xref: /freebsd/sys/netinet6/in6_mcast.c (revision 1de7b4b805ddbf2429da511c053686ac4591ed89)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2009 Bruce Simpson.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. The name of the author may not be used to endorse or promote
16  *    products derived from this software without specific prior written
17  *    permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  */
31 
32 /*
33  * IPv6 multicast socket, group, and socket option processing module.
34  * Normative references: RFC 2292, RFC 3492, RFC 3542, RFC 3678, RFC 3810.
35  */
36 
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
39 
40 #include "opt_inet6.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/protosw.h>
51 #include <sys/sysctl.h>
52 #include <sys/priv.h>
53 #include <sys/ktr.h>
54 #include <sys/tree.h>
55 
56 #include <net/if.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/vnet.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_var.h>
64 #include <netinet6/in6_fib.h>
65 #include <netinet6/in6_var.h>
66 #include <netinet/ip6.h>
67 #include <netinet/icmp6.h>
68 #include <netinet6/ip6_var.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/tcp_var.h>
71 #include <netinet6/nd6.h>
72 #include <netinet6/mld6_var.h>
73 #include <netinet6/scope6_var.h>
74 
75 #ifndef KTR_MLD
76 #define KTR_MLD KTR_INET6
77 #endif
78 
79 #ifndef __SOCKUNION_DECLARED
80 union sockunion {
81 	struct sockaddr_storage	ss;
82 	struct sockaddr		sa;
83 	struct sockaddr_dl	sdl;
84 	struct sockaddr_in6	sin6;
85 };
86 typedef union sockunion sockunion_t;
87 #define __SOCKUNION_DECLARED
88 #endif /* __SOCKUNION_DECLARED */
89 
90 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
91     "IPv6 multicast PCB-layer source filter");
92 static MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
93 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
94 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
95     "IPv6 multicast MLD-layer source filter");
96 
97 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
98 
99 /*
100  * Locking:
101  * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
102  * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
103  *   it can be taken by code in net/if.c also.
104  * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
105  *
106  * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
107  * any need for in6_multi itself to be virtualized -- it is bound to an ifp
108  * anyway no matter what happens.
109  */
110 struct mtx in6_multi_mtx;
111 MTX_SYSINIT(in6_multi_mtx, &in6_multi_mtx, "in6_multi_mtx", MTX_DEF);
112 
113 static void	im6f_commit(struct in6_mfilter *);
114 static int	im6f_get_source(struct in6_mfilter *imf,
115 		    const struct sockaddr_in6 *psin,
116 		    struct in6_msource **);
117 static struct in6_msource *
118 		im6f_graft(struct in6_mfilter *, const uint8_t,
119 		    const struct sockaddr_in6 *);
120 static void	im6f_leave(struct in6_mfilter *);
121 static int	im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
122 static void	im6f_purge(struct in6_mfilter *);
123 static void	im6f_rollback(struct in6_mfilter *);
124 static void	im6f_reap(struct in6_mfilter *);
125 static int	im6o_grow(struct ip6_moptions *);
126 static size_t	im6o_match_group(const struct ip6_moptions *,
127 		    const struct ifnet *, const struct sockaddr *);
128 static struct in6_msource *
129 		im6o_match_source(const struct ip6_moptions *, const size_t,
130 		    const struct sockaddr *);
131 static void	im6s_merge(struct ip6_msource *ims,
132 		    const struct in6_msource *lims, const int rollback);
133 static int	in6_mc_get(struct ifnet *, const struct in6_addr *,
134 		    struct in6_multi **);
135 static int	in6m_get_source(struct in6_multi *inm,
136 		    const struct in6_addr *addr, const int noalloc,
137 		    struct ip6_msource **pims);
138 #ifdef KTR
139 static int	in6m_is_ifp_detached(const struct in6_multi *);
140 #endif
141 static int	in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
142 static void	in6m_purge(struct in6_multi *);
143 static void	in6m_reap(struct in6_multi *);
144 static struct ip6_moptions *
145 		in6p_findmoptions(struct inpcb *);
146 static int	in6p_get_source_filters(struct inpcb *, struct sockopt *);
147 static int	in6p_join_group(struct inpcb *, struct sockopt *);
148 static int	in6p_leave_group(struct inpcb *, struct sockopt *);
149 static struct ifnet *
150 		in6p_lookup_mcast_ifp(const struct inpcb *,
151 		    const struct sockaddr_in6 *);
152 static int	in6p_block_unblock_source(struct inpcb *, struct sockopt *);
153 static int	in6p_set_multicast_if(struct inpcb *, struct sockopt *);
154 static int	in6p_set_source_filters(struct inpcb *, struct sockopt *);
155 static int	sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
156 
157 SYSCTL_DECL(_net_inet6_ip6);	/* XXX Not in any common header. */
158 
159 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0,
160     "IPv6 multicast");
161 
162 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
163 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
164     CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
165     "Max source filters per group");
166 
167 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
168 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
169     CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
170     "Max source filters per socket");
171 
172 /* TODO Virtualize this switch. */
173 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
174 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
175     &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
176 
177 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
178     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
179     "Per-interface stack-wide source filters");
180 
181 #ifdef KTR
182 /*
183  * Inline function which wraps assertions for a valid ifp.
184  * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
185  * is detached.
186  */
187 static int __inline
188 in6m_is_ifp_detached(const struct in6_multi *inm)
189 {
190 	struct ifnet *ifp;
191 
192 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
193 	ifp = inm->in6m_ifma->ifma_ifp;
194 	if (ifp != NULL) {
195 		/*
196 		 * Sanity check that network-layer notion of ifp is the
197 		 * same as that of link-layer.
198 		 */
199 		KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
200 	}
201 
202 	return (ifp == NULL);
203 }
204 #endif
205 
206 /*
207  * Initialize an in6_mfilter structure to a known state at t0, t1
208  * with an empty source filter list.
209  */
210 static __inline void
211 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
212 {
213 	memset(imf, 0, sizeof(struct in6_mfilter));
214 	RB_INIT(&imf->im6f_sources);
215 	imf->im6f_st[0] = st0;
216 	imf->im6f_st[1] = st1;
217 }
218 
219 /*
220  * Resize the ip6_moptions vector to the next power-of-two minus 1.
221  * May be called with locks held; do not sleep.
222  */
223 static int
224 im6o_grow(struct ip6_moptions *imo)
225 {
226 	struct in6_multi	**nmships;
227 	struct in6_multi	**omships;
228 	struct in6_mfilter	 *nmfilters;
229 	struct in6_mfilter	 *omfilters;
230 	size_t			  idx;
231 	size_t			  newmax;
232 	size_t			  oldmax;
233 
234 	nmships = NULL;
235 	nmfilters = NULL;
236 	omships = imo->im6o_membership;
237 	omfilters = imo->im6o_mfilters;
238 	oldmax = imo->im6o_max_memberships;
239 	newmax = ((oldmax + 1) * 2) - 1;
240 
241 	if (newmax <= IPV6_MAX_MEMBERSHIPS) {
242 		nmships = (struct in6_multi **)realloc(omships,
243 		    sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
244 		nmfilters = (struct in6_mfilter *)realloc(omfilters,
245 		    sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
246 		    M_NOWAIT);
247 		if (nmships != NULL && nmfilters != NULL) {
248 			/* Initialize newly allocated source filter heads. */
249 			for (idx = oldmax; idx < newmax; idx++) {
250 				im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
251 				    MCAST_EXCLUDE);
252 			}
253 			imo->im6o_max_memberships = newmax;
254 			imo->im6o_membership = nmships;
255 			imo->im6o_mfilters = nmfilters;
256 		}
257 	}
258 
259 	if (nmships == NULL || nmfilters == NULL) {
260 		if (nmships != NULL)
261 			free(nmships, M_IP6MOPTS);
262 		if (nmfilters != NULL)
263 			free(nmfilters, M_IN6MFILTER);
264 		return (ETOOMANYREFS);
265 	}
266 
267 	return (0);
268 }
269 
270 /*
271  * Find an IPv6 multicast group entry for this ip6_moptions instance
272  * which matches the specified group, and optionally an interface.
273  * Return its index into the array, or -1 if not found.
274  */
275 static size_t
276 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
277     const struct sockaddr *group)
278 {
279 	const struct sockaddr_in6 *gsin6;
280 	struct in6_multi	**pinm;
281 	int		  idx;
282 	int		  nmships;
283 
284 	gsin6 = (const struct sockaddr_in6 *)group;
285 
286 	/* The im6o_membership array may be lazy allocated. */
287 	if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
288 		return (-1);
289 
290 	nmships = imo->im6o_num_memberships;
291 	pinm = &imo->im6o_membership[0];
292 	for (idx = 0; idx < nmships; idx++, pinm++) {
293 		if (*pinm == NULL)
294 			continue;
295 		if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
296 		    IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
297 		    &gsin6->sin6_addr)) {
298 			break;
299 		}
300 	}
301 	if (idx >= nmships)
302 		idx = -1;
303 
304 	return (idx);
305 }
306 
307 /*
308  * Find an IPv6 multicast source entry for this imo which matches
309  * the given group index for this socket, and source address.
310  *
311  * XXX TODO: The scope ID, if present in src, is stripped before
312  * any comparison. We SHOULD enforce scope/zone checks where the source
313  * filter entry has a link scope.
314  *
315  * NOTE: This does not check if the entry is in-mode, merely if
316  * it exists, which may not be the desired behaviour.
317  */
318 static struct in6_msource *
319 im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
320     const struct sockaddr *src)
321 {
322 	struct ip6_msource	 find;
323 	struct in6_mfilter	*imf;
324 	struct ip6_msource	*ims;
325 	const sockunion_t	*psa;
326 
327 	KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
328 	KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
329 	    ("%s: invalid index %d\n", __func__, (int)gidx));
330 
331 	/* The im6o_mfilters array may be lazy allocated. */
332 	if (imo->im6o_mfilters == NULL)
333 		return (NULL);
334 	imf = &imo->im6o_mfilters[gidx];
335 
336 	psa = (const sockunion_t *)src;
337 	find.im6s_addr = psa->sin6.sin6_addr;
338 	in6_clearscope(&find.im6s_addr);		/* XXX */
339 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
340 
341 	return ((struct in6_msource *)ims);
342 }
343 
344 /*
345  * Perform filtering for multicast datagrams on a socket by group and source.
346  *
347  * Returns 0 if a datagram should be allowed through, or various error codes
348  * if the socket was not a member of the group, or the source was muted, etc.
349  */
350 int
351 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
352     const struct sockaddr *group, const struct sockaddr *src)
353 {
354 	size_t gidx;
355 	struct in6_msource *ims;
356 	int mode;
357 
358 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
359 
360 	gidx = im6o_match_group(imo, ifp, group);
361 	if (gidx == -1)
362 		return (MCAST_NOTGMEMBER);
363 
364 	/*
365 	 * Check if the source was included in an (S,G) join.
366 	 * Allow reception on exclusive memberships by default,
367 	 * reject reception on inclusive memberships by default.
368 	 * Exclude source only if an in-mode exclude filter exists.
369 	 * Include source only if an in-mode include filter exists.
370 	 * NOTE: We are comparing group state here at MLD t1 (now)
371 	 * with socket-layer t0 (since last downcall).
372 	 */
373 	mode = imo->im6o_mfilters[gidx].im6f_st[1];
374 	ims = im6o_match_source(imo, gidx, src);
375 
376 	if ((ims == NULL && mode == MCAST_INCLUDE) ||
377 	    (ims != NULL && ims->im6sl_st[0] != mode))
378 		return (MCAST_NOTSMEMBER);
379 
380 	return (MCAST_PASS);
381 }
382 
383 /*
384  * Find and return a reference to an in6_multi record for (ifp, group),
385  * and bump its reference count.
386  * If one does not exist, try to allocate it, and update link-layer multicast
387  * filters on ifp to listen for group.
388  * Assumes the IN6_MULTI lock is held across the call.
389  * Return 0 if successful, otherwise return an appropriate error code.
390  */
391 static int
392 in6_mc_get(struct ifnet *ifp, const struct in6_addr *group,
393     struct in6_multi **pinm)
394 {
395 	struct sockaddr_in6	 gsin6;
396 	struct ifmultiaddr	*ifma;
397 	struct in6_multi	*inm;
398 	int			 error;
399 
400 	error = 0;
401 
402 	/*
403 	 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
404 	 * if_addmulti() takes this mutex itself, so we must drop and
405 	 * re-acquire around the call.
406 	 */
407 	IN6_MULTI_LOCK_ASSERT();
408 	IF_ADDR_WLOCK(ifp);
409 
410 	inm = in6m_lookup_locked(ifp, group);
411 	if (inm != NULL) {
412 		/*
413 		 * If we already joined this group, just bump the
414 		 * refcount and return it.
415 		 */
416 		KASSERT(inm->in6m_refcount >= 1,
417 		    ("%s: bad refcount %d", __func__, inm->in6m_refcount));
418 		++inm->in6m_refcount;
419 		*pinm = inm;
420 		goto out_locked;
421 	}
422 
423 	memset(&gsin6, 0, sizeof(gsin6));
424 	gsin6.sin6_family = AF_INET6;
425 	gsin6.sin6_len = sizeof(struct sockaddr_in6);
426 	gsin6.sin6_addr = *group;
427 
428 	/*
429 	 * Check if a link-layer group is already associated
430 	 * with this network-layer group on the given ifnet.
431 	 */
432 	IF_ADDR_WUNLOCK(ifp);
433 	error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
434 	if (error != 0)
435 		return (error);
436 	IF_ADDR_WLOCK(ifp);
437 
438 	/*
439 	 * If something other than netinet6 is occupying the link-layer
440 	 * group, print a meaningful error message and back out of
441 	 * the allocation.
442 	 * Otherwise, bump the refcount on the existing network-layer
443 	 * group association and return it.
444 	 */
445 	if (ifma->ifma_protospec != NULL) {
446 		inm = (struct in6_multi *)ifma->ifma_protospec;
447 #ifdef INVARIANTS
448 		KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
449 		    __func__));
450 		KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
451 		    ("%s: ifma not AF_INET6", __func__));
452 		KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
453 		if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
454 		    !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
455 			panic("%s: ifma %p is inconsistent with %p (%p)",
456 			    __func__, ifma, inm, group);
457 #endif
458 		++inm->in6m_refcount;
459 		*pinm = inm;
460 		goto out_locked;
461 	}
462 
463 	IF_ADDR_WLOCK_ASSERT(ifp);
464 
465 	/*
466 	 * A new in6_multi record is needed; allocate and initialize it.
467 	 * We DO NOT perform an MLD join as the in6_ layer may need to
468 	 * push an initial source list down to MLD to support SSM.
469 	 *
470 	 * The initial source filter state is INCLUDE, {} as per the RFC.
471 	 * Pending state-changes per group are subject to a bounds check.
472 	 */
473 	inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
474 	if (inm == NULL) {
475 		IF_ADDR_WUNLOCK(ifp);
476 		if_delmulti_ifma(ifma);
477 		return (ENOMEM);
478 	}
479 	inm->in6m_addr = *group;
480 	inm->in6m_ifp = ifp;
481 	inm->in6m_mli = MLD_IFINFO(ifp);
482 	inm->in6m_ifma = ifma;
483 	inm->in6m_refcount = 1;
484 	inm->in6m_state = MLD_NOT_MEMBER;
485 	mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
486 
487 	inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
488 	inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
489 	RB_INIT(&inm->in6m_srcs);
490 
491 	ifma->ifma_protospec = inm;
492 	*pinm = inm;
493 
494 out_locked:
495 	IF_ADDR_WUNLOCK(ifp);
496 	return (error);
497 }
498 
499 /*
500  * Drop a reference to an in6_multi record.
501  *
502  * If the refcount drops to 0, free the in6_multi record and
503  * delete the underlying link-layer membership.
504  */
505 void
506 in6m_release_locked(struct in6_multi *inm)
507 {
508 	struct ifmultiaddr *ifma;
509 
510 	IN6_MULTI_LOCK_ASSERT();
511 
512 	CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
513 
514 	if (--inm->in6m_refcount > 0) {
515 		CTR2(KTR_MLD, "%s: refcount is now %d", __func__,
516 		    inm->in6m_refcount);
517 		return;
518 	}
519 
520 	CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
521 
522 	ifma = inm->in6m_ifma;
523 
524 	/* XXX this access is not covered by IF_ADDR_LOCK */
525 	CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
526 	KASSERT(ifma->ifma_protospec == inm,
527 	    ("%s: ifma_protospec != inm", __func__));
528 	ifma->ifma_protospec = NULL;
529 
530 	in6m_purge(inm);
531 
532 	free(inm, M_IP6MADDR);
533 
534 	if_delmulti_ifma(ifma);
535 }
536 
537 /*
538  * Clear recorded source entries for a group.
539  * Used by the MLD code. Caller must hold the IN6_MULTI lock.
540  * FIXME: Should reap.
541  */
542 void
543 in6m_clear_recorded(struct in6_multi *inm)
544 {
545 	struct ip6_msource	*ims;
546 
547 	IN6_MULTI_LOCK_ASSERT();
548 
549 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
550 		if (ims->im6s_stp) {
551 			ims->im6s_stp = 0;
552 			--inm->in6m_st[1].iss_rec;
553 		}
554 	}
555 	KASSERT(inm->in6m_st[1].iss_rec == 0,
556 	    ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
557 }
558 
559 /*
560  * Record a source as pending for a Source-Group MLDv2 query.
561  * This lives here as it modifies the shared tree.
562  *
563  * inm is the group descriptor.
564  * naddr is the address of the source to record in network-byte order.
565  *
566  * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
567  * lazy-allocate a source node in response to an SG query.
568  * Otherwise, no allocation is performed. This saves some memory
569  * with the trade-off that the source will not be reported to the
570  * router if joined in the window between the query response and
571  * the group actually being joined on the local host.
572  *
573  * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
574  * This turns off the allocation of a recorded source entry if
575  * the group has not been joined.
576  *
577  * Return 0 if the source didn't exist or was already marked as recorded.
578  * Return 1 if the source was marked as recorded by this function.
579  * Return <0 if any error occurred (negated errno code).
580  */
581 int
582 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
583 {
584 	struct ip6_msource	 find;
585 	struct ip6_msource	*ims, *nims;
586 
587 	IN6_MULTI_LOCK_ASSERT();
588 
589 	find.im6s_addr = *addr;
590 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
591 	if (ims && ims->im6s_stp)
592 		return (0);
593 	if (ims == NULL) {
594 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
595 			return (-ENOSPC);
596 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
597 		    M_NOWAIT | M_ZERO);
598 		if (nims == NULL)
599 			return (-ENOMEM);
600 		nims->im6s_addr = find.im6s_addr;
601 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
602 		++inm->in6m_nsrc;
603 		ims = nims;
604 	}
605 
606 	/*
607 	 * Mark the source as recorded and update the recorded
608 	 * source count.
609 	 */
610 	++ims->im6s_stp;
611 	++inm->in6m_st[1].iss_rec;
612 
613 	return (1);
614 }
615 
616 /*
617  * Return a pointer to an in6_msource owned by an in6_mfilter,
618  * given its source address.
619  * Lazy-allocate if needed. If this is a new entry its filter state is
620  * undefined at t0.
621  *
622  * imf is the filter set being modified.
623  * addr is the source address.
624  *
625  * SMPng: May be called with locks held; malloc must not block.
626  */
627 static int
628 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
629     struct in6_msource **plims)
630 {
631 	struct ip6_msource	 find;
632 	struct ip6_msource	*ims, *nims;
633 	struct in6_msource	*lims;
634 	int			 error;
635 
636 	error = 0;
637 	ims = NULL;
638 	lims = NULL;
639 
640 	find.im6s_addr = psin->sin6_addr;
641 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
642 	lims = (struct in6_msource *)ims;
643 	if (lims == NULL) {
644 		if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
645 			return (ENOSPC);
646 		nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
647 		    M_NOWAIT | M_ZERO);
648 		if (nims == NULL)
649 			return (ENOMEM);
650 		lims = (struct in6_msource *)nims;
651 		lims->im6s_addr = find.im6s_addr;
652 		lims->im6sl_st[0] = MCAST_UNDEFINED;
653 		RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
654 		++imf->im6f_nsrc;
655 	}
656 
657 	*plims = lims;
658 
659 	return (error);
660 }
661 
662 /*
663  * Graft a source entry into an existing socket-layer filter set,
664  * maintaining any required invariants and checking allocations.
665  *
666  * The source is marked as being in the new filter mode at t1.
667  *
668  * Return the pointer to the new node, otherwise return NULL.
669  */
670 static struct in6_msource *
671 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
672     const struct sockaddr_in6 *psin)
673 {
674 	struct ip6_msource	*nims;
675 	struct in6_msource	*lims;
676 
677 	nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
678 	    M_NOWAIT | M_ZERO);
679 	if (nims == NULL)
680 		return (NULL);
681 	lims = (struct in6_msource *)nims;
682 	lims->im6s_addr = psin->sin6_addr;
683 	lims->im6sl_st[0] = MCAST_UNDEFINED;
684 	lims->im6sl_st[1] = st1;
685 	RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
686 	++imf->im6f_nsrc;
687 
688 	return (lims);
689 }
690 
691 /*
692  * Prune a source entry from an existing socket-layer filter set,
693  * maintaining any required invariants and checking allocations.
694  *
695  * The source is marked as being left at t1, it is not freed.
696  *
697  * Return 0 if no error occurred, otherwise return an errno value.
698  */
699 static int
700 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
701 {
702 	struct ip6_msource	 find;
703 	struct ip6_msource	*ims;
704 	struct in6_msource	*lims;
705 
706 	find.im6s_addr = psin->sin6_addr;
707 	ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
708 	if (ims == NULL)
709 		return (ENOENT);
710 	lims = (struct in6_msource *)ims;
711 	lims->im6sl_st[1] = MCAST_UNDEFINED;
712 	return (0);
713 }
714 
715 /*
716  * Revert socket-layer filter set deltas at t1 to t0 state.
717  */
718 static void
719 im6f_rollback(struct in6_mfilter *imf)
720 {
721 	struct ip6_msource	*ims, *tims;
722 	struct in6_msource	*lims;
723 
724 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
725 		lims = (struct in6_msource *)ims;
726 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
727 			/* no change at t1 */
728 			continue;
729 		} else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
730 			/* revert change to existing source at t1 */
731 			lims->im6sl_st[1] = lims->im6sl_st[0];
732 		} else {
733 			/* revert source added t1 */
734 			CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
735 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
736 			free(ims, M_IN6MFILTER);
737 			imf->im6f_nsrc--;
738 		}
739 	}
740 	imf->im6f_st[1] = imf->im6f_st[0];
741 }
742 
743 /*
744  * Mark socket-layer filter set as INCLUDE {} at t1.
745  */
746 static void
747 im6f_leave(struct in6_mfilter *imf)
748 {
749 	struct ip6_msource	*ims;
750 	struct in6_msource	*lims;
751 
752 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
753 		lims = (struct in6_msource *)ims;
754 		lims->im6sl_st[1] = MCAST_UNDEFINED;
755 	}
756 	imf->im6f_st[1] = MCAST_INCLUDE;
757 }
758 
759 /*
760  * Mark socket-layer filter set deltas as committed.
761  */
762 static void
763 im6f_commit(struct in6_mfilter *imf)
764 {
765 	struct ip6_msource	*ims;
766 	struct in6_msource	*lims;
767 
768 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
769 		lims = (struct in6_msource *)ims;
770 		lims->im6sl_st[0] = lims->im6sl_st[1];
771 	}
772 	imf->im6f_st[0] = imf->im6f_st[1];
773 }
774 
775 /*
776  * Reap unreferenced sources from socket-layer filter set.
777  */
778 static void
779 im6f_reap(struct in6_mfilter *imf)
780 {
781 	struct ip6_msource	*ims, *tims;
782 	struct in6_msource	*lims;
783 
784 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
785 		lims = (struct in6_msource *)ims;
786 		if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
787 		    (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
788 			CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
789 			RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
790 			free(ims, M_IN6MFILTER);
791 			imf->im6f_nsrc--;
792 		}
793 	}
794 }
795 
796 /*
797  * Purge socket-layer filter set.
798  */
799 static void
800 im6f_purge(struct in6_mfilter *imf)
801 {
802 	struct ip6_msource	*ims, *tims;
803 
804 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
805 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
806 		RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
807 		free(ims, M_IN6MFILTER);
808 		imf->im6f_nsrc--;
809 	}
810 	imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
811 	KASSERT(RB_EMPTY(&imf->im6f_sources),
812 	    ("%s: im6f_sources not empty", __func__));
813 }
814 
815 /*
816  * Look up a source filter entry for a multicast group.
817  *
818  * inm is the group descriptor to work with.
819  * addr is the IPv6 address to look up.
820  * noalloc may be non-zero to suppress allocation of sources.
821  * *pims will be set to the address of the retrieved or allocated source.
822  *
823  * SMPng: NOTE: may be called with locks held.
824  * Return 0 if successful, otherwise return a non-zero error code.
825  */
826 static int
827 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
828     const int noalloc, struct ip6_msource **pims)
829 {
830 	struct ip6_msource	 find;
831 	struct ip6_msource	*ims, *nims;
832 #ifdef KTR
833 	char			 ip6tbuf[INET6_ADDRSTRLEN];
834 #endif
835 
836 	find.im6s_addr = *addr;
837 	ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
838 	if (ims == NULL && !noalloc) {
839 		if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
840 			return (ENOSPC);
841 		nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
842 		    M_NOWAIT | M_ZERO);
843 		if (nims == NULL)
844 			return (ENOMEM);
845 		nims->im6s_addr = *addr;
846 		RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
847 		++inm->in6m_nsrc;
848 		ims = nims;
849 		CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
850 		    ip6_sprintf(ip6tbuf, addr), ims);
851 	}
852 
853 	*pims = ims;
854 	return (0);
855 }
856 
857 /*
858  * Merge socket-layer source into MLD-layer source.
859  * If rollback is non-zero, perform the inverse of the merge.
860  */
861 static void
862 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
863     const int rollback)
864 {
865 	int n = rollback ? -1 : 1;
866 #ifdef KTR
867 	char ip6tbuf[INET6_ADDRSTRLEN];
868 
869 	ip6_sprintf(ip6tbuf, &lims->im6s_addr);
870 #endif
871 
872 	if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
873 		CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
874 		ims->im6s_st[1].ex -= n;
875 	} else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
876 		CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
877 		ims->im6s_st[1].in -= n;
878 	}
879 
880 	if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
881 		CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
882 		ims->im6s_st[1].ex += n;
883 	} else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
884 		CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
885 		ims->im6s_st[1].in += n;
886 	}
887 }
888 
889 /*
890  * Atomically update the global in6_multi state, when a membership's
891  * filter list is being updated in any way.
892  *
893  * imf is the per-inpcb-membership group filter pointer.
894  * A fake imf may be passed for in-kernel consumers.
895  *
896  * XXX This is a candidate for a set-symmetric-difference style loop
897  * which would eliminate the repeated lookup from root of ims nodes,
898  * as they share the same key space.
899  *
900  * If any error occurred this function will back out of refcounts
901  * and return a non-zero value.
902  */
903 static int
904 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
905 {
906 	struct ip6_msource	*ims, *nims;
907 	struct in6_msource	*lims;
908 	int			 schanged, error;
909 	int			 nsrc0, nsrc1;
910 
911 	schanged = 0;
912 	error = 0;
913 	nsrc1 = nsrc0 = 0;
914 
915 	/*
916 	 * Update the source filters first, as this may fail.
917 	 * Maintain count of in-mode filters at t0, t1. These are
918 	 * used to work out if we transition into ASM mode or not.
919 	 * Maintain a count of source filters whose state was
920 	 * actually modified by this operation.
921 	 */
922 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
923 		lims = (struct in6_msource *)ims;
924 		if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
925 		if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
926 		if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
927 		error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
928 		++schanged;
929 		if (error)
930 			break;
931 		im6s_merge(nims, lims, 0);
932 	}
933 	if (error) {
934 		struct ip6_msource *bims;
935 
936 		RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
937 			lims = (struct in6_msource *)ims;
938 			if (lims->im6sl_st[0] == lims->im6sl_st[1])
939 				continue;
940 			(void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
941 			if (bims == NULL)
942 				continue;
943 			im6s_merge(bims, lims, 1);
944 		}
945 		goto out_reap;
946 	}
947 
948 	CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
949 	    __func__, nsrc0, nsrc1);
950 
951 	/* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
952 	if (imf->im6f_st[0] == imf->im6f_st[1] &&
953 	    imf->im6f_st[1] == MCAST_INCLUDE) {
954 		if (nsrc1 == 0) {
955 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
956 			--inm->in6m_st[1].iss_in;
957 		}
958 	}
959 
960 	/* Handle filter mode transition on socket. */
961 	if (imf->im6f_st[0] != imf->im6f_st[1]) {
962 		CTR3(KTR_MLD, "%s: imf transition %d to %d",
963 		    __func__, imf->im6f_st[0], imf->im6f_st[1]);
964 
965 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
966 			CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
967 			--inm->in6m_st[1].iss_ex;
968 		} else if (imf->im6f_st[0] == MCAST_INCLUDE) {
969 			CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
970 			--inm->in6m_st[1].iss_in;
971 		}
972 
973 		if (imf->im6f_st[1] == MCAST_EXCLUDE) {
974 			CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
975 			inm->in6m_st[1].iss_ex++;
976 		} else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
977 			CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
978 			inm->in6m_st[1].iss_in++;
979 		}
980 	}
981 
982 	/*
983 	 * Track inm filter state in terms of listener counts.
984 	 * If there are any exclusive listeners, stack-wide
985 	 * membership is exclusive.
986 	 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
987 	 * If no listeners remain, state is undefined at t1,
988 	 * and the MLD lifecycle for this group should finish.
989 	 */
990 	if (inm->in6m_st[1].iss_ex > 0) {
991 		CTR1(KTR_MLD, "%s: transition to EX", __func__);
992 		inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
993 	} else if (inm->in6m_st[1].iss_in > 0) {
994 		CTR1(KTR_MLD, "%s: transition to IN", __func__);
995 		inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
996 	} else {
997 		CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
998 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
999 	}
1000 
1001 	/* Decrement ASM listener count on transition out of ASM mode. */
1002 	if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1003 		if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1004 		    (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1005 			CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1006 			--inm->in6m_st[1].iss_asm;
1007 		}
1008 	}
1009 
1010 	/* Increment ASM listener count on transition to ASM mode. */
1011 	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1012 		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1013 		inm->in6m_st[1].iss_asm++;
1014 	}
1015 
1016 	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1017 	in6m_print(inm);
1018 
1019 out_reap:
1020 	if (schanged > 0) {
1021 		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1022 		in6m_reap(inm);
1023 	}
1024 	return (error);
1025 }
1026 
1027 /*
1028  * Mark an in6_multi's filter set deltas as committed.
1029  * Called by MLD after a state change has been enqueued.
1030  */
1031 void
1032 in6m_commit(struct in6_multi *inm)
1033 {
1034 	struct ip6_msource	*ims;
1035 
1036 	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1037 	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1038 	in6m_print(inm);
1039 
1040 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1041 		ims->im6s_st[0] = ims->im6s_st[1];
1042 	}
1043 	inm->in6m_st[0] = inm->in6m_st[1];
1044 }
1045 
1046 /*
1047  * Reap unreferenced nodes from an in6_multi's filter set.
1048  */
1049 static void
1050 in6m_reap(struct in6_multi *inm)
1051 {
1052 	struct ip6_msource	*ims, *tims;
1053 
1054 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1055 		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1056 		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1057 		    ims->im6s_stp != 0)
1058 			continue;
1059 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1060 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1061 		free(ims, M_IP6MSOURCE);
1062 		inm->in6m_nsrc--;
1063 	}
1064 }
1065 
1066 /*
1067  * Purge all source nodes from an in6_multi's filter set.
1068  */
1069 static void
1070 in6m_purge(struct in6_multi *inm)
1071 {
1072 	struct ip6_msource	*ims, *tims;
1073 
1074 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1075 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1076 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1077 		free(ims, M_IP6MSOURCE);
1078 		inm->in6m_nsrc--;
1079 	}
1080 	/* Free state-change requests that might be queued. */
1081 	mbufq_drain(&inm->in6m_scq);
1082 }
1083 
1084 /*
1085  * Join a multicast address w/o sources.
1086  * KAME compatibility entry point.
1087  *
1088  * SMPng: Assume no mc locks held by caller.
1089  */
1090 struct in6_multi_mship *
1091 in6_joingroup(struct ifnet *ifp, struct in6_addr *mcaddr,
1092     int *errorp, int delay)
1093 {
1094 	struct in6_multi_mship *imm;
1095 	int error;
1096 
1097 	imm = malloc(sizeof(*imm), M_IP6MADDR, M_NOWAIT);
1098 	if (imm == NULL) {
1099 		*errorp = ENOBUFS;
1100 		return (NULL);
1101 	}
1102 
1103 	delay = (delay * PR_FASTHZ) / hz;
1104 
1105 	error = in6_mc_join(ifp, mcaddr, NULL, &imm->i6mm_maddr, delay);
1106 	if (error) {
1107 		*errorp = error;
1108 		free(imm, M_IP6MADDR);
1109 		return (NULL);
1110 	}
1111 
1112 	return (imm);
1113 }
1114 
1115 /*
1116  * Leave a multicast address w/o sources.
1117  * KAME compatibility entry point.
1118  *
1119  * SMPng: Assume no mc locks held by caller.
1120  */
1121 int
1122 in6_leavegroup(struct in6_multi_mship *imm)
1123 {
1124 
1125 	if (imm->i6mm_maddr != NULL)
1126 		in6_mc_leave(imm->i6mm_maddr, NULL);
1127 	free(imm,  M_IP6MADDR);
1128 	return 0;
1129 }
1130 
1131 /*
1132  * Join a multicast group; unlocked entry point.
1133  *
1134  * SMPng: XXX: in6_mc_join() is called from in6_control() when upper
1135  * locks are not held. Fortunately, ifp is unlikely to have been detached
1136  * at this point, so we assume it's OK to recurse.
1137  */
1138 int
1139 in6_mc_join(struct ifnet *ifp, const struct in6_addr *mcaddr,
1140     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1141     const int delay)
1142 {
1143 	int error;
1144 
1145 	IN6_MULTI_LOCK();
1146 	error = in6_mc_join_locked(ifp, mcaddr, imf, pinm, delay);
1147 	IN6_MULTI_UNLOCK();
1148 
1149 	return (error);
1150 }
1151 
1152 /*
1153  * Join a multicast group; real entry point.
1154  *
1155  * Only preserves atomicity at inm level.
1156  * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1157  *
1158  * If the MLD downcall fails, the group is not joined, and an error
1159  * code is returned.
1160  */
1161 int
1162 in6_mc_join_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1163     /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1164     const int delay)
1165 {
1166 	struct in6_mfilter	 timf;
1167 	struct in6_multi	*inm;
1168 	int			 error;
1169 #ifdef KTR
1170 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1171 #endif
1172 
1173 #ifdef INVARIANTS
1174 	/*
1175 	 * Sanity: Check scope zone ID was set for ifp, if and
1176 	 * only if group is scoped to an interface.
1177 	 */
1178 	KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1179 	    ("%s: not a multicast address", __func__));
1180 	if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1181 	    IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1182 		KASSERT(mcaddr->s6_addr16[1] != 0,
1183 		    ("%s: scope zone ID not set", __func__));
1184 	}
1185 #endif
1186 
1187 	IN6_MULTI_LOCK_ASSERT();
1188 
1189 	CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1190 	    ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1191 
1192 	error = 0;
1193 	inm = NULL;
1194 
1195 	/*
1196 	 * If no imf was specified (i.e. kernel consumer),
1197 	 * fake one up and assume it is an ASM join.
1198 	 */
1199 	if (imf == NULL) {
1200 		im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1201 		imf = &timf;
1202 	}
1203 
1204 	error = in6_mc_get(ifp, mcaddr, &inm);
1205 	if (error) {
1206 		CTR1(KTR_MLD, "%s: in6_mc_get() failure", __func__);
1207 		return (error);
1208 	}
1209 
1210 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1211 	error = in6m_merge(inm, imf);
1212 	if (error) {
1213 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1214 		goto out_in6m_release;
1215 	}
1216 
1217 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1218 	error = mld_change_state(inm, delay);
1219 	if (error) {
1220 		CTR1(KTR_MLD, "%s: failed to update source", __func__);
1221 		goto out_in6m_release;
1222 	}
1223 
1224 out_in6m_release:
1225 	if (error) {
1226 		CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1227 		in6m_release_locked(inm);
1228 	} else {
1229 		*pinm = inm;
1230 	}
1231 
1232 	return (error);
1233 }
1234 
1235 /*
1236  * Leave a multicast group; unlocked entry point.
1237  */
1238 int
1239 in6_mc_leave(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1240 {
1241 	struct ifnet *ifp;
1242 	int error;
1243 
1244 	ifp = inm->in6m_ifp;
1245 
1246 	IN6_MULTI_LOCK();
1247 	error = in6_mc_leave_locked(inm, imf);
1248 	IN6_MULTI_UNLOCK();
1249 
1250 	return (error);
1251 }
1252 
1253 /*
1254  * Leave a multicast group; real entry point.
1255  * All source filters will be expunged.
1256  *
1257  * Only preserves atomicity at inm level.
1258  *
1259  * Holding the write lock for the INP which contains imf
1260  * is highly advisable. We can't assert for it as imf does not
1261  * contain a back-pointer to the owning inp.
1262  *
1263  * Note: This is not the same as in6m_release(*) as this function also
1264  * makes a state change downcall into MLD.
1265  */
1266 int
1267 in6_mc_leave_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1268 {
1269 	struct in6_mfilter	 timf;
1270 	int			 error;
1271 #ifdef KTR
1272 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1273 #endif
1274 
1275 	error = 0;
1276 
1277 	IN6_MULTI_LOCK_ASSERT();
1278 
1279 	CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1280 	    inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1281 	    (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1282 	    imf);
1283 
1284 	/*
1285 	 * If no imf was specified (i.e. kernel consumer),
1286 	 * fake one up and assume it is an ASM join.
1287 	 */
1288 	if (imf == NULL) {
1289 		im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1290 		imf = &timf;
1291 	}
1292 
1293 	/*
1294 	 * Begin state merge transaction at MLD layer.
1295 	 *
1296 	 * As this particular invocation should not cause any memory
1297 	 * to be allocated, and there is no opportunity to roll back
1298 	 * the transaction, it MUST NOT fail.
1299 	 */
1300 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1301 	error = in6m_merge(inm, imf);
1302 	KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1303 
1304 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1305 	error = mld_change_state(inm, 0);
1306 	if (error)
1307 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1308 
1309 	CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1310 	in6m_release_locked(inm);
1311 
1312 	return (error);
1313 }
1314 
1315 /*
1316  * Block or unblock an ASM multicast source on an inpcb.
1317  * This implements the delta-based API described in RFC 3678.
1318  *
1319  * The delta-based API applies only to exclusive-mode memberships.
1320  * An MLD downcall will be performed.
1321  *
1322  * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1323  *
1324  * Return 0 if successful, otherwise return an appropriate error code.
1325  */
1326 static int
1327 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1328 {
1329 	struct group_source_req		 gsr;
1330 	sockunion_t			*gsa, *ssa;
1331 	struct ifnet			*ifp;
1332 	struct in6_mfilter		*imf;
1333 	struct ip6_moptions		*imo;
1334 	struct in6_msource		*ims;
1335 	struct in6_multi			*inm;
1336 	size_t				 idx;
1337 	uint16_t			 fmode;
1338 	int				 error, doblock;
1339 #ifdef KTR
1340 	char				 ip6tbuf[INET6_ADDRSTRLEN];
1341 #endif
1342 
1343 	ifp = NULL;
1344 	error = 0;
1345 	doblock = 0;
1346 
1347 	memset(&gsr, 0, sizeof(struct group_source_req));
1348 	gsa = (sockunion_t *)&gsr.gsr_group;
1349 	ssa = (sockunion_t *)&gsr.gsr_source;
1350 
1351 	switch (sopt->sopt_name) {
1352 	case MCAST_BLOCK_SOURCE:
1353 	case MCAST_UNBLOCK_SOURCE:
1354 		error = sooptcopyin(sopt, &gsr,
1355 		    sizeof(struct group_source_req),
1356 		    sizeof(struct group_source_req));
1357 		if (error)
1358 			return (error);
1359 
1360 		if (gsa->sin6.sin6_family != AF_INET6 ||
1361 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1362 			return (EINVAL);
1363 
1364 		if (ssa->sin6.sin6_family != AF_INET6 ||
1365 		    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1366 			return (EINVAL);
1367 
1368 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1369 			return (EADDRNOTAVAIL);
1370 
1371 		ifp = ifnet_byindex(gsr.gsr_interface);
1372 
1373 		if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1374 			doblock = 1;
1375 		break;
1376 
1377 	default:
1378 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1379 		    __func__, sopt->sopt_name);
1380 		return (EOPNOTSUPP);
1381 		break;
1382 	}
1383 
1384 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1385 		return (EINVAL);
1386 
1387 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1388 
1389 	/*
1390 	 * Check if we are actually a member of this group.
1391 	 */
1392 	imo = in6p_findmoptions(inp);
1393 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1394 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1395 		error = EADDRNOTAVAIL;
1396 		goto out_in6p_locked;
1397 	}
1398 
1399 	KASSERT(imo->im6o_mfilters != NULL,
1400 	    ("%s: im6o_mfilters not allocated", __func__));
1401 	imf = &imo->im6o_mfilters[idx];
1402 	inm = imo->im6o_membership[idx];
1403 
1404 	/*
1405 	 * Attempting to use the delta-based API on an
1406 	 * non exclusive-mode membership is an error.
1407 	 */
1408 	fmode = imf->im6f_st[0];
1409 	if (fmode != MCAST_EXCLUDE) {
1410 		error = EINVAL;
1411 		goto out_in6p_locked;
1412 	}
1413 
1414 	/*
1415 	 * Deal with error cases up-front:
1416 	 *  Asked to block, but already blocked; or
1417 	 *  Asked to unblock, but nothing to unblock.
1418 	 * If adding a new block entry, allocate it.
1419 	 */
1420 	ims = im6o_match_source(imo, idx, &ssa->sa);
1421 	if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1422 		CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1423 		    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1424 		    doblock ? "" : "not ");
1425 		error = EADDRNOTAVAIL;
1426 		goto out_in6p_locked;
1427 	}
1428 
1429 	INP_WLOCK_ASSERT(inp);
1430 
1431 	/*
1432 	 * Begin state merge transaction at socket layer.
1433 	 */
1434 	if (doblock) {
1435 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1436 		ims = im6f_graft(imf, fmode, &ssa->sin6);
1437 		if (ims == NULL)
1438 			error = ENOMEM;
1439 	} else {
1440 		CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1441 		error = im6f_prune(imf, &ssa->sin6);
1442 	}
1443 
1444 	if (error) {
1445 		CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1446 		goto out_im6f_rollback;
1447 	}
1448 
1449 	/*
1450 	 * Begin state merge transaction at MLD layer.
1451 	 */
1452 	IN6_MULTI_LOCK();
1453 
1454 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
1455 	error = in6m_merge(inm, imf);
1456 	if (error)
1457 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1458 	else {
1459 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1460 		error = mld_change_state(inm, 0);
1461 		if (error)
1462 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1463 	}
1464 
1465 	IN6_MULTI_UNLOCK();
1466 
1467 out_im6f_rollback:
1468 	if (error)
1469 		im6f_rollback(imf);
1470 	else
1471 		im6f_commit(imf);
1472 
1473 	im6f_reap(imf);
1474 
1475 out_in6p_locked:
1476 	INP_WUNLOCK(inp);
1477 	return (error);
1478 }
1479 
1480 /*
1481  * Given an inpcb, return its multicast options structure pointer.  Accepts
1482  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1483  *
1484  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1485  * SMPng: NOTE: Returns with the INP write lock held.
1486  */
1487 static struct ip6_moptions *
1488 in6p_findmoptions(struct inpcb *inp)
1489 {
1490 	struct ip6_moptions	 *imo;
1491 	struct in6_multi		**immp;
1492 	struct in6_mfilter	 *imfp;
1493 	size_t			  idx;
1494 
1495 	INP_WLOCK(inp);
1496 	if (inp->in6p_moptions != NULL)
1497 		return (inp->in6p_moptions);
1498 
1499 	INP_WUNLOCK(inp);
1500 
1501 	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1502 	immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1503 	    M_WAITOK | M_ZERO);
1504 	imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1505 	    M_IN6MFILTER, M_WAITOK);
1506 
1507 	imo->im6o_multicast_ifp = NULL;
1508 	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1509 	imo->im6o_multicast_loop = in6_mcast_loop;
1510 	imo->im6o_num_memberships = 0;
1511 	imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1512 	imo->im6o_membership = immp;
1513 
1514 	/* Initialize per-group source filters. */
1515 	for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1516 		im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1517 	imo->im6o_mfilters = imfp;
1518 
1519 	INP_WLOCK(inp);
1520 	if (inp->in6p_moptions != NULL) {
1521 		free(imfp, M_IN6MFILTER);
1522 		free(immp, M_IP6MOPTS);
1523 		free(imo, M_IP6MOPTS);
1524 		return (inp->in6p_moptions);
1525 	}
1526 	inp->in6p_moptions = imo;
1527 	return (imo);
1528 }
1529 
1530 /*
1531  * Discard the IPv6 multicast options (and source filters).
1532  *
1533  * SMPng: NOTE: assumes INP write lock is held.
1534  */
1535 void
1536 ip6_freemoptions(struct ip6_moptions *imo)
1537 {
1538 	struct in6_mfilter	*imf;
1539 	size_t			 idx, nmships;
1540 
1541 	KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1542 
1543 	nmships = imo->im6o_num_memberships;
1544 	for (idx = 0; idx < nmships; ++idx) {
1545 		imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1546 		if (imf)
1547 			im6f_leave(imf);
1548 		/* XXX this will thrash the lock(s) */
1549 		(void)in6_mc_leave(imo->im6o_membership[idx], imf);
1550 		if (imf)
1551 			im6f_purge(imf);
1552 	}
1553 
1554 	if (imo->im6o_mfilters)
1555 		free(imo->im6o_mfilters, M_IN6MFILTER);
1556 	free(imo->im6o_membership, M_IP6MOPTS);
1557 	free(imo, M_IP6MOPTS);
1558 }
1559 
1560 /*
1561  * Atomically get source filters on a socket for an IPv6 multicast group.
1562  * Called with INP lock held; returns with lock released.
1563  */
1564 static int
1565 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1566 {
1567 	struct __msfilterreq	 msfr;
1568 	sockunion_t		*gsa;
1569 	struct ifnet		*ifp;
1570 	struct ip6_moptions	*imo;
1571 	struct in6_mfilter	*imf;
1572 	struct ip6_msource	*ims;
1573 	struct in6_msource	*lims;
1574 	struct sockaddr_in6	*psin;
1575 	struct sockaddr_storage	*ptss;
1576 	struct sockaddr_storage	*tss;
1577 	int			 error;
1578 	size_t			 idx, nsrcs, ncsrcs;
1579 
1580 	INP_WLOCK_ASSERT(inp);
1581 
1582 	imo = inp->in6p_moptions;
1583 	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1584 
1585 	INP_WUNLOCK(inp);
1586 
1587 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1588 	    sizeof(struct __msfilterreq));
1589 	if (error)
1590 		return (error);
1591 
1592 	if (msfr.msfr_group.ss_family != AF_INET6 ||
1593 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1594 		return (EINVAL);
1595 
1596 	gsa = (sockunion_t *)&msfr.msfr_group;
1597 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1598 		return (EINVAL);
1599 
1600 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1601 		return (EADDRNOTAVAIL);
1602 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1603 	if (ifp == NULL)
1604 		return (EADDRNOTAVAIL);
1605 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1606 
1607 	INP_WLOCK(inp);
1608 
1609 	/*
1610 	 * Lookup group on the socket.
1611 	 */
1612 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1613 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1614 		INP_WUNLOCK(inp);
1615 		return (EADDRNOTAVAIL);
1616 	}
1617 	imf = &imo->im6o_mfilters[idx];
1618 
1619 	/*
1620 	 * Ignore memberships which are in limbo.
1621 	 */
1622 	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1623 		INP_WUNLOCK(inp);
1624 		return (EAGAIN);
1625 	}
1626 	msfr.msfr_fmode = imf->im6f_st[1];
1627 
1628 	/*
1629 	 * If the user specified a buffer, copy out the source filter
1630 	 * entries to userland gracefully.
1631 	 * We only copy out the number of entries which userland
1632 	 * has asked for, but we always tell userland how big the
1633 	 * buffer really needs to be.
1634 	 */
1635 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1636 		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1637 	tss = NULL;
1638 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1639 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1640 		    M_TEMP, M_NOWAIT | M_ZERO);
1641 		if (tss == NULL) {
1642 			INP_WUNLOCK(inp);
1643 			return (ENOBUFS);
1644 		}
1645 	}
1646 
1647 	/*
1648 	 * Count number of sources in-mode at t0.
1649 	 * If buffer space exists and remains, copy out source entries.
1650 	 */
1651 	nsrcs = msfr.msfr_nsrcs;
1652 	ncsrcs = 0;
1653 	ptss = tss;
1654 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1655 		lims = (struct in6_msource *)ims;
1656 		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1657 		    lims->im6sl_st[0] != imf->im6f_st[0])
1658 			continue;
1659 		++ncsrcs;
1660 		if (tss != NULL && nsrcs > 0) {
1661 			psin = (struct sockaddr_in6 *)ptss;
1662 			psin->sin6_family = AF_INET6;
1663 			psin->sin6_len = sizeof(struct sockaddr_in6);
1664 			psin->sin6_addr = lims->im6s_addr;
1665 			psin->sin6_port = 0;
1666 			--nsrcs;
1667 			++ptss;
1668 		}
1669 	}
1670 
1671 	INP_WUNLOCK(inp);
1672 
1673 	if (tss != NULL) {
1674 		error = copyout(tss, msfr.msfr_srcs,
1675 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1676 		free(tss, M_TEMP);
1677 		if (error)
1678 			return (error);
1679 	}
1680 
1681 	msfr.msfr_nsrcs = ncsrcs;
1682 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1683 
1684 	return (error);
1685 }
1686 
1687 /*
1688  * Return the IP multicast options in response to user getsockopt().
1689  */
1690 int
1691 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1692 {
1693 	struct ip6_moptions	*im6o;
1694 	int			 error;
1695 	u_int			 optval;
1696 
1697 	INP_WLOCK(inp);
1698 	im6o = inp->in6p_moptions;
1699 	/*
1700 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1701 	 * or is a divert socket, reject it.
1702 	 */
1703 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1704 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1705 	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1706 		INP_WUNLOCK(inp);
1707 		return (EOPNOTSUPP);
1708 	}
1709 
1710 	error = 0;
1711 	switch (sopt->sopt_name) {
1712 	case IPV6_MULTICAST_IF:
1713 		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1714 			optval = 0;
1715 		} else {
1716 			optval = im6o->im6o_multicast_ifp->if_index;
1717 		}
1718 		INP_WUNLOCK(inp);
1719 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1720 		break;
1721 
1722 	case IPV6_MULTICAST_HOPS:
1723 		if (im6o == NULL)
1724 			optval = V_ip6_defmcasthlim;
1725 		else
1726 			optval = im6o->im6o_multicast_hlim;
1727 		INP_WUNLOCK(inp);
1728 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1729 		break;
1730 
1731 	case IPV6_MULTICAST_LOOP:
1732 		if (im6o == NULL)
1733 			optval = in6_mcast_loop; /* XXX VIMAGE */
1734 		else
1735 			optval = im6o->im6o_multicast_loop;
1736 		INP_WUNLOCK(inp);
1737 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1738 		break;
1739 
1740 	case IPV6_MSFILTER:
1741 		if (im6o == NULL) {
1742 			error = EADDRNOTAVAIL;
1743 			INP_WUNLOCK(inp);
1744 		} else {
1745 			error = in6p_get_source_filters(inp, sopt);
1746 		}
1747 		break;
1748 
1749 	default:
1750 		INP_WUNLOCK(inp);
1751 		error = ENOPROTOOPT;
1752 		break;
1753 	}
1754 
1755 	INP_UNLOCK_ASSERT(inp);
1756 
1757 	return (error);
1758 }
1759 
1760 /*
1761  * Look up the ifnet to use for a multicast group membership,
1762  * given the address of an IPv6 group.
1763  *
1764  * This routine exists to support legacy IPv6 multicast applications.
1765  *
1766  * If inp is non-NULL, use this socket's current FIB number for any
1767  * required FIB lookup. Look up the group address in the unicast FIB,
1768  * and use its ifp; usually, this points to the default next-hop.
1769  * If the FIB lookup fails, return NULL.
1770  *
1771  * FUTURE: Support multiple forwarding tables for IPv6.
1772  *
1773  * Returns NULL if no ifp could be found.
1774  */
1775 static struct ifnet *
1776 in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1777     const struct sockaddr_in6 *gsin6)
1778 {
1779 	struct nhop6_basic	nh6;
1780 	struct in6_addr		dst;
1781 	uint32_t		scopeid;
1782 	uint32_t		fibnum;
1783 
1784 	KASSERT(in6p->inp_vflag & INP_IPV6,
1785 	    ("%s: not INP_IPV6 inpcb", __func__));
1786 	KASSERT(gsin6->sin6_family == AF_INET6,
1787 	    ("%s: not AF_INET6 group", __func__));
1788 
1789 	in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1790 	fibnum = in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB;
1791 	if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6) != 0)
1792 		return (NULL);
1793 
1794 	return (nh6.nh_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 	if (ifindex == 0)
2352 		ifp = NULL;
2353 	else {
2354 		ifp = ifnet_byindex(ifindex);
2355 		if (ifp == NULL)
2356 			return (EINVAL);
2357 		if ((ifp->if_flags & IFF_MULTICAST) == 0)
2358 			return (EADDRNOTAVAIL);
2359 	}
2360 	imo = in6p_findmoptions(inp);
2361 	imo->im6o_multicast_ifp = ifp;
2362 	INP_WUNLOCK(inp);
2363 
2364 	return (0);
2365 }
2366 
2367 /*
2368  * Atomically set source filters on a socket for an IPv6 multicast group.
2369  *
2370  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2371  */
2372 static int
2373 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2374 {
2375 	struct __msfilterreq	 msfr;
2376 	sockunion_t		*gsa;
2377 	struct ifnet		*ifp;
2378 	struct in6_mfilter	*imf;
2379 	struct ip6_moptions	*imo;
2380 	struct in6_multi		*inm;
2381 	size_t			 idx;
2382 	int			 error;
2383 
2384 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2385 	    sizeof(struct __msfilterreq));
2386 	if (error)
2387 		return (error);
2388 
2389 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2390 		return (ENOBUFS);
2391 
2392 	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2393 	    msfr.msfr_fmode != MCAST_INCLUDE)
2394 		return (EINVAL);
2395 
2396 	if (msfr.msfr_group.ss_family != AF_INET6 ||
2397 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2398 		return (EINVAL);
2399 
2400 	gsa = (sockunion_t *)&msfr.msfr_group;
2401 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2402 		return (EINVAL);
2403 
2404 	gsa->sin6.sin6_port = 0;	/* ignore port */
2405 
2406 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2407 		return (EADDRNOTAVAIL);
2408 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2409 	if (ifp == NULL)
2410 		return (EADDRNOTAVAIL);
2411 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2412 
2413 	/*
2414 	 * Take the INP write lock.
2415 	 * Check if this socket is a member of this group.
2416 	 */
2417 	imo = in6p_findmoptions(inp);
2418 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2419 	if (idx == -1 || imo->im6o_mfilters == NULL) {
2420 		error = EADDRNOTAVAIL;
2421 		goto out_in6p_locked;
2422 	}
2423 	inm = imo->im6o_membership[idx];
2424 	imf = &imo->im6o_mfilters[idx];
2425 
2426 	/*
2427 	 * Begin state merge transaction at socket layer.
2428 	 */
2429 	INP_WLOCK_ASSERT(inp);
2430 
2431 	imf->im6f_st[1] = msfr.msfr_fmode;
2432 
2433 	/*
2434 	 * Apply any new source filters, if present.
2435 	 * Make a copy of the user-space source vector so
2436 	 * that we may copy them with a single copyin. This
2437 	 * allows us to deal with page faults up-front.
2438 	 */
2439 	if (msfr.msfr_nsrcs > 0) {
2440 		struct in6_msource	*lims;
2441 		struct sockaddr_in6	*psin;
2442 		struct sockaddr_storage	*kss, *pkss;
2443 		int			 i;
2444 
2445 		INP_WUNLOCK(inp);
2446 
2447 		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2448 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2449 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2450 		    M_TEMP, M_WAITOK);
2451 		error = copyin(msfr.msfr_srcs, kss,
2452 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2453 		if (error) {
2454 			free(kss, M_TEMP);
2455 			return (error);
2456 		}
2457 
2458 		INP_WLOCK(inp);
2459 
2460 		/*
2461 		 * Mark all source filters as UNDEFINED at t1.
2462 		 * Restore new group filter mode, as im6f_leave()
2463 		 * will set it to INCLUDE.
2464 		 */
2465 		im6f_leave(imf);
2466 		imf->im6f_st[1] = msfr.msfr_fmode;
2467 
2468 		/*
2469 		 * Update socket layer filters at t1, lazy-allocating
2470 		 * new entries. This saves a bunch of memory at the
2471 		 * cost of one RB_FIND() per source entry; duplicate
2472 		 * entries in the msfr_nsrcs vector are ignored.
2473 		 * If we encounter an error, rollback transaction.
2474 		 *
2475 		 * XXX This too could be replaced with a set-symmetric
2476 		 * difference like loop to avoid walking from root
2477 		 * every time, as the key space is common.
2478 		 */
2479 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2480 			psin = (struct sockaddr_in6 *)pkss;
2481 			if (psin->sin6_family != AF_INET6) {
2482 				error = EAFNOSUPPORT;
2483 				break;
2484 			}
2485 			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2486 				error = EINVAL;
2487 				break;
2488 			}
2489 			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2490 				error = EINVAL;
2491 				break;
2492 			}
2493 			/*
2494 			 * TODO: Validate embedded scope ID in source
2495 			 * list entry against passed-in ifp, if and only
2496 			 * if source list filter entry is iface or node local.
2497 			 */
2498 			in6_clearscope(&psin->sin6_addr);
2499 			error = im6f_get_source(imf, psin, &lims);
2500 			if (error)
2501 				break;
2502 			lims->im6sl_st[1] = imf->im6f_st[1];
2503 		}
2504 		free(kss, M_TEMP);
2505 	}
2506 
2507 	if (error)
2508 		goto out_im6f_rollback;
2509 
2510 	INP_WLOCK_ASSERT(inp);
2511 	IN6_MULTI_LOCK();
2512 
2513 	/*
2514 	 * Begin state merge transaction at MLD layer.
2515 	 */
2516 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2517 	error = in6m_merge(inm, imf);
2518 	if (error)
2519 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2520 	else {
2521 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2522 		error = mld_change_state(inm, 0);
2523 		if (error)
2524 			CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2525 	}
2526 
2527 	IN6_MULTI_UNLOCK();
2528 
2529 out_im6f_rollback:
2530 	if (error)
2531 		im6f_rollback(imf);
2532 	else
2533 		im6f_commit(imf);
2534 
2535 	im6f_reap(imf);
2536 
2537 out_in6p_locked:
2538 	INP_WUNLOCK(inp);
2539 	return (error);
2540 }
2541 
2542 /*
2543  * Set the IP multicast options in response to user setsockopt().
2544  *
2545  * Many of the socket options handled in this function duplicate the
2546  * functionality of socket options in the regular unicast API. However,
2547  * it is not possible to merge the duplicate code, because the idempotence
2548  * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2549  * the effects of these options must be treated as separate and distinct.
2550  *
2551  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2552  */
2553 int
2554 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2555 {
2556 	struct ip6_moptions	*im6o;
2557 	int			 error;
2558 
2559 	error = 0;
2560 
2561 	/*
2562 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2563 	 * or is a divert socket, reject it.
2564 	 */
2565 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2566 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2567 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2568 		return (EOPNOTSUPP);
2569 
2570 	switch (sopt->sopt_name) {
2571 	case IPV6_MULTICAST_IF:
2572 		error = in6p_set_multicast_if(inp, sopt);
2573 		break;
2574 
2575 	case IPV6_MULTICAST_HOPS: {
2576 		int hlim;
2577 
2578 		if (sopt->sopt_valsize != sizeof(int)) {
2579 			error = EINVAL;
2580 			break;
2581 		}
2582 		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2583 		if (error)
2584 			break;
2585 		if (hlim < -1 || hlim > 255) {
2586 			error = EINVAL;
2587 			break;
2588 		} else if (hlim == -1) {
2589 			hlim = V_ip6_defmcasthlim;
2590 		}
2591 		im6o = in6p_findmoptions(inp);
2592 		im6o->im6o_multicast_hlim = hlim;
2593 		INP_WUNLOCK(inp);
2594 		break;
2595 	}
2596 
2597 	case IPV6_MULTICAST_LOOP: {
2598 		u_int loop;
2599 
2600 		/*
2601 		 * Set the loopback flag for outgoing multicast packets.
2602 		 * Must be zero or one.
2603 		 */
2604 		if (sopt->sopt_valsize != sizeof(u_int)) {
2605 			error = EINVAL;
2606 			break;
2607 		}
2608 		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2609 		if (error)
2610 			break;
2611 		if (loop > 1) {
2612 			error = EINVAL;
2613 			break;
2614 		}
2615 		im6o = in6p_findmoptions(inp);
2616 		im6o->im6o_multicast_loop = loop;
2617 		INP_WUNLOCK(inp);
2618 		break;
2619 	}
2620 
2621 	case IPV6_JOIN_GROUP:
2622 	case MCAST_JOIN_GROUP:
2623 	case MCAST_JOIN_SOURCE_GROUP:
2624 		error = in6p_join_group(inp, sopt);
2625 		break;
2626 
2627 	case IPV6_LEAVE_GROUP:
2628 	case MCAST_LEAVE_GROUP:
2629 	case MCAST_LEAVE_SOURCE_GROUP:
2630 		error = in6p_leave_group(inp, sopt);
2631 		break;
2632 
2633 	case MCAST_BLOCK_SOURCE:
2634 	case MCAST_UNBLOCK_SOURCE:
2635 		error = in6p_block_unblock_source(inp, sopt);
2636 		break;
2637 
2638 	case IPV6_MSFILTER:
2639 		error = in6p_set_source_filters(inp, sopt);
2640 		break;
2641 
2642 	default:
2643 		error = EOPNOTSUPP;
2644 		break;
2645 	}
2646 
2647 	INP_UNLOCK_ASSERT(inp);
2648 
2649 	return (error);
2650 }
2651 
2652 /*
2653  * Expose MLD's multicast filter mode and source list(s) to userland,
2654  * keyed by (ifindex, group).
2655  * The filter mode is written out as a uint32_t, followed by
2656  * 0..n of struct in6_addr.
2657  * For use by ifmcstat(8).
2658  * SMPng: NOTE: unlocked read of ifindex space.
2659  */
2660 static int
2661 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2662 {
2663 	struct in6_addr			 mcaddr;
2664 	struct in6_addr			 src;
2665 	struct ifnet			*ifp;
2666 	struct ifmultiaddr		*ifma;
2667 	struct in6_multi		*inm;
2668 	struct ip6_msource		*ims;
2669 	int				*name;
2670 	int				 retval;
2671 	u_int				 namelen;
2672 	uint32_t			 fmode, ifindex;
2673 #ifdef KTR
2674 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2675 #endif
2676 
2677 	name = (int *)arg1;
2678 	namelen = arg2;
2679 
2680 	if (req->newptr != NULL)
2681 		return (EPERM);
2682 
2683 	/* int: ifindex + 4 * 32 bits of IPv6 address */
2684 	if (namelen != 5)
2685 		return (EINVAL);
2686 
2687 	ifindex = name[0];
2688 	if (ifindex <= 0 || ifindex > V_if_index) {
2689 		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2690 		    __func__, ifindex);
2691 		return (ENOENT);
2692 	}
2693 
2694 	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2695 	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2696 		CTR2(KTR_MLD, "%s: group %s is not multicast",
2697 		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2698 		return (EINVAL);
2699 	}
2700 
2701 	ifp = ifnet_byindex(ifindex);
2702 	if (ifp == NULL) {
2703 		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2704 		    __func__, ifindex);
2705 		return (ENOENT);
2706 	}
2707 	/*
2708 	 * Internal MLD lookups require that scope/zone ID is set.
2709 	 */
2710 	(void)in6_setscope(&mcaddr, ifp, NULL);
2711 
2712 	retval = sysctl_wire_old_buffer(req,
2713 	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2714 	if (retval)
2715 		return (retval);
2716 
2717 	IN6_MULTI_LOCK();
2718 
2719 	IF_ADDR_RLOCK(ifp);
2720 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2721 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2722 		    ifma->ifma_protospec == NULL)
2723 			continue;
2724 		inm = (struct in6_multi *)ifma->ifma_protospec;
2725 		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2726 			continue;
2727 		fmode = inm->in6m_st[1].iss_fmode;
2728 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2729 		if (retval != 0)
2730 			break;
2731 		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2732 			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2733 			/*
2734 			 * Only copy-out sources which are in-mode.
2735 			 */
2736 			if (fmode != im6s_get_mode(inm, ims, 1)) {
2737 				CTR1(KTR_MLD, "%s: skip non-in-mode",
2738 				    __func__);
2739 				continue;
2740 			}
2741 			src = ims->im6s_addr;
2742 			retval = SYSCTL_OUT(req, &src,
2743 			    sizeof(struct in6_addr));
2744 			if (retval != 0)
2745 				break;
2746 		}
2747 	}
2748 	IF_ADDR_RUNLOCK(ifp);
2749 
2750 	IN6_MULTI_UNLOCK();
2751 
2752 	return (retval);
2753 }
2754 
2755 #ifdef KTR
2756 
2757 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2758 
2759 static const char *
2760 in6m_mode_str(const int mode)
2761 {
2762 
2763 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2764 		return (in6m_modestrs[mode]);
2765 	return ("??");
2766 }
2767 
2768 static const char *in6m_statestrs[] = {
2769 	"not-member",
2770 	"silent",
2771 	"idle",
2772 	"lazy",
2773 	"sleeping",
2774 	"awakening",
2775 	"query-pending",
2776 	"sg-query-pending",
2777 	"leaving"
2778 };
2779 
2780 static const char *
2781 in6m_state_str(const int state)
2782 {
2783 
2784 	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2785 		return (in6m_statestrs[state]);
2786 	return ("??");
2787 }
2788 
2789 /*
2790  * Dump an in6_multi structure to the console.
2791  */
2792 void
2793 in6m_print(const struct in6_multi *inm)
2794 {
2795 	int t;
2796 	char ip6tbuf[INET6_ADDRSTRLEN];
2797 
2798 	if ((ktr_mask & KTR_MLD) == 0)
2799 		return;
2800 
2801 	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2802 	printf("addr %s ifp %p(%s) ifma %p\n",
2803 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2804 	    inm->in6m_ifp,
2805 	    if_name(inm->in6m_ifp),
2806 	    inm->in6m_ifma);
2807 	printf("timer %u state %s refcount %u scq.len %u\n",
2808 	    inm->in6m_timer,
2809 	    in6m_state_str(inm->in6m_state),
2810 	    inm->in6m_refcount,
2811 	    mbufq_len(&inm->in6m_scq));
2812 	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2813 	    inm->in6m_mli,
2814 	    inm->in6m_nsrc,
2815 	    inm->in6m_sctimer,
2816 	    inm->in6m_scrv);
2817 	for (t = 0; t < 2; t++) {
2818 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2819 		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2820 		    inm->in6m_st[t].iss_asm,
2821 		    inm->in6m_st[t].iss_ex,
2822 		    inm->in6m_st[t].iss_in,
2823 		    inm->in6m_st[t].iss_rec);
2824 	}
2825 	printf("%s: --- end in6m %p ---\n", __func__, inm);
2826 }
2827 
2828 #else /* !KTR */
2829 
2830 void
2831 in6m_print(const struct in6_multi *inm)
2832 {
2833 
2834 }
2835 
2836 #endif /* KTR */
2837