xref: /freebsd/sys/netinet6/in6_mcast.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
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 #include "opt_inet6.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/ktr.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/sysctl.h>
50 #include <sys/priv.h>
51 #include <sys/taskqueue.h>
52 #include <sys/tree.h>
53 
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/if_dl.h>
57 #include <net/if_private.h>
58 #include <net/route.h>
59 #include <net/route/nhop.h>
60 #include <net/vnet.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/udp.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip_var.h>
66 #include <netinet/udp_var.h>
67 #include <netinet6/in6_fib.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet/icmp6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet/in_pcb.h>
73 #include <netinet/tcp_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/mld6_var.h>
76 #include <netinet6/scope6_var.h>
77 
78 #ifndef KTR_MLD
79 #define KTR_MLD KTR_INET6
80 #endif
81 
82 #ifndef __SOCKUNION_DECLARED
83 union sockunion {
84 	struct sockaddr_storage	ss;
85 	struct sockaddr		sa;
86 	struct sockaddr_dl	sdl;
87 	struct sockaddr_in6	sin6;
88 };
89 typedef union sockunion sockunion_t;
90 #define __SOCKUNION_DECLARED
91 #endif /* __SOCKUNION_DECLARED */
92 
93 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
94     "IPv6 multicast PCB-layer source filter");
95 MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
96 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
97 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
98     "IPv6 multicast MLD-layer source filter");
99 
100 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
101 
102 /*
103  * Locking:
104  * - Lock order is: Giant, IN6_MULTI_LOCK, INP_WLOCK,
105  *   IN6_MULTI_LIST_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 static void	im6f_commit(struct in6_mfilter *);
124 static int	im6f_get_source(struct in6_mfilter *imf,
125 		    const struct sockaddr_in6 *psin,
126 		    struct in6_msource **);
127 static struct in6_msource *
128 		im6f_graft(struct in6_mfilter *, const uint8_t,
129 		    const struct sockaddr_in6 *);
130 static void	im6f_leave(struct in6_mfilter *);
131 static int	im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
132 static void	im6f_purge(struct in6_mfilter *);
133 static void	im6f_rollback(struct in6_mfilter *);
134 static void	im6f_reap(struct in6_mfilter *);
135 static struct in6_mfilter *
136 		im6o_match_group(const struct ip6_moptions *,
137 		    const struct ifnet *, const struct sockaddr *);
138 static struct in6_msource *
139 		im6o_match_source(struct in6_mfilter *, const struct sockaddr *);
140 static void	im6s_merge(struct ip6_msource *ims,
141 		    const struct in6_msource *lims, const int rollback);
142 static int	in6_getmulti(struct ifnet *, const struct in6_addr *,
143 		    struct in6_multi **);
144 static int	in6_joingroup_locked(struct ifnet *, const struct in6_addr *,
145 		    struct in6_mfilter *, struct in6_multi **, int);
146 static int	in6m_get_source(struct in6_multi *inm,
147 		    const struct in6_addr *addr, const int noalloc,
148 		    struct ip6_msource **pims);
149 #ifdef KTR
150 static int	in6m_is_ifp_detached(const struct in6_multi *);
151 #endif
152 static int	in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
153 static void	in6m_purge(struct in6_multi *);
154 static void	in6m_reap(struct in6_multi *);
155 static struct ip6_moptions *
156 		in6p_findmoptions(struct inpcb *);
157 static int	in6p_get_source_filters(struct inpcb *, struct sockopt *);
158 static int	in6p_join_group(struct inpcb *, struct sockopt *);
159 static int	in6p_leave_group(struct inpcb *, struct sockopt *);
160 static struct ifnet *
161 		in6p_lookup_mcast_ifp(const struct inpcb *,
162 		    const struct sockaddr_in6 *);
163 static int	in6p_block_unblock_source(struct inpcb *, struct sockopt *);
164 static int	in6p_set_multicast_if(struct inpcb *, struct sockopt *);
165 static int	in6p_set_source_filters(struct inpcb *, struct sockopt *);
166 static int	sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
167 
168 SYSCTL_DECL(_net_inet6_ip6);	/* XXX Not in any common header. */
169 
170 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast,
171     CTLFLAG_RW | CTLFLAG_MPSAFE, 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 #ifdef KTR
194 /*
195  * Inline function which wraps assertions for a valid ifp.
196  * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
197  * is detached.
198  */
199 static int __inline
200 in6m_is_ifp_detached(const struct in6_multi *inm)
201 {
202 	struct ifnet *ifp;
203 
204 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
205 	ifp = inm->in6m_ifma->ifma_ifp;
206 	if (ifp != NULL) {
207 		/*
208 		 * Sanity check that network-layer notion of ifp is the
209 		 * same as that of link-layer.
210 		 */
211 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
212 	}
213 
214 	return (ifp == NULL);
215 }
216 #endif
217 
218 /*
219  * Initialize an in6_mfilter structure to a known state at t0, t1
220  * with an empty source filter list.
221  */
222 static __inline void
223 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
224 {
225 	memset(imf, 0, sizeof(struct in6_mfilter));
226 	RB_INIT(&imf->im6f_sources);
227 	imf->im6f_st[0] = st0;
228 	imf->im6f_st[1] = st1;
229 }
230 
231 struct in6_mfilter *
232 ip6_mfilter_alloc(const int mflags, const int st0, const int st1)
233 {
234 	struct in6_mfilter *imf;
235 
236 	imf = malloc(sizeof(*imf), M_IN6MFILTER, mflags);
237 
238 	if (imf != NULL)
239 		im6f_init(imf, st0, st1);
240 
241 	return (imf);
242 }
243 
244 void
245 ip6_mfilter_free(struct in6_mfilter *imf)
246 {
247 
248 	im6f_purge(imf);
249 	free(imf, M_IN6MFILTER);
250 }
251 
252 /*
253  * Find an IPv6 multicast group entry for this ip6_moptions instance
254  * which matches the specified group, and optionally an interface.
255  * Return its index into the array, or -1 if not found.
256  */
257 static struct in6_mfilter *
258 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
259     const struct sockaddr *group)
260 {
261 	const struct sockaddr_in6 *gsin6;
262         struct in6_mfilter *imf;
263         struct in6_multi *inm;
264 
265         gsin6 = (const struct sockaddr_in6 *)group;
266 
267 	IP6_MFILTER_FOREACH(imf, &imo->im6o_head) {
268 		inm = imf->im6f_in6m;
269 		if (inm == NULL)
270 			continue;
271 		if ((ifp == NULL || (inm->in6m_ifp == ifp)) &&
272 		    IN6_ARE_ADDR_EQUAL(&inm->in6m_addr,
273 		    &gsin6->sin6_addr)) {
274 			break;
275 		}
276 	}
277 	return (imf);
278 }
279 
280 /*
281  * Find an IPv6 multicast source entry for this imo which matches
282  * the given group index for this socket, and source address.
283  *
284  * XXX TODO: The scope ID, if present in src, is stripped before
285  * any comparison. We SHOULD enforce scope/zone checks where the source
286  * filter entry has a link scope.
287  *
288  * NOTE: This does not check if the entry is in-mode, merely if
289  * it exists, which may not be the desired behaviour.
290  */
291 static struct in6_msource *
292 im6o_match_source(struct in6_mfilter *imf, const struct sockaddr *src)
293 {
294 	struct ip6_msource	 find;
295 	struct ip6_msource	*ims;
296 	const sockunion_t	*psa;
297 
298 	KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
299 
300 	psa = (const sockunion_t *)src;
301 	find.im6s_addr = psa->sin6.sin6_addr;
302 	in6_clearscope(&find.im6s_addr);		/* XXX */
303 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
304 
305 	return ((struct in6_msource *)ims);
306 }
307 
308 /*
309  * Perform filtering for multicast datagrams on a socket by group and source.
310  *
311  * Returns 0 if a datagram should be allowed through, or various error codes
312  * if the socket was not a member of the group, or the source was muted, etc.
313  */
314 int
315 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
316     const struct sockaddr *group, const struct sockaddr *src)
317 {
318 	struct in6_mfilter *imf;
319 	struct in6_msource *ims;
320 	int mode;
321 
322 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
323 
324 	imf = im6o_match_group(imo, ifp, group);
325 	if (imf == NULL)
326 		return (MCAST_NOTGMEMBER);
327 
328 	/*
329 	 * Check if the source was included in an (S,G) join.
330 	 * Allow reception on exclusive memberships by default,
331 	 * reject reception on inclusive memberships by default.
332 	 * Exclude source only if an in-mode exclude filter exists.
333 	 * Include source only if an in-mode include filter exists.
334 	 * NOTE: We are comparing group state here at MLD t1 (now)
335 	 * with socket-layer t0 (since last downcall).
336 	 */
337 	mode = imf->im6f_st[1];
338 	ims = im6o_match_source(imf, src);
339 
340 	if ((ims == NULL && mode == MCAST_INCLUDE) ||
341 	    (ims != NULL && ims->im6sl_st[0] != mode))
342 		return (MCAST_NOTSMEMBER);
343 
344 	return (MCAST_PASS);
345 }
346 
347 /*
348  * Look up an in6_multi record for an IPv6 multicast address
349  * on the interface ifp.
350  * If no record found, return NULL.
351  *
352  * SMPng: The IN6_MULTI_LOCK and must be held and must be in network epoch.
353  */
354 struct in6_multi *
355 in6m_lookup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr)
356 {
357 	struct ifmultiaddr *ifma;
358 	struct in6_multi *inm;
359 
360 	NET_EPOCH_ASSERT();
361 
362 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
363 		inm = in6m_ifmultiaddr_get_inm(ifma);
364 		if (inm == NULL)
365 			continue;
366 		if (IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, mcaddr))
367 			return (inm);
368 	}
369 	return (NULL);
370 }
371 
372 /*
373  * Find and return a reference to an in6_multi record for (ifp, group),
374  * and bump its reference count.
375  * If one does not exist, try to allocate it, and update link-layer multicast
376  * filters on ifp to listen for group.
377  * Assumes the IN6_MULTI lock is held across the call.
378  * Return 0 if successful, otherwise return an appropriate error code.
379  */
380 static int
381 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
382     struct in6_multi **pinm)
383 {
384 	struct epoch_tracker	 et;
385 	struct sockaddr_in6	 gsin6;
386 	struct ifmultiaddr	*ifma;
387 	struct in6_multi	*inm;
388 	int			 error;
389 
390 	error = 0;
391 
392 	/*
393 	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
394 	 * if_addmulti() takes this mutex itself, so we must drop and
395 	 * re-acquire around the call.
396 	 */
397 	IN6_MULTI_LOCK_ASSERT();
398 	IN6_MULTI_LIST_LOCK();
399 	IF_ADDR_WLOCK(ifp);
400 	NET_EPOCH_ENTER(et);
401 	/*
402 	 * Does ifp support IPv6 multicasts?
403 	 */
404 	if (ifp->if_afdata[AF_INET6] == NULL)
405 		error = ENODEV;
406 	else
407 		inm = in6m_lookup_locked(ifp, group);
408 	NET_EPOCH_EXIT(et);
409 
410 	if (error != 0)
411 		goto out_locked;
412 
413 	if (inm != NULL) {
414 		/*
415 		 * If we already joined this group, just bump the
416 		 * refcount and return it.
417 		 */
418 		KASSERT(inm->in6m_refcount >= 1,
419 		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
420 		in6m_acquire_locked(inm);
421 		*pinm = inm;
422 		goto out_locked;
423 	}
424 
425 	memset(&gsin6, 0, sizeof(gsin6));
426 	gsin6.sin6_family = AF_INET6;
427 	gsin6.sin6_len = sizeof(struct sockaddr_in6);
428 	gsin6.sin6_addr = *group;
429 
430 	/*
431 	 * Check if a link-layer group is already associated
432 	 * with this network-layer group on the given ifnet.
433 	 */
434 	IN6_MULTI_LIST_UNLOCK();
435 	IF_ADDR_WUNLOCK(ifp);
436 	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
437 	if (error != 0)
438 		return (error);
439 	IN6_MULTI_LIST_LOCK();
440 	IF_ADDR_WLOCK(ifp);
441 
442 	/*
443 	 * If something other than netinet6 is occupying the link-layer
444 	 * group, print a meaningful error message and back out of
445 	 * the allocation.
446 	 * Otherwise, bump the refcount on the existing network-layer
447 	 * group association and return it.
448 	 */
449 	if (ifma->ifma_protospec != NULL) {
450 		inm = (struct in6_multi *)ifma->ifma_protospec;
451 #ifdef INVARIANTS
452 		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
453 		    __func__));
454 		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
455 		    ("%s: ifma not AF_INET6", __func__));
456 		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
457 		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
458 		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
459 			panic("%s: ifma %p is inconsistent with %p (%p)",
460 			    __func__, ifma, inm, group);
461 #endif
462 		in6m_acquire_locked(inm);
463 		*pinm = inm;
464 		goto out_locked;
465 	}
466 
467 	IF_ADDR_WLOCK_ASSERT(ifp);
468 
469 	/*
470 	 * A new in6_multi record is needed; allocate and initialize it.
471 	 * We DO NOT perform an MLD join as the in6_ layer may need to
472 	 * push an initial source list down to MLD to support SSM.
473 	 *
474 	 * The initial source filter state is INCLUDE, {} as per the RFC.
475 	 * Pending state-changes per group are subject to a bounds check.
476 	 */
477 	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
478 	if (inm == NULL) {
479 		IN6_MULTI_LIST_UNLOCK();
480 		IF_ADDR_WUNLOCK(ifp);
481 		if_delmulti_ifma(ifma);
482 		return (ENOMEM);
483 	}
484 	inm->in6m_addr = *group;
485 	inm->in6m_ifp = ifp;
486 	inm->in6m_mli = MLD_IFINFO(ifp);
487 	inm->in6m_ifma = ifma;
488 	inm->in6m_refcount = 1;
489 	inm->in6m_state = MLD_NOT_MEMBER;
490 	mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
491 
492 	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
493 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
494 	RB_INIT(&inm->in6m_srcs);
495 
496 	ifma->ifma_protospec = inm;
497 	*pinm = inm;
498 
499  out_locked:
500 	IN6_MULTI_LIST_UNLOCK();
501 	IF_ADDR_WUNLOCK(ifp);
502 	return (error);
503 }
504 
505 /*
506  * Drop a reference to an in6_multi record.
507  *
508  * If the refcount drops to 0, free the in6_multi record and
509  * delete the underlying link-layer membership.
510  */
511 static void
512 in6m_release(struct in6_multi *inm)
513 {
514 	struct ifmultiaddr *ifma;
515 	struct ifnet *ifp;
516 
517 	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
518 
519 	MPASS(inm->in6m_refcount == 0);
520 	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
521 
522 	ifma = inm->in6m_ifma;
523 	ifp = inm->in6m_ifp;
524 	MPASS(ifma->ifma_llifma == NULL);
525 
526 	/* XXX this access is not covered by IF_ADDR_LOCK */
527 	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
528 	KASSERT(ifma->ifma_protospec == NULL,
529 	    ("%s: ifma_protospec != NULL", __func__));
530 	if (ifp == NULL)
531 		ifp = ifma->ifma_ifp;
532 
533 	if (ifp != NULL) {
534 		CURVNET_SET(ifp->if_vnet);
535 		in6m_purge(inm);
536 		free(inm, M_IP6MADDR);
537 		if_delmulti_ifma_flags(ifma, 1);
538 		CURVNET_RESTORE();
539 		if_rele(ifp);
540 	} else {
541 		in6m_purge(inm);
542 		free(inm, M_IP6MADDR);
543 		if_delmulti_ifma_flags(ifma, 1);
544 	}
545 }
546 
547 /*
548  * Interface detach can happen in a taskqueue thread context, so we must use a
549  * dedicated thread to avoid deadlocks when draining in6m_release tasks.
550  */
551 TASKQUEUE_DEFINE_THREAD(in6m_free);
552 static struct in6_multi_head in6m_free_list = SLIST_HEAD_INITIALIZER();
553 static void in6m_release_task(void *arg __unused, int pending __unused);
554 static struct task in6m_free_task = TASK_INITIALIZER(0, in6m_release_task, NULL);
555 
556 void
557 in6m_release_list_deferred(struct in6_multi_head *inmh)
558 {
559 	if (SLIST_EMPTY(inmh))
560 		return;
561 	mtx_lock(&in6_multi_free_mtx);
562 	SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
563 	mtx_unlock(&in6_multi_free_mtx);
564 	taskqueue_enqueue(taskqueue_in6m_free, &in6m_free_task);
565 }
566 
567 void
568 in6m_release_wait(void *arg __unused)
569 {
570 
571 	/*
572 	 * Make sure all pending multicast addresses are freed before
573 	 * the VNET or network device is destroyed:
574 	 */
575 	taskqueue_drain_all(taskqueue_in6m_free);
576 }
577 #ifdef VIMAGE
578 /* XXX-BZ FIXME, see D24914. */
579 VNET_SYSUNINIT(in6m_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, in6m_release_wait, NULL);
580 #endif
581 
582 void
583 in6m_disconnect_locked(struct in6_multi_head *inmh, struct in6_multi *inm)
584 {
585 	struct ifnet *ifp;
586 	struct ifaddr *ifa;
587 	struct in6_ifaddr *ifa6;
588 	struct in6_multi_mship *imm, *imm_tmp;
589 	struct ifmultiaddr *ifma, *ll_ifma;
590 
591 	IN6_MULTI_LIST_LOCK_ASSERT();
592 
593 	ifp = inm->in6m_ifp;
594 	if (ifp == NULL)
595 		return;		/* already called */
596 
597 	inm->in6m_ifp = NULL;
598 	IF_ADDR_WLOCK_ASSERT(ifp);
599 	ifma = inm->in6m_ifma;
600 	if (ifma == NULL)
601 		return;
602 
603 	if_ref(ifp);
604 	if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
605 		CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
606 		ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
607 	}
608 	MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
609 	if ((ll_ifma = ifma->ifma_llifma) != NULL) {
610 		MPASS(ifma != ll_ifma);
611 		ifma->ifma_llifma = NULL;
612 		MPASS(ll_ifma->ifma_llifma == NULL);
613 		MPASS(ll_ifma->ifma_ifp == ifp);
614 		if (--ll_ifma->ifma_refcount == 0) {
615 			if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
616 				CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
617 				ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
618 			}
619 			MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
620 			if_freemulti(ll_ifma);
621 		}
622 	}
623 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
624 		if (ifa->ifa_addr->sa_family != AF_INET6)
625 			continue;
626 		ifa6 = (void *)ifa;
627 		LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
628 		    i6mm_chain, imm_tmp) {
629 			if (inm == imm->i6mm_maddr) {
630 				LIST_REMOVE(imm, i6mm_chain);
631 				free(imm, M_IP6MADDR);
632 				in6m_rele_locked(inmh, inm);
633 			}
634 		}
635 	}
636 }
637 
638 static void
639 in6m_release_task(void *arg __unused, int pending __unused)
640 {
641 	struct in6_multi_head in6m_free_tmp;
642 	struct in6_multi *inm, *tinm;
643 
644 	SLIST_INIT(&in6m_free_tmp);
645 	mtx_lock(&in6_multi_free_mtx);
646 	SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
647 	mtx_unlock(&in6_multi_free_mtx);
648 	IN6_MULTI_LOCK();
649 	SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
650 		SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
651 		in6m_release(inm);
652 	}
653 	IN6_MULTI_UNLOCK();
654 }
655 
656 /*
657  * Clear recorded source entries for a group.
658  * Used by the MLD code. Caller must hold the IN6_MULTI lock.
659  * FIXME: Should reap.
660  */
661 void
662 in6m_clear_recorded(struct in6_multi *inm)
663 {
664 	struct ip6_msource	*ims;
665 
666 	IN6_MULTI_LIST_LOCK_ASSERT();
667 
668 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
669 		if (ims->im6s_stp) {
670 			ims->im6s_stp = 0;
671 			--inm->in6m_st[1].iss_rec;
672 		}
673 	}
674 	KASSERT(inm->in6m_st[1].iss_rec == 0,
675 	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
676 }
677 
678 /*
679  * Record a source as pending for a Source-Group MLDv2 query.
680  * This lives here as it modifies the shared tree.
681  *
682  * inm is the group descriptor.
683  * naddr is the address of the source to record in network-byte order.
684  *
685  * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
686  * lazy-allocate a source node in response to an SG query.
687  * Otherwise, no allocation is performed. This saves some memory
688  * with the trade-off that the source will not be reported to the
689  * router if joined in the window between the query response and
690  * the group actually being joined on the local host.
691  *
692  * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
693  * This turns off the allocation of a recorded source entry if
694  * the group has not been joined.
695  *
696  * Return 0 if the source didn't exist or was already marked as recorded.
697  * Return 1 if the source was marked as recorded by this function.
698  * Return <0 if any error occurred (negated errno code).
699  */
700 int
701 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
702 {
703 	struct ip6_msource	 find;
704 	struct ip6_msource	*ims, *nims;
705 
706 	IN6_MULTI_LIST_LOCK_ASSERT();
707 
708 	find.im6s_addr = *addr;
709 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
710 	if (ims && ims->im6s_stp)
711 		return (0);
712 	if (ims == NULL) {
713 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
714 			return (-ENOSPC);
715 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
716 		    M_NOWAIT | M_ZERO);
717 		if (nims == NULL)
718 			return (-ENOMEM);
719 		nims->im6s_addr = find.im6s_addr;
720 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
721 		++inm->in6m_nsrc;
722 		ims = nims;
723 	}
724 
725 	/*
726 	 * Mark the source as recorded and update the recorded
727 	 * source count.
728 	 */
729 	++ims->im6s_stp;
730 	++inm->in6m_st[1].iss_rec;
731 
732 	return (1);
733 }
734 
735 /*
736  * Return a pointer to an in6_msource owned by an in6_mfilter,
737  * given its source address.
738  * Lazy-allocate if needed. If this is a new entry its filter state is
739  * undefined at t0.
740  *
741  * imf is the filter set being modified.
742  * addr is the source address.
743  *
744  * SMPng: May be called with locks held; malloc must not block.
745  */
746 static int
747 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
748     struct in6_msource **plims)
749 {
750 	struct ip6_msource	 find;
751 	struct ip6_msource	*ims, *nims;
752 	struct in6_msource	*lims;
753 	int			 error;
754 
755 	error = 0;
756 	ims = NULL;
757 	lims = NULL;
758 
759 	find.im6s_addr = psin->sin6_addr;
760 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
761 	lims = (struct in6_msource *)ims;
762 	if (lims == NULL) {
763 		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
764 			return (ENOSPC);
765 		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
766 		    M_NOWAIT | M_ZERO);
767 		if (nims == NULL)
768 			return (ENOMEM);
769 		lims = (struct in6_msource *)nims;
770 		lims->im6s_addr = find.im6s_addr;
771 		lims->im6sl_st[0] = MCAST_UNDEFINED;
772 		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
773 		++imf->im6f_nsrc;
774 	}
775 
776 	*plims = lims;
777 
778 	return (error);
779 }
780 
781 /*
782  * Graft a source entry into an existing socket-layer filter set,
783  * maintaining any required invariants and checking allocations.
784  *
785  * The source is marked as being in the new filter mode at t1.
786  *
787  * Return the pointer to the new node, otherwise return NULL.
788  */
789 static struct in6_msource *
790 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
791     const struct sockaddr_in6 *psin)
792 {
793 	struct ip6_msource	*nims;
794 	struct in6_msource	*lims;
795 
796 	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
797 	    M_NOWAIT | M_ZERO);
798 	if (nims == NULL)
799 		return (NULL);
800 	lims = (struct in6_msource *)nims;
801 	lims->im6s_addr = psin->sin6_addr;
802 	lims->im6sl_st[0] = MCAST_UNDEFINED;
803 	lims->im6sl_st[1] = st1;
804 	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
805 	++imf->im6f_nsrc;
806 
807 	return (lims);
808 }
809 
810 /*
811  * Prune a source entry from an existing socket-layer filter set,
812  * maintaining any required invariants and checking allocations.
813  *
814  * The source is marked as being left at t1, it is not freed.
815  *
816  * Return 0 if no error occurred, otherwise return an errno value.
817  */
818 static int
819 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
820 {
821 	struct ip6_msource	 find;
822 	struct ip6_msource	*ims;
823 	struct in6_msource	*lims;
824 
825 	find.im6s_addr = psin->sin6_addr;
826 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
827 	if (ims == NULL)
828 		return (ENOENT);
829 	lims = (struct in6_msource *)ims;
830 	lims->im6sl_st[1] = MCAST_UNDEFINED;
831 	return (0);
832 }
833 
834 /*
835  * Revert socket-layer filter set deltas at t1 to t0 state.
836  */
837 static void
838 im6f_rollback(struct in6_mfilter *imf)
839 {
840 	struct ip6_msource	*ims, *tims;
841 	struct in6_msource	*lims;
842 
843 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
844 		lims = (struct in6_msource *)ims;
845 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
846 			/* no change at t1 */
847 			continue;
848 		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
849 			/* revert change to existing source at t1 */
850 			lims->im6sl_st[1] = lims->im6sl_st[0];
851 		} else {
852 			/* revert source added t1 */
853 			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
854 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
855 			free(ims, M_IN6MFILTER);
856 			imf->im6f_nsrc--;
857 		}
858 	}
859 	imf->im6f_st[1] = imf->im6f_st[0];
860 }
861 
862 /*
863  * Mark socket-layer filter set as INCLUDE {} at t1.
864  */
865 static void
866 im6f_leave(struct in6_mfilter *imf)
867 {
868 	struct ip6_msource	*ims;
869 	struct in6_msource	*lims;
870 
871 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
872 		lims = (struct in6_msource *)ims;
873 		lims->im6sl_st[1] = MCAST_UNDEFINED;
874 	}
875 	imf->im6f_st[1] = MCAST_INCLUDE;
876 }
877 
878 /*
879  * Mark socket-layer filter set deltas as committed.
880  */
881 static void
882 im6f_commit(struct in6_mfilter *imf)
883 {
884 	struct ip6_msource	*ims;
885 	struct in6_msource	*lims;
886 
887 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
888 		lims = (struct in6_msource *)ims;
889 		lims->im6sl_st[0] = lims->im6sl_st[1];
890 	}
891 	imf->im6f_st[0] = imf->im6f_st[1];
892 }
893 
894 /*
895  * Reap unreferenced sources from socket-layer filter set.
896  */
897 static void
898 im6f_reap(struct in6_mfilter *imf)
899 {
900 	struct ip6_msource	*ims, *tims;
901 	struct in6_msource	*lims;
902 
903 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
904 		lims = (struct in6_msource *)ims;
905 		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
906 		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
907 			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
908 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
909 			free(ims, M_IN6MFILTER);
910 			imf->im6f_nsrc--;
911 		}
912 	}
913 }
914 
915 /*
916  * Purge socket-layer filter set.
917  */
918 static void
919 im6f_purge(struct in6_mfilter *imf)
920 {
921 	struct ip6_msource	*ims, *tims;
922 
923 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
924 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
925 		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
926 		free(ims, M_IN6MFILTER);
927 		imf->im6f_nsrc--;
928 	}
929 	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
930 	KASSERT(RB_EMPTY(&imf->im6f_sources),
931 	    ("%s: im6f_sources not empty", __func__));
932 }
933 
934 /*
935  * Look up a source filter entry for a multicast group.
936  *
937  * inm is the group descriptor to work with.
938  * addr is the IPv6 address to look up.
939  * noalloc may be non-zero to suppress allocation of sources.
940  * *pims will be set to the address of the retrieved or allocated source.
941  *
942  * SMPng: NOTE: may be called with locks held.
943  * Return 0 if successful, otherwise return a non-zero error code.
944  */
945 static int
946 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
947     const int noalloc, struct ip6_msource **pims)
948 {
949 	struct ip6_msource	 find;
950 	struct ip6_msource	*ims, *nims;
951 #ifdef KTR
952 	char			 ip6tbuf[INET6_ADDRSTRLEN];
953 #endif
954 
955 	find.im6s_addr = *addr;
956 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
957 	if (ims == NULL && !noalloc) {
958 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
959 			return (ENOSPC);
960 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
961 		    M_NOWAIT | M_ZERO);
962 		if (nims == NULL)
963 			return (ENOMEM);
964 		nims->im6s_addr = *addr;
965 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
966 		++inm->in6m_nsrc;
967 		ims = nims;
968 		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
969 		    ip6_sprintf(ip6tbuf, addr), ims);
970 	}
971 
972 	*pims = ims;
973 	return (0);
974 }
975 
976 /*
977  * Merge socket-layer source into MLD-layer source.
978  * If rollback is non-zero, perform the inverse of the merge.
979  */
980 static void
981 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
982     const int rollback)
983 {
984 	int n = rollback ? -1 : 1;
985 #ifdef KTR
986 	char ip6tbuf[INET6_ADDRSTRLEN];
987 
988 	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
989 #endif
990 
991 	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
992 		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
993 		ims->im6s_st[1].ex -= n;
994 	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
995 		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
996 		ims->im6s_st[1].in -= n;
997 	}
998 
999 	if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
1000 		CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
1001 		ims->im6s_st[1].ex += n;
1002 	} else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
1003 		CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
1004 		ims->im6s_st[1].in += n;
1005 	}
1006 }
1007 
1008 /*
1009  * Atomically update the global in6_multi state, when a membership's
1010  * filter list is being updated in any way.
1011  *
1012  * imf is the per-inpcb-membership group filter pointer.
1013  * A fake imf may be passed for in-kernel consumers.
1014  *
1015  * XXX This is a candidate for a set-symmetric-difference style loop
1016  * which would eliminate the repeated lookup from root of ims nodes,
1017  * as they share the same key space.
1018  *
1019  * If any error occurred this function will back out of refcounts
1020  * and return a non-zero value.
1021  */
1022 static int
1023 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1024 {
1025 	struct ip6_msource	*ims, *nims;
1026 	struct in6_msource	*lims;
1027 	int			 schanged, error;
1028 	int			 nsrc0, nsrc1;
1029 
1030 	schanged = 0;
1031 	error = 0;
1032 	nsrc1 = nsrc0 = 0;
1033 	IN6_MULTI_LIST_LOCK_ASSERT();
1034 
1035 	/*
1036 	 * Update the source filters first, as this may fail.
1037 	 * Maintain count of in-mode filters at t0, t1. These are
1038 	 * used to work out if we transition into ASM mode or not.
1039 	 * Maintain a count of source filters whose state was
1040 	 * actually modified by this operation.
1041 	 */
1042 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1043 		lims = (struct in6_msource *)ims;
1044 		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1045 		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1046 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1047 		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1048 		++schanged;
1049 		if (error)
1050 			break;
1051 		im6s_merge(nims, lims, 0);
1052 	}
1053 	if (error) {
1054 		struct ip6_msource *bims;
1055 
1056 		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1057 			lims = (struct in6_msource *)ims;
1058 			if (lims->im6sl_st[0] == lims->im6sl_st[1])
1059 				continue;
1060 			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1061 			if (bims == NULL)
1062 				continue;
1063 			im6s_merge(bims, lims, 1);
1064 		}
1065 		goto out_reap;
1066 	}
1067 
1068 	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1069 	    __func__, nsrc0, nsrc1);
1070 
1071 	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1072 	if (imf->im6f_st[0] == imf->im6f_st[1] &&
1073 	    imf->im6f_st[1] == MCAST_INCLUDE) {
1074 		if (nsrc1 == 0) {
1075 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1076 			--inm->in6m_st[1].iss_in;
1077 		}
1078 	}
1079 
1080 	/* Handle filter mode transition on socket. */
1081 	if (imf->im6f_st[0] != imf->im6f_st[1]) {
1082 		CTR3(KTR_MLD, "%s: imf transition %d to %d",
1083 		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
1084 
1085 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1086 			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1087 			--inm->in6m_st[1].iss_ex;
1088 		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1089 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1090 			--inm->in6m_st[1].iss_in;
1091 		}
1092 
1093 		if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1094 			CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1095 			inm->in6m_st[1].iss_ex++;
1096 		} else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1097 			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1098 			inm->in6m_st[1].iss_in++;
1099 		}
1100 	}
1101 
1102 	/*
1103 	 * Track inm filter state in terms of listener counts.
1104 	 * If there are any exclusive listeners, stack-wide
1105 	 * membership is exclusive.
1106 	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1107 	 * If no listeners remain, state is undefined at t1,
1108 	 * and the MLD lifecycle for this group should finish.
1109 	 */
1110 	if (inm->in6m_st[1].iss_ex > 0) {
1111 		CTR1(KTR_MLD, "%s: transition to EX", __func__);
1112 		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1113 	} else if (inm->in6m_st[1].iss_in > 0) {
1114 		CTR1(KTR_MLD, "%s: transition to IN", __func__);
1115 		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1116 	} else {
1117 		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1118 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1119 	}
1120 
1121 	/* Decrement ASM listener count on transition out of ASM mode. */
1122 	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1123 		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1124 		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1125 			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1126 			--inm->in6m_st[1].iss_asm;
1127 		}
1128 	}
1129 
1130 	/* Increment ASM listener count on transition to ASM mode. */
1131 	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1132 		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1133 		inm->in6m_st[1].iss_asm++;
1134 	}
1135 
1136 	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1137 	in6m_print(inm);
1138 
1139 out_reap:
1140 	if (schanged > 0) {
1141 		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1142 		in6m_reap(inm);
1143 	}
1144 	return (error);
1145 }
1146 
1147 /*
1148  * Mark an in6_multi's filter set deltas as committed.
1149  * Called by MLD after a state change has been enqueued.
1150  */
1151 void
1152 in6m_commit(struct in6_multi *inm)
1153 {
1154 	struct ip6_msource	*ims;
1155 
1156 	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1157 	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1158 	in6m_print(inm);
1159 
1160 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1161 		ims->im6s_st[0] = ims->im6s_st[1];
1162 	}
1163 	inm->in6m_st[0] = inm->in6m_st[1];
1164 }
1165 
1166 /*
1167  * Reap unreferenced nodes from an in6_multi's filter set.
1168  */
1169 static void
1170 in6m_reap(struct in6_multi *inm)
1171 {
1172 	struct ip6_msource	*ims, *tims;
1173 
1174 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1175 		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1176 		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1177 		    ims->im6s_stp != 0)
1178 			continue;
1179 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1180 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1181 		free(ims, M_IP6MSOURCE);
1182 		inm->in6m_nsrc--;
1183 	}
1184 }
1185 
1186 /*
1187  * Purge all source nodes from an in6_multi's filter set.
1188  */
1189 static void
1190 in6m_purge(struct in6_multi *inm)
1191 {
1192 	struct ip6_msource	*ims, *tims;
1193 
1194 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1195 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1196 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1197 		free(ims, M_IP6MSOURCE);
1198 		inm->in6m_nsrc--;
1199 	}
1200 	/* Free state-change requests that might be queued. */
1201 	mbufq_drain(&inm->in6m_scq);
1202 }
1203 
1204 /*
1205  * Join a multicast address w/o sources.
1206  * KAME compatibility entry point.
1207  *
1208  * SMPng: Assume no mc locks held by caller.
1209  */
1210 int
1211 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1212     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1213     const int delay)
1214 {
1215 	int error;
1216 
1217 	IN6_MULTI_LOCK();
1218 	error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1219 	IN6_MULTI_UNLOCK();
1220 	return (error);
1221 }
1222 
1223 /*
1224  * Join a multicast group; real entry point.
1225  *
1226  * Only preserves atomicity at inm level.
1227  * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1228  *
1229  * If the MLD downcall fails, the group is not joined, and an error
1230  * code is returned.
1231  */
1232 static int
1233 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1234     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1235     const int delay)
1236 {
1237 	struct in6_multi_head    inmh;
1238 	struct in6_mfilter	 timf;
1239 	struct in6_multi	*inm;
1240 	struct ifmultiaddr *ifma;
1241 	int			 error;
1242 #ifdef KTR
1243 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1244 #endif
1245 
1246 	/*
1247 	 * Sanity: Check scope zone ID was set for ifp, if and
1248 	 * only if group is scoped to an interface.
1249 	 */
1250 	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1251 	    ("%s: not a multicast address", __func__));
1252 	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1253 	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1254 		KASSERT(mcaddr->s6_addr16[1] != 0,
1255 		    ("%s: scope zone ID not set", __func__));
1256 	}
1257 
1258 	IN6_MULTI_LOCK_ASSERT();
1259 	IN6_MULTI_LIST_UNLOCK_ASSERT();
1260 
1261 	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1262 	    ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1263 
1264 	error = 0;
1265 	inm = NULL;
1266 
1267 	/*
1268 	 * If no imf was specified (i.e. kernel consumer),
1269 	 * fake one up and assume it is an ASM join.
1270 	 */
1271 	if (imf == NULL) {
1272 		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1273 		imf = &timf;
1274 	}
1275 	error = in6_getmulti(ifp, mcaddr, &inm);
1276 	if (error) {
1277 		CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1278 		return (error);
1279 	}
1280 
1281 	IN6_MULTI_LIST_LOCK();
1282 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1283 	error = in6m_merge(inm, imf);
1284 	if (error) {
1285 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1286 		goto out_in6m_release;
1287 	}
1288 
1289 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1290 	error = mld_change_state(inm, delay);
1291 	if (error) {
1292 		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1293 		goto out_in6m_release;
1294 	}
1295 
1296 out_in6m_release:
1297 	SLIST_INIT(&inmh);
1298 	if (error) {
1299 		struct epoch_tracker et;
1300 
1301 		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1302 		IF_ADDR_WLOCK(ifp);
1303 		NET_EPOCH_ENTER(et);
1304 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1305 			if (ifma->ifma_protospec == inm) {
1306 				ifma->ifma_protospec = NULL;
1307 				break;
1308 			}
1309 		}
1310 		in6m_disconnect_locked(&inmh, inm);
1311 		in6m_rele_locked(&inmh, inm);
1312 		NET_EPOCH_EXIT(et);
1313 		IF_ADDR_WUNLOCK(ifp);
1314 	} else {
1315 		*pinm = inm;
1316 	}
1317 	IN6_MULTI_LIST_UNLOCK();
1318 	in6m_release_list_deferred(&inmh);
1319 	return (error);
1320 }
1321 
1322 /*
1323  * Leave a multicast group; unlocked entry point.
1324  */
1325 int
1326 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1327 {
1328 	int error;
1329 
1330 	IN6_MULTI_LOCK();
1331 	error = in6_leavegroup_locked(inm, imf);
1332 	IN6_MULTI_UNLOCK();
1333 	return (error);
1334 }
1335 
1336 /*
1337  * Leave a multicast group; real entry point.
1338  * All source filters will be expunged.
1339  *
1340  * Only preserves atomicity at inm level.
1341  *
1342  * Holding the write lock for the INP which contains imf
1343  * is highly advisable. We can't assert for it as imf does not
1344  * contain a back-pointer to the owning inp.
1345  *
1346  * Note: This is not the same as in6m_release(*) as this function also
1347  * makes a state change downcall into MLD.
1348  */
1349 int
1350 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1351 {
1352 	struct in6_multi_head	 inmh;
1353 	struct in6_mfilter	 timf;
1354 	struct ifnet *ifp;
1355 	int			 error;
1356 #ifdef KTR
1357 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1358 #endif
1359 
1360 	error = 0;
1361 
1362 	IN6_MULTI_LOCK_ASSERT();
1363 
1364 	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1365 	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1366 	    (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1367 	    imf);
1368 
1369 	/*
1370 	 * If no imf was specified (i.e. kernel consumer),
1371 	 * fake one up and assume it is an ASM join.
1372 	 */
1373 	if (imf == NULL) {
1374 		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1375 		imf = &timf;
1376 	}
1377 
1378 	/*
1379 	 * Begin state merge transaction at MLD layer.
1380 	 *
1381 	 * As this particular invocation should not cause any memory
1382 	 * to be allocated, and there is no opportunity to roll back
1383 	 * the transaction, it MUST NOT fail.
1384 	 */
1385 
1386 	ifp = inm->in6m_ifp;
1387 	IN6_MULTI_LIST_LOCK();
1388 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1389 	error = in6m_merge(inm, imf);
1390 	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1391 
1392 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1393 	error = 0;
1394 	if (ifp)
1395 		error = mld_change_state(inm, 0);
1396 	if (error)
1397 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1398 
1399 	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1400 	if (ifp)
1401 		IF_ADDR_WLOCK(ifp);
1402 
1403 	SLIST_INIT(&inmh);
1404 	if (inm->in6m_refcount == 1)
1405 		in6m_disconnect_locked(&inmh, inm);
1406 	in6m_rele_locked(&inmh, inm);
1407 	if (ifp)
1408 		IF_ADDR_WUNLOCK(ifp);
1409 	IN6_MULTI_LIST_UNLOCK();
1410 	in6m_release_list_deferred(&inmh);
1411 	return (error);
1412 }
1413 
1414 /*
1415  * Block or unblock an ASM multicast source on an inpcb.
1416  * This implements the delta-based API described in RFC 3678.
1417  *
1418  * The delta-based API applies only to exclusive-mode memberships.
1419  * An MLD downcall will be performed.
1420  *
1421  * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1422  *
1423  * Return 0 if successful, otherwise return an appropriate error code.
1424  */
1425 static int
1426 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1427 {
1428 	struct group_source_req		 gsr;
1429 	struct epoch_tracker		 et;
1430 	sockunion_t			*gsa, *ssa;
1431 	struct ifnet			*ifp;
1432 	struct in6_mfilter		*imf;
1433 	struct ip6_moptions		*imo;
1434 	struct in6_msource		*ims;
1435 	struct in6_multi			*inm;
1436 	uint16_t			 fmode;
1437 	int				 error, doblock;
1438 #ifdef KTR
1439 	char				 ip6tbuf[INET6_ADDRSTRLEN];
1440 #endif
1441 
1442 	ifp = NULL;
1443 	error = 0;
1444 	doblock = 0;
1445 
1446 	memset(&gsr, 0, sizeof(struct group_source_req));
1447 	gsa = (sockunion_t *)&gsr.gsr_group;
1448 	ssa = (sockunion_t *)&gsr.gsr_source;
1449 
1450 	switch (sopt->sopt_name) {
1451 	case MCAST_BLOCK_SOURCE:
1452 	case MCAST_UNBLOCK_SOURCE:
1453 		error = sooptcopyin(sopt, &gsr,
1454 		    sizeof(struct group_source_req),
1455 		    sizeof(struct group_source_req));
1456 		if (error)
1457 			return (error);
1458 
1459 		if (gsa->sin6.sin6_family != AF_INET6 ||
1460 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1461 			return (EINVAL);
1462 
1463 		if (ssa->sin6.sin6_family != AF_INET6 ||
1464 		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1465 			return (EINVAL);
1466 
1467 		/*
1468 		 * XXXGL: this function should use ifnet_byindex_ref, or
1469 		 * expand the epoch section all the way to where we put
1470 		 * the reference.
1471 		 */
1472 		NET_EPOCH_ENTER(et);
1473 		ifp = ifnet_byindex(gsr.gsr_interface);
1474 		NET_EPOCH_EXIT(et);
1475 		if (ifp == NULL)
1476 			return (EADDRNOTAVAIL);
1477 
1478 		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1479 			doblock = 1;
1480 		break;
1481 
1482 	default:
1483 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1484 		    __func__, sopt->sopt_name);
1485 		return (EOPNOTSUPP);
1486 		break;
1487 	}
1488 
1489 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1490 		return (EINVAL);
1491 
1492 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1493 
1494 	/*
1495 	 * Check if we are actually a member of this group.
1496 	 */
1497 	imo = in6p_findmoptions(inp);
1498 	imf = im6o_match_group(imo, ifp, &gsa->sa);
1499 	if (imf == NULL) {
1500 		error = EADDRNOTAVAIL;
1501 		goto out_in6p_locked;
1502 	}
1503 	inm = imf->im6f_in6m;
1504 
1505 	/*
1506 	 * Attempting to use the delta-based API on an
1507 	 * non exclusive-mode membership is an error.
1508 	 */
1509 	fmode = imf->im6f_st[0];
1510 	if (fmode != MCAST_EXCLUDE) {
1511 		error = EINVAL;
1512 		goto out_in6p_locked;
1513 	}
1514 
1515 	/*
1516 	 * Deal with error cases up-front:
1517 	 *  Asked to block, but already blocked; or
1518 	 *  Asked to unblock, but nothing to unblock.
1519 	 * If adding a new block entry, allocate it.
1520 	 */
1521 	ims = im6o_match_source(imf, &ssa->sa);
1522 	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1523 		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1524 		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1525 		    doblock ? "" : "not ");
1526 		error = EADDRNOTAVAIL;
1527 		goto out_in6p_locked;
1528 	}
1529 
1530 	INP_WLOCK_ASSERT(inp);
1531 
1532 	/*
1533 	 * Begin state merge transaction at socket layer.
1534 	 */
1535 	if (doblock) {
1536 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1537 		ims = im6f_graft(imf, fmode, &ssa->sin6);
1538 		if (ims == NULL)
1539 			error = ENOMEM;
1540 	} else {
1541 		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1542 		error = im6f_prune(imf, &ssa->sin6);
1543 	}
1544 
1545 	if (error) {
1546 		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1547 		goto out_im6f_rollback;
1548 	}
1549 
1550 	/*
1551 	 * Begin state merge transaction at MLD layer.
1552 	 */
1553 	IN6_MULTI_LIST_LOCK();
1554 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1555 	error = in6m_merge(inm, imf);
1556 	if (error)
1557 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1558 	else {
1559 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1560 		error = mld_change_state(inm, 0);
1561 		if (error)
1562 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1563 	}
1564 
1565 	IN6_MULTI_LIST_UNLOCK();
1566 
1567 out_im6f_rollback:
1568 	if (error)
1569 		im6f_rollback(imf);
1570 	else
1571 		im6f_commit(imf);
1572 
1573 	im6f_reap(imf);
1574 
1575 out_in6p_locked:
1576 	INP_WUNLOCK(inp);
1577 	return (error);
1578 }
1579 
1580 /*
1581  * Given an inpcb, return its multicast options structure pointer.  Accepts
1582  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1583  *
1584  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1585  * SMPng: NOTE: Returns with the INP write lock held.
1586  */
1587 static struct ip6_moptions *
1588 in6p_findmoptions(struct inpcb *inp)
1589 {
1590 	struct ip6_moptions	 *imo;
1591 
1592 	INP_WLOCK(inp);
1593 	if (inp->in6p_moptions != NULL)
1594 		return (inp->in6p_moptions);
1595 
1596 	INP_WUNLOCK(inp);
1597 
1598 	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1599 
1600 	imo->im6o_multicast_ifp = NULL;
1601 	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1602 	imo->im6o_multicast_loop = in6_mcast_loop;
1603 	STAILQ_INIT(&imo->im6o_head);
1604 
1605 	INP_WLOCK(inp);
1606 	if (inp->in6p_moptions != NULL) {
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  * XXX can all be safely deferred to epoch_call
1620  *
1621  */
1622 
1623 static void
1624 inp_gcmoptions(struct ip6_moptions *imo)
1625 {
1626 	struct in6_mfilter *imf;
1627 	struct in6_multi *inm;
1628 	struct ifnet *ifp;
1629 
1630 	while ((imf = ip6_mfilter_first(&imo->im6o_head)) != NULL) {
1631                 ip6_mfilter_remove(&imo->im6o_head, imf);
1632 
1633                 im6f_leave(imf);
1634                 if ((inm = imf->im6f_in6m) != NULL) {
1635                         if ((ifp = inm->in6m_ifp) != NULL) {
1636                                 CURVNET_SET(ifp->if_vnet);
1637                                 (void)in6_leavegroup(inm, imf);
1638                                 CURVNET_RESTORE();
1639                         } else {
1640                                 (void)in6_leavegroup(inm, imf);
1641                         }
1642                 }
1643                 ip6_mfilter_free(imf);
1644         }
1645         free(imo, M_IP6MOPTS);
1646 }
1647 
1648 void
1649 ip6_freemoptions(struct ip6_moptions *imo)
1650 {
1651 	if (imo == NULL)
1652 		return;
1653 	inp_gcmoptions(imo);
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 epoch_tracker	 et;
1664 	struct __msfilterreq	 msfr;
1665 	sockunion_t		*gsa;
1666 	struct ifnet		*ifp;
1667 	struct ip6_moptions	*imo;
1668 	struct in6_mfilter	*imf;
1669 	struct ip6_msource	*ims;
1670 	struct in6_msource	*lims;
1671 	struct sockaddr_in6	*psin;
1672 	struct sockaddr_storage	*ptss;
1673 	struct sockaddr_storage	*tss;
1674 	int			 error;
1675 	size_t			 nsrcs, ncsrcs;
1676 
1677 	INP_WLOCK_ASSERT(inp);
1678 
1679 	imo = inp->in6p_moptions;
1680 	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1681 
1682 	INP_WUNLOCK(inp);
1683 
1684 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1685 	    sizeof(struct __msfilterreq));
1686 	if (error)
1687 		return (error);
1688 
1689 	if (msfr.msfr_group.ss_family != AF_INET6 ||
1690 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1691 		return (EINVAL);
1692 
1693 	gsa = (sockunion_t *)&msfr.msfr_group;
1694 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1695 		return (EINVAL);
1696 
1697 	/*
1698 	 * XXXGL: this function should use ifnet_byindex_ref, or expand the
1699 	 * epoch section all the way to where the interface is referenced.
1700 	 */
1701 	NET_EPOCH_ENTER(et);
1702 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1703 	NET_EPOCH_EXIT(et);
1704 	if (ifp == NULL)
1705 		return (EADDRNOTAVAIL);
1706 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1707 
1708 	INP_WLOCK(inp);
1709 
1710 	/*
1711 	 * Lookup group on the socket.
1712 	 */
1713 	imf = im6o_match_group(imo, ifp, &gsa->sa);
1714 	if (imf == NULL) {
1715 		INP_WUNLOCK(inp);
1716 		return (EADDRNOTAVAIL);
1717 	}
1718 
1719 	/*
1720 	 * Ignore memberships which are in limbo.
1721 	 */
1722 	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1723 		INP_WUNLOCK(inp);
1724 		return (EAGAIN);
1725 	}
1726 	msfr.msfr_fmode = imf->im6f_st[1];
1727 
1728 	/*
1729 	 * If the user specified a buffer, copy out the source filter
1730 	 * entries to userland gracefully.
1731 	 * We only copy out the number of entries which userland
1732 	 * has asked for, but we always tell userland how big the
1733 	 * buffer really needs to be.
1734 	 */
1735 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1736 		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1737 	tss = NULL;
1738 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1739 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1740 		    M_TEMP, M_NOWAIT | M_ZERO);
1741 		if (tss == NULL) {
1742 			INP_WUNLOCK(inp);
1743 			return (ENOBUFS);
1744 		}
1745 	}
1746 
1747 	/*
1748 	 * Count number of sources in-mode at t0.
1749 	 * If buffer space exists and remains, copy out source entries.
1750 	 */
1751 	nsrcs = msfr.msfr_nsrcs;
1752 	ncsrcs = 0;
1753 	ptss = tss;
1754 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1755 		lims = (struct in6_msource *)ims;
1756 		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1757 		    lims->im6sl_st[0] != imf->im6f_st[0])
1758 			continue;
1759 		++ncsrcs;
1760 		if (tss != NULL && nsrcs > 0) {
1761 			psin = (struct sockaddr_in6 *)ptss;
1762 			psin->sin6_family = AF_INET6;
1763 			psin->sin6_len = sizeof(struct sockaddr_in6);
1764 			psin->sin6_addr = lims->im6s_addr;
1765 			psin->sin6_port = 0;
1766 			--nsrcs;
1767 			++ptss;
1768 		}
1769 	}
1770 
1771 	INP_WUNLOCK(inp);
1772 
1773 	if (tss != NULL) {
1774 		error = copyout(tss, msfr.msfr_srcs,
1775 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1776 		free(tss, M_TEMP);
1777 		if (error)
1778 			return (error);
1779 	}
1780 
1781 	msfr.msfr_nsrcs = ncsrcs;
1782 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1783 
1784 	return (error);
1785 }
1786 
1787 /*
1788  * Return the IP multicast options in response to user getsockopt().
1789  */
1790 int
1791 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1792 {
1793 	struct ip6_moptions	*im6o;
1794 	int			 error;
1795 	u_int			 optval;
1796 
1797 	INP_WLOCK(inp);
1798 	im6o = inp->in6p_moptions;
1799 	/* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
1800 	if (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  * Use the socket's current FIB number for any required FIB lookup. Look up the
1863  * group address in the unicast FIB, and use its ifp; usually, this points to
1864  * the default next-hop.  If the FIB lookup fails, return NULL.
1865  *
1866  * FUTURE: Support multiple forwarding tables for IPv6.
1867  *
1868  * Returns NULL if no ifp could be found.
1869  */
1870 static struct ifnet *
1871 in6p_lookup_mcast_ifp(const struct inpcb *inp, const struct sockaddr_in6 *gsin6)
1872 {
1873 	struct nhop_object	*nh;
1874 	struct in6_addr		dst;
1875 	uint32_t		scopeid;
1876 	uint32_t		fibnum;
1877 
1878 	KASSERT(gsin6->sin6_family == AF_INET6,
1879 	    ("%s: not AF_INET6 group", __func__));
1880 
1881 	in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1882 	fibnum = inp->inp_inc.inc_fibnum;
1883 	nh = fib6_lookup(fibnum, &dst, scopeid, 0, 0);
1884 
1885 	return (nh ? nh->nh_ifp : NULL);
1886 }
1887 
1888 /*
1889  * Join an IPv6 multicast group, possibly with a source.
1890  *
1891  * FIXME: The KAME use of the unspecified address (::)
1892  * to join *all* multicast groups is currently unsupported.
1893  *
1894  * XXXGL: this function multiple times uses ifnet_byindex() without
1895  * proper protection - staying in epoch, or putting reference on ifnet.
1896  */
1897 static int
1898 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1899 {
1900 	struct in6_multi_head		 inmh;
1901 	struct group_source_req		 gsr;
1902 	struct epoch_tracker		 et;
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 	int				 error, is_new;
1910 
1911 	SLIST_INIT(&inmh);
1912 	ifp = NULL;
1913 	lims = NULL;
1914 	error = 0;
1915 
1916 	memset(&gsr, 0, sizeof(struct group_source_req));
1917 	gsa = (sockunion_t *)&gsr.gsr_group;
1918 	gsa->ss.ss_family = AF_UNSPEC;
1919 	ssa = (sockunion_t *)&gsr.gsr_source;
1920 	ssa->ss.ss_family = AF_UNSPEC;
1921 
1922 	/*
1923 	 * Chew everything into struct group_source_req.
1924 	 * Overwrite the port field if present, as the sockaddr
1925 	 * being copied in may be matched with a binary comparison.
1926 	 * Ignore passed-in scope ID.
1927 	 */
1928 	switch (sopt->sopt_name) {
1929 	case IPV6_JOIN_GROUP: {
1930 		struct ipv6_mreq mreq;
1931 
1932 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1933 		    sizeof(struct ipv6_mreq));
1934 		if (error)
1935 			return (error);
1936 
1937 		gsa->sin6.sin6_family = AF_INET6;
1938 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1939 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1940 
1941 		if (mreq.ipv6mr_interface == 0) {
1942 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1943 		} else {
1944 			NET_EPOCH_ENTER(et);
1945 			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1946 			NET_EPOCH_EXIT(et);
1947 			if (ifp == NULL)
1948 				return (EADDRNOTAVAIL);
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 		NET_EPOCH_ENTER(et);
1988 		ifp = ifnet_byindex(gsr.gsr_interface);
1989 		NET_EPOCH_EXIT(et);
1990 		if (ifp == NULL)
1991 			return (EADDRNOTAVAIL);
1992 		break;
1993 
1994 	default:
1995 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1996 		    __func__, sopt->sopt_name);
1997 		return (EOPNOTSUPP);
1998 		break;
1999 	}
2000 
2001 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2002 		return (EINVAL);
2003 
2004 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2005 		return (EADDRNOTAVAIL);
2006 
2007 	gsa->sin6.sin6_port = 0;
2008 	gsa->sin6.sin6_scope_id = 0;
2009 
2010 	/*
2011 	 * Always set the scope zone ID on memberships created from userland.
2012 	 * Use the passed-in ifp to do this.
2013 	 * XXX The in6_setscope() return value is meaningless.
2014 	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
2015 	 */
2016 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2017 
2018 	IN6_MULTI_LOCK();
2019 
2020 	/*
2021 	 * Find the membership in the membership list.
2022 	 */
2023 	imo = in6p_findmoptions(inp);
2024 	imf = im6o_match_group(imo, ifp, &gsa->sa);
2025 	if (imf == NULL) {
2026 		is_new = 1;
2027 		inm = NULL;
2028 
2029 		if (ip6_mfilter_count(&imo->im6o_head) >= IPV6_MAX_MEMBERSHIPS) {
2030 			error = ENOMEM;
2031 			goto out_in6p_locked;
2032 		}
2033 	} else {
2034 		is_new = 0;
2035 		inm = imf->im6f_in6m;
2036 
2037 		if (ssa->ss.ss_family != AF_UNSPEC) {
2038 			/*
2039 			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2040 			 * is an error. On an existing inclusive membership,
2041 			 * it just adds the source to the filter list.
2042 			 */
2043 			if (imf->im6f_st[1] != MCAST_INCLUDE) {
2044 				error = EINVAL;
2045 				goto out_in6p_locked;
2046 			}
2047 			/*
2048 			 * Throw out duplicates.
2049 			 *
2050 			 * XXX FIXME: This makes a naive assumption that
2051 			 * even if entries exist for *ssa in this imf,
2052 			 * they will be rejected as dupes, even if they
2053 			 * are not valid in the current mode (in-mode).
2054 			 *
2055 			 * in6_msource is transactioned just as for anything
2056 			 * else in SSM -- but note naive use of in6m_graft()
2057 			 * below for allocating new filter entries.
2058 			 *
2059 			 * This is only an issue if someone mixes the
2060 			 * full-state SSM API with the delta-based API,
2061 			 * which is discouraged in the relevant RFCs.
2062 			 */
2063 			lims = im6o_match_source(imf, &ssa->sa);
2064 			if (lims != NULL /*&&
2065 			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2066 				error = EADDRNOTAVAIL;
2067 				goto out_in6p_locked;
2068 			}
2069 		} else {
2070 			/*
2071 			 * MCAST_JOIN_GROUP alone, on any existing membership,
2072 			 * is rejected, to stop the same inpcb tying up
2073 			 * multiple refs to the in_multi.
2074 			 * On an existing inclusive membership, this is also
2075 			 * an error; if you want to change filter mode,
2076 			 * you must use the userland API setsourcefilter().
2077 			 * XXX We don't reject this for imf in UNDEFINED
2078 			 * state at t1, because allocation of a filter
2079 			 * is atomic with allocation of a membership.
2080 			 */
2081 			error = EADDRINUSE;
2082 			goto out_in6p_locked;
2083 		}
2084 	}
2085 
2086 	/*
2087 	 * Begin state merge transaction at socket layer.
2088 	 */
2089 	INP_WLOCK_ASSERT(inp);
2090 
2091 	/*
2092 	 * Graft new source into filter list for this inpcb's
2093 	 * membership of the group. The in6_multi may not have
2094 	 * been allocated yet if this is a new membership, however,
2095 	 * the in_mfilter slot will be allocated and must be initialized.
2096 	 *
2097 	 * Note: Grafting of exclusive mode filters doesn't happen
2098 	 * in this path.
2099 	 * XXX: Should check for non-NULL lims (node exists but may
2100 	 * not be in-mode) for interop with full-state API.
2101 	 */
2102 	if (ssa->ss.ss_family != AF_UNSPEC) {
2103 		/* Membership starts in IN mode */
2104 		if (is_new) {
2105 			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2106 			imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2107 			if (imf == NULL) {
2108 				error = ENOMEM;
2109 				goto out_in6p_locked;
2110 			}
2111 		} else {
2112 			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2113 		}
2114 		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2115 		if (lims == NULL) {
2116 			CTR1(KTR_MLD, "%s: merge imf state failed",
2117 			    __func__);
2118 			error = ENOMEM;
2119 			goto out_in6p_locked;
2120 		}
2121 	} else {
2122 		/* No address specified; Membership starts in EX mode */
2123 		if (is_new) {
2124 			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2125 			imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2126 			if (imf == NULL) {
2127 				error = ENOMEM;
2128 				goto out_in6p_locked;
2129 			}
2130 		}
2131 	}
2132 
2133 	/*
2134 	 * Begin state merge transaction at MLD layer.
2135 	 */
2136 	if (is_new) {
2137 		in_pcbref(inp);
2138 		INP_WUNLOCK(inp);
2139 
2140 		error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2141 		    &imf->im6f_in6m, 0);
2142 
2143 		INP_WLOCK(inp);
2144 		if (in_pcbrele_wlocked(inp)) {
2145 			error = ENXIO;
2146 			goto out_in6p_unlocked;
2147 		}
2148 		if (error) {
2149 			goto out_in6p_locked;
2150 		}
2151 		/*
2152 		 * NOTE: Refcount from in6_joingroup_locked()
2153 		 * is protecting membership.
2154 		 */
2155 		ip6_mfilter_insert(&imo->im6o_head, imf);
2156 	} else {
2157 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2158 		IN6_MULTI_LIST_LOCK();
2159 		error = in6m_merge(inm, imf);
2160 		if (error) {
2161 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2162 			    __func__);
2163 			IN6_MULTI_LIST_UNLOCK();
2164 			im6f_rollback(imf);
2165 			im6f_reap(imf);
2166 			goto out_in6p_locked;
2167 		}
2168 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2169 		error = mld_change_state(inm, 0);
2170 		IN6_MULTI_LIST_UNLOCK();
2171 
2172 		if (error) {
2173 			CTR1(KTR_MLD, "%s: failed mld downcall",
2174 			     __func__);
2175 			im6f_rollback(imf);
2176 			im6f_reap(imf);
2177 			goto out_in6p_locked;
2178 		}
2179 	}
2180 
2181 	im6f_commit(imf);
2182 	imf = NULL;
2183 
2184 out_in6p_locked:
2185 	INP_WUNLOCK(inp);
2186 out_in6p_unlocked:
2187 	IN6_MULTI_UNLOCK();
2188 
2189 	if (is_new && imf) {
2190 		if (imf->im6f_in6m != NULL) {
2191 			struct in6_multi_head inmh;
2192 
2193 			SLIST_INIT(&inmh);
2194 			SLIST_INSERT_HEAD(&inmh, imf->im6f_in6m, in6m_defer);
2195 			in6m_release_list_deferred(&inmh);
2196 		}
2197 		ip6_mfilter_free(imf);
2198 	}
2199 	return (error);
2200 }
2201 
2202 /*
2203  * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2204  */
2205 static int
2206 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2207 {
2208 	struct ipv6_mreq		 mreq;
2209 	struct group_source_req		 gsr;
2210 	struct epoch_tracker		 et;
2211 	sockunion_t			*gsa, *ssa;
2212 	struct ifnet			*ifp;
2213 	struct in6_mfilter		*imf;
2214 	struct ip6_moptions		*imo;
2215 	struct in6_msource		*ims;
2216 	struct in6_multi		*inm;
2217 	uint32_t			 ifindex;
2218 	int				 error;
2219 	bool				 is_final;
2220 #ifdef KTR
2221 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2222 #endif
2223 
2224 	ifp = NULL;
2225 	ifindex = 0;
2226 	error = 0;
2227 	is_final = true;
2228 
2229 	memset(&gsr, 0, sizeof(struct group_source_req));
2230 	gsa = (sockunion_t *)&gsr.gsr_group;
2231 	gsa->ss.ss_family = AF_UNSPEC;
2232 	ssa = (sockunion_t *)&gsr.gsr_source;
2233 	ssa->ss.ss_family = AF_UNSPEC;
2234 
2235 	/*
2236 	 * Chew everything passed in up into a struct group_source_req
2237 	 * as that is easier to process.
2238 	 * Note: Any embedded scope ID in the multicast group passed
2239 	 * in by userland is ignored, the interface index is the recommended
2240 	 * mechanism to specify an interface; see below.
2241 	 */
2242 	switch (sopt->sopt_name) {
2243 	case IPV6_LEAVE_GROUP:
2244 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2245 		    sizeof(struct ipv6_mreq));
2246 		if (error)
2247 			return (error);
2248 		gsa->sin6.sin6_family = AF_INET6;
2249 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2250 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2251 		gsa->sin6.sin6_port = 0;
2252 		gsa->sin6.sin6_scope_id = 0;
2253 		ifindex = mreq.ipv6mr_interface;
2254 		break;
2255 
2256 	case MCAST_LEAVE_GROUP:
2257 	case MCAST_LEAVE_SOURCE_GROUP:
2258 		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2259 			error = sooptcopyin(sopt, &gsr,
2260 			    sizeof(struct group_req),
2261 			    sizeof(struct group_req));
2262 		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2263 			error = sooptcopyin(sopt, &gsr,
2264 			    sizeof(struct group_source_req),
2265 			    sizeof(struct group_source_req));
2266 		}
2267 		if (error)
2268 			return (error);
2269 
2270 		if (gsa->sin6.sin6_family != AF_INET6 ||
2271 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2272 			return (EINVAL);
2273 		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2274 			if (ssa->sin6.sin6_family != AF_INET6 ||
2275 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2276 				return (EINVAL);
2277 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2278 				return (EINVAL);
2279 			/*
2280 			 * TODO: Validate embedded scope ID in source
2281 			 * list entry against passed-in ifp, if and only
2282 			 * if source list filter entry is iface or node local.
2283 			 */
2284 			in6_clearscope(&ssa->sin6.sin6_addr);
2285 		}
2286 		gsa->sin6.sin6_port = 0;
2287 		gsa->sin6.sin6_scope_id = 0;
2288 		ifindex = gsr.gsr_interface;
2289 		break;
2290 
2291 	default:
2292 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2293 		    __func__, sopt->sopt_name);
2294 		return (EOPNOTSUPP);
2295 		break;
2296 	}
2297 
2298 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2299 		return (EINVAL);
2300 
2301 	/*
2302 	 * Validate interface index if provided. If no interface index
2303 	 * was provided separately, attempt to look the membership up
2304 	 * from the default scope as a last resort to disambiguate
2305 	 * the membership we are being asked to leave.
2306 	 * XXX SCOPE6 lock potentially taken here.
2307 	 */
2308 	if (ifindex != 0) {
2309 		NET_EPOCH_ENTER(et);
2310 		ifp = ifnet_byindex(ifindex);
2311 		NET_EPOCH_EXIT(et);	/* XXXGL: unsafe ifp */
2312 		if (ifp == NULL)
2313 			return (EADDRNOTAVAIL);
2314 		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2315 	} else {
2316 		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2317 		if (error)
2318 			return (EADDRNOTAVAIL);
2319 		/*
2320 		 * Some badly behaved applications don't pass an ifindex
2321 		 * or a scope ID, which is an API violation. In this case,
2322 		 * perform a lookup as per a v6 join.
2323 		 *
2324 		 * XXX For now, stomp on zone ID for the corner case.
2325 		 * This is not the 'KAME way', but we need to see the ifp
2326 		 * directly until such time as this implementation is
2327 		 * refactored, assuming the scope IDs are the way to go.
2328 		 */
2329 		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2330 		if (ifindex == 0) {
2331 			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2332 			    "ifp for group %s.", __func__,
2333 			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2334 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2335 		} else {
2336 			NET_EPOCH_ENTER(et);
2337 			ifp = ifnet_byindex(ifindex);
2338 			NET_EPOCH_EXIT(et);	/* XXXGL: unsafe ifp */
2339 		}
2340 		if (ifp == NULL)
2341 			return (EADDRNOTAVAIL);
2342 	}
2343 
2344 	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2345 	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2346 
2347 	IN6_MULTI_LOCK();
2348 
2349 	/*
2350 	 * Find the membership in the membership list.
2351 	 */
2352 	imo = in6p_findmoptions(inp);
2353 	imf = im6o_match_group(imo, ifp, &gsa->sa);
2354 	if (imf == NULL) {
2355 		error = EADDRNOTAVAIL;
2356 		goto out_in6p_locked;
2357 	}
2358 	inm = imf->im6f_in6m;
2359 
2360 	if (ssa->ss.ss_family != AF_UNSPEC)
2361 		is_final = false;
2362 
2363 	/*
2364 	 * Begin state merge transaction at socket layer.
2365 	 */
2366 	INP_WLOCK_ASSERT(inp);
2367 
2368 	/*
2369 	 * If we were instructed only to leave a given source, do so.
2370 	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2371 	 */
2372 	if (is_final) {
2373 		ip6_mfilter_remove(&imo->im6o_head, imf);
2374 		im6f_leave(imf);
2375 
2376 		/*
2377 		 * Give up the multicast address record to which
2378 		 * the membership points.
2379 		 */
2380 		(void)in6_leavegroup_locked(inm, imf);
2381 	} else {
2382 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2383 			error = EADDRNOTAVAIL;
2384 			goto out_in6p_locked;
2385 		}
2386 		ims = im6o_match_source(imf, &ssa->sa);
2387 		if (ims == NULL) {
2388 			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2389 			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2390 			    "not ");
2391 			error = EADDRNOTAVAIL;
2392 			goto out_in6p_locked;
2393 		}
2394 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2395 		error = im6f_prune(imf, &ssa->sin6);
2396 		if (error) {
2397 			CTR1(KTR_MLD, "%s: merge imf state failed",
2398 			    __func__);
2399 			goto out_in6p_locked;
2400 		}
2401 	}
2402 
2403 	/*
2404 	 * Begin state merge transaction at MLD layer.
2405 	 */
2406 	if (!is_final) {
2407 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2408 		IN6_MULTI_LIST_LOCK();
2409 		error = in6m_merge(inm, imf);
2410 		if (error) {
2411 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2412 			    __func__);
2413 			IN6_MULTI_LIST_UNLOCK();
2414 			im6f_rollback(imf);
2415 			im6f_reap(imf);
2416                         goto out_in6p_locked;
2417 		}
2418 
2419 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2420 		error = mld_change_state(inm, 0);
2421 		IN6_MULTI_LIST_UNLOCK();
2422 		if (error) {
2423 			CTR1(KTR_MLD, "%s: failed mld downcall",
2424 			     __func__);
2425 			im6f_rollback(imf);
2426 			im6f_reap(imf);
2427                         goto out_in6p_locked;
2428 		}
2429 	}
2430 
2431 	im6f_commit(imf);
2432 	im6f_reap(imf);
2433 
2434 out_in6p_locked:
2435 	INP_WUNLOCK(inp);
2436 
2437 	if (is_final && imf)
2438 		ip6_mfilter_free(imf);
2439 
2440 	IN6_MULTI_UNLOCK();
2441 	return (error);
2442 }
2443 
2444 /*
2445  * Select the interface for transmitting IPv6 multicast datagrams.
2446  *
2447  * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2448  * may be passed to this socket option. An address of in6addr_any or an
2449  * interface index of 0 is used to remove a previous selection.
2450  * When no interface is selected, one is chosen for every send.
2451  */
2452 static int
2453 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2454 {
2455 	struct epoch_tracker	 et;
2456 	struct ifnet		*ifp;
2457 	struct ip6_moptions	*imo;
2458 	u_int			 ifindex;
2459 	int			 error;
2460 
2461 	if (sopt->sopt_valsize != sizeof(u_int))
2462 		return (EINVAL);
2463 
2464 	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2465 	if (error)
2466 		return (error);
2467 	NET_EPOCH_ENTER(et);
2468 	if (ifindex == 0)
2469 		ifp = NULL;
2470 	else {
2471 		ifp = ifnet_byindex(ifindex);
2472 		if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2473 			NET_EPOCH_EXIT(et);
2474 			return (EADDRNOTAVAIL);
2475 		}
2476 	}
2477 	NET_EPOCH_EXIT(et);	/* XXXGL: unsafe ifp */
2478 	imo = in6p_findmoptions(inp);
2479 	imo->im6o_multicast_ifp = ifp;
2480 	INP_WUNLOCK(inp);
2481 
2482 	return (0);
2483 }
2484 
2485 /*
2486  * Atomically set source filters on a socket for an IPv6 multicast group.
2487  *
2488  * XXXGL: unsafely exits epoch with ifnet pointer
2489  */
2490 static int
2491 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2492 {
2493 	struct __msfilterreq	 msfr;
2494 	struct epoch_tracker	 et;
2495 	sockunion_t		*gsa;
2496 	struct ifnet		*ifp;
2497 	struct in6_mfilter	*imf;
2498 	struct ip6_moptions	*imo;
2499 	struct in6_multi		*inm;
2500 	int			 error;
2501 
2502 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2503 	    sizeof(struct __msfilterreq));
2504 	if (error)
2505 		return (error);
2506 
2507 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2508 		return (ENOBUFS);
2509 
2510 	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2511 	    msfr.msfr_fmode != MCAST_INCLUDE)
2512 		return (EINVAL);
2513 
2514 	if (msfr.msfr_group.ss_family != AF_INET6 ||
2515 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2516 		return (EINVAL);
2517 
2518 	gsa = (sockunion_t *)&msfr.msfr_group;
2519 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2520 		return (EINVAL);
2521 
2522 	gsa->sin6.sin6_port = 0;	/* ignore port */
2523 
2524 	NET_EPOCH_ENTER(et);
2525 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2526 	NET_EPOCH_EXIT(et);
2527 	if (ifp == NULL)
2528 		return (EADDRNOTAVAIL);
2529 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2530 
2531 	/*
2532 	 * Take the INP write lock.
2533 	 * Check if this socket is a member of this group.
2534 	 */
2535 	imo = in6p_findmoptions(inp);
2536 	imf = im6o_match_group(imo, ifp, &gsa->sa);
2537 	if (imf == NULL) {
2538 		error = EADDRNOTAVAIL;
2539 		goto out_in6p_locked;
2540 	}
2541 	inm = imf->im6f_in6m;
2542 
2543 	/*
2544 	 * Begin state merge transaction at socket layer.
2545 	 */
2546 	INP_WLOCK_ASSERT(inp);
2547 
2548 	imf->im6f_st[1] = msfr.msfr_fmode;
2549 
2550 	/*
2551 	 * Apply any new source filters, if present.
2552 	 * Make a copy of the user-space source vector so
2553 	 * that we may copy them with a single copyin. This
2554 	 * allows us to deal with page faults up-front.
2555 	 */
2556 	if (msfr.msfr_nsrcs > 0) {
2557 		struct in6_msource	*lims;
2558 		struct sockaddr_in6	*psin;
2559 		struct sockaddr_storage	*kss, *pkss;
2560 		int			 i;
2561 
2562 		INP_WUNLOCK(inp);
2563 
2564 		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2565 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2566 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2567 		    M_TEMP, M_WAITOK);
2568 		error = copyin(msfr.msfr_srcs, kss,
2569 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2570 		if (error) {
2571 			free(kss, M_TEMP);
2572 			return (error);
2573 		}
2574 
2575 		INP_WLOCK(inp);
2576 
2577 		/*
2578 		 * Mark all source filters as UNDEFINED at t1.
2579 		 * Restore new group filter mode, as im6f_leave()
2580 		 * will set it to INCLUDE.
2581 		 */
2582 		im6f_leave(imf);
2583 		imf->im6f_st[1] = msfr.msfr_fmode;
2584 
2585 		/*
2586 		 * Update socket layer filters at t1, lazy-allocating
2587 		 * new entries. This saves a bunch of memory at the
2588 		 * cost of one RB_FIND() per source entry; duplicate
2589 		 * entries in the msfr_nsrcs vector are ignored.
2590 		 * If we encounter an error, rollback transaction.
2591 		 *
2592 		 * XXX This too could be replaced with a set-symmetric
2593 		 * difference like loop to avoid walking from root
2594 		 * every time, as the key space is common.
2595 		 */
2596 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2597 			psin = (struct sockaddr_in6 *)pkss;
2598 			if (psin->sin6_family != AF_INET6) {
2599 				error = EAFNOSUPPORT;
2600 				break;
2601 			}
2602 			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2603 				error = EINVAL;
2604 				break;
2605 			}
2606 			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2607 				error = EINVAL;
2608 				break;
2609 			}
2610 			/*
2611 			 * TODO: Validate embedded scope ID in source
2612 			 * list entry against passed-in ifp, if and only
2613 			 * if source list filter entry is iface or node local.
2614 			 */
2615 			in6_clearscope(&psin->sin6_addr);
2616 			error = im6f_get_source(imf, psin, &lims);
2617 			if (error)
2618 				break;
2619 			lims->im6sl_st[1] = imf->im6f_st[1];
2620 		}
2621 		free(kss, M_TEMP);
2622 	}
2623 
2624 	if (error)
2625 		goto out_im6f_rollback;
2626 
2627 	INP_WLOCK_ASSERT(inp);
2628 	IN6_MULTI_LIST_LOCK();
2629 
2630 	/*
2631 	 * Begin state merge transaction at MLD layer.
2632 	 */
2633 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2634 	error = in6m_merge(inm, imf);
2635 	if (error)
2636 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2637 	else {
2638 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2639 		error = mld_change_state(inm, 0);
2640 		if (error)
2641 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2642 	}
2643 
2644 	IN6_MULTI_LIST_UNLOCK();
2645 
2646 out_im6f_rollback:
2647 	if (error)
2648 		im6f_rollback(imf);
2649 	else
2650 		im6f_commit(imf);
2651 
2652 	im6f_reap(imf);
2653 
2654 out_in6p_locked:
2655 	INP_WUNLOCK(inp);
2656 	return (error);
2657 }
2658 
2659 /*
2660  * Set the IP multicast options in response to user setsockopt().
2661  *
2662  * Many of the socket options handled in this function duplicate the
2663  * functionality of socket options in the regular unicast API. However,
2664  * it is not possible to merge the duplicate code, because the idempotence
2665  * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2666  * the effects of these options must be treated as separate and distinct.
2667  *
2668  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2669  */
2670 int
2671 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2672 {
2673 	struct ip6_moptions	*im6o;
2674 	int			 error;
2675 
2676 	error = 0;
2677 
2678 	/* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
2679 	if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2680 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)
2681 		return (EOPNOTSUPP);
2682 
2683 	switch (sopt->sopt_name) {
2684 	case IPV6_MULTICAST_IF:
2685 		error = in6p_set_multicast_if(inp, sopt);
2686 		break;
2687 
2688 	case IPV6_MULTICAST_HOPS: {
2689 		int hlim;
2690 
2691 		if (sopt->sopt_valsize != sizeof(int)) {
2692 			error = EINVAL;
2693 			break;
2694 		}
2695 		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2696 		if (error)
2697 			break;
2698 		if (hlim < -1 || hlim > 255) {
2699 			error = EINVAL;
2700 			break;
2701 		} else if (hlim == -1) {
2702 			hlim = V_ip6_defmcasthlim;
2703 		}
2704 		im6o = in6p_findmoptions(inp);
2705 		im6o->im6o_multicast_hlim = hlim;
2706 		INP_WUNLOCK(inp);
2707 		break;
2708 	}
2709 
2710 	case IPV6_MULTICAST_LOOP: {
2711 		u_int loop;
2712 
2713 		/*
2714 		 * Set the loopback flag for outgoing multicast packets.
2715 		 * Must be zero or one.
2716 		 */
2717 		if (sopt->sopt_valsize != sizeof(u_int)) {
2718 			error = EINVAL;
2719 			break;
2720 		}
2721 		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2722 		if (error)
2723 			break;
2724 		if (loop > 1) {
2725 			error = EINVAL;
2726 			break;
2727 		}
2728 		im6o = in6p_findmoptions(inp);
2729 		im6o->im6o_multicast_loop = loop;
2730 		INP_WUNLOCK(inp);
2731 		break;
2732 	}
2733 
2734 	case IPV6_JOIN_GROUP:
2735 	case MCAST_JOIN_GROUP:
2736 	case MCAST_JOIN_SOURCE_GROUP:
2737 		error = in6p_join_group(inp, sopt);
2738 		break;
2739 
2740 	case IPV6_LEAVE_GROUP:
2741 	case MCAST_LEAVE_GROUP:
2742 	case MCAST_LEAVE_SOURCE_GROUP:
2743 		error = in6p_leave_group(inp, sopt);
2744 		break;
2745 
2746 	case MCAST_BLOCK_SOURCE:
2747 	case MCAST_UNBLOCK_SOURCE:
2748 		error = in6p_block_unblock_source(inp, sopt);
2749 		break;
2750 
2751 	case IPV6_MSFILTER:
2752 		error = in6p_set_source_filters(inp, sopt);
2753 		break;
2754 
2755 	default:
2756 		error = EOPNOTSUPP;
2757 		break;
2758 	}
2759 
2760 	INP_UNLOCK_ASSERT(inp);
2761 
2762 	return (error);
2763 }
2764 
2765 /*
2766  * Expose MLD's multicast filter mode and source list(s) to userland,
2767  * keyed by (ifindex, group).
2768  * The filter mode is written out as a uint32_t, followed by
2769  * 0..n of struct in6_addr.
2770  * For use by ifmcstat(8).
2771  * SMPng: NOTE: unlocked read of ifindex space.
2772  */
2773 static int
2774 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2775 {
2776 	struct in6_addr			 mcaddr;
2777 	struct in6_addr			 src;
2778 	struct epoch_tracker		 et;
2779 	struct ifnet			*ifp;
2780 	struct ifmultiaddr		*ifma;
2781 	struct in6_multi		*inm;
2782 	struct ip6_msource		*ims;
2783 	int				*name;
2784 	int				 retval;
2785 	u_int				 namelen;
2786 	uint32_t			 fmode, ifindex;
2787 #ifdef KTR
2788 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2789 #endif
2790 
2791 	name = (int *)arg1;
2792 	namelen = arg2;
2793 
2794 	if (req->newptr != NULL)
2795 		return (EPERM);
2796 
2797 	/* int: ifindex + 4 * 32 bits of IPv6 address */
2798 	if (namelen != 5)
2799 		return (EINVAL);
2800 
2801 	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2802 	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2803 		CTR2(KTR_MLD, "%s: group %s is not multicast",
2804 		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2805 		return (EINVAL);
2806 	}
2807 
2808 	ifindex = name[0];
2809 	NET_EPOCH_ENTER(et);
2810 	ifp = ifnet_byindex(ifindex);
2811 	if (ifp == NULL) {
2812 		NET_EPOCH_EXIT(et);
2813 		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2814 		    __func__, ifindex);
2815 		return (ENOENT);
2816 	}
2817 	/*
2818 	 * Internal MLD lookups require that scope/zone ID is set.
2819 	 */
2820 	(void)in6_setscope(&mcaddr, ifp, NULL);
2821 
2822 	retval = sysctl_wire_old_buffer(req,
2823 	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2824 	if (retval) {
2825 		NET_EPOCH_EXIT(et);
2826 		return (retval);
2827 	}
2828 
2829 	IN6_MULTI_LOCK();
2830 	IN6_MULTI_LIST_LOCK();
2831 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2832 		inm = in6m_ifmultiaddr_get_inm(ifma);
2833 		if (inm == NULL)
2834 			continue;
2835 		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2836 			continue;
2837 		fmode = inm->in6m_st[1].iss_fmode;
2838 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2839 		if (retval != 0)
2840 			break;
2841 		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2842 			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2843 			/*
2844 			 * Only copy-out sources which are in-mode.
2845 			 */
2846 			if (fmode != im6s_get_mode(inm, ims, 1)) {
2847 				CTR1(KTR_MLD, "%s: skip non-in-mode",
2848 				    __func__);
2849 				continue;
2850 			}
2851 			src = ims->im6s_addr;
2852 			retval = SYSCTL_OUT(req, &src,
2853 			    sizeof(struct in6_addr));
2854 			if (retval != 0)
2855 				break;
2856 		}
2857 	}
2858 	IN6_MULTI_LIST_UNLOCK();
2859 	IN6_MULTI_UNLOCK();
2860 	NET_EPOCH_EXIT(et);
2861 
2862 	return (retval);
2863 }
2864 
2865 #ifdef KTR
2866 
2867 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2868 
2869 static const char *
2870 in6m_mode_str(const int mode)
2871 {
2872 
2873 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2874 		return (in6m_modestrs[mode]);
2875 	return ("??");
2876 }
2877 
2878 static const char *in6m_statestrs[] = {
2879 	"not-member",
2880 	"silent",
2881 	"reporting",
2882 	"idle",
2883 	"lazy",
2884 	"sleeping",
2885 	"awakening",
2886 	"query-pending",
2887 	"sg-query-pending",
2888 	"leaving"
2889 };
2890 _Static_assert(nitems(in6m_statestrs) ==
2891     MLD_LEAVING_MEMBER - MLD_NOT_MEMBER + 1, "Missing MLD group state");
2892 
2893 static const char *
2894 in6m_state_str(const int state)
2895 {
2896 
2897 	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2898 		return (in6m_statestrs[state]);
2899 	return ("??");
2900 }
2901 
2902 /*
2903  * Dump an in6_multi structure to the console.
2904  */
2905 void
2906 in6m_print(const struct in6_multi *inm)
2907 {
2908 	int t;
2909 	char ip6tbuf[INET6_ADDRSTRLEN];
2910 
2911 	if ((ktr_mask & KTR_MLD) == 0)
2912 		return;
2913 
2914 	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2915 	printf("addr %s ifp %p(%s) ifma %p\n",
2916 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2917 	    inm->in6m_ifp,
2918 	    if_name(inm->in6m_ifp),
2919 	    inm->in6m_ifma);
2920 	printf("timer %u state %s refcount %u scq.len %u\n",
2921 	    inm->in6m_timer,
2922 	    in6m_state_str(inm->in6m_state),
2923 	    inm->in6m_refcount,
2924 	    mbufq_len(&inm->in6m_scq));
2925 	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2926 	    inm->in6m_mli,
2927 	    inm->in6m_nsrc,
2928 	    inm->in6m_sctimer,
2929 	    inm->in6m_scrv);
2930 	for (t = 0; t < 2; t++) {
2931 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2932 		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2933 		    inm->in6m_st[t].iss_asm,
2934 		    inm->in6m_st[t].iss_ex,
2935 		    inm->in6m_st[t].iss_in,
2936 		    inm->in6m_st[t].iss_rec);
2937 	}
2938 	printf("%s: --- end in6m %p ---\n", __func__, inm);
2939 }
2940 
2941 #else /* !KTR */
2942 
2943 void
2944 in6m_print(const struct in6_multi *inm)
2945 {
2946 
2947 }
2948 
2949 #endif /* KTR */
2950