xref: /freebsd/sys/netinet6/in6_mcast.c (revision 09a53ad8f1318c5daae6cfb19d97f4f6459f0013)
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_fib.h>
63 #include <netinet6/in6_var.h>
64 #include <netinet/ip6.h>
65 #include <netinet/icmp6.h>
66 #include <netinet6/ip6_var.h>
67 #include <netinet/in_pcb.h>
68 #include <netinet/tcp_var.h>
69 #include <netinet6/nd6.h>
70 #include <netinet6/mld6_var.h>
71 #include <netinet6/scope6_var.h>
72 
73 #ifndef KTR_MLD
74 #define KTR_MLD KTR_INET6
75 #endif
76 
77 #ifndef __SOCKUNION_DECLARED
78 union sockunion {
79 	struct sockaddr_storage	ss;
80 	struct sockaddr		sa;
81 	struct sockaddr_dl	sdl;
82 	struct sockaddr_in6	sin6;
83 };
84 typedef union sockunion sockunion_t;
85 #define __SOCKUNION_DECLARED
86 #endif /* __SOCKUNION_DECLARED */
87 
88 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
89     "IPv6 multicast PCB-layer source filter");
90 static MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
91 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
92 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
93     "IPv6 multicast MLD-layer source filter");
94 
95 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
96 
97 /*
98  * Locking:
99  * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
100  * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
101  *   it can be taken by code in net/if.c also.
102  * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
103  *
104  * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
105  * any need for in6_multi itself to be virtualized -- it is bound to an ifp
106  * anyway no matter what happens.
107  */
108 struct mtx in6_multi_mtx;
109 MTX_SYSINIT(in6_multi_mtx, &in6_multi_mtx, "in6_multi_mtx", MTX_DEF);
110 
111 static void	im6f_commit(struct in6_mfilter *);
112 static int	im6f_get_source(struct in6_mfilter *imf,
113 		    const struct sockaddr_in6 *psin,
114 		    struct in6_msource **);
115 static struct in6_msource *
116 		im6f_graft(struct in6_mfilter *, const uint8_t,
117 		    const struct sockaddr_in6 *);
118 static void	im6f_leave(struct in6_mfilter *);
119 static int	im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
120 static void	im6f_purge(struct in6_mfilter *);
121 static void	im6f_rollback(struct in6_mfilter *);
122 static void	im6f_reap(struct in6_mfilter *);
123 static int	im6o_grow(struct ip6_moptions *);
124 static size_t	im6o_match_group(const struct ip6_moptions *,
125 		    const struct ifnet *, const struct sockaddr *);
126 static struct in6_msource *
127 		im6o_match_source(const struct ip6_moptions *, const size_t,
128 		    const struct sockaddr *);
129 static void	im6s_merge(struct ip6_msource *ims,
130 		    const struct in6_msource *lims, const int rollback);
131 static int	in6_mc_get(struct ifnet *, const struct in6_addr *,
132 		    struct in6_multi **);
133 static int	in6m_get_source(struct in6_multi *inm,
134 		    const struct in6_addr *addr, const int noalloc,
135 		    struct ip6_msource **pims);
136 #ifdef KTR
137 static int	in6m_is_ifp_detached(const struct in6_multi *);
138 #endif
139 static int	in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
140 static void	in6m_purge(struct in6_multi *);
141 static void	in6m_reap(struct in6_multi *);
142 static struct ip6_moptions *
143 		in6p_findmoptions(struct inpcb *);
144 static int	in6p_get_source_filters(struct inpcb *, struct sockopt *);
145 static int	in6p_join_group(struct inpcb *, struct sockopt *);
146 static int	in6p_leave_group(struct inpcb *, struct sockopt *);
147 static struct ifnet *
148 		in6p_lookup_mcast_ifp(const struct inpcb *,
149 		    const struct sockaddr_in6 *);
150 static int	in6p_block_unblock_source(struct inpcb *, struct sockopt *);
151 static int	in6p_set_multicast_if(struct inpcb *, struct sockopt *);
152 static int	in6p_set_source_filters(struct inpcb *, struct sockopt *);
153 static int	sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
154 
155 SYSCTL_DECL(_net_inet6_ip6);	/* XXX Not in any common header. */
156 
157 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0,
158     "IPv6 multicast");
159 
160 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
161 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
162     CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
163     "Max source filters per group");
164 
165 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
166 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
167     CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
168     "Max source filters per socket");
169 
170 /* TODO Virtualize this switch. */
171 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
172 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
173     &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
174 
175 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
176     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
177     "Per-interface stack-wide source filters");
178 
179 #ifdef KTR
180 /*
181  * Inline function which wraps assertions for a valid ifp.
182  * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
183  * is detached.
184  */
185 static int __inline
186 in6m_is_ifp_detached(const struct in6_multi *inm)
187 {
188 	struct ifnet *ifp;
189 
190 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
191 	ifp = inm->in6m_ifma->ifma_ifp;
192 	if (ifp != NULL) {
193 		/*
194 		 * Sanity check that network-layer notion of ifp is the
195 		 * same as that of link-layer.
196 		 */
197 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
198 	}
199 
200 	return (ifp == NULL);
201 }
202 #endif
203 
204 /*
205  * Initialize an in6_mfilter structure to a known state at t0, t1
206  * with an empty source filter list.
207  */
208 static __inline void
209 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
210 {
211 	memset(imf, 0, sizeof(struct in6_mfilter));
212 	RB_INIT(&imf->im6f_sources);
213 	imf->im6f_st[0] = st0;
214 	imf->im6f_st[1] = st1;
215 }
216 
217 /*
218  * Resize the ip6_moptions vector to the next power-of-two minus 1.
219  * May be called with locks held; do not sleep.
220  */
221 static int
222 im6o_grow(struct ip6_moptions *imo)
223 {
224 	struct in6_multi	**nmships;
225 	struct in6_multi	**omships;
226 	struct in6_mfilter	 *nmfilters;
227 	struct in6_mfilter	 *omfilters;
228 	size_t			  idx;
229 	size_t			  newmax;
230 	size_t			  oldmax;
231 
232 	nmships = NULL;
233 	nmfilters = NULL;
234 	omships = imo->im6o_membership;
235 	omfilters = imo->im6o_mfilters;
236 	oldmax = imo->im6o_max_memberships;
237 	newmax = ((oldmax + 1) * 2) - 1;
238 
239 	if (newmax <= IPV6_MAX_MEMBERSHIPS) {
240 		nmships = (struct in6_multi **)realloc(omships,
241 		    sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
242 		nmfilters = (struct in6_mfilter *)realloc(omfilters,
243 		    sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
244 		    M_NOWAIT);
245 		if (nmships != NULL && nmfilters != NULL) {
246 			/* Initialize newly allocated source filter heads. */
247 			for (idx = oldmax; idx < newmax; idx++) {
248 				im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
249 				    MCAST_EXCLUDE);
250 			}
251 			imo->im6o_max_memberships = newmax;
252 			imo->im6o_membership = nmships;
253 			imo->im6o_mfilters = nmfilters;
254 		}
255 	}
256 
257 	if (nmships == NULL || nmfilters == NULL) {
258 		if (nmships != NULL)
259 			free(nmships, M_IP6MOPTS);
260 		if (nmfilters != NULL)
261 			free(nmfilters, M_IN6MFILTER);
262 		return (ETOOMANYREFS);
263 	}
264 
265 	return (0);
266 }
267 
268 /*
269  * Find an IPv6 multicast group entry for this ip6_moptions instance
270  * which matches the specified group, and optionally an interface.
271  * Return its index into the array, or -1 if not found.
272  */
273 static size_t
274 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
275     const struct sockaddr *group)
276 {
277 	const struct sockaddr_in6 *gsin6;
278 	struct in6_multi	**pinm;
279 	int		  idx;
280 	int		  nmships;
281 
282 	gsin6 = (const struct sockaddr_in6 *)group;
283 
284 	/* The im6o_membership array may be lazy allocated. */
285 	if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
286 		return (-1);
287 
288 	nmships = imo->im6o_num_memberships;
289 	pinm = &imo->im6o_membership[0];
290 	for (idx = 0; idx < nmships; idx++, pinm++) {
291 		if (*pinm == NULL)
292 			continue;
293 		if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
294 		    IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
295 		    &gsin6->sin6_addr)) {
296 			break;
297 		}
298 	}
299 	if (idx >= nmships)
300 		idx = -1;
301 
302 	return (idx);
303 }
304 
305 /*
306  * Find an IPv6 multicast source entry for this imo which matches
307  * the given group index for this socket, and source address.
308  *
309  * XXX TODO: The scope ID, if present in src, is stripped before
310  * any comparison. We SHOULD enforce scope/zone checks where the source
311  * filter entry has a link scope.
312  *
313  * NOTE: This does not check if the entry is in-mode, merely if
314  * it exists, which may not be the desired behaviour.
315  */
316 static struct in6_msource *
317 im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
318     const struct sockaddr *src)
319 {
320 	struct ip6_msource	 find;
321 	struct in6_mfilter	*imf;
322 	struct ip6_msource	*ims;
323 	const sockunion_t	*psa;
324 
325 	KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
326 	KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
327 	    ("%s: invalid index %d\n", __func__, (int)gidx));
328 
329 	/* The im6o_mfilters array may be lazy allocated. */
330 	if (imo->im6o_mfilters == NULL)
331 		return (NULL);
332 	imf = &imo->im6o_mfilters[gidx];
333 
334 	psa = (const sockunion_t *)src;
335 	find.im6s_addr = psa->sin6.sin6_addr;
336 	in6_clearscope(&find.im6s_addr);		/* XXX */
337 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
338 
339 	return ((struct in6_msource *)ims);
340 }
341 
342 /*
343  * Perform filtering for multicast datagrams on a socket by group and source.
344  *
345  * Returns 0 if a datagram should be allowed through, or various error codes
346  * if the socket was not a member of the group, or the source was muted, etc.
347  */
348 int
349 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
350     const struct sockaddr *group, const struct sockaddr *src)
351 {
352 	size_t gidx;
353 	struct in6_msource *ims;
354 	int mode;
355 
356 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
357 
358 	gidx = im6o_match_group(imo, ifp, group);
359 	if (gidx == -1)
360 		return (MCAST_NOTGMEMBER);
361 
362 	/*
363 	 * Check if the source was included in an (S,G) join.
364 	 * Allow reception on exclusive memberships by default,
365 	 * reject reception on inclusive memberships by default.
366 	 * Exclude source only if an in-mode exclude filter exists.
367 	 * Include source only if an in-mode include filter exists.
368 	 * NOTE: We are comparing group state here at MLD t1 (now)
369 	 * with socket-layer t0 (since last downcall).
370 	 */
371 	mode = imo->im6o_mfilters[gidx].im6f_st[1];
372 	ims = im6o_match_source(imo, gidx, src);
373 
374 	if ((ims == NULL && mode == MCAST_INCLUDE) ||
375 	    (ims != NULL && ims->im6sl_st[0] != mode))
376 		return (MCAST_NOTSMEMBER);
377 
378 	return (MCAST_PASS);
379 }
380 
381 /*
382  * Find and return a reference to an in6_multi record for (ifp, group),
383  * and bump its reference count.
384  * If one does not exist, try to allocate it, and update link-layer multicast
385  * filters on ifp to listen for group.
386  * Assumes the IN6_MULTI lock is held across the call.
387  * Return 0 if successful, otherwise return an appropriate error code.
388  */
389 static int
390 in6_mc_get(struct ifnet *ifp, const struct in6_addr *group,
391     struct in6_multi **pinm)
392 {
393 	struct sockaddr_in6	 gsin6;
394 	struct ifmultiaddr	*ifma;
395 	struct in6_multi	*inm;
396 	int			 error;
397 
398 	error = 0;
399 
400 	/*
401 	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
402 	 * if_addmulti() takes this mutex itself, so we must drop and
403 	 * re-acquire around the call.
404 	 */
405 	IN6_MULTI_LOCK_ASSERT();
406 	IF_ADDR_WLOCK(ifp);
407 
408 	inm = in6m_lookup_locked(ifp, group);
409 	if (inm != NULL) {
410 		/*
411 		 * If we already joined this group, just bump the
412 		 * refcount and return it.
413 		 */
414 		KASSERT(inm->in6m_refcount >= 1,
415 		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
416 		++inm->in6m_refcount;
417 		*pinm = inm;
418 		goto out_locked;
419 	}
420 
421 	memset(&gsin6, 0, sizeof(gsin6));
422 	gsin6.sin6_family = AF_INET6;
423 	gsin6.sin6_len = sizeof(struct sockaddr_in6);
424 	gsin6.sin6_addr = *group;
425 
426 	/*
427 	 * Check if a link-layer group is already associated
428 	 * with this network-layer group on the given ifnet.
429 	 */
430 	IF_ADDR_WUNLOCK(ifp);
431 	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
432 	if (error != 0)
433 		return (error);
434 	IF_ADDR_WLOCK(ifp);
435 
436 	/*
437 	 * If something other than netinet6 is occupying the link-layer
438 	 * group, print a meaningful error message and back out of
439 	 * the allocation.
440 	 * Otherwise, bump the refcount on the existing network-layer
441 	 * group association and return it.
442 	 */
443 	if (ifma->ifma_protospec != NULL) {
444 		inm = (struct in6_multi *)ifma->ifma_protospec;
445 #ifdef INVARIANTS
446 		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
447 		    __func__));
448 		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
449 		    ("%s: ifma not AF_INET6", __func__));
450 		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
451 		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
452 		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
453 			panic("%s: ifma %p is inconsistent with %p (%p)",
454 			    __func__, ifma, inm, group);
455 #endif
456 		++inm->in6m_refcount;
457 		*pinm = inm;
458 		goto out_locked;
459 	}
460 
461 	IF_ADDR_WLOCK_ASSERT(ifp);
462 
463 	/*
464 	 * A new in6_multi record is needed; allocate and initialize it.
465 	 * We DO NOT perform an MLD join as the in6_ layer may need to
466 	 * push an initial source list down to MLD to support SSM.
467 	 *
468 	 * The initial source filter state is INCLUDE, {} as per the RFC.
469 	 * Pending state-changes per group are subject to a bounds check.
470 	 */
471 	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
472 	if (inm == NULL) {
473 		IF_ADDR_WUNLOCK(ifp);
474 		if_delmulti_ifma(ifma);
475 		return (ENOMEM);
476 	}
477 	inm->in6m_addr = *group;
478 	inm->in6m_ifp = ifp;
479 	inm->in6m_mli = MLD_IFINFO(ifp);
480 	inm->in6m_ifma = ifma;
481 	inm->in6m_refcount = 1;
482 	inm->in6m_state = MLD_NOT_MEMBER;
483 	mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
484 
485 	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
486 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
487 	RB_INIT(&inm->in6m_srcs);
488 
489 	ifma->ifma_protospec = inm;
490 	*pinm = inm;
491 
492 out_locked:
493 	IF_ADDR_WUNLOCK(ifp);
494 	return (error);
495 }
496 
497 /*
498  * Drop a reference to an in6_multi record.
499  *
500  * If the refcount drops to 0, free the in6_multi record and
501  * delete the underlying link-layer membership.
502  */
503 void
504 in6m_release_locked(struct in6_multi *inm)
505 {
506 	struct ifmultiaddr *ifma;
507 
508 	IN6_MULTI_LOCK_ASSERT();
509 
510 	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
511 
512 	if (--inm->in6m_refcount > 0) {
513 		CTR2(KTR_MLD, "%s: refcount is now %d", __func__,
514 		    inm->in6m_refcount);
515 		return;
516 	}
517 
518 	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
519 
520 	ifma = inm->in6m_ifma;
521 
522 	/* XXX this access is not covered by IF_ADDR_LOCK */
523 	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
524 	KASSERT(ifma->ifma_protospec == inm,
525 	    ("%s: ifma_protospec != inm", __func__));
526 	ifma->ifma_protospec = NULL;
527 
528 	in6m_purge(inm);
529 
530 	free(inm, M_IP6MADDR);
531 
532 	if_delmulti_ifma(ifma);
533 }
534 
535 /*
536  * Clear recorded source entries for a group.
537  * Used by the MLD code. Caller must hold the IN6_MULTI lock.
538  * FIXME: Should reap.
539  */
540 void
541 in6m_clear_recorded(struct in6_multi *inm)
542 {
543 	struct ip6_msource	*ims;
544 
545 	IN6_MULTI_LOCK_ASSERT();
546 
547 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
548 		if (ims->im6s_stp) {
549 			ims->im6s_stp = 0;
550 			--inm->in6m_st[1].iss_rec;
551 		}
552 	}
553 	KASSERT(inm->in6m_st[1].iss_rec == 0,
554 	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
555 }
556 
557 /*
558  * Record a source as pending for a Source-Group MLDv2 query.
559  * This lives here as it modifies the shared tree.
560  *
561  * inm is the group descriptor.
562  * naddr is the address of the source to record in network-byte order.
563  *
564  * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
565  * lazy-allocate a source node in response to an SG query.
566  * Otherwise, no allocation is performed. This saves some memory
567  * with the trade-off that the source will not be reported to the
568  * router if joined in the window between the query response and
569  * the group actually being joined on the local host.
570  *
571  * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
572  * This turns off the allocation of a recorded source entry if
573  * the group has not been joined.
574  *
575  * Return 0 if the source didn't exist or was already marked as recorded.
576  * Return 1 if the source was marked as recorded by this function.
577  * Return <0 if any error occurred (negated errno code).
578  */
579 int
580 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
581 {
582 	struct ip6_msource	 find;
583 	struct ip6_msource	*ims, *nims;
584 
585 	IN6_MULTI_LOCK_ASSERT();
586 
587 	find.im6s_addr = *addr;
588 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
589 	if (ims && ims->im6s_stp)
590 		return (0);
591 	if (ims == NULL) {
592 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
593 			return (-ENOSPC);
594 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
595 		    M_NOWAIT | M_ZERO);
596 		if (nims == NULL)
597 			return (-ENOMEM);
598 		nims->im6s_addr = find.im6s_addr;
599 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
600 		++inm->in6m_nsrc;
601 		ims = nims;
602 	}
603 
604 	/*
605 	 * Mark the source as recorded and update the recorded
606 	 * source count.
607 	 */
608 	++ims->im6s_stp;
609 	++inm->in6m_st[1].iss_rec;
610 
611 	return (1);
612 }
613 
614 /*
615  * Return a pointer to an in6_msource owned by an in6_mfilter,
616  * given its source address.
617  * Lazy-allocate if needed. If this is a new entry its filter state is
618  * undefined at t0.
619  *
620  * imf is the filter set being modified.
621  * addr is the source address.
622  *
623  * SMPng: May be called with locks held; malloc must not block.
624  */
625 static int
626 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
627     struct in6_msource **plims)
628 {
629 	struct ip6_msource	 find;
630 	struct ip6_msource	*ims, *nims;
631 	struct in6_msource	*lims;
632 	int			 error;
633 
634 	error = 0;
635 	ims = NULL;
636 	lims = NULL;
637 
638 	find.im6s_addr = psin->sin6_addr;
639 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
640 	lims = (struct in6_msource *)ims;
641 	if (lims == NULL) {
642 		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
643 			return (ENOSPC);
644 		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
645 		    M_NOWAIT | M_ZERO);
646 		if (nims == NULL)
647 			return (ENOMEM);
648 		lims = (struct in6_msource *)nims;
649 		lims->im6s_addr = find.im6s_addr;
650 		lims->im6sl_st[0] = MCAST_UNDEFINED;
651 		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
652 		++imf->im6f_nsrc;
653 	}
654 
655 	*plims = lims;
656 
657 	return (error);
658 }
659 
660 /*
661  * Graft a source entry into an existing socket-layer filter set,
662  * maintaining any required invariants and checking allocations.
663  *
664  * The source is marked as being in the new filter mode at t1.
665  *
666  * Return the pointer to the new node, otherwise return NULL.
667  */
668 static struct in6_msource *
669 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
670     const struct sockaddr_in6 *psin)
671 {
672 	struct ip6_msource	*nims;
673 	struct in6_msource	*lims;
674 
675 	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
676 	    M_NOWAIT | M_ZERO);
677 	if (nims == NULL)
678 		return (NULL);
679 	lims = (struct in6_msource *)nims;
680 	lims->im6s_addr = psin->sin6_addr;
681 	lims->im6sl_st[0] = MCAST_UNDEFINED;
682 	lims->im6sl_st[1] = st1;
683 	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
684 	++imf->im6f_nsrc;
685 
686 	return (lims);
687 }
688 
689 /*
690  * Prune a source entry from an existing socket-layer filter set,
691  * maintaining any required invariants and checking allocations.
692  *
693  * The source is marked as being left at t1, it is not freed.
694  *
695  * Return 0 if no error occurred, otherwise return an errno value.
696  */
697 static int
698 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
699 {
700 	struct ip6_msource	 find;
701 	struct ip6_msource	*ims;
702 	struct in6_msource	*lims;
703 
704 	find.im6s_addr = psin->sin6_addr;
705 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
706 	if (ims == NULL)
707 		return (ENOENT);
708 	lims = (struct in6_msource *)ims;
709 	lims->im6sl_st[1] = MCAST_UNDEFINED;
710 	return (0);
711 }
712 
713 /*
714  * Revert socket-layer filter set deltas at t1 to t0 state.
715  */
716 static void
717 im6f_rollback(struct in6_mfilter *imf)
718 {
719 	struct ip6_msource	*ims, *tims;
720 	struct in6_msource	*lims;
721 
722 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
723 		lims = (struct in6_msource *)ims;
724 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
725 			/* no change at t1 */
726 			continue;
727 		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
728 			/* revert change to existing source at t1 */
729 			lims->im6sl_st[1] = lims->im6sl_st[0];
730 		} else {
731 			/* revert source added t1 */
732 			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
733 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
734 			free(ims, M_IN6MFILTER);
735 			imf->im6f_nsrc--;
736 		}
737 	}
738 	imf->im6f_st[1] = imf->im6f_st[0];
739 }
740 
741 /*
742  * Mark socket-layer filter set as INCLUDE {} at t1.
743  */
744 static void
745 im6f_leave(struct in6_mfilter *imf)
746 {
747 	struct ip6_msource	*ims;
748 	struct in6_msource	*lims;
749 
750 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
751 		lims = (struct in6_msource *)ims;
752 		lims->im6sl_st[1] = MCAST_UNDEFINED;
753 	}
754 	imf->im6f_st[1] = MCAST_INCLUDE;
755 }
756 
757 /*
758  * Mark socket-layer filter set deltas as committed.
759  */
760 static void
761 im6f_commit(struct in6_mfilter *imf)
762 {
763 	struct ip6_msource	*ims;
764 	struct in6_msource	*lims;
765 
766 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
767 		lims = (struct in6_msource *)ims;
768 		lims->im6sl_st[0] = lims->im6sl_st[1];
769 	}
770 	imf->im6f_st[0] = imf->im6f_st[1];
771 }
772 
773 /*
774  * Reap unreferenced sources from socket-layer filter set.
775  */
776 static void
777 im6f_reap(struct in6_mfilter *imf)
778 {
779 	struct ip6_msource	*ims, *tims;
780 	struct in6_msource	*lims;
781 
782 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
783 		lims = (struct in6_msource *)ims;
784 		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
785 		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
786 			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
787 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
788 			free(ims, M_IN6MFILTER);
789 			imf->im6f_nsrc--;
790 		}
791 	}
792 }
793 
794 /*
795  * Purge socket-layer filter set.
796  */
797 static void
798 im6f_purge(struct in6_mfilter *imf)
799 {
800 	struct ip6_msource	*ims, *tims;
801 
802 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
803 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
804 		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
805 		free(ims, M_IN6MFILTER);
806 		imf->im6f_nsrc--;
807 	}
808 	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
809 	KASSERT(RB_EMPTY(&imf->im6f_sources),
810 	    ("%s: im6f_sources not empty", __func__));
811 }
812 
813 /*
814  * Look up a source filter entry for a multicast group.
815  *
816  * inm is the group descriptor to work with.
817  * addr is the IPv6 address to look up.
818  * noalloc may be non-zero to suppress allocation of sources.
819  * *pims will be set to the address of the retrieved or allocated source.
820  *
821  * SMPng: NOTE: may be called with locks held.
822  * Return 0 if successful, otherwise return a non-zero error code.
823  */
824 static int
825 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
826     const int noalloc, struct ip6_msource **pims)
827 {
828 	struct ip6_msource	 find;
829 	struct ip6_msource	*ims, *nims;
830 #ifdef KTR
831 	char			 ip6tbuf[INET6_ADDRSTRLEN];
832 #endif
833 
834 	find.im6s_addr = *addr;
835 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
836 	if (ims == NULL && !noalloc) {
837 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
838 			return (ENOSPC);
839 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
840 		    M_NOWAIT | M_ZERO);
841 		if (nims == NULL)
842 			return (ENOMEM);
843 		nims->im6s_addr = *addr;
844 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
845 		++inm->in6m_nsrc;
846 		ims = nims;
847 		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
848 		    ip6_sprintf(ip6tbuf, addr), ims);
849 	}
850 
851 	*pims = ims;
852 	return (0);
853 }
854 
855 /*
856  * Merge socket-layer source into MLD-layer source.
857  * If rollback is non-zero, perform the inverse of the merge.
858  */
859 static void
860 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
861     const int rollback)
862 {
863 	int n = rollback ? -1 : 1;
864 #ifdef KTR
865 	char ip6tbuf[INET6_ADDRSTRLEN];
866 
867 	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
868 #endif
869 
870 	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
871 		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
872 		ims->im6s_st[1].ex -= n;
873 	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
874 		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
875 		ims->im6s_st[1].in -= n;
876 	}
877 
878 	if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
879 		CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
880 		ims->im6s_st[1].ex += n;
881 	} else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
882 		CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
883 		ims->im6s_st[1].in += n;
884 	}
885 }
886 
887 /*
888  * Atomically update the global in6_multi state, when a membership's
889  * filter list is being updated in any way.
890  *
891  * imf is the per-inpcb-membership group filter pointer.
892  * A fake imf may be passed for in-kernel consumers.
893  *
894  * XXX This is a candidate for a set-symmetric-difference style loop
895  * which would eliminate the repeated lookup from root of ims nodes,
896  * as they share the same key space.
897  *
898  * If any error occurred this function will back out of refcounts
899  * and return a non-zero value.
900  */
901 static int
902 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
903 {
904 	struct ip6_msource	*ims, *nims;
905 	struct in6_msource	*lims;
906 	int			 schanged, error;
907 	int			 nsrc0, nsrc1;
908 
909 	schanged = 0;
910 	error = 0;
911 	nsrc1 = nsrc0 = 0;
912 
913 	/*
914 	 * Update the source filters first, as this may fail.
915 	 * Maintain count of in-mode filters at t0, t1. These are
916 	 * used to work out if we transition into ASM mode or not.
917 	 * Maintain a count of source filters whose state was
918 	 * actually modified by this operation.
919 	 */
920 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
921 		lims = (struct in6_msource *)ims;
922 		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
923 		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
924 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
925 		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
926 		++schanged;
927 		if (error)
928 			break;
929 		im6s_merge(nims, lims, 0);
930 	}
931 	if (error) {
932 		struct ip6_msource *bims;
933 
934 		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
935 			lims = (struct in6_msource *)ims;
936 			if (lims->im6sl_st[0] == lims->im6sl_st[1])
937 				continue;
938 			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
939 			if (bims == NULL)
940 				continue;
941 			im6s_merge(bims, lims, 1);
942 		}
943 		goto out_reap;
944 	}
945 
946 	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
947 	    __func__, nsrc0, nsrc1);
948 
949 	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
950 	if (imf->im6f_st[0] == imf->im6f_st[1] &&
951 	    imf->im6f_st[1] == MCAST_INCLUDE) {
952 		if (nsrc1 == 0) {
953 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
954 			--inm->in6m_st[1].iss_in;
955 		}
956 	}
957 
958 	/* Handle filter mode transition on socket. */
959 	if (imf->im6f_st[0] != imf->im6f_st[1]) {
960 		CTR3(KTR_MLD, "%s: imf transition %d to %d",
961 		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
962 
963 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
964 			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
965 			--inm->in6m_st[1].iss_ex;
966 		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
967 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
968 			--inm->in6m_st[1].iss_in;
969 		}
970 
971 		if (imf->im6f_st[1] == MCAST_EXCLUDE) {
972 			CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
973 			inm->in6m_st[1].iss_ex++;
974 		} else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
975 			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
976 			inm->in6m_st[1].iss_in++;
977 		}
978 	}
979 
980 	/*
981 	 * Track inm filter state in terms of listener counts.
982 	 * If there are any exclusive listeners, stack-wide
983 	 * membership is exclusive.
984 	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
985 	 * If no listeners remain, state is undefined at t1,
986 	 * and the MLD lifecycle for this group should finish.
987 	 */
988 	if (inm->in6m_st[1].iss_ex > 0) {
989 		CTR1(KTR_MLD, "%s: transition to EX", __func__);
990 		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
991 	} else if (inm->in6m_st[1].iss_in > 0) {
992 		CTR1(KTR_MLD, "%s: transition to IN", __func__);
993 		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
994 	} else {
995 		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
996 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
997 	}
998 
999 	/* Decrement ASM listener count on transition out of ASM mode. */
1000 	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1001 		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1002 		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1003 			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1004 			--inm->in6m_st[1].iss_asm;
1005 	}
1006 
1007 	/* Increment ASM listener count on transition to ASM mode. */
1008 	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1009 		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1010 		inm->in6m_st[1].iss_asm++;
1011 	}
1012 
1013 	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1014 	in6m_print(inm);
1015 
1016 out_reap:
1017 	if (schanged > 0) {
1018 		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1019 		in6m_reap(inm);
1020 	}
1021 	return (error);
1022 }
1023 
1024 /*
1025  * Mark an in6_multi's filter set deltas as committed.
1026  * Called by MLD after a state change has been enqueued.
1027  */
1028 void
1029 in6m_commit(struct in6_multi *inm)
1030 {
1031 	struct ip6_msource	*ims;
1032 
1033 	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1034 	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1035 	in6m_print(inm);
1036 
1037 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1038 		ims->im6s_st[0] = ims->im6s_st[1];
1039 	}
1040 	inm->in6m_st[0] = inm->in6m_st[1];
1041 }
1042 
1043 /*
1044  * Reap unreferenced nodes from an in6_multi's filter set.
1045  */
1046 static void
1047 in6m_reap(struct in6_multi *inm)
1048 {
1049 	struct ip6_msource	*ims, *tims;
1050 
1051 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1052 		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1053 		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1054 		    ims->im6s_stp != 0)
1055 			continue;
1056 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1057 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1058 		free(ims, M_IP6MSOURCE);
1059 		inm->in6m_nsrc--;
1060 	}
1061 }
1062 
1063 /*
1064  * Purge all source nodes from an in6_multi's filter set.
1065  */
1066 static void
1067 in6m_purge(struct in6_multi *inm)
1068 {
1069 	struct ip6_msource	*ims, *tims;
1070 
1071 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1072 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1073 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1074 		free(ims, M_IP6MSOURCE);
1075 		inm->in6m_nsrc--;
1076 	}
1077 	/* Free state-change requests that might be queued. */
1078 	mbufq_drain(&inm->in6m_scq);
1079 }
1080 
1081 /*
1082  * Join a multicast address w/o sources.
1083  * KAME compatibility entry point.
1084  *
1085  * SMPng: Assume no mc locks held by caller.
1086  */
1087 struct in6_multi_mship *
1088 in6_joingroup(struct ifnet *ifp, struct in6_addr *mcaddr,
1089     int *errorp, int delay)
1090 {
1091 	struct in6_multi_mship *imm;
1092 	int error;
1093 
1094 	imm = malloc(sizeof(*imm), M_IP6MADDR, M_NOWAIT);
1095 	if (imm == NULL) {
1096 		*errorp = ENOBUFS;
1097 		return (NULL);
1098 	}
1099 
1100 	delay = (delay * PR_FASTHZ) / hz;
1101 
1102 	error = in6_mc_join(ifp, mcaddr, NULL, &imm->i6mm_maddr, delay);
1103 	if (error) {
1104 		*errorp = error;
1105 		free(imm, M_IP6MADDR);
1106 		return (NULL);
1107 	}
1108 
1109 	return (imm);
1110 }
1111 
1112 /*
1113  * Leave a multicast address w/o sources.
1114  * KAME compatibility entry point.
1115  *
1116  * SMPng: Assume no mc locks held by caller.
1117  */
1118 int
1119 in6_leavegroup(struct in6_multi_mship *imm)
1120 {
1121 
1122 	if (imm->i6mm_maddr != NULL)
1123 		in6_mc_leave(imm->i6mm_maddr, NULL);
1124 	free(imm,  M_IP6MADDR);
1125 	return 0;
1126 }
1127 
1128 /*
1129  * Join a multicast group; unlocked entry point.
1130  *
1131  * SMPng: XXX: in6_mc_join() is called from in6_control() when upper
1132  * locks are not held. Fortunately, ifp is unlikely to have been detached
1133  * at this point, so we assume it's OK to recurse.
1134  */
1135 int
1136 in6_mc_join(struct ifnet *ifp, const struct in6_addr *mcaddr,
1137     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1138     const int delay)
1139 {
1140 	int error;
1141 
1142 	IN6_MULTI_LOCK();
1143 	error = in6_mc_join_locked(ifp, mcaddr, imf, pinm, delay);
1144 	IN6_MULTI_UNLOCK();
1145 
1146 	return (error);
1147 }
1148 
1149 /*
1150  * Join a multicast group; real entry point.
1151  *
1152  * Only preserves atomicity at inm level.
1153  * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1154  *
1155  * If the MLD downcall fails, the group is not joined, and an error
1156  * code is returned.
1157  */
1158 int
1159 in6_mc_join_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1160     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1161     const int delay)
1162 {
1163 	struct in6_mfilter	 timf;
1164 	struct in6_multi	*inm;
1165 	int			 error;
1166 #ifdef KTR
1167 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1168 #endif
1169 
1170 #ifdef INVARIANTS
1171 	/*
1172 	 * Sanity: Check scope zone ID was set for ifp, if and
1173 	 * only if group is scoped to an interface.
1174 	 */
1175 	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1176 	    ("%s: not a multicast address", __func__));
1177 	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1178 	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1179 		KASSERT(mcaddr->s6_addr16[1] != 0,
1180 		    ("%s: scope zone ID not set", __func__));
1181 	}
1182 #endif
1183 
1184 	IN6_MULTI_LOCK_ASSERT();
1185 
1186 	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1187 	    ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1188 
1189 	error = 0;
1190 	inm = NULL;
1191 
1192 	/*
1193 	 * If no imf was specified (i.e. kernel consumer),
1194 	 * fake one up and assume it is an ASM join.
1195 	 */
1196 	if (imf == NULL) {
1197 		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1198 		imf = &timf;
1199 	}
1200 
1201 	error = in6_mc_get(ifp, mcaddr, &inm);
1202 	if (error) {
1203 		CTR1(KTR_MLD, "%s: in6_mc_get() failure", __func__);
1204 		return (error);
1205 	}
1206 
1207 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1208 	error = in6m_merge(inm, imf);
1209 	if (error) {
1210 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1211 		goto out_in6m_release;
1212 	}
1213 
1214 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1215 	error = mld_change_state(inm, delay);
1216 	if (error) {
1217 		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1218 		goto out_in6m_release;
1219 	}
1220 
1221 out_in6m_release:
1222 	if (error) {
1223 		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1224 		in6m_release_locked(inm);
1225 	} else {
1226 		*pinm = inm;
1227 	}
1228 
1229 	return (error);
1230 }
1231 
1232 /*
1233  * Leave a multicast group; unlocked entry point.
1234  */
1235 int
1236 in6_mc_leave(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1237 {
1238 	struct ifnet *ifp;
1239 	int error;
1240 
1241 	ifp = inm->in6m_ifp;
1242 
1243 	IN6_MULTI_LOCK();
1244 	error = in6_mc_leave_locked(inm, imf);
1245 	IN6_MULTI_UNLOCK();
1246 
1247 	return (error);
1248 }
1249 
1250 /*
1251  * Leave a multicast group; real entry point.
1252  * All source filters will be expunged.
1253  *
1254  * Only preserves atomicity at inm level.
1255  *
1256  * Holding the write lock for the INP which contains imf
1257  * is highly advisable. We can't assert for it as imf does not
1258  * contain a back-pointer to the owning inp.
1259  *
1260  * Note: This is not the same as in6m_release(*) as this function also
1261  * makes a state change downcall into MLD.
1262  */
1263 int
1264 in6_mc_leave_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1265 {
1266 	struct in6_mfilter	 timf;
1267 	int			 error;
1268 #ifdef KTR
1269 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1270 #endif
1271 
1272 	error = 0;
1273 
1274 	IN6_MULTI_LOCK_ASSERT();
1275 
1276 	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1277 	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1278 	    (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1279 	    imf);
1280 
1281 	/*
1282 	 * If no imf was specified (i.e. kernel consumer),
1283 	 * fake one up and assume it is an ASM join.
1284 	 */
1285 	if (imf == NULL) {
1286 		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1287 		imf = &timf;
1288 	}
1289 
1290 	/*
1291 	 * Begin state merge transaction at MLD layer.
1292 	 *
1293 	 * As this particular invocation should not cause any memory
1294 	 * to be allocated, and there is no opportunity to roll back
1295 	 * the transaction, it MUST NOT fail.
1296 	 */
1297 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1298 	error = in6m_merge(inm, imf);
1299 	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1300 
1301 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1302 	error = mld_change_state(inm, 0);
1303 	if (error)
1304 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1305 
1306 	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1307 	in6m_release_locked(inm);
1308 
1309 	return (error);
1310 }
1311 
1312 /*
1313  * Block or unblock an ASM multicast source on an inpcb.
1314  * This implements the delta-based API described in RFC 3678.
1315  *
1316  * The delta-based API applies only to exclusive-mode memberships.
1317  * An MLD downcall will be performed.
1318  *
1319  * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1320  *
1321  * Return 0 if successful, otherwise return an appropriate error code.
1322  */
1323 static int
1324 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1325 {
1326 	struct group_source_req		 gsr;
1327 	sockunion_t			*gsa, *ssa;
1328 	struct ifnet			*ifp;
1329 	struct in6_mfilter		*imf;
1330 	struct ip6_moptions		*imo;
1331 	struct in6_msource		*ims;
1332 	struct in6_multi			*inm;
1333 	size_t				 idx;
1334 	uint16_t			 fmode;
1335 	int				 error, doblock;
1336 #ifdef KTR
1337 	char				 ip6tbuf[INET6_ADDRSTRLEN];
1338 #endif
1339 
1340 	ifp = NULL;
1341 	error = 0;
1342 	doblock = 0;
1343 
1344 	memset(&gsr, 0, sizeof(struct group_source_req));
1345 	gsa = (sockunion_t *)&gsr.gsr_group;
1346 	ssa = (sockunion_t *)&gsr.gsr_source;
1347 
1348 	switch (sopt->sopt_name) {
1349 	case MCAST_BLOCK_SOURCE:
1350 	case MCAST_UNBLOCK_SOURCE:
1351 		error = sooptcopyin(sopt, &gsr,
1352 		    sizeof(struct group_source_req),
1353 		    sizeof(struct group_source_req));
1354 		if (error)
1355 			return (error);
1356 
1357 		if (gsa->sin6.sin6_family != AF_INET6 ||
1358 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1359 			return (EINVAL);
1360 
1361 		if (ssa->sin6.sin6_family != AF_INET6 ||
1362 		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1363 			return (EINVAL);
1364 
1365 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1366 			return (EADDRNOTAVAIL);
1367 
1368 		ifp = ifnet_byindex(gsr.gsr_interface);
1369 
1370 		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1371 			doblock = 1;
1372 		break;
1373 
1374 	default:
1375 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1376 		    __func__, sopt->sopt_name);
1377 		return (EOPNOTSUPP);
1378 		break;
1379 	}
1380 
1381 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1382 		return (EINVAL);
1383 
1384 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1385 
1386 	/*
1387 	 * Check if we are actually a member of this group.
1388 	 */
1389 	imo = in6p_findmoptions(inp);
1390 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1391 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1392 		error = EADDRNOTAVAIL;
1393 		goto out_in6p_locked;
1394 	}
1395 
1396 	KASSERT(imo->im6o_mfilters != NULL,
1397 	    ("%s: im6o_mfilters not allocated", __func__));
1398 	imf = &imo->im6o_mfilters[idx];
1399 	inm = imo->im6o_membership[idx];
1400 
1401 	/*
1402 	 * Attempting to use the delta-based API on an
1403 	 * non exclusive-mode membership is an error.
1404 	 */
1405 	fmode = imf->im6f_st[0];
1406 	if (fmode != MCAST_EXCLUDE) {
1407 		error = EINVAL;
1408 		goto out_in6p_locked;
1409 	}
1410 
1411 	/*
1412 	 * Deal with error cases up-front:
1413 	 *  Asked to block, but already blocked; or
1414 	 *  Asked to unblock, but nothing to unblock.
1415 	 * If adding a new block entry, allocate it.
1416 	 */
1417 	ims = im6o_match_source(imo, idx, &ssa->sa);
1418 	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1419 		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1420 		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1421 		    doblock ? "" : "not ");
1422 		error = EADDRNOTAVAIL;
1423 		goto out_in6p_locked;
1424 	}
1425 
1426 	INP_WLOCK_ASSERT(inp);
1427 
1428 	/*
1429 	 * Begin state merge transaction at socket layer.
1430 	 */
1431 	if (doblock) {
1432 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1433 		ims = im6f_graft(imf, fmode, &ssa->sin6);
1434 		if (ims == NULL)
1435 			error = ENOMEM;
1436 	} else {
1437 		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1438 		error = im6f_prune(imf, &ssa->sin6);
1439 	}
1440 
1441 	if (error) {
1442 		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1443 		goto out_im6f_rollback;
1444 	}
1445 
1446 	/*
1447 	 * Begin state merge transaction at MLD layer.
1448 	 */
1449 	IN6_MULTI_LOCK();
1450 
1451 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1452 	error = in6m_merge(inm, imf);
1453 	if (error)
1454 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1455 	else {
1456 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1457 		error = mld_change_state(inm, 0);
1458 		if (error)
1459 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1460 	}
1461 
1462 	IN6_MULTI_UNLOCK();
1463 
1464 out_im6f_rollback:
1465 	if (error)
1466 		im6f_rollback(imf);
1467 	else
1468 		im6f_commit(imf);
1469 
1470 	im6f_reap(imf);
1471 
1472 out_in6p_locked:
1473 	INP_WUNLOCK(inp);
1474 	return (error);
1475 }
1476 
1477 /*
1478  * Given an inpcb, return its multicast options structure pointer.  Accepts
1479  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1480  *
1481  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1482  * SMPng: NOTE: Returns with the INP write lock held.
1483  */
1484 static struct ip6_moptions *
1485 in6p_findmoptions(struct inpcb *inp)
1486 {
1487 	struct ip6_moptions	 *imo;
1488 	struct in6_multi		**immp;
1489 	struct in6_mfilter	 *imfp;
1490 	size_t			  idx;
1491 
1492 	INP_WLOCK(inp);
1493 	if (inp->in6p_moptions != NULL)
1494 		return (inp->in6p_moptions);
1495 
1496 	INP_WUNLOCK(inp);
1497 
1498 	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1499 	immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1500 	    M_WAITOK | M_ZERO);
1501 	imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1502 	    M_IN6MFILTER, M_WAITOK);
1503 
1504 	imo->im6o_multicast_ifp = NULL;
1505 	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1506 	imo->im6o_multicast_loop = in6_mcast_loop;
1507 	imo->im6o_num_memberships = 0;
1508 	imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1509 	imo->im6o_membership = immp;
1510 
1511 	/* Initialize per-group source filters. */
1512 	for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1513 		im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1514 	imo->im6o_mfilters = imfp;
1515 
1516 	INP_WLOCK(inp);
1517 	if (inp->in6p_moptions != NULL) {
1518 		free(imfp, M_IN6MFILTER);
1519 		free(immp, M_IP6MOPTS);
1520 		free(imo, M_IP6MOPTS);
1521 		return (inp->in6p_moptions);
1522 	}
1523 	inp->in6p_moptions = imo;
1524 	return (imo);
1525 }
1526 
1527 /*
1528  * Discard the IPv6 multicast options (and source filters).
1529  *
1530  * SMPng: NOTE: assumes INP write lock is held.
1531  */
1532 void
1533 ip6_freemoptions(struct ip6_moptions *imo)
1534 {
1535 	struct in6_mfilter	*imf;
1536 	size_t			 idx, nmships;
1537 
1538 	KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1539 
1540 	nmships = imo->im6o_num_memberships;
1541 	for (idx = 0; idx < nmships; ++idx) {
1542 		imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1543 		if (imf)
1544 			im6f_leave(imf);
1545 		/* XXX this will thrash the lock(s) */
1546 		(void)in6_mc_leave(imo->im6o_membership[idx], imf);
1547 		if (imf)
1548 			im6f_purge(imf);
1549 	}
1550 
1551 	if (imo->im6o_mfilters)
1552 		free(imo->im6o_mfilters, M_IN6MFILTER);
1553 	free(imo->im6o_membership, M_IP6MOPTS);
1554 	free(imo, M_IP6MOPTS);
1555 }
1556 
1557 /*
1558  * Atomically get source filters on a socket for an IPv6 multicast group.
1559  * Called with INP lock held; returns with lock released.
1560  */
1561 static int
1562 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1563 {
1564 	struct __msfilterreq	 msfr;
1565 	sockunion_t		*gsa;
1566 	struct ifnet		*ifp;
1567 	struct ip6_moptions	*imo;
1568 	struct in6_mfilter	*imf;
1569 	struct ip6_msource	*ims;
1570 	struct in6_msource	*lims;
1571 	struct sockaddr_in6	*psin;
1572 	struct sockaddr_storage	*ptss;
1573 	struct sockaddr_storage	*tss;
1574 	int			 error;
1575 	size_t			 idx, nsrcs, ncsrcs;
1576 
1577 	INP_WLOCK_ASSERT(inp);
1578 
1579 	imo = inp->in6p_moptions;
1580 	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1581 
1582 	INP_WUNLOCK(inp);
1583 
1584 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1585 	    sizeof(struct __msfilterreq));
1586 	if (error)
1587 		return (error);
1588 
1589 	if (msfr.msfr_group.ss_family != AF_INET6 ||
1590 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1591 		return (EINVAL);
1592 
1593 	gsa = (sockunion_t *)&msfr.msfr_group;
1594 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1595 		return (EINVAL);
1596 
1597 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1598 		return (EADDRNOTAVAIL);
1599 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1600 	if (ifp == NULL)
1601 		return (EADDRNOTAVAIL);
1602 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1603 
1604 	INP_WLOCK(inp);
1605 
1606 	/*
1607 	 * Lookup group on the socket.
1608 	 */
1609 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1610 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1611 		INP_WUNLOCK(inp);
1612 		return (EADDRNOTAVAIL);
1613 	}
1614 	imf = &imo->im6o_mfilters[idx];
1615 
1616 	/*
1617 	 * Ignore memberships which are in limbo.
1618 	 */
1619 	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1620 		INP_WUNLOCK(inp);
1621 		return (EAGAIN);
1622 	}
1623 	msfr.msfr_fmode = imf->im6f_st[1];
1624 
1625 	/*
1626 	 * If the user specified a buffer, copy out the source filter
1627 	 * entries to userland gracefully.
1628 	 * We only copy out the number of entries which userland
1629 	 * has asked for, but we always tell userland how big the
1630 	 * buffer really needs to be.
1631 	 */
1632 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1633 		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1634 	tss = NULL;
1635 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1636 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1637 		    M_TEMP, M_NOWAIT | M_ZERO);
1638 		if (tss == NULL) {
1639 			INP_WUNLOCK(inp);
1640 			return (ENOBUFS);
1641 		}
1642 	}
1643 
1644 	/*
1645 	 * Count number of sources in-mode at t0.
1646 	 * If buffer space exists and remains, copy out source entries.
1647 	 */
1648 	nsrcs = msfr.msfr_nsrcs;
1649 	ncsrcs = 0;
1650 	ptss = tss;
1651 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1652 		lims = (struct in6_msource *)ims;
1653 		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1654 		    lims->im6sl_st[0] != imf->im6f_st[0])
1655 			continue;
1656 		++ncsrcs;
1657 		if (tss != NULL && nsrcs > 0) {
1658 			psin = (struct sockaddr_in6 *)ptss;
1659 			psin->sin6_family = AF_INET6;
1660 			psin->sin6_len = sizeof(struct sockaddr_in6);
1661 			psin->sin6_addr = lims->im6s_addr;
1662 			psin->sin6_port = 0;
1663 			--nsrcs;
1664 			++ptss;
1665 		}
1666 	}
1667 
1668 	INP_WUNLOCK(inp);
1669 
1670 	if (tss != NULL) {
1671 		error = copyout(tss, msfr.msfr_srcs,
1672 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1673 		free(tss, M_TEMP);
1674 		if (error)
1675 			return (error);
1676 	}
1677 
1678 	msfr.msfr_nsrcs = ncsrcs;
1679 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1680 
1681 	return (error);
1682 }
1683 
1684 /*
1685  * Return the IP multicast options in response to user getsockopt().
1686  */
1687 int
1688 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1689 {
1690 	struct ip6_moptions	*im6o;
1691 	int			 error;
1692 	u_int			 optval;
1693 
1694 	INP_WLOCK(inp);
1695 	im6o = inp->in6p_moptions;
1696 	/*
1697 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1698 	 * or is a divert socket, reject it.
1699 	 */
1700 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1701 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1702 	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1703 		INP_WUNLOCK(inp);
1704 		return (EOPNOTSUPP);
1705 	}
1706 
1707 	error = 0;
1708 	switch (sopt->sopt_name) {
1709 	case IPV6_MULTICAST_IF:
1710 		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1711 			optval = 0;
1712 		} else {
1713 			optval = im6o->im6o_multicast_ifp->if_index;
1714 		}
1715 		INP_WUNLOCK(inp);
1716 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1717 		break;
1718 
1719 	case IPV6_MULTICAST_HOPS:
1720 		if (im6o == NULL)
1721 			optval = V_ip6_defmcasthlim;
1722 		else
1723 			optval = im6o->im6o_multicast_hlim;
1724 		INP_WUNLOCK(inp);
1725 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1726 		break;
1727 
1728 	case IPV6_MULTICAST_LOOP:
1729 		if (im6o == NULL)
1730 			optval = in6_mcast_loop; /* XXX VIMAGE */
1731 		else
1732 			optval = im6o->im6o_multicast_loop;
1733 		INP_WUNLOCK(inp);
1734 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1735 		break;
1736 
1737 	case IPV6_MSFILTER:
1738 		if (im6o == NULL) {
1739 			error = EADDRNOTAVAIL;
1740 			INP_WUNLOCK(inp);
1741 		} else {
1742 			error = in6p_get_source_filters(inp, sopt);
1743 		}
1744 		break;
1745 
1746 	default:
1747 		INP_WUNLOCK(inp);
1748 		error = ENOPROTOOPT;
1749 		break;
1750 	}
1751 
1752 	INP_UNLOCK_ASSERT(inp);
1753 
1754 	return (error);
1755 }
1756 
1757 /*
1758  * Look up the ifnet to use for a multicast group membership,
1759  * given the address of an IPv6 group.
1760  *
1761  * This routine exists to support legacy IPv6 multicast applications.
1762  *
1763  * If inp is non-NULL, use this socket's current FIB number for any
1764  * required FIB lookup. Look up the group address in the unicast FIB,
1765  * and use its ifp; usually, this points to the default next-hop.
1766  * If the FIB lookup fails, return NULL.
1767  *
1768  * FUTURE: Support multiple forwarding tables for IPv6.
1769  *
1770  * Returns NULL if no ifp could be found.
1771  */
1772 static struct ifnet *
1773 in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1774     const struct sockaddr_in6 *gsin6)
1775 {
1776 	struct nhop6_basic	nh6;
1777 	struct in6_addr		dst;
1778 	uint32_t		scopeid;
1779 	uint32_t		fibnum;
1780 
1781 	KASSERT(in6p->inp_vflag & INP_IPV6,
1782 	    ("%s: not INP_IPV6 inpcb", __func__));
1783 	KASSERT(gsin6->sin6_family == AF_INET6,
1784 	    ("%s: not AF_INET6 group", __func__));
1785 
1786 	in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1787 	fibnum = in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB;
1788 	if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6) != 0)
1789 		return (NULL);
1790 
1791 	return (nh6.nh_ifp);
1792 }
1793 
1794 /*
1795  * Join an IPv6 multicast group, possibly with a source.
1796  *
1797  * FIXME: The KAME use of the unspecified address (::)
1798  * to join *all* multicast groups is currently unsupported.
1799  */
1800 static int
1801 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1802 {
1803 	struct group_source_req		 gsr;
1804 	sockunion_t			*gsa, *ssa;
1805 	struct ifnet			*ifp;
1806 	struct in6_mfilter		*imf;
1807 	struct ip6_moptions		*imo;
1808 	struct in6_multi		*inm;
1809 	struct in6_msource		*lims;
1810 	size_t				 idx;
1811 	int				 error, is_new;
1812 
1813 	ifp = NULL;
1814 	imf = NULL;
1815 	lims = NULL;
1816 	error = 0;
1817 	is_new = 0;
1818 
1819 	memset(&gsr, 0, sizeof(struct group_source_req));
1820 	gsa = (sockunion_t *)&gsr.gsr_group;
1821 	gsa->ss.ss_family = AF_UNSPEC;
1822 	ssa = (sockunion_t *)&gsr.gsr_source;
1823 	ssa->ss.ss_family = AF_UNSPEC;
1824 
1825 	/*
1826 	 * Chew everything into struct group_source_req.
1827 	 * Overwrite the port field if present, as the sockaddr
1828 	 * being copied in may be matched with a binary comparison.
1829 	 * Ignore passed-in scope ID.
1830 	 */
1831 	switch (sopt->sopt_name) {
1832 	case IPV6_JOIN_GROUP: {
1833 		struct ipv6_mreq mreq;
1834 
1835 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1836 		    sizeof(struct ipv6_mreq));
1837 		if (error)
1838 			return (error);
1839 
1840 		gsa->sin6.sin6_family = AF_INET6;
1841 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1842 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1843 
1844 		if (mreq.ipv6mr_interface == 0) {
1845 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1846 		} else {
1847 			if (V_if_index < mreq.ipv6mr_interface)
1848 				return (EADDRNOTAVAIL);
1849 			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1850 		}
1851 		CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1852 		    __func__, mreq.ipv6mr_interface, ifp);
1853 	} break;
1854 
1855 	case MCAST_JOIN_GROUP:
1856 	case MCAST_JOIN_SOURCE_GROUP:
1857 		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1858 			error = sooptcopyin(sopt, &gsr,
1859 			    sizeof(struct group_req),
1860 			    sizeof(struct group_req));
1861 		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1862 			error = sooptcopyin(sopt, &gsr,
1863 			    sizeof(struct group_source_req),
1864 			    sizeof(struct group_source_req));
1865 		}
1866 		if (error)
1867 			return (error);
1868 
1869 		if (gsa->sin6.sin6_family != AF_INET6 ||
1870 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1871 			return (EINVAL);
1872 
1873 		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1874 			if (ssa->sin6.sin6_family != AF_INET6 ||
1875 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1876 				return (EINVAL);
1877 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1878 				return (EINVAL);
1879 			/*
1880 			 * TODO: Validate embedded scope ID in source
1881 			 * list entry against passed-in ifp, if and only
1882 			 * if source list filter entry is iface or node local.
1883 			 */
1884 			in6_clearscope(&ssa->sin6.sin6_addr);
1885 			ssa->sin6.sin6_port = 0;
1886 			ssa->sin6.sin6_scope_id = 0;
1887 		}
1888 
1889 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1890 			return (EADDRNOTAVAIL);
1891 		ifp = ifnet_byindex(gsr.gsr_interface);
1892 		break;
1893 
1894 	default:
1895 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1896 		    __func__, sopt->sopt_name);
1897 		return (EOPNOTSUPP);
1898 		break;
1899 	}
1900 
1901 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1902 		return (EINVAL);
1903 
1904 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1905 		return (EADDRNOTAVAIL);
1906 
1907 	gsa->sin6.sin6_port = 0;
1908 	gsa->sin6.sin6_scope_id = 0;
1909 
1910 	/*
1911 	 * Always set the scope zone ID on memberships created from userland.
1912 	 * Use the passed-in ifp to do this.
1913 	 * XXX The in6_setscope() return value is meaningless.
1914 	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1915 	 */
1916 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1917 
1918 	imo = in6p_findmoptions(inp);
1919 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1920 	if (idx == -1) {
1921 		is_new = 1;
1922 	} else {
1923 		inm = imo->im6o_membership[idx];
1924 		imf = &imo->im6o_mfilters[idx];
1925 		if (ssa->ss.ss_family != AF_UNSPEC) {
1926 			/*
1927 			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1928 			 * is an error. On an existing inclusive membership,
1929 			 * it just adds the source to the filter list.
1930 			 */
1931 			if (imf->im6f_st[1] != MCAST_INCLUDE) {
1932 				error = EINVAL;
1933 				goto out_in6p_locked;
1934 			}
1935 			/*
1936 			 * Throw out duplicates.
1937 			 *
1938 			 * XXX FIXME: This makes a naive assumption that
1939 			 * even if entries exist for *ssa in this imf,
1940 			 * they will be rejected as dupes, even if they
1941 			 * are not valid in the current mode (in-mode).
1942 			 *
1943 			 * in6_msource is transactioned just as for anything
1944 			 * else in SSM -- but note naive use of in6m_graft()
1945 			 * below for allocating new filter entries.
1946 			 *
1947 			 * This is only an issue if someone mixes the
1948 			 * full-state SSM API with the delta-based API,
1949 			 * which is discouraged in the relevant RFCs.
1950 			 */
1951 			lims = im6o_match_source(imo, idx, &ssa->sa);
1952 			if (lims != NULL /*&&
1953 			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
1954 				error = EADDRNOTAVAIL;
1955 				goto out_in6p_locked;
1956 			}
1957 		} else {
1958 			/*
1959 			 * MCAST_JOIN_GROUP alone, on any existing membership,
1960 			 * is rejected, to stop the same inpcb tying up
1961 			 * multiple refs to the in_multi.
1962 			 * On an existing inclusive membership, this is also
1963 			 * an error; if you want to change filter mode,
1964 			 * you must use the userland API setsourcefilter().
1965 			 * XXX We don't reject this for imf in UNDEFINED
1966 			 * state at t1, because allocation of a filter
1967 			 * is atomic with allocation of a membership.
1968 			 */
1969 			error = EINVAL;
1970 			goto out_in6p_locked;
1971 		}
1972 	}
1973 
1974 	/*
1975 	 * Begin state merge transaction at socket layer.
1976 	 */
1977 	INP_WLOCK_ASSERT(inp);
1978 
1979 	if (is_new) {
1980 		if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
1981 			error = im6o_grow(imo);
1982 			if (error)
1983 				goto out_in6p_locked;
1984 		}
1985 		/*
1986 		 * Allocate the new slot upfront so we can deal with
1987 		 * grafting the new source filter in same code path
1988 		 * as for join-source on existing membership.
1989 		 */
1990 		idx = imo->im6o_num_memberships;
1991 		imo->im6o_membership[idx] = NULL;
1992 		imo->im6o_num_memberships++;
1993 		KASSERT(imo->im6o_mfilters != NULL,
1994 		    ("%s: im6f_mfilters vector was not allocated", __func__));
1995 		imf = &imo->im6o_mfilters[idx];
1996 		KASSERT(RB_EMPTY(&imf->im6f_sources),
1997 		    ("%s: im6f_sources not empty", __func__));
1998 	}
1999 
2000 	/*
2001 	 * Graft new source into filter list for this inpcb's
2002 	 * membership of the group. The in6_multi may not have
2003 	 * been allocated yet if this is a new membership, however,
2004 	 * the in_mfilter slot will be allocated and must be initialized.
2005 	 *
2006 	 * Note: Grafting of exclusive mode filters doesn't happen
2007 	 * in this path.
2008 	 * XXX: Should check for non-NULL lims (node exists but may
2009 	 * not be in-mode) for interop with full-state API.
2010 	 */
2011 	if (ssa->ss.ss_family != AF_UNSPEC) {
2012 		/* Membership starts in IN mode */
2013 		if (is_new) {
2014 			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2015 			im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2016 		} else {
2017 			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2018 		}
2019 		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2020 		if (lims == NULL) {
2021 			CTR1(KTR_MLD, "%s: merge imf state failed",
2022 			    __func__);
2023 			error = ENOMEM;
2024 			goto out_im6o_free;
2025 		}
2026 	} else {
2027 		/* No address specified; Membership starts in EX mode */
2028 		if (is_new) {
2029 			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2030 			im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2031 		}
2032 	}
2033 
2034 	/*
2035 	 * Begin state merge transaction at MLD layer.
2036 	 */
2037 	IN6_MULTI_LOCK();
2038 
2039 	if (is_new) {
2040 		error = in6_mc_join_locked(ifp, &gsa->sin6.sin6_addr, imf,
2041 		    &inm, 0);
2042 		if (error) {
2043 			IN6_MULTI_UNLOCK();
2044 			goto out_im6o_free;
2045 		}
2046 		imo->im6o_membership[idx] = inm;
2047 	} else {
2048 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2049 		error = in6m_merge(inm, imf);
2050 		if (error)
2051 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2052 			    __func__);
2053 		else {
2054 			CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2055 			error = mld_change_state(inm, 0);
2056 			if (error)
2057 				CTR1(KTR_MLD, "%s: failed mld downcall",
2058 				    __func__);
2059 		}
2060 	}
2061 
2062 	IN6_MULTI_UNLOCK();
2063 	INP_WLOCK_ASSERT(inp);
2064 	if (error) {
2065 		im6f_rollback(imf);
2066 		if (is_new)
2067 			im6f_purge(imf);
2068 		else
2069 			im6f_reap(imf);
2070 	} else {
2071 		im6f_commit(imf);
2072 	}
2073 
2074 out_im6o_free:
2075 	if (error && is_new) {
2076 		imo->im6o_membership[idx] = NULL;
2077 		--imo->im6o_num_memberships;
2078 	}
2079 
2080 out_in6p_locked:
2081 	INP_WUNLOCK(inp);
2082 	return (error);
2083 }
2084 
2085 /*
2086  * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2087  */
2088 static int
2089 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2090 {
2091 	struct ipv6_mreq		 mreq;
2092 	struct group_source_req		 gsr;
2093 	sockunion_t			*gsa, *ssa;
2094 	struct ifnet			*ifp;
2095 	struct in6_mfilter		*imf;
2096 	struct ip6_moptions		*imo;
2097 	struct in6_msource		*ims;
2098 	struct in6_multi		*inm;
2099 	uint32_t			 ifindex;
2100 	size_t				 idx;
2101 	int				 error, is_final;
2102 #ifdef KTR
2103 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2104 #endif
2105 
2106 	ifp = NULL;
2107 	ifindex = 0;
2108 	error = 0;
2109 	is_final = 1;
2110 
2111 	memset(&gsr, 0, sizeof(struct group_source_req));
2112 	gsa = (sockunion_t *)&gsr.gsr_group;
2113 	gsa->ss.ss_family = AF_UNSPEC;
2114 	ssa = (sockunion_t *)&gsr.gsr_source;
2115 	ssa->ss.ss_family = AF_UNSPEC;
2116 
2117 	/*
2118 	 * Chew everything passed in up into a struct group_source_req
2119 	 * as that is easier to process.
2120 	 * Note: Any embedded scope ID in the multicast group passed
2121 	 * in by userland is ignored, the interface index is the recommended
2122 	 * mechanism to specify an interface; see below.
2123 	 */
2124 	switch (sopt->sopt_name) {
2125 	case IPV6_LEAVE_GROUP:
2126 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2127 		    sizeof(struct ipv6_mreq));
2128 		if (error)
2129 			return (error);
2130 		gsa->sin6.sin6_family = AF_INET6;
2131 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2132 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2133 		gsa->sin6.sin6_port = 0;
2134 		gsa->sin6.sin6_scope_id = 0;
2135 		ifindex = mreq.ipv6mr_interface;
2136 		break;
2137 
2138 	case MCAST_LEAVE_GROUP:
2139 	case MCAST_LEAVE_SOURCE_GROUP:
2140 		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2141 			error = sooptcopyin(sopt, &gsr,
2142 			    sizeof(struct group_req),
2143 			    sizeof(struct group_req));
2144 		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2145 			error = sooptcopyin(sopt, &gsr,
2146 			    sizeof(struct group_source_req),
2147 			    sizeof(struct group_source_req));
2148 		}
2149 		if (error)
2150 			return (error);
2151 
2152 		if (gsa->sin6.sin6_family != AF_INET6 ||
2153 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2154 			return (EINVAL);
2155 		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2156 			if (ssa->sin6.sin6_family != AF_INET6 ||
2157 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2158 				return (EINVAL);
2159 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2160 				return (EINVAL);
2161 			/*
2162 			 * TODO: Validate embedded scope ID in source
2163 			 * list entry against passed-in ifp, if and only
2164 			 * if source list filter entry is iface or node local.
2165 			 */
2166 			in6_clearscope(&ssa->sin6.sin6_addr);
2167 		}
2168 		gsa->sin6.sin6_port = 0;
2169 		gsa->sin6.sin6_scope_id = 0;
2170 		ifindex = gsr.gsr_interface;
2171 		break;
2172 
2173 	default:
2174 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2175 		    __func__, sopt->sopt_name);
2176 		return (EOPNOTSUPP);
2177 		break;
2178 	}
2179 
2180 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2181 		return (EINVAL);
2182 
2183 	/*
2184 	 * Validate interface index if provided. If no interface index
2185 	 * was provided separately, attempt to look the membership up
2186 	 * from the default scope as a last resort to disambiguate
2187 	 * the membership we are being asked to leave.
2188 	 * XXX SCOPE6 lock potentially taken here.
2189 	 */
2190 	if (ifindex != 0) {
2191 		if (V_if_index < ifindex)
2192 			return (EADDRNOTAVAIL);
2193 		ifp = ifnet_byindex(ifindex);
2194 		if (ifp == NULL)
2195 			return (EADDRNOTAVAIL);
2196 		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2197 	} else {
2198 		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2199 		if (error)
2200 			return (EADDRNOTAVAIL);
2201 		/*
2202 		 * Some badly behaved applications don't pass an ifindex
2203 		 * or a scope ID, which is an API violation. In this case,
2204 		 * perform a lookup as per a v6 join.
2205 		 *
2206 		 * XXX For now, stomp on zone ID for the corner case.
2207 		 * This is not the 'KAME way', but we need to see the ifp
2208 		 * directly until such time as this implementation is
2209 		 * refactored, assuming the scope IDs are the way to go.
2210 		 */
2211 		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2212 		if (ifindex == 0) {
2213 			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2214 			    "ifp for group %s.", __func__,
2215 			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2216 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2217 		} else {
2218 			ifp = ifnet_byindex(ifindex);
2219 		}
2220 		if (ifp == NULL)
2221 			return (EADDRNOTAVAIL);
2222 	}
2223 
2224 	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2225 	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2226 
2227 	/*
2228 	 * Find the membership in the membership array.
2229 	 */
2230 	imo = in6p_findmoptions(inp);
2231 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2232 	if (idx == -1) {
2233 		error = EADDRNOTAVAIL;
2234 		goto out_in6p_locked;
2235 	}
2236 	inm = imo->im6o_membership[idx];
2237 	imf = &imo->im6o_mfilters[idx];
2238 
2239 	if (ssa->ss.ss_family != AF_UNSPEC)
2240 		is_final = 0;
2241 
2242 	/*
2243 	 * Begin state merge transaction at socket layer.
2244 	 */
2245 	INP_WLOCK_ASSERT(inp);
2246 
2247 	/*
2248 	 * If we were instructed only to leave a given source, do so.
2249 	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2250 	 */
2251 	if (is_final) {
2252 		im6f_leave(imf);
2253 	} else {
2254 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2255 			error = EADDRNOTAVAIL;
2256 			goto out_in6p_locked;
2257 		}
2258 		ims = im6o_match_source(imo, idx, &ssa->sa);
2259 		if (ims == NULL) {
2260 			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2261 			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2262 			    "not ");
2263 			error = EADDRNOTAVAIL;
2264 			goto out_in6p_locked;
2265 		}
2266 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2267 		error = im6f_prune(imf, &ssa->sin6);
2268 		if (error) {
2269 			CTR1(KTR_MLD, "%s: merge imf state failed",
2270 			    __func__);
2271 			goto out_in6p_locked;
2272 		}
2273 	}
2274 
2275 	/*
2276 	 * Begin state merge transaction at MLD layer.
2277 	 */
2278 	IN6_MULTI_LOCK();
2279 
2280 	if (is_final) {
2281 		/*
2282 		 * Give up the multicast address record to which
2283 		 * the membership points.
2284 		 */
2285 		(void)in6_mc_leave_locked(inm, imf);
2286 	} else {
2287 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2288 		error = in6m_merge(inm, imf);
2289 		if (error)
2290 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2291 			    __func__);
2292 		else {
2293 			CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2294 			error = mld_change_state(inm, 0);
2295 			if (error)
2296 				CTR1(KTR_MLD, "%s: failed mld downcall",
2297 				    __func__);
2298 		}
2299 	}
2300 
2301 	IN6_MULTI_UNLOCK();
2302 
2303 	if (error)
2304 		im6f_rollback(imf);
2305 	else
2306 		im6f_commit(imf);
2307 
2308 	im6f_reap(imf);
2309 
2310 	if (is_final) {
2311 		/* Remove the gap in the membership array. */
2312 		for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2313 			imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2314 			imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2315 		}
2316 		imo->im6o_num_memberships--;
2317 	}
2318 
2319 out_in6p_locked:
2320 	INP_WUNLOCK(inp);
2321 	return (error);
2322 }
2323 
2324 /*
2325  * Select the interface for transmitting IPv6 multicast datagrams.
2326  *
2327  * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2328  * may be passed to this socket option. An address of in6addr_any or an
2329  * interface index of 0 is used to remove a previous selection.
2330  * When no interface is selected, one is chosen for every send.
2331  */
2332 static int
2333 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2334 {
2335 	struct ifnet		*ifp;
2336 	struct ip6_moptions	*imo;
2337 	u_int			 ifindex;
2338 	int			 error;
2339 
2340 	if (sopt->sopt_valsize != sizeof(u_int))
2341 		return (EINVAL);
2342 
2343 	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2344 	if (error)
2345 		return (error);
2346 	if (V_if_index < ifindex)
2347 		return (EINVAL);
2348 	if (ifindex == 0)
2349 		ifp = NULL;
2350 	else {
2351 		ifp = ifnet_byindex(ifindex);
2352 		if (ifp == NULL)
2353 			return (EINVAL);
2354 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2355 			return (EADDRNOTAVAIL);
2356 	}
2357 	imo = in6p_findmoptions(inp);
2358 	imo->im6o_multicast_ifp = ifp;
2359 	INP_WUNLOCK(inp);
2360 
2361 	return (0);
2362 }
2363 
2364 /*
2365  * Atomically set source filters on a socket for an IPv6 multicast group.
2366  *
2367  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2368  */
2369 static int
2370 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2371 {
2372 	struct __msfilterreq	 msfr;
2373 	sockunion_t		*gsa;
2374 	struct ifnet		*ifp;
2375 	struct in6_mfilter	*imf;
2376 	struct ip6_moptions	*imo;
2377 	struct in6_multi		*inm;
2378 	size_t			 idx;
2379 	int			 error;
2380 
2381 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2382 	    sizeof(struct __msfilterreq));
2383 	if (error)
2384 		return (error);
2385 
2386 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2387 		return (ENOBUFS);
2388 
2389 	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2390 	    msfr.msfr_fmode != MCAST_INCLUDE)
2391 		return (EINVAL);
2392 
2393 	if (msfr.msfr_group.ss_family != AF_INET6 ||
2394 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2395 		return (EINVAL);
2396 
2397 	gsa = (sockunion_t *)&msfr.msfr_group;
2398 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2399 		return (EINVAL);
2400 
2401 	gsa->sin6.sin6_port = 0;	/* ignore port */
2402 
2403 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2404 		return (EADDRNOTAVAIL);
2405 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2406 	if (ifp == NULL)
2407 		return (EADDRNOTAVAIL);
2408 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2409 
2410 	/*
2411 	 * Take the INP write lock.
2412 	 * Check if this socket is a member of this group.
2413 	 */
2414 	imo = in6p_findmoptions(inp);
2415 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2416 	if (idx == -1 || imo->im6o_mfilters == NULL) {
2417 		error = EADDRNOTAVAIL;
2418 		goto out_in6p_locked;
2419 	}
2420 	inm = imo->im6o_membership[idx];
2421 	imf = &imo->im6o_mfilters[idx];
2422 
2423 	/*
2424 	 * Begin state merge transaction at socket layer.
2425 	 */
2426 	INP_WLOCK_ASSERT(inp);
2427 
2428 	imf->im6f_st[1] = msfr.msfr_fmode;
2429 
2430 	/*
2431 	 * Apply any new source filters, if present.
2432 	 * Make a copy of the user-space source vector so
2433 	 * that we may copy them with a single copyin. This
2434 	 * allows us to deal with page faults up-front.
2435 	 */
2436 	if (msfr.msfr_nsrcs > 0) {
2437 		struct in6_msource	*lims;
2438 		struct sockaddr_in6	*psin;
2439 		struct sockaddr_storage	*kss, *pkss;
2440 		int			 i;
2441 
2442 		INP_WUNLOCK(inp);
2443 
2444 		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2445 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2446 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2447 		    M_TEMP, M_WAITOK);
2448 		error = copyin(msfr.msfr_srcs, kss,
2449 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2450 		if (error) {
2451 			free(kss, M_TEMP);
2452 			return (error);
2453 		}
2454 
2455 		INP_WLOCK(inp);
2456 
2457 		/*
2458 		 * Mark all source filters as UNDEFINED at t1.
2459 		 * Restore new group filter mode, as im6f_leave()
2460 		 * will set it to INCLUDE.
2461 		 */
2462 		im6f_leave(imf);
2463 		imf->im6f_st[1] = msfr.msfr_fmode;
2464 
2465 		/*
2466 		 * Update socket layer filters at t1, lazy-allocating
2467 		 * new entries. This saves a bunch of memory at the
2468 		 * cost of one RB_FIND() per source entry; duplicate
2469 		 * entries in the msfr_nsrcs vector are ignored.
2470 		 * If we encounter an error, rollback transaction.
2471 		 *
2472 		 * XXX This too could be replaced with a set-symmetric
2473 		 * difference like loop to avoid walking from root
2474 		 * every time, as the key space is common.
2475 		 */
2476 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2477 			psin = (struct sockaddr_in6 *)pkss;
2478 			if (psin->sin6_family != AF_INET6) {
2479 				error = EAFNOSUPPORT;
2480 				break;
2481 			}
2482 			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2483 				error = EINVAL;
2484 				break;
2485 			}
2486 			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2487 				error = EINVAL;
2488 				break;
2489 			}
2490 			/*
2491 			 * TODO: Validate embedded scope ID in source
2492 			 * list entry against passed-in ifp, if and only
2493 			 * if source list filter entry is iface or node local.
2494 			 */
2495 			in6_clearscope(&psin->sin6_addr);
2496 			error = im6f_get_source(imf, psin, &lims);
2497 			if (error)
2498 				break;
2499 			lims->im6sl_st[1] = imf->im6f_st[1];
2500 		}
2501 		free(kss, M_TEMP);
2502 	}
2503 
2504 	if (error)
2505 		goto out_im6f_rollback;
2506 
2507 	INP_WLOCK_ASSERT(inp);
2508 	IN6_MULTI_LOCK();
2509 
2510 	/*
2511 	 * Begin state merge transaction at MLD layer.
2512 	 */
2513 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2514 	error = in6m_merge(inm, imf);
2515 	if (error)
2516 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2517 	else {
2518 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2519 		error = mld_change_state(inm, 0);
2520 		if (error)
2521 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2522 	}
2523 
2524 	IN6_MULTI_UNLOCK();
2525 
2526 out_im6f_rollback:
2527 	if (error)
2528 		im6f_rollback(imf);
2529 	else
2530 		im6f_commit(imf);
2531 
2532 	im6f_reap(imf);
2533 
2534 out_in6p_locked:
2535 	INP_WUNLOCK(inp);
2536 	return (error);
2537 }
2538 
2539 /*
2540  * Set the IP multicast options in response to user setsockopt().
2541  *
2542  * Many of the socket options handled in this function duplicate the
2543  * functionality of socket options in the regular unicast API. However,
2544  * it is not possible to merge the duplicate code, because the idempotence
2545  * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2546  * the effects of these options must be treated as separate and distinct.
2547  *
2548  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2549  */
2550 int
2551 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2552 {
2553 	struct ip6_moptions	*im6o;
2554 	int			 error;
2555 
2556 	error = 0;
2557 
2558 	/*
2559 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2560 	 * or is a divert socket, reject it.
2561 	 */
2562 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2563 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2564 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2565 		return (EOPNOTSUPP);
2566 
2567 	switch (sopt->sopt_name) {
2568 	case IPV6_MULTICAST_IF:
2569 		error = in6p_set_multicast_if(inp, sopt);
2570 		break;
2571 
2572 	case IPV6_MULTICAST_HOPS: {
2573 		int hlim;
2574 
2575 		if (sopt->sopt_valsize != sizeof(int)) {
2576 			error = EINVAL;
2577 			break;
2578 		}
2579 		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2580 		if (error)
2581 			break;
2582 		if (hlim < -1 || hlim > 255) {
2583 			error = EINVAL;
2584 			break;
2585 		} else if (hlim == -1) {
2586 			hlim = V_ip6_defmcasthlim;
2587 		}
2588 		im6o = in6p_findmoptions(inp);
2589 		im6o->im6o_multicast_hlim = hlim;
2590 		INP_WUNLOCK(inp);
2591 		break;
2592 	}
2593 
2594 	case IPV6_MULTICAST_LOOP: {
2595 		u_int loop;
2596 
2597 		/*
2598 		 * Set the loopback flag for outgoing multicast packets.
2599 		 * Must be zero or one.
2600 		 */
2601 		if (sopt->sopt_valsize != sizeof(u_int)) {
2602 			error = EINVAL;
2603 			break;
2604 		}
2605 		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2606 		if (error)
2607 			break;
2608 		if (loop > 1) {
2609 			error = EINVAL;
2610 			break;
2611 		}
2612 		im6o = in6p_findmoptions(inp);
2613 		im6o->im6o_multicast_loop = loop;
2614 		INP_WUNLOCK(inp);
2615 		break;
2616 	}
2617 
2618 	case IPV6_JOIN_GROUP:
2619 	case MCAST_JOIN_GROUP:
2620 	case MCAST_JOIN_SOURCE_GROUP:
2621 		error = in6p_join_group(inp, sopt);
2622 		break;
2623 
2624 	case IPV6_LEAVE_GROUP:
2625 	case MCAST_LEAVE_GROUP:
2626 	case MCAST_LEAVE_SOURCE_GROUP:
2627 		error = in6p_leave_group(inp, sopt);
2628 		break;
2629 
2630 	case MCAST_BLOCK_SOURCE:
2631 	case MCAST_UNBLOCK_SOURCE:
2632 		error = in6p_block_unblock_source(inp, sopt);
2633 		break;
2634 
2635 	case IPV6_MSFILTER:
2636 		error = in6p_set_source_filters(inp, sopt);
2637 		break;
2638 
2639 	default:
2640 		error = EOPNOTSUPP;
2641 		break;
2642 	}
2643 
2644 	INP_UNLOCK_ASSERT(inp);
2645 
2646 	return (error);
2647 }
2648 
2649 /*
2650  * Expose MLD's multicast filter mode and source list(s) to userland,
2651  * keyed by (ifindex, group).
2652  * The filter mode is written out as a uint32_t, followed by
2653  * 0..n of struct in6_addr.
2654  * For use by ifmcstat(8).
2655  * SMPng: NOTE: unlocked read of ifindex space.
2656  */
2657 static int
2658 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2659 {
2660 	struct in6_addr			 mcaddr;
2661 	struct in6_addr			 src;
2662 	struct ifnet			*ifp;
2663 	struct ifmultiaddr		*ifma;
2664 	struct in6_multi		*inm;
2665 	struct ip6_msource		*ims;
2666 	int				*name;
2667 	int				 retval;
2668 	u_int				 namelen;
2669 	uint32_t			 fmode, ifindex;
2670 #ifdef KTR
2671 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2672 #endif
2673 
2674 	name = (int *)arg1;
2675 	namelen = arg2;
2676 
2677 	if (req->newptr != NULL)
2678 		return (EPERM);
2679 
2680 	/* int: ifindex + 4 * 32 bits of IPv6 address */
2681 	if (namelen != 5)
2682 		return (EINVAL);
2683 
2684 	ifindex = name[0];
2685 	if (ifindex <= 0 || ifindex > V_if_index) {
2686 		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2687 		    __func__, ifindex);
2688 		return (ENOENT);
2689 	}
2690 
2691 	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2692 	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2693 		CTR2(KTR_MLD, "%s: group %s is not multicast",
2694 		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2695 		return (EINVAL);
2696 	}
2697 
2698 	ifp = ifnet_byindex(ifindex);
2699 	if (ifp == NULL) {
2700 		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2701 		    __func__, ifindex);
2702 		return (ENOENT);
2703 	}
2704 	/*
2705 	 * Internal MLD lookups require that scope/zone ID is set.
2706 	 */
2707 	(void)in6_setscope(&mcaddr, ifp, NULL);
2708 
2709 	retval = sysctl_wire_old_buffer(req,
2710 	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2711 	if (retval)
2712 		return (retval);
2713 
2714 	IN6_MULTI_LOCK();
2715 
2716 	IF_ADDR_RLOCK(ifp);
2717 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2718 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2719 		    ifma->ifma_protospec == NULL)
2720 			continue;
2721 		inm = (struct in6_multi *)ifma->ifma_protospec;
2722 		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2723 			continue;
2724 		fmode = inm->in6m_st[1].iss_fmode;
2725 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2726 		if (retval != 0)
2727 			break;
2728 		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2729 			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2730 			/*
2731 			 * Only copy-out sources which are in-mode.
2732 			 */
2733 			if (fmode != im6s_get_mode(inm, ims, 1)) {
2734 				CTR1(KTR_MLD, "%s: skip non-in-mode",
2735 				    __func__);
2736 				continue;
2737 			}
2738 			src = ims->im6s_addr;
2739 			retval = SYSCTL_OUT(req, &src,
2740 			    sizeof(struct in6_addr));
2741 			if (retval != 0)
2742 				break;
2743 		}
2744 	}
2745 	IF_ADDR_RUNLOCK(ifp);
2746 
2747 	IN6_MULTI_UNLOCK();
2748 
2749 	return (retval);
2750 }
2751 
2752 #ifdef KTR
2753 
2754 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2755 
2756 static const char *
2757 in6m_mode_str(const int mode)
2758 {
2759 
2760 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2761 		return (in6m_modestrs[mode]);
2762 	return ("??");
2763 }
2764 
2765 static const char *in6m_statestrs[] = {
2766 	"not-member",
2767 	"silent",
2768 	"idle",
2769 	"lazy",
2770 	"sleeping",
2771 	"awakening",
2772 	"query-pending",
2773 	"sg-query-pending",
2774 	"leaving"
2775 };
2776 
2777 static const char *
2778 in6m_state_str(const int state)
2779 {
2780 
2781 	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2782 		return (in6m_statestrs[state]);
2783 	return ("??");
2784 }
2785 
2786 /*
2787  * Dump an in6_multi structure to the console.
2788  */
2789 void
2790 in6m_print(const struct in6_multi *inm)
2791 {
2792 	int t;
2793 	char ip6tbuf[INET6_ADDRSTRLEN];
2794 
2795 	if ((ktr_mask & KTR_MLD) == 0)
2796 		return;
2797 
2798 	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2799 	printf("addr %s ifp %p(%s) ifma %p\n",
2800 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2801 	    inm->in6m_ifp,
2802 	    if_name(inm->in6m_ifp),
2803 	    inm->in6m_ifma);
2804 	printf("timer %u state %s refcount %u scq.len %u\n",
2805 	    inm->in6m_timer,
2806 	    in6m_state_str(inm->in6m_state),
2807 	    inm->in6m_refcount,
2808 	    mbufq_len(&inm->in6m_scq));
2809 	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2810 	    inm->in6m_mli,
2811 	    inm->in6m_nsrc,
2812 	    inm->in6m_sctimer,
2813 	    inm->in6m_scrv);
2814 	for (t = 0; t < 2; t++) {
2815 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2816 		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2817 		    inm->in6m_st[t].iss_asm,
2818 		    inm->in6m_st[t].iss_ex,
2819 		    inm->in6m_st[t].iss_in,
2820 		    inm->in6m_st[t].iss_rec);
2821 	}
2822 	printf("%s: --- end in6m %p ---\n", __func__, inm);
2823 }
2824 
2825 #else /* !KTR */
2826 
2827 void
2828 in6m_print(const struct in6_multi *inm)
2829 {
2830 
2831 }
2832 
2833 #endif /* KTR */
2834