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: releng/12.0/sys/netinet6/in6_mcast.c 338178 2018-08-22 04:09:55Z mmacy $");
39
40 #include "opt_inet6.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/gtaskqueue.h>
45 #include <sys/kernel.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.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
63 #include <netinet/in.h>
64 #include <netinet/udp.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/udp_var.h>
68 #include <netinet6/in6_fib.h>
69 #include <netinet6/in6_var.h>
70 #include <netinet/ip6.h>
71 #include <netinet/icmp6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet/in_pcb.h>
74 #include <netinet/tcp_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet6/mld6_var.h>
77 #include <netinet6/scope6_var.h>
78
79 #ifndef KTR_MLD
80 #define KTR_MLD KTR_INET6
81 #endif
82
83 #ifndef __SOCKUNION_DECLARED
84 union sockunion {
85 struct sockaddr_storage ss;
86 struct sockaddr sa;
87 struct sockaddr_dl sdl;
88 struct sockaddr_in6 sin6;
89 };
90 typedef union sockunion sockunion_t;
91 #define __SOCKUNION_DECLARED
92 #endif /* __SOCKUNION_DECLARED */
93
94 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
95 "IPv6 multicast PCB-layer source filter");
96 MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
97 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
98 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
99 "IPv6 multicast MLD-layer source filter");
100
101 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
102
103 /*
104 * Locking:
105 * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
106 * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
107 * it can be taken by code in net/if.c also.
108 * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
109 *
110 * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
111 * any need for in6_multi itself to be virtualized -- it is bound to an ifp
112 * anyway no matter what happens.
113 */
114 struct mtx in6_multi_list_mtx;
115 MTX_SYSINIT(in6_multi_mtx, &in6_multi_list_mtx, "in6_multi_list_mtx", MTX_DEF);
116
117 struct mtx in6_multi_free_mtx;
118 MTX_SYSINIT(in6_multi_free_mtx, &in6_multi_free_mtx, "in6_multi_free_mtx", MTX_DEF);
119
120 struct sx in6_multi_sx;
121 SX_SYSINIT(in6_multi_sx, &in6_multi_sx, "in6_multi_sx");
122
123
124
125 static void im6f_commit(struct in6_mfilter *);
126 static int im6f_get_source(struct in6_mfilter *imf,
127 const struct sockaddr_in6 *psin,
128 struct in6_msource **);
129 static struct in6_msource *
130 im6f_graft(struct in6_mfilter *, const uint8_t,
131 const struct sockaddr_in6 *);
132 static void im6f_leave(struct in6_mfilter *);
133 static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
134 static void im6f_purge(struct in6_mfilter *);
135 static void im6f_rollback(struct in6_mfilter *);
136 static void im6f_reap(struct in6_mfilter *);
137 static int im6o_grow(struct ip6_moptions *);
138 static size_t im6o_match_group(const struct ip6_moptions *,
139 const struct ifnet *, const struct sockaddr *);
140 static struct in6_msource *
141 im6o_match_source(const struct ip6_moptions *, const size_t,
142 const struct sockaddr *);
143 static void im6s_merge(struct ip6_msource *ims,
144 const struct in6_msource *lims, const int rollback);
145 static int in6_getmulti(struct ifnet *, const struct in6_addr *,
146 struct in6_multi **);
147 static int in6m_get_source(struct in6_multi *inm,
148 const struct in6_addr *addr, const int noalloc,
149 struct ip6_msource **pims);
150 #ifdef KTR
151 static int in6m_is_ifp_detached(const struct in6_multi *);
152 #endif
153 static int in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
154 static void in6m_purge(struct in6_multi *);
155 static void in6m_reap(struct in6_multi *);
156 static struct ip6_moptions *
157 in6p_findmoptions(struct inpcb *);
158 static int in6p_get_source_filters(struct inpcb *, struct sockopt *);
159 static int in6p_join_group(struct inpcb *, struct sockopt *);
160 static int in6p_leave_group(struct inpcb *, struct sockopt *);
161 static struct ifnet *
162 in6p_lookup_mcast_ifp(const struct inpcb *,
163 const struct sockaddr_in6 *);
164 static int in6p_block_unblock_source(struct inpcb *, struct sockopt *);
165 static int in6p_set_multicast_if(struct inpcb *, struct sockopt *);
166 static int in6p_set_source_filters(struct inpcb *, struct sockopt *);
167 static int sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
168
169 SYSCTL_DECL(_net_inet6_ip6); /* XXX Not in any common header. */
170
171 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0,
172 "IPv6 multicast");
173
174 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
175 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
176 CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
177 "Max source filters per group");
178
179 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
180 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
181 CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
182 "Max source filters per socket");
183
184 /* TODO Virtualize this switch. */
185 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
186 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
187 &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
188
189 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
190 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
191 "Per-interface stack-wide source filters");
192
193 int ifma6_restart = 0;
194 #ifdef KTR
195 /*
196 * Inline function which wraps assertions for a valid ifp.
197 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
198 * is detached.
199 */
200 static int __inline
201 in6m_is_ifp_detached(const struct in6_multi *inm)
202 {
203 struct ifnet *ifp;
204
205 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
206 ifp = inm->in6m_ifma->ifma_ifp;
207 if (ifp != NULL) {
208 /*
209 * Sanity check that network-layer notion of ifp is the
210 * same as that of link-layer.
211 */
212 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
213 }
214
215 return (ifp == NULL);
216 }
217 #endif
218
219 /*
220 * Initialize an in6_mfilter structure to a known state at t0, t1
221 * with an empty source filter list.
222 */
223 static __inline void
224 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
225 {
226 memset(imf, 0, sizeof(struct in6_mfilter));
227 RB_INIT(&imf->im6f_sources);
228 imf->im6f_st[0] = st0;
229 imf->im6f_st[1] = st1;
230 }
231
232 /*
233 * Resize the ip6_moptions vector to the next power-of-two minus 1.
234 * May be called with locks held; do not sleep.
235 */
236 static int
237 im6o_grow(struct ip6_moptions *imo)
238 {
239 struct in6_multi **nmships;
240 struct in6_multi **omships;
241 struct in6_mfilter *nmfilters;
242 struct in6_mfilter *omfilters;
243 size_t idx;
244 size_t newmax;
245 size_t oldmax;
246
247 nmships = NULL;
248 nmfilters = NULL;
249 omships = imo->im6o_membership;
250 omfilters = imo->im6o_mfilters;
251 oldmax = imo->im6o_max_memberships;
252 newmax = ((oldmax + 1) * 2) - 1;
253
254 if (newmax <= IPV6_MAX_MEMBERSHIPS) {
255 nmships = (struct in6_multi **)realloc(omships,
256 sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
257 nmfilters = (struct in6_mfilter *)realloc(omfilters,
258 sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
259 M_NOWAIT);
260 if (nmships != NULL && nmfilters != NULL) {
261 /* Initialize newly allocated source filter heads. */
262 for (idx = oldmax; idx < newmax; idx++) {
263 im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
264 MCAST_EXCLUDE);
265 }
266 imo->im6o_max_memberships = newmax;
267 imo->im6o_membership = nmships;
268 imo->im6o_mfilters = nmfilters;
269 }
270 }
271
272 if (nmships == NULL || nmfilters == NULL) {
273 if (nmships != NULL)
274 free(nmships, M_IP6MOPTS);
275 if (nmfilters != NULL)
276 free(nmfilters, M_IN6MFILTER);
277 return (ETOOMANYREFS);
278 }
279
280 return (0);
281 }
282
283 /*
284 * Find an IPv6 multicast group entry for this ip6_moptions instance
285 * which matches the specified group, and optionally an interface.
286 * Return its index into the array, or -1 if not found.
287 */
288 static size_t
289 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
290 const struct sockaddr *group)
291 {
292 const struct sockaddr_in6 *gsin6;
293 struct in6_multi **pinm;
294 int idx;
295 int nmships;
296
297 gsin6 = (const struct sockaddr_in6 *)group;
298
299 /* The im6o_membership array may be lazy allocated. */
300 if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
301 return (-1);
302
303 nmships = imo->im6o_num_memberships;
304 pinm = &imo->im6o_membership[0];
305 for (idx = 0; idx < nmships; idx++, pinm++) {
306 if (*pinm == NULL)
307 continue;
308 if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
309 IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
310 &gsin6->sin6_addr)) {
311 break;
312 }
313 }
314 if (idx >= nmships)
315 idx = -1;
316
317 return (idx);
318 }
319
320 /*
321 * Find an IPv6 multicast source entry for this imo which matches
322 * the given group index for this socket, and source address.
323 *
324 * XXX TODO: The scope ID, if present in src, is stripped before
325 * any comparison. We SHOULD enforce scope/zone checks where the source
326 * filter entry has a link scope.
327 *
328 * NOTE: This does not check if the entry is in-mode, merely if
329 * it exists, which may not be the desired behaviour.
330 */
331 static struct in6_msource *
332 im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
333 const struct sockaddr *src)
334 {
335 struct ip6_msource find;
336 struct in6_mfilter *imf;
337 struct ip6_msource *ims;
338 const sockunion_t *psa;
339
340 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
341 KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
342 ("%s: invalid index %d\n", __func__, (int)gidx));
343
344 /* The im6o_mfilters array may be lazy allocated. */
345 if (imo->im6o_mfilters == NULL)
346 return (NULL);
347 imf = &imo->im6o_mfilters[gidx];
348
349 psa = (const sockunion_t *)src;
350 find.im6s_addr = psa->sin6.sin6_addr;
351 in6_clearscope(&find.im6s_addr); /* XXX */
352 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
353
354 return ((struct in6_msource *)ims);
355 }
356
357 /*
358 * Perform filtering for multicast datagrams on a socket by group and source.
359 *
360 * Returns 0 if a datagram should be allowed through, or various error codes
361 * if the socket was not a member of the group, or the source was muted, etc.
362 */
363 int
364 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
365 const struct sockaddr *group, const struct sockaddr *src)
366 {
367 size_t gidx;
368 struct in6_msource *ims;
369 int mode;
370
371 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
372
373 gidx = im6o_match_group(imo, ifp, group);
374 if (gidx == -1)
375 return (MCAST_NOTGMEMBER);
376
377 /*
378 * Check if the source was included in an (S,G) join.
379 * Allow reception on exclusive memberships by default,
380 * reject reception on inclusive memberships by default.
381 * Exclude source only if an in-mode exclude filter exists.
382 * Include source only if an in-mode include filter exists.
383 * NOTE: We are comparing group state here at MLD t1 (now)
384 * with socket-layer t0 (since last downcall).
385 */
386 mode = imo->im6o_mfilters[gidx].im6f_st[1];
387 ims = im6o_match_source(imo, gidx, src);
388
389 if ((ims == NULL && mode == MCAST_INCLUDE) ||
390 (ims != NULL && ims->im6sl_st[0] != mode))
391 return (MCAST_NOTSMEMBER);
392
393 return (MCAST_PASS);
394 }
395
396 /*
397 * Find and return a reference to an in6_multi record for (ifp, group),
398 * and bump its reference count.
399 * If one does not exist, try to allocate it, and update link-layer multicast
400 * filters on ifp to listen for group.
401 * Assumes the IN6_MULTI lock is held across the call.
402 * Return 0 if successful, otherwise return an appropriate error code.
403 */
404 static int
405 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
406 struct in6_multi **pinm)
407 {
408 struct sockaddr_in6 gsin6;
409 struct ifmultiaddr *ifma;
410 struct in6_multi *inm;
411 int error;
412
413 error = 0;
414
415 /*
416 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
417 * if_addmulti() takes this mutex itself, so we must drop and
418 * re-acquire around the call.
419 */
420 IN6_MULTI_LOCK_ASSERT();
421 IN6_MULTI_LIST_LOCK();
422 IF_ADDR_WLOCK(ifp);
423 inm = in6m_lookup_locked(ifp, group);
424 if (inm != NULL) {
425 /*
426 * If we already joined this group, just bump the
427 * refcount and return it.
428 */
429 KASSERT(inm->in6m_refcount >= 1,
430 ("%s: bad refcount %d", __func__, inm->in6m_refcount));
431 in6m_acquire_locked(inm);
432 *pinm = inm;
433 goto out_locked;
434 }
435
436 memset(&gsin6, 0, sizeof(gsin6));
437 gsin6.sin6_family = AF_INET6;
438 gsin6.sin6_len = sizeof(struct sockaddr_in6);
439 gsin6.sin6_addr = *group;
440
441 /*
442 * Check if a link-layer group is already associated
443 * with this network-layer group on the given ifnet.
444 */
445 IN6_MULTI_LIST_UNLOCK();
446 IF_ADDR_WUNLOCK(ifp);
447 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
448 if (error != 0)
449 return (error);
450 IN6_MULTI_LIST_LOCK();
451 IF_ADDR_WLOCK(ifp);
452
453 /*
454 * If something other than netinet6 is occupying the link-layer
455 * group, print a meaningful error message and back out of
456 * the allocation.
457 * Otherwise, bump the refcount on the existing network-layer
458 * group association and return it.
459 */
460 if (ifma->ifma_protospec != NULL) {
461 inm = (struct in6_multi *)ifma->ifma_protospec;
462 #ifdef INVARIANTS
463 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
464 __func__));
465 KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
466 ("%s: ifma not AF_INET6", __func__));
467 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
468 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
469 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
470 panic("%s: ifma %p is inconsistent with %p (%p)",
471 __func__, ifma, inm, group);
472 #endif
473 in6m_acquire_locked(inm);
474 *pinm = inm;
475 goto out_locked;
476 }
477
478 IF_ADDR_WLOCK_ASSERT(ifp);
479
480 /*
481 * A new in6_multi record is needed; allocate and initialize it.
482 * We DO NOT perform an MLD join as the in6_ layer may need to
483 * push an initial source list down to MLD to support SSM.
484 *
485 * The initial source filter state is INCLUDE, {} as per the RFC.
486 * Pending state-changes per group are subject to a bounds check.
487 */
488 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
489 if (inm == NULL) {
490 IN6_MULTI_LIST_UNLOCK();
491 IF_ADDR_WUNLOCK(ifp);
492 if_delmulti_ifma(ifma);
493 return (ENOMEM);
494 }
495 inm->in6m_addr = *group;
496 inm->in6m_ifp = ifp;
497 inm->in6m_mli = MLD_IFINFO(ifp);
498 inm->in6m_ifma = ifma;
499 inm->in6m_refcount = 1;
500 inm->in6m_state = MLD_NOT_MEMBER;
501 mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
502
503 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
504 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
505 RB_INIT(&inm->in6m_srcs);
506
507 ifma->ifma_protospec = inm;
508 *pinm = inm;
509
510 out_locked:
511 IN6_MULTI_LIST_UNLOCK();
512 IF_ADDR_WUNLOCK(ifp);
513 return (error);
514 }
515
516 /*
517 * Drop a reference to an in6_multi record.
518 *
519 * If the refcount drops to 0, free the in6_multi record and
520 * delete the underlying link-layer membership.
521 */
522 static void
523 in6m_release(struct in6_multi *inm)
524 {
525 struct ifmultiaddr *ifma;
526 struct ifnet *ifp;
527
528 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
529
530 MPASS(inm->in6m_refcount == 0);
531 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
532
533 ifma = inm->in6m_ifma;
534 ifp = inm->in6m_ifp;
535 MPASS(ifma->ifma_llifma == NULL);
536
537 /* XXX this access is not covered by IF_ADDR_LOCK */
538 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
539 KASSERT(ifma->ifma_protospec == NULL,
540 ("%s: ifma_protospec != NULL", __func__));
541 if (ifp == NULL)
542 ifp = ifma->ifma_ifp;
543
544 if (ifp != NULL) {
545 CURVNET_SET(ifp->if_vnet);
546 in6m_purge(inm);
547 free(inm, M_IP6MADDR);
548 if_delmulti_ifma_flags(ifma, 1);
549 CURVNET_RESTORE();
550 if_rele(ifp);
551 } else {
552 in6m_purge(inm);
553 free(inm, M_IP6MADDR);
554 if_delmulti_ifma_flags(ifma, 1);
555 }
556 }
557
558 static struct grouptask free_gtask;
559 static struct in6_multi_head in6m_free_list;
560 static void in6m_release_task(void *arg __unused);
561 static void in6m_init(void)
562 {
563 SLIST_INIT(&in6m_free_list);
564 taskqgroup_config_gtask_init(NULL, &free_gtask, in6m_release_task, "in6m release task");
565 }
566
567 #ifdef EARLY_AP_STARTUP
568 SYSINIT(in6m_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
569 in6m_init, NULL);
570 #else
571 SYSINIT(in6m_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_SECOND,
572 in6m_init, NULL);
573 #endif
574
575
576 void
577 in6m_release_list_deferred(struct in6_multi_head *inmh)
578 {
579 if (SLIST_EMPTY(inmh))
580 return;
581 mtx_lock(&in6_multi_free_mtx);
582 SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
583 mtx_unlock(&in6_multi_free_mtx);
584 GROUPTASK_ENQUEUE(&free_gtask);
585 }
586
587 void
588 in6m_disconnect(struct in6_multi *inm)
589 {
590 struct ifnet *ifp;
591 struct ifaddr *ifa;
592 struct in6_ifaddr *ifa6;
593 struct in6_multi_mship *imm, *imm_tmp;
594 struct ifmultiaddr *ifma, *ll_ifma;
595
596 ifp = inm->in6m_ifp;
597
598 if (ifp == NULL)
599 return;
600 inm->in6m_ifp = NULL;
601 IF_ADDR_WLOCK_ASSERT(ifp);
602 ifma = inm->in6m_ifma;
603 if (ifma == NULL)
604 return;
605
606 if_ref(ifp);
607 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
608 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
609 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
610 }
611 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
612 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
613 MPASS(ifma != ll_ifma);
614 ifma->ifma_llifma = NULL;
615 MPASS(ll_ifma->ifma_llifma == NULL);
616 MPASS(ll_ifma->ifma_ifp == ifp);
617 if (--ll_ifma->ifma_refcount == 0) {
618 ifma6_restart = true;
619 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
620 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
621 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
622 }
623 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
624 if_freemulti(ll_ifma);
625 }
626 }
627 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
628 if (ifa->ifa_addr->sa_family != AF_INET6)
629 continue;
630 ifa6 = (void *)ifa;
631 LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
632 i6mm_chain, imm_tmp) {
633 if (inm == imm->i6mm_maddr) {
634 LIST_REMOVE(imm, i6mm_chain);
635 free(imm, M_IP6MADDR);
636 }
637 }
638 }
639 }
640
641 void
642 in6m_release_deferred(struct in6_multi *inm)
643 {
644 struct in6_multi_head tmp;
645
646 IN6_MULTI_LIST_LOCK_ASSERT();
647 KASSERT(inm->in6m_refcount > 0, ("refcount == %d inm: %p", inm->in6m_refcount, inm));
648 if (--inm->in6m_refcount == 0) {
649 MPASS(inm->in6m_ifp == NULL);
650 SLIST_INIT(&tmp);
651 inm->in6m_ifma->ifma_protospec = NULL;
652 MPASS(inm->in6m_ifma->ifma_llifma == NULL);
653 SLIST_INSERT_HEAD(&tmp, inm, in6m_nrele);
654 in6m_release_list_deferred(&tmp);
655 }
656 }
657
658 static void
659 in6m_release_task(void *arg __unused)
660 {
661 struct in6_multi_head in6m_free_tmp;
662 struct in6_multi *inm, *tinm;
663
664 SLIST_INIT(&in6m_free_tmp);
665 mtx_lock(&in6_multi_free_mtx);
666 SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
667 mtx_unlock(&in6_multi_free_mtx);
668 IN6_MULTI_LOCK();
669 SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
670 SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
671 in6m_release(inm);
672 }
673 IN6_MULTI_UNLOCK();
674 }
675
676 /*
677 * Clear recorded source entries for a group.
678 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
679 * FIXME: Should reap.
680 */
681 void
682 in6m_clear_recorded(struct in6_multi *inm)
683 {
684 struct ip6_msource *ims;
685
686 IN6_MULTI_LIST_LOCK_ASSERT();
687
688 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
689 if (ims->im6s_stp) {
690 ims->im6s_stp = 0;
691 --inm->in6m_st[1].iss_rec;
692 }
693 }
694 KASSERT(inm->in6m_st[1].iss_rec == 0,
695 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
696 }
697
698 /*
699 * Record a source as pending for a Source-Group MLDv2 query.
700 * This lives here as it modifies the shared tree.
701 *
702 * inm is the group descriptor.
703 * naddr is the address of the source to record in network-byte order.
704 *
705 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
706 * lazy-allocate a source node in response to an SG query.
707 * Otherwise, no allocation is performed. This saves some memory
708 * with the trade-off that the source will not be reported to the
709 * router if joined in the window between the query response and
710 * the group actually being joined on the local host.
711 *
712 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
713 * This turns off the allocation of a recorded source entry if
714 * the group has not been joined.
715 *
716 * Return 0 if the source didn't exist or was already marked as recorded.
717 * Return 1 if the source was marked as recorded by this function.
718 * Return <0 if any error occurred (negated errno code).
719 */
720 int
721 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
722 {
723 struct ip6_msource find;
724 struct ip6_msource *ims, *nims;
725
726 IN6_MULTI_LIST_LOCK_ASSERT();
727
728 find.im6s_addr = *addr;
729 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
730 if (ims && ims->im6s_stp)
731 return (0);
732 if (ims == NULL) {
733 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
734 return (-ENOSPC);
735 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
736 M_NOWAIT | M_ZERO);
737 if (nims == NULL)
738 return (-ENOMEM);
739 nims->im6s_addr = find.im6s_addr;
740 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
741 ++inm->in6m_nsrc;
742 ims = nims;
743 }
744
745 /*
746 * Mark the source as recorded and update the recorded
747 * source count.
748 */
749 ++ims->im6s_stp;
750 ++inm->in6m_st[1].iss_rec;
751
752 return (1);
753 }
754
755 /*
756 * Return a pointer to an in6_msource owned by an in6_mfilter,
757 * given its source address.
758 * Lazy-allocate if needed. If this is a new entry its filter state is
759 * undefined at t0.
760 *
761 * imf is the filter set being modified.
762 * addr is the source address.
763 *
764 * SMPng: May be called with locks held; malloc must not block.
765 */
766 static int
767 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
768 struct in6_msource **plims)
769 {
770 struct ip6_msource find;
771 struct ip6_msource *ims, *nims;
772 struct in6_msource *lims;
773 int error;
774
775 error = 0;
776 ims = NULL;
777 lims = NULL;
778
779 find.im6s_addr = psin->sin6_addr;
780 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
781 lims = (struct in6_msource *)ims;
782 if (lims == NULL) {
783 if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
784 return (ENOSPC);
785 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
786 M_NOWAIT | M_ZERO);
787 if (nims == NULL)
788 return (ENOMEM);
789 lims = (struct in6_msource *)nims;
790 lims->im6s_addr = find.im6s_addr;
791 lims->im6sl_st[0] = MCAST_UNDEFINED;
792 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
793 ++imf->im6f_nsrc;
794 }
795
796 *plims = lims;
797
798 return (error);
799 }
800
801 /*
802 * Graft a source entry into an existing socket-layer filter set,
803 * maintaining any required invariants and checking allocations.
804 *
805 * The source is marked as being in the new filter mode at t1.
806 *
807 * Return the pointer to the new node, otherwise return NULL.
808 */
809 static struct in6_msource *
810 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
811 const struct sockaddr_in6 *psin)
812 {
813 struct ip6_msource *nims;
814 struct in6_msource *lims;
815
816 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
817 M_NOWAIT | M_ZERO);
818 if (nims == NULL)
819 return (NULL);
820 lims = (struct in6_msource *)nims;
821 lims->im6s_addr = psin->sin6_addr;
822 lims->im6sl_st[0] = MCAST_UNDEFINED;
823 lims->im6sl_st[1] = st1;
824 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
825 ++imf->im6f_nsrc;
826
827 return (lims);
828 }
829
830 /*
831 * Prune a source entry from an existing socket-layer filter set,
832 * maintaining any required invariants and checking allocations.
833 *
834 * The source is marked as being left at t1, it is not freed.
835 *
836 * Return 0 if no error occurred, otherwise return an errno value.
837 */
838 static int
839 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
840 {
841 struct ip6_msource find;
842 struct ip6_msource *ims;
843 struct in6_msource *lims;
844
845 find.im6s_addr = psin->sin6_addr;
846 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
847 if (ims == NULL)
848 return (ENOENT);
849 lims = (struct in6_msource *)ims;
850 lims->im6sl_st[1] = MCAST_UNDEFINED;
851 return (0);
852 }
853
854 /*
855 * Revert socket-layer filter set deltas at t1 to t0 state.
856 */
857 static void
858 im6f_rollback(struct in6_mfilter *imf)
859 {
860 struct ip6_msource *ims, *tims;
861 struct in6_msource *lims;
862
863 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
864 lims = (struct in6_msource *)ims;
865 if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
866 /* no change at t1 */
867 continue;
868 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
869 /* revert change to existing source at t1 */
870 lims->im6sl_st[1] = lims->im6sl_st[0];
871 } else {
872 /* revert source added t1 */
873 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
874 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
875 free(ims, M_IN6MFILTER);
876 imf->im6f_nsrc--;
877 }
878 }
879 imf->im6f_st[1] = imf->im6f_st[0];
880 }
881
882 /*
883 * Mark socket-layer filter set as INCLUDE {} at t1.
884 */
885 static void
886 im6f_leave(struct in6_mfilter *imf)
887 {
888 struct ip6_msource *ims;
889 struct in6_msource *lims;
890
891 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
892 lims = (struct in6_msource *)ims;
893 lims->im6sl_st[1] = MCAST_UNDEFINED;
894 }
895 imf->im6f_st[1] = MCAST_INCLUDE;
896 }
897
898 /*
899 * Mark socket-layer filter set deltas as committed.
900 */
901 static void
902 im6f_commit(struct in6_mfilter *imf)
903 {
904 struct ip6_msource *ims;
905 struct in6_msource *lims;
906
907 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
908 lims = (struct in6_msource *)ims;
909 lims->im6sl_st[0] = lims->im6sl_st[1];
910 }
911 imf->im6f_st[0] = imf->im6f_st[1];
912 }
913
914 /*
915 * Reap unreferenced sources from socket-layer filter set.
916 */
917 static void
918 im6f_reap(struct in6_mfilter *imf)
919 {
920 struct ip6_msource *ims, *tims;
921 struct in6_msource *lims;
922
923 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
924 lims = (struct in6_msource *)ims;
925 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
926 (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
927 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
928 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
929 free(ims, M_IN6MFILTER);
930 imf->im6f_nsrc--;
931 }
932 }
933 }
934
935 /*
936 * Purge socket-layer filter set.
937 */
938 static void
939 im6f_purge(struct in6_mfilter *imf)
940 {
941 struct ip6_msource *ims, *tims;
942
943 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
944 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
945 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
946 free(ims, M_IN6MFILTER);
947 imf->im6f_nsrc--;
948 }
949 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
950 KASSERT(RB_EMPTY(&imf->im6f_sources),
951 ("%s: im6f_sources not empty", __func__));
952 }
953
954 /*
955 * Look up a source filter entry for a multicast group.
956 *
957 * inm is the group descriptor to work with.
958 * addr is the IPv6 address to look up.
959 * noalloc may be non-zero to suppress allocation of sources.
960 * *pims will be set to the address of the retrieved or allocated source.
961 *
962 * SMPng: NOTE: may be called with locks held.
963 * Return 0 if successful, otherwise return a non-zero error code.
964 */
965 static int
966 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
967 const int noalloc, struct ip6_msource **pims)
968 {
969 struct ip6_msource find;
970 struct ip6_msource *ims, *nims;
971 #ifdef KTR
972 char ip6tbuf[INET6_ADDRSTRLEN];
973 #endif
974
975 find.im6s_addr = *addr;
976 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
977 if (ims == NULL && !noalloc) {
978 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
979 return (ENOSPC);
980 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
981 M_NOWAIT | M_ZERO);
982 if (nims == NULL)
983 return (ENOMEM);
984 nims->im6s_addr = *addr;
985 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
986 ++inm->in6m_nsrc;
987 ims = nims;
988 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
989 ip6_sprintf(ip6tbuf, addr), ims);
990 }
991
992 *pims = ims;
993 return (0);
994 }
995
996 /*
997 * Merge socket-layer source into MLD-layer source.
998 * If rollback is non-zero, perform the inverse of the merge.
999 */
1000 static void
1001 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
1002 const int rollback)
1003 {
1004 int n = rollback ? -1 : 1;
1005 #ifdef KTR
1006 char ip6tbuf[INET6_ADDRSTRLEN];
1007
1008 ip6_sprintf(ip6tbuf, &lims->im6s_addr);
1009 #endif
1010
1011 if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
1012 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
1013 ims->im6s_st[1].ex -= n;
1014 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
1015 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
1016 ims->im6s_st[1].in -= n;
1017 }
1018
1019 if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
1020 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
1021 ims->im6s_st[1].ex += n;
1022 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
1023 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
1024 ims->im6s_st[1].in += n;
1025 }
1026 }
1027
1028 /*
1029 * Atomically update the global in6_multi state, when a membership's
1030 * filter list is being updated in any way.
1031 *
1032 * imf is the per-inpcb-membership group filter pointer.
1033 * A fake imf may be passed for in-kernel consumers.
1034 *
1035 * XXX This is a candidate for a set-symmetric-difference style loop
1036 * which would eliminate the repeated lookup from root of ims nodes,
1037 * as they share the same key space.
1038 *
1039 * If any error occurred this function will back out of refcounts
1040 * and return a non-zero value.
1041 */
1042 static int
1043 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1044 {
1045 struct ip6_msource *ims, *nims;
1046 struct in6_msource *lims;
1047 int schanged, error;
1048 int nsrc0, nsrc1;
1049
1050 schanged = 0;
1051 error = 0;
1052 nsrc1 = nsrc0 = 0;
1053 IN6_MULTI_LIST_LOCK_ASSERT();
1054
1055 /*
1056 * Update the source filters first, as this may fail.
1057 * Maintain count of in-mode filters at t0, t1. These are
1058 * used to work out if we transition into ASM mode or not.
1059 * Maintain a count of source filters whose state was
1060 * actually modified by this operation.
1061 */
1062 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1063 lims = (struct in6_msource *)ims;
1064 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1065 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1066 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1067 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1068 ++schanged;
1069 if (error)
1070 break;
1071 im6s_merge(nims, lims, 0);
1072 }
1073 if (error) {
1074 struct ip6_msource *bims;
1075
1076 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1077 lims = (struct in6_msource *)ims;
1078 if (lims->im6sl_st[0] == lims->im6sl_st[1])
1079 continue;
1080 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1081 if (bims == NULL)
1082 continue;
1083 im6s_merge(bims, lims, 1);
1084 }
1085 goto out_reap;
1086 }
1087
1088 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1089 __func__, nsrc0, nsrc1);
1090
1091 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1092 if (imf->im6f_st[0] == imf->im6f_st[1] &&
1093 imf->im6f_st[1] == MCAST_INCLUDE) {
1094 if (nsrc1 == 0) {
1095 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1096 --inm->in6m_st[1].iss_in;
1097 }
1098 }
1099
1100 /* Handle filter mode transition on socket. */
1101 if (imf->im6f_st[0] != imf->im6f_st[1]) {
1102 CTR3(KTR_MLD, "%s: imf transition %d to %d",
1103 __func__, imf->im6f_st[0], imf->im6f_st[1]);
1104
1105 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1106 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1107 --inm->in6m_st[1].iss_ex;
1108 } else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1109 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1110 --inm->in6m_st[1].iss_in;
1111 }
1112
1113 if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1114 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1115 inm->in6m_st[1].iss_ex++;
1116 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1117 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1118 inm->in6m_st[1].iss_in++;
1119 }
1120 }
1121
1122 /*
1123 * Track inm filter state in terms of listener counts.
1124 * If there are any exclusive listeners, stack-wide
1125 * membership is exclusive.
1126 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1127 * If no listeners remain, state is undefined at t1,
1128 * and the MLD lifecycle for this group should finish.
1129 */
1130 if (inm->in6m_st[1].iss_ex > 0) {
1131 CTR1(KTR_MLD, "%s: transition to EX", __func__);
1132 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1133 } else if (inm->in6m_st[1].iss_in > 0) {
1134 CTR1(KTR_MLD, "%s: transition to IN", __func__);
1135 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1136 } else {
1137 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1138 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1139 }
1140
1141 /* Decrement ASM listener count on transition out of ASM mode. */
1142 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1143 if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1144 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1145 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1146 --inm->in6m_st[1].iss_asm;
1147 }
1148 }
1149
1150 /* Increment ASM listener count on transition to ASM mode. */
1151 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1152 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1153 inm->in6m_st[1].iss_asm++;
1154 }
1155
1156 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1157 in6m_print(inm);
1158
1159 out_reap:
1160 if (schanged > 0) {
1161 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1162 in6m_reap(inm);
1163 }
1164 return (error);
1165 }
1166
1167 /*
1168 * Mark an in6_multi's filter set deltas as committed.
1169 * Called by MLD after a state change has been enqueued.
1170 */
1171 void
1172 in6m_commit(struct in6_multi *inm)
1173 {
1174 struct ip6_msource *ims;
1175
1176 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1177 CTR1(KTR_MLD, "%s: pre commit:", __func__);
1178 in6m_print(inm);
1179
1180 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1181 ims->im6s_st[0] = ims->im6s_st[1];
1182 }
1183 inm->in6m_st[0] = inm->in6m_st[1];
1184 }
1185
1186 /*
1187 * Reap unreferenced nodes from an in6_multi's filter set.
1188 */
1189 static void
1190 in6m_reap(struct in6_multi *inm)
1191 {
1192 struct ip6_msource *ims, *tims;
1193
1194 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1195 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1196 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1197 ims->im6s_stp != 0)
1198 continue;
1199 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1200 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1201 free(ims, M_IP6MSOURCE);
1202 inm->in6m_nsrc--;
1203 }
1204 }
1205
1206 /*
1207 * Purge all source nodes from an in6_multi's filter set.
1208 */
1209 static void
1210 in6m_purge(struct in6_multi *inm)
1211 {
1212 struct ip6_msource *ims, *tims;
1213
1214 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1215 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1216 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1217 free(ims, M_IP6MSOURCE);
1218 inm->in6m_nsrc--;
1219 }
1220 /* Free state-change requests that might be queued. */
1221 mbufq_drain(&inm->in6m_scq);
1222 }
1223
1224 /*
1225 * Join a multicast address w/o sources.
1226 * KAME compatibility entry point.
1227 *
1228 * SMPng: Assume no mc locks held by caller.
1229 */
1230 int
1231 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1232 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1233 const int delay)
1234 {
1235 int error;
1236
1237 IN6_MULTI_LOCK();
1238 error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1239 IN6_MULTI_UNLOCK();
1240 return (error);
1241 }
1242
1243 /*
1244 * Join a multicast group; real entry point.
1245 *
1246 * Only preserves atomicity at inm level.
1247 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1248 *
1249 * If the MLD downcall fails, the group is not joined, and an error
1250 * code is returned.
1251 */
1252 int
1253 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1254 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1255 const int delay)
1256 {
1257 struct in6_mfilter timf;
1258 struct in6_multi *inm;
1259 struct ifmultiaddr *ifma;
1260 int error;
1261 #ifdef KTR
1262 char ip6tbuf[INET6_ADDRSTRLEN];
1263 #endif
1264
1265 #ifdef INVARIANTS
1266 /*
1267 * Sanity: Check scope zone ID was set for ifp, if and
1268 * only if group is scoped to an interface.
1269 */
1270 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1271 ("%s: not a multicast address", __func__));
1272 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1273 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1274 KASSERT(mcaddr->s6_addr16[1] != 0,
1275 ("%s: scope zone ID not set", __func__));
1276 }
1277 #endif
1278
1279 IN6_MULTI_LOCK_ASSERT();
1280 IN6_MULTI_LIST_UNLOCK_ASSERT();
1281
1282 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1283 ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1284
1285 error = 0;
1286 inm = NULL;
1287
1288 /*
1289 * If no imf was specified (i.e. kernel consumer),
1290 * fake one up and assume it is an ASM join.
1291 */
1292 if (imf == NULL) {
1293 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1294 imf = &timf;
1295 }
1296 error = in6_getmulti(ifp, mcaddr, &inm);
1297 if (error) {
1298 CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1299 return (error);
1300 }
1301
1302 IN6_MULTI_LIST_LOCK();
1303 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1304 error = in6m_merge(inm, imf);
1305 if (error) {
1306 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1307 goto out_in6m_release;
1308 }
1309
1310 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1311 error = mld_change_state(inm, delay);
1312 if (error) {
1313 CTR1(KTR_MLD, "%s: failed to update source", __func__);
1314 goto out_in6m_release;
1315 }
1316
1317 out_in6m_release:
1318 if (error) {
1319 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1320 IF_ADDR_RLOCK(ifp);
1321 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1322 if (ifma->ifma_protospec == inm) {
1323 ifma->ifma_protospec = NULL;
1324 break;
1325 }
1326 }
1327 in6m_disconnect(inm);
1328 in6m_release_deferred(inm);
1329 IF_ADDR_RUNLOCK(ifp);
1330 } else {
1331 *pinm = inm;
1332 }
1333 IN6_MULTI_LIST_UNLOCK();
1334 return (error);
1335 }
1336
1337 /*
1338 * Leave a multicast group; unlocked entry point.
1339 */
1340 int
1341 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1342 {
1343 int error;
1344
1345 IN6_MULTI_LOCK();
1346 error = in6_leavegroup_locked(inm, imf);
1347 IN6_MULTI_UNLOCK();
1348 return (error);
1349 }
1350
1351 /*
1352 * Leave a multicast group; real entry point.
1353 * All source filters will be expunged.
1354 *
1355 * Only preserves atomicity at inm level.
1356 *
1357 * Holding the write lock for the INP which contains imf
1358 * is highly advisable. We can't assert for it as imf does not
1359 * contain a back-pointer to the owning inp.
1360 *
1361 * Note: This is not the same as in6m_release(*) as this function also
1362 * makes a state change downcall into MLD.
1363 */
1364 int
1365 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1366 {
1367 struct in6_mfilter timf;
1368 struct ifnet *ifp;
1369 int error;
1370 #ifdef KTR
1371 char ip6tbuf[INET6_ADDRSTRLEN];
1372 #endif
1373
1374 error = 0;
1375
1376 IN6_MULTI_LOCK_ASSERT();
1377
1378 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1379 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1380 (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1381 imf);
1382
1383 /*
1384 * If no imf was specified (i.e. kernel consumer),
1385 * fake one up and assume it is an ASM join.
1386 */
1387 if (imf == NULL) {
1388 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1389 imf = &timf;
1390 }
1391
1392 /*
1393 * Begin state merge transaction at MLD layer.
1394 *
1395 * As this particular invocation should not cause any memory
1396 * to be allocated, and there is no opportunity to roll back
1397 * the transaction, it MUST NOT fail.
1398 */
1399
1400 ifp = inm->in6m_ifp;
1401 IN6_MULTI_LIST_LOCK();
1402 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1403 error = in6m_merge(inm, imf);
1404 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1405
1406 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1407 error = 0;
1408 if (ifp)
1409 error = mld_change_state(inm, 0);
1410 if (error)
1411 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1412
1413 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1414 if (ifp)
1415 IF_ADDR_WLOCK(ifp);
1416 if (inm->in6m_refcount == 1 && inm->in6m_ifp != NULL)
1417 in6m_disconnect(inm);
1418 in6m_release_deferred(inm);
1419 if (ifp)
1420 IF_ADDR_WUNLOCK(ifp);
1421 IN6_MULTI_LIST_UNLOCK();
1422
1423 return (error);
1424 }
1425
1426
1427 /*
1428 * Block or unblock an ASM multicast source on an inpcb.
1429 * This implements the delta-based API described in RFC 3678.
1430 *
1431 * The delta-based API applies only to exclusive-mode memberships.
1432 * An MLD downcall will be performed.
1433 *
1434 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1435 *
1436 * Return 0 if successful, otherwise return an appropriate error code.
1437 */
1438 static int
1439 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1440 {
1441 struct group_source_req gsr;
1442 sockunion_t *gsa, *ssa;
1443 struct ifnet *ifp;
1444 struct in6_mfilter *imf;
1445 struct ip6_moptions *imo;
1446 struct in6_msource *ims;
1447 struct in6_multi *inm;
1448 size_t idx;
1449 uint16_t fmode;
1450 int error, doblock;
1451 #ifdef KTR
1452 char ip6tbuf[INET6_ADDRSTRLEN];
1453 #endif
1454
1455 ifp = NULL;
1456 error = 0;
1457 doblock = 0;
1458
1459 memset(&gsr, 0, sizeof(struct group_source_req));
1460 gsa = (sockunion_t *)&gsr.gsr_group;
1461 ssa = (sockunion_t *)&gsr.gsr_source;
1462
1463 switch (sopt->sopt_name) {
1464 case MCAST_BLOCK_SOURCE:
1465 case MCAST_UNBLOCK_SOURCE:
1466 error = sooptcopyin(sopt, &gsr,
1467 sizeof(struct group_source_req),
1468 sizeof(struct group_source_req));
1469 if (error)
1470 return (error);
1471
1472 if (gsa->sin6.sin6_family != AF_INET6 ||
1473 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1474 return (EINVAL);
1475
1476 if (ssa->sin6.sin6_family != AF_INET6 ||
1477 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1478 return (EINVAL);
1479
1480 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1481 return (EADDRNOTAVAIL);
1482
1483 ifp = ifnet_byindex(gsr.gsr_interface);
1484
1485 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1486 doblock = 1;
1487 break;
1488
1489 default:
1490 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1491 __func__, sopt->sopt_name);
1492 return (EOPNOTSUPP);
1493 break;
1494 }
1495
1496 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1497 return (EINVAL);
1498
1499 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1500
1501 /*
1502 * Check if we are actually a member of this group.
1503 */
1504 imo = in6p_findmoptions(inp);
1505 idx = im6o_match_group(imo, ifp, &gsa->sa);
1506 if (idx == -1 || imo->im6o_mfilters == NULL) {
1507 error = EADDRNOTAVAIL;
1508 goto out_in6p_locked;
1509 }
1510
1511 KASSERT(imo->im6o_mfilters != NULL,
1512 ("%s: im6o_mfilters not allocated", __func__));
1513 imf = &imo->im6o_mfilters[idx];
1514 inm = imo->im6o_membership[idx];
1515
1516 /*
1517 * Attempting to use the delta-based API on an
1518 * non exclusive-mode membership is an error.
1519 */
1520 fmode = imf->im6f_st[0];
1521 if (fmode != MCAST_EXCLUDE) {
1522 error = EINVAL;
1523 goto out_in6p_locked;
1524 }
1525
1526 /*
1527 * Deal with error cases up-front:
1528 * Asked to block, but already blocked; or
1529 * Asked to unblock, but nothing to unblock.
1530 * If adding a new block entry, allocate it.
1531 */
1532 ims = im6o_match_source(imo, idx, &ssa->sa);
1533 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1534 CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1535 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1536 doblock ? "" : "not ");
1537 error = EADDRNOTAVAIL;
1538 goto out_in6p_locked;
1539 }
1540
1541 INP_WLOCK_ASSERT(inp);
1542
1543 /*
1544 * Begin state merge transaction at socket layer.
1545 */
1546 if (doblock) {
1547 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1548 ims = im6f_graft(imf, fmode, &ssa->sin6);
1549 if (ims == NULL)
1550 error = ENOMEM;
1551 } else {
1552 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1553 error = im6f_prune(imf, &ssa->sin6);
1554 }
1555
1556 if (error) {
1557 CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1558 goto out_im6f_rollback;
1559 }
1560
1561 /*
1562 * Begin state merge transaction at MLD layer.
1563 */
1564 IN6_MULTI_LIST_LOCK();
1565 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1566 error = in6m_merge(inm, imf);
1567 if (error)
1568 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1569 else {
1570 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1571 error = mld_change_state(inm, 0);
1572 if (error)
1573 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1574 }
1575
1576 IN6_MULTI_LIST_UNLOCK();
1577
1578 out_im6f_rollback:
1579 if (error)
1580 im6f_rollback(imf);
1581 else
1582 im6f_commit(imf);
1583
1584 im6f_reap(imf);
1585
1586 out_in6p_locked:
1587 INP_WUNLOCK(inp);
1588 return (error);
1589 }
1590
1591 /*
1592 * Given an inpcb, return its multicast options structure pointer. Accepts
1593 * an unlocked inpcb pointer, but will return it locked. May sleep.
1594 *
1595 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1596 * SMPng: NOTE: Returns with the INP write lock held.
1597 */
1598 static struct ip6_moptions *
1599 in6p_findmoptions(struct inpcb *inp)
1600 {
1601 struct ip6_moptions *imo;
1602 struct in6_multi **immp;
1603 struct in6_mfilter *imfp;
1604 size_t idx;
1605
1606 INP_WLOCK(inp);
1607 if (inp->in6p_moptions != NULL)
1608 return (inp->in6p_moptions);
1609
1610 INP_WUNLOCK(inp);
1611
1612 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1613 immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1614 M_WAITOK | M_ZERO);
1615 imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1616 M_IN6MFILTER, M_WAITOK);
1617
1618 imo->im6o_multicast_ifp = NULL;
1619 imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1620 imo->im6o_multicast_loop = in6_mcast_loop;
1621 imo->im6o_num_memberships = 0;
1622 imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1623 imo->im6o_membership = immp;
1624
1625 /* Initialize per-group source filters. */
1626 for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1627 im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1628 imo->im6o_mfilters = imfp;
1629
1630 INP_WLOCK(inp);
1631 if (inp->in6p_moptions != NULL) {
1632 free(imfp, M_IN6MFILTER);
1633 free(immp, M_IP6MOPTS);
1634 free(imo, M_IP6MOPTS);
1635 return (inp->in6p_moptions);
1636 }
1637 inp->in6p_moptions = imo;
1638 return (imo);
1639 }
1640
1641 /*
1642 * Discard the IPv6 multicast options (and source filters).
1643 *
1644 * SMPng: NOTE: assumes INP write lock is held.
1645 *
1646 * XXX can all be safely deferred to epoch_call
1647 *
1648 */
1649
1650 static void
1651 inp_gcmoptions(struct ip6_moptions *imo)
1652 {
1653 struct in6_mfilter *imf;
1654 struct in6_multi *inm;
1655 struct ifnet *ifp;
1656 size_t idx, nmships;
1657
1658 nmships = imo->im6o_num_memberships;
1659 for (idx = 0; idx < nmships; ++idx) {
1660 imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1661 if (imf)
1662 im6f_leave(imf);
1663 inm = imo->im6o_membership[idx];
1664 ifp = inm->in6m_ifp;
1665 if (ifp != NULL) {
1666 CURVNET_SET(ifp->if_vnet);
1667 (void)in6_leavegroup(inm, imf);
1668 CURVNET_RESTORE();
1669 } else {
1670 (void)in6_leavegroup(inm, imf);
1671 }
1672 if (imf)
1673 im6f_purge(imf);
1674 }
1675
1676 if (imo->im6o_mfilters)
1677 free(imo->im6o_mfilters, M_IN6MFILTER);
1678 free(imo->im6o_membership, M_IP6MOPTS);
1679 free(imo, M_IP6MOPTS);
1680 }
1681
1682 void
1683 ip6_freemoptions(struct ip6_moptions *imo)
1684 {
1685 if (imo == NULL)
1686 return;
1687 inp_gcmoptions(imo);
1688 }
1689
1690 /*
1691 * Atomically get source filters on a socket for an IPv6 multicast group.
1692 * Called with INP lock held; returns with lock released.
1693 */
1694 static int
1695 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1696 {
1697 struct __msfilterreq msfr;
1698 sockunion_t *gsa;
1699 struct ifnet *ifp;
1700 struct ip6_moptions *imo;
1701 struct in6_mfilter *imf;
1702 struct ip6_msource *ims;
1703 struct in6_msource *lims;
1704 struct sockaddr_in6 *psin;
1705 struct sockaddr_storage *ptss;
1706 struct sockaddr_storage *tss;
1707 int error;
1708 size_t idx, nsrcs, ncsrcs;
1709
1710 INP_WLOCK_ASSERT(inp);
1711
1712 imo = inp->in6p_moptions;
1713 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1714
1715 INP_WUNLOCK(inp);
1716
1717 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1718 sizeof(struct __msfilterreq));
1719 if (error)
1720 return (error);
1721
1722 if (msfr.msfr_group.ss_family != AF_INET6 ||
1723 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1724 return (EINVAL);
1725
1726 gsa = (sockunion_t *)&msfr.msfr_group;
1727 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1728 return (EINVAL);
1729
1730 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1731 return (EADDRNOTAVAIL);
1732 ifp = ifnet_byindex(msfr.msfr_ifindex);
1733 if (ifp == NULL)
1734 return (EADDRNOTAVAIL);
1735 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1736
1737 INP_WLOCK(inp);
1738
1739 /*
1740 * Lookup group on the socket.
1741 */
1742 idx = im6o_match_group(imo, ifp, &gsa->sa);
1743 if (idx == -1 || imo->im6o_mfilters == NULL) {
1744 INP_WUNLOCK(inp);
1745 return (EADDRNOTAVAIL);
1746 }
1747 imf = &imo->im6o_mfilters[idx];
1748
1749 /*
1750 * Ignore memberships which are in limbo.
1751 */
1752 if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1753 INP_WUNLOCK(inp);
1754 return (EAGAIN);
1755 }
1756 msfr.msfr_fmode = imf->im6f_st[1];
1757
1758 /*
1759 * If the user specified a buffer, copy out the source filter
1760 * entries to userland gracefully.
1761 * We only copy out the number of entries which userland
1762 * has asked for, but we always tell userland how big the
1763 * buffer really needs to be.
1764 */
1765 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1766 msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1767 tss = NULL;
1768 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1769 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1770 M_TEMP, M_NOWAIT | M_ZERO);
1771 if (tss == NULL) {
1772 INP_WUNLOCK(inp);
1773 return (ENOBUFS);
1774 }
1775 }
1776
1777 /*
1778 * Count number of sources in-mode at t0.
1779 * If buffer space exists and remains, copy out source entries.
1780 */
1781 nsrcs = msfr.msfr_nsrcs;
1782 ncsrcs = 0;
1783 ptss = tss;
1784 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1785 lims = (struct in6_msource *)ims;
1786 if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1787 lims->im6sl_st[0] != imf->im6f_st[0])
1788 continue;
1789 ++ncsrcs;
1790 if (tss != NULL && nsrcs > 0) {
1791 psin = (struct sockaddr_in6 *)ptss;
1792 psin->sin6_family = AF_INET6;
1793 psin->sin6_len = sizeof(struct sockaddr_in6);
1794 psin->sin6_addr = lims->im6s_addr;
1795 psin->sin6_port = 0;
1796 --nsrcs;
1797 ++ptss;
1798 }
1799 }
1800
1801 INP_WUNLOCK(inp);
1802
1803 if (tss != NULL) {
1804 error = copyout(tss, msfr.msfr_srcs,
1805 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1806 free(tss, M_TEMP);
1807 if (error)
1808 return (error);
1809 }
1810
1811 msfr.msfr_nsrcs = ncsrcs;
1812 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1813
1814 return (error);
1815 }
1816
1817 /*
1818 * Return the IP multicast options in response to user getsockopt().
1819 */
1820 int
1821 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1822 {
1823 struct ip6_moptions *im6o;
1824 int error;
1825 u_int optval;
1826
1827 INP_WLOCK(inp);
1828 im6o = inp->in6p_moptions;
1829 /*
1830 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1831 * or is a divert socket, reject it.
1832 */
1833 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1834 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1835 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1836 INP_WUNLOCK(inp);
1837 return (EOPNOTSUPP);
1838 }
1839
1840 error = 0;
1841 switch (sopt->sopt_name) {
1842 case IPV6_MULTICAST_IF:
1843 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1844 optval = 0;
1845 } else {
1846 optval = im6o->im6o_multicast_ifp->if_index;
1847 }
1848 INP_WUNLOCK(inp);
1849 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1850 break;
1851
1852 case IPV6_MULTICAST_HOPS:
1853 if (im6o == NULL)
1854 optval = V_ip6_defmcasthlim;
1855 else
1856 optval = im6o->im6o_multicast_hlim;
1857 INP_WUNLOCK(inp);
1858 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1859 break;
1860
1861 case IPV6_MULTICAST_LOOP:
1862 if (im6o == NULL)
1863 optval = in6_mcast_loop; /* XXX VIMAGE */
1864 else
1865 optval = im6o->im6o_multicast_loop;
1866 INP_WUNLOCK(inp);
1867 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1868 break;
1869
1870 case IPV6_MSFILTER:
1871 if (im6o == NULL) {
1872 error = EADDRNOTAVAIL;
1873 INP_WUNLOCK(inp);
1874 } else {
1875 error = in6p_get_source_filters(inp, sopt);
1876 }
1877 break;
1878
1879 default:
1880 INP_WUNLOCK(inp);
1881 error = ENOPROTOOPT;
1882 break;
1883 }
1884
1885 INP_UNLOCK_ASSERT(inp);
1886
1887 return (error);
1888 }
1889
1890 /*
1891 * Look up the ifnet to use for a multicast group membership,
1892 * given the address of an IPv6 group.
1893 *
1894 * This routine exists to support legacy IPv6 multicast applications.
1895 *
1896 * If inp is non-NULL, use this socket's current FIB number for any
1897 * required FIB lookup. Look up the group address in the unicast FIB,
1898 * and use its ifp; usually, this points to the default next-hop.
1899 * If the FIB lookup fails, return NULL.
1900 *
1901 * FUTURE: Support multiple forwarding tables for IPv6.
1902 *
1903 * Returns NULL if no ifp could be found.
1904 */
1905 static struct ifnet *
1906 in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1907 const struct sockaddr_in6 *gsin6)
1908 {
1909 struct nhop6_basic nh6;
1910 struct in6_addr dst;
1911 uint32_t scopeid;
1912 uint32_t fibnum;
1913
1914 KASSERT(in6p->inp_vflag & INP_IPV6,
1915 ("%s: not INP_IPV6 inpcb", __func__));
1916 KASSERT(gsin6->sin6_family == AF_INET6,
1917 ("%s: not AF_INET6 group", __func__));
1918
1919 in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1920 fibnum = in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB;
1921 if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6) != 0)
1922 return (NULL);
1923
1924 return (nh6.nh_ifp);
1925 }
1926
1927 /*
1928 * Join an IPv6 multicast group, possibly with a source.
1929 *
1930 * FIXME: The KAME use of the unspecified address (::)
1931 * to join *all* multicast groups is currently unsupported.
1932 */
1933 static int
1934 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1935 {
1936 struct group_source_req gsr;
1937 sockunion_t *gsa, *ssa;
1938 struct ifnet *ifp;
1939 struct in6_mfilter *imf;
1940 struct ip6_moptions *imo;
1941 struct in6_multi *inm;
1942 struct in6_msource *lims;
1943 size_t idx;
1944 int error, is_new;
1945
1946 ifp = NULL;
1947 imf = NULL;
1948 lims = NULL;
1949 error = 0;
1950 is_new = 0;
1951
1952 memset(&gsr, 0, sizeof(struct group_source_req));
1953 gsa = (sockunion_t *)&gsr.gsr_group;
1954 gsa->ss.ss_family = AF_UNSPEC;
1955 ssa = (sockunion_t *)&gsr.gsr_source;
1956 ssa->ss.ss_family = AF_UNSPEC;
1957
1958 /*
1959 * Chew everything into struct group_source_req.
1960 * Overwrite the port field if present, as the sockaddr
1961 * being copied in may be matched with a binary comparison.
1962 * Ignore passed-in scope ID.
1963 */
1964 switch (sopt->sopt_name) {
1965 case IPV6_JOIN_GROUP: {
1966 struct ipv6_mreq mreq;
1967
1968 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1969 sizeof(struct ipv6_mreq));
1970 if (error)
1971 return (error);
1972
1973 gsa->sin6.sin6_family = AF_INET6;
1974 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1975 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1976
1977 if (mreq.ipv6mr_interface == 0) {
1978 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1979 } else {
1980 if (V_if_index < mreq.ipv6mr_interface)
1981 return (EADDRNOTAVAIL);
1982 ifp = ifnet_byindex(mreq.ipv6mr_interface);
1983 }
1984 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1985 __func__, mreq.ipv6mr_interface, ifp);
1986 } break;
1987
1988 case MCAST_JOIN_GROUP:
1989 case MCAST_JOIN_SOURCE_GROUP:
1990 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1991 error = sooptcopyin(sopt, &gsr,
1992 sizeof(struct group_req),
1993 sizeof(struct group_req));
1994 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1995 error = sooptcopyin(sopt, &gsr,
1996 sizeof(struct group_source_req),
1997 sizeof(struct group_source_req));
1998 }
1999 if (error)
2000 return (error);
2001
2002 if (gsa->sin6.sin6_family != AF_INET6 ||
2003 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2004 return (EINVAL);
2005
2006 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2007 if (ssa->sin6.sin6_family != AF_INET6 ||
2008 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2009 return (EINVAL);
2010 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2011 return (EINVAL);
2012 /*
2013 * TODO: Validate embedded scope ID in source
2014 * list entry against passed-in ifp, if and only
2015 * if source list filter entry is iface or node local.
2016 */
2017 in6_clearscope(&ssa->sin6.sin6_addr);
2018 ssa->sin6.sin6_port = 0;
2019 ssa->sin6.sin6_scope_id = 0;
2020 }
2021
2022 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2023 return (EADDRNOTAVAIL);
2024 ifp = ifnet_byindex(gsr.gsr_interface);
2025 break;
2026
2027 default:
2028 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2029 __func__, sopt->sopt_name);
2030 return (EOPNOTSUPP);
2031 break;
2032 }
2033
2034 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2035 return (EINVAL);
2036
2037 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2038 return (EADDRNOTAVAIL);
2039
2040 gsa->sin6.sin6_port = 0;
2041 gsa->sin6.sin6_scope_id = 0;
2042
2043 /*
2044 * Always set the scope zone ID on memberships created from userland.
2045 * Use the passed-in ifp to do this.
2046 * XXX The in6_setscope() return value is meaningless.
2047 * XXX SCOPE6_LOCK() is taken by in6_setscope().
2048 */
2049 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2050
2051 imo = in6p_findmoptions(inp);
2052 idx = im6o_match_group(imo, ifp, &gsa->sa);
2053 if (idx == -1) {
2054 is_new = 1;
2055 } else {
2056 inm = imo->im6o_membership[idx];
2057 imf = &imo->im6o_mfilters[idx];
2058 if (ssa->ss.ss_family != AF_UNSPEC) {
2059 /*
2060 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2061 * is an error. On an existing inclusive membership,
2062 * it just adds the source to the filter list.
2063 */
2064 if (imf->im6f_st[1] != MCAST_INCLUDE) {
2065 error = EINVAL;
2066 goto out_in6p_locked;
2067 }
2068 /*
2069 * Throw out duplicates.
2070 *
2071 * XXX FIXME: This makes a naive assumption that
2072 * even if entries exist for *ssa in this imf,
2073 * they will be rejected as dupes, even if they
2074 * are not valid in the current mode (in-mode).
2075 *
2076 * in6_msource is transactioned just as for anything
2077 * else in SSM -- but note naive use of in6m_graft()
2078 * below for allocating new filter entries.
2079 *
2080 * This is only an issue if someone mixes the
2081 * full-state SSM API with the delta-based API,
2082 * which is discouraged in the relevant RFCs.
2083 */
2084 lims = im6o_match_source(imo, idx, &ssa->sa);
2085 if (lims != NULL /*&&
2086 lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2087 error = EADDRNOTAVAIL;
2088 goto out_in6p_locked;
2089 }
2090 } else {
2091 /*
2092 * MCAST_JOIN_GROUP alone, on any existing membership,
2093 * is rejected, to stop the same inpcb tying up
2094 * multiple refs to the in_multi.
2095 * On an existing inclusive membership, this is also
2096 * an error; if you want to change filter mode,
2097 * you must use the userland API setsourcefilter().
2098 * XXX We don't reject this for imf in UNDEFINED
2099 * state at t1, because allocation of a filter
2100 * is atomic with allocation of a membership.
2101 */
2102 error = EINVAL;
2103 goto out_in6p_locked;
2104 }
2105 }
2106
2107 /*
2108 * Begin state merge transaction at socket layer.
2109 */
2110 INP_WLOCK_ASSERT(inp);
2111
2112 if (is_new) {
2113 if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
2114 error = im6o_grow(imo);
2115 if (error)
2116 goto out_in6p_locked;
2117 }
2118 /*
2119 * Allocate the new slot upfront so we can deal with
2120 * grafting the new source filter in same code path
2121 * as for join-source on existing membership.
2122 */
2123 idx = imo->im6o_num_memberships;
2124 imo->im6o_membership[idx] = NULL;
2125 imo->im6o_num_memberships++;
2126 KASSERT(imo->im6o_mfilters != NULL,
2127 ("%s: im6f_mfilters vector was not allocated", __func__));
2128 imf = &imo->im6o_mfilters[idx];
2129 KASSERT(RB_EMPTY(&imf->im6f_sources),
2130 ("%s: im6f_sources not empty", __func__));
2131 }
2132
2133 /*
2134 * Graft new source into filter list for this inpcb's
2135 * membership of the group. The in6_multi may not have
2136 * been allocated yet if this is a new membership, however,
2137 * the in_mfilter slot will be allocated and must be initialized.
2138 *
2139 * Note: Grafting of exclusive mode filters doesn't happen
2140 * in this path.
2141 * XXX: Should check for non-NULL lims (node exists but may
2142 * not be in-mode) for interop with full-state API.
2143 */
2144 if (ssa->ss.ss_family != AF_UNSPEC) {
2145 /* Membership starts in IN mode */
2146 if (is_new) {
2147 CTR1(KTR_MLD, "%s: new join w/source", __func__);
2148 im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2149 } else {
2150 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2151 }
2152 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2153 if (lims == NULL) {
2154 CTR1(KTR_MLD, "%s: merge imf state failed",
2155 __func__);
2156 error = ENOMEM;
2157 goto out_im6o_free;
2158 }
2159 } else {
2160 /* No address specified; Membership starts in EX mode */
2161 if (is_new) {
2162 CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2163 im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2164 }
2165 }
2166
2167 /*
2168 * Begin state merge transaction at MLD layer.
2169 */
2170 in_pcbref(inp);
2171 INP_WUNLOCK(inp);
2172 IN6_MULTI_LOCK();
2173
2174 if (is_new) {
2175 error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2176 &inm, 0);
2177 if (error) {
2178 IN6_MULTI_UNLOCK();
2179 goto out_im6o_free;
2180 }
2181 in6m_acquire(inm);
2182 imo->im6o_membership[idx] = inm;
2183 } else {
2184 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2185 IN6_MULTI_LIST_LOCK();
2186 error = in6m_merge(inm, imf);
2187 if (error)
2188 CTR1(KTR_MLD, "%s: failed to merge inm state",
2189 __func__);
2190 else {
2191 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2192 error = mld_change_state(inm, 0);
2193 if (error)
2194 CTR1(KTR_MLD, "%s: failed mld downcall",
2195 __func__);
2196 }
2197 IN6_MULTI_LIST_UNLOCK();
2198 }
2199
2200 IN6_MULTI_UNLOCK();
2201 INP_WLOCK(inp);
2202 if (in_pcbrele_wlocked(inp))
2203 return (ENXIO);
2204 if (error) {
2205 im6f_rollback(imf);
2206 if (is_new)
2207 im6f_purge(imf);
2208 else
2209 im6f_reap(imf);
2210 } else {
2211 im6f_commit(imf);
2212 }
2213
2214 out_im6o_free:
2215 if (error && is_new) {
2216 inm = imo->im6o_membership[idx];
2217 if (inm != NULL) {
2218 IN6_MULTI_LIST_LOCK();
2219 in6m_release_deferred(inm);
2220 IN6_MULTI_LIST_UNLOCK();
2221 }
2222 imo->im6o_membership[idx] = NULL;
2223 --imo->im6o_num_memberships;
2224 }
2225
2226 out_in6p_locked:
2227 INP_WUNLOCK(inp);
2228 return (error);
2229 }
2230
2231 /*
2232 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2233 */
2234 static int
2235 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2236 {
2237 struct ipv6_mreq mreq;
2238 struct group_source_req gsr;
2239 sockunion_t *gsa, *ssa;
2240 struct ifnet *ifp;
2241 struct in6_mfilter *imf;
2242 struct ip6_moptions *imo;
2243 struct in6_msource *ims;
2244 struct in6_multi *inm;
2245 uint32_t ifindex;
2246 size_t idx;
2247 int error, is_final;
2248 #ifdef KTR
2249 char ip6tbuf[INET6_ADDRSTRLEN];
2250 #endif
2251
2252 ifp = NULL;
2253 ifindex = 0;
2254 error = 0;
2255 is_final = 1;
2256
2257 memset(&gsr, 0, sizeof(struct group_source_req));
2258 gsa = (sockunion_t *)&gsr.gsr_group;
2259 gsa->ss.ss_family = AF_UNSPEC;
2260 ssa = (sockunion_t *)&gsr.gsr_source;
2261 ssa->ss.ss_family = AF_UNSPEC;
2262
2263 /*
2264 * Chew everything passed in up into a struct group_source_req
2265 * as that is easier to process.
2266 * Note: Any embedded scope ID in the multicast group passed
2267 * in by userland is ignored, the interface index is the recommended
2268 * mechanism to specify an interface; see below.
2269 */
2270 switch (sopt->sopt_name) {
2271 case IPV6_LEAVE_GROUP:
2272 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2273 sizeof(struct ipv6_mreq));
2274 if (error)
2275 return (error);
2276 gsa->sin6.sin6_family = AF_INET6;
2277 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2278 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2279 gsa->sin6.sin6_port = 0;
2280 gsa->sin6.sin6_scope_id = 0;
2281 ifindex = mreq.ipv6mr_interface;
2282 break;
2283
2284 case MCAST_LEAVE_GROUP:
2285 case MCAST_LEAVE_SOURCE_GROUP:
2286 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2287 error = sooptcopyin(sopt, &gsr,
2288 sizeof(struct group_req),
2289 sizeof(struct group_req));
2290 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2291 error = sooptcopyin(sopt, &gsr,
2292 sizeof(struct group_source_req),
2293 sizeof(struct group_source_req));
2294 }
2295 if (error)
2296 return (error);
2297
2298 if (gsa->sin6.sin6_family != AF_INET6 ||
2299 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2300 return (EINVAL);
2301 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2302 if (ssa->sin6.sin6_family != AF_INET6 ||
2303 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2304 return (EINVAL);
2305 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2306 return (EINVAL);
2307 /*
2308 * TODO: Validate embedded scope ID in source
2309 * list entry against passed-in ifp, if and only
2310 * if source list filter entry is iface or node local.
2311 */
2312 in6_clearscope(&ssa->sin6.sin6_addr);
2313 }
2314 gsa->sin6.sin6_port = 0;
2315 gsa->sin6.sin6_scope_id = 0;
2316 ifindex = gsr.gsr_interface;
2317 break;
2318
2319 default:
2320 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2321 __func__, sopt->sopt_name);
2322 return (EOPNOTSUPP);
2323 break;
2324 }
2325
2326 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2327 return (EINVAL);
2328
2329 /*
2330 * Validate interface index if provided. If no interface index
2331 * was provided separately, attempt to look the membership up
2332 * from the default scope as a last resort to disambiguate
2333 * the membership we are being asked to leave.
2334 * XXX SCOPE6 lock potentially taken here.
2335 */
2336 if (ifindex != 0) {
2337 if (V_if_index < ifindex)
2338 return (EADDRNOTAVAIL);
2339 ifp = ifnet_byindex(ifindex);
2340 if (ifp == NULL)
2341 return (EADDRNOTAVAIL);
2342 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2343 } else {
2344 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2345 if (error)
2346 return (EADDRNOTAVAIL);
2347 /*
2348 * Some badly behaved applications don't pass an ifindex
2349 * or a scope ID, which is an API violation. In this case,
2350 * perform a lookup as per a v6 join.
2351 *
2352 * XXX For now, stomp on zone ID for the corner case.
2353 * This is not the 'KAME way', but we need to see the ifp
2354 * directly until such time as this implementation is
2355 * refactored, assuming the scope IDs are the way to go.
2356 */
2357 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2358 if (ifindex == 0) {
2359 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2360 "ifp for group %s.", __func__,
2361 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2362 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2363 } else {
2364 ifp = ifnet_byindex(ifindex);
2365 }
2366 if (ifp == NULL)
2367 return (EADDRNOTAVAIL);
2368 }
2369
2370 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2371 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2372
2373 /*
2374 * Find the membership in the membership array.
2375 */
2376 imo = in6p_findmoptions(inp);
2377 idx = im6o_match_group(imo, ifp, &gsa->sa);
2378 if (idx == -1) {
2379 error = EADDRNOTAVAIL;
2380 goto out_in6p_locked;
2381 }
2382 inm = imo->im6o_membership[idx];
2383 imf = &imo->im6o_mfilters[idx];
2384
2385 if (ssa->ss.ss_family != AF_UNSPEC)
2386 is_final = 0;
2387
2388 /*
2389 * Begin state merge transaction at socket layer.
2390 */
2391 INP_WLOCK_ASSERT(inp);
2392
2393 /*
2394 * If we were instructed only to leave a given source, do so.
2395 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2396 */
2397 if (is_final) {
2398 im6f_leave(imf);
2399 } else {
2400 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2401 error = EADDRNOTAVAIL;
2402 goto out_in6p_locked;
2403 }
2404 ims = im6o_match_source(imo, idx, &ssa->sa);
2405 if (ims == NULL) {
2406 CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2407 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2408 "not ");
2409 error = EADDRNOTAVAIL;
2410 goto out_in6p_locked;
2411 }
2412 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2413 error = im6f_prune(imf, &ssa->sin6);
2414 if (error) {
2415 CTR1(KTR_MLD, "%s: merge imf state failed",
2416 __func__);
2417 goto out_in6p_locked;
2418 }
2419 }
2420
2421 /*
2422 * Begin state merge transaction at MLD layer.
2423 */
2424 in_pcbref(inp);
2425 INP_WUNLOCK(inp);
2426 IN6_MULTI_LOCK();
2427
2428 if (is_final) {
2429 /*
2430 * Give up the multicast address record to which
2431 * the membership points.
2432 */
2433 (void)in6_leavegroup_locked(inm, imf);
2434 } else {
2435 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2436 IN6_MULTI_LIST_LOCK();
2437 error = in6m_merge(inm, imf);
2438 if (error)
2439 CTR1(KTR_MLD, "%s: failed to merge inm state",
2440 __func__);
2441 else {
2442 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2443 error = mld_change_state(inm, 0);
2444 if (error)
2445 CTR1(KTR_MLD, "%s: failed mld downcall",
2446 __func__);
2447 }
2448 IN6_MULTI_LIST_UNLOCK();
2449 }
2450
2451 IN6_MULTI_UNLOCK();
2452 INP_WLOCK(inp);
2453 if (in_pcbrele_wlocked(inp))
2454 return (ENXIO);
2455
2456 if (error)
2457 im6f_rollback(imf);
2458 else
2459 im6f_commit(imf);
2460
2461 im6f_reap(imf);
2462
2463 if (is_final) {
2464 /* Remove the gap in the membership array. */
2465 for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2466 imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2467 imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2468 }
2469 imo->im6o_num_memberships--;
2470 }
2471
2472 out_in6p_locked:
2473 INP_WUNLOCK(inp);
2474 return (error);
2475 }
2476
2477 /*
2478 * Select the interface for transmitting IPv6 multicast datagrams.
2479 *
2480 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2481 * may be passed to this socket option. An address of in6addr_any or an
2482 * interface index of 0 is used to remove a previous selection.
2483 * When no interface is selected, one is chosen for every send.
2484 */
2485 static int
2486 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2487 {
2488 struct ifnet *ifp;
2489 struct ip6_moptions *imo;
2490 u_int ifindex;
2491 int error;
2492
2493 if (sopt->sopt_valsize != sizeof(u_int))
2494 return (EINVAL);
2495
2496 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2497 if (error)
2498 return (error);
2499 if (V_if_index < ifindex)
2500 return (EINVAL);
2501 if (ifindex == 0)
2502 ifp = NULL;
2503 else {
2504 ifp = ifnet_byindex(ifindex);
2505 if (ifp == NULL)
2506 return (EINVAL);
2507 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2508 return (EADDRNOTAVAIL);
2509 }
2510 imo = in6p_findmoptions(inp);
2511 imo->im6o_multicast_ifp = ifp;
2512 INP_WUNLOCK(inp);
2513
2514 return (0);
2515 }
2516
2517 /*
2518 * Atomically set source filters on a socket for an IPv6 multicast group.
2519 *
2520 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2521 */
2522 static int
2523 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2524 {
2525 struct __msfilterreq msfr;
2526 sockunion_t *gsa;
2527 struct ifnet *ifp;
2528 struct in6_mfilter *imf;
2529 struct ip6_moptions *imo;
2530 struct in6_multi *inm;
2531 size_t idx;
2532 int error;
2533
2534 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2535 sizeof(struct __msfilterreq));
2536 if (error)
2537 return (error);
2538
2539 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2540 return (ENOBUFS);
2541
2542 if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2543 msfr.msfr_fmode != MCAST_INCLUDE)
2544 return (EINVAL);
2545
2546 if (msfr.msfr_group.ss_family != AF_INET6 ||
2547 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2548 return (EINVAL);
2549
2550 gsa = (sockunion_t *)&msfr.msfr_group;
2551 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2552 return (EINVAL);
2553
2554 gsa->sin6.sin6_port = 0; /* ignore port */
2555
2556 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2557 return (EADDRNOTAVAIL);
2558 ifp = ifnet_byindex(msfr.msfr_ifindex);
2559 if (ifp == NULL)
2560 return (EADDRNOTAVAIL);
2561 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2562
2563 /*
2564 * Take the INP write lock.
2565 * Check if this socket is a member of this group.
2566 */
2567 imo = in6p_findmoptions(inp);
2568 idx = im6o_match_group(imo, ifp, &gsa->sa);
2569 if (idx == -1 || imo->im6o_mfilters == NULL) {
2570 error = EADDRNOTAVAIL;
2571 goto out_in6p_locked;
2572 }
2573 inm = imo->im6o_membership[idx];
2574 imf = &imo->im6o_mfilters[idx];
2575
2576 /*
2577 * Begin state merge transaction at socket layer.
2578 */
2579 INP_WLOCK_ASSERT(inp);
2580
2581 imf->im6f_st[1] = msfr.msfr_fmode;
2582
2583 /*
2584 * Apply any new source filters, if present.
2585 * Make a copy of the user-space source vector so
2586 * that we may copy them with a single copyin. This
2587 * allows us to deal with page faults up-front.
2588 */
2589 if (msfr.msfr_nsrcs > 0) {
2590 struct in6_msource *lims;
2591 struct sockaddr_in6 *psin;
2592 struct sockaddr_storage *kss, *pkss;
2593 int i;
2594
2595 INP_WUNLOCK(inp);
2596
2597 CTR2(KTR_MLD, "%s: loading %lu source list entries",
2598 __func__, (unsigned long)msfr.msfr_nsrcs);
2599 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2600 M_TEMP, M_WAITOK);
2601 error = copyin(msfr.msfr_srcs, kss,
2602 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2603 if (error) {
2604 free(kss, M_TEMP);
2605 return (error);
2606 }
2607
2608 INP_WLOCK(inp);
2609
2610 /*
2611 * Mark all source filters as UNDEFINED at t1.
2612 * Restore new group filter mode, as im6f_leave()
2613 * will set it to INCLUDE.
2614 */
2615 im6f_leave(imf);
2616 imf->im6f_st[1] = msfr.msfr_fmode;
2617
2618 /*
2619 * Update socket layer filters at t1, lazy-allocating
2620 * new entries. This saves a bunch of memory at the
2621 * cost of one RB_FIND() per source entry; duplicate
2622 * entries in the msfr_nsrcs vector are ignored.
2623 * If we encounter an error, rollback transaction.
2624 *
2625 * XXX This too could be replaced with a set-symmetric
2626 * difference like loop to avoid walking from root
2627 * every time, as the key space is common.
2628 */
2629 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2630 psin = (struct sockaddr_in6 *)pkss;
2631 if (psin->sin6_family != AF_INET6) {
2632 error = EAFNOSUPPORT;
2633 break;
2634 }
2635 if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2636 error = EINVAL;
2637 break;
2638 }
2639 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2640 error = EINVAL;
2641 break;
2642 }
2643 /*
2644 * TODO: Validate embedded scope ID in source
2645 * list entry against passed-in ifp, if and only
2646 * if source list filter entry is iface or node local.
2647 */
2648 in6_clearscope(&psin->sin6_addr);
2649 error = im6f_get_source(imf, psin, &lims);
2650 if (error)
2651 break;
2652 lims->im6sl_st[1] = imf->im6f_st[1];
2653 }
2654 free(kss, M_TEMP);
2655 }
2656
2657 if (error)
2658 goto out_im6f_rollback;
2659
2660 INP_WLOCK_ASSERT(inp);
2661 IN6_MULTI_LIST_LOCK();
2662
2663 /*
2664 * Begin state merge transaction at MLD layer.
2665 */
2666 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2667 error = in6m_merge(inm, imf);
2668 if (error)
2669 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2670 else {
2671 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2672 error = mld_change_state(inm, 0);
2673 if (error)
2674 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2675 }
2676
2677 IN6_MULTI_LIST_UNLOCK();
2678
2679 out_im6f_rollback:
2680 if (error)
2681 im6f_rollback(imf);
2682 else
2683 im6f_commit(imf);
2684
2685 im6f_reap(imf);
2686
2687 out_in6p_locked:
2688 INP_WUNLOCK(inp);
2689 return (error);
2690 }
2691
2692 /*
2693 * Set the IP multicast options in response to user setsockopt().
2694 *
2695 * Many of the socket options handled in this function duplicate the
2696 * functionality of socket options in the regular unicast API. However,
2697 * it is not possible to merge the duplicate code, because the idempotence
2698 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2699 * the effects of these options must be treated as separate and distinct.
2700 *
2701 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2702 */
2703 int
2704 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2705 {
2706 struct ip6_moptions *im6o;
2707 int error;
2708
2709 error = 0;
2710
2711 /*
2712 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2713 * or is a divert socket, reject it.
2714 */
2715 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2716 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2717 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2718 return (EOPNOTSUPP);
2719
2720 switch (sopt->sopt_name) {
2721 case IPV6_MULTICAST_IF:
2722 error = in6p_set_multicast_if(inp, sopt);
2723 break;
2724
2725 case IPV6_MULTICAST_HOPS: {
2726 int hlim;
2727
2728 if (sopt->sopt_valsize != sizeof(int)) {
2729 error = EINVAL;
2730 break;
2731 }
2732 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2733 if (error)
2734 break;
2735 if (hlim < -1 || hlim > 255) {
2736 error = EINVAL;
2737 break;
2738 } else if (hlim == -1) {
2739 hlim = V_ip6_defmcasthlim;
2740 }
2741 im6o = in6p_findmoptions(inp);
2742 im6o->im6o_multicast_hlim = hlim;
2743 INP_WUNLOCK(inp);
2744 break;
2745 }
2746
2747 case IPV6_MULTICAST_LOOP: {
2748 u_int loop;
2749
2750 /*
2751 * Set the loopback flag for outgoing multicast packets.
2752 * Must be zero or one.
2753 */
2754 if (sopt->sopt_valsize != sizeof(u_int)) {
2755 error = EINVAL;
2756 break;
2757 }
2758 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2759 if (error)
2760 break;
2761 if (loop > 1) {
2762 error = EINVAL;
2763 break;
2764 }
2765 im6o = in6p_findmoptions(inp);
2766 im6o->im6o_multicast_loop = loop;
2767 INP_WUNLOCK(inp);
2768 break;
2769 }
2770
2771 case IPV6_JOIN_GROUP:
2772 case MCAST_JOIN_GROUP:
2773 case MCAST_JOIN_SOURCE_GROUP:
2774 error = in6p_join_group(inp, sopt);
2775 break;
2776
2777 case IPV6_LEAVE_GROUP:
2778 case MCAST_LEAVE_GROUP:
2779 case MCAST_LEAVE_SOURCE_GROUP:
2780 error = in6p_leave_group(inp, sopt);
2781 break;
2782
2783 case MCAST_BLOCK_SOURCE:
2784 case MCAST_UNBLOCK_SOURCE:
2785 error = in6p_block_unblock_source(inp, sopt);
2786 break;
2787
2788 case IPV6_MSFILTER:
2789 error = in6p_set_source_filters(inp, sopt);
2790 break;
2791
2792 default:
2793 error = EOPNOTSUPP;
2794 break;
2795 }
2796
2797 INP_UNLOCK_ASSERT(inp);
2798
2799 return (error);
2800 }
2801
2802 /*
2803 * Expose MLD's multicast filter mode and source list(s) to userland,
2804 * keyed by (ifindex, group).
2805 * The filter mode is written out as a uint32_t, followed by
2806 * 0..n of struct in6_addr.
2807 * For use by ifmcstat(8).
2808 * SMPng: NOTE: unlocked read of ifindex space.
2809 */
2810 static int
2811 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2812 {
2813 struct in6_addr mcaddr;
2814 struct in6_addr src;
2815 struct ifnet *ifp;
2816 struct ifmultiaddr *ifma;
2817 struct in6_multi *inm;
2818 struct ip6_msource *ims;
2819 int *name;
2820 int retval;
2821 u_int namelen;
2822 uint32_t fmode, ifindex;
2823 #ifdef KTR
2824 char ip6tbuf[INET6_ADDRSTRLEN];
2825 #endif
2826
2827 name = (int *)arg1;
2828 namelen = arg2;
2829
2830 if (req->newptr != NULL)
2831 return (EPERM);
2832
2833 /* int: ifindex + 4 * 32 bits of IPv6 address */
2834 if (namelen != 5)
2835 return (EINVAL);
2836
2837 ifindex = name[0];
2838 if (ifindex <= 0 || ifindex > V_if_index) {
2839 CTR2(KTR_MLD, "%s: ifindex %u out of range",
2840 __func__, ifindex);
2841 return (ENOENT);
2842 }
2843
2844 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2845 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2846 CTR2(KTR_MLD, "%s: group %s is not multicast",
2847 __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2848 return (EINVAL);
2849 }
2850
2851 ifp = ifnet_byindex(ifindex);
2852 if (ifp == NULL) {
2853 CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2854 __func__, ifindex);
2855 return (ENOENT);
2856 }
2857 /*
2858 * Internal MLD lookups require that scope/zone ID is set.
2859 */
2860 (void)in6_setscope(&mcaddr, ifp, NULL);
2861
2862 retval = sysctl_wire_old_buffer(req,
2863 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2864 if (retval)
2865 return (retval);
2866
2867 IN6_MULTI_LOCK();
2868 IN6_MULTI_LIST_LOCK();
2869 IF_ADDR_RLOCK(ifp);
2870 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2871 if (ifma->ifma_addr->sa_family != AF_INET6 ||
2872 ifma->ifma_protospec == NULL)
2873 continue;
2874 inm = (struct in6_multi *)ifma->ifma_protospec;
2875 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2876 continue;
2877 fmode = inm->in6m_st[1].iss_fmode;
2878 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2879 if (retval != 0)
2880 break;
2881 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2882 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2883 /*
2884 * Only copy-out sources which are in-mode.
2885 */
2886 if (fmode != im6s_get_mode(inm, ims, 1)) {
2887 CTR1(KTR_MLD, "%s: skip non-in-mode",
2888 __func__);
2889 continue;
2890 }
2891 src = ims->im6s_addr;
2892 retval = SYSCTL_OUT(req, &src,
2893 sizeof(struct in6_addr));
2894 if (retval != 0)
2895 break;
2896 }
2897 }
2898 IF_ADDR_RUNLOCK(ifp);
2899
2900 IN6_MULTI_LIST_UNLOCK();
2901 IN6_MULTI_UNLOCK();
2902
2903 return (retval);
2904 }
2905
2906 #ifdef KTR
2907
2908 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2909
2910 static const char *
2911 in6m_mode_str(const int mode)
2912 {
2913
2914 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2915 return (in6m_modestrs[mode]);
2916 return ("??");
2917 }
2918
2919 static const char *in6m_statestrs[] = {
2920 "not-member",
2921 "silent",
2922 "idle",
2923 "lazy",
2924 "sleeping",
2925 "awakening",
2926 "query-pending",
2927 "sg-query-pending",
2928 "leaving"
2929 };
2930
2931 static const char *
2932 in6m_state_str(const int state)
2933 {
2934
2935 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2936 return (in6m_statestrs[state]);
2937 return ("??");
2938 }
2939
2940 /*
2941 * Dump an in6_multi structure to the console.
2942 */
2943 void
2944 in6m_print(const struct in6_multi *inm)
2945 {
2946 int t;
2947 char ip6tbuf[INET6_ADDRSTRLEN];
2948
2949 if ((ktr_mask & KTR_MLD) == 0)
2950 return;
2951
2952 printf("%s: --- begin in6m %p ---\n", __func__, inm);
2953 printf("addr %s ifp %p(%s) ifma %p\n",
2954 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2955 inm->in6m_ifp,
2956 if_name(inm->in6m_ifp),
2957 inm->in6m_ifma);
2958 printf("timer %u state %s refcount %u scq.len %u\n",
2959 inm->in6m_timer,
2960 in6m_state_str(inm->in6m_state),
2961 inm->in6m_refcount,
2962 mbufq_len(&inm->in6m_scq));
2963 printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2964 inm->in6m_mli,
2965 inm->in6m_nsrc,
2966 inm->in6m_sctimer,
2967 inm->in6m_scrv);
2968 for (t = 0; t < 2; t++) {
2969 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2970 in6m_mode_str(inm->in6m_st[t].iss_fmode),
2971 inm->in6m_st[t].iss_asm,
2972 inm->in6m_st[t].iss_ex,
2973 inm->in6m_st[t].iss_in,
2974 inm->in6m_st[t].iss_rec);
2975 }
2976 printf("%s: --- end in6m %p ---\n", __func__, inm);
2977 }
2978
2979 #else /* !KTR */
2980
2981 void
2982 in6m_print(const struct in6_multi *inm)
2983 {
2984
2985 }
2986
2987 #endif /* KTR */
Cache object: 530574f7e19c7b13193ad8777880e352
|