xref: /freebsd/sys/netinet6/in6_mcast.c (revision 38d120bc13ac1de5b739b67b87016b9122149374)
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_var.h>
56 #include <net/if_dl.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_var.h>
62 #include <netinet6/in6_var.h>
63 #include <netinet/ip6.h>
64 #include <netinet/icmp6.h>
65 #include <netinet6/ip6_var.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/tcp_var.h>
68 #include <netinet6/nd6.h>
69 #include <netinet6/mld6_var.h>
70 #include <netinet6/scope6_var.h>
71 
72 #ifndef KTR_MLD
73 #define KTR_MLD KTR_INET6
74 #endif
75 
76 #ifndef __SOCKUNION_DECLARED
77 union sockunion {
78 	struct sockaddr_storage	ss;
79 	struct sockaddr		sa;
80 	struct sockaddr_dl	sdl;
81 	struct sockaddr_in6	sin6;
82 };
83 typedef union sockunion sockunion_t;
84 #define __SOCKUNION_DECLARED
85 #endif /* __SOCKUNION_DECLARED */
86 
87 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
88     "IPv6 multicast PCB-layer source filter");
89 static MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
90 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
91 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
92     "IPv6 multicast MLD-layer source filter");
93 
94 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
95 
96 /*
97  * Locking:
98  * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
99  * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
100  *   it can be taken by code in net/if.c also.
101  * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
102  *
103  * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
104  * any need for in6_multi itself to be virtualized -- it is bound to an ifp
105  * anyway no matter what happens.
106  */
107 struct mtx in6_multi_mtx;
108 MTX_SYSINIT(in6_multi_mtx, &in6_multi_mtx, "in6_multi_mtx", MTX_DEF);
109 
110 static void	im6f_commit(struct in6_mfilter *);
111 static int	im6f_get_source(struct in6_mfilter *imf,
112 		    const struct sockaddr_in6 *psin,
113 		    struct in6_msource **);
114 static struct in6_msource *
115 		im6f_graft(struct in6_mfilter *, const uint8_t,
116 		    const struct sockaddr_in6 *);
117 static void	im6f_leave(struct in6_mfilter *);
118 static int	im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
119 static void	im6f_purge(struct in6_mfilter *);
120 static void	im6f_rollback(struct in6_mfilter *);
121 static void	im6f_reap(struct in6_mfilter *);
122 static int	im6o_grow(struct ip6_moptions *);
123 static size_t	im6o_match_group(const struct ip6_moptions *,
124 		    const struct ifnet *, const struct sockaddr *);
125 static struct in6_msource *
126 		im6o_match_source(const struct ip6_moptions *, const size_t,
127 		    const struct sockaddr *);
128 static void	im6s_merge(struct ip6_msource *ims,
129 		    const struct in6_msource *lims, const int rollback);
130 static int	in6_mc_get(struct ifnet *, const struct in6_addr *,
131 		    struct in6_multi **);
132 static int	in6m_get_source(struct in6_multi *inm,
133 		    const struct in6_addr *addr, const int noalloc,
134 		    struct ip6_msource **pims);
135 #ifdef KTR
136 static int	in6m_is_ifp_detached(const struct in6_multi *);
137 #endif
138 static int	in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
139 static void	in6m_purge(struct in6_multi *);
140 static void	in6m_reap(struct in6_multi *);
141 static struct ip6_moptions *
142 		in6p_findmoptions(struct inpcb *);
143 static int	in6p_get_source_filters(struct inpcb *, struct sockopt *);
144 static int	in6p_join_group(struct inpcb *, struct sockopt *);
145 static int	in6p_leave_group(struct inpcb *, struct sockopt *);
146 static struct ifnet *
147 		in6p_lookup_mcast_ifp(const struct inpcb *,
148 		    const struct sockaddr_in6 *);
149 static int	in6p_block_unblock_source(struct inpcb *, struct sockopt *);
150 static int	in6p_set_multicast_if(struct inpcb *, struct sockopt *);
151 static int	in6p_set_source_filters(struct inpcb *, struct sockopt *);
152 static int	sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
153 
154 SYSCTL_DECL(_net_inet6_ip6);	/* XXX Not in any common header. */
155 
156 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0,
157     "IPv6 multicast");
158 
159 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
160 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
161     CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
162     "Max source filters per group");
163 
164 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
165 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
166     CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
167     "Max source filters per socket");
168 
169 /* TODO Virtualize this switch. */
170 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
171 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
172     &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
173 
174 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
175     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
176     "Per-interface stack-wide source filters");
177 
178 #ifdef KTR
179 /*
180  * Inline function which wraps assertions for a valid ifp.
181  * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
182  * is detached.
183  */
184 static int __inline
185 in6m_is_ifp_detached(const struct in6_multi *inm)
186 {
187 	struct ifnet *ifp;
188 
189 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
190 	ifp = inm->in6m_ifma->ifma_ifp;
191 	if (ifp != NULL) {
192 		/*
193 		 * Sanity check that network-layer notion of ifp is the
194 		 * same as that of link-layer.
195 		 */
196 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
197 	}
198 
199 	return (ifp == NULL);
200 }
201 #endif
202 
203 /*
204  * Initialize an in6_mfilter structure to a known state at t0, t1
205  * with an empty source filter list.
206  */
207 static __inline void
208 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
209 {
210 	memset(imf, 0, sizeof(struct in6_mfilter));
211 	RB_INIT(&imf->im6f_sources);
212 	imf->im6f_st[0] = st0;
213 	imf->im6f_st[1] = st1;
214 }
215 
216 /*
217  * Resize the ip6_moptions vector to the next power-of-two minus 1.
218  * May be called with locks held; do not sleep.
219  */
220 static int
221 im6o_grow(struct ip6_moptions *imo)
222 {
223 	struct in6_multi	**nmships;
224 	struct in6_multi	**omships;
225 	struct in6_mfilter	 *nmfilters;
226 	struct in6_mfilter	 *omfilters;
227 	size_t			  idx;
228 	size_t			  newmax;
229 	size_t			  oldmax;
230 
231 	nmships = NULL;
232 	nmfilters = NULL;
233 	omships = imo->im6o_membership;
234 	omfilters = imo->im6o_mfilters;
235 	oldmax = imo->im6o_max_memberships;
236 	newmax = ((oldmax + 1) * 2) - 1;
237 
238 	if (newmax <= IPV6_MAX_MEMBERSHIPS) {
239 		nmships = (struct in6_multi **)realloc(omships,
240 		    sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
241 		nmfilters = (struct in6_mfilter *)realloc(omfilters,
242 		    sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
243 		    M_NOWAIT);
244 		if (nmships != NULL && nmfilters != NULL) {
245 			/* Initialize newly allocated source filter heads. */
246 			for (idx = oldmax; idx < newmax; idx++) {
247 				im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
248 				    MCAST_EXCLUDE);
249 			}
250 			imo->im6o_max_memberships = newmax;
251 			imo->im6o_membership = nmships;
252 			imo->im6o_mfilters = nmfilters;
253 		}
254 	}
255 
256 	if (nmships == NULL || nmfilters == NULL) {
257 		if (nmships != NULL)
258 			free(nmships, M_IP6MOPTS);
259 		if (nmfilters != NULL)
260 			free(nmfilters, M_IN6MFILTER);
261 		return (ETOOMANYREFS);
262 	}
263 
264 	return (0);
265 }
266 
267 /*
268  * Find an IPv6 multicast group entry for this ip6_moptions instance
269  * which matches the specified group, and optionally an interface.
270  * Return its index into the array, or -1 if not found.
271  */
272 static size_t
273 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
274     const struct sockaddr *group)
275 {
276 	const struct sockaddr_in6 *gsin6;
277 	struct in6_multi	**pinm;
278 	int		  idx;
279 	int		  nmships;
280 
281 	gsin6 = (const struct sockaddr_in6 *)group;
282 
283 	/* The im6o_membership array may be lazy allocated. */
284 	if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
285 		return (-1);
286 
287 	nmships = imo->im6o_num_memberships;
288 	pinm = &imo->im6o_membership[0];
289 	for (idx = 0; idx < nmships; idx++, pinm++) {
290 		if (*pinm == NULL)
291 			continue;
292 		if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
293 		    IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
294 		    &gsin6->sin6_addr)) {
295 			break;
296 		}
297 	}
298 	if (idx >= nmships)
299 		idx = -1;
300 
301 	return (idx);
302 }
303 
304 /*
305  * Find an IPv6 multicast source entry for this imo which matches
306  * the given group index for this socket, and source address.
307  *
308  * XXX TODO: The scope ID, if present in src, is stripped before
309  * any comparison. We SHOULD enforce scope/zone checks where the source
310  * filter entry has a link scope.
311  *
312  * NOTE: This does not check if the entry is in-mode, merely if
313  * it exists, which may not be the desired behaviour.
314  */
315 static struct in6_msource *
316 im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
317     const struct sockaddr *src)
318 {
319 	struct ip6_msource	 find;
320 	struct in6_mfilter	*imf;
321 	struct ip6_msource	*ims;
322 	const sockunion_t	*psa;
323 
324 	KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
325 	KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
326 	    ("%s: invalid index %d\n", __func__, (int)gidx));
327 
328 	/* The im6o_mfilters array may be lazy allocated. */
329 	if (imo->im6o_mfilters == NULL)
330 		return (NULL);
331 	imf = &imo->im6o_mfilters[gidx];
332 
333 	psa = (const sockunion_t *)src;
334 	find.im6s_addr = psa->sin6.sin6_addr;
335 	in6_clearscope(&find.im6s_addr);		/* XXX */
336 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
337 
338 	return ((struct in6_msource *)ims);
339 }
340 
341 /*
342  * Perform filtering for multicast datagrams on a socket by group and source.
343  *
344  * Returns 0 if a datagram should be allowed through, or various error codes
345  * if the socket was not a member of the group, or the source was muted, etc.
346  */
347 int
348 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
349     const struct sockaddr *group, const struct sockaddr *src)
350 {
351 	size_t gidx;
352 	struct in6_msource *ims;
353 	int mode;
354 
355 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
356 
357 	gidx = im6o_match_group(imo, ifp, group);
358 	if (gidx == -1)
359 		return (MCAST_NOTGMEMBER);
360 
361 	/*
362 	 * Check if the source was included in an (S,G) join.
363 	 * Allow reception on exclusive memberships by default,
364 	 * reject reception on inclusive memberships by default.
365 	 * Exclude source only if an in-mode exclude filter exists.
366 	 * Include source only if an in-mode include filter exists.
367 	 * NOTE: We are comparing group state here at MLD t1 (now)
368 	 * with socket-layer t0 (since last downcall).
369 	 */
370 	mode = imo->im6o_mfilters[gidx].im6f_st[1];
371 	ims = im6o_match_source(imo, gidx, src);
372 
373 	if ((ims == NULL && mode == MCAST_INCLUDE) ||
374 	    (ims != NULL && ims->im6sl_st[0] != mode))
375 		return (MCAST_NOTSMEMBER);
376 
377 	return (MCAST_PASS);
378 }
379 
380 /*
381  * Find and return a reference to an in6_multi record for (ifp, group),
382  * and bump its reference count.
383  * If one does not exist, try to allocate it, and update link-layer multicast
384  * filters on ifp to listen for group.
385  * Assumes the IN6_MULTI lock is held across the call.
386  * Return 0 if successful, otherwise return an appropriate error code.
387  */
388 static int
389 in6_mc_get(struct ifnet *ifp, const struct in6_addr *group,
390     struct in6_multi **pinm)
391 {
392 	struct sockaddr_in6	 gsin6;
393 	struct ifmultiaddr	*ifma;
394 	struct in6_multi	*inm;
395 	int			 error;
396 
397 	error = 0;
398 
399 	/*
400 	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
401 	 * if_addmulti() takes this mutex itself, so we must drop and
402 	 * re-acquire around the call.
403 	 */
404 	IN6_MULTI_LOCK_ASSERT();
405 	IF_ADDR_WLOCK(ifp);
406 
407 	inm = in6m_lookup_locked(ifp, group);
408 	if (inm != NULL) {
409 		/*
410 		 * If we already joined this group, just bump the
411 		 * refcount and return it.
412 		 */
413 		KASSERT(inm->in6m_refcount >= 1,
414 		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
415 		++inm->in6m_refcount;
416 		*pinm = inm;
417 		goto out_locked;
418 	}
419 
420 	memset(&gsin6, 0, sizeof(gsin6));
421 	gsin6.sin6_family = AF_INET6;
422 	gsin6.sin6_len = sizeof(struct sockaddr_in6);
423 	gsin6.sin6_addr = *group;
424 
425 	/*
426 	 * Check if a link-layer group is already associated
427 	 * with this network-layer group on the given ifnet.
428 	 */
429 	IF_ADDR_WUNLOCK(ifp);
430 	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
431 	if (error != 0)
432 		return (error);
433 	IF_ADDR_WLOCK(ifp);
434 
435 	/*
436 	 * If something other than netinet6 is occupying the link-layer
437 	 * group, print a meaningful error message and back out of
438 	 * the allocation.
439 	 * Otherwise, bump the refcount on the existing network-layer
440 	 * group association and return it.
441 	 */
442 	if (ifma->ifma_protospec != NULL) {
443 		inm = (struct in6_multi *)ifma->ifma_protospec;
444 #ifdef INVARIANTS
445 		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
446 		    __func__));
447 		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
448 		    ("%s: ifma not AF_INET6", __func__));
449 		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
450 		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
451 		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
452 			panic("%s: ifma %p is inconsistent with %p (%p)",
453 			    __func__, ifma, inm, group);
454 #endif
455 		++inm->in6m_refcount;
456 		*pinm = inm;
457 		goto out_locked;
458 	}
459 
460 	IF_ADDR_WLOCK_ASSERT(ifp);
461 
462 	/*
463 	 * A new in6_multi record is needed; allocate and initialize it.
464 	 * We DO NOT perform an MLD join as the in6_ layer may need to
465 	 * push an initial source list down to MLD to support SSM.
466 	 *
467 	 * The initial source filter state is INCLUDE, {} as per the RFC.
468 	 * Pending state-changes per group are subject to a bounds check.
469 	 */
470 	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
471 	if (inm == NULL) {
472 		if_delmulti_ifma(ifma);
473 		error = ENOMEM;
474 		goto out_locked;
475 	}
476 	inm->in6m_addr = *group;
477 	inm->in6m_ifp = ifp;
478 	inm->in6m_mli = MLD_IFINFO(ifp);
479 	inm->in6m_ifma = ifma;
480 	inm->in6m_refcount = 1;
481 	inm->in6m_state = MLD_NOT_MEMBER;
482 	IFQ_SET_MAXLEN(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
483 
484 	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
485 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
486 	RB_INIT(&inm->in6m_srcs);
487 
488 	ifma->ifma_protospec = inm;
489 	*pinm = inm;
490 
491 out_locked:
492 	IF_ADDR_WUNLOCK(ifp);
493 	return (error);
494 }
495 
496 /*
497  * Drop a reference to an in6_multi record.
498  *
499  * If the refcount drops to 0, free the in6_multi record and
500  * delete the underlying link-layer membership.
501  */
502 void
503 in6m_release_locked(struct in6_multi *inm)
504 {
505 	struct ifmultiaddr *ifma;
506 
507 	IN6_MULTI_LOCK_ASSERT();
508 
509 	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
510 
511 	if (--inm->in6m_refcount > 0) {
512 		CTR2(KTR_MLD, "%s: refcount is now %d", __func__,
513 		    inm->in6m_refcount);
514 		return;
515 	}
516 
517 	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
518 
519 	ifma = inm->in6m_ifma;
520 
521 	/* XXX this access is not covered by IF_ADDR_LOCK */
522 	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
523 	KASSERT(ifma->ifma_protospec == inm,
524 	    ("%s: ifma_protospec != inm", __func__));
525 	ifma->ifma_protospec = NULL;
526 
527 	in6m_purge(inm);
528 
529 	free(inm, M_IP6MADDR);
530 
531 	if_delmulti_ifma(ifma);
532 }
533 
534 /*
535  * Clear recorded source entries for a group.
536  * Used by the MLD code. Caller must hold the IN6_MULTI lock.
537  * FIXME: Should reap.
538  */
539 void
540 in6m_clear_recorded(struct in6_multi *inm)
541 {
542 	struct ip6_msource	*ims;
543 
544 	IN6_MULTI_LOCK_ASSERT();
545 
546 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
547 		if (ims->im6s_stp) {
548 			ims->im6s_stp = 0;
549 			--inm->in6m_st[1].iss_rec;
550 		}
551 	}
552 	KASSERT(inm->in6m_st[1].iss_rec == 0,
553 	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
554 }
555 
556 /*
557  * Record a source as pending for a Source-Group MLDv2 query.
558  * This lives here as it modifies the shared tree.
559  *
560  * inm is the group descriptor.
561  * naddr is the address of the source to record in network-byte order.
562  *
563  * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
564  * lazy-allocate a source node in response to an SG query.
565  * Otherwise, no allocation is performed. This saves some memory
566  * with the trade-off that the source will not be reported to the
567  * router if joined in the window between the query response and
568  * the group actually being joined on the local host.
569  *
570  * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
571  * This turns off the allocation of a recorded source entry if
572  * the group has not been joined.
573  *
574  * Return 0 if the source didn't exist or was already marked as recorded.
575  * Return 1 if the source was marked as recorded by this function.
576  * Return <0 if any error occured (negated errno code).
577  */
578 int
579 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
580 {
581 	struct ip6_msource	 find;
582 	struct ip6_msource	*ims, *nims;
583 
584 	IN6_MULTI_LOCK_ASSERT();
585 
586 	find.im6s_addr = *addr;
587 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
588 	if (ims && ims->im6s_stp)
589 		return (0);
590 	if (ims == NULL) {
591 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
592 			return (-ENOSPC);
593 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
594 		    M_NOWAIT | M_ZERO);
595 		if (nims == NULL)
596 			return (-ENOMEM);
597 		nims->im6s_addr = find.im6s_addr;
598 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
599 		++inm->in6m_nsrc;
600 		ims = nims;
601 	}
602 
603 	/*
604 	 * Mark the source as recorded and update the recorded
605 	 * source count.
606 	 */
607 	++ims->im6s_stp;
608 	++inm->in6m_st[1].iss_rec;
609 
610 	return (1);
611 }
612 
613 /*
614  * Return a pointer to an in6_msource owned by an in6_mfilter,
615  * given its source address.
616  * Lazy-allocate if needed. If this is a new entry its filter state is
617  * undefined at t0.
618  *
619  * imf is the filter set being modified.
620  * addr is the source address.
621  *
622  * SMPng: May be called with locks held; malloc must not block.
623  */
624 static int
625 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
626     struct in6_msource **plims)
627 {
628 	struct ip6_msource	 find;
629 	struct ip6_msource	*ims, *nims;
630 	struct in6_msource	*lims;
631 	int			 error;
632 
633 	error = 0;
634 	ims = NULL;
635 	lims = NULL;
636 
637 	find.im6s_addr = psin->sin6_addr;
638 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
639 	lims = (struct in6_msource *)ims;
640 	if (lims == NULL) {
641 		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
642 			return (ENOSPC);
643 		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
644 		    M_NOWAIT | M_ZERO);
645 		if (nims == NULL)
646 			return (ENOMEM);
647 		lims = (struct in6_msource *)nims;
648 		lims->im6s_addr = find.im6s_addr;
649 		lims->im6sl_st[0] = MCAST_UNDEFINED;
650 		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
651 		++imf->im6f_nsrc;
652 	}
653 
654 	*plims = lims;
655 
656 	return (error);
657 }
658 
659 /*
660  * Graft a source entry into an existing socket-layer filter set,
661  * maintaining any required invariants and checking allocations.
662  *
663  * The source is marked as being in the new filter mode at t1.
664  *
665  * Return the pointer to the new node, otherwise return NULL.
666  */
667 static struct in6_msource *
668 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
669     const struct sockaddr_in6 *psin)
670 {
671 	struct ip6_msource	*nims;
672 	struct in6_msource	*lims;
673 
674 	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
675 	    M_NOWAIT | M_ZERO);
676 	if (nims == NULL)
677 		return (NULL);
678 	lims = (struct in6_msource *)nims;
679 	lims->im6s_addr = psin->sin6_addr;
680 	lims->im6sl_st[0] = MCAST_UNDEFINED;
681 	lims->im6sl_st[1] = st1;
682 	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
683 	++imf->im6f_nsrc;
684 
685 	return (lims);
686 }
687 
688 /*
689  * Prune a source entry from an existing socket-layer filter set,
690  * maintaining any required invariants and checking allocations.
691  *
692  * The source is marked as being left at t1, it is not freed.
693  *
694  * Return 0 if no error occurred, otherwise return an errno value.
695  */
696 static int
697 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
698 {
699 	struct ip6_msource	 find;
700 	struct ip6_msource	*ims;
701 	struct in6_msource	*lims;
702 
703 	find.im6s_addr = psin->sin6_addr;
704 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
705 	if (ims == NULL)
706 		return (ENOENT);
707 	lims = (struct in6_msource *)ims;
708 	lims->im6sl_st[1] = MCAST_UNDEFINED;
709 	return (0);
710 }
711 
712 /*
713  * Revert socket-layer filter set deltas at t1 to t0 state.
714  */
715 static void
716 im6f_rollback(struct in6_mfilter *imf)
717 {
718 	struct ip6_msource	*ims, *tims;
719 	struct in6_msource	*lims;
720 
721 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
722 		lims = (struct in6_msource *)ims;
723 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
724 			/* no change at t1 */
725 			continue;
726 		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
727 			/* revert change to existing source at t1 */
728 			lims->im6sl_st[1] = lims->im6sl_st[0];
729 		} else {
730 			/* revert source added t1 */
731 			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
732 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
733 			free(ims, M_IN6MFILTER);
734 			imf->im6f_nsrc--;
735 		}
736 	}
737 	imf->im6f_st[1] = imf->im6f_st[0];
738 }
739 
740 /*
741  * Mark socket-layer filter set as INCLUDE {} at t1.
742  */
743 static void
744 im6f_leave(struct in6_mfilter *imf)
745 {
746 	struct ip6_msource	*ims;
747 	struct in6_msource	*lims;
748 
749 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
750 		lims = (struct in6_msource *)ims;
751 		lims->im6sl_st[1] = MCAST_UNDEFINED;
752 	}
753 	imf->im6f_st[1] = MCAST_INCLUDE;
754 }
755 
756 /*
757  * Mark socket-layer filter set deltas as committed.
758  */
759 static void
760 im6f_commit(struct in6_mfilter *imf)
761 {
762 	struct ip6_msource	*ims;
763 	struct in6_msource	*lims;
764 
765 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
766 		lims = (struct in6_msource *)ims;
767 		lims->im6sl_st[0] = lims->im6sl_st[1];
768 	}
769 	imf->im6f_st[0] = imf->im6f_st[1];
770 }
771 
772 /*
773  * Reap unreferenced sources from socket-layer filter set.
774  */
775 static void
776 im6f_reap(struct in6_mfilter *imf)
777 {
778 	struct ip6_msource	*ims, *tims;
779 	struct in6_msource	*lims;
780 
781 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
782 		lims = (struct in6_msource *)ims;
783 		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
784 		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
785 			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
786 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
787 			free(ims, M_IN6MFILTER);
788 			imf->im6f_nsrc--;
789 		}
790 	}
791 }
792 
793 /*
794  * Purge socket-layer filter set.
795  */
796 static void
797 im6f_purge(struct in6_mfilter *imf)
798 {
799 	struct ip6_msource	*ims, *tims;
800 
801 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
802 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
803 		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
804 		free(ims, M_IN6MFILTER);
805 		imf->im6f_nsrc--;
806 	}
807 	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
808 	KASSERT(RB_EMPTY(&imf->im6f_sources),
809 	    ("%s: im6f_sources not empty", __func__));
810 }
811 
812 /*
813  * Look up a source filter entry for a multicast group.
814  *
815  * inm is the group descriptor to work with.
816  * addr is the IPv6 address to look up.
817  * noalloc may be non-zero to suppress allocation of sources.
818  * *pims will be set to the address of the retrieved or allocated source.
819  *
820  * SMPng: NOTE: may be called with locks held.
821  * Return 0 if successful, otherwise return a non-zero error code.
822  */
823 static int
824 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
825     const int noalloc, struct ip6_msource **pims)
826 {
827 	struct ip6_msource	 find;
828 	struct ip6_msource	*ims, *nims;
829 #ifdef KTR
830 	char			 ip6tbuf[INET6_ADDRSTRLEN];
831 #endif
832 
833 	find.im6s_addr = *addr;
834 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
835 	if (ims == NULL && !noalloc) {
836 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
837 			return (ENOSPC);
838 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
839 		    M_NOWAIT | M_ZERO);
840 		if (nims == NULL)
841 			return (ENOMEM);
842 		nims->im6s_addr = *addr;
843 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
844 		++inm->in6m_nsrc;
845 		ims = nims;
846 		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
847 		    ip6_sprintf(ip6tbuf, addr), ims);
848 	}
849 
850 	*pims = ims;
851 	return (0);
852 }
853 
854 /*
855  * Merge socket-layer source into MLD-layer source.
856  * If rollback is non-zero, perform the inverse of the merge.
857  */
858 static void
859 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
860     const int rollback)
861 {
862 	int n = rollback ? -1 : 1;
863 #ifdef KTR
864 	char ip6tbuf[INET6_ADDRSTRLEN];
865 
866 	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
867 #endif
868 
869 	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
870 		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
871 		ims->im6s_st[1].ex -= n;
872 	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
873 		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
874 		ims->im6s_st[1].in -= n;
875 	}
876 
877 	if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
878 		CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
879 		ims->im6s_st[1].ex += n;
880 	} else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
881 		CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
882 		ims->im6s_st[1].in += n;
883 	}
884 }
885 
886 /*
887  * Atomically update the global in6_multi state, when a membership's
888  * filter list is being updated in any way.
889  *
890  * imf is the per-inpcb-membership group filter pointer.
891  * A fake imf may be passed for in-kernel consumers.
892  *
893  * XXX This is a candidate for a set-symmetric-difference style loop
894  * which would eliminate the repeated lookup from root of ims nodes,
895  * as they share the same key space.
896  *
897  * If any error occurred this function will back out of refcounts
898  * and return a non-zero value.
899  */
900 static int
901 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
902 {
903 	struct ip6_msource	*ims, *nims;
904 	struct in6_msource	*lims;
905 	int			 schanged, error;
906 	int			 nsrc0, nsrc1;
907 
908 	schanged = 0;
909 	error = 0;
910 	nsrc1 = nsrc0 = 0;
911 
912 	/*
913 	 * Update the source filters first, as this may fail.
914 	 * Maintain count of in-mode filters at t0, t1. These are
915 	 * used to work out if we transition into ASM mode or not.
916 	 * Maintain a count of source filters whose state was
917 	 * actually modified by this operation.
918 	 */
919 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
920 		lims = (struct in6_msource *)ims;
921 		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
922 		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
923 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
924 		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
925 		++schanged;
926 		if (error)
927 			break;
928 		im6s_merge(nims, lims, 0);
929 	}
930 	if (error) {
931 		struct ip6_msource *bims;
932 
933 		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
934 			lims = (struct in6_msource *)ims;
935 			if (lims->im6sl_st[0] == lims->im6sl_st[1])
936 				continue;
937 			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
938 			if (bims == NULL)
939 				continue;
940 			im6s_merge(bims, lims, 1);
941 		}
942 		goto out_reap;
943 	}
944 
945 	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
946 	    __func__, nsrc0, nsrc1);
947 
948 	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
949 	if (imf->im6f_st[0] == imf->im6f_st[1] &&
950 	    imf->im6f_st[1] == MCAST_INCLUDE) {
951 		if (nsrc1 == 0) {
952 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
953 			--inm->in6m_st[1].iss_in;
954 		}
955 	}
956 
957 	/* Handle filter mode transition on socket. */
958 	if (imf->im6f_st[0] != imf->im6f_st[1]) {
959 		CTR3(KTR_MLD, "%s: imf transition %d to %d",
960 		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
961 
962 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
963 			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
964 			--inm->in6m_st[1].iss_ex;
965 		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
966 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
967 			--inm->in6m_st[1].iss_in;
968 		}
969 
970 		if (imf->im6f_st[1] == MCAST_EXCLUDE) {
971 			CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
972 			inm->in6m_st[1].iss_ex++;
973 		} else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
974 			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
975 			inm->in6m_st[1].iss_in++;
976 		}
977 	}
978 
979 	/*
980 	 * Track inm filter state in terms of listener counts.
981 	 * If there are any exclusive listeners, stack-wide
982 	 * membership is exclusive.
983 	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
984 	 * If no listeners remain, state is undefined at t1,
985 	 * and the MLD lifecycle for this group should finish.
986 	 */
987 	if (inm->in6m_st[1].iss_ex > 0) {
988 		CTR1(KTR_MLD, "%s: transition to EX", __func__);
989 		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
990 	} else if (inm->in6m_st[1].iss_in > 0) {
991 		CTR1(KTR_MLD, "%s: transition to IN", __func__);
992 		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
993 	} else {
994 		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
995 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
996 	}
997 
998 	/* Decrement ASM listener count on transition out of ASM mode. */
999 	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1000 		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1001 		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1002 			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1003 			--inm->in6m_st[1].iss_asm;
1004 	}
1005 
1006 	/* Increment ASM listener count on transition to ASM mode. */
1007 	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1008 		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1009 		inm->in6m_st[1].iss_asm++;
1010 	}
1011 
1012 	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1013 	in6m_print(inm);
1014 
1015 out_reap:
1016 	if (schanged > 0) {
1017 		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1018 		in6m_reap(inm);
1019 	}
1020 	return (error);
1021 }
1022 
1023 /*
1024  * Mark an in6_multi's filter set deltas as committed.
1025  * Called by MLD after a state change has been enqueued.
1026  */
1027 void
1028 in6m_commit(struct in6_multi *inm)
1029 {
1030 	struct ip6_msource	*ims;
1031 
1032 	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1033 	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1034 	in6m_print(inm);
1035 
1036 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1037 		ims->im6s_st[0] = ims->im6s_st[1];
1038 	}
1039 	inm->in6m_st[0] = inm->in6m_st[1];
1040 }
1041 
1042 /*
1043  * Reap unreferenced nodes from an in6_multi's filter set.
1044  */
1045 static void
1046 in6m_reap(struct in6_multi *inm)
1047 {
1048 	struct ip6_msource	*ims, *tims;
1049 
1050 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1051 		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1052 		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1053 		    ims->im6s_stp != 0)
1054 			continue;
1055 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1056 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1057 		free(ims, M_IP6MSOURCE);
1058 		inm->in6m_nsrc--;
1059 	}
1060 }
1061 
1062 /*
1063  * Purge all source nodes from an in6_multi's filter set.
1064  */
1065 static void
1066 in6m_purge(struct in6_multi *inm)
1067 {
1068 	struct ip6_msource	*ims, *tims;
1069 
1070 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1071 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1072 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1073 		free(ims, M_IP6MSOURCE);
1074 		inm->in6m_nsrc--;
1075 	}
1076 	/* Free state-change requests that might be queued. */
1077 	_IF_DRAIN(&inm->in6m_scq);
1078 }
1079 
1080 /*
1081  * Join a multicast address w/o sources.
1082  * KAME compatibility entry point.
1083  *
1084  * SMPng: Assume no mc locks held by caller.
1085  */
1086 struct in6_multi_mship *
1087 in6_joingroup(struct ifnet *ifp, struct in6_addr *mcaddr,
1088     int *errorp, int delay)
1089 {
1090 	struct in6_multi_mship *imm;
1091 	int error;
1092 
1093 	imm = malloc(sizeof(*imm), M_IP6MADDR, M_NOWAIT);
1094 	if (imm == NULL) {
1095 		*errorp = ENOBUFS;
1096 		return (NULL);
1097 	}
1098 
1099 	delay = (delay * PR_FASTHZ) / hz;
1100 
1101 	error = in6_mc_join(ifp, mcaddr, NULL, &imm->i6mm_maddr, delay);
1102 	if (error) {
1103 		*errorp = error;
1104 		free(imm, M_IP6MADDR);
1105 		return (NULL);
1106 	}
1107 
1108 	return (imm);
1109 }
1110 
1111 /*
1112  * Leave a multicast address w/o sources.
1113  * KAME compatibility entry point.
1114  *
1115  * SMPng: Assume no mc locks held by caller.
1116  */
1117 int
1118 in6_leavegroup(struct in6_multi_mship *imm)
1119 {
1120 
1121 	if (imm->i6mm_maddr != NULL)
1122 		in6_mc_leave(imm->i6mm_maddr, NULL);
1123 	free(imm,  M_IP6MADDR);
1124 	return 0;
1125 }
1126 
1127 /*
1128  * Join a multicast group; unlocked entry point.
1129  *
1130  * SMPng: XXX: in6_mc_join() is called from in6_control() when upper
1131  * locks are not held. Fortunately, ifp is unlikely to have been detached
1132  * at this point, so we assume it's OK to recurse.
1133  */
1134 int
1135 in6_mc_join(struct ifnet *ifp, const struct in6_addr *mcaddr,
1136     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1137     const int delay)
1138 {
1139 	int error;
1140 
1141 	IN6_MULTI_LOCK();
1142 	error = in6_mc_join_locked(ifp, mcaddr, imf, pinm, delay);
1143 	IN6_MULTI_UNLOCK();
1144 
1145 	return (error);
1146 }
1147 
1148 /*
1149  * Join a multicast group; real entry point.
1150  *
1151  * Only preserves atomicity at inm level.
1152  * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1153  *
1154  * If the MLD downcall fails, the group is not joined, and an error
1155  * code is returned.
1156  */
1157 int
1158 in6_mc_join_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1159     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1160     const int delay)
1161 {
1162 	struct in6_mfilter	 timf;
1163 	struct in6_multi	*inm;
1164 	int			 error;
1165 #ifdef KTR
1166 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1167 #endif
1168 
1169 #ifdef INVARIANTS
1170 	/*
1171 	 * Sanity: Check scope zone ID was set for ifp, if and
1172 	 * only if group is scoped to an interface.
1173 	 */
1174 	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1175 	    ("%s: not a multicast address", __func__));
1176 	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1177 	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1178 		KASSERT(mcaddr->s6_addr16[1] != 0,
1179 		    ("%s: scope zone ID not set", __func__));
1180 	}
1181 #endif
1182 
1183 	IN6_MULTI_LOCK_ASSERT();
1184 
1185 	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1186 	    ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1187 
1188 	error = 0;
1189 	inm = NULL;
1190 
1191 	/*
1192 	 * If no imf was specified (i.e. kernel consumer),
1193 	 * fake one up and assume it is an ASM join.
1194 	 */
1195 	if (imf == NULL) {
1196 		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1197 		imf = &timf;
1198 	}
1199 
1200 	error = in6_mc_get(ifp, mcaddr, &inm);
1201 	if (error) {
1202 		CTR1(KTR_MLD, "%s: in6_mc_get() failure", __func__);
1203 		return (error);
1204 	}
1205 
1206 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1207 	error = in6m_merge(inm, imf);
1208 	if (error) {
1209 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1210 		goto out_in6m_release;
1211 	}
1212 
1213 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1214 	error = mld_change_state(inm, delay);
1215 	if (error) {
1216 		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1217 		goto out_in6m_release;
1218 	}
1219 
1220 out_in6m_release:
1221 	if (error) {
1222 		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1223 		in6m_release_locked(inm);
1224 	} else {
1225 		*pinm = inm;
1226 	}
1227 
1228 	return (error);
1229 }
1230 
1231 /*
1232  * Leave a multicast group; unlocked entry point.
1233  */
1234 int
1235 in6_mc_leave(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1236 {
1237 	struct ifnet *ifp;
1238 	int error;
1239 
1240 	ifp = inm->in6m_ifp;
1241 
1242 	IN6_MULTI_LOCK();
1243 	error = in6_mc_leave_locked(inm, imf);
1244 	IN6_MULTI_UNLOCK();
1245 
1246 	return (error);
1247 }
1248 
1249 /*
1250  * Leave a multicast group; real entry point.
1251  * All source filters will be expunged.
1252  *
1253  * Only preserves atomicity at inm level.
1254  *
1255  * Holding the write lock for the INP which contains imf
1256  * is highly advisable. We can't assert for it as imf does not
1257  * contain a back-pointer to the owning inp.
1258  *
1259  * Note: This is not the same as in6m_release(*) as this function also
1260  * makes a state change downcall into MLD.
1261  */
1262 int
1263 in6_mc_leave_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1264 {
1265 	struct in6_mfilter	 timf;
1266 	int			 error;
1267 #ifdef KTR
1268 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1269 #endif
1270 
1271 	error = 0;
1272 
1273 	IN6_MULTI_LOCK_ASSERT();
1274 
1275 	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1276 	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1277 	    (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1278 	    imf);
1279 
1280 	/*
1281 	 * If no imf was specified (i.e. kernel consumer),
1282 	 * fake one up and assume it is an ASM join.
1283 	 */
1284 	if (imf == NULL) {
1285 		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1286 		imf = &timf;
1287 	}
1288 
1289 	/*
1290 	 * Begin state merge transaction at MLD layer.
1291 	 *
1292 	 * As this particular invocation should not cause any memory
1293 	 * to be allocated, and there is no opportunity to roll back
1294 	 * the transaction, it MUST NOT fail.
1295 	 */
1296 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1297 	error = in6m_merge(inm, imf);
1298 	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1299 
1300 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1301 	error = mld_change_state(inm, 0);
1302 	if (error)
1303 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1304 
1305 	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1306 	in6m_release_locked(inm);
1307 
1308 	return (error);
1309 }
1310 
1311 /*
1312  * Block or unblock an ASM multicast source on an inpcb.
1313  * This implements the delta-based API described in RFC 3678.
1314  *
1315  * The delta-based API applies only to exclusive-mode memberships.
1316  * An MLD downcall will be performed.
1317  *
1318  * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1319  *
1320  * Return 0 if successful, otherwise return an appropriate error code.
1321  */
1322 static int
1323 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1324 {
1325 	struct group_source_req		 gsr;
1326 	sockunion_t			*gsa, *ssa;
1327 	struct ifnet			*ifp;
1328 	struct in6_mfilter		*imf;
1329 	struct ip6_moptions		*imo;
1330 	struct in6_msource		*ims;
1331 	struct in6_multi			*inm;
1332 	size_t				 idx;
1333 	uint16_t			 fmode;
1334 	int				 error, doblock;
1335 #ifdef KTR
1336 	char				 ip6tbuf[INET6_ADDRSTRLEN];
1337 #endif
1338 
1339 	ifp = NULL;
1340 	error = 0;
1341 	doblock = 0;
1342 
1343 	memset(&gsr, 0, sizeof(struct group_source_req));
1344 	gsa = (sockunion_t *)&gsr.gsr_group;
1345 	ssa = (sockunion_t *)&gsr.gsr_source;
1346 
1347 	switch (sopt->sopt_name) {
1348 	case MCAST_BLOCK_SOURCE:
1349 	case MCAST_UNBLOCK_SOURCE:
1350 		error = sooptcopyin(sopt, &gsr,
1351 		    sizeof(struct group_source_req),
1352 		    sizeof(struct group_source_req));
1353 		if (error)
1354 			return (error);
1355 
1356 		if (gsa->sin6.sin6_family != AF_INET6 ||
1357 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1358 			return (EINVAL);
1359 
1360 		if (ssa->sin6.sin6_family != AF_INET6 ||
1361 		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1362 			return (EINVAL);
1363 
1364 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1365 			return (EADDRNOTAVAIL);
1366 
1367 		ifp = ifnet_byindex(gsr.gsr_interface);
1368 
1369 		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1370 			doblock = 1;
1371 		break;
1372 
1373 	default:
1374 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1375 		    __func__, sopt->sopt_name);
1376 		return (EOPNOTSUPP);
1377 		break;
1378 	}
1379 
1380 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1381 		return (EINVAL);
1382 
1383 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1384 
1385 	/*
1386 	 * Check if we are actually a member of this group.
1387 	 */
1388 	imo = in6p_findmoptions(inp);
1389 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1390 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1391 		error = EADDRNOTAVAIL;
1392 		goto out_in6p_locked;
1393 	}
1394 
1395 	KASSERT(imo->im6o_mfilters != NULL,
1396 	    ("%s: im6o_mfilters not allocated", __func__));
1397 	imf = &imo->im6o_mfilters[idx];
1398 	inm = imo->im6o_membership[idx];
1399 
1400 	/*
1401 	 * Attempting to use the delta-based API on an
1402 	 * non exclusive-mode membership is an error.
1403 	 */
1404 	fmode = imf->im6f_st[0];
1405 	if (fmode != MCAST_EXCLUDE) {
1406 		error = EINVAL;
1407 		goto out_in6p_locked;
1408 	}
1409 
1410 	/*
1411 	 * Deal with error cases up-front:
1412 	 *  Asked to block, but already blocked; or
1413 	 *  Asked to unblock, but nothing to unblock.
1414 	 * If adding a new block entry, allocate it.
1415 	 */
1416 	ims = im6o_match_source(imo, idx, &ssa->sa);
1417 	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1418 		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1419 		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1420 		    doblock ? "" : "not ");
1421 		error = EADDRNOTAVAIL;
1422 		goto out_in6p_locked;
1423 	}
1424 
1425 	INP_WLOCK_ASSERT(inp);
1426 
1427 	/*
1428 	 * Begin state merge transaction at socket layer.
1429 	 */
1430 	if (doblock) {
1431 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1432 		ims = im6f_graft(imf, fmode, &ssa->sin6);
1433 		if (ims == NULL)
1434 			error = ENOMEM;
1435 	} else {
1436 		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1437 		error = im6f_prune(imf, &ssa->sin6);
1438 	}
1439 
1440 	if (error) {
1441 		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1442 		goto out_im6f_rollback;
1443 	}
1444 
1445 	/*
1446 	 * Begin state merge transaction at MLD layer.
1447 	 */
1448 	IN6_MULTI_LOCK();
1449 
1450 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1451 	error = in6m_merge(inm, imf);
1452 	if (error)
1453 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1454 	else {
1455 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1456 		error = mld_change_state(inm, 0);
1457 		if (error)
1458 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1459 	}
1460 
1461 	IN6_MULTI_UNLOCK();
1462 
1463 out_im6f_rollback:
1464 	if (error)
1465 		im6f_rollback(imf);
1466 	else
1467 		im6f_commit(imf);
1468 
1469 	im6f_reap(imf);
1470 
1471 out_in6p_locked:
1472 	INP_WUNLOCK(inp);
1473 	return (error);
1474 }
1475 
1476 /*
1477  * Given an inpcb, return its multicast options structure pointer.  Accepts
1478  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1479  *
1480  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1481  * SMPng: NOTE: Returns with the INP write lock held.
1482  */
1483 static struct ip6_moptions *
1484 in6p_findmoptions(struct inpcb *inp)
1485 {
1486 	struct ip6_moptions	 *imo;
1487 	struct in6_multi		**immp;
1488 	struct in6_mfilter	 *imfp;
1489 	size_t			  idx;
1490 
1491 	INP_WLOCK(inp);
1492 	if (inp->in6p_moptions != NULL)
1493 		return (inp->in6p_moptions);
1494 
1495 	INP_WUNLOCK(inp);
1496 
1497 	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1498 	immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1499 	    M_WAITOK | M_ZERO);
1500 	imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1501 	    M_IN6MFILTER, M_WAITOK);
1502 
1503 	imo->im6o_multicast_ifp = NULL;
1504 	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1505 	imo->im6o_multicast_loop = in6_mcast_loop;
1506 	imo->im6o_num_memberships = 0;
1507 	imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1508 	imo->im6o_membership = immp;
1509 
1510 	/* Initialize per-group source filters. */
1511 	for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1512 		im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1513 	imo->im6o_mfilters = imfp;
1514 
1515 	INP_WLOCK(inp);
1516 	if (inp->in6p_moptions != NULL) {
1517 		free(imfp, M_IN6MFILTER);
1518 		free(immp, M_IP6MOPTS);
1519 		free(imo, M_IP6MOPTS);
1520 		return (inp->in6p_moptions);
1521 	}
1522 	inp->in6p_moptions = imo;
1523 	return (imo);
1524 }
1525 
1526 /*
1527  * Discard the IPv6 multicast options (and source filters).
1528  *
1529  * SMPng: NOTE: assumes INP write lock is held.
1530  */
1531 void
1532 ip6_freemoptions(struct ip6_moptions *imo)
1533 {
1534 	struct in6_mfilter	*imf;
1535 	size_t			 idx, nmships;
1536 
1537 	KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1538 
1539 	nmships = imo->im6o_num_memberships;
1540 	for (idx = 0; idx < nmships; ++idx) {
1541 		imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1542 		if (imf)
1543 			im6f_leave(imf);
1544 		/* XXX this will thrash the lock(s) */
1545 		(void)in6_mc_leave(imo->im6o_membership[idx], imf);
1546 		if (imf)
1547 			im6f_purge(imf);
1548 	}
1549 
1550 	if (imo->im6o_mfilters)
1551 		free(imo->im6o_mfilters, M_IN6MFILTER);
1552 	free(imo->im6o_membership, M_IP6MOPTS);
1553 	free(imo, M_IP6MOPTS);
1554 }
1555 
1556 /*
1557  * Atomically get source filters on a socket for an IPv6 multicast group.
1558  * Called with INP lock held; returns with lock released.
1559  */
1560 static int
1561 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1562 {
1563 	struct __msfilterreq	 msfr;
1564 	sockunion_t		*gsa;
1565 	struct ifnet		*ifp;
1566 	struct ip6_moptions	*imo;
1567 	struct in6_mfilter	*imf;
1568 	struct ip6_msource	*ims;
1569 	struct in6_msource	*lims;
1570 	struct sockaddr_in6	*psin;
1571 	struct sockaddr_storage	*ptss;
1572 	struct sockaddr_storage	*tss;
1573 	int			 error;
1574 	size_t			 idx, nsrcs, ncsrcs;
1575 
1576 	INP_WLOCK_ASSERT(inp);
1577 
1578 	imo = inp->in6p_moptions;
1579 	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1580 
1581 	INP_WUNLOCK(inp);
1582 
1583 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1584 	    sizeof(struct __msfilterreq));
1585 	if (error)
1586 		return (error);
1587 
1588 	if (msfr.msfr_group.ss_family != AF_INET6 ||
1589 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1590 		return (EINVAL);
1591 
1592 	gsa = (sockunion_t *)&msfr.msfr_group;
1593 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1594 		return (EINVAL);
1595 
1596 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1597 		return (EADDRNOTAVAIL);
1598 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1599 	if (ifp == NULL)
1600 		return (EADDRNOTAVAIL);
1601 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1602 
1603 	INP_WLOCK(inp);
1604 
1605 	/*
1606 	 * Lookup group on the socket.
1607 	 */
1608 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1609 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1610 		INP_WUNLOCK(inp);
1611 		return (EADDRNOTAVAIL);
1612 	}
1613 	imf = &imo->im6o_mfilters[idx];
1614 
1615 	/*
1616 	 * Ignore memberships which are in limbo.
1617 	 */
1618 	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1619 		INP_WUNLOCK(inp);
1620 		return (EAGAIN);
1621 	}
1622 	msfr.msfr_fmode = imf->im6f_st[1];
1623 
1624 	/*
1625 	 * If the user specified a buffer, copy out the source filter
1626 	 * entries to userland gracefully.
1627 	 * We only copy out the number of entries which userland
1628 	 * has asked for, but we always tell userland how big the
1629 	 * buffer really needs to be.
1630 	 */
1631 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1632 		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1633 	tss = NULL;
1634 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1635 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1636 		    M_TEMP, M_NOWAIT | M_ZERO);
1637 		if (tss == NULL) {
1638 			INP_WUNLOCK(inp);
1639 			return (ENOBUFS);
1640 		}
1641 	}
1642 
1643 	/*
1644 	 * Count number of sources in-mode at t0.
1645 	 * If buffer space exists and remains, copy out source entries.
1646 	 */
1647 	nsrcs = msfr.msfr_nsrcs;
1648 	ncsrcs = 0;
1649 	ptss = tss;
1650 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1651 		lims = (struct in6_msource *)ims;
1652 		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1653 		    lims->im6sl_st[0] != imf->im6f_st[0])
1654 			continue;
1655 		++ncsrcs;
1656 		if (tss != NULL && nsrcs > 0) {
1657 			psin = (struct sockaddr_in6 *)ptss;
1658 			psin->sin6_family = AF_INET6;
1659 			psin->sin6_len = sizeof(struct sockaddr_in6);
1660 			psin->sin6_addr = lims->im6s_addr;
1661 			psin->sin6_port = 0;
1662 			--nsrcs;
1663 			++ptss;
1664 		}
1665 	}
1666 
1667 	INP_WUNLOCK(inp);
1668 
1669 	if (tss != NULL) {
1670 		error = copyout(tss, msfr.msfr_srcs,
1671 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1672 		free(tss, M_TEMP);
1673 		if (error)
1674 			return (error);
1675 	}
1676 
1677 	msfr.msfr_nsrcs = ncsrcs;
1678 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1679 
1680 	return (error);
1681 }
1682 
1683 /*
1684  * Return the IP multicast options in response to user getsockopt().
1685  */
1686 int
1687 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1688 {
1689 	struct ip6_moptions	*im6o;
1690 	int			 error;
1691 	u_int			 optval;
1692 
1693 	INP_WLOCK(inp);
1694 	im6o = inp->in6p_moptions;
1695 	/*
1696 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1697 	 * or is a divert socket, reject it.
1698 	 */
1699 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1700 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1701 	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1702 		INP_WUNLOCK(inp);
1703 		return (EOPNOTSUPP);
1704 	}
1705 
1706 	error = 0;
1707 	switch (sopt->sopt_name) {
1708 	case IPV6_MULTICAST_IF:
1709 		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1710 			optval = 0;
1711 		} else {
1712 			optval = im6o->im6o_multicast_ifp->if_index;
1713 		}
1714 		INP_WUNLOCK(inp);
1715 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1716 		break;
1717 
1718 	case IPV6_MULTICAST_HOPS:
1719 		if (im6o == NULL)
1720 			optval = V_ip6_defmcasthlim;
1721 		else
1722 			optval = im6o->im6o_multicast_hlim;
1723 		INP_WUNLOCK(inp);
1724 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1725 		break;
1726 
1727 	case IPV6_MULTICAST_LOOP:
1728 		if (im6o == NULL)
1729 			optval = in6_mcast_loop; /* XXX VIMAGE */
1730 		else
1731 			optval = im6o->im6o_multicast_loop;
1732 		INP_WUNLOCK(inp);
1733 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1734 		break;
1735 
1736 	case IPV6_MSFILTER:
1737 		if (im6o == NULL) {
1738 			error = EADDRNOTAVAIL;
1739 			INP_WUNLOCK(inp);
1740 		} else {
1741 			error = in6p_get_source_filters(inp, sopt);
1742 		}
1743 		break;
1744 
1745 	default:
1746 		INP_WUNLOCK(inp);
1747 		error = ENOPROTOOPT;
1748 		break;
1749 	}
1750 
1751 	INP_UNLOCK_ASSERT(inp);
1752 
1753 	return (error);
1754 }
1755 
1756 /*
1757  * Look up the ifnet to use for a multicast group membership,
1758  * given the address of an IPv6 group.
1759  *
1760  * This routine exists to support legacy IPv6 multicast applications.
1761  *
1762  * If inp is non-NULL, use this socket's current FIB number for any
1763  * required FIB lookup. Look up the group address in the unicast FIB,
1764  * and use its ifp; usually, this points to the default next-hop.
1765  * If the FIB lookup fails, return NULL.
1766  *
1767  * FUTURE: Support multiple forwarding tables for IPv6.
1768  *
1769  * Returns NULL if no ifp could be found.
1770  */
1771 static struct ifnet *
1772 in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1773     const struct sockaddr_in6 *gsin6)
1774 {
1775 	struct route_in6	 ro6;
1776 	struct ifnet		*ifp;
1777 
1778 	KASSERT(in6p->inp_vflag & INP_IPV6,
1779 	    ("%s: not INP_IPV6 inpcb", __func__));
1780 	KASSERT(gsin6->sin6_family == AF_INET6,
1781 	    ("%s: not AF_INET6 group", __func__));
1782 
1783 	ifp = NULL;
1784 	memset(&ro6, 0, sizeof(struct route_in6));
1785 	memcpy(&ro6.ro_dst, gsin6, sizeof(struct sockaddr_in6));
1786 	rtalloc_ign_fib((struct route *)&ro6, 0,
1787 	    in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB);
1788 	if (ro6.ro_rt != NULL) {
1789 		ifp = ro6.ro_rt->rt_ifp;
1790 		KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1791 		RTFREE(ro6.ro_rt);
1792 	}
1793 
1794 	return (ifp);
1795 }
1796 
1797 /*
1798  * Join an IPv6 multicast group, possibly with a source.
1799  *
1800  * FIXME: The KAME use of the unspecified address (::)
1801  * to join *all* multicast groups is currently unsupported.
1802  */
1803 static int
1804 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1805 {
1806 	struct group_source_req		 gsr;
1807 	sockunion_t			*gsa, *ssa;
1808 	struct ifnet			*ifp;
1809 	struct in6_mfilter		*imf;
1810 	struct ip6_moptions		*imo;
1811 	struct in6_multi		*inm;
1812 	struct in6_msource		*lims;
1813 	size_t				 idx;
1814 	int				 error, is_new;
1815 
1816 	ifp = NULL;
1817 	imf = NULL;
1818 	lims = NULL;
1819 	error = 0;
1820 	is_new = 0;
1821 
1822 	memset(&gsr, 0, sizeof(struct group_source_req));
1823 	gsa = (sockunion_t *)&gsr.gsr_group;
1824 	gsa->ss.ss_family = AF_UNSPEC;
1825 	ssa = (sockunion_t *)&gsr.gsr_source;
1826 	ssa->ss.ss_family = AF_UNSPEC;
1827 
1828 	/*
1829 	 * Chew everything into struct group_source_req.
1830 	 * Overwrite the port field if present, as the sockaddr
1831 	 * being copied in may be matched with a binary comparison.
1832 	 * Ignore passed-in scope ID.
1833 	 */
1834 	switch (sopt->sopt_name) {
1835 	case IPV6_JOIN_GROUP: {
1836 		struct ipv6_mreq mreq;
1837 
1838 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1839 		    sizeof(struct ipv6_mreq));
1840 		if (error)
1841 			return (error);
1842 
1843 		gsa->sin6.sin6_family = AF_INET6;
1844 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1845 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1846 
1847 		if (mreq.ipv6mr_interface == 0) {
1848 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1849 		} else {
1850 			if (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 			IN6_MULTI_UNLOCK();
2047 			goto out_im6o_free;
2048 		}
2049 		imo->im6o_membership[idx] = inm;
2050 	} else {
2051 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2052 		error = in6m_merge(inm, imf);
2053 		if (error)
2054 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2055 			    __func__);
2056 		else {
2057 			CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2058 			error = mld_change_state(inm, 0);
2059 			if (error)
2060 				CTR1(KTR_MLD, "%s: failed mld downcall",
2061 				    __func__);
2062 		}
2063 	}
2064 
2065 	IN6_MULTI_UNLOCK();
2066 	INP_WLOCK_ASSERT(inp);
2067 	if (error) {
2068 		im6f_rollback(imf);
2069 		if (is_new)
2070 			im6f_purge(imf);
2071 		else
2072 			im6f_reap(imf);
2073 	} else {
2074 		im6f_commit(imf);
2075 	}
2076 
2077 out_im6o_free:
2078 	if (error && is_new) {
2079 		imo->im6o_membership[idx] = NULL;
2080 		--imo->im6o_num_memberships;
2081 	}
2082 
2083 out_in6p_locked:
2084 	INP_WUNLOCK(inp);
2085 	return (error);
2086 }
2087 
2088 /*
2089  * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2090  */
2091 static int
2092 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2093 {
2094 	struct ipv6_mreq		 mreq;
2095 	struct group_source_req		 gsr;
2096 	sockunion_t			*gsa, *ssa;
2097 	struct ifnet			*ifp;
2098 	struct in6_mfilter		*imf;
2099 	struct ip6_moptions		*imo;
2100 	struct in6_msource		*ims;
2101 	struct in6_multi		*inm;
2102 	uint32_t			 ifindex;
2103 	size_t				 idx;
2104 	int				 error, is_final;
2105 #ifdef KTR
2106 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2107 #endif
2108 
2109 	ifp = NULL;
2110 	ifindex = 0;
2111 	error = 0;
2112 	is_final = 1;
2113 
2114 	memset(&gsr, 0, sizeof(struct group_source_req));
2115 	gsa = (sockunion_t *)&gsr.gsr_group;
2116 	gsa->ss.ss_family = AF_UNSPEC;
2117 	ssa = (sockunion_t *)&gsr.gsr_source;
2118 	ssa->ss.ss_family = AF_UNSPEC;
2119 
2120 	/*
2121 	 * Chew everything passed in up into a struct group_source_req
2122 	 * as that is easier to process.
2123 	 * Note: Any embedded scope ID in the multicast group passed
2124 	 * in by userland is ignored, the interface index is the recommended
2125 	 * mechanism to specify an interface; see below.
2126 	 */
2127 	switch (sopt->sopt_name) {
2128 	case IPV6_LEAVE_GROUP:
2129 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2130 		    sizeof(struct ipv6_mreq));
2131 		if (error)
2132 			return (error);
2133 		gsa->sin6.sin6_family = AF_INET6;
2134 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2135 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2136 		gsa->sin6.sin6_port = 0;
2137 		gsa->sin6.sin6_scope_id = 0;
2138 		ifindex = mreq.ipv6mr_interface;
2139 		break;
2140 
2141 	case MCAST_LEAVE_GROUP:
2142 	case MCAST_LEAVE_SOURCE_GROUP:
2143 		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2144 			error = sooptcopyin(sopt, &gsr,
2145 			    sizeof(struct group_req),
2146 			    sizeof(struct group_req));
2147 		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2148 			error = sooptcopyin(sopt, &gsr,
2149 			    sizeof(struct group_source_req),
2150 			    sizeof(struct group_source_req));
2151 		}
2152 		if (error)
2153 			return (error);
2154 
2155 		if (gsa->sin6.sin6_family != AF_INET6 ||
2156 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2157 			return (EINVAL);
2158 		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2159 			if (ssa->sin6.sin6_family != AF_INET6 ||
2160 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2161 				return (EINVAL);
2162 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2163 				return (EINVAL);
2164 			/*
2165 			 * TODO: Validate embedded scope ID in source
2166 			 * list entry against passed-in ifp, if and only
2167 			 * if source list filter entry is iface or node local.
2168 			 */
2169 			in6_clearscope(&ssa->sin6.sin6_addr);
2170 		}
2171 		gsa->sin6.sin6_port = 0;
2172 		gsa->sin6.sin6_scope_id = 0;
2173 		ifindex = gsr.gsr_interface;
2174 		break;
2175 
2176 	default:
2177 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2178 		    __func__, sopt->sopt_name);
2179 		return (EOPNOTSUPP);
2180 		break;
2181 	}
2182 
2183 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2184 		return (EINVAL);
2185 
2186 	/*
2187 	 * Validate interface index if provided. If no interface index
2188 	 * was provided separately, attempt to look the membership up
2189 	 * from the default scope as a last resort to disambiguate
2190 	 * the membership we are being asked to leave.
2191 	 * XXX SCOPE6 lock potentially taken here.
2192 	 */
2193 	if (ifindex != 0) {
2194 		if (V_if_index < ifindex)
2195 			return (EADDRNOTAVAIL);
2196 		ifp = ifnet_byindex(ifindex);
2197 		if (ifp == NULL)
2198 			return (EADDRNOTAVAIL);
2199 		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2200 	} else {
2201 		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2202 		if (error)
2203 			return (EADDRNOTAVAIL);
2204 		/*
2205 		 * Some badly behaved applications don't pass an ifindex
2206 		 * or a scope ID, which is an API violation. In this case,
2207 		 * perform a lookup as per a v6 join.
2208 		 *
2209 		 * XXX For now, stomp on zone ID for the corner case.
2210 		 * This is not the 'KAME way', but we need to see the ifp
2211 		 * directly until such time as this implementation is
2212 		 * refactored, assuming the scope IDs are the way to go.
2213 		 */
2214 		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2215 		if (ifindex == 0) {
2216 			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2217 			    "ifp for group %s.", __func__,
2218 			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2219 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2220 		} else {
2221 			ifp = ifnet_byindex(ifindex);
2222 		}
2223 		if (ifp == NULL)
2224 			return (EADDRNOTAVAIL);
2225 	}
2226 
2227 	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2228 	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2229 
2230 	/*
2231 	 * Find the membership in the membership array.
2232 	 */
2233 	imo = in6p_findmoptions(inp);
2234 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2235 	if (idx == -1) {
2236 		error = EADDRNOTAVAIL;
2237 		goto out_in6p_locked;
2238 	}
2239 	inm = imo->im6o_membership[idx];
2240 	imf = &imo->im6o_mfilters[idx];
2241 
2242 	if (ssa->ss.ss_family != AF_UNSPEC)
2243 		is_final = 0;
2244 
2245 	/*
2246 	 * Begin state merge transaction at socket layer.
2247 	 */
2248 	INP_WLOCK_ASSERT(inp);
2249 
2250 	/*
2251 	 * If we were instructed only to leave a given source, do so.
2252 	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2253 	 */
2254 	if (is_final) {
2255 		im6f_leave(imf);
2256 	} else {
2257 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2258 			error = EADDRNOTAVAIL;
2259 			goto out_in6p_locked;
2260 		}
2261 		ims = im6o_match_source(imo, idx, &ssa->sa);
2262 		if (ims == NULL) {
2263 			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2264 			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2265 			    "not ");
2266 			error = EADDRNOTAVAIL;
2267 			goto out_in6p_locked;
2268 		}
2269 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2270 		error = im6f_prune(imf, &ssa->sin6);
2271 		if (error) {
2272 			CTR1(KTR_MLD, "%s: merge imf state failed",
2273 			    __func__);
2274 			goto out_in6p_locked;
2275 		}
2276 	}
2277 
2278 	/*
2279 	 * Begin state merge transaction at MLD layer.
2280 	 */
2281 	IN6_MULTI_LOCK();
2282 
2283 	if (is_final) {
2284 		/*
2285 		 * Give up the multicast address record to which
2286 		 * the membership points.
2287 		 */
2288 		(void)in6_mc_leave_locked(inm, imf);
2289 	} else {
2290 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2291 		error = in6m_merge(inm, imf);
2292 		if (error)
2293 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2294 			    __func__);
2295 		else {
2296 			CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2297 			error = mld_change_state(inm, 0);
2298 			if (error)
2299 				CTR1(KTR_MLD, "%s: failed mld downcall",
2300 				    __func__);
2301 		}
2302 	}
2303 
2304 	IN6_MULTI_UNLOCK();
2305 
2306 	if (error)
2307 		im6f_rollback(imf);
2308 	else
2309 		im6f_commit(imf);
2310 
2311 	im6f_reap(imf);
2312 
2313 	if (is_final) {
2314 		/* Remove the gap in the membership array. */
2315 		for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2316 			imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2317 			imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2318 		}
2319 		imo->im6o_num_memberships--;
2320 	}
2321 
2322 out_in6p_locked:
2323 	INP_WUNLOCK(inp);
2324 	return (error);
2325 }
2326 
2327 /*
2328  * Select the interface for transmitting IPv6 multicast datagrams.
2329  *
2330  * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2331  * may be passed to this socket option. An address of in6addr_any or an
2332  * interface index of 0 is used to remove a previous selection.
2333  * When no interface is selected, one is chosen for every send.
2334  */
2335 static int
2336 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2337 {
2338 	struct ifnet		*ifp;
2339 	struct ip6_moptions	*imo;
2340 	u_int			 ifindex;
2341 	int			 error;
2342 
2343 	if (sopt->sopt_valsize != sizeof(u_int))
2344 		return (EINVAL);
2345 
2346 	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2347 	if (error)
2348 		return (error);
2349 	if (V_if_index < ifindex)
2350 		return (EINVAL);
2351 
2352 	ifp = ifnet_byindex(ifindex);
2353 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2354 		return (EADDRNOTAVAIL);
2355 
2356 	imo = in6p_findmoptions(inp);
2357 	imo->im6o_multicast_ifp = ifp;
2358 	INP_WUNLOCK(inp);
2359 
2360 	return (0);
2361 }
2362 
2363 /*
2364  * Atomically set source filters on a socket for an IPv6 multicast group.
2365  *
2366  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2367  */
2368 static int
2369 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2370 {
2371 	struct __msfilterreq	 msfr;
2372 	sockunion_t		*gsa;
2373 	struct ifnet		*ifp;
2374 	struct in6_mfilter	*imf;
2375 	struct ip6_moptions	*imo;
2376 	struct in6_multi		*inm;
2377 	size_t			 idx;
2378 	int			 error;
2379 
2380 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2381 	    sizeof(struct __msfilterreq));
2382 	if (error)
2383 		return (error);
2384 
2385 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2386 		return (ENOBUFS);
2387 
2388 	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2389 	    msfr.msfr_fmode != MCAST_INCLUDE)
2390 		return (EINVAL);
2391 
2392 	if (msfr.msfr_group.ss_family != AF_INET6 ||
2393 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2394 		return (EINVAL);
2395 
2396 	gsa = (sockunion_t *)&msfr.msfr_group;
2397 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2398 		return (EINVAL);
2399 
2400 	gsa->sin6.sin6_port = 0;	/* ignore port */
2401 
2402 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2403 		return (EADDRNOTAVAIL);
2404 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2405 	if (ifp == NULL)
2406 		return (EADDRNOTAVAIL);
2407 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2408 
2409 	/*
2410 	 * Take the INP write lock.
2411 	 * Check if this socket is a member of this group.
2412 	 */
2413 	imo = in6p_findmoptions(inp);
2414 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2415 	if (idx == -1 || imo->im6o_mfilters == NULL) {
2416 		error = EADDRNOTAVAIL;
2417 		goto out_in6p_locked;
2418 	}
2419 	inm = imo->im6o_membership[idx];
2420 	imf = &imo->im6o_mfilters[idx];
2421 
2422 	/*
2423 	 * Begin state merge transaction at socket layer.
2424 	 */
2425 	INP_WLOCK_ASSERT(inp);
2426 
2427 	imf->im6f_st[1] = msfr.msfr_fmode;
2428 
2429 	/*
2430 	 * Apply any new source filters, if present.
2431 	 * Make a copy of the user-space source vector so
2432 	 * that we may copy them with a single copyin. This
2433 	 * allows us to deal with page faults up-front.
2434 	 */
2435 	if (msfr.msfr_nsrcs > 0) {
2436 		struct in6_msource	*lims;
2437 		struct sockaddr_in6	*psin;
2438 		struct sockaddr_storage	*kss, *pkss;
2439 		int			 i;
2440 
2441 		INP_WUNLOCK(inp);
2442 
2443 		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2444 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2445 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2446 		    M_TEMP, M_WAITOK);
2447 		error = copyin(msfr.msfr_srcs, kss,
2448 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2449 		if (error) {
2450 			free(kss, M_TEMP);
2451 			return (error);
2452 		}
2453 
2454 		INP_WLOCK(inp);
2455 
2456 		/*
2457 		 * Mark all source filters as UNDEFINED at t1.
2458 		 * Restore new group filter mode, as im6f_leave()
2459 		 * will set it to INCLUDE.
2460 		 */
2461 		im6f_leave(imf);
2462 		imf->im6f_st[1] = msfr.msfr_fmode;
2463 
2464 		/*
2465 		 * Update socket layer filters at t1, lazy-allocating
2466 		 * new entries. This saves a bunch of memory at the
2467 		 * cost of one RB_FIND() per source entry; duplicate
2468 		 * entries in the msfr_nsrcs vector are ignored.
2469 		 * If we encounter an error, rollback transaction.
2470 		 *
2471 		 * XXX This too could be replaced with a set-symmetric
2472 		 * difference like loop to avoid walking from root
2473 		 * every time, as the key space is common.
2474 		 */
2475 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2476 			psin = (struct sockaddr_in6 *)pkss;
2477 			if (psin->sin6_family != AF_INET6) {
2478 				error = EAFNOSUPPORT;
2479 				break;
2480 			}
2481 			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2482 				error = EINVAL;
2483 				break;
2484 			}
2485 			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2486 				error = EINVAL;
2487 				break;
2488 			}
2489 			/*
2490 			 * TODO: Validate embedded scope ID in source
2491 			 * list entry against passed-in ifp, if and only
2492 			 * if source list filter entry is iface or node local.
2493 			 */
2494 			in6_clearscope(&psin->sin6_addr);
2495 			error = im6f_get_source(imf, psin, &lims);
2496 			if (error)
2497 				break;
2498 			lims->im6sl_st[1] = imf->im6f_st[1];
2499 		}
2500 		free(kss, M_TEMP);
2501 	}
2502 
2503 	if (error)
2504 		goto out_im6f_rollback;
2505 
2506 	INP_WLOCK_ASSERT(inp);
2507 	IN6_MULTI_LOCK();
2508 
2509 	/*
2510 	 * Begin state merge transaction at MLD layer.
2511 	 */
2512 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2513 	error = in6m_merge(inm, imf);
2514 	if (error)
2515 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2516 	else {
2517 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2518 		error = mld_change_state(inm, 0);
2519 		if (error)
2520 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2521 	}
2522 
2523 	IN6_MULTI_UNLOCK();
2524 
2525 out_im6f_rollback:
2526 	if (error)
2527 		im6f_rollback(imf);
2528 	else
2529 		im6f_commit(imf);
2530 
2531 	im6f_reap(imf);
2532 
2533 out_in6p_locked:
2534 	INP_WUNLOCK(inp);
2535 	return (error);
2536 }
2537 
2538 /*
2539  * Set the IP multicast options in response to user setsockopt().
2540  *
2541  * Many of the socket options handled in this function duplicate the
2542  * functionality of socket options in the regular unicast API. However,
2543  * it is not possible to merge the duplicate code, because the idempotence
2544  * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2545  * the effects of these options must be treated as separate and distinct.
2546  *
2547  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2548  */
2549 int
2550 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2551 {
2552 	struct ip6_moptions	*im6o;
2553 	int			 error;
2554 
2555 	error = 0;
2556 
2557 	/*
2558 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2559 	 * or is a divert socket, reject it.
2560 	 */
2561 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2562 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2563 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2564 		return (EOPNOTSUPP);
2565 
2566 	switch (sopt->sopt_name) {
2567 	case IPV6_MULTICAST_IF:
2568 		error = in6p_set_multicast_if(inp, sopt);
2569 		break;
2570 
2571 	case IPV6_MULTICAST_HOPS: {
2572 		int hlim;
2573 
2574 		if (sopt->sopt_valsize != sizeof(int)) {
2575 			error = EINVAL;
2576 			break;
2577 		}
2578 		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2579 		if (error)
2580 			break;
2581 		if (hlim < -1 || hlim > 255) {
2582 			error = EINVAL;
2583 			break;
2584 		} else if (hlim == -1) {
2585 			hlim = V_ip6_defmcasthlim;
2586 		}
2587 		im6o = in6p_findmoptions(inp);
2588 		im6o->im6o_multicast_hlim = hlim;
2589 		INP_WUNLOCK(inp);
2590 		break;
2591 	}
2592 
2593 	case IPV6_MULTICAST_LOOP: {
2594 		u_int loop;
2595 
2596 		/*
2597 		 * Set the loopback flag for outgoing multicast packets.
2598 		 * Must be zero or one.
2599 		 */
2600 		if (sopt->sopt_valsize != sizeof(u_int)) {
2601 			error = EINVAL;
2602 			break;
2603 		}
2604 		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2605 		if (error)
2606 			break;
2607 		if (loop > 1) {
2608 			error = EINVAL;
2609 			break;
2610 		}
2611 		im6o = in6p_findmoptions(inp);
2612 		im6o->im6o_multicast_loop = loop;
2613 		INP_WUNLOCK(inp);
2614 		break;
2615 	}
2616 
2617 	case IPV6_JOIN_GROUP:
2618 	case MCAST_JOIN_GROUP:
2619 	case MCAST_JOIN_SOURCE_GROUP:
2620 		error = in6p_join_group(inp, sopt);
2621 		break;
2622 
2623 	case IPV6_LEAVE_GROUP:
2624 	case MCAST_LEAVE_GROUP:
2625 	case MCAST_LEAVE_SOURCE_GROUP:
2626 		error = in6p_leave_group(inp, sopt);
2627 		break;
2628 
2629 	case MCAST_BLOCK_SOURCE:
2630 	case MCAST_UNBLOCK_SOURCE:
2631 		error = in6p_block_unblock_source(inp, sopt);
2632 		break;
2633 
2634 	case IPV6_MSFILTER:
2635 		error = in6p_set_source_filters(inp, sopt);
2636 		break;
2637 
2638 	default:
2639 		error = EOPNOTSUPP;
2640 		break;
2641 	}
2642 
2643 	INP_UNLOCK_ASSERT(inp);
2644 
2645 	return (error);
2646 }
2647 
2648 /*
2649  * Expose MLD's multicast filter mode and source list(s) to userland,
2650  * keyed by (ifindex, group).
2651  * The filter mode is written out as a uint32_t, followed by
2652  * 0..n of struct in6_addr.
2653  * For use by ifmcstat(8).
2654  * SMPng: NOTE: unlocked read of ifindex space.
2655  */
2656 static int
2657 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2658 {
2659 	struct in6_addr			 mcaddr;
2660 	struct in6_addr			 src;
2661 	struct ifnet			*ifp;
2662 	struct ifmultiaddr		*ifma;
2663 	struct in6_multi		*inm;
2664 	struct ip6_msource		*ims;
2665 	int				*name;
2666 	int				 retval;
2667 	u_int				 namelen;
2668 	uint32_t			 fmode, ifindex;
2669 #ifdef KTR
2670 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2671 #endif
2672 
2673 	name = (int *)arg1;
2674 	namelen = arg2;
2675 
2676 	if (req->newptr != NULL)
2677 		return (EPERM);
2678 
2679 	/* int: ifindex + 4 * 32 bits of IPv6 address */
2680 	if (namelen != 5)
2681 		return (EINVAL);
2682 
2683 	ifindex = name[0];
2684 	if (ifindex <= 0 || ifindex > V_if_index) {
2685 		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2686 		    __func__, ifindex);
2687 		return (ENOENT);
2688 	}
2689 
2690 	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2691 	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2692 		CTR2(KTR_MLD, "%s: group %s is not multicast",
2693 		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2694 		return (EINVAL);
2695 	}
2696 
2697 	ifp = ifnet_byindex(ifindex);
2698 	if (ifp == NULL) {
2699 		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2700 		    __func__, ifindex);
2701 		return (ENOENT);
2702 	}
2703 	/*
2704 	 * Internal MLD lookups require that scope/zone ID is set.
2705 	 */
2706 	(void)in6_setscope(&mcaddr, ifp, NULL);
2707 
2708 	retval = sysctl_wire_old_buffer(req,
2709 	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2710 	if (retval)
2711 		return (retval);
2712 
2713 	IN6_MULTI_LOCK();
2714 
2715 	IF_ADDR_RLOCK(ifp);
2716 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2717 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2718 		    ifma->ifma_protospec == NULL)
2719 			continue;
2720 		inm = (struct in6_multi *)ifma->ifma_protospec;
2721 		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2722 			continue;
2723 		fmode = inm->in6m_st[1].iss_fmode;
2724 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2725 		if (retval != 0)
2726 			break;
2727 		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2728 			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2729 			/*
2730 			 * Only copy-out sources which are in-mode.
2731 			 */
2732 			if (fmode != im6s_get_mode(inm, ims, 1)) {
2733 				CTR1(KTR_MLD, "%s: skip non-in-mode",
2734 				    __func__);
2735 				continue;
2736 			}
2737 			src = ims->im6s_addr;
2738 			retval = SYSCTL_OUT(req, &src,
2739 			    sizeof(struct in6_addr));
2740 			if (retval != 0)
2741 				break;
2742 		}
2743 	}
2744 	IF_ADDR_RUNLOCK(ifp);
2745 
2746 	IN6_MULTI_UNLOCK();
2747 
2748 	return (retval);
2749 }
2750 
2751 #ifdef KTR
2752 
2753 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2754 
2755 static const char *
2756 in6m_mode_str(const int mode)
2757 {
2758 
2759 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2760 		return (in6m_modestrs[mode]);
2761 	return ("??");
2762 }
2763 
2764 static const char *in6m_statestrs[] = {
2765 	"not-member",
2766 	"silent",
2767 	"idle",
2768 	"lazy",
2769 	"sleeping",
2770 	"awakening",
2771 	"query-pending",
2772 	"sg-query-pending",
2773 	"leaving"
2774 };
2775 
2776 static const char *
2777 in6m_state_str(const int state)
2778 {
2779 
2780 	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2781 		return (in6m_statestrs[state]);
2782 	return ("??");
2783 }
2784 
2785 /*
2786  * Dump an in6_multi structure to the console.
2787  */
2788 void
2789 in6m_print(const struct in6_multi *inm)
2790 {
2791 	int t;
2792 	char ip6tbuf[INET6_ADDRSTRLEN];
2793 
2794 	if ((ktr_mask & KTR_MLD) == 0)
2795 		return;
2796 
2797 	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2798 	printf("addr %s ifp %p(%s) ifma %p\n",
2799 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2800 	    inm->in6m_ifp,
2801 	    if_name(inm->in6m_ifp),
2802 	    inm->in6m_ifma);
2803 	printf("timer %u state %s refcount %u scq.len %u\n",
2804 	    inm->in6m_timer,
2805 	    in6m_state_str(inm->in6m_state),
2806 	    inm->in6m_refcount,
2807 	    inm->in6m_scq.ifq_len);
2808 	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2809 	    inm->in6m_mli,
2810 	    inm->in6m_nsrc,
2811 	    inm->in6m_sctimer,
2812 	    inm->in6m_scrv);
2813 	for (t = 0; t < 2; t++) {
2814 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2815 		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2816 		    inm->in6m_st[t].iss_asm,
2817 		    inm->in6m_st[t].iss_ex,
2818 		    inm->in6m_st[t].iss_in,
2819 		    inm->in6m_st[t].iss_rec);
2820 	}
2821 	printf("%s: --- end in6m %p ---\n", __func__, inm);
2822 }
2823 
2824 #else /* !KTR */
2825 
2826 void
2827 in6m_print(const struct in6_multi *inm)
2828 {
2829 
2830 }
2831 
2832 #endif /* KTR */
2833