xref: /freebsd/sys/netinet6/in6_mcast.c (revision 68d75eff68281c1b445e3010bb975eae07aac225)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2009 Bruce Simpson.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. The name of the author may not be used to endorse or promote
16  *    products derived from this software without specific prior written
17  *    permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*
33  * IPv6 multicast socket, group, and socket option processing module.
34  * Normative references: RFC 2292, RFC 3492, RFC 3542, RFC 3678, RFC 3810.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_inet6.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/gtaskqueue.h>
45 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/priv.h>
53 #include <sys/ktr.h>
54 #include <sys/tree.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/vnet.h>
61 
62 
63 #include <netinet/in.h>
64 #include <netinet/udp.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/udp_var.h>
68 #include <netinet6/in6_fib.h>
69 #include <netinet6/in6_var.h>
70 #include <netinet/ip6.h>
71 #include <netinet/icmp6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet/in_pcb.h>
74 #include <netinet/tcp_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet6/mld6_var.h>
77 #include <netinet6/scope6_var.h>
78 
79 #ifndef KTR_MLD
80 #define KTR_MLD KTR_INET6
81 #endif
82 
83 #ifndef __SOCKUNION_DECLARED
84 union sockunion {
85 	struct sockaddr_storage	ss;
86 	struct sockaddr		sa;
87 	struct sockaddr_dl	sdl;
88 	struct sockaddr_in6	sin6;
89 };
90 typedef union sockunion sockunion_t;
91 #define __SOCKUNION_DECLARED
92 #endif /* __SOCKUNION_DECLARED */
93 
94 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
95     "IPv6 multicast PCB-layer source filter");
96 MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
97 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
98 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
99     "IPv6 multicast MLD-layer source filter");
100 
101 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
102 
103 /*
104  * Locking:
105  * - Lock order is: Giant, 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, CTLFLAG_RW, 0,
172     "IPv6 multicast");
173 
174 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
175 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
176     CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
177     "Max source filters per group");
178 
179 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
180 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
181     CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
182     "Max source filters per socket");
183 
184 /* TODO Virtualize this switch. */
185 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
186 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
187     &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
188 
189 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
190     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
191     "Per-interface stack-wide source filters");
192 
193 #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  * Find and return a reference to an in6_multi record for (ifp, group),
349  * and bump its reference count.
350  * If one does not exist, try to allocate it, and update link-layer multicast
351  * filters on ifp to listen for group.
352  * Assumes the IN6_MULTI lock is held across the call.
353  * Return 0 if successful, otherwise return an appropriate error code.
354  */
355 static int
356 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
357     struct in6_multi **pinm)
358 {
359 	struct epoch_tracker	 et;
360 	struct sockaddr_in6	 gsin6;
361 	struct ifmultiaddr	*ifma;
362 	struct in6_multi	*inm;
363 	int			 error;
364 
365 	error = 0;
366 
367 	/*
368 	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
369 	 * if_addmulti() takes this mutex itself, so we must drop and
370 	 * re-acquire around the call.
371 	 */
372 	IN6_MULTI_LOCK_ASSERT();
373 	IN6_MULTI_LIST_LOCK();
374 	IF_ADDR_WLOCK(ifp);
375 	NET_EPOCH_ENTER(et);
376 	inm = in6m_lookup_locked(ifp, group);
377 	NET_EPOCH_EXIT(et);
378 
379 	if (inm != NULL) {
380 		/*
381 		 * If we already joined this group, just bump the
382 		 * refcount and return it.
383 		 */
384 		KASSERT(inm->in6m_refcount >= 1,
385 		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
386 		in6m_acquire_locked(inm);
387 		*pinm = inm;
388 		goto out_locked;
389 	}
390 
391 	memset(&gsin6, 0, sizeof(gsin6));
392 	gsin6.sin6_family = AF_INET6;
393 	gsin6.sin6_len = sizeof(struct sockaddr_in6);
394 	gsin6.sin6_addr = *group;
395 
396 	/*
397 	 * Check if a link-layer group is already associated
398 	 * with this network-layer group on the given ifnet.
399 	 */
400 	IN6_MULTI_LIST_UNLOCK();
401 	IF_ADDR_WUNLOCK(ifp);
402 	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
403 	if (error != 0)
404 		return (error);
405 	IN6_MULTI_LIST_LOCK();
406 	IF_ADDR_WLOCK(ifp);
407 
408 	/*
409 	 * If something other than netinet6 is occupying the link-layer
410 	 * group, print a meaningful error message and back out of
411 	 * the allocation.
412 	 * Otherwise, bump the refcount on the existing network-layer
413 	 * group association and return it.
414 	 */
415 	if (ifma->ifma_protospec != NULL) {
416 		inm = (struct in6_multi *)ifma->ifma_protospec;
417 #ifdef INVARIANTS
418 		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
419 		    __func__));
420 		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
421 		    ("%s: ifma not AF_INET6", __func__));
422 		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
423 		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
424 		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
425 			panic("%s: ifma %p is inconsistent with %p (%p)",
426 			    __func__, ifma, inm, group);
427 #endif
428 		in6m_acquire_locked(inm);
429 		*pinm = inm;
430 		goto out_locked;
431 	}
432 
433 	IF_ADDR_WLOCK_ASSERT(ifp);
434 
435 	/*
436 	 * A new in6_multi record is needed; allocate and initialize it.
437 	 * We DO NOT perform an MLD join as the in6_ layer may need to
438 	 * push an initial source list down to MLD to support SSM.
439 	 *
440 	 * The initial source filter state is INCLUDE, {} as per the RFC.
441 	 * Pending state-changes per group are subject to a bounds check.
442 	 */
443 	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
444 	if (inm == NULL) {
445 		IN6_MULTI_LIST_UNLOCK();
446 		IF_ADDR_WUNLOCK(ifp);
447 		if_delmulti_ifma(ifma);
448 		return (ENOMEM);
449 	}
450 	inm->in6m_addr = *group;
451 	inm->in6m_ifp = ifp;
452 	inm->in6m_mli = MLD_IFINFO(ifp);
453 	inm->in6m_ifma = ifma;
454 	inm->in6m_refcount = 1;
455 	inm->in6m_state = MLD_NOT_MEMBER;
456 	mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
457 
458 	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
459 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
460 	RB_INIT(&inm->in6m_srcs);
461 
462 	ifma->ifma_protospec = inm;
463 	*pinm = inm;
464 
465  out_locked:
466 	IN6_MULTI_LIST_UNLOCK();
467 	IF_ADDR_WUNLOCK(ifp);
468 	return (error);
469 }
470 
471 /*
472  * Drop a reference to an in6_multi record.
473  *
474  * If the refcount drops to 0, free the in6_multi record and
475  * delete the underlying link-layer membership.
476  */
477 static void
478 in6m_release(struct in6_multi *inm)
479 {
480 	struct ifmultiaddr *ifma;
481 	struct ifnet *ifp;
482 
483 	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
484 
485 	MPASS(inm->in6m_refcount == 0);
486 	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
487 
488 	ifma = inm->in6m_ifma;
489 	ifp = inm->in6m_ifp;
490 	MPASS(ifma->ifma_llifma == NULL);
491 
492 	/* XXX this access is not covered by IF_ADDR_LOCK */
493 	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
494 	KASSERT(ifma->ifma_protospec == NULL,
495 	    ("%s: ifma_protospec != NULL", __func__));
496 	if (ifp == NULL)
497 		ifp = ifma->ifma_ifp;
498 
499 	if (ifp != NULL) {
500 		CURVNET_SET(ifp->if_vnet);
501 		in6m_purge(inm);
502 		free(inm, M_IP6MADDR);
503 		if_delmulti_ifma_flags(ifma, 1);
504 		CURVNET_RESTORE();
505 		if_rele(ifp);
506 	} else {
507 		in6m_purge(inm);
508 		free(inm, M_IP6MADDR);
509 		if_delmulti_ifma_flags(ifma, 1);
510 	}
511 }
512 
513 static struct grouptask free_gtask;
514 static struct in6_multi_head in6m_free_list;
515 static void in6m_release_task(void *arg __unused);
516 static void in6m_init(void)
517 {
518 	SLIST_INIT(&in6m_free_list);
519 	taskqgroup_config_gtask_init(NULL, &free_gtask, in6m_release_task, "in6m release task");
520 }
521 
522 #ifdef EARLY_AP_STARTUP
523 SYSINIT(in6m_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
524 	in6m_init, NULL);
525 #else
526 SYSINIT(in6m_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_SECOND,
527 	in6m_init, NULL);
528 #endif
529 
530 
531 void
532 in6m_release_list_deferred(struct in6_multi_head *inmh)
533 {
534 	if (SLIST_EMPTY(inmh))
535 		return;
536 	mtx_lock(&in6_multi_free_mtx);
537 	SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
538 	mtx_unlock(&in6_multi_free_mtx);
539 	GROUPTASK_ENQUEUE(&free_gtask);
540 }
541 
542 void
543 in6m_release_wait(void)
544 {
545 
546 	/* Wait for all jobs to complete. */
547 	gtaskqueue_drain_all(free_gtask.gt_taskqueue);
548 }
549 
550 void
551 in6m_disconnect_locked(struct in6_multi_head *inmh, struct in6_multi *inm)
552 {
553 	struct ifnet *ifp;
554 	struct ifaddr *ifa;
555 	struct in6_ifaddr *ifa6;
556 	struct in6_multi_mship *imm, *imm_tmp;
557 	struct ifmultiaddr *ifma, *ll_ifma;
558 
559 	IN6_MULTI_LIST_LOCK_ASSERT();
560 
561 	ifp = inm->in6m_ifp;
562 	if (ifp == NULL)
563 		return;		/* already called */
564 
565 	inm->in6m_ifp = NULL;
566 	IF_ADDR_WLOCK_ASSERT(ifp);
567 	ifma = inm->in6m_ifma;
568 	if (ifma == NULL)
569 		return;
570 
571 	if_ref(ifp);
572 	if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
573 		CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
574 		ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
575 	}
576 	MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
577 	if ((ll_ifma = ifma->ifma_llifma) != NULL) {
578 		MPASS(ifma != ll_ifma);
579 		ifma->ifma_llifma = NULL;
580 		MPASS(ll_ifma->ifma_llifma == NULL);
581 		MPASS(ll_ifma->ifma_ifp == ifp);
582 		if (--ll_ifma->ifma_refcount == 0) {
583 			if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
584 				CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
585 				ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
586 			}
587 			MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
588 			if_freemulti(ll_ifma);
589 		}
590 	}
591 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
592 		if (ifa->ifa_addr->sa_family != AF_INET6)
593 			continue;
594 		ifa6 = (void *)ifa;
595 		LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
596 		    i6mm_chain, imm_tmp) {
597 			if (inm == imm->i6mm_maddr) {
598 				LIST_REMOVE(imm, i6mm_chain);
599 				free(imm, M_IP6MADDR);
600 				in6m_rele_locked(inmh, inm);
601 			}
602 		}
603 	}
604 }
605 
606 static void
607 in6m_release_task(void *arg __unused)
608 {
609 	struct in6_multi_head in6m_free_tmp;
610 	struct in6_multi *inm, *tinm;
611 
612 	SLIST_INIT(&in6m_free_tmp);
613 	mtx_lock(&in6_multi_free_mtx);
614 	SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
615 	mtx_unlock(&in6_multi_free_mtx);
616 	IN6_MULTI_LOCK();
617 	SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
618 		SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
619 		in6m_release(inm);
620 	}
621 	IN6_MULTI_UNLOCK();
622 }
623 
624 /*
625  * Clear recorded source entries for a group.
626  * Used by the MLD code. Caller must hold the IN6_MULTI lock.
627  * FIXME: Should reap.
628  */
629 void
630 in6m_clear_recorded(struct in6_multi *inm)
631 {
632 	struct ip6_msource	*ims;
633 
634 	IN6_MULTI_LIST_LOCK_ASSERT();
635 
636 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
637 		if (ims->im6s_stp) {
638 			ims->im6s_stp = 0;
639 			--inm->in6m_st[1].iss_rec;
640 		}
641 	}
642 	KASSERT(inm->in6m_st[1].iss_rec == 0,
643 	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
644 }
645 
646 /*
647  * Record a source as pending for a Source-Group MLDv2 query.
648  * This lives here as it modifies the shared tree.
649  *
650  * inm is the group descriptor.
651  * naddr is the address of the source to record in network-byte order.
652  *
653  * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
654  * lazy-allocate a source node in response to an SG query.
655  * Otherwise, no allocation is performed. This saves some memory
656  * with the trade-off that the source will not be reported to the
657  * router if joined in the window between the query response and
658  * the group actually being joined on the local host.
659  *
660  * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
661  * This turns off the allocation of a recorded source entry if
662  * the group has not been joined.
663  *
664  * Return 0 if the source didn't exist or was already marked as recorded.
665  * Return 1 if the source was marked as recorded by this function.
666  * Return <0 if any error occurred (negated errno code).
667  */
668 int
669 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
670 {
671 	struct ip6_msource	 find;
672 	struct ip6_msource	*ims, *nims;
673 
674 	IN6_MULTI_LIST_LOCK_ASSERT();
675 
676 	find.im6s_addr = *addr;
677 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
678 	if (ims && ims->im6s_stp)
679 		return (0);
680 	if (ims == NULL) {
681 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
682 			return (-ENOSPC);
683 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
684 		    M_NOWAIT | M_ZERO);
685 		if (nims == NULL)
686 			return (-ENOMEM);
687 		nims->im6s_addr = find.im6s_addr;
688 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
689 		++inm->in6m_nsrc;
690 		ims = nims;
691 	}
692 
693 	/*
694 	 * Mark the source as recorded and update the recorded
695 	 * source count.
696 	 */
697 	++ims->im6s_stp;
698 	++inm->in6m_st[1].iss_rec;
699 
700 	return (1);
701 }
702 
703 /*
704  * Return a pointer to an in6_msource owned by an in6_mfilter,
705  * given its source address.
706  * Lazy-allocate if needed. If this is a new entry its filter state is
707  * undefined at t0.
708  *
709  * imf is the filter set being modified.
710  * addr is the source address.
711  *
712  * SMPng: May be called with locks held; malloc must not block.
713  */
714 static int
715 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
716     struct in6_msource **plims)
717 {
718 	struct ip6_msource	 find;
719 	struct ip6_msource	*ims, *nims;
720 	struct in6_msource	*lims;
721 	int			 error;
722 
723 	error = 0;
724 	ims = NULL;
725 	lims = NULL;
726 
727 	find.im6s_addr = psin->sin6_addr;
728 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
729 	lims = (struct in6_msource *)ims;
730 	if (lims == NULL) {
731 		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
732 			return (ENOSPC);
733 		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
734 		    M_NOWAIT | M_ZERO);
735 		if (nims == NULL)
736 			return (ENOMEM);
737 		lims = (struct in6_msource *)nims;
738 		lims->im6s_addr = find.im6s_addr;
739 		lims->im6sl_st[0] = MCAST_UNDEFINED;
740 		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
741 		++imf->im6f_nsrc;
742 	}
743 
744 	*plims = lims;
745 
746 	return (error);
747 }
748 
749 /*
750  * Graft a source entry into an existing socket-layer filter set,
751  * maintaining any required invariants and checking allocations.
752  *
753  * The source is marked as being in the new filter mode at t1.
754  *
755  * Return the pointer to the new node, otherwise return NULL.
756  */
757 static struct in6_msource *
758 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
759     const struct sockaddr_in6 *psin)
760 {
761 	struct ip6_msource	*nims;
762 	struct in6_msource	*lims;
763 
764 	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
765 	    M_NOWAIT | M_ZERO);
766 	if (nims == NULL)
767 		return (NULL);
768 	lims = (struct in6_msource *)nims;
769 	lims->im6s_addr = psin->sin6_addr;
770 	lims->im6sl_st[0] = MCAST_UNDEFINED;
771 	lims->im6sl_st[1] = st1;
772 	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
773 	++imf->im6f_nsrc;
774 
775 	return (lims);
776 }
777 
778 /*
779  * Prune a source entry from an existing socket-layer filter set,
780  * maintaining any required invariants and checking allocations.
781  *
782  * The source is marked as being left at t1, it is not freed.
783  *
784  * Return 0 if no error occurred, otherwise return an errno value.
785  */
786 static int
787 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
788 {
789 	struct ip6_msource	 find;
790 	struct ip6_msource	*ims;
791 	struct in6_msource	*lims;
792 
793 	find.im6s_addr = psin->sin6_addr;
794 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
795 	if (ims == NULL)
796 		return (ENOENT);
797 	lims = (struct in6_msource *)ims;
798 	lims->im6sl_st[1] = MCAST_UNDEFINED;
799 	return (0);
800 }
801 
802 /*
803  * Revert socket-layer filter set deltas at t1 to t0 state.
804  */
805 static void
806 im6f_rollback(struct in6_mfilter *imf)
807 {
808 	struct ip6_msource	*ims, *tims;
809 	struct in6_msource	*lims;
810 
811 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
812 		lims = (struct in6_msource *)ims;
813 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
814 			/* no change at t1 */
815 			continue;
816 		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
817 			/* revert change to existing source at t1 */
818 			lims->im6sl_st[1] = lims->im6sl_st[0];
819 		} else {
820 			/* revert source added t1 */
821 			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
822 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
823 			free(ims, M_IN6MFILTER);
824 			imf->im6f_nsrc--;
825 		}
826 	}
827 	imf->im6f_st[1] = imf->im6f_st[0];
828 }
829 
830 /*
831  * Mark socket-layer filter set as INCLUDE {} at t1.
832  */
833 static void
834 im6f_leave(struct in6_mfilter *imf)
835 {
836 	struct ip6_msource	*ims;
837 	struct in6_msource	*lims;
838 
839 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
840 		lims = (struct in6_msource *)ims;
841 		lims->im6sl_st[1] = MCAST_UNDEFINED;
842 	}
843 	imf->im6f_st[1] = MCAST_INCLUDE;
844 }
845 
846 /*
847  * Mark socket-layer filter set deltas as committed.
848  */
849 static void
850 im6f_commit(struct in6_mfilter *imf)
851 {
852 	struct ip6_msource	*ims;
853 	struct in6_msource	*lims;
854 
855 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
856 		lims = (struct in6_msource *)ims;
857 		lims->im6sl_st[0] = lims->im6sl_st[1];
858 	}
859 	imf->im6f_st[0] = imf->im6f_st[1];
860 }
861 
862 /*
863  * Reap unreferenced sources from socket-layer filter set.
864  */
865 static void
866 im6f_reap(struct in6_mfilter *imf)
867 {
868 	struct ip6_msource	*ims, *tims;
869 	struct in6_msource	*lims;
870 
871 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
872 		lims = (struct in6_msource *)ims;
873 		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
874 		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
875 			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
876 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
877 			free(ims, M_IN6MFILTER);
878 			imf->im6f_nsrc--;
879 		}
880 	}
881 }
882 
883 /*
884  * Purge socket-layer filter set.
885  */
886 static void
887 im6f_purge(struct in6_mfilter *imf)
888 {
889 	struct ip6_msource	*ims, *tims;
890 
891 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
892 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
893 		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
894 		free(ims, M_IN6MFILTER);
895 		imf->im6f_nsrc--;
896 	}
897 	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
898 	KASSERT(RB_EMPTY(&imf->im6f_sources),
899 	    ("%s: im6f_sources not empty", __func__));
900 }
901 
902 /*
903  * Look up a source filter entry for a multicast group.
904  *
905  * inm is the group descriptor to work with.
906  * addr is the IPv6 address to look up.
907  * noalloc may be non-zero to suppress allocation of sources.
908  * *pims will be set to the address of the retrieved or allocated source.
909  *
910  * SMPng: NOTE: may be called with locks held.
911  * Return 0 if successful, otherwise return a non-zero error code.
912  */
913 static int
914 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
915     const int noalloc, struct ip6_msource **pims)
916 {
917 	struct ip6_msource	 find;
918 	struct ip6_msource	*ims, *nims;
919 #ifdef KTR
920 	char			 ip6tbuf[INET6_ADDRSTRLEN];
921 #endif
922 
923 	find.im6s_addr = *addr;
924 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
925 	if (ims == NULL && !noalloc) {
926 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
927 			return (ENOSPC);
928 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
929 		    M_NOWAIT | M_ZERO);
930 		if (nims == NULL)
931 			return (ENOMEM);
932 		nims->im6s_addr = *addr;
933 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
934 		++inm->in6m_nsrc;
935 		ims = nims;
936 		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
937 		    ip6_sprintf(ip6tbuf, addr), ims);
938 	}
939 
940 	*pims = ims;
941 	return (0);
942 }
943 
944 /*
945  * Merge socket-layer source into MLD-layer source.
946  * If rollback is non-zero, perform the inverse of the merge.
947  */
948 static void
949 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
950     const int rollback)
951 {
952 	int n = rollback ? -1 : 1;
953 #ifdef KTR
954 	char ip6tbuf[INET6_ADDRSTRLEN];
955 
956 	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
957 #endif
958 
959 	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
960 		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
961 		ims->im6s_st[1].ex -= n;
962 	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
963 		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
964 		ims->im6s_st[1].in -= n;
965 	}
966 
967 	if (lims->im6sl_st[1] == 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[1] == 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 
976 /*
977  * Atomically update the global in6_multi state, when a membership's
978  * filter list is being updated in any way.
979  *
980  * imf is the per-inpcb-membership group filter pointer.
981  * A fake imf may be passed for in-kernel consumers.
982  *
983  * XXX This is a candidate for a set-symmetric-difference style loop
984  * which would eliminate the repeated lookup from root of ims nodes,
985  * as they share the same key space.
986  *
987  * If any error occurred this function will back out of refcounts
988  * and return a non-zero value.
989  */
990 static int
991 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
992 {
993 	struct ip6_msource	*ims, *nims;
994 	struct in6_msource	*lims;
995 	int			 schanged, error;
996 	int			 nsrc0, nsrc1;
997 
998 	schanged = 0;
999 	error = 0;
1000 	nsrc1 = nsrc0 = 0;
1001 	IN6_MULTI_LIST_LOCK_ASSERT();
1002 
1003 	/*
1004 	 * Update the source filters first, as this may fail.
1005 	 * Maintain count of in-mode filters at t0, t1. These are
1006 	 * used to work out if we transition into ASM mode or not.
1007 	 * Maintain a count of source filters whose state was
1008 	 * actually modified by this operation.
1009 	 */
1010 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1011 		lims = (struct in6_msource *)ims;
1012 		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1013 		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1014 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1015 		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1016 		++schanged;
1017 		if (error)
1018 			break;
1019 		im6s_merge(nims, lims, 0);
1020 	}
1021 	if (error) {
1022 		struct ip6_msource *bims;
1023 
1024 		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1025 			lims = (struct in6_msource *)ims;
1026 			if (lims->im6sl_st[0] == lims->im6sl_st[1])
1027 				continue;
1028 			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1029 			if (bims == NULL)
1030 				continue;
1031 			im6s_merge(bims, lims, 1);
1032 		}
1033 		goto out_reap;
1034 	}
1035 
1036 	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1037 	    __func__, nsrc0, nsrc1);
1038 
1039 	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1040 	if (imf->im6f_st[0] == imf->im6f_st[1] &&
1041 	    imf->im6f_st[1] == MCAST_INCLUDE) {
1042 		if (nsrc1 == 0) {
1043 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1044 			--inm->in6m_st[1].iss_in;
1045 		}
1046 	}
1047 
1048 	/* Handle filter mode transition on socket. */
1049 	if (imf->im6f_st[0] != imf->im6f_st[1]) {
1050 		CTR3(KTR_MLD, "%s: imf transition %d to %d",
1051 		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
1052 
1053 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1054 			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1055 			--inm->in6m_st[1].iss_ex;
1056 		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1057 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1058 			--inm->in6m_st[1].iss_in;
1059 		}
1060 
1061 		if (imf->im6f_st[1] == 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[1] == MCAST_INCLUDE && nsrc1 > 0) {
1065 			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1066 			inm->in6m_st[1].iss_in++;
1067 		}
1068 	}
1069 
1070 	/*
1071 	 * Track inm filter state in terms of listener counts.
1072 	 * If there are any exclusive listeners, stack-wide
1073 	 * membership is exclusive.
1074 	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1075 	 * If no listeners remain, state is undefined at t1,
1076 	 * and the MLD lifecycle for this group should finish.
1077 	 */
1078 	if (inm->in6m_st[1].iss_ex > 0) {
1079 		CTR1(KTR_MLD, "%s: transition to EX", __func__);
1080 		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1081 	} else if (inm->in6m_st[1].iss_in > 0) {
1082 		CTR1(KTR_MLD, "%s: transition to IN", __func__);
1083 		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1084 	} else {
1085 		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1086 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1087 	}
1088 
1089 	/* Decrement ASM listener count on transition out of ASM mode. */
1090 	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1091 		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1092 		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1093 			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1094 			--inm->in6m_st[1].iss_asm;
1095 		}
1096 	}
1097 
1098 	/* Increment ASM listener count on transition to ASM mode. */
1099 	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1100 		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1101 		inm->in6m_st[1].iss_asm++;
1102 	}
1103 
1104 	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1105 	in6m_print(inm);
1106 
1107 out_reap:
1108 	if (schanged > 0) {
1109 		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1110 		in6m_reap(inm);
1111 	}
1112 	return (error);
1113 }
1114 
1115 /*
1116  * Mark an in6_multi's filter set deltas as committed.
1117  * Called by MLD after a state change has been enqueued.
1118  */
1119 void
1120 in6m_commit(struct in6_multi *inm)
1121 {
1122 	struct ip6_msource	*ims;
1123 
1124 	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1125 	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1126 	in6m_print(inm);
1127 
1128 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1129 		ims->im6s_st[0] = ims->im6s_st[1];
1130 	}
1131 	inm->in6m_st[0] = inm->in6m_st[1];
1132 }
1133 
1134 /*
1135  * Reap unreferenced nodes from an in6_multi's filter set.
1136  */
1137 static void
1138 in6m_reap(struct in6_multi *inm)
1139 {
1140 	struct ip6_msource	*ims, *tims;
1141 
1142 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1143 		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1144 		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1145 		    ims->im6s_stp != 0)
1146 			continue;
1147 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1148 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1149 		free(ims, M_IP6MSOURCE);
1150 		inm->in6m_nsrc--;
1151 	}
1152 }
1153 
1154 /*
1155  * Purge all source nodes from an in6_multi's filter set.
1156  */
1157 static void
1158 in6m_purge(struct in6_multi *inm)
1159 {
1160 	struct ip6_msource	*ims, *tims;
1161 
1162 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1163 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1164 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1165 		free(ims, M_IP6MSOURCE);
1166 		inm->in6m_nsrc--;
1167 	}
1168 	/* Free state-change requests that might be queued. */
1169 	mbufq_drain(&inm->in6m_scq);
1170 }
1171 
1172 /*
1173  * Join a multicast address w/o sources.
1174  * KAME compatibility entry point.
1175  *
1176  * SMPng: Assume no mc locks held by caller.
1177  */
1178 int
1179 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1180     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1181     const int delay)
1182 {
1183 	int error;
1184 
1185 	IN6_MULTI_LOCK();
1186 	error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1187 	IN6_MULTI_UNLOCK();
1188 	return (error);
1189 }
1190 
1191 /*
1192  * Join a multicast group; real entry point.
1193  *
1194  * Only preserves atomicity at inm level.
1195  * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1196  *
1197  * If the MLD downcall fails, the group is not joined, and an error
1198  * code is returned.
1199  */
1200 static int
1201 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1202     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1203     const int delay)
1204 {
1205 	struct in6_multi_head    inmh;
1206 	struct in6_mfilter	 timf;
1207 	struct in6_multi	*inm;
1208 	struct ifmultiaddr *ifma;
1209 	int			 error;
1210 #ifdef KTR
1211 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1212 #endif
1213 
1214 	/*
1215 	 * Sanity: Check scope zone ID was set for ifp, if and
1216 	 * only if group is scoped to an interface.
1217 	 */
1218 	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1219 	    ("%s: not a multicast address", __func__));
1220 	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1221 	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1222 		KASSERT(mcaddr->s6_addr16[1] != 0,
1223 		    ("%s: scope zone ID not set", __func__));
1224 	}
1225 
1226 	IN6_MULTI_LOCK_ASSERT();
1227 	IN6_MULTI_LIST_UNLOCK_ASSERT();
1228 
1229 	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1230 	    ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1231 
1232 	error = 0;
1233 	inm = NULL;
1234 
1235 	/*
1236 	 * If no imf was specified (i.e. kernel consumer),
1237 	 * fake one up and assume it is an ASM join.
1238 	 */
1239 	if (imf == NULL) {
1240 		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1241 		imf = &timf;
1242 	}
1243 	error = in6_getmulti(ifp, mcaddr, &inm);
1244 	if (error) {
1245 		CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1246 		return (error);
1247 	}
1248 
1249 	IN6_MULTI_LIST_LOCK();
1250 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1251 	error = in6m_merge(inm, imf);
1252 	if (error) {
1253 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1254 		goto out_in6m_release;
1255 	}
1256 
1257 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1258 	error = mld_change_state(inm, delay);
1259 	if (error) {
1260 		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1261 		goto out_in6m_release;
1262 	}
1263 
1264 out_in6m_release:
1265 	SLIST_INIT(&inmh);
1266 	if (error) {
1267 		struct epoch_tracker et;
1268 
1269 		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1270 		IF_ADDR_WLOCK(ifp);
1271 		NET_EPOCH_ENTER(et);
1272 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1273 			if (ifma->ifma_protospec == inm) {
1274 				ifma->ifma_protospec = NULL;
1275 				break;
1276 			}
1277 		}
1278 		in6m_disconnect_locked(&inmh, inm);
1279 		in6m_rele_locked(&inmh, inm);
1280 		NET_EPOCH_EXIT(et);
1281 		IF_ADDR_WUNLOCK(ifp);
1282 	} else {
1283 		*pinm = inm;
1284 	}
1285 	IN6_MULTI_LIST_UNLOCK();
1286 	in6m_release_list_deferred(&inmh);
1287 	return (error);
1288 }
1289 
1290 /*
1291  * Leave a multicast group; unlocked entry point.
1292  */
1293 int
1294 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1295 {
1296 	int error;
1297 
1298 	IN6_MULTI_LOCK();
1299 	error = in6_leavegroup_locked(inm, imf);
1300 	IN6_MULTI_UNLOCK();
1301 	return (error);
1302 }
1303 
1304 /*
1305  * Leave a multicast group; real entry point.
1306  * All source filters will be expunged.
1307  *
1308  * Only preserves atomicity at inm level.
1309  *
1310  * Holding the write lock for the INP which contains imf
1311  * is highly advisable. We can't assert for it as imf does not
1312  * contain a back-pointer to the owning inp.
1313  *
1314  * Note: This is not the same as in6m_release(*) as this function also
1315  * makes a state change downcall into MLD.
1316  */
1317 int
1318 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1319 {
1320 	struct in6_multi_head	 inmh;
1321 	struct in6_mfilter	 timf;
1322 	struct ifnet *ifp;
1323 	int			 error;
1324 #ifdef KTR
1325 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1326 #endif
1327 
1328 	error = 0;
1329 
1330 	IN6_MULTI_LOCK_ASSERT();
1331 
1332 	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1333 	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1334 	    (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1335 	    imf);
1336 
1337 	/*
1338 	 * If no imf was specified (i.e. kernel consumer),
1339 	 * fake one up and assume it is an ASM join.
1340 	 */
1341 	if (imf == NULL) {
1342 		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1343 		imf = &timf;
1344 	}
1345 
1346 	/*
1347 	 * Begin state merge transaction at MLD layer.
1348 	 *
1349 	 * As this particular invocation should not cause any memory
1350 	 * to be allocated, and there is no opportunity to roll back
1351 	 * the transaction, it MUST NOT fail.
1352 	 */
1353 
1354 	ifp = inm->in6m_ifp;
1355 	IN6_MULTI_LIST_LOCK();
1356 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1357 	error = in6m_merge(inm, imf);
1358 	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1359 
1360 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1361 	error = 0;
1362 	if (ifp)
1363 		error = mld_change_state(inm, 0);
1364 	if (error)
1365 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1366 
1367 	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1368 	if (ifp)
1369 		IF_ADDR_WLOCK(ifp);
1370 
1371 	SLIST_INIT(&inmh);
1372 	if (inm->in6m_refcount == 1)
1373 		in6m_disconnect_locked(&inmh, inm);
1374 	in6m_rele_locked(&inmh, inm);
1375 	if (ifp)
1376 		IF_ADDR_WUNLOCK(ifp);
1377 	IN6_MULTI_LIST_UNLOCK();
1378 	in6m_release_list_deferred(&inmh);
1379 	return (error);
1380 }
1381 
1382 
1383 /*
1384  * Block or unblock an ASM multicast source on an inpcb.
1385  * This implements the delta-based API described in RFC 3678.
1386  *
1387  * The delta-based API applies only to exclusive-mode memberships.
1388  * An MLD downcall will be performed.
1389  *
1390  * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1391  *
1392  * Return 0 if successful, otherwise return an appropriate error code.
1393  */
1394 static int
1395 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1396 {
1397 	struct group_source_req		 gsr;
1398 	sockunion_t			*gsa, *ssa;
1399 	struct ifnet			*ifp;
1400 	struct in6_mfilter		*imf;
1401 	struct ip6_moptions		*imo;
1402 	struct in6_msource		*ims;
1403 	struct in6_multi			*inm;
1404 	uint16_t			 fmode;
1405 	int				 error, doblock;
1406 #ifdef KTR
1407 	char				 ip6tbuf[INET6_ADDRSTRLEN];
1408 #endif
1409 
1410 	ifp = NULL;
1411 	error = 0;
1412 	doblock = 0;
1413 
1414 	memset(&gsr, 0, sizeof(struct group_source_req));
1415 	gsa = (sockunion_t *)&gsr.gsr_group;
1416 	ssa = (sockunion_t *)&gsr.gsr_source;
1417 
1418 	switch (sopt->sopt_name) {
1419 	case MCAST_BLOCK_SOURCE:
1420 	case MCAST_UNBLOCK_SOURCE:
1421 		error = sooptcopyin(sopt, &gsr,
1422 		    sizeof(struct group_source_req),
1423 		    sizeof(struct group_source_req));
1424 		if (error)
1425 			return (error);
1426 
1427 		if (gsa->sin6.sin6_family != AF_INET6 ||
1428 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1429 			return (EINVAL);
1430 
1431 		if (ssa->sin6.sin6_family != AF_INET6 ||
1432 		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1433 			return (EINVAL);
1434 
1435 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1436 			return (EADDRNOTAVAIL);
1437 
1438 		ifp = ifnet_byindex(gsr.gsr_interface);
1439 
1440 		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1441 			doblock = 1;
1442 		break;
1443 
1444 	default:
1445 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1446 		    __func__, sopt->sopt_name);
1447 		return (EOPNOTSUPP);
1448 		break;
1449 	}
1450 
1451 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1452 		return (EINVAL);
1453 
1454 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1455 
1456 	/*
1457 	 * Check if we are actually a member of this group.
1458 	 */
1459 	imo = in6p_findmoptions(inp);
1460 	imf = im6o_match_group(imo, ifp, &gsa->sa);
1461 	if (imf == NULL) {
1462 		error = EADDRNOTAVAIL;
1463 		goto out_in6p_locked;
1464 	}
1465 	inm = imf->im6f_in6m;
1466 
1467 	/*
1468 	 * Attempting to use the delta-based API on an
1469 	 * non exclusive-mode membership is an error.
1470 	 */
1471 	fmode = imf->im6f_st[0];
1472 	if (fmode != MCAST_EXCLUDE) {
1473 		error = EINVAL;
1474 		goto out_in6p_locked;
1475 	}
1476 
1477 	/*
1478 	 * Deal with error cases up-front:
1479 	 *  Asked to block, but already blocked; or
1480 	 *  Asked to unblock, but nothing to unblock.
1481 	 * If adding a new block entry, allocate it.
1482 	 */
1483 	ims = im6o_match_source(imf, &ssa->sa);
1484 	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1485 		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1486 		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1487 		    doblock ? "" : "not ");
1488 		error = EADDRNOTAVAIL;
1489 		goto out_in6p_locked;
1490 	}
1491 
1492 	INP_WLOCK_ASSERT(inp);
1493 
1494 	/*
1495 	 * Begin state merge transaction at socket layer.
1496 	 */
1497 	if (doblock) {
1498 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1499 		ims = im6f_graft(imf, fmode, &ssa->sin6);
1500 		if (ims == NULL)
1501 			error = ENOMEM;
1502 	} else {
1503 		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1504 		error = im6f_prune(imf, &ssa->sin6);
1505 	}
1506 
1507 	if (error) {
1508 		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1509 		goto out_im6f_rollback;
1510 	}
1511 
1512 	/*
1513 	 * Begin state merge transaction at MLD layer.
1514 	 */
1515 	IN6_MULTI_LIST_LOCK();
1516 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1517 	error = in6m_merge(inm, imf);
1518 	if (error)
1519 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1520 	else {
1521 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1522 		error = mld_change_state(inm, 0);
1523 		if (error)
1524 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1525 	}
1526 
1527 	IN6_MULTI_LIST_UNLOCK();
1528 
1529 out_im6f_rollback:
1530 	if (error)
1531 		im6f_rollback(imf);
1532 	else
1533 		im6f_commit(imf);
1534 
1535 	im6f_reap(imf);
1536 
1537 out_in6p_locked:
1538 	INP_WUNLOCK(inp);
1539 	return (error);
1540 }
1541 
1542 /*
1543  * Given an inpcb, return its multicast options structure pointer.  Accepts
1544  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1545  *
1546  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1547  * SMPng: NOTE: Returns with the INP write lock held.
1548  */
1549 static struct ip6_moptions *
1550 in6p_findmoptions(struct inpcb *inp)
1551 {
1552 	struct ip6_moptions	 *imo;
1553 
1554 	INP_WLOCK(inp);
1555 	if (inp->in6p_moptions != NULL)
1556 		return (inp->in6p_moptions);
1557 
1558 	INP_WUNLOCK(inp);
1559 
1560 	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1561 
1562 	imo->im6o_multicast_ifp = NULL;
1563 	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1564 	imo->im6o_multicast_loop = in6_mcast_loop;
1565 	STAILQ_INIT(&imo->im6o_head);
1566 
1567 	INP_WLOCK(inp);
1568 	if (inp->in6p_moptions != NULL) {
1569 		free(imo, M_IP6MOPTS);
1570 		return (inp->in6p_moptions);
1571 	}
1572 	inp->in6p_moptions = imo;
1573 	return (imo);
1574 }
1575 
1576 /*
1577  * Discard the IPv6 multicast options (and source filters).
1578  *
1579  * SMPng: NOTE: assumes INP write lock is held.
1580  *
1581  * XXX can all be safely deferred to epoch_call
1582  *
1583  */
1584 
1585 static void
1586 inp_gcmoptions(struct ip6_moptions *imo)
1587 {
1588 	struct in6_mfilter *imf;
1589 	struct in6_multi *inm;
1590 	struct ifnet *ifp;
1591 
1592 	while ((imf = ip6_mfilter_first(&imo->im6o_head)) != NULL) {
1593                 ip6_mfilter_remove(&imo->im6o_head, imf);
1594 
1595                 im6f_leave(imf);
1596                 if ((inm = imf->im6f_in6m) != NULL) {
1597                         if ((ifp = inm->in6m_ifp) != NULL) {
1598                                 CURVNET_SET(ifp->if_vnet);
1599                                 (void)in6_leavegroup(inm, imf);
1600                                 CURVNET_RESTORE();
1601                         } else {
1602                                 (void)in6_leavegroup(inm, imf);
1603                         }
1604                 }
1605                 ip6_mfilter_free(imf);
1606         }
1607         free(imo, M_IP6MOPTS);
1608 }
1609 
1610 void
1611 ip6_freemoptions(struct ip6_moptions *imo)
1612 {
1613 	if (imo == NULL)
1614 		return;
1615 	inp_gcmoptions(imo);
1616 }
1617 
1618 /*
1619  * Atomically get source filters on a socket for an IPv6 multicast group.
1620  * Called with INP lock held; returns with lock released.
1621  */
1622 static int
1623 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1624 {
1625 	struct __msfilterreq	 msfr;
1626 	sockunion_t		*gsa;
1627 	struct ifnet		*ifp;
1628 	struct ip6_moptions	*imo;
1629 	struct in6_mfilter	*imf;
1630 	struct ip6_msource	*ims;
1631 	struct in6_msource	*lims;
1632 	struct sockaddr_in6	*psin;
1633 	struct sockaddr_storage	*ptss;
1634 	struct sockaddr_storage	*tss;
1635 	int			 error;
1636 	size_t			 nsrcs, ncsrcs;
1637 
1638 	INP_WLOCK_ASSERT(inp);
1639 
1640 	imo = inp->in6p_moptions;
1641 	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1642 
1643 	INP_WUNLOCK(inp);
1644 
1645 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1646 	    sizeof(struct __msfilterreq));
1647 	if (error)
1648 		return (error);
1649 
1650 	if (msfr.msfr_group.ss_family != AF_INET6 ||
1651 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1652 		return (EINVAL);
1653 
1654 	gsa = (sockunion_t *)&msfr.msfr_group;
1655 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1656 		return (EINVAL);
1657 
1658 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1659 		return (EADDRNOTAVAIL);
1660 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1661 	if (ifp == NULL)
1662 		return (EADDRNOTAVAIL);
1663 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1664 
1665 	INP_WLOCK(inp);
1666 
1667 	/*
1668 	 * Lookup group on the socket.
1669 	 */
1670 	imf = im6o_match_group(imo, ifp, &gsa->sa);
1671 	if (imf == NULL) {
1672 		INP_WUNLOCK(inp);
1673 		return (EADDRNOTAVAIL);
1674 	}
1675 
1676 	/*
1677 	 * Ignore memberships which are in limbo.
1678 	 */
1679 	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1680 		INP_WUNLOCK(inp);
1681 		return (EAGAIN);
1682 	}
1683 	msfr.msfr_fmode = imf->im6f_st[1];
1684 
1685 	/*
1686 	 * If the user specified a buffer, copy out the source filter
1687 	 * entries to userland gracefully.
1688 	 * We only copy out the number of entries which userland
1689 	 * has asked for, but we always tell userland how big the
1690 	 * buffer really needs to be.
1691 	 */
1692 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1693 		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1694 	tss = NULL;
1695 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1696 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1697 		    M_TEMP, M_NOWAIT | M_ZERO);
1698 		if (tss == NULL) {
1699 			INP_WUNLOCK(inp);
1700 			return (ENOBUFS);
1701 		}
1702 	}
1703 
1704 	/*
1705 	 * Count number of sources in-mode at t0.
1706 	 * If buffer space exists and remains, copy out source entries.
1707 	 */
1708 	nsrcs = msfr.msfr_nsrcs;
1709 	ncsrcs = 0;
1710 	ptss = tss;
1711 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1712 		lims = (struct in6_msource *)ims;
1713 		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1714 		    lims->im6sl_st[0] != imf->im6f_st[0])
1715 			continue;
1716 		++ncsrcs;
1717 		if (tss != NULL && nsrcs > 0) {
1718 			psin = (struct sockaddr_in6 *)ptss;
1719 			psin->sin6_family = AF_INET6;
1720 			psin->sin6_len = sizeof(struct sockaddr_in6);
1721 			psin->sin6_addr = lims->im6s_addr;
1722 			psin->sin6_port = 0;
1723 			--nsrcs;
1724 			++ptss;
1725 		}
1726 	}
1727 
1728 	INP_WUNLOCK(inp);
1729 
1730 	if (tss != NULL) {
1731 		error = copyout(tss, msfr.msfr_srcs,
1732 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1733 		free(tss, M_TEMP);
1734 		if (error)
1735 			return (error);
1736 	}
1737 
1738 	msfr.msfr_nsrcs = ncsrcs;
1739 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1740 
1741 	return (error);
1742 }
1743 
1744 /*
1745  * Return the IP multicast options in response to user getsockopt().
1746  */
1747 int
1748 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1749 {
1750 	struct ip6_moptions	*im6o;
1751 	int			 error;
1752 	u_int			 optval;
1753 
1754 	INP_WLOCK(inp);
1755 	im6o = inp->in6p_moptions;
1756 	/*
1757 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1758 	 * or is a divert socket, reject it.
1759 	 */
1760 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1761 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1762 	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1763 		INP_WUNLOCK(inp);
1764 		return (EOPNOTSUPP);
1765 	}
1766 
1767 	error = 0;
1768 	switch (sopt->sopt_name) {
1769 	case IPV6_MULTICAST_IF:
1770 		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1771 			optval = 0;
1772 		} else {
1773 			optval = im6o->im6o_multicast_ifp->if_index;
1774 		}
1775 		INP_WUNLOCK(inp);
1776 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1777 		break;
1778 
1779 	case IPV6_MULTICAST_HOPS:
1780 		if (im6o == NULL)
1781 			optval = V_ip6_defmcasthlim;
1782 		else
1783 			optval = im6o->im6o_multicast_hlim;
1784 		INP_WUNLOCK(inp);
1785 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1786 		break;
1787 
1788 	case IPV6_MULTICAST_LOOP:
1789 		if (im6o == NULL)
1790 			optval = in6_mcast_loop; /* XXX VIMAGE */
1791 		else
1792 			optval = im6o->im6o_multicast_loop;
1793 		INP_WUNLOCK(inp);
1794 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1795 		break;
1796 
1797 	case IPV6_MSFILTER:
1798 		if (im6o == NULL) {
1799 			error = EADDRNOTAVAIL;
1800 			INP_WUNLOCK(inp);
1801 		} else {
1802 			error = in6p_get_source_filters(inp, sopt);
1803 		}
1804 		break;
1805 
1806 	default:
1807 		INP_WUNLOCK(inp);
1808 		error = ENOPROTOOPT;
1809 		break;
1810 	}
1811 
1812 	INP_UNLOCK_ASSERT(inp);
1813 
1814 	return (error);
1815 }
1816 
1817 /*
1818  * Look up the ifnet to use for a multicast group membership,
1819  * given the address of an IPv6 group.
1820  *
1821  * This routine exists to support legacy IPv6 multicast applications.
1822  *
1823  * If inp is non-NULL, use this socket's current FIB number for any
1824  * required FIB lookup. Look up the group address in the unicast FIB,
1825  * and use its ifp; usually, this points to the default next-hop.
1826  * If the FIB lookup fails, return NULL.
1827  *
1828  * FUTURE: Support multiple forwarding tables for IPv6.
1829  *
1830  * Returns NULL if no ifp could be found.
1831  */
1832 static struct ifnet *
1833 in6p_lookup_mcast_ifp(const struct inpcb *inp,
1834     const struct sockaddr_in6 *gsin6)
1835 {
1836 	struct nhop6_basic	nh6;
1837 	struct in6_addr		dst;
1838 	uint32_t		scopeid;
1839 	uint32_t		fibnum;
1840 
1841 	KASSERT(inp->inp_vflag & INP_IPV6,
1842 	    ("%s: not INP_IPV6 inpcb", __func__));
1843 	KASSERT(gsin6->sin6_family == AF_INET6,
1844 	    ("%s: not AF_INET6 group", __func__));
1845 
1846 	in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1847 	fibnum = inp ? inp->inp_inc.inc_fibnum : RT_DEFAULT_FIB;
1848 	if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6) != 0)
1849 		return (NULL);
1850 
1851 	return (nh6.nh_ifp);
1852 }
1853 
1854 /*
1855  * Join an IPv6 multicast group, possibly with a source.
1856  *
1857  * FIXME: The KAME use of the unspecified address (::)
1858  * to join *all* multicast groups is currently unsupported.
1859  */
1860 static int
1861 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1862 {
1863 	struct in6_multi_head		 inmh;
1864 	struct group_source_req		 gsr;
1865 	sockunion_t			*gsa, *ssa;
1866 	struct ifnet			*ifp;
1867 	struct in6_mfilter		*imf;
1868 	struct ip6_moptions		*imo;
1869 	struct in6_multi		*inm;
1870 	struct in6_msource		*lims;
1871 	int				 error, is_new;
1872 
1873 	SLIST_INIT(&inmh);
1874 	ifp = NULL;
1875 	lims = NULL;
1876 	error = 0;
1877 
1878 	memset(&gsr, 0, sizeof(struct group_source_req));
1879 	gsa = (sockunion_t *)&gsr.gsr_group;
1880 	gsa->ss.ss_family = AF_UNSPEC;
1881 	ssa = (sockunion_t *)&gsr.gsr_source;
1882 	ssa->ss.ss_family = AF_UNSPEC;
1883 
1884 	/*
1885 	 * Chew everything into struct group_source_req.
1886 	 * Overwrite the port field if present, as the sockaddr
1887 	 * being copied in may be matched with a binary comparison.
1888 	 * Ignore passed-in scope ID.
1889 	 */
1890 	switch (sopt->sopt_name) {
1891 	case IPV6_JOIN_GROUP: {
1892 		struct ipv6_mreq mreq;
1893 
1894 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1895 		    sizeof(struct ipv6_mreq));
1896 		if (error)
1897 			return (error);
1898 
1899 		gsa->sin6.sin6_family = AF_INET6;
1900 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1901 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1902 
1903 		if (mreq.ipv6mr_interface == 0) {
1904 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1905 		} else {
1906 			if (V_if_index < mreq.ipv6mr_interface)
1907 				return (EADDRNOTAVAIL);
1908 			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1909 		}
1910 		CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1911 		    __func__, mreq.ipv6mr_interface, ifp);
1912 	} break;
1913 
1914 	case MCAST_JOIN_GROUP:
1915 	case MCAST_JOIN_SOURCE_GROUP:
1916 		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1917 			error = sooptcopyin(sopt, &gsr,
1918 			    sizeof(struct group_req),
1919 			    sizeof(struct group_req));
1920 		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1921 			error = sooptcopyin(sopt, &gsr,
1922 			    sizeof(struct group_source_req),
1923 			    sizeof(struct group_source_req));
1924 		}
1925 		if (error)
1926 			return (error);
1927 
1928 		if (gsa->sin6.sin6_family != AF_INET6 ||
1929 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1930 			return (EINVAL);
1931 
1932 		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1933 			if (ssa->sin6.sin6_family != AF_INET6 ||
1934 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1935 				return (EINVAL);
1936 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1937 				return (EINVAL);
1938 			/*
1939 			 * TODO: Validate embedded scope ID in source
1940 			 * list entry against passed-in ifp, if and only
1941 			 * if source list filter entry is iface or node local.
1942 			 */
1943 			in6_clearscope(&ssa->sin6.sin6_addr);
1944 			ssa->sin6.sin6_port = 0;
1945 			ssa->sin6.sin6_scope_id = 0;
1946 		}
1947 
1948 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1949 			return (EADDRNOTAVAIL);
1950 		ifp = ifnet_byindex(gsr.gsr_interface);
1951 		break;
1952 
1953 	default:
1954 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1955 		    __func__, sopt->sopt_name);
1956 		return (EOPNOTSUPP);
1957 		break;
1958 	}
1959 
1960 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1961 		return (EINVAL);
1962 
1963 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1964 		return (EADDRNOTAVAIL);
1965 
1966 	gsa->sin6.sin6_port = 0;
1967 	gsa->sin6.sin6_scope_id = 0;
1968 
1969 	/*
1970 	 * Always set the scope zone ID on memberships created from userland.
1971 	 * Use the passed-in ifp to do this.
1972 	 * XXX The in6_setscope() return value is meaningless.
1973 	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1974 	 */
1975 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1976 
1977 	IN6_MULTI_LOCK();
1978 
1979 	/*
1980 	 * Find the membership in the membership list.
1981 	 */
1982 	imo = in6p_findmoptions(inp);
1983 	imf = im6o_match_group(imo, ifp, &gsa->sa);
1984 	if (imf == NULL) {
1985 		is_new = 1;
1986 		inm = NULL;
1987 
1988 		if (ip6_mfilter_count(&imo->im6o_head) >= IPV6_MAX_MEMBERSHIPS) {
1989 			error = ENOMEM;
1990 			goto out_in6p_locked;
1991 		}
1992 	} else {
1993 		is_new = 0;
1994 		inm = imf->im6f_in6m;
1995 
1996 		if (ssa->ss.ss_family != AF_UNSPEC) {
1997 			/*
1998 			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1999 			 * is an error. On an existing inclusive membership,
2000 			 * it just adds the source to the filter list.
2001 			 */
2002 			if (imf->im6f_st[1] != MCAST_INCLUDE) {
2003 				error = EINVAL;
2004 				goto out_in6p_locked;
2005 			}
2006 			/*
2007 			 * Throw out duplicates.
2008 			 *
2009 			 * XXX FIXME: This makes a naive assumption that
2010 			 * even if entries exist for *ssa in this imf,
2011 			 * they will be rejected as dupes, even if they
2012 			 * are not valid in the current mode (in-mode).
2013 			 *
2014 			 * in6_msource is transactioned just as for anything
2015 			 * else in SSM -- but note naive use of in6m_graft()
2016 			 * below for allocating new filter entries.
2017 			 *
2018 			 * This is only an issue if someone mixes the
2019 			 * full-state SSM API with the delta-based API,
2020 			 * which is discouraged in the relevant RFCs.
2021 			 */
2022 			lims = im6o_match_source(imf, &ssa->sa);
2023 			if (lims != NULL /*&&
2024 			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2025 				error = EADDRNOTAVAIL;
2026 				goto out_in6p_locked;
2027 			}
2028 		} else {
2029 			/*
2030 			 * MCAST_JOIN_GROUP alone, on any existing membership,
2031 			 * is rejected, to stop the same inpcb tying up
2032 			 * multiple refs to the in_multi.
2033 			 * On an existing inclusive membership, this is also
2034 			 * an error; if you want to change filter mode,
2035 			 * you must use the userland API setsourcefilter().
2036 			 * XXX We don't reject this for imf in UNDEFINED
2037 			 * state at t1, because allocation of a filter
2038 			 * is atomic with allocation of a membership.
2039 			 */
2040 			error = EINVAL;
2041 			goto out_in6p_locked;
2042 		}
2043 	}
2044 
2045 	/*
2046 	 * Begin state merge transaction at socket layer.
2047 	 */
2048 	INP_WLOCK_ASSERT(inp);
2049 
2050 	/*
2051 	 * Graft new source into filter list for this inpcb's
2052 	 * membership of the group. The in6_multi may not have
2053 	 * been allocated yet if this is a new membership, however,
2054 	 * the in_mfilter slot will be allocated and must be initialized.
2055 	 *
2056 	 * Note: Grafting of exclusive mode filters doesn't happen
2057 	 * in this path.
2058 	 * XXX: Should check for non-NULL lims (node exists but may
2059 	 * not be in-mode) for interop with full-state API.
2060 	 */
2061 	if (ssa->ss.ss_family != AF_UNSPEC) {
2062 		/* Membership starts in IN mode */
2063 		if (is_new) {
2064 			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2065 			imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2066 			if (imf == NULL) {
2067 				error = ENOMEM;
2068 				goto out_in6p_locked;
2069 			}
2070 		} else {
2071 			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2072 		}
2073 		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2074 		if (lims == NULL) {
2075 			CTR1(KTR_MLD, "%s: merge imf state failed",
2076 			    __func__);
2077 			error = ENOMEM;
2078 			goto out_in6p_locked;
2079 		}
2080 	} else {
2081 		/* No address specified; Membership starts in EX mode */
2082 		if (is_new) {
2083 			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2084 			imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2085 			if (imf == NULL) {
2086 				error = ENOMEM;
2087 				goto out_in6p_locked;
2088 			}
2089 		}
2090 	}
2091 
2092 	/*
2093 	 * Begin state merge transaction at MLD layer.
2094 	 */
2095 	if (is_new) {
2096 		in_pcbref(inp);
2097 		INP_WUNLOCK(inp);
2098 
2099 		error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2100 		    &imf->im6f_in6m, 0);
2101 
2102 		INP_WLOCK(inp);
2103 		if (in_pcbrele_wlocked(inp)) {
2104 			error = ENXIO;
2105 			goto out_in6p_unlocked;
2106 		}
2107 		if (error) {
2108 			goto out_in6p_locked;
2109 		}
2110 		/*
2111 		 * NOTE: Refcount from in6_joingroup_locked()
2112 		 * is protecting membership.
2113 		 */
2114 		ip6_mfilter_insert(&imo->im6o_head, imf);
2115 	} else {
2116 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2117 		IN6_MULTI_LIST_LOCK();
2118 		error = in6m_merge(inm, imf);
2119 		if (error) {
2120 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2121 			    __func__);
2122 			IN6_MULTI_LIST_UNLOCK();
2123 			im6f_rollback(imf);
2124 			im6f_reap(imf);
2125 			goto out_in6p_locked;
2126 		}
2127 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2128 		error = mld_change_state(inm, 0);
2129 		IN6_MULTI_LIST_UNLOCK();
2130 
2131 		if (error) {
2132 			CTR1(KTR_MLD, "%s: failed mld downcall",
2133 			     __func__);
2134 			im6f_rollback(imf);
2135 			im6f_reap(imf);
2136 			goto out_in6p_locked;
2137 		}
2138 	}
2139 
2140 	im6f_commit(imf);
2141 	imf = NULL;
2142 
2143 out_in6p_locked:
2144 	INP_WUNLOCK(inp);
2145 out_in6p_unlocked:
2146 	IN6_MULTI_UNLOCK();
2147 
2148 	if (is_new && imf) {
2149 		if (imf->im6f_in6m != NULL) {
2150 			struct in6_multi_head inmh;
2151 
2152 			SLIST_INIT(&inmh);
2153 			SLIST_INSERT_HEAD(&inmh, imf->im6f_in6m, in6m_defer);
2154 			in6m_release_list_deferred(&inmh);
2155 		}
2156 		ip6_mfilter_free(imf);
2157 	}
2158 	return (error);
2159 }
2160 
2161 /*
2162  * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2163  */
2164 static int
2165 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2166 {
2167 	struct ipv6_mreq		 mreq;
2168 	struct group_source_req		 gsr;
2169 	sockunion_t			*gsa, *ssa;
2170 	struct ifnet			*ifp;
2171 	struct in6_mfilter		*imf;
2172 	struct ip6_moptions		*imo;
2173 	struct in6_msource		*ims;
2174 	struct in6_multi		*inm;
2175 	uint32_t			 ifindex;
2176 	int				 error;
2177 	bool				 is_final;
2178 #ifdef KTR
2179 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2180 #endif
2181 
2182 	ifp = NULL;
2183 	ifindex = 0;
2184 	error = 0;
2185 	is_final = true;
2186 
2187 	memset(&gsr, 0, sizeof(struct group_source_req));
2188 	gsa = (sockunion_t *)&gsr.gsr_group;
2189 	gsa->ss.ss_family = AF_UNSPEC;
2190 	ssa = (sockunion_t *)&gsr.gsr_source;
2191 	ssa->ss.ss_family = AF_UNSPEC;
2192 
2193 	/*
2194 	 * Chew everything passed in up into a struct group_source_req
2195 	 * as that is easier to process.
2196 	 * Note: Any embedded scope ID in the multicast group passed
2197 	 * in by userland is ignored, the interface index is the recommended
2198 	 * mechanism to specify an interface; see below.
2199 	 */
2200 	switch (sopt->sopt_name) {
2201 	case IPV6_LEAVE_GROUP:
2202 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2203 		    sizeof(struct ipv6_mreq));
2204 		if (error)
2205 			return (error);
2206 		gsa->sin6.sin6_family = AF_INET6;
2207 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2208 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2209 		gsa->sin6.sin6_port = 0;
2210 		gsa->sin6.sin6_scope_id = 0;
2211 		ifindex = mreq.ipv6mr_interface;
2212 		break;
2213 
2214 	case MCAST_LEAVE_GROUP:
2215 	case MCAST_LEAVE_SOURCE_GROUP:
2216 		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2217 			error = sooptcopyin(sopt, &gsr,
2218 			    sizeof(struct group_req),
2219 			    sizeof(struct group_req));
2220 		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2221 			error = sooptcopyin(sopt, &gsr,
2222 			    sizeof(struct group_source_req),
2223 			    sizeof(struct group_source_req));
2224 		}
2225 		if (error)
2226 			return (error);
2227 
2228 		if (gsa->sin6.sin6_family != AF_INET6 ||
2229 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2230 			return (EINVAL);
2231 		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2232 			if (ssa->sin6.sin6_family != AF_INET6 ||
2233 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2234 				return (EINVAL);
2235 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2236 				return (EINVAL);
2237 			/*
2238 			 * TODO: Validate embedded scope ID in source
2239 			 * list entry against passed-in ifp, if and only
2240 			 * if source list filter entry is iface or node local.
2241 			 */
2242 			in6_clearscope(&ssa->sin6.sin6_addr);
2243 		}
2244 		gsa->sin6.sin6_port = 0;
2245 		gsa->sin6.sin6_scope_id = 0;
2246 		ifindex = gsr.gsr_interface;
2247 		break;
2248 
2249 	default:
2250 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2251 		    __func__, sopt->sopt_name);
2252 		return (EOPNOTSUPP);
2253 		break;
2254 	}
2255 
2256 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2257 		return (EINVAL);
2258 
2259 	/*
2260 	 * Validate interface index if provided. If no interface index
2261 	 * was provided separately, attempt to look the membership up
2262 	 * from the default scope as a last resort to disambiguate
2263 	 * the membership we are being asked to leave.
2264 	 * XXX SCOPE6 lock potentially taken here.
2265 	 */
2266 	if (ifindex != 0) {
2267 		if (V_if_index < ifindex)
2268 			return (EADDRNOTAVAIL);
2269 		ifp = ifnet_byindex(ifindex);
2270 		if (ifp == NULL)
2271 			return (EADDRNOTAVAIL);
2272 		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2273 	} else {
2274 		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2275 		if (error)
2276 			return (EADDRNOTAVAIL);
2277 		/*
2278 		 * Some badly behaved applications don't pass an ifindex
2279 		 * or a scope ID, which is an API violation. In this case,
2280 		 * perform a lookup as per a v6 join.
2281 		 *
2282 		 * XXX For now, stomp on zone ID for the corner case.
2283 		 * This is not the 'KAME way', but we need to see the ifp
2284 		 * directly until such time as this implementation is
2285 		 * refactored, assuming the scope IDs are the way to go.
2286 		 */
2287 		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2288 		if (ifindex == 0) {
2289 			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2290 			    "ifp for group %s.", __func__,
2291 			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2292 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2293 		} else {
2294 			ifp = ifnet_byindex(ifindex);
2295 		}
2296 		if (ifp == NULL)
2297 			return (EADDRNOTAVAIL);
2298 	}
2299 
2300 	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2301 	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2302 
2303 	IN6_MULTI_LOCK();
2304 
2305 	/*
2306 	 * Find the membership in the membership list.
2307 	 */
2308 	imo = in6p_findmoptions(inp);
2309 	imf = im6o_match_group(imo, ifp, &gsa->sa);
2310 	if (imf == NULL) {
2311 		error = EADDRNOTAVAIL;
2312 		goto out_in6p_locked;
2313 	}
2314 	inm = imf->im6f_in6m;
2315 
2316 	if (ssa->ss.ss_family != AF_UNSPEC)
2317 		is_final = false;
2318 
2319 	/*
2320 	 * Begin state merge transaction at socket layer.
2321 	 */
2322 	INP_WLOCK_ASSERT(inp);
2323 
2324 	/*
2325 	 * If we were instructed only to leave a given source, do so.
2326 	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2327 	 */
2328 	if (is_final) {
2329 		ip6_mfilter_remove(&imo->im6o_head, imf);
2330 		im6f_leave(imf);
2331 
2332 		/*
2333 		 * Give up the multicast address record to which
2334 		 * the membership points.
2335 		 */
2336 		(void)in6_leavegroup_locked(inm, imf);
2337 	} else {
2338 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2339 			error = EADDRNOTAVAIL;
2340 			goto out_in6p_locked;
2341 		}
2342 		ims = im6o_match_source(imf, &ssa->sa);
2343 		if (ims == NULL) {
2344 			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2345 			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2346 			    "not ");
2347 			error = EADDRNOTAVAIL;
2348 			goto out_in6p_locked;
2349 		}
2350 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2351 		error = im6f_prune(imf, &ssa->sin6);
2352 		if (error) {
2353 			CTR1(KTR_MLD, "%s: merge imf state failed",
2354 			    __func__);
2355 			goto out_in6p_locked;
2356 		}
2357 	}
2358 
2359 	/*
2360 	 * Begin state merge transaction at MLD layer.
2361 	 */
2362 	if (!is_final) {
2363 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2364 		IN6_MULTI_LIST_LOCK();
2365 		error = in6m_merge(inm, imf);
2366 		if (error) {
2367 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2368 			    __func__);
2369 			IN6_MULTI_LIST_UNLOCK();
2370 			im6f_rollback(imf);
2371 			im6f_reap(imf);
2372                         goto out_in6p_locked;
2373 		}
2374 
2375 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2376 		error = mld_change_state(inm, 0);
2377 		IN6_MULTI_LIST_UNLOCK();
2378 		if (error) {
2379 			CTR1(KTR_MLD, "%s: failed mld downcall",
2380 			     __func__);
2381 			im6f_rollback(imf);
2382 			im6f_reap(imf);
2383                         goto out_in6p_locked;
2384 		}
2385 	}
2386 
2387 	im6f_commit(imf);
2388 	im6f_reap(imf);
2389 
2390 out_in6p_locked:
2391 	INP_WUNLOCK(inp);
2392 
2393 	if (is_final && imf)
2394 		ip6_mfilter_free(imf);
2395 
2396 	IN6_MULTI_UNLOCK();
2397 	return (error);
2398 }
2399 
2400 /*
2401  * Select the interface for transmitting IPv6 multicast datagrams.
2402  *
2403  * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2404  * may be passed to this socket option. An address of in6addr_any or an
2405  * interface index of 0 is used to remove a previous selection.
2406  * When no interface is selected, one is chosen for every send.
2407  */
2408 static int
2409 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2410 {
2411 	struct ifnet		*ifp;
2412 	struct ip6_moptions	*imo;
2413 	u_int			 ifindex;
2414 	int			 error;
2415 
2416 	if (sopt->sopt_valsize != sizeof(u_int))
2417 		return (EINVAL);
2418 
2419 	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2420 	if (error)
2421 		return (error);
2422 	if (V_if_index < ifindex)
2423 		return (EINVAL);
2424 	if (ifindex == 0)
2425 		ifp = NULL;
2426 	else {
2427 		ifp = ifnet_byindex(ifindex);
2428 		if (ifp == NULL)
2429 			return (EINVAL);
2430 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2431 			return (EADDRNOTAVAIL);
2432 	}
2433 	imo = in6p_findmoptions(inp);
2434 	imo->im6o_multicast_ifp = ifp;
2435 	INP_WUNLOCK(inp);
2436 
2437 	return (0);
2438 }
2439 
2440 /*
2441  * Atomically set source filters on a socket for an IPv6 multicast group.
2442  *
2443  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2444  */
2445 static int
2446 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2447 {
2448 	struct __msfilterreq	 msfr;
2449 	sockunion_t		*gsa;
2450 	struct ifnet		*ifp;
2451 	struct in6_mfilter	*imf;
2452 	struct ip6_moptions	*imo;
2453 	struct in6_multi		*inm;
2454 	int			 error;
2455 
2456 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2457 	    sizeof(struct __msfilterreq));
2458 	if (error)
2459 		return (error);
2460 
2461 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2462 		return (ENOBUFS);
2463 
2464 	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2465 	    msfr.msfr_fmode != MCAST_INCLUDE)
2466 		return (EINVAL);
2467 
2468 	if (msfr.msfr_group.ss_family != AF_INET6 ||
2469 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2470 		return (EINVAL);
2471 
2472 	gsa = (sockunion_t *)&msfr.msfr_group;
2473 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2474 		return (EINVAL);
2475 
2476 	gsa->sin6.sin6_port = 0;	/* ignore port */
2477 
2478 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2479 		return (EADDRNOTAVAIL);
2480 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2481 	if (ifp == NULL)
2482 		return (EADDRNOTAVAIL);
2483 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2484 
2485 	/*
2486 	 * Take the INP write lock.
2487 	 * Check if this socket is a member of this group.
2488 	 */
2489 	imo = in6p_findmoptions(inp);
2490 	imf = im6o_match_group(imo, ifp, &gsa->sa);
2491 	if (imf == NULL) {
2492 		error = EADDRNOTAVAIL;
2493 		goto out_in6p_locked;
2494 	}
2495 	inm = imf->im6f_in6m;
2496 
2497 	/*
2498 	 * Begin state merge transaction at socket layer.
2499 	 */
2500 	INP_WLOCK_ASSERT(inp);
2501 
2502 	imf->im6f_st[1] = msfr.msfr_fmode;
2503 
2504 	/*
2505 	 * Apply any new source filters, if present.
2506 	 * Make a copy of the user-space source vector so
2507 	 * that we may copy them with a single copyin. This
2508 	 * allows us to deal with page faults up-front.
2509 	 */
2510 	if (msfr.msfr_nsrcs > 0) {
2511 		struct in6_msource	*lims;
2512 		struct sockaddr_in6	*psin;
2513 		struct sockaddr_storage	*kss, *pkss;
2514 		int			 i;
2515 
2516 		INP_WUNLOCK(inp);
2517 
2518 		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2519 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2520 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2521 		    M_TEMP, M_WAITOK);
2522 		error = copyin(msfr.msfr_srcs, kss,
2523 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2524 		if (error) {
2525 			free(kss, M_TEMP);
2526 			return (error);
2527 		}
2528 
2529 		INP_WLOCK(inp);
2530 
2531 		/*
2532 		 * Mark all source filters as UNDEFINED at t1.
2533 		 * Restore new group filter mode, as im6f_leave()
2534 		 * will set it to INCLUDE.
2535 		 */
2536 		im6f_leave(imf);
2537 		imf->im6f_st[1] = msfr.msfr_fmode;
2538 
2539 		/*
2540 		 * Update socket layer filters at t1, lazy-allocating
2541 		 * new entries. This saves a bunch of memory at the
2542 		 * cost of one RB_FIND() per source entry; duplicate
2543 		 * entries in the msfr_nsrcs vector are ignored.
2544 		 * If we encounter an error, rollback transaction.
2545 		 *
2546 		 * XXX This too could be replaced with a set-symmetric
2547 		 * difference like loop to avoid walking from root
2548 		 * every time, as the key space is common.
2549 		 */
2550 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2551 			psin = (struct sockaddr_in6 *)pkss;
2552 			if (psin->sin6_family != AF_INET6) {
2553 				error = EAFNOSUPPORT;
2554 				break;
2555 			}
2556 			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2557 				error = EINVAL;
2558 				break;
2559 			}
2560 			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2561 				error = EINVAL;
2562 				break;
2563 			}
2564 			/*
2565 			 * TODO: Validate embedded scope ID in source
2566 			 * list entry against passed-in ifp, if and only
2567 			 * if source list filter entry is iface or node local.
2568 			 */
2569 			in6_clearscope(&psin->sin6_addr);
2570 			error = im6f_get_source(imf, psin, &lims);
2571 			if (error)
2572 				break;
2573 			lims->im6sl_st[1] = imf->im6f_st[1];
2574 		}
2575 		free(kss, M_TEMP);
2576 	}
2577 
2578 	if (error)
2579 		goto out_im6f_rollback;
2580 
2581 	INP_WLOCK_ASSERT(inp);
2582 	IN6_MULTI_LIST_LOCK();
2583 
2584 	/*
2585 	 * Begin state merge transaction at MLD layer.
2586 	 */
2587 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2588 	error = in6m_merge(inm, imf);
2589 	if (error)
2590 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2591 	else {
2592 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2593 		error = mld_change_state(inm, 0);
2594 		if (error)
2595 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2596 	}
2597 
2598 	IN6_MULTI_LIST_UNLOCK();
2599 
2600 out_im6f_rollback:
2601 	if (error)
2602 		im6f_rollback(imf);
2603 	else
2604 		im6f_commit(imf);
2605 
2606 	im6f_reap(imf);
2607 
2608 out_in6p_locked:
2609 	INP_WUNLOCK(inp);
2610 	return (error);
2611 }
2612 
2613 /*
2614  * Set the IP multicast options in response to user setsockopt().
2615  *
2616  * Many of the socket options handled in this function duplicate the
2617  * functionality of socket options in the regular unicast API. However,
2618  * it is not possible to merge the duplicate code, because the idempotence
2619  * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2620  * the effects of these options must be treated as separate and distinct.
2621  *
2622  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2623  */
2624 int
2625 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2626 {
2627 	struct ip6_moptions	*im6o;
2628 	int			 error;
2629 
2630 	error = 0;
2631 
2632 	/*
2633 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2634 	 * or is a divert socket, reject it.
2635 	 */
2636 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2637 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2638 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2639 		return (EOPNOTSUPP);
2640 
2641 	switch (sopt->sopt_name) {
2642 	case IPV6_MULTICAST_IF:
2643 		error = in6p_set_multicast_if(inp, sopt);
2644 		break;
2645 
2646 	case IPV6_MULTICAST_HOPS: {
2647 		int hlim;
2648 
2649 		if (sopt->sopt_valsize != sizeof(int)) {
2650 			error = EINVAL;
2651 			break;
2652 		}
2653 		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2654 		if (error)
2655 			break;
2656 		if (hlim < -1 || hlim > 255) {
2657 			error = EINVAL;
2658 			break;
2659 		} else if (hlim == -1) {
2660 			hlim = V_ip6_defmcasthlim;
2661 		}
2662 		im6o = in6p_findmoptions(inp);
2663 		im6o->im6o_multicast_hlim = hlim;
2664 		INP_WUNLOCK(inp);
2665 		break;
2666 	}
2667 
2668 	case IPV6_MULTICAST_LOOP: {
2669 		u_int loop;
2670 
2671 		/*
2672 		 * Set the loopback flag for outgoing multicast packets.
2673 		 * Must be zero or one.
2674 		 */
2675 		if (sopt->sopt_valsize != sizeof(u_int)) {
2676 			error = EINVAL;
2677 			break;
2678 		}
2679 		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2680 		if (error)
2681 			break;
2682 		if (loop > 1) {
2683 			error = EINVAL;
2684 			break;
2685 		}
2686 		im6o = in6p_findmoptions(inp);
2687 		im6o->im6o_multicast_loop = loop;
2688 		INP_WUNLOCK(inp);
2689 		break;
2690 	}
2691 
2692 	case IPV6_JOIN_GROUP:
2693 	case MCAST_JOIN_GROUP:
2694 	case MCAST_JOIN_SOURCE_GROUP:
2695 		error = in6p_join_group(inp, sopt);
2696 		break;
2697 
2698 	case IPV6_LEAVE_GROUP:
2699 	case MCAST_LEAVE_GROUP:
2700 	case MCAST_LEAVE_SOURCE_GROUP:
2701 		error = in6p_leave_group(inp, sopt);
2702 		break;
2703 
2704 	case MCAST_BLOCK_SOURCE:
2705 	case MCAST_UNBLOCK_SOURCE:
2706 		error = in6p_block_unblock_source(inp, sopt);
2707 		break;
2708 
2709 	case IPV6_MSFILTER:
2710 		error = in6p_set_source_filters(inp, sopt);
2711 		break;
2712 
2713 	default:
2714 		error = EOPNOTSUPP;
2715 		break;
2716 	}
2717 
2718 	INP_UNLOCK_ASSERT(inp);
2719 
2720 	return (error);
2721 }
2722 
2723 /*
2724  * Expose MLD's multicast filter mode and source list(s) to userland,
2725  * keyed by (ifindex, group).
2726  * The filter mode is written out as a uint32_t, followed by
2727  * 0..n of struct in6_addr.
2728  * For use by ifmcstat(8).
2729  * SMPng: NOTE: unlocked read of ifindex space.
2730  */
2731 static int
2732 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2733 {
2734 	struct in6_addr			 mcaddr;
2735 	struct in6_addr			 src;
2736 	struct epoch_tracker		 et;
2737 	struct ifnet			*ifp;
2738 	struct ifmultiaddr		*ifma;
2739 	struct in6_multi		*inm;
2740 	struct ip6_msource		*ims;
2741 	int				*name;
2742 	int				 retval;
2743 	u_int				 namelen;
2744 	uint32_t			 fmode, ifindex;
2745 #ifdef KTR
2746 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2747 #endif
2748 
2749 	name = (int *)arg1;
2750 	namelen = arg2;
2751 
2752 	if (req->newptr != NULL)
2753 		return (EPERM);
2754 
2755 	/* int: ifindex + 4 * 32 bits of IPv6 address */
2756 	if (namelen != 5)
2757 		return (EINVAL);
2758 
2759 	ifindex = name[0];
2760 	if (ifindex <= 0 || ifindex > V_if_index) {
2761 		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2762 		    __func__, ifindex);
2763 		return (ENOENT);
2764 	}
2765 
2766 	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2767 	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2768 		CTR2(KTR_MLD, "%s: group %s is not multicast",
2769 		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2770 		return (EINVAL);
2771 	}
2772 
2773 	NET_EPOCH_ENTER(et);
2774 	ifp = ifnet_byindex(ifindex);
2775 	if (ifp == NULL) {
2776 		NET_EPOCH_EXIT(et);
2777 		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2778 		    __func__, ifindex);
2779 		return (ENOENT);
2780 	}
2781 	/*
2782 	 * Internal MLD lookups require that scope/zone ID is set.
2783 	 */
2784 	(void)in6_setscope(&mcaddr, ifp, NULL);
2785 
2786 	retval = sysctl_wire_old_buffer(req,
2787 	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2788 	if (retval) {
2789 		NET_EPOCH_EXIT(et);
2790 		return (retval);
2791 	}
2792 
2793 	IN6_MULTI_LOCK();
2794 	IN6_MULTI_LIST_LOCK();
2795 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2796 		inm = in6m_ifmultiaddr_get_inm(ifma);
2797 		if (inm == NULL)
2798 			continue;
2799 		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2800 			continue;
2801 		fmode = inm->in6m_st[1].iss_fmode;
2802 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2803 		if (retval != 0)
2804 			break;
2805 		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2806 			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2807 			/*
2808 			 * Only copy-out sources which are in-mode.
2809 			 */
2810 			if (fmode != im6s_get_mode(inm, ims, 1)) {
2811 				CTR1(KTR_MLD, "%s: skip non-in-mode",
2812 				    __func__);
2813 				continue;
2814 			}
2815 			src = ims->im6s_addr;
2816 			retval = SYSCTL_OUT(req, &src,
2817 			    sizeof(struct in6_addr));
2818 			if (retval != 0)
2819 				break;
2820 		}
2821 	}
2822 	IN6_MULTI_LIST_UNLOCK();
2823 	IN6_MULTI_UNLOCK();
2824 	NET_EPOCH_EXIT(et);
2825 
2826 	return (retval);
2827 }
2828 
2829 #ifdef KTR
2830 
2831 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2832 
2833 static const char *
2834 in6m_mode_str(const int mode)
2835 {
2836 
2837 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2838 		return (in6m_modestrs[mode]);
2839 	return ("??");
2840 }
2841 
2842 static const char *in6m_statestrs[] = {
2843 	"not-member",
2844 	"silent",
2845 	"idle",
2846 	"lazy",
2847 	"sleeping",
2848 	"awakening",
2849 	"query-pending",
2850 	"sg-query-pending",
2851 	"leaving"
2852 };
2853 
2854 static const char *
2855 in6m_state_str(const int state)
2856 {
2857 
2858 	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2859 		return (in6m_statestrs[state]);
2860 	return ("??");
2861 }
2862 
2863 /*
2864  * Dump an in6_multi structure to the console.
2865  */
2866 void
2867 in6m_print(const struct in6_multi *inm)
2868 {
2869 	int t;
2870 	char ip6tbuf[INET6_ADDRSTRLEN];
2871 
2872 	if ((ktr_mask & KTR_MLD) == 0)
2873 		return;
2874 
2875 	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2876 	printf("addr %s ifp %p(%s) ifma %p\n",
2877 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2878 	    inm->in6m_ifp,
2879 	    if_name(inm->in6m_ifp),
2880 	    inm->in6m_ifma);
2881 	printf("timer %u state %s refcount %u scq.len %u\n",
2882 	    inm->in6m_timer,
2883 	    in6m_state_str(inm->in6m_state),
2884 	    inm->in6m_refcount,
2885 	    mbufq_len(&inm->in6m_scq));
2886 	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2887 	    inm->in6m_mli,
2888 	    inm->in6m_nsrc,
2889 	    inm->in6m_sctimer,
2890 	    inm->in6m_scrv);
2891 	for (t = 0; t < 2; t++) {
2892 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2893 		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2894 		    inm->in6m_st[t].iss_asm,
2895 		    inm->in6m_st[t].iss_ex,
2896 		    inm->in6m_st[t].iss_in,
2897 		    inm->in6m_st[t].iss_rec);
2898 	}
2899 	printf("%s: --- end in6m %p ---\n", __func__, inm);
2900 }
2901 
2902 #else /* !KTR */
2903 
2904 void
2905 in6m_print(const struct in6_multi *inm)
2906 {
2907 
2908 }
2909 
2910 #endif /* KTR */
2911