xref: /freebsd/sys/netinet6/in6_mcast.c (revision b197d4b893974c9eb4d7b38704c6d5c486235d6f)
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/kernel.h>
45 #include <sys/ktr.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/taskqueue.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/route/nhop.h>
61 #include <net/vnet.h>
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, IN6_MULTI_LOCK, INP_WLOCK,
106  *   IN6_MULTI_LIST_LOCK, MLD_LOCK, IF_ADDR_LOCK.
107  * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
108  *   it can be taken by code in net/if.c also.
109  * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
110  *
111  * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
112  * any need for in6_multi itself to be virtualized -- it is bound to an ifp
113  * anyway no matter what happens.
114  */
115 struct mtx in6_multi_list_mtx;
116 MTX_SYSINIT(in6_multi_mtx, &in6_multi_list_mtx, "in6_multi_list_mtx", MTX_DEF);
117 
118 struct mtx in6_multi_free_mtx;
119 MTX_SYSINIT(in6_multi_free_mtx, &in6_multi_free_mtx, "in6_multi_free_mtx", MTX_DEF);
120 
121 struct sx in6_multi_sx;
122 SX_SYSINIT(in6_multi_sx, &in6_multi_sx, "in6_multi_sx");
123 
124 static void	im6f_commit(struct in6_mfilter *);
125 static int	im6f_get_source(struct in6_mfilter *imf,
126 		    const struct sockaddr_in6 *psin,
127 		    struct in6_msource **);
128 static struct in6_msource *
129 		im6f_graft(struct in6_mfilter *, const uint8_t,
130 		    const struct sockaddr_in6 *);
131 static void	im6f_leave(struct in6_mfilter *);
132 static int	im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
133 static void	im6f_purge(struct in6_mfilter *);
134 static void	im6f_rollback(struct in6_mfilter *);
135 static void	im6f_reap(struct in6_mfilter *);
136 static struct in6_mfilter *
137 		im6o_match_group(const struct ip6_moptions *,
138 		    const struct ifnet *, const struct sockaddr *);
139 static struct in6_msource *
140 		im6o_match_source(struct in6_mfilter *, const struct sockaddr *);
141 static void	im6s_merge(struct ip6_msource *ims,
142 		    const struct in6_msource *lims, const int rollback);
143 static int	in6_getmulti(struct ifnet *, const struct in6_addr *,
144 		    struct in6_multi **);
145 static int	in6_joingroup_locked(struct ifnet *, const struct in6_addr *,
146 		    struct in6_mfilter *, struct in6_multi **, int);
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,
172     CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
173     "IPv6 multicast");
174 
175 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
176 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
177     CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
178     "Max source filters per group");
179 
180 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
181 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
182     CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
183     "Max source filters per socket");
184 
185 /* TODO Virtualize this switch. */
186 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
187 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
188     &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
189 
190 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
191     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
192     "Per-interface stack-wide source filters");
193 
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 struct in6_mfilter *
233 ip6_mfilter_alloc(const int mflags, const int st0, const int st1)
234 {
235 	struct in6_mfilter *imf;
236 
237 	imf = malloc(sizeof(*imf), M_IN6MFILTER, mflags);
238 
239 	if (imf != NULL)
240 		im6f_init(imf, st0, st1);
241 
242 	return (imf);
243 }
244 
245 void
246 ip6_mfilter_free(struct in6_mfilter *imf)
247 {
248 
249 	im6f_purge(imf);
250 	free(imf, M_IN6MFILTER);
251 }
252 
253 /*
254  * Find an IPv6 multicast group entry for this ip6_moptions instance
255  * which matches the specified group, and optionally an interface.
256  * Return its index into the array, or -1 if not found.
257  */
258 static struct in6_mfilter *
259 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
260     const struct sockaddr *group)
261 {
262 	const struct sockaddr_in6 *gsin6;
263         struct in6_mfilter *imf;
264         struct in6_multi *inm;
265 
266         gsin6 = (const struct sockaddr_in6 *)group;
267 
268 	IP6_MFILTER_FOREACH(imf, &imo->im6o_head) {
269 		inm = imf->im6f_in6m;
270 		if (inm == NULL)
271 			continue;
272 		if ((ifp == NULL || (inm->in6m_ifp == ifp)) &&
273 		    IN6_ARE_ADDR_EQUAL(&inm->in6m_addr,
274 		    &gsin6->sin6_addr)) {
275 			break;
276 		}
277 	}
278 	return (imf);
279 }
280 
281 /*
282  * Find an IPv6 multicast source entry for this imo which matches
283  * the given group index for this socket, and source address.
284  *
285  * XXX TODO: The scope ID, if present in src, is stripped before
286  * any comparison. We SHOULD enforce scope/zone checks where the source
287  * filter entry has a link scope.
288  *
289  * NOTE: This does not check if the entry is in-mode, merely if
290  * it exists, which may not be the desired behaviour.
291  */
292 static struct in6_msource *
293 im6o_match_source(struct in6_mfilter *imf, const struct sockaddr *src)
294 {
295 	struct ip6_msource	 find;
296 	struct ip6_msource	*ims;
297 	const sockunion_t	*psa;
298 
299 	KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
300 
301 	psa = (const sockunion_t *)src;
302 	find.im6s_addr = psa->sin6.sin6_addr;
303 	in6_clearscope(&find.im6s_addr);		/* XXX */
304 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
305 
306 	return ((struct in6_msource *)ims);
307 }
308 
309 /*
310  * Perform filtering for multicast datagrams on a socket by group and source.
311  *
312  * Returns 0 if a datagram should be allowed through, or various error codes
313  * if the socket was not a member of the group, or the source was muted, etc.
314  */
315 int
316 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
317     const struct sockaddr *group, const struct sockaddr *src)
318 {
319 	struct in6_mfilter *imf;
320 	struct in6_msource *ims;
321 	int mode;
322 
323 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
324 
325 	imf = im6o_match_group(imo, ifp, group);
326 	if (imf == NULL)
327 		return (MCAST_NOTGMEMBER);
328 
329 	/*
330 	 * Check if the source was included in an (S,G) join.
331 	 * Allow reception on exclusive memberships by default,
332 	 * reject reception on inclusive memberships by default.
333 	 * Exclude source only if an in-mode exclude filter exists.
334 	 * Include source only if an in-mode include filter exists.
335 	 * NOTE: We are comparing group state here at MLD t1 (now)
336 	 * with socket-layer t0 (since last downcall).
337 	 */
338 	mode = imf->im6f_st[1];
339 	ims = im6o_match_source(imf, src);
340 
341 	if ((ims == NULL && mode == MCAST_INCLUDE) ||
342 	    (ims != NULL && ims->im6sl_st[0] != mode))
343 		return (MCAST_NOTSMEMBER);
344 
345 	return (MCAST_PASS);
346 }
347 
348 /*
349  * Find and return a reference to an in6_multi record for (ifp, group),
350  * and bump its reference count.
351  * If one does not exist, try to allocate it, and update link-layer multicast
352  * filters on ifp to listen for group.
353  * Assumes the IN6_MULTI lock is held across the call.
354  * Return 0 if successful, otherwise return an appropriate error code.
355  */
356 static int
357 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
358     struct in6_multi **pinm)
359 {
360 	struct epoch_tracker	 et;
361 	struct sockaddr_in6	 gsin6;
362 	struct ifmultiaddr	*ifma;
363 	struct in6_multi	*inm;
364 	int			 error;
365 
366 	error = 0;
367 
368 	/*
369 	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
370 	 * if_addmulti() takes this mutex itself, so we must drop and
371 	 * re-acquire around the call.
372 	 */
373 	IN6_MULTI_LOCK_ASSERT();
374 	IN6_MULTI_LIST_LOCK();
375 	IF_ADDR_WLOCK(ifp);
376 	NET_EPOCH_ENTER(et);
377 	/*
378 	 * Does ifp support IPv6 multicasts?
379 	 */
380 	if (ifp->if_afdata[AF_INET6] == NULL)
381 		error = ENODEV;
382 	else
383 		inm = in6m_lookup_locked(ifp, group);
384 	NET_EPOCH_EXIT(et);
385 
386 	if (error != 0)
387 		goto out_locked;
388 
389 	if (inm != NULL) {
390 		/*
391 		 * If we already joined this group, just bump the
392 		 * refcount and return it.
393 		 */
394 		KASSERT(inm->in6m_refcount >= 1,
395 		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
396 		in6m_acquire_locked(inm);
397 		*pinm = inm;
398 		goto out_locked;
399 	}
400 
401 	memset(&gsin6, 0, sizeof(gsin6));
402 	gsin6.sin6_family = AF_INET6;
403 	gsin6.sin6_len = sizeof(struct sockaddr_in6);
404 	gsin6.sin6_addr = *group;
405 
406 	/*
407 	 * Check if a link-layer group is already associated
408 	 * with this network-layer group on the given ifnet.
409 	 */
410 	IN6_MULTI_LIST_UNLOCK();
411 	IF_ADDR_WUNLOCK(ifp);
412 	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
413 	if (error != 0)
414 		return (error);
415 	IN6_MULTI_LIST_LOCK();
416 	IF_ADDR_WLOCK(ifp);
417 
418 	/*
419 	 * If something other than netinet6 is occupying the link-layer
420 	 * group, print a meaningful error message and back out of
421 	 * the allocation.
422 	 * Otherwise, bump the refcount on the existing network-layer
423 	 * group association and return it.
424 	 */
425 	if (ifma->ifma_protospec != NULL) {
426 		inm = (struct in6_multi *)ifma->ifma_protospec;
427 #ifdef INVARIANTS
428 		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
429 		    __func__));
430 		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
431 		    ("%s: ifma not AF_INET6", __func__));
432 		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
433 		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
434 		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
435 			panic("%s: ifma %p is inconsistent with %p (%p)",
436 			    __func__, ifma, inm, group);
437 #endif
438 		in6m_acquire_locked(inm);
439 		*pinm = inm;
440 		goto out_locked;
441 	}
442 
443 	IF_ADDR_WLOCK_ASSERT(ifp);
444 
445 	/*
446 	 * A new in6_multi record is needed; allocate and initialize it.
447 	 * We DO NOT perform an MLD join as the in6_ layer may need to
448 	 * push an initial source list down to MLD to support SSM.
449 	 *
450 	 * The initial source filter state is INCLUDE, {} as per the RFC.
451 	 * Pending state-changes per group are subject to a bounds check.
452 	 */
453 	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
454 	if (inm == NULL) {
455 		IN6_MULTI_LIST_UNLOCK();
456 		IF_ADDR_WUNLOCK(ifp);
457 		if_delmulti_ifma(ifma);
458 		return (ENOMEM);
459 	}
460 	inm->in6m_addr = *group;
461 	inm->in6m_ifp = ifp;
462 	inm->in6m_mli = MLD_IFINFO(ifp);
463 	inm->in6m_ifma = ifma;
464 	inm->in6m_refcount = 1;
465 	inm->in6m_state = MLD_NOT_MEMBER;
466 	mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
467 
468 	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
469 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
470 	RB_INIT(&inm->in6m_srcs);
471 
472 	ifma->ifma_protospec = inm;
473 	*pinm = inm;
474 
475  out_locked:
476 	IN6_MULTI_LIST_UNLOCK();
477 	IF_ADDR_WUNLOCK(ifp);
478 	return (error);
479 }
480 
481 /*
482  * Drop a reference to an in6_multi record.
483  *
484  * If the refcount drops to 0, free the in6_multi record and
485  * delete the underlying link-layer membership.
486  */
487 static void
488 in6m_release(struct in6_multi *inm)
489 {
490 	struct ifmultiaddr *ifma;
491 	struct ifnet *ifp;
492 
493 	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
494 
495 	MPASS(inm->in6m_refcount == 0);
496 	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
497 
498 	ifma = inm->in6m_ifma;
499 	ifp = inm->in6m_ifp;
500 	MPASS(ifma->ifma_llifma == NULL);
501 
502 	/* XXX this access is not covered by IF_ADDR_LOCK */
503 	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
504 	KASSERT(ifma->ifma_protospec == NULL,
505 	    ("%s: ifma_protospec != NULL", __func__));
506 	if (ifp == NULL)
507 		ifp = ifma->ifma_ifp;
508 
509 	if (ifp != NULL) {
510 		CURVNET_SET(ifp->if_vnet);
511 		in6m_purge(inm);
512 		free(inm, M_IP6MADDR);
513 		if_delmulti_ifma_flags(ifma, 1);
514 		CURVNET_RESTORE();
515 		if_rele(ifp);
516 	} else {
517 		in6m_purge(inm);
518 		free(inm, M_IP6MADDR);
519 		if_delmulti_ifma_flags(ifma, 1);
520 	}
521 }
522 
523 /*
524  * Interface detach can happen in a taskqueue thread context, so we must use a
525  * dedicated thread to avoid deadlocks when draining in6m_release tasks.
526  */
527 TASKQUEUE_DEFINE_THREAD(in6m_free);
528 static struct in6_multi_head in6m_free_list = SLIST_HEAD_INITIALIZER();
529 static void in6m_release_task(void *arg __unused, int pending __unused);
530 static struct task in6m_free_task = TASK_INITIALIZER(0, in6m_release_task, NULL);
531 
532 void
533 in6m_release_list_deferred(struct in6_multi_head *inmh)
534 {
535 	if (SLIST_EMPTY(inmh))
536 		return;
537 	mtx_lock(&in6_multi_free_mtx);
538 	SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
539 	mtx_unlock(&in6_multi_free_mtx);
540 	taskqueue_enqueue(taskqueue_in6m_free, &in6m_free_task);
541 }
542 
543 void
544 in6m_release_wait(void *arg __unused)
545 {
546 
547 	/*
548 	 * Make sure all pending multicast addresses are freed before
549 	 * the VNET or network device is destroyed:
550 	 */
551 	taskqueue_drain_all(taskqueue_in6m_free);
552 }
553 #ifdef VIMAGE
554 /* XXX-BZ FIXME, see D24914. */
555 VNET_SYSUNINIT(in6m_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, in6m_release_wait, NULL);
556 #endif
557 
558 void
559 in6m_disconnect_locked(struct in6_multi_head *inmh, struct in6_multi *inm)
560 {
561 	struct ifnet *ifp;
562 	struct ifaddr *ifa;
563 	struct in6_ifaddr *ifa6;
564 	struct in6_multi_mship *imm, *imm_tmp;
565 	struct ifmultiaddr *ifma, *ll_ifma;
566 
567 	IN6_MULTI_LIST_LOCK_ASSERT();
568 
569 	ifp = inm->in6m_ifp;
570 	if (ifp == NULL)
571 		return;		/* already called */
572 
573 	inm->in6m_ifp = NULL;
574 	IF_ADDR_WLOCK_ASSERT(ifp);
575 	ifma = inm->in6m_ifma;
576 	if (ifma == NULL)
577 		return;
578 
579 	if_ref(ifp);
580 	if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
581 		CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
582 		ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
583 	}
584 	MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
585 	if ((ll_ifma = ifma->ifma_llifma) != NULL) {
586 		MPASS(ifma != ll_ifma);
587 		ifma->ifma_llifma = NULL;
588 		MPASS(ll_ifma->ifma_llifma == NULL);
589 		MPASS(ll_ifma->ifma_ifp == ifp);
590 		if (--ll_ifma->ifma_refcount == 0) {
591 			if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
592 				CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
593 				ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
594 			}
595 			MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
596 			if_freemulti(ll_ifma);
597 		}
598 	}
599 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
600 		if (ifa->ifa_addr->sa_family != AF_INET6)
601 			continue;
602 		ifa6 = (void *)ifa;
603 		LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
604 		    i6mm_chain, imm_tmp) {
605 			if (inm == imm->i6mm_maddr) {
606 				LIST_REMOVE(imm, i6mm_chain);
607 				free(imm, M_IP6MADDR);
608 				in6m_rele_locked(inmh, inm);
609 			}
610 		}
611 	}
612 }
613 
614 static void
615 in6m_release_task(void *arg __unused, int pending __unused)
616 {
617 	struct in6_multi_head in6m_free_tmp;
618 	struct in6_multi *inm, *tinm;
619 
620 	SLIST_INIT(&in6m_free_tmp);
621 	mtx_lock(&in6_multi_free_mtx);
622 	SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
623 	mtx_unlock(&in6_multi_free_mtx);
624 	IN6_MULTI_LOCK();
625 	SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
626 		SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
627 		in6m_release(inm);
628 	}
629 	IN6_MULTI_UNLOCK();
630 }
631 
632 /*
633  * Clear recorded source entries for a group.
634  * Used by the MLD code. Caller must hold the IN6_MULTI lock.
635  * FIXME: Should reap.
636  */
637 void
638 in6m_clear_recorded(struct in6_multi *inm)
639 {
640 	struct ip6_msource	*ims;
641 
642 	IN6_MULTI_LIST_LOCK_ASSERT();
643 
644 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
645 		if (ims->im6s_stp) {
646 			ims->im6s_stp = 0;
647 			--inm->in6m_st[1].iss_rec;
648 		}
649 	}
650 	KASSERT(inm->in6m_st[1].iss_rec == 0,
651 	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
652 }
653 
654 /*
655  * Record a source as pending for a Source-Group MLDv2 query.
656  * This lives here as it modifies the shared tree.
657  *
658  * inm is the group descriptor.
659  * naddr is the address of the source to record in network-byte order.
660  *
661  * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
662  * lazy-allocate a source node in response to an SG query.
663  * Otherwise, no allocation is performed. This saves some memory
664  * with the trade-off that the source will not be reported to the
665  * router if joined in the window between the query response and
666  * the group actually being joined on the local host.
667  *
668  * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
669  * This turns off the allocation of a recorded source entry if
670  * the group has not been joined.
671  *
672  * Return 0 if the source didn't exist or was already marked as recorded.
673  * Return 1 if the source was marked as recorded by this function.
674  * Return <0 if any error occurred (negated errno code).
675  */
676 int
677 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
678 {
679 	struct ip6_msource	 find;
680 	struct ip6_msource	*ims, *nims;
681 
682 	IN6_MULTI_LIST_LOCK_ASSERT();
683 
684 	find.im6s_addr = *addr;
685 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
686 	if (ims && ims->im6s_stp)
687 		return (0);
688 	if (ims == NULL) {
689 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
690 			return (-ENOSPC);
691 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
692 		    M_NOWAIT | M_ZERO);
693 		if (nims == NULL)
694 			return (-ENOMEM);
695 		nims->im6s_addr = find.im6s_addr;
696 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
697 		++inm->in6m_nsrc;
698 		ims = nims;
699 	}
700 
701 	/*
702 	 * Mark the source as recorded and update the recorded
703 	 * source count.
704 	 */
705 	++ims->im6s_stp;
706 	++inm->in6m_st[1].iss_rec;
707 
708 	return (1);
709 }
710 
711 /*
712  * Return a pointer to an in6_msource owned by an in6_mfilter,
713  * given its source address.
714  * Lazy-allocate if needed. If this is a new entry its filter state is
715  * undefined at t0.
716  *
717  * imf is the filter set being modified.
718  * addr is the source address.
719  *
720  * SMPng: May be called with locks held; malloc must not block.
721  */
722 static int
723 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
724     struct in6_msource **plims)
725 {
726 	struct ip6_msource	 find;
727 	struct ip6_msource	*ims, *nims;
728 	struct in6_msource	*lims;
729 	int			 error;
730 
731 	error = 0;
732 	ims = NULL;
733 	lims = NULL;
734 
735 	find.im6s_addr = psin->sin6_addr;
736 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
737 	lims = (struct in6_msource *)ims;
738 	if (lims == NULL) {
739 		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
740 			return (ENOSPC);
741 		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
742 		    M_NOWAIT | M_ZERO);
743 		if (nims == NULL)
744 			return (ENOMEM);
745 		lims = (struct in6_msource *)nims;
746 		lims->im6s_addr = find.im6s_addr;
747 		lims->im6sl_st[0] = MCAST_UNDEFINED;
748 		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
749 		++imf->im6f_nsrc;
750 	}
751 
752 	*plims = lims;
753 
754 	return (error);
755 }
756 
757 /*
758  * Graft a source entry into an existing socket-layer filter set,
759  * maintaining any required invariants and checking allocations.
760  *
761  * The source is marked as being in the new filter mode at t1.
762  *
763  * Return the pointer to the new node, otherwise return NULL.
764  */
765 static struct in6_msource *
766 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
767     const struct sockaddr_in6 *psin)
768 {
769 	struct ip6_msource	*nims;
770 	struct in6_msource	*lims;
771 
772 	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
773 	    M_NOWAIT | M_ZERO);
774 	if (nims == NULL)
775 		return (NULL);
776 	lims = (struct in6_msource *)nims;
777 	lims->im6s_addr = psin->sin6_addr;
778 	lims->im6sl_st[0] = MCAST_UNDEFINED;
779 	lims->im6sl_st[1] = st1;
780 	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
781 	++imf->im6f_nsrc;
782 
783 	return (lims);
784 }
785 
786 /*
787  * Prune a source entry from an existing socket-layer filter set,
788  * maintaining any required invariants and checking allocations.
789  *
790  * The source is marked as being left at t1, it is not freed.
791  *
792  * Return 0 if no error occurred, otherwise return an errno value.
793  */
794 static int
795 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
796 {
797 	struct ip6_msource	 find;
798 	struct ip6_msource	*ims;
799 	struct in6_msource	*lims;
800 
801 	find.im6s_addr = psin->sin6_addr;
802 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
803 	if (ims == NULL)
804 		return (ENOENT);
805 	lims = (struct in6_msource *)ims;
806 	lims->im6sl_st[1] = MCAST_UNDEFINED;
807 	return (0);
808 }
809 
810 /*
811  * Revert socket-layer filter set deltas at t1 to t0 state.
812  */
813 static void
814 im6f_rollback(struct in6_mfilter *imf)
815 {
816 	struct ip6_msource	*ims, *tims;
817 	struct in6_msource	*lims;
818 
819 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
820 		lims = (struct in6_msource *)ims;
821 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
822 			/* no change at t1 */
823 			continue;
824 		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
825 			/* revert change to existing source at t1 */
826 			lims->im6sl_st[1] = lims->im6sl_st[0];
827 		} else {
828 			/* revert source added t1 */
829 			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
830 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
831 			free(ims, M_IN6MFILTER);
832 			imf->im6f_nsrc--;
833 		}
834 	}
835 	imf->im6f_st[1] = imf->im6f_st[0];
836 }
837 
838 /*
839  * Mark socket-layer filter set as INCLUDE {} at t1.
840  */
841 static void
842 im6f_leave(struct in6_mfilter *imf)
843 {
844 	struct ip6_msource	*ims;
845 	struct in6_msource	*lims;
846 
847 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
848 		lims = (struct in6_msource *)ims;
849 		lims->im6sl_st[1] = MCAST_UNDEFINED;
850 	}
851 	imf->im6f_st[1] = MCAST_INCLUDE;
852 }
853 
854 /*
855  * Mark socket-layer filter set deltas as committed.
856  */
857 static void
858 im6f_commit(struct in6_mfilter *imf)
859 {
860 	struct ip6_msource	*ims;
861 	struct in6_msource	*lims;
862 
863 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
864 		lims = (struct in6_msource *)ims;
865 		lims->im6sl_st[0] = lims->im6sl_st[1];
866 	}
867 	imf->im6f_st[0] = imf->im6f_st[1];
868 }
869 
870 /*
871  * Reap unreferenced sources from socket-layer filter set.
872  */
873 static void
874 im6f_reap(struct in6_mfilter *imf)
875 {
876 	struct ip6_msource	*ims, *tims;
877 	struct in6_msource	*lims;
878 
879 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
880 		lims = (struct in6_msource *)ims;
881 		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
882 		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
883 			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
884 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
885 			free(ims, M_IN6MFILTER);
886 			imf->im6f_nsrc--;
887 		}
888 	}
889 }
890 
891 /*
892  * Purge socket-layer filter set.
893  */
894 static void
895 im6f_purge(struct in6_mfilter *imf)
896 {
897 	struct ip6_msource	*ims, *tims;
898 
899 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
900 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
901 		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
902 		free(ims, M_IN6MFILTER);
903 		imf->im6f_nsrc--;
904 	}
905 	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
906 	KASSERT(RB_EMPTY(&imf->im6f_sources),
907 	    ("%s: im6f_sources not empty", __func__));
908 }
909 
910 /*
911  * Look up a source filter entry for a multicast group.
912  *
913  * inm is the group descriptor to work with.
914  * addr is the IPv6 address to look up.
915  * noalloc may be non-zero to suppress allocation of sources.
916  * *pims will be set to the address of the retrieved or allocated source.
917  *
918  * SMPng: NOTE: may be called with locks held.
919  * Return 0 if successful, otherwise return a non-zero error code.
920  */
921 static int
922 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
923     const int noalloc, struct ip6_msource **pims)
924 {
925 	struct ip6_msource	 find;
926 	struct ip6_msource	*ims, *nims;
927 #ifdef KTR
928 	char			 ip6tbuf[INET6_ADDRSTRLEN];
929 #endif
930 
931 	find.im6s_addr = *addr;
932 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
933 	if (ims == NULL && !noalloc) {
934 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
935 			return (ENOSPC);
936 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
937 		    M_NOWAIT | M_ZERO);
938 		if (nims == NULL)
939 			return (ENOMEM);
940 		nims->im6s_addr = *addr;
941 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
942 		++inm->in6m_nsrc;
943 		ims = nims;
944 		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
945 		    ip6_sprintf(ip6tbuf, addr), ims);
946 	}
947 
948 	*pims = ims;
949 	return (0);
950 }
951 
952 /*
953  * Merge socket-layer source into MLD-layer source.
954  * If rollback is non-zero, perform the inverse of the merge.
955  */
956 static void
957 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
958     const int rollback)
959 {
960 	int n = rollback ? -1 : 1;
961 #ifdef KTR
962 	char ip6tbuf[INET6_ADDRSTRLEN];
963 
964 	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
965 #endif
966 
967 	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
968 		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
969 		ims->im6s_st[1].ex -= n;
970 	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
971 		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
972 		ims->im6s_st[1].in -= n;
973 	}
974 
975 	if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
976 		CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
977 		ims->im6s_st[1].ex += n;
978 	} else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
979 		CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
980 		ims->im6s_st[1].in += n;
981 	}
982 }
983 
984 /*
985  * Atomically update the global in6_multi state, when a membership's
986  * filter list is being updated in any way.
987  *
988  * imf is the per-inpcb-membership group filter pointer.
989  * A fake imf may be passed for in-kernel consumers.
990  *
991  * XXX This is a candidate for a set-symmetric-difference style loop
992  * which would eliminate the repeated lookup from root of ims nodes,
993  * as they share the same key space.
994  *
995  * If any error occurred this function will back out of refcounts
996  * and return a non-zero value.
997  */
998 static int
999 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1000 {
1001 	struct ip6_msource	*ims, *nims;
1002 	struct in6_msource	*lims;
1003 	int			 schanged, error;
1004 	int			 nsrc0, nsrc1;
1005 
1006 	schanged = 0;
1007 	error = 0;
1008 	nsrc1 = nsrc0 = 0;
1009 	IN6_MULTI_LIST_LOCK_ASSERT();
1010 
1011 	/*
1012 	 * Update the source filters first, as this may fail.
1013 	 * Maintain count of in-mode filters at t0, t1. These are
1014 	 * used to work out if we transition into ASM mode or not.
1015 	 * Maintain a count of source filters whose state was
1016 	 * actually modified by this operation.
1017 	 */
1018 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1019 		lims = (struct in6_msource *)ims;
1020 		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1021 		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1022 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1023 		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1024 		++schanged;
1025 		if (error)
1026 			break;
1027 		im6s_merge(nims, lims, 0);
1028 	}
1029 	if (error) {
1030 		struct ip6_msource *bims;
1031 
1032 		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1033 			lims = (struct in6_msource *)ims;
1034 			if (lims->im6sl_st[0] == lims->im6sl_st[1])
1035 				continue;
1036 			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1037 			if (bims == NULL)
1038 				continue;
1039 			im6s_merge(bims, lims, 1);
1040 		}
1041 		goto out_reap;
1042 	}
1043 
1044 	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1045 	    __func__, nsrc0, nsrc1);
1046 
1047 	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1048 	if (imf->im6f_st[0] == imf->im6f_st[1] &&
1049 	    imf->im6f_st[1] == MCAST_INCLUDE) {
1050 		if (nsrc1 == 0) {
1051 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1052 			--inm->in6m_st[1].iss_in;
1053 		}
1054 	}
1055 
1056 	/* Handle filter mode transition on socket. */
1057 	if (imf->im6f_st[0] != imf->im6f_st[1]) {
1058 		CTR3(KTR_MLD, "%s: imf transition %d to %d",
1059 		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
1060 
1061 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1062 			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1063 			--inm->in6m_st[1].iss_ex;
1064 		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1065 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1066 			--inm->in6m_st[1].iss_in;
1067 		}
1068 
1069 		if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1070 			CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1071 			inm->in6m_st[1].iss_ex++;
1072 		} else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1073 			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1074 			inm->in6m_st[1].iss_in++;
1075 		}
1076 	}
1077 
1078 	/*
1079 	 * Track inm filter state in terms of listener counts.
1080 	 * If there are any exclusive listeners, stack-wide
1081 	 * membership is exclusive.
1082 	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1083 	 * If no listeners remain, state is undefined at t1,
1084 	 * and the MLD lifecycle for this group should finish.
1085 	 */
1086 	if (inm->in6m_st[1].iss_ex > 0) {
1087 		CTR1(KTR_MLD, "%s: transition to EX", __func__);
1088 		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1089 	} else if (inm->in6m_st[1].iss_in > 0) {
1090 		CTR1(KTR_MLD, "%s: transition to IN", __func__);
1091 		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1092 	} else {
1093 		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1094 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1095 	}
1096 
1097 	/* Decrement ASM listener count on transition out of ASM mode. */
1098 	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1099 		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1100 		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1101 			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1102 			--inm->in6m_st[1].iss_asm;
1103 		}
1104 	}
1105 
1106 	/* Increment ASM listener count on transition to ASM mode. */
1107 	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1108 		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1109 		inm->in6m_st[1].iss_asm++;
1110 	}
1111 
1112 	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1113 	in6m_print(inm);
1114 
1115 out_reap:
1116 	if (schanged > 0) {
1117 		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1118 		in6m_reap(inm);
1119 	}
1120 	return (error);
1121 }
1122 
1123 /*
1124  * Mark an in6_multi's filter set deltas as committed.
1125  * Called by MLD after a state change has been enqueued.
1126  */
1127 void
1128 in6m_commit(struct in6_multi *inm)
1129 {
1130 	struct ip6_msource	*ims;
1131 
1132 	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1133 	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1134 	in6m_print(inm);
1135 
1136 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1137 		ims->im6s_st[0] = ims->im6s_st[1];
1138 	}
1139 	inm->in6m_st[0] = inm->in6m_st[1];
1140 }
1141 
1142 /*
1143  * Reap unreferenced nodes from an in6_multi's filter set.
1144  */
1145 static void
1146 in6m_reap(struct in6_multi *inm)
1147 {
1148 	struct ip6_msource	*ims, *tims;
1149 
1150 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1151 		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1152 		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1153 		    ims->im6s_stp != 0)
1154 			continue;
1155 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1156 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1157 		free(ims, M_IP6MSOURCE);
1158 		inm->in6m_nsrc--;
1159 	}
1160 }
1161 
1162 /*
1163  * Purge all source nodes from an in6_multi's filter set.
1164  */
1165 static void
1166 in6m_purge(struct in6_multi *inm)
1167 {
1168 	struct ip6_msource	*ims, *tims;
1169 
1170 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1171 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1172 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1173 		free(ims, M_IP6MSOURCE);
1174 		inm->in6m_nsrc--;
1175 	}
1176 	/* Free state-change requests that might be queued. */
1177 	mbufq_drain(&inm->in6m_scq);
1178 }
1179 
1180 /*
1181  * Join a multicast address w/o sources.
1182  * KAME compatibility entry point.
1183  *
1184  * SMPng: Assume no mc locks held by caller.
1185  */
1186 int
1187 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1188     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1189     const int delay)
1190 {
1191 	int error;
1192 
1193 	IN6_MULTI_LOCK();
1194 	error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1195 	IN6_MULTI_UNLOCK();
1196 	return (error);
1197 }
1198 
1199 /*
1200  * Join a multicast group; real entry point.
1201  *
1202  * Only preserves atomicity at inm level.
1203  * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1204  *
1205  * If the MLD downcall fails, the group is not joined, and an error
1206  * code is returned.
1207  */
1208 static int
1209 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1210     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1211     const int delay)
1212 {
1213 	struct in6_multi_head    inmh;
1214 	struct in6_mfilter	 timf;
1215 	struct in6_multi	*inm;
1216 	struct ifmultiaddr *ifma;
1217 	int			 error;
1218 #ifdef KTR
1219 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1220 #endif
1221 
1222 	/*
1223 	 * Sanity: Check scope zone ID was set for ifp, if and
1224 	 * only if group is scoped to an interface.
1225 	 */
1226 	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1227 	    ("%s: not a multicast address", __func__));
1228 	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1229 	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1230 		KASSERT(mcaddr->s6_addr16[1] != 0,
1231 		    ("%s: scope zone ID not set", __func__));
1232 	}
1233 
1234 	IN6_MULTI_LOCK_ASSERT();
1235 	IN6_MULTI_LIST_UNLOCK_ASSERT();
1236 
1237 	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1238 	    ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1239 
1240 	error = 0;
1241 	inm = NULL;
1242 
1243 	/*
1244 	 * If no imf was specified (i.e. kernel consumer),
1245 	 * fake one up and assume it is an ASM join.
1246 	 */
1247 	if (imf == NULL) {
1248 		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1249 		imf = &timf;
1250 	}
1251 	error = in6_getmulti(ifp, mcaddr, &inm);
1252 	if (error) {
1253 		CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1254 		return (error);
1255 	}
1256 
1257 	IN6_MULTI_LIST_LOCK();
1258 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1259 	error = in6m_merge(inm, imf);
1260 	if (error) {
1261 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1262 		goto out_in6m_release;
1263 	}
1264 
1265 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1266 	error = mld_change_state(inm, delay);
1267 	if (error) {
1268 		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1269 		goto out_in6m_release;
1270 	}
1271 
1272 out_in6m_release:
1273 	SLIST_INIT(&inmh);
1274 	if (error) {
1275 		struct epoch_tracker et;
1276 
1277 		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1278 		IF_ADDR_WLOCK(ifp);
1279 		NET_EPOCH_ENTER(et);
1280 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1281 			if (ifma->ifma_protospec == inm) {
1282 				ifma->ifma_protospec = NULL;
1283 				break;
1284 			}
1285 		}
1286 		in6m_disconnect_locked(&inmh, inm);
1287 		in6m_rele_locked(&inmh, inm);
1288 		NET_EPOCH_EXIT(et);
1289 		IF_ADDR_WUNLOCK(ifp);
1290 	} else {
1291 		*pinm = inm;
1292 	}
1293 	IN6_MULTI_LIST_UNLOCK();
1294 	in6m_release_list_deferred(&inmh);
1295 	return (error);
1296 }
1297 
1298 /*
1299  * Leave a multicast group; unlocked entry point.
1300  */
1301 int
1302 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1303 {
1304 	int error;
1305 
1306 	IN6_MULTI_LOCK();
1307 	error = in6_leavegroup_locked(inm, imf);
1308 	IN6_MULTI_UNLOCK();
1309 	return (error);
1310 }
1311 
1312 /*
1313  * Leave a multicast group; real entry point.
1314  * All source filters will be expunged.
1315  *
1316  * Only preserves atomicity at inm level.
1317  *
1318  * Holding the write lock for the INP which contains imf
1319  * is highly advisable. We can't assert for it as imf does not
1320  * contain a back-pointer to the owning inp.
1321  *
1322  * Note: This is not the same as in6m_release(*) as this function also
1323  * makes a state change downcall into MLD.
1324  */
1325 int
1326 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1327 {
1328 	struct in6_multi_head	 inmh;
1329 	struct in6_mfilter	 timf;
1330 	struct ifnet *ifp;
1331 	int			 error;
1332 #ifdef KTR
1333 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1334 #endif
1335 
1336 	error = 0;
1337 
1338 	IN6_MULTI_LOCK_ASSERT();
1339 
1340 	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1341 	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1342 	    (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1343 	    imf);
1344 
1345 	/*
1346 	 * If no imf was specified (i.e. kernel consumer),
1347 	 * fake one up and assume it is an ASM join.
1348 	 */
1349 	if (imf == NULL) {
1350 		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1351 		imf = &timf;
1352 	}
1353 
1354 	/*
1355 	 * Begin state merge transaction at MLD layer.
1356 	 *
1357 	 * As this particular invocation should not cause any memory
1358 	 * to be allocated, and there is no opportunity to roll back
1359 	 * the transaction, it MUST NOT fail.
1360 	 */
1361 
1362 	ifp = inm->in6m_ifp;
1363 	IN6_MULTI_LIST_LOCK();
1364 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1365 	error = in6m_merge(inm, imf);
1366 	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1367 
1368 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1369 	error = 0;
1370 	if (ifp)
1371 		error = mld_change_state(inm, 0);
1372 	if (error)
1373 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1374 
1375 	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1376 	if (ifp)
1377 		IF_ADDR_WLOCK(ifp);
1378 
1379 	SLIST_INIT(&inmh);
1380 	if (inm->in6m_refcount == 1)
1381 		in6m_disconnect_locked(&inmh, inm);
1382 	in6m_rele_locked(&inmh, inm);
1383 	if (ifp)
1384 		IF_ADDR_WUNLOCK(ifp);
1385 	IN6_MULTI_LIST_UNLOCK();
1386 	in6m_release_list_deferred(&inmh);
1387 	return (error);
1388 }
1389 
1390 /*
1391  * Block or unblock an ASM multicast source on an inpcb.
1392  * This implements the delta-based API described in RFC 3678.
1393  *
1394  * The delta-based API applies only to exclusive-mode memberships.
1395  * An MLD downcall will be performed.
1396  *
1397  * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1398  *
1399  * Return 0 if successful, otherwise return an appropriate error code.
1400  */
1401 static int
1402 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1403 {
1404 	struct group_source_req		 gsr;
1405 	struct epoch_tracker		 et;
1406 	sockunion_t			*gsa, *ssa;
1407 	struct ifnet			*ifp;
1408 	struct in6_mfilter		*imf;
1409 	struct ip6_moptions		*imo;
1410 	struct in6_msource		*ims;
1411 	struct in6_multi			*inm;
1412 	uint16_t			 fmode;
1413 	int				 error, doblock;
1414 #ifdef KTR
1415 	char				 ip6tbuf[INET6_ADDRSTRLEN];
1416 #endif
1417 
1418 	ifp = NULL;
1419 	error = 0;
1420 	doblock = 0;
1421 
1422 	memset(&gsr, 0, sizeof(struct group_source_req));
1423 	gsa = (sockunion_t *)&gsr.gsr_group;
1424 	ssa = (sockunion_t *)&gsr.gsr_source;
1425 
1426 	switch (sopt->sopt_name) {
1427 	case MCAST_BLOCK_SOURCE:
1428 	case MCAST_UNBLOCK_SOURCE:
1429 		error = sooptcopyin(sopt, &gsr,
1430 		    sizeof(struct group_source_req),
1431 		    sizeof(struct group_source_req));
1432 		if (error)
1433 			return (error);
1434 
1435 		if (gsa->sin6.sin6_family != AF_INET6 ||
1436 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1437 			return (EINVAL);
1438 
1439 		if (ssa->sin6.sin6_family != AF_INET6 ||
1440 		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1441 			return (EINVAL);
1442 
1443 		/*
1444 		 * XXXGL: this function should use ifnet_byindex_ref, or
1445 		 * expand the epoch section all the way to where we put
1446 		 * the reference.
1447 		 */
1448 		NET_EPOCH_ENTER(et);
1449 		ifp = ifnet_byindex(gsr.gsr_interface);
1450 		NET_EPOCH_EXIT(et);
1451 		if (ifp == NULL)
1452 			return (EADDRNOTAVAIL);
1453 
1454 		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1455 			doblock = 1;
1456 		break;
1457 
1458 	default:
1459 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1460 		    __func__, sopt->sopt_name);
1461 		return (EOPNOTSUPP);
1462 		break;
1463 	}
1464 
1465 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1466 		return (EINVAL);
1467 
1468 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1469 
1470 	/*
1471 	 * Check if we are actually a member of this group.
1472 	 */
1473 	imo = in6p_findmoptions(inp);
1474 	imf = im6o_match_group(imo, ifp, &gsa->sa);
1475 	if (imf == NULL) {
1476 		error = EADDRNOTAVAIL;
1477 		goto out_in6p_locked;
1478 	}
1479 	inm = imf->im6f_in6m;
1480 
1481 	/*
1482 	 * Attempting to use the delta-based API on an
1483 	 * non exclusive-mode membership is an error.
1484 	 */
1485 	fmode = imf->im6f_st[0];
1486 	if (fmode != MCAST_EXCLUDE) {
1487 		error = EINVAL;
1488 		goto out_in6p_locked;
1489 	}
1490 
1491 	/*
1492 	 * Deal with error cases up-front:
1493 	 *  Asked to block, but already blocked; or
1494 	 *  Asked to unblock, but nothing to unblock.
1495 	 * If adding a new block entry, allocate it.
1496 	 */
1497 	ims = im6o_match_source(imf, &ssa->sa);
1498 	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1499 		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1500 		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1501 		    doblock ? "" : "not ");
1502 		error = EADDRNOTAVAIL;
1503 		goto out_in6p_locked;
1504 	}
1505 
1506 	INP_WLOCK_ASSERT(inp);
1507 
1508 	/*
1509 	 * Begin state merge transaction at socket layer.
1510 	 */
1511 	if (doblock) {
1512 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1513 		ims = im6f_graft(imf, fmode, &ssa->sin6);
1514 		if (ims == NULL)
1515 			error = ENOMEM;
1516 	} else {
1517 		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1518 		error = im6f_prune(imf, &ssa->sin6);
1519 	}
1520 
1521 	if (error) {
1522 		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1523 		goto out_im6f_rollback;
1524 	}
1525 
1526 	/*
1527 	 * Begin state merge transaction at MLD layer.
1528 	 */
1529 	IN6_MULTI_LIST_LOCK();
1530 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1531 	error = in6m_merge(inm, imf);
1532 	if (error)
1533 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1534 	else {
1535 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1536 		error = mld_change_state(inm, 0);
1537 		if (error)
1538 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1539 	}
1540 
1541 	IN6_MULTI_LIST_UNLOCK();
1542 
1543 out_im6f_rollback:
1544 	if (error)
1545 		im6f_rollback(imf);
1546 	else
1547 		im6f_commit(imf);
1548 
1549 	im6f_reap(imf);
1550 
1551 out_in6p_locked:
1552 	INP_WUNLOCK(inp);
1553 	return (error);
1554 }
1555 
1556 /*
1557  * Given an inpcb, return its multicast options structure pointer.  Accepts
1558  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1559  *
1560  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1561  * SMPng: NOTE: Returns with the INP write lock held.
1562  */
1563 static struct ip6_moptions *
1564 in6p_findmoptions(struct inpcb *inp)
1565 {
1566 	struct ip6_moptions	 *imo;
1567 
1568 	INP_WLOCK(inp);
1569 	if (inp->in6p_moptions != NULL)
1570 		return (inp->in6p_moptions);
1571 
1572 	INP_WUNLOCK(inp);
1573 
1574 	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1575 
1576 	imo->im6o_multicast_ifp = NULL;
1577 	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1578 	imo->im6o_multicast_loop = in6_mcast_loop;
1579 	STAILQ_INIT(&imo->im6o_head);
1580 
1581 	INP_WLOCK(inp);
1582 	if (inp->in6p_moptions != NULL) {
1583 		free(imo, M_IP6MOPTS);
1584 		return (inp->in6p_moptions);
1585 	}
1586 	inp->in6p_moptions = imo;
1587 	return (imo);
1588 }
1589 
1590 /*
1591  * Discard the IPv6 multicast options (and source filters).
1592  *
1593  * SMPng: NOTE: assumes INP write lock is held.
1594  *
1595  * XXX can all be safely deferred to epoch_call
1596  *
1597  */
1598 
1599 static void
1600 inp_gcmoptions(struct ip6_moptions *imo)
1601 {
1602 	struct in6_mfilter *imf;
1603 	struct in6_multi *inm;
1604 	struct ifnet *ifp;
1605 
1606 	while ((imf = ip6_mfilter_first(&imo->im6o_head)) != NULL) {
1607                 ip6_mfilter_remove(&imo->im6o_head, imf);
1608 
1609                 im6f_leave(imf);
1610                 if ((inm = imf->im6f_in6m) != NULL) {
1611                         if ((ifp = inm->in6m_ifp) != NULL) {
1612                                 CURVNET_SET(ifp->if_vnet);
1613                                 (void)in6_leavegroup(inm, imf);
1614                                 CURVNET_RESTORE();
1615                         } else {
1616                                 (void)in6_leavegroup(inm, imf);
1617                         }
1618                 }
1619                 ip6_mfilter_free(imf);
1620         }
1621         free(imo, M_IP6MOPTS);
1622 }
1623 
1624 void
1625 ip6_freemoptions(struct ip6_moptions *imo)
1626 {
1627 	if (imo == NULL)
1628 		return;
1629 	inp_gcmoptions(imo);
1630 }
1631 
1632 /*
1633  * Atomically get source filters on a socket for an IPv6 multicast group.
1634  * Called with INP lock held; returns with lock released.
1635  */
1636 static int
1637 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1638 {
1639 	struct epoch_tracker	 et;
1640 	struct __msfilterreq	 msfr;
1641 	sockunion_t		*gsa;
1642 	struct ifnet		*ifp;
1643 	struct ip6_moptions	*imo;
1644 	struct in6_mfilter	*imf;
1645 	struct ip6_msource	*ims;
1646 	struct in6_msource	*lims;
1647 	struct sockaddr_in6	*psin;
1648 	struct sockaddr_storage	*ptss;
1649 	struct sockaddr_storage	*tss;
1650 	int			 error;
1651 	size_t			 nsrcs, ncsrcs;
1652 
1653 	INP_WLOCK_ASSERT(inp);
1654 
1655 	imo = inp->in6p_moptions;
1656 	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1657 
1658 	INP_WUNLOCK(inp);
1659 
1660 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1661 	    sizeof(struct __msfilterreq));
1662 	if (error)
1663 		return (error);
1664 
1665 	if (msfr.msfr_group.ss_family != AF_INET6 ||
1666 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1667 		return (EINVAL);
1668 
1669 	gsa = (sockunion_t *)&msfr.msfr_group;
1670 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1671 		return (EINVAL);
1672 
1673 	/*
1674 	 * XXXGL: this function should use ifnet_byindex_ref, or expand the
1675 	 * epoch section all the way to where the interface is referenced.
1676 	 */
1677 	NET_EPOCH_ENTER(et);
1678 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1679 	NET_EPOCH_EXIT(et);
1680 	if (ifp == NULL)
1681 		return (EADDRNOTAVAIL);
1682 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1683 
1684 	INP_WLOCK(inp);
1685 
1686 	/*
1687 	 * Lookup group on the socket.
1688 	 */
1689 	imf = im6o_match_group(imo, ifp, &gsa->sa);
1690 	if (imf == NULL) {
1691 		INP_WUNLOCK(inp);
1692 		return (EADDRNOTAVAIL);
1693 	}
1694 
1695 	/*
1696 	 * Ignore memberships which are in limbo.
1697 	 */
1698 	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1699 		INP_WUNLOCK(inp);
1700 		return (EAGAIN);
1701 	}
1702 	msfr.msfr_fmode = imf->im6f_st[1];
1703 
1704 	/*
1705 	 * If the user specified a buffer, copy out the source filter
1706 	 * entries to userland gracefully.
1707 	 * We only copy out the number of entries which userland
1708 	 * has asked for, but we always tell userland how big the
1709 	 * buffer really needs to be.
1710 	 */
1711 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1712 		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1713 	tss = NULL;
1714 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1715 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1716 		    M_TEMP, M_NOWAIT | M_ZERO);
1717 		if (tss == NULL) {
1718 			INP_WUNLOCK(inp);
1719 			return (ENOBUFS);
1720 		}
1721 	}
1722 
1723 	/*
1724 	 * Count number of sources in-mode at t0.
1725 	 * If buffer space exists and remains, copy out source entries.
1726 	 */
1727 	nsrcs = msfr.msfr_nsrcs;
1728 	ncsrcs = 0;
1729 	ptss = tss;
1730 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1731 		lims = (struct in6_msource *)ims;
1732 		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1733 		    lims->im6sl_st[0] != imf->im6f_st[0])
1734 			continue;
1735 		++ncsrcs;
1736 		if (tss != NULL && nsrcs > 0) {
1737 			psin = (struct sockaddr_in6 *)ptss;
1738 			psin->sin6_family = AF_INET6;
1739 			psin->sin6_len = sizeof(struct sockaddr_in6);
1740 			psin->sin6_addr = lims->im6s_addr;
1741 			psin->sin6_port = 0;
1742 			--nsrcs;
1743 			++ptss;
1744 		}
1745 	}
1746 
1747 	INP_WUNLOCK(inp);
1748 
1749 	if (tss != NULL) {
1750 		error = copyout(tss, msfr.msfr_srcs,
1751 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1752 		free(tss, M_TEMP);
1753 		if (error)
1754 			return (error);
1755 	}
1756 
1757 	msfr.msfr_nsrcs = ncsrcs;
1758 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1759 
1760 	return (error);
1761 }
1762 
1763 /*
1764  * Return the IP multicast options in response to user getsockopt().
1765  */
1766 int
1767 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1768 {
1769 	struct ip6_moptions	*im6o;
1770 	int			 error;
1771 	u_int			 optval;
1772 
1773 	INP_WLOCK(inp);
1774 	im6o = inp->in6p_moptions;
1775 	/* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
1776 	if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1777 	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM) {
1778 		INP_WUNLOCK(inp);
1779 		return (EOPNOTSUPP);
1780 	}
1781 
1782 	error = 0;
1783 	switch (sopt->sopt_name) {
1784 	case IPV6_MULTICAST_IF:
1785 		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1786 			optval = 0;
1787 		} else {
1788 			optval = im6o->im6o_multicast_ifp->if_index;
1789 		}
1790 		INP_WUNLOCK(inp);
1791 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1792 		break;
1793 
1794 	case IPV6_MULTICAST_HOPS:
1795 		if (im6o == NULL)
1796 			optval = V_ip6_defmcasthlim;
1797 		else
1798 			optval = im6o->im6o_multicast_hlim;
1799 		INP_WUNLOCK(inp);
1800 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1801 		break;
1802 
1803 	case IPV6_MULTICAST_LOOP:
1804 		if (im6o == NULL)
1805 			optval = in6_mcast_loop; /* XXX VIMAGE */
1806 		else
1807 			optval = im6o->im6o_multicast_loop;
1808 		INP_WUNLOCK(inp);
1809 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1810 		break;
1811 
1812 	case IPV6_MSFILTER:
1813 		if (im6o == NULL) {
1814 			error = EADDRNOTAVAIL;
1815 			INP_WUNLOCK(inp);
1816 		} else {
1817 			error = in6p_get_source_filters(inp, sopt);
1818 		}
1819 		break;
1820 
1821 	default:
1822 		INP_WUNLOCK(inp);
1823 		error = ENOPROTOOPT;
1824 		break;
1825 	}
1826 
1827 	INP_UNLOCK_ASSERT(inp);
1828 
1829 	return (error);
1830 }
1831 
1832 /*
1833  * Look up the ifnet to use for a multicast group membership,
1834  * given the address of an IPv6 group.
1835  *
1836  * This routine exists to support legacy IPv6 multicast applications.
1837  *
1838  * Use the socket's current FIB number for any required FIB lookup. Look up the
1839  * group address in the unicast FIB, and use its ifp; usually, this points to
1840  * the default next-hop.  If the FIB lookup fails, return NULL.
1841  *
1842  * FUTURE: Support multiple forwarding tables for IPv6.
1843  *
1844  * Returns NULL if no ifp could be found.
1845  */
1846 static struct ifnet *
1847 in6p_lookup_mcast_ifp(const struct inpcb *inp, const struct sockaddr_in6 *gsin6)
1848 {
1849 	struct nhop_object	*nh;
1850 	struct in6_addr		dst;
1851 	uint32_t		scopeid;
1852 	uint32_t		fibnum;
1853 
1854 	KASSERT(gsin6->sin6_family == AF_INET6,
1855 	    ("%s: not AF_INET6 group", __func__));
1856 
1857 	in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1858 	fibnum = inp->inp_inc.inc_fibnum;
1859 	nh = fib6_lookup(fibnum, &dst, scopeid, 0, 0);
1860 
1861 	return (nh ? nh->nh_ifp : NULL);
1862 }
1863 
1864 /*
1865  * Join an IPv6 multicast group, possibly with a source.
1866  *
1867  * FIXME: The KAME use of the unspecified address (::)
1868  * to join *all* multicast groups is currently unsupported.
1869  *
1870  * XXXGL: this function multiple times uses ifnet_byindex() without
1871  * proper protection - staying in epoch, or putting reference on ifnet.
1872  */
1873 static int
1874 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1875 {
1876 	struct in6_multi_head		 inmh;
1877 	struct group_source_req		 gsr;
1878 	struct epoch_tracker		 et;
1879 	sockunion_t			*gsa, *ssa;
1880 	struct ifnet			*ifp;
1881 	struct in6_mfilter		*imf;
1882 	struct ip6_moptions		*imo;
1883 	struct in6_multi		*inm;
1884 	struct in6_msource		*lims;
1885 	int				 error, is_new;
1886 
1887 	SLIST_INIT(&inmh);
1888 	ifp = NULL;
1889 	lims = NULL;
1890 	error = 0;
1891 
1892 	memset(&gsr, 0, sizeof(struct group_source_req));
1893 	gsa = (sockunion_t *)&gsr.gsr_group;
1894 	gsa->ss.ss_family = AF_UNSPEC;
1895 	ssa = (sockunion_t *)&gsr.gsr_source;
1896 	ssa->ss.ss_family = AF_UNSPEC;
1897 
1898 	/*
1899 	 * Chew everything into struct group_source_req.
1900 	 * Overwrite the port field if present, as the sockaddr
1901 	 * being copied in may be matched with a binary comparison.
1902 	 * Ignore passed-in scope ID.
1903 	 */
1904 	switch (sopt->sopt_name) {
1905 	case IPV6_JOIN_GROUP: {
1906 		struct ipv6_mreq mreq;
1907 
1908 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1909 		    sizeof(struct ipv6_mreq));
1910 		if (error)
1911 			return (error);
1912 
1913 		gsa->sin6.sin6_family = AF_INET6;
1914 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1915 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1916 
1917 		if (mreq.ipv6mr_interface == 0) {
1918 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1919 		} else {
1920 			NET_EPOCH_ENTER(et);
1921 			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1922 			NET_EPOCH_EXIT(et);
1923 			if (ifp == NULL)
1924 				return (EADDRNOTAVAIL);
1925 		}
1926 		CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1927 		    __func__, mreq.ipv6mr_interface, ifp);
1928 	} break;
1929 
1930 	case MCAST_JOIN_GROUP:
1931 	case MCAST_JOIN_SOURCE_GROUP:
1932 		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1933 			error = sooptcopyin(sopt, &gsr,
1934 			    sizeof(struct group_req),
1935 			    sizeof(struct group_req));
1936 		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1937 			error = sooptcopyin(sopt, &gsr,
1938 			    sizeof(struct group_source_req),
1939 			    sizeof(struct group_source_req));
1940 		}
1941 		if (error)
1942 			return (error);
1943 
1944 		if (gsa->sin6.sin6_family != AF_INET6 ||
1945 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1946 			return (EINVAL);
1947 
1948 		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1949 			if (ssa->sin6.sin6_family != AF_INET6 ||
1950 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1951 				return (EINVAL);
1952 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1953 				return (EINVAL);
1954 			/*
1955 			 * TODO: Validate embedded scope ID in source
1956 			 * list entry against passed-in ifp, if and only
1957 			 * if source list filter entry is iface or node local.
1958 			 */
1959 			in6_clearscope(&ssa->sin6.sin6_addr);
1960 			ssa->sin6.sin6_port = 0;
1961 			ssa->sin6.sin6_scope_id = 0;
1962 		}
1963 		NET_EPOCH_ENTER(et);
1964 		ifp = ifnet_byindex(gsr.gsr_interface);
1965 		NET_EPOCH_EXIT(et);
1966 		if (ifp == NULL)
1967 			return (EADDRNOTAVAIL);
1968 		break;
1969 
1970 	default:
1971 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1972 		    __func__, sopt->sopt_name);
1973 		return (EOPNOTSUPP);
1974 		break;
1975 	}
1976 
1977 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1978 		return (EINVAL);
1979 
1980 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1981 		return (EADDRNOTAVAIL);
1982 
1983 	gsa->sin6.sin6_port = 0;
1984 	gsa->sin6.sin6_scope_id = 0;
1985 
1986 	/*
1987 	 * Always set the scope zone ID on memberships created from userland.
1988 	 * Use the passed-in ifp to do this.
1989 	 * XXX The in6_setscope() return value is meaningless.
1990 	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1991 	 */
1992 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1993 
1994 	IN6_MULTI_LOCK();
1995 
1996 	/*
1997 	 * Find the membership in the membership list.
1998 	 */
1999 	imo = in6p_findmoptions(inp);
2000 	imf = im6o_match_group(imo, ifp, &gsa->sa);
2001 	if (imf == NULL) {
2002 		is_new = 1;
2003 		inm = NULL;
2004 
2005 		if (ip6_mfilter_count(&imo->im6o_head) >= IPV6_MAX_MEMBERSHIPS) {
2006 			error = ENOMEM;
2007 			goto out_in6p_locked;
2008 		}
2009 	} else {
2010 		is_new = 0;
2011 		inm = imf->im6f_in6m;
2012 
2013 		if (ssa->ss.ss_family != AF_UNSPEC) {
2014 			/*
2015 			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2016 			 * is an error. On an existing inclusive membership,
2017 			 * it just adds the source to the filter list.
2018 			 */
2019 			if (imf->im6f_st[1] != MCAST_INCLUDE) {
2020 				error = EINVAL;
2021 				goto out_in6p_locked;
2022 			}
2023 			/*
2024 			 * Throw out duplicates.
2025 			 *
2026 			 * XXX FIXME: This makes a naive assumption that
2027 			 * even if entries exist for *ssa in this imf,
2028 			 * they will be rejected as dupes, even if they
2029 			 * are not valid in the current mode (in-mode).
2030 			 *
2031 			 * in6_msource is transactioned just as for anything
2032 			 * else in SSM -- but note naive use of in6m_graft()
2033 			 * below for allocating new filter entries.
2034 			 *
2035 			 * This is only an issue if someone mixes the
2036 			 * full-state SSM API with the delta-based API,
2037 			 * which is discouraged in the relevant RFCs.
2038 			 */
2039 			lims = im6o_match_source(imf, &ssa->sa);
2040 			if (lims != NULL /*&&
2041 			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2042 				error = EADDRNOTAVAIL;
2043 				goto out_in6p_locked;
2044 			}
2045 		} else {
2046 			/*
2047 			 * MCAST_JOIN_GROUP alone, on any existing membership,
2048 			 * is rejected, to stop the same inpcb tying up
2049 			 * multiple refs to the in_multi.
2050 			 * On an existing inclusive membership, this is also
2051 			 * an error; if you want to change filter mode,
2052 			 * you must use the userland API setsourcefilter().
2053 			 * XXX We don't reject this for imf in UNDEFINED
2054 			 * state at t1, because allocation of a filter
2055 			 * is atomic with allocation of a membership.
2056 			 */
2057 			error = EADDRINUSE;
2058 			goto out_in6p_locked;
2059 		}
2060 	}
2061 
2062 	/*
2063 	 * Begin state merge transaction at socket layer.
2064 	 */
2065 	INP_WLOCK_ASSERT(inp);
2066 
2067 	/*
2068 	 * Graft new source into filter list for this inpcb's
2069 	 * membership of the group. The in6_multi may not have
2070 	 * been allocated yet if this is a new membership, however,
2071 	 * the in_mfilter slot will be allocated and must be initialized.
2072 	 *
2073 	 * Note: Grafting of exclusive mode filters doesn't happen
2074 	 * in this path.
2075 	 * XXX: Should check for non-NULL lims (node exists but may
2076 	 * not be in-mode) for interop with full-state API.
2077 	 */
2078 	if (ssa->ss.ss_family != AF_UNSPEC) {
2079 		/* Membership starts in IN mode */
2080 		if (is_new) {
2081 			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2082 			imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2083 			if (imf == NULL) {
2084 				error = ENOMEM;
2085 				goto out_in6p_locked;
2086 			}
2087 		} else {
2088 			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2089 		}
2090 		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2091 		if (lims == NULL) {
2092 			CTR1(KTR_MLD, "%s: merge imf state failed",
2093 			    __func__);
2094 			error = ENOMEM;
2095 			goto out_in6p_locked;
2096 		}
2097 	} else {
2098 		/* No address specified; Membership starts in EX mode */
2099 		if (is_new) {
2100 			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2101 			imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2102 			if (imf == NULL) {
2103 				error = ENOMEM;
2104 				goto out_in6p_locked;
2105 			}
2106 		}
2107 	}
2108 
2109 	/*
2110 	 * Begin state merge transaction at MLD layer.
2111 	 */
2112 	if (is_new) {
2113 		in_pcbref(inp);
2114 		INP_WUNLOCK(inp);
2115 
2116 		error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2117 		    &imf->im6f_in6m, 0);
2118 
2119 		INP_WLOCK(inp);
2120 		if (in_pcbrele_wlocked(inp)) {
2121 			error = ENXIO;
2122 			goto out_in6p_unlocked;
2123 		}
2124 		if (error) {
2125 			goto out_in6p_locked;
2126 		}
2127 		/*
2128 		 * NOTE: Refcount from in6_joingroup_locked()
2129 		 * is protecting membership.
2130 		 */
2131 		ip6_mfilter_insert(&imo->im6o_head, imf);
2132 	} else {
2133 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2134 		IN6_MULTI_LIST_LOCK();
2135 		error = in6m_merge(inm, imf);
2136 		if (error) {
2137 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2138 			    __func__);
2139 			IN6_MULTI_LIST_UNLOCK();
2140 			im6f_rollback(imf);
2141 			im6f_reap(imf);
2142 			goto out_in6p_locked;
2143 		}
2144 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2145 		error = mld_change_state(inm, 0);
2146 		IN6_MULTI_LIST_UNLOCK();
2147 
2148 		if (error) {
2149 			CTR1(KTR_MLD, "%s: failed mld downcall",
2150 			     __func__);
2151 			im6f_rollback(imf);
2152 			im6f_reap(imf);
2153 			goto out_in6p_locked;
2154 		}
2155 	}
2156 
2157 	im6f_commit(imf);
2158 	imf = NULL;
2159 
2160 out_in6p_locked:
2161 	INP_WUNLOCK(inp);
2162 out_in6p_unlocked:
2163 	IN6_MULTI_UNLOCK();
2164 
2165 	if (is_new && imf) {
2166 		if (imf->im6f_in6m != NULL) {
2167 			struct in6_multi_head inmh;
2168 
2169 			SLIST_INIT(&inmh);
2170 			SLIST_INSERT_HEAD(&inmh, imf->im6f_in6m, in6m_defer);
2171 			in6m_release_list_deferred(&inmh);
2172 		}
2173 		ip6_mfilter_free(imf);
2174 	}
2175 	return (error);
2176 }
2177 
2178 /*
2179  * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2180  */
2181 static int
2182 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2183 {
2184 	struct ipv6_mreq		 mreq;
2185 	struct group_source_req		 gsr;
2186 	struct epoch_tracker		 et;
2187 	sockunion_t			*gsa, *ssa;
2188 	struct ifnet			*ifp;
2189 	struct in6_mfilter		*imf;
2190 	struct ip6_moptions		*imo;
2191 	struct in6_msource		*ims;
2192 	struct in6_multi		*inm;
2193 	uint32_t			 ifindex;
2194 	int				 error;
2195 	bool				 is_final;
2196 #ifdef KTR
2197 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2198 #endif
2199 
2200 	ifp = NULL;
2201 	ifindex = 0;
2202 	error = 0;
2203 	is_final = true;
2204 
2205 	memset(&gsr, 0, sizeof(struct group_source_req));
2206 	gsa = (sockunion_t *)&gsr.gsr_group;
2207 	gsa->ss.ss_family = AF_UNSPEC;
2208 	ssa = (sockunion_t *)&gsr.gsr_source;
2209 	ssa->ss.ss_family = AF_UNSPEC;
2210 
2211 	/*
2212 	 * Chew everything passed in up into a struct group_source_req
2213 	 * as that is easier to process.
2214 	 * Note: Any embedded scope ID in the multicast group passed
2215 	 * in by userland is ignored, the interface index is the recommended
2216 	 * mechanism to specify an interface; see below.
2217 	 */
2218 	switch (sopt->sopt_name) {
2219 	case IPV6_LEAVE_GROUP:
2220 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2221 		    sizeof(struct ipv6_mreq));
2222 		if (error)
2223 			return (error);
2224 		gsa->sin6.sin6_family = AF_INET6;
2225 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2226 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2227 		gsa->sin6.sin6_port = 0;
2228 		gsa->sin6.sin6_scope_id = 0;
2229 		ifindex = mreq.ipv6mr_interface;
2230 		break;
2231 
2232 	case MCAST_LEAVE_GROUP:
2233 	case MCAST_LEAVE_SOURCE_GROUP:
2234 		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2235 			error = sooptcopyin(sopt, &gsr,
2236 			    sizeof(struct group_req),
2237 			    sizeof(struct group_req));
2238 		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2239 			error = sooptcopyin(sopt, &gsr,
2240 			    sizeof(struct group_source_req),
2241 			    sizeof(struct group_source_req));
2242 		}
2243 		if (error)
2244 			return (error);
2245 
2246 		if (gsa->sin6.sin6_family != AF_INET6 ||
2247 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2248 			return (EINVAL);
2249 		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2250 			if (ssa->sin6.sin6_family != AF_INET6 ||
2251 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2252 				return (EINVAL);
2253 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2254 				return (EINVAL);
2255 			/*
2256 			 * TODO: Validate embedded scope ID in source
2257 			 * list entry against passed-in ifp, if and only
2258 			 * if source list filter entry is iface or node local.
2259 			 */
2260 			in6_clearscope(&ssa->sin6.sin6_addr);
2261 		}
2262 		gsa->sin6.sin6_port = 0;
2263 		gsa->sin6.sin6_scope_id = 0;
2264 		ifindex = gsr.gsr_interface;
2265 		break;
2266 
2267 	default:
2268 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2269 		    __func__, sopt->sopt_name);
2270 		return (EOPNOTSUPP);
2271 		break;
2272 	}
2273 
2274 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2275 		return (EINVAL);
2276 
2277 	/*
2278 	 * Validate interface index if provided. If no interface index
2279 	 * was provided separately, attempt to look the membership up
2280 	 * from the default scope as a last resort to disambiguate
2281 	 * the membership we are being asked to leave.
2282 	 * XXX SCOPE6 lock potentially taken here.
2283 	 */
2284 	if (ifindex != 0) {
2285 		NET_EPOCH_ENTER(et);
2286 		ifp = ifnet_byindex(ifindex);
2287 		NET_EPOCH_EXIT(et);	/* XXXGL: unsafe ifp */
2288 		if (ifp == NULL)
2289 			return (EADDRNOTAVAIL);
2290 		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2291 	} else {
2292 		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2293 		if (error)
2294 			return (EADDRNOTAVAIL);
2295 		/*
2296 		 * Some badly behaved applications don't pass an ifindex
2297 		 * or a scope ID, which is an API violation. In this case,
2298 		 * perform a lookup as per a v6 join.
2299 		 *
2300 		 * XXX For now, stomp on zone ID for the corner case.
2301 		 * This is not the 'KAME way', but we need to see the ifp
2302 		 * directly until such time as this implementation is
2303 		 * refactored, assuming the scope IDs are the way to go.
2304 		 */
2305 		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2306 		if (ifindex == 0) {
2307 			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2308 			    "ifp for group %s.", __func__,
2309 			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2310 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2311 		} else {
2312 			NET_EPOCH_ENTER(et);
2313 			ifp = ifnet_byindex(ifindex);
2314 			NET_EPOCH_EXIT(et);	/* XXXGL: unsafe ifp */
2315 		}
2316 		if (ifp == NULL)
2317 			return (EADDRNOTAVAIL);
2318 	}
2319 
2320 	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2321 	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2322 
2323 	IN6_MULTI_LOCK();
2324 
2325 	/*
2326 	 * Find the membership in the membership list.
2327 	 */
2328 	imo = in6p_findmoptions(inp);
2329 	imf = im6o_match_group(imo, ifp, &gsa->sa);
2330 	if (imf == NULL) {
2331 		error = EADDRNOTAVAIL;
2332 		goto out_in6p_locked;
2333 	}
2334 	inm = imf->im6f_in6m;
2335 
2336 	if (ssa->ss.ss_family != AF_UNSPEC)
2337 		is_final = false;
2338 
2339 	/*
2340 	 * Begin state merge transaction at socket layer.
2341 	 */
2342 	INP_WLOCK_ASSERT(inp);
2343 
2344 	/*
2345 	 * If we were instructed only to leave a given source, do so.
2346 	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2347 	 */
2348 	if (is_final) {
2349 		ip6_mfilter_remove(&imo->im6o_head, imf);
2350 		im6f_leave(imf);
2351 
2352 		/*
2353 		 * Give up the multicast address record to which
2354 		 * the membership points.
2355 		 */
2356 		(void)in6_leavegroup_locked(inm, imf);
2357 	} else {
2358 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2359 			error = EADDRNOTAVAIL;
2360 			goto out_in6p_locked;
2361 		}
2362 		ims = im6o_match_source(imf, &ssa->sa);
2363 		if (ims == NULL) {
2364 			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2365 			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2366 			    "not ");
2367 			error = EADDRNOTAVAIL;
2368 			goto out_in6p_locked;
2369 		}
2370 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2371 		error = im6f_prune(imf, &ssa->sin6);
2372 		if (error) {
2373 			CTR1(KTR_MLD, "%s: merge imf state failed",
2374 			    __func__);
2375 			goto out_in6p_locked;
2376 		}
2377 	}
2378 
2379 	/*
2380 	 * Begin state merge transaction at MLD layer.
2381 	 */
2382 	if (!is_final) {
2383 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2384 		IN6_MULTI_LIST_LOCK();
2385 		error = in6m_merge(inm, imf);
2386 		if (error) {
2387 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2388 			    __func__);
2389 			IN6_MULTI_LIST_UNLOCK();
2390 			im6f_rollback(imf);
2391 			im6f_reap(imf);
2392                         goto out_in6p_locked;
2393 		}
2394 
2395 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2396 		error = mld_change_state(inm, 0);
2397 		IN6_MULTI_LIST_UNLOCK();
2398 		if (error) {
2399 			CTR1(KTR_MLD, "%s: failed mld downcall",
2400 			     __func__);
2401 			im6f_rollback(imf);
2402 			im6f_reap(imf);
2403                         goto out_in6p_locked;
2404 		}
2405 	}
2406 
2407 	im6f_commit(imf);
2408 	im6f_reap(imf);
2409 
2410 out_in6p_locked:
2411 	INP_WUNLOCK(inp);
2412 
2413 	if (is_final && imf)
2414 		ip6_mfilter_free(imf);
2415 
2416 	IN6_MULTI_UNLOCK();
2417 	return (error);
2418 }
2419 
2420 /*
2421  * Select the interface for transmitting IPv6 multicast datagrams.
2422  *
2423  * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2424  * may be passed to this socket option. An address of in6addr_any or an
2425  * interface index of 0 is used to remove a previous selection.
2426  * When no interface is selected, one is chosen for every send.
2427  */
2428 static int
2429 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2430 {
2431 	struct epoch_tracker	 et;
2432 	struct ifnet		*ifp;
2433 	struct ip6_moptions	*imo;
2434 	u_int			 ifindex;
2435 	int			 error;
2436 
2437 	if (sopt->sopt_valsize != sizeof(u_int))
2438 		return (EINVAL);
2439 
2440 	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2441 	if (error)
2442 		return (error);
2443 	NET_EPOCH_ENTER(et);
2444 	if (ifindex == 0)
2445 		ifp = NULL;
2446 	else {
2447 		ifp = ifnet_byindex(ifindex);
2448 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2449 			NET_EPOCH_EXIT(et);
2450 			return (EADDRNOTAVAIL);
2451 		}
2452 	}
2453 	NET_EPOCH_EXIT(et);	/* XXXGL: unsafe ifp */
2454 	imo = in6p_findmoptions(inp);
2455 	imo->im6o_multicast_ifp = ifp;
2456 	INP_WUNLOCK(inp);
2457 
2458 	return (0);
2459 }
2460 
2461 /*
2462  * Atomically set source filters on a socket for an IPv6 multicast group.
2463  *
2464  * XXXGL: unsafely exits epoch with ifnet pointer
2465  */
2466 static int
2467 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2468 {
2469 	struct __msfilterreq	 msfr;
2470 	struct epoch_tracker	 et;
2471 	sockunion_t		*gsa;
2472 	struct ifnet		*ifp;
2473 	struct in6_mfilter	*imf;
2474 	struct ip6_moptions	*imo;
2475 	struct in6_multi		*inm;
2476 	int			 error;
2477 
2478 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2479 	    sizeof(struct __msfilterreq));
2480 	if (error)
2481 		return (error);
2482 
2483 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2484 		return (ENOBUFS);
2485 
2486 	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2487 	    msfr.msfr_fmode != MCAST_INCLUDE)
2488 		return (EINVAL);
2489 
2490 	if (msfr.msfr_group.ss_family != AF_INET6 ||
2491 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2492 		return (EINVAL);
2493 
2494 	gsa = (sockunion_t *)&msfr.msfr_group;
2495 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2496 		return (EINVAL);
2497 
2498 	gsa->sin6.sin6_port = 0;	/* ignore port */
2499 
2500 	NET_EPOCH_ENTER(et);
2501 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2502 	NET_EPOCH_EXIT(et);
2503 	if (ifp == NULL)
2504 		return (EADDRNOTAVAIL);
2505 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2506 
2507 	/*
2508 	 * Take the INP write lock.
2509 	 * Check if this socket is a member of this group.
2510 	 */
2511 	imo = in6p_findmoptions(inp);
2512 	imf = im6o_match_group(imo, ifp, &gsa->sa);
2513 	if (imf == NULL) {
2514 		error = EADDRNOTAVAIL;
2515 		goto out_in6p_locked;
2516 	}
2517 	inm = imf->im6f_in6m;
2518 
2519 	/*
2520 	 * Begin state merge transaction at socket layer.
2521 	 */
2522 	INP_WLOCK_ASSERT(inp);
2523 
2524 	imf->im6f_st[1] = msfr.msfr_fmode;
2525 
2526 	/*
2527 	 * Apply any new source filters, if present.
2528 	 * Make a copy of the user-space source vector so
2529 	 * that we may copy them with a single copyin. This
2530 	 * allows us to deal with page faults up-front.
2531 	 */
2532 	if (msfr.msfr_nsrcs > 0) {
2533 		struct in6_msource	*lims;
2534 		struct sockaddr_in6	*psin;
2535 		struct sockaddr_storage	*kss, *pkss;
2536 		int			 i;
2537 
2538 		INP_WUNLOCK(inp);
2539 
2540 		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2541 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2542 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2543 		    M_TEMP, M_WAITOK);
2544 		error = copyin(msfr.msfr_srcs, kss,
2545 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2546 		if (error) {
2547 			free(kss, M_TEMP);
2548 			return (error);
2549 		}
2550 
2551 		INP_WLOCK(inp);
2552 
2553 		/*
2554 		 * Mark all source filters as UNDEFINED at t1.
2555 		 * Restore new group filter mode, as im6f_leave()
2556 		 * will set it to INCLUDE.
2557 		 */
2558 		im6f_leave(imf);
2559 		imf->im6f_st[1] = msfr.msfr_fmode;
2560 
2561 		/*
2562 		 * Update socket layer filters at t1, lazy-allocating
2563 		 * new entries. This saves a bunch of memory at the
2564 		 * cost of one RB_FIND() per source entry; duplicate
2565 		 * entries in the msfr_nsrcs vector are ignored.
2566 		 * If we encounter an error, rollback transaction.
2567 		 *
2568 		 * XXX This too could be replaced with a set-symmetric
2569 		 * difference like loop to avoid walking from root
2570 		 * every time, as the key space is common.
2571 		 */
2572 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2573 			psin = (struct sockaddr_in6 *)pkss;
2574 			if (psin->sin6_family != AF_INET6) {
2575 				error = EAFNOSUPPORT;
2576 				break;
2577 			}
2578 			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2579 				error = EINVAL;
2580 				break;
2581 			}
2582 			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2583 				error = EINVAL;
2584 				break;
2585 			}
2586 			/*
2587 			 * TODO: Validate embedded scope ID in source
2588 			 * list entry against passed-in ifp, if and only
2589 			 * if source list filter entry is iface or node local.
2590 			 */
2591 			in6_clearscope(&psin->sin6_addr);
2592 			error = im6f_get_source(imf, psin, &lims);
2593 			if (error)
2594 				break;
2595 			lims->im6sl_st[1] = imf->im6f_st[1];
2596 		}
2597 		free(kss, M_TEMP);
2598 	}
2599 
2600 	if (error)
2601 		goto out_im6f_rollback;
2602 
2603 	INP_WLOCK_ASSERT(inp);
2604 	IN6_MULTI_LIST_LOCK();
2605 
2606 	/*
2607 	 * Begin state merge transaction at MLD layer.
2608 	 */
2609 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2610 	error = in6m_merge(inm, imf);
2611 	if (error)
2612 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2613 	else {
2614 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2615 		error = mld_change_state(inm, 0);
2616 		if (error)
2617 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2618 	}
2619 
2620 	IN6_MULTI_LIST_UNLOCK();
2621 
2622 out_im6f_rollback:
2623 	if (error)
2624 		im6f_rollback(imf);
2625 	else
2626 		im6f_commit(imf);
2627 
2628 	im6f_reap(imf);
2629 
2630 out_in6p_locked:
2631 	INP_WUNLOCK(inp);
2632 	return (error);
2633 }
2634 
2635 /*
2636  * Set the IP multicast options in response to user setsockopt().
2637  *
2638  * Many of the socket options handled in this function duplicate the
2639  * functionality of socket options in the regular unicast API. However,
2640  * it is not possible to merge the duplicate code, because the idempotence
2641  * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2642  * the effects of these options must be treated as separate and distinct.
2643  *
2644  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2645  */
2646 int
2647 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2648 {
2649 	struct ip6_moptions	*im6o;
2650 	int			 error;
2651 
2652 	error = 0;
2653 
2654 	/* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
2655 	if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2656 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)
2657 		return (EOPNOTSUPP);
2658 
2659 	switch (sopt->sopt_name) {
2660 	case IPV6_MULTICAST_IF:
2661 		error = in6p_set_multicast_if(inp, sopt);
2662 		break;
2663 
2664 	case IPV6_MULTICAST_HOPS: {
2665 		int hlim;
2666 
2667 		if (sopt->sopt_valsize != sizeof(int)) {
2668 			error = EINVAL;
2669 			break;
2670 		}
2671 		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2672 		if (error)
2673 			break;
2674 		if (hlim < -1 || hlim > 255) {
2675 			error = EINVAL;
2676 			break;
2677 		} else if (hlim == -1) {
2678 			hlim = V_ip6_defmcasthlim;
2679 		}
2680 		im6o = in6p_findmoptions(inp);
2681 		im6o->im6o_multicast_hlim = hlim;
2682 		INP_WUNLOCK(inp);
2683 		break;
2684 	}
2685 
2686 	case IPV6_MULTICAST_LOOP: {
2687 		u_int loop;
2688 
2689 		/*
2690 		 * Set the loopback flag for outgoing multicast packets.
2691 		 * Must be zero or one.
2692 		 */
2693 		if (sopt->sopt_valsize != sizeof(u_int)) {
2694 			error = EINVAL;
2695 			break;
2696 		}
2697 		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2698 		if (error)
2699 			break;
2700 		if (loop > 1) {
2701 			error = EINVAL;
2702 			break;
2703 		}
2704 		im6o = in6p_findmoptions(inp);
2705 		im6o->im6o_multicast_loop = loop;
2706 		INP_WUNLOCK(inp);
2707 		break;
2708 	}
2709 
2710 	case IPV6_JOIN_GROUP:
2711 	case MCAST_JOIN_GROUP:
2712 	case MCAST_JOIN_SOURCE_GROUP:
2713 		error = in6p_join_group(inp, sopt);
2714 		break;
2715 
2716 	case IPV6_LEAVE_GROUP:
2717 	case MCAST_LEAVE_GROUP:
2718 	case MCAST_LEAVE_SOURCE_GROUP:
2719 		error = in6p_leave_group(inp, sopt);
2720 		break;
2721 
2722 	case MCAST_BLOCK_SOURCE:
2723 	case MCAST_UNBLOCK_SOURCE:
2724 		error = in6p_block_unblock_source(inp, sopt);
2725 		break;
2726 
2727 	case IPV6_MSFILTER:
2728 		error = in6p_set_source_filters(inp, sopt);
2729 		break;
2730 
2731 	default:
2732 		error = EOPNOTSUPP;
2733 		break;
2734 	}
2735 
2736 	INP_UNLOCK_ASSERT(inp);
2737 
2738 	return (error);
2739 }
2740 
2741 /*
2742  * Expose MLD's multicast filter mode and source list(s) to userland,
2743  * keyed by (ifindex, group).
2744  * The filter mode is written out as a uint32_t, followed by
2745  * 0..n of struct in6_addr.
2746  * For use by ifmcstat(8).
2747  * SMPng: NOTE: unlocked read of ifindex space.
2748  */
2749 static int
2750 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2751 {
2752 	struct in6_addr			 mcaddr;
2753 	struct in6_addr			 src;
2754 	struct epoch_tracker		 et;
2755 	struct ifnet			*ifp;
2756 	struct ifmultiaddr		*ifma;
2757 	struct in6_multi		*inm;
2758 	struct ip6_msource		*ims;
2759 	int				*name;
2760 	int				 retval;
2761 	u_int				 namelen;
2762 	uint32_t			 fmode, ifindex;
2763 #ifdef KTR
2764 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2765 #endif
2766 
2767 	name = (int *)arg1;
2768 	namelen = arg2;
2769 
2770 	if (req->newptr != NULL)
2771 		return (EPERM);
2772 
2773 	/* int: ifindex + 4 * 32 bits of IPv6 address */
2774 	if (namelen != 5)
2775 		return (EINVAL);
2776 
2777 	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2778 	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2779 		CTR2(KTR_MLD, "%s: group %s is not multicast",
2780 		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2781 		return (EINVAL);
2782 	}
2783 
2784 	ifindex = name[0];
2785 	NET_EPOCH_ENTER(et);
2786 	ifp = ifnet_byindex(ifindex);
2787 	if (ifp == NULL) {
2788 		NET_EPOCH_EXIT(et);
2789 		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2790 		    __func__, ifindex);
2791 		return (ENOENT);
2792 	}
2793 	/*
2794 	 * Internal MLD lookups require that scope/zone ID is set.
2795 	 */
2796 	(void)in6_setscope(&mcaddr, ifp, NULL);
2797 
2798 	retval = sysctl_wire_old_buffer(req,
2799 	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2800 	if (retval) {
2801 		NET_EPOCH_EXIT(et);
2802 		return (retval);
2803 	}
2804 
2805 	IN6_MULTI_LOCK();
2806 	IN6_MULTI_LIST_LOCK();
2807 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2808 		inm = in6m_ifmultiaddr_get_inm(ifma);
2809 		if (inm == NULL)
2810 			continue;
2811 		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2812 			continue;
2813 		fmode = inm->in6m_st[1].iss_fmode;
2814 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2815 		if (retval != 0)
2816 			break;
2817 		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2818 			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2819 			/*
2820 			 * Only copy-out sources which are in-mode.
2821 			 */
2822 			if (fmode != im6s_get_mode(inm, ims, 1)) {
2823 				CTR1(KTR_MLD, "%s: skip non-in-mode",
2824 				    __func__);
2825 				continue;
2826 			}
2827 			src = ims->im6s_addr;
2828 			retval = SYSCTL_OUT(req, &src,
2829 			    sizeof(struct in6_addr));
2830 			if (retval != 0)
2831 				break;
2832 		}
2833 	}
2834 	IN6_MULTI_LIST_UNLOCK();
2835 	IN6_MULTI_UNLOCK();
2836 	NET_EPOCH_EXIT(et);
2837 
2838 	return (retval);
2839 }
2840 
2841 #ifdef KTR
2842 
2843 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2844 
2845 static const char *
2846 in6m_mode_str(const int mode)
2847 {
2848 
2849 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2850 		return (in6m_modestrs[mode]);
2851 	return ("??");
2852 }
2853 
2854 static const char *in6m_statestrs[] = {
2855 	"not-member",
2856 	"silent",
2857 	"idle",
2858 	"lazy",
2859 	"sleeping",
2860 	"awakening",
2861 	"query-pending",
2862 	"sg-query-pending",
2863 	"leaving"
2864 };
2865 
2866 static const char *
2867 in6m_state_str(const int state)
2868 {
2869 
2870 	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2871 		return (in6m_statestrs[state]);
2872 	return ("??");
2873 }
2874 
2875 /*
2876  * Dump an in6_multi structure to the console.
2877  */
2878 void
2879 in6m_print(const struct in6_multi *inm)
2880 {
2881 	int t;
2882 	char ip6tbuf[INET6_ADDRSTRLEN];
2883 
2884 	if ((ktr_mask & KTR_MLD) == 0)
2885 		return;
2886 
2887 	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2888 	printf("addr %s ifp %p(%s) ifma %p\n",
2889 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2890 	    inm->in6m_ifp,
2891 	    if_name(inm->in6m_ifp),
2892 	    inm->in6m_ifma);
2893 	printf("timer %u state %s refcount %u scq.len %u\n",
2894 	    inm->in6m_timer,
2895 	    in6m_state_str(inm->in6m_state),
2896 	    inm->in6m_refcount,
2897 	    mbufq_len(&inm->in6m_scq));
2898 	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2899 	    inm->in6m_mli,
2900 	    inm->in6m_nsrc,
2901 	    inm->in6m_sctimer,
2902 	    inm->in6m_scrv);
2903 	for (t = 0; t < 2; t++) {
2904 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2905 		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2906 		    inm->in6m_st[t].iss_asm,
2907 		    inm->in6m_st[t].iss_ex,
2908 		    inm->in6m_st[t].iss_in,
2909 		    inm->in6m_st[t].iss_rec);
2910 	}
2911 	printf("%s: --- end in6m %p ---\n", __func__, inm);
2912 }
2913 
2914 #else /* !KTR */
2915 
2916 void
2917 in6m_print(const struct in6_multi *inm)
2918 {
2919 
2920 }
2921 
2922 #endif /* KTR */
2923