1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 2007-2009 Bruce Simpson.
5 * Copyright (c) 2005 Robert N. M. Watson.
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote
17 * products derived from this software without specific prior written
18 * permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 */
32
33 /*
34 * IPv4 multicast socket, group, and socket option processing module.
35 */
36
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/lock.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/protosw.h>
47 #include <sys/sysctl.h>
48 #include <sys/ktr.h>
49 #include <sys/taskqueue.h>
50 #include <sys/tree.h>
51
52 #include <net/if.h>
53 #include <net/if_var.h>
54 #include <net/if_dl.h>
55 #include <net/route.h>
56 #include <net/route/nhop.h>
57 #include <net/vnet.h>
58
59 #include <net/ethernet.h>
60
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/in_fib.h>
64 #include <netinet/in_pcb.h>
65 #include <netinet/in_var.h>
66 #include <net/if_private.h>
67 #include <netinet/ip_var.h>
68 #include <netinet/igmp_var.h>
69
70 #ifndef KTR_IGMPV3
71 #define KTR_IGMPV3 KTR_INET
72 #endif
73
74 #ifndef __SOCKUNION_DECLARED
75 union sockunion {
76 struct sockaddr_storage ss;
77 struct sockaddr sa;
78 struct sockaddr_dl sdl;
79 struct sockaddr_in sin;
80 };
81 typedef union sockunion sockunion_t;
82 #define __SOCKUNION_DECLARED
83 #endif /* __SOCKUNION_DECLARED */
84
85 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
86 "IPv4 multicast PCB-layer source filter");
87 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
88 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
89 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
90 "IPv4 multicast IGMP-layer source filter");
91
92 /*
93 * Locking:
94 *
95 * - Lock order is: IN_MULTI_LOCK, INP_WLOCK, IN_MULTI_LIST_LOCK, IGMP_LOCK,
96 * IF_ADDR_LOCK.
97 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
98 * it can be taken by code in net/if.c also.
99 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
100 *
101 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
102 * any need for in_multi itself to be virtualized -- it is bound to an ifp
103 * anyway no matter what happens.
104 */
105 struct mtx in_multi_list_mtx;
106 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
107
108 struct mtx in_multi_free_mtx;
109 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
110
111 struct sx in_multi_sx;
112 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
113
114 /*
115 * Functions with non-static linkage defined in this file should be
116 * declared in in_var.h:
117 * imo_multi_filter()
118 * in_joingroup()
119 * in_joingroup_locked()
120 * in_leavegroup()
121 * in_leavegroup_locked()
122 * and ip_var.h:
123 * inp_freemoptions()
124 * inp_getmoptions()
125 * inp_setmoptions()
126 */
127 static void imf_commit(struct in_mfilter *);
128 static int imf_get_source(struct in_mfilter *imf,
129 const struct sockaddr_in *psin,
130 struct in_msource **);
131 static struct in_msource *
132 imf_graft(struct in_mfilter *, const uint8_t,
133 const struct sockaddr_in *);
134 static void imf_leave(struct in_mfilter *);
135 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
136 static void imf_purge(struct in_mfilter *);
137 static void imf_rollback(struct in_mfilter *);
138 static void imf_reap(struct in_mfilter *);
139 static struct in_mfilter *
140 imo_match_group(const struct ip_moptions *,
141 const struct ifnet *, const struct sockaddr *);
142 static struct in_msource *
143 imo_match_source(struct in_mfilter *, const struct sockaddr *);
144 static void ims_merge(struct ip_msource *ims,
145 const struct in_msource *lims, const int rollback);
146 static int in_getmulti(struct ifnet *, const struct in_addr *,
147 struct in_multi **);
148 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
149 const int noalloc, struct ip_msource **pims);
150 #ifdef KTR
151 static int inm_is_ifp_detached(const struct in_multi *);
152 #endif
153 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
154 static void inm_purge(struct in_multi *);
155 static void inm_reap(struct in_multi *);
156 static void inm_release(struct in_multi *);
157 static struct ip_moptions *
158 inp_findmoptions(struct inpcb *);
159 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
160 static int inp_join_group(struct inpcb *, struct sockopt *);
161 static int inp_leave_group(struct inpcb *, struct sockopt *);
162 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
163 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
164 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
165 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
166
167 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast,
168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
169 "IPv4 multicast");
170
171 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
172 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
173 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
174 "Max source filters per group");
175
176 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
177 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
178 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
179 "Max source filters per socket");
180
181 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
182 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
183 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
184
185 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
186 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
187 "Per-interface stack-wide source filters");
188
189 #ifdef KTR
190 /*
191 * Inline function which wraps assertions for a valid ifp.
192 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
193 * is detached.
194 */
195 static int __inline
inm_is_ifp_detached(const struct in_multi * inm)196 inm_is_ifp_detached(const struct in_multi *inm)
197 {
198 struct ifnet *ifp;
199
200 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
201 ifp = inm->inm_ifma->ifma_ifp;
202 if (ifp != NULL) {
203 /*
204 * Sanity check that netinet's notion of ifp is the
205 * same as net's.
206 */
207 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
208 }
209
210 return (ifp == NULL);
211 }
212 #endif
213
214 /*
215 * Interface detach can happen in a taskqueue thread context, so we must use a
216 * dedicated thread to avoid deadlocks when draining inm_release tasks.
217 */
218 TASKQUEUE_DEFINE_THREAD(inm_free);
219 static struct in_multi_head inm_free_list = SLIST_HEAD_INITIALIZER();
220 static void inm_release_task(void *arg __unused, int pending __unused);
221 static struct task inm_free_task = TASK_INITIALIZER(0, inm_release_task, NULL);
222
223 void
inm_release_wait(void * arg __unused)224 inm_release_wait(void *arg __unused)
225 {
226
227 /*
228 * Make sure all pending multicast addresses are freed before
229 * the VNET or network device is destroyed:
230 */
231 taskqueue_drain(taskqueue_inm_free, &inm_free_task);
232 }
233 #ifdef VIMAGE
234 /* XXX-BZ FIXME, see D24914. */
235 VNET_SYSUNINIT(inm_release_wait, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST, inm_release_wait, NULL);
236 #endif
237
238 void
inm_release_list_deferred(struct in_multi_head * inmh)239 inm_release_list_deferred(struct in_multi_head *inmh)
240 {
241
242 if (SLIST_EMPTY(inmh))
243 return;
244 mtx_lock(&in_multi_free_mtx);
245 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
246 mtx_unlock(&in_multi_free_mtx);
247 taskqueue_enqueue(taskqueue_inm_free, &inm_free_task);
248 }
249
250 void
inm_disconnect(struct in_multi * inm)251 inm_disconnect(struct in_multi *inm)
252 {
253 struct ifnet *ifp;
254 struct ifmultiaddr *ifma, *ll_ifma;
255
256 ifp = inm->inm_ifp;
257 IF_ADDR_WLOCK_ASSERT(ifp);
258 ifma = inm->inm_ifma;
259
260 if_ref(ifp);
261 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
262 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
263 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
264 }
265 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
266 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
267 MPASS(ifma != ll_ifma);
268 ifma->ifma_llifma = NULL;
269 MPASS(ll_ifma->ifma_llifma == NULL);
270 MPASS(ll_ifma->ifma_ifp == ifp);
271 if (--ll_ifma->ifma_refcount == 0) {
272 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
273 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
274 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
275 }
276 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
277 if_freemulti(ll_ifma);
278 }
279 }
280 }
281
282 void
inm_release_deferred(struct in_multi * inm)283 inm_release_deferred(struct in_multi *inm)
284 {
285 struct in_multi_head tmp;
286
287 IN_MULTI_LIST_LOCK_ASSERT();
288 MPASS(inm->inm_refcount > 0);
289 if (--inm->inm_refcount == 0) {
290 SLIST_INIT(&tmp);
291 inm_disconnect(inm);
292 inm->inm_ifma->ifma_protospec = NULL;
293 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
294 inm_release_list_deferred(&tmp);
295 }
296 }
297
298 static void
inm_release_task(void * arg __unused,int pending __unused)299 inm_release_task(void *arg __unused, int pending __unused)
300 {
301 struct in_multi_head inm_free_tmp;
302 struct in_multi *inm, *tinm;
303
304 SLIST_INIT(&inm_free_tmp);
305 mtx_lock(&in_multi_free_mtx);
306 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
307 mtx_unlock(&in_multi_free_mtx);
308 IN_MULTI_LOCK();
309 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
310 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
311 MPASS(inm);
312 inm_release(inm);
313 }
314 IN_MULTI_UNLOCK();
315 }
316
317 /*
318 * Initialize an in_mfilter structure to a known state at t0, t1
319 * with an empty source filter list.
320 */
321 static __inline void
imf_init(struct in_mfilter * imf,const int st0,const int st1)322 imf_init(struct in_mfilter *imf, const int st0, const int st1)
323 {
324 memset(imf, 0, sizeof(struct in_mfilter));
325 RB_INIT(&imf->imf_sources);
326 imf->imf_st[0] = st0;
327 imf->imf_st[1] = st1;
328 }
329
330 struct in_mfilter *
ip_mfilter_alloc(const int mflags,const int st0,const int st1)331 ip_mfilter_alloc(const int mflags, const int st0, const int st1)
332 {
333 struct in_mfilter *imf;
334
335 imf = malloc(sizeof(*imf), M_INMFILTER, mflags);
336 if (imf != NULL)
337 imf_init(imf, st0, st1);
338
339 return (imf);
340 }
341
342 void
ip_mfilter_free(struct in_mfilter * imf)343 ip_mfilter_free(struct in_mfilter *imf)
344 {
345
346 imf_purge(imf);
347 free(imf, M_INMFILTER);
348 }
349
350 /*
351 * Function for looking up an in_multi record for an IPv4 multicast address
352 * on a given interface. ifp must be valid. If no record found, return NULL.
353 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
354 */
355 struct in_multi *
inm_lookup_locked(struct ifnet * ifp,const struct in_addr ina)356 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
357 {
358 struct ifmultiaddr *ifma;
359 struct in_multi *inm;
360
361 IN_MULTI_LIST_LOCK_ASSERT();
362 IF_ADDR_LOCK_ASSERT(ifp);
363
364 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
365 inm = inm_ifmultiaddr_get_inm(ifma);
366 if (inm == NULL)
367 continue;
368 if (inm->inm_addr.s_addr == ina.s_addr)
369 return (inm);
370 }
371 return (NULL);
372 }
373
374 /*
375 * Wrapper for inm_lookup_locked().
376 * The IF_ADDR_LOCK will be taken on ifp and released on return.
377 */
378 struct in_multi *
inm_lookup(struct ifnet * ifp,const struct in_addr ina)379 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
380 {
381 struct epoch_tracker et;
382 struct in_multi *inm;
383
384 IN_MULTI_LIST_LOCK_ASSERT();
385 NET_EPOCH_ENTER(et);
386
387 inm = inm_lookup_locked(ifp, ina);
388 NET_EPOCH_EXIT(et);
389
390 return (inm);
391 }
392
393 /*
394 * Find an IPv4 multicast group entry for this ip_moptions instance
395 * which matches the specified group, and optionally an interface.
396 * Return its index into the array, or -1 if not found.
397 */
398 static struct in_mfilter *
imo_match_group(const struct ip_moptions * imo,const struct ifnet * ifp,const struct sockaddr * group)399 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
400 const struct sockaddr *group)
401 {
402 const struct sockaddr_in *gsin;
403 struct in_mfilter *imf;
404 struct in_multi *inm;
405
406 gsin = (const struct sockaddr_in *)group;
407
408 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
409 inm = imf->imf_inm;
410 if (inm == NULL)
411 continue;
412 if ((ifp == NULL || (inm->inm_ifp == ifp)) &&
413 in_hosteq(inm->inm_addr, gsin->sin_addr)) {
414 break;
415 }
416 }
417 return (imf);
418 }
419
420 /*
421 * Find an IPv4 multicast source entry for this imo which matches
422 * the given group index for this socket, and source address.
423 *
424 * NOTE: This does not check if the entry is in-mode, merely if
425 * it exists, which may not be the desired behaviour.
426 */
427 static struct in_msource *
imo_match_source(struct in_mfilter * imf,const struct sockaddr * src)428 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src)
429 {
430 struct ip_msource find;
431 struct ip_msource *ims;
432 const sockunion_t *psa;
433
434 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
435
436 /* Source trees are keyed in host byte order. */
437 psa = (const sockunion_t *)src;
438 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
439 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
440
441 return ((struct in_msource *)ims);
442 }
443
444 /*
445 * Perform filtering for multicast datagrams on a socket by group and source.
446 *
447 * Returns 0 if a datagram should be allowed through, or various error codes
448 * if the socket was not a member of the group, or the source was muted, etc.
449 */
450 int
imo_multi_filter(const struct ip_moptions * imo,const struct ifnet * ifp,const struct sockaddr * group,const struct sockaddr * src)451 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
452 const struct sockaddr *group, const struct sockaddr *src)
453 {
454 struct in_mfilter *imf;
455 struct in_msource *ims;
456 int mode;
457
458 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
459
460 imf = imo_match_group(imo, ifp, group);
461 if (imf == NULL)
462 return (MCAST_NOTGMEMBER);
463
464 /*
465 * Check if the source was included in an (S,G) join.
466 * Allow reception on exclusive memberships by default,
467 * reject reception on inclusive memberships by default.
468 * Exclude source only if an in-mode exclude filter exists.
469 * Include source only if an in-mode include filter exists.
470 * NOTE: We are comparing group state here at IGMP t1 (now)
471 * with socket-layer t0 (since last downcall).
472 */
473 mode = imf->imf_st[1];
474 ims = imo_match_source(imf, src);
475
476 if ((ims == NULL && mode == MCAST_INCLUDE) ||
477 (ims != NULL && ims->imsl_st[0] == MCAST_EXCLUDE))
478 return (MCAST_NOTSMEMBER);
479
480 return (MCAST_PASS);
481 }
482
483 /*
484 * Find and return a reference to an in_multi record for (ifp, group),
485 * and bump its reference count.
486 * If one does not exist, try to allocate it, and update link-layer multicast
487 * filters on ifp to listen for group.
488 * Assumes the IN_MULTI lock is held across the call.
489 * Return 0 if successful, otherwise return an appropriate error code.
490 */
491 static int
in_getmulti(struct ifnet * ifp,const struct in_addr * group,struct in_multi ** pinm)492 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
493 struct in_multi **pinm)
494 {
495 struct sockaddr_in gsin;
496 struct ifmultiaddr *ifma;
497 struct in_ifinfo *ii;
498 struct in_multi *inm;
499 int error;
500
501 IN_MULTI_LOCK_ASSERT();
502
503 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
504 IN_MULTI_LIST_LOCK();
505 inm = inm_lookup(ifp, *group);
506 if (inm != NULL) {
507 /*
508 * If we already joined this group, just bump the
509 * refcount and return it.
510 */
511 KASSERT(inm->inm_refcount >= 1,
512 ("%s: bad refcount %d", __func__, inm->inm_refcount));
513 inm_acquire_locked(inm);
514 *pinm = inm;
515 }
516 IN_MULTI_LIST_UNLOCK();
517 if (inm != NULL)
518 return (0);
519
520 memset(&gsin, 0, sizeof(gsin));
521 gsin.sin_family = AF_INET;
522 gsin.sin_len = sizeof(struct sockaddr_in);
523 gsin.sin_addr = *group;
524
525 /*
526 * Check if a link-layer group is already associated
527 * with this network-layer group on the given ifnet.
528 */
529 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
530 if (error != 0)
531 return (error);
532
533 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
534 IN_MULTI_LIST_LOCK();
535 IF_ADDR_WLOCK(ifp);
536
537 /*
538 * If something other than netinet is occupying the link-layer
539 * group, print a meaningful error message and back out of
540 * the allocation.
541 * Otherwise, bump the refcount on the existing network-layer
542 * group association and return it.
543 */
544 if (ifma->ifma_protospec != NULL) {
545 inm = (struct in_multi *)ifma->ifma_protospec;
546 #ifdef INVARIANTS
547 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
548 __func__));
549 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
550 ("%s: ifma not AF_INET", __func__));
551 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
552 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
553 !in_hosteq(inm->inm_addr, *group)) {
554 char addrbuf[INET_ADDRSTRLEN];
555
556 panic("%s: ifma %p is inconsistent with %p (%s)",
557 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
558 }
559 #endif
560 inm_acquire_locked(inm);
561 *pinm = inm;
562 goto out_locked;
563 }
564
565 IF_ADDR_WLOCK_ASSERT(ifp);
566
567 /*
568 * A new in_multi record is needed; allocate and initialize it.
569 * We DO NOT perform an IGMP join as the in_ layer may need to
570 * push an initial source list down to IGMP to support SSM.
571 *
572 * The initial source filter state is INCLUDE, {} as per the RFC.
573 */
574 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
575 if (inm == NULL) {
576 IF_ADDR_WUNLOCK(ifp);
577 IN_MULTI_LIST_UNLOCK();
578 if_delmulti_ifma(ifma);
579 return (ENOMEM);
580 }
581 inm->inm_addr = *group;
582 inm->inm_ifp = ifp;
583 inm->inm_igi = ii->ii_igmp;
584 inm->inm_ifma = ifma;
585 inm->inm_refcount = 1;
586 inm->inm_state = IGMP_NOT_MEMBER;
587 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
588 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
589 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
590 RB_INIT(&inm->inm_srcs);
591
592 ifma->ifma_protospec = inm;
593
594 *pinm = inm;
595 out_locked:
596 IF_ADDR_WUNLOCK(ifp);
597 IN_MULTI_LIST_UNLOCK();
598 return (0);
599 }
600
601 /*
602 * Drop a reference to an in_multi record.
603 *
604 * If the refcount drops to 0, free the in_multi record and
605 * delete the underlying link-layer membership.
606 */
607 static void
inm_release(struct in_multi * inm)608 inm_release(struct in_multi *inm)
609 {
610 struct ifmultiaddr *ifma;
611 struct ifnet *ifp;
612
613 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
614 MPASS(inm->inm_refcount == 0);
615 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
616
617 ifma = inm->inm_ifma;
618 ifp = inm->inm_ifp;
619
620 /* XXX this access is not covered by IF_ADDR_LOCK */
621 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
622 if (ifp != NULL) {
623 CURVNET_SET(ifp->if_vnet);
624 inm_purge(inm);
625 free(inm, M_IPMADDR);
626 if_delmulti_ifma_flags(ifma, 1);
627 CURVNET_RESTORE();
628 if_rele(ifp);
629 } else {
630 inm_purge(inm);
631 free(inm, M_IPMADDR);
632 if_delmulti_ifma_flags(ifma, 1);
633 }
634 }
635
636 /*
637 * Clear recorded source entries for a group.
638 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
639 * FIXME: Should reap.
640 */
641 void
inm_clear_recorded(struct in_multi * inm)642 inm_clear_recorded(struct in_multi *inm)
643 {
644 struct ip_msource *ims;
645
646 IN_MULTI_LIST_LOCK_ASSERT();
647
648 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
649 if (ims->ims_stp) {
650 ims->ims_stp = 0;
651 --inm->inm_st[1].iss_rec;
652 }
653 }
654 KASSERT(inm->inm_st[1].iss_rec == 0,
655 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
656 }
657
658 /*
659 * Record a source as pending for a Source-Group IGMPv3 query.
660 * This lives here as it modifies the shared tree.
661 *
662 * inm is the group descriptor.
663 * naddr is the address of the source to record in network-byte order.
664 *
665 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
666 * lazy-allocate a source node in response to an SG query.
667 * Otherwise, no allocation is performed. This saves some memory
668 * with the trade-off that the source will not be reported to the
669 * router if joined in the window between the query response and
670 * the group actually being joined on the local host.
671 *
672 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
673 * This turns off the allocation of a recorded source entry if
674 * the group has not been joined.
675 *
676 * Return 0 if the source didn't exist or was already marked as recorded.
677 * Return 1 if the source was marked as recorded by this function.
678 * Return <0 if any error occurred (negated errno code).
679 */
680 int
inm_record_source(struct in_multi * inm,const in_addr_t naddr)681 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
682 {
683 struct ip_msource find;
684 struct ip_msource *ims, *nims;
685
686 IN_MULTI_LIST_LOCK_ASSERT();
687
688 find.ims_haddr = ntohl(naddr);
689 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
690 if (ims && ims->ims_stp)
691 return (0);
692 if (ims == NULL) {
693 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
694 return (-ENOSPC);
695 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
696 M_NOWAIT | M_ZERO);
697 if (nims == NULL)
698 return (-ENOMEM);
699 nims->ims_haddr = find.ims_haddr;
700 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
701 ++inm->inm_nsrc;
702 ims = nims;
703 }
704
705 /*
706 * Mark the source as recorded and update the recorded
707 * source count.
708 */
709 ++ims->ims_stp;
710 ++inm->inm_st[1].iss_rec;
711
712 return (1);
713 }
714
715 /*
716 * Return a pointer to an in_msource owned by an in_mfilter,
717 * given its source address.
718 * Lazy-allocate if needed. If this is a new entry its filter state is
719 * undefined at t0.
720 *
721 * imf is the filter set being modified.
722 * haddr is the source address in *host* byte-order.
723 *
724 * SMPng: May be called with locks held; malloc must not block.
725 */
726 static int
imf_get_source(struct in_mfilter * imf,const struct sockaddr_in * psin,struct in_msource ** plims)727 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
728 struct in_msource **plims)
729 {
730 struct ip_msource find;
731 struct ip_msource *ims, *nims;
732 struct in_msource *lims;
733 int error;
734
735 error = 0;
736 ims = NULL;
737 lims = NULL;
738
739 /* key is host byte order */
740 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
741 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
742 lims = (struct in_msource *)ims;
743 if (lims == NULL) {
744 if (imf->imf_nsrc == in_mcast_maxsocksrc)
745 return (ENOSPC);
746 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
747 M_NOWAIT | M_ZERO);
748 if (nims == NULL)
749 return (ENOMEM);
750 lims = (struct in_msource *)nims;
751 lims->ims_haddr = find.ims_haddr;
752 lims->imsl_st[0] = MCAST_UNDEFINED;
753 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
754 ++imf->imf_nsrc;
755 }
756
757 *plims = lims;
758
759 return (error);
760 }
761
762 /*
763 * Graft a source entry into an existing socket-layer filter set,
764 * maintaining any required invariants and checking allocations.
765 *
766 * The source is marked as being in the new filter mode at t1.
767 *
768 * Return the pointer to the new node, otherwise return NULL.
769 */
770 static struct in_msource *
imf_graft(struct in_mfilter * imf,const uint8_t st1,const struct sockaddr_in * psin)771 imf_graft(struct in_mfilter *imf, const uint8_t st1,
772 const struct sockaddr_in *psin)
773 {
774 struct ip_msource *nims;
775 struct in_msource *lims;
776
777 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
778 M_NOWAIT | M_ZERO);
779 if (nims == NULL)
780 return (NULL);
781 lims = (struct in_msource *)nims;
782 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
783 lims->imsl_st[0] = MCAST_UNDEFINED;
784 lims->imsl_st[1] = st1;
785 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
786 ++imf->imf_nsrc;
787
788 return (lims);
789 }
790
791 /*
792 * Prune a source entry from an existing socket-layer filter set,
793 * maintaining any required invariants and checking allocations.
794 *
795 * The source is marked as being left at t1, it is not freed.
796 *
797 * Return 0 if no error occurred, otherwise return an errno value.
798 */
799 static int
imf_prune(struct in_mfilter * imf,const struct sockaddr_in * psin)800 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
801 {
802 struct ip_msource find;
803 struct ip_msource *ims;
804 struct in_msource *lims;
805
806 /* key is host byte order */
807 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
808 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
809 if (ims == NULL)
810 return (ENOENT);
811 lims = (struct in_msource *)ims;
812 lims->imsl_st[1] = MCAST_UNDEFINED;
813 return (0);
814 }
815
816 /*
817 * Revert socket-layer filter set deltas at t1 to t0 state.
818 */
819 static void
imf_rollback(struct in_mfilter * imf)820 imf_rollback(struct in_mfilter *imf)
821 {
822 struct ip_msource *ims, *tims;
823 struct in_msource *lims;
824
825 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
826 lims = (struct in_msource *)ims;
827 if (lims->imsl_st[0] == lims->imsl_st[1]) {
828 /* no change at t1 */
829 continue;
830 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
831 /* revert change to existing source at t1 */
832 lims->imsl_st[1] = lims->imsl_st[0];
833 } else {
834 /* revert source added t1 */
835 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
836 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
837 free(ims, M_INMFILTER);
838 imf->imf_nsrc--;
839 }
840 }
841 imf->imf_st[1] = imf->imf_st[0];
842 }
843
844 /*
845 * Mark socket-layer filter set as INCLUDE {} at t1.
846 */
847 static void
imf_leave(struct in_mfilter * imf)848 imf_leave(struct in_mfilter *imf)
849 {
850 struct ip_msource *ims;
851 struct in_msource *lims;
852
853 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
854 lims = (struct in_msource *)ims;
855 lims->imsl_st[1] = MCAST_UNDEFINED;
856 }
857 imf->imf_st[1] = MCAST_INCLUDE;
858 }
859
860 /*
861 * Mark socket-layer filter set deltas as committed.
862 */
863 static void
imf_commit(struct in_mfilter * imf)864 imf_commit(struct in_mfilter *imf)
865 {
866 struct ip_msource *ims;
867 struct in_msource *lims;
868
869 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
870 lims = (struct in_msource *)ims;
871 lims->imsl_st[0] = lims->imsl_st[1];
872 }
873 imf->imf_st[0] = imf->imf_st[1];
874 }
875
876 /*
877 * Reap unreferenced sources from socket-layer filter set.
878 */
879 static void
imf_reap(struct in_mfilter * imf)880 imf_reap(struct in_mfilter *imf)
881 {
882 struct ip_msource *ims, *tims;
883 struct in_msource *lims;
884
885 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
886 lims = (struct in_msource *)ims;
887 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
888 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
889 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
890 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
891 free(ims, M_INMFILTER);
892 imf->imf_nsrc--;
893 }
894 }
895 }
896
897 /*
898 * Purge socket-layer filter set.
899 */
900 static void
imf_purge(struct in_mfilter * imf)901 imf_purge(struct in_mfilter *imf)
902 {
903 struct ip_msource *ims, *tims;
904
905 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
906 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
907 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
908 free(ims, M_INMFILTER);
909 imf->imf_nsrc--;
910 }
911 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
912 KASSERT(RB_EMPTY(&imf->imf_sources),
913 ("%s: imf_sources not empty", __func__));
914 }
915
916 /*
917 * Look up a source filter entry for a multicast group.
918 *
919 * inm is the group descriptor to work with.
920 * haddr is the host-byte-order IPv4 address to look up.
921 * noalloc may be non-zero to suppress allocation of sources.
922 * *pims will be set to the address of the retrieved or allocated source.
923 *
924 * SMPng: NOTE: may be called with locks held.
925 * Return 0 if successful, otherwise return a non-zero error code.
926 */
927 static int
inm_get_source(struct in_multi * inm,const in_addr_t haddr,const int noalloc,struct ip_msource ** pims)928 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
929 const int noalloc, struct ip_msource **pims)
930 {
931 struct ip_msource find;
932 struct ip_msource *ims, *nims;
933
934 find.ims_haddr = haddr;
935 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
936 if (ims == NULL && !noalloc) {
937 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
938 return (ENOSPC);
939 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
940 M_NOWAIT | M_ZERO);
941 if (nims == NULL)
942 return (ENOMEM);
943 nims->ims_haddr = haddr;
944 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
945 ++inm->inm_nsrc;
946 ims = nims;
947 #ifdef KTR
948 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
949 haddr, ims);
950 #endif
951 }
952
953 *pims = ims;
954 return (0);
955 }
956
957 /*
958 * Merge socket-layer source into IGMP-layer source.
959 * If rollback is non-zero, perform the inverse of the merge.
960 */
961 static void
ims_merge(struct ip_msource * ims,const struct in_msource * lims,const int rollback)962 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
963 const int rollback)
964 {
965 int n = rollback ? -1 : 1;
966
967 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
968 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
969 __func__, n, ims->ims_haddr);
970 ims->ims_st[1].ex -= n;
971 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
972 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
973 __func__, n, ims->ims_haddr);
974 ims->ims_st[1].in -= n;
975 }
976
977 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
978 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
979 __func__, n, ims->ims_haddr);
980 ims->ims_st[1].ex += n;
981 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
982 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
983 __func__, n, ims->ims_haddr);
984 ims->ims_st[1].in += n;
985 }
986 }
987
988 /*
989 * Atomically update the global in_multi state, when a membership's
990 * filter list is being updated in any way.
991 *
992 * imf is the per-inpcb-membership group filter pointer.
993 * A fake imf may be passed for in-kernel consumers.
994 *
995 * XXX This is a candidate for a set-symmetric-difference style loop
996 * which would eliminate the repeated lookup from root of ims nodes,
997 * as they share the same key space.
998 *
999 * If any error occurred this function will back out of refcounts
1000 * and return a non-zero value.
1001 */
1002 static int
inm_merge(struct in_multi * inm,struct in_mfilter * imf)1003 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1004 {
1005 struct ip_msource *ims, *nims;
1006 struct in_msource *lims;
1007 int schanged, error;
1008 int nsrc0, nsrc1;
1009
1010 schanged = 0;
1011 error = 0;
1012 nsrc1 = nsrc0 = 0;
1013 IN_MULTI_LIST_LOCK_ASSERT();
1014
1015 /*
1016 * Update the source filters first, as this may fail.
1017 * Maintain count of in-mode filters at t0, t1. These are
1018 * used to work out if we transition into ASM mode or not.
1019 * Maintain a count of source filters whose state was
1020 * actually modified by this operation.
1021 */
1022 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1023 lims = (struct in_msource *)ims;
1024 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1025 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1026 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1027 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1028 ++schanged;
1029 if (error)
1030 break;
1031 ims_merge(nims, lims, 0);
1032 }
1033 if (error) {
1034 struct ip_msource *bims;
1035
1036 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1037 lims = (struct in_msource *)ims;
1038 if (lims->imsl_st[0] == lims->imsl_st[1])
1039 continue;
1040 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1041 if (bims == NULL)
1042 continue;
1043 ims_merge(bims, lims, 1);
1044 }
1045 goto out_reap;
1046 }
1047
1048 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1049 __func__, nsrc0, nsrc1);
1050
1051 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1052 if (imf->imf_st[0] == imf->imf_st[1] &&
1053 imf->imf_st[1] == MCAST_INCLUDE) {
1054 if (nsrc1 == 0) {
1055 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1056 --inm->inm_st[1].iss_in;
1057 }
1058 }
1059
1060 /* Handle filter mode transition on socket. */
1061 if (imf->imf_st[0] != imf->imf_st[1]) {
1062 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1063 __func__, imf->imf_st[0], imf->imf_st[1]);
1064
1065 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1066 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1067 --inm->inm_st[1].iss_ex;
1068 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1069 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1070 --inm->inm_st[1].iss_in;
1071 }
1072
1073 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1074 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1075 inm->inm_st[1].iss_ex++;
1076 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1077 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1078 inm->inm_st[1].iss_in++;
1079 }
1080 }
1081
1082 /*
1083 * Track inm filter state in terms of listener counts.
1084 * If there are any exclusive listeners, stack-wide
1085 * membership is exclusive.
1086 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1087 * If no listeners remain, state is undefined at t1,
1088 * and the IGMP lifecycle for this group should finish.
1089 */
1090 if (inm->inm_st[1].iss_ex > 0) {
1091 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1092 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1093 } else if (inm->inm_st[1].iss_in > 0) {
1094 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1095 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1096 } else {
1097 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1098 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1099 }
1100
1101 /* Decrement ASM listener count on transition out of ASM mode. */
1102 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1103 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1104 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1105 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1106 --inm->inm_st[1].iss_asm;
1107 }
1108 }
1109
1110 /* Increment ASM listener count on transition to ASM mode. */
1111 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1112 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1113 inm->inm_st[1].iss_asm++;
1114 }
1115
1116 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1117 inm_print(inm);
1118
1119 out_reap:
1120 if (schanged > 0) {
1121 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1122 inm_reap(inm);
1123 }
1124 return (error);
1125 }
1126
1127 /*
1128 * Mark an in_multi's filter set deltas as committed.
1129 * Called by IGMP after a state change has been enqueued.
1130 */
1131 void
inm_commit(struct in_multi * inm)1132 inm_commit(struct in_multi *inm)
1133 {
1134 struct ip_msource *ims;
1135
1136 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1137 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1138 inm_print(inm);
1139
1140 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1141 ims->ims_st[0] = ims->ims_st[1];
1142 }
1143 inm->inm_st[0] = inm->inm_st[1];
1144 }
1145
1146 /*
1147 * Reap unreferenced nodes from an in_multi's filter set.
1148 */
1149 static void
inm_reap(struct in_multi * inm)1150 inm_reap(struct in_multi *inm)
1151 {
1152 struct ip_msource *ims, *tims;
1153
1154 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1155 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1156 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1157 ims->ims_stp != 0)
1158 continue;
1159 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1160 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1161 free(ims, M_IPMSOURCE);
1162 inm->inm_nsrc--;
1163 }
1164 }
1165
1166 /*
1167 * Purge all source nodes from an in_multi's filter set.
1168 */
1169 static void
inm_purge(struct in_multi * inm)1170 inm_purge(struct in_multi *inm)
1171 {
1172 struct ip_msource *ims, *tims;
1173
1174 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1175 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1176 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1177 free(ims, M_IPMSOURCE);
1178 inm->inm_nsrc--;
1179 }
1180 mbufq_drain(&inm->inm_scq);
1181 }
1182
1183 /*
1184 * Join a multicast group; unlocked entry point.
1185 *
1186 * SMPng: XXX: in_joingroup() is called from in_control(). Fortunately,
1187 * ifp is unlikely to have been detached at this point, so we assume
1188 * it's OK to recurse.
1189 */
1190 int
in_joingroup(struct ifnet * ifp,const struct in_addr * gina,struct in_mfilter * imf,struct in_multi ** pinm)1191 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1192 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1193 {
1194 int error;
1195
1196 IN_MULTI_LOCK();
1197 error = in_joingroup_locked(ifp, gina, imf, pinm);
1198 IN_MULTI_UNLOCK();
1199
1200 return (error);
1201 }
1202
1203 /*
1204 * Join a multicast group; real entry point.
1205 *
1206 * Only preserves atomicity at inm level.
1207 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1208 *
1209 * If the IGMP downcall fails, the group is not joined, and an error
1210 * code is returned.
1211 */
1212 int
in_joingroup_locked(struct ifnet * ifp,const struct in_addr * gina,struct in_mfilter * imf,struct in_multi ** pinm)1213 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1214 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1215 {
1216 struct in_mfilter timf;
1217 struct in_multi *inm;
1218 int error;
1219
1220 IN_MULTI_LOCK_ASSERT();
1221 IN_MULTI_LIST_UNLOCK_ASSERT();
1222
1223 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1224 ntohl(gina->s_addr), ifp, ifp->if_xname);
1225
1226 error = 0;
1227 inm = NULL;
1228
1229 /*
1230 * If no imf was specified (i.e. kernel consumer),
1231 * fake one up and assume it is an ASM join.
1232 */
1233 if (imf == NULL) {
1234 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1235 imf = &timf;
1236 }
1237
1238 error = in_getmulti(ifp, gina, &inm);
1239 if (error) {
1240 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1241 return (error);
1242 }
1243 IN_MULTI_LIST_LOCK();
1244 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1245 error = inm_merge(inm, imf);
1246 if (error) {
1247 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1248 goto out_inm_release;
1249 }
1250
1251 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1252 error = igmp_change_state(inm);
1253 if (error) {
1254 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1255 goto out_inm_release;
1256 }
1257
1258 out_inm_release:
1259 if (error) {
1260 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1261 IF_ADDR_WLOCK(ifp);
1262 inm_release_deferred(inm);
1263 IF_ADDR_WUNLOCK(ifp);
1264 } else {
1265 *pinm = inm;
1266 }
1267 IN_MULTI_LIST_UNLOCK();
1268
1269 return (error);
1270 }
1271
1272 /*
1273 * Leave a multicast group; unlocked entry point.
1274 */
1275 int
in_leavegroup(struct in_multi * inm,struct in_mfilter * imf)1276 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1277 {
1278 int error;
1279
1280 IN_MULTI_LOCK();
1281 error = in_leavegroup_locked(inm, imf);
1282 IN_MULTI_UNLOCK();
1283
1284 return (error);
1285 }
1286
1287 /*
1288 * Leave a multicast group; real entry point.
1289 * All source filters will be expunged.
1290 *
1291 * Only preserves atomicity at inm level.
1292 *
1293 * Holding the write lock for the INP which contains imf
1294 * is highly advisable. We can't assert for it as imf does not
1295 * contain a back-pointer to the owning inp.
1296 *
1297 * Note: This is not the same as inm_release(*) as this function also
1298 * makes a state change downcall into IGMP.
1299 */
1300 int
in_leavegroup_locked(struct in_multi * inm,struct in_mfilter * imf)1301 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1302 {
1303 struct in_mfilter timf;
1304 int error;
1305
1306 IN_MULTI_LOCK_ASSERT();
1307 IN_MULTI_LIST_UNLOCK_ASSERT();
1308
1309 error = 0;
1310
1311 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1312 inm, ntohl(inm->inm_addr.s_addr),
1313 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1314 imf);
1315
1316 /*
1317 * If no imf was specified (i.e. kernel consumer),
1318 * fake one up and assume it is an ASM join.
1319 */
1320 if (imf == NULL) {
1321 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1322 imf = &timf;
1323 }
1324
1325 /*
1326 * Begin state merge transaction at IGMP layer.
1327 *
1328 * As this particular invocation should not cause any memory
1329 * to be allocated, and there is no opportunity to roll back
1330 * the transaction, it MUST NOT fail.
1331 */
1332 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1333 IN_MULTI_LIST_LOCK();
1334 error = inm_merge(inm, imf);
1335 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1336
1337 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1338 CURVNET_SET(inm->inm_ifp->if_vnet);
1339 error = igmp_change_state(inm);
1340 IF_ADDR_WLOCK(inm->inm_ifp);
1341 inm_release_deferred(inm);
1342 IF_ADDR_WUNLOCK(inm->inm_ifp);
1343 IN_MULTI_LIST_UNLOCK();
1344 CURVNET_RESTORE();
1345 if (error)
1346 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1347
1348 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1349
1350 return (error);
1351 }
1352
1353 /*#ifndef BURN_BRIDGES*/
1354
1355 /*
1356 * Block or unblock an ASM multicast source on an inpcb.
1357 * This implements the delta-based API described in RFC 3678.
1358 *
1359 * The delta-based API applies only to exclusive-mode memberships.
1360 * An IGMP downcall will be performed.
1361 *
1362 * Return 0 if successful, otherwise return an appropriate error code.
1363 */
1364 static int
inp_block_unblock_source(struct inpcb * inp,struct sockopt * sopt)1365 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1366 {
1367 struct epoch_tracker et;
1368 struct group_source_req gsr;
1369 sockunion_t *gsa, *ssa;
1370 struct ifnet *ifp;
1371 struct in_mfilter *imf;
1372 struct ip_moptions *imo;
1373 struct in_msource *ims;
1374 struct in_multi *inm;
1375 uint16_t fmode;
1376 int error, doblock;
1377
1378 ifp = NULL;
1379 error = 0;
1380 doblock = 0;
1381
1382 memset(&gsr, 0, sizeof(struct group_source_req));
1383 gsa = (sockunion_t *)&gsr.gsr_group;
1384 ssa = (sockunion_t *)&gsr.gsr_source;
1385
1386 switch (sopt->sopt_name) {
1387 case IP_BLOCK_SOURCE:
1388 case IP_UNBLOCK_SOURCE: {
1389 struct ip_mreq_source mreqs;
1390
1391 error = sooptcopyin(sopt, &mreqs,
1392 sizeof(struct ip_mreq_source),
1393 sizeof(struct ip_mreq_source));
1394 if (error)
1395 return (error);
1396
1397 gsa->sin.sin_family = AF_INET;
1398 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1399 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1400
1401 ssa->sin.sin_family = AF_INET;
1402 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1403 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1404
1405 if (!in_nullhost(mreqs.imr_interface)) {
1406 NET_EPOCH_ENTER(et);
1407 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1408 /* XXXGL: ifref? */
1409 NET_EPOCH_EXIT(et);
1410 }
1411 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1412 doblock = 1;
1413
1414 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1415 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1416 break;
1417 }
1418
1419 case MCAST_BLOCK_SOURCE:
1420 case MCAST_UNBLOCK_SOURCE:
1421 error = sooptcopyin(sopt, &gsr,
1422 sizeof(struct group_source_req),
1423 sizeof(struct group_source_req));
1424 if (error)
1425 return (error);
1426
1427 if (gsa->sin.sin_family != AF_INET ||
1428 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1429 return (EINVAL);
1430
1431 if (ssa->sin.sin_family != AF_INET ||
1432 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1433 return (EINVAL);
1434
1435 NET_EPOCH_ENTER(et);
1436 ifp = ifnet_byindex(gsr.gsr_interface);
1437 NET_EPOCH_EXIT(et);
1438 if (ifp == NULL)
1439 return (EADDRNOTAVAIL);
1440
1441 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1442 doblock = 1;
1443 break;
1444
1445 default:
1446 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1447 __func__, sopt->sopt_name);
1448 return (EOPNOTSUPP);
1449 break;
1450 }
1451
1452 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1453 return (EINVAL);
1454
1455 IN_MULTI_LOCK();
1456
1457 /*
1458 * Check if we are actually a member of this group.
1459 */
1460 imo = inp_findmoptions(inp);
1461 imf = imo_match_group(imo, ifp, &gsa->sa);
1462 if (imf == NULL) {
1463 error = EADDRNOTAVAIL;
1464 goto out_inp_locked;
1465 }
1466 inm = imf->imf_inm;
1467
1468 /*
1469 * Attempting to use the delta-based API on an
1470 * non exclusive-mode membership is an error.
1471 */
1472 fmode = imf->imf_st[0];
1473 if (fmode != MCAST_EXCLUDE) {
1474 error = EINVAL;
1475 goto out_inp_locked;
1476 }
1477
1478 /*
1479 * Deal with error cases up-front:
1480 * Asked to block, but already blocked; or
1481 * Asked to unblock, but nothing to unblock.
1482 * If adding a new block entry, allocate it.
1483 */
1484 ims = imo_match_source(imf, &ssa->sa);
1485 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1486 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1487 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1488 error = EADDRNOTAVAIL;
1489 goto out_inp_locked;
1490 }
1491
1492 INP_WLOCK_ASSERT(inp);
1493
1494 /*
1495 * Begin state merge transaction at socket layer.
1496 */
1497 if (doblock) {
1498 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1499 ims = imf_graft(imf, fmode, &ssa->sin);
1500 if (ims == NULL)
1501 error = ENOMEM;
1502 } else {
1503 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1504 error = imf_prune(imf, &ssa->sin);
1505 }
1506
1507 if (error) {
1508 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1509 goto out_imf_rollback;
1510 }
1511
1512 /*
1513 * Begin state merge transaction at IGMP layer.
1514 */
1515 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1516 IN_MULTI_LIST_LOCK();
1517 error = inm_merge(inm, imf);
1518 if (error) {
1519 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1520 IN_MULTI_LIST_UNLOCK();
1521 goto out_imf_rollback;
1522 }
1523
1524 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1525 error = igmp_change_state(inm);
1526 IN_MULTI_LIST_UNLOCK();
1527 if (error)
1528 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1529
1530 out_imf_rollback:
1531 if (error)
1532 imf_rollback(imf);
1533 else
1534 imf_commit(imf);
1535
1536 imf_reap(imf);
1537
1538 out_inp_locked:
1539 INP_WUNLOCK(inp);
1540 IN_MULTI_UNLOCK();
1541 return (error);
1542 }
1543
1544 /*
1545 * Given an inpcb, return its multicast options structure pointer. Accepts
1546 * an unlocked inpcb pointer, but will return it locked. May sleep.
1547 *
1548 * SMPng: NOTE: Returns with the INP write lock held.
1549 */
1550 static struct ip_moptions *
inp_findmoptions(struct inpcb * inp)1551 inp_findmoptions(struct inpcb *inp)
1552 {
1553 struct ip_moptions *imo;
1554
1555 INP_WLOCK(inp);
1556 if (inp->inp_moptions != NULL)
1557 return (inp->inp_moptions);
1558
1559 INP_WUNLOCK(inp);
1560
1561 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1562
1563 imo->imo_multicast_ifp = NULL;
1564 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1565 imo->imo_multicast_vif = -1;
1566 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1567 imo->imo_multicast_loop = in_mcast_loop;
1568 STAILQ_INIT(&imo->imo_head);
1569
1570 INP_WLOCK(inp);
1571 if (inp->inp_moptions != NULL) {
1572 free(imo, M_IPMOPTS);
1573 return (inp->inp_moptions);
1574 }
1575 inp->inp_moptions = imo;
1576 return (imo);
1577 }
1578
1579 void
inp_freemoptions(struct ip_moptions * imo)1580 inp_freemoptions(struct ip_moptions *imo)
1581 {
1582 struct in_mfilter *imf;
1583 struct in_multi *inm;
1584 struct ifnet *ifp;
1585
1586 if (imo == NULL)
1587 return;
1588
1589 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1590 ip_mfilter_remove(&imo->imo_head, imf);
1591
1592 imf_leave(imf);
1593 if ((inm = imf->imf_inm) != NULL) {
1594 if ((ifp = inm->inm_ifp) != NULL) {
1595 CURVNET_SET(ifp->if_vnet);
1596 (void)in_leavegroup(inm, imf);
1597 CURVNET_RESTORE();
1598 } else {
1599 (void)in_leavegroup(inm, imf);
1600 }
1601 }
1602 ip_mfilter_free(imf);
1603 }
1604 free(imo, M_IPMOPTS);
1605 }
1606
1607 /*
1608 * Atomically get source filters on a socket for an IPv4 multicast group.
1609 * Called with INP lock held; returns with lock released.
1610 */
1611 static int
inp_get_source_filters(struct inpcb * inp,struct sockopt * sopt)1612 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1613 {
1614 struct epoch_tracker et;
1615 struct __msfilterreq msfr;
1616 sockunion_t *gsa;
1617 struct ifnet *ifp;
1618 struct ip_moptions *imo;
1619 struct in_mfilter *imf;
1620 struct ip_msource *ims;
1621 struct in_msource *lims;
1622 struct sockaddr_in *psin;
1623 struct sockaddr_storage *ptss;
1624 struct sockaddr_storage *tss;
1625 int error;
1626 size_t nsrcs, ncsrcs;
1627
1628 INP_WLOCK_ASSERT(inp);
1629
1630 imo = inp->inp_moptions;
1631 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1632
1633 INP_WUNLOCK(inp);
1634
1635 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1636 sizeof(struct __msfilterreq));
1637 if (error)
1638 return (error);
1639
1640 NET_EPOCH_ENTER(et);
1641 ifp = ifnet_byindex(msfr.msfr_ifindex);
1642 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifnet pointer left */
1643 if (ifp == NULL)
1644 return (EINVAL);
1645
1646 INP_WLOCK(inp);
1647
1648 /*
1649 * Lookup group on the socket.
1650 */
1651 gsa = (sockunion_t *)&msfr.msfr_group;
1652 imf = imo_match_group(imo, ifp, &gsa->sa);
1653 if (imf == NULL) {
1654 INP_WUNLOCK(inp);
1655 return (EADDRNOTAVAIL);
1656 }
1657
1658 /*
1659 * Ignore memberships which are in limbo.
1660 */
1661 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1662 INP_WUNLOCK(inp);
1663 return (EAGAIN);
1664 }
1665 msfr.msfr_fmode = imf->imf_st[1];
1666
1667 /*
1668 * If the user specified a buffer, copy out the source filter
1669 * entries to userland gracefully.
1670 * We only copy out the number of entries which userland
1671 * has asked for, but we always tell userland how big the
1672 * buffer really needs to be.
1673 */
1674 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1675 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1676 tss = NULL;
1677 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1678 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1679 M_TEMP, M_NOWAIT | M_ZERO);
1680 if (tss == NULL) {
1681 INP_WUNLOCK(inp);
1682 return (ENOBUFS);
1683 }
1684 }
1685
1686 /*
1687 * Count number of sources in-mode at t0.
1688 * If buffer space exists and remains, copy out source entries.
1689 */
1690 nsrcs = msfr.msfr_nsrcs;
1691 ncsrcs = 0;
1692 ptss = tss;
1693 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1694 lims = (struct in_msource *)ims;
1695 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1696 lims->imsl_st[0] != imf->imf_st[0])
1697 continue;
1698 ++ncsrcs;
1699 if (tss != NULL && nsrcs > 0) {
1700 psin = (struct sockaddr_in *)ptss;
1701 psin->sin_family = AF_INET;
1702 psin->sin_len = sizeof(struct sockaddr_in);
1703 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1704 psin->sin_port = 0;
1705 ++ptss;
1706 --nsrcs;
1707 }
1708 }
1709
1710 INP_WUNLOCK(inp);
1711
1712 if (tss != NULL) {
1713 error = copyout(tss, msfr.msfr_srcs,
1714 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1715 free(tss, M_TEMP);
1716 if (error)
1717 return (error);
1718 }
1719
1720 msfr.msfr_nsrcs = ncsrcs;
1721 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1722
1723 return (error);
1724 }
1725
1726 /*
1727 * Return the IP multicast options in response to user getsockopt().
1728 */
1729 int
inp_getmoptions(struct inpcb * inp,struct sockopt * sopt)1730 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1731 {
1732 struct ip_mreqn mreqn;
1733 struct ip_moptions *imo;
1734 struct ifnet *ifp;
1735 struct in_ifaddr *ia;
1736 int error, optval;
1737 u_char coptval;
1738
1739 INP_WLOCK(inp);
1740 imo = inp->inp_moptions;
1741 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM reject it. */
1742 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1743 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM) {
1744 INP_WUNLOCK(inp);
1745 return (EOPNOTSUPP);
1746 }
1747
1748 error = 0;
1749 switch (sopt->sopt_name) {
1750 case IP_MULTICAST_VIF:
1751 if (imo != NULL)
1752 optval = imo->imo_multicast_vif;
1753 else
1754 optval = -1;
1755 INP_WUNLOCK(inp);
1756 error = sooptcopyout(sopt, &optval, sizeof(int));
1757 break;
1758
1759 case IP_MULTICAST_IF:
1760 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1761 if (imo != NULL) {
1762 ifp = imo->imo_multicast_ifp;
1763 if (!in_nullhost(imo->imo_multicast_addr)) {
1764 mreqn.imr_address = imo->imo_multicast_addr;
1765 } else if (ifp != NULL) {
1766 struct epoch_tracker et;
1767
1768 mreqn.imr_ifindex = ifp->if_index;
1769 NET_EPOCH_ENTER(et);
1770 IFP_TO_IA(ifp, ia);
1771 if (ia != NULL)
1772 mreqn.imr_address =
1773 IA_SIN(ia)->sin_addr;
1774 NET_EPOCH_EXIT(et);
1775 }
1776 }
1777 INP_WUNLOCK(inp);
1778 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1779 error = sooptcopyout(sopt, &mreqn,
1780 sizeof(struct ip_mreqn));
1781 } else {
1782 error = sooptcopyout(sopt, &mreqn.imr_address,
1783 sizeof(struct in_addr));
1784 }
1785 break;
1786
1787 case IP_MULTICAST_TTL:
1788 if (imo == NULL)
1789 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1790 else
1791 optval = coptval = imo->imo_multicast_ttl;
1792 INP_WUNLOCK(inp);
1793 if (sopt->sopt_valsize == sizeof(u_char))
1794 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1795 else
1796 error = sooptcopyout(sopt, &optval, sizeof(int));
1797 break;
1798
1799 case IP_MULTICAST_LOOP:
1800 if (imo == NULL)
1801 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1802 else
1803 optval = coptval = imo->imo_multicast_loop;
1804 INP_WUNLOCK(inp);
1805 if (sopt->sopt_valsize == sizeof(u_char))
1806 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1807 else
1808 error = sooptcopyout(sopt, &optval, sizeof(int));
1809 break;
1810
1811 case IP_MSFILTER:
1812 if (imo == NULL) {
1813 error = EADDRNOTAVAIL;
1814 INP_WUNLOCK(inp);
1815 } else {
1816 error = inp_get_source_filters(inp, sopt);
1817 }
1818 break;
1819
1820 default:
1821 INP_WUNLOCK(inp);
1822 error = ENOPROTOOPT;
1823 break;
1824 }
1825
1826 INP_UNLOCK_ASSERT(inp);
1827
1828 return (error);
1829 }
1830
1831 /*
1832 * Look up the ifnet to join a multicast group membership via legacy
1833 * IP_ADD_MEMBERSHIP or via more modern MCAST_JOIN_GROUP.
1834 *
1835 * If the interface index was specified explicitly, just use it. If the
1836 * address was specified (legacy), try to find matching interface. Else
1837 * (index == 0 && no address) do a route lookup. If that fails for a modern
1838 * MCAST_JOIN_GROUP return failure, for legacy IP_ADD_MEMBERSHIP find first
1839 * multicast capable interface.
1840 */
1841 static struct ifnet *
inp_lookup_mcast_ifp(const struct inpcb * inp,const struct in_addr maddr,const struct in_addr * ina,const u_int index)1842 inp_lookup_mcast_ifp(const struct inpcb *inp, const struct in_addr maddr,
1843 const struct in_addr *ina, const u_int index)
1844 {
1845 struct ifnet *ifp;
1846 struct nhop_object *nh;
1847
1848 NET_EPOCH_ASSERT();
1849
1850 if (index != 0)
1851 return (ifnet_byindex_ref(index));
1852
1853 if (ina != NULL && !in_nullhost(*ina)) {
1854 INADDR_TO_IFP(*ina, ifp);
1855 if (ifp != NULL)
1856 if_ref(ifp);
1857 return (ifp);
1858 }
1859
1860 nh = fib4_lookup(inp->inp_inc.inc_fibnum, maddr, 0, NHR_NONE, 0);
1861 if (nh != NULL) {
1862 ifp = nh->nh_ifp;
1863 if_ref(ifp);
1864 return (ifp);
1865 }
1866
1867 if (ina != NULL) {
1868 struct in_ifaddr *ia;
1869
1870 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1871 if (!(ia->ia_ifp->if_flags & IFF_LOOPBACK) &&
1872 (ia->ia_ifp->if_flags & IFF_MULTICAST)) {
1873 ifp = ia->ia_ifp;
1874 if_ref(ifp);
1875 return (ifp);
1876 }
1877 }
1878 }
1879
1880 return (NULL);
1881 }
1882
1883 /*
1884 * Join an IPv4 multicast group, possibly with a source.
1885 */
1886 static int
inp_join_group(struct inpcb * inp,struct sockopt * sopt)1887 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1888 {
1889 struct group_source_req gsr;
1890 sockunion_t *gsa, *ssa;
1891 struct ifnet *ifp;
1892 struct in_mfilter *imf;
1893 struct ip_moptions *imo;
1894 struct in_multi *inm;
1895 struct in_msource *lims;
1896 struct epoch_tracker et;
1897 int error, is_new;
1898
1899 ifp = NULL;
1900 lims = NULL;
1901 error = 0;
1902
1903 memset(&gsr, 0, sizeof(struct group_source_req));
1904 gsa = (sockunion_t *)&gsr.gsr_group;
1905 gsa->ss.ss_family = AF_UNSPEC;
1906 ssa = (sockunion_t *)&gsr.gsr_source;
1907 ssa->ss.ss_family = AF_UNSPEC;
1908
1909 switch (sopt->sopt_name) {
1910 case IP_ADD_MEMBERSHIP: {
1911 struct ip_mreqn mreqn;
1912 bool mreq;
1913
1914 mreq = (sopt->sopt_valsize != sizeof(struct ip_mreqn));
1915
1916 error = sooptcopyin(sopt, &mreqn,
1917 mreq ? sizeof(struct ip_mreq) : sizeof(struct ip_mreqn),
1918 mreq ? sizeof(struct ip_mreq) : sizeof(struct ip_mreqn));
1919 if (error)
1920 return (error);
1921
1922 gsa->sin.sin_family = AF_INET;
1923 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1924 gsa->sin.sin_addr = mreqn.imr_multiaddr;
1925 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1926 return (EINVAL);
1927
1928 NET_EPOCH_ENTER(et);
1929 ifp = inp_lookup_mcast_ifp(inp, mreqn.imr_multiaddr,
1930 mreq ? &mreqn.imr_address : NULL,
1931 mreq ? 0 : mreqn.imr_ifindex);
1932 NET_EPOCH_EXIT(et);
1933 break;
1934 }
1935 case IP_ADD_SOURCE_MEMBERSHIP: {
1936 struct ip_mreq_source mreqs;
1937
1938 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source),
1939 sizeof(struct ip_mreq_source));
1940 if (error)
1941 return (error);
1942
1943 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET;
1944 gsa->sin.sin_len = ssa->sin.sin_len =
1945 sizeof(struct sockaddr_in);
1946
1947 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1948 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1949 return (EINVAL);
1950
1951 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1952
1953 NET_EPOCH_ENTER(et);
1954 ifp = inp_lookup_mcast_ifp(inp, mreqs.imr_multiaddr,
1955 &mreqs.imr_interface, 0);
1956 NET_EPOCH_EXIT(et);
1957 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1958 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1959 break;
1960 }
1961
1962 case MCAST_JOIN_GROUP:
1963 case MCAST_JOIN_SOURCE_GROUP:
1964 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1965 error = sooptcopyin(sopt, &gsr,
1966 sizeof(struct group_req),
1967 sizeof(struct group_req));
1968 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1969 error = sooptcopyin(sopt, &gsr,
1970 sizeof(struct group_source_req),
1971 sizeof(struct group_source_req));
1972 }
1973 if (error)
1974 return (error);
1975
1976 if (gsa->sin.sin_family != AF_INET ||
1977 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1978 return (EINVAL);
1979
1980 /*
1981 * Overwrite the port field if present, as the sockaddr
1982 * being copied in may be matched with a binary comparison.
1983 */
1984 gsa->sin.sin_port = 0;
1985 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1986 if (ssa->sin.sin_family != AF_INET ||
1987 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1988 return (EINVAL);
1989 ssa->sin.sin_port = 0;
1990 }
1991
1992 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1993 return (EINVAL);
1994
1995 NET_EPOCH_ENTER(et);
1996 ifp = inp_lookup_mcast_ifp(inp, gsa->sin.sin_addr, NULL,
1997 gsr.gsr_interface);
1998 NET_EPOCH_EXIT(et);
1999 if (ifp == NULL)
2000 return (EADDRNOTAVAIL);
2001 break;
2002
2003 default:
2004 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2005 __func__, sopt->sopt_name);
2006 return (EOPNOTSUPP);
2007 break;
2008 }
2009
2010 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0) {
2011 if (ifp != NULL)
2012 if_rele(ifp);
2013 return (EADDRNOTAVAIL);
2014 }
2015
2016 IN_MULTI_LOCK();
2017
2018 /*
2019 * Find the membership in the membership list.
2020 */
2021 imo = inp_findmoptions(inp);
2022 imf = imo_match_group(imo, ifp, &gsa->sa);
2023 if (imf == NULL) {
2024 is_new = 1;
2025 inm = NULL;
2026
2027 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) {
2028 error = ENOMEM;
2029 goto out_inp_locked;
2030 }
2031 } else {
2032 is_new = 0;
2033 inm = imf->imf_inm;
2034
2035 if (ssa->ss.ss_family != AF_UNSPEC) {
2036 /*
2037 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2038 * is an error. On an existing inclusive membership,
2039 * it just adds the source to the filter list.
2040 */
2041 if (imf->imf_st[1] != MCAST_INCLUDE) {
2042 error = EINVAL;
2043 goto out_inp_locked;
2044 }
2045 /*
2046 * Throw out duplicates.
2047 *
2048 * XXX FIXME: This makes a naive assumption that
2049 * even if entries exist for *ssa in this imf,
2050 * they will be rejected as dupes, even if they
2051 * are not valid in the current mode (in-mode).
2052 *
2053 * in_msource is transactioned just as for anything
2054 * else in SSM -- but note naive use of inm_graft()
2055 * below for allocating new filter entries.
2056 *
2057 * This is only an issue if someone mixes the
2058 * full-state SSM API with the delta-based API,
2059 * which is discouraged in the relevant RFCs.
2060 */
2061 lims = imo_match_source(imf, &ssa->sa);
2062 if (lims != NULL /*&&
2063 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2064 error = EADDRNOTAVAIL;
2065 goto out_inp_locked;
2066 }
2067 } else {
2068 /*
2069 * MCAST_JOIN_GROUP on an existing exclusive
2070 * membership is an error; return EADDRINUSE
2071 * to preserve 4.4BSD API idempotence, and
2072 * avoid tedious detour to code below.
2073 * NOTE: This is bending RFC 3678 a bit.
2074 *
2075 * On an existing inclusive membership, this is also
2076 * an error; if you want to change filter mode,
2077 * you must use the userland API setsourcefilter().
2078 * XXX We don't reject this for imf in UNDEFINED
2079 * state at t1, because allocation of a filter
2080 * is atomic with allocation of a membership.
2081 */
2082 error = EINVAL;
2083 if (imf->imf_st[1] == MCAST_EXCLUDE)
2084 error = EADDRINUSE;
2085 goto out_inp_locked;
2086 }
2087 }
2088
2089 /*
2090 * Begin state merge transaction at socket layer.
2091 */
2092 INP_WLOCK_ASSERT(inp);
2093
2094 /*
2095 * Graft new source into filter list for this inpcb's
2096 * membership of the group. The in_multi may not have
2097 * been allocated yet if this is a new membership, however,
2098 * the in_mfilter slot will be allocated and must be initialized.
2099 *
2100 * Note: Grafting of exclusive mode filters doesn't happen
2101 * in this path.
2102 * XXX: Should check for non-NULL lims (node exists but may
2103 * not be in-mode) for interop with full-state API.
2104 */
2105 if (ssa->ss.ss_family != AF_UNSPEC) {
2106 /* Membership starts in IN mode */
2107 if (is_new) {
2108 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2109 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2110 if (imf == NULL) {
2111 error = ENOMEM;
2112 goto out_inp_locked;
2113 }
2114 } else {
2115 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2116 }
2117 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2118 if (lims == NULL) {
2119 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2120 __func__);
2121 error = ENOMEM;
2122 goto out_inp_locked;
2123 }
2124 } else {
2125 /* No address specified; Membership starts in EX mode */
2126 if (is_new) {
2127 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2128 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2129 if (imf == NULL) {
2130 error = ENOMEM;
2131 goto out_inp_locked;
2132 }
2133 }
2134 }
2135
2136 /*
2137 * Begin state merge transaction at IGMP layer.
2138 */
2139 if (is_new) {
2140 in_pcbref(inp);
2141 INP_WUNLOCK(inp);
2142
2143 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2144 &imf->imf_inm);
2145
2146 INP_WLOCK(inp);
2147 if (in_pcbrele_wlocked(inp)) {
2148 error = ENXIO;
2149 goto out_inp_unlocked;
2150 }
2151 if (error) {
2152 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2153 __func__);
2154 goto out_inp_locked;
2155 }
2156 /*
2157 * NOTE: Refcount from in_joingroup_locked()
2158 * is protecting membership.
2159 */
2160 ip_mfilter_insert(&imo->imo_head, imf);
2161 } else {
2162 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2163 IN_MULTI_LIST_LOCK();
2164 error = inm_merge(inm, imf);
2165 if (error) {
2166 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2167 __func__);
2168 IN_MULTI_LIST_UNLOCK();
2169 imf_rollback(imf);
2170 imf_reap(imf);
2171 goto out_inp_locked;
2172 }
2173 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2174 error = igmp_change_state(inm);
2175 IN_MULTI_LIST_UNLOCK();
2176 if (error) {
2177 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2178 __func__);
2179 imf_rollback(imf);
2180 imf_reap(imf);
2181 goto out_inp_locked;
2182 }
2183 }
2184
2185 imf_commit(imf);
2186 imf = NULL;
2187
2188 out_inp_locked:
2189 INP_WUNLOCK(inp);
2190 out_inp_unlocked:
2191 IN_MULTI_UNLOCK();
2192
2193 if (is_new && imf) {
2194 if (imf->imf_inm != NULL) {
2195 IN_MULTI_LIST_LOCK();
2196 IF_ADDR_WLOCK(ifp);
2197 inm_release_deferred(imf->imf_inm);
2198 IF_ADDR_WUNLOCK(ifp);
2199 IN_MULTI_LIST_UNLOCK();
2200 }
2201 ip_mfilter_free(imf);
2202 }
2203 if_rele(ifp);
2204 return (error);
2205 }
2206
2207 /*
2208 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2209 */
2210 static int
inp_leave_group(struct inpcb * inp,struct sockopt * sopt)2211 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2212 {
2213 struct epoch_tracker et;
2214 struct group_source_req gsr;
2215 struct ip_mreq_source mreqs;
2216 sockunion_t *gsa, *ssa;
2217 struct ifnet *ifp;
2218 struct in_mfilter *imf;
2219 struct ip_moptions *imo;
2220 struct in_msource *ims;
2221 struct in_multi *inm;
2222 int error;
2223 bool is_final;
2224
2225 ifp = NULL;
2226 error = 0;
2227 is_final = true;
2228
2229 memset(&gsr, 0, sizeof(struct group_source_req));
2230 gsa = (sockunion_t *)&gsr.gsr_group;
2231 gsa->ss.ss_family = AF_UNSPEC;
2232 ssa = (sockunion_t *)&gsr.gsr_source;
2233 ssa->ss.ss_family = AF_UNSPEC;
2234
2235 switch (sopt->sopt_name) {
2236 case IP_DROP_MEMBERSHIP:
2237 case IP_DROP_SOURCE_MEMBERSHIP:
2238 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2239 error = sooptcopyin(sopt, &mreqs,
2240 sizeof(struct ip_mreq),
2241 sizeof(struct ip_mreq));
2242 /*
2243 * Swap interface and sourceaddr arguments,
2244 * as ip_mreq and ip_mreq_source are laid
2245 * out differently.
2246 */
2247 mreqs.imr_interface = mreqs.imr_sourceaddr;
2248 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2249 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2250 error = sooptcopyin(sopt, &mreqs,
2251 sizeof(struct ip_mreq_source),
2252 sizeof(struct ip_mreq_source));
2253 }
2254 if (error)
2255 return (error);
2256
2257 gsa->sin.sin_family = AF_INET;
2258 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2259 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2260
2261 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2262 ssa->sin.sin_family = AF_INET;
2263 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2264 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2265 }
2266
2267 /*
2268 * Attempt to look up hinted ifp from interface address.
2269 * Fallthrough with null ifp iff lookup fails, to
2270 * preserve 4.4BSD mcast API idempotence.
2271 * XXX NOTE WELL: The RFC 3678 API is preferred because
2272 * using an IPv4 address as a key is racy.
2273 */
2274 if (!in_nullhost(mreqs.imr_interface)) {
2275 NET_EPOCH_ENTER(et);
2276 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2277 /* XXXGL ifref? */
2278 NET_EPOCH_EXIT(et);
2279 }
2280 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2281 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2282
2283 break;
2284
2285 case MCAST_LEAVE_GROUP:
2286 case MCAST_LEAVE_SOURCE_GROUP:
2287 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2288 error = sooptcopyin(sopt, &gsr,
2289 sizeof(struct group_req),
2290 sizeof(struct group_req));
2291 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2292 error = sooptcopyin(sopt, &gsr,
2293 sizeof(struct group_source_req),
2294 sizeof(struct group_source_req));
2295 }
2296 if (error)
2297 return (error);
2298
2299 if (gsa->sin.sin_family != AF_INET ||
2300 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2301 return (EINVAL);
2302
2303 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2304 if (ssa->sin.sin_family != AF_INET ||
2305 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2306 return (EINVAL);
2307 }
2308
2309 NET_EPOCH_ENTER(et);
2310 ifp = ifnet_byindex(gsr.gsr_interface);
2311 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2312 if (ifp == NULL)
2313 return (EADDRNOTAVAIL);
2314 break;
2315
2316 default:
2317 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2318 __func__, sopt->sopt_name);
2319 return (EOPNOTSUPP);
2320 break;
2321 }
2322
2323 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2324 return (EINVAL);
2325
2326 IN_MULTI_LOCK();
2327
2328 /*
2329 * Find the membership in the membership list.
2330 */
2331 imo = inp_findmoptions(inp);
2332 imf = imo_match_group(imo, ifp, &gsa->sa);
2333 if (imf == NULL) {
2334 error = EADDRNOTAVAIL;
2335 goto out_inp_locked;
2336 }
2337 inm = imf->imf_inm;
2338
2339 if (ssa->ss.ss_family != AF_UNSPEC)
2340 is_final = false;
2341
2342 /*
2343 * Begin state merge transaction at socket layer.
2344 */
2345 INP_WLOCK_ASSERT(inp);
2346
2347 /*
2348 * If we were instructed only to leave a given source, do so.
2349 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2350 */
2351 if (is_final) {
2352 ip_mfilter_remove(&imo->imo_head, imf);
2353 imf_leave(imf);
2354
2355 /*
2356 * Give up the multicast address record to which
2357 * the membership points.
2358 */
2359 (void) in_leavegroup_locked(imf->imf_inm, imf);
2360 } else {
2361 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2362 error = EADDRNOTAVAIL;
2363 goto out_inp_locked;
2364 }
2365 ims = imo_match_source(imf, &ssa->sa);
2366 if (ims == NULL) {
2367 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2368 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2369 error = EADDRNOTAVAIL;
2370 goto out_inp_locked;
2371 }
2372 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2373 error = imf_prune(imf, &ssa->sin);
2374 if (error) {
2375 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2376 __func__);
2377 goto out_inp_locked;
2378 }
2379 }
2380
2381 /*
2382 * Begin state merge transaction at IGMP layer.
2383 */
2384 if (!is_final) {
2385 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2386 IN_MULTI_LIST_LOCK();
2387 error = inm_merge(inm, imf);
2388 if (error) {
2389 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2390 __func__);
2391 IN_MULTI_LIST_UNLOCK();
2392 imf_rollback(imf);
2393 imf_reap(imf);
2394 goto out_inp_locked;
2395 }
2396
2397 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2398 error = igmp_change_state(inm);
2399 IN_MULTI_LIST_UNLOCK();
2400 if (error) {
2401 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2402 __func__);
2403 imf_rollback(imf);
2404 imf_reap(imf);
2405 goto out_inp_locked;
2406 }
2407 }
2408 imf_commit(imf);
2409 imf_reap(imf);
2410
2411 out_inp_locked:
2412 INP_WUNLOCK(inp);
2413
2414 if (is_final && imf)
2415 ip_mfilter_free(imf);
2416
2417 IN_MULTI_UNLOCK();
2418 return (error);
2419 }
2420
2421 /*
2422 * Select the interface for transmitting IPv4 multicast datagrams.
2423 *
2424 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2425 * may be passed to this socket option. An address of INADDR_ANY or an
2426 * interface index of 0 is used to remove a previous selection.
2427 * When no interface is selected, one is chosen for every send.
2428 */
2429 static int
inp_set_multicast_if(struct inpcb * inp,struct sockopt * sopt)2430 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2431 {
2432 struct in_addr addr;
2433 struct ip_mreqn mreqn;
2434 struct ifnet *ifp;
2435 struct ip_moptions *imo;
2436 int error;
2437
2438 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2439 /*
2440 * An interface index was specified using the
2441 * Linux-derived ip_mreqn structure.
2442 */
2443 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2444 sizeof(struct ip_mreqn));
2445 if (error)
2446 return (error);
2447
2448 if (mreqn.imr_ifindex < 0)
2449 return (EINVAL);
2450
2451 if (mreqn.imr_ifindex == 0) {
2452 ifp = NULL;
2453 } else {
2454 struct epoch_tracker et;
2455
2456 NET_EPOCH_ENTER(et);
2457 ifp = ifnet_byindex(mreqn.imr_ifindex);
2458 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2459 if (ifp == NULL)
2460 return (EADDRNOTAVAIL);
2461 }
2462 } else {
2463 /*
2464 * An interface was specified by IPv4 address.
2465 * This is the traditional BSD usage.
2466 */
2467 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2468 sizeof(struct in_addr));
2469 if (error)
2470 return (error);
2471 if (in_nullhost(addr)) {
2472 ifp = NULL;
2473 } else {
2474 struct epoch_tracker et;
2475
2476 NET_EPOCH_ENTER(et);
2477 INADDR_TO_IFP(addr, ifp);
2478 /* XXXGL ifref? */
2479 NET_EPOCH_EXIT(et);
2480 if (ifp == NULL)
2481 return (EADDRNOTAVAIL);
2482 }
2483 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2484 ntohl(addr.s_addr));
2485 }
2486
2487 /* Reject interfaces which do not support multicast. */
2488 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2489 return (EOPNOTSUPP);
2490
2491 imo = inp_findmoptions(inp);
2492 imo->imo_multicast_ifp = ifp;
2493 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2494 INP_WUNLOCK(inp);
2495
2496 return (0);
2497 }
2498
2499 /*
2500 * Atomically set source filters on a socket for an IPv4 multicast group.
2501 */
2502 static int
inp_set_source_filters(struct inpcb * inp,struct sockopt * sopt)2503 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2504 {
2505 struct epoch_tracker et;
2506 struct __msfilterreq msfr;
2507 sockunion_t *gsa;
2508 struct ifnet *ifp;
2509 struct in_mfilter *imf;
2510 struct ip_moptions *imo;
2511 struct in_multi *inm;
2512 int error;
2513
2514 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2515 sizeof(struct __msfilterreq));
2516 if (error)
2517 return (error);
2518
2519 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2520 return (ENOBUFS);
2521
2522 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2523 msfr.msfr_fmode != MCAST_INCLUDE))
2524 return (EINVAL);
2525
2526 if (msfr.msfr_group.ss_family != AF_INET ||
2527 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2528 return (EINVAL);
2529
2530 gsa = (sockunion_t *)&msfr.msfr_group;
2531 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2532 return (EINVAL);
2533
2534 gsa->sin.sin_port = 0; /* ignore port */
2535
2536 NET_EPOCH_ENTER(et);
2537 ifp = ifnet_byindex(msfr.msfr_ifindex);
2538 NET_EPOCH_EXIT(et); /* XXXGL: unsafe ifp */
2539 if (ifp == NULL)
2540 return (EADDRNOTAVAIL);
2541
2542 IN_MULTI_LOCK();
2543
2544 /*
2545 * Take the INP write lock.
2546 * Check if this socket is a member of this group.
2547 */
2548 imo = inp_findmoptions(inp);
2549 imf = imo_match_group(imo, ifp, &gsa->sa);
2550 if (imf == NULL) {
2551 error = EADDRNOTAVAIL;
2552 goto out_inp_locked;
2553 }
2554 inm = imf->imf_inm;
2555
2556 /*
2557 * Begin state merge transaction at socket layer.
2558 */
2559 INP_WLOCK_ASSERT(inp);
2560
2561 imf->imf_st[1] = msfr.msfr_fmode;
2562
2563 /*
2564 * Apply any new source filters, if present.
2565 * Make a copy of the user-space source vector so
2566 * that we may copy them with a single copyin. This
2567 * allows us to deal with page faults up-front.
2568 */
2569 if (msfr.msfr_nsrcs > 0) {
2570 struct in_msource *lims;
2571 struct sockaddr_in *psin;
2572 struct sockaddr_storage *kss, *pkss;
2573 int i;
2574
2575 INP_WUNLOCK(inp);
2576
2577 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2578 __func__, (unsigned long)msfr.msfr_nsrcs);
2579 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2580 M_TEMP, M_WAITOK);
2581 error = copyin(msfr.msfr_srcs, kss,
2582 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2583 if (error) {
2584 free(kss, M_TEMP);
2585 return (error);
2586 }
2587
2588 INP_WLOCK(inp);
2589
2590 /*
2591 * Mark all source filters as UNDEFINED at t1.
2592 * Restore new group filter mode, as imf_leave()
2593 * will set it to INCLUDE.
2594 */
2595 imf_leave(imf);
2596 imf->imf_st[1] = msfr.msfr_fmode;
2597
2598 /*
2599 * Update socket layer filters at t1, lazy-allocating
2600 * new entries. This saves a bunch of memory at the
2601 * cost of one RB_FIND() per source entry; duplicate
2602 * entries in the msfr_nsrcs vector are ignored.
2603 * If we encounter an error, rollback transaction.
2604 *
2605 * XXX This too could be replaced with a set-symmetric
2606 * difference like loop to avoid walking from root
2607 * every time, as the key space is common.
2608 */
2609 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2610 psin = (struct sockaddr_in *)pkss;
2611 if (psin->sin_family != AF_INET) {
2612 error = EAFNOSUPPORT;
2613 break;
2614 }
2615 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2616 error = EINVAL;
2617 break;
2618 }
2619 error = imf_get_source(imf, psin, &lims);
2620 if (error)
2621 break;
2622 lims->imsl_st[1] = imf->imf_st[1];
2623 }
2624 free(kss, M_TEMP);
2625 }
2626
2627 if (error)
2628 goto out_imf_rollback;
2629
2630 INP_WLOCK_ASSERT(inp);
2631
2632 /*
2633 * Begin state merge transaction at IGMP layer.
2634 */
2635 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2636 IN_MULTI_LIST_LOCK();
2637 error = inm_merge(inm, imf);
2638 if (error) {
2639 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2640 IN_MULTI_LIST_UNLOCK();
2641 goto out_imf_rollback;
2642 }
2643
2644 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2645 error = igmp_change_state(inm);
2646 IN_MULTI_LIST_UNLOCK();
2647 if (error)
2648 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2649
2650 out_imf_rollback:
2651 if (error)
2652 imf_rollback(imf);
2653 else
2654 imf_commit(imf);
2655
2656 imf_reap(imf);
2657
2658 out_inp_locked:
2659 INP_WUNLOCK(inp);
2660 IN_MULTI_UNLOCK();
2661 return (error);
2662 }
2663
2664 /*
2665 * Set the IP multicast options in response to user setsockopt().
2666 *
2667 * Many of the socket options handled in this function duplicate the
2668 * functionality of socket options in the regular unicast API. However,
2669 * it is not possible to merge the duplicate code, because the idempotence
2670 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2671 * the effects of these options must be treated as separate and distinct.
2672 *
2673 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2674 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2675 * is refactored to no longer use vifs.
2676 */
2677 int
inp_setmoptions(struct inpcb * inp,struct sockopt * sopt)2678 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2679 {
2680 struct ip_moptions *imo;
2681 int error;
2682
2683 error = 0;
2684
2685 /* If socket is neither of type SOCK_RAW or SOCK_DGRAM, reject it. */
2686 if (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2687 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)
2688 return (EOPNOTSUPP);
2689
2690 switch (sopt->sopt_name) {
2691 case IP_MULTICAST_VIF: {
2692 int vifi;
2693 /*
2694 * Select a multicast VIF for transmission.
2695 * Only useful if multicast forwarding is active.
2696 */
2697 if (legal_vif_num == NULL) {
2698 error = EOPNOTSUPP;
2699 break;
2700 }
2701 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2702 if (error)
2703 break;
2704 if (!legal_vif_num(vifi) && (vifi != -1)) {
2705 error = EINVAL;
2706 break;
2707 }
2708 imo = inp_findmoptions(inp);
2709 imo->imo_multicast_vif = vifi;
2710 INP_WUNLOCK(inp);
2711 break;
2712 }
2713
2714 case IP_MULTICAST_IF:
2715 error = inp_set_multicast_if(inp, sopt);
2716 break;
2717
2718 case IP_MULTICAST_TTL: {
2719 u_char ttl;
2720
2721 /*
2722 * Set the IP time-to-live for outgoing multicast packets.
2723 * The original multicast API required a char argument,
2724 * which is inconsistent with the rest of the socket API.
2725 * We allow either a char or an int.
2726 */
2727 if (sopt->sopt_valsize == sizeof(u_char)) {
2728 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2729 sizeof(u_char));
2730 if (error)
2731 break;
2732 } else {
2733 u_int ittl;
2734
2735 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2736 sizeof(u_int));
2737 if (error)
2738 break;
2739 if (ittl > 255) {
2740 error = EINVAL;
2741 break;
2742 }
2743 ttl = (u_char)ittl;
2744 }
2745 imo = inp_findmoptions(inp);
2746 imo->imo_multicast_ttl = ttl;
2747 INP_WUNLOCK(inp);
2748 break;
2749 }
2750
2751 case IP_MULTICAST_LOOP: {
2752 u_char loop;
2753
2754 /*
2755 * Set the loopback flag for outgoing multicast packets.
2756 * Must be zero or one. The original multicast API required a
2757 * char argument, which is inconsistent with the rest
2758 * of the socket API. We allow either a char or an int.
2759 */
2760 if (sopt->sopt_valsize == sizeof(u_char)) {
2761 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2762 sizeof(u_char));
2763 if (error)
2764 break;
2765 } else {
2766 u_int iloop;
2767
2768 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2769 sizeof(u_int));
2770 if (error)
2771 break;
2772 loop = (u_char)iloop;
2773 }
2774 imo = inp_findmoptions(inp);
2775 imo->imo_multicast_loop = !!loop;
2776 INP_WUNLOCK(inp);
2777 break;
2778 }
2779
2780 case IP_ADD_MEMBERSHIP:
2781 case IP_ADD_SOURCE_MEMBERSHIP:
2782 case MCAST_JOIN_GROUP:
2783 case MCAST_JOIN_SOURCE_GROUP:
2784 error = inp_join_group(inp, sopt);
2785 break;
2786
2787 case IP_DROP_MEMBERSHIP:
2788 case IP_DROP_SOURCE_MEMBERSHIP:
2789 case MCAST_LEAVE_GROUP:
2790 case MCAST_LEAVE_SOURCE_GROUP:
2791 error = inp_leave_group(inp, sopt);
2792 break;
2793
2794 case IP_BLOCK_SOURCE:
2795 case IP_UNBLOCK_SOURCE:
2796 case MCAST_BLOCK_SOURCE:
2797 case MCAST_UNBLOCK_SOURCE:
2798 error = inp_block_unblock_source(inp, sopt);
2799 break;
2800
2801 case IP_MSFILTER:
2802 error = inp_set_source_filters(inp, sopt);
2803 break;
2804
2805 default:
2806 error = EOPNOTSUPP;
2807 break;
2808 }
2809
2810 INP_UNLOCK_ASSERT(inp);
2811
2812 return (error);
2813 }
2814
2815 /*
2816 * Expose IGMP's multicast filter mode and source list(s) to userland,
2817 * keyed by (ifindex, group).
2818 * The filter mode is written out as a uint32_t, followed by
2819 * 0..n of struct in_addr.
2820 * For use by ifmcstat(8).
2821 * SMPng: NOTE: unlocked read of ifindex space.
2822 */
2823 static int
sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)2824 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2825 {
2826 struct in_addr src, group;
2827 struct epoch_tracker et;
2828 struct ifnet *ifp;
2829 struct ifmultiaddr *ifma;
2830 struct in_multi *inm;
2831 struct ip_msource *ims;
2832 int *name;
2833 int retval;
2834 u_int namelen;
2835 uint32_t fmode, ifindex;
2836
2837 name = (int *)arg1;
2838 namelen = arg2;
2839
2840 if (req->newptr != NULL)
2841 return (EPERM);
2842
2843 if (namelen != 2)
2844 return (EINVAL);
2845
2846 group.s_addr = name[1];
2847 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2848 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2849 __func__, ntohl(group.s_addr));
2850 return (EINVAL);
2851 }
2852
2853 ifindex = name[0];
2854 NET_EPOCH_ENTER(et);
2855 ifp = ifnet_byindex(ifindex);
2856 if (ifp == NULL) {
2857 NET_EPOCH_EXIT(et);
2858 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2859 __func__, ifindex);
2860 return (ENOENT);
2861 }
2862
2863 retval = sysctl_wire_old_buffer(req,
2864 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2865 if (retval) {
2866 NET_EPOCH_EXIT(et);
2867 return (retval);
2868 }
2869
2870 IN_MULTI_LIST_LOCK();
2871
2872 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2873 inm = inm_ifmultiaddr_get_inm(ifma);
2874 if (inm == NULL)
2875 continue;
2876 if (!in_hosteq(inm->inm_addr, group))
2877 continue;
2878 fmode = inm->inm_st[1].iss_fmode;
2879 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2880 if (retval != 0)
2881 break;
2882 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2883 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2884 ims->ims_haddr);
2885 /*
2886 * Only copy-out sources which are in-mode.
2887 */
2888 if (fmode != ims_get_mode(inm, ims, 1)) {
2889 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2890 __func__);
2891 continue;
2892 }
2893 src.s_addr = htonl(ims->ims_haddr);
2894 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2895 if (retval != 0)
2896 break;
2897 }
2898 }
2899
2900 IN_MULTI_LIST_UNLOCK();
2901 NET_EPOCH_EXIT(et);
2902
2903 return (retval);
2904 }
2905
2906 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2907
2908 static const char *inm_modestrs[] = {
2909 [MCAST_UNDEFINED] = "un",
2910 [MCAST_INCLUDE] = "in",
2911 [MCAST_EXCLUDE] = "ex",
2912 };
2913 _Static_assert(MCAST_UNDEFINED == 0 &&
2914 MCAST_EXCLUDE + 1 == nitems(inm_modestrs),
2915 "inm_modestrs: no longer matches #defines");
2916
2917 static const char *
inm_mode_str(const int mode)2918 inm_mode_str(const int mode)
2919 {
2920
2921 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2922 return (inm_modestrs[mode]);
2923 return ("??");
2924 }
2925
2926 static const char *inm_statestrs[] = {
2927 [IGMP_NOT_MEMBER] = "not-member",
2928 [IGMP_SILENT_MEMBER] = "silent",
2929 [IGMP_REPORTING_MEMBER] = "reporting",
2930 [IGMP_IDLE_MEMBER] = "idle",
2931 [IGMP_LAZY_MEMBER] = "lazy",
2932 [IGMP_SLEEPING_MEMBER] = "sleeping",
2933 [IGMP_AWAKENING_MEMBER] = "awakening",
2934 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending",
2935 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending",
2936 [IGMP_LEAVING_MEMBER] = "leaving",
2937 };
2938 _Static_assert(IGMP_NOT_MEMBER == 0 &&
2939 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs),
2940 "inm_statetrs: no longer matches #defines");
2941
2942 static const char *
inm_state_str(const int state)2943 inm_state_str(const int state)
2944 {
2945
2946 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2947 return (inm_statestrs[state]);
2948 return ("??");
2949 }
2950
2951 /*
2952 * Dump an in_multi structure to the console.
2953 */
2954 void
inm_print(const struct in_multi * inm)2955 inm_print(const struct in_multi *inm)
2956 {
2957 int t;
2958 char addrbuf[INET_ADDRSTRLEN];
2959
2960 if ((ktr_mask & KTR_IGMPV3) == 0)
2961 return;
2962
2963 printf("%s: --- begin inm %p ---\n", __func__, inm);
2964 printf("addr %s ifp %p(%s) ifma %p\n",
2965 inet_ntoa_r(inm->inm_addr, addrbuf),
2966 inm->inm_ifp,
2967 inm->inm_ifp->if_xname,
2968 inm->inm_ifma);
2969 printf("timer %u state %s refcount %u scq.len %u\n",
2970 inm->inm_timer,
2971 inm_state_str(inm->inm_state),
2972 inm->inm_refcount,
2973 inm->inm_scq.mq_len);
2974 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2975 inm->inm_igi,
2976 inm->inm_nsrc,
2977 inm->inm_sctimer,
2978 inm->inm_scrv);
2979 for (t = 0; t < 2; t++) {
2980 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2981 inm_mode_str(inm->inm_st[t].iss_fmode),
2982 inm->inm_st[t].iss_asm,
2983 inm->inm_st[t].iss_ex,
2984 inm->inm_st[t].iss_in,
2985 inm->inm_st[t].iss_rec);
2986 }
2987 printf("%s: --- end inm %p ---\n", __func__, inm);
2988 }
2989
2990 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
2991
2992 void
inm_print(const struct in_multi * inm)2993 inm_print(const struct in_multi *inm)
2994 {
2995
2996 }
2997
2998 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
2999
3000 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);
3001