_UNDEFINED;
800 	}
801 	imf->imf_st[1] = MCAST_INCLUDE;
802 }
803 
804 /*
805  * Mark socket-layer filter set deltas as committed.
806  */
807 static void
808 imf_commit(struct in_mfilter *imf)
809 {
810 	struct ip_msource	*ims;
811 	struct in_msource	*lims;
812 
813 	RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
814 		lims = (struct in_msource *)ims;
815 		lims->imsl_st[0] = lims->imsl_st[1];
816 	}
817 	imf->imf_st[0] = imf->imf_st[1];
818 }
819 
820 /*
821  * Reap unreferenced sources from socket-layer filter set.
822  */
823 static void
824 imf_reap(struct in_mfilter *imf)
825 {
826 	struct ip_msource	*ims, *tims;
827 	struct in_msource	*lims;
828 
829 	RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
830 		lims = (struct in_msource *)ims;
831 		if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
832 		    (lims->imsl_st[1] == MCAST_UNDEFINED)) {
833 			CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
834 			RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
835 			free(ims, M_INMFILTER);
836 			imf->imf_nsrc--;
837 		}
838 	}
839 }
840 
841 /*
842  * Purge socket-layer filter set.
843  */
844 static void
845 imf_purge(struct in_mfilter *imf)
846 {
847 	struct ip_msource	*ims, *tims;
848 
849 	RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
850 		CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
851 		RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
852 		free(ims, M_INMFILTER);
853 		imf->imf_nsrc--;
854 	}
855 	imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
856 	KASSERT(RB_EMPTY(&imf->imf_sources),
857 	    ("%s: imf_sources not empty", __func__));
858 }
859 
860 /*
861  * Look up a source filter entry for a multicast group.
862  *
863  * inm is the group descriptor to work with.
864  * haddr is the host-byte-order IPv4 address to look up.
865  * noalloc may be non-zero to suppress allocation of sources.
866  * *pims will be set to the address of the retrieved or allocated source.
867  *
868  * SMPng: NOTE: may be called with locks held.
869  * Return 0 if successful, otherwise return a non-zero error code.
870  */
871 static int
872 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
873     const int noalloc, struct ip_msource **pims)
874 {
875 	struct ip_msource	 find;
876 	struct ip_msource	*ims, *nims;
877 #ifdef KTR
878 	struct in_addr ia;
879 #endif
880 
881 	find.ims_haddr = haddr;
882 	ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
883 	if (ims == NULL && !noalloc) {
884 		if (inm->inm_nsrc == in_mcast_maxgrpsrc)
885 			return (ENOSPC);
886 		nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
887 		    M_NOWAIT | M_ZERO);
888 		if (nims == NULL)
889 			return (ENOMEM);
890 		nims->ims_haddr = haddr;
891 		RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
892 		++inm->inm_nsrc;
893 		ims = nims;
894 #ifdef KTR
895 		ia.s_addr = htonl(haddr);
896 		CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__,
897 		    inet_ntoa(ia), ims);
898 #endif
899 	}
900 
901 	*pims = ims;
902 	return (0);
903 }
904 
905 /*
906  * Merge socket-layer source into IGMP-layer source.
907  * If rollback is non-zero, perform the inverse of the merge.
908  */
909 static void
910 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
911     const int rollback)
912 {
913 	int n = rollback ? -1 : 1;
914 #ifdef KTR
915 	struct in_addr ia;
916 
917 	ia.s_addr = htonl(ims->ims_haddr);
918 #endif
919 
920 	if (lims->imsl_st[0] == MCAST_EXCLUDE) {
921 		CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s",
922 		    __func__, n, inet_ntoa(ia));
923 		ims->ims_st[1].ex -= n;
924 	} else if (lims->imsl_st[0] == MCAST_INCLUDE) {
925 		CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s",
926 		    __func__, n, inet_ntoa(ia));
927 		ims->ims_st[1].in -= n;
928 	}
929 
930 	if (lims->imsl_st[1] == MCAST_EXCLUDE) {
931 		CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s",
932 		    __func__, n, inet_ntoa(ia));
933 		ims->ims_st[1].ex += n;
934 	} else if (lims->imsl_st[1] == MCAST_INCLUDE) {
935 		CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s",
936 		    __func__, n, inet_ntoa(ia));
937 		ims->ims_st[1].in += n;
938 	}
939 }
940 
941 /*
942  * Atomically update the global in_multi state, when a membership's
943  * filter list is being updated in any way.
944  *
945  * imf is the per-inpcb-membership group filter pointer.
946  * A fake imf may be passed for in-kernel consumers.
947  *
948  * XXX This is a candidate for a set-symmetric-difference style loop
949  * which would eliminate the repeated lookup from root of ims nodes,
950  * as they share the same key space.
951  *
952  * If any error occurred this function will back out of refcounts
953  * and return a non-zero value.
954  */
955 static int
956 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
957 {
958 	struct ip_msource	*ims, *nims;
959 	struct in_msource	*lims;
960 	int			 schanged, error;
961 	int			 nsrc0, nsrc1;
962 
963 	schanged = 0;
964 	error = 0;
965 	nsrc1 = nsrc0 = 0;
966 
967 	/*
968 	 * Update the source filters first, as this may fail.
969 	 * Maintain count of in-mode filters at t0, t1. These are
970 	 * used to work out if we transition into ASM mode or not.
971 	 * Maintain a count of source filters whose state was
972 	 * actually modified by this operation.
973 	 */
974 	RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
975 		lims = (struct in_msource *)ims;
976 		if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
977 		if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
978 		if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
979 		error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
980 		++schanged;
981 		if (error)
982 			break;
983 		ims_merge(nims, lims, 0);
984 	}
985 	if (error) {
986 		struct ip_msource *bims;
987 
988 		RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
989 			lims = (struct in_msource *)ims;
990 			if (lims->imsl_st[0] == lims->imsl_st[1])
991 				continue;
992 			(void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
993 			if (bims == NULL)
994 				continue;
995 			ims_merge(bims, lims, 1);
996 		}
997 		goto out_reap;
998 	}
999 
1000 	CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1001 	    __func__, nsrc0, nsrc1);
1002 
1003 	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1004 	if (imf->imf_st[0] == imf->imf_st[1] &&
1005 	    imf->imf_st[1] == MCAST_INCLUDE) {
1006 		if (nsrc1 == 0) {
1007 			CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1008 			--inm->inm_st[1].iss_in;
1009 		}
1010 	}
1011 
1012 	/* Handle filter mode transition on socket. */
1013 	if (imf->imf_st[0] != imf->imf_st[1]) {
1014 		CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1015 		    __func__, imf->imf_st[0], imf->imf_st[1]);
1016 
1017 		if (imf->imf_st[0] == MCAST_EXCLUDE) {
1018 			CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1019 			--inm->inm_st[1].iss_ex;
1020 		} else if (imf->imf_st[0] == MCAST_INCLUDE) {
1021 			CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1022 			--inm->inm_st[1].iss_in;
1023 		}
1024 
1025 		if (imf->imf_st[1] == MCAST_EXCLUDE) {
1026 			CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1027 			inm->inm_st[1].iss_ex++;
1028 		} else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1029 			CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1030 			inm->inm_st[1].iss_in++;
1031 		}
1032 	}
1033 
1034 	/*
1035 	 * Track inm filter state in terms of listener counts.
1036 	 * If there are any exclusive listeners, stack-wide
1037 	 * membership is exclusive.
1038 	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1039 	 * If no listeners remain, state is undefined at t1,
1040 	 * and the IGMP lifecycle for this group should finish.
1041 	 */
1042 	if (inm->inm_st[1].iss_ex > 0) {
1043 		CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1044 		inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1045 	} else if (inm->inm_st[1].iss_in > 0) {
1046 		CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1047 		inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1048 	} else {
1049 		CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1050 		inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1051 	}
1052 
1053 	/* Decrement ASM listener count on transition out of ASM mode. */
1054 	if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1055 		if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1056 		    (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1057 			CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1058 			--inm->inm_st[1].iss_asm;
1059 	}
1060 
1061 	/* Increment ASM listener count on transition to ASM mode. */
1062 	if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1063 		CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1064 		inm->inm_st[1].iss_asm++;
1065 	}
1066 
1067 	CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1068 	inm_print(inm);
1069 
1070 out_reap:
1071 	if (schanged > 0) {
1072 		CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1073 		inm_reap(inm);
1074 	}
1075 	return (error);
1076 }
1077 
1078 /*
1079  * Mark an in_multi's filter set deltas as committed.
1080  * Called by IGMP after a state change has been enqueued.
1081  */
1082 void
1083 inm_commit(struct in_multi *inm)
1084 {
1085 	struct ip_msource	*ims;
1086 
1087 	CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1088 	CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1089 	inm_print(inm);
1090 
1091 	RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1092 		ims->ims_st[0] = ims->ims_st[1];
1093 	}
1094 	inm->inm_st[0] = inm->inm_st[1];
1095 }
1096 
1097 /*
1098  * Reap unreferenced nodes from an in_multi's filter set.
1099  */
1100 static void
1101 inm_reap(struct in_multi *inm)
1102 {
1103 	struct ip_msource	*ims, *tims;
1104 
1105 	RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1106 		if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1107 		    ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1108 		    ims->ims_stp != 0)
1109 			continue;
1110 		CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1111 		RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1112 		free(ims, M_IPMSOURCE);
1113 		inm->inm_nsrc--;
1114 	}
1115 }
1116 
1117 /*
1118  * Purge all source nodes from an in_multi's filter set.
1119  */
1120 static void
1121 inm_purge(struct in_multi *inm)
1122 {
1123 	struct ip_msource	*ims, *tims;
1124 
1125 	RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1126 		CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1127 		RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1128 		free(ims, M_IPMSOURCE);
1129 		inm->inm_nsrc--;
1130 	}
1131 }
1132 
1133 /*
1134  * Join a multicast group; unlocked entry point.
1135  *
1136  * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1137  * is not held. Fortunately, ifp is unlikely to have been detached
1138  * at this point, so we assume it's OK to recurse.
1139  */
1140 int
1141 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1142     /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1143 {
1144 	int error;
1145 
1146 	IN_MULTI_LOCK();
1147 	error = in_joingroup_locked(ifp, gina, imf, pinm);
1148 	IN_MULTI_UNLOCK();
1149 
1150 	return (error);
1151 }
1152 
1153 /*
1154  * Join a multicast group; real entry point.
1155  *
1156  * Only preserves atomicity at inm level.
1157  * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1158  *
1159  * If the IGMP downcall fails, the group is not joined, and an error
1160  * code is returned.
1161  */
1162 int
1163 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1164     /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1165 {
1166 	struct in_mfilter	 timf;
1167 	struct in_multi		*inm;
1168 	int			 error;
1169 
1170 	IN_MULTI_LOCK_ASSERT();
1171 
1172 	CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__,
1173 	    inet_ntoa(*gina), ifp, ifp->if_xname);
1174 
1175 	error = 0;
1176 	inm = NULL;
1177 
1178 	/*
1179 	 * If no imf was specified (i.e. kernel consumer),
1180 	 * fake one up and assume it is an ASM join.
1181 	 */
1182 	if (imf == NULL) {
1183 		imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1184 		imf = &timf;
1185 	}
1186 
1187 	error = in_getmulti(ifp, gina, &inm);
1188 	if (error) {
1189 		CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1190 		return (error);
1191 	}
1192 
1193 	CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1194 	error = inm_merge(inm, imf);
1195 	if (error) {
1196 		CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1197 		goto out_inm_release;
1198 	}
1199 
1200 	CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1201 	error = igmp_change_state(inm);
1202 	if (error) {
1203 		CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1204 		goto out_inm_release;
1205 	}
1206 
1207 out_inm_release:
1208 	if (error) {
1209 		CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1210 		inm_release_locked(inm);
1211 	} else {
1212 		*pinm = inm;
1213 	}
1214 
1215 	return (error);
1216 }
1217 
1218 /*
1219  * Leave a multicast group; unlocked entry point.
1220  */
1221 int
1222 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1223 {
1224 	int error;
1225 
1226 	IN_MULTI_LOCK();
1227 	error = in_leavegroup_locked(inm, imf);
1228 	IN_MULTI_UNLOCK();
1229 
1230 	return (error);
1231 }
1232 
1233 /*
1234  * Leave a multicast group; real entry point.
1235  * All source filters will be expunged.
1236  *
1237  * Only preserves atomicity at inm level.
1238  *
1239  * Holding the write lock for the INP which contains imf
1240  * is highly advisable. We can't assert for it as imf does not
1241  * contain a back-pointer to the owning inp.
1242  *
1243  * Note: This is not the same as inm_release(*) as this function also
1244  * makes a state change downcall into IGMP.
1245  */
1246 int
1247 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1248 {
1249 	struct in_mfilter	 timf;
1250 	int			 error;
1251 
1252 	error = 0;
1253 
1254 	IN_MULTI_LOCK_ASSERT();
1255 
1256 	CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__,
1257 	    inm, inet_ntoa(inm->inm_addr),
1258 	    (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1259 	    imf);
1260 
1261 	/*
1262 	 * If no imf was specified (i.e. kernel consumer),
1263 	 * fake one up and assume it is an ASM join.
1264 	 */
1265 	if (imf == NULL) {
1266 		imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1267 		imf = &timf;
1268 	}
1269 
1270 	/*
1271 	 * Begin state merge transaction at IGMP layer.
1272 	 *
1273 	 * As this particular invocation should not cause any memory
1274 	 * to be allocated, and there is no opportunity to roll back
1275 	 * the transaction, it MUST NOT fail.
1276 	 */
1277 	CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1278 	error = inm_merge(inm, imf);
1279 	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1280 
1281 	CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1282 	CURVNET_SET(inm->inm_ifp->if_vnet);
1283 	error = igmp_change_state(inm);
1284 	CURVNET_RESTORE();
1285 	if (error)
1286 		CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1287 
1288 	CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1289 	inm_release_locked(inm);
1290 
1291 	return (error);
1292 }
1293 
1294 /*#ifndef BURN_BRIDGES*/
1295 /*
1296  * Join an IPv4 multicast group in (*,G) exclusive mode.
1297  * The group must be a 224.0.0.0/24 link-scope group.
1298  * This KPI is for legacy kernel consumers only.
1299  */
1300 struct in_multi *
1301 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1302 {
1303 	struct in_multi *pinm;
1304 	int error;
1305 
1306 	KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1307 	    ("%s: %s not in 224.0.0.0/24", __func__, inet_ntoa(*ap)));
1308 
1309 	error = in_joingroup(ifp, ap, NULL, &pinm);
1310 	if (error != 0)
1311 		pinm = NULL;
1312 
1313 	return (pinm);
1314 }
1315 
1316 /*
1317  * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1318  * This KPI is for legacy kernel consumers only.
1319  */
1320 void
1321 in_delmulti(struct in_multi *inm)
1322 {
1323 
1324 	(void)in_leavegroup(inm, NULL);
1325 }
1326 /*#endif*/
1327 
1328 /*
1329  * Block or unblock an ASM multicast source on an inpcb.
1330  * This implements the delta-based API described in RFC 3678.
1331  *
1332  * The delta-based API applies only to exclusive-mode memberships.
1333  * An IGMP downcall will be performed.
1334  *
1335  * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1336  *
1337  * Return 0 if successful, otherwise return an appropriate error code.
1338  */
1339 static int
1340 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1341 {
1342 	struct group_source_req		 gsr;
1343 	sockunion_t			*gsa, *ssa;
1344 	struct ifnet			*ifp;
1345 	struct in_mfilter		*imf;
1346 	struct ip_moptions		*imo;
1347 	struct in_msource		*ims;
1348 	struct in_multi			*inm;
1349 	size_t				 idx;
1350 	uint16_t			 fmode;
1351 	int				 error, doblock;
1352 
1353 	ifp = NULL;
1354 	error = 0;
1355 	doblock = 0;
1356 
1357 	memset(&gsr, 0, sizeof(struct group_source_req));
1358 	gsa = (sockunion_t *)&gsr.gsr_group;
1359 	ssa = (sockunion_t *)&gsr.gsr_source;
1360 
1361 	switch (sopt->sopt_name) {
1362 	case IP_BLOCK_SOURCE:
1363 	case IP_UNBLOCK_SOURCE: {
1364 		struct ip_mreq_source	 mreqs;
1365 
1366 		error = sooptcopyin(sopt, &mreqs,
1367 		    sizeof(struct ip_mreq_source),
1368 		    sizeof(struct ip_mreq_source));
1369 		if (error)
1370 			return (error);
1371 
1372 		gsa->sin.sin_family = AF_INET;
1373 		gsa->sin.sin_len = sizeof(struct sockaddr_in);
1374 		gsa->sin.sin_addr = mreqs.imr_multiaddr;
1375 
1376 		ssa->sin.sin_family = AF_INET;
1377 		ssa->sin.sin_len = sizeof(struct sockaddr_in);
1378 		ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1379 
1380 		if (!in_nullhost(mreqs.imr_interface))
1381 			INADDR_TO_IFP(mreqs.imr_interface, ifp);
1382 
1383 		if (sopt->sopt_name == IP_BLOCK_SOURCE)
1384 			doblock = 1;
1385 
1386 		CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1387 		    __func__, inet_ntoa(mreqs.imr_interface), ifp);
1388 		break;
1389 	    }
1390 
1391 	case MCAST_BLOCK_SOURCE:
1392 	case MCAST_UNBLOCK_SOURCE:
1393 		error = sooptcopyin(sopt, &gsr,
1394 		    sizeof(struct group_source_req),
1395 		    sizeof(struct group_source_req));
1396 		if (error)
1397 			return (error);
1398 
1399 		if (gsa->sin.sin_family != AF_INET ||
1400 		    gsa->sin.sin_len != sizeof(struct sockaddr_in))
1401 			return (EINVAL);
1402 
1403 		if (ssa->sin.sin_family != AF_INET ||
1404 		    ssa->sin.sin_len != sizeof(struct sockaddr_in))
1405 			return (EINVAL);
1406 
1407 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1408 			return (EADDRNOTAVAIL);
1409 
1410 		ifp = ifnet_byindex(gsr.gsr_interface);
1411 
1412 		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1413 			doblock = 1;
1414 		break;
1415 
1416 	default:
1417 		CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1418 		    __func__, sopt->sopt_name);
1419 		return (EOPNOTSUPP);
1420 		break;
1421 	}
1422 
1423 	if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1424 		return (EINVAL);
1425 
1426 	/*
1427 	 * Check if we are actually a member of this group.
1428 	 */
1429 	imo = inp_findmoptions(inp);
1430 	idx = imo_match_group(imo, ifp, &gsa->sa);
1431 	if (idx == -1 || imo->imo_mfilters == NULL) {
1432 		error = EADDRNOTAVAIL;
1433 		goto out_inp_locked;
1434 	}
1435 
1436 	KASSERT(imo->imo_mfilters != NULL,
1437 	    ("%s: imo_mfilters not allocated", __func__));
1438 	imf = &imo->imo_mfilters[idx];
1439 	inm = imo->imo_membership[idx];
1440 
1441 	/*
1442 	 * Attempting to use the delta-based API on an
1443 	 * non exclusive-mode membership is an error.
1444 	 */
1445 	fmode = imf->imf_st[0];
1446 	if (fmode != MCAST_EXCLUDE) {
1447 		error = EINVAL;
1448 		goto out_inp_locked;
1449 	}
1450 
1451 	/*
1452 	 * Deal with error cases up-front:
1453 	 *  Asked to block, but already blocked; or
1454 	 *  Asked to unblock, but nothing to unblock.
1455 	 * If adding a new block entry, allocate it.
1456 	 */
1457 	ims = imo_match_source(imo, idx, &ssa->sa);
1458 	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1459 		CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
1460 		    inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not ");
1461 		error = EADDRNOTAVAIL;
1462 		goto out_inp_locked;
1463 	}
1464 
1465 	INP_WLOCK_ASSERT(inp);
1466 
1467 	/*
1468 	 * Begin state merge transaction at socket layer.
1469 	 */
1470 	if (doblock) {
1471 		CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1472 		ims = imf_graft(imf, fmode, &ssa->sin);
1473 		if (ims == NULL)
1474 			error = ENOMEM;
1475 	} else {
1476 		CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1477 		error = imf_prune(imf, &ssa->sin);
1478 	}
1479 
1480 	if (error) {
1481 		CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1482 		goto out_imf_rollback;
1483 	}
1484 
1485 	/*
1486 	 * Begin state merge transaction at IGMP layer.
1487 	 */
1488 	IN_MULTI_LOCK();
1489 
1490 	CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1491 	error = inm_merge(inm, imf);
1492 	if (error) {
1493 		CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1494 		goto out_in_multi_locked;
1495 	}
1496 
1497 	CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1498 	error = igmp_change_state(inm);
1499 	if (error)
1500 		CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1501 
1502 out_in_multi_locked:
1503 
1504 	IN_MULTI_UNLOCK();
1505 
1506 out_imf_rollback:
1507 	if (error)
1508 		imf_rollback(imf);
1509 	else
1510 		imf_commit(imf);
1511 
1512 	imf_reap(imf);
1513 
1514 out_inp_locked:
1515 	INP_WUNLOCK(inp);
1516 	return (error);
1517 }
1518 
1519 /*
1520  * Given an inpcb, return its multicast options structure pointer.  Accepts
1521  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1522  *
1523  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1524  * SMPng: NOTE: Returns with the INP write lock held.
1525  */
1526 static struct ip_moptions *
1527 inp_findmoptions(struct inpcb *inp)
1528 {
1529 	struct ip_moptions	 *imo;
1530 	struct in_multi		**immp;
1531 	struct in_mfilter	 *imfp;
1532 	size_t			  idx;
1533 
1534 	INP_WLOCK(inp);
1535 	if (inp->inp_moptions != NULL)
1536 		return (inp->inp_moptions);
1537 
1538 	INP_WUNLOCK(inp);
1539 
1540 	imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1541 	immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1542 	    M_WAITOK | M_ZERO);
1543 	imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1544 	    M_INMFILTER, M_WAITOK);
1545 
1546 	imo->imo_multicast_ifp = NULL;
1547 	imo->imo_multicast_addr.s_addr = INADDR_ANY;
1548 	imo->imo_multicast_vif = -1;
1549 	imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1550 	imo->imo_multicast_loop = in_mcast_loop;
1551 	imo->imo_num_memberships = 0;
1552 	imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1553 	imo->imo_membership = immp;
1554 
1555 	/* Initialize per-group source filters. */
1556 	for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1557 		imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1558 	imo->imo_mfilters = imfp;
1559 
1560 	INP_WLOCK(inp);
1561 	if (inp->inp_moptions != NULL) {
1562 		free(imfp, M_INMFILTER);
1563 		free(immp, M_IPMOPTS);
1564 		free(imo, M_IPMOPTS);
1565 		return (inp->inp_moptions);
1566 	}
1567 	inp->inp_moptions = imo;
1568 	return (imo);
1569 }
1570 
1571 /*
1572  * Discard the IP multicast options (and source filters).  To minimize
1573  * the amount of work done while holding locks such as the INP's
1574  * pcbinfo lock (which is used in the receive path), the free
1575  * operation is performed asynchronously in a separate task.
1576  *
1577  * SMPng: NOTE: assumes INP write lock is held.
1578  */
1579 void
1580 inp_freemoptions(struct ip_moptions *imo)
1581 {
1582 
1583 	KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1584 	IN_MULTI_LOCK();
1585 	STAILQ_INSERT_TAIL(&imo_gc_list, imo, imo_link);
1586 	IN_MULTI_UNLOCK();
1587 	taskqueue_enqueue(taskqueue_thread, &imo_gc_task);
1588 }
1589 
1590 static void
1591 inp_freemoptions_internal(struct ip_moptions *imo)
1592 {
1593 	struct in_mfilter	*imf;
1594 	size_t			 idx, nmships;
1595 
1596 	nmships = imo->imo_num_memberships;
1597 	for (idx = 0; idx < nmships; ++idx) {
1598 		imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1599 		if (imf)
1600 			imf_leave(imf);
1601 		(void)in_leavegroup(imo->imo_membership[idx], imf);
1602 		if (imf)
1603 			imf_purge(imf);
1604 	}
1605 
1606 	if (imo->imo_mfilters)
1607 		free(imo->imo_mfilters, M_INMFILTER);
1608 	free(imo->imo_membership, M_IPMOPTS);
1609 	free(imo, M_IPMOPTS);
1610 }
1611 
1612 static void
1613 inp_gcmoptions(void *context, int pending)
1614 {
1615 	struct ip_moptions *imo;
1616 
1617 	IN_MULTI_LOCK();
1618 	while (!STAILQ_EMPTY(&imo_gc_list)) {
1619 		imo = STAILQ_FIRST(&imo_gc_list);
1620 		STAILQ_REMOVE_HEAD(&imo_gc_list, imo_link);
1621 		IN_MULTI_UNLOCK();
1622 		inp_freemoptions_internal(imo);
1623 		IN_MULTI_LOCK();
1624 	}
1625 	IN_MULTI_UNLOCK();
1626 }
1627 
1628 /*
1629  * Atomically get source filters on a socket for an IPv4 multicast group.
1630  * Called with INP lock held; returns with lock released.
1631  */
1632 static int
1633 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1634 {
1635 	struct __msfilterreq	 msfr;
1636 	sockunion_t		*gsa;
1637 	struct ifnet		*ifp;
1638 	struct ip_moptions	*imo;
1639 	struct in_mfilter	*imf;
1640 	struct ip_msource	*ims;
1641 	struct in_msource	*lims;
1642 	struct sockaddr_in	*psin;
1643 	struct sockaddr_storage	*ptss;
1644 	struct sockaddr_storage	*tss;
1645 	int			 error;
1646 	size_t			 idx, nsrcs, ncsrcs;
1647 
1648 	INP_WLOCK_ASSERT(inp);
1649 
1650 	imo = inp->inp_moptions;
1651 	KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1652 
1653 	INP_WUNLOCK(inp);
1654 
1655 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1656 	    sizeof(struct __msfilterreq));
1657 	if (error)
1658 		return (error);
1659 
1660 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1661 		return (EINVAL);
1662 
1663 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1664 	if (ifp == NULL)
1665 		return (EINVAL);
1666 
1667 	INP_WLOCK(inp);
1668 
1669 	/*
1670 	 * Lookup group on the socket.
1671 	 */
1672 	gsa = (sockunion_t *)&msfr.msfr_group;
1673 	idx = imo_match_group(imo, ifp, &gsa->sa);
1674 	if (idx == -1 || imo->imo_mfilters == NULL) {
1675 		INP_WUNLOCK(inp);
1676 		return (EADDRNOTAVAIL);
1677 	}
1678 	imf = &imo->imo_mfilters[idx];
1679 
1680 	/*
1681 	 * Ignore memberships which are in limbo.
1682 	 */
1683 	if (imf->imf_st[1] == MCAST_UNDEFINED) {
1684 		INP_WUNLOCK(inp);
1685 		return (EAGAIN);
1686 	}
1687 	msfr.msfr_fmode = imf->imf_st[1];
1688 
1689 	/*
1690 	 * If the user specified a buffer, copy out the source filter
1691 	 * entries to userland gracefully.
1692 	 * We only copy out the number of entries which userland
1693 	 * has asked for, but we always tell userland how big the
1694 	 * buffer really needs to be.
1695 	 */
1696 	if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1697 		msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1698 	tss = NULL;
1699 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1700 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1701 		    M_TEMP, M_NOWAIT | M_ZERO);
1702 		if (tss == NULL) {
1703 			INP_WUNLOCK(inp);
1704 			return (ENOBUFS);
1705 		}
1706 	}
1707 
1708 	/*
1709 	 * Count number of sources in-mode at t0.
1710 	 * If buffer space exists and remains, copy out source entries.
1711 	 */
1712 	nsrcs = msfr.msfr_nsrcs;
1713 	ncsrcs = 0;
1714 	ptss = tss;
1715 	RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1716 		lims = (struct in_msource *)ims;
1717 		if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1718 		    lims->imsl_st[0] != imf->imf_st[0])
1719 			continue;
1720 		++ncsrcs;
1721 		if (tss != NULL && nsrcs > 0) {
1722 			psin = (struct sockaddr_in *)ptss;
1723 			psin->sin_family = AF_INET;
1724 			psin->sin_len = sizeof(struct sockaddr_in);
1725 			psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1726 			psin->sin_port = 0;
1727 			++ptss;
1728 			--nsrcs;
1729 		}
1730 	}
1731 
1732 	INP_WUNLOCK(inp);
1733 
1734 	if (tss != NULL) {
1735 		error = copyout(tss, msfr.msfr_srcs,
1736 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1737 		free(tss, M_TEMP);
1738 		if (error)
1739 			return (error);
1740 	}
1741 
1742 	msfr.msfr_nsrcs = ncsrcs;
1743 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1744 
1745 	return (error);
1746 }
1747 
1748 /*
1749  * Return the IP multicast options in response to user getsockopt().
1750  */
1751 int
1752 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1753 {
1754 	struct rm_priotracker	 in_ifa_tracker;
1755 	struct ip_mreqn		 mreqn;
1756 	struct ip_moptions	*imo;
1757 	struct ifnet		*ifp;
1758 	struct in_ifaddr	*ia;
1759 	int			 error, optval;
1760 	u_char			 coptval;
1761 
1762 	INP_WLOCK(inp);
1763 	imo = inp->inp_moptions;
1764 	/*
1765 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1766 	 * or is a divert socket, reject it.
1767 	 */
1768 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1769 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1770 	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1771 		INP_WUNLOCK(inp);
1772 		return (EOPNOTSUPP);
1773 	}
1774 
1775 	error = 0;
1776 	switch (sopt->sopt_name) {
1777 	case IP_MULTICAST_VIF:
1778 		if (imo != NULL)
1779 			optval = imo->imo_multicast_vif;
1780 		else
1781 			optval = -1;
1782 		INP_WUNLOCK(inp);
1783 		error = sooptcopyout(sopt, &optval, sizeof(int));
1784 		break;
1785 
1786 	case IP_MULTICAST_IF:
1787 		memset(&mreqn, 0, sizeof(struct ip_mreqn));
1788 		if (imo != NULL) {
1789 			ifp = imo->imo_multicast_ifp;
1790 			if (!in_nullhost(imo->imo_multicast_addr)) {
1791 				mreqn.imr_address = imo->imo_multicast_addr;
1792 			} else if (ifp != NULL) {
1793 				mreqn.imr_ifindex = ifp->if_index;
1794 				IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1795 				if (ia != NULL) {
1796 					mreqn.imr_address =
1797 					    IA_SIN(ia)->sin_addr;
1798 					ifa_free(&ia->ia_ifa);
1799 				}
1800 			}
1801 		}
1802 		INP_WUNLOCK(inp);
1803 		if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1804 			error = sooptcopyout(sopt, &mreqn,
1805 			    sizeof(struct ip_mreqn));
1806 		} else {
1807 			error = sooptcopyout(sopt, &mreqn.imr_address,
1808 			    sizeof(struct in_addr));
1809 		}
1810 		break;
1811 
1812 	case IP_MULTICAST_TTL:
1813 		if (imo == NULL)
1814 			optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1815 		else
1816 			optval = coptval = imo->imo_multicast_ttl;
1817 		INP_WUNLOCK(inp);
1818 		if (sopt->sopt_valsize == sizeof(u_char))
1819 			error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1820 		else
1821 			error = sooptcopyout(sopt, &optval, sizeof(int));
1822 		break;
1823 
1824 	case IP_MULTICAST_LOOP:
1825 		if (imo == NULL)
1826 			optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1827 		else
1828 			optval = coptval = imo->imo_multicast_loop;
1829 		INP_WUNLOCK(inp);
1830 		if (sopt->sopt_valsize == sizeof(u_char))
1831 			error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1832 		else
1833 			error = sooptcopyout(sopt, &optval, sizeof(int));
1834 		break;
1835 
1836 	case IP_MSFILTER:
1837 		if (imo == NULL) {
1838 			error = EADDRNOTAVAIL;
1839 			INP_WUNLOCK(inp);
1840 		} else {
1841 			error = inp_get_source_filters(inp, sopt);
1842 		}
1843 		break;
1844 
1845 	default:
1846 		INP_WUNLOCK(inp);
1847 		error = ENOPROTOOPT;
1848 		break;
1849 	}
1850 
1851 	INP_UNLOCK_ASSERT(inp);
1852 
1853 	return (error);
1854 }
1855 
1856 /*
1857  * Look up the ifnet to use for a multicast group membership,
1858  * given the IPv4 address of an interface, and the IPv4 group address.
1859  *
1860  * This routine exists to support legacy multicast applications
1861  * which do not understand that multicast memberships are scoped to
1862  * specific physical links in the networking stack, or which need
1863  * to join link-scope groups before IPv4 addresses are configured.
1864  *
1865  * If inp is non-NULL, use this socket's current FIB number for any
1866  * required FIB lookup.
1867  * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1868  * and use its ifp; usually, this points to the default next-hop.
1869  *
1870  * If the FIB lookup fails, attempt to use the first non-loopback
1871  * interface with multicast capability in the system as a
1872  * last resort. The legacy IPv4 ASM API requires that we do
1873  * this in order to allow groups to be joined when the routing
1874  * table has not yet been populated during boot.
1875  *
1876  * Returns NULL if no ifp could be found.
1877  *
1878  * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1879  * FUTURE: Implement IPv4 source-address selection.
1880  */
1881 static struct ifnet *
1882 inp_lookup_mcast_ifp(const struct inpcb *inp,
1883     const struct sockaddr_in *gsin, const struct in_addr ina)
1884 {
1885 	struct rm_priotracker in_ifa_tracker;
1886 	struct ifnet *ifp;
1887 	struct nhop4_basic nh4;
1888 	uint32_t fibnum;
1889 
1890 	KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1891 	KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1892 	    ("%s: not multicast", __func__));
1893 
1894 	ifp = NULL;
1895 	if (!in_nullhost(ina)) {
1896 		INADDR_TO_IFP(ina, ifp);
1897 	} else {
1898 		fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1899 		if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1900 			ifp = nh4.nh_ifp;
1901 		else {
1902 			struct in_ifaddr *ia;
1903 			struct ifnet *mifp;
1904 
1905 			mifp = NULL;
1906 			IN_IFADDR_RLOCK(&in_ifa_tracker);
1907 			TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1908 				mifp = ia->ia_ifp;
1909 				if (!(mifp->if_flags & IFF_LOOPBACK) &&
1910 				     (mifp->if_flags & IFF_MULTICAST)) {
1911 					ifp = mifp;
1912 					break;
1913 				}
1914 			}
1915 			IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1916 		}
1917 	}
1918 
1919 	return (ifp);
1920 }
1921 
1922 /*
1923  * Join an IPv4 multicast group, possibly with a source.
1924  */
1925 static int
1926 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1927 {
1928 	struct group_source_req		 gsr;
1929 	sockunion_t			*gsa, *ssa;
1930 	struct ifnet			*ifp;
1931 	struct in_mfilter		*imf;
1932 	struct ip_moptions		*imo;
1933 	struct in_multi			*inm;
1934 	struct in_msource		*lims;
1935 	size_t				 idx;
1936 	int				 error, is_new;
1937 
1938 	ifp = NULL;
1939 	imf = NULL;
1940 	lims = NULL;
1941 	error = 0;
1942 	is_new = 0;
1943 
1944 	memset(&gsr, 0, sizeof(struct group_source_req));
1945 	gsa = (sockunion_t *)&gsr.gsr_group;
1946 	gsa->ss.ss_family = AF_UNSPEC;
1947 	ssa = (sockunion_t *)&gsr.gsr_source;
1948 	ssa->ss.ss_family = AF_UNSPEC;
1949 
1950 	switch (sopt->sopt_name) {
1951 	case IP_ADD_MEMBERSHIP:
1952 	case IP_ADD_SOURCE_MEMBERSHIP: {
1953 		struct ip_mreq_source	 mreqs;
1954 
1955 		if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1956 			error = sooptcopyin(sopt, &mreqs,
1957 			    sizeof(struct ip_mreq),
1958 			    sizeof(struct ip_mreq));
1959 			/*
1960 			 * Do argument switcharoo from ip_mreq into
1961 			 * ip_mreq_source to avoid using two instances.
1962 			 */
1963 			mreqs.imr_interface = mreqs.imr_sourceaddr;
1964 			mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1965 		} else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1966 			error = sooptcopyin(sopt, &mreqs,
1967 			    sizeof(struct ip_mreq_source),
1968 			    sizeof(struct ip_mreq_source));
1969 		}
1970 		if (error)
1971 			return (error);
1972 
1973 		gsa->sin.sin_family = AF_INET;
1974 		gsa->sin.sin_len = sizeof(struct sockaddr_in);
1975 		gsa->sin.sin_addr = mreqs.imr_multiaddr;
1976 
1977 		if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1978 			ssa->sin.sin_family = AF_INET;
1979 			ssa->sin.sin_len = sizeof(struct sockaddr_in);
1980 			ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1981 		}
1982 
1983 		if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1984 			return (EINVAL);
1985 
1986 		ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1987 		    mreqs.imr_interface);
1988 		CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1989 		    __func__, inet_ntoa(mreqs.imr_interface), ifp);
1990 		break;
1991 	}
1992 
1993 	case MCAST_JOIN_GROUP:
1994 	case MCAST_JOIN_SOURCE_GROUP:
1995 		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1996 			error = sooptcopyin(sopt, &gsr,
1997 			    sizeof(struct group_req),
1998 			    sizeof(struct group_req));
1999 		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2000 			error = sooptcopyin(sopt, &gsr,
2001 			    sizeof(struct group_source_req),
2002 			    sizeof(struct group_source_req));
2003 		}
2004 		if (error)
2005 			return (error);
2006 
2007 		if (gsa->sin.sin_family != AF_INET ||
2008 		    gsa->sin.sin_len != sizeof(struct sockaddr_in))
2009 			return (EINVAL);
2010 
2011 		/*
2012 		 * Overwrite the port field if present, as the sockaddr
2013 		 * being copied in may be matched with a binary comparison.
2014 		 */
2015 		gsa->sin.sin_port = 0;
2016 		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2017 			if (ssa->sin.sin_family != AF_INET ||
2018 			    ssa->sin.sin_len != sizeof(struct sockaddr_in))
2019 				return (EINVAL);
2020 			ssa->sin.sin_port = 0;
2021 		}
2022 
2023 		if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2024 			return (EINVAL);
2025 
2026 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2027 			return (EADDRNOTAVAIL);
2028 		ifp = ifnet_byindex(gsr.gsr_interface);
2029 		break;
2030 
2031 	default:
2032 		CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2033 		    __func__, sopt->sopt_name);
2034 		return (EOPNOTSUPP);
2035 		break;
2036 	}
2037 
2038 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2039 		return (EADDRNOTAVAIL);
2040 
2041 	imo = inp_findmoptions(inp);
2042 	idx = imo_match_group(imo, ifp, &gsa->sa);
2043 	if (idx == -1) {
2044 		is_new = 1;
2045 	} else {
2046 		inm = imo->imo_membership[idx];
2047 		imf = &imo->imo_mfilters[idx];
2048 		if (ssa->ss.ss_family != AF_UNSPEC) {
2049 			/*
2050 			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2051 			 * is an error. On an existing inclusive membership,
2052 			 * it just adds the source to the filter list.
2053 			 */
2054 			if (imf->imf_st[1] != MCAST_INCLUDE) {
2055 				error = EINVAL;
2056 				goto out_inp_locked;
2057 			}
2058 			/*
2059 			 * Throw out duplicates.
2060 			 *
2061 			 * XXX FIXME: This makes a naive assumption that
2062 			 * even if entries exist for *ssa in this imf,
2063 			 * they will be rejected as dupes, even if they
2064 			 * are not valid in the current mode (in-mode).
2065 			 *
2066 			 * in_msource is transactioned just as for anything
2067 			 * else in SSM -- but note naive use of inm_graft()
2068 			 * below for allocating new filter entries.
2069 			 *
2070 			 * This is only an issue if someone mixes the
2071 			 * full-state SSM API with the delta-based API,
2072 			 * which is discouraged in the relevant RFCs.
2073 			 */
2074 			lims = imo_match_source(imo, idx, &ssa->sa);
2075 			if (lims != NULL /*&&
2076 			    lims->imsl_st[1] == MCAST_INCLUDE*/) {
2077 				error = EADDRNOTAVAIL;
2078 				goto out_inp_locked;
2079 			}
2080 		} else {
2081 			/*
2082 			 * MCAST_JOIN_GROUP on an existing exclusive
2083 			 * membership is an error; return EADDRINUSE
2084 			 * to preserve 4.4BSD API idempotence, and
2085 			 * avoid tedious detour to code below.
2086 			 * NOTE: This is bending RFC 3678 a bit.
2087 			 *
2088 			 * On an existing inclusive membership, this is also
2089 			 * an error; if you want to change filter mode,
2090 			 * you must use the userland API setsourcefilter().
2091 			 * XXX We don't reject this for imf in UNDEFINED
2092 			 * state at t1, because allocation of a filter
2093 			 * is atomic with allocation of a membership.
2094 			 */
2095 			error = EINVAL;
2096 			if (imf->imf_st[1] == MCAST_EXCLUDE)
2097 				error = EADDRINUSE;
2098 			goto out_inp_locked;
2099 		}
2100 	}
2101 
2102 	/*
2103 	 * Begin state merge transaction at socket layer.
2104 	 */
2105 	INP_WLOCK_ASSERT(inp);
2106 
2107 	if (is_new) {
2108 		if (imo->imo_num_memberships == imo->imo_max_memberships) {
2109 			error = imo_grow(imo);
2110 			if (error)
2111 				goto out_inp_locked;
2112 		}
2113 		/*
2114 		 * Allocate the new slot upfront so we can deal with
2115 		 * grafting the new source filter in same code path
2116 		 * as for join-source on existing membership.
2117 		 */
2118 		idx = imo->imo_num_memberships;
2119 		imo->imo_membership[idx] = NULL;
2120 		imo->imo_num_memberships++;
2121 		KASSERT(imo->imo_mfilters != NULL,
2122 		    ("%s: imf_mfilters vector was not allocated", __func__));
2123 		imf = &imo->imo_mfilters[idx];
2124 		KASSERT(RB_EMPTY(&imf->imf_sources),
2125 		    ("%s: imf_sources not empty", __func__));
2126 	}
2127 
2128 	/*
2129 	 * Graft new source into filter list for this inpcb's
2130 	 * membership of the group. The in_multi may not have
2131 	 * been allocated yet if this is a new membership, however,
2132 	 * the in_mfilter slot will be allocated and must be initialized.
2133 	 *
2134 	 * Note: Grafting of exclusive mode filters doesn't happen
2135 	 * in this path.
2136 	 * XXX: Should check for non-NULL lims (node exists but may
2137 	 * not be in-mode) for interop with full-state API.
2138 	 */
2139 	if (ssa->ss.ss_family != AF_UNSPEC) {
2140 		/* Membership starts in IN mode */
2141 		if (is_new) {
2142 			CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2143 			imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2144 		} else {
2145 			CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2146 		}
2147 		lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2148 		if (lims == NULL) {
2149 			CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2150 			    __func__);
2151 			error = ENOMEM;
2152 			goto out_imo_free;
2153 		}
2154 	} else {
2155 		/* No address specified; Membership starts in EX mode */
2156 		if (is_new) {
2157 			CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2158 			imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2159 		}
2160 	}
2161 
2162 	/*
2163 	 * Begin state merge transaction at IGMP layer.
2164 	 */
2165 	IN_MULTI_LOCK();
2166 
2167 	if (is_new) {
2168 		error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2169 		    &inm);
2170 		if (error) {
2171                         CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2172                             __func__);
2173                         IN_MULTI_UNLOCK();
2174 			goto out_imo_free;
2175                 }
2176 		imo->imo_membership[idx] = inm;
2177 	} else {
2178 		CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2179 		error = inm_merge(inm, imf);
2180 		if (error) {
2181 			CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2182 			    __func__);
2183 			goto out_in_multi_locked;
2184 		}
2185 		CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2186 		error = igmp_change_state(inm);
2187 		if (error) {
2188 			CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2189 			    __func__);
2190 			goto out_in_multi_locked;
2191 		}
2192 	}
2193 
2194 out_in_multi_locked:
2195 
2196 	IN_MULTI_UNLOCK();
2197 
2198 	INP_WLOCK_ASSERT(inp);
2199 	if (error) {
2200 		imf_rollback(imf);
2201 		if (is_new)
2202 			imf_purge(imf);
2203 		else
2204 			imf_reap(imf);
2205 	} else {
2206 		imf_commit(imf);
2207 	}
2208 
2209 out_imo_free:
2210 	if (error && is_new) {
2211 		imo->imo_membership[idx] = NULL;
2212 		--imo->imo_num_memberships;
2213 	}
2214 
2215 out_inp_locked:
2216 	INP_WUNLOCK(inp);
2217 	return (error);
2218 }
2219 
2220 /*
2221  * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2222  */
2223 static int
2224 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2225 {
2226 	struct group_source_req		 gsr;
2227 	struct ip_mreq_source		 mreqs;
2228 	sockunion_t			*gsa, *ssa;
2229 	struct ifnet			*ifp;
2230 	struct in_mfilter		*imf;
2231 	struct ip_moptions		*imo;
2232 	struct in_msource		*ims;
2233 	struct in_multi			*inm;
2234 	size_t				 idx;
2235 	int				 error, is_final;
2236 
2237 	ifp = NULL;
2238 	error = 0;
2239 	is_final = 1;
2240 
2241 	memset(&gsr, 0, sizeof(struct group_source_req));
2242 	gsa = (sockunion_t *)&gsr.gsr_group;
2243 	gsa->ss.ss_family = AF_UNSPEC;
2244 	ssa = (sockunion_t *)&gsr.gsr_source;
2245 	ssa->ss.ss_family = AF_UNSPEC;
2246 
2247 	switch (sopt->sopt_name) {
2248 	case IP_DROP_MEMBERSHIP:
2249 	case IP_DROP_SOURCE_MEMBERSHIP:
2250 		if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2251 			error = sooptcopyin(sopt, &mreqs,
2252 			    sizeof(struct ip_mreq),
2253 			    sizeof(struct ip_mreq));
2254 			/*
2255 			 * Swap interface and sourceaddr arguments,
2256 			 * as ip_mreq and ip_mreq_source are laid
2257 			 * out differently.
2258 			 */
2259 			mreqs.imr_interface = mreqs.imr_sourceaddr;
2260 			mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2261 		} else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2262 			error = sooptcopyin(sopt, &mreqs,
2263 			    sizeof(struct ip_mreq_source),
2264 			    sizeof(struct ip_mreq_source));
2265 		}
2266 		if (error)
2267 			return (error);
2268 
2269 		gsa->sin.sin_family = AF_INET;
2270 		gsa->sin.sin_len = sizeof(struct sockaddr_in);
2271 		gsa->sin.sin_addr = mreqs.imr_multiaddr;
2272 
2273 		if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2274 			ssa->sin.sin_family = AF_INET;
2275 			ssa->sin.sin_len = sizeof(struct sockaddr_in);
2276 			ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2277 		}
2278 
2279 		/*
2280 		 * Attempt to look up hinted ifp from interface address.
2281 		 * Fallthrough with null ifp iff lookup fails, to
2282 		 * preserve 4.4BSD mcast API idempotence.
2283 		 * XXX NOTE WELL: The RFC 3678 API is preferred because
2284 		 * using an IPv4 address as a key is racy.
2285 		 */
2286 		if (!in_nullhost(mreqs.imr_interface))
2287 			INADDR_TO_IFP(mreqs.imr_interface, ifp);
2288 
2289 		CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
2290 		    __func__, inet_ntoa(mreqs.imr_interface), ifp);
2291 
2292 		break;
2293 
2294 	case MCAST_LEAVE_GROUP:
2295 	case MCAST_LEAVE_SOURCE_GROUP:
2296 		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2297 			error = sooptcopyin(sopt, &gsr,
2298 			    sizeof(struct group_req),
2299 			    sizeof(struct group_req));
2300 		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2301 			error = sooptcopyin(sopt, &gsr,
2302 			    sizeof(struct group_source_req),
2303 			    sizeof(struct group_source_req));
2304 		}
2305 		if (error)
2306 			return (error);
2307 
2308 		if (gsa->sin.sin_family != AF_INET ||
2309 		    gsa->sin.sin_len != sizeof(struct sockaddr_in))
2310 			return (EINVAL);
2311 
2312 		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2313 			if (ssa->sin.sin_family != AF_INET ||
2314 			    ssa->sin.sin_len != sizeof(struct sockaddr_in))
2315 				return (EINVAL);
2316 		}
2317 
2318 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2319 			return (EADDRNOTAVAIL);
2320 
2321 		ifp = ifnet_byindex(gsr.gsr_interface);
2322 
2323 		if (ifp == NULL)
2324 			return (EADDRNOTAVAIL);
2325 		break;
2326 
2327 	default:
2328 		CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2329 		    __func__, sopt->sopt_name);
2330 		return (EOPNOTSUPP);
2331 		break;
2332 	}
2333 
2334 	if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2335 		return (EINVAL);
2336 
2337 	/*
2338 	 * Find the membership in the membership array.
2339 	 */
2340 	imo = inp_findmoptions(inp);
2341 	idx = imo_match_group(imo, ifp, &gsa->sa);
2342 	if (idx == -1) {
2343 		error = EADDRNOTAVAIL;
2344 		goto out_inp_locked;
2345 	}
2346 	inm = imo->imo_membership[idx];
2347 	imf = &imo->imo_mfilters[idx];
2348 
2349 	if (ssa->ss.ss_family != AF_UNSPEC)
2350 		is_final = 0;
2351 
2352 	/*
2353 	 * Begin state merge transaction at socket layer.
2354 	 */
2355 	INP_WLOCK_ASSERT(inp);
2356 
2357 	/*
2358 	 * If we were instructed only to leave a given source, do so.
2359 	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2360 	 */
2361 	if (is_final) {
2362 		imf_leave(imf);
2363 	} else {
2364 		if (imf->imf_st[0] == MCAST_EXCLUDE) {
2365 			error = EADDRNOTAVAIL;
2366 			goto out_inp_locked;
2367 		}
2368 		ims = imo_match_source(imo, idx, &ssa->sa);
2369 		if (ims == NULL) {
2370 			CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
2371 			    inet_ntoa(ssa->sin.sin_addr), "not ");
2372 			error = EADDRNOTAVAIL;
2373 			goto out_inp_locked;
2374 		}
2375 		CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2376 		error = imf_prune(imf, &ssa->sin);
2377 		if (error) {
2378 			CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2379 			    __func__);
2380 			goto out_inp_locked;
2381 		}
2382 	}
2383 
2384 	/*
2385 	 * Begin state merge transaction at IGMP layer.
2386 	 */
2387 	IN_MULTI_LOCK();
2388 
2389 	if (is_final) {
2390 		/*
2391 		 * Give up the multicast address record to which
2392 		 * the membership points.
2393 		 */
2394 		(void)in_leavegroup_locked(inm, imf);
2395 	} else {
2396 		CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2397 		error = inm_merge(inm, imf);
2398 		if (error) {
2399 			CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2400 			    __func__);
2401 			goto out_in_multi_locked;
2402 		}
2403 
2404 		CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2405 		error = igmp_change_state(inm);
2406 		if (error) {
2407 			CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2408 			    __func__);
2409 		}
2410 	}
2411 
2412 out_in_multi_locked:
2413 
2414 	IN_MULTI_UNLOCK();
2415 
2416 	if (error)
2417 		imf_rollback(imf);
2418 	else
2419 		imf_commit(imf);
2420 
2421 	imf_reap(imf);
2422 
2423 	if (is_final) {
2424 		/* Remove the gap in the membership and filter array. */
2425 		for (++idx; idx < imo->imo_num_memberships; ++idx) {
2426 			imo->imo_membership[idx-1] = imo->imo_membership[idx];
2427 			imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2428 		}
2429 		imo->imo_num_memberships--;
2430 	}
2431 
2432 out_inp_locked:
2433 	INP_WUNLOCK(inp);
2434 	return (error);
2435 }
2436 
2437 /*
2438  * Select the interface for transmitting IPv4 multicast datagrams.
2439  *
2440  * Either an instance of struct in_addr or an instance of struct ip_mreqn
2441  * may be passed to this socket option. An address of INADDR_ANY or an
2442  * interface index of 0 is used to remove a previous selection.
2443  * When no interface is selected, one is chosen for every send.
2444  */
2445 static int
2446 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2447 {
2448 	struct in_addr		 addr;
2449 	struct ip_mreqn		 mreqn;
2450 	struct ifnet		*ifp;
2451 	struct ip_moptions	*imo;
2452 	int			 error;
2453 
2454 	if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2455 		/*
2456 		 * An interface index was specified using the
2457 		 * Linux-derived ip_mreqn structure.
2458 		 */
2459 		error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2460 		    sizeof(struct ip_mreqn));
2461 		if (error)
2462 			return (error);
2463 
2464 		if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2465 			return (EINVAL);
2466 
2467 		if (mreqn.imr_ifindex == 0) {
2468 			ifp = NULL;
2469 		} else {
2470 			ifp = ifnet_byindex(mreqn.imr_ifindex);
2471 			if (ifp == NULL)
2472 				return (EADDRNOTAVAIL);
2473 		}
2474 	} else {
2475 		/*
2476 		 * An interface was specified by IPv4 address.
2477 		 * This is the traditional BSD usage.
2478 		 */
2479 		error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2480 		    sizeof(struct in_addr));
2481 		if (error)
2482 			return (error);
2483 		if (in_nullhost(addr)) {
2484 			ifp = NULL;
2485 		} else {
2486 			INADDR_TO_IFP(addr, ifp);
2487 			if (ifp == NULL)
2488 				return (EADDRNOTAVAIL);
2489 		}
2490 		CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp,
2491 		    inet_ntoa(addr));
2492 	}
2493 
2494 	/* Reject interfaces which do not support multicast. */
2495 	if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2496 		return (EOPNOTSUPP);
2497 
2498 	imo = inp_findmoptions(inp);
2499 	imo->imo_multicast_ifp = ifp;
2500 	imo->imo_multicast_addr.s_addr = INADDR_ANY;
2501 	INP_WUNLOCK(inp);
2502 
2503 	return (0);
2504 }
2505 
2506 /*
2507  * Atomically set source filters on a socket for an IPv4 multicast group.
2508  *
2509  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2510  */
2511 static int
2512 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2513 {
2514 	struct __msfilterreq	 msfr;
2515 	sockunion_t		*gsa;
2516 	struct ifnet		*ifp;
2517 	struct in_mfilter	*imf;
2518 	struct ip_moptions	*imo;
2519 	struct in_multi		*inm;
2520 	size_t			 idx;
2521 	int			 error;
2522 
2523 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2524 	    sizeof(struct __msfilterreq));
2525 	if (error)
2526 		return (error);
2527 
2528 	if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2529 		return (ENOBUFS);
2530 
2531 	if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2532 	     msfr.msfr_fmode != MCAST_INCLUDE))
2533 		return (EINVAL);
2534 
2535 	if (msfr.msfr_group.ss_family != AF_INET ||
2536 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2537 		return (EINVAL);
2538 
2539 	gsa = (sockunion_t *)&msfr.msfr_group;
2540 	if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2541 		return (EINVAL);
2542 
2543 	gsa->sin.sin_port = 0;	/* ignore port */
2544 
2545 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2546 		return (EADDRNOTAVAIL);
2547 
2548 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2549 	if (ifp == NULL)
2550 		return (EADDRNOTAVAIL);
2551 
2552 	/*
2553 	 * Take the INP write lock.
2554 	 * Check if this socket is a member of this group.
2555 	 */
2556 	imo = inp_findmoptions(inp);
2557 	idx = imo_match_group(imo, ifp, &gsa->sa);
2558 	if (idx == -1 || imo->imo_mfilters == NULL) {
2559 		error = EADDRNOTAVAIL;
2560 		goto out_inp_locked;
2561 	}
2562 	inm = imo->imo_membership[idx];
2563 	imf = &imo->imo_mfilters[idx];
2564 
2565 	/*
2566 	 * Begin state merge transaction at socket layer.
2567 	 */
2568 	INP_WLOCK_ASSERT(inp);
2569 
2570 	imf->imf_st[1] = msfr.msfr_fmode;
2571 
2572 	/*
2573 	 * Apply any new source filters, if present.
2574 	 * Make a copy of the user-space source vector so
2575 	 * that we may copy them with a single copyin. This
2576 	 * allows us to deal with page faults up-front.
2577 	 */
2578 	if (msfr.msfr_nsrcs > 0) {
2579 		struct in_msource	*lims;
2580 		struct sockaddr_in	*psin;
2581 		struct sockaddr_storage	*kss, *pkss;
2582 		int			 i;
2583 
2584 		INP_WUNLOCK(inp);
2585 
2586 		CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2587 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2588 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2589 		    M_TEMP, M_WAITOK);
2590 		error = copyin(msfr.msfr_srcs, kss,
2591 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2592 		if (error) {
2593 			free(kss, M_TEMP);
2594 			return (error);
2595 		}
2596 
2597 		INP_WLOCK(inp);
2598 
2599 		/*
2600 		 * Mark all source filters as UNDEFINED at t1.
2601 		 * Restore new group filter mode, as imf_leave()
2602 		 * will set it to INCLUDE.
2603 		 */
2604 		imf_leave(imf);
2605 		imf->imf_st[1] = msfr.msfr_fmode;
2606 
2607 		/*
2608 		 * Update socket layer filters at t1, lazy-allocating
2609 		 * new entries. This saves a bunch of memory at the
2610 		 * cost of one RB_FIND() per source entry; duplicate
2611 		 * entries in the msfr_nsrcs vector are ignored.
2612 		 * If we encounter an error, rollback transaction.
2613 		 *
2614 		 * XXX This too could be replaced with a set-symmetric
2615 		 * difference like loop to avoid walking from root
2616 		 * every time, as the key space is common.
2617 		 */
2618 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2619 			psin = (struct sockaddr_in *)pkss;
2620 			if (psin->sin_family != AF_INET) {
2621 				error = EAFNOSUPPORT;
2622 				break;
2623 			}
2624 			if (psin->sin_len != sizeof(struct sockaddr_in)) {
2625 				error = EINVAL;
2626 				break;
2627 			}
2628 			error = imf_get_source(imf, psin, &lims);
2629 			if (error)
2630 				break;
2631 			lims->imsl_st[1] = imf->imf_st[1];
2632 		}
2633 		free(kss, M_TEMP);
2634 	}
2635 
2636 	if (error)
2637 		goto out_imf_rollback;
2638 
2639 	INP_WLOCK_ASSERT(inp);
2640 	IN_MULTI_LOCK();
2641 
2642 	/*
2643 	 * Begin state merge transaction at IGMP layer.
2644 	 */
2645 	CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2646 	error = inm_merge(inm, imf);
2647 	if (error) {
2648 		CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2649 		goto out_in_multi_locked;
2650 	}
2651 
2652 	CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2653 	error = igmp_change_state(inm);
2654 	if (error)
2655 		CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2656 
2657 out_in_multi_locked:
2658 
2659 	IN_MULTI_UNLOCK();
2660 
2661 out_imf_rollback:
2662 	if (error)
2663 		imf_rollback(imf);
2664 	else
2665 		imf_commit(imf);
2666 
2667 	imf_reap(imf);
2668 
2669 out_inp_locked:
2670 	INP_WUNLOCK(inp);
2671 	return (error);
2672 }
2673 
2674 /*
2675  * Set the IP multicast options in response to user setsockopt().
2676  *
2677  * Many of the socket options handled in this function duplicate the
2678  * functionality of socket options in the regular unicast API. However,
2679  * it is not possible to merge the duplicate code, because the idempotence
2680  * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2681  * the effects of these options must be treated as separate and distinct.
2682  *
2683  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2684  * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2685  * is refactored to no longer use vifs.
2686  */
2687 int
2688 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2689 {
2690 	struct ip_moptions	*imo;
2691 	int			 error;
2692 
2693 	error = 0;
2694 
2695 	/*
2696 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2697 	 * or is a divert socket, reject it.
2698 	 */
2699 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2700 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2701 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2702 		return (EOPNOTSUPP);
2703 
2704 	switch (sopt->sopt_name) {
2705 	case IP_MULTICAST_VIF: {
2706 		int vifi;
2707 		/*
2708 		 * Select a multicast VIF for transmission.
2709 		 * Only useful if multicast forwarding is active.
2710 		 */
2711 		if (legal_vif_num == NULL) {
2712 			error = EOPNOTSUPP;
2713 			break;
2714 		}
2715 		error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2716 		if (error)
2717 			break;
2718 		if (!legal_vif_num(vifi) && (vifi != -1)) {
2719 			error = EINVAL;
2720 			break;
2721 		}
2722 		imo = inp_findmoptions(inp);
2723 		imo->imo_multicast_vif = vifi;
2724 		INP_WUNLOCK(inp);
2725 		break;
2726 	}
2727 
2728 	case IP_MULTICAST_IF:
2729 		error = inp_set_multicast_if(inp, sopt);
2730 		break;
2731 
2732 	case IP_MULTICAST_TTL: {
2733 		u_char ttl;
2734 
2735 		/*
2736 		 * Set the IP time-to-live for outgoing multicast packets.
2737 		 * The original multicast API required a char argument,
2738 		 * which is inconsistent with the rest of the socket API.
2739 		 * We allow either a char or an int.
2740 		 */
2741 		if (sopt->sopt_valsize == sizeof(u_char)) {
2742 			error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2743 			    sizeof(u_char));
2744 			if (error)
2745 				break;
2746 		} else {
2747 			u_int ittl;
2748 
2749 			error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2750 			    sizeof(u_int));
2751 			if (error)
2752 				break;
2753 			if (ittl > 255) {
2754 				error = EINVAL;
2755 				break;
2756 			}
2757 			ttl = (u_char)ittl;
2758 		}
2759 		imo = inp_findmoptions(inp);
2760 		imo->imo_multicast_ttl = ttl;
2761 		INP_WUNLOCK(inp);
2762 		break;
2763 	}
2764 
2765 	case IP_MULTICAST_LOOP: {
2766 		u_char loop;
2767 
2768 		/*
2769 		 * Set the loopback flag for outgoing multicast packets.
2770 		 * Must be zero or one.  The original multicast API required a
2771 		 * char argument, which is inconsistent with the rest
2772 		 * of the socket API.  We allow either a char or an int.
2773 		 */
2774 		if (sopt->sopt_valsize == sizeof(u_char)) {
2775 			error = sooptcopyin(sopt, &loop, sizeof(u_char),
2776 			    sizeof(u_char));
2777 			if (error)
2778 				break;
2779 		} else {
2780 			u_int iloop;
2781 
2782 			error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2783 					    sizeof(u_int));
2784 			if (error)
2785 				break;
2786 			loop = (u_char)iloop;
2787 		}
2788 		imo = inp_findmoptions(inp);
2789 		imo->imo_multicast_loop = !!loop;
2790 		INP_WUNLOCK(inp);
2791 		break;
2792 	}
2793 
2794 	case IP_ADD_MEMBERSHIP:
2795 	case IP_ADD_SOURCE_MEMBERSHIP:
2796 	case MCAST_JOIN_GROUP:
2797 	case MCAST_JOIN_SOURCE_GROUP:
2798 		error = inp_join_group(inp, sopt);
2799 		break;
2800 
2801 	case IP_DROP_MEMBERSHIP:
2802 	case IP_DROP_SOURCE_MEMBERSHIP:
2803 	case MCAST_LEAVE_GROUP:
2804 	case MCAST_LEAVE_SOURCE_GROUP:
2805 		error = inp_leave_group(inp, sopt);
2806 		break;
2807 
2808 	case IP_BLOCK_SOURCE:
2809 	case IP_UNBLOCK_SOURCE:
2810 	case MCAST_BLOCK_SOURCE:
2811 	case MCAST_UNBLOCK_SOURCE:
2812 		error = inp_block_unblock_source(inp, sopt);
2813 		break;
2814 
2815 	case IP_MSFILTER:
2816 		error = inp_set_source_filters(inp, sopt);
2817 		break;
2818 
2819 	default:
2820 		error = EOPNOTSUPP;
2821 		break;
2822 	}
2823 
2824 	INP_UNLOCK_ASSERT(inp);
2825 
2826 	return (error);
2827 }
2828 
2829 /*
2830  * Expose IGMP's multicast filter mode and source list(s) to userland,
2831  * keyed by (ifindex, group).
2832  * The filter mode is written out as a uint32_t, followed by
2833  * 0..n of struct in_addr.
2834  * For use by ifmcstat(8).
2835  * SMPng: NOTE: unlocked read of ifindex space.
2836  */
2837 static int
2838 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2839 {
2840 	struct in_addr			 src, group;
2841 	struct ifnet			*ifp;
2842 	struct ifmultiaddr		*ifma;
2843 	struct in_multi			*inm;
2844 	struct ip_msource		*ims;
2845 	int				*name;
2846 	int				 retval;
2847 	u_int				 namelen;
2848 	uint32_t			 fmode, ifindex;
2849 
2850 	name = (int *)arg1;
2851 	namelen = arg2;
2852 
2853 	if (req->newptr != NULL)
2854 		return (EPERM);
2855 
2856 	if (namelen != 2)
2857 		return (EINVAL);
2858 
2859 	ifindex = name[0];
2860 	if (ifindex <= 0 || ifindex > V_if_index) {
2861 		CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2862 		    __func__, ifindex);
2863 		return (ENOENT);
2864 	}
2865 
2866 	group.s_addr = name[1];
2867 	if (!IN_MULTICAST(ntohl(group.s_addr))) {
2868 		CTR2(KTR_IGMPV3, "%s: group %s is not multicast",
2869 		    __func__, inet_ntoa(group));
2870 		return (EINVAL);
2871 	}
2872 
2873 	ifp = ifnet_byindex(ifindex);
2874 	if (ifp == NULL) {
2875 		CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2876 		    __func__, ifindex);
2877 		return (ENOENT);
2878 	}
2879 
2880 	retval = sysctl_wire_old_buffer(req,
2881 	    sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2882 	if (retval)
2883 		return (retval);
2884 
2885 	IN_MULTI_LOCK();
2886 
2887 	IF_ADDR_RLOCK(ifp);
2888 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2889 		if (ifma->ifma_addr->sa_family != AF_INET ||
2890 		    ifma->ifma_protospec == NULL)
2891 			continue;
2892 		inm = (struct in_multi *)ifma->ifma_protospec;
2893 		if (!in_hosteq(inm->inm_addr, group))
2894 			continue;
2895 		fmode = inm->inm_st[1].iss_fmode;
2896 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2897 		if (retval != 0)
2898 			break;
2899 		RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2900 #ifdef KTR
2901 			struct in_addr ina;
2902 			ina.s_addr = htonl(ims->ims_haddr);
2903 			CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2904 			    inet_ntoa(ina));
2905 #endif
2906 			/*
2907 			 * Only copy-out sources which are in-mode.
2908 			 */
2909 			if (fmode != ims_get_mode(inm, ims, 1)) {
2910 				CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2911 				    __func__);
2912 				continue;
2913 			}
2914 			src.s_addr = htonl(ims->ims_haddr);
2915 			retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2916 			if (retval != 0)
2917 				break;
2918 		}
2919 	}
2920 	IF_ADDR_RUNLOCK(ifp);
2921 
2922 	IN_MULTI_UNLOCK();
2923 
2924 	return (retval);
2925 }
2926 
2927 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2928 
2929 static const char *inm_modestrs[] = { "un", "in", "ex" };
2930 
2931 static const char *
2932 inm_mode_str(const int mode)
2933 {
2934 
2935 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2936 		return (inm_modestrs[mode]);
2937 	return ("??");
2938 }
2939 
2940 static const char *inm_statestrs[] = {
2941 	"not-member",
2942 	"silent",
2943 	"idle",
2944 	"lazy",
2945 	"sleeping",
2946 	"awakening",
2947 	"query-pending",
2948 	"sg-query-pending",
2949 	"leaving"
2950 };
2951 
2952 static const char *
2953 inm_state_str(const int state)
2954 {
2955 
2956 	if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2957 		return (inm_statestrs[state]);
2958 	return ("??");
2959 }
2960 
2961 /*
2962  * Dump an in_multi structure to the console.
2963  */
2964 void
2965 inm_print(const struct in_multi *inm)
2966 {
2967 	int t;
2968 
2969 	if ((ktr_mask & KTR_IGMPV3) == 0)
2970 		return;
2971 
2972 	printf("%s: --- begin inm %p ---\n", __func__, inm);
2973 	printf("addr %s ifp %p(%s) ifma %p\n",
2974 	    inet_ntoa(inm->inm_addr),
2975 	    inm->inm_ifp,
2976 	    inm->inm_ifp->if_xname,
2977 	    inm->inm_ifma);
2978 	printf("timer %u state %s refcount %u scq.len %u\n",
2979 	    inm->inm_timer,
2980 	    inm_state_str(inm->inm_state),
2981 	    inm->inm_refcount,
2982 	    inm->inm_scq.mq_len);
2983 	printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2984 	    inm->inm_igi,
2985 	    inm->inm_nsrc,
2986 	    inm->inm_sctimer,
2987 	    inm->inm_scrv);
2988 	for (t = 0; t < 2; t++) {
2989 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2990 		    inm_mode_str(inm->inm_st[t].iss_fmode),
2991 		    inm->inm_st[t].iss_asm,
2992 		    inm->inm_st[t].iss_ex,
2993 		    inm->inm_st[t].iss_in,
2994 		    inm->inm_st[t].iss_rec);
2995 	}
2996 	printf("%s: --- end inm %p ---\n", __func__, inm);
2997 }
2998 
2999 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3000 
3001 void
3002 inm_print(const struct in_multi *inm)
3003 {
3004 
3005 }
3006 
3007 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3008 
3009 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);
3010