xref: /freebsd/sys/netinet6/in6_mcast.c (revision 7661de35d15f582ab33e3bd6b8d909601557e436)
1 /*
2  * Copyright (c) 2009 Bruce Simpson.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. The name of the author may not be used to endorse or promote
14  *    products derived from this software without specific prior written
15  *    permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /*
31  * IPv6 multicast socket, group, and socket option processing module.
32  * Normative references: RFC 2292, RFC 3492, RFC 3542, RFC 3678, RFC 3810.
33  */
34 
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD$");
37 
38 #include "opt_inet6.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/protosw.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/protosw.h>
49 #include <sys/sysctl.h>
50 #include <sys/priv.h>
51 #include <sys/ktr.h>
52 #include <sys/tree.h>
53 
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/if_dl.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_var.h>
62 #include <netinet6/in6_var.h>
63 #include <netinet/ip6.h>
64 #include <netinet/icmp6.h>
65 #include <netinet6/ip6_var.h>
66 #include <netinet/in_pcb.h>
67 #include <netinet/tcp_var.h>
68 #include <netinet6/nd6.h>
69 #include <netinet6/mld6_var.h>
70 #include <netinet6/scope6_var.h>
71 
72 #ifndef KTR_MLD
73 #define KTR_MLD KTR_INET6
74 #endif
75 
76 #ifndef __SOCKUNION_DECLARED
77 union sockunion {
78 	struct sockaddr_storage	ss;
79 	struct sockaddr		sa;
80 	struct sockaddr_dl	sdl;
81 	struct sockaddr_in6	sin6;
82 };
83 typedef union sockunion sockunion_t;
84 #define __SOCKUNION_DECLARED
85 #endif /* __SOCKUNION_DECLARED */
86 
87 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
88     "IPv6 multicast PCB-layer source filter");
89 static MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
90 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
91 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
92     "IPv6 multicast MLD-layer source filter");
93 
94 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
95 
96 /*
97  * Locking:
98  * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
99  * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
100  *   it can be taken by code in net/if.c also.
101  * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
102  *
103  * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
104  * any need for in6_multi itself to be virtualized -- it is bound to an ifp
105  * anyway no matter what happens.
106  */
107 struct mtx in6_multi_mtx;
108 MTX_SYSINIT(in6_multi_mtx, &in6_multi_mtx, "in6_multi_mtx", MTX_DEF);
109 
110 static void	im6f_commit(struct in6_mfilter *);
111 static int	im6f_get_source(struct in6_mfilter *imf,
112 		    const struct sockaddr_in6 *psin,
113 		    struct in6_msource **);
114 static struct in6_msource *
115 		im6f_graft(struct in6_mfilter *, const uint8_t,
116 		    const struct sockaddr_in6 *);
117 static void	im6f_leave(struct in6_mfilter *);
118 static int	im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
119 static void	im6f_purge(struct in6_mfilter *);
120 static void	im6f_rollback(struct in6_mfilter *);
121 static void	im6f_reap(struct in6_mfilter *);
122 static int	im6o_grow(struct ip6_moptions *);
123 static size_t	im6o_match_group(const struct ip6_moptions *,
124 		    const struct ifnet *, const struct sockaddr *);
125 static struct in6_msource *
126 		im6o_match_source(const struct ip6_moptions *, const size_t,
127 		    const struct sockaddr *);
128 static void	im6s_merge(struct ip6_msource *ims,
129 		    const struct in6_msource *lims, const int rollback);
130 static int	in6_mc_get(struct ifnet *, const struct in6_addr *,
131 		    struct in6_multi **);
132 static int	in6m_get_source(struct in6_multi *inm,
133 		    const struct in6_addr *addr, const int noalloc,
134 		    struct ip6_msource **pims);
135 #ifdef KTR
136 static int	in6m_is_ifp_detached(const struct in6_multi *);
137 #endif
138 static int	in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
139 static void	in6m_purge(struct in6_multi *);
140 static void	in6m_reap(struct in6_multi *);
141 static struct ip6_moptions *
142 		in6p_findmoptions(struct inpcb *);
143 static int	in6p_get_source_filters(struct inpcb *, struct sockopt *);
144 static int	in6p_join_group(struct inpcb *, struct sockopt *);
145 static int	in6p_leave_group(struct inpcb *, struct sockopt *);
146 static struct ifnet *
147 		in6p_lookup_mcast_ifp(const struct inpcb *,
148 		    const struct sockaddr_in6 *);
149 static int	in6p_block_unblock_source(struct inpcb *, struct sockopt *);
150 static int	in6p_set_multicast_if(struct inpcb *, struct sockopt *);
151 static int	in6p_set_source_filters(struct inpcb *, struct sockopt *);
152 static int	sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
153 
154 SYSCTL_DECL(_net_inet6_ip6);	/* XXX Not in any common header. */
155 
156 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0,
157     "IPv6 multicast");
158 
159 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
160 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
161     CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxgrpsrc, 0,
162     "Max source filters per group");
163 TUNABLE_ULONG("net.inet6.ip6.mcast.maxgrpsrc", &in6_mcast_maxgrpsrc);
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_RW | CTLFLAG_TUN, &in6_mcast_maxsocksrc, 0,
168     "Max source filters per socket");
169 TUNABLE_ULONG("net.inet6.ip6.mcast.maxsocksrc", &in6_mcast_maxsocksrc);
170 
171 /* TODO Virtualize this switch. */
172 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
173 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
174     &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
175 TUNABLE_INT("net.inet6.ip6.mcast.loop", &in6_mcast_loop);
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_delmulti_ifma(ifma);
476 		error = ENOMEM;
477 		goto out_locked;
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 	IFQ_SET_MAXLEN(&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 occured (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 	/* Increment ASM listener count on transition to ASM mode. */
1010 	if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1011 		CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1012 		inm->in6m_st[1].iss_asm++;
1013 	}
1014 
1015 	CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1016 	in6m_print(inm);
1017 
1018 out_reap:
1019 	if (schanged > 0) {
1020 		CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1021 		in6m_reap(inm);
1022 	}
1023 	return (error);
1024 }
1025 
1026 /*
1027  * Mark an in6_multi's filter set deltas as committed.
1028  * Called by MLD after a state change has been enqueued.
1029  */
1030 void
1031 in6m_commit(struct in6_multi *inm)
1032 {
1033 	struct ip6_msource	*ims;
1034 
1035 	CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1036 	CTR1(KTR_MLD, "%s: pre commit:", __func__);
1037 	in6m_print(inm);
1038 
1039 	RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1040 		ims->im6s_st[0] = ims->im6s_st[1];
1041 	}
1042 	inm->in6m_st[0] = inm->in6m_st[1];
1043 }
1044 
1045 /*
1046  * Reap unreferenced nodes from an in6_multi's filter set.
1047  */
1048 static void
1049 in6m_reap(struct in6_multi *inm)
1050 {
1051 	struct ip6_msource	*ims, *tims;
1052 
1053 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1054 		if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1055 		    ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1056 		    ims->im6s_stp != 0)
1057 			continue;
1058 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1059 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1060 		free(ims, M_IP6MSOURCE);
1061 		inm->in6m_nsrc--;
1062 	}
1063 }
1064 
1065 /*
1066  * Purge all source nodes from an in6_multi's filter set.
1067  */
1068 static void
1069 in6m_purge(struct in6_multi *inm)
1070 {
1071 	struct ip6_msource	*ims, *tims;
1072 
1073 	RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1074 		CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1075 		RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1076 		free(ims, M_IP6MSOURCE);
1077 		inm->in6m_nsrc--;
1078 	}
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 		goto out_im6f_rollback;
1456 	}
1457 
1458 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1459 	error = mld_change_state(inm, 0);
1460 	if (error)
1461 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1462 
1463 	IN6_MULTI_UNLOCK();
1464 
1465 out_im6f_rollback:
1466 	if (error)
1467 		im6f_rollback(imf);
1468 	else
1469 		im6f_commit(imf);
1470 
1471 	im6f_reap(imf);
1472 
1473 out_in6p_locked:
1474 	INP_WUNLOCK(inp);
1475 	return (error);
1476 }
1477 
1478 /*
1479  * Given an inpcb, return its multicast options structure pointer.  Accepts
1480  * an unlocked inpcb pointer, but will return it locked.  May sleep.
1481  *
1482  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1483  * SMPng: NOTE: Returns with the INP write lock held.
1484  */
1485 static struct ip6_moptions *
1486 in6p_findmoptions(struct inpcb *inp)
1487 {
1488 	struct ip6_moptions	 *imo;
1489 	struct in6_multi		**immp;
1490 	struct in6_mfilter	 *imfp;
1491 	size_t			  idx;
1492 
1493 	INP_WLOCK(inp);
1494 	if (inp->in6p_moptions != NULL)
1495 		return (inp->in6p_moptions);
1496 
1497 	INP_WUNLOCK(inp);
1498 
1499 	imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1500 	immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1501 	    M_WAITOK | M_ZERO);
1502 	imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1503 	    M_IN6MFILTER, M_WAITOK);
1504 
1505 	imo->im6o_multicast_ifp = NULL;
1506 	imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1507 	imo->im6o_multicast_loop = in6_mcast_loop;
1508 	imo->im6o_num_memberships = 0;
1509 	imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1510 	imo->im6o_membership = immp;
1511 
1512 	/* Initialize per-group source filters. */
1513 	for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1514 		im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1515 	imo->im6o_mfilters = imfp;
1516 
1517 	INP_WLOCK(inp);
1518 	if (inp->in6p_moptions != NULL) {
1519 		free(imfp, M_IN6MFILTER);
1520 		free(immp, M_IP6MOPTS);
1521 		free(imo, M_IP6MOPTS);
1522 		return (inp->in6p_moptions);
1523 	}
1524 	inp->in6p_moptions = imo;
1525 	return (imo);
1526 }
1527 
1528 /*
1529  * Discard the IPv6 multicast options (and source filters).
1530  *
1531  * SMPng: NOTE: assumes INP write lock is held.
1532  */
1533 void
1534 ip6_freemoptions(struct ip6_moptions *imo)
1535 {
1536 	struct in6_mfilter	*imf;
1537 	size_t			 idx, nmships;
1538 
1539 	KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1540 
1541 	nmships = imo->im6o_num_memberships;
1542 	for (idx = 0; idx < nmships; ++idx) {
1543 		imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1544 		if (imf)
1545 			im6f_leave(imf);
1546 		/* XXX this will thrash the lock(s) */
1547 		(void)in6_mc_leave(imo->im6o_membership[idx], imf);
1548 		if (imf)
1549 			im6f_purge(imf);
1550 	}
1551 
1552 	if (imo->im6o_mfilters)
1553 		free(imo->im6o_mfilters, M_IN6MFILTER);
1554 	free(imo->im6o_membership, M_IP6MOPTS);
1555 	free(imo, M_IP6MOPTS);
1556 }
1557 
1558 /*
1559  * Atomically get source filters on a socket for an IPv6 multicast group.
1560  * Called with INP lock held; returns with lock released.
1561  */
1562 static int
1563 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1564 {
1565 	struct __msfilterreq	 msfr;
1566 	sockunion_t		*gsa;
1567 	struct ifnet		*ifp;
1568 	struct ip6_moptions	*imo;
1569 	struct in6_mfilter	*imf;
1570 	struct ip6_msource	*ims;
1571 	struct in6_msource	*lims;
1572 	struct sockaddr_in6	*psin;
1573 	struct sockaddr_storage	*ptss;
1574 	struct sockaddr_storage	*tss;
1575 	int			 error;
1576 	size_t			 idx, nsrcs, ncsrcs;
1577 
1578 	INP_WLOCK_ASSERT(inp);
1579 
1580 	imo = inp->in6p_moptions;
1581 	KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1582 
1583 	INP_WUNLOCK(inp);
1584 
1585 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1586 	    sizeof(struct __msfilterreq));
1587 	if (error)
1588 		return (error);
1589 
1590 	if (msfr.msfr_group.ss_family != AF_INET6 ||
1591 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1592 		return (EINVAL);
1593 
1594 	gsa = (sockunion_t *)&msfr.msfr_group;
1595 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1596 		return (EINVAL);
1597 
1598 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1599 		return (EADDRNOTAVAIL);
1600 	ifp = ifnet_byindex(msfr.msfr_ifindex);
1601 	if (ifp == NULL)
1602 		return (EADDRNOTAVAIL);
1603 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1604 
1605 	INP_WLOCK(inp);
1606 
1607 	/*
1608 	 * Lookup group on the socket.
1609 	 */
1610 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1611 	if (idx == -1 || imo->im6o_mfilters == NULL) {
1612 		INP_WUNLOCK(inp);
1613 		return (EADDRNOTAVAIL);
1614 	}
1615 	imf = &imo->im6o_mfilters[idx];
1616 
1617 	/*
1618 	 * Ignore memberships which are in limbo.
1619 	 */
1620 	if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1621 		INP_WUNLOCK(inp);
1622 		return (EAGAIN);
1623 	}
1624 	msfr.msfr_fmode = imf->im6f_st[1];
1625 
1626 	/*
1627 	 * If the user specified a buffer, copy out the source filter
1628 	 * entries to userland gracefully.
1629 	 * We only copy out the number of entries which userland
1630 	 * has asked for, but we always tell userland how big the
1631 	 * buffer really needs to be.
1632 	 */
1633 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1634 		msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1635 	tss = NULL;
1636 	if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1637 		tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1638 		    M_TEMP, M_NOWAIT | M_ZERO);
1639 		if (tss == NULL) {
1640 			INP_WUNLOCK(inp);
1641 			return (ENOBUFS);
1642 		}
1643 	}
1644 
1645 	/*
1646 	 * Count number of sources in-mode at t0.
1647 	 * If buffer space exists and remains, copy out source entries.
1648 	 */
1649 	nsrcs = msfr.msfr_nsrcs;
1650 	ncsrcs = 0;
1651 	ptss = tss;
1652 	RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1653 		lims = (struct in6_msource *)ims;
1654 		if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1655 		    lims->im6sl_st[0] != imf->im6f_st[0])
1656 			continue;
1657 		++ncsrcs;
1658 		if (tss != NULL && nsrcs > 0) {
1659 			psin = (struct sockaddr_in6 *)ptss;
1660 			psin->sin6_family = AF_INET6;
1661 			psin->sin6_len = sizeof(struct sockaddr_in6);
1662 			psin->sin6_addr = lims->im6s_addr;
1663 			psin->sin6_port = 0;
1664 			--nsrcs;
1665 			++ptss;
1666 		}
1667 	}
1668 
1669 	INP_WUNLOCK(inp);
1670 
1671 	if (tss != NULL) {
1672 		error = copyout(tss, msfr.msfr_srcs,
1673 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1674 		free(tss, M_TEMP);
1675 		if (error)
1676 			return (error);
1677 	}
1678 
1679 	msfr.msfr_nsrcs = ncsrcs;
1680 	error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1681 
1682 	return (error);
1683 }
1684 
1685 /*
1686  * Return the IP multicast options in response to user getsockopt().
1687  */
1688 int
1689 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1690 {
1691 	struct ip6_moptions	*im6o;
1692 	int			 error;
1693 	u_int			 optval;
1694 
1695 	INP_WLOCK(inp);
1696 	im6o = inp->in6p_moptions;
1697 	/*
1698 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1699 	 * or is a divert socket, reject it.
1700 	 */
1701 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1702 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1703 	    inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1704 		INP_WUNLOCK(inp);
1705 		return (EOPNOTSUPP);
1706 	}
1707 
1708 	error = 0;
1709 	switch (sopt->sopt_name) {
1710 	case IPV6_MULTICAST_IF:
1711 		if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1712 			optval = 0;
1713 		} else {
1714 			optval = im6o->im6o_multicast_ifp->if_index;
1715 		}
1716 		INP_WUNLOCK(inp);
1717 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1718 		break;
1719 
1720 	case IPV6_MULTICAST_HOPS:
1721 		if (im6o == NULL)
1722 			optval = V_ip6_defmcasthlim;
1723 		else
1724 			optval = im6o->im6o_multicast_hlim;
1725 		INP_WUNLOCK(inp);
1726 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1727 		break;
1728 
1729 	case IPV6_MULTICAST_LOOP:
1730 		if (im6o == NULL)
1731 			optval = in6_mcast_loop; /* XXX VIMAGE */
1732 		else
1733 			optval = im6o->im6o_multicast_loop;
1734 		INP_WUNLOCK(inp);
1735 		error = sooptcopyout(sopt, &optval, sizeof(u_int));
1736 		break;
1737 
1738 	case IPV6_MSFILTER:
1739 		if (im6o == NULL) {
1740 			error = EADDRNOTAVAIL;
1741 			INP_WUNLOCK(inp);
1742 		} else {
1743 			error = in6p_get_source_filters(inp, sopt);
1744 		}
1745 		break;
1746 
1747 	default:
1748 		INP_WUNLOCK(inp);
1749 		error = ENOPROTOOPT;
1750 		break;
1751 	}
1752 
1753 	INP_UNLOCK_ASSERT(inp);
1754 
1755 	return (error);
1756 }
1757 
1758 /*
1759  * Look up the ifnet to use for a multicast group membership,
1760  * given the address of an IPv6 group.
1761  *
1762  * This routine exists to support legacy IPv6 multicast applications.
1763  *
1764  * If inp is non-NULL, use this socket's current FIB number for any
1765  * required FIB lookup. Look up the group address in the unicast FIB,
1766  * and use its ifp; usually, this points to the default next-hop.
1767  * If the FIB lookup fails, return NULL.
1768  *
1769  * FUTURE: Support multiple forwarding tables for IPv6.
1770  *
1771  * Returns NULL if no ifp could be found.
1772  */
1773 static struct ifnet *
1774 in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1775     const struct sockaddr_in6 *gsin6)
1776 {
1777 	struct route_in6	 ro6;
1778 	struct ifnet		*ifp;
1779 
1780 	KASSERT(in6p->inp_vflag & INP_IPV6,
1781 	    ("%s: not INP_IPV6 inpcb", __func__));
1782 	KASSERT(gsin6->sin6_family == AF_INET6,
1783 	    ("%s: not AF_INET6 group", __func__));
1784 
1785 	ifp = NULL;
1786 	memset(&ro6, 0, sizeof(struct route_in6));
1787 	memcpy(&ro6.ro_dst, gsin6, sizeof(struct sockaddr_in6));
1788 	rtalloc_ign_fib((struct route *)&ro6, 0,
1789 	    in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB);
1790 	if (ro6.ro_rt != NULL) {
1791 		ifp = ro6.ro_rt->rt_ifp;
1792 		KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1793 		RTFREE(ro6.ro_rt);
1794 	}
1795 
1796 	return (ifp);
1797 }
1798 
1799 /*
1800  * Join an IPv6 multicast group, possibly with a source.
1801  *
1802  * FIXME: The KAME use of the unspecified address (::)
1803  * to join *all* multicast groups is currently unsupported.
1804  */
1805 static int
1806 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1807 {
1808 	struct group_source_req		 gsr;
1809 	sockunion_t			*gsa, *ssa;
1810 	struct ifnet			*ifp;
1811 	struct in6_mfilter		*imf;
1812 	struct ip6_moptions		*imo;
1813 	struct in6_multi		*inm;
1814 	struct in6_msource		*lims;
1815 	size_t				 idx;
1816 	int				 error, is_new;
1817 
1818 	ifp = NULL;
1819 	imf = NULL;
1820 	lims = NULL;
1821 	error = 0;
1822 	is_new = 0;
1823 
1824 	memset(&gsr, 0, sizeof(struct group_source_req));
1825 	gsa = (sockunion_t *)&gsr.gsr_group;
1826 	gsa->ss.ss_family = AF_UNSPEC;
1827 	ssa = (sockunion_t *)&gsr.gsr_source;
1828 	ssa->ss.ss_family = AF_UNSPEC;
1829 
1830 	/*
1831 	 * Chew everything into struct group_source_req.
1832 	 * Overwrite the port field if present, as the sockaddr
1833 	 * being copied in may be matched with a binary comparison.
1834 	 * Ignore passed-in scope ID.
1835 	 */
1836 	switch (sopt->sopt_name) {
1837 	case IPV6_JOIN_GROUP: {
1838 		struct ipv6_mreq mreq;
1839 
1840 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1841 		    sizeof(struct ipv6_mreq));
1842 		if (error)
1843 			return (error);
1844 
1845 		gsa->sin6.sin6_family = AF_INET6;
1846 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1847 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1848 
1849 		if (mreq.ipv6mr_interface == 0) {
1850 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1851 		} else {
1852 			if (V_if_index < mreq.ipv6mr_interface)
1853 				return (EADDRNOTAVAIL);
1854 			ifp = ifnet_byindex(mreq.ipv6mr_interface);
1855 		}
1856 		CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1857 		    __func__, mreq.ipv6mr_interface, ifp);
1858 	} break;
1859 
1860 	case MCAST_JOIN_GROUP:
1861 	case MCAST_JOIN_SOURCE_GROUP:
1862 		if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1863 			error = sooptcopyin(sopt, &gsr,
1864 			    sizeof(struct group_req),
1865 			    sizeof(struct group_req));
1866 		} else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1867 			error = sooptcopyin(sopt, &gsr,
1868 			    sizeof(struct group_source_req),
1869 			    sizeof(struct group_source_req));
1870 		}
1871 		if (error)
1872 			return (error);
1873 
1874 		if (gsa->sin6.sin6_family != AF_INET6 ||
1875 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1876 			return (EINVAL);
1877 
1878 		if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1879 			if (ssa->sin6.sin6_family != AF_INET6 ||
1880 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1881 				return (EINVAL);
1882 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1883 				return (EINVAL);
1884 			/*
1885 			 * TODO: Validate embedded scope ID in source
1886 			 * list entry against passed-in ifp, if and only
1887 			 * if source list filter entry is iface or node local.
1888 			 */
1889 			in6_clearscope(&ssa->sin6.sin6_addr);
1890 			ssa->sin6.sin6_port = 0;
1891 			ssa->sin6.sin6_scope_id = 0;
1892 		}
1893 
1894 		if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1895 			return (EADDRNOTAVAIL);
1896 		ifp = ifnet_byindex(gsr.gsr_interface);
1897 		break;
1898 
1899 	default:
1900 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1901 		    __func__, sopt->sopt_name);
1902 		return (EOPNOTSUPP);
1903 		break;
1904 	}
1905 
1906 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1907 		return (EINVAL);
1908 
1909 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1910 		return (EADDRNOTAVAIL);
1911 
1912 	gsa->sin6.sin6_port = 0;
1913 	gsa->sin6.sin6_scope_id = 0;
1914 
1915 	/*
1916 	 * Always set the scope zone ID on memberships created from userland.
1917 	 * Use the passed-in ifp to do this.
1918 	 * XXX The in6_setscope() return value is meaningless.
1919 	 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1920 	 */
1921 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1922 
1923 	imo = in6p_findmoptions(inp);
1924 	idx = im6o_match_group(imo, ifp, &gsa->sa);
1925 	if (idx == -1) {
1926 		is_new = 1;
1927 	} else {
1928 		inm = imo->im6o_membership[idx];
1929 		imf = &imo->im6o_mfilters[idx];
1930 		if (ssa->ss.ss_family != AF_UNSPEC) {
1931 			/*
1932 			 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1933 			 * is an error. On an existing inclusive membership,
1934 			 * it just adds the source to the filter list.
1935 			 */
1936 			if (imf->im6f_st[1] != MCAST_INCLUDE) {
1937 				error = EINVAL;
1938 				goto out_in6p_locked;
1939 			}
1940 			/*
1941 			 * Throw out duplicates.
1942 			 *
1943 			 * XXX FIXME: This makes a naive assumption that
1944 			 * even if entries exist for *ssa in this imf,
1945 			 * they will be rejected as dupes, even if they
1946 			 * are not valid in the current mode (in-mode).
1947 			 *
1948 			 * in6_msource is transactioned just as for anything
1949 			 * else in SSM -- but note naive use of in6m_graft()
1950 			 * below for allocating new filter entries.
1951 			 *
1952 			 * This is only an issue if someone mixes the
1953 			 * full-state SSM API with the delta-based API,
1954 			 * which is discouraged in the relevant RFCs.
1955 			 */
1956 			lims = im6o_match_source(imo, idx, &ssa->sa);
1957 			if (lims != NULL /*&&
1958 			    lims->im6sl_st[1] == MCAST_INCLUDE*/) {
1959 				error = EADDRNOTAVAIL;
1960 				goto out_in6p_locked;
1961 			}
1962 		} else {
1963 			/*
1964 			 * MCAST_JOIN_GROUP alone, on any existing membership,
1965 			 * is rejected, to stop the same inpcb tying up
1966 			 * multiple refs to the in_multi.
1967 			 * On an existing inclusive membership, this is also
1968 			 * an error; if you want to change filter mode,
1969 			 * you must use the userland API setsourcefilter().
1970 			 * XXX We don't reject this for imf in UNDEFINED
1971 			 * state at t1, because allocation of a filter
1972 			 * is atomic with allocation of a membership.
1973 			 */
1974 			error = EINVAL;
1975 			goto out_in6p_locked;
1976 		}
1977 	}
1978 
1979 	/*
1980 	 * Begin state merge transaction at socket layer.
1981 	 */
1982 	INP_WLOCK_ASSERT(inp);
1983 
1984 	if (is_new) {
1985 		if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
1986 			error = im6o_grow(imo);
1987 			if (error)
1988 				goto out_in6p_locked;
1989 		}
1990 		/*
1991 		 * Allocate the new slot upfront so we can deal with
1992 		 * grafting the new source filter in same code path
1993 		 * as for join-source on existing membership.
1994 		 */
1995 		idx = imo->im6o_num_memberships;
1996 		imo->im6o_membership[idx] = NULL;
1997 		imo->im6o_num_memberships++;
1998 		KASSERT(imo->im6o_mfilters != NULL,
1999 		    ("%s: im6f_mfilters vector was not allocated", __func__));
2000 		imf = &imo->im6o_mfilters[idx];
2001 		KASSERT(RB_EMPTY(&imf->im6f_sources),
2002 		    ("%s: im6f_sources not empty", __func__));
2003 	}
2004 
2005 	/*
2006 	 * Graft new source into filter list for this inpcb's
2007 	 * membership of the group. The in6_multi may not have
2008 	 * been allocated yet if this is a new membership, however,
2009 	 * the in_mfilter slot will be allocated and must be initialized.
2010 	 *
2011 	 * Note: Grafting of exclusive mode filters doesn't happen
2012 	 * in this path.
2013 	 * XXX: Should check for non-NULL lims (node exists but may
2014 	 * not be in-mode) for interop with full-state API.
2015 	 */
2016 	if (ssa->ss.ss_family != AF_UNSPEC) {
2017 		/* Membership starts in IN mode */
2018 		if (is_new) {
2019 			CTR1(KTR_MLD, "%s: new join w/source", __func__);
2020 			im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2021 		} else {
2022 			CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2023 		}
2024 		lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2025 		if (lims == NULL) {
2026 			CTR1(KTR_MLD, "%s: merge imf state failed",
2027 			    __func__);
2028 			error = ENOMEM;
2029 			goto out_im6o_free;
2030 		}
2031 	} else {
2032 		/* No address specified; Membership starts in EX mode */
2033 		if (is_new) {
2034 			CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2035 			im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2036 		}
2037 	}
2038 
2039 	/*
2040 	 * Begin state merge transaction at MLD layer.
2041 	 */
2042 	IN6_MULTI_LOCK();
2043 
2044 	if (is_new) {
2045 		error = in6_mc_join_locked(ifp, &gsa->sin6.sin6_addr, imf,
2046 		    &inm, 0);
2047 		if (error)
2048 			goto out_im6o_free;
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 			goto out_im6f_rollback;
2057 		}
2058 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2059 		error = mld_change_state(inm, 0);
2060 		if (error) {
2061 			CTR1(KTR_MLD, "%s: failed mld downcall",
2062 			    __func__);
2063 			goto out_im6f_rollback;
2064 		}
2065 	}
2066 
2067 	IN6_MULTI_UNLOCK();
2068 
2069 out_im6f_rollback:
2070 	INP_WLOCK_ASSERT(inp);
2071 	if (error) {
2072 		im6f_rollback(imf);
2073 		if (is_new)
2074 			im6f_purge(imf);
2075 		else
2076 			im6f_reap(imf);
2077 	} else {
2078 		im6f_commit(imf);
2079 	}
2080 
2081 out_im6o_free:
2082 	if (error && is_new) {
2083 		imo->im6o_membership[idx] = NULL;
2084 		--imo->im6o_num_memberships;
2085 	}
2086 
2087 out_in6p_locked:
2088 	INP_WUNLOCK(inp);
2089 	return (error);
2090 }
2091 
2092 /*
2093  * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2094  */
2095 static int
2096 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2097 {
2098 	struct ipv6_mreq		 mreq;
2099 	struct group_source_req		 gsr;
2100 	sockunion_t			*gsa, *ssa;
2101 	struct ifnet			*ifp;
2102 	struct in6_mfilter		*imf;
2103 	struct ip6_moptions		*imo;
2104 	struct in6_msource		*ims;
2105 	struct in6_multi		*inm;
2106 	uint32_t			 ifindex;
2107 	size_t				 idx;
2108 	int				 error, is_final;
2109 #ifdef KTR
2110 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2111 #endif
2112 
2113 	ifp = NULL;
2114 	ifindex = 0;
2115 	error = 0;
2116 	is_final = 1;
2117 
2118 	memset(&gsr, 0, sizeof(struct group_source_req));
2119 	gsa = (sockunion_t *)&gsr.gsr_group;
2120 	gsa->ss.ss_family = AF_UNSPEC;
2121 	ssa = (sockunion_t *)&gsr.gsr_source;
2122 	ssa->ss.ss_family = AF_UNSPEC;
2123 
2124 	/*
2125 	 * Chew everything passed in up into a struct group_source_req
2126 	 * as that is easier to process.
2127 	 * Note: Any embedded scope ID in the multicast group passed
2128 	 * in by userland is ignored, the interface index is the recommended
2129 	 * mechanism to specify an interface; see below.
2130 	 */
2131 	switch (sopt->sopt_name) {
2132 	case IPV6_LEAVE_GROUP:
2133 		error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2134 		    sizeof(struct ipv6_mreq));
2135 		if (error)
2136 			return (error);
2137 		gsa->sin6.sin6_family = AF_INET6;
2138 		gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2139 		gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2140 		gsa->sin6.sin6_port = 0;
2141 		gsa->sin6.sin6_scope_id = 0;
2142 		ifindex = mreq.ipv6mr_interface;
2143 		break;
2144 
2145 	case MCAST_LEAVE_GROUP:
2146 	case MCAST_LEAVE_SOURCE_GROUP:
2147 		if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2148 			error = sooptcopyin(sopt, &gsr,
2149 			    sizeof(struct group_req),
2150 			    sizeof(struct group_req));
2151 		} else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2152 			error = sooptcopyin(sopt, &gsr,
2153 			    sizeof(struct group_source_req),
2154 			    sizeof(struct group_source_req));
2155 		}
2156 		if (error)
2157 			return (error);
2158 
2159 		if (gsa->sin6.sin6_family != AF_INET6 ||
2160 		    gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2161 			return (EINVAL);
2162 		if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2163 			if (ssa->sin6.sin6_family != AF_INET6 ||
2164 			    ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2165 				return (EINVAL);
2166 			if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2167 				return (EINVAL);
2168 			/*
2169 			 * TODO: Validate embedded scope ID in source
2170 			 * list entry against passed-in ifp, if and only
2171 			 * if source list filter entry is iface or node local.
2172 			 */
2173 			in6_clearscope(&ssa->sin6.sin6_addr);
2174 		}
2175 		gsa->sin6.sin6_port = 0;
2176 		gsa->sin6.sin6_scope_id = 0;
2177 		ifindex = gsr.gsr_interface;
2178 		break;
2179 
2180 	default:
2181 		CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2182 		    __func__, sopt->sopt_name);
2183 		return (EOPNOTSUPP);
2184 		break;
2185 	}
2186 
2187 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2188 		return (EINVAL);
2189 
2190 	/*
2191 	 * Validate interface index if provided. If no interface index
2192 	 * was provided separately, attempt to look the membership up
2193 	 * from the default scope as a last resort to disambiguate
2194 	 * the membership we are being asked to leave.
2195 	 * XXX SCOPE6 lock potentially taken here.
2196 	 */
2197 	if (ifindex != 0) {
2198 		if (V_if_index < ifindex)
2199 			return (EADDRNOTAVAIL);
2200 		ifp = ifnet_byindex(ifindex);
2201 		if (ifp == NULL)
2202 			return (EADDRNOTAVAIL);
2203 		(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2204 	} else {
2205 		error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2206 		if (error)
2207 			return (EADDRNOTAVAIL);
2208 		/*
2209 		 * Some badly behaved applications don't pass an ifindex
2210 		 * or a scope ID, which is an API violation. In this case,
2211 		 * perform a lookup as per a v6 join.
2212 		 *
2213 		 * XXX For now, stomp on zone ID for the corner case.
2214 		 * This is not the 'KAME way', but we need to see the ifp
2215 		 * directly until such time as this implementation is
2216 		 * refactored, assuming the scope IDs are the way to go.
2217 		 */
2218 		ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2219 		if (ifindex == 0) {
2220 			CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2221 			    "ifp for group %s.", __func__,
2222 			    ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2223 			ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2224 		} else {
2225 			ifp = ifnet_byindex(ifindex);
2226 		}
2227 		if (ifp == NULL)
2228 			return (EADDRNOTAVAIL);
2229 	}
2230 
2231 	CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2232 	KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2233 
2234 	/*
2235 	 * Find the membership in the membership array.
2236 	 */
2237 	imo = in6p_findmoptions(inp);
2238 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2239 	if (idx == -1) {
2240 		error = EADDRNOTAVAIL;
2241 		goto out_in6p_locked;
2242 	}
2243 	inm = imo->im6o_membership[idx];
2244 	imf = &imo->im6o_mfilters[idx];
2245 
2246 	if (ssa->ss.ss_family != AF_UNSPEC)
2247 		is_final = 0;
2248 
2249 	/*
2250 	 * Begin state merge transaction at socket layer.
2251 	 */
2252 	INP_WLOCK_ASSERT(inp);
2253 
2254 	/*
2255 	 * If we were instructed only to leave a given source, do so.
2256 	 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2257 	 */
2258 	if (is_final) {
2259 		im6f_leave(imf);
2260 	} else {
2261 		if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2262 			error = EADDRNOTAVAIL;
2263 			goto out_in6p_locked;
2264 		}
2265 		ims = im6o_match_source(imo, idx, &ssa->sa);
2266 		if (ims == NULL) {
2267 			CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2268 			    ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2269 			    "not ");
2270 			error = EADDRNOTAVAIL;
2271 			goto out_in6p_locked;
2272 		}
2273 		CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2274 		error = im6f_prune(imf, &ssa->sin6);
2275 		if (error) {
2276 			CTR1(KTR_MLD, "%s: merge imf state failed",
2277 			    __func__);
2278 			goto out_in6p_locked;
2279 		}
2280 	}
2281 
2282 	/*
2283 	 * Begin state merge transaction at MLD layer.
2284 	 */
2285 	IN6_MULTI_LOCK();
2286 
2287 	if (is_final) {
2288 		/*
2289 		 * Give up the multicast address record to which
2290 		 * the membership points.
2291 		 */
2292 		(void)in6_mc_leave_locked(inm, imf);
2293 	} else {
2294 		CTR1(KTR_MLD, "%s: merge inm state", __func__);
2295 		error = in6m_merge(inm, imf);
2296 		if (error) {
2297 			CTR1(KTR_MLD, "%s: failed to merge inm state",
2298 			    __func__);
2299 			goto out_im6f_rollback;
2300 		}
2301 
2302 		CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2303 		error = mld_change_state(inm, 0);
2304 		if (error) {
2305 			CTR1(KTR_MLD, "%s: failed mld downcall",
2306 			    __func__);
2307 		}
2308 	}
2309 
2310 	IN6_MULTI_UNLOCK();
2311 
2312 out_im6f_rollback:
2313 	if (error)
2314 		im6f_rollback(imf);
2315 	else
2316 		im6f_commit(imf);
2317 
2318 	im6f_reap(imf);
2319 
2320 	if (is_final) {
2321 		/* Remove the gap in the membership array. */
2322 		for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2323 			imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2324 			imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2325 		}
2326 		imo->im6o_num_memberships--;
2327 	}
2328 
2329 out_in6p_locked:
2330 	INP_WUNLOCK(inp);
2331 	return (error);
2332 }
2333 
2334 /*
2335  * Select the interface for transmitting IPv6 multicast datagrams.
2336  *
2337  * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2338  * may be passed to this socket option. An address of in6addr_any or an
2339  * interface index of 0 is used to remove a previous selection.
2340  * When no interface is selected, one is chosen for every send.
2341  */
2342 static int
2343 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2344 {
2345 	struct ifnet		*ifp;
2346 	struct ip6_moptions	*imo;
2347 	u_int			 ifindex;
2348 	int			 error;
2349 
2350 	if (sopt->sopt_valsize != sizeof(u_int))
2351 		return (EINVAL);
2352 
2353 	error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2354 	if (error)
2355 		return (error);
2356 	if (V_if_index < ifindex)
2357 		return (EINVAL);
2358 
2359 	ifp = ifnet_byindex(ifindex);
2360 	if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2361 		return (EADDRNOTAVAIL);
2362 
2363 	imo = in6p_findmoptions(inp);
2364 	imo->im6o_multicast_ifp = ifp;
2365 	INP_WUNLOCK(inp);
2366 
2367 	return (0);
2368 }
2369 
2370 /*
2371  * Atomically set source filters on a socket for an IPv6 multicast group.
2372  *
2373  * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2374  */
2375 static int
2376 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2377 {
2378 	struct __msfilterreq	 msfr;
2379 	sockunion_t		*gsa;
2380 	struct ifnet		*ifp;
2381 	struct in6_mfilter	*imf;
2382 	struct ip6_moptions	*imo;
2383 	struct in6_multi		*inm;
2384 	size_t			 idx;
2385 	int			 error;
2386 
2387 	error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2388 	    sizeof(struct __msfilterreq));
2389 	if (error)
2390 		return (error);
2391 
2392 	if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2393 		return (ENOBUFS);
2394 
2395 	if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2396 	    msfr.msfr_fmode != MCAST_INCLUDE)
2397 		return (EINVAL);
2398 
2399 	if (msfr.msfr_group.ss_family != AF_INET6 ||
2400 	    msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2401 		return (EINVAL);
2402 
2403 	gsa = (sockunion_t *)&msfr.msfr_group;
2404 	if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2405 		return (EINVAL);
2406 
2407 	gsa->sin6.sin6_port = 0;	/* ignore port */
2408 
2409 	if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2410 		return (EADDRNOTAVAIL);
2411 	ifp = ifnet_byindex(msfr.msfr_ifindex);
2412 	if (ifp == NULL)
2413 		return (EADDRNOTAVAIL);
2414 	(void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2415 
2416 	/*
2417 	 * Take the INP write lock.
2418 	 * Check if this socket is a member of this group.
2419 	 */
2420 	imo = in6p_findmoptions(inp);
2421 	idx = im6o_match_group(imo, ifp, &gsa->sa);
2422 	if (idx == -1 || imo->im6o_mfilters == NULL) {
2423 		error = EADDRNOTAVAIL;
2424 		goto out_in6p_locked;
2425 	}
2426 	inm = imo->im6o_membership[idx];
2427 	imf = &imo->im6o_mfilters[idx];
2428 
2429 	/*
2430 	 * Begin state merge transaction at socket layer.
2431 	 */
2432 	INP_WLOCK_ASSERT(inp);
2433 
2434 	imf->im6f_st[1] = msfr.msfr_fmode;
2435 
2436 	/*
2437 	 * Apply any new source filters, if present.
2438 	 * Make a copy of the user-space source vector so
2439 	 * that we may copy them with a single copyin. This
2440 	 * allows us to deal with page faults up-front.
2441 	 */
2442 	if (msfr.msfr_nsrcs > 0) {
2443 		struct in6_msource	*lims;
2444 		struct sockaddr_in6	*psin;
2445 		struct sockaddr_storage	*kss, *pkss;
2446 		int			 i;
2447 
2448 		INP_WUNLOCK(inp);
2449 
2450 		CTR2(KTR_MLD, "%s: loading %lu source list entries",
2451 		    __func__, (unsigned long)msfr.msfr_nsrcs);
2452 		kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2453 		    M_TEMP, M_WAITOK);
2454 		error = copyin(msfr.msfr_srcs, kss,
2455 		    sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2456 		if (error) {
2457 			free(kss, M_TEMP);
2458 			return (error);
2459 		}
2460 
2461 		INP_WLOCK(inp);
2462 
2463 		/*
2464 		 * Mark all source filters as UNDEFINED at t1.
2465 		 * Restore new group filter mode, as im6f_leave()
2466 		 * will set it to INCLUDE.
2467 		 */
2468 		im6f_leave(imf);
2469 		imf->im6f_st[1] = msfr.msfr_fmode;
2470 
2471 		/*
2472 		 * Update socket layer filters at t1, lazy-allocating
2473 		 * new entries. This saves a bunch of memory at the
2474 		 * cost of one RB_FIND() per source entry; duplicate
2475 		 * entries in the msfr_nsrcs vector are ignored.
2476 		 * If we encounter an error, rollback transaction.
2477 		 *
2478 		 * XXX This too could be replaced with a set-symmetric
2479 		 * difference like loop to avoid walking from root
2480 		 * every time, as the key space is common.
2481 		 */
2482 		for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2483 			psin = (struct sockaddr_in6 *)pkss;
2484 			if (psin->sin6_family != AF_INET6) {
2485 				error = EAFNOSUPPORT;
2486 				break;
2487 			}
2488 			if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2489 				error = EINVAL;
2490 				break;
2491 			}
2492 			if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2493 				error = EINVAL;
2494 				break;
2495 			}
2496 			/*
2497 			 * TODO: Validate embedded scope ID in source
2498 			 * list entry against passed-in ifp, if and only
2499 			 * if source list filter entry is iface or node local.
2500 			 */
2501 			in6_clearscope(&psin->sin6_addr);
2502 			error = im6f_get_source(imf, psin, &lims);
2503 			if (error)
2504 				break;
2505 			lims->im6sl_st[1] = imf->im6f_st[1];
2506 		}
2507 		free(kss, M_TEMP);
2508 	}
2509 
2510 	if (error)
2511 		goto out_im6f_rollback;
2512 
2513 	INP_WLOCK_ASSERT(inp);
2514 	IN6_MULTI_LOCK();
2515 
2516 	/*
2517 	 * Begin state merge transaction at MLD layer.
2518 	 */
2519 	CTR1(KTR_MLD, "%s: merge inm state", __func__);
2520 	error = in6m_merge(inm, imf);
2521 	if (error) {
2522 		CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2523 		goto out_im6f_rollback;
2524 	}
2525 
2526 	CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2527 	error = mld_change_state(inm, 0);
2528 	if (error)
2529 		CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2530 
2531 	IN6_MULTI_UNLOCK();
2532 
2533 out_im6f_rollback:
2534 	if (error)
2535 		im6f_rollback(imf);
2536 	else
2537 		im6f_commit(imf);
2538 
2539 	im6f_reap(imf);
2540 
2541 out_in6p_locked:
2542 	INP_WUNLOCK(inp);
2543 	return (error);
2544 }
2545 
2546 /*
2547  * Set the IP multicast options in response to user setsockopt().
2548  *
2549  * Many of the socket options handled in this function duplicate the
2550  * functionality of socket options in the regular unicast API. However,
2551  * it is not possible to merge the duplicate code, because the idempotence
2552  * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2553  * the effects of these options must be treated as separate and distinct.
2554  *
2555  * SMPng: XXX: Unlocked read of inp_socket believed OK.
2556  */
2557 int
2558 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2559 {
2560 	struct ip6_moptions	*im6o;
2561 	int			 error;
2562 
2563 	error = 0;
2564 
2565 	/*
2566 	 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2567 	 * or is a divert socket, reject it.
2568 	 */
2569 	if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2570 	    (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2571 	     inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2572 		return (EOPNOTSUPP);
2573 
2574 	switch (sopt->sopt_name) {
2575 	case IPV6_MULTICAST_IF:
2576 		error = in6p_set_multicast_if(inp, sopt);
2577 		break;
2578 
2579 	case IPV6_MULTICAST_HOPS: {
2580 		int hlim;
2581 
2582 		if (sopt->sopt_valsize != sizeof(int)) {
2583 			error = EINVAL;
2584 			break;
2585 		}
2586 		error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2587 		if (error)
2588 			break;
2589 		if (hlim < -1 || hlim > 255) {
2590 			error = EINVAL;
2591 			break;
2592 		} else if (hlim == -1) {
2593 			hlim = V_ip6_defmcasthlim;
2594 		}
2595 		im6o = in6p_findmoptions(inp);
2596 		im6o->im6o_multicast_hlim = hlim;
2597 		INP_WUNLOCK(inp);
2598 		break;
2599 	}
2600 
2601 	case IPV6_MULTICAST_LOOP: {
2602 		u_int loop;
2603 
2604 		/*
2605 		 * Set the loopback flag for outgoing multicast packets.
2606 		 * Must be zero or one.
2607 		 */
2608 		if (sopt->sopt_valsize != sizeof(u_int)) {
2609 			error = EINVAL;
2610 			break;
2611 		}
2612 		error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2613 		if (error)
2614 			break;
2615 		if (loop > 1) {
2616 			error = EINVAL;
2617 			break;
2618 		}
2619 		im6o = in6p_findmoptions(inp);
2620 		im6o->im6o_multicast_loop = loop;
2621 		INP_WUNLOCK(inp);
2622 		break;
2623 	}
2624 
2625 	case IPV6_JOIN_GROUP:
2626 	case MCAST_JOIN_GROUP:
2627 	case MCAST_JOIN_SOURCE_GROUP:
2628 		error = in6p_join_group(inp, sopt);
2629 		break;
2630 
2631 	case IPV6_LEAVE_GROUP:
2632 	case MCAST_LEAVE_GROUP:
2633 	case MCAST_LEAVE_SOURCE_GROUP:
2634 		error = in6p_leave_group(inp, sopt);
2635 		break;
2636 
2637 	case MCAST_BLOCK_SOURCE:
2638 	case MCAST_UNBLOCK_SOURCE:
2639 		error = in6p_block_unblock_source(inp, sopt);
2640 		break;
2641 
2642 	case IPV6_MSFILTER:
2643 		error = in6p_set_source_filters(inp, sopt);
2644 		break;
2645 
2646 	default:
2647 		error = EOPNOTSUPP;
2648 		break;
2649 	}
2650 
2651 	INP_UNLOCK_ASSERT(inp);
2652 
2653 	return (error);
2654 }
2655 
2656 /*
2657  * Expose MLD's multicast filter mode and source list(s) to userland,
2658  * keyed by (ifindex, group).
2659  * The filter mode is written out as a uint32_t, followed by
2660  * 0..n of struct in6_addr.
2661  * For use by ifmcstat(8).
2662  * SMPng: NOTE: unlocked read of ifindex space.
2663  */
2664 static int
2665 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2666 {
2667 	struct in6_addr			 mcaddr;
2668 	struct in6_addr			 src;
2669 	struct ifnet			*ifp;
2670 	struct ifmultiaddr		*ifma;
2671 	struct in6_multi		*inm;
2672 	struct ip6_msource		*ims;
2673 	int				*name;
2674 	int				 retval;
2675 	u_int				 namelen;
2676 	uint32_t			 fmode, ifindex;
2677 #ifdef KTR
2678 	char				 ip6tbuf[INET6_ADDRSTRLEN];
2679 #endif
2680 
2681 	name = (int *)arg1;
2682 	namelen = arg2;
2683 
2684 	if (req->newptr != NULL)
2685 		return (EPERM);
2686 
2687 	/* int: ifindex + 4 * 32 bits of IPv6 address */
2688 	if (namelen != 5)
2689 		return (EINVAL);
2690 
2691 	ifindex = name[0];
2692 	if (ifindex <= 0 || ifindex > V_if_index) {
2693 		CTR2(KTR_MLD, "%s: ifindex %u out of range",
2694 		    __func__, ifindex);
2695 		return (ENOENT);
2696 	}
2697 
2698 	memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2699 	if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2700 		CTR2(KTR_MLD, "%s: group %s is not multicast",
2701 		    __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2702 		return (EINVAL);
2703 	}
2704 
2705 	ifp = ifnet_byindex(ifindex);
2706 	if (ifp == NULL) {
2707 		CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2708 		    __func__, ifindex);
2709 		return (ENOENT);
2710 	}
2711 	/*
2712 	 * Internal MLD lookups require that scope/zone ID is set.
2713 	 */
2714 	(void)in6_setscope(&mcaddr, ifp, NULL);
2715 
2716 	retval = sysctl_wire_old_buffer(req,
2717 	    sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2718 	if (retval)
2719 		return (retval);
2720 
2721 	IN6_MULTI_LOCK();
2722 
2723 	IF_ADDR_RLOCK(ifp);
2724 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2725 		if (ifma->ifma_addr->sa_family != AF_INET6 ||
2726 		    ifma->ifma_protospec == NULL)
2727 			continue;
2728 		inm = (struct in6_multi *)ifma->ifma_protospec;
2729 		if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2730 			continue;
2731 		fmode = inm->in6m_st[1].iss_fmode;
2732 		retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2733 		if (retval != 0)
2734 			break;
2735 		RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2736 			CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2737 			/*
2738 			 * Only copy-out sources which are in-mode.
2739 			 */
2740 			if (fmode != im6s_get_mode(inm, ims, 1)) {
2741 				CTR1(KTR_MLD, "%s: skip non-in-mode",
2742 				    __func__);
2743 				continue;
2744 			}
2745 			src = ims->im6s_addr;
2746 			retval = SYSCTL_OUT(req, &src,
2747 			    sizeof(struct in6_addr));
2748 			if (retval != 0)
2749 				break;
2750 		}
2751 	}
2752 	IF_ADDR_RUNLOCK(ifp);
2753 
2754 	IN6_MULTI_UNLOCK();
2755 
2756 	return (retval);
2757 }
2758 
2759 #ifdef KTR
2760 
2761 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2762 
2763 static const char *
2764 in6m_mode_str(const int mode)
2765 {
2766 
2767 	if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2768 		return (in6m_modestrs[mode]);
2769 	return ("??");
2770 }
2771 
2772 static const char *in6m_statestrs[] = {
2773 	"not-member",
2774 	"silent",
2775 	"idle",
2776 	"lazy",
2777 	"sleeping",
2778 	"awakening",
2779 	"query-pending",
2780 	"sg-query-pending",
2781 	"leaving"
2782 };
2783 
2784 static const char *
2785 in6m_state_str(const int state)
2786 {
2787 
2788 	if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2789 		return (in6m_statestrs[state]);
2790 	return ("??");
2791 }
2792 
2793 /*
2794  * Dump an in6_multi structure to the console.
2795  */
2796 void
2797 in6m_print(const struct in6_multi *inm)
2798 {
2799 	int t;
2800 	char ip6tbuf[INET6_ADDRSTRLEN];
2801 
2802 	if ((ktr_mask & KTR_MLD) == 0)
2803 		return;
2804 
2805 	printf("%s: --- begin in6m %p ---\n", __func__, inm);
2806 	printf("addr %s ifp %p(%s) ifma %p\n",
2807 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2808 	    inm->in6m_ifp,
2809 	    if_name(inm->in6m_ifp),
2810 	    inm->in6m_ifma);
2811 	printf("timer %u state %s refcount %u scq.len %u\n",
2812 	    inm->in6m_timer,
2813 	    in6m_state_str(inm->in6m_state),
2814 	    inm->in6m_refcount,
2815 	    inm->in6m_scq.ifq_len);
2816 	printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2817 	    inm->in6m_mli,
2818 	    inm->in6m_nsrc,
2819 	    inm->in6m_sctimer,
2820 	    inm->in6m_scrv);
2821 	for (t = 0; t < 2; t++) {
2822 		printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2823 		    in6m_mode_str(inm->in6m_st[t].iss_fmode),
2824 		    inm->in6m_st[t].iss_asm,
2825 		    inm->in6m_st[t].iss_ex,
2826 		    inm->in6m_st[t].iss_in,
2827 		    inm->in6m_st[t].iss_rec);
2828 	}
2829 	printf("%s: --- end in6m %p ---\n", __func__, inm);
2830 }
2831 
2832 #else /* !KTR */
2833 
2834 void
2835 in6m_print(const struct in6_multi *inm)
2836 {
2837 
2838 }
2839 
2840 #endif /* KTR */
2841