in6_mcast.c revision 259983
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 259983 2013-12-28 01:19:48Z dim $");
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);
134#ifdef KTR
135static int	in6m_is_ifp_detached(const struct in6_multi *);
136#endif
137static int	in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
138static void	in6m_purge(struct in6_multi *);
139static void	in6m_reap(struct in6_multi *);
140static struct ip6_moptions *
141		in6p_findmoptions(struct inpcb *);
142static int	in6p_get_source_filters(struct inpcb *, struct sockopt *);
143static int	in6p_join_group(struct inpcb *, struct sockopt *);
144static int	in6p_leave_group(struct inpcb *, struct sockopt *);
145static struct ifnet *
146		in6p_lookup_mcast_ifp(const struct inpcb *,
147		    const struct sockaddr_in6 *);
148static int	in6p_block_unblock_source(struct inpcb *, struct sockopt *);
149static int	in6p_set_multicast_if(struct inpcb *, struct sockopt *);
150static int	in6p_set_source_filters(struct inpcb *, struct sockopt *);
151static int	sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
152
153SYSCTL_DECL(_net_inet6_ip6);	/* XXX Not in any common header. */
154
155static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0,
156    "IPv6 multicast");
157
158static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
159SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
160    CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxgrpsrc, 0,
161    "Max source filters per group");
162TUNABLE_ULONG("net.inet6.ip6.mcast.maxgrpsrc", &in6_mcast_maxgrpsrc);
163
164static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
165SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
166    CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxsocksrc, 0,
167    "Max source filters per socket");
168TUNABLE_ULONG("net.inet6.ip6.mcast.maxsocksrc", &in6_mcast_maxsocksrc);
169
170/* TODO Virtualize this switch. */
171int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
172SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
173    &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
174TUNABLE_INT("net.inet6.ip6.mcast.loop", &in6_mcast_loop);
175
176static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
177    CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
178    "Per-interface stack-wide source filters");
179
180#ifdef KTR
181/*
182 * Inline function which wraps assertions for a valid ifp.
183 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
184 * is detached.
185 */
186static int __inline
187in6m_is_ifp_detached(const struct in6_multi *inm)
188{
189	struct ifnet *ifp;
190
191	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
192	ifp = inm->in6m_ifma->ifma_ifp;
193	if (ifp != NULL) {
194		/*
195		 * Sanity check that network-layer notion of ifp is the
196		 * same as that of link-layer.
197		 */
198		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
199	}
200
201	return (ifp == NULL);
202}
203#endif
204
205/*
206 * Initialize an in6_mfilter structure to a known state at t0, t1
207 * with an empty source filter list.
208 */
209static __inline void
210im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
211{
212	memset(imf, 0, sizeof(struct in6_mfilter));
213	RB_INIT(&imf->im6f_sources);
214	imf->im6f_st[0] = st0;
215	imf->im6f_st[1] = st1;
216}
217
218/*
219 * Resize the ip6_moptions vector to the next power-of-two minus 1.
220 * May be called with locks held; do not sleep.
221 */
222static int
223im6o_grow(struct ip6_moptions *imo)
224{
225	struct in6_multi	**nmships;
226	struct in6_multi	**omships;
227	struct in6_mfilter	 *nmfilters;
228	struct in6_mfilter	 *omfilters;
229	size_t			  idx;
230	size_t			  newmax;
231	size_t			  oldmax;
232
233	nmships = NULL;
234	nmfilters = NULL;
235	omships = imo->im6o_membership;
236	omfilters = imo->im6o_mfilters;
237	oldmax = imo->im6o_max_memberships;
238	newmax = ((oldmax + 1) * 2) - 1;
239
240	if (newmax <= IPV6_MAX_MEMBERSHIPS) {
241		nmships = (struct in6_multi **)realloc(omships,
242		    sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
243		nmfilters = (struct in6_mfilter *)realloc(omfilters,
244		    sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
245		    M_NOWAIT);
246		if (nmships != NULL && nmfilters != NULL) {
247			/* Initialize newly allocated source filter heads. */
248			for (idx = oldmax; idx < newmax; idx++) {
249				im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
250				    MCAST_EXCLUDE);
251			}
252			imo->im6o_max_memberships = newmax;
253			imo->im6o_membership = nmships;
254			imo->im6o_mfilters = nmfilters;
255		}
256	}
257
258	if (nmships == NULL || nmfilters == NULL) {
259		if (nmships != NULL)
260			free(nmships, M_IP6MOPTS);
261		if (nmfilters != NULL)
262			free(nmfilters, M_IN6MFILTER);
263		return (ETOOMANYREFS);
264	}
265
266	return (0);
267}
268
269/*
270 * Find an IPv6 multicast group entry for this ip6_moptions instance
271 * which matches the specified group, and optionally an interface.
272 * Return its index into the array, or -1 if not found.
273 */
274static size_t
275im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
276    const struct sockaddr *group)
277{
278	const struct sockaddr_in6 *gsin6;
279	struct in6_multi	**pinm;
280	int		  idx;
281	int		  nmships;
282
283	gsin6 = (const struct sockaddr_in6 *)group;
284
285	/* The im6o_membership array may be lazy allocated. */
286	if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
287		return (-1);
288
289	nmships = imo->im6o_num_memberships;
290	pinm = &imo->im6o_membership[0];
291	for (idx = 0; idx < nmships; idx++, pinm++) {
292		if (*pinm == NULL)
293			continue;
294		if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
295		    IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
296		    &gsin6->sin6_addr)) {
297			break;
298		}
299	}
300	if (idx >= nmships)
301		idx = -1;
302
303	return (idx);
304}
305
306/*
307 * Find an IPv6 multicast source entry for this imo which matches
308 * the given group index for this socket, and source address.
309 *
310 * XXX TODO: The scope ID, if present in src, is stripped before
311 * any comparison. We SHOULD enforce scope/zone checks where the source
312 * filter entry has a link scope.
313 *
314 * NOTE: This does not check if the entry is in-mode, merely if
315 * it exists, which may not be the desired behaviour.
316 */
317static struct in6_msource *
318im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
319    const struct sockaddr *src)
320{
321	struct ip6_msource	 find;
322	struct in6_mfilter	*imf;
323	struct ip6_msource	*ims;
324	const sockunion_t	*psa;
325
326	KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
327	KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
328	    ("%s: invalid index %d\n", __func__, (int)gidx));
329
330	/* The im6o_mfilters array may be lazy allocated. */
331	if (imo->im6o_mfilters == NULL)
332		return (NULL);
333	imf = &imo->im6o_mfilters[gidx];
334
335	psa = (const sockunion_t *)src;
336	find.im6s_addr = psa->sin6.sin6_addr;
337	in6_clearscope(&find.im6s_addr);		/* XXX */
338	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
339
340	return ((struct in6_msource *)ims);
341}
342
343/*
344 * Perform filtering for multicast datagrams on a socket by group and source.
345 *
346 * Returns 0 if a datagram should be allowed through, or various error codes
347 * if the socket was not a member of the group, or the source was muted, etc.
348 */
349int
350im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
351    const struct sockaddr *group, const struct sockaddr *src)
352{
353	size_t gidx;
354	struct in6_msource *ims;
355	int mode;
356
357	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
358
359	gidx = im6o_match_group(imo, ifp, group);
360	if (gidx == -1)
361		return (MCAST_NOTGMEMBER);
362
363	/*
364	 * Check if the source was included in an (S,G) join.
365	 * Allow reception on exclusive memberships by default,
366	 * reject reception on inclusive memberships by default.
367	 * Exclude source only if an in-mode exclude filter exists.
368	 * Include source only if an in-mode include filter exists.
369	 * NOTE: We are comparing group state here at MLD t1 (now)
370	 * with socket-layer t0 (since last downcall).
371	 */
372	mode = imo->im6o_mfilters[gidx].im6f_st[1];
373	ims = im6o_match_source(imo, gidx, src);
374
375	if ((ims == NULL && mode == MCAST_INCLUDE) ||
376	    (ims != NULL && ims->im6sl_st[0] != mode))
377		return (MCAST_NOTSMEMBER);
378
379	return (MCAST_PASS);
380}
381
382/*
383 * Find and return a reference to an in6_multi record for (ifp, group),
384 * and bump its reference count.
385 * If one does not exist, try to allocate it, and update link-layer multicast
386 * filters on ifp to listen for group.
387 * Assumes the IN6_MULTI lock is held across the call.
388 * Return 0 if successful, otherwise return an appropriate error code.
389 */
390static int
391in6_mc_get(struct ifnet *ifp, const struct in6_addr *group,
392    struct in6_multi **pinm)
393{
394	struct sockaddr_in6	 gsin6;
395	struct ifmultiaddr	*ifma;
396	struct in6_multi	*inm;
397	int			 error;
398
399	error = 0;
400
401	/*
402	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
403	 * if_addmulti() takes this mutex itself, so we must drop and
404	 * re-acquire around the call.
405	 */
406	IN6_MULTI_LOCK_ASSERT();
407	IF_ADDR_WLOCK(ifp);
408
409	inm = in6m_lookup_locked(ifp, group);
410	if (inm != NULL) {
411		/*
412		 * If we already joined this group, just bump the
413		 * refcount and return it.
414		 */
415		KASSERT(inm->in6m_refcount >= 1,
416		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
417		++inm->in6m_refcount;
418		*pinm = inm;
419		goto out_locked;
420	}
421
422	memset(&gsin6, 0, sizeof(gsin6));
423	gsin6.sin6_family = AF_INET6;
424	gsin6.sin6_len = sizeof(struct sockaddr_in6);
425	gsin6.sin6_addr = *group;
426
427	/*
428	 * Check if a link-layer group is already associated
429	 * with this network-layer group on the given ifnet.
430	 */
431	IF_ADDR_WUNLOCK(ifp);
432	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
433	if (error != 0)
434		return (error);
435	IF_ADDR_WLOCK(ifp);
436
437	/*
438	 * If something other than netinet6 is occupying the link-layer
439	 * group, print a meaningful error message and back out of
440	 * the allocation.
441	 * Otherwise, bump the refcount on the existing network-layer
442	 * group association and return it.
443	 */
444	if (ifma->ifma_protospec != NULL) {
445		inm = (struct in6_multi *)ifma->ifma_protospec;
446#ifdef INVARIANTS
447		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
448		    __func__));
449		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
450		    ("%s: ifma not AF_INET6", __func__));
451		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
452		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
453		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
454			panic("%s: ifma %p is inconsistent with %p (%p)",
455			    __func__, ifma, inm, group);
456#endif
457		++inm->in6m_refcount;
458		*pinm = inm;
459		goto out_locked;
460	}
461
462	IF_ADDR_WLOCK_ASSERT(ifp);
463
464	/*
465	 * A new in6_multi record is needed; allocate and initialize it.
466	 * We DO NOT perform an MLD join as the in6_ layer may need to
467	 * push an initial source list down to MLD to support SSM.
468	 *
469	 * The initial source filter state is INCLUDE, {} as per the RFC.
470	 * Pending state-changes per group are subject to a bounds check.
471	 */
472	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
473	if (inm == NULL) {
474		if_delmulti_ifma(ifma);
475		error = ENOMEM;
476		goto out_locked;
477	}
478	inm->in6m_addr = *group;
479	inm->in6m_ifp = ifp;
480	inm->in6m_mli = MLD_IFINFO(ifp);
481	inm->in6m_ifma = ifma;
482	inm->in6m_refcount = 1;
483	inm->in6m_state = MLD_NOT_MEMBER;
484	IFQ_SET_MAXLEN(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
485
486	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
487	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
488	RB_INIT(&inm->in6m_srcs);
489
490	ifma->ifma_protospec = inm;
491	*pinm = inm;
492
493out_locked:
494	IF_ADDR_WUNLOCK(ifp);
495	return (error);
496}
497
498/*
499 * Drop a reference to an in6_multi record.
500 *
501 * If the refcount drops to 0, free the in6_multi record and
502 * delete the underlying link-layer membership.
503 */
504void
505in6m_release_locked(struct in6_multi *inm)
506{
507	struct ifmultiaddr *ifma;
508
509	IN6_MULTI_LOCK_ASSERT();
510
511	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
512
513	if (--inm->in6m_refcount > 0) {
514		CTR2(KTR_MLD, "%s: refcount is now %d", __func__,
515		    inm->in6m_refcount);
516		return;
517	}
518
519	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
520
521	ifma = inm->in6m_ifma;
522
523	/* XXX this access is not covered by IF_ADDR_LOCK */
524	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
525	KASSERT(ifma->ifma_protospec == inm,
526	    ("%s: ifma_protospec != inm", __func__));
527	ifma->ifma_protospec = NULL;
528
529	in6m_purge(inm);
530
531	free(inm, M_IP6MADDR);
532
533	if_delmulti_ifma(ifma);
534}
535
536/*
537 * Clear recorded source entries for a group.
538 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
539 * FIXME: Should reap.
540 */
541void
542in6m_clear_recorded(struct in6_multi *inm)
543{
544	struct ip6_msource	*ims;
545
546	IN6_MULTI_LOCK_ASSERT();
547
548	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
549		if (ims->im6s_stp) {
550			ims->im6s_stp = 0;
551			--inm->in6m_st[1].iss_rec;
552		}
553	}
554	KASSERT(inm->in6m_st[1].iss_rec == 0,
555	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
556}
557
558/*
559 * Record a source as pending for a Source-Group MLDv2 query.
560 * This lives here as it modifies the shared tree.
561 *
562 * inm is the group descriptor.
563 * naddr is the address of the source to record in network-byte order.
564 *
565 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
566 * lazy-allocate a source node in response to an SG query.
567 * Otherwise, no allocation is performed. This saves some memory
568 * with the trade-off that the source will not be reported to the
569 * router if joined in the window between the query response and
570 * the group actually being joined on the local host.
571 *
572 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
573 * This turns off the allocation of a recorded source entry if
574 * the group has not been joined.
575 *
576 * Return 0 if the source didn't exist or was already marked as recorded.
577 * Return 1 if the source was marked as recorded by this function.
578 * Return <0 if any error occured (negated errno code).
579 */
580int
581in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
582{
583	struct ip6_msource	 find;
584	struct ip6_msource	*ims, *nims;
585
586	IN6_MULTI_LOCK_ASSERT();
587
588	find.im6s_addr = *addr;
589	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
590	if (ims && ims->im6s_stp)
591		return (0);
592	if (ims == NULL) {
593		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
594			return (-ENOSPC);
595		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
596		    M_NOWAIT | M_ZERO);
597		if (nims == NULL)
598			return (-ENOMEM);
599		nims->im6s_addr = find.im6s_addr;
600		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
601		++inm->in6m_nsrc;
602		ims = nims;
603	}
604
605	/*
606	 * Mark the source as recorded and update the recorded
607	 * source count.
608	 */
609	++ims->im6s_stp;
610	++inm->in6m_st[1].iss_rec;
611
612	return (1);
613}
614
615/*
616 * Return a pointer to an in6_msource owned by an in6_mfilter,
617 * given its source address.
618 * Lazy-allocate if needed. If this is a new entry its filter state is
619 * undefined at t0.
620 *
621 * imf is the filter set being modified.
622 * addr is the source address.
623 *
624 * SMPng: May be called with locks held; malloc must not block.
625 */
626static int
627im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
628    struct in6_msource **plims)
629{
630	struct ip6_msource	 find;
631	struct ip6_msource	*ims, *nims;
632	struct in6_msource	*lims;
633	int			 error;
634
635	error = 0;
636	ims = NULL;
637	lims = NULL;
638
639	find.im6s_addr = psin->sin6_addr;
640	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
641	lims = (struct in6_msource *)ims;
642	if (lims == NULL) {
643		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
644			return (ENOSPC);
645		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
646		    M_NOWAIT | M_ZERO);
647		if (nims == NULL)
648			return (ENOMEM);
649		lims = (struct in6_msource *)nims;
650		lims->im6s_addr = find.im6s_addr;
651		lims->im6sl_st[0] = MCAST_UNDEFINED;
652		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
653		++imf->im6f_nsrc;
654	}
655
656	*plims = lims;
657
658	return (error);
659}
660
661/*
662 * Graft a source entry into an existing socket-layer filter set,
663 * maintaining any required invariants and checking allocations.
664 *
665 * The source is marked as being in the new filter mode at t1.
666 *
667 * Return the pointer to the new node, otherwise return NULL.
668 */
669static struct in6_msource *
670im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
671    const struct sockaddr_in6 *psin)
672{
673	struct ip6_msource	*nims;
674	struct in6_msource	*lims;
675
676	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
677	    M_NOWAIT | M_ZERO);
678	if (nims == NULL)
679		return (NULL);
680	lims = (struct in6_msource *)nims;
681	lims->im6s_addr = psin->sin6_addr;
682	lims->im6sl_st[0] = MCAST_UNDEFINED;
683	lims->im6sl_st[1] = st1;
684	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
685	++imf->im6f_nsrc;
686
687	return (lims);
688}
689
690/*
691 * Prune a source entry from an existing socket-layer filter set,
692 * maintaining any required invariants and checking allocations.
693 *
694 * The source is marked as being left at t1, it is not freed.
695 *
696 * Return 0 if no error occurred, otherwise return an errno value.
697 */
698static int
699im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
700{
701	struct ip6_msource	 find;
702	struct ip6_msource	*ims;
703	struct in6_msource	*lims;
704
705	find.im6s_addr = psin->sin6_addr;
706	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
707	if (ims == NULL)
708		return (ENOENT);
709	lims = (struct in6_msource *)ims;
710	lims->im6sl_st[1] = MCAST_UNDEFINED;
711	return (0);
712}
713
714/*
715 * Revert socket-layer filter set deltas at t1 to t0 state.
716 */
717static void
718im6f_rollback(struct in6_mfilter *imf)
719{
720	struct ip6_msource	*ims, *tims;
721	struct in6_msource	*lims;
722
723	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
724		lims = (struct in6_msource *)ims;
725		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
726			/* no change at t1 */
727			continue;
728		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
729			/* revert change to existing source at t1 */
730			lims->im6sl_st[1] = lims->im6sl_st[0];
731		} else {
732			/* revert source added t1 */
733			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
734			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
735			free(ims, M_IN6MFILTER);
736			imf->im6f_nsrc--;
737		}
738	}
739	imf->im6f_st[1] = imf->im6f_st[0];
740}
741
742/*
743 * Mark socket-layer filter set as INCLUDE {} at t1.
744 */
745static void
746im6f_leave(struct in6_mfilter *imf)
747{
748	struct ip6_msource	*ims;
749	struct in6_msource	*lims;
750
751	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
752		lims = (struct in6_msource *)ims;
753		lims->im6sl_st[1] = MCAST_UNDEFINED;
754	}
755	imf->im6f_st[1] = MCAST_INCLUDE;
756}
757
758/*
759 * Mark socket-layer filter set deltas as committed.
760 */
761static void
762im6f_commit(struct in6_mfilter *imf)
763{
764	struct ip6_msource	*ims;
765	struct in6_msource	*lims;
766
767	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
768		lims = (struct in6_msource *)ims;
769		lims->im6sl_st[0] = lims->im6sl_st[1];
770	}
771	imf->im6f_st[0] = imf->im6f_st[1];
772}
773
774/*
775 * Reap unreferenced sources from socket-layer filter set.
776 */
777static void
778im6f_reap(struct in6_mfilter *imf)
779{
780	struct ip6_msource	*ims, *tims;
781	struct in6_msource	*lims;
782
783	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
784		lims = (struct in6_msource *)ims;
785		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
786		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
787			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
788			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
789			free(ims, M_IN6MFILTER);
790			imf->im6f_nsrc--;
791		}
792	}
793}
794
795/*
796 * Purge socket-layer filter set.
797 */
798static void
799im6f_purge(struct in6_mfilter *imf)
800{
801	struct ip6_msource	*ims, *tims;
802
803	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
804		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
805		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
806		free(ims, M_IN6MFILTER);
807		imf->im6f_nsrc--;
808	}
809	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
810	KASSERT(RB_EMPTY(&imf->im6f_sources),
811	    ("%s: im6f_sources not empty", __func__));
812}
813
814/*
815 * Look up a source filter entry for a multicast group.
816 *
817 * inm is the group descriptor to work with.
818 * addr is the IPv6 address to look up.
819 * noalloc may be non-zero to suppress allocation of sources.
820 * *pims will be set to the address of the retrieved or allocated source.
821 *
822 * SMPng: NOTE: may be called with locks held.
823 * Return 0 if successful, otherwise return a non-zero error code.
824 */
825static int
826in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
827    const int noalloc, struct ip6_msource **pims)
828{
829	struct ip6_msource	 find;
830	struct ip6_msource	*ims, *nims;
831#ifdef KTR
832	char			 ip6tbuf[INET6_ADDRSTRLEN];
833#endif
834
835	find.im6s_addr = *addr;
836	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
837	if (ims == NULL && !noalloc) {
838		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
839			return (ENOSPC);
840		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
841		    M_NOWAIT | M_ZERO);
842		if (nims == NULL)
843			return (ENOMEM);
844		nims->im6s_addr = *addr;
845		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
846		++inm->in6m_nsrc;
847		ims = nims;
848		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
849		    ip6_sprintf(ip6tbuf, addr), ims);
850	}
851
852	*pims = ims;
853	return (0);
854}
855
856/*
857 * Merge socket-layer source into MLD-layer source.
858 * If rollback is non-zero, perform the inverse of the merge.
859 */
860static void
861im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
862    const int rollback)
863{
864	int n = rollback ? -1 : 1;
865#ifdef KTR
866	char ip6tbuf[INET6_ADDRSTRLEN];
867
868	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
869#endif
870
871	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
872		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
873		ims->im6s_st[1].ex -= n;
874	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
875		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
876		ims->im6s_st[1].in -= n;
877	}
878
879	if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
880		CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
881		ims->im6s_st[1].ex += n;
882	} else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
883		CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
884		ims->im6s_st[1].in += n;
885	}
886}
887
888/*
889 * Atomically update the global in6_multi state, when a membership's
890 * filter list is being updated in any way.
891 *
892 * imf is the per-inpcb-membership group filter pointer.
893 * A fake imf may be passed for in-kernel consumers.
894 *
895 * XXX This is a candidate for a set-symmetric-difference style loop
896 * which would eliminate the repeated lookup from root of ims nodes,
897 * as they share the same key space.
898 *
899 * If any error occurred this function will back out of refcounts
900 * and return a non-zero value.
901 */
902static int
903in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
904{
905	struct ip6_msource	*ims, *nims;
906	struct in6_msource	*lims;
907	int			 schanged, error;
908	int			 nsrc0, nsrc1;
909
910	schanged = 0;
911	error = 0;
912	nsrc1 = nsrc0 = 0;
913
914	/*
915	 * Update the source filters first, as this may fail.
916	 * Maintain count of in-mode filters at t0, t1. These are
917	 * used to work out if we transition into ASM mode or not.
918	 * Maintain a count of source filters whose state was
919	 * actually modified by this operation.
920	 */
921	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
922		lims = (struct in6_msource *)ims;
923		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
924		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
925		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
926		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
927		++schanged;
928		if (error)
929			break;
930		im6s_merge(nims, lims, 0);
931	}
932	if (error) {
933		struct ip6_msource *bims;
934
935		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
936			lims = (struct in6_msource *)ims;
937			if (lims->im6sl_st[0] == lims->im6sl_st[1])
938				continue;
939			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
940			if (bims == NULL)
941				continue;
942			im6s_merge(bims, lims, 1);
943		}
944		goto out_reap;
945	}
946
947	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
948	    __func__, nsrc0, nsrc1);
949
950	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
951	if (imf->im6f_st[0] == imf->im6f_st[1] &&
952	    imf->im6f_st[1] == MCAST_INCLUDE) {
953		if (nsrc1 == 0) {
954			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
955			--inm->in6m_st[1].iss_in;
956		}
957	}
958
959	/* Handle filter mode transition on socket. */
960	if (imf->im6f_st[0] != imf->im6f_st[1]) {
961		CTR3(KTR_MLD, "%s: imf transition %d to %d",
962		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
963
964		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
965			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
966			--inm->in6m_st[1].iss_ex;
967		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
968			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
969			--inm->in6m_st[1].iss_in;
970		}
971
972		if (imf->im6f_st[1] == MCAST_EXCLUDE) {
973			CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
974			inm->in6m_st[1].iss_ex++;
975		} else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
976			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
977			inm->in6m_st[1].iss_in++;
978		}
979	}
980
981	/*
982	 * Track inm filter state in terms of listener counts.
983	 * If there are any exclusive listeners, stack-wide
984	 * membership is exclusive.
985	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
986	 * If no listeners remain, state is undefined at t1,
987	 * and the MLD lifecycle for this group should finish.
988	 */
989	if (inm->in6m_st[1].iss_ex > 0) {
990		CTR1(KTR_MLD, "%s: transition to EX", __func__);
991		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
992	} else if (inm->in6m_st[1].iss_in > 0) {
993		CTR1(KTR_MLD, "%s: transition to IN", __func__);
994		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
995	} else {
996		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
997		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
998	}
999
1000	/* Decrement ASM listener count on transition out of ASM mode. */
1001	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1002		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1003		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1004			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1005			--inm->in6m_st[1].iss_asm;
1006	}
1007
1008	/* Increment ASM listener count on transition to ASM mode. */
1009	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1010		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1011		inm->in6m_st[1].iss_asm++;
1012	}
1013
1014	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1015	in6m_print(inm);
1016
1017out_reap:
1018	if (schanged > 0) {
1019		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1020		in6m_reap(inm);
1021	}
1022	return (error);
1023}
1024
1025/*
1026 * Mark an in6_multi's filter set deltas as committed.
1027 * Called by MLD after a state change has been enqueued.
1028 */
1029void
1030in6m_commit(struct in6_multi *inm)
1031{
1032	struct ip6_msource	*ims;
1033
1034	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1035	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1036	in6m_print(inm);
1037
1038	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1039		ims->im6s_st[0] = ims->im6s_st[1];
1040	}
1041	inm->in6m_st[0] = inm->in6m_st[1];
1042}
1043
1044/*
1045 * Reap unreferenced nodes from an in6_multi's filter set.
1046 */
1047static void
1048in6m_reap(struct in6_multi *inm)
1049{
1050	struct ip6_msource	*ims, *tims;
1051
1052	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1053		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1054		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1055		    ims->im6s_stp != 0)
1056			continue;
1057		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1058		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1059		free(ims, M_IP6MSOURCE);
1060		inm->in6m_nsrc--;
1061	}
1062}
1063
1064/*
1065 * Purge all source nodes from an in6_multi's filter set.
1066 */
1067static void
1068in6m_purge(struct in6_multi *inm)
1069{
1070	struct ip6_msource	*ims, *tims;
1071
1072	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1073		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1074		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1075		free(ims, M_IP6MSOURCE);
1076		inm->in6m_nsrc--;
1077	}
1078}
1079
1080/*
1081 * Join a multicast address w/o sources.
1082 * KAME compatibility entry point.
1083 *
1084 * SMPng: Assume no mc locks held by caller.
1085 */
1086struct in6_multi_mship *
1087in6_joingroup(struct ifnet *ifp, struct in6_addr *mcaddr,
1088    int *errorp, int delay)
1089{
1090	struct in6_multi_mship *imm;
1091	int error;
1092
1093	imm = malloc(sizeof(*imm), M_IP6MADDR, M_NOWAIT);
1094	if (imm == NULL) {
1095		*errorp = ENOBUFS;
1096		return (NULL);
1097	}
1098
1099	delay = (delay * PR_FASTHZ) / hz;
1100
1101	error = in6_mc_join(ifp, mcaddr, NULL, &imm->i6mm_maddr, delay);
1102	if (error) {
1103		*errorp = error;
1104		free(imm, M_IP6MADDR);
1105		return (NULL);
1106	}
1107
1108	return (imm);
1109}
1110
1111/*
1112 * Leave a multicast address w/o sources.
1113 * KAME compatibility entry point.
1114 *
1115 * SMPng: Assume no mc locks held by caller.
1116 */
1117int
1118in6_leavegroup(struct in6_multi_mship *imm)
1119{
1120
1121	if (imm->i6mm_maddr != NULL)
1122		in6_mc_leave(imm->i6mm_maddr, NULL);
1123	free(imm,  M_IP6MADDR);
1124	return 0;
1125}
1126
1127/*
1128 * Join a multicast group; unlocked entry point.
1129 *
1130 * SMPng: XXX: in6_mc_join() is called from in6_control() when upper
1131 * locks are not held. Fortunately, ifp is unlikely to have been detached
1132 * at this point, so we assume it's OK to recurse.
1133 */
1134int
1135in6_mc_join(struct ifnet *ifp, const struct in6_addr *mcaddr,
1136    /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1137    const int delay)
1138{
1139	int error;
1140
1141	IN6_MULTI_LOCK();
1142	error = in6_mc_join_locked(ifp, mcaddr, imf, pinm, delay);
1143	IN6_MULTI_UNLOCK();
1144
1145	return (error);
1146}
1147
1148/*
1149 * Join a multicast group; real entry point.
1150 *
1151 * Only preserves atomicity at inm level.
1152 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1153 *
1154 * If the MLD downcall fails, the group is not joined, and an error
1155 * code is returned.
1156 */
1157int
1158in6_mc_join_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1159    /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1160    const int delay)
1161{
1162	struct in6_mfilter	 timf;
1163	struct in6_multi	*inm;
1164	int			 error;
1165#ifdef KTR
1166	char			 ip6tbuf[INET6_ADDRSTRLEN];
1167#endif
1168
1169#ifdef INVARIANTS
1170	/*
1171	 * Sanity: Check scope zone ID was set for ifp, if and
1172	 * only if group is scoped to an interface.
1173	 */
1174	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1175	    ("%s: not a multicast address", __func__));
1176	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1177	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1178		KASSERT(mcaddr->s6_addr16[1] != 0,
1179		    ("%s: scope zone ID not set", __func__));
1180	}
1181#endif
1182
1183	IN6_MULTI_LOCK_ASSERT();
1184
1185	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1186	    ip6_sprintf(ip6tbuf, mcaddr), ifp, ifp->if_xname);
1187
1188	error = 0;
1189	inm = NULL;
1190
1191	/*
1192	 * If no imf was specified (i.e. kernel consumer),
1193	 * fake one up and assume it is an ASM join.
1194	 */
1195	if (imf == NULL) {
1196		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1197		imf = &timf;
1198	}
1199
1200	error = in6_mc_get(ifp, mcaddr, &inm);
1201	if (error) {
1202		CTR1(KTR_MLD, "%s: in6_mc_get() failure", __func__);
1203		return (error);
1204	}
1205
1206	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1207	error = in6m_merge(inm, imf);
1208	if (error) {
1209		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1210		goto out_in6m_release;
1211	}
1212
1213	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1214	error = mld_change_state(inm, delay);
1215	if (error) {
1216		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1217		goto out_in6m_release;
1218	}
1219
1220out_in6m_release:
1221	if (error) {
1222		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1223		in6m_release_locked(inm);
1224	} else {
1225		*pinm = inm;
1226	}
1227
1228	return (error);
1229}
1230
1231/*
1232 * Leave a multicast group; unlocked entry point.
1233 */
1234int
1235in6_mc_leave(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1236{
1237	struct ifnet *ifp;
1238	int error;
1239
1240	ifp = inm->in6m_ifp;
1241
1242	IN6_MULTI_LOCK();
1243	error = in6_mc_leave_locked(inm, imf);
1244	IN6_MULTI_UNLOCK();
1245
1246	return (error);
1247}
1248
1249/*
1250 * Leave a multicast group; real entry point.
1251 * All source filters will be expunged.
1252 *
1253 * Only preserves atomicity at inm level.
1254 *
1255 * Holding the write lock for the INP which contains imf
1256 * is highly advisable. We can't assert for it as imf does not
1257 * contain a back-pointer to the owning inp.
1258 *
1259 * Note: This is not the same as in6m_release(*) as this function also
1260 * makes a state change downcall into MLD.
1261 */
1262int
1263in6_mc_leave_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1264{
1265	struct in6_mfilter	 timf;
1266	int			 error;
1267#ifdef KTR
1268	char			 ip6tbuf[INET6_ADDRSTRLEN];
1269#endif
1270
1271	error = 0;
1272
1273	IN6_MULTI_LOCK_ASSERT();
1274
1275	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1276	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1277	    (in6m_is_ifp_detached(inm) ? "null" : inm->in6m_ifp->if_xname),
1278	    imf);
1279
1280	/*
1281	 * If no imf was specified (i.e. kernel consumer),
1282	 * fake one up and assume it is an ASM join.
1283	 */
1284	if (imf == NULL) {
1285		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1286		imf = &timf;
1287	}
1288
1289	/*
1290	 * Begin state merge transaction at MLD layer.
1291	 *
1292	 * As this particular invocation should not cause any memory
1293	 * to be allocated, and there is no opportunity to roll back
1294	 * the transaction, it MUST NOT fail.
1295	 */
1296	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1297	error = in6m_merge(inm, imf);
1298	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1299
1300	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1301	error = mld_change_state(inm, 0);
1302	if (error)
1303		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1304
1305	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1306	in6m_release_locked(inm);
1307
1308	return (error);
1309}
1310
1311/*
1312 * Block or unblock an ASM multicast source on an inpcb.
1313 * This implements the delta-based API described in RFC 3678.
1314 *
1315 * The delta-based API applies only to exclusive-mode memberships.
1316 * An MLD downcall will be performed.
1317 *
1318 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1319 *
1320 * Return 0 if successful, otherwise return an appropriate error code.
1321 */
1322static int
1323in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1324{
1325	struct group_source_req		 gsr;
1326	sockunion_t			*gsa, *ssa;
1327	struct ifnet			*ifp;
1328	struct in6_mfilter		*imf;
1329	struct ip6_moptions		*imo;
1330	struct in6_msource		*ims;
1331	struct in6_multi			*inm;
1332	size_t				 idx;
1333	uint16_t			 fmode;
1334	int				 error, doblock;
1335#ifdef KTR
1336	char				 ip6tbuf[INET6_ADDRSTRLEN];
1337#endif
1338
1339	ifp = NULL;
1340	error = 0;
1341	doblock = 0;
1342
1343	memset(&gsr, 0, sizeof(struct group_source_req));
1344	gsa = (sockunion_t *)&gsr.gsr_group;
1345	ssa = (sockunion_t *)&gsr.gsr_source;
1346
1347	switch (sopt->sopt_name) {
1348	case MCAST_BLOCK_SOURCE:
1349	case MCAST_UNBLOCK_SOURCE:
1350		error = sooptcopyin(sopt, &gsr,
1351		    sizeof(struct group_source_req),
1352		    sizeof(struct group_source_req));
1353		if (error)
1354			return (error);
1355
1356		if (gsa->sin6.sin6_family != AF_INET6 ||
1357		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1358			return (EINVAL);
1359
1360		if (ssa->sin6.sin6_family != AF_INET6 ||
1361		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1362			return (EINVAL);
1363
1364		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1365			return (EADDRNOTAVAIL);
1366
1367		ifp = ifnet_byindex(gsr.gsr_interface);
1368
1369		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1370			doblock = 1;
1371		break;
1372
1373	default:
1374		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1375		    __func__, sopt->sopt_name);
1376		return (EOPNOTSUPP);
1377		break;
1378	}
1379
1380	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1381		return (EINVAL);
1382
1383	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1384
1385	/*
1386	 * Check if we are actually a member of this group.
1387	 */
1388	imo = in6p_findmoptions(inp);
1389	idx = im6o_match_group(imo, ifp, &gsa->sa);
1390	if (idx == -1 || imo->im6o_mfilters == NULL) {
1391		error = EADDRNOTAVAIL;
1392		goto out_in6p_locked;
1393	}
1394
1395	KASSERT(imo->im6o_mfilters != NULL,
1396	    ("%s: im6o_mfilters not allocated", __func__));
1397	imf = &imo->im6o_mfilters[idx];
1398	inm = imo->im6o_membership[idx];
1399
1400	/*
1401	 * Attempting to use the delta-based API on an
1402	 * non exclusive-mode membership is an error.
1403	 */
1404	fmode = imf->im6f_st[0];
1405	if (fmode != MCAST_EXCLUDE) {
1406		error = EINVAL;
1407		goto out_in6p_locked;
1408	}
1409
1410	/*
1411	 * Deal with error cases up-front:
1412	 *  Asked to block, but already blocked; or
1413	 *  Asked to unblock, but nothing to unblock.
1414	 * If adding a new block entry, allocate it.
1415	 */
1416	ims = im6o_match_source(imo, idx, &ssa->sa);
1417	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1418		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1419		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1420		    doblock ? "" : "not ");
1421		error = EADDRNOTAVAIL;
1422		goto out_in6p_locked;
1423	}
1424
1425	INP_WLOCK_ASSERT(inp);
1426
1427	/*
1428	 * Begin state merge transaction at socket layer.
1429	 */
1430	if (doblock) {
1431		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1432		ims = im6f_graft(imf, fmode, &ssa->sin6);
1433		if (ims == NULL)
1434			error = ENOMEM;
1435	} else {
1436		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1437		error = im6f_prune(imf, &ssa->sin6);
1438	}
1439
1440	if (error) {
1441		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1442		goto out_im6f_rollback;
1443	}
1444
1445	/*
1446	 * Begin state merge transaction at MLD layer.
1447	 */
1448	IN6_MULTI_LOCK();
1449
1450	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1451	error = in6m_merge(inm, imf);
1452	if (error) {
1453		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1454		goto out_im6f_rollback;
1455	}
1456
1457	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1458	error = mld_change_state(inm, 0);
1459	if (error)
1460		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1461
1462	IN6_MULTI_UNLOCK();
1463
1464out_im6f_rollback:
1465	if (error)
1466		im6f_rollback(imf);
1467	else
1468		im6f_commit(imf);
1469
1470	im6f_reap(imf);
1471
1472out_in6p_locked:
1473	INP_WUNLOCK(inp);
1474	return (error);
1475}
1476
1477/*
1478 * Given an inpcb, return its multicast options structure pointer.  Accepts
1479 * an unlocked inpcb pointer, but will return it locked.  May sleep.
1480 *
1481 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1482 * SMPng: NOTE: Returns with the INP write lock held.
1483 */
1484static struct ip6_moptions *
1485in6p_findmoptions(struct inpcb *inp)
1486{
1487	struct ip6_moptions	 *imo;
1488	struct in6_multi		**immp;
1489	struct in6_mfilter	 *imfp;
1490	size_t			  idx;
1491
1492	INP_WLOCK(inp);
1493	if (inp->in6p_moptions != NULL)
1494		return (inp->in6p_moptions);
1495
1496	INP_WUNLOCK(inp);
1497
1498	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1499	immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1500	    M_WAITOK | M_ZERO);
1501	imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1502	    M_IN6MFILTER, M_WAITOK);
1503
1504	imo->im6o_multicast_ifp = NULL;
1505	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1506	imo->im6o_multicast_loop = in6_mcast_loop;
1507	imo->im6o_num_memberships = 0;
1508	imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1509	imo->im6o_membership = immp;
1510
1511	/* Initialize per-group source filters. */
1512	for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1513		im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1514	imo->im6o_mfilters = imfp;
1515
1516	INP_WLOCK(inp);
1517	if (inp->in6p_moptions != NULL) {
1518		free(imfp, M_IN6MFILTER);
1519		free(immp, M_IP6MOPTS);
1520		free(imo, M_IP6MOPTS);
1521		return (inp->in6p_moptions);
1522	}
1523	inp->in6p_moptions = imo;
1524	return (imo);
1525}
1526
1527/*
1528 * Discard the IPv6 multicast options (and source filters).
1529 *
1530 * SMPng: NOTE: assumes INP write lock is held.
1531 */
1532void
1533ip6_freemoptions(struct ip6_moptions *imo)
1534{
1535	struct in6_mfilter	*imf;
1536	size_t			 idx, nmships;
1537
1538	KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1539
1540	nmships = imo->im6o_num_memberships;
1541	for (idx = 0; idx < nmships; ++idx) {
1542		imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1543		if (imf)
1544			im6f_leave(imf);
1545		/* XXX this will thrash the lock(s) */
1546		(void)in6_mc_leave(imo->im6o_membership[idx], imf);
1547		if (imf)
1548			im6f_purge(imf);
1549	}
1550
1551	if (imo->im6o_mfilters)
1552		free(imo->im6o_mfilters, M_IN6MFILTER);
1553	free(imo->im6o_membership, M_IP6MOPTS);
1554	free(imo, M_IP6MOPTS);
1555}
1556
1557/*
1558 * Atomically get source filters on a socket for an IPv6 multicast group.
1559 * Called with INP lock held; returns with lock released.
1560 */
1561static int
1562in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1563{
1564	struct __msfilterreq	 msfr;
1565	sockunion_t		*gsa;
1566	struct ifnet		*ifp;
1567	struct ip6_moptions	*imo;
1568	struct in6_mfilter	*imf;
1569	struct ip6_msource	*ims;
1570	struct in6_msource	*lims;
1571	struct sockaddr_in6	*psin;
1572	struct sockaddr_storage	*ptss;
1573	struct sockaddr_storage	*tss;
1574	int			 error;
1575	size_t			 idx, nsrcs, ncsrcs;
1576
1577	INP_WLOCK_ASSERT(inp);
1578
1579	imo = inp->in6p_moptions;
1580	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1581
1582	INP_WUNLOCK(inp);
1583
1584	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1585	    sizeof(struct __msfilterreq));
1586	if (error)
1587		return (error);
1588
1589	if (msfr.msfr_group.ss_family != AF_INET6 ||
1590	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1591		return (EINVAL);
1592
1593	gsa = (sockunion_t *)&msfr.msfr_group;
1594	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1595		return (EINVAL);
1596
1597	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1598		return (EADDRNOTAVAIL);
1599	ifp = ifnet_byindex(msfr.msfr_ifindex);
1600	if (ifp == NULL)
1601		return (EADDRNOTAVAIL);
1602	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1603
1604	INP_WLOCK(inp);
1605
1606	/*
1607	 * Lookup group on the socket.
1608	 */
1609	idx = im6o_match_group(imo, ifp, &gsa->sa);
1610	if (idx == -1 || imo->im6o_mfilters == NULL) {
1611		INP_WUNLOCK(inp);
1612		return (EADDRNOTAVAIL);
1613	}
1614	imf = &imo->im6o_mfilters[idx];
1615
1616	/*
1617	 * Ignore memberships which are in limbo.
1618	 */
1619	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1620		INP_WUNLOCK(inp);
1621		return (EAGAIN);
1622	}
1623	msfr.msfr_fmode = imf->im6f_st[1];
1624
1625	/*
1626	 * If the user specified a buffer, copy out the source filter
1627	 * entries to userland gracefully.
1628	 * We only copy out the number of entries which userland
1629	 * has asked for, but we always tell userland how big the
1630	 * buffer really needs to be.
1631	 */
1632	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1633		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1634	tss = NULL;
1635	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1636		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1637		    M_TEMP, M_NOWAIT | M_ZERO);
1638		if (tss == NULL) {
1639			INP_WUNLOCK(inp);
1640			return (ENOBUFS);
1641		}
1642	}
1643
1644	/*
1645	 * Count number of sources in-mode at t0.
1646	 * If buffer space exists and remains, copy out source entries.
1647	 */
1648	nsrcs = msfr.msfr_nsrcs;
1649	ncsrcs = 0;
1650	ptss = tss;
1651	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1652		lims = (struct in6_msource *)ims;
1653		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1654		    lims->im6sl_st[0] != imf->im6f_st[0])
1655			continue;
1656		++ncsrcs;
1657		if (tss != NULL && nsrcs > 0) {
1658			psin = (struct sockaddr_in6 *)ptss;
1659			psin->sin6_family = AF_INET6;
1660			psin->sin6_len = sizeof(struct sockaddr_in6);
1661			psin->sin6_addr = lims->im6s_addr;
1662			psin->sin6_port = 0;
1663			--nsrcs;
1664			++ptss;
1665		}
1666	}
1667
1668	INP_WUNLOCK(inp);
1669
1670	if (tss != NULL) {
1671		error = copyout(tss, msfr.msfr_srcs,
1672		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1673		free(tss, M_TEMP);
1674		if (error)
1675			return (error);
1676	}
1677
1678	msfr.msfr_nsrcs = ncsrcs;
1679	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1680
1681	return (error);
1682}
1683
1684/*
1685 * Return the IP multicast options in response to user getsockopt().
1686 */
1687int
1688ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1689{
1690	struct ip6_moptions	*im6o;
1691	int			 error;
1692	u_int			 optval;
1693
1694	INP_WLOCK(inp);
1695	im6o = inp->in6p_moptions;
1696	/*
1697	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1698	 * or is a divert socket, reject it.
1699	 */
1700	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1701	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1702	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1703		INP_WUNLOCK(inp);
1704		return (EOPNOTSUPP);
1705	}
1706
1707	error = 0;
1708	switch (sopt->sopt_name) {
1709	case IPV6_MULTICAST_IF:
1710		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1711			optval = 0;
1712		} else {
1713			optval = im6o->im6o_multicast_ifp->if_index;
1714		}
1715		INP_WUNLOCK(inp);
1716		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1717		break;
1718
1719	case IPV6_MULTICAST_HOPS:
1720		if (im6o == NULL)
1721			optval = V_ip6_defmcasthlim;
1722		else
1723			optval = im6o->im6o_multicast_hlim;
1724		INP_WUNLOCK(inp);
1725		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1726		break;
1727
1728	case IPV6_MULTICAST_LOOP:
1729		if (im6o == NULL)
1730			optval = in6_mcast_loop; /* XXX VIMAGE */
1731		else
1732			optval = im6o->im6o_multicast_loop;
1733		INP_WUNLOCK(inp);
1734		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1735		break;
1736
1737	case IPV6_MSFILTER:
1738		if (im6o == NULL) {
1739			error = EADDRNOTAVAIL;
1740			INP_WUNLOCK(inp);
1741		} else {
1742			error = in6p_get_source_filters(inp, sopt);
1743		}
1744		break;
1745
1746	default:
1747		INP_WUNLOCK(inp);
1748		error = ENOPROTOOPT;
1749		break;
1750	}
1751
1752	INP_UNLOCK_ASSERT(inp);
1753
1754	return (error);
1755}
1756
1757/*
1758 * Look up the ifnet to use for a multicast group membership,
1759 * given the address of an IPv6 group.
1760 *
1761 * This routine exists to support legacy IPv6 multicast applications.
1762 *
1763 * If inp is non-NULL, use this socket's current FIB number for any
1764 * required FIB lookup. Look up the group address in the unicast FIB,
1765 * and use its ifp; usually, this points to the default next-hop.
1766 * If the FIB lookup fails, return NULL.
1767 *
1768 * FUTURE: Support multiple forwarding tables for IPv6.
1769 *
1770 * Returns NULL if no ifp could be found.
1771 */
1772static struct ifnet *
1773in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1774    const struct sockaddr_in6 *gsin6)
1775{
1776	struct route_in6	 ro6;
1777	struct ifnet		*ifp;
1778
1779	KASSERT(in6p->inp_vflag & INP_IPV6,
1780	    ("%s: not INP_IPV6 inpcb", __func__));
1781	KASSERT(gsin6->sin6_family == AF_INET6,
1782	    ("%s: not AF_INET6 group", __func__));
1783	KASSERT(IN6_IS_ADDR_MULTICAST(&gsin6->sin6_addr),
1784	    ("%s: not multicast", __func__));
1785
1786	ifp = NULL;
1787	memset(&ro6, 0, sizeof(struct route_in6));
1788	memcpy(&ro6.ro_dst, gsin6, sizeof(struct sockaddr_in6));
1789	rtalloc_ign_fib((struct route *)&ro6, 0,
1790	    in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB);
1791	if (ro6.ro_rt != NULL) {
1792		ifp = ro6.ro_rt->rt_ifp;
1793		KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1794		RTFREE(ro6.ro_rt);
1795	}
1796
1797	return (ifp);
1798}
1799
1800/*
1801 * Join an IPv6 multicast group, possibly with a source.
1802 *
1803 * FIXME: The KAME use of the unspecified address (::)
1804 * to join *all* multicast groups is currently unsupported.
1805 */
1806static int
1807in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1808{
1809	struct group_source_req		 gsr;
1810	sockunion_t			*gsa, *ssa;
1811	struct ifnet			*ifp;
1812	struct in6_mfilter		*imf;
1813	struct ip6_moptions		*imo;
1814	struct in6_multi		*inm;
1815	struct in6_msource		*lims;
1816	size_t				 idx;
1817	int				 error, is_new;
1818
1819	ifp = NULL;
1820	imf = NULL;
1821	lims = NULL;
1822	error = 0;
1823	is_new = 0;
1824
1825	memset(&gsr, 0, sizeof(struct group_source_req));
1826	gsa = (sockunion_t *)&gsr.gsr_group;
1827	gsa->ss.ss_family = AF_UNSPEC;
1828	ssa = (sockunion_t *)&gsr.gsr_source;
1829	ssa->ss.ss_family = AF_UNSPEC;
1830
1831	/*
1832	 * Chew everything into struct group_source_req.
1833	 * Overwrite the port field if present, as the sockaddr
1834	 * being copied in may be matched with a binary comparison.
1835	 * Ignore passed-in scope ID.
1836	 */
1837	switch (sopt->sopt_name) {
1838	case IPV6_JOIN_GROUP: {
1839		struct ipv6_mreq mreq;
1840
1841		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1842		    sizeof(struct ipv6_mreq));
1843		if (error)
1844			return (error);
1845
1846		gsa->sin6.sin6_family = AF_INET6;
1847		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1848		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1849
1850		if (mreq.ipv6mr_interface == 0) {
1851			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1852		} else {
1853			if (mreq.ipv6mr_interface < 0 ||
1854			    V_if_index < mreq.ipv6mr_interface)
1855				return (EADDRNOTAVAIL);
1856			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1857		}
1858		CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1859		    __func__, mreq.ipv6mr_interface, ifp);
1860	} break;
1861
1862	case MCAST_JOIN_GROUP:
1863	case MCAST_JOIN_SOURCE_GROUP:
1864		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1865			error = sooptcopyin(sopt, &gsr,
1866			    sizeof(struct group_req),
1867			    sizeof(struct group_req));
1868		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1869			error = sooptcopyin(sopt, &gsr,
1870			    sizeof(struct group_source_req),
1871			    sizeof(struct group_source_req));
1872		}
1873		if (error)
1874			return (error);
1875
1876		if (gsa->sin6.sin6_family != AF_INET6 ||
1877		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1878			return (EINVAL);
1879
1880		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1881			if (ssa->sin6.sin6_family != AF_INET6 ||
1882			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1883				return (EINVAL);
1884			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1885				return (EINVAL);
1886			/*
1887			 * TODO: Validate embedded scope ID in source
1888			 * list entry against passed-in ifp, if and only
1889			 * if source list filter entry is iface or node local.
1890			 */
1891			in6_clearscope(&ssa->sin6.sin6_addr);
1892			ssa->sin6.sin6_port = 0;
1893			ssa->sin6.sin6_scope_id = 0;
1894		}
1895
1896		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1897			return (EADDRNOTAVAIL);
1898		ifp = ifnet_byindex(gsr.gsr_interface);
1899		break;
1900
1901	default:
1902		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1903		    __func__, sopt->sopt_name);
1904		return (EOPNOTSUPP);
1905		break;
1906	}
1907
1908	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1909		return (EINVAL);
1910
1911	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1912		return (EADDRNOTAVAIL);
1913
1914	gsa->sin6.sin6_port = 0;
1915	gsa->sin6.sin6_scope_id = 0;
1916
1917	/*
1918	 * Always set the scope zone ID on memberships created from userland.
1919	 * Use the passed-in ifp to do this.
1920	 * XXX The in6_setscope() return value is meaningless.
1921	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1922	 */
1923	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1924
1925	imo = in6p_findmoptions(inp);
1926	idx = im6o_match_group(imo, ifp, &gsa->sa);
1927	if (idx == -1) {
1928		is_new = 1;
1929	} else {
1930		inm = imo->im6o_membership[idx];
1931		imf = &imo->im6o_mfilters[idx];
1932		if (ssa->ss.ss_family != AF_UNSPEC) {
1933			/*
1934			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1935			 * is an error. On an existing inclusive membership,
1936			 * it just adds the source to the filter list.
1937			 */
1938			if (imf->im6f_st[1] != MCAST_INCLUDE) {
1939				error = EINVAL;
1940				goto out_in6p_locked;
1941			}
1942			/*
1943			 * Throw out duplicates.
1944			 *
1945			 * XXX FIXME: This makes a naive assumption that
1946			 * even if entries exist for *ssa in this imf,
1947			 * they will be rejected as dupes, even if they
1948			 * are not valid in the current mode (in-mode).
1949			 *
1950			 * in6_msource is transactioned just as for anything
1951			 * else in SSM -- but note naive use of in6m_graft()
1952			 * below for allocating new filter entries.
1953			 *
1954			 * This is only an issue if someone mixes the
1955			 * full-state SSM API with the delta-based API,
1956			 * which is discouraged in the relevant RFCs.
1957			 */
1958			lims = im6o_match_source(imo, idx, &ssa->sa);
1959			if (lims != NULL /*&&
1960			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
1961				error = EADDRNOTAVAIL;
1962				goto out_in6p_locked;
1963			}
1964		} else {
1965			/*
1966			 * MCAST_JOIN_GROUP alone, on any existing membership,
1967			 * is rejected, to stop the same inpcb tying up
1968			 * multiple refs to the in_multi.
1969			 * On an existing inclusive membership, this is also
1970			 * an error; if you want to change filter mode,
1971			 * you must use the userland API setsourcefilter().
1972			 * XXX We don't reject this for imf in UNDEFINED
1973			 * state at t1, because allocation of a filter
1974			 * is atomic with allocation of a membership.
1975			 */
1976			error = EINVAL;
1977			goto out_in6p_locked;
1978		}
1979	}
1980
1981	/*
1982	 * Begin state merge transaction at socket layer.
1983	 */
1984	INP_WLOCK_ASSERT(inp);
1985
1986	if (is_new) {
1987		if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
1988			error = im6o_grow(imo);
1989			if (error)
1990				goto out_in6p_locked;
1991		}
1992		/*
1993		 * Allocate the new slot upfront so we can deal with
1994		 * grafting the new source filter in same code path
1995		 * as for join-source on existing membership.
1996		 */
1997		idx = imo->im6o_num_memberships;
1998		imo->im6o_membership[idx] = NULL;
1999		imo->im6o_num_memberships++;
2000		KASSERT(imo->im6o_mfilters != NULL,
2001		    ("%s: im6f_mfilters vector was not allocated", __func__));
2002		imf = &imo->im6o_mfilters[idx];
2003		KASSERT(RB_EMPTY(&imf->im6f_sources),
2004		    ("%s: im6f_sources not empty", __func__));
2005	}
2006
2007	/*
2008	 * Graft new source into filter list for this inpcb's
2009	 * membership of the group. The in6_multi may not have
2010	 * been allocated yet if this is a new membership, however,
2011	 * the in_mfilter slot will be allocated and must be initialized.
2012	 *
2013	 * Note: Grafting of exclusive mode filters doesn't happen
2014	 * in this path.
2015	 * XXX: Should check for non-NULL lims (node exists but may
2016	 * not be in-mode) for interop with full-state API.
2017	 */
2018	if (ssa->ss.ss_family != AF_UNSPEC) {
2019		/* Membership starts in IN mode */
2020		if (is_new) {
2021			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2022			im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2023		} else {
2024			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2025		}
2026		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2027		if (lims == NULL) {
2028			CTR1(KTR_MLD, "%s: merge imf state failed",
2029			    __func__);
2030			error = ENOMEM;
2031			goto out_im6o_free;
2032		}
2033	} else {
2034		/* No address specified; Membership starts in EX mode */
2035		if (is_new) {
2036			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2037			im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2038		}
2039	}
2040
2041	/*
2042	 * Begin state merge transaction at MLD layer.
2043	 */
2044	IN6_MULTI_LOCK();
2045
2046	if (is_new) {
2047		error = in6_mc_join_locked(ifp, &gsa->sin6.sin6_addr, imf,
2048		    &inm, 0);
2049		if (error)
2050			goto out_im6o_free;
2051		imo->im6o_membership[idx] = inm;
2052	} else {
2053		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2054		error = in6m_merge(inm, imf);
2055		if (error) {
2056			CTR1(KTR_MLD, "%s: failed to merge inm state",
2057			    __func__);
2058			goto out_im6f_rollback;
2059		}
2060		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2061		error = mld_change_state(inm, 0);
2062		if (error) {
2063			CTR1(KTR_MLD, "%s: failed mld downcall",
2064			    __func__);
2065			goto out_im6f_rollback;
2066		}
2067	}
2068
2069	IN6_MULTI_UNLOCK();
2070
2071out_im6f_rollback:
2072	INP_WLOCK_ASSERT(inp);
2073	if (error) {
2074		im6f_rollback(imf);
2075		if (is_new)
2076			im6f_purge(imf);
2077		else
2078			im6f_reap(imf);
2079	} else {
2080		im6f_commit(imf);
2081	}
2082
2083out_im6o_free:
2084	if (error && is_new) {
2085		imo->im6o_membership[idx] = NULL;
2086		--imo->im6o_num_memberships;
2087	}
2088
2089out_in6p_locked:
2090	INP_WUNLOCK(inp);
2091	return (error);
2092}
2093
2094/*
2095 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2096 */
2097static int
2098in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2099{
2100	struct ipv6_mreq		 mreq;
2101	struct group_source_req		 gsr;
2102	sockunion_t			*gsa, *ssa;
2103	struct ifnet			*ifp;
2104	struct in6_mfilter		*imf;
2105	struct ip6_moptions		*imo;
2106	struct in6_msource		*ims;
2107	struct in6_multi		*inm;
2108	uint32_t			 ifindex;
2109	size_t				 idx;
2110	int				 error, is_final;
2111#ifdef KTR
2112	char				 ip6tbuf[INET6_ADDRSTRLEN];
2113#endif
2114
2115	ifp = NULL;
2116	ifindex = 0;
2117	error = 0;
2118	is_final = 1;
2119
2120	memset(&gsr, 0, sizeof(struct group_source_req));
2121	gsa = (sockunion_t *)&gsr.gsr_group;
2122	gsa->ss.ss_family = AF_UNSPEC;
2123	ssa = (sockunion_t *)&gsr.gsr_source;
2124	ssa->ss.ss_family = AF_UNSPEC;
2125
2126	/*
2127	 * Chew everything passed in up into a struct group_source_req
2128	 * as that is easier to process.
2129	 * Note: Any embedded scope ID in the multicast group passed
2130	 * in by userland is ignored, the interface index is the recommended
2131	 * mechanism to specify an interface; see below.
2132	 */
2133	switch (sopt->sopt_name) {
2134	case IPV6_LEAVE_GROUP:
2135		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2136		    sizeof(struct ipv6_mreq));
2137		if (error)
2138			return (error);
2139		gsa->sin6.sin6_family = AF_INET6;
2140		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2141		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2142		gsa->sin6.sin6_port = 0;
2143		gsa->sin6.sin6_scope_id = 0;
2144		ifindex = mreq.ipv6mr_interface;
2145		break;
2146
2147	case MCAST_LEAVE_GROUP:
2148	case MCAST_LEAVE_SOURCE_GROUP:
2149		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2150			error = sooptcopyin(sopt, &gsr,
2151			    sizeof(struct group_req),
2152			    sizeof(struct group_req));
2153		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2154			error = sooptcopyin(sopt, &gsr,
2155			    sizeof(struct group_source_req),
2156			    sizeof(struct group_source_req));
2157		}
2158		if (error)
2159			return (error);
2160
2161		if (gsa->sin6.sin6_family != AF_INET6 ||
2162		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2163			return (EINVAL);
2164		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2165			if (ssa->sin6.sin6_family != AF_INET6 ||
2166			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2167				return (EINVAL);
2168			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2169				return (EINVAL);
2170			/*
2171			 * TODO: Validate embedded scope ID in source
2172			 * list entry against passed-in ifp, if and only
2173			 * if source list filter entry is iface or node local.
2174			 */
2175			in6_clearscope(&ssa->sin6.sin6_addr);
2176		}
2177		gsa->sin6.sin6_port = 0;
2178		gsa->sin6.sin6_scope_id = 0;
2179		ifindex = gsr.gsr_interface;
2180		break;
2181
2182	default:
2183		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2184		    __func__, sopt->sopt_name);
2185		return (EOPNOTSUPP);
2186		break;
2187	}
2188
2189	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2190		return (EINVAL);
2191
2192	/*
2193	 * Validate interface index if provided. If no interface index
2194	 * was provided separately, attempt to look the membership up
2195	 * from the default scope as a last resort to disambiguate
2196	 * the membership we are being asked to leave.
2197	 * XXX SCOPE6 lock potentially taken here.
2198	 */
2199	if (ifindex != 0) {
2200		if (ifindex < 0 || V_if_index < ifindex)
2201			return (EADDRNOTAVAIL);
2202		ifp = ifnet_byindex(ifindex);
2203		if (ifp == NULL)
2204			return (EADDRNOTAVAIL);
2205		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2206	} else {
2207		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2208		if (error)
2209			return (EADDRNOTAVAIL);
2210		/*
2211		 * Some badly behaved applications don't pass an ifindex
2212		 * or a scope ID, which is an API violation. In this case,
2213		 * perform a lookup as per a v6 join.
2214		 *
2215		 * XXX For now, stomp on zone ID for the corner case.
2216		 * This is not the 'KAME way', but we need to see the ifp
2217		 * directly until such time as this implementation is
2218		 * refactored, assuming the scope IDs are the way to go.
2219		 */
2220		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2221		if (ifindex == 0) {
2222			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2223			    "ifp for group %s.", __func__,
2224			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2225			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2226		} else {
2227			ifp = ifnet_byindex(ifindex);
2228		}
2229		if (ifp == NULL)
2230			return (EADDRNOTAVAIL);
2231	}
2232
2233	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2234	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2235
2236	/*
2237	 * Find the membership in the membership array.
2238	 */
2239	imo = in6p_findmoptions(inp);
2240	idx = im6o_match_group(imo, ifp, &gsa->sa);
2241	if (idx == -1) {
2242		error = EADDRNOTAVAIL;
2243		goto out_in6p_locked;
2244	}
2245	inm = imo->im6o_membership[idx];
2246	imf = &imo->im6o_mfilters[idx];
2247
2248	if (ssa->ss.ss_family != AF_UNSPEC)
2249		is_final = 0;
2250
2251	/*
2252	 * Begin state merge transaction at socket layer.
2253	 */
2254	INP_WLOCK_ASSERT(inp);
2255
2256	/*
2257	 * If we were instructed only to leave a given source, do so.
2258	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2259	 */
2260	if (is_final) {
2261		im6f_leave(imf);
2262	} else {
2263		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2264			error = EADDRNOTAVAIL;
2265			goto out_in6p_locked;
2266		}
2267		ims = im6o_match_source(imo, idx, &ssa->sa);
2268		if (ims == NULL) {
2269			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2270			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2271			    "not ");
2272			error = EADDRNOTAVAIL;
2273			goto out_in6p_locked;
2274		}
2275		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2276		error = im6f_prune(imf, &ssa->sin6);
2277		if (error) {
2278			CTR1(KTR_MLD, "%s: merge imf state failed",
2279			    __func__);
2280			goto out_in6p_locked;
2281		}
2282	}
2283
2284	/*
2285	 * Begin state merge transaction at MLD layer.
2286	 */
2287	IN6_MULTI_LOCK();
2288
2289	if (is_final) {
2290		/*
2291		 * Give up the multicast address record to which
2292		 * the membership points.
2293		 */
2294		(void)in6_mc_leave_locked(inm, imf);
2295	} else {
2296		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2297		error = in6m_merge(inm, imf);
2298		if (error) {
2299			CTR1(KTR_MLD, "%s: failed to merge inm state",
2300			    __func__);
2301			goto out_im6f_rollback;
2302		}
2303
2304		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2305		error = mld_change_state(inm, 0);
2306		if (error) {
2307			CTR1(KTR_MLD, "%s: failed mld downcall",
2308			    __func__);
2309		}
2310	}
2311
2312	IN6_MULTI_UNLOCK();
2313
2314out_im6f_rollback:
2315	if (error)
2316		im6f_rollback(imf);
2317	else
2318		im6f_commit(imf);
2319
2320	im6f_reap(imf);
2321
2322	if (is_final) {
2323		/* Remove the gap in the membership array. */
2324		for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2325			imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2326			imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2327		}
2328		imo->im6o_num_memberships--;
2329	}
2330
2331out_in6p_locked:
2332	INP_WUNLOCK(inp);
2333	return (error);
2334}
2335
2336/*
2337 * Select the interface for transmitting IPv6 multicast datagrams.
2338 *
2339 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2340 * may be passed to this socket option. An address of in6addr_any or an
2341 * interface index of 0 is used to remove a previous selection.
2342 * When no interface is selected, one is chosen for every send.
2343 */
2344static int
2345in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2346{
2347	struct ifnet		*ifp;
2348	struct ip6_moptions	*imo;
2349	u_int			 ifindex;
2350	int			 error;
2351
2352	if (sopt->sopt_valsize != sizeof(u_int))
2353		return (EINVAL);
2354
2355	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2356	if (error)
2357		return (error);
2358	if (ifindex < 0 || V_if_index < ifindex)
2359		return (EINVAL);
2360
2361	ifp = ifnet_byindex(ifindex);
2362	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2363		return (EADDRNOTAVAIL);
2364
2365	imo = in6p_findmoptions(inp);
2366	imo->im6o_multicast_ifp = ifp;
2367	INP_WUNLOCK(inp);
2368
2369	return (0);
2370}
2371
2372/*
2373 * Atomically set source filters on a socket for an IPv6 multicast group.
2374 *
2375 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2376 */
2377static int
2378in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2379{
2380	struct __msfilterreq	 msfr;
2381	sockunion_t		*gsa;
2382	struct ifnet		*ifp;
2383	struct in6_mfilter	*imf;
2384	struct ip6_moptions	*imo;
2385	struct in6_multi		*inm;
2386	size_t			 idx;
2387	int			 error;
2388
2389	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2390	    sizeof(struct __msfilterreq));
2391	if (error)
2392		return (error);
2393
2394	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2395		return (ENOBUFS);
2396
2397	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2398	    msfr.msfr_fmode != MCAST_INCLUDE)
2399		return (EINVAL);
2400
2401	if (msfr.msfr_group.ss_family != AF_INET6 ||
2402	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2403		return (EINVAL);
2404
2405	gsa = (sockunion_t *)&msfr.msfr_group;
2406	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2407		return (EINVAL);
2408
2409	gsa->sin6.sin6_port = 0;	/* ignore port */
2410
2411	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2412		return (EADDRNOTAVAIL);
2413	ifp = ifnet_byindex(msfr.msfr_ifindex);
2414	if (ifp == NULL)
2415		return (EADDRNOTAVAIL);
2416	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2417
2418	/*
2419	 * Take the INP write lock.
2420	 * Check if this socket is a member of this group.
2421	 */
2422	imo = in6p_findmoptions(inp);
2423	idx = im6o_match_group(imo, ifp, &gsa->sa);
2424	if (idx == -1 || imo->im6o_mfilters == NULL) {
2425		error = EADDRNOTAVAIL;
2426		goto out_in6p_locked;
2427	}
2428	inm = imo->im6o_membership[idx];
2429	imf = &imo->im6o_mfilters[idx];
2430
2431	/*
2432	 * Begin state merge transaction at socket layer.
2433	 */
2434	INP_WLOCK_ASSERT(inp);
2435
2436	imf->im6f_st[1] = msfr.msfr_fmode;
2437
2438	/*
2439	 * Apply any new source filters, if present.
2440	 * Make a copy of the user-space source vector so
2441	 * that we may copy them with a single copyin. This
2442	 * allows us to deal with page faults up-front.
2443	 */
2444	if (msfr.msfr_nsrcs > 0) {
2445		struct in6_msource	*lims;
2446		struct sockaddr_in6	*psin;
2447		struct sockaddr_storage	*kss, *pkss;
2448		int			 i;
2449
2450		INP_WUNLOCK(inp);
2451
2452		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2453		    __func__, (unsigned long)msfr.msfr_nsrcs);
2454		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2455		    M_TEMP, M_WAITOK);
2456		error = copyin(msfr.msfr_srcs, kss,
2457		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2458		if (error) {
2459			free(kss, M_TEMP);
2460			return (error);
2461		}
2462
2463		INP_WLOCK(inp);
2464
2465		/*
2466		 * Mark all source filters as UNDEFINED at t1.
2467		 * Restore new group filter mode, as im6f_leave()
2468		 * will set it to INCLUDE.
2469		 */
2470		im6f_leave(imf);
2471		imf->im6f_st[1] = msfr.msfr_fmode;
2472
2473		/*
2474		 * Update socket layer filters at t1, lazy-allocating
2475		 * new entries. This saves a bunch of memory at the
2476		 * cost of one RB_FIND() per source entry; duplicate
2477		 * entries in the msfr_nsrcs vector are ignored.
2478		 * If we encounter an error, rollback transaction.
2479		 *
2480		 * XXX This too could be replaced with a set-symmetric
2481		 * difference like loop to avoid walking from root
2482		 * every time, as the key space is common.
2483		 */
2484		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2485			psin = (struct sockaddr_in6 *)pkss;
2486			if (psin->sin6_family != AF_INET6) {
2487				error = EAFNOSUPPORT;
2488				break;
2489			}
2490			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2491				error = EINVAL;
2492				break;
2493			}
2494			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2495				error = EINVAL;
2496				break;
2497			}
2498			/*
2499			 * TODO: Validate embedded scope ID in source
2500			 * list entry against passed-in ifp, if and only
2501			 * if source list filter entry is iface or node local.
2502			 */
2503			in6_clearscope(&psin->sin6_addr);
2504			error = im6f_get_source(imf, psin, &lims);
2505			if (error)
2506				break;
2507			lims->im6sl_st[1] = imf->im6f_st[1];
2508		}
2509		free(kss, M_TEMP);
2510	}
2511
2512	if (error)
2513		goto out_im6f_rollback;
2514
2515	INP_WLOCK_ASSERT(inp);
2516	IN6_MULTI_LOCK();
2517
2518	/*
2519	 * Begin state merge transaction at MLD layer.
2520	 */
2521	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2522	error = in6m_merge(inm, imf);
2523	if (error) {
2524		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2525		goto out_im6f_rollback;
2526	}
2527
2528	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2529	error = mld_change_state(inm, 0);
2530	if (error)
2531		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2532
2533	IN6_MULTI_UNLOCK();
2534
2535out_im6f_rollback:
2536	if (error)
2537		im6f_rollback(imf);
2538	else
2539		im6f_commit(imf);
2540
2541	im6f_reap(imf);
2542
2543out_in6p_locked:
2544	INP_WUNLOCK(inp);
2545	return (error);
2546}
2547
2548/*
2549 * Set the IP multicast options in response to user setsockopt().
2550 *
2551 * Many of the socket options handled in this function duplicate the
2552 * functionality of socket options in the regular unicast API. However,
2553 * it is not possible to merge the duplicate code, because the idempotence
2554 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2555 * the effects of these options must be treated as separate and distinct.
2556 *
2557 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2558 */
2559int
2560ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2561{
2562	struct ip6_moptions	*im6o;
2563	int			 error;
2564
2565	error = 0;
2566
2567	/*
2568	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2569	 * or is a divert socket, reject it.
2570	 */
2571	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2572	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2573	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2574		return (EOPNOTSUPP);
2575
2576	switch (sopt->sopt_name) {
2577	case IPV6_MULTICAST_IF:
2578		error = in6p_set_multicast_if(inp, sopt);
2579		break;
2580
2581	case IPV6_MULTICAST_HOPS: {
2582		int hlim;
2583
2584		if (sopt->sopt_valsize != sizeof(int)) {
2585			error = EINVAL;
2586			break;
2587		}
2588		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2589		if (error)
2590			break;
2591		if (hlim < -1 || hlim > 255) {
2592			error = EINVAL;
2593			break;
2594		} else if (hlim == -1) {
2595			hlim = V_ip6_defmcasthlim;
2596		}
2597		im6o = in6p_findmoptions(inp);
2598		im6o->im6o_multicast_hlim = hlim;
2599		INP_WUNLOCK(inp);
2600		break;
2601	}
2602
2603	case IPV6_MULTICAST_LOOP: {
2604		u_int loop;
2605
2606		/*
2607		 * Set the loopback flag for outgoing multicast packets.
2608		 * Must be zero or one.
2609		 */
2610		if (sopt->sopt_valsize != sizeof(u_int)) {
2611			error = EINVAL;
2612			break;
2613		}
2614		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2615		if (error)
2616			break;
2617		if (loop > 1) {
2618			error = EINVAL;
2619			break;
2620		}
2621		im6o = in6p_findmoptions(inp);
2622		im6o->im6o_multicast_loop = loop;
2623		INP_WUNLOCK(inp);
2624		break;
2625	}
2626
2627	case IPV6_JOIN_GROUP:
2628	case MCAST_JOIN_GROUP:
2629	case MCAST_JOIN_SOURCE_GROUP:
2630		error = in6p_join_group(inp, sopt);
2631		break;
2632
2633	case IPV6_LEAVE_GROUP:
2634	case MCAST_LEAVE_GROUP:
2635	case MCAST_LEAVE_SOURCE_GROUP:
2636		error = in6p_leave_group(inp, sopt);
2637		break;
2638
2639	case MCAST_BLOCK_SOURCE:
2640	case MCAST_UNBLOCK_SOURCE:
2641		error = in6p_block_unblock_source(inp, sopt);
2642		break;
2643
2644	case IPV6_MSFILTER:
2645		error = in6p_set_source_filters(inp, sopt);
2646		break;
2647
2648	default:
2649		error = EOPNOTSUPP;
2650		break;
2651	}
2652
2653	INP_UNLOCK_ASSERT(inp);
2654
2655	return (error);
2656}
2657
2658/*
2659 * Expose MLD's multicast filter mode and source list(s) to userland,
2660 * keyed by (ifindex, group).
2661 * The filter mode is written out as a uint32_t, followed by
2662 * 0..n of struct in6_addr.
2663 * For use by ifmcstat(8).
2664 * SMPng: NOTE: unlocked read of ifindex space.
2665 */
2666static int
2667sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2668{
2669	struct in6_addr			 mcaddr;
2670	struct in6_addr			 src;
2671	struct ifnet			*ifp;
2672	struct ifmultiaddr		*ifma;
2673	struct in6_multi		*inm;
2674	struct ip6_msource		*ims;
2675	int				*name;
2676	int				 retval;
2677	u_int				 namelen;
2678	uint32_t			 fmode, ifindex;
2679#ifdef KTR
2680	char				 ip6tbuf[INET6_ADDRSTRLEN];
2681#endif
2682
2683	name = (int *)arg1;
2684	namelen = arg2;
2685
2686	if (req->newptr != NULL)
2687		return (EPERM);
2688
2689	/* int: ifindex + 4 * 32 bits of IPv6 address */
2690	if (namelen != 5)
2691		return (EINVAL);
2692
2693	ifindex = name[0];
2694	if (ifindex <= 0 || ifindex > V_if_index) {
2695		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2696		    __func__, ifindex);
2697		return (ENOENT);
2698	}
2699
2700	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2701	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2702		CTR2(KTR_MLD, "%s: group %s is not multicast",
2703		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2704		return (EINVAL);
2705	}
2706
2707	ifp = ifnet_byindex(ifindex);
2708	if (ifp == NULL) {
2709		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2710		    __func__, ifindex);
2711		return (ENOENT);
2712	}
2713	/*
2714	 * Internal MLD lookups require that scope/zone ID is set.
2715	 */
2716	(void)in6_setscope(&mcaddr, ifp, NULL);
2717
2718	retval = sysctl_wire_old_buffer(req,
2719	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2720	if (retval)
2721		return (retval);
2722
2723	IN6_MULTI_LOCK();
2724
2725	IF_ADDR_RLOCK(ifp);
2726	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2727		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2728		    ifma->ifma_protospec == NULL)
2729			continue;
2730		inm = (struct in6_multi *)ifma->ifma_protospec;
2731		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2732			continue;
2733		fmode = inm->in6m_st[1].iss_fmode;
2734		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2735		if (retval != 0)
2736			break;
2737		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2738			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2739			/*
2740			 * Only copy-out sources which are in-mode.
2741			 */
2742			if (fmode != im6s_get_mode(inm, ims, 1)) {
2743				CTR1(KTR_MLD, "%s: skip non-in-mode",
2744				    __func__);
2745				continue;
2746			}
2747			src = ims->im6s_addr;
2748			retval = SYSCTL_OUT(req, &src,
2749			    sizeof(struct in6_addr));
2750			if (retval != 0)
2751				break;
2752		}
2753	}
2754	IF_ADDR_RUNLOCK(ifp);
2755
2756	IN6_MULTI_UNLOCK();
2757
2758	return (retval);
2759}
2760
2761#ifdef KTR
2762
2763static const char *in6m_modestrs[] = { "un", "in", "ex" };
2764
2765static const char *
2766in6m_mode_str(const int mode)
2767{
2768
2769	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2770		return (in6m_modestrs[mode]);
2771	return ("??");
2772}
2773
2774static const char *in6m_statestrs[] = {
2775	"not-member",
2776	"silent",
2777	"idle",
2778	"lazy",
2779	"sleeping",
2780	"awakening",
2781	"query-pending",
2782	"sg-query-pending",
2783	"leaving"
2784};
2785
2786static const char *
2787in6m_state_str(const int state)
2788{
2789
2790	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2791		return (in6m_statestrs[state]);
2792	return ("??");
2793}
2794
2795/*
2796 * Dump an in6_multi structure to the console.
2797 */
2798void
2799in6m_print(const struct in6_multi *inm)
2800{
2801	int t;
2802	char ip6tbuf[INET6_ADDRSTRLEN];
2803
2804	if ((ktr_mask & KTR_MLD) == 0)
2805		return;
2806
2807	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2808	printf("addr %s ifp %p(%s) ifma %p\n",
2809	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2810	    inm->in6m_ifp,
2811	    inm->in6m_ifp->if_xname,
2812	    inm->in6m_ifma);
2813	printf("timer %u state %s refcount %u scq.len %u\n",
2814	    inm->in6m_timer,
2815	    in6m_state_str(inm->in6m_state),
2816	    inm->in6m_refcount,
2817	    inm->in6m_scq.ifq_len);
2818	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2819	    inm->in6m_mli,
2820	    inm->in6m_nsrc,
2821	    inm->in6m_sctimer,
2822	    inm->in6m_scrv);
2823	for (t = 0; t < 2; t++) {
2824		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2825		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2826		    inm->in6m_st[t].iss_asm,
2827		    inm->in6m_st[t].iss_ex,
2828		    inm->in6m_st[t].iss_in,
2829		    inm->in6m_st[t].iss_rec);
2830	}
2831	printf("%s: --- end in6m %p ---\n", __func__, inm);
2832}
2833
2834#else /* !KTR */
2835
2836void
2837in6m_print(const struct in6_multi *inm)
2838{
2839
2840}
2841
2842#endif /* KTR */
2843