in6_mcast.c revision 321134
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 321134 2017-07-18 16:58:52Z ngie $");
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
1009	/* Increment ASM listener count on transition to ASM mode. */
1010	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1011		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1012		inm->in6m_st[1].iss_asm++;
1013	}
1014
1015	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1016	in6m_print(inm);
1017
1018out_reap:
1019	if (schanged > 0) {
1020		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1021		in6m_reap(inm);
1022	}
1023	return (error);
1024}
1025
1026/*
1027 * Mark an in6_multi's filter set deltas as committed.
1028 * Called by MLD after a state change has been enqueued.
1029 */
1030void
1031in6m_commit(struct in6_multi *inm)
1032{
1033	struct ip6_msource	*ims;
1034
1035	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1036	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1037	in6m_print(inm);
1038
1039	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1040		ims->im6s_st[0] = ims->im6s_st[1];
1041	}
1042	inm->in6m_st[0] = inm->in6m_st[1];
1043}
1044
1045/*
1046 * Reap unreferenced nodes from an in6_multi's filter set.
1047 */
1048static void
1049in6m_reap(struct in6_multi *inm)
1050{
1051	struct ip6_msource	*ims, *tims;
1052
1053	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1054		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1055		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1056		    ims->im6s_stp != 0)
1057			continue;
1058		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1059		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1060		free(ims, M_IP6MSOURCE);
1061		inm->in6m_nsrc--;
1062	}
1063}
1064
1065/*
1066 * Purge all source nodes from an in6_multi's filter set.
1067 */
1068static void
1069in6m_purge(struct in6_multi *inm)
1070{
1071	struct ip6_msource	*ims, *tims;
1072
1073	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1074		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1075		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1076		free(ims, M_IP6MSOURCE);
1077		inm->in6m_nsrc--;
1078	}
1079	/* Free state-change requests that might be queued. */
1080	_IF_DRAIN(&inm->in6m_scq);
1081}
1082
1083/*
1084 * Join a multicast address w/o sources.
1085 * KAME compatibility entry point.
1086 *
1087 * SMPng: Assume no mc locks held by caller.
1088 */
1089struct in6_multi_mship *
1090in6_joingroup(struct ifnet *ifp, struct in6_addr *mcaddr,
1091    int *errorp, int delay)
1092{
1093	struct in6_multi_mship *imm;
1094	int error;
1095
1096	imm = malloc(sizeof(*imm), M_IP6MADDR, M_NOWAIT);
1097	if (imm == NULL) {
1098		*errorp = ENOBUFS;
1099		return (NULL);
1100	}
1101
1102	delay = (delay * PR_FASTHZ) / hz;
1103
1104	error = in6_mc_join(ifp, mcaddr, NULL, &imm->i6mm_maddr, delay);
1105	if (error) {
1106		*errorp = error;
1107		free(imm, M_IP6MADDR);
1108		return (NULL);
1109	}
1110
1111	return (imm);
1112}
1113
1114/*
1115 * Leave a multicast address w/o sources.
1116 * KAME compatibility entry point.
1117 *
1118 * SMPng: Assume no mc locks held by caller.
1119 */
1120int
1121in6_leavegroup(struct in6_multi_mship *imm)
1122{
1123
1124	if (imm->i6mm_maddr != NULL)
1125		in6_mc_leave(imm->i6mm_maddr, NULL);
1126	free(imm,  M_IP6MADDR);
1127	return 0;
1128}
1129
1130/*
1131 * Join a multicast group; unlocked entry point.
1132 *
1133 * SMPng: XXX: in6_mc_join() is called from in6_control() when upper
1134 * locks are not held. Fortunately, ifp is unlikely to have been detached
1135 * at this point, so we assume it's OK to recurse.
1136 */
1137int
1138in6_mc_join(struct ifnet *ifp, const struct in6_addr *mcaddr,
1139    /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1140    const int delay)
1141{
1142	int error;
1143
1144	IN6_MULTI_LOCK();
1145	error = in6_mc_join_locked(ifp, mcaddr, imf, pinm, delay);
1146	IN6_MULTI_UNLOCK();
1147
1148	return (error);
1149}
1150
1151/*
1152 * Join a multicast group; real entry point.
1153 *
1154 * Only preserves atomicity at inm level.
1155 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1156 *
1157 * If the MLD downcall fails, the group is not joined, and an error
1158 * code is returned.
1159 */
1160int
1161in6_mc_join_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1162    /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1163    const int delay)
1164{
1165	struct in6_mfilter	 timf;
1166	struct in6_multi	*inm;
1167	int			 error;
1168#ifdef KTR
1169	char			 ip6tbuf[INET6_ADDRSTRLEN];
1170#endif
1171
1172#ifdef INVARIANTS
1173	/*
1174	 * Sanity: Check scope zone ID was set for ifp, if and
1175	 * only if group is scoped to an interface.
1176	 */
1177	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1178	    ("%s: not a multicast address", __func__));
1179	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1180	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1181		KASSERT(mcaddr->s6_addr16[1] != 0,
1182		    ("%s: scope zone ID not set", __func__));
1183	}
1184#endif
1185
1186	IN6_MULTI_LOCK_ASSERT();
1187
1188	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1189	    ip6_sprintf(ip6tbuf, mcaddr), ifp, ifp->if_xname);
1190
1191	error = 0;
1192	inm = NULL;
1193
1194	/*
1195	 * If no imf was specified (i.e. kernel consumer),
1196	 * fake one up and assume it is an ASM join.
1197	 */
1198	if (imf == NULL) {
1199		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1200		imf = &timf;
1201	}
1202
1203	error = in6_mc_get(ifp, mcaddr, &inm);
1204	if (error) {
1205		CTR1(KTR_MLD, "%s: in6_mc_get() failure", __func__);
1206		return (error);
1207	}
1208
1209	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1210	error = in6m_merge(inm, imf);
1211	if (error) {
1212		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1213		goto out_in6m_release;
1214	}
1215
1216	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1217	error = mld_change_state(inm, delay);
1218	if (error) {
1219		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1220		goto out_in6m_release;
1221	}
1222
1223out_in6m_release:
1224	if (error) {
1225		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1226		in6m_release_locked(inm);
1227	} else {
1228		*pinm = inm;
1229	}
1230
1231	return (error);
1232}
1233
1234/*
1235 * Leave a multicast group; unlocked entry point.
1236 */
1237int
1238in6_mc_leave(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1239{
1240	struct ifnet *ifp;
1241	int error;
1242
1243	ifp = inm->in6m_ifp;
1244
1245	IN6_MULTI_LOCK();
1246	error = in6_mc_leave_locked(inm, imf);
1247	IN6_MULTI_UNLOCK();
1248
1249	return (error);
1250}
1251
1252/*
1253 * Leave a multicast group; real entry point.
1254 * All source filters will be expunged.
1255 *
1256 * Only preserves atomicity at inm level.
1257 *
1258 * Holding the write lock for the INP which contains imf
1259 * is highly advisable. We can't assert for it as imf does not
1260 * contain a back-pointer to the owning inp.
1261 *
1262 * Note: This is not the same as in6m_release(*) as this function also
1263 * makes a state change downcall into MLD.
1264 */
1265int
1266in6_mc_leave_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1267{
1268	struct in6_mfilter	 timf;
1269	int			 error;
1270#ifdef KTR
1271	char			 ip6tbuf[INET6_ADDRSTRLEN];
1272#endif
1273
1274	error = 0;
1275
1276	IN6_MULTI_LOCK_ASSERT();
1277
1278	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1279	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1280	    (in6m_is_ifp_detached(inm) ? "null" : inm->in6m_ifp->if_xname),
1281	    imf);
1282
1283	/*
1284	 * If no imf was specified (i.e. kernel consumer),
1285	 * fake one up and assume it is an ASM join.
1286	 */
1287	if (imf == NULL) {
1288		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1289		imf = &timf;
1290	}
1291
1292	/*
1293	 * Begin state merge transaction at MLD layer.
1294	 *
1295	 * As this particular invocation should not cause any memory
1296	 * to be allocated, and there is no opportunity to roll back
1297	 * the transaction, it MUST NOT fail.
1298	 */
1299	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1300	error = in6m_merge(inm, imf);
1301	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1302
1303	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1304	error = mld_change_state(inm, 0);
1305	if (error)
1306		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1307
1308	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1309	in6m_release_locked(inm);
1310
1311	return (error);
1312}
1313
1314/*
1315 * Block or unblock an ASM multicast source on an inpcb.
1316 * This implements the delta-based API described in RFC 3678.
1317 *
1318 * The delta-based API applies only to exclusive-mode memberships.
1319 * An MLD downcall will be performed.
1320 *
1321 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1322 *
1323 * Return 0 if successful, otherwise return an appropriate error code.
1324 */
1325static int
1326in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1327{
1328	struct group_source_req		 gsr;
1329	sockunion_t			*gsa, *ssa;
1330	struct ifnet			*ifp;
1331	struct in6_mfilter		*imf;
1332	struct ip6_moptions		*imo;
1333	struct in6_msource		*ims;
1334	struct in6_multi			*inm;
1335	size_t				 idx;
1336	uint16_t			 fmode;
1337	int				 error, doblock;
1338#ifdef KTR
1339	char				 ip6tbuf[INET6_ADDRSTRLEN];
1340#endif
1341
1342	ifp = NULL;
1343	error = 0;
1344	doblock = 0;
1345
1346	memset(&gsr, 0, sizeof(struct group_source_req));
1347	gsa = (sockunion_t *)&gsr.gsr_group;
1348	ssa = (sockunion_t *)&gsr.gsr_source;
1349
1350	switch (sopt->sopt_name) {
1351	case MCAST_BLOCK_SOURCE:
1352	case MCAST_UNBLOCK_SOURCE:
1353		error = sooptcopyin(sopt, &gsr,
1354		    sizeof(struct group_source_req),
1355		    sizeof(struct group_source_req));
1356		if (error)
1357			return (error);
1358
1359		if (gsa->sin6.sin6_family != AF_INET6 ||
1360		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1361			return (EINVAL);
1362
1363		if (ssa->sin6.sin6_family != AF_INET6 ||
1364		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1365			return (EINVAL);
1366
1367		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1368			return (EADDRNOTAVAIL);
1369
1370		ifp = ifnet_byindex(gsr.gsr_interface);
1371
1372		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1373			doblock = 1;
1374		break;
1375
1376	default:
1377		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1378		    __func__, sopt->sopt_name);
1379		return (EOPNOTSUPP);
1380		break;
1381	}
1382
1383	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1384		return (EINVAL);
1385
1386	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1387
1388	/*
1389	 * Check if we are actually a member of this group.
1390	 */
1391	imo = in6p_findmoptions(inp);
1392	idx = im6o_match_group(imo, ifp, &gsa->sa);
1393	if (idx == -1 || imo->im6o_mfilters == NULL) {
1394		error = EADDRNOTAVAIL;
1395		goto out_in6p_locked;
1396	}
1397
1398	KASSERT(imo->im6o_mfilters != NULL,
1399	    ("%s: im6o_mfilters not allocated", __func__));
1400	imf = &imo->im6o_mfilters[idx];
1401	inm = imo->im6o_membership[idx];
1402
1403	/*
1404	 * Attempting to use the delta-based API on an
1405	 * non exclusive-mode membership is an error.
1406	 */
1407	fmode = imf->im6f_st[0];
1408	if (fmode != MCAST_EXCLUDE) {
1409		error = EINVAL;
1410		goto out_in6p_locked;
1411	}
1412
1413	/*
1414	 * Deal with error cases up-front:
1415	 *  Asked to block, but already blocked; or
1416	 *  Asked to unblock, but nothing to unblock.
1417	 * If adding a new block entry, allocate it.
1418	 */
1419	ims = im6o_match_source(imo, idx, &ssa->sa);
1420	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1421		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1422		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1423		    doblock ? "" : "not ");
1424		error = EADDRNOTAVAIL;
1425		goto out_in6p_locked;
1426	}
1427
1428	INP_WLOCK_ASSERT(inp);
1429
1430	/*
1431	 * Begin state merge transaction at socket layer.
1432	 */
1433	if (doblock) {
1434		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1435		ims = im6f_graft(imf, fmode, &ssa->sin6);
1436		if (ims == NULL)
1437			error = ENOMEM;
1438	} else {
1439		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1440		error = im6f_prune(imf, &ssa->sin6);
1441	}
1442
1443	if (error) {
1444		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1445		goto out_im6f_rollback;
1446	}
1447
1448	/*
1449	 * Begin state merge transaction at MLD layer.
1450	 */
1451	IN6_MULTI_LOCK();
1452
1453	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1454	error = in6m_merge(inm, imf);
1455	if (error)
1456		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1457	else {
1458		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1459		error = mld_change_state(inm, 0);
1460		if (error)
1461			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1462	}
1463
1464	IN6_MULTI_UNLOCK();
1465
1466out_im6f_rollback:
1467	if (error)
1468		im6f_rollback(imf);
1469	else
1470		im6f_commit(imf);
1471
1472	im6f_reap(imf);
1473
1474out_in6p_locked:
1475	INP_WUNLOCK(inp);
1476	return (error);
1477}
1478
1479/*
1480 * Given an inpcb, return its multicast options structure pointer.  Accepts
1481 * an unlocked inpcb pointer, but will return it locked.  May sleep.
1482 *
1483 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1484 * SMPng: NOTE: Returns with the INP write lock held.
1485 */
1486static struct ip6_moptions *
1487in6p_findmoptions(struct inpcb *inp)
1488{
1489	struct ip6_moptions	 *imo;
1490	struct in6_multi		**immp;
1491	struct in6_mfilter	 *imfp;
1492	size_t			  idx;
1493
1494	INP_WLOCK(inp);
1495	if (inp->in6p_moptions != NULL)
1496		return (inp->in6p_moptions);
1497
1498	INP_WUNLOCK(inp);
1499
1500	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1501	immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1502	    M_WAITOK | M_ZERO);
1503	imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1504	    M_IN6MFILTER, M_WAITOK);
1505
1506	imo->im6o_multicast_ifp = NULL;
1507	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1508	imo->im6o_multicast_loop = in6_mcast_loop;
1509	imo->im6o_num_memberships = 0;
1510	imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1511	imo->im6o_membership = immp;
1512
1513	/* Initialize per-group source filters. */
1514	for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1515		im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1516	imo->im6o_mfilters = imfp;
1517
1518	INP_WLOCK(inp);
1519	if (inp->in6p_moptions != NULL) {
1520		free(imfp, M_IN6MFILTER);
1521		free(immp, M_IP6MOPTS);
1522		free(imo, M_IP6MOPTS);
1523		return (inp->in6p_moptions);
1524	}
1525	inp->in6p_moptions = imo;
1526	return (imo);
1527}
1528
1529/*
1530 * Discard the IPv6 multicast options (and source filters).
1531 *
1532 * SMPng: NOTE: assumes INP write lock is held.
1533 */
1534void
1535ip6_freemoptions(struct ip6_moptions *imo)
1536{
1537	struct in6_mfilter	*imf;
1538	size_t			 idx, nmships;
1539
1540	KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1541
1542	nmships = imo->im6o_num_memberships;
1543	for (idx = 0; idx < nmships; ++idx) {
1544		imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1545		if (imf)
1546			im6f_leave(imf);
1547		/* XXX this will thrash the lock(s) */
1548		(void)in6_mc_leave(imo->im6o_membership[idx], imf);
1549		if (imf)
1550			im6f_purge(imf);
1551	}
1552
1553	if (imo->im6o_mfilters)
1554		free(imo->im6o_mfilters, M_IN6MFILTER);
1555	free(imo->im6o_membership, M_IP6MOPTS);
1556	free(imo, M_IP6MOPTS);
1557}
1558
1559/*
1560 * Atomically get source filters on a socket for an IPv6 multicast group.
1561 * Called with INP lock held; returns with lock released.
1562 */
1563static int
1564in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1565{
1566	struct __msfilterreq	 msfr;
1567	sockunion_t		*gsa;
1568	struct ifnet		*ifp;
1569	struct ip6_moptions	*imo;
1570	struct in6_mfilter	*imf;
1571	struct ip6_msource	*ims;
1572	struct in6_msource	*lims;
1573	struct sockaddr_in6	*psin;
1574	struct sockaddr_storage	*ptss;
1575	struct sockaddr_storage	*tss;
1576	int			 error;
1577	size_t			 idx, nsrcs, ncsrcs;
1578
1579	INP_WLOCK_ASSERT(inp);
1580
1581	imo = inp->in6p_moptions;
1582	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1583
1584	INP_WUNLOCK(inp);
1585
1586	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1587	    sizeof(struct __msfilterreq));
1588	if (error)
1589		return (error);
1590
1591	if (msfr.msfr_group.ss_family != AF_INET6 ||
1592	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1593		return (EINVAL);
1594
1595	gsa = (sockunion_t *)&msfr.msfr_group;
1596	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1597		return (EINVAL);
1598
1599	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1600		return (EADDRNOTAVAIL);
1601	ifp = ifnet_byindex(msfr.msfr_ifindex);
1602	if (ifp == NULL)
1603		return (EADDRNOTAVAIL);
1604	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1605
1606	INP_WLOCK(inp);
1607
1608	/*
1609	 * Lookup group on the socket.
1610	 */
1611	idx = im6o_match_group(imo, ifp, &gsa->sa);
1612	if (idx == -1 || imo->im6o_mfilters == NULL) {
1613		INP_WUNLOCK(inp);
1614		return (EADDRNOTAVAIL);
1615	}
1616	imf = &imo->im6o_mfilters[idx];
1617
1618	/*
1619	 * Ignore memberships which are in limbo.
1620	 */
1621	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1622		INP_WUNLOCK(inp);
1623		return (EAGAIN);
1624	}
1625	msfr.msfr_fmode = imf->im6f_st[1];
1626
1627	/*
1628	 * If the user specified a buffer, copy out the source filter
1629	 * entries to userland gracefully.
1630	 * We only copy out the number of entries which userland
1631	 * has asked for, but we always tell userland how big the
1632	 * buffer really needs to be.
1633	 */
1634	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1635		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1636	tss = NULL;
1637	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1638		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1639		    M_TEMP, M_NOWAIT | M_ZERO);
1640		if (tss == NULL) {
1641			INP_WUNLOCK(inp);
1642			return (ENOBUFS);
1643		}
1644	}
1645
1646	/*
1647	 * Count number of sources in-mode at t0.
1648	 * If buffer space exists and remains, copy out source entries.
1649	 */
1650	nsrcs = msfr.msfr_nsrcs;
1651	ncsrcs = 0;
1652	ptss = tss;
1653	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1654		lims = (struct in6_msource *)ims;
1655		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1656		    lims->im6sl_st[0] != imf->im6f_st[0])
1657			continue;
1658		++ncsrcs;
1659		if (tss != NULL && nsrcs > 0) {
1660			psin = (struct sockaddr_in6 *)ptss;
1661			psin->sin6_family = AF_INET6;
1662			psin->sin6_len = sizeof(struct sockaddr_in6);
1663			psin->sin6_addr = lims->im6s_addr;
1664			psin->sin6_port = 0;
1665			--nsrcs;
1666			++ptss;
1667		}
1668	}
1669
1670	INP_WUNLOCK(inp);
1671
1672	if (tss != NULL) {
1673		error = copyout(tss, msfr.msfr_srcs,
1674		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1675		free(tss, M_TEMP);
1676		if (error)
1677			return (error);
1678	}
1679
1680	msfr.msfr_nsrcs = ncsrcs;
1681	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1682
1683	return (error);
1684}
1685
1686/*
1687 * Return the IP multicast options in response to user getsockopt().
1688 */
1689int
1690ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1691{
1692	struct ip6_moptions	*im6o;
1693	int			 error;
1694	u_int			 optval;
1695
1696	INP_WLOCK(inp);
1697	im6o = inp->in6p_moptions;
1698	/*
1699	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1700	 * or is a divert socket, reject it.
1701	 */
1702	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1703	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1704	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1705		INP_WUNLOCK(inp);
1706		return (EOPNOTSUPP);
1707	}
1708
1709	error = 0;
1710	switch (sopt->sopt_name) {
1711	case IPV6_MULTICAST_IF:
1712		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1713			optval = 0;
1714		} else {
1715			optval = im6o->im6o_multicast_ifp->if_index;
1716		}
1717		INP_WUNLOCK(inp);
1718		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1719		break;
1720
1721	case IPV6_MULTICAST_HOPS:
1722		if (im6o == NULL)
1723			optval = V_ip6_defmcasthlim;
1724		else
1725			optval = im6o->im6o_multicast_hlim;
1726		INP_WUNLOCK(inp);
1727		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1728		break;
1729
1730	case IPV6_MULTICAST_LOOP:
1731		if (im6o == NULL)
1732			optval = in6_mcast_loop; /* XXX VIMAGE */
1733		else
1734			optval = im6o->im6o_multicast_loop;
1735		INP_WUNLOCK(inp);
1736		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1737		break;
1738
1739	case IPV6_MSFILTER:
1740		if (im6o == NULL) {
1741			error = EADDRNOTAVAIL;
1742			INP_WUNLOCK(inp);
1743		} else {
1744			error = in6p_get_source_filters(inp, sopt);
1745		}
1746		break;
1747
1748	default:
1749		INP_WUNLOCK(inp);
1750		error = ENOPROTOOPT;
1751		break;
1752	}
1753
1754	INP_UNLOCK_ASSERT(inp);
1755
1756	return (error);
1757}
1758
1759/*
1760 * Look up the ifnet to use for a multicast group membership,
1761 * given the address of an IPv6 group.
1762 *
1763 * This routine exists to support legacy IPv6 multicast applications.
1764 *
1765 * If inp is non-NULL, use this socket's current FIB number for any
1766 * required FIB lookup. Look up the group address in the unicast FIB,
1767 * and use its ifp; usually, this points to the default next-hop.
1768 * If the FIB lookup fails, return NULL.
1769 *
1770 * FUTURE: Support multiple forwarding tables for IPv6.
1771 *
1772 * Returns NULL if no ifp could be found.
1773 */
1774static struct ifnet *
1775in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1776    const struct sockaddr_in6 *gsin6)
1777{
1778	struct route_in6	 ro6;
1779	struct ifnet		*ifp;
1780
1781	KASSERT(in6p->inp_vflag & INP_IPV6,
1782	    ("%s: not INP_IPV6 inpcb", __func__));
1783	KASSERT(gsin6->sin6_family == AF_INET6,
1784	    ("%s: not AF_INET6 group", __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 (V_if_index < mreq.ipv6mr_interface)
1854				return (EADDRNOTAVAIL);
1855			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1856		}
1857		CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1858		    __func__, mreq.ipv6mr_interface, ifp);
1859	} break;
1860
1861	case MCAST_JOIN_GROUP:
1862	case MCAST_JOIN_SOURCE_GROUP:
1863		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1864			error = sooptcopyin(sopt, &gsr,
1865			    sizeof(struct group_req),
1866			    sizeof(struct group_req));
1867		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1868			error = sooptcopyin(sopt, &gsr,
1869			    sizeof(struct group_source_req),
1870			    sizeof(struct group_source_req));
1871		}
1872		if (error)
1873			return (error);
1874
1875		if (gsa->sin6.sin6_family != AF_INET6 ||
1876		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1877			return (EINVAL);
1878
1879		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1880			if (ssa->sin6.sin6_family != AF_INET6 ||
1881			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1882				return (EINVAL);
1883			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1884				return (EINVAL);
1885			/*
1886			 * TODO: Validate embedded scope ID in source
1887			 * list entry against passed-in ifp, if and only
1888			 * if source list filter entry is iface or node local.
1889			 */
1890			in6_clearscope(&ssa->sin6.sin6_addr);
1891			ssa->sin6.sin6_port = 0;
1892			ssa->sin6.sin6_scope_id = 0;
1893		}
1894
1895		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1896			return (EADDRNOTAVAIL);
1897		ifp = ifnet_byindex(gsr.gsr_interface);
1898		break;
1899
1900	default:
1901		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1902		    __func__, sopt->sopt_name);
1903		return (EOPNOTSUPP);
1904		break;
1905	}
1906
1907	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1908		return (EINVAL);
1909
1910	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1911		return (EADDRNOTAVAIL);
1912
1913	gsa->sin6.sin6_port = 0;
1914	gsa->sin6.sin6_scope_id = 0;
1915
1916	/*
1917	 * Always set the scope zone ID on memberships created from userland.
1918	 * Use the passed-in ifp to do this.
1919	 * XXX The in6_setscope() return value is meaningless.
1920	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1921	 */
1922	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1923
1924	imo = in6p_findmoptions(inp);
1925	idx = im6o_match_group(imo, ifp, &gsa->sa);
1926	if (idx == -1) {
1927		is_new = 1;
1928	} else {
1929		inm = imo->im6o_membership[idx];
1930		imf = &imo->im6o_mfilters[idx];
1931		if (ssa->ss.ss_family != AF_UNSPEC) {
1932			/*
1933			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1934			 * is an error. On an existing inclusive membership,
1935			 * it just adds the source to the filter list.
1936			 */
1937			if (imf->im6f_st[1] != MCAST_INCLUDE) {
1938				error = EINVAL;
1939				goto out_in6p_locked;
1940			}
1941			/*
1942			 * Throw out duplicates.
1943			 *
1944			 * XXX FIXME: This makes a naive assumption that
1945			 * even if entries exist for *ssa in this imf,
1946			 * they will be rejected as dupes, even if they
1947			 * are not valid in the current mode (in-mode).
1948			 *
1949			 * in6_msource is transactioned just as for anything
1950			 * else in SSM -- but note naive use of in6m_graft()
1951			 * below for allocating new filter entries.
1952			 *
1953			 * This is only an issue if someone mixes the
1954			 * full-state SSM API with the delta-based API,
1955			 * which is discouraged in the relevant RFCs.
1956			 */
1957			lims = im6o_match_source(imo, idx, &ssa->sa);
1958			if (lims != NULL /*&&
1959			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
1960				error = EADDRNOTAVAIL;
1961				goto out_in6p_locked;
1962			}
1963		} else {
1964			/*
1965			 * MCAST_JOIN_GROUP alone, on any existing membership,
1966			 * is rejected, to stop the same inpcb tying up
1967			 * multiple refs to the in_multi.
1968			 * On an existing inclusive membership, this is also
1969			 * an error; if you want to change filter mode,
1970			 * you must use the userland API setsourcefilter().
1971			 * XXX We don't reject this for imf in UNDEFINED
1972			 * state at t1, because allocation of a filter
1973			 * is atomic with allocation of a membership.
1974			 */
1975			error = EINVAL;
1976			goto out_in6p_locked;
1977		}
1978	}
1979
1980	/*
1981	 * Begin state merge transaction at socket layer.
1982	 */
1983	INP_WLOCK_ASSERT(inp);
1984
1985	if (is_new) {
1986		if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
1987			error = im6o_grow(imo);
1988			if (error)
1989				goto out_in6p_locked;
1990		}
1991		/*
1992		 * Allocate the new slot upfront so we can deal with
1993		 * grafting the new source filter in same code path
1994		 * as for join-source on existing membership.
1995		 */
1996		idx = imo->im6o_num_memberships;
1997		imo->im6o_membership[idx] = NULL;
1998		imo->im6o_num_memberships++;
1999		KASSERT(imo->im6o_mfilters != NULL,
2000		    ("%s: im6f_mfilters vector was not allocated", __func__));
2001		imf = &imo->im6o_mfilters[idx];
2002		KASSERT(RB_EMPTY(&imf->im6f_sources),
2003		    ("%s: im6f_sources not empty", __func__));
2004	}
2005
2006	/*
2007	 * Graft new source into filter list for this inpcb's
2008	 * membership of the group. The in6_multi may not have
2009	 * been allocated yet if this is a new membership, however,
2010	 * the in_mfilter slot will be allocated and must be initialized.
2011	 *
2012	 * Note: Grafting of exclusive mode filters doesn't happen
2013	 * in this path.
2014	 * XXX: Should check for non-NULL lims (node exists but may
2015	 * not be in-mode) for interop with full-state API.
2016	 */
2017	if (ssa->ss.ss_family != AF_UNSPEC) {
2018		/* Membership starts in IN mode */
2019		if (is_new) {
2020			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2021			im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2022		} else {
2023			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2024		}
2025		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2026		if (lims == NULL) {
2027			CTR1(KTR_MLD, "%s: merge imf state failed",
2028			    __func__);
2029			error = ENOMEM;
2030			goto out_im6o_free;
2031		}
2032	} else {
2033		/* No address specified; Membership starts in EX mode */
2034		if (is_new) {
2035			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2036			im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2037		}
2038	}
2039
2040	/*
2041	 * Begin state merge transaction at MLD layer.
2042	 */
2043	IN6_MULTI_LOCK();
2044
2045	if (is_new) {
2046		error = in6_mc_join_locked(ifp, &gsa->sin6.sin6_addr, imf,
2047		    &inm, 0);
2048		if (error) {
2049			IN6_MULTI_UNLOCK();
2050			goto out_im6o_free;
2051		}
2052		imo->im6o_membership[idx] = inm;
2053	} else {
2054		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2055		error = in6m_merge(inm, imf);
2056		if (error)
2057			CTR1(KTR_MLD, "%s: failed to merge inm state",
2058			    __func__);
2059		else {
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		}
2066	}
2067
2068	IN6_MULTI_UNLOCK();
2069	INP_WLOCK_ASSERT(inp);
2070	if (error) {
2071		im6f_rollback(imf);
2072		if (is_new)
2073			im6f_purge(imf);
2074		else
2075			im6f_reap(imf);
2076	} else {
2077		im6f_commit(imf);
2078	}
2079
2080out_im6o_free:
2081	if (error && is_new) {
2082		imo->im6o_membership[idx] = NULL;
2083		--imo->im6o_num_memberships;
2084	}
2085
2086out_in6p_locked:
2087	INP_WUNLOCK(inp);
2088	return (error);
2089}
2090
2091/*
2092 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2093 */
2094static int
2095in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2096{
2097	struct ipv6_mreq		 mreq;
2098	struct group_source_req		 gsr;
2099	sockunion_t			*gsa, *ssa;
2100	struct ifnet			*ifp;
2101	struct in6_mfilter		*imf;
2102	struct ip6_moptions		*imo;
2103	struct in6_msource		*ims;
2104	struct in6_multi		*inm;
2105	uint32_t			 ifindex;
2106	size_t				 idx;
2107	int				 error, is_final;
2108#ifdef KTR
2109	char				 ip6tbuf[INET6_ADDRSTRLEN];
2110#endif
2111
2112	ifp = NULL;
2113	ifindex = 0;
2114	error = 0;
2115	is_final = 1;
2116
2117	memset(&gsr, 0, sizeof(struct group_source_req));
2118	gsa = (sockunion_t *)&gsr.gsr_group;
2119	gsa->ss.ss_family = AF_UNSPEC;
2120	ssa = (sockunion_t *)&gsr.gsr_source;
2121	ssa->ss.ss_family = AF_UNSPEC;
2122
2123	/*
2124	 * Chew everything passed in up into a struct group_source_req
2125	 * as that is easier to process.
2126	 * Note: Any embedded scope ID in the multicast group passed
2127	 * in by userland is ignored, the interface index is the recommended
2128	 * mechanism to specify an interface; see below.
2129	 */
2130	switch (sopt->sopt_name) {
2131	case IPV6_LEAVE_GROUP:
2132		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2133		    sizeof(struct ipv6_mreq));
2134		if (error)
2135			return (error);
2136		gsa->sin6.sin6_family = AF_INET6;
2137		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2138		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2139		gsa->sin6.sin6_port = 0;
2140		gsa->sin6.sin6_scope_id = 0;
2141		ifindex = mreq.ipv6mr_interface;
2142		break;
2143
2144	case MCAST_LEAVE_GROUP:
2145	case MCAST_LEAVE_SOURCE_GROUP:
2146		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2147			error = sooptcopyin(sopt, &gsr,
2148			    sizeof(struct group_req),
2149			    sizeof(struct group_req));
2150		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2151			error = sooptcopyin(sopt, &gsr,
2152			    sizeof(struct group_source_req),
2153			    sizeof(struct group_source_req));
2154		}
2155		if (error)
2156			return (error);
2157
2158		if (gsa->sin6.sin6_family != AF_INET6 ||
2159		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2160			return (EINVAL);
2161		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2162			if (ssa->sin6.sin6_family != AF_INET6 ||
2163			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2164				return (EINVAL);
2165			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2166				return (EINVAL);
2167			/*
2168			 * TODO: Validate embedded scope ID in source
2169			 * list entry against passed-in ifp, if and only
2170			 * if source list filter entry is iface or node local.
2171			 */
2172			in6_clearscope(&ssa->sin6.sin6_addr);
2173		}
2174		gsa->sin6.sin6_port = 0;
2175		gsa->sin6.sin6_scope_id = 0;
2176		ifindex = gsr.gsr_interface;
2177		break;
2178
2179	default:
2180		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2181		    __func__, sopt->sopt_name);
2182		return (EOPNOTSUPP);
2183		break;
2184	}
2185
2186	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2187		return (EINVAL);
2188
2189	/*
2190	 * Validate interface index if provided. If no interface index
2191	 * was provided separately, attempt to look the membership up
2192	 * from the default scope as a last resort to disambiguate
2193	 * the membership we are being asked to leave.
2194	 * XXX SCOPE6 lock potentially taken here.
2195	 */
2196	if (ifindex != 0) {
2197		if (V_if_index < ifindex)
2198			return (EADDRNOTAVAIL);
2199		ifp = ifnet_byindex(ifindex);
2200		if (ifp == NULL)
2201			return (EADDRNOTAVAIL);
2202		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2203	} else {
2204		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2205		if (error)
2206			return (EADDRNOTAVAIL);
2207		/*
2208		 * Some badly behaved applications don't pass an ifindex
2209		 * or a scope ID, which is an API violation. In this case,
2210		 * perform a lookup as per a v6 join.
2211		 *
2212		 * XXX For now, stomp on zone ID for the corner case.
2213		 * This is not the 'KAME way', but we need to see the ifp
2214		 * directly until such time as this implementation is
2215		 * refactored, assuming the scope IDs are the way to go.
2216		 */
2217		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2218		if (ifindex == 0) {
2219			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2220			    "ifp for group %s.", __func__,
2221			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2222			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2223		} else {
2224			ifp = ifnet_byindex(ifindex);
2225		}
2226		if (ifp == NULL)
2227			return (EADDRNOTAVAIL);
2228	}
2229
2230	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2231	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2232
2233	/*
2234	 * Find the membership in the membership array.
2235	 */
2236	imo = in6p_findmoptions(inp);
2237	idx = im6o_match_group(imo, ifp, &gsa->sa);
2238	if (idx == -1) {
2239		error = EADDRNOTAVAIL;
2240		goto out_in6p_locked;
2241	}
2242	inm = imo->im6o_membership[idx];
2243	imf = &imo->im6o_mfilters[idx];
2244
2245	if (ssa->ss.ss_family != AF_UNSPEC)
2246		is_final = 0;
2247
2248	/*
2249	 * Begin state merge transaction at socket layer.
2250	 */
2251	INP_WLOCK_ASSERT(inp);
2252
2253	/*
2254	 * If we were instructed only to leave a given source, do so.
2255	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2256	 */
2257	if (is_final) {
2258		im6f_leave(imf);
2259	} else {
2260		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2261			error = EADDRNOTAVAIL;
2262			goto out_in6p_locked;
2263		}
2264		ims = im6o_match_source(imo, idx, &ssa->sa);
2265		if (ims == NULL) {
2266			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2267			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2268			    "not ");
2269			error = EADDRNOTAVAIL;
2270			goto out_in6p_locked;
2271		}
2272		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2273		error = im6f_prune(imf, &ssa->sin6);
2274		if (error) {
2275			CTR1(KTR_MLD, "%s: merge imf state failed",
2276			    __func__);
2277			goto out_in6p_locked;
2278		}
2279	}
2280
2281	/*
2282	 * Begin state merge transaction at MLD layer.
2283	 */
2284	IN6_MULTI_LOCK();
2285
2286	if (is_final) {
2287		/*
2288		 * Give up the multicast address record to which
2289		 * the membership points.
2290		 */
2291		(void)in6_mc_leave_locked(inm, imf);
2292	} else {
2293		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2294		error = in6m_merge(inm, imf);
2295		if (error)
2296			CTR1(KTR_MLD, "%s: failed to merge inm state",
2297			    __func__);
2298		else {
2299			CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2300			error = mld_change_state(inm, 0);
2301			if (error)
2302				CTR1(KTR_MLD, "%s: failed mld downcall",
2303				    __func__);
2304		}
2305	}
2306
2307	IN6_MULTI_UNLOCK();
2308
2309	if (error)
2310		im6f_rollback(imf);
2311	else
2312		im6f_commit(imf);
2313
2314	im6f_reap(imf);
2315
2316	if (is_final) {
2317		/* Remove the gap in the membership array. */
2318		for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2319			imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2320			imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2321		}
2322		imo->im6o_num_memberships--;
2323	}
2324
2325out_in6p_locked:
2326	INP_WUNLOCK(inp);
2327	return (error);
2328}
2329
2330/*
2331 * Select the interface for transmitting IPv6 multicast datagrams.
2332 *
2333 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2334 * may be passed to this socket option. An address of in6addr_any or an
2335 * interface index of 0 is used to remove a previous selection.
2336 * When no interface is selected, one is chosen for every send.
2337 */
2338static int
2339in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2340{
2341	struct ifnet		*ifp;
2342	struct ip6_moptions	*imo;
2343	u_int			 ifindex;
2344	int			 error;
2345
2346	if (sopt->sopt_valsize != sizeof(u_int))
2347		return (EINVAL);
2348
2349	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2350	if (error)
2351		return (error);
2352	if (V_if_index < ifindex)
2353		return (EINVAL);
2354	if (ifindex == 0)
2355		ifp = NULL;
2356	else {
2357		ifp = ifnet_byindex(ifindex);
2358		if (ifp == NULL)
2359			return (EINVAL);
2360		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2361			return (EADDRNOTAVAIL);
2362	}
2363	imo = in6p_findmoptions(inp);
2364	imo->im6o_multicast_ifp = ifp;
2365	INP_WUNLOCK(inp);
2366
2367	return (0);
2368}
2369
2370/*
2371 * Atomically set source filters on a socket for an IPv6 multicast group.
2372 *
2373 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2374 */
2375static int
2376in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2377{
2378	struct __msfilterreq	 msfr;
2379	sockunion_t		*gsa;
2380	struct ifnet		*ifp;
2381	struct in6_mfilter	*imf;
2382	struct ip6_moptions	*imo;
2383	struct in6_multi		*inm;
2384	size_t			 idx;
2385	int			 error;
2386
2387	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2388	    sizeof(struct __msfilterreq));
2389	if (error)
2390		return (error);
2391
2392	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2393		return (ENOBUFS);
2394
2395	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2396	    msfr.msfr_fmode != MCAST_INCLUDE)
2397		return (EINVAL);
2398
2399	if (msfr.msfr_group.ss_family != AF_INET6 ||
2400	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2401		return (EINVAL);
2402
2403	gsa = (sockunion_t *)&msfr.msfr_group;
2404	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2405		return (EINVAL);
2406
2407	gsa->sin6.sin6_port = 0;	/* ignore port */
2408
2409	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2410		return (EADDRNOTAVAIL);
2411	ifp = ifnet_byindex(msfr.msfr_ifindex);
2412	if (ifp == NULL)
2413		return (EADDRNOTAVAIL);
2414	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2415
2416	/*
2417	 * Take the INP write lock.
2418	 * Check if this socket is a member of this group.
2419	 */
2420	imo = in6p_findmoptions(inp);
2421	idx = im6o_match_group(imo, ifp, &gsa->sa);
2422	if (idx == -1 || imo->im6o_mfilters == NULL) {
2423		error = EADDRNOTAVAIL;
2424		goto out_in6p_locked;
2425	}
2426	inm = imo->im6o_membership[idx];
2427	imf = &imo->im6o_mfilters[idx];
2428
2429	/*
2430	 * Begin state merge transaction at socket layer.
2431	 */
2432	INP_WLOCK_ASSERT(inp);
2433
2434	imf->im6f_st[1] = msfr.msfr_fmode;
2435
2436	/*
2437	 * Apply any new source filters, if present.
2438	 * Make a copy of the user-space source vector so
2439	 * that we may copy them with a single copyin. This
2440	 * allows us to deal with page faults up-front.
2441	 */
2442	if (msfr.msfr_nsrcs > 0) {
2443		struct in6_msource	*lims;
2444		struct sockaddr_in6	*psin;
2445		struct sockaddr_storage	*kss, *pkss;
2446		int			 i;
2447
2448		INP_WUNLOCK(inp);
2449
2450		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2451		    __func__, (unsigned long)msfr.msfr_nsrcs);
2452		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2453		    M_TEMP, M_WAITOK);
2454		error = copyin(msfr.msfr_srcs, kss,
2455		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2456		if (error) {
2457			free(kss, M_TEMP);
2458			return (error);
2459		}
2460
2461		INP_WLOCK(inp);
2462
2463		/*
2464		 * Mark all source filters as UNDEFINED at t1.
2465		 * Restore new group filter mode, as im6f_leave()
2466		 * will set it to INCLUDE.
2467		 */
2468		im6f_leave(imf);
2469		imf->im6f_st[1] = msfr.msfr_fmode;
2470
2471		/*
2472		 * Update socket layer filters at t1, lazy-allocating
2473		 * new entries. This saves a bunch of memory at the
2474		 * cost of one RB_FIND() per source entry; duplicate
2475		 * entries in the msfr_nsrcs vector are ignored.
2476		 * If we encounter an error, rollback transaction.
2477		 *
2478		 * XXX This too could be replaced with a set-symmetric
2479		 * difference like loop to avoid walking from root
2480		 * every time, as the key space is common.
2481		 */
2482		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2483			psin = (struct sockaddr_in6 *)pkss;
2484			if (psin->sin6_family != AF_INET6) {
2485				error = EAFNOSUPPORT;
2486				break;
2487			}
2488			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2489				error = EINVAL;
2490				break;
2491			}
2492			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2493				error = EINVAL;
2494				break;
2495			}
2496			/*
2497			 * TODO: Validate embedded scope ID in source
2498			 * list entry against passed-in ifp, if and only
2499			 * if source list filter entry is iface or node local.
2500			 */
2501			in6_clearscope(&psin->sin6_addr);
2502			error = im6f_get_source(imf, psin, &lims);
2503			if (error)
2504				break;
2505			lims->im6sl_st[1] = imf->im6f_st[1];
2506		}
2507		free(kss, M_TEMP);
2508	}
2509
2510	if (error)
2511		goto out_im6f_rollback;
2512
2513	INP_WLOCK_ASSERT(inp);
2514	IN6_MULTI_LOCK();
2515
2516	/*
2517	 * Begin state merge transaction at MLD layer.
2518	 */
2519	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2520	error = in6m_merge(inm, imf);
2521	if (error)
2522		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2523	else {
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
2530	IN6_MULTI_UNLOCK();
2531
2532out_im6f_rollback:
2533	if (error)
2534		im6f_rollback(imf);
2535	else
2536		im6f_commit(imf);
2537
2538	im6f_reap(imf);
2539
2540out_in6p_locked:
2541	INP_WUNLOCK(inp);
2542	return (error);
2543}
2544
2545/*
2546 * Set the IP multicast options in response to user setsockopt().
2547 *
2548 * Many of the socket options handled in this function duplicate the
2549 * functionality of socket options in the regular unicast API. However,
2550 * it is not possible to merge the duplicate code, because the idempotence
2551 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2552 * the effects of these options must be treated as separate and distinct.
2553 *
2554 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2555 */
2556int
2557ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2558{
2559	struct ip6_moptions	*im6o;
2560	int			 error;
2561
2562	error = 0;
2563
2564	/*
2565	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2566	 * or is a divert socket, reject it.
2567	 */
2568	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2569	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2570	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2571		return (EOPNOTSUPP);
2572
2573	switch (sopt->sopt_name) {
2574	case IPV6_MULTICAST_IF:
2575		error = in6p_set_multicast_if(inp, sopt);
2576		break;
2577
2578	case IPV6_MULTICAST_HOPS: {
2579		int hlim;
2580
2581		if (sopt->sopt_valsize != sizeof(int)) {
2582			error = EINVAL;
2583			break;
2584		}
2585		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2586		if (error)
2587			break;
2588		if (hlim < -1 || hlim > 255) {
2589			error = EINVAL;
2590			break;
2591		} else if (hlim == -1) {
2592			hlim = V_ip6_defmcasthlim;
2593		}
2594		im6o = in6p_findmoptions(inp);
2595		im6o->im6o_multicast_hlim = hlim;
2596		INP_WUNLOCK(inp);
2597		break;
2598	}
2599
2600	case IPV6_MULTICAST_LOOP: {
2601		u_int loop;
2602
2603		/*
2604		 * Set the loopback flag for outgoing multicast packets.
2605		 * Must be zero or one.
2606		 */
2607		if (sopt->sopt_valsize != sizeof(u_int)) {
2608			error = EINVAL;
2609			break;
2610		}
2611		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2612		if (error)
2613			break;
2614		if (loop > 1) {
2615			error = EINVAL;
2616			break;
2617		}
2618		im6o = in6p_findmoptions(inp);
2619		im6o->im6o_multicast_loop = loop;
2620		INP_WUNLOCK(inp);
2621		break;
2622	}
2623
2624	case IPV6_JOIN_GROUP:
2625	case MCAST_JOIN_GROUP:
2626	case MCAST_JOIN_SOURCE_GROUP:
2627		error = in6p_join_group(inp, sopt);
2628		break;
2629
2630	case IPV6_LEAVE_GROUP:
2631	case MCAST_LEAVE_GROUP:
2632	case MCAST_LEAVE_SOURCE_GROUP:
2633		error = in6p_leave_group(inp, sopt);
2634		break;
2635
2636	case MCAST_BLOCK_SOURCE:
2637	case MCAST_UNBLOCK_SOURCE:
2638		error = in6p_block_unblock_source(inp, sopt);
2639		break;
2640
2641	case IPV6_MSFILTER:
2642		error = in6p_set_source_filters(inp, sopt);
2643		break;
2644
2645	default:
2646		error = EOPNOTSUPP;
2647		break;
2648	}
2649
2650	INP_UNLOCK_ASSERT(inp);
2651
2652	return (error);
2653}
2654
2655/*
2656 * Expose MLD's multicast filter mode and source list(s) to userland,
2657 * keyed by (ifindex, group).
2658 * The filter mode is written out as a uint32_t, followed by
2659 * 0..n of struct in6_addr.
2660 * For use by ifmcstat(8).
2661 * SMPng: NOTE: unlocked read of ifindex space.
2662 */
2663static int
2664sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2665{
2666	struct in6_addr			 mcaddr;
2667	struct in6_addr			 src;
2668	struct ifnet			*ifp;
2669	struct ifmultiaddr		*ifma;
2670	struct in6_multi		*inm;
2671	struct ip6_msource		*ims;
2672	int				*name;
2673	int				 retval;
2674	u_int				 namelen;
2675	uint32_t			 fmode, ifindex;
2676#ifdef KTR
2677	char				 ip6tbuf[INET6_ADDRSTRLEN];
2678#endif
2679
2680	name = (int *)arg1;
2681	namelen = arg2;
2682
2683	if (req->newptr != NULL)
2684		return (EPERM);
2685
2686	/* int: ifindex + 4 * 32 bits of IPv6 address */
2687	if (namelen != 5)
2688		return (EINVAL);
2689
2690	ifindex = name[0];
2691	if (ifindex <= 0 || ifindex > V_if_index) {
2692		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2693		    __func__, ifindex);
2694		return (ENOENT);
2695	}
2696
2697	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2698	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2699		CTR2(KTR_MLD, "%s: group %s is not multicast",
2700		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2701		return (EINVAL);
2702	}
2703
2704	ifp = ifnet_byindex(ifindex);
2705	if (ifp == NULL) {
2706		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2707		    __func__, ifindex);
2708		return (ENOENT);
2709	}
2710	/*
2711	 * Internal MLD lookups require that scope/zone ID is set.
2712	 */
2713	(void)in6_setscope(&mcaddr, ifp, NULL);
2714
2715	retval = sysctl_wire_old_buffer(req,
2716	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2717	if (retval)
2718		return (retval);
2719
2720	IN6_MULTI_LOCK();
2721
2722	IF_ADDR_RLOCK(ifp);
2723	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2724		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2725		    ifma->ifma_protospec == NULL)
2726			continue;
2727		inm = (struct in6_multi *)ifma->ifma_protospec;
2728		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2729			continue;
2730		fmode = inm->in6m_st[1].iss_fmode;
2731		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2732		if (retval != 0)
2733			break;
2734		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2735			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2736			/*
2737			 * Only copy-out sources which are in-mode.
2738			 */
2739			if (fmode != im6s_get_mode(inm, ims, 1)) {
2740				CTR1(KTR_MLD, "%s: skip non-in-mode",
2741				    __func__);
2742				continue;
2743			}
2744			src = ims->im6s_addr;
2745			retval = SYSCTL_OUT(req, &src,
2746			    sizeof(struct in6_addr));
2747			if (retval != 0)
2748				break;
2749		}
2750	}
2751	IF_ADDR_RUNLOCK(ifp);
2752
2753	IN6_MULTI_UNLOCK();
2754
2755	return (retval);
2756}
2757
2758#ifdef KTR
2759
2760static const char *in6m_modestrs[] = { "un", "in", "ex" };
2761
2762static const char *
2763in6m_mode_str(const int mode)
2764{
2765
2766	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2767		return (in6m_modestrs[mode]);
2768	return ("??");
2769}
2770
2771static const char *in6m_statestrs[] = {
2772	"not-member",
2773	"silent",
2774	"idle",
2775	"lazy",
2776	"sleeping",
2777	"awakening",
2778	"query-pending",
2779	"sg-query-pending",
2780	"leaving"
2781};
2782
2783static const char *
2784in6m_state_str(const int state)
2785{
2786
2787	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2788		return (in6m_statestrs[state]);
2789	return ("??");
2790}
2791
2792/*
2793 * Dump an in6_multi structure to the console.
2794 */
2795void
2796in6m_print(const struct in6_multi *inm)
2797{
2798	int t;
2799	char ip6tbuf[INET6_ADDRSTRLEN];
2800
2801	if ((ktr_mask & KTR_MLD) == 0)
2802		return;
2803
2804	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2805	printf("addr %s ifp %p(%s) ifma %p\n",
2806	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2807	    inm->in6m_ifp,
2808	    inm->in6m_ifp->if_xname,
2809	    inm->in6m_ifma);
2810	printf("timer %u state %s refcount %u scq.len %u\n",
2811	    inm->in6m_timer,
2812	    in6m_state_str(inm->in6m_state),
2813	    inm->in6m_refcount,
2814	    inm->in6m_scq.ifq_len);
2815	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2816	    inm->in6m_mli,
2817	    inm->in6m_nsrc,
2818	    inm->in6m_sctimer,
2819	    inm->in6m_scrv);
2820	for (t = 0; t < 2; t++) {
2821		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2822		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2823		    inm->in6m_st[t].iss_asm,
2824		    inm->in6m_st[t].iss_ex,
2825		    inm->in6m_st[t].iss_in,
2826		    inm->in6m_st[t].iss_rec);
2827	}
2828	printf("%s: --- end in6m %p ---\n", __func__, inm);
2829}
2830
2831#else /* !KTR */
2832
2833void
2834in6m_print(const struct in6_multi *inm)
2835{
2836
2837}
2838
2839#endif /* KTR */
2840