in6_mcast.c revision 256281
1/*
2 * Copyright (c) 2009 Bruce Simpson.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 * 3. The name of the author may not be used to endorse or promote
14 *    products derived from this software without specific prior written
15 *    permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 */
29
30/*
31 * IPv6 multicast socket, group, and socket option processing module.
32 * Normative references: RFC 2292, RFC 3492, RFC 3542, RFC 3678, RFC 3810.
33 */
34
35#include <sys/cdefs.h>
36__FBSDID("$FreeBSD: stable/10/sys/netinet6/in6_mcast.c 254629 2013-08-22 00:51:37Z delphij $");
37
38#include "opt_inet6.h"
39
40#include <sys/param.h>
41#include <sys/systm.h>
42#include <sys/kernel.h>
43#include <sys/malloc.h>
44#include <sys/mbuf.h>
45#include <sys/protosw.h>
46#include <sys/socket.h>
47#include <sys/socketvar.h>
48#include <sys/protosw.h>
49#include <sys/sysctl.h>
50#include <sys/priv.h>
51#include <sys/ktr.h>
52#include <sys/tree.h>
53
54#include <net/if.h>
55#include <net/if_dl.h>
56#include <net/route.h>
57#include <net/vnet.h>
58
59#include <netinet/in.h>
60#include <netinet/in_var.h>
61#include <netinet6/in6_var.h>
62#include <netinet/ip6.h>
63#include <netinet/icmp6.h>
64#include <netinet6/ip6_var.h>
65#include <netinet/in_pcb.h>
66#include <netinet/tcp_var.h>
67#include <netinet6/nd6.h>
68#include <netinet6/mld6_var.h>
69#include <netinet6/scope6_var.h>
70
71#ifndef KTR_MLD
72#define KTR_MLD KTR_INET6
73#endif
74
75#ifndef __SOCKUNION_DECLARED
76union sockunion {
77	struct sockaddr_storage	ss;
78	struct sockaddr		sa;
79	struct sockaddr_dl	sdl;
80	struct sockaddr_in6	sin6;
81};
82typedef union sockunion sockunion_t;
83#define __SOCKUNION_DECLARED
84#endif /* __SOCKUNION_DECLARED */
85
86static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
87    "IPv6 multicast PCB-layer source filter");
88static MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
89static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
90static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
91    "IPv6 multicast MLD-layer source filter");
92
93RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
94
95/*
96 * Locking:
97 * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
98 * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
99 *   it can be taken by code in net/if.c also.
100 * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
101 *
102 * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
103 * any need for in6_multi itself to be virtualized -- it is bound to an ifp
104 * anyway no matter what happens.
105 */
106struct mtx in6_multi_mtx;
107MTX_SYSINIT(in6_multi_mtx, &in6_multi_mtx, "in6_multi_mtx", MTX_DEF);
108
109static void	im6f_commit(struct in6_mfilter *);
110static int	im6f_get_source(struct in6_mfilter *imf,
111		    const struct sockaddr_in6 *psin,
112		    struct in6_msource **);
113static struct in6_msource *
114		im6f_graft(struct in6_mfilter *, const uint8_t,
115		    const struct sockaddr_in6 *);
116static void	im6f_leave(struct in6_mfilter *);
117static int	im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
118static void	im6f_purge(struct in6_mfilter *);
119static void	im6f_rollback(struct in6_mfilter *);
120static void	im6f_reap(struct in6_mfilter *);
121static int	im6o_grow(struct ip6_moptions *);
122static size_t	im6o_match_group(const struct ip6_moptions *,
123		    const struct ifnet *, const struct sockaddr *);
124static struct in6_msource *
125		im6o_match_source(const struct ip6_moptions *, const size_t,
126		    const struct sockaddr *);
127static void	im6s_merge(struct ip6_msource *ims,
128		    const struct in6_msource *lims, const int rollback);
129static int	in6_mc_get(struct ifnet *, const struct in6_addr *,
130		    struct in6_multi **);
131static int	in6m_get_source(struct in6_multi *inm,
132		    const struct in6_addr *addr, const int noalloc,
133		    struct ip6_msource **pims);
134static int	in6m_is_ifp_detached(const struct in6_multi *);
135static int	in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
136static void	in6m_purge(struct in6_multi *);
137static void	in6m_reap(struct in6_multi *);
138static struct ip6_moptions *
139		in6p_findmoptions(struct inpcb *);
140static int	in6p_get_source_filters(struct inpcb *, struct sockopt *);
141static int	in6p_join_group(struct inpcb *, struct sockopt *);
142static int	in6p_leave_group(struct inpcb *, struct sockopt *);
143static struct ifnet *
144		in6p_lookup_mcast_ifp(const struct inpcb *,
145		    const struct sockaddr_in6 *);
146static int	in6p_block_unblock_source(struct inpcb *, struct sockopt *);
147static int	in6p_set_multicast_if(struct inpcb *, struct sockopt *);
148static int	in6p_set_source_filters(struct inpcb *, struct sockopt *);
149static int	sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
150
151SYSCTL_DECL(_net_inet6_ip6);	/* XXX Not in any common header. */
152
153static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0,
154    "IPv6 multicast");
155
156static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
157SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
158    CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxgrpsrc, 0,
159    "Max source filters per group");
160TUNABLE_ULONG("net.inet6.ip6.mcast.maxgrpsrc", &in6_mcast_maxgrpsrc);
161
162static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
163SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
164    CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxsocksrc, 0,
165    "Max source filters per socket");
166TUNABLE_ULONG("net.inet6.ip6.mcast.maxsocksrc", &in6_mcast_maxsocksrc);
167
168/* TODO Virtualize this switch. */
169int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
170SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
171    &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
172TUNABLE_INT("net.inet6.ip6.mcast.loop", &in6_mcast_loop);
173
174static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
175    CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
176    "Per-interface stack-wide source filters");
177
178/*
179 * Inline function which wraps assertions for a valid ifp.
180 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
181 * is detached.
182 */
183static int __inline
184in6m_is_ifp_detached(const struct in6_multi *inm)
185{
186	struct ifnet *ifp;
187
188	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
189	ifp = inm->in6m_ifma->ifma_ifp;
190	if (ifp != NULL) {
191		/*
192		 * Sanity check that network-layer notion of ifp is the
193		 * same as that of link-layer.
194		 */
195		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
196	}
197
198	return (ifp == NULL);
199}
200
201/*
202 * Initialize an in6_mfilter structure to a known state at t0, t1
203 * with an empty source filter list.
204 */
205static __inline void
206im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
207{
208	memset(imf, 0, sizeof(struct in6_mfilter));
209	RB_INIT(&imf->im6f_sources);
210	imf->im6f_st[0] = st0;
211	imf->im6f_st[1] = st1;
212}
213
214/*
215 * Resize the ip6_moptions vector to the next power-of-two minus 1.
216 * May be called with locks held; do not sleep.
217 */
218static int
219im6o_grow(struct ip6_moptions *imo)
220{
221	struct in6_multi	**nmships;
222	struct in6_multi	**omships;
223	struct in6_mfilter	 *nmfilters;
224	struct in6_mfilter	 *omfilters;
225	size_t			  idx;
226	size_t			  newmax;
227	size_t			  oldmax;
228
229	nmships = NULL;
230	nmfilters = NULL;
231	omships = imo->im6o_membership;
232	omfilters = imo->im6o_mfilters;
233	oldmax = imo->im6o_max_memberships;
234	newmax = ((oldmax + 1) * 2) - 1;
235
236	if (newmax <= IPV6_MAX_MEMBERSHIPS) {
237		nmships = (struct in6_multi **)realloc(omships,
238		    sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
239		nmfilters = (struct in6_mfilter *)realloc(omfilters,
240		    sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
241		    M_NOWAIT);
242		if (nmships != NULL && nmfilters != NULL) {
243			/* Initialize newly allocated source filter heads. */
244			for (idx = oldmax; idx < newmax; idx++) {
245				im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
246				    MCAST_EXCLUDE);
247			}
248			imo->im6o_max_memberships = newmax;
249			imo->im6o_membership = nmships;
250			imo->im6o_mfilters = nmfilters;
251		}
252	}
253
254	if (nmships == NULL || nmfilters == NULL) {
255		if (nmships != NULL)
256			free(nmships, M_IP6MOPTS);
257		if (nmfilters != NULL)
258			free(nmfilters, M_IN6MFILTER);
259		return (ETOOMANYREFS);
260	}
261
262	return (0);
263}
264
265/*
266 * Find an IPv6 multicast group entry for this ip6_moptions instance
267 * which matches the specified group, and optionally an interface.
268 * Return its index into the array, or -1 if not found.
269 */
270static size_t
271im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
272    const struct sockaddr *group)
273{
274	const struct sockaddr_in6 *gsin6;
275	struct in6_multi	**pinm;
276	int		  idx;
277	int		  nmships;
278
279	gsin6 = (const struct sockaddr_in6 *)group;
280
281	/* The im6o_membership array may be lazy allocated. */
282	if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
283		return (-1);
284
285	nmships = imo->im6o_num_memberships;
286	pinm = &imo->im6o_membership[0];
287	for (idx = 0; idx < nmships; idx++, pinm++) {
288		if (*pinm == NULL)
289			continue;
290		if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
291		    IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
292		    &gsin6->sin6_addr)) {
293			break;
294		}
295	}
296	if (idx >= nmships)
297		idx = -1;
298
299	return (idx);
300}
301
302/*
303 * Find an IPv6 multicast source entry for this imo which matches
304 * the given group index for this socket, and source address.
305 *
306 * XXX TODO: The scope ID, if present in src, is stripped before
307 * any comparison. We SHOULD enforce scope/zone checks where the source
308 * filter entry has a link scope.
309 *
310 * NOTE: This does not check if the entry is in-mode, merely if
311 * it exists, which may not be the desired behaviour.
312 */
313static struct in6_msource *
314im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
315    const struct sockaddr *src)
316{
317	struct ip6_msource	 find;
318	struct in6_mfilter	*imf;
319	struct ip6_msource	*ims;
320	const sockunion_t	*psa;
321
322	KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
323	KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
324	    ("%s: invalid index %d\n", __func__, (int)gidx));
325
326	/* The im6o_mfilters array may be lazy allocated. */
327	if (imo->im6o_mfilters == NULL)
328		return (NULL);
329	imf = &imo->im6o_mfilters[gidx];
330
331	psa = (const sockunion_t *)src;
332	find.im6s_addr = psa->sin6.sin6_addr;
333	in6_clearscope(&find.im6s_addr);		/* XXX */
334	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
335
336	return ((struct in6_msource *)ims);
337}
338
339/*
340 * Perform filtering for multicast datagrams on a socket by group and source.
341 *
342 * Returns 0 if a datagram should be allowed through, or various error codes
343 * if the socket was not a member of the group, or the source was muted, etc.
344 */
345int
346im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
347    const struct sockaddr *group, const struct sockaddr *src)
348{
349	size_t gidx;
350	struct in6_msource *ims;
351	int mode;
352
353	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
354
355	gidx = im6o_match_group(imo, ifp, group);
356	if (gidx == -1)
357		return (MCAST_NOTGMEMBER);
358
359	/*
360	 * Check if the source was included in an (S,G) join.
361	 * Allow reception on exclusive memberships by default,
362	 * reject reception on inclusive memberships by default.
363	 * Exclude source only if an in-mode exclude filter exists.
364	 * Include source only if an in-mode include filter exists.
365	 * NOTE: We are comparing group state here at MLD t1 (now)
366	 * with socket-layer t0 (since last downcall).
367	 */
368	mode = imo->im6o_mfilters[gidx].im6f_st[1];
369	ims = im6o_match_source(imo, gidx, src);
370
371	if ((ims == NULL && mode == MCAST_INCLUDE) ||
372	    (ims != NULL && ims->im6sl_st[0] != mode))
373		return (MCAST_NOTSMEMBER);
374
375	return (MCAST_PASS);
376}
377
378/*
379 * Find and return a reference to an in6_multi record for (ifp, group),
380 * and bump its reference count.
381 * If one does not exist, try to allocate it, and update link-layer multicast
382 * filters on ifp to listen for group.
383 * Assumes the IN6_MULTI lock is held across the call.
384 * Return 0 if successful, otherwise return an appropriate error code.
385 */
386static int
387in6_mc_get(struct ifnet *ifp, const struct in6_addr *group,
388    struct in6_multi **pinm)
389{
390	struct sockaddr_in6	 gsin6;
391	struct ifmultiaddr	*ifma;
392	struct in6_multi	*inm;
393	int			 error;
394
395	error = 0;
396
397	/*
398	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
399	 * if_addmulti() takes this mutex itself, so we must drop and
400	 * re-acquire around the call.
401	 */
402	IN6_MULTI_LOCK_ASSERT();
403	IF_ADDR_WLOCK(ifp);
404
405	inm = in6m_lookup_locked(ifp, group);
406	if (inm != NULL) {
407		/*
408		 * If we already joined this group, just bump the
409		 * refcount and return it.
410		 */
411		KASSERT(inm->in6m_refcount >= 1,
412		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
413		++inm->in6m_refcount;
414		*pinm = inm;
415		goto out_locked;
416	}
417
418	memset(&gsin6, 0, sizeof(gsin6));
419	gsin6.sin6_family = AF_INET6;
420	gsin6.sin6_len = sizeof(struct sockaddr_in6);
421	gsin6.sin6_addr = *group;
422
423	/*
424	 * Check if a link-layer group is already associated
425	 * with this network-layer group on the given ifnet.
426	 */
427	IF_ADDR_WUNLOCK(ifp);
428	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
429	if (error != 0)
430		return (error);
431	IF_ADDR_WLOCK(ifp);
432
433	/*
434	 * If something other than netinet6 is occupying the link-layer
435	 * group, print a meaningful error message and back out of
436	 * the allocation.
437	 * Otherwise, bump the refcount on the existing network-layer
438	 * group association and return it.
439	 */
440	if (ifma->ifma_protospec != NULL) {
441		inm = (struct in6_multi *)ifma->ifma_protospec;
442#ifdef INVARIANTS
443		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
444		    __func__));
445		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
446		    ("%s: ifma not AF_INET6", __func__));
447		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
448		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
449		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
450			panic("%s: ifma %p is inconsistent with %p (%p)",
451			    __func__, ifma, inm, group);
452#endif
453		++inm->in6m_refcount;
454		*pinm = inm;
455		goto out_locked;
456	}
457
458	IF_ADDR_WLOCK_ASSERT(ifp);
459
460	/*
461	 * A new in6_multi record is needed; allocate and initialize it.
462	 * We DO NOT perform an MLD join as the in6_ layer may need to
463	 * push an initial source list down to MLD to support SSM.
464	 *
465	 * The initial source filter state is INCLUDE, {} as per the RFC.
466	 * Pending state-changes per group are subject to a bounds check.
467	 */
468	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
469	if (inm == NULL) {
470		if_delmulti_ifma(ifma);
471		error = ENOMEM;
472		goto out_locked;
473	}
474	inm->in6m_addr = *group;
475	inm->in6m_ifp = ifp;
476	inm->in6m_mli = MLD_IFINFO(ifp);
477	inm->in6m_ifma = ifma;
478	inm->in6m_refcount = 1;
479	inm->in6m_state = MLD_NOT_MEMBER;
480	IFQ_SET_MAXLEN(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
481
482	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
483	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
484	RB_INIT(&inm->in6m_srcs);
485
486	ifma->ifma_protospec = inm;
487	*pinm = inm;
488
489out_locked:
490	IF_ADDR_WUNLOCK(ifp);
491	return (error);
492}
493
494/*
495 * Drop a reference to an in6_multi record.
496 *
497 * If the refcount drops to 0, free the in6_multi record and
498 * delete the underlying link-layer membership.
499 */
500void
501in6m_release_locked(struct in6_multi *inm)
502{
503	struct ifmultiaddr *ifma;
504
505	IN6_MULTI_LOCK_ASSERT();
506
507	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
508
509	if (--inm->in6m_refcount > 0) {
510		CTR2(KTR_MLD, "%s: refcount is now %d", __func__,
511		    inm->in6m_refcount);
512		return;
513	}
514
515	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
516
517	ifma = inm->in6m_ifma;
518
519	/* XXX this access is not covered by IF_ADDR_LOCK */
520	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
521	KASSERT(ifma->ifma_protospec == inm,
522	    ("%s: ifma_protospec != inm", __func__));
523	ifma->ifma_protospec = NULL;
524
525	in6m_purge(inm);
526
527	free(inm, M_IP6MADDR);
528
529	if_delmulti_ifma(ifma);
530}
531
532/*
533 * Clear recorded source entries for a group.
534 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
535 * FIXME: Should reap.
536 */
537void
538in6m_clear_recorded(struct in6_multi *inm)
539{
540	struct ip6_msource	*ims;
541
542	IN6_MULTI_LOCK_ASSERT();
543
544	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
545		if (ims->im6s_stp) {
546			ims->im6s_stp = 0;
547			--inm->in6m_st[1].iss_rec;
548		}
549	}
550	KASSERT(inm->in6m_st[1].iss_rec == 0,
551	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
552}
553
554/*
555 * Record a source as pending for a Source-Group MLDv2 query.
556 * This lives here as it modifies the shared tree.
557 *
558 * inm is the group descriptor.
559 * naddr is the address of the source to record in network-byte order.
560 *
561 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
562 * lazy-allocate a source node in response to an SG query.
563 * Otherwise, no allocation is performed. This saves some memory
564 * with the trade-off that the source will not be reported to the
565 * router if joined in the window between the query response and
566 * the group actually being joined on the local host.
567 *
568 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
569 * This turns off the allocation of a recorded source entry if
570 * the group has not been joined.
571 *
572 * Return 0 if the source didn't exist or was already marked as recorded.
573 * Return 1 if the source was marked as recorded by this function.
574 * Return <0 if any error occured (negated errno code).
575 */
576int
577in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
578{
579	struct ip6_msource	 find;
580	struct ip6_msource	*ims, *nims;
581
582	IN6_MULTI_LOCK_ASSERT();
583
584	find.im6s_addr = *addr;
585	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
586	if (ims && ims->im6s_stp)
587		return (0);
588	if (ims == NULL) {
589		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
590			return (-ENOSPC);
591		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
592		    M_NOWAIT | M_ZERO);
593		if (nims == NULL)
594			return (-ENOMEM);
595		nims->im6s_addr = find.im6s_addr;
596		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
597		++inm->in6m_nsrc;
598		ims = nims;
599	}
600
601	/*
602	 * Mark the source as recorded and update the recorded
603	 * source count.
604	 */
605	++ims->im6s_stp;
606	++inm->in6m_st[1].iss_rec;
607
608	return (1);
609}
610
611/*
612 * Return a pointer to an in6_msource owned by an in6_mfilter,
613 * given its source address.
614 * Lazy-allocate if needed. If this is a new entry its filter state is
615 * undefined at t0.
616 *
617 * imf is the filter set being modified.
618 * addr is the source address.
619 *
620 * SMPng: May be called with locks held; malloc must not block.
621 */
622static int
623im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
624    struct in6_msource **plims)
625{
626	struct ip6_msource	 find;
627	struct ip6_msource	*ims, *nims;
628	struct in6_msource	*lims;
629	int			 error;
630
631	error = 0;
632	ims = NULL;
633	lims = NULL;
634
635	find.im6s_addr = psin->sin6_addr;
636	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
637	lims = (struct in6_msource *)ims;
638	if (lims == NULL) {
639		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
640			return (ENOSPC);
641		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
642		    M_NOWAIT | M_ZERO);
643		if (nims == NULL)
644			return (ENOMEM);
645		lims = (struct in6_msource *)nims;
646		lims->im6s_addr = find.im6s_addr;
647		lims->im6sl_st[0] = MCAST_UNDEFINED;
648		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
649		++imf->im6f_nsrc;
650	}
651
652	*plims = lims;
653
654	return (error);
655}
656
657/*
658 * Graft a source entry into an existing socket-layer filter set,
659 * maintaining any required invariants and checking allocations.
660 *
661 * The source is marked as being in the new filter mode at t1.
662 *
663 * Return the pointer to the new node, otherwise return NULL.
664 */
665static struct in6_msource *
666im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
667    const struct sockaddr_in6 *psin)
668{
669	struct ip6_msource	*nims;
670	struct in6_msource	*lims;
671
672	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
673	    M_NOWAIT | M_ZERO);
674	if (nims == NULL)
675		return (NULL);
676	lims = (struct in6_msource *)nims;
677	lims->im6s_addr = psin->sin6_addr;
678	lims->im6sl_st[0] = MCAST_UNDEFINED;
679	lims->im6sl_st[1] = st1;
680	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
681	++imf->im6f_nsrc;
682
683	return (lims);
684}
685
686/*
687 * Prune a source entry from an existing socket-layer filter set,
688 * maintaining any required invariants and checking allocations.
689 *
690 * The source is marked as being left at t1, it is not freed.
691 *
692 * Return 0 if no error occurred, otherwise return an errno value.
693 */
694static int
695im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
696{
697	struct ip6_msource	 find;
698	struct ip6_msource	*ims;
699	struct in6_msource	*lims;
700
701	find.im6s_addr = psin->sin6_addr;
702	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
703	if (ims == NULL)
704		return (ENOENT);
705	lims = (struct in6_msource *)ims;
706	lims->im6sl_st[1] = MCAST_UNDEFINED;
707	return (0);
708}
709
710/*
711 * Revert socket-layer filter set deltas at t1 to t0 state.
712 */
713static void
714im6f_rollback(struct in6_mfilter *imf)
715{
716	struct ip6_msource	*ims, *tims;
717	struct in6_msource	*lims;
718
719	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
720		lims = (struct in6_msource *)ims;
721		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
722			/* no change at t1 */
723			continue;
724		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
725			/* revert change to existing source at t1 */
726			lims->im6sl_st[1] = lims->im6sl_st[0];
727		} else {
728			/* revert source added t1 */
729			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
730			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
731			free(ims, M_IN6MFILTER);
732			imf->im6f_nsrc--;
733		}
734	}
735	imf->im6f_st[1] = imf->im6f_st[0];
736}
737
738/*
739 * Mark socket-layer filter set as INCLUDE {} at t1.
740 */
741static void
742im6f_leave(struct in6_mfilter *imf)
743{
744	struct ip6_msource	*ims;
745	struct in6_msource	*lims;
746
747	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
748		lims = (struct in6_msource *)ims;
749		lims->im6sl_st[1] = MCAST_UNDEFINED;
750	}
751	imf->im6f_st[1] = MCAST_INCLUDE;
752}
753
754/*
755 * Mark socket-layer filter set deltas as committed.
756 */
757static void
758im6f_commit(struct in6_mfilter *imf)
759{
760	struct ip6_msource	*ims;
761	struct in6_msource	*lims;
762
763	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
764		lims = (struct in6_msource *)ims;
765		lims->im6sl_st[0] = lims->im6sl_st[1];
766	}
767	imf->im6f_st[0] = imf->im6f_st[1];
768}
769
770/*
771 * Reap unreferenced sources from socket-layer filter set.
772 */
773static void
774im6f_reap(struct in6_mfilter *imf)
775{
776	struct ip6_msource	*ims, *tims;
777	struct in6_msource	*lims;
778
779	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
780		lims = (struct in6_msource *)ims;
781		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
782		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
783			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
784			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
785			free(ims, M_IN6MFILTER);
786			imf->im6f_nsrc--;
787		}
788	}
789}
790
791/*
792 * Purge socket-layer filter set.
793 */
794static void
795im6f_purge(struct in6_mfilter *imf)
796{
797	struct ip6_msource	*ims, *tims;
798
799	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
800		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
801		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
802		free(ims, M_IN6MFILTER);
803		imf->im6f_nsrc--;
804	}
805	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
806	KASSERT(RB_EMPTY(&imf->im6f_sources),
807	    ("%s: im6f_sources not empty", __func__));
808}
809
810/*
811 * Look up a source filter entry for a multicast group.
812 *
813 * inm is the group descriptor to work with.
814 * addr is the IPv6 address to look up.
815 * noalloc may be non-zero to suppress allocation of sources.
816 * *pims will be set to the address of the retrieved or allocated source.
817 *
818 * SMPng: NOTE: may be called with locks held.
819 * Return 0 if successful, otherwise return a non-zero error code.
820 */
821static int
822in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
823    const int noalloc, struct ip6_msource **pims)
824{
825	struct ip6_msource	 find;
826	struct ip6_msource	*ims, *nims;
827#ifdef KTR
828	char			 ip6tbuf[INET6_ADDRSTRLEN];
829#endif
830
831	find.im6s_addr = *addr;
832	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
833	if (ims == NULL && !noalloc) {
834		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
835			return (ENOSPC);
836		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
837		    M_NOWAIT | M_ZERO);
838		if (nims == NULL)
839			return (ENOMEM);
840		nims->im6s_addr = *addr;
841		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
842		++inm->in6m_nsrc;
843		ims = nims;
844		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
845		    ip6_sprintf(ip6tbuf, addr), ims);
846	}
847
848	*pims = ims;
849	return (0);
850}
851
852/*
853 * Merge socket-layer source into MLD-layer source.
854 * If rollback is non-zero, perform the inverse of the merge.
855 */
856static void
857im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
858    const int rollback)
859{
860	int n = rollback ? -1 : 1;
861#ifdef KTR
862	char ip6tbuf[INET6_ADDRSTRLEN];
863
864	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
865#endif
866
867	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
868		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
869		ims->im6s_st[1].ex -= n;
870	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
871		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
872		ims->im6s_st[1].in -= n;
873	}
874
875	if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
876		CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
877		ims->im6s_st[1].ex += n;
878	} else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
879		CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
880		ims->im6s_st[1].in += n;
881	}
882}
883
884/*
885 * Atomically update the global in6_multi state, when a membership's
886 * filter list is being updated in any way.
887 *
888 * imf is the per-inpcb-membership group filter pointer.
889 * A fake imf may be passed for in-kernel consumers.
890 *
891 * XXX This is a candidate for a set-symmetric-difference style loop
892 * which would eliminate the repeated lookup from root of ims nodes,
893 * as they share the same key space.
894 *
895 * If any error occurred this function will back out of refcounts
896 * and return a non-zero value.
897 */
898static int
899in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
900{
901	struct ip6_msource	*ims, *nims;
902	struct in6_msource	*lims;
903	int			 schanged, error;
904	int			 nsrc0, nsrc1;
905
906	schanged = 0;
907	error = 0;
908	nsrc1 = nsrc0 = 0;
909
910	/*
911	 * Update the source filters first, as this may fail.
912	 * Maintain count of in-mode filters at t0, t1. These are
913	 * used to work out if we transition into ASM mode or not.
914	 * Maintain a count of source filters whose state was
915	 * actually modified by this operation.
916	 */
917	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
918		lims = (struct in6_msource *)ims;
919		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
920		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
921		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
922		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
923		++schanged;
924		if (error)
925			break;
926		im6s_merge(nims, lims, 0);
927	}
928	if (error) {
929		struct ip6_msource *bims;
930
931		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
932			lims = (struct in6_msource *)ims;
933			if (lims->im6sl_st[0] == lims->im6sl_st[1])
934				continue;
935			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
936			if (bims == NULL)
937				continue;
938			im6s_merge(bims, lims, 1);
939		}
940		goto out_reap;
941	}
942
943	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
944	    __func__, nsrc0, nsrc1);
945
946	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
947	if (imf->im6f_st[0] == imf->im6f_st[1] &&
948	    imf->im6f_st[1] == MCAST_INCLUDE) {
949		if (nsrc1 == 0) {
950			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
951			--inm->in6m_st[1].iss_in;
952		}
953	}
954
955	/* Handle filter mode transition on socket. */
956	if (imf->im6f_st[0] != imf->im6f_st[1]) {
957		CTR3(KTR_MLD, "%s: imf transition %d to %d",
958		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
959
960		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
961			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
962			--inm->in6m_st[1].iss_ex;
963		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
964			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
965			--inm->in6m_st[1].iss_in;
966		}
967
968		if (imf->im6f_st[1] == MCAST_EXCLUDE) {
969			CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
970			inm->in6m_st[1].iss_ex++;
971		} else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
972			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
973			inm->in6m_st[1].iss_in++;
974		}
975	}
976
977	/*
978	 * Track inm filter state in terms of listener counts.
979	 * If there are any exclusive listeners, stack-wide
980	 * membership is exclusive.
981	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
982	 * If no listeners remain, state is undefined at t1,
983	 * and the MLD lifecycle for this group should finish.
984	 */
985	if (inm->in6m_st[1].iss_ex > 0) {
986		CTR1(KTR_MLD, "%s: transition to EX", __func__);
987		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
988	} else if (inm->in6m_st[1].iss_in > 0) {
989		CTR1(KTR_MLD, "%s: transition to IN", __func__);
990		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
991	} else {
992		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
993		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
994	}
995
996	/* Decrement ASM listener count on transition out of ASM mode. */
997	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
998		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
999		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1000			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1001			--inm->in6m_st[1].iss_asm;
1002	}
1003
1004	/* Increment ASM listener count on transition to ASM mode. */
1005	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1006		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1007		inm->in6m_st[1].iss_asm++;
1008	}
1009
1010	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1011	in6m_print(inm);
1012
1013out_reap:
1014	if (schanged > 0) {
1015		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1016		in6m_reap(inm);
1017	}
1018	return (error);
1019}
1020
1021/*
1022 * Mark an in6_multi's filter set deltas as committed.
1023 * Called by MLD after a state change has been enqueued.
1024 */
1025void
1026in6m_commit(struct in6_multi *inm)
1027{
1028	struct ip6_msource	*ims;
1029
1030	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1031	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1032	in6m_print(inm);
1033
1034	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1035		ims->im6s_st[0] = ims->im6s_st[1];
1036	}
1037	inm->in6m_st[0] = inm->in6m_st[1];
1038}
1039
1040/*
1041 * Reap unreferenced nodes from an in6_multi's filter set.
1042 */
1043static void
1044in6m_reap(struct in6_multi *inm)
1045{
1046	struct ip6_msource	*ims, *tims;
1047
1048	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1049		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1050		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1051		    ims->im6s_stp != 0)
1052			continue;
1053		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1054		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1055		free(ims, M_IP6MSOURCE);
1056		inm->in6m_nsrc--;
1057	}
1058}
1059
1060/*
1061 * Purge all source nodes from an in6_multi's filter set.
1062 */
1063static void
1064in6m_purge(struct in6_multi *inm)
1065{
1066	struct ip6_msource	*ims, *tims;
1067
1068	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1069		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1070		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1071		free(ims, M_IP6MSOURCE);
1072		inm->in6m_nsrc--;
1073	}
1074}
1075
1076/*
1077 * Join a multicast address w/o sources.
1078 * KAME compatibility entry point.
1079 *
1080 * SMPng: Assume no mc locks held by caller.
1081 */
1082struct in6_multi_mship *
1083in6_joingroup(struct ifnet *ifp, struct in6_addr *mcaddr,
1084    int *errorp, int delay)
1085{
1086	struct in6_multi_mship *imm;
1087	int error;
1088
1089	imm = malloc(sizeof(*imm), M_IP6MADDR, M_NOWAIT);
1090	if (imm == NULL) {
1091		*errorp = ENOBUFS;
1092		return (NULL);
1093	}
1094
1095	delay = (delay * PR_FASTHZ) / hz;
1096
1097	error = in6_mc_join(ifp, mcaddr, NULL, &imm->i6mm_maddr, delay);
1098	if (error) {
1099		*errorp = error;
1100		free(imm, M_IP6MADDR);
1101		return (NULL);
1102	}
1103
1104	return (imm);
1105}
1106
1107/*
1108 * Leave a multicast address w/o sources.
1109 * KAME compatibility entry point.
1110 *
1111 * SMPng: Assume no mc locks held by caller.
1112 */
1113int
1114in6_leavegroup(struct in6_multi_mship *imm)
1115{
1116
1117	if (imm->i6mm_maddr != NULL)
1118		in6_mc_leave(imm->i6mm_maddr, NULL);
1119	free(imm,  M_IP6MADDR);
1120	return 0;
1121}
1122
1123/*
1124 * Join a multicast group; unlocked entry point.
1125 *
1126 * SMPng: XXX: in6_mc_join() is called from in6_control() when upper
1127 * locks are not held. Fortunately, ifp is unlikely to have been detached
1128 * at this point, so we assume it's OK to recurse.
1129 */
1130int
1131in6_mc_join(struct ifnet *ifp, const struct in6_addr *mcaddr,
1132    /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1133    const int delay)
1134{
1135	int error;
1136
1137	IN6_MULTI_LOCK();
1138	error = in6_mc_join_locked(ifp, mcaddr, imf, pinm, delay);
1139	IN6_MULTI_UNLOCK();
1140
1141	return (error);
1142}
1143
1144/*
1145 * Join a multicast group; real entry point.
1146 *
1147 * Only preserves atomicity at inm level.
1148 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1149 *
1150 * If the MLD downcall fails, the group is not joined, and an error
1151 * code is returned.
1152 */
1153int
1154in6_mc_join_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1155    /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1156    const int delay)
1157{
1158	struct in6_mfilter	 timf;
1159	struct in6_multi	*inm;
1160	int			 error;
1161#ifdef KTR
1162	char			 ip6tbuf[INET6_ADDRSTRLEN];
1163#endif
1164
1165#ifdef INVARIANTS
1166	/*
1167	 * Sanity: Check scope zone ID was set for ifp, if and
1168	 * only if group is scoped to an interface.
1169	 */
1170	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1171	    ("%s: not a multicast address", __func__));
1172	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1173	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1174		KASSERT(mcaddr->s6_addr16[1] != 0,
1175		    ("%s: scope zone ID not set", __func__));
1176	}
1177#endif
1178
1179	IN6_MULTI_LOCK_ASSERT();
1180
1181	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1182	    ip6_sprintf(ip6tbuf, mcaddr), ifp, ifp->if_xname);
1183
1184	error = 0;
1185	inm = NULL;
1186
1187	/*
1188	 * If no imf was specified (i.e. kernel consumer),
1189	 * fake one up and assume it is an ASM join.
1190	 */
1191	if (imf == NULL) {
1192		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1193		imf = &timf;
1194	}
1195
1196	error = in6_mc_get(ifp, mcaddr, &inm);
1197	if (error) {
1198		CTR1(KTR_MLD, "%s: in6_mc_get() failure", __func__);
1199		return (error);
1200	}
1201
1202	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1203	error = in6m_merge(inm, imf);
1204	if (error) {
1205		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1206		goto out_in6m_release;
1207	}
1208
1209	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1210	error = mld_change_state(inm, delay);
1211	if (error) {
1212		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1213		goto out_in6m_release;
1214	}
1215
1216out_in6m_release:
1217	if (error) {
1218		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1219		in6m_release_locked(inm);
1220	} else {
1221		*pinm = inm;
1222	}
1223
1224	return (error);
1225}
1226
1227/*
1228 * Leave a multicast group; unlocked entry point.
1229 */
1230int
1231in6_mc_leave(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1232{
1233	struct ifnet *ifp;
1234	int error;
1235
1236	ifp = inm->in6m_ifp;
1237
1238	IN6_MULTI_LOCK();
1239	error = in6_mc_leave_locked(inm, imf);
1240	IN6_MULTI_UNLOCK();
1241
1242	return (error);
1243}
1244
1245/*
1246 * Leave a multicast group; real entry point.
1247 * All source filters will be expunged.
1248 *
1249 * Only preserves atomicity at inm level.
1250 *
1251 * Holding the write lock for the INP which contains imf
1252 * is highly advisable. We can't assert for it as imf does not
1253 * contain a back-pointer to the owning inp.
1254 *
1255 * Note: This is not the same as in6m_release(*) as this function also
1256 * makes a state change downcall into MLD.
1257 */
1258int
1259in6_mc_leave_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1260{
1261	struct in6_mfilter	 timf;
1262	int			 error;
1263#ifdef KTR
1264	char			 ip6tbuf[INET6_ADDRSTRLEN];
1265#endif
1266
1267	error = 0;
1268
1269	IN6_MULTI_LOCK_ASSERT();
1270
1271	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1272	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1273	    (in6m_is_ifp_detached(inm) ? "null" : inm->in6m_ifp->if_xname),
1274	    imf);
1275
1276	/*
1277	 * If no imf was specified (i.e. kernel consumer),
1278	 * fake one up and assume it is an ASM join.
1279	 */
1280	if (imf == NULL) {
1281		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1282		imf = &timf;
1283	}
1284
1285	/*
1286	 * Begin state merge transaction at MLD layer.
1287	 *
1288	 * As this particular invocation should not cause any memory
1289	 * to be allocated, and there is no opportunity to roll back
1290	 * the transaction, it MUST NOT fail.
1291	 */
1292	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1293	error = in6m_merge(inm, imf);
1294	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1295
1296	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1297	error = mld_change_state(inm, 0);
1298	if (error)
1299		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1300
1301	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1302	in6m_release_locked(inm);
1303
1304	return (error);
1305}
1306
1307/*
1308 * Block or unblock an ASM multicast source on an inpcb.
1309 * This implements the delta-based API described in RFC 3678.
1310 *
1311 * The delta-based API applies only to exclusive-mode memberships.
1312 * An MLD downcall will be performed.
1313 *
1314 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1315 *
1316 * Return 0 if successful, otherwise return an appropriate error code.
1317 */
1318static int
1319in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1320{
1321	struct group_source_req		 gsr;
1322	sockunion_t			*gsa, *ssa;
1323	struct ifnet			*ifp;
1324	struct in6_mfilter		*imf;
1325	struct ip6_moptions		*imo;
1326	struct in6_msource		*ims;
1327	struct in6_multi			*inm;
1328	size_t				 idx;
1329	uint16_t			 fmode;
1330	int				 error, doblock;
1331#ifdef KTR
1332	char				 ip6tbuf[INET6_ADDRSTRLEN];
1333#endif
1334
1335	ifp = NULL;
1336	error = 0;
1337	doblock = 0;
1338
1339	memset(&gsr, 0, sizeof(struct group_source_req));
1340	gsa = (sockunion_t *)&gsr.gsr_group;
1341	ssa = (sockunion_t *)&gsr.gsr_source;
1342
1343	switch (sopt->sopt_name) {
1344	case MCAST_BLOCK_SOURCE:
1345	case MCAST_UNBLOCK_SOURCE:
1346		error = sooptcopyin(sopt, &gsr,
1347		    sizeof(struct group_source_req),
1348		    sizeof(struct group_source_req));
1349		if (error)
1350			return (error);
1351
1352		if (gsa->sin6.sin6_family != AF_INET6 ||
1353		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1354			return (EINVAL);
1355
1356		if (ssa->sin6.sin6_family != AF_INET6 ||
1357		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1358			return (EINVAL);
1359
1360		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1361			return (EADDRNOTAVAIL);
1362
1363		ifp = ifnet_byindex(gsr.gsr_interface);
1364
1365		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1366			doblock = 1;
1367		break;
1368
1369	default:
1370		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1371		    __func__, sopt->sopt_name);
1372		return (EOPNOTSUPP);
1373		break;
1374	}
1375
1376	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1377		return (EINVAL);
1378
1379	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1380
1381	/*
1382	 * Check if we are actually a member of this group.
1383	 */
1384	imo = in6p_findmoptions(inp);
1385	idx = im6o_match_group(imo, ifp, &gsa->sa);
1386	if (idx == -1 || imo->im6o_mfilters == NULL) {
1387		error = EADDRNOTAVAIL;
1388		goto out_in6p_locked;
1389	}
1390
1391	KASSERT(imo->im6o_mfilters != NULL,
1392	    ("%s: im6o_mfilters not allocated", __func__));
1393	imf = &imo->im6o_mfilters[idx];
1394	inm = imo->im6o_membership[idx];
1395
1396	/*
1397	 * Attempting to use the delta-based API on an
1398	 * non exclusive-mode membership is an error.
1399	 */
1400	fmode = imf->im6f_st[0];
1401	if (fmode != MCAST_EXCLUDE) {
1402		error = EINVAL;
1403		goto out_in6p_locked;
1404	}
1405
1406	/*
1407	 * Deal with error cases up-front:
1408	 *  Asked to block, but already blocked; or
1409	 *  Asked to unblock, but nothing to unblock.
1410	 * If adding a new block entry, allocate it.
1411	 */
1412	ims = im6o_match_source(imo, idx, &ssa->sa);
1413	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1414		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1415		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1416		    doblock ? "" : "not ");
1417		error = EADDRNOTAVAIL;
1418		goto out_in6p_locked;
1419	}
1420
1421	INP_WLOCK_ASSERT(inp);
1422
1423	/*
1424	 * Begin state merge transaction at socket layer.
1425	 */
1426	if (doblock) {
1427		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1428		ims = im6f_graft(imf, fmode, &ssa->sin6);
1429		if (ims == NULL)
1430			error = ENOMEM;
1431	} else {
1432		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1433		error = im6f_prune(imf, &ssa->sin6);
1434	}
1435
1436	if (error) {
1437		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1438		goto out_im6f_rollback;
1439	}
1440
1441	/*
1442	 * Begin state merge transaction at MLD layer.
1443	 */
1444	IN6_MULTI_LOCK();
1445
1446	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1447	error = in6m_merge(inm, imf);
1448	if (error) {
1449		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1450		goto out_im6f_rollback;
1451	}
1452
1453	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1454	error = mld_change_state(inm, 0);
1455	if (error)
1456		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1457
1458	IN6_MULTI_UNLOCK();
1459
1460out_im6f_rollback:
1461	if (error)
1462		im6f_rollback(imf);
1463	else
1464		im6f_commit(imf);
1465
1466	im6f_reap(imf);
1467
1468out_in6p_locked:
1469	INP_WUNLOCK(inp);
1470	return (error);
1471}
1472
1473/*
1474 * Given an inpcb, return its multicast options structure pointer.  Accepts
1475 * an unlocked inpcb pointer, but will return it locked.  May sleep.
1476 *
1477 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1478 * SMPng: NOTE: Returns with the INP write lock held.
1479 */
1480static struct ip6_moptions *
1481in6p_findmoptions(struct inpcb *inp)
1482{
1483	struct ip6_moptions	 *imo;
1484	struct in6_multi		**immp;
1485	struct in6_mfilter	 *imfp;
1486	size_t			  idx;
1487
1488	INP_WLOCK(inp);
1489	if (inp->in6p_moptions != NULL)
1490		return (inp->in6p_moptions);
1491
1492	INP_WUNLOCK(inp);
1493
1494	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1495	immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1496	    M_WAITOK | M_ZERO);
1497	imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1498	    M_IN6MFILTER, M_WAITOK);
1499
1500	imo->im6o_multicast_ifp = NULL;
1501	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1502	imo->im6o_multicast_loop = in6_mcast_loop;
1503	imo->im6o_num_memberships = 0;
1504	imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1505	imo->im6o_membership = immp;
1506
1507	/* Initialize per-group source filters. */
1508	for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1509		im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1510	imo->im6o_mfilters = imfp;
1511
1512	INP_WLOCK(inp);
1513	if (inp->in6p_moptions != NULL) {
1514		free(imfp, M_IN6MFILTER);
1515		free(immp, M_IP6MOPTS);
1516		free(imo, M_IP6MOPTS);
1517		return (inp->in6p_moptions);
1518	}
1519	inp->in6p_moptions = imo;
1520	return (imo);
1521}
1522
1523/*
1524 * Discard the IPv6 multicast options (and source filters).
1525 *
1526 * SMPng: NOTE: assumes INP write lock is held.
1527 */
1528void
1529ip6_freemoptions(struct ip6_moptions *imo)
1530{
1531	struct in6_mfilter	*imf;
1532	size_t			 idx, nmships;
1533
1534	KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1535
1536	nmships = imo->im6o_num_memberships;
1537	for (idx = 0; idx < nmships; ++idx) {
1538		imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1539		if (imf)
1540			im6f_leave(imf);
1541		/* XXX this will thrash the lock(s) */
1542		(void)in6_mc_leave(imo->im6o_membership[idx], imf);
1543		if (imf)
1544			im6f_purge(imf);
1545	}
1546
1547	if (imo->im6o_mfilters)
1548		free(imo->im6o_mfilters, M_IN6MFILTER);
1549	free(imo->im6o_membership, M_IP6MOPTS);
1550	free(imo, M_IP6MOPTS);
1551}
1552
1553/*
1554 * Atomically get source filters on a socket for an IPv6 multicast group.
1555 * Called with INP lock held; returns with lock released.
1556 */
1557static int
1558in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1559{
1560	struct __msfilterreq	 msfr;
1561	sockunion_t		*gsa;
1562	struct ifnet		*ifp;
1563	struct ip6_moptions	*imo;
1564	struct in6_mfilter	*imf;
1565	struct ip6_msource	*ims;
1566	struct in6_msource	*lims;
1567	struct sockaddr_in6	*psin;
1568	struct sockaddr_storage	*ptss;
1569	struct sockaddr_storage	*tss;
1570	int			 error;
1571	size_t			 idx, nsrcs, ncsrcs;
1572
1573	INP_WLOCK_ASSERT(inp);
1574
1575	imo = inp->in6p_moptions;
1576	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1577
1578	INP_WUNLOCK(inp);
1579
1580	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1581	    sizeof(struct __msfilterreq));
1582	if (error)
1583		return (error);
1584
1585	if (msfr.msfr_group.ss_family != AF_INET6 ||
1586	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1587		return (EINVAL);
1588
1589	gsa = (sockunion_t *)&msfr.msfr_group;
1590	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1591		return (EINVAL);
1592
1593	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1594		return (EADDRNOTAVAIL);
1595	ifp = ifnet_byindex(msfr.msfr_ifindex);
1596	if (ifp == NULL)
1597		return (EADDRNOTAVAIL);
1598	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1599
1600	INP_WLOCK(inp);
1601
1602	/*
1603	 * Lookup group on the socket.
1604	 */
1605	idx = im6o_match_group(imo, ifp, &gsa->sa);
1606	if (idx == -1 || imo->im6o_mfilters == NULL) {
1607		INP_WUNLOCK(inp);
1608		return (EADDRNOTAVAIL);
1609	}
1610	imf = &imo->im6o_mfilters[idx];
1611
1612	/*
1613	 * Ignore memberships which are in limbo.
1614	 */
1615	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1616		INP_WUNLOCK(inp);
1617		return (EAGAIN);
1618	}
1619	msfr.msfr_fmode = imf->im6f_st[1];
1620
1621	/*
1622	 * If the user specified a buffer, copy out the source filter
1623	 * entries to userland gracefully.
1624	 * We only copy out the number of entries which userland
1625	 * has asked for, but we always tell userland how big the
1626	 * buffer really needs to be.
1627	 */
1628	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1629		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1630	tss = NULL;
1631	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1632		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1633		    M_TEMP, M_NOWAIT | M_ZERO);
1634		if (tss == NULL) {
1635			INP_WUNLOCK(inp);
1636			return (ENOBUFS);
1637		}
1638	}
1639
1640	/*
1641	 * Count number of sources in-mode at t0.
1642	 * If buffer space exists and remains, copy out source entries.
1643	 */
1644	nsrcs = msfr.msfr_nsrcs;
1645	ncsrcs = 0;
1646	ptss = tss;
1647	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1648		lims = (struct in6_msource *)ims;
1649		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1650		    lims->im6sl_st[0] != imf->im6f_st[0])
1651			continue;
1652		++ncsrcs;
1653		if (tss != NULL && nsrcs > 0) {
1654			psin = (struct sockaddr_in6 *)ptss;
1655			psin->sin6_family = AF_INET6;
1656			psin->sin6_len = sizeof(struct sockaddr_in6);
1657			psin->sin6_addr = lims->im6s_addr;
1658			psin->sin6_port = 0;
1659			--nsrcs;
1660			++ptss;
1661		}
1662	}
1663
1664	INP_WUNLOCK(inp);
1665
1666	if (tss != NULL) {
1667		error = copyout(tss, msfr.msfr_srcs,
1668		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1669		free(tss, M_TEMP);
1670		if (error)
1671			return (error);
1672	}
1673
1674	msfr.msfr_nsrcs = ncsrcs;
1675	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1676
1677	return (error);
1678}
1679
1680/*
1681 * Return the IP multicast options in response to user getsockopt().
1682 */
1683int
1684ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1685{
1686	struct ip6_moptions	*im6o;
1687	int			 error;
1688	u_int			 optval;
1689
1690	INP_WLOCK(inp);
1691	im6o = inp->in6p_moptions;
1692	/*
1693	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1694	 * or is a divert socket, reject it.
1695	 */
1696	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1697	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1698	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1699		INP_WUNLOCK(inp);
1700		return (EOPNOTSUPP);
1701	}
1702
1703	error = 0;
1704	switch (sopt->sopt_name) {
1705	case IPV6_MULTICAST_IF:
1706		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1707			optval = 0;
1708		} else {
1709			optval = im6o->im6o_multicast_ifp->if_index;
1710		}
1711		INP_WUNLOCK(inp);
1712		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1713		break;
1714
1715	case IPV6_MULTICAST_HOPS:
1716		if (im6o == NULL)
1717			optval = V_ip6_defmcasthlim;
1718		else
1719			optval = im6o->im6o_multicast_hlim;
1720		INP_WUNLOCK(inp);
1721		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1722		break;
1723
1724	case IPV6_MULTICAST_LOOP:
1725		if (im6o == NULL)
1726			optval = in6_mcast_loop; /* XXX VIMAGE */
1727		else
1728			optval = im6o->im6o_multicast_loop;
1729		INP_WUNLOCK(inp);
1730		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1731		break;
1732
1733	case IPV6_MSFILTER:
1734		if (im6o == NULL) {
1735			error = EADDRNOTAVAIL;
1736			INP_WUNLOCK(inp);
1737		} else {
1738			error = in6p_get_source_filters(inp, sopt);
1739		}
1740		break;
1741
1742	default:
1743		INP_WUNLOCK(inp);
1744		error = ENOPROTOOPT;
1745		break;
1746	}
1747
1748	INP_UNLOCK_ASSERT(inp);
1749
1750	return (error);
1751}
1752
1753/*
1754 * Look up the ifnet to use for a multicast group membership,
1755 * given the address of an IPv6 group.
1756 *
1757 * This routine exists to support legacy IPv6 multicast applications.
1758 *
1759 * If inp is non-NULL, use this socket's current FIB number for any
1760 * required FIB lookup. Look up the group address in the unicast FIB,
1761 * and use its ifp; usually, this points to the default next-hop.
1762 * If the FIB lookup fails, return NULL.
1763 *
1764 * FUTURE: Support multiple forwarding tables for IPv6.
1765 *
1766 * Returns NULL if no ifp could be found.
1767 */
1768static struct ifnet *
1769in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1770    const struct sockaddr_in6 *gsin6)
1771{
1772	struct route_in6	 ro6;
1773	struct ifnet		*ifp;
1774
1775	KASSERT(in6p->inp_vflag & INP_IPV6,
1776	    ("%s: not INP_IPV6 inpcb", __func__));
1777	KASSERT(gsin6->sin6_family == AF_INET6,
1778	    ("%s: not AF_INET6 group", __func__));
1779	KASSERT(IN6_IS_ADDR_MULTICAST(&gsin6->sin6_addr),
1780	    ("%s: not multicast", __func__));
1781
1782	ifp = NULL;
1783	memset(&ro6, 0, sizeof(struct route_in6));
1784	memcpy(&ro6.ro_dst, gsin6, sizeof(struct sockaddr_in6));
1785	rtalloc_ign_fib((struct route *)&ro6, 0,
1786	    in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB);
1787	if (ro6.ro_rt != NULL) {
1788		ifp = ro6.ro_rt->rt_ifp;
1789		KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1790		RTFREE(ro6.ro_rt);
1791	}
1792
1793	return (ifp);
1794}
1795
1796/*
1797 * Join an IPv6 multicast group, possibly with a source.
1798 *
1799 * FIXME: The KAME use of the unspecified address (::)
1800 * to join *all* multicast groups is currently unsupported.
1801 */
1802static int
1803in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1804{
1805	struct group_source_req		 gsr;
1806	sockunion_t			*gsa, *ssa;
1807	struct ifnet			*ifp;
1808	struct in6_mfilter		*imf;
1809	struct ip6_moptions		*imo;
1810	struct in6_multi		*inm;
1811	struct in6_msource		*lims;
1812	size_t				 idx;
1813	int				 error, is_new;
1814
1815	ifp = NULL;
1816	imf = NULL;
1817	lims = NULL;
1818	error = 0;
1819	is_new = 0;
1820
1821	memset(&gsr, 0, sizeof(struct group_source_req));
1822	gsa = (sockunion_t *)&gsr.gsr_group;
1823	gsa->ss.ss_family = AF_UNSPEC;
1824	ssa = (sockunion_t *)&gsr.gsr_source;
1825	ssa->ss.ss_family = AF_UNSPEC;
1826
1827	/*
1828	 * Chew everything into struct group_source_req.
1829	 * Overwrite the port field if present, as the sockaddr
1830	 * being copied in may be matched with a binary comparison.
1831	 * Ignore passed-in scope ID.
1832	 */
1833	switch (sopt->sopt_name) {
1834	case IPV6_JOIN_GROUP: {
1835		struct ipv6_mreq mreq;
1836
1837		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1838		    sizeof(struct ipv6_mreq));
1839		if (error)
1840			return (error);
1841
1842		gsa->sin6.sin6_family = AF_INET6;
1843		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1844		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1845
1846		if (mreq.ipv6mr_interface == 0) {
1847			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1848		} else {
1849			if (mreq.ipv6mr_interface < 0 ||
1850			    V_if_index < mreq.ipv6mr_interface)
1851				return (EADDRNOTAVAIL);
1852			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1853		}
1854		CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1855		    __func__, mreq.ipv6mr_interface, ifp);
1856	} break;
1857
1858	case MCAST_JOIN_GROUP:
1859	case MCAST_JOIN_SOURCE_GROUP:
1860		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1861			error = sooptcopyin(sopt, &gsr,
1862			    sizeof(struct group_req),
1863			    sizeof(struct group_req));
1864		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1865			error = sooptcopyin(sopt, &gsr,
1866			    sizeof(struct group_source_req),
1867			    sizeof(struct group_source_req));
1868		}
1869		if (error)
1870			return (error);
1871
1872		if (gsa->sin6.sin6_family != AF_INET6 ||
1873		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1874			return (EINVAL);
1875
1876		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1877			if (ssa->sin6.sin6_family != AF_INET6 ||
1878			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1879				return (EINVAL);
1880			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1881				return (EINVAL);
1882			/*
1883			 * TODO: Validate embedded scope ID in source
1884			 * list entry against passed-in ifp, if and only
1885			 * if source list filter entry is iface or node local.
1886			 */
1887			in6_clearscope(&ssa->sin6.sin6_addr);
1888			ssa->sin6.sin6_port = 0;
1889			ssa->sin6.sin6_scope_id = 0;
1890		}
1891
1892		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1893			return (EADDRNOTAVAIL);
1894		ifp = ifnet_byindex(gsr.gsr_interface);
1895		break;
1896
1897	default:
1898		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1899		    __func__, sopt->sopt_name);
1900		return (EOPNOTSUPP);
1901		break;
1902	}
1903
1904	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1905		return (EINVAL);
1906
1907	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1908		return (EADDRNOTAVAIL);
1909
1910	gsa->sin6.sin6_port = 0;
1911	gsa->sin6.sin6_scope_id = 0;
1912
1913	/*
1914	 * Always set the scope zone ID on memberships created from userland.
1915	 * Use the passed-in ifp to do this.
1916	 * XXX The in6_setscope() return value is meaningless.
1917	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1918	 */
1919	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1920
1921	imo = in6p_findmoptions(inp);
1922	idx = im6o_match_group(imo, ifp, &gsa->sa);
1923	if (idx == -1) {
1924		is_new = 1;
1925	} else {
1926		inm = imo->im6o_membership[idx];
1927		imf = &imo->im6o_mfilters[idx];
1928		if (ssa->ss.ss_family != AF_UNSPEC) {
1929			/*
1930			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1931			 * is an error. On an existing inclusive membership,
1932			 * it just adds the source to the filter list.
1933			 */
1934			if (imf->im6f_st[1] != MCAST_INCLUDE) {
1935				error = EINVAL;
1936				goto out_in6p_locked;
1937			}
1938			/*
1939			 * Throw out duplicates.
1940			 *
1941			 * XXX FIXME: This makes a naive assumption that
1942			 * even if entries exist for *ssa in this imf,
1943			 * they will be rejected as dupes, even if they
1944			 * are not valid in the current mode (in-mode).
1945			 *
1946			 * in6_msource is transactioned just as for anything
1947			 * else in SSM -- but note naive use of in6m_graft()
1948			 * below for allocating new filter entries.
1949			 *
1950			 * This is only an issue if someone mixes the
1951			 * full-state SSM API with the delta-based API,
1952			 * which is discouraged in the relevant RFCs.
1953			 */
1954			lims = im6o_match_source(imo, idx, &ssa->sa);
1955			if (lims != NULL /*&&
1956			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
1957				error = EADDRNOTAVAIL;
1958				goto out_in6p_locked;
1959			}
1960		} else {
1961			/*
1962			 * MCAST_JOIN_GROUP alone, on any existing membership,
1963			 * is rejected, to stop the same inpcb tying up
1964			 * multiple refs to the in_multi.
1965			 * On an existing inclusive membership, this is also
1966			 * an error; if you want to change filter mode,
1967			 * you must use the userland API setsourcefilter().
1968			 * XXX We don't reject this for imf in UNDEFINED
1969			 * state at t1, because allocation of a filter
1970			 * is atomic with allocation of a membership.
1971			 */
1972			error = EINVAL;
1973			goto out_in6p_locked;
1974		}
1975	}
1976
1977	/*
1978	 * Begin state merge transaction at socket layer.
1979	 */
1980	INP_WLOCK_ASSERT(inp);
1981
1982	if (is_new) {
1983		if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
1984			error = im6o_grow(imo);
1985			if (error)
1986				goto out_in6p_locked;
1987		}
1988		/*
1989		 * Allocate the new slot upfront so we can deal with
1990		 * grafting the new source filter in same code path
1991		 * as for join-source on existing membership.
1992		 */
1993		idx = imo->im6o_num_memberships;
1994		imo->im6o_membership[idx] = NULL;
1995		imo->im6o_num_memberships++;
1996		KASSERT(imo->im6o_mfilters != NULL,
1997		    ("%s: im6f_mfilters vector was not allocated", __func__));
1998		imf = &imo->im6o_mfilters[idx];
1999		KASSERT(RB_EMPTY(&imf->im6f_sources),
2000		    ("%s: im6f_sources not empty", __func__));
2001	}
2002
2003	/*
2004	 * Graft new source into filter list for this inpcb's
2005	 * membership of the group. The in6_multi may not have
2006	 * been allocated yet if this is a new membership, however,
2007	 * the in_mfilter slot will be allocated and must be initialized.
2008	 *
2009	 * Note: Grafting of exclusive mode filters doesn't happen
2010	 * in this path.
2011	 * XXX: Should check for non-NULL lims (node exists but may
2012	 * not be in-mode) for interop with full-state API.
2013	 */
2014	if (ssa->ss.ss_family != AF_UNSPEC) {
2015		/* Membership starts in IN mode */
2016		if (is_new) {
2017			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2018			im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2019		} else {
2020			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2021		}
2022		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2023		if (lims == NULL) {
2024			CTR1(KTR_MLD, "%s: merge imf state failed",
2025			    __func__);
2026			error = ENOMEM;
2027			goto out_im6o_free;
2028		}
2029	} else {
2030		/* No address specified; Membership starts in EX mode */
2031		if (is_new) {
2032			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2033			im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2034		}
2035	}
2036
2037	/*
2038	 * Begin state merge transaction at MLD layer.
2039	 */
2040	IN6_MULTI_LOCK();
2041
2042	if (is_new) {
2043		error = in6_mc_join_locked(ifp, &gsa->sin6.sin6_addr, imf,
2044		    &inm, 0);
2045		if (error)
2046			goto out_im6o_free;
2047		imo->im6o_membership[idx] = inm;
2048	} else {
2049		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2050		error = in6m_merge(inm, imf);
2051		if (error) {
2052			CTR1(KTR_MLD, "%s: failed to merge inm state",
2053			    __func__);
2054			goto out_im6f_rollback;
2055		}
2056		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2057		error = mld_change_state(inm, 0);
2058		if (error) {
2059			CTR1(KTR_MLD, "%s: failed mld downcall",
2060			    __func__);
2061			goto out_im6f_rollback;
2062		}
2063	}
2064
2065	IN6_MULTI_UNLOCK();
2066
2067out_im6f_rollback:
2068	INP_WLOCK_ASSERT(inp);
2069	if (error) {
2070		im6f_rollback(imf);
2071		if (is_new)
2072			im6f_purge(imf);
2073		else
2074			im6f_reap(imf);
2075	} else {
2076		im6f_commit(imf);
2077	}
2078
2079out_im6o_free:
2080	if (error && is_new) {
2081		imo->im6o_membership[idx] = NULL;
2082		--imo->im6o_num_memberships;
2083	}
2084
2085out_in6p_locked:
2086	INP_WUNLOCK(inp);
2087	return (error);
2088}
2089
2090/*
2091 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2092 */
2093static int
2094in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2095{
2096	struct ipv6_mreq		 mreq;
2097	struct group_source_req		 gsr;
2098	sockunion_t			*gsa, *ssa;
2099	struct ifnet			*ifp;
2100	struct in6_mfilter		*imf;
2101	struct ip6_moptions		*imo;
2102	struct in6_msource		*ims;
2103	struct in6_multi		*inm;
2104	uint32_t			 ifindex;
2105	size_t				 idx;
2106	int				 error, is_final;
2107#ifdef KTR
2108	char				 ip6tbuf[INET6_ADDRSTRLEN];
2109#endif
2110
2111	ifp = NULL;
2112	ifindex = 0;
2113	error = 0;
2114	is_final = 1;
2115
2116	memset(&gsr, 0, sizeof(struct group_source_req));
2117	gsa = (sockunion_t *)&gsr.gsr_group;
2118	gsa->ss.ss_family = AF_UNSPEC;
2119	ssa = (sockunion_t *)&gsr.gsr_source;
2120	ssa->ss.ss_family = AF_UNSPEC;
2121
2122	/*
2123	 * Chew everything passed in up into a struct group_source_req
2124	 * as that is easier to process.
2125	 * Note: Any embedded scope ID in the multicast group passed
2126	 * in by userland is ignored, the interface index is the recommended
2127	 * mechanism to specify an interface; see below.
2128	 */
2129	switch (sopt->sopt_name) {
2130	case IPV6_LEAVE_GROUP:
2131		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2132		    sizeof(struct ipv6_mreq));
2133		if (error)
2134			return (error);
2135		gsa->sin6.sin6_family = AF_INET6;
2136		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2137		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2138		gsa->sin6.sin6_port = 0;
2139		gsa->sin6.sin6_scope_id = 0;
2140		ifindex = mreq.ipv6mr_interface;
2141		break;
2142
2143	case MCAST_LEAVE_GROUP:
2144	case MCAST_LEAVE_SOURCE_GROUP:
2145		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2146			error = sooptcopyin(sopt, &gsr,
2147			    sizeof(struct group_req),
2148			    sizeof(struct group_req));
2149		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2150			error = sooptcopyin(sopt, &gsr,
2151			    sizeof(struct group_source_req),
2152			    sizeof(struct group_source_req));
2153		}
2154		if (error)
2155			return (error);
2156
2157		if (gsa->sin6.sin6_family != AF_INET6 ||
2158		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2159			return (EINVAL);
2160		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2161			if (ssa->sin6.sin6_family != AF_INET6 ||
2162			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2163				return (EINVAL);
2164			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2165				return (EINVAL);
2166			/*
2167			 * TODO: Validate embedded scope ID in source
2168			 * list entry against passed-in ifp, if and only
2169			 * if source list filter entry is iface or node local.
2170			 */
2171			in6_clearscope(&ssa->sin6.sin6_addr);
2172		}
2173		gsa->sin6.sin6_port = 0;
2174		gsa->sin6.sin6_scope_id = 0;
2175		ifindex = gsr.gsr_interface;
2176		break;
2177
2178	default:
2179		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2180		    __func__, sopt->sopt_name);
2181		return (EOPNOTSUPP);
2182		break;
2183	}
2184
2185	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2186		return (EINVAL);
2187
2188	/*
2189	 * Validate interface index if provided. If no interface index
2190	 * was provided separately, attempt to look the membership up
2191	 * from the default scope as a last resort to disambiguate
2192	 * the membership we are being asked to leave.
2193	 * XXX SCOPE6 lock potentially taken here.
2194	 */
2195	if (ifindex != 0) {
2196		if (ifindex < 0 || V_if_index < ifindex)
2197			return (EADDRNOTAVAIL);
2198		ifp = ifnet_byindex(ifindex);
2199		if (ifp == NULL)
2200			return (EADDRNOTAVAIL);
2201		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2202	} else {
2203		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2204		if (error)
2205			return (EADDRNOTAVAIL);
2206		/*
2207		 * Some badly behaved applications don't pass an ifindex
2208		 * or a scope ID, which is an API violation. In this case,
2209		 * perform a lookup as per a v6 join.
2210		 *
2211		 * XXX For now, stomp on zone ID for the corner case.
2212		 * This is not the 'KAME way', but we need to see the ifp
2213		 * directly until such time as this implementation is
2214		 * refactored, assuming the scope IDs are the way to go.
2215		 */
2216		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2217		if (ifindex == 0) {
2218			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2219			    "ifp for group %s.", __func__,
2220			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2221			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2222		} else {
2223			ifp = ifnet_byindex(ifindex);
2224		}
2225		if (ifp == NULL)
2226			return (EADDRNOTAVAIL);
2227	}
2228
2229	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2230	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2231
2232	/*
2233	 * Find the membership in the membership array.
2234	 */
2235	imo = in6p_findmoptions(inp);
2236	idx = im6o_match_group(imo, ifp, &gsa->sa);
2237	if (idx == -1) {
2238		error = EADDRNOTAVAIL;
2239		goto out_in6p_locked;
2240	}
2241	inm = imo->im6o_membership[idx];
2242	imf = &imo->im6o_mfilters[idx];
2243
2244	if (ssa->ss.ss_family != AF_UNSPEC)
2245		is_final = 0;
2246
2247	/*
2248	 * Begin state merge transaction at socket layer.
2249	 */
2250	INP_WLOCK_ASSERT(inp);
2251
2252	/*
2253	 * If we were instructed only to leave a given source, do so.
2254	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2255	 */
2256	if (is_final) {
2257		im6f_leave(imf);
2258	} else {
2259		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2260			error = EADDRNOTAVAIL;
2261			goto out_in6p_locked;
2262		}
2263		ims = im6o_match_source(imo, idx, &ssa->sa);
2264		if (ims == NULL) {
2265			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2266			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2267			    "not ");
2268			error = EADDRNOTAVAIL;
2269			goto out_in6p_locked;
2270		}
2271		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2272		error = im6f_prune(imf, &ssa->sin6);
2273		if (error) {
2274			CTR1(KTR_MLD, "%s: merge imf state failed",
2275			    __func__);
2276			goto out_in6p_locked;
2277		}
2278	}
2279
2280	/*
2281	 * Begin state merge transaction at MLD layer.
2282	 */
2283	IN6_MULTI_LOCK();
2284
2285	if (is_final) {
2286		/*
2287		 * Give up the multicast address record to which
2288		 * the membership points.
2289		 */
2290		(void)in6_mc_leave_locked(inm, imf);
2291	} else {
2292		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2293		error = in6m_merge(inm, imf);
2294		if (error) {
2295			CTR1(KTR_MLD, "%s: failed to merge inm state",
2296			    __func__);
2297			goto out_im6f_rollback;
2298		}
2299
2300		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2301		error = mld_change_state(inm, 0);
2302		if (error) {
2303			CTR1(KTR_MLD, "%s: failed mld downcall",
2304			    __func__);
2305		}
2306	}
2307
2308	IN6_MULTI_UNLOCK();
2309
2310out_im6f_rollback:
2311	if (error)
2312		im6f_rollback(imf);
2313	else
2314		im6f_commit(imf);
2315
2316	im6f_reap(imf);
2317
2318	if (is_final) {
2319		/* Remove the gap in the membership array. */
2320		for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2321			imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2322			imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2323		}
2324		imo->im6o_num_memberships--;
2325	}
2326
2327out_in6p_locked:
2328	INP_WUNLOCK(inp);
2329	return (error);
2330}
2331
2332/*
2333 * Select the interface for transmitting IPv6 multicast datagrams.
2334 *
2335 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2336 * may be passed to this socket option. An address of in6addr_any or an
2337 * interface index of 0 is used to remove a previous selection.
2338 * When no interface is selected, one is chosen for every send.
2339 */
2340static int
2341in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2342{
2343	struct ifnet		*ifp;
2344	struct ip6_moptions	*imo;
2345	u_int			 ifindex;
2346	int			 error;
2347
2348	if (sopt->sopt_valsize != sizeof(u_int))
2349		return (EINVAL);
2350
2351	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2352	if (error)
2353		return (error);
2354	if (ifindex < 0 || V_if_index < ifindex)
2355		return (EINVAL);
2356
2357	ifp = ifnet_byindex(ifindex);
2358	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2359		return (EADDRNOTAVAIL);
2360
2361	imo = in6p_findmoptions(inp);
2362	imo->im6o_multicast_ifp = ifp;
2363	INP_WUNLOCK(inp);
2364
2365	return (0);
2366}
2367
2368/*
2369 * Atomically set source filters on a socket for an IPv6 multicast group.
2370 *
2371 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2372 */
2373static int
2374in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2375{
2376	struct __msfilterreq	 msfr;
2377	sockunion_t		*gsa;
2378	struct ifnet		*ifp;
2379	struct in6_mfilter	*imf;
2380	struct ip6_moptions	*imo;
2381	struct in6_multi		*inm;
2382	size_t			 idx;
2383	int			 error;
2384
2385	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2386	    sizeof(struct __msfilterreq));
2387	if (error)
2388		return (error);
2389
2390	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2391		return (ENOBUFS);
2392
2393	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2394	    msfr.msfr_fmode != MCAST_INCLUDE)
2395		return (EINVAL);
2396
2397	if (msfr.msfr_group.ss_family != AF_INET6 ||
2398	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2399		return (EINVAL);
2400
2401	gsa = (sockunion_t *)&msfr.msfr_group;
2402	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2403		return (EINVAL);
2404
2405	gsa->sin6.sin6_port = 0;	/* ignore port */
2406
2407	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2408		return (EADDRNOTAVAIL);
2409	ifp = ifnet_byindex(msfr.msfr_ifindex);
2410	if (ifp == NULL)
2411		return (EADDRNOTAVAIL);
2412	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2413
2414	/*
2415	 * Take the INP write lock.
2416	 * Check if this socket is a member of this group.
2417	 */
2418	imo = in6p_findmoptions(inp);
2419	idx = im6o_match_group(imo, ifp, &gsa->sa);
2420	if (idx == -1 || imo->im6o_mfilters == NULL) {
2421		error = EADDRNOTAVAIL;
2422		goto out_in6p_locked;
2423	}
2424	inm = imo->im6o_membership[idx];
2425	imf = &imo->im6o_mfilters[idx];
2426
2427	/*
2428	 * Begin state merge transaction at socket layer.
2429	 */
2430	INP_WLOCK_ASSERT(inp);
2431
2432	imf->im6f_st[1] = msfr.msfr_fmode;
2433
2434	/*
2435	 * Apply any new source filters, if present.
2436	 * Make a copy of the user-space source vector so
2437	 * that we may copy them with a single copyin. This
2438	 * allows us to deal with page faults up-front.
2439	 */
2440	if (msfr.msfr_nsrcs > 0) {
2441		struct in6_msource	*lims;
2442		struct sockaddr_in6	*psin;
2443		struct sockaddr_storage	*kss, *pkss;
2444		int			 i;
2445
2446		INP_WUNLOCK(inp);
2447
2448		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2449		    __func__, (unsigned long)msfr.msfr_nsrcs);
2450		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2451		    M_TEMP, M_WAITOK);
2452		error = copyin(msfr.msfr_srcs, kss,
2453		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2454		if (error) {
2455			free(kss, M_TEMP);
2456			return (error);
2457		}
2458
2459		INP_WLOCK(inp);
2460
2461		/*
2462		 * Mark all source filters as UNDEFINED at t1.
2463		 * Restore new group filter mode, as im6f_leave()
2464		 * will set it to INCLUDE.
2465		 */
2466		im6f_leave(imf);
2467		imf->im6f_st[1] = msfr.msfr_fmode;
2468
2469		/*
2470		 * Update socket layer filters at t1, lazy-allocating
2471		 * new entries. This saves a bunch of memory at the
2472		 * cost of one RB_FIND() per source entry; duplicate
2473		 * entries in the msfr_nsrcs vector are ignored.
2474		 * If we encounter an error, rollback transaction.
2475		 *
2476		 * XXX This too could be replaced with a set-symmetric
2477		 * difference like loop to avoid walking from root
2478		 * every time, as the key space is common.
2479		 */
2480		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2481			psin = (struct sockaddr_in6 *)pkss;
2482			if (psin->sin6_family != AF_INET6) {
2483				error = EAFNOSUPPORT;
2484				break;
2485			}
2486			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2487				error = EINVAL;
2488				break;
2489			}
2490			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2491				error = EINVAL;
2492				break;
2493			}
2494			/*
2495			 * TODO: Validate embedded scope ID in source
2496			 * list entry against passed-in ifp, if and only
2497			 * if source list filter entry is iface or node local.
2498			 */
2499			in6_clearscope(&psin->sin6_addr);
2500			error = im6f_get_source(imf, psin, &lims);
2501			if (error)
2502				break;
2503			lims->im6sl_st[1] = imf->im6f_st[1];
2504		}
2505		free(kss, M_TEMP);
2506	}
2507
2508	if (error)
2509		goto out_im6f_rollback;
2510
2511	INP_WLOCK_ASSERT(inp);
2512	IN6_MULTI_LOCK();
2513
2514	/*
2515	 * Begin state merge transaction at MLD layer.
2516	 */
2517	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2518	error = in6m_merge(inm, imf);
2519	if (error) {
2520		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2521		goto out_im6f_rollback;
2522	}
2523
2524	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2525	error = mld_change_state(inm, 0);
2526	if (error)
2527		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2528
2529	IN6_MULTI_UNLOCK();
2530
2531out_im6f_rollback:
2532	if (error)
2533		im6f_rollback(imf);
2534	else
2535		im6f_commit(imf);
2536
2537	im6f_reap(imf);
2538
2539out_in6p_locked:
2540	INP_WUNLOCK(inp);
2541	return (error);
2542}
2543
2544/*
2545 * Set the IP multicast options in response to user setsockopt().
2546 *
2547 * Many of the socket options handled in this function duplicate the
2548 * functionality of socket options in the regular unicast API. However,
2549 * it is not possible to merge the duplicate code, because the idempotence
2550 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2551 * the effects of these options must be treated as separate and distinct.
2552 *
2553 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2554 */
2555int
2556ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2557{
2558	struct ip6_moptions	*im6o;
2559	int			 error;
2560
2561	error = 0;
2562
2563	/*
2564	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2565	 * or is a divert socket, reject it.
2566	 */
2567	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2568	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2569	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2570		return (EOPNOTSUPP);
2571
2572	switch (sopt->sopt_name) {
2573	case IPV6_MULTICAST_IF:
2574		error = in6p_set_multicast_if(inp, sopt);
2575		break;
2576
2577	case IPV6_MULTICAST_HOPS: {
2578		int hlim;
2579
2580		if (sopt->sopt_valsize != sizeof(int)) {
2581			error = EINVAL;
2582			break;
2583		}
2584		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2585		if (error)
2586			break;
2587		if (hlim < -1 || hlim > 255) {
2588			error = EINVAL;
2589			break;
2590		} else if (hlim == -1) {
2591			hlim = V_ip6_defmcasthlim;
2592		}
2593		im6o = in6p_findmoptions(inp);
2594		im6o->im6o_multicast_hlim = hlim;
2595		INP_WUNLOCK(inp);
2596		break;
2597	}
2598
2599	case IPV6_MULTICAST_LOOP: {
2600		u_int loop;
2601
2602		/*
2603		 * Set the loopback flag for outgoing multicast packets.
2604		 * Must be zero or one.
2605		 */
2606		if (sopt->sopt_valsize != sizeof(u_int)) {
2607			error = EINVAL;
2608			break;
2609		}
2610		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2611		if (error)
2612			break;
2613		if (loop > 1) {
2614			error = EINVAL;
2615			break;
2616		}
2617		im6o = in6p_findmoptions(inp);
2618		im6o->im6o_multicast_loop = loop;
2619		INP_WUNLOCK(inp);
2620		break;
2621	}
2622
2623	case IPV6_JOIN_GROUP:
2624	case MCAST_JOIN_GROUP:
2625	case MCAST_JOIN_SOURCE_GROUP:
2626		error = in6p_join_group(inp, sopt);
2627		break;
2628
2629	case IPV6_LEAVE_GROUP:
2630	case MCAST_LEAVE_GROUP:
2631	case MCAST_LEAVE_SOURCE_GROUP:
2632		error = in6p_leave_group(inp, sopt);
2633		break;
2634
2635	case MCAST_BLOCK_SOURCE:
2636	case MCAST_UNBLOCK_SOURCE:
2637		error = in6p_block_unblock_source(inp, sopt);
2638		break;
2639
2640	case IPV6_MSFILTER:
2641		error = in6p_set_source_filters(inp, sopt);
2642		break;
2643
2644	default:
2645		error = EOPNOTSUPP;
2646		break;
2647	}
2648
2649	INP_UNLOCK_ASSERT(inp);
2650
2651	return (error);
2652}
2653
2654/*
2655 * Expose MLD's multicast filter mode and source list(s) to userland,
2656 * keyed by (ifindex, group).
2657 * The filter mode is written out as a uint32_t, followed by
2658 * 0..n of struct in6_addr.
2659 * For use by ifmcstat(8).
2660 * SMPng: NOTE: unlocked read of ifindex space.
2661 */
2662static int
2663sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2664{
2665	struct in6_addr			 mcaddr;
2666	struct in6_addr			 src;
2667	struct ifnet			*ifp;
2668	struct ifmultiaddr		*ifma;
2669	struct in6_multi		*inm;
2670	struct ip6_msource		*ims;
2671	int				*name;
2672	int				 retval;
2673	u_int				 namelen;
2674	uint32_t			 fmode, ifindex;
2675#ifdef KTR
2676	char				 ip6tbuf[INET6_ADDRSTRLEN];
2677#endif
2678
2679	name = (int *)arg1;
2680	namelen = arg2;
2681
2682	if (req->newptr != NULL)
2683		return (EPERM);
2684
2685	/* int: ifindex + 4 * 32 bits of IPv6 address */
2686	if (namelen != 5)
2687		return (EINVAL);
2688
2689	ifindex = name[0];
2690	if (ifindex <= 0 || ifindex > V_if_index) {
2691		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2692		    __func__, ifindex);
2693		return (ENOENT);
2694	}
2695
2696	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2697	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2698		CTR2(KTR_MLD, "%s: group %s is not multicast",
2699		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2700		return (EINVAL);
2701	}
2702
2703	ifp = ifnet_byindex(ifindex);
2704	if (ifp == NULL) {
2705		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2706		    __func__, ifindex);
2707		return (ENOENT);
2708	}
2709	/*
2710	 * Internal MLD lookups require that scope/zone ID is set.
2711	 */
2712	(void)in6_setscope(&mcaddr, ifp, NULL);
2713
2714	retval = sysctl_wire_old_buffer(req,
2715	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2716	if (retval)
2717		return (retval);
2718
2719	IN6_MULTI_LOCK();
2720
2721	IF_ADDR_RLOCK(ifp);
2722	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2723		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2724		    ifma->ifma_protospec == NULL)
2725			continue;
2726		inm = (struct in6_multi *)ifma->ifma_protospec;
2727		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2728			continue;
2729		fmode = inm->in6m_st[1].iss_fmode;
2730		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2731		if (retval != 0)
2732			break;
2733		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2734			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2735			/*
2736			 * Only copy-out sources which are in-mode.
2737			 */
2738			if (fmode != im6s_get_mode(inm, ims, 1)) {
2739				CTR1(KTR_MLD, "%s: skip non-in-mode",
2740				    __func__);
2741				continue;
2742			}
2743			src = ims->im6s_addr;
2744			retval = SYSCTL_OUT(req, &src,
2745			    sizeof(struct in6_addr));
2746			if (retval != 0)
2747				break;
2748		}
2749	}
2750	IF_ADDR_RUNLOCK(ifp);
2751
2752	IN6_MULTI_UNLOCK();
2753
2754	return (retval);
2755}
2756
2757#ifdef KTR
2758
2759static const char *in6m_modestrs[] = { "un", "in", "ex" };
2760
2761static const char *
2762in6m_mode_str(const int mode)
2763{
2764
2765	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2766		return (in6m_modestrs[mode]);
2767	return ("??");
2768}
2769
2770static const char *in6m_statestrs[] = {
2771	"not-member",
2772	"silent",
2773	"idle",
2774	"lazy",
2775	"sleeping",
2776	"awakening",
2777	"query-pending",
2778	"sg-query-pending",
2779	"leaving"
2780};
2781
2782static const char *
2783in6m_state_str(const int state)
2784{
2785
2786	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2787		return (in6m_statestrs[state]);
2788	return ("??");
2789}
2790
2791/*
2792 * Dump an in6_multi structure to the console.
2793 */
2794void
2795in6m_print(const struct in6_multi *inm)
2796{
2797	int t;
2798	char ip6tbuf[INET6_ADDRSTRLEN];
2799
2800	if ((ktr_mask & KTR_MLD) == 0)
2801		return;
2802
2803	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2804	printf("addr %s ifp %p(%s) ifma %p\n",
2805	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2806	    inm->in6m_ifp,
2807	    inm->in6m_ifp->if_xname,
2808	    inm->in6m_ifma);
2809	printf("timer %u state %s refcount %u scq.len %u\n",
2810	    inm->in6m_timer,
2811	    in6m_state_str(inm->in6m_state),
2812	    inm->in6m_refcount,
2813	    inm->in6m_scq.ifq_len);
2814	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2815	    inm->in6m_mli,
2816	    inm->in6m_nsrc,
2817	    inm->in6m_sctimer,
2818	    inm->in6m_scrv);
2819	for (t = 0; t < 2; t++) {
2820		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2821		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2822		    inm->in6m_st[t].iss_asm,
2823		    inm->in6m_st[t].iss_ex,
2824		    inm->in6m_st[t].iss_in,
2825		    inm->in6m_st[t].iss_rec);
2826	}
2827	printf("%s: --- end in6m %p ---\n", __func__, inm);
2828}
2829
2830#else /* !KTR */
2831
2832void
2833in6m_print(const struct in6_multi *inm)
2834{
2835
2836}
2837
2838#endif /* KTR */
2839