xref: /freebsd/sys/netinet6/in6_mcast.c (revision 4f52dfbb8d6c4d446500c5b097e3806ec219fbd4)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2009 Bruce Simpson.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. The name of the author may not be used to endorse or promote
16  *    products derived from this software without specific prior written
17  *    permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*
33  * IPv6 multicast socket, group, and socket option processing module.
34  * Normative references: RFC 2292, RFC 3492, RFC 3542, RFC 3678, RFC 3810.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_inet6.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/gtaskqueue.h>
45 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/priv.h>
53 #include <sys/ktr.h>
54 #include <sys/tree.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/vnet.h>
61 
62 
63 #include <netinet/in.h>
64 #include <netinet/udp.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/udp_var.h>
68 #include <netinet6/in6_fib.h>
69 #include <netinet6/in6_var.h>
70 #include <netinet/ip6.h>
71 #include <netinet/icmp6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet/in_pcb.h>
74 #include <netinet/tcp_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet6/mld6_var.h>
77 #include <netinet6/scope6_var.h>
78 
79 #ifndef KTR_MLD
80 #define KTR_MLD KTR_INET6
81 #endif
82 
83 #ifndef __SOCKUNION_DECLARED
84 union sockunion {
85 	struct sockaddr_storage	ss;
86 	struct sockaddr		sa;
87 	struct sockaddr_dl	sdl;
88 	struct sockaddr_in6	sin6;
89 };
90 typedef union sockunion sockunion_t;
91 #define __SOCKUNION_DECLARED
92 #endif /* __SOCKUNION_DECLARED */
93 
94 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
95     "IPv6 multicast PCB-layer source filter");
96 MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
97 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
98 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
99     "IPv6 multicast MLD-layer source filter");
100 
101 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
102 
103 /*
104  * Locking:
105  * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
106  * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
107  *   it can be taken by code in net/if.c also.
108  * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
109  *
110  * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
111  * any need for in6_multi itself to be virtualized -- it is bound to an ifp
112  * anyway no matter what happens.
113  */
114 struct mtx in6_multi_list_mtx;
115 MTX_SYSINIT(in6_multi_mtx, &in6_multi_list_mtx, "in6_multi_list_mtx", MTX_DEF);
116 
117 struct mtx in6_multi_free_mtx;
118 MTX_SYSINIT(in6_multi_free_mtx, &in6_multi_free_mtx, "in6_multi_free_mtx", MTX_DEF);
119 
120 struct sx in6_multi_sx;
121 SX_SYSINIT(in6_multi_sx, &in6_multi_sx, "in6_multi_sx");
122 
123 
124 
125 static void	im6f_commit(struct in6_mfilter *);
126 static int	im6f_get_source(struct in6_mfilter *imf,
127 		    const struct sockaddr_in6 *psin,
128 		    struct in6_msource **);
129 static struct in6_msource *
130 		im6f_graft(struct in6_mfilter *, const uint8_t,
131 		    const struct sockaddr_in6 *);
132 static void	im6f_leave(struct in6_mfilter *);
133 static int	im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
134 static void	im6f_purge(struct in6_mfilter *);
135 static void	im6f_rollback(struct in6_mfilter *);
136 static void	im6f_reap(struct in6_mfilter *);
137 static int	im6o_grow(struct ip6_moptions *);
138 static size_t	im6o_match_group(const struct ip6_moptions *,
139 		    const struct ifnet *, const struct sockaddr *);
140 static struct in6_msource *
141 		im6o_match_source(const struct ip6_moptions *, const size_t,
142 		    const struct sockaddr *);
143 static void	im6s_merge(struct ip6_msource *ims,
144 		    const struct in6_msource *lims, const int rollback);
145 static int	in6_getmulti(struct ifnet *, const struct in6_addr *,
146 		    struct in6_multi **);
147 static int	in6m_get_source(struct in6_multi *inm,
148 		    const struct in6_addr *addr, const int noalloc,
149 		    struct ip6_msource **pims);
150 #ifdef KTR
151 static int	in6m_is_ifp_detached(const struct in6_multi *);
152 #endif
153 static int	in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
154 static void	in6m_purge(struct in6_multi *);
155 static void	in6m_reap(struct in6_multi *);
156 static struct ip6_moptions *
157 		in6p_findmoptions(struct inpcb *);
158 static int	in6p_get_source_filters(struct inpcb *, struct sockopt *);
159 static int	in6p_join_group(struct inpcb *, struct sockopt *);
160 static int	in6p_leave_group(struct inpcb *, struct sockopt *);
161 static struct ifnet *
162 		in6p_lookup_mcast_ifp(const struct inpcb *,
163 		    const struct sockaddr_in6 *);
164 static int	in6p_block_unblock_source(struct inpcb *, struct sockopt *);
165 static int	in6p_set_multicast_if(struct inpcb *, struct sockopt *);
166 static int	in6p_set_source_filters(struct inpcb *, struct sockopt *);
167 static int	sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
168 
169 SYSCTL_DECL(_net_inet6_ip6);	/* XXX Not in any common header. */
170 
171 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0,
172     "IPv6 multicast");
173 
174 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
175 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
176     CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
177     "Max source filters per group");
178 
179 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
180 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
181     CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
182     "Max source filters per socket");
183 
184 /* TODO Virtualize this switch. */
185 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
186 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
187     &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
188 
189 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
190     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
191     "Per-interface stack-wide source filters");
192 
193 int ifma6_restart = 0;
194 #ifdef KTR
195 /*
196  * Inline function which wraps assertions for a valid ifp.
197  * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
198  * is detached.
199  */
200 static int __inline
201 in6m_is_ifp_detached(const struct in6_multi *inm)
202 {
203 	struct ifnet *ifp;
204 
205 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
206 	ifp = inm->in6m_ifma->ifma_ifp;
207 	if (ifp != NULL) {
208 		/*
209 		 * Sanity check that network-layer notion of ifp is the
210 		 * same as that of link-layer.
211 		 */
212 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
213 	}
214 
215 	return (ifp == NULL);
216 }
217 #endif
218 
219 /*
220  * Initialize an in6_mfilter structure to a known state at t0, t1
221  * with an empty source filter list.
222  */
223 static __inline void
224 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
225 {
226 	memset(imf, 0, sizeof(struct in6_mfilter));
227 	RB_INIT(&imf->im6f_sources);
228 	imf->im6f_st[0] = st0;
229 	imf->im6f_st[1] = st1;
230 }
231 
232 /*
233  * Resize the ip6_moptions vector to the next power-of-two minus 1.
234  * May be called with locks held; do not sleep.
235  */
236 static int
237 im6o_grow(struct ip6_moptions *imo)
238 {
239 	struct in6_multi	**nmships;
240 	struct in6_multi	**omships;
241 	struct in6_mfilter	 *nmfilters;
242 	struct in6_mfilter	 *omfilters;
243 	size_t			  idx;
244 	size_t			  newmax;
245 	size_t			  oldmax;
246 
247 	nmships = NULL;
248 	nmfilters = NULL;
249 	omships = imo->im6o_membership;
250 	omfilters = imo->im6o_mfilters;
251 	oldmax = imo->im6o_max_memberships;
252 	newmax = ((oldmax + 1) * 2) - 1;
253 
254 	if (newmax <= IPV6_MAX_MEMBERSHIPS) {
255 		nmships = (struct in6_multi **)realloc(omships,
256 		    sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
257 		nmfilters = (struct in6_mfilter *)realloc(omfilters,
258 		    sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
259 		    M_NOWAIT);
260 		if (nmships != NULL && nmfilters != NULL) {
261 			/* Initialize newly allocated source filter heads. */
262 			for (idx = oldmax; idx < newmax; idx++) {
263 				im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
264 				    MCAST_EXCLUDE);
265 			}
266 			imo->im6o_max_memberships = newmax;
267 			imo->im6o_membership = nmships;
268 			imo->im6o_mfilters = nmfilters;
269 		}
270 	}
271 
272 	if (nmships == NULL || nmfilters == NULL) {
273 		if (nmships != NULL)
274 			free(nmships, M_IP6MOPTS);
275 		if (nmfilters != NULL)
276 			free(nmfilters, M_IN6MFILTER);
277 		return (ETOOMANYREFS);
278 	}
279 
280 	return (0);
281 }
282 
283 /*
284  * Find an IPv6 multicast group entry for this ip6_moptions instance
285  * which matches the specified group, and optionally an interface.
286  * Return its index into the array, or -1 if not found.
287  */
288 static size_t
289 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
290     const struct sockaddr *group)
291 {
292 	const struct sockaddr_in6 *gsin6;
293 	struct in6_multi	**pinm;
294 	int		  idx;
295 	int		  nmships;
296 
297 	gsin6 = (const struct sockaddr_in6 *)group;
298 
299 	/* The im6o_membership array may be lazy allocated. */
300 	if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
301 		return (-1);
302 
303 	nmships = imo->im6o_num_memberships;
304 	pinm = &imo->im6o_membership[0];
305 	for (idx = 0; idx < nmships; idx++, pinm++) {
306 		if (*pinm == NULL)
307 			continue;
308 		if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
309 		    IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
310 		    &gsin6->sin6_addr)) {
311 			break;
312 		}
313 	}
314 	if (idx >= nmships)
315 		idx = -1;
316 
317 	return (idx);
318 }
319 
320 /*
321  * Find an IPv6 multicast source entry for this imo which matches
322  * the given group index for this socket, and source address.
323  *
324  * XXX TODO: The scope ID, if present in src, is stripped before
325  * any comparison. We SHOULD enforce scope/zone checks where the source
326  * filter entry has a link scope.
327  *
328  * NOTE: This does not check if the entry is in-mode, merely if
329  * it exists, which may not be the desired behaviour.
330  */
331 static struct in6_msource *
332 im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
333     const struct sockaddr *src)
334 {
335 	struct ip6_msource	 find;
336 	struct in6_mfilter	*imf;
337 	struct ip6_msource	*ims;
338 	const sockunion_t	*psa;
339 
340 	KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
341 	KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
342 	    ("%s: invalid index %d\n", __func__, (int)gidx));
343 
344 	/* The im6o_mfilters array may be lazy allocated. */
345 	if (imo->im6o_mfilters == NULL)
346 		return (NULL);
347 	imf = &imo->im6o_mfilters[gidx];
348 
349 	psa = (const sockunion_t *)src;
350 	find.im6s_addr = psa->sin6.sin6_addr;
351 	in6_clearscope(&find.im6s_addr);		/* XXX */
352 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
353 
354 	return ((struct in6_msource *)ims);
355 }
356 
357 /*
358  * Perform filtering for multicast datagrams on a socket by group and source.
359  *
360  * Returns 0 if a datagram should be allowed through, or various error codes
361  * if the socket was not a member of the group, or the source was muted, etc.
362  */
363 int
364 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
365     const struct sockaddr *group, const struct sockaddr *src)
366 {
367 	size_t gidx;
368 	struct in6_msource *ims;
369 	int mode;
370 
371 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
372 
373 	gidx = im6o_match_group(imo, ifp, group);
374 	if (gidx == -1)
375 		return (MCAST_NOTGMEMBER);
376 
377 	/*
378 	 * Check if the source was included in an (S,G) join.
379 	 * Allow reception on exclusive memberships by default,
380 	 * reject reception on inclusive memberships by default.
381 	 * Exclude source only if an in-mode exclude filter exists.
382 	 * Include source only if an in-mode include filter exists.
383 	 * NOTE: We are comparing group state here at MLD t1 (now)
384 	 * with socket-layer t0 (since last downcall).
385 	 */
386 	mode = imo->im6o_mfilters[gidx].im6f_st[1];
387 	ims = im6o_match_source(imo, gidx, src);
388 
389 	if ((ims == NULL && mode == MCAST_INCLUDE) ||
390 	    (ims != NULL && ims->im6sl_st[0] != mode))
391 		return (MCAST_NOTSMEMBER);
392 
393 	return (MCAST_PASS);
394 }
395 
396 /*
397  * Find and return a reference to an in6_multi record for (ifp, group),
398  * and bump its reference count.
399  * If one does not exist, try to allocate it, and update link-layer multicast
400  * filters on ifp to listen for group.
401  * Assumes the IN6_MULTI lock is held across the call.
402  * Return 0 if successful, otherwise return an appropriate error code.
403  */
404 static int
405 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
406     struct in6_multi **pinm)
407 {
408 	struct sockaddr_in6	 gsin6;
409 	struct ifmultiaddr	*ifma;
410 	struct in6_multi	*inm;
411 	int			 error;
412 
413 	error = 0;
414 
415 	/*
416 	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
417 	 * if_addmulti() takes this mutex itself, so we must drop and
418 	 * re-acquire around the call.
419 	 */
420 	IN6_MULTI_LOCK_ASSERT();
421 	IN6_MULTI_LIST_LOCK();
422 	IF_ADDR_WLOCK(ifp);
423 	inm = in6m_lookup_locked(ifp, group);
424 	if (inm != NULL) {
425 		/*
426 		 * If we already joined this group, just bump the
427 		 * refcount and return it.
428 		 */
429 		KASSERT(inm->in6m_refcount >= 1,
430 		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
431 		in6m_acquire_locked(inm);
432 		*pinm = inm;
433 		goto out_locked;
434 	}
435 
436 	memset(&gsin6, 0, sizeof(gsin6));
437 	gsin6.sin6_family = AF_INET6;
438 	gsin6.sin6_len = sizeof(struct sockaddr_in6);
439 	gsin6.sin6_addr = *group;
440 
441 	/*
442 	 * Check if a link-layer group is already associated
443 	 * with this network-layer group on the given ifnet.
444 	 */
445 	IN6_MULTI_LIST_UNLOCK();
446 	IF_ADDR_WUNLOCK(ifp);
447 	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
448 	if (error != 0)
449 		return (error);
450 	IN6_MULTI_LIST_LOCK();
451 	IF_ADDR_WLOCK(ifp);
452 
453 	/*
454 	 * If something other than netinet6 is occupying the link-layer
455 	 * group, print a meaningful error message and back out of
456 	 * the allocation.
457 	 * Otherwise, bump the refcount on the existing network-layer
458 	 * group association and return it.
459 	 */
460 	if (ifma->ifma_protospec != NULL) {
461 		inm = (struct in6_multi *)ifma->ifma_protospec;
462 #ifdef INVARIANTS
463 		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
464 		    __func__));
465 		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
466 		    ("%s: ifma not AF_INET6", __func__));
467 		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
468 		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
469 		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
470 			panic("%s: ifma %p is inconsistent with %p (%p)",
471 			    __func__, ifma, inm, group);
472 #endif
473 		in6m_acquire_locked(inm);
474 		*pinm = inm;
475 		goto out_locked;
476 	}
477 
478 	IF_ADDR_WLOCK_ASSERT(ifp);
479 
480 	/*
481 	 * A new in6_multi record is needed; allocate and initialize it.
482 	 * We DO NOT perform an MLD join as the in6_ layer may need to
483 	 * push an initial source list down to MLD to support SSM.
484 	 *
485 	 * The initial source filter state is INCLUDE, {} as per the RFC.
486 	 * Pending state-changes per group are subject to a bounds check.
487 	 */
488 	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
489 	if (inm == NULL) {
490 		IN6_MULTI_LIST_UNLOCK();
491 		IF_ADDR_WUNLOCK(ifp);
492 		if_delmulti_ifma(ifma);
493 		return (ENOMEM);
494 	}
495 	inm->in6m_addr = *group;
496 	inm->in6m_ifp = ifp;
497 	inm->in6m_mli = MLD_IFINFO(ifp);
498 	inm->in6m_ifma = ifma;
499 	inm->in6m_refcount = 1;
500 	inm->in6m_state = MLD_NOT_MEMBER;
501 	mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
502 
503 	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
504 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
505 	RB_INIT(&inm->in6m_srcs);
506 
507 	ifma->ifma_protospec = inm;
508 	*pinm = inm;
509 
510  out_locked:
511 	IN6_MULTI_LIST_UNLOCK();
512 	IF_ADDR_WUNLOCK(ifp);
513 	return (error);
514 }
515 
516 /*
517  * Drop a reference to an in6_multi record.
518  *
519  * If the refcount drops to 0, free the in6_multi record and
520  * delete the underlying link-layer membership.
521  */
522 static void
523 in6m_release(struct in6_multi *inm)
524 {
525 	struct ifmultiaddr *ifma;
526 	struct ifnet *ifp;
527 
528 	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
529 
530 	MPASS(inm->in6m_refcount == 0);
531 	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
532 
533 	ifma = inm->in6m_ifma;
534 	ifp = inm->in6m_ifp;
535 	MPASS(ifma->ifma_llifma == NULL);
536 
537 	/* XXX this access is not covered by IF_ADDR_LOCK */
538 	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
539 	KASSERT(ifma->ifma_protospec == NULL,
540 	    ("%s: ifma_protospec != NULL", __func__));
541 
542 	if (ifp)
543 		CURVNET_SET(ifp->if_vnet);
544 	in6m_purge(inm);
545 	free(inm, M_IP6MADDR);
546 
547 	if_delmulti_ifma_flags(ifma, 1);
548 	if (ifp) {
549 		CURVNET_RESTORE();
550 		if_rele(ifp);
551 	}
552 }
553 
554 static struct grouptask free_gtask;
555 static struct in6_multi_head in6m_free_list;
556 static void in6m_release_task(void *arg __unused);
557 static void in6m_init(void)
558 {
559 	SLIST_INIT(&in6m_free_list);
560 	taskqgroup_config_gtask_init(NULL, &free_gtask, in6m_release_task, "in6m release task");
561 }
562 
563 SYSINIT(in6m_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
564 	in6m_init, NULL);
565 
566 
567 void
568 in6m_release_list_deferred(struct in6_multi_head *inmh)
569 {
570 	if (SLIST_EMPTY(inmh))
571 		return;
572 	mtx_lock(&in6_multi_free_mtx);
573 	SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
574 	mtx_unlock(&in6_multi_free_mtx);
575 	GROUPTASK_ENQUEUE(&free_gtask);
576 }
577 
578 void
579 in6m_disconnect(struct in6_multi *inm)
580 {
581 	struct ifnet *ifp;
582 	struct ifaddr *ifa;
583 	struct in6_ifaddr *ifa6;
584 	struct in6_multi_mship *imm;
585 	struct ifmultiaddr *ifma, *ll_ifma;
586 
587 	ifp = inm->in6m_ifp;
588 	IF_ADDR_WLOCK_ASSERT(ifp);
589 	ifma = inm->in6m_ifma;
590 
591 	if_ref(ifp);
592 	TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
593 	MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
594 	if ((ll_ifma = ifma->ifma_llifma) != NULL) {
595 		MPASS(ifma != ll_ifma);
596 		ifma->ifma_llifma = NULL;
597 		MPASS(ll_ifma->ifma_llifma == NULL);
598 		MPASS(ll_ifma->ifma_ifp == ifp);
599 		if (--ll_ifma->ifma_refcount == 0) {
600 			ifma6_restart = true;
601 			TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifma_link);
602 			MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
603 			if_freemulti(ll_ifma);
604 		}
605 	}
606 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
607 		if (ifa->ifa_addr->sa_family != AF_INET6)
608 			continue;
609 		ifa6 = (void *)ifa;
610 		LIST_FOREACH(imm, &ifa6->ia6_memberships, i6mm_chain) {
611 			if (inm == imm->i6mm_maddr) {
612 				LIST_REMOVE(imm, i6mm_chain);
613 				free(imm, M_IP6MADDR);
614 			}
615 		}
616 	}
617 }
618 
619 void
620 in6m_release_deferred(struct in6_multi *inm)
621 {
622 	struct in6_multi_head tmp;
623 
624 	IN6_MULTI_LIST_LOCK_ASSERT();
625 	KASSERT(inm->in6m_refcount > 0, ("refcount == %d inm: %p", inm->in6m_refcount, inm));
626 	if (--inm->in6m_refcount == 0) {
627 		in6m_disconnect(inm);
628 		SLIST_INIT(&tmp);
629 		inm->in6m_ifma->ifma_protospec = NULL;
630 		MPASS(inm->in6m_ifma->ifma_llifma == NULL);
631 		SLIST_INSERT_HEAD(&tmp, inm, in6m_nrele);
632 		in6m_release_list_deferred(&tmp);
633 	}
634 }
635 
636 static void
637 in6m_release_task(void *arg __unused)
638 {
639 	struct in6_multi_head in6m_free_tmp;
640 	struct in6_multi *inm, *tinm;
641 
642 	SLIST_INIT(&in6m_free_tmp);
643 	mtx_lock(&in6_multi_free_mtx);
644 	SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
645 	mtx_unlock(&in6_multi_free_mtx);
646 	IN6_MULTI_LOCK();
647 	SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
648 		SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
649 		in6m_release(inm);
650 	}
651 	IN6_MULTI_UNLOCK();
652 }
653 
654 /*
655  * Clear recorded source entries for a group.
656  * Used by the MLD code. Caller must hold the IN6_MULTI lock.
657  * FIXME: Should reap.
658  */
659 void
660 in6m_clear_recorded(struct in6_multi *inm)
661 {
662 	struct ip6_msource	*ims;
663 
664 	IN6_MULTI_LIST_LOCK_ASSERT();
665 
666 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
667 		if (ims->im6s_stp) {
668 			ims->im6s_stp = 0;
669 			--inm->in6m_st[1].iss_rec;
670 		}
671 	}
672 	KASSERT(inm->in6m_st[1].iss_rec == 0,
673 	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
674 }
675 
676 /*
677  * Record a source as pending for a Source-Group MLDv2 query.
678  * This lives here as it modifies the shared tree.
679  *
680  * inm is the group descriptor.
681  * naddr is the address of the source to record in network-byte order.
682  *
683  * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
684  * lazy-allocate a source node in response to an SG query.
685  * Otherwise, no allocation is performed. This saves some memory
686  * with the trade-off that the source will not be reported to the
687  * router if joined in the window between the query response and
688  * the group actually being joined on the local host.
689  *
690  * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
691  * This turns off the allocation of a recorded source entry if
692  * the group has not been joined.
693  *
694  * Return 0 if the source didn't exist or was already marked as recorded.
695  * Return 1 if the source was marked as recorded by this function.
696  * Return <0 if any error occurred (negated errno code).
697  */
698 int
699 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
700 {
701 	struct ip6_msource	 find;
702 	struct ip6_msource	*ims, *nims;
703 
704 	IN6_MULTI_LIST_LOCK_ASSERT();
705 
706 	find.im6s_addr = *addr;
707 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
708 	if (ims && ims->im6s_stp)
709 		return (0);
710 	if (ims == NULL) {
711 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
712 			return (-ENOSPC);
713 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
714 		    M_NOWAIT | M_ZERO);
715 		if (nims == NULL)
716 			return (-ENOMEM);
717 		nims->im6s_addr = find.im6s_addr;
718 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
719 		++inm->in6m_nsrc;
720 		ims = nims;
721 	}
722 
723 	/*
724 	 * Mark the source as recorded and update the recorded
725 	 * source count.
726 	 */
727 	++ims->im6s_stp;
728 	++inm->in6m_st[1].iss_rec;
729 
730 	return (1);
731 }
732 
733 /*
734  * Return a pointer to an in6_msource owned by an in6_mfilter,
735  * given its source address.
736  * Lazy-allocate if needed. If this is a new entry its filter state is
737  * undefined at t0.
738  *
739  * imf is the filter set being modified.
740  * addr is the source address.
741  *
742  * SMPng: May be called with locks held; malloc must not block.
743  */
744 static int
745 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
746     struct in6_msource **plims)
747 {
748 	struct ip6_msource	 find;
749 	struct ip6_msource	*ims, *nims;
750 	struct in6_msource	*lims;
751 	int			 error;
752 
753 	error = 0;
754 	ims = NULL;
755 	lims = NULL;
756 
757 	find.im6s_addr = psin->sin6_addr;
758 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
759 	lims = (struct in6_msource *)ims;
760 	if (lims == NULL) {
761 		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
762 			return (ENOSPC);
763 		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
764 		    M_NOWAIT | M_ZERO);
765 		if (nims == NULL)
766 			return (ENOMEM);
767 		lims = (struct in6_msource *)nims;
768 		lims->im6s_addr = find.im6s_addr;
769 		lims->im6sl_st[0] = MCAST_UNDEFINED;
770 		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
771 		++imf->im6f_nsrc;
772 	}
773 
774 	*plims = lims;
775 
776 	return (error);
777 }
778 
779 /*
780  * Graft a source entry into an existing socket-layer filter set,
781  * maintaining any required invariants and checking allocations.
782  *
783  * The source is marked as being in the new filter mode at t1.
784  *
785  * Return the pointer to the new node, otherwise return NULL.
786  */
787 static struct in6_msource *
788 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
789     const struct sockaddr_in6 *psin)
790 {
791 	struct ip6_msource	*nims;
792 	struct in6_msource	*lims;
793 
794 	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
795 	    M_NOWAIT | M_ZERO);
796 	if (nims == NULL)
797 		return (NULL);
798 	lims = (struct in6_msource *)nims;
799 	lims->im6s_addr = psin->sin6_addr;
800 	lims->im6sl_st[0] = MCAST_UNDEFINED;
801 	lims->im6sl_st[1] = st1;
802 	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
803 	++imf->im6f_nsrc;
804 
805 	return (lims);
806 }
807 
808 /*
809  * Prune a source entry from an existing socket-layer filter set,
810  * maintaining any required invariants and checking allocations.
811  *
812  * The source is marked as being left at t1, it is not freed.
813  *
814  * Return 0 if no error occurred, otherwise return an errno value.
815  */
816 static int
817 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
818 {
819 	struct ip6_msource	 find;
820 	struct ip6_msource	*ims;
821 	struct in6_msource	*lims;
822 
823 	find.im6s_addr = psin->sin6_addr;
824 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
825 	if (ims == NULL)
826 		return (ENOENT);
827 	lims = (struct in6_msource *)ims;
828 	lims->im6sl_st[1] = MCAST_UNDEFINED;
829 	return (0);
830 }
831 
832 /*
833  * Revert socket-layer filter set deltas at t1 to t0 state.
834  */
835 static void
836 im6f_rollback(struct in6_mfilter *imf)
837 {
838 	struct ip6_msource	*ims, *tims;
839 	struct in6_msource	*lims;
840 
841 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
842 		lims = (struct in6_msource *)ims;
843 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
844 			/* no change at t1 */
845 			continue;
846 		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
847 			/* revert change to existing source at t1 */
848 			lims->im6sl_st[1] = lims->im6sl_st[0];
849 		} else {
850 			/* revert source added t1 */
851 			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
852 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
853 			free(ims, M_IN6MFILTER);
854 			imf->im6f_nsrc--;
855 		}
856 	}
857 	imf->im6f_st[1] = imf->im6f_st[0];
858 }
859 
860 /*
861  * Mark socket-layer filter set as INCLUDE {} at t1.
862  */
863 static void
864 im6f_leave(struct in6_mfilter *imf)
865 {
866 	struct ip6_msource	*ims;
867 	struct in6_msource	*lims;
868 
869 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
870 		lims = (struct in6_msource *)ims;
871 		lims->im6sl_st[1] = MCAST_UNDEFINED;
872 	}
873 	imf->im6f_st[1] = MCAST_INCLUDE;
874 }
875 
876 /*
877  * Mark socket-layer filter set deltas as committed.
878  */
879 static void
880 im6f_commit(struct in6_mfilter *imf)
881 {
882 	struct ip6_msource	*ims;
883 	struct in6_msource	*lims;
884 
885 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
886 		lims = (struct in6_msource *)ims;
887 		lims->im6sl_st[0] = lims->im6sl_st[1];
888 	}
889 	imf->im6f_st[0] = imf->im6f_st[1];
890 }
891 
892 /*
893  * Reap unreferenced sources from socket-layer filter set.
894  */
895 static void
896 im6f_reap(struct in6_mfilter *imf)
897 {
898 	struct ip6_msource	*ims, *tims;
899 	struct in6_msource	*lims;
900 
901 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
902 		lims = (struct in6_msource *)ims;
903 		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
904 		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
905 			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
906 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
907 			free(ims, M_IN6MFILTER);
908 			imf->im6f_nsrc--;
909 		}
910 	}
911 }
912 
913 /*
914  * Purge socket-layer filter set.
915  */
916 static void
917 im6f_purge(struct in6_mfilter *imf)
918 {
919 	struct ip6_msource	*ims, *tims;
920 
921 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
922 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
923 		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
924 		free(ims, M_IN6MFILTER);
925 		imf->im6f_nsrc--;
926 	}
927 	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
928 	KASSERT(RB_EMPTY(&imf->im6f_sources),
929 	    ("%s: im6f_sources not empty", __func__));
930 }
931 
932 /*
933  * Look up a source filter entry for a multicast group.
934  *
935  * inm is the group descriptor to work with.
936  * addr is the IPv6 address to look up.
937  * noalloc may be non-zero to suppress allocation of sources.
938  * *pims will be set to the address of the retrieved or allocated source.
939  *
940  * SMPng: NOTE: may be called with locks held.
941  * Return 0 if successful, otherwise return a non-zero error code.
942  */
943 static int
944 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
945     const int noalloc, struct ip6_msource **pims)
946 {
947 	struct ip6_msource	 find;
948 	struct ip6_msource	*ims, *nims;
949 #ifdef KTR
950 	char			 ip6tbuf[INET6_ADDRSTRLEN];
951 #endif
952 
953 	find.im6s_addr = *addr;
954 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
955 	if (ims == NULL && !noalloc) {
956 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
957 			return (ENOSPC);
958 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
959 		    M_NOWAIT | M_ZERO);
960 		if (nims == NULL)
961 			return (ENOMEM);
962 		nims->im6s_addr = *addr;
963 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
964 		++inm->in6m_nsrc;
965 		ims = nims;
966 		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
967 		    ip6_sprintf(ip6tbuf, addr), ims);
968 	}
969 
970 	*pims = ims;
971 	return (0);
972 }
973 
974 /*
975  * Merge socket-layer source into MLD-layer source.
976  * If rollback is non-zero, perform the inverse of the merge.
977  */
978 static void
979 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
980     const int rollback)
981 {
982 	int n = rollback ? -1 : 1;
983 #ifdef KTR
984 	char ip6tbuf[INET6_ADDRSTRLEN];
985 
986 	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
987 #endif
988 
989 	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
990 		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
991 		ims->im6s_st[1].ex -= n;
992 	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
993 		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
994 		ims->im6s_st[1].in -= n;
995 	}
996 
997 	if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
998 		CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
999 		ims->im6s_st[1].ex += n;
1000 	} else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
1001 		CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
1002 		ims->im6s_st[1].in += n;
1003 	}
1004 }
1005 
1006 /*
1007  * Atomically update the global in6_multi state, when a membership's
1008  * filter list is being updated in any way.
1009  *
1010  * imf is the per-inpcb-membership group filter pointer.
1011  * A fake imf may be passed for in-kernel consumers.
1012  *
1013  * XXX This is a candidate for a set-symmetric-difference style loop
1014  * which would eliminate the repeated lookup from root of ims nodes,
1015  * as they share the same key space.
1016  *
1017  * If any error occurred this function will back out of refcounts
1018  * and return a non-zero value.
1019  */
1020 static int
1021 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1022 {
1023 	struct ip6_msource	*ims, *nims;
1024 	struct in6_msource	*lims;
1025 	int			 schanged, error;
1026 	int			 nsrc0, nsrc1;
1027 
1028 	schanged = 0;
1029 	error = 0;
1030 	nsrc1 = nsrc0 = 0;
1031 	IN6_MULTI_LIST_LOCK_ASSERT();
1032 
1033 	/*
1034 	 * Update the source filters first, as this may fail.
1035 	 * Maintain count of in-mode filters at t0, t1. These are
1036 	 * used to work out if we transition into ASM mode or not.
1037 	 * Maintain a count of source filters whose state was
1038 	 * actually modified by this operation.
1039 	 */
1040 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1041 		lims = (struct in6_msource *)ims;
1042 		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1043 		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1044 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1045 		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1046 		++schanged;
1047 		if (error)
1048 			break;
1049 		im6s_merge(nims, lims, 0);
1050 	}
1051 	if (error) {
1052 		struct ip6_msource *bims;
1053 
1054 		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1055 			lims = (struct in6_msource *)ims;
1056 			if (lims->im6sl_st[0] == lims->im6sl_st[1])
1057 				continue;
1058 			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1059 			if (bims == NULL)
1060 				continue;
1061 			im6s_merge(bims, lims, 1);
1062 		}
1063 		goto out_reap;
1064 	}
1065 
1066 	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1067 	    __func__, nsrc0, nsrc1);
1068 
1069 	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1070 	if (imf->im6f_st[0] == imf->im6f_st[1] &&
1071 	    imf->im6f_st[1] == MCAST_INCLUDE) {
1072 		if (nsrc1 == 0) {
1073 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1074 			--inm->in6m_st[1].iss_in;
1075 		}
1076 	}
1077 
1078 	/* Handle filter mode transition on socket. */
1079 	if (imf->im6f_st[0] != imf->im6f_st[1]) {
1080 		CTR3(KTR_MLD, "%s: imf transition %d to %d",
1081 		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
1082 
1083 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1084 			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1085 			--inm->in6m_st[1].iss_ex;
1086 		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1087 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1088 			--inm->in6m_st[1].iss_in;
1089 		}
1090 
1091 		if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1092 			CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1093 			inm->in6m_st[1].iss_ex++;
1094 		} else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1095 			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1096 			inm->in6m_st[1].iss_in++;
1097 		}
1098 	}
1099 
1100 	/*
1101 	 * Track inm filter state in terms of listener counts.
1102 	 * If there are any exclusive listeners, stack-wide
1103 	 * membership is exclusive.
1104 	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1105 	 * If no listeners remain, state is undefined at t1,
1106 	 * and the MLD lifecycle for this group should finish.
1107 	 */
1108 	if (inm->in6m_st[1].iss_ex > 0) {
1109 		CTR1(KTR_MLD, "%s: transition to EX", __func__);
1110 		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1111 	} else if (inm->in6m_st[1].iss_in > 0) {
1112 		CTR1(KTR_MLD, "%s: transition to IN", __func__);
1113 		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1114 	} else {
1115 		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1116 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1117 	}
1118 
1119 	/* Decrement ASM listener count on transition out of ASM mode. */
1120 	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1121 		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1122 		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1123 			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1124 			--inm->in6m_st[1].iss_asm;
1125 		}
1126 	}
1127 
1128 	/* Increment ASM listener count on transition to ASM mode. */
1129 	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1130 		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1131 		inm->in6m_st[1].iss_asm++;
1132 	}
1133 
1134 	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1135 	in6m_print(inm);
1136 
1137 out_reap:
1138 	if (schanged > 0) {
1139 		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1140 		in6m_reap(inm);
1141 	}
1142 	return (error);
1143 }
1144 
1145 /*
1146  * Mark an in6_multi's filter set deltas as committed.
1147  * Called by MLD after a state change has been enqueued.
1148  */
1149 void
1150 in6m_commit(struct in6_multi *inm)
1151 {
1152 	struct ip6_msource	*ims;
1153 
1154 	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1155 	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1156 	in6m_print(inm);
1157 
1158 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1159 		ims->im6s_st[0] = ims->im6s_st[1];
1160 	}
1161 	inm->in6m_st[0] = inm->in6m_st[1];
1162 }
1163 
1164 /*
1165  * Reap unreferenced nodes from an in6_multi's filter set.
1166  */
1167 static void
1168 in6m_reap(struct in6_multi *inm)
1169 {
1170 	struct ip6_msource	*ims, *tims;
1171 
1172 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1173 		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1174 		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1175 		    ims->im6s_stp != 0)
1176 			continue;
1177 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1178 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1179 		free(ims, M_IP6MSOURCE);
1180 		inm->in6m_nsrc--;
1181 	}
1182 }
1183 
1184 /*
1185  * Purge all source nodes from an in6_multi's filter set.
1186  */
1187 static void
1188 in6m_purge(struct in6_multi *inm)
1189 {
1190 	struct ip6_msource	*ims, *tims;
1191 
1192 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1193 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1194 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1195 		free(ims, M_IP6MSOURCE);
1196 		inm->in6m_nsrc--;
1197 	}
1198 	/* Free state-change requests that might be queued. */
1199 	mbufq_drain(&inm->in6m_scq);
1200 }
1201 
1202 /*
1203  * Join a multicast address w/o sources.
1204  * KAME compatibility entry point.
1205  *
1206  * SMPng: Assume no mc locks held by caller.
1207  */
1208 int
1209 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1210     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1211     const int delay)
1212 {
1213 	int error;
1214 
1215 	IN6_MULTI_LOCK();
1216 	error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1217 	IN6_MULTI_UNLOCK();
1218 	return (error);
1219 }
1220 
1221 /*
1222  * Join a multicast group; real entry point.
1223  *
1224  * Only preserves atomicity at inm level.
1225  * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1226  *
1227  * If the MLD downcall fails, the group is not joined, and an error
1228  * code is returned.
1229  */
1230 int
1231 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1232     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1233     const int delay)
1234 {
1235 	struct in6_mfilter	 timf;
1236 	struct in6_multi	*inm;
1237 	struct ifmultiaddr *ifma;
1238 	int			 error;
1239 #ifdef KTR
1240 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1241 #endif
1242 
1243 #ifdef INVARIANTS
1244 	/*
1245 	 * Sanity: Check scope zone ID was set for ifp, if and
1246 	 * only if group is scoped to an interface.
1247 	 */
1248 	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1249 	    ("%s: not a multicast address", __func__));
1250 	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1251 	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1252 		KASSERT(mcaddr->s6_addr16[1] != 0,
1253 		    ("%s: scope zone ID not set", __func__));
1254 	}
1255 #endif
1256 
1257 	IN6_MULTI_LOCK_ASSERT();
1258 	IN6_MULTI_LIST_UNLOCK_ASSERT();
1259 
1260 	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1261 	    ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1262 
1263 	error = 0;
1264 	inm = NULL;
1265 
1266 	/*
1267 	 * If no imf was specified (i.e. kernel consumer),
1268 	 * fake one up and assume it is an ASM join.
1269 	 */
1270 	if (imf == NULL) {
1271 		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1272 		imf = &timf;
1273 	}
1274 	error = in6_getmulti(ifp, mcaddr, &inm);
1275 	if (error) {
1276 		CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1277 		return (error);
1278 	}
1279 
1280 	IN6_MULTI_LIST_LOCK();
1281 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1282 	error = in6m_merge(inm, imf);
1283 	if (error) {
1284 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1285 		goto out_in6m_release;
1286 	}
1287 
1288 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1289 	error = mld_change_state(inm, delay);
1290 	if (error) {
1291 		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1292 		goto out_in6m_release;
1293 	}
1294 
1295 out_in6m_release:
1296 	if (error) {
1297 		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1298 		IF_ADDR_RLOCK(ifp);
1299 		TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1300 			if (ifma->ifma_protospec == inm) {
1301 				ifma->ifma_protospec = NULL;
1302 				break;
1303 			}
1304 		}
1305 		in6m_release_deferred(inm);
1306 		IF_ADDR_RUNLOCK(ifp);
1307 	} else {
1308 		*pinm = inm;
1309 	}
1310 	IN6_MULTI_LIST_UNLOCK();
1311 	return (error);
1312 }
1313 
1314 /*
1315  * Leave a multicast group; unlocked entry point.
1316  */
1317 int
1318 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1319 {
1320 	int error;
1321 
1322 	IN6_MULTI_LOCK();
1323 	error = in6_leavegroup_locked(inm, imf);
1324 	IN6_MULTI_UNLOCK();
1325 	return (error);
1326 }
1327 
1328 /*
1329  * Leave a multicast group; real entry point.
1330  * All source filters will be expunged.
1331  *
1332  * Only preserves atomicity at inm level.
1333  *
1334  * Holding the write lock for the INP which contains imf
1335  * is highly advisable. We can't assert for it as imf does not
1336  * contain a back-pointer to the owning inp.
1337  *
1338  * Note: This is not the same as in6m_release(*) as this function also
1339  * makes a state change downcall into MLD.
1340  */
1341 int
1342 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1343 {
1344 	struct in6_mfilter	 timf;
1345 	struct ifnet *ifp;
1346 	int			 error;
1347 #ifdef KTR
1348 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1349 #endif
1350 
1351 	error = 0;
1352 
1353 	IN6_MULTI_LOCK_ASSERT();
1354 
1355 	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1356 	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1357 	    (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1358 	    imf);
1359 
1360 	/*
1361 	 * If no imf was specified (i.e. kernel consumer),
1362 	 * fake one up and assume it is an ASM join.
1363 	 */
1364 	if (imf == NULL) {
1365 		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1366 		imf = &timf;
1367 	}
1368 
1369 	/*
1370 	 * Begin state merge transaction at MLD layer.
1371 	 *
1372 	 * As this particular invocation should not cause any memory
1373 	 * to be allocated, and there is no opportunity to roll back
1374 	 * the transaction, it MUST NOT fail.
1375 	 */
1376 
1377 	ifp = inm->in6m_ifp;
1378 	IN6_MULTI_LIST_LOCK();
1379 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1380 	error = in6m_merge(inm, imf);
1381 	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1382 
1383 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1384 	error = mld_change_state(inm, 0);
1385 	if (error)
1386 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1387 
1388 	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1389 	if (ifp)
1390 		IF_ADDR_WLOCK(ifp);
1391 	in6m_release_deferred(inm);
1392 	if (ifp)
1393 		IF_ADDR_WUNLOCK(ifp);
1394 	IN6_MULTI_LIST_UNLOCK();
1395 
1396 	return (error);
1397 }
1398 
1399 
1400 /*
1401  * Block or unblock an ASM multicast source on an inpcb.
1402  * This implements the delta-based API described in RFC 3678.
1403  *
1404  * The delta-based API applies only to exclusive-mode memberships.
1405  * An MLD downcall will be performed.
1406  *
1407  * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1408  *
1409  * Return 0 if successful, otherwise return an appropriate error code.
1410  */
1411 static int
1412 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1413 {
1414 	struct group_source_req		 gsr;
1415 	sockunion_t			*gsa, *ssa;
1416 	struct ifnet			*ifp;
1417 	struct in6_mfilter		*imf;
1418 	struct ip6_moptions		*imo;
1419 	struct in6_msource		*ims;
1420 	struct in6_multi			*inm;
1421 	size_t				 idx;
1422 	uint16_t			 fmode;
1423 	int				 error, doblock;
1424 #ifdef KTR
1425 	char				 ip6tbuf[INET6_ADDRSTRLEN];
1426 #endif
1427 
1428 	ifp = NULL;
1429 	error = 0;
1430 	doblock = 0;
1431 
1432 	memset(&gsr, 0, sizeof(struct group_source_req));
1433 	gsa = (sockunion_t *)&gsr.gsr_group;
1434 	ssa = (sockunion_t *)&gsr.gsr_source;
1435 
1436 	switch (sopt->sopt_name) {
1437 	case MCAST_BLOCK_SOURCE:
1438 	case MCAST_UNBLOCK_SOURCE:
1439 		error = sooptcopyin(sopt, &gsr,
1440 		    sizeof(struct group_source_req),
1441 		    sizeof(struct group_source_req));
1442 		if (error)
1443 			return (error);
1444 
1445 		if (gsa->sin6.sin6_family != AF_INET6 ||
1446 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1447 			return (EINVAL);
1448 
1449 		if (ssa->sin6.sin6_family != AF_INET6 ||
1450 		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1451 			return (EINVAL);
1452 
1453 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1454 			return (EADDRNOTAVAIL);
1455 
1456 		ifp = ifnet_byindex(gsr.gsr_interface);
1457 
1458 		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1459 			doblock = 1;
1460 		break;
1461 
1462 	default:
1463 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1464 		    __func__, sopt->sopt_name);
1465 		return (EOPNOTSUPP);
1466 		break;
1467 	}
1468 
1469 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1470 		return (EINVAL);
1471 
1472 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1473 
1474 	/*
1475 	 * Check if we are actually a member of this group.
1476 	 */
1477 	imo = in6p_findmoptions(inp);
1478 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1479 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1480 		error = EADDRNOTAVAIL;
1481 		goto out_in6p_locked;
1482 	}
1483 
1484 	KASSERT(imo->im6o_mfilters != NULL,
1485 	    ("%s: im6o_mfilters not allocated", __func__));
1486 	imf = &imo->im6o_mfilters[idx];
1487 	inm = imo->im6o_membership[idx];
1488 
1489 	/*
1490 	 * Attempting to use the delta-based API on an
1491 	 * non exclusive-mode membership is an error.
1492 	 */
1493 	fmode = imf->im6f_st[0];
1494 	if (fmode != MCAST_EXCLUDE) {
1495 		error = EINVAL;
1496 		goto out_in6p_locked;
1497 	}
1498 
1499 	/*
1500 	 * Deal with error cases up-front:
1501 	 *  Asked to block, but already blocked; or
1502 	 *  Asked to unblock, but nothing to unblock.
1503 	 * If adding a new block entry, allocate it.
1504 	 */
1505 	ims = im6o_match_source(imo, idx, &ssa->sa);
1506 	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1507 		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1508 		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1509 		    doblock ? "" : "not ");
1510 		error = EADDRNOTAVAIL;
1511 		goto out_in6p_locked;
1512 	}
1513 
1514 	INP_WLOCK_ASSERT(inp);
1515 
1516 	/*
1517 	 * Begin state merge transaction at socket layer.
1518 	 */
1519 	if (doblock) {
1520 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1521 		ims = im6f_graft(imf, fmode, &ssa->sin6);
1522 		if (ims == NULL)
1523 			error = ENOMEM;
1524 	} else {
1525 		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1526 		error = im6f_prune(imf, &ssa->sin6);
1527 	}
1528 
1529 	if (error) {
1530 		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1531 		goto out_im6f_rollback;
1532 	}
1533 
1534 	/*
1535 	 * Begin state merge transaction at MLD layer.
1536 	 */
1537 	IN6_MULTI_LIST_LOCK();
1538 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1539 	error = in6m_merge(inm, imf);
1540 	if (error)
1541 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1542 	else {
1543 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1544 		error = mld_change_state(inm, 0);
1545 		if (error)
1546 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1547 	}
1548 
1549 	IN6_MULTI_LIST_UNLOCK();
1550 
1551 out_im6f_rollback:
1552 	if (error)
1553 		im6f_rollback(imf);
1554 	else
1555 		im6f_commit(imf);
1556 
1557 	im6f_reap(imf);
1558 
1559 out_in6p_locked:
1560 	INP_WUNLOCK(inp);
1561 	return (error);
1562 }
1563 
1564 /*
1565  * Given an inpcb, return its multicast options structure pointer.  Accepts
1566  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1567  *
1568  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1569  * SMPng: NOTE: Returns with the INP write lock held.
1570  */
1571 static struct ip6_moptions *
1572 in6p_findmoptions(struct inpcb *inp)
1573 {
1574 	struct ip6_moptions	 *imo;
1575 	struct in6_multi		**immp;
1576 	struct in6_mfilter	 *imfp;
1577 	size_t			  idx;
1578 
1579 	INP_WLOCK(inp);
1580 	if (inp->in6p_moptions != NULL)
1581 		return (inp->in6p_moptions);
1582 
1583 	INP_WUNLOCK(inp);
1584 
1585 	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1586 	immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1587 	    M_WAITOK | M_ZERO);
1588 	imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1589 	    M_IN6MFILTER, M_WAITOK);
1590 
1591 	imo->im6o_multicast_ifp = NULL;
1592 	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1593 	imo->im6o_multicast_loop = in6_mcast_loop;
1594 	imo->im6o_num_memberships = 0;
1595 	imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1596 	imo->im6o_membership = immp;
1597 
1598 	/* Initialize per-group source filters. */
1599 	for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1600 		im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1601 	imo->im6o_mfilters = imfp;
1602 
1603 	INP_WLOCK(inp);
1604 	if (inp->in6p_moptions != NULL) {
1605 		free(imfp, M_IN6MFILTER);
1606 		free(immp, M_IP6MOPTS);
1607 		free(imo, M_IP6MOPTS);
1608 		return (inp->in6p_moptions);
1609 	}
1610 	inp->in6p_moptions = imo;
1611 	return (imo);
1612 }
1613 
1614 /*
1615  * Discard the IPv6 multicast options (and source filters).
1616  *
1617  * SMPng: NOTE: assumes INP write lock is held.
1618  */
1619 void
1620 ip6_freemoptions(struct ip6_moptions *imo, struct inpcbinfo *pcbinfo)
1621 {
1622 	struct in6_mfilter	*imf;
1623 	size_t			 idx, nmships;
1624 	int wlock;
1625 
1626 	if (imo == NULL)
1627 		return;
1628 	INP_INFO_LOCK_ASSERT(pcbinfo);
1629 	wlock = INP_INFO_WLOCKED(pcbinfo);
1630 	if (wlock)
1631 		INP_INFO_WUNLOCK(pcbinfo);
1632 	else
1633 		INP_INFO_RUNLOCK(pcbinfo);
1634 
1635 	nmships = imo->im6o_num_memberships;
1636 	for (idx = 0; idx < nmships; ++idx) {
1637 		imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1638 		if (imf)
1639 			im6f_leave(imf);
1640 		/* XXX this will thrash the lock(s) */
1641 		(void)in6_leavegroup(imo->im6o_membership[idx], imf);
1642 		if (imf)
1643 			im6f_purge(imf);
1644 	}
1645 
1646 	if (imo->im6o_mfilters)
1647 		free(imo->im6o_mfilters, M_IN6MFILTER);
1648 	free(imo->im6o_membership, M_IP6MOPTS);
1649 	free(imo, M_IP6MOPTS);
1650 	if (wlock)
1651 		INP_INFO_WLOCK(pcbinfo);
1652 	else
1653 		INP_INFO_RLOCK(pcbinfo);
1654 }
1655 
1656 /*
1657  * Atomically get source filters on a socket for an IPv6 multicast group.
1658  * Called with INP lock held; returns with lock released.
1659  */
1660 static int
1661 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1662 {
1663 	struct __msfilterreq	 msfr;
1664 	sockunion_t		*gsa;
1665 	struct ifnet		*ifp;
1666 	struct ip6_moptions	*imo;
1667 	struct in6_mfilter	*imf;
1668 	struct ip6_msource	*ims;
1669 	struct in6_msource	*lims;
1670 	struct sockaddr_in6	*psin;
1671 	struct sockaddr_storage	*ptss;
1672 	struct sockaddr_storage	*tss;
1673 	int			 error;
1674 	size_t			 idx, nsrcs, ncsrcs;
1675 
1676 	INP_WLOCK_ASSERT(inp);
1677 
1678 	imo = inp->in6p_moptions;
1679 	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1680 
1681 	INP_WUNLOCK(inp);
1682 
1683 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1684 	    sizeof(struct __msfilterreq));
1685 	if (error)
1686 		return (error);
1687 
1688 	if (msfr.msfr_group.ss_family != AF_INET6 ||
1689 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1690 		return (EINVAL);
1691 
1692 	gsa = (sockunion_t *)&msfr.msfr_group;
1693 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1694 		return (EINVAL);
1695 
1696 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1697 		return (EADDRNOTAVAIL);
1698 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1699 	if (ifp == NULL)
1700 		return (EADDRNOTAVAIL);
1701 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1702 
1703 	INP_WLOCK(inp);
1704 
1705 	/*
1706 	 * Lookup group on the socket.
1707 	 */
1708 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1709 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1710 		INP_WUNLOCK(inp);
1711 		return (EADDRNOTAVAIL);
1712 	}
1713 	imf = &imo->im6o_mfilters[idx];
1714 
1715 	/*
1716 	 * Ignore memberships which are in limbo.
1717 	 */
1718 	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1719 		INP_WUNLOCK(inp);
1720 		return (EAGAIN);
1721 	}
1722 	msfr.msfr_fmode = imf->im6f_st[1];
1723 
1724 	/*
1725 	 * If the user specified a buffer, copy out the source filter
1726 	 * entries to userland gracefully.
1727 	 * We only copy out the number of entries which userland
1728 	 * has asked for, but we always tell userland how big the
1729 	 * buffer really needs to be.
1730 	 */
1731 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1732 		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1733 	tss = NULL;
1734 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1735 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1736 		    M_TEMP, M_NOWAIT | M_ZERO);
1737 		if (tss == NULL) {
1738 			INP_WUNLOCK(inp);
1739 			return (ENOBUFS);
1740 		}
1741 	}
1742 
1743 	/*
1744 	 * Count number of sources in-mode at t0.
1745 	 * If buffer space exists and remains, copy out source entries.
1746 	 */
1747 	nsrcs = msfr.msfr_nsrcs;
1748 	ncsrcs = 0;
1749 	ptss = tss;
1750 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1751 		lims = (struct in6_msource *)ims;
1752 		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1753 		    lims->im6sl_st[0] != imf->im6f_st[0])
1754 			continue;
1755 		++ncsrcs;
1756 		if (tss != NULL && nsrcs > 0) {
1757 			psin = (struct sockaddr_in6 *)ptss;
1758 			psin->sin6_family = AF_INET6;
1759 			psin->sin6_len = sizeof(struct sockaddr_in6);
1760 			psin->sin6_addr = lims->im6s_addr;
1761 			psin->sin6_port = 0;
1762 			--nsrcs;
1763 			++ptss;
1764 		}
1765 	}
1766 
1767 	INP_WUNLOCK(inp);
1768 
1769 	if (tss != NULL) {
1770 		error = copyout(tss, msfr.msfr_srcs,
1771 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1772 		free(tss, M_TEMP);
1773 		if (error)
1774 			return (error);
1775 	}
1776 
1777 	msfr.msfr_nsrcs = ncsrcs;
1778 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1779 
1780 	return (error);
1781 }
1782 
1783 /*
1784  * Return the IP multicast options in response to user getsockopt().
1785  */
1786 int
1787 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1788 {
1789 	struct ip6_moptions	*im6o;
1790 	int			 error;
1791 	u_int			 optval;
1792 
1793 	INP_WLOCK(inp);
1794 	im6o = inp->in6p_moptions;
1795 	/*
1796 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1797 	 * or is a divert socket, reject it.
1798 	 */
1799 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1800 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1801 	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1802 		INP_WUNLOCK(inp);
1803 		return (EOPNOTSUPP);
1804 	}
1805 
1806 	error = 0;
1807 	switch (sopt->sopt_name) {
1808 	case IPV6_MULTICAST_IF:
1809 		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1810 			optval = 0;
1811 		} else {
1812 			optval = im6o->im6o_multicast_ifp->if_index;
1813 		}
1814 		INP_WUNLOCK(inp);
1815 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1816 		break;
1817 
1818 	case IPV6_MULTICAST_HOPS:
1819 		if (im6o == NULL)
1820 			optval = V_ip6_defmcasthlim;
1821 		else
1822 			optval = im6o->im6o_multicast_hlim;
1823 		INP_WUNLOCK(inp);
1824 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1825 		break;
1826 
1827 	case IPV6_MULTICAST_LOOP:
1828 		if (im6o == NULL)
1829 			optval = in6_mcast_loop; /* XXX VIMAGE */
1830 		else
1831 			optval = im6o->im6o_multicast_loop;
1832 		INP_WUNLOCK(inp);
1833 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1834 		break;
1835 
1836 	case IPV6_MSFILTER:
1837 		if (im6o == NULL) {
1838 			error = EADDRNOTAVAIL;
1839 			INP_WUNLOCK(inp);
1840 		} else {
1841 			error = in6p_get_source_filters(inp, sopt);
1842 		}
1843 		break;
1844 
1845 	default:
1846 		INP_WUNLOCK(inp);
1847 		error = ENOPROTOOPT;
1848 		break;
1849 	}
1850 
1851 	INP_UNLOCK_ASSERT(inp);
1852 
1853 	return (error);
1854 }
1855 
1856 /*
1857  * Look up the ifnet to use for a multicast group membership,
1858  * given the address of an IPv6 group.
1859  *
1860  * This routine exists to support legacy IPv6 multicast applications.
1861  *
1862  * If inp is non-NULL, use this socket's current FIB number for any
1863  * required FIB lookup. Look up the group address in the unicast FIB,
1864  * and use its ifp; usually, this points to the default next-hop.
1865  * If the FIB lookup fails, return NULL.
1866  *
1867  * FUTURE: Support multiple forwarding tables for IPv6.
1868  *
1869  * Returns NULL if no ifp could be found.
1870  */
1871 static struct ifnet *
1872 in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1873     const struct sockaddr_in6 *gsin6)
1874 {
1875 	struct nhop6_basic	nh6;
1876 	struct in6_addr		dst;
1877 	uint32_t		scopeid;
1878 	uint32_t		fibnum;
1879 
1880 	KASSERT(in6p->inp_vflag & INP_IPV6,
1881 	    ("%s: not INP_IPV6 inpcb", __func__));
1882 	KASSERT(gsin6->sin6_family == AF_INET6,
1883 	    ("%s: not AF_INET6 group", __func__));
1884 
1885 	in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1886 	fibnum = in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB;
1887 	if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6) != 0)
1888 		return (NULL);
1889 
1890 	return (nh6.nh_ifp);
1891 }
1892 
1893 /*
1894  * Join an IPv6 multicast group, possibly with a source.
1895  *
1896  * FIXME: The KAME use of the unspecified address (::)
1897  * to join *all* multicast groups is currently unsupported.
1898  */
1899 static int
1900 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1901 {
1902 	struct group_source_req		 gsr;
1903 	sockunion_t			*gsa, *ssa;
1904 	struct ifnet			*ifp;
1905 	struct in6_mfilter		*imf;
1906 	struct ip6_moptions		*imo;
1907 	struct in6_multi		*inm;
1908 	struct in6_msource		*lims;
1909 	size_t				 idx;
1910 	int				 error, is_new;
1911 
1912 	ifp = NULL;
1913 	imf = NULL;
1914 	lims = NULL;
1915 	error = 0;
1916 	is_new = 0;
1917 
1918 	memset(&gsr, 0, sizeof(struct group_source_req));
1919 	gsa = (sockunion_t *)&gsr.gsr_group;
1920 	gsa->ss.ss_family = AF_UNSPEC;
1921 	ssa = (sockunion_t *)&gsr.gsr_source;
1922 	ssa->ss.ss_family = AF_UNSPEC;
1923 
1924 	/*
1925 	 * Chew everything into struct group_source_req.
1926 	 * Overwrite the port field if present, as the sockaddr
1927 	 * being copied in may be matched with a binary comparison.
1928 	 * Ignore passed-in scope ID.
1929 	 */
1930 	switch (sopt->sopt_name) {
1931 	case IPV6_JOIN_GROUP: {
1932 		struct ipv6_mreq mreq;
1933 
1934 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1935 		    sizeof(struct ipv6_mreq));
1936 		if (error)
1937 			return (error);
1938 
1939 		gsa->sin6.sin6_family = AF_INET6;
1940 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1941 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1942 
1943 		if (mreq.ipv6mr_interface == 0) {
1944 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1945 		} else {
1946 			if (V_if_index < mreq.ipv6mr_interface)
1947 				return (EADDRNOTAVAIL);
1948 			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1949 		}
1950 		CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1951 		    __func__, mreq.ipv6mr_interface, ifp);
1952 	} break;
1953 
1954 	case MCAST_JOIN_GROUP:
1955 	case MCAST_JOIN_SOURCE_GROUP:
1956 		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1957 			error = sooptcopyin(sopt, &gsr,
1958 			    sizeof(struct group_req),
1959 			    sizeof(struct group_req));
1960 		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1961 			error = sooptcopyin(sopt, &gsr,
1962 			    sizeof(struct group_source_req),
1963 			    sizeof(struct group_source_req));
1964 		}
1965 		if (error)
1966 			return (error);
1967 
1968 		if (gsa->sin6.sin6_family != AF_INET6 ||
1969 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1970 			return (EINVAL);
1971 
1972 		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1973 			if (ssa->sin6.sin6_family != AF_INET6 ||
1974 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1975 				return (EINVAL);
1976 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1977 				return (EINVAL);
1978 			/*
1979 			 * TODO: Validate embedded scope ID in source
1980 			 * list entry against passed-in ifp, if and only
1981 			 * if source list filter entry is iface or node local.
1982 			 */
1983 			in6_clearscope(&ssa->sin6.sin6_addr);
1984 			ssa->sin6.sin6_port = 0;
1985 			ssa->sin6.sin6_scope_id = 0;
1986 		}
1987 
1988 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1989 			return (EADDRNOTAVAIL);
1990 		ifp = ifnet_byindex(gsr.gsr_interface);
1991 		break;
1992 
1993 	default:
1994 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1995 		    __func__, sopt->sopt_name);
1996 		return (EOPNOTSUPP);
1997 		break;
1998 	}
1999 
2000 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2001 		return (EINVAL);
2002 
2003 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2004 		return (EADDRNOTAVAIL);
2005 
2006 	gsa->sin6.sin6_port = 0;
2007 	gsa->sin6.sin6_scope_id = 0;
2008 
2009 	/*
2010 	 * Always set the scope zone ID on memberships created from userland.
2011 	 * Use the passed-in ifp to do this.
2012 	 * XXX The in6_setscope() return value is meaningless.
2013 	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
2014 	 */
2015 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2016 
2017 	imo = in6p_findmoptions(inp);
2018 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2019 	if (idx == -1) {
2020 		is_new = 1;
2021 	} else {
2022 		inm = imo->im6o_membership[idx];
2023 		imf = &imo->im6o_mfilters[idx];
2024 		if (ssa->ss.ss_family != AF_UNSPEC) {
2025 			/*
2026 			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2027 			 * is an error. On an existing inclusive membership,
2028 			 * it just adds the source to the filter list.
2029 			 */
2030 			if (imf->im6f_st[1] != MCAST_INCLUDE) {
2031 				error = EINVAL;
2032 				goto out_in6p_locked;
2033 			}
2034 			/*
2035 			 * Throw out duplicates.
2036 			 *
2037 			 * XXX FIXME: This makes a naive assumption that
2038 			 * even if entries exist for *ssa in this imf,
2039 			 * they will be rejected as dupes, even if they
2040 			 * are not valid in the current mode (in-mode).
2041 			 *
2042 			 * in6_msource is transactioned just as for anything
2043 			 * else in SSM -- but note naive use of in6m_graft()
2044 			 * below for allocating new filter entries.
2045 			 *
2046 			 * This is only an issue if someone mixes the
2047 			 * full-state SSM API with the delta-based API,
2048 			 * which is discouraged in the relevant RFCs.
2049 			 */
2050 			lims = im6o_match_source(imo, idx, &ssa->sa);
2051 			if (lims != NULL /*&&
2052 			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2053 				error = EADDRNOTAVAIL;
2054 				goto out_in6p_locked;
2055 			}
2056 		} else {
2057 			/*
2058 			 * MCAST_JOIN_GROUP alone, on any existing membership,
2059 			 * is rejected, to stop the same inpcb tying up
2060 			 * multiple refs to the in_multi.
2061 			 * On an existing inclusive membership, this is also
2062 			 * an error; if you want to change filter mode,
2063 			 * you must use the userland API setsourcefilter().
2064 			 * XXX We don't reject this for imf in UNDEFINED
2065 			 * state at t1, because allocation of a filter
2066 			 * is atomic with allocation of a membership.
2067 			 */
2068 			error = EINVAL;
2069 			goto out_in6p_locked;
2070 		}
2071 	}
2072 
2073 	/*
2074 	 * Begin state merge transaction at socket layer.
2075 	 */
2076 	INP_WLOCK_ASSERT(inp);
2077 
2078 	if (is_new) {
2079 		if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
2080 			error = im6o_grow(imo);
2081 			if (error)
2082 				goto out_in6p_locked;
2083 		}
2084 		/*
2085 		 * Allocate the new slot upfront so we can deal with
2086 		 * grafting the new source filter in same code path
2087 		 * as for join-source on existing membership.
2088 		 */
2089 		idx = imo->im6o_num_memberships;
2090 		imo->im6o_membership[idx] = NULL;
2091 		imo->im6o_num_memberships++;
2092 		KASSERT(imo->im6o_mfilters != NULL,
2093 		    ("%s: im6f_mfilters vector was not allocated", __func__));
2094 		imf = &imo->im6o_mfilters[idx];
2095 		KASSERT(RB_EMPTY(&imf->im6f_sources),
2096 		    ("%s: im6f_sources not empty", __func__));
2097 	}
2098 
2099 	/*
2100 	 * Graft new source into filter list for this inpcb's
2101 	 * membership of the group. The in6_multi may not have
2102 	 * been allocated yet if this is a new membership, however,
2103 	 * the in_mfilter slot will be allocated and must be initialized.
2104 	 *
2105 	 * Note: Grafting of exclusive mode filters doesn't happen
2106 	 * in this path.
2107 	 * XXX: Should check for non-NULL lims (node exists but may
2108 	 * not be in-mode) for interop with full-state API.
2109 	 */
2110 	if (ssa->ss.ss_family != AF_UNSPEC) {
2111 		/* Membership starts in IN mode */
2112 		if (is_new) {
2113 			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2114 			im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2115 		} else {
2116 			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2117 		}
2118 		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2119 		if (lims == NULL) {
2120 			CTR1(KTR_MLD, "%s: merge imf state failed",
2121 			    __func__);
2122 			error = ENOMEM;
2123 			goto out_im6o_free;
2124 		}
2125 	} else {
2126 		/* No address specified; Membership starts in EX mode */
2127 		if (is_new) {
2128 			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2129 			im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2130 		}
2131 	}
2132 
2133 	/*
2134 	 * Begin state merge transaction at MLD layer.
2135 	 */
2136 	in_pcbref(inp);
2137 	INP_WUNLOCK(inp);
2138 	IN6_MULTI_LOCK();
2139 
2140 	if (is_new) {
2141 		error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2142 		    &inm, 0);
2143 		if (error) {
2144 			IN6_MULTI_UNLOCK();
2145 			goto out_im6o_free;
2146 		}
2147 		imo->im6o_membership[idx] = inm;
2148 	} else {
2149 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2150 		IN6_MULTI_LIST_LOCK();
2151 		error = in6m_merge(inm, imf);
2152 		if (error)
2153 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2154 			    __func__);
2155 		else {
2156 			CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2157 			error = mld_change_state(inm, 0);
2158 			if (error)
2159 				CTR1(KTR_MLD, "%s: failed mld downcall",
2160 				    __func__);
2161 		}
2162 		IN6_MULTI_LIST_UNLOCK();
2163 	}
2164 
2165 	IN6_MULTI_UNLOCK();
2166 	INP_WLOCK(inp);
2167 	if (in_pcbrele_wlocked(inp))
2168 		return (ENXIO);
2169 	if (error) {
2170 		im6f_rollback(imf);
2171 		if (is_new)
2172 			im6f_purge(imf);
2173 		else
2174 			im6f_reap(imf);
2175 	} else {
2176 		im6f_commit(imf);
2177 	}
2178 
2179 out_im6o_free:
2180 	if (error && is_new) {
2181 		imo->im6o_membership[idx] = NULL;
2182 		--imo->im6o_num_memberships;
2183 	}
2184 
2185 out_in6p_locked:
2186 	INP_WUNLOCK(inp);
2187 	return (error);
2188 }
2189 
2190 /*
2191  * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2192  */
2193 static int
2194 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2195 {
2196 	struct ipv6_mreq		 mreq;
2197 	struct group_source_req		 gsr;
2198 	sockunion_t			*gsa, *ssa;
2199 	struct ifnet			*ifp;
2200 	struct in6_mfilter		*imf;
2201 	struct ip6_moptions		*imo;
2202 	struct in6_msource		*ims;
2203 	struct in6_multi		*inm;
2204 	uint32_t			 ifindex;
2205 	size_t				 idx;
2206 	int				 error, is_final;
2207 #ifdef KTR
2208 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2209 #endif
2210 
2211 	ifp = NULL;
2212 	ifindex = 0;
2213 	error = 0;
2214 	is_final = 1;
2215 
2216 	memset(&gsr, 0, sizeof(struct group_source_req));
2217 	gsa = (sockunion_t *)&gsr.gsr_group;
2218 	gsa->ss.ss_family = AF_UNSPEC;
2219 	ssa = (sockunion_t *)&gsr.gsr_source;
2220 	ssa->ss.ss_family = AF_UNSPEC;
2221 
2222 	/*
2223 	 * Chew everything passed in up into a struct group_source_req
2224 	 * as that is easier to process.
2225 	 * Note: Any embedded scope ID in the multicast group passed
2226 	 * in by userland is ignored, the interface index is the recommended
2227 	 * mechanism to specify an interface; see below.
2228 	 */
2229 	switch (sopt->sopt_name) {
2230 	case IPV6_LEAVE_GROUP:
2231 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2232 		    sizeof(struct ipv6_mreq));
2233 		if (error)
2234 			return (error);
2235 		gsa->sin6.sin6_family = AF_INET6;
2236 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2237 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2238 		gsa->sin6.sin6_port = 0;
2239 		gsa->sin6.sin6_scope_id = 0;
2240 		ifindex = mreq.ipv6mr_interface;
2241 		break;
2242 
2243 	case MCAST_LEAVE_GROUP:
2244 	case MCAST_LEAVE_SOURCE_GROUP:
2245 		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2246 			error = sooptcopyin(sopt, &gsr,
2247 			    sizeof(struct group_req),
2248 			    sizeof(struct group_req));
2249 		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2250 			error = sooptcopyin(sopt, &gsr,
2251 			    sizeof(struct group_source_req),
2252 			    sizeof(struct group_source_req));
2253 		}
2254 		if (error)
2255 			return (error);
2256 
2257 		if (gsa->sin6.sin6_family != AF_INET6 ||
2258 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2259 			return (EINVAL);
2260 		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2261 			if (ssa->sin6.sin6_family != AF_INET6 ||
2262 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2263 				return (EINVAL);
2264 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2265 				return (EINVAL);
2266 			/*
2267 			 * TODO: Validate embedded scope ID in source
2268 			 * list entry against passed-in ifp, if and only
2269 			 * if source list filter entry is iface or node local.
2270 			 */
2271 			in6_clearscope(&ssa->sin6.sin6_addr);
2272 		}
2273 		gsa->sin6.sin6_port = 0;
2274 		gsa->sin6.sin6_scope_id = 0;
2275 		ifindex = gsr.gsr_interface;
2276 		break;
2277 
2278 	default:
2279 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2280 		    __func__, sopt->sopt_name);
2281 		return (EOPNOTSUPP);
2282 		break;
2283 	}
2284 
2285 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2286 		return (EINVAL);
2287 
2288 	/*
2289 	 * Validate interface index if provided. If no interface index
2290 	 * was provided separately, attempt to look the membership up
2291 	 * from the default scope as a last resort to disambiguate
2292 	 * the membership we are being asked to leave.
2293 	 * XXX SCOPE6 lock potentially taken here.
2294 	 */
2295 	if (ifindex != 0) {
2296 		if (V_if_index < ifindex)
2297 			return (EADDRNOTAVAIL);
2298 		ifp = ifnet_byindex(ifindex);
2299 		if (ifp == NULL)
2300 			return (EADDRNOTAVAIL);
2301 		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2302 	} else {
2303 		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2304 		if (error)
2305 			return (EADDRNOTAVAIL);
2306 		/*
2307 		 * Some badly behaved applications don't pass an ifindex
2308 		 * or a scope ID, which is an API violation. In this case,
2309 		 * perform a lookup as per a v6 join.
2310 		 *
2311 		 * XXX For now, stomp on zone ID for the corner case.
2312 		 * This is not the 'KAME way', but we need to see the ifp
2313 		 * directly until such time as this implementation is
2314 		 * refactored, assuming the scope IDs are the way to go.
2315 		 */
2316 		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2317 		if (ifindex == 0) {
2318 			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2319 			    "ifp for group %s.", __func__,
2320 			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2321 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2322 		} else {
2323 			ifp = ifnet_byindex(ifindex);
2324 		}
2325 		if (ifp == NULL)
2326 			return (EADDRNOTAVAIL);
2327 	}
2328 
2329 	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2330 	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2331 
2332 	/*
2333 	 * Find the membership in the membership array.
2334 	 */
2335 	imo = in6p_findmoptions(inp);
2336 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2337 	if (idx == -1) {
2338 		error = EADDRNOTAVAIL;
2339 		goto out_in6p_locked;
2340 	}
2341 	inm = imo->im6o_membership[idx];
2342 	imf = &imo->im6o_mfilters[idx];
2343 
2344 	if (ssa->ss.ss_family != AF_UNSPEC)
2345 		is_final = 0;
2346 
2347 	/*
2348 	 * Begin state merge transaction at socket layer.
2349 	 */
2350 	INP_WLOCK_ASSERT(inp);
2351 
2352 	/*
2353 	 * If we were instructed only to leave a given source, do so.
2354 	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2355 	 */
2356 	if (is_final) {
2357 		im6f_leave(imf);
2358 	} else {
2359 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2360 			error = EADDRNOTAVAIL;
2361 			goto out_in6p_locked;
2362 		}
2363 		ims = im6o_match_source(imo, idx, &ssa->sa);
2364 		if (ims == NULL) {
2365 			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2366 			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2367 			    "not ");
2368 			error = EADDRNOTAVAIL;
2369 			goto out_in6p_locked;
2370 		}
2371 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2372 		error = im6f_prune(imf, &ssa->sin6);
2373 		if (error) {
2374 			CTR1(KTR_MLD, "%s: merge imf state failed",
2375 			    __func__);
2376 			goto out_in6p_locked;
2377 		}
2378 	}
2379 
2380 	/*
2381 	 * Begin state merge transaction at MLD layer.
2382 	 */
2383 	IN6_MULTI_LOCK();
2384 
2385 	if (is_final) {
2386 		/*
2387 		 * Give up the multicast address record to which
2388 		 * the membership points.
2389 		 */
2390 		(void)in6_leavegroup_locked(inm, imf);
2391 	} else {
2392 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2393 		IN6_MULTI_LIST_LOCK();
2394 		error = in6m_merge(inm, imf);
2395 		if (error)
2396 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2397 			    __func__);
2398 		else {
2399 			CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2400 			error = mld_change_state(inm, 0);
2401 			if (error)
2402 				CTR1(KTR_MLD, "%s: failed mld downcall",
2403 				    __func__);
2404 		}
2405 		IN6_MULTI_LIST_UNLOCK();
2406 	}
2407 
2408 	IN6_MULTI_UNLOCK();
2409 
2410 	if (error)
2411 		im6f_rollback(imf);
2412 	else
2413 		im6f_commit(imf);
2414 
2415 	im6f_reap(imf);
2416 
2417 	if (is_final) {
2418 		/* Remove the gap in the membership array. */
2419 		for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2420 			imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2421 			imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2422 		}
2423 		imo->im6o_num_memberships--;
2424 	}
2425 
2426 out_in6p_locked:
2427 	INP_WUNLOCK(inp);
2428 	return (error);
2429 }
2430 
2431 /*
2432  * Select the interface for transmitting IPv6 multicast datagrams.
2433  *
2434  * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2435  * may be passed to this socket option. An address of in6addr_any or an
2436  * interface index of 0 is used to remove a previous selection.
2437  * When no interface is selected, one is chosen for every send.
2438  */
2439 static int
2440 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2441 {
2442 	struct ifnet		*ifp;
2443 	struct ip6_moptions	*imo;
2444 	u_int			 ifindex;
2445 	int			 error;
2446 
2447 	if (sopt->sopt_valsize != sizeof(u_int))
2448 		return (EINVAL);
2449 
2450 	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2451 	if (error)
2452 		return (error);
2453 	if (V_if_index < ifindex)
2454 		return (EINVAL);
2455 	if (ifindex == 0)
2456 		ifp = NULL;
2457 	else {
2458 		ifp = ifnet_byindex(ifindex);
2459 		if (ifp == NULL)
2460 			return (EINVAL);
2461 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2462 			return (EADDRNOTAVAIL);
2463 	}
2464 	imo = in6p_findmoptions(inp);
2465 	imo->im6o_multicast_ifp = ifp;
2466 	INP_WUNLOCK(inp);
2467 
2468 	return (0);
2469 }
2470 
2471 /*
2472  * Atomically set source filters on a socket for an IPv6 multicast group.
2473  *
2474  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2475  */
2476 static int
2477 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2478 {
2479 	struct __msfilterreq	 msfr;
2480 	sockunion_t		*gsa;
2481 	struct ifnet		*ifp;
2482 	struct in6_mfilter	*imf;
2483 	struct ip6_moptions	*imo;
2484 	struct in6_multi		*inm;
2485 	size_t			 idx;
2486 	int			 error;
2487 
2488 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2489 	    sizeof(struct __msfilterreq));
2490 	if (error)
2491 		return (error);
2492 
2493 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2494 		return (ENOBUFS);
2495 
2496 	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2497 	    msfr.msfr_fmode != MCAST_INCLUDE)
2498 		return (EINVAL);
2499 
2500 	if (msfr.msfr_group.ss_family != AF_INET6 ||
2501 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2502 		return (EINVAL);
2503 
2504 	gsa = (sockunion_t *)&msfr.msfr_group;
2505 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2506 		return (EINVAL);
2507 
2508 	gsa->sin6.sin6_port = 0;	/* ignore port */
2509 
2510 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2511 		return (EADDRNOTAVAIL);
2512 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2513 	if (ifp == NULL)
2514 		return (EADDRNOTAVAIL);
2515 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2516 
2517 	/*
2518 	 * Take the INP write lock.
2519 	 * Check if this socket is a member of this group.
2520 	 */
2521 	imo = in6p_findmoptions(inp);
2522 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2523 	if (idx == -1 || imo->im6o_mfilters == NULL) {
2524 		error = EADDRNOTAVAIL;
2525 		goto out_in6p_locked;
2526 	}
2527 	inm = imo->im6o_membership[idx];
2528 	imf = &imo->im6o_mfilters[idx];
2529 
2530 	/*
2531 	 * Begin state merge transaction at socket layer.
2532 	 */
2533 	INP_WLOCK_ASSERT(inp);
2534 
2535 	imf->im6f_st[1] = msfr.msfr_fmode;
2536 
2537 	/*
2538 	 * Apply any new source filters, if present.
2539 	 * Make a copy of the user-space source vector so
2540 	 * that we may copy them with a single copyin. This
2541 	 * allows us to deal with page faults up-front.
2542 	 */
2543 	if (msfr.msfr_nsrcs > 0) {
2544 		struct in6_msource	*lims;
2545 		struct sockaddr_in6	*psin;
2546 		struct sockaddr_storage	*kss, *pkss;
2547 		int			 i;
2548 
2549 		INP_WUNLOCK(inp);
2550 
2551 		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2552 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2553 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2554 		    M_TEMP, M_WAITOK);
2555 		error = copyin(msfr.msfr_srcs, kss,
2556 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2557 		if (error) {
2558 			free(kss, M_TEMP);
2559 			return (error);
2560 		}
2561 
2562 		INP_WLOCK(inp);
2563 
2564 		/*
2565 		 * Mark all source filters as UNDEFINED at t1.
2566 		 * Restore new group filter mode, as im6f_leave()
2567 		 * will set it to INCLUDE.
2568 		 */
2569 		im6f_leave(imf);
2570 		imf->im6f_st[1] = msfr.msfr_fmode;
2571 
2572 		/*
2573 		 * Update socket layer filters at t1, lazy-allocating
2574 		 * new entries. This saves a bunch of memory at the
2575 		 * cost of one RB_FIND() per source entry; duplicate
2576 		 * entries in the msfr_nsrcs vector are ignored.
2577 		 * If we encounter an error, rollback transaction.
2578 		 *
2579 		 * XXX This too could be replaced with a set-symmetric
2580 		 * difference like loop to avoid walking from root
2581 		 * every time, as the key space is common.
2582 		 */
2583 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2584 			psin = (struct sockaddr_in6 *)pkss;
2585 			if (psin->sin6_family != AF_INET6) {
2586 				error = EAFNOSUPPORT;
2587 				break;
2588 			}
2589 			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2590 				error = EINVAL;
2591 				break;
2592 			}
2593 			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2594 				error = EINVAL;
2595 				break;
2596 			}
2597 			/*
2598 			 * TODO: Validate embedded scope ID in source
2599 			 * list entry against passed-in ifp, if and only
2600 			 * if source list filter entry is iface or node local.
2601 			 */
2602 			in6_clearscope(&psin->sin6_addr);
2603 			error = im6f_get_source(imf, psin, &lims);
2604 			if (error)
2605 				break;
2606 			lims->im6sl_st[1] = imf->im6f_st[1];
2607 		}
2608 		free(kss, M_TEMP);
2609 	}
2610 
2611 	if (error)
2612 		goto out_im6f_rollback;
2613 
2614 	INP_WLOCK_ASSERT(inp);
2615 	IN6_MULTI_LIST_LOCK();
2616 
2617 	/*
2618 	 * Begin state merge transaction at MLD layer.
2619 	 */
2620 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2621 	error = in6m_merge(inm, imf);
2622 	if (error)
2623 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2624 	else {
2625 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2626 		error = mld_change_state(inm, 0);
2627 		if (error)
2628 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2629 	}
2630 
2631 	IN6_MULTI_LIST_UNLOCK();
2632 
2633 out_im6f_rollback:
2634 	if (error)
2635 		im6f_rollback(imf);
2636 	else
2637 		im6f_commit(imf);
2638 
2639 	im6f_reap(imf);
2640 
2641 out_in6p_locked:
2642 	INP_WUNLOCK(inp);
2643 	return (error);
2644 }
2645 
2646 /*
2647  * Set the IP multicast options in response to user setsockopt().
2648  *
2649  * Many of the socket options handled in this function duplicate the
2650  * functionality of socket options in the regular unicast API. However,
2651  * it is not possible to merge the duplicate code, because the idempotence
2652  * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2653  * the effects of these options must be treated as separate and distinct.
2654  *
2655  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2656  */
2657 int
2658 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2659 {
2660 	struct ip6_moptions	*im6o;
2661 	int			 error;
2662 
2663 	error = 0;
2664 
2665 	/*
2666 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2667 	 * or is a divert socket, reject it.
2668 	 */
2669 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2670 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2671 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2672 		return (EOPNOTSUPP);
2673 
2674 	switch (sopt->sopt_name) {
2675 	case IPV6_MULTICAST_IF:
2676 		error = in6p_set_multicast_if(inp, sopt);
2677 		break;
2678 
2679 	case IPV6_MULTICAST_HOPS: {
2680 		int hlim;
2681 
2682 		if (sopt->sopt_valsize != sizeof(int)) {
2683 			error = EINVAL;
2684 			break;
2685 		}
2686 		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2687 		if (error)
2688 			break;
2689 		if (hlim < -1 || hlim > 255) {
2690 			error = EINVAL;
2691 			break;
2692 		} else if (hlim == -1) {
2693 			hlim = V_ip6_defmcasthlim;
2694 		}
2695 		im6o = in6p_findmoptions(inp);
2696 		im6o->im6o_multicast_hlim = hlim;
2697 		INP_WUNLOCK(inp);
2698 		break;
2699 	}
2700 
2701 	case IPV6_MULTICAST_LOOP: {
2702 		u_int loop;
2703 
2704 		/*
2705 		 * Set the loopback flag for outgoing multicast packets.
2706 		 * Must be zero or one.
2707 		 */
2708 		if (sopt->sopt_valsize != sizeof(u_int)) {
2709 			error = EINVAL;
2710 			break;
2711 		}
2712 		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2713 		if (error)
2714 			break;
2715 		if (loop > 1) {
2716 			error = EINVAL;
2717 			break;
2718 		}
2719 		im6o = in6p_findmoptions(inp);
2720 		im6o->im6o_multicast_loop = loop;
2721 		INP_WUNLOCK(inp);
2722 		break;
2723 	}
2724 
2725 	case IPV6_JOIN_GROUP:
2726 	case MCAST_JOIN_GROUP:
2727 	case MCAST_JOIN_SOURCE_GROUP:
2728 		error = in6p_join_group(inp, sopt);
2729 		break;
2730 
2731 	case IPV6_LEAVE_GROUP:
2732 	case MCAST_LEAVE_GROUP:
2733 	case MCAST_LEAVE_SOURCE_GROUP:
2734 		error = in6p_leave_group(inp, sopt);
2735 		break;
2736 
2737 	case MCAST_BLOCK_SOURCE:
2738 	case MCAST_UNBLOCK_SOURCE:
2739 		error = in6p_block_unblock_source(inp, sopt);
2740 		break;
2741 
2742 	case IPV6_MSFILTER:
2743 		error = in6p_set_source_filters(inp, sopt);
2744 		break;
2745 
2746 	default:
2747 		error = EOPNOTSUPP;
2748 		break;
2749 	}
2750 
2751 	INP_UNLOCK_ASSERT(inp);
2752 
2753 	return (error);
2754 }
2755 
2756 /*
2757  * Expose MLD's multicast filter mode and source list(s) to userland,
2758  * keyed by (ifindex, group).
2759  * The filter mode is written out as a uint32_t, followed by
2760  * 0..n of struct in6_addr.
2761  * For use by ifmcstat(8).
2762  * SMPng: NOTE: unlocked read of ifindex space.
2763  */
2764 static int
2765 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2766 {
2767 	struct in6_addr			 mcaddr;
2768 	struct in6_addr			 src;
2769 	struct ifnet			*ifp;
2770 	struct ifmultiaddr		*ifma;
2771 	struct in6_multi		*inm;
2772 	struct ip6_msource		*ims;
2773 	int				*name;
2774 	int				 retval;
2775 	u_int				 namelen;
2776 	uint32_t			 fmode, ifindex;
2777 #ifdef KTR
2778 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2779 #endif
2780 
2781 	name = (int *)arg1;
2782 	namelen = arg2;
2783 
2784 	if (req->newptr != NULL)
2785 		return (EPERM);
2786 
2787 	/* int: ifindex + 4 * 32 bits of IPv6 address */
2788 	if (namelen != 5)
2789 		return (EINVAL);
2790 
2791 	ifindex = name[0];
2792 	if (ifindex <= 0 || ifindex > V_if_index) {
2793 		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2794 		    __func__, ifindex);
2795 		return (ENOENT);
2796 	}
2797 
2798 	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2799 	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2800 		CTR2(KTR_MLD, "%s: group %s is not multicast",
2801 		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2802 		return (EINVAL);
2803 	}
2804 
2805 	ifp = ifnet_byindex(ifindex);
2806 	if (ifp == NULL) {
2807 		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2808 		    __func__, ifindex);
2809 		return (ENOENT);
2810 	}
2811 	/*
2812 	 * Internal MLD lookups require that scope/zone ID is set.
2813 	 */
2814 	(void)in6_setscope(&mcaddr, ifp, NULL);
2815 
2816 	retval = sysctl_wire_old_buffer(req,
2817 	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2818 	if (retval)
2819 		return (retval);
2820 
2821 	IN6_MULTI_LOCK();
2822 	IN6_MULTI_LIST_LOCK();
2823 	IF_ADDR_RLOCK(ifp);
2824 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2825 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2826 		    ifma->ifma_protospec == NULL)
2827 			continue;
2828 		inm = (struct in6_multi *)ifma->ifma_protospec;
2829 		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2830 			continue;
2831 		fmode = inm->in6m_st[1].iss_fmode;
2832 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2833 		if (retval != 0)
2834 			break;
2835 		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2836 			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2837 			/*
2838 			 * Only copy-out sources which are in-mode.
2839 			 */
2840 			if (fmode != im6s_get_mode(inm, ims, 1)) {
2841 				CTR1(KTR_MLD, "%s: skip non-in-mode",
2842 				    __func__);
2843 				continue;
2844 			}
2845 			src = ims->im6s_addr;
2846 			retval = SYSCTL_OUT(req, &src,
2847 			    sizeof(struct in6_addr));
2848 			if (retval != 0)
2849 				break;
2850 		}
2851 	}
2852 	IF_ADDR_RUNLOCK(ifp);
2853 
2854 	IN6_MULTI_LIST_UNLOCK();
2855 	IN6_MULTI_UNLOCK();
2856 
2857 	return (retval);
2858 }
2859 
2860 #ifdef KTR
2861 
2862 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2863 
2864 static const char *
2865 in6m_mode_str(const int mode)
2866 {
2867 
2868 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2869 		return (in6m_modestrs[mode]);
2870 	return ("??");
2871 }
2872 
2873 static const char *in6m_statestrs[] = {
2874 	"not-member",
2875 	"silent",
2876 	"idle",
2877 	"lazy",
2878 	"sleeping",
2879 	"awakening",
2880 	"query-pending",
2881 	"sg-query-pending",
2882 	"leaving"
2883 };
2884 
2885 static const char *
2886 in6m_state_str(const int state)
2887 {
2888 
2889 	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2890 		return (in6m_statestrs[state]);
2891 	return ("??");
2892 }
2893 
2894 /*
2895  * Dump an in6_multi structure to the console.
2896  */
2897 void
2898 in6m_print(const struct in6_multi *inm)
2899 {
2900 	int t;
2901 	char ip6tbuf[INET6_ADDRSTRLEN];
2902 
2903 	if ((ktr_mask & KTR_MLD) == 0)
2904 		return;
2905 
2906 	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2907 	printf("addr %s ifp %p(%s) ifma %p\n",
2908 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2909 	    inm->in6m_ifp,
2910 	    if_name(inm->in6m_ifp),
2911 	    inm->in6m_ifma);
2912 	printf("timer %u state %s refcount %u scq.len %u\n",
2913 	    inm->in6m_timer,
2914 	    in6m_state_str(inm->in6m_state),
2915 	    inm->in6m_refcount,
2916 	    mbufq_len(&inm->in6m_scq));
2917 	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2918 	    inm->in6m_mli,
2919 	    inm->in6m_nsrc,
2920 	    inm->in6m_sctimer,
2921 	    inm->in6m_scrv);
2922 	for (t = 0; t < 2; t++) {
2923 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2924 		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2925 		    inm->in6m_st[t].iss_asm,
2926 		    inm->in6m_st[t].iss_ex,
2927 		    inm->in6m_st[t].iss_in,
2928 		    inm->in6m_st[t].iss_rec);
2929 	}
2930 	printf("%s: --- end in6m %p ---\n", __func__, inm);
2931 }
2932 
2933 #else /* !KTR */
2934 
2935 void
2936 in6m_print(const struct in6_multi *inm)
2937 {
2938 
2939 }
2940 
2941 #endif /* KTR */
2942