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 "opt_inet6.h"
38
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/kernel.h>
42 #include <sys/ktr.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/protosw.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/sysctl.h>
49 #include <sys/priv.h>
50 #include <sys/taskqueue.h>
51 #include <sys/tree.h>
52
53 #include <net/if.h>
54 #include <net/if_var.h>
55 #include <net/if_dl.h>
56 #include <net/if_private.h>
57 #include <net/route.h>
58 #include <net/route/nhop.h>
59 #include <net/vnet.h>
60
61 #include <netinet/in.h>
62 #include <netinet/udp.h>
63 #include <netinet/in_var.h>
64 #include <netinet/ip_var.h>
65 #include <netinet/udp_var.h>
66 #include <netinet6/in6_fib.h>
67 #include <netinet6/in6_var.h>
68 #include <netinet/ip6.h>
69 #include <netinet/icmp6.h>
70 #include <netinet6/ip6_var.h>
71 #include <netinet/in_pcb.h>
72 #include <netinet/tcp_var.h>
73 #include <netinet6/nd6.h>
74 #include <netinet6/mld6_var.h>
75 #include <netinet6/scope6_var.h>
76
77 #ifndef KTR_MLD
78 #define KTR_MLD KTR_INET6
79 #endif
80
81 #ifndef __SOCKUNION_DECLARED
82 union sockunion {
83 struct sockaddr_storage ss;
84 struct sockaddr sa;
85 struct sockaddr_dl sdl;
86 struct sockaddr_in6 sin6;
87 };
88 typedef union sockunion sockunion_t;
89 #define __SOCKUNION_DECLARED
90 #endif /* __SOCKUNION_DECLARED */
91
92 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
93 "IPv6 multicast PCB-layer source filter");
94 MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
95 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
96 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
97 "IPv6 multicast MLD-layer source filter");
98
99 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
100
101 /*
102 * Locking:
103 * - Lock order is: IN6_MULTI_LOCK, INP_WLOCK, IN6_MULTI_LIST_LOCK, MLD_LOCK,
104 * IF_ADDR_LOCK.
105 * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
106 * it can be taken by code in net/if.c also.
107 * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
108 *
109 * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
110 * any need for in6_multi itself to be virtualized -- it is bound to an ifp
111 * anyway no matter what happens.
112 */
113 struct mtx in6_multi_list_mtx;
114 MTX_SYSINIT(in6_multi_mtx, &in6_multi_list_mtx, "in6_multi_list_mtx", MTX_DEF);
115
116 struct mtx in6_multi_free_mtx;
117 MTX_SYSINIT(in6_multi_free_mtx, &in6_multi_free_mtx, "in6_multi_free_mtx", MTX_DEF);
118
119 struct sx in6_multi_sx;
120 SX_SYSINIT(in6_multi_sx, &in6_multi_sx, "in6_multi_sx");
121
122 static void im6f_commit(struct in6_mfilter *);
123 static int im6f_get_source(struct in6_mfilter *imf,
124 const struct sockaddr_in6 *psin,
125 struct in6_msource **);
126 static struct in6_msource *
127 im6f_graft(struct in6_mfilter *, const uint8_t,
128 const struct sockaddr_in6 *);
129 static void im6f_leave(struct in6_mfilter *);
130 static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
131 static void im6f_purge(struct in6_mfilter *);
132 static void im6f_rollback(struct in6_mfilter *);
133 static void im6f_reap(struct in6_mfilter *);
134 static struct in6_mfilter *
135 im6o_match_group(const struct ip6_moptions *,
136 const struct ifnet *, const struct sockaddr *);
137 static struct in6_msource *
138 im6o_match_source(struct in6_mfilter *, const struct sockaddr *);
139 static void im6s_merge(struct ip6_msource *ims,
140 const struct in6_msource *lims, const int rollback);
141 static int in6_getmulti(struct ifnet *, const struct in6_addr *,
142 struct in6_multi **);
143 static int in6_joingroup_locked(struct ifnet *, const struct in6_addr *,
144 struct in6_mfilter *, struct in6_multi **, int);
145 static int in6m_get_source(struct in6_multi *inm,
146 const struct in6_addr *addr, const int noalloc,
147 struct ip6_msource **pims);
148 #ifdef KTR
149 static int in6m_is_ifp_detached(const struct in6_multi *);
150 #endif
151 static int in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
152 static void in6m_purge(struct in6_multi *);
153 static void in6m_reap(struct in6_multi *);
154 static struct ip6_moptions *
155 in6p_findmoptions(struct inpcb *);
156 static int in6p_get_source_filters(struct inpcb *, struct sockopt *);
157 static int in6p_join_group(struct inpcb *, struct sockopt *);
158 static int in6p_leave_group(struct inpcb *, struct sockopt *);
159 static struct ifnet *
160 in6p_lookup_mcast_ifp(const struct inpcb *,
161 const struct sockaddr_in6 *);
162 static int in6p_block_unblock_source(struct inpcb *, struct sockopt *);
163 static int in6p_set_multicast_if(struct inpcb *, struct sockopt *);
164 static int in6p_set_source_filters(struct inpcb *, struct sockopt *);
165 static int sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
166
167 SYSCTL_DECL(_net_inet6_ip6); /* XXX Not in any common header. */
168
169 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast,
170 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
171 "IPv6 multicast");
172
173 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
174 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
175 CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
176 "Max source filters per group");
177
178 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
179 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
180 CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
181 "Max source filters per socket");
182
183 /* TODO Virtualize this switch. */
184 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
185 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
186 &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
187
188 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
189 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
190 "Per-interface stack-wide source filters");
191
192 #ifdef KTR
193 /*
194 * Inline function which wraps assertions for a valid ifp.
195 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
196 * is detached.
197 */
198 static int __inline
in6m_is_ifp_detached(const struct in6_multi * inm)199 in6m_is_ifp_detached(const struct in6_multi *inm)
200 {
201 struct ifnet *ifp;
202
203 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
204 ifp = inm->in6m_ifma->ifma_ifp;
205 if (ifp != NULL) {
206 /*
207 * Sanity check that network-layer notion of ifp is the
208 * same as that of link-layer.
209 */
210 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
211 }
212
213 return (ifp == NULL);
214 }
215 #endif
216
217 /*
218 * Initialize an in6_mfilter structure to a known state at t0, t1
219 * with an empty source filter list.
220 */
221 static __inline void
im6f_init(struct in6_mfilter * imf,const int st0,const int st1)222 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
223 {
224 memset(imf, 0, sizeof(struct in6_mfilter));
225 RB_INIT(&imf->im6f_sources);
226 imf->im6f_st[0] = st0;
227 imf->im6f_st[1] = st1;
228 }
229
230 struct in6_mfilter *
ip6_mfilter_alloc(const int mflags,const int st0,const int st1)231 ip6_mfilter_alloc(const int mflags, const int st0, const int st1)
232 {
233 struct in6_mfilter *imf;
234
235 imf = malloc(sizeof(*imf), M_IN6MFILTER, mflags);
236
237 if (imf != NULL)
238 im6f_init(imf, st0, st1);
239
240 return (imf);
241 }
242
243 void
ip6_mfilter_free(struct in6_mfilter * imf)244 ip6_mfilter_free(struct in6_mfilter *imf)
245 {
246
247 im6f_purge(imf);
248 free(imf, M_IN6MFILTER);
249 }
250
251 /*
252 * Find an IPv6 multicast group entry for this ip6_moptions instance
253 * which matches the specified group, and optionally an interface.
254 * Return its index into the array, or -1 if not found.
255 */
256 static struct in6_mfilter *
im6o_match_group(const struct ip6_moptions * imo,const struct ifnet * ifp,const struct sockaddr * group)257 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
258 const struct sockaddr *group)
259 {
260 const struct sockaddr_in6 *gsin6;
261 struct in6_mfilter *imf;
262 struct in6_multi *inm;
263
264 gsin6 = (const struct sockaddr_in6 *)group;
265
266 IP6_MFILTER_FOREACH(imf, &imo->im6o_head) {
267 inm = imf->im6f_in6m;
268 if (inm == NULL)
269 continue;
270 if ((ifp == NULL || (inm->in6m_ifp == ifp)) &&
271 IN6_ARE_ADDR_EQUAL(&inm->in6m_addr,
272 &gsin6->sin6_addr)) {
273 break;
274 }
275 }
276 return (imf);
277 }
278
279 /*
280 * Find an IPv6 multicast source entry for this imo which matches
281 * the given group index for this socket, and source address.
282 *
283 * XXX TODO: The scope ID, if present in src, is stripped before
284 * any comparison. We SHOULD enforce scope/zone checks where the source
285 * filter entry has a link scope.
286 *
287 * NOTE: This does not check if the entry is in-mode, merely if
288 * it exists, which may not be the desired behaviour.
289 */
290 static struct in6_msource *
im6o_match_source(struct in6_mfilter * imf,const struct sockaddr * src)291 im6o_match_source(struct in6_mfilter *imf, const struct sockaddr *src)
292 {
293 struct ip6_msource find;
294 struct ip6_msource *ims;
295 const sockunion_t *psa;
296
297 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
298
299 psa = (const sockunion_t *)src;
300 find.im6s_addr = psa->sin6.sin6_addr;
301 in6_clearscope(&find.im6s_addr); /* XXX */
302 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
303
304 return ((struct in6_msource *)ims);
305 }
306
307 /*
308 * Perform filtering for multicast datagrams on a socket by group and source.
309 *
310 * Returns 0 if a datagram should be allowed through, or various error codes
311 * if the socket was not a member of the group, or the source was muted, etc.
312 */
313 int
im6o_mc_filter(const struct ip6_moptions * imo,const struct ifnet * ifp,const struct sockaddr * group,const struct sockaddr * src)314 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
315 const struct sockaddr *group, const struct sockaddr *src)
316 {
317 struct in6_mfilter *imf;
318 struct in6_msource *ims;
319 int mode;
320
321 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
322
323 imf = im6o_match_group(imo, ifp, group);
324 if (imf == NULL)
325 return (MCAST_NOTGMEMBER);
326
327 /*
328 * Check if the source was included in an (S,G) join.
329 * Allow reception on exclusive memberships by default,
330 * reject reception on inclusive memberships by default.
331 * Exclude source only if an in-mode exclude filter exists.
332 * Include source only if an in-mode include filter exists.
333 * NOTE: We are comparing group state here at MLD t1 (now)
334 * with socket-layer t0 (since last downcall).
335 */
336 mode = imf->im6f_st[1];
337 ims = im6o_match_source(imf, src);
338
339 if ((ims == NULL && mode == MCAST_INCLUDE) ||
340 (ims != NULL && ims->im6sl_st[0] != mode))
341 return (MCAST_NOTSMEMBER);
342
343 return (MCAST_PASS);
344 }
345
346 /*
347 * Look up an in6_multi record for an IPv6 multicast address
348 * on the interface ifp.
349 * If no record found, return NULL.
350 *
351 * SMPng: The IN6_MULTI_LOCK and must be held and must be in network epoch.
352 */
353 struct in6_multi *
in6m_lookup_locked(struct ifnet * ifp,const struct in6_addr * mcaddr)354 in6m_lookup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr)
355 {
356 struct ifmultiaddr *ifma;
357 struct in6_multi *inm;
358
359 NET_EPOCH_ASSERT();
360
361 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
362 inm = in6m_ifmultiaddr_get_inm(ifma);
363 if (inm == NULL)
364 continue;
365 if (IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, mcaddr))
366 return (inm);
367 }
368 return (NULL);
369 }
370
371 /*
372 * Find and return a reference to an in6_multi record for (ifp, group),
373 * and bump its reference count.
374 * If one does not exist, try to allocate it, and update link-layer multicast
375 * filters on ifp to listen for group.
376 * Assumes the IN6_MULTI lock is held across the call.
377 * Return 0 if successful, otherwise return an appropriate error code.
378 */
379 static int
in6_getmulti(struct ifnet * ifp,const struct in6_addr * group,struct in6_multi ** pinm)380 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
381 struct in6_multi **pinm)
382 {
383 struct epoch_tracker et;
384 struct sockaddr_in6 gsin6;
385 struct ifmultiaddr *ifma;
386 struct in6_multi *inm;
387 int error;
388
389 error = 0;
390
391 /*
392 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
393 * if_addmulti() takes this mutex itself, so we must drop and
394 * re-acquire around the call.
395 */
396 IN6_MULTI_LOCK_ASSERT();
397 IN6_MULTI_LIST_LOCK();
398 IF_ADDR_WLOCK(ifp);
399 NET_EPOCH_ENTER(et);
400 /*
401 * Does ifp support IPv6 multicasts?
402 */
403 if (ifp->if_afdata[AF_INET6] == NULL)
404 error = ENODEV;
405 else
406 inm = in6m_lookup_locked(ifp, group);
407 NET_EPOCH_EXIT(et);
408
409 if (error != 0)
410 goto out_locked;
411
412 if (inm != NULL) {
413 /*
414 * If we already joined this group, just bump the
415 * refcount and return it.
416 */
417 KASSERT(inm->in6m_refcount >= 1,
418 ("%s: bad refcount %d", __func__, inm->in6m_refcount));
419 in6m_acquire_locked(inm);
420 *pinm = inm;
421 goto out_locked;
422 }
423
424 memset(&gsin6, 0, sizeof(gsin6));
425 gsin6.sin6_family = AF_INET6;
426 gsin6.sin6_len = sizeof(struct sockaddr_in6);
427 gsin6.sin6_addr = *group;
428
429 /*
430 * Check if a link-layer group is already associated
431 * with this network-layer group on the given ifnet.
432 */
433 IN6_MULTI_LIST_UNLOCK();
434 IF_ADDR_WUNLOCK(ifp);
435 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
436 if (error != 0)
437 return (error);
438 IN6_MULTI_LIST_LOCK();
439 IF_ADDR_WLOCK(ifp);
440
441 /*
442 * If something other than netinet6 is occupying the link-layer
443 * group, print a meaningful error message and back out of
444 * the allocation.
445 * Otherwise, bump the refcount on the existing network-layer
446 * group association and return it.
447 */
448 if (ifma->ifma_protospec != NULL) {
449 inm = (struct in6_multi *)ifma->ifma_protospec;
450 #ifdef INVARIANTS
451 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
452 __func__));
453 KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
454 ("%s: ifma not AF_INET6", __func__));
455 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
456 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
457 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
458 panic("%s: ifma %p is inconsistent with %p (%p)",
459 __func__, ifma, inm, group);
460 #endif
461 in6m_acquire_locked(inm);
462 *pinm = inm;
463 goto out_locked;
464 }
465
466 IF_ADDR_WLOCK_ASSERT(ifp);
467
468 /*
469 * A new in6_multi record is needed; allocate and initialize it.
470 * We DO NOT perform an MLD join as the in6_ layer may need to
471 * push an initial source list down to MLD to support SSM.
472 *
473 * The initial source filter state is INCLUDE, {} as per the RFC.
474 * Pending state-changes per group are subject to a bounds check.
475 */
476 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
477 if (inm == NULL) {
478 IN6_MULTI_LIST_UNLOCK();
479 IF_ADDR_WUNLOCK(ifp);
480 if_delmulti_ifma(ifma);
481 return (ENOMEM);
482 }
483 inm->in6m_addr = *group;
484 inm->in6m_ifp = ifp;
485 inm->in6m_mli = MLD_IFINFO(ifp);
486 inm->in6m_ifma = ifma;
487 inm->in6m_refcount = 1;
488 inm->in6m_state = MLD_NOT_MEMBER;
489 mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
490
491 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
492 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
493 RB_INIT(&inm->in6m_srcs);
494
495 ifma->ifma_protospec = inm;
496 *pinm = inm;
497
498 out_locked:
499 IN6_MULTI_LIST_UNLOCK();
500 IF_ADDR_WUNLOCK(ifp);
501 return (error);
502 }
503
504 /*
505 * Drop a reference to an in6_multi record.
506 *
507 * If the refcount drops to 0, free the in6_multi record and
508 * delete the underlying link-layer membership.
509 */
510 static void
in6m_release(struct in6_multi * inm)511 in6m_release(struct in6_multi *inm)
512 {
513 struct ifmultiaddr *ifma;
514 struct ifnet *ifp;
515
516 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
517
518 MPASS(inm->in6m_refcount == 0);
519 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
520
521 ifma = inm->in6m_ifma;
522 ifp = inm->in6m_ifp;
523 MPASS(ifma->ifma_llifma == NULL);
524
525 /* XXX this access is not covered by IF_ADDR_LOCK */
526 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
527 KASSERT(ifma->ifma_protospec == NULL,
528 ("%s: ifma_protospec != NULL", __func__));
529 if (ifp == NULL)
530 ifp = ifma->ifma_ifp;
531
532 if (ifp != NULL) {
533 CURVNET_SET(ifp->if_vnet);
534 in6m_purge(inm);
535 free(inm, M_IP6MADDR);
536 if_delmulti_ifma_flags(ifma, 1);
537 CURVNET_RESTORE();
538 if_rele(ifp);
539 } else {
540 in6m_purge(inm);
541 free(inm, M_IP6MADDR);
542 if_delmulti_ifma_flags(ifma, 1);
543 }
544 }
545
546 /*
547 * Interface detach can happen in a taskqueue thread context, so we must use a
548 * dedicated thread to avoid deadlocks when draining in6m_release tasks.
549 */
550 TASKQUEUE_DEFINE_THREAD(in6m_free);
551 static struct in6_multi_head in6m_free_list = SLIST_HEAD_INITIALIZER();
552 static void in6m_release_task(void *arg __unused, int pending __unused);
553 static struct task in6m_free_task = TASK_INITIALIZER(0, in6m_release_task, NULL);
554
555 void
in6m_release_list_deferred(struct in6_multi_head * inmh)556 in6m_release_list_deferred(struct in6_multi_head *inmh)
557 {
558 if (SLIST_EMPTY(inmh))
559 return;
560 mtx_lock(&in6_multi_free_mtx);
561 SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
562 mtx_unlock(&in6_multi_free_mtx);
563 taskqueue_enqueue(taskqueue_in6m_free, &in6m_free_task);
564 }
565
566 void
in6m_release_wait(void * arg __unused)567 in6m_release_wait(void *arg __unused)
568 {
569
570 /*
571 * Make sure all pending multicast addresses are freed before
572 * the VNET or network device is destroyed:
573 */
574 taskqueue_drain_all(taskqueue_in6m_free);
575 }
576 #ifdef VIMAGE
577 /* XXX-BZ FIXME, see D24914. */
578 VNET_SYSUNINIT(in6m_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, in6m_release_wait, NULL);
579 #endif
580
581 void
in6m_disconnect_locked(struct in6_multi_head * inmh,struct in6_multi * inm)582 in6m_disconnect_locked(struct in6_multi_head *inmh, struct in6_multi *inm)
583 {
584 struct ifnet *ifp;
585 struct ifaddr *ifa;
586 struct in6_ifaddr *ifa6;
587 struct in6_multi_mship *imm, *imm_tmp;
588 struct ifmultiaddr *ifma, *ll_ifma;
589
590 IN6_MULTI_LIST_LOCK_ASSERT();
591
592 ifp = inm->in6m_ifp;
593 if (ifp == NULL)
594 return; /* already called */
595
596 inm->in6m_ifp = NULL;
597 IF_ADDR_WLOCK_ASSERT(ifp);
598 ifma = inm->in6m_ifma;
599 if (ifma == NULL)
600 return;
601
602 if_ref(ifp);
603 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
604 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
605 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
606 }
607 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
608 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
609 MPASS(ifma != ll_ifma);
610 ifma->ifma_llifma = NULL;
611 MPASS(ll_ifma->ifma_llifma == NULL);
612 MPASS(ll_ifma->ifma_ifp == ifp);
613 if (--ll_ifma->ifma_refcount == 0) {
614 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
615 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
616 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
617 }
618 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
619 if_freemulti(ll_ifma);
620 }
621 }
622 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
623 if (ifa->ifa_addr->sa_family != AF_INET6)
624 continue;
625 ifa6 = (void *)ifa;
626 LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
627 i6mm_chain, imm_tmp) {
628 if (inm == imm->i6mm_maddr) {
629 LIST_REMOVE(imm, i6mm_chain);
630 free(imm, M_IP6MADDR);
631 in6m_rele_locked(inmh, inm);
632 }
633 }
634 }
635 }
636
637 static void
in6m_release_task(void * arg __unused,int pending __unused)638 in6m_release_task(void *arg __unused, int pending __unused)
639 {
640 struct in6_multi_head in6m_free_tmp;
641 struct in6_multi *inm, *tinm;
642
643 SLIST_INIT(&in6m_free_tmp);
644 mtx_lock(&in6_multi_free_mtx);
645 SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
646 mtx_unlock(&in6_multi_free_mtx);
647 IN6_MULTI_LOCK();
648 SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
649 SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
650 in6m_release(inm);
651 }
652 IN6_MULTI_UNLOCK();
653 }
654
655 /*
656 * Clear recorded source entries for a group.
657 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
658 * FIXME: Should reap.
659 */
660 void
in6m_clear_recorded(struct in6_multi * inm)661 in6m_clear_recorded(struct in6_multi *inm)
662 {
663 struct ip6_msource *ims;
664
665 IN6_MULTI_LIST_LOCK_ASSERT();
666
667 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
668 if (ims->im6s_stp) {
669 ims->im6s_stp = 0;
670 --inm->in6m_st[1].iss_rec;
671 }
672 }
673 KASSERT(inm->in6m_st[1].iss_rec == 0,
674 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
675 }
676
677 /*
678 * Record a source as pending for a Source-Group MLDv2 query.
679 * This lives here as it modifies the shared tree.
680 *
681 * inm is the group descriptor.
682 * naddr is the address of the source to record in network-byte order.
683 *
684 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
685 * lazy-allocate a source node in response to an SG query.
686 * Otherwise, no allocation is performed. This saves some memory
687 * with the trade-off that the source will not be reported to the
688 * router if joined in the window between the query response and
689 * the group actually being joined on the local host.
690 *
691 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
692 * This turns off the allocation of a recorded source entry if
693 * the group has not been joined.
694 *
695 * Return 0 if the source didn't exist or was already marked as recorded.
696 * Return 1 if the source was marked as recorded by this function.
697 * Return <0 if any error occurred (negated errno code).
698 */
699 int
in6m_record_source(struct in6_multi * inm,const struct in6_addr * addr)700 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
701 {
702 struct ip6_msource find;
703 struct ip6_msource *ims, *nims;
704
705 IN6_MULTI_LIST_LOCK_ASSERT();
706
707 find.im6s_addr = *addr;
708 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
709 if (ims && ims->im6s_stp)
710 return (0);
711 if (ims == NULL) {
712 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
713 return (-ENOSPC);
714 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
715 M_NOWAIT | M_ZERO);
716 if (nims == NULL)
717 return (-ENOMEM);
718 nims->im6s_addr = find.im6s_addr;
719 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
720 ++inm->in6m_nsrc;
721 ims = nims;
722 }
723
724 /*
725 * Mark the source as recorded and update the recorded
726 * source count.
727 */
728 ++ims->im6s_stp;
729 ++inm->in6m_st[1].iss_rec;
730
731 return (1);
732 }
733
734 /*
735 * Return a pointer to an in6_msource owned by an in6_mfilter,
736 * given its source address.
737 * Lazy-allocate if needed. If this is a new entry its filter state is
738 * undefined at t0.
739 *
740 * imf is the filter set being modified.
741 * addr is the source address.
742 *
743 * SMPng: May be called with locks held; malloc must not block.
744 */
745 static int
im6f_get_source(struct in6_mfilter * imf,const struct sockaddr_in6 * psin,struct in6_msource ** plims)746 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
747 struct in6_msource **plims)
748 {
749 struct ip6_msource find;
750 struct ip6_msource *ims, *nims;
751 struct in6_msource *lims;
752 int error;
753
754 error = 0;
755 ims = NULL;
756 lims = NULL;
757
758 find.im6s_addr = psin->sin6_addr;
759 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
760 lims = (struct in6_msource *)ims;
761 if (lims == NULL) {
762 if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
763 return (ENOSPC);
764 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
765 M_NOWAIT | M_ZERO);
766 if (nims == NULL)
767 return (ENOMEM);
768 lims = (struct in6_msource *)nims;
769 lims->im6s_addr = find.im6s_addr;
770 lims->im6sl_st[0] = MCAST_UNDEFINED;
771 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
772 ++imf->im6f_nsrc;
773 }
774
775 *plims = lims;
776
777 return (error);
778 }
779
780 /*
781 * Graft a source entry into an existing socket-layer filter set,
782 * maintaining any required invariants and checking allocations.
783 *
784 * The source is marked as being in the new filter mode at t1.
785 *
786 * Return the pointer to the new node, otherwise return NULL.
787 */
788 static struct in6_msource *
im6f_graft(struct in6_mfilter * imf,const uint8_t st1,const struct sockaddr_in6 * psin)789 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
790 const struct sockaddr_in6 *psin)
791 {
792 struct ip6_msource *nims;
793 struct in6_msource *lims;
794
795 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
796 M_NOWAIT | M_ZERO);
797 if (nims == NULL)
798 return (NULL);
799 lims = (struct in6_msource *)nims;
800 lims->im6s_addr = psin->sin6_addr;
801 lims->im6sl_st[0] = MCAST_UNDEFINED;
802 lims->im6sl_st[1] = st1;
803 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
804 ++imf->im6f_nsrc;
805
806 return (lims);
807 }
808
809 /*
810 * Prune a source entry from an existing socket-layer filter set,
811 * maintaining any required invariants and checking allocations.
812 *
813 * The source is marked as being left at t1, it is not freed.
814 *
815 * Return 0 if no error occurred, otherwise return an errno value.
816 */
817 static int
im6f_prune(struct in6_mfilter * imf,const struct sockaddr_in6 * psin)818 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
819 {
820 struct ip6_msource find;
821 struct ip6_msource *ims;
822 struct in6_msource *lims;
823
824 find.im6s_addr = psin->sin6_addr;
825 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
826 if (ims == NULL)
827 return (ENOENT);
828 lims = (struct in6_msource *)ims;
829 lims->im6sl_st[1] = MCAST_UNDEFINED;
830 return (0);
831 }
832
833 /*
834 * Revert socket-layer filter set deltas at t1 to t0 state.
835 */
836 static void
im6f_rollback(struct in6_mfilter * imf)837 im6f_rollback(struct in6_mfilter *imf)
838 {
839 struct ip6_msource *ims, *tims;
840 struct in6_msource *lims;
841
842 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
843 lims = (struct in6_msource *)ims;
844 if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
845 /* no change at t1 */
846 continue;
847 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
848 /* revert change to existing source at t1 */
849 lims->im6sl_st[1] = lims->im6sl_st[0];
850 } else {
851 /* revert source added t1 */
852 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
853 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
854 free(ims, M_IN6MFILTER);
855 imf->im6f_nsrc--;
856 }
857 }
858 imf->im6f_st[1] = imf->im6f_st[0];
859 }
860
861 /*
862 * Mark socket-layer filter set as INCLUDE {} at t1.
863 */
864 static void
im6f_leave(struct in6_mfilter * imf)865 im6f_leave(struct in6_mfilter *imf)
866 {
867 struct ip6_msource *ims;
868 struct in6_msource *lims;
869
870 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
871 lims = (struct in6_msource *)ims;
872 lims->im6sl_st[1] = MCAST_UNDEFINED;
873 }
874 imf->im6f_st[1] = MCAST_INCLUDE;
875 }
876
877 /*
878 * Mark socket-layer filter set deltas as committed.
879 */
880 static void
im6f_commit(struct in6_mfilter * imf)881 im6f_commit(struct in6_mfilter *imf)
882 {
883 struct ip6_msource *ims;
884 struct in6_msource *lims;
885
886 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
887 lims = (struct in6_msource *)ims;
888 lims->im6sl_st[0] = lims->im6sl_st[1];
889 }
890 imf->im6f_st[0] = imf->im6f_st[1];
891 }
892
893 /*
894 * Reap unreferenced sources from socket-layer filter set.
895 */
896 static void
im6f_reap(struct in6_mfilter * imf)897 im6f_reap(struct in6_mfilter *imf)
898 {
899 struct ip6_msource *ims, *tims;
900 struct in6_msource *lims;
901
902 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
903 lims = (struct in6_msource *)ims;
904 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
905 (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
906 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
907 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
908 free(ims, M_IN6MFILTER);
909 imf->im6f_nsrc--;
910 }
911 }
912 }
913
914 /*
915 * Purge socket-layer filter set.
916 */
917 static void
im6f_purge(struct in6_mfilter * imf)918 im6f_purge(struct in6_mfilter *imf)
919 {
920 struct ip6_msource *ims, *tims;
921
922 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
923 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
924 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
925 free(ims, M_IN6MFILTER);
926 imf->im6f_nsrc--;
927 }
928 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
929 KASSERT(RB_EMPTY(&imf->im6f_sources),
930 ("%s: im6f_sources not empty", __func__));
931 }
932
933 /*
934 * Look up a source filter entry for a multicast group.
935 *
936 * inm is the group descriptor to work with.
937 * addr is the IPv6 address to look up.
938 * noalloc may be non-zero to suppress allocation of sources.
939 * *pims will be set to the address of the retrieved or allocated source.
940 *
941 * SMPng: NOTE: may be called with locks held.
942 * Return 0 if successful, otherwise return a non-zero error code.
943 */
944 static int
in6m_get_source(struct in6_multi * inm,const struct in6_addr * addr,const int noalloc,struct ip6_msource ** pims)945 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
946 const int noalloc, struct ip6_msource **pims)
947 {
948 struct ip6_msource find;
949 struct ip6_msource *ims, *nims;
950 #ifdef KTR
951 char ip6tbuf[INET6_ADDRSTRLEN];
952 #endif
953
954 find.im6s_addr = *addr;
955 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
956 if (ims == NULL && !noalloc) {
957 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
958 return (ENOSPC);
959 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
960 M_NOWAIT | M_ZERO);
961 if (nims == NULL)
962 return (ENOMEM);
963 nims->im6s_addr = *addr;
964 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
965 ++inm->in6m_nsrc;
966 ims = nims;
967 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
968 ip6_sprintf(ip6tbuf, addr), ims);
969 }
970
971 *pims = ims;
972 return (0);
973 }
974
975 /*
976 * Merge socket-layer source into MLD-layer source.
977 * If rollback is non-zero, perform the inverse of the merge.
978 */
979 static void
im6s_merge(struct ip6_msource * ims,const struct in6_msource * lims,const int rollback)980 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
981 const int rollback)
982 {
983 int n = rollback ? -1 : 1;
984 #ifdef KTR
985 char ip6tbuf[INET6_ADDRSTRLEN];
986
987 ip6_sprintf(ip6tbuf, &lims->im6s_addr);
988 #endif
989
990 if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
991 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
992 ims->im6s_st[1].ex -= n;
993 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
994 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
995 ims->im6s_st[1].in -= n;
996 }
997
998 if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
999 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
1000 ims->im6s_st[1].ex += n;
1001 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
1002 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
1003 ims->im6s_st[1].in += n;
1004 }
1005 }
1006
1007 /*
1008 * Atomically update the global in6_multi state, when a membership's
1009 * filter list is being updated in any way.
1010 *
1011 * imf is the per-inpcb-membership group filter pointer.
1012 * A fake imf may be passed for in-kernel consumers.
1013 *
1014 * XXX This is a candidate for a set-symmetric-difference style loop
1015 * which would eliminate the repeated lookup from root of ims nodes,
1016 * as they share the same key space.
1017 *
1018 * If any error occurred this function will back out of refcounts
1019 * and return a non-zero value.
1020 */
1021 static int
in6m_merge(struct in6_multi * inm,struct in6_mfilter * imf)1022 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1023 {
1024 struct ip6_msource *ims, *nims;
1025 struct in6_msource *lims;
1026 int schanged, error;
1027 int nsrc0, nsrc1;
1028
1029 schanged = 0;
1030 error = 0;
1031 nsrc1 = nsrc0 = 0;
1032 IN6_MULTI_LIST_LOCK_ASSERT();
1033
1034 /*
1035 * Update the source filters first, as this may fail.
1036 * Maintain count of in-mode filters at t0, t1. These are
1037 * used to work out if we transition into ASM mode or not.
1038 * Maintain a count of source filters whose state was
1039 * actually modified by this operation.
1040 */
1041 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1042 lims = (struct in6_msource *)ims;
1043 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1044 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1045 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1046 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1047 ++schanged;
1048 if (error)
1049 break;
1050 im6s_merge(nims, lims, 0);
1051 }
1052 if (error) {
1053 struct ip6_msource *bims;
1054
1055 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1056 lims = (struct in6_msource *)ims;
1057 if (lims->im6sl_st[0] == lims->im6sl_st[1])
1058 continue;
1059 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1060 if (bims == NULL)
1061 continue;
1062 im6s_merge(bims, lims, 1);
1063 }
1064 goto out_reap;
1065 }
1066
1067 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1068 __func__, nsrc0, nsrc1);
1069
1070 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1071 if (imf->im6f_st[0] == imf->im6f_st[1] &&
1072 imf->im6f_st[1] == MCAST_INCLUDE) {
1073 if (nsrc1 == 0) {
1074 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1075 --inm->in6m_st[1].iss_in;
1076 }
1077 }
1078
1079 /* Handle filter mode transition on socket. */
1080 if (imf->im6f_st[0] != imf->im6f_st[1]) {
1081 CTR3(KTR_MLD, "%s: imf transition %d to %d",
1082 __func__, imf->im6f_st[0], imf->im6f_st[1]);
1083
1084 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1085 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1086 --inm->in6m_st[1].iss_ex;
1087 } else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1088 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1089 --inm->in6m_st[1].iss_in;
1090 }
1091
1092 if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1093 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1094 inm->in6m_st[1].iss_ex++;
1095 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1096 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1097 inm->in6m_st[1].iss_in++;
1098 }
1099 }
1100
1101 /*
1102 * Track inm filter state in terms of listener counts.
1103 * If there are any exclusive listeners, stack-wide
1104 * membership is exclusive.
1105 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1106 * If no listeners remain, state is undefined at t1,
1107 * and the MLD lifecycle for this group should finish.
1108 */
1109 if (inm->in6m_st[1].iss_ex > 0) {
1110 CTR1(KTR_MLD, "%s: transition to EX", __func__);
1111 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1112 } else if (inm->in6m_st[1].iss_in > 0) {
1113 CTR1(KTR_MLD, "%s: transition to IN", __func__);
1114 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1115 } else {
1116 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1117 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1118 }
1119
1120 /* Decrement ASM listener count on transition out of ASM mode. */
1121 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1122 if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1123 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1124 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1125 --inm->in6m_st[1].iss_asm;
1126 }
1127 }
1128
1129 /* Increment ASM listener count on transition to ASM mode. */
1130 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1131 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1132 inm->in6m_st[1].iss_asm++;
1133 }
1134
1135 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1136 in6m_print(inm);
1137
1138 out_reap:
1139 if (schanged > 0) {
1140 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1141 in6m_reap(inm);
1142 }
1143 return (error);
1144 }
1145
1146 /*
1147 * Mark an in6_multi's filter set deltas as committed.
1148 * Called by MLD after a state change has been enqueued.
1149 */
1150 void
in6m_commit(struct in6_multi * inm)1151 in6m_commit(struct in6_multi *inm)
1152 {
1153 struct ip6_msource *ims;
1154
1155 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1156 CTR1(KTR_MLD, "%s: pre commit:", __func__);
1157 in6m_print(inm);
1158
1159 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1160 ims->im6s_st[0] = ims->im6s_st[1];
1161 }
1162 inm->in6m_st[0] = inm->in6m_st[1];
1163 }
1164
1165 /*
1166 * Reap unreferenced nodes from an in6_multi's filter set.
1167 */
1168 static void
in6m_reap(struct in6_multi * inm)1169 in6m_reap(struct in6_multi *inm)
1170 {
1171 struct ip6_msource *ims, *tims;
1172
1173 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1174 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1175 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1176 ims->im6s_stp != 0)
1177 continue;
1178 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1179 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1180 free(ims, M_IP6MSOURCE);
1181 inm->in6m_nsrc--;
1182 }
1183 }
1184
1185 /*
1186 * Purge all source nodes from an in6_multi's filter set.
1187 */
1188 static void
in6m_purge(struct in6_multi * inm)1189 in6m_purge(struct in6_multi *inm)
1190 {
1191 struct ip6_msource *ims, *tims;
1192
1193 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1194 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1195 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1196 free(ims, M_IP6MSOURCE);
1197 inm->in6m_nsrc--;
1198 }
1199 /* Free state-change requests that might be queued. */
1200 mbufq_drain(&inm->in6m_scq);
1201 }
1202
1203 /*
1204 * Join a multicast address w/o sources.
1205 * KAME compatibility entry point.
1206 *
1207 * SMPng: Assume no mc locks held by caller.
1208 */
1209 int
in6_joingroup(struct ifnet * ifp,const struct in6_addr * mcaddr,struct in6_mfilter * imf,struct in6_multi ** pinm,const int delay)1210 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1211 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1212 const int delay)
1213 {
1214 int error;
1215
1216 IN6_MULTI_LOCK();
1217 error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1218 IN6_MULTI_UNLOCK();
1219 return (error);
1220 }
1221
1222 /*
1223 * Join a multicast group; real entry point.
1224 *
1225 * Only preserves atomicity at inm level.
1226 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1227 *
1228 * If the MLD downcall fails, the group is not joined, and an error
1229 * code is returned.
1230 */
1231 static int
in6_joingroup_locked(struct ifnet * ifp,const struct in6_addr * mcaddr,struct in6_mfilter * imf,struct in6_multi ** pinm,const int delay)1232 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1233 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1234 const int delay)
1235 {
1236 struct in6_multi_head inmh;
1237 struct in6_mfilter timf;
1238 struct in6_multi *inm;
1239 struct ifmultiaddr *ifma;
1240 int error;
1241 #ifdef KTR
1242 char ip6tbuf[INET6_ADDRSTRLEN];
1243 #endif
1244
1245 /*
1246 * Sanity: Check scope zone ID was set for ifp, if and
1247 * only if group is scoped to an interface.
1248 */
1249 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1250 ("%s: not a multicast address", __func__));
1251 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1252 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1253 KASSERT(mcaddr->s6_addr16[1] != 0,
1254 ("%s: scope zone ID not set", __func__));
1255 }
1256
1257 IN6_MULTI_LOCK_ASSERT();
1258 IN6_MULTI_LIST_UNLOCK_ASSERT();
1259
1260 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1261 ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1262
1263 error = 0;
1264 inm = NULL;
1265
1266 /*
1267 * If no imf was specified (i.e. kernel consumer),
1268 * fake one up and assume it is an ASM join.
1269 */
1270 if (imf == NULL) {
1271 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1272 imf = &timf;
1273 }
1274 error = in6_getmulti(ifp, mcaddr, &inm);
1275 if (error) {
1276 CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1277 return (error);
1278 }
1279
1280 IN6_MULTI_LIST_LOCK();
1281 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1282 error = in6m_merge(inm, imf);
1283 if (error) {
1284 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1285 goto out_in6m_release;
1286 }
1287
1288 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1289 error = mld_change_state(inm, delay);
1290 if (error) {
1291 CTR1(KTR_MLD, "%s: failed to update source", __func__);
1292 goto out_in6m_release;
1293 }
1294
1295 out_in6m_release:
1296 SLIST_INIT(&inmh);
1297 if (error) {
1298 struct epoch_tracker et;
1299
1300 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1301 IF_ADDR_WLOCK(ifp);
1302 NET_EPOCH_ENTER(et);
1303 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1304 if (ifma->ifma_protospec == inm) {
1305 ifma->ifma_protospec = NULL;
1306 break;
1307 }
1308 }
1309 in6m_disconnect_locked(&inmh, inm);
1310 in6m_rele_locked(&inmh, inm);
1311 NET_EPOCH_EXIT(et);
1312 IF_ADDR_WUNLOCK(ifp);
1313 } else {
1314 *pinm = inm;
1315 }
1316 IN6_MULTI_LIST_UNLOCK();
1317 in6m_release_list_deferred(&inmh);
1318 return (error);
1319 }
1320
1321 /*
1322 * Leave a multicast group; unlocked entry point.
1323 */
1324 int
in6_leavegroup(struct in6_multi * inm,struct in6_mfilter * imf)1325 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1326 {
1327 int error;
1328
1329 IN6_MULTI_LOCK();
1330 error = in6_leavegroup_locked(inm, imf);
1331 IN6_MULTI_UNLOCK();
1332 return (error);
1333 }
1334
1335 /*
1336 * Leave a multicast group; real entry point.
1337 * All source filters will be expunged.
1338 *
1339 * Only preserves atomicity at inm level.
1340 *
1341 * Holding the write lock for the INP which contains imf
1342 * is highly advisable. We can't assert for it as imf does not
1343 * contain a back-pointer to the owning inp.
1344 *
1345 * Note: This is not the same as in6m_release(*) as this function also
1346 * makes a state change downcall into MLD.
1347 */
1348 int
in6_leavegroup_locked(struct in6_multi * inm,struct in6_mfilter * imf)1349 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1350 {
1351 struct in6_multi_head inmh;
1352 struct in6_mfilter timf;
1353 struct ifnet *ifp;
1354 int error;
1355 #ifdef KTR
1356 char ip6tbuf[INET6_ADDRSTRLEN];
1357 #endif
1358
1359 error = 0;
1360
1361 IN6_MULTI_LOCK_ASSERT();
1362
1363 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1364 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1365 (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1366 imf);
1367
1368 /*
1369 * If no imf was specified (i.e. kernel consumer),
1370 * fake one up and assume it is an ASM join.
1371 */
1372 if (imf == NULL) {
1373 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1374 imf = &timf;
1375 }
1376
1377 /*
1378 * Begin state merge transaction at MLD layer.
1379 *
1380 * As this particular invocation should not cause any memory
1381 * to be allocated, and there is no opportunity to roll back
1382 * the transaction, it MUST NOT fail.
1383 */
1384
1385 ifp = inm->in6m_ifp;
1386 IN6_MULTI_LIST_LOCK();
1387 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1388 error = in6m_merge(inm, imf);
1389 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1390
1391 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1392 error = 0;
1393 if (ifp)
1394 error = mld_change_state(inm, 0);
1395 if (error)
1396 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1397
1398 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1399 if (ifp)
1400 IF_ADDR_WLOCK(ifp);
1401
1402 SLIST_INIT(&inmh);
1403 if (inm->in6m_refcount == 1)
1404 in6m_disconnect_locked(&inmh, inm);
1405 in6m_rele_locked(&inmh, inm);
1406 if (ifp)
1407 IF_ADDR_WUNLOCK(ifp);
1408 IN6_MULTI_LIST_UNLOCK();
1409 in6m_release_list_deferred(&inmh);
1410 return (error);
1411 }
1412
1413 /*
1414 * Block or unblock an ASM multicast source on an inpcb.
1415 * This implements the delta-based API described in RFC 3678.
1416 *
1417 * The delta-based API applies only to exclusive-mode memberships.
1418 * An MLD downcall will be performed.
1419 *
1420 * Return 0 if successful, otherwise return an appropriate error code.
1421 */
1422 static int
in6p_block_unblock_source(struct inpcb * inp,struct sockopt * sopt)1423 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1424 {
1425 struct group_source_req gsr;
1426 struct epoch_tracker et;
1427 sockunion_t *gsa, *ssa;
1428 struct ifnet *ifp;
1429 struct in6_mfilter *imf;
1430 struct ip6_moptions *imo;
1431 struct in6_msource *ims;
1432 struct in6_multi *inm;
1433 uint16_t fmode;
1434 int error, doblock;
1435 #ifdef KTR
1436 char ip6tbuf[INET6_ADDRSTRLEN];
1437 #endif
1438
1439 ifp = NULL;
1440 error = 0;
1441 doblock = 0;
1442
1443 memset(&gsr, 0, sizeof(struct group_source_req));
1444 gsa = (sockunion_t *)&gsr.gsr_group;
1445 ssa = (sockunion_t *)&gsr.gsr_source;
1446
1447 switch (sopt->sopt_name) {
1448 case MCAST_BLOCK_SOURCE:
1449 case MCAST_UNBLOCK_SOURCE:
1450 error = sooptcopyin(sopt, &gsr,
1451 sizeof(struct group_source_req),
1452 sizeof(struct group_source_req));
1453 if (error)
1454 return (error);
1455
1456 if (gsa->sin6.sin6_family != AF_INET6 ||
1457 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1458 return (EINVAL);
1459
1460 if (ssa->sin6.sin6_family != AF_INET6 ||
1461 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1462 return (EINVAL);
1463
1464 /*
1465 * XXXGL: this function should use ifnet_byindex_ref, or
1466 * expand the epoch section all the way to where we put
1467 * the reference.
1468 */
1469 NET_EPOCH_ENTER(et);
1470 ifp = ifnet_byindex(gsr.gsr_interface);
1471 NET_EPOCH_EXIT(et);
1472 if (ifp == NULL)
1473 return (EADDRNOTAVAIL);
1474
1475 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1476 doblock = 1;
1477 break;
1478
1479 default:
1480 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1481 __func__, sopt->sopt_name);
1482 return (EOPNOTSUPP);
1483 break;
1484 }
1485
1486 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1487 return (EINVAL);
1488
1489 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1490
1491 /*
1492 * Check if we are actually a member of this group.
1493 */
1494 imo = in6p_findmoptions(inp);
1495 imf = im6o_match_group(imo, ifp, &gsa->sa);
1496 if (imf == NULL) {
1497 error = EADDRNOTAVAIL;
1498 goto out_in6p_locked;
1499 }
1500 inm = imf->im6f_in6m;
1501
1502 /*
1503 * Attempting to use the delta-based API on an
1504 * non exclusive-mode membership is an error.
1505 */
1506 fmode = imf->im6f_st[0];
1507 if (fmode != MCAST_EXCLUDE) {
1508 error = EINVAL;
1509 goto out_in6p_locked;
1510 }
1511
1512 /*
1513 * Deal with error cases up-front:
1514 * Asked to block, but already blocked; or
1515 * Asked to unblock, but nothing to unblock.
1516 * If adding a new block entry, allocate it.
1517 */
1518 ims = im6o_match_source(imf, &ssa->sa);
1519 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1520 CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1521 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1522 doblock ? "" : "not ");
1523 error = EADDRNOTAVAIL;
1524 goto out_in6p_locked;
1525 }
1526
1527 INP_WLOCK_ASSERT(inp);
1528
1529 /*
1530 * Begin state merge transaction at socket layer.
1531 */
1532 if (doblock) {
1533 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1534 ims = im6f_graft(imf, fmode, &ssa->sin6);
1535 if (ims == NULL)
1536 error = ENOMEM;
1537 } else {
1538 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1539 error = im6f_prune(imf, &ssa->sin6);
1540 }
1541
1542 if (error) {
1543 CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1544 goto out_im6f_rollback;
1545 }
1546
1547 /*
1548 * Begin state merge transaction at MLD layer.
1549 */
1550 IN6_MULTI_LIST_LOCK();
1551 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1552 error = in6m_merge(inm, imf);
1553 if (error)
1554 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1555 else {
1556 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1557 error = mld_change_state(inm, 0);
1558 if (error)
1559 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1560 }
1561
1562 IN6_MULTI_LIST_UNLOCK();
1563
1564 out_im6f_rollback:
1565 if (error)
1566 im6f_rollback(imf);
1567 else
1568 im6f_commit(imf);
1569
1570 im6f_reap(imf);
1571
1572 out_in6p_locked:
1573 INP_WUNLOCK(inp);
1574 return (error);
1575 }
1576
1577 /*
1578 * Given an inpcb, return its multicast options structure pointer. Accepts
1579 * an unlocked inpcb pointer, but will return it locked. May sleep.
1580 *
1581 * SMPng: NOTE: Returns with the INP write lock held.
1582 */
1583 static struct ip6_moptions *
in6p_findmoptions(struct inpcb * inp)1584 in6p_findmoptions(struct inpcb *inp)
1585 {
1586 struct ip6_moptions *imo;
1587
1588 INP_WLOCK(inp);
1589 if (inp->in6p_moptions != NULL)
1590 return (inp->in6p_moptions);
1591
1592 INP_WUNLOCK(inp);
1593
1594 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1595
1596 imo->im6o_multicast_ifp = NULL;
1597 imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1598 imo->im6o_multicast_loop = in6_mcast_loop;
1599 STAILQ_INIT(&imo->im6o_head);
1600
1601 INP_WLOCK(inp);
1602 if (inp->in6p_moptions != NULL) {
1603 free(imo, M_IP6MOPTS);
1604 return (inp->in6p_moptions);
1605 }
1606 inp->in6p_moptions = imo;
1607 return (imo);
1608 }
1609
1610 /*
1611 * Discard the IPv6 multicast options (and source filters).
1612 *
1613 * SMPng: NOTE: assumes INP write lock is held.
1614 *
1615 * XXX can all be safely deferred to epoch_call
1616 *
1617 */
1618
1619 static void
inp_gcmoptions(struct ip6_moptions * imo)1620 inp_gcmoptions(struct ip6_moptions *imo)
1621 {
1622 struct in6_mfilter *imf;
1623 struct in6_multi *inm;
1624 struct ifnet *ifp;
1625
1626 while ((imf = ip6_mfilter_first(&imo->im6o_head)) != NULL) {
1627 ip6_mfilter_remove(&imo->im6o_head, imf);
1628
1629 im6f_leave(imf);
1630 if ((inm = imf->im6f_in6m) != NULL) {
1631 if ((ifp = inm->in6m_ifp) != NULL) {
1632 CURVNET_SET(ifp->if_vnet);
1633 (void)in6_leavegroup(inm, imf);
1634 CURVNET_RESTORE();
1635 } else {
1636 (void)in6_leavegroup(inm, imf);
1637 }
1638 }
1639 ip6_mfilter_free(imf);
1640 }
1641 free(imo, M_IP6MOPTS);
1642 }
1643
1644 void
ip6_freemoptions(struct ip6_moptions * imo)1645 ip6_freemoptions(struct ip6_moptions *imo)
1646 {
1647 if (imo == NULL)
1648 return;
1649 inp_gcmoptions(imo);
1650 }
1651
1652 /*
1653 * Atomically get source filters on a socket for an IPv6 multicast group.
1654 * Called with INP lock held; returns with lock released.
1655 */
1656 static int
in6p_get_source_filters(struct inpcb * inp,struct sockopt * sopt)1657 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1658 {
1659 struct epoch_tracker et;
1660 struct __msfilterreq msfr;
1661 sockunion_t *gsa;
1662 struct ifnet *ifp;
1663 struct ip6_moptions *imo;
1664 struct in6_mfilter *imf;
1665 struct ip6_msource *ims;
1666 struct in6_msource *lims;
1667 struct sockaddr_in6 *psin;
1668 struct sockaddr_storage *ptss;
1669 struct sockaddr_storage *tss;
1670 int error;
1671 size_t nsrcs, ncsrcs;
1672
1673 INP_WLOCK_ASSERT(inp);
1674
1675 imo = inp->in6p_moptions;
1676 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1677
1678 INP_WUNLOCK(inp);
1679
1680 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1681 sizeof(struct __msfilterreq));
1682 if (error)
1683 return (error);
1684
1685 if (msfr.msfr_group.ss_family != AF_INET6 ||
1686 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1687 return (EINVAL);
1688
1689 gsa = (sockunion_t *)&msfr.msfr_group;
1690 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1691 return (EINVAL);
1692
1693 /*
1694 * XXXGL: this function should use ifnet_byindex_ref, or expand the
1695 * epoch section all the way to where the interface is referenced.
1696 */
1697 NET_EPOCH_ENTER(et);
1698 ifp = ifnet_byindex(msfr.msfr_ifindex);
1699 NET_EPOCH_EXIT(et);
1700 if (ifp == NULL)
1701 return (EADDRNOTAVAIL);
1702 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1703
1704 INP_WLOCK(inp);
1705
1706 /*
1707 * Lookup group on the socket.
1708 */
1709 imf = im6o_match_group(imo, ifp, &gsa->sa);
1710 if (imf == NULL) {
1711 INP_WUNLOCK(inp);
1712 return (EADDRNOTAVAIL);
1713 }
1714
1715 /*
1716 * Ignore memberships which are in limbo.
1717 */
1718 if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1719 INP_WUNLOCK(inp);
1720 return (EAGAIN);
1721 }
1722 msfr.msfr_fmode = imf->im6f_st[1];
1723
1724 /*
1725 * If the user specified a buffer, copy out the source filter
1726 * entries to userland gracefully.
1727 * We only copy out the number of entries which userland
1728 * has asked for, but we always tell userland how big the
1729 * buffer really needs to be.
1730 */
1731 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1732 msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1733 tss = NULL;
1734 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1735 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1736 M_TEMP, M_NOWAIT | M_ZERO);
1737 if (tss == NULL) {
1738 INP_WUNLOCK(inp);
1739 return (ENOBUFS);
1740 }
1741 }
1742
1743 /*
1744 * Count number of sources in-mode at t0.
1745 * If buffer space exists and remains, copy out source entries.
1746 */
1747 nsrcs = msfr.msfr_nsrcs;
1748 ncsrcs = 0;
1749 ptss = tss;
1750 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1751 lims = (struct in6_msource *)ims;
1752 if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1753 lims->im6sl_st[0] != imf->im6f_st[0])
1754 continue;
1755 ++ncsrcs;
1756 if (tss != NULL && nsrcs > 0) {
1757 psin = (struct sockaddr_in6 *)ptss;
1758 psin->sin6_family = AF_INET6;
1759 psin->sin6_len = sizeof(struct sockaddr_in6);
1760 psin->sin6_addr = lims->im6s_addr;
1761 psin->sin6_port = 0;
1762 --nsrcs;
1763 ++ptss;
1764 }
1765 }
1766
1767 INP_WUNLOCK(inp);
1768
1769 if (tss != NULL) {
1770 error = copyout(tss, msfr.msfr_srcs,
1771 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1772 free(tss, M_TEMP);
1773 if (error)
1774 return (error);
1775 }
1776
1777 msfr.msfr_nsrcs = ncsrcs;
1778 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1779
1780 return (error);
1781 }
1782
1783 /*
1784 * Return the IP multicast options in response to user getsockopt().
1785 */
1786 int
ip6_getmoptions(struct inpcb * inp,struct sockopt * sopt)1787 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1788 {
1789 struct ip6_moptions *im6o;
1790 int error;
1791 u_int optval;
1792
1793 INP_WLOCK(inp);
1794 im6o = inp->in6p_moptions;
1795 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
1796 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1797 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM) {
1798 INP_WUNLOCK(inp);
1799 return (EOPNOTSUPP);
1800 }
1801
1802 error = 0;
1803 switch (sopt->sopt_name) {
1804 case IPV6_MULTICAST_IF:
1805 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1806 optval = 0;
1807 } else {
1808 optval = im6o->im6o_multicast_ifp->if_index;
1809 }
1810 INP_WUNLOCK(inp);
1811 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1812 break;
1813
1814 case IPV6_MULTICAST_HOPS:
1815 if (im6o == NULL)
1816 optval = V_ip6_defmcasthlim;
1817 else
1818 optval = im6o->im6o_multicast_hlim;
1819 INP_WUNLOCK(inp);
1820 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1821 break;
1822
1823 case IPV6_MULTICAST_LOOP:
1824 if (im6o == NULL)
1825 optval = in6_mcast_loop; /* XXX VIMAGE */
1826 else
1827 optval = im6o->im6o_multicast_loop;
1828 INP_WUNLOCK(inp);
1829 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1830 break;
1831
1832 case IPV6_MSFILTER:
1833 if (im6o == NULL) {
1834 error = EADDRNOTAVAIL;
1835 INP_WUNLOCK(inp);
1836 } else {
1837 error = in6p_get_source_filters(inp, sopt);
1838 }
1839 break;
1840
1841 default:
1842 INP_WUNLOCK(inp);
1843 error = ENOPROTOOPT;
1844 break;
1845 }
1846
1847 INP_UNLOCK_ASSERT(inp);
1848
1849 return (error);
1850 }
1851
1852 /*
1853 * Look up the ifnet to use for a multicast group membership,
1854 * given the address of an IPv6 group.
1855 *
1856 * This routine exists to support legacy IPv6 multicast applications.
1857 *
1858 * Use the socket's current FIB number for any required FIB lookup. Look up the
1859 * group address in the unicast FIB, and use its ifp; usually, this points to
1860 * the default next-hop. If the FIB lookup fails, return NULL.
1861 *
1862 * FUTURE: Support multiple forwarding tables for IPv6.
1863 *
1864 * Returns NULL if no ifp could be found.
1865 */
1866 static struct ifnet *
in6p_lookup_mcast_ifp(const struct inpcb * inp,const struct sockaddr_in6 * gsin6)1867 in6p_lookup_mcast_ifp(const struct inpcb *inp, const struct sockaddr_in6 *gsin6)
1868 {
1869 struct nhop_object *nh;
1870 struct in6_addr dst;
1871 uint32_t scopeid;
1872 uint32_t fibnum;
1873
1874 KASSERT(gsin6->sin6_family == AF_INET6,
1875 ("%s: not AF_INET6 group", __func__));
1876
1877 in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1878 fibnum = inp->inp_inc.inc_fibnum;
1879 nh = fib6_lookup(fibnum, &dst, scopeid, 0, 0);
1880
1881 return (nh ? nh->nh_ifp : NULL);
1882 }
1883
1884 /*
1885 * Join an IPv6 multicast group, possibly with a source.
1886 *
1887 * FIXME: The KAME use of the unspecified address (::)
1888 * to join *all* multicast groups is currently unsupported.
1889 *
1890 * XXXGL: this function multiple times uses ifnet_byindex() without
1891 * proper protection - staying in epoch, or putting reference on ifnet.
1892 */
1893 static int
in6p_join_group(struct inpcb * inp,struct sockopt * sopt)1894 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1895 {
1896 struct in6_multi_head inmh;
1897 struct group_source_req gsr;
1898 struct epoch_tracker et;
1899 sockunion_t *gsa, *ssa;
1900 struct ifnet *ifp;
1901 struct in6_mfilter *imf;
1902 struct ip6_moptions *imo;
1903 struct in6_multi *inm;
1904 struct in6_msource *lims;
1905 int error, is_new;
1906
1907 SLIST_INIT(&inmh);
1908 ifp = NULL;
1909 lims = NULL;
1910 error = 0;
1911
1912 memset(&gsr, 0, sizeof(struct group_source_req));
1913 gsa = (sockunion_t *)&gsr.gsr_group;
1914 gsa->ss.ss_family = AF_UNSPEC;
1915 ssa = (sockunion_t *)&gsr.gsr_source;
1916 ssa->ss.ss_family = AF_UNSPEC;
1917
1918 /*
1919 * Chew everything into struct group_source_req.
1920 * Overwrite the port field if present, as the sockaddr
1921 * being copied in may be matched with a binary comparison.
1922 * Ignore passed-in scope ID.
1923 */
1924 switch (sopt->sopt_name) {
1925 case IPV6_JOIN_GROUP: {
1926 struct ipv6_mreq mreq;
1927
1928 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1929 sizeof(struct ipv6_mreq));
1930 if (error)
1931 return (error);
1932
1933 gsa->sin6.sin6_family = AF_INET6;
1934 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1935 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1936
1937 if (mreq.ipv6mr_interface == 0) {
1938 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1939 } else {
1940 NET_EPOCH_ENTER(et);
1941 ifp = ifnet_byindex(mreq.ipv6mr_interface);
1942 NET_EPOCH_EXIT(et);
1943 if (ifp == NULL)
1944 return (EADDRNOTAVAIL);
1945 }
1946 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1947 __func__, mreq.ipv6mr_interface, ifp);
1948 } break;
1949
1950 case MCAST_JOIN_GROUP:
1951 case MCAST_JOIN_SOURCE_GROUP:
1952 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1953 error = sooptcopyin(sopt, &gsr,
1954 sizeof(struct group_req),
1955 sizeof(struct group_req));
1956 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1957 error = sooptcopyin(sopt, &gsr,
1958 sizeof(struct group_source_req),
1959 sizeof(struct group_source_req));
1960 }
1961 if (error)
1962 return (error);
1963
1964 if (gsa->sin6.sin6_family != AF_INET6 ||
1965 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1966 return (EINVAL);
1967
1968 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1969 if (ssa->sin6.sin6_family != AF_INET6 ||
1970 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1971 return (EINVAL);
1972 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1973 return (EINVAL);
1974 /*
1975 * TODO: Validate embedded scope ID in source
1976 * list entry against passed-in ifp, if and only
1977 * if source list filter entry is iface or node local.
1978 */
1979 in6_clearscope(&ssa->sin6.sin6_addr);
1980 ssa->sin6.sin6_port = 0;
1981 ssa->sin6.sin6_scope_id = 0;
1982 }
1983 NET_EPOCH_ENTER(et);
1984 ifp = ifnet_byindex(gsr.gsr_interface);
1985 NET_EPOCH_EXIT(et);
1986 if (ifp == NULL)
1987 return (EADDRNOTAVAIL);
1988 break;
1989
1990 default:
1991 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1992 __func__, sopt->sopt_name);
1993 return (EOPNOTSUPP);
1994 break;
1995 }
1996
1997 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1998 return (EINVAL);
1999
2000 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2001 return (EADDRNOTAVAIL);
2002
2003 gsa->sin6.sin6_port = 0;
2004 gsa->sin6.sin6_scope_id = 0;
2005
2006 /*
2007 * Always set the scope zone ID on memberships created from userland.
2008 * Use the passed-in ifp to do this.
2009 * XXX The in6_setscope() return value is meaningless.
2010 * XXX SCOPE6_LOCK() is taken by in6_setscope().
2011 */
2012 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2013
2014 IN6_MULTI_LOCK();
2015
2016 /*
2017 * Find the membership in the membership list.
2018 */
2019 imo = in6p_findmoptions(inp);
2020 imf = im6o_match_group(imo, ifp, &gsa->sa);
2021 if (imf == NULL) {
2022 is_new = 1;
2023 inm = NULL;
2024
2025 if (ip6_mfilter_count(&imo->im6o_head) >= IPV6_MAX_MEMBERSHIPS) {
2026 error = ENOMEM;
2027 goto out_in6p_locked;
2028 }
2029 } else {
2030 is_new = 0;
2031 inm = imf->im6f_in6m;
2032
2033 if (ssa->ss.ss_family != AF_UNSPEC) {
2034 /*
2035 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2036 * is an error. On an existing inclusive membership,
2037 * it just adds the source to the filter list.
2038 */
2039 if (imf->im6f_st[1] != MCAST_INCLUDE) {
2040 error = EINVAL;
2041 goto out_in6p_locked;
2042 }
2043 /*
2044 * Throw out duplicates.
2045 *
2046 * XXX FIXME: This makes a naive assumption that
2047 * even if entries exist for *ssa in this imf,
2048 * they will be rejected as dupes, even if they
2049 * are not valid in the current mode (in-mode).
2050 *
2051 * in6_msource is transactioned just as for anything
2052 * else in SSM -- but note naive use of in6m_graft()
2053 * below for allocating new filter entries.
2054 *
2055 * This is only an issue if someone mixes the
2056 * full-state SSM API with the delta-based API,
2057 * which is discouraged in the relevant RFCs.
2058 */
2059 lims = im6o_match_source(imf, &ssa->sa);
2060 if (lims != NULL /*&&
2061 lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2062 error = EADDRNOTAVAIL;
2063 goto out_in6p_locked;
2064 }
2065 } else {
2066 /*
2067 * MCAST_JOIN_GROUP alone, on any existing membership,
2068 * is rejected, to stop the same inpcb tying up
2069 * multiple refs to the in_multi.
2070 * On an existing inclusive membership, this is also
2071 * an error; if you want to change filter mode,
2072 * you must use the userland API setsourcefilter().
2073 * XXX We don't reject this for imf in UNDEFINED
2074 * state at t1, because allocation of a filter
2075 * is atomic with allocation of a membership.
2076 */
2077 error = EADDRINUSE;
2078 goto out_in6p_locked;
2079 }
2080 }
2081
2082 /*
2083 * Begin state merge transaction at socket layer.
2084 */
2085 INP_WLOCK_ASSERT(inp);
2086
2087 /*
2088 * Graft new source into filter list for this inpcb's
2089 * membership of the group. The in6_multi may not have
2090 * been allocated yet if this is a new membership, however,
2091 * the in_mfilter slot will be allocated and must be initialized.
2092 *
2093 * Note: Grafting of exclusive mode filters doesn't happen
2094 * in this path.
2095 * XXX: Should check for non-NULL lims (node exists but may
2096 * not be in-mode) for interop with full-state API.
2097 */
2098 if (ssa->ss.ss_family != AF_UNSPEC) {
2099 /* Membership starts in IN mode */
2100 if (is_new) {
2101 CTR1(KTR_MLD, "%s: new join w/source", __func__);
2102 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2103 if (imf == NULL) {
2104 error = ENOMEM;
2105 goto out_in6p_locked;
2106 }
2107 } else {
2108 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2109 }
2110 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2111 if (lims == NULL) {
2112 CTR1(KTR_MLD, "%s: merge imf state failed",
2113 __func__);
2114 error = ENOMEM;
2115 goto out_in6p_locked;
2116 }
2117 } else {
2118 /* No address specified; Membership starts in EX mode */
2119 if (is_new) {
2120 CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2121 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2122 if (imf == NULL) {
2123 error = ENOMEM;
2124 goto out_in6p_locked;
2125 }
2126 }
2127 }
2128
2129 /*
2130 * Begin state merge transaction at MLD layer.
2131 */
2132 if (is_new) {
2133 in_pcbref(inp);
2134 INP_WUNLOCK(inp);
2135
2136 error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2137 &imf->im6f_in6m, 0);
2138
2139 INP_WLOCK(inp);
2140 if (in_pcbrele_wlocked(inp)) {
2141 error = ENXIO;
2142 goto out_in6p_unlocked;
2143 }
2144 if (error) {
2145 goto out_in6p_locked;
2146 }
2147 /*
2148 * NOTE: Refcount from in6_joingroup_locked()
2149 * is protecting membership.
2150 */
2151 ip6_mfilter_insert(&imo->im6o_head, imf);
2152 } else {
2153 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2154 IN6_MULTI_LIST_LOCK();
2155 error = in6m_merge(inm, imf);
2156 if (error) {
2157 CTR1(KTR_MLD, "%s: failed to merge inm state",
2158 __func__);
2159 IN6_MULTI_LIST_UNLOCK();
2160 im6f_rollback(imf);
2161 im6f_reap(imf);
2162 goto out_in6p_locked;
2163 }
2164 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2165 error = mld_change_state(inm, 0);
2166 IN6_MULTI_LIST_UNLOCK();
2167
2168 if (error) {
2169 CTR1(KTR_MLD, "%s: failed mld downcall",
2170 __func__);
2171 im6f_rollback(imf);
2172 im6f_reap(imf);
2173 goto out_in6p_locked;
2174 }
2175 }
2176
2177 im6f_commit(imf);
2178 imf = NULL;
2179
2180 out_in6p_locked:
2181 INP_WUNLOCK(inp);
2182 out_in6p_unlocked:
2183 IN6_MULTI_UNLOCK();
2184
2185 if (is_new && imf) {
2186 if (imf->im6f_in6m != NULL) {
2187 struct in6_multi_head inmh;
2188
2189 SLIST_INIT(&inmh);
2190 SLIST_INSERT_HEAD(&inmh, imf->im6f_in6m, in6m_defer);
2191 in6m_release_list_deferred(&inmh);
2192 }
2193 ip6_mfilter_free(imf);
2194 }
2195 return (error);
2196 }
2197
2198 /*
2199 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2200 */
2201 static int
in6p_leave_group(struct inpcb * inp,struct sockopt * sopt)2202 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2203 {
2204 struct ipv6_mreq mreq;
2205 struct group_source_req gsr;
2206 struct epoch_tracker et;
2207 sockunion_t *gsa, *ssa;
2208 struct ifnet *ifp;
2209 struct in6_mfilter *imf;
2210 struct ip6_moptions *imo;
2211 struct in6_msource *ims;
2212 struct in6_multi *inm;
2213 uint32_t ifindex;
2214 int error;
2215 bool is_final;
2216 #ifdef KTR
2217 char ip6tbuf[INET6_ADDRSTRLEN];
2218 #endif
2219
2220 ifp = NULL;
2221 ifindex = 0;
2222 error = 0;
2223 is_final = true;
2224
2225 memset(&gsr, 0, sizeof(struct group_source_req));
2226 gsa = (sockunion_t *)&gsr.gsr_group;
2227 gsa->ss.ss_family = AF_UNSPEC;
2228 ssa = (sockunion_t *)&gsr.gsr_source;
2229 ssa->ss.ss_family = AF_UNSPEC;
2230
2231 /*
2232 * Chew everything passed in up into a struct group_source_req
2233 * as that is easier to process.
2234 * Note: Any embedded scope ID in the multicast group passed
2235 * in by userland is ignored, the interface index is the recommended
2236 * mechanism to specify an interface; see below.
2237 */
2238 switch (sopt->sopt_name) {
2239 case IPV6_LEAVE_GROUP:
2240 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2241 sizeof(struct ipv6_mreq));
2242 if (error)
2243 return (error);
2244 gsa->sin6.sin6_family = AF_INET6;
2245 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2246 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2247 gsa->sin6.sin6_port = 0;
2248 gsa->sin6.sin6_scope_id = 0;
2249 ifindex = mreq.ipv6mr_interface;
2250 break;
2251
2252 case MCAST_LEAVE_GROUP:
2253 case MCAST_LEAVE_SOURCE_GROUP:
2254 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2255 error = sooptcopyin(sopt, &gsr,
2256 sizeof(struct group_req),
2257 sizeof(struct group_req));
2258 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2259 error = sooptcopyin(sopt, &gsr,
2260 sizeof(struct group_source_req),
2261 sizeof(struct group_source_req));
2262 }
2263 if (error)
2264 return (error);
2265
2266 if (gsa->sin6.sin6_family != AF_INET6 ||
2267 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2268 return (EINVAL);
2269 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2270 if (ssa->sin6.sin6_family != AF_INET6 ||
2271 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2272 return (EINVAL);
2273 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2274 return (EINVAL);
2275 /*
2276 * TODO: Validate embedded scope ID in source
2277 * list entry against passed-in ifp, if and only
2278 * if source list filter entry is iface or node local.
2279 */
2280 in6_clearscope(&ssa->sin6.sin6_addr);
2281 }
2282 gsa->sin6.sin6_port = 0;
2283 gsa->sin6.sin6_scope_id = 0;
2284 ifindex = gsr.gsr_interface;
2285 break;
2286
2287 default:
2288 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2289 __func__, sopt->sopt_name);
2290 return (EOPNOTSUPP);
2291 break;
2292 }
2293
2294 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2295 return (EINVAL);
2296
2297 /*
2298 * Validate interface index if provided. If no interface index
2299 * was provided separately, attempt to look the membership up
2300 * from the default scope as a last resort to disambiguate
2301 * the membership we are being asked to leave.
2302 * XXX SCOPE6 lock potentially taken here.
2303 */
2304 if (ifindex != 0) {
2305 NET_EPOCH_ENTER(et);
2306 ifp = ifnet_byindex(ifindex);
2307 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2308 if (ifp == NULL)
2309 return (EADDRNOTAVAIL);
2310 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2311 } else {
2312 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2313 if (error)
2314 return (EADDRNOTAVAIL);
2315 /*
2316 * Some badly behaved applications don't pass an ifindex
2317 * or a scope ID, which is an API violation. In this case,
2318 * perform a lookup as per a v6 join.
2319 *
2320 * XXX For now, stomp on zone ID for the corner case.
2321 * This is not the 'KAME way', but we need to see the ifp
2322 * directly until such time as this implementation is
2323 * refactored, assuming the scope IDs are the way to go.
2324 */
2325 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2326 if (ifindex == 0) {
2327 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2328 "ifp for group %s.", __func__,
2329 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2330 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2331 } else {
2332 NET_EPOCH_ENTER(et);
2333 ifp = ifnet_byindex(ifindex);
2334 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2335 }
2336 if (ifp == NULL)
2337 return (EADDRNOTAVAIL);
2338 }
2339
2340 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2341 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2342
2343 IN6_MULTI_LOCK();
2344
2345 /*
2346 * Find the membership in the membership list.
2347 */
2348 imo = in6p_findmoptions(inp);
2349 imf = im6o_match_group(imo, ifp, &gsa->sa);
2350 if (imf == NULL) {
2351 error = EADDRNOTAVAIL;
2352 goto out_in6p_locked;
2353 }
2354 inm = imf->im6f_in6m;
2355
2356 if (ssa->ss.ss_family != AF_UNSPEC)
2357 is_final = false;
2358
2359 /*
2360 * Begin state merge transaction at socket layer.
2361 */
2362 INP_WLOCK_ASSERT(inp);
2363
2364 /*
2365 * If we were instructed only to leave a given source, do so.
2366 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2367 */
2368 if (is_final) {
2369 ip6_mfilter_remove(&imo->im6o_head, imf);
2370 im6f_leave(imf);
2371
2372 /*
2373 * Give up the multicast address record to which
2374 * the membership points.
2375 */
2376 (void)in6_leavegroup_locked(inm, imf);
2377 } else {
2378 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2379 error = EADDRNOTAVAIL;
2380 goto out_in6p_locked;
2381 }
2382 ims = im6o_match_source(imf, &ssa->sa);
2383 if (ims == NULL) {
2384 CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2385 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2386 "not ");
2387 error = EADDRNOTAVAIL;
2388 goto out_in6p_locked;
2389 }
2390 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2391 error = im6f_prune(imf, &ssa->sin6);
2392 if (error) {
2393 CTR1(KTR_MLD, "%s: merge imf state failed",
2394 __func__);
2395 goto out_in6p_locked;
2396 }
2397 }
2398
2399 /*
2400 * Begin state merge transaction at MLD layer.
2401 */
2402 if (!is_final) {
2403 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2404 IN6_MULTI_LIST_LOCK();
2405 error = in6m_merge(inm, imf);
2406 if (error) {
2407 CTR1(KTR_MLD, "%s: failed to merge inm state",
2408 __func__);
2409 IN6_MULTI_LIST_UNLOCK();
2410 im6f_rollback(imf);
2411 im6f_reap(imf);
2412 goto out_in6p_locked;
2413 }
2414
2415 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2416 error = mld_change_state(inm, 0);
2417 IN6_MULTI_LIST_UNLOCK();
2418 if (error) {
2419 CTR1(KTR_MLD, "%s: failed mld downcall",
2420 __func__);
2421 im6f_rollback(imf);
2422 im6f_reap(imf);
2423 goto out_in6p_locked;
2424 }
2425 }
2426
2427 im6f_commit(imf);
2428 im6f_reap(imf);
2429
2430 out_in6p_locked:
2431 INP_WUNLOCK(inp);
2432
2433 if (is_final && imf)
2434 ip6_mfilter_free(imf);
2435
2436 IN6_MULTI_UNLOCK();
2437 return (error);
2438 }
2439
2440 /*
2441 * Select the interface for transmitting IPv6 multicast datagrams.
2442 *
2443 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2444 * may be passed to this socket option. An address of in6addr_any or an
2445 * interface index of 0 is used to remove a previous selection.
2446 * When no interface is selected, one is chosen for every send.
2447 */
2448 static int
in6p_set_multicast_if(struct inpcb * inp,struct sockopt * sopt)2449 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2450 {
2451 struct epoch_tracker et;
2452 struct ifnet *ifp;
2453 struct ip6_moptions *imo;
2454 u_int ifindex;
2455 int error;
2456
2457 if (sopt->sopt_valsize != sizeof(u_int))
2458 return (EINVAL);
2459
2460 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2461 if (error)
2462 return (error);
2463 NET_EPOCH_ENTER(et);
2464 if (ifindex == 0)
2465 ifp = NULL;
2466 else {
2467 ifp = ifnet_byindex(ifindex);
2468 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2469 NET_EPOCH_EXIT(et);
2470 return (EADDRNOTAVAIL);
2471 }
2472 }
2473 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2474 imo = in6p_findmoptions(inp);
2475 imo->im6o_multicast_ifp = ifp;
2476 INP_WUNLOCK(inp);
2477
2478 return (0);
2479 }
2480
2481 /*
2482 * Atomically set source filters on a socket for an IPv6 multicast group.
2483 *
2484 * XXXGL: unsafely exits epoch with ifnet pointer
2485 */
2486 static int
in6p_set_source_filters(struct inpcb * inp,struct sockopt * sopt)2487 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2488 {
2489 struct __msfilterreq msfr;
2490 struct epoch_tracker et;
2491 sockunion_t *gsa;
2492 struct ifnet *ifp;
2493 struct in6_mfilter *imf;
2494 struct ip6_moptions *imo;
2495 struct in6_multi *inm;
2496 int error;
2497
2498 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2499 sizeof(struct __msfilterreq));
2500 if (error)
2501 return (error);
2502
2503 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2504 return (ENOBUFS);
2505
2506 if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2507 msfr.msfr_fmode != MCAST_INCLUDE)
2508 return (EINVAL);
2509
2510 if (msfr.msfr_group.ss_family != AF_INET6 ||
2511 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2512 return (EINVAL);
2513
2514 gsa = (sockunion_t *)&msfr.msfr_group;
2515 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2516 return (EINVAL);
2517
2518 gsa->sin6.sin6_port = 0; /* ignore port */
2519
2520 NET_EPOCH_ENTER(et);
2521 ifp = ifnet_byindex(msfr.msfr_ifindex);
2522 NET_EPOCH_EXIT(et);
2523 if (ifp == NULL)
2524 return (EADDRNOTAVAIL);
2525 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2526
2527 /*
2528 * Take the INP write lock.
2529 * Check if this socket is a member of this group.
2530 */
2531 imo = in6p_findmoptions(inp);
2532 imf = im6o_match_group(imo, ifp, &gsa->sa);
2533 if (imf == NULL) {
2534 error = EADDRNOTAVAIL;
2535 goto out_in6p_locked;
2536 }
2537 inm = imf->im6f_in6m;
2538
2539 /*
2540 * Begin state merge transaction at socket layer.
2541 */
2542 INP_WLOCK_ASSERT(inp);
2543
2544 imf->im6f_st[1] = msfr.msfr_fmode;
2545
2546 /*
2547 * Apply any new source filters, if present.
2548 * Make a copy of the user-space source vector so
2549 * that we may copy them with a single copyin. This
2550 * allows us to deal with page faults up-front.
2551 */
2552 if (msfr.msfr_nsrcs > 0) {
2553 struct in6_msource *lims;
2554 struct sockaddr_in6 *psin;
2555 struct sockaddr_storage *kss, *pkss;
2556 int i;
2557
2558 INP_WUNLOCK(inp);
2559
2560 CTR2(KTR_MLD, "%s: loading %lu source list entries",
2561 __func__, (unsigned long)msfr.msfr_nsrcs);
2562 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2563 M_TEMP, M_WAITOK);
2564 error = copyin(msfr.msfr_srcs, kss,
2565 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2566 if (error) {
2567 free(kss, M_TEMP);
2568 return (error);
2569 }
2570
2571 INP_WLOCK(inp);
2572
2573 /*
2574 * Mark all source filters as UNDEFINED at t1.
2575 * Restore new group filter mode, as im6f_leave()
2576 * will set it to INCLUDE.
2577 */
2578 im6f_leave(imf);
2579 imf->im6f_st[1] = msfr.msfr_fmode;
2580
2581 /*
2582 * Update socket layer filters at t1, lazy-allocating
2583 * new entries. This saves a bunch of memory at the
2584 * cost of one RB_FIND() per source entry; duplicate
2585 * entries in the msfr_nsrcs vector are ignored.
2586 * If we encounter an error, rollback transaction.
2587 *
2588 * XXX This too could be replaced with a set-symmetric
2589 * difference like loop to avoid walking from root
2590 * every time, as the key space is common.
2591 */
2592 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2593 psin = (struct sockaddr_in6 *)pkss;
2594 if (psin->sin6_family != AF_INET6) {
2595 error = EAFNOSUPPORT;
2596 break;
2597 }
2598 if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2599 error = EINVAL;
2600 break;
2601 }
2602 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2603 error = EINVAL;
2604 break;
2605 }
2606 /*
2607 * TODO: Validate embedded scope ID in source
2608 * list entry against passed-in ifp, if and only
2609 * if source list filter entry is iface or node local.
2610 */
2611 in6_clearscope(&psin->sin6_addr);
2612 error = im6f_get_source(imf, psin, &lims);
2613 if (error)
2614 break;
2615 lims->im6sl_st[1] = imf->im6f_st[1];
2616 }
2617 free(kss, M_TEMP);
2618 }
2619
2620 if (error)
2621 goto out_im6f_rollback;
2622
2623 INP_WLOCK_ASSERT(inp);
2624 IN6_MULTI_LIST_LOCK();
2625
2626 /*
2627 * Begin state merge transaction at MLD layer.
2628 */
2629 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2630 error = in6m_merge(inm, imf);
2631 if (error)
2632 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2633 else {
2634 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2635 error = mld_change_state(inm, 0);
2636 if (error)
2637 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2638 }
2639
2640 IN6_MULTI_LIST_UNLOCK();
2641
2642 out_im6f_rollback:
2643 if (error)
2644 im6f_rollback(imf);
2645 else
2646 im6f_commit(imf);
2647
2648 im6f_reap(imf);
2649
2650 out_in6p_locked:
2651 INP_WUNLOCK(inp);
2652 return (error);
2653 }
2654
2655 /*
2656 * Set the IP multicast options in response to user setsockopt().
2657 *
2658 * Many of the socket options handled in this function duplicate the
2659 * functionality of socket options in the regular unicast API. However,
2660 * it is not possible to merge the duplicate code, because the idempotence
2661 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2662 * the effects of these options must be treated as separate and distinct.
2663 *
2664 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2665 */
2666 int
ip6_setmoptions(struct inpcb * inp,struct sockopt * sopt)2667 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2668 {
2669 struct ip6_moptions *im6o;
2670 int error;
2671
2672 error = 0;
2673
2674 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
2675 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2676 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)
2677 return (EOPNOTSUPP);
2678
2679 switch (sopt->sopt_name) {
2680 case IPV6_MULTICAST_IF:
2681 error = in6p_set_multicast_if(inp, sopt);
2682 break;
2683
2684 case IPV6_MULTICAST_HOPS: {
2685 int hlim;
2686
2687 if (sopt->sopt_valsize != sizeof(int)) {
2688 error = EINVAL;
2689 break;
2690 }
2691 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2692 if (error)
2693 break;
2694 if (hlim < -1 || hlim > 255) {
2695 error = EINVAL;
2696 break;
2697 } else if (hlim == -1) {
2698 hlim = V_ip6_defmcasthlim;
2699 }
2700 im6o = in6p_findmoptions(inp);
2701 im6o->im6o_multicast_hlim = hlim;
2702 INP_WUNLOCK(inp);
2703 break;
2704 }
2705
2706 case IPV6_MULTICAST_LOOP: {
2707 u_int loop;
2708
2709 /*
2710 * Set the loopback flag for outgoing multicast packets.
2711 * Must be zero or one.
2712 */
2713 if (sopt->sopt_valsize != sizeof(u_int)) {
2714 error = EINVAL;
2715 break;
2716 }
2717 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2718 if (error)
2719 break;
2720 if (loop > 1) {
2721 error = EINVAL;
2722 break;
2723 }
2724 im6o = in6p_findmoptions(inp);
2725 im6o->im6o_multicast_loop = loop;
2726 INP_WUNLOCK(inp);
2727 break;
2728 }
2729
2730 case IPV6_JOIN_GROUP:
2731 case MCAST_JOIN_GROUP:
2732 case MCAST_JOIN_SOURCE_GROUP:
2733 error = in6p_join_group(inp, sopt);
2734 break;
2735
2736 case IPV6_LEAVE_GROUP:
2737 case MCAST_LEAVE_GROUP:
2738 case MCAST_LEAVE_SOURCE_GROUP:
2739 error = in6p_leave_group(inp, sopt);
2740 break;
2741
2742 case MCAST_BLOCK_SOURCE:
2743 case MCAST_UNBLOCK_SOURCE:
2744 error = in6p_block_unblock_source(inp, sopt);
2745 break;
2746
2747 case IPV6_MSFILTER:
2748 error = in6p_set_source_filters(inp, sopt);
2749 break;
2750
2751 default:
2752 error = EOPNOTSUPP;
2753 break;
2754 }
2755
2756 INP_UNLOCK_ASSERT(inp);
2757
2758 return (error);
2759 }
2760
2761 /*
2762 * Expose MLD's multicast filter mode and source list(s) to userland,
2763 * keyed by (ifindex, group).
2764 * The filter mode is written out as a uint32_t, followed by
2765 * 0..n of struct in6_addr.
2766 * For use by ifmcstat(8).
2767 * SMPng: NOTE: unlocked read of ifindex space.
2768 */
2769 static int
sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)2770 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2771 {
2772 struct in6_addr mcaddr;
2773 struct in6_addr src;
2774 struct epoch_tracker et;
2775 struct ifnet *ifp;
2776 struct ifmultiaddr *ifma;
2777 struct in6_multi *inm;
2778 struct ip6_msource *ims;
2779 int *name;
2780 int retval;
2781 u_int namelen;
2782 uint32_t fmode, ifindex;
2783 #ifdef KTR
2784 char ip6tbuf[INET6_ADDRSTRLEN];
2785 #endif
2786
2787 name = (int *)arg1;
2788 namelen = arg2;
2789
2790 if (req->newptr != NULL)
2791 return (EPERM);
2792
2793 /* int: ifindex + 4 * 32 bits of IPv6 address */
2794 if (namelen != 5)
2795 return (EINVAL);
2796
2797 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2798 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2799 CTR2(KTR_MLD, "%s: group %s is not multicast",
2800 __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2801 return (EINVAL);
2802 }
2803
2804 ifindex = name[0];
2805 NET_EPOCH_ENTER(et);
2806 ifp = ifnet_byindex(ifindex);
2807 if (ifp == NULL) {
2808 NET_EPOCH_EXIT(et);
2809 CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2810 __func__, ifindex);
2811 return (ENOENT);
2812 }
2813 /*
2814 * Internal MLD lookups require that scope/zone ID is set.
2815 */
2816 (void)in6_setscope(&mcaddr, ifp, NULL);
2817
2818 retval = sysctl_wire_old_buffer(req,
2819 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2820 if (retval) {
2821 NET_EPOCH_EXIT(et);
2822 return (retval);
2823 }
2824
2825 IN6_MULTI_LOCK();
2826 IN6_MULTI_LIST_LOCK();
2827 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2828 inm = in6m_ifmultiaddr_get_inm(ifma);
2829 if (inm == NULL)
2830 continue;
2831 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2832 continue;
2833 fmode = inm->in6m_st[1].iss_fmode;
2834 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2835 if (retval != 0)
2836 break;
2837 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2838 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2839 /*
2840 * Only copy-out sources which are in-mode.
2841 */
2842 if (fmode != im6s_get_mode(inm, ims, 1)) {
2843 CTR1(KTR_MLD, "%s: skip non-in-mode",
2844 __func__);
2845 continue;
2846 }
2847 src = ims->im6s_addr;
2848 retval = SYSCTL_OUT(req, &src,
2849 sizeof(struct in6_addr));
2850 if (retval != 0)
2851 break;
2852 }
2853 }
2854 IN6_MULTI_LIST_UNLOCK();
2855 IN6_MULTI_UNLOCK();
2856 NET_EPOCH_EXIT(et);
2857
2858 return (retval);
2859 }
2860
2861 #ifdef KTR
2862
2863 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2864
2865 static const char *
in6m_mode_str(const int mode)2866 in6m_mode_str(const int mode)
2867 {
2868
2869 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2870 return (in6m_modestrs[mode]);
2871 return ("??");
2872 }
2873
2874 static const char *in6m_statestrs[] = {
2875 "not-member",
2876 "silent",
2877 "reporting",
2878 "idle",
2879 "lazy",
2880 "sleeping",
2881 "awakening",
2882 "query-pending",
2883 "sg-query-pending",
2884 "leaving"
2885 };
2886 _Static_assert(nitems(in6m_statestrs) ==
2887 MLD_LEAVING_MEMBER - MLD_NOT_MEMBER + 1, "Missing MLD group state");
2888
2889 static const char *
in6m_state_str(const int state)2890 in6m_state_str(const int state)
2891 {
2892
2893 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2894 return (in6m_statestrs[state]);
2895 return ("??");
2896 }
2897
2898 /*
2899 * Dump an in6_multi structure to the console.
2900 */
2901 void
in6m_print(const struct in6_multi * inm)2902 in6m_print(const struct in6_multi *inm)
2903 {
2904 int t;
2905 char ip6tbuf[INET6_ADDRSTRLEN];
2906
2907 if ((ktr_mask & KTR_MLD) == 0)
2908 return;
2909
2910 printf("%s: --- begin in6m %p ---\n", __func__, inm);
2911 printf("addr %s ifp %p(%s) ifma %p\n",
2912 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2913 inm->in6m_ifp,
2914 if_name(inm->in6m_ifp),
2915 inm->in6m_ifma);
2916 printf("timer %u state %s refcount %u scq.len %u\n",
2917 inm->in6m_timer,
2918 in6m_state_str(inm->in6m_state),
2919 inm->in6m_refcount,
2920 mbufq_len(&inm->in6m_scq));
2921 printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2922 inm->in6m_mli,
2923 inm->in6m_nsrc,
2924 inm->in6m_sctimer,
2925 inm->in6m_scrv);
2926 for (t = 0; t < 2; t++) {
2927 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2928 in6m_mode_str(inm->in6m_st[t].iss_fmode),
2929 inm->in6m_st[t].iss_asm,
2930 inm->in6m_st[t].iss_ex,
2931 inm->in6m_st[t].iss_in,
2932 inm->in6m_st[t].iss_rec);
2933 }
2934 printf("%s: --- end in6m %p ---\n", __func__, inm);
2935 }
2936
2937 #else /* !KTR */
2938
2939 void
in6m_print(const struct in6_multi * inm)2940 in6m_print(const struct in6_multi *inm)
2941 {
2942
2943 }
2944
2945 #endif /* KTR */
2946