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 #include "opt_inet6.h"
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/ktr.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/protosw.h>
47 #include <sys/socket.h>
48 #include <sys/socketvar.h>
49 #include <sys/sysctl.h>
50 #include <sys/priv.h>
51 #include <sys/taskqueue.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/if_private.h>
58 #include <net/route.h>
59 #include <net/route/nhop.h>
60 #include <net/vnet.h>
61
62 #include <netinet/in.h>
63 #include <netinet/udp.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip_var.h>
66 #include <netinet/udp_var.h>
67 #include <netinet6/in6_fib.h>
68 #include <netinet6/in6_var.h>
69 #include <netinet/ip6.h>
70 #include <netinet/icmp6.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet/in_pcb.h>
73 #include <netinet/tcp_var.h>
74 #include <netinet6/nd6.h>
75 #include <netinet6/mld6_var.h>
76 #include <netinet6/scope6_var.h>
77
78 #ifndef KTR_MLD
79 #define KTR_MLD KTR_INET6
80 #endif
81
82 #ifndef __SOCKUNION_DECLARED
83 union sockunion {
84 struct sockaddr_storage ss;
85 struct sockaddr sa;
86 struct sockaddr_dl sdl;
87 struct sockaddr_in6 sin6;
88 };
89 typedef union sockunion sockunion_t;
90 #define __SOCKUNION_DECLARED
91 #endif /* __SOCKUNION_DECLARED */
92
93 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
94 "IPv6 multicast PCB-layer source filter");
95 MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
96 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
97 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
98 "IPv6 multicast MLD-layer source filter");
99
100 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
101
102 /*
103 * Locking:
104 * - Lock order is: IN6_MULTI_LOCK, INP_WLOCK, IN6_MULTI_LIST_LOCK, MLD_LOCK,
105 * 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 static void im6f_commit(struct in6_mfilter *);
124 static int im6f_get_source(struct in6_mfilter *imf,
125 const struct sockaddr_in6 *psin,
126 struct in6_msource **);
127 static struct in6_msource *
128 im6f_graft(struct in6_mfilter *, const uint8_t,
129 const struct sockaddr_in6 *);
130 static void im6f_leave(struct in6_mfilter *);
131 static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
132 static void im6f_purge(struct in6_mfilter *);
133 static void im6f_rollback(struct in6_mfilter *);
134 static void im6f_reap(struct in6_mfilter *);
135 static struct in6_mfilter *
136 im6o_match_group(const struct ip6_moptions *,
137 const struct ifnet *, const struct sockaddr *);
138 static struct in6_msource *
139 im6o_match_source(struct in6_mfilter *, const struct sockaddr *);
140 static void im6s_merge(struct ip6_msource *ims,
141 const struct in6_msource *lims, const int rollback);
142 static int in6_getmulti(struct ifnet *, const struct in6_addr *,
143 struct in6_multi **);
144 static int in6_joingroup_locked(struct ifnet *, const struct in6_addr *,
145 struct in6_mfilter *, struct in6_multi **, int);
146 static int in6m_get_source(struct in6_multi *inm,
147 const struct in6_addr *addr, const int noalloc,
148 struct ip6_msource **pims);
149 #ifdef KTR
150 static int in6m_is_ifp_detached(const struct in6_multi *);
151 #endif
152 static int in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
153 static void in6m_purge(struct in6_multi *);
154 static void in6m_reap(struct in6_multi *);
155 static struct ip6_moptions *
156 in6p_findmoptions(struct inpcb *);
157 static int in6p_get_source_filters(struct inpcb *, struct sockopt *);
158 static int in6p_join_group(struct inpcb *, struct sockopt *);
159 static int in6p_leave_group(struct inpcb *, struct sockopt *);
160 static struct ifnet *
161 in6p_lookup_mcast_ifp(const struct inpcb *,
162 const struct sockaddr_in6 *);
163 static int in6p_block_unblock_source(struct inpcb *, struct sockopt *);
164 static int in6p_set_multicast_if(struct inpcb *, struct sockopt *);
165 static int in6p_set_source_filters(struct inpcb *, struct sockopt *);
166 static int sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
167
168 SYSCTL_DECL(_net_inet6_ip6); /* XXX Not in any common header. */
169
170 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast,
171 CTLFLAG_RW | CTLFLAG_MPSAFE, 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 #ifdef KTR
194 /*
195 * Inline function which wraps assertions for a valid ifp.
196 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
197 * is detached.
198 */
199 static int __inline
in6m_is_ifp_detached(const struct in6_multi * inm)200 in6m_is_ifp_detached(const struct in6_multi *inm)
201 {
202 struct ifnet *ifp;
203
204 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
205 ifp = inm->in6m_ifma->ifma_ifp;
206 if (ifp != NULL) {
207 /*
208 * Sanity check that network-layer notion of ifp is the
209 * same as that of link-layer.
210 */
211 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
212 }
213
214 return (ifp == NULL);
215 }
216 #endif
217
218 /*
219 * Initialize an in6_mfilter structure to a known state at t0, t1
220 * with an empty source filter list.
221 */
222 static __inline void
im6f_init(struct in6_mfilter * imf,const int st0,const int st1)223 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
224 {
225 memset(imf, 0, sizeof(struct in6_mfilter));
226 RB_INIT(&imf->im6f_sources);
227 imf->im6f_st[0] = st0;
228 imf->im6f_st[1] = st1;
229 }
230
231 struct in6_mfilter *
ip6_mfilter_alloc(const int mflags,const int st0,const int st1)232 ip6_mfilter_alloc(const int mflags, const int st0, const int st1)
233 {
234 struct in6_mfilter *imf;
235
236 imf = malloc(sizeof(*imf), M_IN6MFILTER, mflags);
237
238 if (imf != NULL)
239 im6f_init(imf, st0, st1);
240
241 return (imf);
242 }
243
244 void
ip6_mfilter_free(struct in6_mfilter * imf)245 ip6_mfilter_free(struct in6_mfilter *imf)
246 {
247
248 im6f_purge(imf);
249 free(imf, M_IN6MFILTER);
250 }
251
252 /*
253 * Find an IPv6 multicast group entry for this ip6_moptions instance
254 * which matches the specified group, and optionally an interface.
255 * Return its index into the array, or -1 if not found.
256 */
257 static struct in6_mfilter *
im6o_match_group(const struct ip6_moptions * imo,const struct ifnet * ifp,const struct sockaddr * group)258 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
259 const struct sockaddr *group)
260 {
261 const struct sockaddr_in6 *gsin6;
262 struct in6_mfilter *imf;
263 struct in6_multi *inm;
264
265 gsin6 = (const struct sockaddr_in6 *)group;
266
267 IP6_MFILTER_FOREACH(imf, &imo->im6o_head) {
268 inm = imf->im6f_in6m;
269 if (inm == NULL)
270 continue;
271 if ((ifp == NULL || (inm->in6m_ifp == ifp)) &&
272 IN6_ARE_ADDR_EQUAL(&inm->in6m_addr,
273 &gsin6->sin6_addr)) {
274 break;
275 }
276 }
277 return (imf);
278 }
279
280 /*
281 * Find an IPv6 multicast source entry for this imo which matches
282 * the given group index for this socket, and source address.
283 *
284 * XXX TODO: The scope ID, if present in src, is stripped before
285 * any comparison. We SHOULD enforce scope/zone checks where the source
286 * filter entry has a link scope.
287 *
288 * NOTE: This does not check if the entry is in-mode, merely if
289 * it exists, which may not be the desired behaviour.
290 */
291 static struct in6_msource *
im6o_match_source(struct in6_mfilter * imf,const struct sockaddr * src)292 im6o_match_source(struct in6_mfilter *imf, const struct sockaddr *src)
293 {
294 struct ip6_msource find;
295 struct ip6_msource *ims;
296 const sockunion_t *psa;
297
298 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
299
300 psa = (const sockunion_t *)src;
301 find.im6s_addr = psa->sin6.sin6_addr;
302 in6_clearscope(&find.im6s_addr); /* XXX */
303 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
304
305 return ((struct in6_msource *)ims);
306 }
307
308 /*
309 * Perform filtering for multicast datagrams on a socket by group and source.
310 *
311 * Returns 0 if a datagram should be allowed through, or various error codes
312 * if the socket was not a member of the group, or the source was muted, etc.
313 */
314 int
im6o_mc_filter(const struct ip6_moptions * imo,const struct ifnet * ifp,const struct sockaddr * group,const struct sockaddr * src)315 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
316 const struct sockaddr *group, const struct sockaddr *src)
317 {
318 struct in6_mfilter *imf;
319 struct in6_msource *ims;
320 int mode;
321
322 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
323
324 imf = im6o_match_group(imo, ifp, group);
325 if (imf == NULL)
326 return (MCAST_NOTGMEMBER);
327
328 /*
329 * Check if the source was included in an (S,G) join.
330 * Allow reception on exclusive memberships by default,
331 * reject reception on inclusive memberships by default.
332 * Exclude source only if an in-mode exclude filter exists.
333 * Include source only if an in-mode include filter exists.
334 * NOTE: We are comparing group state here at MLD t1 (now)
335 * with socket-layer t0 (since last downcall).
336 */
337 mode = imf->im6f_st[1];
338 ims = im6o_match_source(imf, src);
339
340 if ((ims == NULL && mode == MCAST_INCLUDE) ||
341 (ims != NULL && ims->im6sl_st[0] != mode))
342 return (MCAST_NOTSMEMBER);
343
344 return (MCAST_PASS);
345 }
346
347 /*
348 * Look up an in6_multi record for an IPv6 multicast address
349 * on the interface ifp.
350 * If no record found, return NULL.
351 *
352 * SMPng: The IN6_MULTI_LOCK and must be held and must be in network epoch.
353 */
354 struct in6_multi *
in6m_lookup_locked(struct ifnet * ifp,const struct in6_addr * mcaddr)355 in6m_lookup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr)
356 {
357 struct ifmultiaddr *ifma;
358 struct in6_multi *inm;
359
360 NET_EPOCH_ASSERT();
361
362 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
363 inm = in6m_ifmultiaddr_get_inm(ifma);
364 if (inm == NULL)
365 continue;
366 if (IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, mcaddr))
367 return (inm);
368 }
369 return (NULL);
370 }
371
372 /*
373 * Find and return a reference to an in6_multi record for (ifp, group),
374 * and bump its reference count.
375 * If one does not exist, try to allocate it, and update link-layer multicast
376 * filters on ifp to listen for group.
377 * Assumes the IN6_MULTI lock is held across the call.
378 * Return 0 if successful, otherwise return an appropriate error code.
379 */
380 static int
in6_getmulti(struct ifnet * ifp,const struct in6_addr * group,struct in6_multi ** pinm)381 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
382 struct in6_multi **pinm)
383 {
384 struct epoch_tracker et;
385 struct sockaddr_in6 gsin6;
386 struct ifmultiaddr *ifma;
387 struct in6_multi *inm;
388 int error;
389
390 error = 0;
391
392 /*
393 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
394 * if_addmulti() takes this mutex itself, so we must drop and
395 * re-acquire around the call.
396 */
397 IN6_MULTI_LOCK_ASSERT();
398 IN6_MULTI_LIST_LOCK();
399 IF_ADDR_WLOCK(ifp);
400 NET_EPOCH_ENTER(et);
401 /*
402 * Does ifp support IPv6 multicasts?
403 */
404 if (ifp->if_afdata[AF_INET6] == NULL)
405 error = ENODEV;
406 else
407 inm = in6m_lookup_locked(ifp, group);
408 NET_EPOCH_EXIT(et);
409
410 if (error != 0)
411 goto out_locked;
412
413 if (inm != NULL) {
414 /*
415 * If we already joined this group, just bump the
416 * refcount and return it.
417 */
418 KASSERT(inm->in6m_refcount >= 1,
419 ("%s: bad refcount %d", __func__, inm->in6m_refcount));
420 in6m_acquire_locked(inm);
421 *pinm = inm;
422 goto out_locked;
423 }
424
425 memset(&gsin6, 0, sizeof(gsin6));
426 gsin6.sin6_family = AF_INET6;
427 gsin6.sin6_len = sizeof(struct sockaddr_in6);
428 gsin6.sin6_addr = *group;
429
430 /*
431 * Check if a link-layer group is already associated
432 * with this network-layer group on the given ifnet.
433 */
434 IN6_MULTI_LIST_UNLOCK();
435 IF_ADDR_WUNLOCK(ifp);
436 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
437 if (error != 0)
438 return (error);
439 IN6_MULTI_LIST_LOCK();
440 IF_ADDR_WLOCK(ifp);
441
442 /*
443 * If something other than netinet6 is occupying the link-layer
444 * group, print a meaningful error message and back out of
445 * the allocation.
446 * Otherwise, bump the refcount on the existing network-layer
447 * group association and return it.
448 */
449 if (ifma->ifma_protospec != NULL) {
450 inm = (struct in6_multi *)ifma->ifma_protospec;
451 #ifdef INVARIANTS
452 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
453 __func__));
454 KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
455 ("%s: ifma not AF_INET6", __func__));
456 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
457 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
458 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
459 panic("%s: ifma %p is inconsistent with %p (%p)",
460 __func__, ifma, inm, group);
461 #endif
462 in6m_acquire_locked(inm);
463 *pinm = inm;
464 goto out_locked;
465 }
466
467 IF_ADDR_WLOCK_ASSERT(ifp);
468
469 /*
470 * A new in6_multi record is needed; allocate and initialize it.
471 * We DO NOT perform an MLD join as the in6_ layer may need to
472 * push an initial source list down to MLD to support SSM.
473 *
474 * The initial source filter state is INCLUDE, {} as per the RFC.
475 * Pending state-changes per group are subject to a bounds check.
476 */
477 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
478 if (inm == NULL) {
479 IN6_MULTI_LIST_UNLOCK();
480 IF_ADDR_WUNLOCK(ifp);
481 if_delmulti_ifma(ifma);
482 return (ENOMEM);
483 }
484 inm->in6m_addr = *group;
485 inm->in6m_ifp = ifp;
486 inm->in6m_mli = MLD_IFINFO(ifp);
487 inm->in6m_ifma = ifma;
488 inm->in6m_refcount = 1;
489 inm->in6m_state = MLD_NOT_MEMBER;
490 mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
491
492 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
493 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
494 RB_INIT(&inm->in6m_srcs);
495
496 ifma->ifma_protospec = inm;
497 *pinm = inm;
498
499 out_locked:
500 IN6_MULTI_LIST_UNLOCK();
501 IF_ADDR_WUNLOCK(ifp);
502 return (error);
503 }
504
505 /*
506 * Drop a reference to an in6_multi record.
507 *
508 * If the refcount drops to 0, free the in6_multi record and
509 * delete the underlying link-layer membership.
510 */
511 static void
in6m_release(struct in6_multi * inm)512 in6m_release(struct in6_multi *inm)
513 {
514 struct ifmultiaddr *ifma;
515 struct ifnet *ifp;
516
517 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
518
519 MPASS(inm->in6m_refcount == 0);
520 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
521
522 ifma = inm->in6m_ifma;
523 ifp = inm->in6m_ifp;
524 MPASS(ifma->ifma_llifma == NULL);
525
526 /* XXX this access is not covered by IF_ADDR_LOCK */
527 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
528 KASSERT(ifma->ifma_protospec == NULL,
529 ("%s: ifma_protospec != NULL", __func__));
530 if (ifp == NULL)
531 ifp = ifma->ifma_ifp;
532
533 if (ifp != NULL) {
534 CURVNET_SET(ifp->if_vnet);
535 in6m_purge(inm);
536 free(inm, M_IP6MADDR);
537 if_delmulti_ifma_flags(ifma, 1);
538 CURVNET_RESTORE();
539 if_rele(ifp);
540 } else {
541 in6m_purge(inm);
542 free(inm, M_IP6MADDR);
543 if_delmulti_ifma_flags(ifma, 1);
544 }
545 }
546
547 /*
548 * Interface detach can happen in a taskqueue thread context, so we must use a
549 * dedicated thread to avoid deadlocks when draining in6m_release tasks.
550 */
551 TASKQUEUE_DEFINE_THREAD(in6m_free);
552 static struct in6_multi_head in6m_free_list = SLIST_HEAD_INITIALIZER();
553 static void in6m_release_task(void *arg __unused, int pending __unused);
554 static struct task in6m_free_task = TASK_INITIALIZER(0, in6m_release_task, NULL);
555
556 void
in6m_release_list_deferred(struct in6_multi_head * inmh)557 in6m_release_list_deferred(struct in6_multi_head *inmh)
558 {
559 if (SLIST_EMPTY(inmh))
560 return;
561 mtx_lock(&in6_multi_free_mtx);
562 SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
563 mtx_unlock(&in6_multi_free_mtx);
564 taskqueue_enqueue(taskqueue_in6m_free, &in6m_free_task);
565 }
566
567 void
in6m_release_wait(void * arg __unused)568 in6m_release_wait(void *arg __unused)
569 {
570
571 /*
572 * Make sure all pending multicast addresses are freed before
573 * the VNET or network device is destroyed:
574 */
575 taskqueue_drain_all(taskqueue_in6m_free);
576 }
577 #ifdef VIMAGE
578 /* XXX-BZ FIXME, see D24914. */
579 VNET_SYSUNINIT(in6m_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, in6m_release_wait, NULL);
580 #endif
581
582 void
in6m_disconnect_locked(struct in6_multi_head * inmh,struct in6_multi * inm)583 in6m_disconnect_locked(struct in6_multi_head *inmh, struct in6_multi *inm)
584 {
585 struct ifnet *ifp;
586 struct ifaddr *ifa;
587 struct in6_ifaddr *ifa6;
588 struct in6_multi_mship *imm, *imm_tmp;
589 struct ifmultiaddr *ifma, *ll_ifma;
590
591 IN6_MULTI_LIST_LOCK_ASSERT();
592
593 ifp = inm->in6m_ifp;
594 if (ifp == NULL)
595 return; /* already called */
596
597 inm->in6m_ifp = NULL;
598 IF_ADDR_WLOCK_ASSERT(ifp);
599 ifma = inm->in6m_ifma;
600 if (ifma == NULL)
601 return;
602
603 if_ref(ifp);
604 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
605 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
606 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
607 }
608 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
609 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
610 MPASS(ifma != ll_ifma);
611 ifma->ifma_llifma = NULL;
612 MPASS(ll_ifma->ifma_llifma == NULL);
613 MPASS(ll_ifma->ifma_ifp == ifp);
614 if (--ll_ifma->ifma_refcount == 0) {
615 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
616 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
617 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
618 }
619 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
620 if_freemulti(ll_ifma);
621 }
622 }
623 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
624 if (ifa->ifa_addr->sa_family != AF_INET6)
625 continue;
626 ifa6 = (void *)ifa;
627 LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
628 i6mm_chain, imm_tmp) {
629 if (inm == imm->i6mm_maddr) {
630 LIST_REMOVE(imm, i6mm_chain);
631 free(imm, M_IP6MADDR);
632 in6m_rele_locked(inmh, inm);
633 }
634 }
635 }
636 }
637
638 static void
in6m_release_task(void * arg __unused,int pending __unused)639 in6m_release_task(void *arg __unused, int pending __unused)
640 {
641 struct in6_multi_head in6m_free_tmp;
642 struct in6_multi *inm, *tinm;
643
644 SLIST_INIT(&in6m_free_tmp);
645 mtx_lock(&in6_multi_free_mtx);
646 SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
647 mtx_unlock(&in6_multi_free_mtx);
648 IN6_MULTI_LOCK();
649 SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
650 SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
651 in6m_release(inm);
652 }
653 IN6_MULTI_UNLOCK();
654 }
655
656 /*
657 * Clear recorded source entries for a group.
658 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
659 * FIXME: Should reap.
660 */
661 void
in6m_clear_recorded(struct in6_multi * inm)662 in6m_clear_recorded(struct in6_multi *inm)
663 {
664 struct ip6_msource *ims;
665
666 IN6_MULTI_LIST_LOCK_ASSERT();
667
668 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
669 if (ims->im6s_stp) {
670 ims->im6s_stp = 0;
671 --inm->in6m_st[1].iss_rec;
672 }
673 }
674 KASSERT(inm->in6m_st[1].iss_rec == 0,
675 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
676 }
677
678 /*
679 * Record a source as pending for a Source-Group MLDv2 query.
680 * This lives here as it modifies the shared tree.
681 *
682 * inm is the group descriptor.
683 * naddr is the address of the source to record in network-byte order.
684 *
685 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
686 * lazy-allocate a source node in response to an SG query.
687 * Otherwise, no allocation is performed. This saves some memory
688 * with the trade-off that the source will not be reported to the
689 * router if joined in the window between the query response and
690 * the group actually being joined on the local host.
691 *
692 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
693 * This turns off the allocation of a recorded source entry if
694 * the group has not been joined.
695 *
696 * Return 0 if the source didn't exist or was already marked as recorded.
697 * Return 1 if the source was marked as recorded by this function.
698 * Return <0 if any error occurred (negated errno code).
699 */
700 int
in6m_record_source(struct in6_multi * inm,const struct in6_addr * addr)701 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
702 {
703 struct ip6_msource find;
704 struct ip6_msource *ims, *nims;
705
706 IN6_MULTI_LIST_LOCK_ASSERT();
707
708 find.im6s_addr = *addr;
709 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
710 if (ims && ims->im6s_stp)
711 return (0);
712 if (ims == NULL) {
713 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
714 return (-ENOSPC);
715 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
716 M_NOWAIT | M_ZERO);
717 if (nims == NULL)
718 return (-ENOMEM);
719 nims->im6s_addr = find.im6s_addr;
720 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
721 ++inm->in6m_nsrc;
722 ims = nims;
723 }
724
725 /*
726 * Mark the source as recorded and update the recorded
727 * source count.
728 */
729 ++ims->im6s_stp;
730 ++inm->in6m_st[1].iss_rec;
731
732 return (1);
733 }
734
735 /*
736 * Return a pointer to an in6_msource owned by an in6_mfilter,
737 * given its source address.
738 * Lazy-allocate if needed. If this is a new entry its filter state is
739 * undefined at t0.
740 *
741 * imf is the filter set being modified.
742 * addr is the source address.
743 *
744 * SMPng: May be called with locks held; malloc must not block.
745 */
746 static int
im6f_get_source(struct in6_mfilter * imf,const struct sockaddr_in6 * psin,struct in6_msource ** plims)747 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
748 struct in6_msource **plims)
749 {
750 struct ip6_msource find;
751 struct ip6_msource *ims, *nims;
752 struct in6_msource *lims;
753 int error;
754
755 error = 0;
756 ims = NULL;
757 lims = NULL;
758
759 find.im6s_addr = psin->sin6_addr;
760 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
761 lims = (struct in6_msource *)ims;
762 if (lims == NULL) {
763 if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
764 return (ENOSPC);
765 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
766 M_NOWAIT | M_ZERO);
767 if (nims == NULL)
768 return (ENOMEM);
769 lims = (struct in6_msource *)nims;
770 lims->im6s_addr = find.im6s_addr;
771 lims->im6sl_st[0] = MCAST_UNDEFINED;
772 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
773 ++imf->im6f_nsrc;
774 }
775
776 *plims = lims;
777
778 return (error);
779 }
780
781 /*
782 * Graft a source entry into an existing socket-layer filter set,
783 * maintaining any required invariants and checking allocations.
784 *
785 * The source is marked as being in the new filter mode at t1.
786 *
787 * Return the pointer to the new node, otherwise return NULL.
788 */
789 static struct in6_msource *
im6f_graft(struct in6_mfilter * imf,const uint8_t st1,const struct sockaddr_in6 * psin)790 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
791 const struct sockaddr_in6 *psin)
792 {
793 struct ip6_msource *nims;
794 struct in6_msource *lims;
795
796 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
797 M_NOWAIT | M_ZERO);
798 if (nims == NULL)
799 return (NULL);
800 lims = (struct in6_msource *)nims;
801 lims->im6s_addr = psin->sin6_addr;
802 lims->im6sl_st[0] = MCAST_UNDEFINED;
803 lims->im6sl_st[1] = st1;
804 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
805 ++imf->im6f_nsrc;
806
807 return (lims);
808 }
809
810 /*
811 * Prune a source entry from an existing socket-layer filter set,
812 * maintaining any required invariants and checking allocations.
813 *
814 * The source is marked as being left at t1, it is not freed.
815 *
816 * Return 0 if no error occurred, otherwise return an errno value.
817 */
818 static int
im6f_prune(struct in6_mfilter * imf,const struct sockaddr_in6 * psin)819 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
820 {
821 struct ip6_msource find;
822 struct ip6_msource *ims;
823 struct in6_msource *lims;
824
825 find.im6s_addr = psin->sin6_addr;
826 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
827 if (ims == NULL)
828 return (ENOENT);
829 lims = (struct in6_msource *)ims;
830 lims->im6sl_st[1] = MCAST_UNDEFINED;
831 return (0);
832 }
833
834 /*
835 * Revert socket-layer filter set deltas at t1 to t0 state.
836 */
837 static void
im6f_rollback(struct in6_mfilter * imf)838 im6f_rollback(struct in6_mfilter *imf)
839 {
840 struct ip6_msource *ims, *tims;
841 struct in6_msource *lims;
842
843 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
844 lims = (struct in6_msource *)ims;
845 if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
846 /* no change at t1 */
847 continue;
848 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
849 /* revert change to existing source at t1 */
850 lims->im6sl_st[1] = lims->im6sl_st[0];
851 } else {
852 /* revert source added t1 */
853 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
854 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
855 free(ims, M_IN6MFILTER);
856 imf->im6f_nsrc--;
857 }
858 }
859 imf->im6f_st[1] = imf->im6f_st[0];
860 }
861
862 /*
863 * Mark socket-layer filter set as INCLUDE {} at t1.
864 */
865 static void
im6f_leave(struct in6_mfilter * imf)866 im6f_leave(struct in6_mfilter *imf)
867 {
868 struct ip6_msource *ims;
869 struct in6_msource *lims;
870
871 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
872 lims = (struct in6_msource *)ims;
873 lims->im6sl_st[1] = MCAST_UNDEFINED;
874 }
875 imf->im6f_st[1] = MCAST_INCLUDE;
876 }
877
878 /*
879 * Mark socket-layer filter set deltas as committed.
880 */
881 static void
im6f_commit(struct in6_mfilter * imf)882 im6f_commit(struct in6_mfilter *imf)
883 {
884 struct ip6_msource *ims;
885 struct in6_msource *lims;
886
887 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
888 lims = (struct in6_msource *)ims;
889 lims->im6sl_st[0] = lims->im6sl_st[1];
890 }
891 imf->im6f_st[0] = imf->im6f_st[1];
892 }
893
894 /*
895 * Reap unreferenced sources from socket-layer filter set.
896 */
897 static void
im6f_reap(struct in6_mfilter * imf)898 im6f_reap(struct in6_mfilter *imf)
899 {
900 struct ip6_msource *ims, *tims;
901 struct in6_msource *lims;
902
903 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
904 lims = (struct in6_msource *)ims;
905 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
906 (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
907 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
908 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
909 free(ims, M_IN6MFILTER);
910 imf->im6f_nsrc--;
911 }
912 }
913 }
914
915 /*
916 * Purge socket-layer filter set.
917 */
918 static void
im6f_purge(struct in6_mfilter * imf)919 im6f_purge(struct in6_mfilter *imf)
920 {
921 struct ip6_msource *ims, *tims;
922
923 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
924 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
925 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
926 free(ims, M_IN6MFILTER);
927 imf->im6f_nsrc--;
928 }
929 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
930 KASSERT(RB_EMPTY(&imf->im6f_sources),
931 ("%s: im6f_sources not empty", __func__));
932 }
933
934 /*
935 * Look up a source filter entry for a multicast group.
936 *
937 * inm is the group descriptor to work with.
938 * addr is the IPv6 address to look up.
939 * noalloc may be non-zero to suppress allocation of sources.
940 * *pims will be set to the address of the retrieved or allocated source.
941 *
942 * SMPng: NOTE: may be called with locks held.
943 * Return 0 if successful, otherwise return a non-zero error code.
944 */
945 static int
in6m_get_source(struct in6_multi * inm,const struct in6_addr * addr,const int noalloc,struct ip6_msource ** pims)946 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
947 const int noalloc, struct ip6_msource **pims)
948 {
949 struct ip6_msource find;
950 struct ip6_msource *ims, *nims;
951 #ifdef KTR
952 char ip6tbuf[INET6_ADDRSTRLEN];
953 #endif
954
955 find.im6s_addr = *addr;
956 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
957 if (ims == NULL && !noalloc) {
958 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
959 return (ENOSPC);
960 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
961 M_NOWAIT | M_ZERO);
962 if (nims == NULL)
963 return (ENOMEM);
964 nims->im6s_addr = *addr;
965 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
966 ++inm->in6m_nsrc;
967 ims = nims;
968 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
969 ip6_sprintf(ip6tbuf, addr), ims);
970 }
971
972 *pims = ims;
973 return (0);
974 }
975
976 /*
977 * Merge socket-layer source into MLD-layer source.
978 * If rollback is non-zero, perform the inverse of the merge.
979 */
980 static void
im6s_merge(struct ip6_msource * ims,const struct in6_msource * lims,const int rollback)981 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
982 const int rollback)
983 {
984 int n = rollback ? -1 : 1;
985 #ifdef KTR
986 char ip6tbuf[INET6_ADDRSTRLEN];
987
988 ip6_sprintf(ip6tbuf, &lims->im6s_addr);
989 #endif
990
991 if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
992 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
993 ims->im6s_st[1].ex -= n;
994 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
995 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
996 ims->im6s_st[1].in -= n;
997 }
998
999 if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
1000 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
1001 ims->im6s_st[1].ex += n;
1002 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
1003 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
1004 ims->im6s_st[1].in += n;
1005 }
1006 }
1007
1008 /*
1009 * Atomically update the global in6_multi state, when a membership's
1010 * filter list is being updated in any way.
1011 *
1012 * imf is the per-inpcb-membership group filter pointer.
1013 * A fake imf may be passed for in-kernel consumers.
1014 *
1015 * XXX This is a candidate for a set-symmetric-difference style loop
1016 * which would eliminate the repeated lookup from root of ims nodes,
1017 * as they share the same key space.
1018 *
1019 * If any error occurred this function will back out of refcounts
1020 * and return a non-zero value.
1021 */
1022 static int
in6m_merge(struct in6_multi * inm,struct in6_mfilter * imf)1023 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1024 {
1025 struct ip6_msource *ims, *nims;
1026 struct in6_msource *lims;
1027 int schanged, error;
1028 int nsrc0, nsrc1;
1029
1030 schanged = 0;
1031 error = 0;
1032 nsrc1 = nsrc0 = 0;
1033 IN6_MULTI_LIST_LOCK_ASSERT();
1034
1035 /*
1036 * Update the source filters first, as this may fail.
1037 * Maintain count of in-mode filters at t0, t1. These are
1038 * used to work out if we transition into ASM mode or not.
1039 * Maintain a count of source filters whose state was
1040 * actually modified by this operation.
1041 */
1042 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1043 lims = (struct in6_msource *)ims;
1044 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1045 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1046 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1047 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1048 ++schanged;
1049 if (error)
1050 break;
1051 im6s_merge(nims, lims, 0);
1052 }
1053 if (error) {
1054 struct ip6_msource *bims;
1055
1056 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1057 lims = (struct in6_msource *)ims;
1058 if (lims->im6sl_st[0] == lims->im6sl_st[1])
1059 continue;
1060 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1061 if (bims == NULL)
1062 continue;
1063 im6s_merge(bims, lims, 1);
1064 }
1065 goto out_reap;
1066 }
1067
1068 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1069 __func__, nsrc0, nsrc1);
1070
1071 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1072 if (imf->im6f_st[0] == imf->im6f_st[1] &&
1073 imf->im6f_st[1] == MCAST_INCLUDE) {
1074 if (nsrc1 == 0) {
1075 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1076 --inm->in6m_st[1].iss_in;
1077 }
1078 }
1079
1080 /* Handle filter mode transition on socket. */
1081 if (imf->im6f_st[0] != imf->im6f_st[1]) {
1082 CTR3(KTR_MLD, "%s: imf transition %d to %d",
1083 __func__, imf->im6f_st[0], imf->im6f_st[1]);
1084
1085 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1086 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1087 --inm->in6m_st[1].iss_ex;
1088 } else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1089 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1090 --inm->in6m_st[1].iss_in;
1091 }
1092
1093 if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1094 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1095 inm->in6m_st[1].iss_ex++;
1096 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1097 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1098 inm->in6m_st[1].iss_in++;
1099 }
1100 }
1101
1102 /*
1103 * Track inm filter state in terms of listener counts.
1104 * If there are any exclusive listeners, stack-wide
1105 * membership is exclusive.
1106 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1107 * If no listeners remain, state is undefined at t1,
1108 * and the MLD lifecycle for this group should finish.
1109 */
1110 if (inm->in6m_st[1].iss_ex > 0) {
1111 CTR1(KTR_MLD, "%s: transition to EX", __func__);
1112 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1113 } else if (inm->in6m_st[1].iss_in > 0) {
1114 CTR1(KTR_MLD, "%s: transition to IN", __func__);
1115 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1116 } else {
1117 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1118 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1119 }
1120
1121 /* Decrement ASM listener count on transition out of ASM mode. */
1122 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1123 if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1124 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1125 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1126 --inm->in6m_st[1].iss_asm;
1127 }
1128 }
1129
1130 /* Increment ASM listener count on transition to ASM mode. */
1131 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1132 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1133 inm->in6m_st[1].iss_asm++;
1134 }
1135
1136 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1137 in6m_print(inm);
1138
1139 out_reap:
1140 if (schanged > 0) {
1141 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1142 in6m_reap(inm);
1143 }
1144 return (error);
1145 }
1146
1147 /*
1148 * Mark an in6_multi's filter set deltas as committed.
1149 * Called by MLD after a state change has been enqueued.
1150 */
1151 void
in6m_commit(struct in6_multi * inm)1152 in6m_commit(struct in6_multi *inm)
1153 {
1154 struct ip6_msource *ims;
1155
1156 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1157 CTR1(KTR_MLD, "%s: pre commit:", __func__);
1158 in6m_print(inm);
1159
1160 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1161 ims->im6s_st[0] = ims->im6s_st[1];
1162 }
1163 inm->in6m_st[0] = inm->in6m_st[1];
1164 }
1165
1166 /*
1167 * Reap unreferenced nodes from an in6_multi's filter set.
1168 */
1169 static void
in6m_reap(struct in6_multi * inm)1170 in6m_reap(struct in6_multi *inm)
1171 {
1172 struct ip6_msource *ims, *tims;
1173
1174 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1175 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1176 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1177 ims->im6s_stp != 0)
1178 continue;
1179 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1180 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1181 free(ims, M_IP6MSOURCE);
1182 inm->in6m_nsrc--;
1183 }
1184 }
1185
1186 /*
1187 * Purge all source nodes from an in6_multi's filter set.
1188 */
1189 static void
in6m_purge(struct in6_multi * inm)1190 in6m_purge(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 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1196 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1197 free(ims, M_IP6MSOURCE);
1198 inm->in6m_nsrc--;
1199 }
1200 /* Free state-change requests that might be queued. */
1201 mbufq_drain(&inm->in6m_scq);
1202 }
1203
1204 /*
1205 * Join a multicast address w/o sources.
1206 * KAME compatibility entry point.
1207 *
1208 * SMPng: Assume no mc locks held by caller.
1209 */
1210 int
in6_joingroup(struct ifnet * ifp,const struct in6_addr * mcaddr,struct in6_mfilter * imf,struct in6_multi ** pinm,const int delay)1211 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1212 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1213 const int delay)
1214 {
1215 int error;
1216
1217 IN6_MULTI_LOCK();
1218 error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1219 IN6_MULTI_UNLOCK();
1220 return (error);
1221 }
1222
1223 /*
1224 * Join a multicast group; real entry point.
1225 *
1226 * Only preserves atomicity at inm level.
1227 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1228 *
1229 * If the MLD downcall fails, the group is not joined, and an error
1230 * code is returned.
1231 */
1232 static int
in6_joingroup_locked(struct ifnet * ifp,const struct in6_addr * mcaddr,struct in6_mfilter * imf,struct in6_multi ** pinm,const int delay)1233 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1234 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1235 const int delay)
1236 {
1237 struct in6_multi_head inmh;
1238 struct in6_mfilter timf;
1239 struct in6_multi *inm;
1240 struct ifmultiaddr *ifma;
1241 int error;
1242 #ifdef KTR
1243 char ip6tbuf[INET6_ADDRSTRLEN];
1244 #endif
1245
1246 /*
1247 * Sanity: Check scope zone ID was set for ifp, if and
1248 * only if group is scoped to an interface.
1249 */
1250 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1251 ("%s: not a multicast address", __func__));
1252 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1253 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1254 KASSERT(mcaddr->s6_addr16[1] != 0,
1255 ("%s: scope zone ID not set", __func__));
1256 }
1257
1258 IN6_MULTI_LOCK_ASSERT();
1259 IN6_MULTI_LIST_UNLOCK_ASSERT();
1260
1261 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1262 ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1263
1264 error = 0;
1265 inm = NULL;
1266
1267 /*
1268 * If no imf was specified (i.e. kernel consumer),
1269 * fake one up and assume it is an ASM join.
1270 */
1271 if (imf == NULL) {
1272 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1273 imf = &timf;
1274 }
1275 error = in6_getmulti(ifp, mcaddr, &inm);
1276 if (error) {
1277 CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1278 return (error);
1279 }
1280
1281 IN6_MULTI_LIST_LOCK();
1282 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1283 error = in6m_merge(inm, imf);
1284 if (error) {
1285 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1286 goto out_in6m_release;
1287 }
1288
1289 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1290 error = mld_change_state(inm, delay);
1291 if (error) {
1292 CTR1(KTR_MLD, "%s: failed to update source", __func__);
1293 goto out_in6m_release;
1294 }
1295
1296 out_in6m_release:
1297 SLIST_INIT(&inmh);
1298 if (error) {
1299 struct epoch_tracker et;
1300
1301 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1302 IF_ADDR_WLOCK(ifp);
1303 NET_EPOCH_ENTER(et);
1304 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1305 if (ifma->ifma_protospec == inm) {
1306 ifma->ifma_protospec = NULL;
1307 break;
1308 }
1309 }
1310 in6m_disconnect_locked(&inmh, inm);
1311 in6m_rele_locked(&inmh, inm);
1312 NET_EPOCH_EXIT(et);
1313 IF_ADDR_WUNLOCK(ifp);
1314 } else {
1315 *pinm = inm;
1316 }
1317 IN6_MULTI_LIST_UNLOCK();
1318 in6m_release_list_deferred(&inmh);
1319 return (error);
1320 }
1321
1322 /*
1323 * Leave a multicast group; unlocked entry point.
1324 */
1325 int
in6_leavegroup(struct in6_multi * inm,struct in6_mfilter * imf)1326 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1327 {
1328 int error;
1329
1330 IN6_MULTI_LOCK();
1331 error = in6_leavegroup_locked(inm, imf);
1332 IN6_MULTI_UNLOCK();
1333 return (error);
1334 }
1335
1336 /*
1337 * Leave a multicast group; real entry point.
1338 * All source filters will be expunged.
1339 *
1340 * Only preserves atomicity at inm level.
1341 *
1342 * Holding the write lock for the INP which contains imf
1343 * is highly advisable. We can't assert for it as imf does not
1344 * contain a back-pointer to the owning inp.
1345 *
1346 * Note: This is not the same as in6m_release(*) as this function also
1347 * makes a state change downcall into MLD.
1348 */
1349 int
in6_leavegroup_locked(struct in6_multi * inm,struct in6_mfilter * imf)1350 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1351 {
1352 struct in6_multi_head inmh;
1353 struct in6_mfilter timf;
1354 struct ifnet *ifp;
1355 int error;
1356 #ifdef KTR
1357 char ip6tbuf[INET6_ADDRSTRLEN];
1358 #endif
1359
1360 error = 0;
1361
1362 IN6_MULTI_LOCK_ASSERT();
1363
1364 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1365 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1366 (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1367 imf);
1368
1369 /*
1370 * If no imf was specified (i.e. kernel consumer),
1371 * fake one up and assume it is an ASM join.
1372 */
1373 if (imf == NULL) {
1374 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1375 imf = &timf;
1376 }
1377
1378 /*
1379 * Begin state merge transaction at MLD layer.
1380 *
1381 * As this particular invocation should not cause any memory
1382 * to be allocated, and there is no opportunity to roll back
1383 * the transaction, it MUST NOT fail.
1384 */
1385
1386 ifp = inm->in6m_ifp;
1387 IN6_MULTI_LIST_LOCK();
1388 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1389 error = in6m_merge(inm, imf);
1390 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1391
1392 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1393 error = 0;
1394 if (ifp)
1395 error = mld_change_state(inm, 0);
1396 if (error)
1397 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1398
1399 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1400 if (ifp)
1401 IF_ADDR_WLOCK(ifp);
1402
1403 SLIST_INIT(&inmh);
1404 if (inm->in6m_refcount == 1)
1405 in6m_disconnect_locked(&inmh, inm);
1406 in6m_rele_locked(&inmh, inm);
1407 if (ifp)
1408 IF_ADDR_WUNLOCK(ifp);
1409 IN6_MULTI_LIST_UNLOCK();
1410 in6m_release_list_deferred(&inmh);
1411 return (error);
1412 }
1413
1414 /*
1415 * Block or unblock an ASM multicast source on an inpcb.
1416 * This implements the delta-based API described in RFC 3678.
1417 *
1418 * The delta-based API applies only to exclusive-mode memberships.
1419 * An MLD downcall will be performed.
1420 *
1421 * Return 0 if successful, otherwise return an appropriate error code.
1422 */
1423 static int
in6p_block_unblock_source(struct inpcb * inp,struct sockopt * sopt)1424 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1425 {
1426 struct group_source_req gsr;
1427 struct epoch_tracker et;
1428 sockunion_t *gsa, *ssa;
1429 struct ifnet *ifp;
1430 struct in6_mfilter *imf;
1431 struct ip6_moptions *imo;
1432 struct in6_msource *ims;
1433 struct in6_multi *inm;
1434 uint16_t fmode;
1435 int error, doblock;
1436 #ifdef KTR
1437 char ip6tbuf[INET6_ADDRSTRLEN];
1438 #endif
1439
1440 ifp = NULL;
1441 error = 0;
1442 doblock = 0;
1443
1444 memset(&gsr, 0, sizeof(struct group_source_req));
1445 gsa = (sockunion_t *)&gsr.gsr_group;
1446 ssa = (sockunion_t *)&gsr.gsr_source;
1447
1448 switch (sopt->sopt_name) {
1449 case MCAST_BLOCK_SOURCE:
1450 case MCAST_UNBLOCK_SOURCE:
1451 error = sooptcopyin(sopt, &gsr,
1452 sizeof(struct group_source_req),
1453 sizeof(struct group_source_req));
1454 if (error)
1455 return (error);
1456
1457 if (gsa->sin6.sin6_family != AF_INET6 ||
1458 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1459 return (EINVAL);
1460
1461 if (ssa->sin6.sin6_family != AF_INET6 ||
1462 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1463 return (EINVAL);
1464
1465 /*
1466 * XXXGL: this function should use ifnet_byindex_ref, or
1467 * expand the epoch section all the way to where we put
1468 * the reference.
1469 */
1470 NET_EPOCH_ENTER(et);
1471 ifp = ifnet_byindex(gsr.gsr_interface);
1472 NET_EPOCH_EXIT(et);
1473 if (ifp == NULL)
1474 return (EADDRNOTAVAIL);
1475
1476 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1477 doblock = 1;
1478 break;
1479
1480 default:
1481 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1482 __func__, sopt->sopt_name);
1483 return (EOPNOTSUPP);
1484 break;
1485 }
1486
1487 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1488 return (EINVAL);
1489
1490 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1491
1492 /*
1493 * Check if we are actually a member of this group.
1494 */
1495 imo = in6p_findmoptions(inp);
1496 imf = im6o_match_group(imo, ifp, &gsa->sa);
1497 if (imf == NULL) {
1498 error = EADDRNOTAVAIL;
1499 goto out_in6p_locked;
1500 }
1501 inm = imf->im6f_in6m;
1502
1503 /*
1504 * Attempting to use the delta-based API on an
1505 * non exclusive-mode membership is an error.
1506 */
1507 fmode = imf->im6f_st[0];
1508 if (fmode != MCAST_EXCLUDE) {
1509 error = EINVAL;
1510 goto out_in6p_locked;
1511 }
1512
1513 /*
1514 * Deal with error cases up-front:
1515 * Asked to block, but already blocked; or
1516 * Asked to unblock, but nothing to unblock.
1517 * If adding a new block entry, allocate it.
1518 */
1519 ims = im6o_match_source(imf, &ssa->sa);
1520 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1521 CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1522 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1523 doblock ? "" : "not ");
1524 error = EADDRNOTAVAIL;
1525 goto out_in6p_locked;
1526 }
1527
1528 INP_WLOCK_ASSERT(inp);
1529
1530 /*
1531 * Begin state merge transaction at socket layer.
1532 */
1533 if (doblock) {
1534 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1535 ims = im6f_graft(imf, fmode, &ssa->sin6);
1536 if (ims == NULL)
1537 error = ENOMEM;
1538 } else {
1539 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1540 error = im6f_prune(imf, &ssa->sin6);
1541 }
1542
1543 if (error) {
1544 CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1545 goto out_im6f_rollback;
1546 }
1547
1548 /*
1549 * Begin state merge transaction at MLD layer.
1550 */
1551 IN6_MULTI_LIST_LOCK();
1552 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1553 error = in6m_merge(inm, imf);
1554 if (error)
1555 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1556 else {
1557 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1558 error = mld_change_state(inm, 0);
1559 if (error)
1560 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1561 }
1562
1563 IN6_MULTI_LIST_UNLOCK();
1564
1565 out_im6f_rollback:
1566 if (error)
1567 im6f_rollback(imf);
1568 else
1569 im6f_commit(imf);
1570
1571 im6f_reap(imf);
1572
1573 out_in6p_locked:
1574 INP_WUNLOCK(inp);
1575 return (error);
1576 }
1577
1578 /*
1579 * Given an inpcb, return its multicast options structure pointer. Accepts
1580 * an unlocked inpcb pointer, but will return it locked. May sleep.
1581 *
1582 * SMPng: NOTE: Returns with the INP write lock held.
1583 */
1584 static struct ip6_moptions *
in6p_findmoptions(struct inpcb * inp)1585 in6p_findmoptions(struct inpcb *inp)
1586 {
1587 struct ip6_moptions *imo;
1588
1589 INP_WLOCK(inp);
1590 if (inp->in6p_moptions != NULL)
1591 return (inp->in6p_moptions);
1592
1593 INP_WUNLOCK(inp);
1594
1595 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1596
1597 imo->im6o_multicast_ifp = NULL;
1598 imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1599 imo->im6o_multicast_loop = in6_mcast_loop;
1600 STAILQ_INIT(&imo->im6o_head);
1601
1602 INP_WLOCK(inp);
1603 if (inp->in6p_moptions != NULL) {
1604 free(imo, M_IP6MOPTS);
1605 return (inp->in6p_moptions);
1606 }
1607 inp->in6p_moptions = imo;
1608 return (imo);
1609 }
1610
1611 /*
1612 * Discard the IPv6 multicast options (and source filters).
1613 *
1614 * SMPng: NOTE: assumes INP write lock is held.
1615 *
1616 * XXX can all be safely deferred to epoch_call
1617 *
1618 */
1619
1620 static void
inp_gcmoptions(struct ip6_moptions * imo)1621 inp_gcmoptions(struct ip6_moptions *imo)
1622 {
1623 struct in6_mfilter *imf;
1624 struct in6_multi *inm;
1625 struct ifnet *ifp;
1626
1627 while ((imf = ip6_mfilter_first(&imo->im6o_head)) != NULL) {
1628 ip6_mfilter_remove(&imo->im6o_head, imf);
1629
1630 im6f_leave(imf);
1631 if ((inm = imf->im6f_in6m) != NULL) {
1632 if ((ifp = inm->in6m_ifp) != NULL) {
1633 CURVNET_SET(ifp->if_vnet);
1634 (void)in6_leavegroup(inm, imf);
1635 CURVNET_RESTORE();
1636 } else {
1637 (void)in6_leavegroup(inm, imf);
1638 }
1639 }
1640 ip6_mfilter_free(imf);
1641 }
1642 free(imo, M_IP6MOPTS);
1643 }
1644
1645 void
ip6_freemoptions(struct ip6_moptions * imo)1646 ip6_freemoptions(struct ip6_moptions *imo)
1647 {
1648 if (imo == NULL)
1649 return;
1650 inp_gcmoptions(imo);
1651 }
1652
1653 /*
1654 * Atomically get source filters on a socket for an IPv6 multicast group.
1655 * Called with INP lock held; returns with lock released.
1656 */
1657 static int
in6p_get_source_filters(struct inpcb * inp,struct sockopt * sopt)1658 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1659 {
1660 struct epoch_tracker et;
1661 struct __msfilterreq msfr;
1662 sockunion_t *gsa;
1663 struct ifnet *ifp;
1664 struct ip6_moptions *imo;
1665 struct in6_mfilter *imf;
1666 struct ip6_msource *ims;
1667 struct in6_msource *lims;
1668 struct sockaddr_in6 *psin;
1669 struct sockaddr_storage *ptss;
1670 struct sockaddr_storage *tss;
1671 int error;
1672 size_t nsrcs, ncsrcs;
1673
1674 INP_WLOCK_ASSERT(inp);
1675
1676 imo = inp->in6p_moptions;
1677 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1678
1679 INP_WUNLOCK(inp);
1680
1681 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1682 sizeof(struct __msfilterreq));
1683 if (error)
1684 return (error);
1685
1686 if (msfr.msfr_group.ss_family != AF_INET6 ||
1687 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1688 return (EINVAL);
1689
1690 gsa = (sockunion_t *)&msfr.msfr_group;
1691 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1692 return (EINVAL);
1693
1694 /*
1695 * XXXGL: this function should use ifnet_byindex_ref, or expand the
1696 * epoch section all the way to where the interface is referenced.
1697 */
1698 NET_EPOCH_ENTER(et);
1699 ifp = ifnet_byindex(msfr.msfr_ifindex);
1700 NET_EPOCH_EXIT(et);
1701 if (ifp == NULL)
1702 return (EADDRNOTAVAIL);
1703 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1704
1705 INP_WLOCK(inp);
1706
1707 /*
1708 * Lookup group on the socket.
1709 */
1710 imf = im6o_match_group(imo, ifp, &gsa->sa);
1711 if (imf == NULL) {
1712 INP_WUNLOCK(inp);
1713 return (EADDRNOTAVAIL);
1714 }
1715
1716 /*
1717 * Ignore memberships which are in limbo.
1718 */
1719 if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1720 INP_WUNLOCK(inp);
1721 return (EAGAIN);
1722 }
1723 msfr.msfr_fmode = imf->im6f_st[1];
1724
1725 /*
1726 * If the user specified a buffer, copy out the source filter
1727 * entries to userland gracefully.
1728 * We only copy out the number of entries which userland
1729 * has asked for, but we always tell userland how big the
1730 * buffer really needs to be.
1731 */
1732 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1733 msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1734 tss = NULL;
1735 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1736 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1737 M_TEMP, M_NOWAIT | M_ZERO);
1738 if (tss == NULL) {
1739 INP_WUNLOCK(inp);
1740 return (ENOBUFS);
1741 }
1742 }
1743
1744 /*
1745 * Count number of sources in-mode at t0.
1746 * If buffer space exists and remains, copy out source entries.
1747 */
1748 nsrcs = msfr.msfr_nsrcs;
1749 ncsrcs = 0;
1750 ptss = tss;
1751 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1752 lims = (struct in6_msource *)ims;
1753 if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1754 lims->im6sl_st[0] != imf->im6f_st[0])
1755 continue;
1756 ++ncsrcs;
1757 if (tss != NULL && nsrcs > 0) {
1758 psin = (struct sockaddr_in6 *)ptss;
1759 psin->sin6_family = AF_INET6;
1760 psin->sin6_len = sizeof(struct sockaddr_in6);
1761 psin->sin6_addr = lims->im6s_addr;
1762 psin->sin6_port = 0;
1763 --nsrcs;
1764 ++ptss;
1765 }
1766 }
1767
1768 INP_WUNLOCK(inp);
1769
1770 if (tss != NULL) {
1771 error = copyout(tss, msfr.msfr_srcs,
1772 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1773 free(tss, M_TEMP);
1774 if (error)
1775 return (error);
1776 }
1777
1778 msfr.msfr_nsrcs = ncsrcs;
1779 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1780
1781 return (error);
1782 }
1783
1784 /*
1785 * Return the IP multicast options in response to user getsockopt().
1786 */
1787 int
ip6_getmoptions(struct inpcb * inp,struct sockopt * sopt)1788 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1789 {
1790 struct ip6_moptions *im6o;
1791 int error;
1792 u_int optval;
1793
1794 INP_WLOCK(inp);
1795 im6o = inp->in6p_moptions;
1796 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
1797 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1798 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM) {
1799 INP_WUNLOCK(inp);
1800 return (EOPNOTSUPP);
1801 }
1802
1803 error = 0;
1804 switch (sopt->sopt_name) {
1805 case IPV6_MULTICAST_IF:
1806 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1807 optval = 0;
1808 } else {
1809 optval = im6o->im6o_multicast_ifp->if_index;
1810 }
1811 INP_WUNLOCK(inp);
1812 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1813 break;
1814
1815 case IPV6_MULTICAST_HOPS:
1816 if (im6o == NULL)
1817 optval = V_ip6_defmcasthlim;
1818 else
1819 optval = im6o->im6o_multicast_hlim;
1820 INP_WUNLOCK(inp);
1821 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1822 break;
1823
1824 case IPV6_MULTICAST_LOOP:
1825 if (im6o == NULL)
1826 optval = in6_mcast_loop; /* XXX VIMAGE */
1827 else
1828 optval = im6o->im6o_multicast_loop;
1829 INP_WUNLOCK(inp);
1830 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1831 break;
1832
1833 case IPV6_MSFILTER:
1834 if (im6o == NULL) {
1835 error = EADDRNOTAVAIL;
1836 INP_WUNLOCK(inp);
1837 } else {
1838 error = in6p_get_source_filters(inp, sopt);
1839 }
1840 break;
1841
1842 default:
1843 INP_WUNLOCK(inp);
1844 error = ENOPROTOOPT;
1845 break;
1846 }
1847
1848 INP_UNLOCK_ASSERT(inp);
1849
1850 return (error);
1851 }
1852
1853 /*
1854 * Look up the ifnet to use for a multicast group membership,
1855 * given the address of an IPv6 group.
1856 *
1857 * This routine exists to support legacy IPv6 multicast applications.
1858 *
1859 * Use the socket's current FIB number for any required FIB lookup. Look up the
1860 * group address in the unicast FIB, and use its ifp; usually, this points to
1861 * the default next-hop. If the FIB lookup fails, return NULL.
1862 *
1863 * FUTURE: Support multiple forwarding tables for IPv6.
1864 *
1865 * Returns NULL if no ifp could be found.
1866 */
1867 static struct ifnet *
in6p_lookup_mcast_ifp(const struct inpcb * inp,const struct sockaddr_in6 * gsin6)1868 in6p_lookup_mcast_ifp(const struct inpcb *inp, const struct sockaddr_in6 *gsin6)
1869 {
1870 struct nhop_object *nh;
1871 struct in6_addr dst;
1872 uint32_t scopeid;
1873 uint32_t fibnum;
1874
1875 KASSERT(gsin6->sin6_family == AF_INET6,
1876 ("%s: not AF_INET6 group", __func__));
1877
1878 in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1879 fibnum = inp->inp_inc.inc_fibnum;
1880 nh = fib6_lookup(fibnum, &dst, scopeid, 0, 0);
1881
1882 return (nh ? nh->nh_ifp : NULL);
1883 }
1884
1885 /*
1886 * Join an IPv6 multicast group, possibly with a source.
1887 *
1888 * FIXME: The KAME use of the unspecified address (::)
1889 * to join *all* multicast groups is currently unsupported.
1890 *
1891 * XXXGL: this function multiple times uses ifnet_byindex() without
1892 * proper protection - staying in epoch, or putting reference on ifnet.
1893 */
1894 static int
in6p_join_group(struct inpcb * inp,struct sockopt * sopt)1895 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1896 {
1897 struct in6_multi_head inmh;
1898 struct group_source_req gsr;
1899 struct epoch_tracker et;
1900 sockunion_t *gsa, *ssa;
1901 struct ifnet *ifp;
1902 struct in6_mfilter *imf;
1903 struct ip6_moptions *imo;
1904 struct in6_multi *inm;
1905 struct in6_msource *lims;
1906 int error, is_new;
1907
1908 SLIST_INIT(&inmh);
1909 ifp = NULL;
1910 lims = NULL;
1911 error = 0;
1912
1913 memset(&gsr, 0, sizeof(struct group_source_req));
1914 gsa = (sockunion_t *)&gsr.gsr_group;
1915 gsa->ss.ss_family = AF_UNSPEC;
1916 ssa = (sockunion_t *)&gsr.gsr_source;
1917 ssa->ss.ss_family = AF_UNSPEC;
1918
1919 /*
1920 * Chew everything into struct group_source_req.
1921 * Overwrite the port field if present, as the sockaddr
1922 * being copied in may be matched with a binary comparison.
1923 * Ignore passed-in scope ID.
1924 */
1925 switch (sopt->sopt_name) {
1926 case IPV6_JOIN_GROUP: {
1927 struct ipv6_mreq mreq;
1928
1929 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1930 sizeof(struct ipv6_mreq));
1931 if (error)
1932 return (error);
1933
1934 gsa->sin6.sin6_family = AF_INET6;
1935 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1936 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1937
1938 if (mreq.ipv6mr_interface == 0) {
1939 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1940 } else {
1941 NET_EPOCH_ENTER(et);
1942 ifp = ifnet_byindex(mreq.ipv6mr_interface);
1943 NET_EPOCH_EXIT(et);
1944 if (ifp == NULL)
1945 return (EADDRNOTAVAIL);
1946 }
1947 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1948 __func__, mreq.ipv6mr_interface, ifp);
1949 } break;
1950
1951 case MCAST_JOIN_GROUP:
1952 case MCAST_JOIN_SOURCE_GROUP:
1953 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1954 error = sooptcopyin(sopt, &gsr,
1955 sizeof(struct group_req),
1956 sizeof(struct group_req));
1957 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1958 error = sooptcopyin(sopt, &gsr,
1959 sizeof(struct group_source_req),
1960 sizeof(struct group_source_req));
1961 }
1962 if (error)
1963 return (error);
1964
1965 if (gsa->sin6.sin6_family != AF_INET6 ||
1966 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1967 return (EINVAL);
1968
1969 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1970 if (ssa->sin6.sin6_family != AF_INET6 ||
1971 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1972 return (EINVAL);
1973 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1974 return (EINVAL);
1975 /*
1976 * TODO: Validate embedded scope ID in source
1977 * list entry against passed-in ifp, if and only
1978 * if source list filter entry is iface or node local.
1979 */
1980 in6_clearscope(&ssa->sin6.sin6_addr);
1981 ssa->sin6.sin6_port = 0;
1982 ssa->sin6.sin6_scope_id = 0;
1983 }
1984 NET_EPOCH_ENTER(et);
1985 ifp = ifnet_byindex(gsr.gsr_interface);
1986 NET_EPOCH_EXIT(et);
1987 if (ifp == NULL)
1988 return (EADDRNOTAVAIL);
1989 break;
1990
1991 default:
1992 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1993 __func__, sopt->sopt_name);
1994 return (EOPNOTSUPP);
1995 break;
1996 }
1997
1998 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1999 return (EINVAL);
2000
2001 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2002 return (EADDRNOTAVAIL);
2003
2004 gsa->sin6.sin6_port = 0;
2005 gsa->sin6.sin6_scope_id = 0;
2006
2007 /*
2008 * Always set the scope zone ID on memberships created from userland.
2009 * Use the passed-in ifp to do this.
2010 * XXX The in6_setscope() return value is meaningless.
2011 * XXX SCOPE6_LOCK() is taken by in6_setscope().
2012 */
2013 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2014
2015 IN6_MULTI_LOCK();
2016
2017 /*
2018 * Find the membership in the membership list.
2019 */
2020 imo = in6p_findmoptions(inp);
2021 imf = im6o_match_group(imo, ifp, &gsa->sa);
2022 if (imf == NULL) {
2023 is_new = 1;
2024 inm = NULL;
2025
2026 if (ip6_mfilter_count(&imo->im6o_head) >= IPV6_MAX_MEMBERSHIPS) {
2027 error = ENOMEM;
2028 goto out_in6p_locked;
2029 }
2030 } else {
2031 is_new = 0;
2032 inm = imf->im6f_in6m;
2033
2034 if (ssa->ss.ss_family != AF_UNSPEC) {
2035 /*
2036 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2037 * is an error. On an existing inclusive membership,
2038 * it just adds the source to the filter list.
2039 */
2040 if (imf->im6f_st[1] != MCAST_INCLUDE) {
2041 error = EINVAL;
2042 goto out_in6p_locked;
2043 }
2044 /*
2045 * Throw out duplicates.
2046 *
2047 * XXX FIXME: This makes a naive assumption that
2048 * even if entries exist for *ssa in this imf,
2049 * they will be rejected as dupes, even if they
2050 * are not valid in the current mode (in-mode).
2051 *
2052 * in6_msource is transactioned just as for anything
2053 * else in SSM -- but note naive use of in6m_graft()
2054 * below for allocating new filter entries.
2055 *
2056 * This is only an issue if someone mixes the
2057 * full-state SSM API with the delta-based API,
2058 * which is discouraged in the relevant RFCs.
2059 */
2060 lims = im6o_match_source(imf, &ssa->sa);
2061 if (lims != NULL /*&&
2062 lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2063 error = EADDRNOTAVAIL;
2064 goto out_in6p_locked;
2065 }
2066 } else {
2067 /*
2068 * MCAST_JOIN_GROUP alone, on any existing membership,
2069 * is rejected, to stop the same inpcb tying up
2070 * multiple refs to the in_multi.
2071 * On an existing inclusive membership, this is also
2072 * an error; if you want to change filter mode,
2073 * you must use the userland API setsourcefilter().
2074 * XXX We don't reject this for imf in UNDEFINED
2075 * state at t1, because allocation of a filter
2076 * is atomic with allocation of a membership.
2077 */
2078 error = EADDRINUSE;
2079 goto out_in6p_locked;
2080 }
2081 }
2082
2083 /*
2084 * Begin state merge transaction at socket layer.
2085 */
2086 INP_WLOCK_ASSERT(inp);
2087
2088 /*
2089 * Graft new source into filter list for this inpcb's
2090 * membership of the group. The in6_multi may not have
2091 * been allocated yet if this is a new membership, however,
2092 * the in_mfilter slot will be allocated and must be initialized.
2093 *
2094 * Note: Grafting of exclusive mode filters doesn't happen
2095 * in this path.
2096 * XXX: Should check for non-NULL lims (node exists but may
2097 * not be in-mode) for interop with full-state API.
2098 */
2099 if (ssa->ss.ss_family != AF_UNSPEC) {
2100 /* Membership starts in IN mode */
2101 if (is_new) {
2102 CTR1(KTR_MLD, "%s: new join w/source", __func__);
2103 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2104 if (imf == NULL) {
2105 error = ENOMEM;
2106 goto out_in6p_locked;
2107 }
2108 } else {
2109 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2110 }
2111 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2112 if (lims == NULL) {
2113 CTR1(KTR_MLD, "%s: merge imf state failed",
2114 __func__);
2115 error = ENOMEM;
2116 goto out_in6p_locked;
2117 }
2118 } else {
2119 /* No address specified; Membership starts in EX mode */
2120 if (is_new) {
2121 CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2122 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2123 if (imf == NULL) {
2124 error = ENOMEM;
2125 goto out_in6p_locked;
2126 }
2127 }
2128 }
2129
2130 /*
2131 * Begin state merge transaction at MLD layer.
2132 */
2133 if (is_new) {
2134 in_pcbref(inp);
2135 INP_WUNLOCK(inp);
2136
2137 error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2138 &imf->im6f_in6m, 0);
2139
2140 INP_WLOCK(inp);
2141 if (in_pcbrele_wlocked(inp)) {
2142 error = ENXIO;
2143 goto out_in6p_unlocked;
2144 }
2145 if (error) {
2146 goto out_in6p_locked;
2147 }
2148 /*
2149 * NOTE: Refcount from in6_joingroup_locked()
2150 * is protecting membership.
2151 */
2152 ip6_mfilter_insert(&imo->im6o_head, imf);
2153 } else {
2154 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2155 IN6_MULTI_LIST_LOCK();
2156 error = in6m_merge(inm, imf);
2157 if (error) {
2158 CTR1(KTR_MLD, "%s: failed to merge inm state",
2159 __func__);
2160 IN6_MULTI_LIST_UNLOCK();
2161 im6f_rollback(imf);
2162 im6f_reap(imf);
2163 goto out_in6p_locked;
2164 }
2165 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2166 error = mld_change_state(inm, 0);
2167 IN6_MULTI_LIST_UNLOCK();
2168
2169 if (error) {
2170 CTR1(KTR_MLD, "%s: failed mld downcall",
2171 __func__);
2172 im6f_rollback(imf);
2173 im6f_reap(imf);
2174 goto out_in6p_locked;
2175 }
2176 }
2177
2178 im6f_commit(imf);
2179 imf = NULL;
2180
2181 out_in6p_locked:
2182 INP_WUNLOCK(inp);
2183 out_in6p_unlocked:
2184 IN6_MULTI_UNLOCK();
2185
2186 if (is_new && imf) {
2187 if (imf->im6f_in6m != NULL) {
2188 struct in6_multi_head inmh;
2189
2190 SLIST_INIT(&inmh);
2191 SLIST_INSERT_HEAD(&inmh, imf->im6f_in6m, in6m_defer);
2192 in6m_release_list_deferred(&inmh);
2193 }
2194 ip6_mfilter_free(imf);
2195 }
2196 return (error);
2197 }
2198
2199 /*
2200 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2201 */
2202 static int
in6p_leave_group(struct inpcb * inp,struct sockopt * sopt)2203 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2204 {
2205 struct ipv6_mreq mreq;
2206 struct group_source_req gsr;
2207 struct epoch_tracker et;
2208 sockunion_t *gsa, *ssa;
2209 struct ifnet *ifp;
2210 struct in6_mfilter *imf;
2211 struct ip6_moptions *imo;
2212 struct in6_msource *ims;
2213 struct in6_multi *inm;
2214 uint32_t ifindex;
2215 int error;
2216 bool is_final;
2217 #ifdef KTR
2218 char ip6tbuf[INET6_ADDRSTRLEN];
2219 #endif
2220
2221 ifp = NULL;
2222 ifindex = 0;
2223 error = 0;
2224 is_final = true;
2225
2226 memset(&gsr, 0, sizeof(struct group_source_req));
2227 gsa = (sockunion_t *)&gsr.gsr_group;
2228 gsa->ss.ss_family = AF_UNSPEC;
2229 ssa = (sockunion_t *)&gsr.gsr_source;
2230 ssa->ss.ss_family = AF_UNSPEC;
2231
2232 /*
2233 * Chew everything passed in up into a struct group_source_req
2234 * as that is easier to process.
2235 * Note: Any embedded scope ID in the multicast group passed
2236 * in by userland is ignored, the interface index is the recommended
2237 * mechanism to specify an interface; see below.
2238 */
2239 switch (sopt->sopt_name) {
2240 case IPV6_LEAVE_GROUP:
2241 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2242 sizeof(struct ipv6_mreq));
2243 if (error)
2244 return (error);
2245 gsa->sin6.sin6_family = AF_INET6;
2246 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2247 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2248 gsa->sin6.sin6_port = 0;
2249 gsa->sin6.sin6_scope_id = 0;
2250 ifindex = mreq.ipv6mr_interface;
2251 break;
2252
2253 case MCAST_LEAVE_GROUP:
2254 case MCAST_LEAVE_SOURCE_GROUP:
2255 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2256 error = sooptcopyin(sopt, &gsr,
2257 sizeof(struct group_req),
2258 sizeof(struct group_req));
2259 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2260 error = sooptcopyin(sopt, &gsr,
2261 sizeof(struct group_source_req),
2262 sizeof(struct group_source_req));
2263 }
2264 if (error)
2265 return (error);
2266
2267 if (gsa->sin6.sin6_family != AF_INET6 ||
2268 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2269 return (EINVAL);
2270 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2271 if (ssa->sin6.sin6_family != AF_INET6 ||
2272 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2273 return (EINVAL);
2274 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2275 return (EINVAL);
2276 /*
2277 * TODO: Validate embedded scope ID in source
2278 * list entry against passed-in ifp, if and only
2279 * if source list filter entry is iface or node local.
2280 */
2281 in6_clearscope(&ssa->sin6.sin6_addr);
2282 }
2283 gsa->sin6.sin6_port = 0;
2284 gsa->sin6.sin6_scope_id = 0;
2285 ifindex = gsr.gsr_interface;
2286 break;
2287
2288 default:
2289 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2290 __func__, sopt->sopt_name);
2291 return (EOPNOTSUPP);
2292 break;
2293 }
2294
2295 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2296 return (EINVAL);
2297
2298 /*
2299 * Validate interface index if provided. If no interface index
2300 * was provided separately, attempt to look the membership up
2301 * from the default scope as a last resort to disambiguate
2302 * the membership we are being asked to leave.
2303 * XXX SCOPE6 lock potentially taken here.
2304 */
2305 if (ifindex != 0) {
2306 NET_EPOCH_ENTER(et);
2307 ifp = ifnet_byindex(ifindex);
2308 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2309 if (ifp == NULL)
2310 return (EADDRNOTAVAIL);
2311 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2312 } else {
2313 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2314 if (error)
2315 return (EADDRNOTAVAIL);
2316 /*
2317 * Some badly behaved applications don't pass an ifindex
2318 * or a scope ID, which is an API violation. In this case,
2319 * perform a lookup as per a v6 join.
2320 *
2321 * XXX For now, stomp on zone ID for the corner case.
2322 * This is not the 'KAME way', but we need to see the ifp
2323 * directly until such time as this implementation is
2324 * refactored, assuming the scope IDs are the way to go.
2325 */
2326 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2327 if (ifindex == 0) {
2328 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2329 "ifp for group %s.", __func__,
2330 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2331 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2332 } else {
2333 NET_EPOCH_ENTER(et);
2334 ifp = ifnet_byindex(ifindex);
2335 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2336 }
2337 if (ifp == NULL)
2338 return (EADDRNOTAVAIL);
2339 }
2340
2341 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2342 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2343
2344 IN6_MULTI_LOCK();
2345
2346 /*
2347 * Find the membership in the membership list.
2348 */
2349 imo = in6p_findmoptions(inp);
2350 imf = im6o_match_group(imo, ifp, &gsa->sa);
2351 if (imf == NULL) {
2352 error = EADDRNOTAVAIL;
2353 goto out_in6p_locked;
2354 }
2355 inm = imf->im6f_in6m;
2356
2357 if (ssa->ss.ss_family != AF_UNSPEC)
2358 is_final = false;
2359
2360 /*
2361 * Begin state merge transaction at socket layer.
2362 */
2363 INP_WLOCK_ASSERT(inp);
2364
2365 /*
2366 * If we were instructed only to leave a given source, do so.
2367 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2368 */
2369 if (is_final) {
2370 ip6_mfilter_remove(&imo->im6o_head, imf);
2371 im6f_leave(imf);
2372
2373 /*
2374 * Give up the multicast address record to which
2375 * the membership points.
2376 */
2377 (void)in6_leavegroup_locked(inm, imf);
2378 } else {
2379 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2380 error = EADDRNOTAVAIL;
2381 goto out_in6p_locked;
2382 }
2383 ims = im6o_match_source(imf, &ssa->sa);
2384 if (ims == NULL) {
2385 CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2386 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2387 "not ");
2388 error = EADDRNOTAVAIL;
2389 goto out_in6p_locked;
2390 }
2391 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2392 error = im6f_prune(imf, &ssa->sin6);
2393 if (error) {
2394 CTR1(KTR_MLD, "%s: merge imf state failed",
2395 __func__);
2396 goto out_in6p_locked;
2397 }
2398 }
2399
2400 /*
2401 * Begin state merge transaction at MLD layer.
2402 */
2403 if (!is_final) {
2404 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2405 IN6_MULTI_LIST_LOCK();
2406 error = in6m_merge(inm, imf);
2407 if (error) {
2408 CTR1(KTR_MLD, "%s: failed to merge inm state",
2409 __func__);
2410 IN6_MULTI_LIST_UNLOCK();
2411 im6f_rollback(imf);
2412 im6f_reap(imf);
2413 goto out_in6p_locked;
2414 }
2415
2416 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2417 error = mld_change_state(inm, 0);
2418 IN6_MULTI_LIST_UNLOCK();
2419 if (error) {
2420 CTR1(KTR_MLD, "%s: failed mld downcall",
2421 __func__);
2422 im6f_rollback(imf);
2423 im6f_reap(imf);
2424 goto out_in6p_locked;
2425 }
2426 }
2427
2428 im6f_commit(imf);
2429 im6f_reap(imf);
2430
2431 out_in6p_locked:
2432 INP_WUNLOCK(inp);
2433
2434 if (is_final && imf)
2435 ip6_mfilter_free(imf);
2436
2437 IN6_MULTI_UNLOCK();
2438 return (error);
2439 }
2440
2441 /*
2442 * Select the interface for transmitting IPv6 multicast datagrams.
2443 *
2444 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2445 * may be passed to this socket option. An address of in6addr_any or an
2446 * interface index of 0 is used to remove a previous selection.
2447 * When no interface is selected, one is chosen for every send.
2448 */
2449 static int
in6p_set_multicast_if(struct inpcb * inp,struct sockopt * sopt)2450 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2451 {
2452 struct epoch_tracker et;
2453 struct ifnet *ifp;
2454 struct ip6_moptions *imo;
2455 u_int ifindex;
2456 int error;
2457
2458 if (sopt->sopt_valsize != sizeof(u_int))
2459 return (EINVAL);
2460
2461 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2462 if (error)
2463 return (error);
2464 NET_EPOCH_ENTER(et);
2465 if (ifindex == 0)
2466 ifp = NULL;
2467 else {
2468 ifp = ifnet_byindex(ifindex);
2469 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2470 NET_EPOCH_EXIT(et);
2471 return (EADDRNOTAVAIL);
2472 }
2473 }
2474 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2475 imo = in6p_findmoptions(inp);
2476 imo->im6o_multicast_ifp = ifp;
2477 INP_WUNLOCK(inp);
2478
2479 return (0);
2480 }
2481
2482 /*
2483 * Atomically set source filters on a socket for an IPv6 multicast group.
2484 *
2485 * XXXGL: unsafely exits epoch with ifnet pointer
2486 */
2487 static int
in6p_set_source_filters(struct inpcb * inp,struct sockopt * sopt)2488 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2489 {
2490 struct __msfilterreq msfr;
2491 struct epoch_tracker et;
2492 sockunion_t *gsa;
2493 struct ifnet *ifp;
2494 struct in6_mfilter *imf;
2495 struct ip6_moptions *imo;
2496 struct in6_multi *inm;
2497 int error;
2498
2499 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2500 sizeof(struct __msfilterreq));
2501 if (error)
2502 return (error);
2503
2504 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2505 return (ENOBUFS);
2506
2507 if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2508 msfr.msfr_fmode != MCAST_INCLUDE)
2509 return (EINVAL);
2510
2511 if (msfr.msfr_group.ss_family != AF_INET6 ||
2512 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2513 return (EINVAL);
2514
2515 gsa = (sockunion_t *)&msfr.msfr_group;
2516 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2517 return (EINVAL);
2518
2519 gsa->sin6.sin6_port = 0; /* ignore port */
2520
2521 NET_EPOCH_ENTER(et);
2522 ifp = ifnet_byindex(msfr.msfr_ifindex);
2523 NET_EPOCH_EXIT(et);
2524 if (ifp == NULL)
2525 return (EADDRNOTAVAIL);
2526 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2527
2528 /*
2529 * Take the INP write lock.
2530 * Check if this socket is a member of this group.
2531 */
2532 imo = in6p_findmoptions(inp);
2533 imf = im6o_match_group(imo, ifp, &gsa->sa);
2534 if (imf == NULL) {
2535 error = EADDRNOTAVAIL;
2536 goto out_in6p_locked;
2537 }
2538 inm = imf->im6f_in6m;
2539
2540 /*
2541 * Begin state merge transaction at socket layer.
2542 */
2543 INP_WLOCK_ASSERT(inp);
2544
2545 imf->im6f_st[1] = msfr.msfr_fmode;
2546
2547 /*
2548 * Apply any new source filters, if present.
2549 * Make a copy of the user-space source vector so
2550 * that we may copy them with a single copyin. This
2551 * allows us to deal with page faults up-front.
2552 */
2553 if (msfr.msfr_nsrcs > 0) {
2554 struct in6_msource *lims;
2555 struct sockaddr_in6 *psin;
2556 struct sockaddr_storage *kss, *pkss;
2557 int i;
2558
2559 INP_WUNLOCK(inp);
2560
2561 CTR2(KTR_MLD, "%s: loading %lu source list entries",
2562 __func__, (unsigned long)msfr.msfr_nsrcs);
2563 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2564 M_TEMP, M_WAITOK);
2565 error = copyin(msfr.msfr_srcs, kss,
2566 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2567 if (error) {
2568 free(kss, M_TEMP);
2569 return (error);
2570 }
2571
2572 INP_WLOCK(inp);
2573
2574 /*
2575 * Mark all source filters as UNDEFINED at t1.
2576 * Restore new group filter mode, as im6f_leave()
2577 * will set it to INCLUDE.
2578 */
2579 im6f_leave(imf);
2580 imf->im6f_st[1] = msfr.msfr_fmode;
2581
2582 /*
2583 * Update socket layer filters at t1, lazy-allocating
2584 * new entries. This saves a bunch of memory at the
2585 * cost of one RB_FIND() per source entry; duplicate
2586 * entries in the msfr_nsrcs vector are ignored.
2587 * If we encounter an error, rollback transaction.
2588 *
2589 * XXX This too could be replaced with a set-symmetric
2590 * difference like loop to avoid walking from root
2591 * every time, as the key space is common.
2592 */
2593 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2594 psin = (struct sockaddr_in6 *)pkss;
2595 if (psin->sin6_family != AF_INET6) {
2596 error = EAFNOSUPPORT;
2597 break;
2598 }
2599 if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2600 error = EINVAL;
2601 break;
2602 }
2603 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2604 error = EINVAL;
2605 break;
2606 }
2607 /*
2608 * TODO: Validate embedded scope ID in source
2609 * list entry against passed-in ifp, if and only
2610 * if source list filter entry is iface or node local.
2611 */
2612 in6_clearscope(&psin->sin6_addr);
2613 error = im6f_get_source(imf, psin, &lims);
2614 if (error)
2615 break;
2616 lims->im6sl_st[1] = imf->im6f_st[1];
2617 }
2618 free(kss, M_TEMP);
2619 }
2620
2621 if (error)
2622 goto out_im6f_rollback;
2623
2624 INP_WLOCK_ASSERT(inp);
2625 IN6_MULTI_LIST_LOCK();
2626
2627 /*
2628 * Begin state merge transaction at MLD layer.
2629 */
2630 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2631 error = in6m_merge(inm, imf);
2632 if (error)
2633 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2634 else {
2635 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2636 error = mld_change_state(inm, 0);
2637 if (error)
2638 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2639 }
2640
2641 IN6_MULTI_LIST_UNLOCK();
2642
2643 out_im6f_rollback:
2644 if (error)
2645 im6f_rollback(imf);
2646 else
2647 im6f_commit(imf);
2648
2649 im6f_reap(imf);
2650
2651 out_in6p_locked:
2652 INP_WUNLOCK(inp);
2653 return (error);
2654 }
2655
2656 /*
2657 * Set the IP multicast options in response to user setsockopt().
2658 *
2659 * Many of the socket options handled in this function duplicate the
2660 * functionality of socket options in the regular unicast API. However,
2661 * it is not possible to merge the duplicate code, because the idempotence
2662 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2663 * the effects of these options must be treated as separate and distinct.
2664 *
2665 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2666 */
2667 int
ip6_setmoptions(struct inpcb * inp,struct sockopt * sopt)2668 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2669 {
2670 struct ip6_moptions *im6o;
2671 int error;
2672
2673 error = 0;
2674
2675 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
2676 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2677 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)
2678 return (EOPNOTSUPP);
2679
2680 switch (sopt->sopt_name) {
2681 case IPV6_MULTICAST_IF:
2682 error = in6p_set_multicast_if(inp, sopt);
2683 break;
2684
2685 case IPV6_MULTICAST_HOPS: {
2686 int hlim;
2687
2688 if (sopt->sopt_valsize != sizeof(int)) {
2689 error = EINVAL;
2690 break;
2691 }
2692 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2693 if (error)
2694 break;
2695 if (hlim < -1 || hlim > 255) {
2696 error = EINVAL;
2697 break;
2698 } else if (hlim == -1) {
2699 hlim = V_ip6_defmcasthlim;
2700 }
2701 im6o = in6p_findmoptions(inp);
2702 im6o->im6o_multicast_hlim = hlim;
2703 INP_WUNLOCK(inp);
2704 break;
2705 }
2706
2707 case IPV6_MULTICAST_LOOP: {
2708 u_int loop;
2709
2710 /*
2711 * Set the loopback flag for outgoing multicast packets.
2712 * Must be zero or one.
2713 */
2714 if (sopt->sopt_valsize != sizeof(u_int)) {
2715 error = EINVAL;
2716 break;
2717 }
2718 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2719 if (error)
2720 break;
2721 if (loop > 1) {
2722 error = EINVAL;
2723 break;
2724 }
2725 im6o = in6p_findmoptions(inp);
2726 im6o->im6o_multicast_loop = loop;
2727 INP_WUNLOCK(inp);
2728 break;
2729 }
2730
2731 case IPV6_JOIN_GROUP:
2732 case MCAST_JOIN_GROUP:
2733 case MCAST_JOIN_SOURCE_GROUP:
2734 error = in6p_join_group(inp, sopt);
2735 break;
2736
2737 case IPV6_LEAVE_GROUP:
2738 case MCAST_LEAVE_GROUP:
2739 case MCAST_LEAVE_SOURCE_GROUP:
2740 error = in6p_leave_group(inp, sopt);
2741 break;
2742
2743 case MCAST_BLOCK_SOURCE:
2744 case MCAST_UNBLOCK_SOURCE:
2745 error = in6p_block_unblock_source(inp, sopt);
2746 break;
2747
2748 case IPV6_MSFILTER:
2749 error = in6p_set_source_filters(inp, sopt);
2750 break;
2751
2752 default:
2753 error = EOPNOTSUPP;
2754 break;
2755 }
2756
2757 INP_UNLOCK_ASSERT(inp);
2758
2759 return (error);
2760 }
2761
2762 /*
2763 * Expose MLD's multicast filter mode and source list(s) to userland,
2764 * keyed by (ifindex, group).
2765 * The filter mode is written out as a uint32_t, followed by
2766 * 0..n of struct in6_addr.
2767 * For use by ifmcstat(8).
2768 * SMPng: NOTE: unlocked read of ifindex space.
2769 */
2770 static int
sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)2771 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2772 {
2773 struct in6_addr mcaddr;
2774 struct in6_addr src;
2775 struct epoch_tracker et;
2776 struct ifnet *ifp;
2777 struct ifmultiaddr *ifma;
2778 struct in6_multi *inm;
2779 struct ip6_msource *ims;
2780 int *name;
2781 int retval;
2782 u_int namelen;
2783 uint32_t fmode, ifindex;
2784 #ifdef KTR
2785 char ip6tbuf[INET6_ADDRSTRLEN];
2786 #endif
2787
2788 name = (int *)arg1;
2789 namelen = arg2;
2790
2791 if (req->newptr != NULL)
2792 return (EPERM);
2793
2794 /* int: ifindex + 4 * 32 bits of IPv6 address */
2795 if (namelen != 5)
2796 return (EINVAL);
2797
2798 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2799 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2800 CTR2(KTR_MLD, "%s: group %s is not multicast",
2801 __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2802 return (EINVAL);
2803 }
2804
2805 ifindex = name[0];
2806 NET_EPOCH_ENTER(et);
2807 ifp = ifnet_byindex(ifindex);
2808 if (ifp == NULL) {
2809 NET_EPOCH_EXIT(et);
2810 CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2811 __func__, ifindex);
2812 return (ENOENT);
2813 }
2814 /*
2815 * Internal MLD lookups require that scope/zone ID is set.
2816 */
2817 (void)in6_setscope(&mcaddr, ifp, NULL);
2818
2819 retval = sysctl_wire_old_buffer(req,
2820 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2821 if (retval) {
2822 NET_EPOCH_EXIT(et);
2823 return (retval);
2824 }
2825
2826 IN6_MULTI_LOCK();
2827 IN6_MULTI_LIST_LOCK();
2828 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2829 inm = in6m_ifmultiaddr_get_inm(ifma);
2830 if (inm == NULL)
2831 continue;
2832 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2833 continue;
2834 fmode = inm->in6m_st[1].iss_fmode;
2835 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2836 if (retval != 0)
2837 break;
2838 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2839 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2840 /*
2841 * Only copy-out sources which are in-mode.
2842 */
2843 if (fmode != im6s_get_mode(inm, ims, 1)) {
2844 CTR1(KTR_MLD, "%s: skip non-in-mode",
2845 __func__);
2846 continue;
2847 }
2848 src = ims->im6s_addr;
2849 retval = SYSCTL_OUT(req, &src,
2850 sizeof(struct in6_addr));
2851 if (retval != 0)
2852 break;
2853 }
2854 }
2855 IN6_MULTI_LIST_UNLOCK();
2856 IN6_MULTI_UNLOCK();
2857 NET_EPOCH_EXIT(et);
2858
2859 return (retval);
2860 }
2861
2862 #ifdef KTR
2863
2864 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2865
2866 static const char *
in6m_mode_str(const int mode)2867 in6m_mode_str(const int mode)
2868 {
2869
2870 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2871 return (in6m_modestrs[mode]);
2872 return ("??");
2873 }
2874
2875 static const char *in6m_statestrs[] = {
2876 "not-member",
2877 "silent",
2878 "reporting",
2879 "idle",
2880 "lazy",
2881 "sleeping",
2882 "awakening",
2883 "query-pending",
2884 "sg-query-pending",
2885 "leaving"
2886 };
2887 _Static_assert(nitems(in6m_statestrs) ==
2888 MLD_LEAVING_MEMBER - MLD_NOT_MEMBER + 1, "Missing MLD group state");
2889
2890 static const char *
in6m_state_str(const int state)2891 in6m_state_str(const int state)
2892 {
2893
2894 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2895 return (in6m_statestrs[state]);
2896 return ("??");
2897 }
2898
2899 /*
2900 * Dump an in6_multi structure to the console.
2901 */
2902 void
in6m_print(const struct in6_multi * inm)2903 in6m_print(const struct in6_multi *inm)
2904 {
2905 int t;
2906 char ip6tbuf[INET6_ADDRSTRLEN];
2907
2908 if ((ktr_mask & KTR_MLD) == 0)
2909 return;
2910
2911 printf("%s: --- begin in6m %p ---\n", __func__, inm);
2912 printf("addr %s ifp %p(%s) ifma %p\n",
2913 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2914 inm->in6m_ifp,
2915 if_name(inm->in6m_ifp),
2916 inm->in6m_ifma);
2917 printf("timer %u state %s refcount %u scq.len %u\n",
2918 inm->in6m_timer,
2919 in6m_state_str(inm->in6m_state),
2920 inm->in6m_refcount,
2921 mbufq_len(&inm->in6m_scq));
2922 printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2923 inm->in6m_mli,
2924 inm->in6m_nsrc,
2925 inm->in6m_sctimer,
2926 inm->in6m_scrv);
2927 for (t = 0; t < 2; t++) {
2928 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2929 in6m_mode_str(inm->in6m_st[t].iss_fmode),
2930 inm->in6m_st[t].iss_asm,
2931 inm->in6m_st[t].iss_ex,
2932 inm->in6m_st[t].iss_in,
2933 inm->in6m_st[t].iss_rec);
2934 }
2935 printf("%s: --- end in6m %p ---\n", __func__, inm);
2936 }
2937
2938 #else /* !KTR */
2939
2940 void
in6m_print(const struct in6_multi * inm)2941 in6m_print(const struct in6_multi *inm)
2942 {
2943
2944 }
2945
2946 #endif /* KTR */
2947