xref: /freebsd/sys/netinet6/in6_mcast.c (revision cf7d8dea39cb00887476ad1a42480bce00b0b842)
1 /*
2  * Copyright (c) 2009 Bruce Simpson.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. The name of the author may not be used to endorse or promote
14  *    products derived from this software without specific prior written
15  *    permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  * IPv6 multicast socket, group, and socket option processing module.
32  * Normative references: RFC 2292, RFC 3492, RFC 3542, RFC 3678, RFC 3810.
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 #include "opt_inet6.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/protosw.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/protosw.h>
49 #include <sys/sysctl.h>
50 #include <sys/priv.h>
51 #include <sys/ktr.h>
52 #include <sys/tree.h>
53 
54 #include <net/if.h>
55 #include <net/if_dl.h>
56 #include <net/route.h>
57 #include <net/vnet.h>
58 
59 #include <netinet/in.h>
60 #include <netinet/in_var.h>
61 #include <netinet6/in6_var.h>
62 #include <netinet/ip6.h>
63 #include <netinet/icmp6.h>
64 #include <netinet6/ip6_var.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/tcp_var.h>
67 #include <netinet6/nd6.h>
68 #include <netinet6/mld6_var.h>
69 #include <netinet6/scope6_var.h>
70 
71 #ifndef KTR_MLD
72 #define KTR_MLD KTR_INET6
73 #endif
74 
75 #ifndef __SOCKUNION_DECLARED
76 union sockunion {
77 	struct sockaddr_storage	ss;
78 	struct sockaddr		sa;
79 	struct sockaddr_dl	sdl;
80 	struct sockaddr_in6	sin6;
81 };
82 typedef union sockunion sockunion_t;
83 #define __SOCKUNION_DECLARED
84 #endif /* __SOCKUNION_DECLARED */
85 
86 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
87     "IPv6 multicast PCB-layer source filter");
88 static MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
89 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
90 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
91     "IPv6 multicast MLD-layer source filter");
92 
93 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
94 
95 /*
96  * Locking:
97  * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
98  * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
99  *   it can be taken by code in net/if.c also.
100  * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
101  *
102  * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
103  * any need for in6_multi itself to be virtualized -- it is bound to an ifp
104  * anyway no matter what happens.
105  */
106 struct mtx in6_multi_mtx;
107 MTX_SYSINIT(in6_multi_mtx, &in6_multi_mtx, "in6_multi_mtx", MTX_DEF);
108 
109 static void	im6f_commit(struct in6_mfilter *);
110 static int	im6f_get_source(struct in6_mfilter *imf,
111 		    const struct sockaddr_in6 *psin,
112 		    struct in6_msource **);
113 static struct in6_msource *
114 		im6f_graft(struct in6_mfilter *, const uint8_t,
115 		    const struct sockaddr_in6 *);
116 static void	im6f_leave(struct in6_mfilter *);
117 static int	im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
118 static void	im6f_purge(struct in6_mfilter *);
119 static void	im6f_rollback(struct in6_mfilter *);
120 static void	im6f_reap(struct in6_mfilter *);
121 static int	im6o_grow(struct ip6_moptions *);
122 static size_t	im6o_match_group(const struct ip6_moptions *,
123 		    const struct ifnet *, const struct sockaddr *);
124 static struct in6_msource *
125 		im6o_match_source(const struct ip6_moptions *, const size_t,
126 		    const struct sockaddr *);
127 static void	im6s_merge(struct ip6_msource *ims,
128 		    const struct in6_msource *lims, const int rollback);
129 static int	in6_mc_get(struct ifnet *, const struct in6_addr *,
130 		    struct in6_multi **);
131 static int	in6m_get_source(struct in6_multi *inm,
132 		    const struct in6_addr *addr, const int noalloc,
133 		    struct ip6_msource **pims);
134 static int	in6m_is_ifp_detached(const struct in6_multi *);
135 static int	in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
136 static void	in6m_purge(struct in6_multi *);
137 static void	in6m_reap(struct in6_multi *);
138 static struct ip6_moptions *
139 		in6p_findmoptions(struct inpcb *);
140 static int	in6p_get_source_filters(struct inpcb *, struct sockopt *);
141 static int	in6p_join_group(struct inpcb *, struct sockopt *);
142 static int	in6p_leave_group(struct inpcb *, struct sockopt *);
143 static struct ifnet *
144 		in6p_lookup_mcast_ifp(const struct inpcb *,
145 		    const struct sockaddr_in6 *);
146 static int	in6p_block_unblock_source(struct inpcb *, struct sockopt *);
147 static int	in6p_set_multicast_if(struct inpcb *, struct sockopt *);
148 static int	in6p_set_source_filters(struct inpcb *, struct sockopt *);
149 static int	sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
150 
151 SYSCTL_DECL(_net_inet6_ip6);	/* XXX Not in any common header. */
152 
153 SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0, "IPv6 multicast");
154 
155 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
156 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
157     CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxgrpsrc, 0,
158     "Max source filters per group");
159 TUNABLE_ULONG("net.inet6.ip6.mcast.maxgrpsrc", &in6_mcast_maxgrpsrc);
160 
161 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
162 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
163     CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxsocksrc, 0,
164     "Max source filters per socket");
165 TUNABLE_ULONG("net.inet6.ip6.mcast.maxsocksrc", &in6_mcast_maxsocksrc);
166 
167 /* TODO Virtualize this switch. */
168 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
169 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
170     &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
171 TUNABLE_INT("net.inet6.ip6.mcast.loop", &in6_mcast_loop);
172 
173 SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
174     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
175     "Per-interface stack-wide source filters");
176 
177 /*
178  * Inline function which wraps assertions for a valid ifp.
179  * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
180  * is detached.
181  */
182 static int __inline
183 in6m_is_ifp_detached(const struct in6_multi *inm)
184 {
185 	struct ifnet *ifp;
186 
187 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
188 	ifp = inm->in6m_ifma->ifma_ifp;
189 	if (ifp != NULL) {
190 		/*
191 		 * Sanity check that network-layer notion of ifp is the
192 		 * same as that of link-layer.
193 		 */
194 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
195 	}
196 
197 	return (ifp == NULL);
198 }
199 
200 /*
201  * Initialize an in6_mfilter structure to a known state at t0, t1
202  * with an empty source filter list.
203  */
204 static __inline void
205 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
206 {
207 	memset(imf, 0, sizeof(struct in6_mfilter));
208 	RB_INIT(&imf->im6f_sources);
209 	imf->im6f_st[0] = st0;
210 	imf->im6f_st[1] = st1;
211 }
212 
213 /*
214  * Resize the ip6_moptions vector to the next power-of-two minus 1.
215  * May be called with locks held; do not sleep.
216  */
217 static int
218 im6o_grow(struct ip6_moptions *imo)
219 {
220 	struct in6_multi	**nmships;
221 	struct in6_multi	**omships;
222 	struct in6_mfilter	 *nmfilters;
223 	struct in6_mfilter	 *omfilters;
224 	size_t			  idx;
225 	size_t			  newmax;
226 	size_t			  oldmax;
227 
228 	nmships = NULL;
229 	nmfilters = NULL;
230 	omships = imo->im6o_membership;
231 	omfilters = imo->im6o_mfilters;
232 	oldmax = imo->im6o_max_memberships;
233 	newmax = ((oldmax + 1) * 2) - 1;
234 
235 	if (newmax <= IPV6_MAX_MEMBERSHIPS) {
236 		nmships = (struct in6_multi **)realloc(omships,
237 		    sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
238 		nmfilters = (struct in6_mfilter *)realloc(omfilters,
239 		    sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
240 		    M_NOWAIT);
241 		if (nmships != NULL && nmfilters != NULL) {
242 			/* Initialize newly allocated source filter heads. */
243 			for (idx = oldmax; idx < newmax; idx++) {
244 				im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
245 				    MCAST_EXCLUDE);
246 			}
247 			imo->im6o_max_memberships = newmax;
248 			imo->im6o_membership = nmships;
249 			imo->im6o_mfilters = nmfilters;
250 		}
251 	}
252 
253 	if (nmships == NULL || nmfilters == NULL) {
254 		if (nmships != NULL)
255 			free(nmships, M_IP6MOPTS);
256 		if (nmfilters != NULL)
257 			free(nmfilters, M_IN6MFILTER);
258 		return (ETOOMANYREFS);
259 	}
260 
261 	return (0);
262 }
263 
264 /*
265  * Find an IPv6 multicast group entry for this ip6_moptions instance
266  * which matches the specified group, and optionally an interface.
267  * Return its index into the array, or -1 if not found.
268  */
269 static size_t
270 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
271     const struct sockaddr *group)
272 {
273 	const struct sockaddr_in6 *gsin6;
274 	struct in6_multi	**pinm;
275 	int		  idx;
276 	int		  nmships;
277 
278 	gsin6 = (const struct sockaddr_in6 *)group;
279 
280 	/* The im6o_membership array may be lazy allocated. */
281 	if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
282 		return (-1);
283 
284 	nmships = imo->im6o_num_memberships;
285 	pinm = &imo->im6o_membership[0];
286 	for (idx = 0; idx < nmships; idx++, pinm++) {
287 		if (*pinm == NULL)
288 			continue;
289 		if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
290 		    IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
291 		    &gsin6->sin6_addr)) {
292 			break;
293 		}
294 	}
295 	if (idx >= nmships)
296 		idx = -1;
297 
298 	return (idx);
299 }
300 
301 /*
302  * Find an IPv6 multicast source entry for this imo which matches
303  * the given group index for this socket, and source address.
304  *
305  * XXX TODO: The scope ID, if present in src, is stripped before
306  * any comparison. We SHOULD enforce scope/zone checks where the source
307  * filter entry has a link scope.
308  *
309  * NOTE: This does not check if the entry is in-mode, merely if
310  * it exists, which may not be the desired behaviour.
311  */
312 static struct in6_msource *
313 im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
314     const struct sockaddr *src)
315 {
316 	struct ip6_msource	 find;
317 	struct in6_mfilter	*imf;
318 	struct ip6_msource	*ims;
319 	const sockunion_t	*psa;
320 
321 	KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
322 	KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
323 	    ("%s: invalid index %d\n", __func__, (int)gidx));
324 
325 	/* The im6o_mfilters array may be lazy allocated. */
326 	if (imo->im6o_mfilters == NULL)
327 		return (NULL);
328 	imf = &imo->im6o_mfilters[gidx];
329 
330 	psa = (const sockunion_t *)src;
331 	find.im6s_addr = psa->sin6.sin6_addr;
332 	in6_clearscope(&find.im6s_addr);		/* XXX */
333 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
334 
335 	return ((struct in6_msource *)ims);
336 }
337 
338 /*
339  * Perform filtering for multicast datagrams on a socket by group and source.
340  *
341  * Returns 0 if a datagram should be allowed through, or various error codes
342  * if the socket was not a member of the group, or the source was muted, etc.
343  */
344 int
345 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
346     const struct sockaddr *group, const struct sockaddr *src)
347 {
348 	size_t gidx;
349 	struct in6_msource *ims;
350 	int mode;
351 
352 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
353 
354 	gidx = im6o_match_group(imo, ifp, group);
355 	if (gidx == -1)
356 		return (MCAST_NOTGMEMBER);
357 
358 	/*
359 	 * Check if the source was included in an (S,G) join.
360 	 * Allow reception on exclusive memberships by default,
361 	 * reject reception on inclusive memberships by default.
362 	 * Exclude source only if an in-mode exclude filter exists.
363 	 * Include source only if an in-mode include filter exists.
364 	 * NOTE: We are comparing group state here at MLD t1 (now)
365 	 * with socket-layer t0 (since last downcall).
366 	 */
367 	mode = imo->im6o_mfilters[gidx].im6f_st[1];
368 	ims = im6o_match_source(imo, gidx, src);
369 
370 	if ((ims == NULL && mode == MCAST_INCLUDE) ||
371 	    (ims != NULL && ims->im6sl_st[0] != mode))
372 		return (MCAST_NOTSMEMBER);
373 
374 	return (MCAST_PASS);
375 }
376 
377 /*
378  * Find and return a reference to an in6_multi record for (ifp, group),
379  * and bump its reference count.
380  * If one does not exist, try to allocate it, and update link-layer multicast
381  * filters on ifp to listen for group.
382  * Assumes the IN6_MULTI lock is held across the call.
383  * Return 0 if successful, otherwise return an appropriate error code.
384  */
385 static int
386 in6_mc_get(struct ifnet *ifp, const struct in6_addr *group,
387     struct in6_multi **pinm)
388 {
389 	struct sockaddr_in6	 gsin6;
390 	struct ifmultiaddr	*ifma;
391 	struct in6_multi	*inm;
392 	int			 error;
393 
394 	error = 0;
395 
396 	/*
397 	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
398 	 * if_addmulti() takes this mutex itself, so we must drop and
399 	 * re-acquire around the call.
400 	 */
401 	IN6_MULTI_LOCK_ASSERT();
402 	IF_ADDR_LOCK(ifp);
403 
404 	inm = in6m_lookup_locked(ifp, group);
405 	if (inm != NULL) {
406 		/*
407 		 * If we already joined this group, just bump the
408 		 * refcount and return it.
409 		 */
410 		KASSERT(inm->in6m_refcount >= 1,
411 		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
412 		++inm->in6m_refcount;
413 		*pinm = inm;
414 		goto out_locked;
415 	}
416 
417 	memset(&gsin6, 0, sizeof(gsin6));
418 	gsin6.sin6_family = AF_INET6;
419 	gsin6.sin6_len = sizeof(struct sockaddr_in6);
420 	gsin6.sin6_addr = *group;
421 
422 	/*
423 	 * Check if a link-layer group is already associated
424 	 * with this network-layer group on the given ifnet.
425 	 */
426 	IF_ADDR_UNLOCK(ifp);
427 	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
428 	if (error != 0)
429 		return (error);
430 	IF_ADDR_LOCK(ifp);
431 
432 	/*
433 	 * If something other than netinet6 is occupying the link-layer
434 	 * group, print a meaningful error message and back out of
435 	 * the allocation.
436 	 * Otherwise, bump the refcount on the existing network-layer
437 	 * group association and return it.
438 	 */
439 	if (ifma->ifma_protospec != NULL) {
440 		inm = (struct in6_multi *)ifma->ifma_protospec;
441 #ifdef INVARIANTS
442 		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
443 		    __func__));
444 		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
445 		    ("%s: ifma not AF_INET6", __func__));
446 		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
447 		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
448 		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
449 			panic("%s: ifma %p is inconsistent with %p (%p)",
450 			    __func__, ifma, inm, group);
451 #endif
452 		++inm->in6m_refcount;
453 		*pinm = inm;
454 		goto out_locked;
455 	}
456 
457 	IF_ADDR_LOCK_ASSERT(ifp);
458 
459 	/*
460 	 * A new in6_multi record is needed; allocate and initialize it.
461 	 * We DO NOT perform an MLD join as the in6_ layer may need to
462 	 * push an initial source list down to MLD to support SSM.
463 	 *
464 	 * The initial source filter state is INCLUDE, {} as per the RFC.
465 	 * Pending state-changes per group are subject to a bounds check.
466 	 */
467 	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
468 	if (inm == NULL) {
469 		if_delmulti_ifma(ifma);
470 		error = ENOMEM;
471 		goto out_locked;
472 	}
473 	inm->in6m_addr = *group;
474 	inm->in6m_ifp = ifp;
475 	inm->in6m_mli = MLD_IFINFO(ifp);
476 	inm->in6m_ifma = ifma;
477 	inm->in6m_refcount = 1;
478 	inm->in6m_state = MLD_NOT_MEMBER;
479 	IFQ_SET_MAXLEN(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
480 
481 	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
482 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
483 	RB_INIT(&inm->in6m_srcs);
484 
485 	ifma->ifma_protospec = inm;
486 	*pinm = inm;
487 
488 out_locked:
489 	IF_ADDR_UNLOCK(ifp);
490 	return (error);
491 }
492 
493 /*
494  * Drop a reference to an in6_multi record.
495  *
496  * If the refcount drops to 0, free the in6_multi record and
497  * delete the underlying link-layer membership.
498  */
499 void
500 in6m_release_locked(struct in6_multi *inm)
501 {
502 	struct ifmultiaddr *ifma;
503 
504 	IN6_MULTI_LOCK_ASSERT();
505 
506 	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
507 
508 	if (--inm->in6m_refcount > 0) {
509 		CTR2(KTR_MLD, "%s: refcount is now %d", __func__,
510 		    inm->in6m_refcount);
511 		return;
512 	}
513 
514 	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
515 
516 	ifma = inm->in6m_ifma;
517 
518 	/* XXX this access is not covered by IF_ADDR_LOCK */
519 	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
520 	KASSERT(ifma->ifma_protospec == inm,
521 	    ("%s: ifma_protospec != inm", __func__));
522 	ifma->ifma_protospec = NULL;
523 
524 	in6m_purge(inm);
525 
526 	free(inm, M_IP6MADDR);
527 
528 	if_delmulti_ifma(ifma);
529 }
530 
531 /*
532  * Clear recorded source entries for a group.
533  * Used by the MLD code. Caller must hold the IN6_MULTI lock.
534  * FIXME: Should reap.
535  */
536 void
537 in6m_clear_recorded(struct in6_multi *inm)
538 {
539 	struct ip6_msource	*ims;
540 
541 	IN6_MULTI_LOCK_ASSERT();
542 
543 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
544 		if (ims->im6s_stp) {
545 			ims->im6s_stp = 0;
546 			--inm->in6m_st[1].iss_rec;
547 		}
548 	}
549 	KASSERT(inm->in6m_st[1].iss_rec == 0,
550 	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
551 }
552 
553 /*
554  * Record a source as pending for a Source-Group MLDv2 query.
555  * This lives here as it modifies the shared tree.
556  *
557  * inm is the group descriptor.
558  * naddr is the address of the source to record in network-byte order.
559  *
560  * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
561  * lazy-allocate a source node in response to an SG query.
562  * Otherwise, no allocation is performed. This saves some memory
563  * with the trade-off that the source will not be reported to the
564  * router if joined in the window between the query response and
565  * the group actually being joined on the local host.
566  *
567  * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
568  * This turns off the allocation of a recorded source entry if
569  * the group has not been joined.
570  *
571  * Return 0 if the source didn't exist or was already marked as recorded.
572  * Return 1 if the source was marked as recorded by this function.
573  * Return <0 if any error occured (negated errno code).
574  */
575 int
576 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
577 {
578 	struct ip6_msource	 find;
579 	struct ip6_msource	*ims, *nims;
580 
581 	IN6_MULTI_LOCK_ASSERT();
582 
583 	find.im6s_addr = *addr;
584 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
585 	if (ims && ims->im6s_stp)
586 		return (0);
587 	if (ims == NULL) {
588 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
589 			return (-ENOSPC);
590 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
591 		    M_NOWAIT | M_ZERO);
592 		if (nims == NULL)
593 			return (-ENOMEM);
594 		nims->im6s_addr = find.im6s_addr;
595 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
596 		++inm->in6m_nsrc;
597 		ims = nims;
598 	}
599 
600 	/*
601 	 * Mark the source as recorded and update the recorded
602 	 * source count.
603 	 */
604 	++ims->im6s_stp;
605 	++inm->in6m_st[1].iss_rec;
606 
607 	return (1);
608 }
609 
610 /*
611  * Return a pointer to an in6_msource owned by an in6_mfilter,
612  * given its source address.
613  * Lazy-allocate if needed. If this is a new entry its filter state is
614  * undefined at t0.
615  *
616  * imf is the filter set being modified.
617  * addr is the source address.
618  *
619  * SMPng: May be called with locks held; malloc must not block.
620  */
621 static int
622 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
623     struct in6_msource **plims)
624 {
625 	struct ip6_msource	 find;
626 	struct ip6_msource	*ims, *nims;
627 	struct in6_msource	*lims;
628 	int			 error;
629 
630 	error = 0;
631 	ims = NULL;
632 	lims = NULL;
633 
634 	find.im6s_addr = psin->sin6_addr;
635 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
636 	lims = (struct in6_msource *)ims;
637 	if (lims == NULL) {
638 		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
639 			return (ENOSPC);
640 		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
641 		    M_NOWAIT | M_ZERO);
642 		if (nims == NULL)
643 			return (ENOMEM);
644 		lims = (struct in6_msource *)nims;
645 		lims->im6s_addr = find.im6s_addr;
646 		lims->im6sl_st[0] = MCAST_UNDEFINED;
647 		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
648 		++imf->im6f_nsrc;
649 	}
650 
651 	*plims = lims;
652 
653 	return (error);
654 }
655 
656 /*
657  * Graft a source entry into an existing socket-layer filter set,
658  * maintaining any required invariants and checking allocations.
659  *
660  * The source is marked as being in the new filter mode at t1.
661  *
662  * Return the pointer to the new node, otherwise return NULL.
663  */
664 static struct in6_msource *
665 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
666     const struct sockaddr_in6 *psin)
667 {
668 	struct ip6_msource	*nims;
669 	struct in6_msource	*lims;
670 
671 	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
672 	    M_NOWAIT | M_ZERO);
673 	if (nims == NULL)
674 		return (NULL);
675 	lims = (struct in6_msource *)nims;
676 	lims->im6s_addr = psin->sin6_addr;
677 	lims->im6sl_st[0] = MCAST_UNDEFINED;
678 	lims->im6sl_st[1] = st1;
679 	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
680 	++imf->im6f_nsrc;
681 
682 	return (lims);
683 }
684 
685 /*
686  * Prune a source entry from an existing socket-layer filter set,
687  * maintaining any required invariants and checking allocations.
688  *
689  * The source is marked as being left at t1, it is not freed.
690  *
691  * Return 0 if no error occurred, otherwise return an errno value.
692  */
693 static int
694 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
695 {
696 	struct ip6_msource	 find;
697 	struct ip6_msource	*ims;
698 	struct in6_msource	*lims;
699 
700 	find.im6s_addr = psin->sin6_addr;
701 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
702 	if (ims == NULL)
703 		return (ENOENT);
704 	lims = (struct in6_msource *)ims;
705 	lims->im6sl_st[1] = MCAST_UNDEFINED;
706 	return (0);
707 }
708 
709 /*
710  * Revert socket-layer filter set deltas at t1 to t0 state.
711  */
712 static void
713 im6f_rollback(struct in6_mfilter *imf)
714 {
715 	struct ip6_msource	*ims, *tims;
716 	struct in6_msource	*lims;
717 
718 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
719 		lims = (struct in6_msource *)ims;
720 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
721 			/* no change at t1 */
722 			continue;
723 		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
724 			/* revert change to existing source at t1 */
725 			lims->im6sl_st[1] = lims->im6sl_st[0];
726 		} else {
727 			/* revert source added t1 */
728 			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
729 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
730 			free(ims, M_IN6MFILTER);
731 			imf->im6f_nsrc--;
732 		}
733 	}
734 	imf->im6f_st[1] = imf->im6f_st[0];
735 }
736 
737 /*
738  * Mark socket-layer filter set as INCLUDE {} at t1.
739  */
740 static void
741 im6f_leave(struct in6_mfilter *imf)
742 {
743 	struct ip6_msource	*ims;
744 	struct in6_msource	*lims;
745 
746 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
747 		lims = (struct in6_msource *)ims;
748 		lims->im6sl_st[1] = MCAST_UNDEFINED;
749 	}
750 	imf->im6f_st[1] = MCAST_INCLUDE;
751 }
752 
753 /*
754  * Mark socket-layer filter set deltas as committed.
755  */
756 static void
757 im6f_commit(struct in6_mfilter *imf)
758 {
759 	struct ip6_msource	*ims;
760 	struct in6_msource	*lims;
761 
762 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
763 		lims = (struct in6_msource *)ims;
764 		lims->im6sl_st[0] = lims->im6sl_st[1];
765 	}
766 	imf->im6f_st[0] = imf->im6f_st[1];
767 }
768 
769 /*
770  * Reap unreferenced sources from socket-layer filter set.
771  */
772 static void
773 im6f_reap(struct in6_mfilter *imf)
774 {
775 	struct ip6_msource	*ims, *tims;
776 	struct in6_msource	*lims;
777 
778 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
779 		lims = (struct in6_msource *)ims;
780 		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
781 		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
782 			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
783 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
784 			free(ims, M_IN6MFILTER);
785 			imf->im6f_nsrc--;
786 		}
787 	}
788 }
789 
790 /*
791  * Purge socket-layer filter set.
792  */
793 static void
794 im6f_purge(struct in6_mfilter *imf)
795 {
796 	struct ip6_msource	*ims, *tims;
797 
798 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
799 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
800 		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
801 		free(ims, M_IN6MFILTER);
802 		imf->im6f_nsrc--;
803 	}
804 	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
805 	KASSERT(RB_EMPTY(&imf->im6f_sources),
806 	    ("%s: im6f_sources not empty", __func__));
807 }
808 
809 /*
810  * Look up a source filter entry for a multicast group.
811  *
812  * inm is the group descriptor to work with.
813  * addr is the IPv6 address to look up.
814  * noalloc may be non-zero to suppress allocation of sources.
815  * *pims will be set to the address of the retrieved or allocated source.
816  *
817  * SMPng: NOTE: may be called with locks held.
818  * Return 0 if successful, otherwise return a non-zero error code.
819  */
820 static int
821 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
822     const int noalloc, struct ip6_msource **pims)
823 {
824 	struct ip6_msource	 find;
825 	struct ip6_msource	*ims, *nims;
826 #ifdef KTR
827 	char			 ip6tbuf[INET6_ADDRSTRLEN];
828 #endif
829 
830 	find.im6s_addr = *addr;
831 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
832 	if (ims == NULL && !noalloc) {
833 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
834 			return (ENOSPC);
835 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
836 		    M_NOWAIT | M_ZERO);
837 		if (nims == NULL)
838 			return (ENOMEM);
839 		nims->im6s_addr = *addr;
840 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
841 		++inm->in6m_nsrc;
842 		ims = nims;
843 		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
844 		    ip6_sprintf(ip6tbuf, addr), ims);
845 	}
846 
847 	*pims = ims;
848 	return (0);
849 }
850 
851 /*
852  * Merge socket-layer source into MLD-layer source.
853  * If rollback is non-zero, perform the inverse of the merge.
854  */
855 static void
856 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
857     const int rollback)
858 {
859 	int n = rollback ? -1 : 1;
860 #ifdef KTR
861 	char ip6tbuf[INET6_ADDRSTRLEN];
862 
863 	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
864 #endif
865 
866 	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
867 		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
868 		ims->im6s_st[1].ex -= n;
869 	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
870 		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
871 		ims->im6s_st[1].in -= n;
872 	}
873 
874 	if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
875 		CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
876 		ims->im6s_st[1].ex += n;
877 	} else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
878 		CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
879 		ims->im6s_st[1].in += n;
880 	}
881 }
882 
883 /*
884  * Atomically update the global in6_multi state, when a membership's
885  * filter list is being updated in any way.
886  *
887  * imf is the per-inpcb-membership group filter pointer.
888  * A fake imf may be passed for in-kernel consumers.
889  *
890  * XXX This is a candidate for a set-symmetric-difference style loop
891  * which would eliminate the repeated lookup from root of ims nodes,
892  * as they share the same key space.
893  *
894  * If any error occurred this function will back out of refcounts
895  * and return a non-zero value.
896  */
897 static int
898 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
899 {
900 	struct ip6_msource	*ims, *nims;
901 	struct in6_msource	*lims;
902 	int			 schanged, error;
903 	int			 nsrc0, nsrc1;
904 
905 	schanged = 0;
906 	error = 0;
907 	nsrc1 = nsrc0 = 0;
908 
909 	/*
910 	 * Update the source filters first, as this may fail.
911 	 * Maintain count of in-mode filters at t0, t1. These are
912 	 * used to work out if we transition into ASM mode or not.
913 	 * Maintain a count of source filters whose state was
914 	 * actually modified by this operation.
915 	 */
916 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
917 		lims = (struct in6_msource *)ims;
918 		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
919 		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
920 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
921 		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
922 		++schanged;
923 		if (error)
924 			break;
925 		im6s_merge(nims, lims, 0);
926 	}
927 	if (error) {
928 		struct ip6_msource *bims;
929 
930 		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
931 			lims = (struct in6_msource *)ims;
932 			if (lims->im6sl_st[0] == lims->im6sl_st[1])
933 				continue;
934 			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
935 			if (bims == NULL)
936 				continue;
937 			im6s_merge(bims, lims, 1);
938 		}
939 		goto out_reap;
940 	}
941 
942 	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
943 	    __func__, nsrc0, nsrc1);
944 
945 	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
946 	if (imf->im6f_st[0] == imf->im6f_st[1] &&
947 	    imf->im6f_st[1] == MCAST_INCLUDE) {
948 		if (nsrc1 == 0) {
949 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
950 			--inm->in6m_st[1].iss_in;
951 		}
952 	}
953 
954 	/* Handle filter mode transition on socket. */
955 	if (imf->im6f_st[0] != imf->im6f_st[1]) {
956 		CTR3(KTR_MLD, "%s: imf transition %d to %d",
957 		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
958 
959 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
960 			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
961 			--inm->in6m_st[1].iss_ex;
962 		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
963 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
964 			--inm->in6m_st[1].iss_in;
965 		}
966 
967 		if (imf->im6f_st[1] == MCAST_EXCLUDE) {
968 			CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
969 			inm->in6m_st[1].iss_ex++;
970 		} else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
971 			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
972 			inm->in6m_st[1].iss_in++;
973 		}
974 	}
975 
976 	/*
977 	 * Track inm filter state in terms of listener counts.
978 	 * If there are any exclusive listeners, stack-wide
979 	 * membership is exclusive.
980 	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
981 	 * If no listeners remain, state is undefined at t1,
982 	 * and the MLD lifecycle for this group should finish.
983 	 */
984 	if (inm->in6m_st[1].iss_ex > 0) {
985 		CTR1(KTR_MLD, "%s: transition to EX", __func__);
986 		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
987 	} else if (inm->in6m_st[1].iss_in > 0) {
988 		CTR1(KTR_MLD, "%s: transition to IN", __func__);
989 		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
990 	} else {
991 		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
992 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
993 	}
994 
995 	/* Decrement ASM listener count on transition out of ASM mode. */
996 	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
997 		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
998 		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
999 			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1000 			--inm->in6m_st[1].iss_asm;
1001 	}
1002 
1003 	/* Increment ASM listener count on transition to ASM mode. */
1004 	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1005 		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1006 		inm->in6m_st[1].iss_asm++;
1007 	}
1008 
1009 	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1010 	in6m_print(inm);
1011 
1012 out_reap:
1013 	if (schanged > 0) {
1014 		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1015 		in6m_reap(inm);
1016 	}
1017 	return (error);
1018 }
1019 
1020 /*
1021  * Mark an in6_multi's filter set deltas as committed.
1022  * Called by MLD after a state change has been enqueued.
1023  */
1024 void
1025 in6m_commit(struct in6_multi *inm)
1026 {
1027 	struct ip6_msource	*ims;
1028 
1029 	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1030 	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1031 	in6m_print(inm);
1032 
1033 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1034 		ims->im6s_st[0] = ims->im6s_st[1];
1035 	}
1036 	inm->in6m_st[0] = inm->in6m_st[1];
1037 }
1038 
1039 /*
1040  * Reap unreferenced nodes from an in6_multi's filter set.
1041  */
1042 static void
1043 in6m_reap(struct in6_multi *inm)
1044 {
1045 	struct ip6_msource	*ims, *tims;
1046 
1047 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1048 		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1049 		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1050 		    ims->im6s_stp != 0)
1051 			continue;
1052 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1053 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1054 		free(ims, M_IP6MSOURCE);
1055 		inm->in6m_nsrc--;
1056 	}
1057 }
1058 
1059 /*
1060  * Purge all source nodes from an in6_multi's filter set.
1061  */
1062 static void
1063 in6m_purge(struct in6_multi *inm)
1064 {
1065 	struct ip6_msource	*ims, *tims;
1066 
1067 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1068 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1069 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1070 		free(ims, M_IP6MSOURCE);
1071 		inm->in6m_nsrc--;
1072 	}
1073 }
1074 
1075 /*
1076  * Join a multicast address w/o sources.
1077  * KAME compatibility entry point.
1078  *
1079  * SMPng: Assume no mc locks held by caller.
1080  */
1081 struct in6_multi_mship *
1082 in6_joingroup(struct ifnet *ifp, struct in6_addr *mcaddr,
1083     int *errorp, int delay)
1084 {
1085 	struct in6_multi_mship *imm;
1086 	int error;
1087 
1088 	imm = malloc(sizeof(*imm), M_IP6MADDR, M_NOWAIT);
1089 	if (imm == NULL) {
1090 		*errorp = ENOBUFS;
1091 		return (NULL);
1092 	}
1093 
1094 	delay = (delay * PR_FASTHZ) / hz;
1095 
1096 	error = in6_mc_join(ifp, mcaddr, NULL, &imm->i6mm_maddr, delay);
1097 	if (error) {
1098 		*errorp = error;
1099 		free(imm, M_IP6MADDR);
1100 		return (NULL);
1101 	}
1102 
1103 	return (imm);
1104 }
1105 
1106 /*
1107  * Leave a multicast address w/o sources.
1108  * KAME compatibility entry point.
1109  *
1110  * SMPng: Assume no mc locks held by caller.
1111  */
1112 int
1113 in6_leavegroup(struct in6_multi_mship *imm)
1114 {
1115 
1116 	if (imm->i6mm_maddr != NULL)
1117 		in6_mc_leave(imm->i6mm_maddr, NULL);
1118 	free(imm,  M_IP6MADDR);
1119 	return 0;
1120 }
1121 
1122 /*
1123  * Join a multicast group; unlocked entry point.
1124  *
1125  * SMPng: XXX: in6_mc_join() is called from in6_control() when upper
1126  * locks are not held. Fortunately, ifp is unlikely to have been detached
1127  * at this point, so we assume it's OK to recurse.
1128  */
1129 int
1130 in6_mc_join(struct ifnet *ifp, const struct in6_addr *mcaddr,
1131     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1132     const int delay)
1133 {
1134 	int error;
1135 
1136 	IN6_MULTI_LOCK();
1137 	error = in6_mc_join_locked(ifp, mcaddr, imf, pinm, delay);
1138 	IN6_MULTI_UNLOCK();
1139 
1140 	return (error);
1141 }
1142 
1143 /*
1144  * Join a multicast group; real entry point.
1145  *
1146  * Only preserves atomicity at inm level.
1147  * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1148  *
1149  * If the MLD downcall fails, the group is not joined, and an error
1150  * code is returned.
1151  */
1152 int
1153 in6_mc_join_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1154     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1155     const int delay)
1156 {
1157 	struct in6_mfilter	 timf;
1158 	struct in6_multi	*inm;
1159 	int			 error;
1160 #ifdef KTR
1161 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1162 #endif
1163 
1164 #ifdef INVARIANTS
1165 	/*
1166 	 * Sanity: Check scope zone ID was set for ifp, if and
1167 	 * only if group is scoped to an interface.
1168 	 */
1169 	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1170 	    ("%s: not a multicast address", __func__));
1171 	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1172 	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1173 		KASSERT(mcaddr->s6_addr16[1] != 0,
1174 		    ("%s: scope zone ID not set", __func__));
1175 	}
1176 #endif
1177 
1178 	IN6_MULTI_LOCK_ASSERT();
1179 
1180 	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1181 	    ip6_sprintf(ip6tbuf, mcaddr), ifp, ifp->if_xname);
1182 
1183 	error = 0;
1184 	inm = NULL;
1185 
1186 	/*
1187 	 * If no imf was specified (i.e. kernel consumer),
1188 	 * fake one up and assume it is an ASM join.
1189 	 */
1190 	if (imf == NULL) {
1191 		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1192 		imf = &timf;
1193 	}
1194 
1195 	error = in6_mc_get(ifp, mcaddr, &inm);
1196 	if (error) {
1197 		CTR1(KTR_MLD, "%s: in6_mc_get() failure", __func__);
1198 		return (error);
1199 	}
1200 
1201 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1202 	error = in6m_merge(inm, imf);
1203 	if (error) {
1204 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1205 		goto out_in6m_release;
1206 	}
1207 
1208 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1209 	error = mld_change_state(inm, delay);
1210 	if (error) {
1211 		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1212 		goto out_in6m_release;
1213 	}
1214 
1215 out_in6m_release:
1216 	if (error) {
1217 		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1218 		in6m_release_locked(inm);
1219 	} else {
1220 		*pinm = inm;
1221 	}
1222 
1223 	return (error);
1224 }
1225 
1226 /*
1227  * Leave a multicast group; unlocked entry point.
1228  */
1229 int
1230 in6_mc_leave(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1231 {
1232 	struct ifnet *ifp;
1233 	int error;
1234 
1235 	ifp = inm->in6m_ifp;
1236 
1237 	IN6_MULTI_LOCK();
1238 	error = in6_mc_leave_locked(inm, imf);
1239 	IN6_MULTI_UNLOCK();
1240 
1241 	return (error);
1242 }
1243 
1244 /*
1245  * Leave a multicast group; real entry point.
1246  * All source filters will be expunged.
1247  *
1248  * Only preserves atomicity at inm level.
1249  *
1250  * Holding the write lock for the INP which contains imf
1251  * is highly advisable. We can't assert for it as imf does not
1252  * contain a back-pointer to the owning inp.
1253  *
1254  * Note: This is not the same as in6m_release(*) as this function also
1255  * makes a state change downcall into MLD.
1256  */
1257 int
1258 in6_mc_leave_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1259 {
1260 	struct in6_mfilter	 timf;
1261 	int			 error;
1262 #ifdef KTR
1263 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1264 #endif
1265 
1266 	error = 0;
1267 
1268 	IN6_MULTI_LOCK_ASSERT();
1269 
1270 	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1271 	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1272 	    (in6m_is_ifp_detached(inm) ? "null" : inm->in6m_ifp->if_xname),
1273 	    imf);
1274 
1275 	/*
1276 	 * If no imf was specified (i.e. kernel consumer),
1277 	 * fake one up and assume it is an ASM join.
1278 	 */
1279 	if (imf == NULL) {
1280 		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1281 		imf = &timf;
1282 	}
1283 
1284 	/*
1285 	 * Begin state merge transaction at MLD layer.
1286 	 *
1287 	 * As this particular invocation should not cause any memory
1288 	 * to be allocated, and there is no opportunity to roll back
1289 	 * the transaction, it MUST NOT fail.
1290 	 */
1291 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1292 	error = in6m_merge(inm, imf);
1293 	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1294 
1295 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1296 	error = mld_change_state(inm, 0);
1297 	if (error)
1298 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1299 
1300 	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1301 	in6m_release_locked(inm);
1302 
1303 	return (error);
1304 }
1305 
1306 /*
1307  * Block or unblock an ASM multicast source on an inpcb.
1308  * This implements the delta-based API described in RFC 3678.
1309  *
1310  * The delta-based API applies only to exclusive-mode memberships.
1311  * An MLD downcall will be performed.
1312  *
1313  * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1314  *
1315  * Return 0 if successful, otherwise return an appropriate error code.
1316  */
1317 static int
1318 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1319 {
1320 	struct group_source_req		 gsr;
1321 	sockunion_t			*gsa, *ssa;
1322 	struct ifnet			*ifp;
1323 	struct in6_mfilter		*imf;
1324 	struct ip6_moptions		*imo;
1325 	struct in6_msource		*ims;
1326 	struct in6_multi			*inm;
1327 	size_t				 idx;
1328 	uint16_t			 fmode;
1329 	int				 error, doblock;
1330 #ifdef KTR
1331 	char				 ip6tbuf[INET6_ADDRSTRLEN];
1332 #endif
1333 
1334 	ifp = NULL;
1335 	error = 0;
1336 	doblock = 0;
1337 
1338 	memset(&gsr, 0, sizeof(struct group_source_req));
1339 	gsa = (sockunion_t *)&gsr.gsr_group;
1340 	ssa = (sockunion_t *)&gsr.gsr_source;
1341 
1342 	switch (sopt->sopt_name) {
1343 	case MCAST_BLOCK_SOURCE:
1344 	case MCAST_UNBLOCK_SOURCE:
1345 		error = sooptcopyin(sopt, &gsr,
1346 		    sizeof(struct group_source_req),
1347 		    sizeof(struct group_source_req));
1348 		if (error)
1349 			return (error);
1350 
1351 		if (gsa->sin6.sin6_family != AF_INET6 ||
1352 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1353 			return (EINVAL);
1354 
1355 		if (ssa->sin6.sin6_family != AF_INET6 ||
1356 		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1357 			return (EINVAL);
1358 
1359 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1360 			return (EADDRNOTAVAIL);
1361 
1362 		ifp = ifnet_byindex(gsr.gsr_interface);
1363 
1364 		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1365 			doblock = 1;
1366 		break;
1367 
1368 	default:
1369 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1370 		    __func__, sopt->sopt_name);
1371 		return (EOPNOTSUPP);
1372 		break;
1373 	}
1374 
1375 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1376 		return (EINVAL);
1377 
1378 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1379 
1380 	/*
1381 	 * Check if we are actually a member of this group.
1382 	 */
1383 	imo = in6p_findmoptions(inp);
1384 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1385 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1386 		error = EADDRNOTAVAIL;
1387 		goto out_in6p_locked;
1388 	}
1389 
1390 	KASSERT(imo->im6o_mfilters != NULL,
1391 	    ("%s: im6o_mfilters not allocated", __func__));
1392 	imf = &imo->im6o_mfilters[idx];
1393 	inm = imo->im6o_membership[idx];
1394 
1395 	/*
1396 	 * Attempting to use the delta-based API on an
1397 	 * non exclusive-mode membership is an error.
1398 	 */
1399 	fmode = imf->im6f_st[0];
1400 	if (fmode != MCAST_EXCLUDE) {
1401 		error = EINVAL;
1402 		goto out_in6p_locked;
1403 	}
1404 
1405 	/*
1406 	 * Deal with error cases up-front:
1407 	 *  Asked to block, but already blocked; or
1408 	 *  Asked to unblock, but nothing to unblock.
1409 	 * If adding a new block entry, allocate it.
1410 	 */
1411 	ims = im6o_match_source(imo, idx, &ssa->sa);
1412 	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1413 		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1414 		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1415 		    doblock ? "" : "not ");
1416 		error = EADDRNOTAVAIL;
1417 		goto out_in6p_locked;
1418 	}
1419 
1420 	INP_WLOCK_ASSERT(inp);
1421 
1422 	/*
1423 	 * Begin state merge transaction at socket layer.
1424 	 */
1425 	if (doblock) {
1426 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1427 		ims = im6f_graft(imf, fmode, &ssa->sin6);
1428 		if (ims == NULL)
1429 			error = ENOMEM;
1430 	} else {
1431 		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1432 		error = im6f_prune(imf, &ssa->sin6);
1433 	}
1434 
1435 	if (error) {
1436 		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1437 		goto out_im6f_rollback;
1438 	}
1439 
1440 	/*
1441 	 * Begin state merge transaction at MLD layer.
1442 	 */
1443 	IN6_MULTI_LOCK();
1444 
1445 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1446 	error = in6m_merge(inm, imf);
1447 	if (error) {
1448 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1449 		goto out_im6f_rollback;
1450 	}
1451 
1452 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1453 	error = mld_change_state(inm, 0);
1454 	if (error)
1455 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1456 
1457 	IN6_MULTI_UNLOCK();
1458 
1459 out_im6f_rollback:
1460 	if (error)
1461 		im6f_rollback(imf);
1462 	else
1463 		im6f_commit(imf);
1464 
1465 	im6f_reap(imf);
1466 
1467 out_in6p_locked:
1468 	INP_WUNLOCK(inp);
1469 	return (error);
1470 }
1471 
1472 /*
1473  * Given an inpcb, return its multicast options structure pointer.  Accepts
1474  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1475  *
1476  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1477  * SMPng: NOTE: Returns with the INP write lock held.
1478  */
1479 static struct ip6_moptions *
1480 in6p_findmoptions(struct inpcb *inp)
1481 {
1482 	struct ip6_moptions	 *imo;
1483 	struct in6_multi		**immp;
1484 	struct in6_mfilter	 *imfp;
1485 	size_t			  idx;
1486 
1487 	INP_WLOCK(inp);
1488 	if (inp->in6p_moptions != NULL)
1489 		return (inp->in6p_moptions);
1490 
1491 	INP_WUNLOCK(inp);
1492 
1493 	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1494 	immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1495 	    M_WAITOK | M_ZERO);
1496 	imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1497 	    M_IN6MFILTER, M_WAITOK);
1498 
1499 	imo->im6o_multicast_ifp = NULL;
1500 	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1501 	imo->im6o_multicast_loop = in6_mcast_loop;
1502 	imo->im6o_num_memberships = 0;
1503 	imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1504 	imo->im6o_membership = immp;
1505 
1506 	/* Initialize per-group source filters. */
1507 	for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1508 		im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1509 	imo->im6o_mfilters = imfp;
1510 
1511 	INP_WLOCK(inp);
1512 	if (inp->in6p_moptions != NULL) {
1513 		free(imfp, M_IN6MFILTER);
1514 		free(immp, M_IP6MOPTS);
1515 		free(imo, M_IP6MOPTS);
1516 		return (inp->in6p_moptions);
1517 	}
1518 	inp->in6p_moptions = imo;
1519 	return (imo);
1520 }
1521 
1522 /*
1523  * Discard the IPv6 multicast options (and source filters).
1524  *
1525  * SMPng: NOTE: assumes INP write lock is held.
1526  */
1527 void
1528 ip6_freemoptions(struct ip6_moptions *imo)
1529 {
1530 	struct in6_mfilter	*imf;
1531 	size_t			 idx, nmships;
1532 
1533 	KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1534 
1535 	nmships = imo->im6o_num_memberships;
1536 	for (idx = 0; idx < nmships; ++idx) {
1537 		imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1538 		if (imf)
1539 			im6f_leave(imf);
1540 		/* XXX this will thrash the lock(s) */
1541 		(void)in6_mc_leave(imo->im6o_membership[idx], imf);
1542 		if (imf)
1543 			im6f_purge(imf);
1544 	}
1545 
1546 	if (imo->im6o_mfilters)
1547 		free(imo->im6o_mfilters, M_IN6MFILTER);
1548 	free(imo->im6o_membership, M_IP6MOPTS);
1549 	free(imo, M_IP6MOPTS);
1550 }
1551 
1552 /*
1553  * Atomically get source filters on a socket for an IPv6 multicast group.
1554  * Called with INP lock held; returns with lock released.
1555  */
1556 static int
1557 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1558 {
1559 	struct __msfilterreq	 msfr;
1560 	sockunion_t		*gsa;
1561 	struct ifnet		*ifp;
1562 	struct ip6_moptions	*imo;
1563 	struct in6_mfilter	*imf;
1564 	struct ip6_msource	*ims;
1565 	struct in6_msource	*lims;
1566 	struct sockaddr_in6	*psin;
1567 	struct sockaddr_storage	*ptss;
1568 	struct sockaddr_storage	*tss;
1569 	int			 error;
1570 	size_t			 idx, nsrcs, ncsrcs;
1571 
1572 	INP_WLOCK_ASSERT(inp);
1573 
1574 	imo = inp->in6p_moptions;
1575 	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1576 
1577 	INP_WUNLOCK(inp);
1578 
1579 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1580 	    sizeof(struct __msfilterreq));
1581 	if (error)
1582 		return (error);
1583 
1584 	if (msfr.msfr_group.ss_family != AF_INET6 ||
1585 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1586 		return (EINVAL);
1587 
1588 	gsa = (sockunion_t *)&msfr.msfr_group;
1589 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1590 		return (EINVAL);
1591 
1592 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1593 		return (EADDRNOTAVAIL);
1594 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1595 	if (ifp == NULL)
1596 		return (EADDRNOTAVAIL);
1597 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1598 
1599 	INP_WLOCK(inp);
1600 
1601 	/*
1602 	 * Lookup group on the socket.
1603 	 */
1604 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1605 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1606 		INP_WUNLOCK(inp);
1607 		return (EADDRNOTAVAIL);
1608 	}
1609 	imf = &imo->im6o_mfilters[idx];
1610 
1611 	/*
1612 	 * Ignore memberships which are in limbo.
1613 	 */
1614 	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1615 		INP_WUNLOCK(inp);
1616 		return (EAGAIN);
1617 	}
1618 	msfr.msfr_fmode = imf->im6f_st[1];
1619 
1620 	/*
1621 	 * If the user specified a buffer, copy out the source filter
1622 	 * entries to userland gracefully.
1623 	 * We only copy out the number of entries which userland
1624 	 * has asked for, but we always tell userland how big the
1625 	 * buffer really needs to be.
1626 	 */
1627 	tss = NULL;
1628 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1629 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1630 		    M_TEMP, M_NOWAIT | M_ZERO);
1631 		if (tss == NULL) {
1632 			INP_WUNLOCK(inp);
1633 			return (ENOBUFS);
1634 		}
1635 	}
1636 
1637 	/*
1638 	 * Count number of sources in-mode at t0.
1639 	 * If buffer space exists and remains, copy out source entries.
1640 	 */
1641 	nsrcs = msfr.msfr_nsrcs;
1642 	ncsrcs = 0;
1643 	ptss = tss;
1644 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1645 		lims = (struct in6_msource *)ims;
1646 		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1647 		    lims->im6sl_st[0] != imf->im6f_st[0])
1648 			continue;
1649 		++ncsrcs;
1650 		if (tss != NULL && nsrcs > 0) {
1651 			psin = (struct sockaddr_in6 *)ptss;
1652 			psin->sin6_family = AF_INET6;
1653 			psin->sin6_len = sizeof(struct sockaddr_in6);
1654 			psin->sin6_addr = lims->im6s_addr;
1655 			psin->sin6_port = 0;
1656 			--nsrcs;
1657 			++ptss;
1658 		}
1659 	}
1660 
1661 	INP_WUNLOCK(inp);
1662 
1663 	if (tss != NULL) {
1664 		error = copyout(tss, msfr.msfr_srcs,
1665 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1666 		free(tss, M_TEMP);
1667 		if (error)
1668 			return (error);
1669 	}
1670 
1671 	msfr.msfr_nsrcs = ncsrcs;
1672 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1673 
1674 	return (error);
1675 }
1676 
1677 /*
1678  * Return the IP multicast options in response to user getsockopt().
1679  */
1680 int
1681 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1682 {
1683 	struct ip6_moptions	*im6o;
1684 	int			 error;
1685 	u_int			 optval;
1686 
1687 	INP_WLOCK(inp);
1688 	im6o = inp->in6p_moptions;
1689 	/*
1690 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1691 	 * or is a divert socket, reject it.
1692 	 */
1693 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1694 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1695 	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1696 		INP_WUNLOCK(inp);
1697 		return (EOPNOTSUPP);
1698 	}
1699 
1700 	error = 0;
1701 	switch (sopt->sopt_name) {
1702 	case IPV6_MULTICAST_IF:
1703 		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1704 			optval = 0;
1705 		} else {
1706 			optval = im6o->im6o_multicast_ifp->if_index;
1707 		}
1708 		INP_WUNLOCK(inp);
1709 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1710 		break;
1711 
1712 	case IPV6_MULTICAST_HOPS:
1713 		if (im6o == NULL)
1714 			optval = V_ip6_defmcasthlim;
1715 		else
1716 			optval = im6o->im6o_multicast_loop;
1717 		INP_WUNLOCK(inp);
1718 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1719 		break;
1720 
1721 	case IPV6_MULTICAST_LOOP:
1722 		if (im6o == NULL)
1723 			optval = in6_mcast_loop; /* XXX VIMAGE */
1724 		else
1725 			optval = im6o->im6o_multicast_loop;
1726 		INP_WUNLOCK(inp);
1727 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1728 		break;
1729 
1730 	case IPV6_MSFILTER:
1731 		if (im6o == NULL) {
1732 			error = EADDRNOTAVAIL;
1733 			INP_WUNLOCK(inp);
1734 		} else {
1735 			error = in6p_get_source_filters(inp, sopt);
1736 		}
1737 		break;
1738 
1739 	default:
1740 		INP_WUNLOCK(inp);
1741 		error = ENOPROTOOPT;
1742 		break;
1743 	}
1744 
1745 	INP_UNLOCK_ASSERT(inp);
1746 
1747 	return (error);
1748 }
1749 
1750 /*
1751  * Look up the ifnet to use for a multicast group membership,
1752  * given the address of an IPv6 group.
1753  *
1754  * This routine exists to support legacy IPv6 multicast applications.
1755  *
1756  * If inp is non-NULL, use this socket's current FIB number for any
1757  * required FIB lookup. Look up the group address in the unicast FIB,
1758  * and use its ifp; usually, this points to the default next-hop.
1759  * If the FIB lookup fails, return NULL.
1760  *
1761  * FUTURE: Support multiple forwarding tables for IPv6.
1762  *
1763  * Returns NULL if no ifp could be found.
1764  */
1765 static struct ifnet *
1766 in6p_lookup_mcast_ifp(const struct inpcb *in6p __unused,
1767     const struct sockaddr_in6 *gsin6)
1768 {
1769 	struct route_in6	 ro6;
1770 	struct ifnet		*ifp;
1771 
1772 	KASSERT(in6p->inp_vflag & INP_IPV6,
1773 	    ("%s: not INP_IPV6 inpcb", __func__));
1774 	KASSERT(gsin6->sin6_family == AF_INET6,
1775 	    ("%s: not AF_INET6 group", __func__));
1776 	KASSERT(IN6_IS_ADDR_MULTICAST(&gsin6->sin6_addr),
1777 	    ("%s: not multicast", __func__));
1778 
1779 	ifp = NULL;
1780 	memset(&ro6, 0, sizeof(struct route_in6));
1781 	memcpy(&ro6.ro_dst, gsin6, sizeof(struct sockaddr_in6));
1782 #ifdef notyet
1783 	rtalloc_ign_fib(&ro6, 0, inp ? inp->inp_inc.inc_fibnum : 0);
1784 #else
1785 	rtalloc_ign((struct route *)&ro6, 0);
1786 #endif
1787 	if (ro6.ro_rt != NULL) {
1788 		ifp = ro6.ro_rt->rt_ifp;
1789 		KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1790 		RTFREE(ro6.ro_rt);
1791 	}
1792 
1793 	return (ifp);
1794 }
1795 
1796 /*
1797  * Join an IPv6 multicast group, possibly with a source.
1798  *
1799  * FIXME: The KAME use of the unspecified address (::)
1800  * to join *all* multicast groups is currently unsupported.
1801  */
1802 static int
1803 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1804 {
1805 	struct group_source_req		 gsr;
1806 	sockunion_t			*gsa, *ssa;
1807 	struct ifnet			*ifp;
1808 	struct in6_mfilter		*imf;
1809 	struct ip6_moptions		*imo;
1810 	struct in6_multi		*inm;
1811 	struct in6_msource		*lims;
1812 	size_t				 idx;
1813 	int				 error, is_new;
1814 
1815 	ifp = NULL;
1816 	imf = NULL;
1817 	lims = NULL;
1818 	error = 0;
1819 	is_new = 0;
1820 
1821 	memset(&gsr, 0, sizeof(struct group_source_req));
1822 	gsa = (sockunion_t *)&gsr.gsr_group;
1823 	gsa->ss.ss_family = AF_UNSPEC;
1824 	ssa = (sockunion_t *)&gsr.gsr_source;
1825 	ssa->ss.ss_family = AF_UNSPEC;
1826 
1827 	/*
1828 	 * Chew everything into struct group_source_req.
1829 	 * Overwrite the port field if present, as the sockaddr
1830 	 * being copied in may be matched with a binary comparison.
1831 	 * Ignore passed-in scope ID.
1832 	 */
1833 	switch (sopt->sopt_name) {
1834 	case IPV6_JOIN_GROUP: {
1835 		struct ipv6_mreq mreq;
1836 
1837 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1838 		    sizeof(struct ipv6_mreq));
1839 		if (error)
1840 			return (error);
1841 
1842 		gsa->sin6.sin6_family = AF_INET6;
1843 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1844 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1845 
1846 		if (mreq.ipv6mr_interface == 0) {
1847 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1848 		} else {
1849 			if (mreq.ipv6mr_interface < 0 ||
1850 			    V_if_index < mreq.ipv6mr_interface)
1851 				return (EADDRNOTAVAIL);
1852 			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1853 		}
1854 		CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1855 		    __func__, mreq.ipv6mr_interface, ifp);
1856 	} break;
1857 
1858 	case MCAST_JOIN_GROUP:
1859 	case MCAST_JOIN_SOURCE_GROUP:
1860 		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1861 			error = sooptcopyin(sopt, &gsr,
1862 			    sizeof(struct group_req),
1863 			    sizeof(struct group_req));
1864 		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1865 			error = sooptcopyin(sopt, &gsr,
1866 			    sizeof(struct group_source_req),
1867 			    sizeof(struct group_source_req));
1868 		}
1869 		if (error)
1870 			return (error);
1871 
1872 		if (gsa->sin6.sin6_family != AF_INET6 ||
1873 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1874 			return (EINVAL);
1875 
1876 		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1877 			if (ssa->sin6.sin6_family != AF_INET6 ||
1878 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1879 				return (EINVAL);
1880 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1881 				return (EINVAL);
1882 			/*
1883 			 * TODO: Validate embedded scope ID in source
1884 			 * list entry against passed-in ifp, if and only
1885 			 * if source list filter entry is iface or node local.
1886 			 */
1887 			in6_clearscope(&ssa->sin6.sin6_addr);
1888 			ssa->sin6.sin6_port = 0;
1889 			ssa->sin6.sin6_scope_id = 0;
1890 		}
1891 
1892 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1893 			return (EADDRNOTAVAIL);
1894 		ifp = ifnet_byindex(gsr.gsr_interface);
1895 		break;
1896 
1897 	default:
1898 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1899 		    __func__, sopt->sopt_name);
1900 		return (EOPNOTSUPP);
1901 		break;
1902 	}
1903 
1904 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1905 		return (EINVAL);
1906 
1907 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1908 		return (EADDRNOTAVAIL);
1909 
1910 	gsa->sin6.sin6_port = 0;
1911 	gsa->sin6.sin6_scope_id = 0;
1912 
1913 	/*
1914 	 * Always set the scope zone ID on memberships created from userland.
1915 	 * Use the passed-in ifp to do this.
1916 	 * XXX The in6_setscope() return value is meaningless.
1917 	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1918 	 */
1919 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1920 
1921 	imo = in6p_findmoptions(inp);
1922 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1923 	if (idx == -1) {
1924 		is_new = 1;
1925 	} else {
1926 		inm = imo->im6o_membership[idx];
1927 		imf = &imo->im6o_mfilters[idx];
1928 		if (ssa->ss.ss_family != AF_UNSPEC) {
1929 			/*
1930 			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1931 			 * is an error. On an existing inclusive membership,
1932 			 * it just adds the source to the filter list.
1933 			 */
1934 			if (imf->im6f_st[1] != MCAST_INCLUDE) {
1935 				error = EINVAL;
1936 				goto out_in6p_locked;
1937 			}
1938 			/*
1939 			 * Throw out duplicates.
1940 			 *
1941 			 * XXX FIXME: This makes a naive assumption that
1942 			 * even if entries exist for *ssa in this imf,
1943 			 * they will be rejected as dupes, even if they
1944 			 * are not valid in the current mode (in-mode).
1945 			 *
1946 			 * in6_msource is transactioned just as for anything
1947 			 * else in SSM -- but note naive use of in6m_graft()
1948 			 * below for allocating new filter entries.
1949 			 *
1950 			 * This is only an issue if someone mixes the
1951 			 * full-state SSM API with the delta-based API,
1952 			 * which is discouraged in the relevant RFCs.
1953 			 */
1954 			lims = im6o_match_source(imo, idx, &ssa->sa);
1955 			if (lims != NULL /*&&
1956 			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
1957 				error = EADDRNOTAVAIL;
1958 				goto out_in6p_locked;
1959 			}
1960 		} else {
1961 			/*
1962 			 * MCAST_JOIN_GROUP alone, on any existing membership,
1963 			 * is rejected, to stop the same inpcb tying up
1964 			 * multiple refs to the in_multi.
1965 			 * On an existing inclusive membership, this is also
1966 			 * an error; if you want to change filter mode,
1967 			 * you must use the userland API setsourcefilter().
1968 			 * XXX We don't reject this for imf in UNDEFINED
1969 			 * state at t1, because allocation of a filter
1970 			 * is atomic with allocation of a membership.
1971 			 */
1972 			error = EINVAL;
1973 			goto out_in6p_locked;
1974 		}
1975 	}
1976 
1977 	/*
1978 	 * Begin state merge transaction at socket layer.
1979 	 */
1980 	INP_WLOCK_ASSERT(inp);
1981 
1982 	if (is_new) {
1983 		if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
1984 			error = im6o_grow(imo);
1985 			if (error)
1986 				goto out_in6p_locked;
1987 		}
1988 		/*
1989 		 * Allocate the new slot upfront so we can deal with
1990 		 * grafting the new source filter in same code path
1991 		 * as for join-source on existing membership.
1992 		 */
1993 		idx = imo->im6o_num_memberships;
1994 		imo->im6o_membership[idx] = NULL;
1995 		imo->im6o_num_memberships++;
1996 		KASSERT(imo->im6o_mfilters != NULL,
1997 		    ("%s: im6f_mfilters vector was not allocated", __func__));
1998 		imf = &imo->im6o_mfilters[idx];
1999 		KASSERT(RB_EMPTY(&imf->im6f_sources),
2000 		    ("%s: im6f_sources not empty", __func__));
2001 	}
2002 
2003 	/*
2004 	 * Graft new source into filter list for this inpcb's
2005 	 * membership of the group. The in6_multi may not have
2006 	 * been allocated yet if this is a new membership, however,
2007 	 * the in_mfilter slot will be allocated and must be initialized.
2008 	 *
2009 	 * Note: Grafting of exclusive mode filters doesn't happen
2010 	 * in this path.
2011 	 * XXX: Should check for non-NULL lims (node exists but may
2012 	 * not be in-mode) for interop with full-state API.
2013 	 */
2014 	if (ssa->ss.ss_family != AF_UNSPEC) {
2015 		/* Membership starts in IN mode */
2016 		if (is_new) {
2017 			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2018 			im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2019 		} else {
2020 			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2021 		}
2022 		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2023 		if (lims == NULL) {
2024 			CTR1(KTR_MLD, "%s: merge imf state failed",
2025 			    __func__);
2026 			error = ENOMEM;
2027 			goto out_im6o_free;
2028 		}
2029 	} else {
2030 		/* No address specified; Membership starts in EX mode */
2031 		if (is_new) {
2032 			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2033 			im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2034 		}
2035 	}
2036 
2037 	/*
2038 	 * Begin state merge transaction at MLD layer.
2039 	 */
2040 	IN6_MULTI_LOCK();
2041 
2042 	if (is_new) {
2043 		error = in6_mc_join_locked(ifp, &gsa->sin6.sin6_addr, imf,
2044 		    &inm, 0);
2045 		if (error)
2046 			goto out_im6o_free;
2047 		imo->im6o_membership[idx] = inm;
2048 	} else {
2049 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2050 		error = in6m_merge(inm, imf);
2051 		if (error) {
2052 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2053 			    __func__);
2054 			goto out_im6f_rollback;
2055 		}
2056 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2057 		error = mld_change_state(inm, 0);
2058 		if (error) {
2059 			CTR1(KTR_MLD, "%s: failed mld downcall",
2060 			    __func__);
2061 			goto out_im6f_rollback;
2062 		}
2063 	}
2064 
2065 	IN6_MULTI_UNLOCK();
2066 
2067 out_im6f_rollback:
2068 	INP_WLOCK_ASSERT(inp);
2069 	if (error) {
2070 		im6f_rollback(imf);
2071 		if (is_new)
2072 			im6f_purge(imf);
2073 		else
2074 			im6f_reap(imf);
2075 	} else {
2076 		im6f_commit(imf);
2077 	}
2078 
2079 out_im6o_free:
2080 	if (error && is_new) {
2081 		imo->im6o_membership[idx] = NULL;
2082 		--imo->im6o_num_memberships;
2083 	}
2084 
2085 out_in6p_locked:
2086 	INP_WUNLOCK(inp);
2087 	return (error);
2088 }
2089 
2090 /*
2091  * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2092  */
2093 static int
2094 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2095 {
2096 	struct ipv6_mreq		 mreq;
2097 	struct group_source_req		 gsr;
2098 	sockunion_t			*gsa, *ssa;
2099 	struct ifnet			*ifp;
2100 	struct in6_mfilter		*imf;
2101 	struct ip6_moptions		*imo;
2102 	struct in6_msource		*ims;
2103 	struct in6_multi		*inm;
2104 	uint32_t			 ifindex;
2105 	size_t				 idx;
2106 	int				 error, is_final;
2107 #ifdef KTR
2108 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2109 #endif
2110 
2111 	ifp = NULL;
2112 	ifindex = 0;
2113 	error = 0;
2114 	is_final = 1;
2115 
2116 	memset(&gsr, 0, sizeof(struct group_source_req));
2117 	gsa = (sockunion_t *)&gsr.gsr_group;
2118 	gsa->ss.ss_family = AF_UNSPEC;
2119 	ssa = (sockunion_t *)&gsr.gsr_source;
2120 	ssa->ss.ss_family = AF_UNSPEC;
2121 
2122 	/*
2123 	 * Chew everything passed in up into a struct group_source_req
2124 	 * as that is easier to process.
2125 	 * Note: Any embedded scope ID in the multicast group passed
2126 	 * in by userland is ignored, the interface index is the recommended
2127 	 * mechanism to specify an interface; see below.
2128 	 */
2129 	switch (sopt->sopt_name) {
2130 	case IPV6_LEAVE_GROUP:
2131 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2132 		    sizeof(struct ipv6_mreq));
2133 		if (error)
2134 			return (error);
2135 		gsa->sin6.sin6_family = AF_INET6;
2136 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2137 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2138 		gsa->sin6.sin6_port = 0;
2139 		gsa->sin6.sin6_scope_id = 0;
2140 		ifindex = mreq.ipv6mr_interface;
2141 		break;
2142 
2143 	case MCAST_LEAVE_GROUP:
2144 	case MCAST_LEAVE_SOURCE_GROUP:
2145 		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2146 			error = sooptcopyin(sopt, &gsr,
2147 			    sizeof(struct group_req),
2148 			    sizeof(struct group_req));
2149 		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2150 			error = sooptcopyin(sopt, &gsr,
2151 			    sizeof(struct group_source_req),
2152 			    sizeof(struct group_source_req));
2153 		}
2154 		if (error)
2155 			return (error);
2156 
2157 		if (gsa->sin6.sin6_family != AF_INET6 ||
2158 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2159 			return (EINVAL);
2160 		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2161 			if (ssa->sin6.sin6_family != AF_INET6 ||
2162 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2163 				return (EINVAL);
2164 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2165 				return (EINVAL);
2166 			/*
2167 			 * TODO: Validate embedded scope ID in source
2168 			 * list entry against passed-in ifp, if and only
2169 			 * if source list filter entry is iface or node local.
2170 			 */
2171 			in6_clearscope(&ssa->sin6.sin6_addr);
2172 		}
2173 		gsa->sin6.sin6_port = 0;
2174 		gsa->sin6.sin6_scope_id = 0;
2175 		ifindex = gsr.gsr_interface;
2176 		break;
2177 
2178 	default:
2179 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2180 		    __func__, sopt->sopt_name);
2181 		return (EOPNOTSUPP);
2182 		break;
2183 	}
2184 
2185 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2186 		return (EINVAL);
2187 
2188 	/*
2189 	 * Validate interface index if provided. If no interface index
2190 	 * was provided separately, attempt to look the membership up
2191 	 * from the default scope as a last resort to disambiguate
2192 	 * the membership we are being asked to leave.
2193 	 * XXX SCOPE6 lock potentially taken here.
2194 	 */
2195 	if (ifindex != 0) {
2196 		if (ifindex < 0 || V_if_index < ifindex)
2197 			return (EADDRNOTAVAIL);
2198 		ifp = ifnet_byindex(ifindex);
2199 		if (ifp == NULL)
2200 			return (EADDRNOTAVAIL);
2201 		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2202 	} else {
2203 		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2204 		if (error)
2205 			return (EADDRNOTAVAIL);
2206 		/*
2207 		 * Some badly behaved applications don't pass an ifindex
2208 		 * or a scope ID, which is an API violation. In this case,
2209 		 * perform a lookup as per a v6 join.
2210 		 *
2211 		 * XXX For now, stomp on zone ID for the corner case.
2212 		 * This is not the 'KAME way', but we need to see the ifp
2213 		 * directly until such time as this implementation is
2214 		 * refactored, assuming the scope IDs are the way to go.
2215 		 */
2216 		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2217 		if (ifindex == 0) {
2218 			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2219 			    "ifp for group %s.", __func__,
2220 			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2221 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2222 		} else {
2223 			ifp = ifnet_byindex(ifindex);
2224 		}
2225 		if (ifp == NULL)
2226 			return (EADDRNOTAVAIL);
2227 	}
2228 
2229 	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2230 	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2231 
2232 	/*
2233 	 * Find the membership in the membership array.
2234 	 */
2235 	imo = in6p_findmoptions(inp);
2236 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2237 	if (idx == -1) {
2238 		error = EADDRNOTAVAIL;
2239 		goto out_in6p_locked;
2240 	}
2241 	inm = imo->im6o_membership[idx];
2242 	imf = &imo->im6o_mfilters[idx];
2243 
2244 	if (ssa->ss.ss_family != AF_UNSPEC)
2245 		is_final = 0;
2246 
2247 	/*
2248 	 * Begin state merge transaction at socket layer.
2249 	 */
2250 	INP_WLOCK_ASSERT(inp);
2251 
2252 	/*
2253 	 * If we were instructed only to leave a given source, do so.
2254 	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2255 	 */
2256 	if (is_final) {
2257 		im6f_leave(imf);
2258 	} else {
2259 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2260 			error = EADDRNOTAVAIL;
2261 			goto out_in6p_locked;
2262 		}
2263 		ims = im6o_match_source(imo, idx, &ssa->sa);
2264 		if (ims == NULL) {
2265 			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2266 			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2267 			    "not ");
2268 			error = EADDRNOTAVAIL;
2269 			goto out_in6p_locked;
2270 		}
2271 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2272 		error = im6f_prune(imf, &ssa->sin6);
2273 		if (error) {
2274 			CTR1(KTR_MLD, "%s: merge imf state failed",
2275 			    __func__);
2276 			goto out_in6p_locked;
2277 		}
2278 	}
2279 
2280 	/*
2281 	 * Begin state merge transaction at MLD layer.
2282 	 */
2283 	IN6_MULTI_LOCK();
2284 
2285 	if (is_final) {
2286 		/*
2287 		 * Give up the multicast address record to which
2288 		 * the membership points.
2289 		 */
2290 		(void)in6_mc_leave_locked(inm, imf);
2291 	} else {
2292 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2293 		error = in6m_merge(inm, imf);
2294 		if (error) {
2295 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2296 			    __func__);
2297 			goto out_im6f_rollback;
2298 		}
2299 
2300 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2301 		error = mld_change_state(inm, 0);
2302 		if (error) {
2303 			CTR1(KTR_MLD, "%s: failed mld downcall",
2304 			    __func__);
2305 		}
2306 	}
2307 
2308 	IN6_MULTI_UNLOCK();
2309 
2310 out_im6f_rollback:
2311 	if (error)
2312 		im6f_rollback(imf);
2313 	else
2314 		im6f_commit(imf);
2315 
2316 	im6f_reap(imf);
2317 
2318 	if (is_final) {
2319 		/* Remove the gap in the membership array. */
2320 		for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2321 			imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2322 			imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2323 		}
2324 		imo->im6o_num_memberships--;
2325 	}
2326 
2327 out_in6p_locked:
2328 	INP_WUNLOCK(inp);
2329 	return (error);
2330 }
2331 
2332 /*
2333  * Select the interface for transmitting IPv6 multicast datagrams.
2334  *
2335  * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2336  * may be passed to this socket option. An address of in6addr_any or an
2337  * interface index of 0 is used to remove a previous selection.
2338  * When no interface is selected, one is chosen for every send.
2339  */
2340 static int
2341 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2342 {
2343 	struct ifnet		*ifp;
2344 	struct ip6_moptions	*imo;
2345 	u_int			 ifindex;
2346 	int			 error;
2347 
2348 	if (sopt->sopt_valsize != sizeof(u_int))
2349 		return (EINVAL);
2350 
2351 	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2352 	if (error)
2353 		return (error);
2354 	if (ifindex < 0 || V_if_index < ifindex)
2355 		return (EINVAL);
2356 
2357 	ifp = ifnet_byindex(ifindex);
2358 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2359 		return (EADDRNOTAVAIL);
2360 
2361 	imo = in6p_findmoptions(inp);
2362 	imo->im6o_multicast_ifp = ifp;
2363 	INP_WUNLOCK(inp);
2364 
2365 	return (0);
2366 }
2367 
2368 /*
2369  * Atomically set source filters on a socket for an IPv6 multicast group.
2370  *
2371  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2372  */
2373 static int
2374 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2375 {
2376 	struct __msfilterreq	 msfr;
2377 	sockunion_t		*gsa;
2378 	struct ifnet		*ifp;
2379 	struct in6_mfilter	*imf;
2380 	struct ip6_moptions	*imo;
2381 	struct in6_multi		*inm;
2382 	size_t			 idx;
2383 	int			 error;
2384 
2385 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2386 	    sizeof(struct __msfilterreq));
2387 	if (error)
2388 		return (error);
2389 
2390 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2391 		return (ENOBUFS);
2392 
2393 	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2394 	    msfr.msfr_fmode != MCAST_INCLUDE)
2395 		return (EINVAL);
2396 
2397 	if (msfr.msfr_group.ss_family != AF_INET6 ||
2398 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2399 		return (EINVAL);
2400 
2401 	gsa = (sockunion_t *)&msfr.msfr_group;
2402 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2403 		return (EINVAL);
2404 
2405 	gsa->sin6.sin6_port = 0;	/* ignore port */
2406 
2407 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2408 		return (EADDRNOTAVAIL);
2409 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2410 	if (ifp == NULL)
2411 		return (EADDRNOTAVAIL);
2412 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2413 
2414 	/*
2415 	 * Take the INP write lock.
2416 	 * Check if this socket is a member of this group.
2417 	 */
2418 	imo = in6p_findmoptions(inp);
2419 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2420 	if (idx == -1 || imo->im6o_mfilters == NULL) {
2421 		error = EADDRNOTAVAIL;
2422 		goto out_in6p_locked;
2423 	}
2424 	inm = imo->im6o_membership[idx];
2425 	imf = &imo->im6o_mfilters[idx];
2426 
2427 	/*
2428 	 * Begin state merge transaction at socket layer.
2429 	 */
2430 	INP_WLOCK_ASSERT(inp);
2431 
2432 	imf->im6f_st[1] = msfr.msfr_fmode;
2433 
2434 	/*
2435 	 * Apply any new source filters, if present.
2436 	 * Make a copy of the user-space source vector so
2437 	 * that we may copy them with a single copyin. This
2438 	 * allows us to deal with page faults up-front.
2439 	 */
2440 	if (msfr.msfr_nsrcs > 0) {
2441 		struct in6_msource	*lims;
2442 		struct sockaddr_in6	*psin;
2443 		struct sockaddr_storage	*kss, *pkss;
2444 		int			 i;
2445 
2446 		INP_WUNLOCK(inp);
2447 
2448 		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2449 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2450 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2451 		    M_TEMP, M_WAITOK);
2452 		error = copyin(msfr.msfr_srcs, kss,
2453 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2454 		if (error) {
2455 			free(kss, M_TEMP);
2456 			return (error);
2457 		}
2458 
2459 		INP_WLOCK(inp);
2460 
2461 		/*
2462 		 * Mark all source filters as UNDEFINED at t1.
2463 		 * Restore new group filter mode, as im6f_leave()
2464 		 * will set it to INCLUDE.
2465 		 */
2466 		im6f_leave(imf);
2467 		imf->im6f_st[1] = msfr.msfr_fmode;
2468 
2469 		/*
2470 		 * Update socket layer filters at t1, lazy-allocating
2471 		 * new entries. This saves a bunch of memory at the
2472 		 * cost of one RB_FIND() per source entry; duplicate
2473 		 * entries in the msfr_nsrcs vector are ignored.
2474 		 * If we encounter an error, rollback transaction.
2475 		 *
2476 		 * XXX This too could be replaced with a set-symmetric
2477 		 * difference like loop to avoid walking from root
2478 		 * every time, as the key space is common.
2479 		 */
2480 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2481 			psin = (struct sockaddr_in6 *)pkss;
2482 			if (psin->sin6_family != AF_INET6) {
2483 				error = EAFNOSUPPORT;
2484 				break;
2485 			}
2486 			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2487 				error = EINVAL;
2488 				break;
2489 			}
2490 			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2491 				error = EINVAL;
2492 				break;
2493 			}
2494 			/*
2495 			 * TODO: Validate embedded scope ID in source
2496 			 * list entry against passed-in ifp, if and only
2497 			 * if source list filter entry is iface or node local.
2498 			 */
2499 			in6_clearscope(&psin->sin6_addr);
2500 			error = im6f_get_source(imf, psin, &lims);
2501 			if (error)
2502 				break;
2503 			lims->im6sl_st[1] = imf->im6f_st[1];
2504 		}
2505 		free(kss, M_TEMP);
2506 	}
2507 
2508 	if (error)
2509 		goto out_im6f_rollback;
2510 
2511 	INP_WLOCK_ASSERT(inp);
2512 	IN6_MULTI_LOCK();
2513 
2514 	/*
2515 	 * Begin state merge transaction at MLD layer.
2516 	 */
2517 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2518 	error = in6m_merge(inm, imf);
2519 	if (error) {
2520 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2521 		goto out_im6f_rollback;
2522 	}
2523 
2524 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2525 	error = mld_change_state(inm, 0);
2526 	if (error)
2527 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2528 
2529 	IN6_MULTI_UNLOCK();
2530 
2531 out_im6f_rollback:
2532 	if (error)
2533 		im6f_rollback(imf);
2534 	else
2535 		im6f_commit(imf);
2536 
2537 	im6f_reap(imf);
2538 
2539 out_in6p_locked:
2540 	INP_WUNLOCK(inp);
2541 	return (error);
2542 }
2543 
2544 /*
2545  * Set the IP multicast options in response to user setsockopt().
2546  *
2547  * Many of the socket options handled in this function duplicate the
2548  * functionality of socket options in the regular unicast API. However,
2549  * it is not possible to merge the duplicate code, because the idempotence
2550  * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2551  * the effects of these options must be treated as separate and distinct.
2552  *
2553  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2554  */
2555 int
2556 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2557 {
2558 	struct ip6_moptions	*im6o;
2559 	int			 error;
2560 
2561 	error = 0;
2562 
2563 	/*
2564 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2565 	 * or is a divert socket, reject it.
2566 	 */
2567 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2568 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2569 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2570 		return (EOPNOTSUPP);
2571 
2572 	switch (sopt->sopt_name) {
2573 	case IPV6_MULTICAST_IF:
2574 		error = in6p_set_multicast_if(inp, sopt);
2575 		break;
2576 
2577 	case IPV6_MULTICAST_HOPS: {
2578 		int hlim;
2579 
2580 		if (sopt->sopt_valsize != sizeof(int)) {
2581 			error = EINVAL;
2582 			break;
2583 		}
2584 		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2585 		if (error)
2586 			break;
2587 		if (hlim < -1 || hlim > 255) {
2588 			error = EINVAL;
2589 			break;
2590 		} else if (hlim == -1) {
2591 			hlim = V_ip6_defmcasthlim;
2592 		}
2593 		im6o = in6p_findmoptions(inp);
2594 		im6o->im6o_multicast_hlim = hlim;
2595 		INP_WUNLOCK(inp);
2596 		break;
2597 	}
2598 
2599 	case IPV6_MULTICAST_LOOP: {
2600 		u_int loop;
2601 
2602 		/*
2603 		 * Set the loopback flag for outgoing multicast packets.
2604 		 * Must be zero or one.
2605 		 */
2606 		if (sopt->sopt_valsize != sizeof(u_int)) {
2607 			error = EINVAL;
2608 			break;
2609 		}
2610 		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2611 		if (error)
2612 			break;
2613 		if (loop > 1) {
2614 			error = EINVAL;
2615 			break;
2616 		}
2617 		im6o = in6p_findmoptions(inp);
2618 		im6o->im6o_multicast_loop = loop;
2619 		INP_WUNLOCK(inp);
2620 		break;
2621 	}
2622 
2623 	case IPV6_JOIN_GROUP:
2624 	case MCAST_JOIN_GROUP:
2625 	case MCAST_JOIN_SOURCE_GROUP:
2626 		error = in6p_join_group(inp, sopt);
2627 		break;
2628 
2629 	case IPV6_LEAVE_GROUP:
2630 	case MCAST_LEAVE_GROUP:
2631 	case MCAST_LEAVE_SOURCE_GROUP:
2632 		error = in6p_leave_group(inp, sopt);
2633 		break;
2634 
2635 	case MCAST_BLOCK_SOURCE:
2636 	case MCAST_UNBLOCK_SOURCE:
2637 		error = in6p_block_unblock_source(inp, sopt);
2638 		break;
2639 
2640 	case IPV6_MSFILTER:
2641 		error = in6p_set_source_filters(inp, sopt);
2642 		break;
2643 
2644 	default:
2645 		error = EOPNOTSUPP;
2646 		break;
2647 	}
2648 
2649 	INP_UNLOCK_ASSERT(inp);
2650 
2651 	return (error);
2652 }
2653 
2654 /*
2655  * Expose MLD's multicast filter mode and source list(s) to userland,
2656  * keyed by (ifindex, group).
2657  * The filter mode is written out as a uint32_t, followed by
2658  * 0..n of struct in6_addr.
2659  * For use by ifmcstat(8).
2660  * SMPng: NOTE: unlocked read of ifindex space.
2661  */
2662 static int
2663 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2664 {
2665 	struct in6_addr			 mcaddr;
2666 	struct in6_addr			 src;
2667 	struct ifnet			*ifp;
2668 	struct ifmultiaddr		*ifma;
2669 	struct in6_multi		*inm;
2670 	struct ip6_msource		*ims;
2671 	int				*name;
2672 	int				 retval;
2673 	u_int				 namelen;
2674 	uint32_t			 fmode, ifindex;
2675 #ifdef KTR
2676 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2677 #endif
2678 
2679 	name = (int *)arg1;
2680 	namelen = arg2;
2681 
2682 	if (req->newptr != NULL)
2683 		return (EPERM);
2684 
2685 	/* int: ifindex + 4 * 32 bits of IPv6 address */
2686 	if (namelen != 5)
2687 		return (EINVAL);
2688 
2689 	ifindex = name[0];
2690 	if (ifindex <= 0 || ifindex > V_if_index) {
2691 		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2692 		    __func__, ifindex);
2693 		return (ENOENT);
2694 	}
2695 
2696 	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2697 	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2698 		CTR2(KTR_MLD, "%s: group %s is not multicast",
2699 		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2700 		return (EINVAL);
2701 	}
2702 
2703 	ifp = ifnet_byindex(ifindex);
2704 	if (ifp == NULL) {
2705 		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2706 		    __func__, ifindex);
2707 		return (ENOENT);
2708 	}
2709 	/*
2710 	 * Internal MLD lookups require that scope/zone ID is set.
2711 	 */
2712 	(void)in6_setscope(&mcaddr, ifp, NULL);
2713 
2714 	retval = sysctl_wire_old_buffer(req,
2715 	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2716 	if (retval)
2717 		return (retval);
2718 
2719 	IN6_MULTI_LOCK();
2720 
2721 	IF_ADDR_LOCK(ifp);
2722 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2723 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2724 		    ifma->ifma_protospec == NULL)
2725 			continue;
2726 		inm = (struct in6_multi *)ifma->ifma_protospec;
2727 		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2728 			continue;
2729 		fmode = inm->in6m_st[1].iss_fmode;
2730 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2731 		if (retval != 0)
2732 			break;
2733 		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2734 			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2735 			/*
2736 			 * Only copy-out sources which are in-mode.
2737 			 */
2738 			if (fmode != im6s_get_mode(inm, ims, 1)) {
2739 				CTR1(KTR_MLD, "%s: skip non-in-mode",
2740 				    __func__);
2741 				continue;
2742 			}
2743 			src = ims->im6s_addr;
2744 			retval = SYSCTL_OUT(req, &src,
2745 			    sizeof(struct in6_addr));
2746 			if (retval != 0)
2747 				break;
2748 		}
2749 	}
2750 	IF_ADDR_UNLOCK(ifp);
2751 
2752 	IN6_MULTI_UNLOCK();
2753 
2754 	return (retval);
2755 }
2756 
2757 #ifdef KTR
2758 
2759 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2760 
2761 static const char *
2762 in6m_mode_str(const int mode)
2763 {
2764 
2765 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2766 		return (in6m_modestrs[mode]);
2767 	return ("??");
2768 }
2769 
2770 static const char *in6m_statestrs[] = {
2771 	"not-member",
2772 	"silent",
2773 	"idle",
2774 	"lazy",
2775 	"sleeping",
2776 	"awakening",
2777 	"query-pending",
2778 	"sg-query-pending",
2779 	"leaving"
2780 };
2781 
2782 static const char *
2783 in6m_state_str(const int state)
2784 {
2785 
2786 	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2787 		return (in6m_statestrs[state]);
2788 	return ("??");
2789 }
2790 
2791 /*
2792  * Dump an in6_multi structure to the console.
2793  */
2794 void
2795 in6m_print(const struct in6_multi *inm)
2796 {
2797 	int t;
2798 	char ip6tbuf[INET6_ADDRSTRLEN];
2799 
2800 	if ((ktr_mask & KTR_MLD) == 0)
2801 		return;
2802 
2803 	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2804 	printf("addr %s ifp %p(%s) ifma %p\n",
2805 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2806 	    inm->in6m_ifp,
2807 	    inm->in6m_ifp->if_xname,
2808 	    inm->in6m_ifma);
2809 	printf("timer %u state %s refcount %u scq.len %u\n",
2810 	    inm->in6m_timer,
2811 	    in6m_state_str(inm->in6m_state),
2812 	    inm->in6m_refcount,
2813 	    inm->in6m_scq.ifq_len);
2814 	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2815 	    inm->in6m_mli,
2816 	    inm->in6m_nsrc,
2817 	    inm->in6m_sctimer,
2818 	    inm->in6m_scrv);
2819 	for (t = 0; t < 2; t++) {
2820 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2821 		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2822 		    inm->in6m_st[t].iss_asm,
2823 		    inm->in6m_st[t].iss_ex,
2824 		    inm->in6m_st[t].iss_in,
2825 		    inm->in6m_st[t].iss_rec);
2826 	}
2827 	printf("%s: --- end in6m %p ---\n", __func__, inm);
2828 }
2829 
2830 #else /* !KTR */
2831 
2832 void
2833 in6m_print(const struct in6_multi *inm)
2834 {
2835 
2836 }
2837 
2838 #endif /* KTR */
2839