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/cdefs.h>
38 __FBSDID("$FreeBSD$");
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/protosw.h>
47 #include <sys/rmlock.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/protosw.h>
51 #include <sys/sysctl.h>
52 #include <sys/ktr.h>
53 #include <sys/taskqueue.h>
54 #include <sys/gtaskqueue.h>
55 #include <sys/tree.h>
56
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
61 #include <net/vnet.h>
62
63 #include <net/ethernet.h>
64
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_fib.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/igmp_var.h>
72
73 #ifndef KTR_IGMPV3
74 #define KTR_IGMPV3 KTR_INET
75 #endif
76
77 #ifndef __SOCKUNION_DECLARED
78 union sockunion {
79 struct sockaddr_storage ss;
80 struct sockaddr sa;
81 struct sockaddr_dl sdl;
82 struct sockaddr_in sin;
83 };
84 typedef union sockunion sockunion_t;
85 #define __SOCKUNION_DECLARED
86 #endif /* __SOCKUNION_DECLARED */
87
88 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
89 "IPv4 multicast PCB-layer source filter");
90 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
91 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
92 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
93 "IPv4 multicast IGMP-layer source filter");
94
95 /*
96 * Locking:
97 *
98 * - Lock order is: Giant, IN_MULTI_LOCK, INP_WLOCK,
99 * IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
100 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
101 * it can be taken by code in net/if.c also.
102 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
103 *
104 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
105 * any need for in_multi itself to be virtualized -- it is bound to an ifp
106 * anyway no matter what happens.
107 */
108 struct mtx in_multi_list_mtx;
109 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
110
111 struct mtx in_multi_free_mtx;
112 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
113
114 struct sx in_multi_sx;
115 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
116
117 int ifma_restart;
118
119 /*
120 * Functions with non-static linkage defined in this file should be
121 * declared in in_var.h:
122 * imo_multi_filter()
123 * in_addmulti()
124 * in_delmulti()
125 * in_joingroup()
126 * in_joingroup_locked()
127 * in_leavegroup()
128 * in_leavegroup_locked()
129 * and ip_var.h:
130 * inp_freemoptions()
131 * inp_getmoptions()
132 * inp_setmoptions()
133 *
134 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
135 * and in_delmulti().
136 */
137 static void imf_commit(struct in_mfilter *);
138 static int imf_get_source(struct in_mfilter *imf,
139 const struct sockaddr_in *psin,
140 struct in_msource **);
141 static struct in_msource *
142 imf_graft(struct in_mfilter *, const uint8_t,
143 const struct sockaddr_in *);
144 static void imf_leave(struct in_mfilter *);
145 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
146 static void imf_purge(struct in_mfilter *);
147 static void imf_rollback(struct in_mfilter *);
148 static void imf_reap(struct in_mfilter *);
149 static struct in_mfilter *
150 imo_match_group(const struct ip_moptions *,
151 const struct ifnet *, const struct sockaddr *);
152 static struct in_msource *
153 imo_match_source(struct in_mfilter *, const struct sockaddr *);
154 static void ims_merge(struct ip_msource *ims,
155 const struct in_msource *lims, const int rollback);
156 static int in_getmulti(struct ifnet *, const struct in_addr *,
157 struct in_multi **);
158 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
159 const int noalloc, struct ip_msource **pims);
160 #ifdef KTR
161 static int inm_is_ifp_detached(const struct in_multi *);
162 #endif
163 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
164 static void inm_purge(struct in_multi *);
165 static void inm_reap(struct in_multi *);
166 static void inm_release(struct in_multi *);
167 static struct ip_moptions *
168 inp_findmoptions(struct inpcb *);
169 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
170 static int inp_join_group(struct inpcb *, struct sockopt *);
171 static int inp_leave_group(struct inpcb *, struct sockopt *);
172 static struct ifnet *
173 inp_lookup_mcast_ifp(const struct inpcb *,
174 const struct sockaddr_in *, const struct in_addr);
175 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
176 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
177 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
178 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
179
180 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
181 "IPv4 multicast");
182
183 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
184 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
185 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
186 "Max source filters per group");
187
188 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
189 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
190 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
191 "Max source filters per socket");
192
193 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
194 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
195 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
196
197 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
198 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
199 "Per-interface stack-wide source filters");
200
201 #ifdef KTR
202 /*
203 * Inline function which wraps assertions for a valid ifp.
204 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
205 * is detached.
206 */
207 static int __inline
inm_is_ifp_detached(const struct in_multi * inm)208 inm_is_ifp_detached(const struct in_multi *inm)
209 {
210 struct ifnet *ifp;
211
212 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
213 ifp = inm->inm_ifma->ifma_ifp;
214 if (ifp != NULL) {
215 /*
216 * Sanity check that netinet's notion of ifp is the
217 * same as net's.
218 */
219 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
220 }
221
222 return (ifp == NULL);
223 }
224 #endif
225
226 static struct grouptask free_gtask;
227 static struct in_multi_head inm_free_list;
228 static void inm_release_task(void *arg __unused);
inm_init(void)229 static void inm_init(void)
230 {
231 SLIST_INIT(&inm_free_list);
232 taskqgroup_config_gtask_init(NULL, &free_gtask, inm_release_task, "inm release task");
233 }
234
235 #ifdef EARLY_AP_STARTUP
236 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
237 inm_init, NULL);
238 #else
239 SYSINIT(inm_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_FIRST,
240 inm_init, NULL);
241 #endif
242
243
244 void
inm_release_list_deferred(struct in_multi_head * inmh)245 inm_release_list_deferred(struct in_multi_head *inmh)
246 {
247
248 if (SLIST_EMPTY(inmh))
249 return;
250 mtx_lock(&in_multi_free_mtx);
251 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
252 mtx_unlock(&in_multi_free_mtx);
253 GROUPTASK_ENQUEUE(&free_gtask);
254 }
255
256 void
inm_disconnect(struct in_multi * inm)257 inm_disconnect(struct in_multi *inm)
258 {
259 struct ifnet *ifp;
260 struct ifmultiaddr *ifma, *ll_ifma;
261
262 ifp = inm->inm_ifp;
263 IF_ADDR_WLOCK_ASSERT(ifp);
264 ifma = inm->inm_ifma;
265
266 if_ref(ifp);
267 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
268 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
269 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
270 }
271 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
272 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
273 MPASS(ifma != ll_ifma);
274 ifma->ifma_llifma = NULL;
275 MPASS(ll_ifma->ifma_llifma == NULL);
276 MPASS(ll_ifma->ifma_ifp == ifp);
277 if (--ll_ifma->ifma_refcount == 0) {
278 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
279 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
280 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
281 }
282 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
283 if_freemulti(ll_ifma);
284 ifma_restart = true;
285 }
286 }
287 }
288
289 void
inm_release_deferred(struct in_multi * inm)290 inm_release_deferred(struct in_multi *inm)
291 {
292 struct in_multi_head tmp;
293
294 IN_MULTI_LIST_LOCK_ASSERT();
295 MPASS(inm->inm_refcount > 0);
296 if (--inm->inm_refcount == 0) {
297 SLIST_INIT(&tmp);
298 inm_disconnect(inm);
299 inm->inm_ifma->ifma_protospec = NULL;
300 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
301 inm_release_list_deferred(&tmp);
302 }
303 }
304
305 static void
inm_release_task(void * arg __unused)306 inm_release_task(void *arg __unused)
307 {
308 struct in_multi_head inm_free_tmp;
309 struct in_multi *inm, *tinm;
310
311 SLIST_INIT(&inm_free_tmp);
312 mtx_lock(&in_multi_free_mtx);
313 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
314 mtx_unlock(&in_multi_free_mtx);
315 IN_MULTI_LOCK();
316 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
317 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
318 MPASS(inm);
319 inm_release(inm);
320 }
321 IN_MULTI_UNLOCK();
322 }
323
324 /*
325 * Initialize an in_mfilter structure to a known state at t0, t1
326 * with an empty source filter list.
327 */
328 static __inline void
imf_init(struct in_mfilter * imf,const int st0,const int st1)329 imf_init(struct in_mfilter *imf, const int st0, const int st1)
330 {
331 memset(imf, 0, sizeof(struct in_mfilter));
332 RB_INIT(&imf->imf_sources);
333 imf->imf_st[0] = st0;
334 imf->imf_st[1] = st1;
335 }
336
337 struct in_mfilter *
ip_mfilter_alloc(const int mflags,const int st0,const int st1)338 ip_mfilter_alloc(const int mflags, const int st0, const int st1)
339 {
340 struct in_mfilter *imf;
341
342 imf = malloc(sizeof(*imf), M_INMFILTER, mflags);
343 if (imf != NULL)
344 imf_init(imf, st0, st1);
345
346 return (imf);
347 }
348
349 void
ip_mfilter_free(struct in_mfilter * imf)350 ip_mfilter_free(struct in_mfilter *imf)
351 {
352
353 imf_purge(imf);
354 free(imf, M_INMFILTER);
355 }
356
357 /*
358 * Function for looking up an in_multi record for an IPv4 multicast address
359 * on a given interface. ifp must be valid. If no record found, return NULL.
360 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
361 */
362 struct in_multi *
inm_lookup_locked(struct ifnet * ifp,const struct in_addr ina)363 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
364 {
365 struct ifmultiaddr *ifma;
366 struct in_multi *inm;
367
368 IN_MULTI_LIST_LOCK_ASSERT();
369 IF_ADDR_LOCK_ASSERT(ifp);
370
371 inm = NULL;
372 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
373 if (ifma->ifma_addr->sa_family != AF_INET ||
374 ifma->ifma_protospec == NULL)
375 continue;
376 inm = (struct in_multi *)ifma->ifma_protospec;
377 if (inm->inm_addr.s_addr == ina.s_addr)
378 break;
379 inm = NULL;
380 }
381 return (inm);
382 }
383
384 /*
385 * Wrapper for inm_lookup_locked().
386 * The IF_ADDR_LOCK will be taken on ifp and released on return.
387 */
388 struct in_multi *
inm_lookup(struct ifnet * ifp,const struct in_addr ina)389 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
390 {
391 struct in_multi *inm;
392
393 IN_MULTI_LIST_LOCK_ASSERT();
394 IF_ADDR_RLOCK(ifp);
395 inm = inm_lookup_locked(ifp, ina);
396 IF_ADDR_RUNLOCK(ifp);
397
398 return (inm);
399 }
400
401 /*
402 * Find an IPv4 multicast group entry for this ip_moptions instance
403 * which matches the specified group, and optionally an interface.
404 * Return its index into the array, or -1 if not found.
405 */
406 static struct in_mfilter *
imo_match_group(const struct ip_moptions * imo,const struct ifnet * ifp,const struct sockaddr * group)407 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
408 const struct sockaddr *group)
409 {
410 const struct sockaddr_in *gsin;
411 struct in_mfilter *imf;
412 struct in_multi *inm;
413
414 gsin = (const struct sockaddr_in *)group;
415
416 IP_MFILTER_FOREACH(imf, &imo->imo_head) {
417 inm = imf->imf_inm;
418 if (inm == NULL)
419 continue;
420 if ((ifp == NULL || (inm->inm_ifp == ifp)) &&
421 in_hosteq(inm->inm_addr, gsin->sin_addr)) {
422 break;
423 }
424 }
425 return (imf);
426 }
427
428 /*
429 * Find an IPv4 multicast source entry for this imo which matches
430 * the given group index for this socket, and source address.
431 *
432 * NOTE: This does not check if the entry is in-mode, merely if
433 * it exists, which may not be the desired behaviour.
434 */
435 static struct in_msource *
imo_match_source(struct in_mfilter * imf,const struct sockaddr * src)436 imo_match_source(struct in_mfilter *imf, const struct sockaddr *src)
437 {
438 struct ip_msource find;
439 struct ip_msource *ims;
440 const sockunion_t *psa;
441
442 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
443
444 /* Source trees are keyed in host byte order. */
445 psa = (const sockunion_t *)src;
446 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
447 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
448
449 return ((struct in_msource *)ims);
450 }
451
452 /*
453 * Perform filtering for multicast datagrams on a socket by group and source.
454 *
455 * Returns 0 if a datagram should be allowed through, or various error codes
456 * if the socket was not a member of the group, or the source was muted, etc.
457 */
458 int
imo_multi_filter(const struct ip_moptions * imo,const struct ifnet * ifp,const struct sockaddr * group,const struct sockaddr * src)459 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
460 const struct sockaddr *group, const struct sockaddr *src)
461 {
462 struct in_mfilter *imf;
463 struct in_msource *ims;
464 int mode;
465
466 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
467
468 imf = imo_match_group(imo, ifp, group);
469 if (imf == NULL)
470 return (MCAST_NOTGMEMBER);
471
472 /*
473 * Check if the source was included in an (S,G) join.
474 * Allow reception on exclusive memberships by default,
475 * reject reception on inclusive memberships by default.
476 * Exclude source only if an in-mode exclude filter exists.
477 * Include source only if an in-mode include filter exists.
478 * NOTE: We are comparing group state here at IGMP t1 (now)
479 * with socket-layer t0 (since last downcall).
480 */
481 mode = imf->imf_st[1];
482 ims = imo_match_source(imf, src);
483
484 if ((ims == NULL && mode == MCAST_INCLUDE) ||
485 (ims != NULL && ims->imsl_st[0] != mode))
486 return (MCAST_NOTSMEMBER);
487
488 return (MCAST_PASS);
489 }
490
491 /*
492 * Find and return a reference to an in_multi record for (ifp, group),
493 * and bump its reference count.
494 * If one does not exist, try to allocate it, and update link-layer multicast
495 * filters on ifp to listen for group.
496 * Assumes the IN_MULTI lock is held across the call.
497 * Return 0 if successful, otherwise return an appropriate error code.
498 */
499 static int
in_getmulti(struct ifnet * ifp,const struct in_addr * group,struct in_multi ** pinm)500 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
501 struct in_multi **pinm)
502 {
503 struct sockaddr_in gsin;
504 struct ifmultiaddr *ifma;
505 struct in_ifinfo *ii;
506 struct in_multi *inm;
507 int error;
508
509 IN_MULTI_LOCK_ASSERT();
510
511 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
512 IN_MULTI_LIST_LOCK();
513 inm = inm_lookup(ifp, *group);
514 if (inm != NULL) {
515 /*
516 * If we already joined this group, just bump the
517 * refcount and return it.
518 */
519 KASSERT(inm->inm_refcount >= 1,
520 ("%s: bad refcount %d", __func__, inm->inm_refcount));
521 inm_acquire_locked(inm);
522 *pinm = inm;
523 }
524 IN_MULTI_LIST_UNLOCK();
525 if (inm != NULL)
526 return (0);
527
528 memset(&gsin, 0, sizeof(gsin));
529 gsin.sin_family = AF_INET;
530 gsin.sin_len = sizeof(struct sockaddr_in);
531 gsin.sin_addr = *group;
532
533 /*
534 * Check if a link-layer group is already associated
535 * with this network-layer group on the given ifnet.
536 */
537 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
538 if (error != 0)
539 return (error);
540
541 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
542 IN_MULTI_LIST_LOCK();
543 IF_ADDR_WLOCK(ifp);
544
545 /*
546 * If something other than netinet is occupying the link-layer
547 * group, print a meaningful error message and back out of
548 * the allocation.
549 * Otherwise, bump the refcount on the existing network-layer
550 * group association and return it.
551 */
552 if (ifma->ifma_protospec != NULL) {
553 inm = (struct in_multi *)ifma->ifma_protospec;
554 #ifdef INVARIANTS
555 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
556 __func__));
557 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
558 ("%s: ifma not AF_INET", __func__));
559 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
560 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
561 !in_hosteq(inm->inm_addr, *group)) {
562 char addrbuf[INET_ADDRSTRLEN];
563
564 panic("%s: ifma %p is inconsistent with %p (%s)",
565 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
566 }
567 #endif
568 inm_acquire_locked(inm);
569 *pinm = inm;
570 goto out_locked;
571 }
572
573 IF_ADDR_WLOCK_ASSERT(ifp);
574
575 /*
576 * A new in_multi record is needed; allocate and initialize it.
577 * We DO NOT perform an IGMP join as the in_ layer may need to
578 * push an initial source list down to IGMP to support SSM.
579 *
580 * The initial source filter state is INCLUDE, {} as per the RFC.
581 */
582 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
583 if (inm == NULL) {
584 IF_ADDR_WUNLOCK(ifp);
585 IN_MULTI_LIST_UNLOCK();
586 if_delmulti_ifma(ifma);
587 return (ENOMEM);
588 }
589 inm->inm_addr = *group;
590 inm->inm_ifp = ifp;
591 inm->inm_igi = ii->ii_igmp;
592 inm->inm_ifma = ifma;
593 inm->inm_refcount = 1;
594 inm->inm_state = IGMP_NOT_MEMBER;
595 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
596 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
597 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
598 RB_INIT(&inm->inm_srcs);
599
600 ifma->ifma_protospec = inm;
601
602 *pinm = inm;
603 out_locked:
604 IF_ADDR_WUNLOCK(ifp);
605 IN_MULTI_LIST_UNLOCK();
606 return (0);
607 }
608
609 /*
610 * Drop a reference to an in_multi record.
611 *
612 * If the refcount drops to 0, free the in_multi record and
613 * delete the underlying link-layer membership.
614 */
615 static void
inm_release(struct in_multi * inm)616 inm_release(struct in_multi *inm)
617 {
618 struct ifmultiaddr *ifma;
619 struct ifnet *ifp;
620
621 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
622 MPASS(inm->inm_refcount == 0);
623 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
624
625 ifma = inm->inm_ifma;
626 ifp = inm->inm_ifp;
627
628 /* XXX this access is not covered by IF_ADDR_LOCK */
629 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
630 if (ifp != NULL) {
631 CURVNET_SET(ifp->if_vnet);
632 inm_purge(inm);
633 free(inm, M_IPMADDR);
634 if_delmulti_ifma_flags(ifma, 1);
635 CURVNET_RESTORE();
636 if_rele(ifp);
637 } else {
638 inm_purge(inm);
639 free(inm, M_IPMADDR);
640 if_delmulti_ifma_flags(ifma, 1);
641 }
642 }
643
644 /*
645 * Clear recorded source entries for a group.
646 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
647 * FIXME: Should reap.
648 */
649 void
inm_clear_recorded(struct in_multi * inm)650 inm_clear_recorded(struct in_multi *inm)
651 {
652 struct ip_msource *ims;
653
654 IN_MULTI_LIST_LOCK_ASSERT();
655
656 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
657 if (ims->ims_stp) {
658 ims->ims_stp = 0;
659 --inm->inm_st[1].iss_rec;
660 }
661 }
662 KASSERT(inm->inm_st[1].iss_rec == 0,
663 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
664 }
665
666 /*
667 * Record a source as pending for a Source-Group IGMPv3 query.
668 * This lives here as it modifies the shared tree.
669 *
670 * inm is the group descriptor.
671 * naddr is the address of the source to record in network-byte order.
672 *
673 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
674 * lazy-allocate a source node in response to an SG query.
675 * Otherwise, no allocation is performed. This saves some memory
676 * with the trade-off that the source will not be reported to the
677 * router if joined in the window between the query response and
678 * the group actually being joined on the local host.
679 *
680 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
681 * This turns off the allocation of a recorded source entry if
682 * the group has not been joined.
683 *
684 * Return 0 if the source didn't exist or was already marked as recorded.
685 * Return 1 if the source was marked as recorded by this function.
686 * Return <0 if any error occurred (negated errno code).
687 */
688 int
inm_record_source(struct in_multi * inm,const in_addr_t naddr)689 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
690 {
691 struct ip_msource find;
692 struct ip_msource *ims, *nims;
693
694 IN_MULTI_LIST_LOCK_ASSERT();
695
696 find.ims_haddr = ntohl(naddr);
697 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
698 if (ims && ims->ims_stp)
699 return (0);
700 if (ims == NULL) {
701 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
702 return (-ENOSPC);
703 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
704 M_NOWAIT | M_ZERO);
705 if (nims == NULL)
706 return (-ENOMEM);
707 nims->ims_haddr = find.ims_haddr;
708 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
709 ++inm->inm_nsrc;
710 ims = nims;
711 }
712
713 /*
714 * Mark the source as recorded and update the recorded
715 * source count.
716 */
717 ++ims->ims_stp;
718 ++inm->inm_st[1].iss_rec;
719
720 return (1);
721 }
722
723 /*
724 * Return a pointer to an in_msource owned by an in_mfilter,
725 * given its source address.
726 * Lazy-allocate if needed. If this is a new entry its filter state is
727 * undefined at t0.
728 *
729 * imf is the filter set being modified.
730 * haddr is the source address in *host* byte-order.
731 *
732 * SMPng: May be called with locks held; malloc must not block.
733 */
734 static int
imf_get_source(struct in_mfilter * imf,const struct sockaddr_in * psin,struct in_msource ** plims)735 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
736 struct in_msource **plims)
737 {
738 struct ip_msource find;
739 struct ip_msource *ims, *nims;
740 struct in_msource *lims;
741 int error;
742
743 error = 0;
744 ims = NULL;
745 lims = NULL;
746
747 /* key is host byte order */
748 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
749 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
750 lims = (struct in_msource *)ims;
751 if (lims == NULL) {
752 if (imf->imf_nsrc == in_mcast_maxsocksrc)
753 return (ENOSPC);
754 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
755 M_NOWAIT | M_ZERO);
756 if (nims == NULL)
757 return (ENOMEM);
758 lims = (struct in_msource *)nims;
759 lims->ims_haddr = find.ims_haddr;
760 lims->imsl_st[0] = MCAST_UNDEFINED;
761 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
762 ++imf->imf_nsrc;
763 }
764
765 *plims = lims;
766
767 return (error);
768 }
769
770 /*
771 * Graft a source entry into an existing socket-layer filter set,
772 * maintaining any required invariants and checking allocations.
773 *
774 * The source is marked as being in the new filter mode at t1.
775 *
776 * Return the pointer to the new node, otherwise return NULL.
777 */
778 static struct in_msource *
imf_graft(struct in_mfilter * imf,const uint8_t st1,const struct sockaddr_in * psin)779 imf_graft(struct in_mfilter *imf, const uint8_t st1,
780 const struct sockaddr_in *psin)
781 {
782 struct ip_msource *nims;
783 struct in_msource *lims;
784
785 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
786 M_NOWAIT | M_ZERO);
787 if (nims == NULL)
788 return (NULL);
789 lims = (struct in_msource *)nims;
790 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
791 lims->imsl_st[0] = MCAST_UNDEFINED;
792 lims->imsl_st[1] = st1;
793 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
794 ++imf->imf_nsrc;
795
796 return (lims);
797 }
798
799 /*
800 * Prune a source entry from an existing socket-layer filter set,
801 * maintaining any required invariants and checking allocations.
802 *
803 * The source is marked as being left at t1, it is not freed.
804 *
805 * Return 0 if no error occurred, otherwise return an errno value.
806 */
807 static int
imf_prune(struct in_mfilter * imf,const struct sockaddr_in * psin)808 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
809 {
810 struct ip_msource find;
811 struct ip_msource *ims;
812 struct in_msource *lims;
813
814 /* key is host byte order */
815 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
816 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
817 if (ims == NULL)
818 return (ENOENT);
819 lims = (struct in_msource *)ims;
820 lims->imsl_st[1] = MCAST_UNDEFINED;
821 return (0);
822 }
823
824 /*
825 * Revert socket-layer filter set deltas at t1 to t0 state.
826 */
827 static void
imf_rollback(struct in_mfilter * imf)828 imf_rollback(struct in_mfilter *imf)
829 {
830 struct ip_msource *ims, *tims;
831 struct in_msource *lims;
832
833 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
834 lims = (struct in_msource *)ims;
835 if (lims->imsl_st[0] == lims->imsl_st[1]) {
836 /* no change at t1 */
837 continue;
838 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
839 /* revert change to existing source at t1 */
840 lims->imsl_st[1] = lims->imsl_st[0];
841 } else {
842 /* revert source added t1 */
843 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
844 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
845 free(ims, M_INMFILTER);
846 imf->imf_nsrc--;
847 }
848 }
849 imf->imf_st[1] = imf->imf_st[0];
850 }
851
852 /*
853 * Mark socket-layer filter set as INCLUDE {} at t1.
854 */
855 static void
imf_leave(struct in_mfilter * imf)856 imf_leave(struct in_mfilter *imf)
857 {
858 struct ip_msource *ims;
859 struct in_msource *lims;
860
861 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
862 lims = (struct in_msource *)ims;
863 lims->imsl_st[1] = MCAST_UNDEFINED;
864 }
865 imf->imf_st[1] = MCAST_INCLUDE;
866 }
867
868 /*
869 * Mark socket-layer filter set deltas as committed.
870 */
871 static void
imf_commit(struct in_mfilter * imf)872 imf_commit(struct in_mfilter *imf)
873 {
874 struct ip_msource *ims;
875 struct in_msource *lims;
876
877 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
878 lims = (struct in_msource *)ims;
879 lims->imsl_st[0] = lims->imsl_st[1];
880 }
881 imf->imf_st[0] = imf->imf_st[1];
882 }
883
884 /*
885 * Reap unreferenced sources from socket-layer filter set.
886 */
887 static void
imf_reap(struct in_mfilter * imf)888 imf_reap(struct in_mfilter *imf)
889 {
890 struct ip_msource *ims, *tims;
891 struct in_msource *lims;
892
893 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
894 lims = (struct in_msource *)ims;
895 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
896 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
897 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
898 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
899 free(ims, M_INMFILTER);
900 imf->imf_nsrc--;
901 }
902 }
903 }
904
905 /*
906 * Purge socket-layer filter set.
907 */
908 static void
imf_purge(struct in_mfilter * imf)909 imf_purge(struct in_mfilter *imf)
910 {
911 struct ip_msource *ims, *tims;
912
913 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
914 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
915 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
916 free(ims, M_INMFILTER);
917 imf->imf_nsrc--;
918 }
919 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
920 KASSERT(RB_EMPTY(&imf->imf_sources),
921 ("%s: imf_sources not empty", __func__));
922 }
923
924 /*
925 * Look up a source filter entry for a multicast group.
926 *
927 * inm is the group descriptor to work with.
928 * haddr is the host-byte-order IPv4 address to look up.
929 * noalloc may be non-zero to suppress allocation of sources.
930 * *pims will be set to the address of the retrieved or allocated source.
931 *
932 * SMPng: NOTE: may be called with locks held.
933 * Return 0 if successful, otherwise return a non-zero error code.
934 */
935 static int
inm_get_source(struct in_multi * inm,const in_addr_t haddr,const int noalloc,struct ip_msource ** pims)936 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
937 const int noalloc, struct ip_msource **pims)
938 {
939 struct ip_msource find;
940 struct ip_msource *ims, *nims;
941
942 find.ims_haddr = haddr;
943 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
944 if (ims == NULL && !noalloc) {
945 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
946 return (ENOSPC);
947 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
948 M_NOWAIT | M_ZERO);
949 if (nims == NULL)
950 return (ENOMEM);
951 nims->ims_haddr = haddr;
952 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
953 ++inm->inm_nsrc;
954 ims = nims;
955 #ifdef KTR
956 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
957 haddr, ims);
958 #endif
959 }
960
961 *pims = ims;
962 return (0);
963 }
964
965 /*
966 * Merge socket-layer source into IGMP-layer source.
967 * If rollback is non-zero, perform the inverse of the merge.
968 */
969 static void
ims_merge(struct ip_msource * ims,const struct in_msource * lims,const int rollback)970 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
971 const int rollback)
972 {
973 int n = rollback ? -1 : 1;
974
975 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
976 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
977 __func__, n, ims->ims_haddr);
978 ims->ims_st[1].ex -= n;
979 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
980 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
981 __func__, n, ims->ims_haddr);
982 ims->ims_st[1].in -= n;
983 }
984
985 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
986 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
987 __func__, n, ims->ims_haddr);
988 ims->ims_st[1].ex += n;
989 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
990 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
991 __func__, n, ims->ims_haddr);
992 ims->ims_st[1].in += n;
993 }
994 }
995
996 /*
997 * Atomically update the global in_multi state, when a membership's
998 * filter list is being updated in any way.
999 *
1000 * imf is the per-inpcb-membership group filter pointer.
1001 * A fake imf may be passed for in-kernel consumers.
1002 *
1003 * XXX This is a candidate for a set-symmetric-difference style loop
1004 * which would eliminate the repeated lookup from root of ims nodes,
1005 * as they share the same key space.
1006 *
1007 * If any error occurred this function will back out of refcounts
1008 * and return a non-zero value.
1009 */
1010 static int
inm_merge(struct in_multi * inm,struct in_mfilter * imf)1011 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1012 {
1013 struct ip_msource *ims, *nims;
1014 struct in_msource *lims;
1015 int schanged, error;
1016 int nsrc0, nsrc1;
1017
1018 schanged = 0;
1019 error = 0;
1020 nsrc1 = nsrc0 = 0;
1021 IN_MULTI_LIST_LOCK_ASSERT();
1022
1023 /*
1024 * Update the source filters first, as this may fail.
1025 * Maintain count of in-mode filters at t0, t1. These are
1026 * used to work out if we transition into ASM mode or not.
1027 * Maintain a count of source filters whose state was
1028 * actually modified by this operation.
1029 */
1030 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1031 lims = (struct in_msource *)ims;
1032 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1033 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1034 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1035 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1036 ++schanged;
1037 if (error)
1038 break;
1039 ims_merge(nims, lims, 0);
1040 }
1041 if (error) {
1042 struct ip_msource *bims;
1043
1044 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1045 lims = (struct in_msource *)ims;
1046 if (lims->imsl_st[0] == lims->imsl_st[1])
1047 continue;
1048 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1049 if (bims == NULL)
1050 continue;
1051 ims_merge(bims, lims, 1);
1052 }
1053 goto out_reap;
1054 }
1055
1056 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1057 __func__, nsrc0, nsrc1);
1058
1059 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1060 if (imf->imf_st[0] == imf->imf_st[1] &&
1061 imf->imf_st[1] == MCAST_INCLUDE) {
1062 if (nsrc1 == 0) {
1063 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1064 --inm->inm_st[1].iss_in;
1065 }
1066 }
1067
1068 /* Handle filter mode transition on socket. */
1069 if (imf->imf_st[0] != imf->imf_st[1]) {
1070 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1071 __func__, imf->imf_st[0], imf->imf_st[1]);
1072
1073 if (imf->imf_st[0] == 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[0] == MCAST_INCLUDE) {
1077 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1078 --inm->inm_st[1].iss_in;
1079 }
1080
1081 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1082 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1083 inm->inm_st[1].iss_ex++;
1084 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1085 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1086 inm->inm_st[1].iss_in++;
1087 }
1088 }
1089
1090 /*
1091 * Track inm filter state in terms of listener counts.
1092 * If there are any exclusive listeners, stack-wide
1093 * membership is exclusive.
1094 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1095 * If no listeners remain, state is undefined at t1,
1096 * and the IGMP lifecycle for this group should finish.
1097 */
1098 if (inm->inm_st[1].iss_ex > 0) {
1099 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1100 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1101 } else if (inm->inm_st[1].iss_in > 0) {
1102 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1103 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1104 } else {
1105 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1106 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1107 }
1108
1109 /* Decrement ASM listener count on transition out of ASM mode. */
1110 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1111 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1112 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1113 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1114 --inm->inm_st[1].iss_asm;
1115 }
1116 }
1117
1118 /* Increment ASM listener count on transition to ASM mode. */
1119 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1120 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1121 inm->inm_st[1].iss_asm++;
1122 }
1123
1124 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1125 inm_print(inm);
1126
1127 out_reap:
1128 if (schanged > 0) {
1129 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1130 inm_reap(inm);
1131 }
1132 return (error);
1133 }
1134
1135 /*
1136 * Mark an in_multi's filter set deltas as committed.
1137 * Called by IGMP after a state change has been enqueued.
1138 */
1139 void
inm_commit(struct in_multi * inm)1140 inm_commit(struct in_multi *inm)
1141 {
1142 struct ip_msource *ims;
1143
1144 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1145 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1146 inm_print(inm);
1147
1148 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1149 ims->ims_st[0] = ims->ims_st[1];
1150 }
1151 inm->inm_st[0] = inm->inm_st[1];
1152 }
1153
1154 /*
1155 * Reap unreferenced nodes from an in_multi's filter set.
1156 */
1157 static void
inm_reap(struct in_multi * inm)1158 inm_reap(struct in_multi *inm)
1159 {
1160 struct ip_msource *ims, *tims;
1161
1162 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1163 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1164 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1165 ims->ims_stp != 0)
1166 continue;
1167 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1168 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1169 free(ims, M_IPMSOURCE);
1170 inm->inm_nsrc--;
1171 }
1172 }
1173
1174 /*
1175 * Purge all source nodes from an in_multi's filter set.
1176 */
1177 static void
inm_purge(struct in_multi * inm)1178 inm_purge(struct in_multi *inm)
1179 {
1180 struct ip_msource *ims, *tims;
1181
1182 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1183 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1184 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1185 free(ims, M_IPMSOURCE);
1186 inm->inm_nsrc--;
1187 }
1188 }
1189
1190 /*
1191 * Join a multicast group; unlocked entry point.
1192 *
1193 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1194 * is not held. Fortunately, ifp is unlikely to have been detached
1195 * at this point, so we assume it's OK to recurse.
1196 */
1197 int
in_joingroup(struct ifnet * ifp,const struct in_addr * gina,struct in_mfilter * imf,struct in_multi ** pinm)1198 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1199 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1200 {
1201 int error;
1202
1203 IN_MULTI_LOCK();
1204 error = in_joingroup_locked(ifp, gina, imf, pinm);
1205 IN_MULTI_UNLOCK();
1206
1207 return (error);
1208 }
1209
1210 /*
1211 * Join a multicast group; real entry point.
1212 *
1213 * Only preserves atomicity at inm level.
1214 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1215 *
1216 * If the IGMP downcall fails, the group is not joined, and an error
1217 * code is returned.
1218 */
1219 int
in_joingroup_locked(struct ifnet * ifp,const struct in_addr * gina,struct in_mfilter * imf,struct in_multi ** pinm)1220 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1221 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1222 {
1223 struct in_mfilter timf;
1224 struct in_multi *inm;
1225 int error;
1226
1227 IN_MULTI_LOCK_ASSERT();
1228 IN_MULTI_LIST_UNLOCK_ASSERT();
1229
1230 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1231 ntohl(gina->s_addr), ifp, ifp->if_xname);
1232
1233 error = 0;
1234 inm = NULL;
1235
1236 /*
1237 * If no imf was specified (i.e. kernel consumer),
1238 * fake one up and assume it is an ASM join.
1239 */
1240 if (imf == NULL) {
1241 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1242 imf = &timf;
1243 }
1244
1245 error = in_getmulti(ifp, gina, &inm);
1246 if (error) {
1247 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1248 return (error);
1249 }
1250 IN_MULTI_LIST_LOCK();
1251 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1252 error = inm_merge(inm, imf);
1253 if (error) {
1254 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1255 goto out_inm_release;
1256 }
1257
1258 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1259 error = igmp_change_state(inm);
1260 if (error) {
1261 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1262 goto out_inm_release;
1263 }
1264
1265 out_inm_release:
1266 if (error) {
1267
1268 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1269 inm_release_deferred(inm);
1270 } else {
1271 *pinm = inm;
1272 }
1273 IN_MULTI_LIST_UNLOCK();
1274
1275 return (error);
1276 }
1277
1278 /*
1279 * Leave a multicast group; unlocked entry point.
1280 */
1281 int
in_leavegroup(struct in_multi * inm,struct in_mfilter * imf)1282 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1283 {
1284 int error;
1285
1286 IN_MULTI_LOCK();
1287 error = in_leavegroup_locked(inm, imf);
1288 IN_MULTI_UNLOCK();
1289
1290 return (error);
1291 }
1292
1293 /*
1294 * Leave a multicast group; real entry point.
1295 * All source filters will be expunged.
1296 *
1297 * Only preserves atomicity at inm level.
1298 *
1299 * Holding the write lock for the INP which contains imf
1300 * is highly advisable. We can't assert for it as imf does not
1301 * contain a back-pointer to the owning inp.
1302 *
1303 * Note: This is not the same as inm_release(*) as this function also
1304 * makes a state change downcall into IGMP.
1305 */
1306 int
in_leavegroup_locked(struct in_multi * inm,struct in_mfilter * imf)1307 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1308 {
1309 struct in_mfilter timf;
1310 int error;
1311
1312 error = 0;
1313
1314 IN_MULTI_LOCK_ASSERT();
1315 IN_MULTI_LIST_UNLOCK_ASSERT();
1316
1317 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1318 inm, ntohl(inm->inm_addr.s_addr),
1319 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1320 imf);
1321
1322 /*
1323 * If no imf was specified (i.e. kernel consumer),
1324 * fake one up and assume it is an ASM join.
1325 */
1326 if (imf == NULL) {
1327 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1328 imf = &timf;
1329 }
1330
1331 /*
1332 * Begin state merge transaction at IGMP layer.
1333 *
1334 * As this particular invocation should not cause any memory
1335 * to be allocated, and there is no opportunity to roll back
1336 * the transaction, it MUST NOT fail.
1337 */
1338 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1339 IN_MULTI_LIST_LOCK();
1340 error = inm_merge(inm, imf);
1341 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1342
1343 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1344 CURVNET_SET(inm->inm_ifp->if_vnet);
1345 error = igmp_change_state(inm);
1346 IF_ADDR_WLOCK(inm->inm_ifp);
1347 inm_release_deferred(inm);
1348 IF_ADDR_WUNLOCK(inm->inm_ifp);
1349 IN_MULTI_LIST_UNLOCK();
1350 CURVNET_RESTORE();
1351 if (error)
1352 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1353
1354 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1355
1356 return (error);
1357 }
1358
1359 /*#ifndef BURN_BRIDGES*/
1360 /*
1361 * Join an IPv4 multicast group in (*,G) exclusive mode.
1362 * The group must be a 224.0.0.0/24 link-scope group.
1363 * This KPI is for legacy kernel consumers only.
1364 */
1365 struct in_multi *
in_addmulti(struct in_addr * ap,struct ifnet * ifp)1366 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1367 {
1368 struct in_multi *pinm;
1369 int error;
1370 #ifdef INVARIANTS
1371 char addrbuf[INET_ADDRSTRLEN];
1372 #endif
1373
1374 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1375 ("%s: %s not in 224.0.0.0/24", __func__,
1376 inet_ntoa_r(*ap, addrbuf)));
1377
1378 error = in_joingroup(ifp, ap, NULL, &pinm);
1379 if (error != 0)
1380 pinm = NULL;
1381
1382 return (pinm);
1383 }
1384
1385 /*
1386 * Block or unblock an ASM multicast source on an inpcb.
1387 * This implements the delta-based API described in RFC 3678.
1388 *
1389 * The delta-based API applies only to exclusive-mode memberships.
1390 * An IGMP downcall will be performed.
1391 *
1392 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1393 *
1394 * Return 0 if successful, otherwise return an appropriate error code.
1395 */
1396 static int
inp_block_unblock_source(struct inpcb * inp,struct sockopt * sopt)1397 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1398 {
1399 struct group_source_req gsr;
1400 struct rm_priotracker in_ifa_tracker;
1401 sockunion_t *gsa, *ssa;
1402 struct ifnet *ifp;
1403 struct in_mfilter *imf;
1404 struct ip_moptions *imo;
1405 struct in_msource *ims;
1406 struct in_multi *inm;
1407 uint16_t fmode;
1408 int error, doblock;
1409
1410 ifp = NULL;
1411 error = 0;
1412 doblock = 0;
1413
1414 memset(&gsr, 0, sizeof(struct group_source_req));
1415 gsa = (sockunion_t *)&gsr.gsr_group;
1416 ssa = (sockunion_t *)&gsr.gsr_source;
1417
1418 switch (sopt->sopt_name) {
1419 case IP_BLOCK_SOURCE:
1420 case IP_UNBLOCK_SOURCE: {
1421 struct ip_mreq_source mreqs;
1422
1423 error = sooptcopyin(sopt, &mreqs,
1424 sizeof(struct ip_mreq_source),
1425 sizeof(struct ip_mreq_source));
1426 if (error)
1427 return (error);
1428
1429 gsa->sin.sin_family = AF_INET;
1430 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1431 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1432
1433 ssa->sin.sin_family = AF_INET;
1434 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1435 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1436
1437 if (!in_nullhost(mreqs.imr_interface)) {
1438 IN_IFADDR_RLOCK(&in_ifa_tracker);
1439 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1440 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1441 }
1442 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1443 doblock = 1;
1444
1445 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1446 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1447 break;
1448 }
1449
1450 case MCAST_BLOCK_SOURCE:
1451 case MCAST_UNBLOCK_SOURCE:
1452 error = sooptcopyin(sopt, &gsr,
1453 sizeof(struct group_source_req),
1454 sizeof(struct group_source_req));
1455 if (error)
1456 return (error);
1457
1458 if (gsa->sin.sin_family != AF_INET ||
1459 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1460 return (EINVAL);
1461
1462 if (ssa->sin.sin_family != AF_INET ||
1463 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1464 return (EINVAL);
1465
1466 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1467 return (EADDRNOTAVAIL);
1468
1469 ifp = ifnet_byindex(gsr.gsr_interface);
1470
1471 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1472 doblock = 1;
1473 break;
1474
1475 default:
1476 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1477 __func__, sopt->sopt_name);
1478 return (EOPNOTSUPP);
1479 break;
1480 }
1481
1482 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1483 return (EINVAL);
1484
1485 IN_MULTI_LOCK();
1486
1487 /*
1488 * Check if we are actually a member of this group.
1489 */
1490 imo = inp_findmoptions(inp);
1491 imf = imo_match_group(imo, ifp, &gsa->sa);
1492 if (imf == NULL) {
1493 error = EADDRNOTAVAIL;
1494 goto out_inp_locked;
1495 }
1496 inm = imf->imf_inm;
1497
1498 /*
1499 * Attempting to use the delta-based API on an
1500 * non exclusive-mode membership is an error.
1501 */
1502 fmode = imf->imf_st[0];
1503 if (fmode != MCAST_EXCLUDE) {
1504 error = EINVAL;
1505 goto out_inp_locked;
1506 }
1507
1508 /*
1509 * Deal with error cases up-front:
1510 * Asked to block, but already blocked; or
1511 * Asked to unblock, but nothing to unblock.
1512 * If adding a new block entry, allocate it.
1513 */
1514 ims = imo_match_source(imf, &ssa->sa);
1515 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1516 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1517 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1518 error = EADDRNOTAVAIL;
1519 goto out_inp_locked;
1520 }
1521
1522 INP_WLOCK_ASSERT(inp);
1523
1524 /*
1525 * Begin state merge transaction at socket layer.
1526 */
1527 if (doblock) {
1528 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1529 ims = imf_graft(imf, fmode, &ssa->sin);
1530 if (ims == NULL)
1531 error = ENOMEM;
1532 } else {
1533 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1534 error = imf_prune(imf, &ssa->sin);
1535 }
1536
1537 if (error) {
1538 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1539 goto out_imf_rollback;
1540 }
1541
1542 /*
1543 * Begin state merge transaction at IGMP layer.
1544 */
1545 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1546 IN_MULTI_LIST_LOCK();
1547 error = inm_merge(inm, imf);
1548 if (error) {
1549 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1550 IN_MULTI_LIST_UNLOCK();
1551 goto out_imf_rollback;
1552 }
1553
1554 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1555 error = igmp_change_state(inm);
1556 IN_MULTI_LIST_UNLOCK();
1557 if (error)
1558 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1559
1560 out_imf_rollback:
1561 if (error)
1562 imf_rollback(imf);
1563 else
1564 imf_commit(imf);
1565
1566 imf_reap(imf);
1567
1568 out_inp_locked:
1569 INP_WUNLOCK(inp);
1570 IN_MULTI_UNLOCK();
1571 return (error);
1572 }
1573
1574 /*
1575 * Given an inpcb, return its multicast options structure pointer. Accepts
1576 * an unlocked inpcb pointer, but will return it locked. May sleep.
1577 *
1578 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1579 * SMPng: NOTE: Returns with the INP write lock held.
1580 */
1581 static struct ip_moptions *
inp_findmoptions(struct inpcb * inp)1582 inp_findmoptions(struct inpcb *inp)
1583 {
1584 struct ip_moptions *imo;
1585
1586 INP_WLOCK(inp);
1587 if (inp->inp_moptions != NULL)
1588 return (inp->inp_moptions);
1589
1590 INP_WUNLOCK(inp);
1591
1592 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1593
1594 imo->imo_multicast_ifp = NULL;
1595 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1596 imo->imo_multicast_vif = -1;
1597 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1598 imo->imo_multicast_loop = in_mcast_loop;
1599 STAILQ_INIT(&imo->imo_head);
1600
1601 INP_WLOCK(inp);
1602 if (inp->inp_moptions != NULL) {
1603 free(imo, M_IPMOPTS);
1604 return (inp->inp_moptions);
1605 }
1606 inp->inp_moptions = imo;
1607 return (imo);
1608 }
1609
1610 static void
inp_gcmoptions(struct ip_moptions * imo)1611 inp_gcmoptions(struct ip_moptions *imo)
1612 {
1613 struct in_mfilter *imf;
1614 struct in_multi *inm;
1615 struct ifnet *ifp;
1616
1617 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1618 ip_mfilter_remove(&imo->imo_head, imf);
1619
1620 imf_leave(imf);
1621 if ((inm = imf->imf_inm) != NULL) {
1622 if ((ifp = inm->inm_ifp) != NULL) {
1623 CURVNET_SET(ifp->if_vnet);
1624 (void)in_leavegroup(inm, imf);
1625 CURVNET_RESTORE();
1626 } else {
1627 (void)in_leavegroup(inm, imf);
1628 }
1629 }
1630 ip_mfilter_free(imf);
1631 }
1632 free(imo, M_IPMOPTS);
1633 }
1634
1635 /*
1636 * Discard the IP multicast options (and source filters). To minimize
1637 * the amount of work done while holding locks such as the INP's
1638 * pcbinfo lock (which is used in the receive path), the free
1639 * operation is deferred to the epoch callback task.
1640 */
1641 void
inp_freemoptions(struct ip_moptions * imo)1642 inp_freemoptions(struct ip_moptions *imo)
1643 {
1644 if (imo == NULL)
1645 return;
1646 inp_gcmoptions(imo);
1647 }
1648
1649 /*
1650 * Atomically get source filters on a socket for an IPv4 multicast group.
1651 * Called with INP lock held; returns with lock released.
1652 */
1653 static int
inp_get_source_filters(struct inpcb * inp,struct sockopt * sopt)1654 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1655 {
1656 struct __msfilterreq msfr;
1657 sockunion_t *gsa;
1658 struct ifnet *ifp;
1659 struct ip_moptions *imo;
1660 struct in_mfilter *imf;
1661 struct ip_msource *ims;
1662 struct in_msource *lims;
1663 struct sockaddr_in *psin;
1664 struct sockaddr_storage *ptss;
1665 struct sockaddr_storage *tss;
1666 int error;
1667 size_t nsrcs, ncsrcs;
1668
1669 INP_WLOCK_ASSERT(inp);
1670
1671 imo = inp->inp_moptions;
1672 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1673
1674 INP_WUNLOCK(inp);
1675
1676 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1677 sizeof(struct __msfilterreq));
1678 if (error)
1679 return (error);
1680
1681 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1682 return (EINVAL);
1683
1684 ifp = ifnet_byindex(msfr.msfr_ifindex);
1685 if (ifp == NULL)
1686 return (EINVAL);
1687
1688 INP_WLOCK(inp);
1689
1690 /*
1691 * Lookup group on the socket.
1692 */
1693 gsa = (sockunion_t *)&msfr.msfr_group;
1694 imf = imo_match_group(imo, ifp, &gsa->sa);
1695 if (imf == NULL) {
1696 INP_WUNLOCK(inp);
1697 return (EADDRNOTAVAIL);
1698 }
1699
1700 /*
1701 * Ignore memberships which are in limbo.
1702 */
1703 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1704 INP_WUNLOCK(inp);
1705 return (EAGAIN);
1706 }
1707 msfr.msfr_fmode = imf->imf_st[1];
1708
1709 /*
1710 * If the user specified a buffer, copy out the source filter
1711 * entries to userland gracefully.
1712 * We only copy out the number of entries which userland
1713 * has asked for, but we always tell userland how big the
1714 * buffer really needs to be.
1715 */
1716 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1717 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1718 tss = NULL;
1719 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1720 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1721 M_TEMP, M_NOWAIT | M_ZERO);
1722 if (tss == NULL) {
1723 INP_WUNLOCK(inp);
1724 return (ENOBUFS);
1725 }
1726 }
1727
1728 /*
1729 * Count number of sources in-mode at t0.
1730 * If buffer space exists and remains, copy out source entries.
1731 */
1732 nsrcs = msfr.msfr_nsrcs;
1733 ncsrcs = 0;
1734 ptss = tss;
1735 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1736 lims = (struct in_msource *)ims;
1737 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1738 lims->imsl_st[0] != imf->imf_st[0])
1739 continue;
1740 ++ncsrcs;
1741 if (tss != NULL && nsrcs > 0) {
1742 psin = (struct sockaddr_in *)ptss;
1743 psin->sin_family = AF_INET;
1744 psin->sin_len = sizeof(struct sockaddr_in);
1745 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1746 psin->sin_port = 0;
1747 ++ptss;
1748 --nsrcs;
1749 }
1750 }
1751
1752 INP_WUNLOCK(inp);
1753
1754 if (tss != NULL) {
1755 error = copyout(tss, msfr.msfr_srcs,
1756 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1757 free(tss, M_TEMP);
1758 if (error)
1759 return (error);
1760 }
1761
1762 msfr.msfr_nsrcs = ncsrcs;
1763 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1764
1765 return (error);
1766 }
1767
1768 /*
1769 * Return the IP multicast options in response to user getsockopt().
1770 */
1771 int
inp_getmoptions(struct inpcb * inp,struct sockopt * sopt)1772 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1773 {
1774 struct rm_priotracker in_ifa_tracker;
1775 struct ip_mreqn mreqn;
1776 struct ip_moptions *imo;
1777 struct ifnet *ifp;
1778 struct in_ifaddr *ia;
1779 int error, optval;
1780 u_char coptval;
1781
1782 INP_WLOCK(inp);
1783 imo = inp->inp_moptions;
1784 /*
1785 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1786 * or is a divert socket, reject it.
1787 */
1788 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1789 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1790 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1791 INP_WUNLOCK(inp);
1792 return (EOPNOTSUPP);
1793 }
1794
1795 error = 0;
1796 switch (sopt->sopt_name) {
1797 case IP_MULTICAST_VIF:
1798 if (imo != NULL)
1799 optval = imo->imo_multicast_vif;
1800 else
1801 optval = -1;
1802 INP_WUNLOCK(inp);
1803 error = sooptcopyout(sopt, &optval, sizeof(int));
1804 break;
1805
1806 case IP_MULTICAST_IF:
1807 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1808 if (imo != NULL) {
1809 ifp = imo->imo_multicast_ifp;
1810 if (!in_nullhost(imo->imo_multicast_addr)) {
1811 mreqn.imr_address = imo->imo_multicast_addr;
1812 } else if (ifp != NULL) {
1813 mreqn.imr_ifindex = ifp->if_index;
1814 NET_EPOCH_ENTER();
1815 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1816 if (ia != NULL)
1817 mreqn.imr_address =
1818 IA_SIN(ia)->sin_addr;
1819 NET_EPOCH_EXIT();
1820 }
1821 }
1822 INP_WUNLOCK(inp);
1823 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1824 error = sooptcopyout(sopt, &mreqn,
1825 sizeof(struct ip_mreqn));
1826 } else {
1827 error = sooptcopyout(sopt, &mreqn.imr_address,
1828 sizeof(struct in_addr));
1829 }
1830 break;
1831
1832 case IP_MULTICAST_TTL:
1833 if (imo == NULL)
1834 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1835 else
1836 optval = coptval = imo->imo_multicast_ttl;
1837 INP_WUNLOCK(inp);
1838 if (sopt->sopt_valsize == sizeof(u_char))
1839 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1840 else
1841 error = sooptcopyout(sopt, &optval, sizeof(int));
1842 break;
1843
1844 case IP_MULTICAST_LOOP:
1845 if (imo == NULL)
1846 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1847 else
1848 optval = coptval = imo->imo_multicast_loop;
1849 INP_WUNLOCK(inp);
1850 if (sopt->sopt_valsize == sizeof(u_char))
1851 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1852 else
1853 error = sooptcopyout(sopt, &optval, sizeof(int));
1854 break;
1855
1856 case IP_MSFILTER:
1857 if (imo == NULL) {
1858 error = EADDRNOTAVAIL;
1859 INP_WUNLOCK(inp);
1860 } else {
1861 error = inp_get_source_filters(inp, sopt);
1862 }
1863 break;
1864
1865 default:
1866 INP_WUNLOCK(inp);
1867 error = ENOPROTOOPT;
1868 break;
1869 }
1870
1871 INP_UNLOCK_ASSERT(inp);
1872
1873 return (error);
1874 }
1875
1876 /*
1877 * Look up the ifnet to use for a multicast group membership,
1878 * given the IPv4 address of an interface, and the IPv4 group address.
1879 *
1880 * This routine exists to support legacy multicast applications
1881 * which do not understand that multicast memberships are scoped to
1882 * specific physical links in the networking stack, or which need
1883 * to join link-scope groups before IPv4 addresses are configured.
1884 *
1885 * If inp is non-NULL, use this socket's current FIB number for any
1886 * required FIB lookup.
1887 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1888 * and use its ifp; usually, this points to the default next-hop.
1889 *
1890 * If the FIB lookup fails, attempt to use the first non-loopback
1891 * interface with multicast capability in the system as a
1892 * last resort. The legacy IPv4 ASM API requires that we do
1893 * this in order to allow groups to be joined when the routing
1894 * table has not yet been populated during boot.
1895 *
1896 * Returns NULL if no ifp could be found.
1897 *
1898 * FUTURE: Implement IPv4 source-address selection.
1899 */
1900 static struct ifnet *
inp_lookup_mcast_ifp(const struct inpcb * inp,const struct sockaddr_in * gsin,const struct in_addr ina)1901 inp_lookup_mcast_ifp(const struct inpcb *inp,
1902 const struct sockaddr_in *gsin, const struct in_addr ina)
1903 {
1904 struct rm_priotracker in_ifa_tracker;
1905 struct ifnet *ifp;
1906 struct nhop4_basic nh4;
1907 uint32_t fibnum;
1908
1909 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1910 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1911 ("%s: not multicast", __func__));
1912
1913 ifp = NULL;
1914 if (!in_nullhost(ina)) {
1915 IN_IFADDR_RLOCK(&in_ifa_tracker);
1916 INADDR_TO_IFP(ina, ifp);
1917 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1918 } else {
1919 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1920 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1921 ifp = nh4.nh_ifp;
1922 else {
1923 struct in_ifaddr *ia;
1924 struct ifnet *mifp;
1925
1926 mifp = NULL;
1927 IN_IFADDR_RLOCK(&in_ifa_tracker);
1928 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1929 mifp = ia->ia_ifp;
1930 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1931 (mifp->if_flags & IFF_MULTICAST)) {
1932 ifp = mifp;
1933 break;
1934 }
1935 }
1936 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1937 }
1938 }
1939
1940 return (ifp);
1941 }
1942
1943 /*
1944 * Join an IPv4 multicast group, possibly with a source.
1945 */
1946 static int
inp_join_group(struct inpcb * inp,struct sockopt * sopt)1947 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1948 {
1949 struct group_source_req gsr;
1950 sockunion_t *gsa, *ssa;
1951 struct ifnet *ifp;
1952 struct in_mfilter *imf;
1953 struct ip_moptions *imo;
1954 struct in_multi *inm;
1955 struct in_msource *lims;
1956 int error, is_new;
1957
1958 ifp = NULL;
1959 lims = NULL;
1960 error = 0;
1961
1962 memset(&gsr, 0, sizeof(struct group_source_req));
1963 gsa = (sockunion_t *)&gsr.gsr_group;
1964 gsa->ss.ss_family = AF_UNSPEC;
1965 ssa = (sockunion_t *)&gsr.gsr_source;
1966 ssa->ss.ss_family = AF_UNSPEC;
1967
1968 switch (sopt->sopt_name) {
1969 case IP_ADD_MEMBERSHIP: {
1970 struct ip_mreqn mreqn;
1971
1972 if (sopt->sopt_valsize == sizeof(struct ip_mreqn))
1973 error = sooptcopyin(sopt, &mreqn,
1974 sizeof(struct ip_mreqn), sizeof(struct ip_mreqn));
1975 else
1976 error = sooptcopyin(sopt, &mreqn,
1977 sizeof(struct ip_mreq), sizeof(struct ip_mreq));
1978 if (error)
1979 return (error);
1980
1981 gsa->sin.sin_family = AF_INET;
1982 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1983 gsa->sin.sin_addr = mreqn.imr_multiaddr;
1984 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1985 return (EINVAL);
1986
1987 if (sopt->sopt_valsize == sizeof(struct ip_mreqn) &&
1988 mreqn.imr_ifindex != 0)
1989 ifp = ifnet_byindex(mreqn.imr_ifindex);
1990 else
1991 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1992 mreqn.imr_address);
1993 break;
1994 }
1995 case IP_ADD_SOURCE_MEMBERSHIP: {
1996 struct ip_mreq_source mreqs;
1997
1998 error = sooptcopyin(sopt, &mreqs, sizeof(struct ip_mreq_source),
1999 sizeof(struct ip_mreq_source));
2000 if (error)
2001 return (error);
2002
2003 gsa->sin.sin_family = ssa->sin.sin_family = AF_INET;
2004 gsa->sin.sin_len = ssa->sin.sin_len =
2005 sizeof(struct sockaddr_in);
2006
2007 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2008 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2009 return (EINVAL);
2010
2011 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2012
2013 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2014 mreqs.imr_interface);
2015 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2016 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2017 break;
2018 }
2019
2020 case MCAST_JOIN_GROUP:
2021 case MCAST_JOIN_SOURCE_GROUP:
2022 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2023 error = sooptcopyin(sopt, &gsr,
2024 sizeof(struct group_req),
2025 sizeof(struct group_req));
2026 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2027 error = sooptcopyin(sopt, &gsr,
2028 sizeof(struct group_source_req),
2029 sizeof(struct group_source_req));
2030 }
2031 if (error)
2032 return (error);
2033
2034 if (gsa->sin.sin_family != AF_INET ||
2035 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2036 return (EINVAL);
2037
2038 /*
2039 * Overwrite the port field if present, as the sockaddr
2040 * being copied in may be matched with a binary comparison.
2041 */
2042 gsa->sin.sin_port = 0;
2043 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2044 if (ssa->sin.sin_family != AF_INET ||
2045 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2046 return (EINVAL);
2047 ssa->sin.sin_port = 0;
2048 }
2049
2050 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2051 return (EINVAL);
2052
2053 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2054 return (EADDRNOTAVAIL);
2055 ifp = ifnet_byindex(gsr.gsr_interface);
2056 break;
2057
2058 default:
2059 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2060 __func__, sopt->sopt_name);
2061 return (EOPNOTSUPP);
2062 break;
2063 }
2064
2065 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2066 return (EADDRNOTAVAIL);
2067
2068 IN_MULTI_LOCK();
2069
2070 /*
2071 * Find the membership in the membership list.
2072 */
2073 imo = inp_findmoptions(inp);
2074 imf = imo_match_group(imo, ifp, &gsa->sa);
2075 if (imf == NULL) {
2076 is_new = 1;
2077 inm = NULL;
2078
2079 if (ip_mfilter_count(&imo->imo_head) >= IP_MAX_MEMBERSHIPS) {
2080 error = ENOMEM;
2081 goto out_inp_locked;
2082 }
2083 } else {
2084 is_new = 0;
2085 inm = imf->imf_inm;
2086
2087 if (ssa->ss.ss_family != AF_UNSPEC) {
2088 /*
2089 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2090 * is an error. On an existing inclusive membership,
2091 * it just adds the source to the filter list.
2092 */
2093 if (imf->imf_st[1] != MCAST_INCLUDE) {
2094 error = EINVAL;
2095 goto out_inp_locked;
2096 }
2097 /*
2098 * Throw out duplicates.
2099 *
2100 * XXX FIXME: This makes a naive assumption that
2101 * even if entries exist for *ssa in this imf,
2102 * they will be rejected as dupes, even if they
2103 * are not valid in the current mode (in-mode).
2104 *
2105 * in_msource is transactioned just as for anything
2106 * else in SSM -- but note naive use of inm_graft()
2107 * below for allocating new filter entries.
2108 *
2109 * This is only an issue if someone mixes the
2110 * full-state SSM API with the delta-based API,
2111 * which is discouraged in the relevant RFCs.
2112 */
2113 lims = imo_match_source(imf, &ssa->sa);
2114 if (lims != NULL /*&&
2115 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2116 error = EADDRNOTAVAIL;
2117 goto out_inp_locked;
2118 }
2119 } else {
2120 /*
2121 * MCAST_JOIN_GROUP on an existing exclusive
2122 * membership is an error; return EADDRINUSE
2123 * to preserve 4.4BSD API idempotence, and
2124 * avoid tedious detour to code below.
2125 * NOTE: This is bending RFC 3678 a bit.
2126 *
2127 * On an existing inclusive membership, this is also
2128 * an error; if you want to change filter mode,
2129 * you must use the userland API setsourcefilter().
2130 * XXX We don't reject this for imf in UNDEFINED
2131 * state at t1, because allocation of a filter
2132 * is atomic with allocation of a membership.
2133 */
2134 error = EINVAL;
2135 if (imf->imf_st[1] == MCAST_EXCLUDE)
2136 error = EADDRINUSE;
2137 goto out_inp_locked;
2138 }
2139 }
2140
2141 /*
2142 * Begin state merge transaction at socket layer.
2143 */
2144 INP_WLOCK_ASSERT(inp);
2145
2146 /*
2147 * Graft new source into filter list for this inpcb's
2148 * membership of the group. The in_multi may not have
2149 * been allocated yet if this is a new membership, however,
2150 * the in_mfilter slot will be allocated and must be initialized.
2151 *
2152 * Note: Grafting of exclusive mode filters doesn't happen
2153 * in this path.
2154 * XXX: Should check for non-NULL lims (node exists but may
2155 * not be in-mode) for interop with full-state API.
2156 */
2157 if (ssa->ss.ss_family != AF_UNSPEC) {
2158 /* Membership starts in IN mode */
2159 if (is_new) {
2160 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2161 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2162 if (imf == NULL) {
2163 error = ENOMEM;
2164 goto out_inp_locked;
2165 }
2166 } else {
2167 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2168 }
2169 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2170 if (lims == NULL) {
2171 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2172 __func__);
2173 error = ENOMEM;
2174 goto out_inp_locked;
2175 }
2176 } else {
2177 /* No address specified; Membership starts in EX mode */
2178 if (is_new) {
2179 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2180 imf = ip_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2181 if (imf == NULL) {
2182 error = ENOMEM;
2183 goto out_inp_locked;
2184 }
2185 }
2186 }
2187
2188 /*
2189 * Begin state merge transaction at IGMP layer.
2190 */
2191 if (is_new) {
2192 in_pcbref(inp);
2193 INP_WUNLOCK(inp);
2194
2195 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2196 &imf->imf_inm);
2197
2198 INP_WLOCK(inp);
2199 if (in_pcbrele_wlocked(inp)) {
2200 error = ENXIO;
2201 goto out_inp_unlocked;
2202 }
2203 if (error) {
2204 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2205 __func__);
2206 goto out_inp_locked;
2207 }
2208 inm_acquire(imf->imf_inm);
2209 } else {
2210 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2211 IN_MULTI_LIST_LOCK();
2212 error = inm_merge(inm, imf);
2213 if (error) {
2214 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2215 __func__);
2216 IN_MULTI_LIST_UNLOCK();
2217 imf_rollback(imf);
2218 imf_reap(imf);
2219 goto out_inp_locked;
2220 }
2221 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2222 error = igmp_change_state(inm);
2223 IN_MULTI_LIST_UNLOCK();
2224 if (error) {
2225 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2226 __func__);
2227 imf_rollback(imf);
2228 imf_reap(imf);
2229 goto out_inp_locked;
2230 }
2231 }
2232 if (is_new)
2233 ip_mfilter_insert(&imo->imo_head, imf);
2234
2235 imf_commit(imf);
2236 imf = NULL;
2237
2238 out_inp_locked:
2239 INP_WUNLOCK(inp);
2240 out_inp_unlocked:
2241 IN_MULTI_UNLOCK();
2242
2243 if (is_new && imf) {
2244 if (imf->imf_inm != NULL) {
2245 IN_MULTI_LIST_LOCK();
2246 inm_release_deferred(imf->imf_inm);
2247 IN_MULTI_LIST_UNLOCK();
2248 }
2249 ip_mfilter_free(imf);
2250 }
2251 return (error);
2252 }
2253
2254 /*
2255 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2256 */
2257 static int
inp_leave_group(struct inpcb * inp,struct sockopt * sopt)2258 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2259 {
2260 struct group_source_req gsr;
2261 struct ip_mreq_source mreqs;
2262 struct rm_priotracker in_ifa_tracker;
2263 sockunion_t *gsa, *ssa;
2264 struct ifnet *ifp;
2265 struct in_mfilter *imf;
2266 struct ip_moptions *imo;
2267 struct in_msource *ims;
2268 struct in_multi *inm;
2269 int error;
2270 bool is_final;
2271
2272 ifp = NULL;
2273 error = 0;
2274 is_final = true;
2275
2276 memset(&gsr, 0, sizeof(struct group_source_req));
2277 gsa = (sockunion_t *)&gsr.gsr_group;
2278 gsa->ss.ss_family = AF_UNSPEC;
2279 ssa = (sockunion_t *)&gsr.gsr_source;
2280 ssa->ss.ss_family = AF_UNSPEC;
2281
2282 switch (sopt->sopt_name) {
2283 case IP_DROP_MEMBERSHIP:
2284 case IP_DROP_SOURCE_MEMBERSHIP:
2285 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2286 error = sooptcopyin(sopt, &mreqs,
2287 sizeof(struct ip_mreq),
2288 sizeof(struct ip_mreq));
2289 /*
2290 * Swap interface and sourceaddr arguments,
2291 * as ip_mreq and ip_mreq_source are laid
2292 * out differently.
2293 */
2294 mreqs.imr_interface = mreqs.imr_sourceaddr;
2295 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2296 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2297 error = sooptcopyin(sopt, &mreqs,
2298 sizeof(struct ip_mreq_source),
2299 sizeof(struct ip_mreq_source));
2300 }
2301 if (error)
2302 return (error);
2303
2304 gsa->sin.sin_family = AF_INET;
2305 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2306 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2307
2308 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2309 ssa->sin.sin_family = AF_INET;
2310 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2311 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2312 }
2313
2314 /*
2315 * Attempt to look up hinted ifp from interface address.
2316 * Fallthrough with null ifp iff lookup fails, to
2317 * preserve 4.4BSD mcast API idempotence.
2318 * XXX NOTE WELL: The RFC 3678 API is preferred because
2319 * using an IPv4 address as a key is racy.
2320 */
2321 if (!in_nullhost(mreqs.imr_interface)) {
2322 IN_IFADDR_RLOCK(&in_ifa_tracker);
2323 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2324 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2325 }
2326 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2327 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2328
2329 break;
2330
2331 case MCAST_LEAVE_GROUP:
2332 case MCAST_LEAVE_SOURCE_GROUP:
2333 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2334 error = sooptcopyin(sopt, &gsr,
2335 sizeof(struct group_req),
2336 sizeof(struct group_req));
2337 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2338 error = sooptcopyin(sopt, &gsr,
2339 sizeof(struct group_source_req),
2340 sizeof(struct group_source_req));
2341 }
2342 if (error)
2343 return (error);
2344
2345 if (gsa->sin.sin_family != AF_INET ||
2346 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2347 return (EINVAL);
2348
2349 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2350 if (ssa->sin.sin_family != AF_INET ||
2351 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2352 return (EINVAL);
2353 }
2354
2355 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2356 return (EADDRNOTAVAIL);
2357
2358 ifp = ifnet_byindex(gsr.gsr_interface);
2359
2360 if (ifp == NULL)
2361 return (EADDRNOTAVAIL);
2362 break;
2363
2364 default:
2365 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2366 __func__, sopt->sopt_name);
2367 return (EOPNOTSUPP);
2368 break;
2369 }
2370
2371 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2372 return (EINVAL);
2373
2374 IN_MULTI_LOCK();
2375
2376 /*
2377 * Find the membership in the membership list.
2378 */
2379 imo = inp_findmoptions(inp);
2380 imf = imo_match_group(imo, ifp, &gsa->sa);
2381 if (imf == NULL) {
2382 error = EADDRNOTAVAIL;
2383 goto out_inp_locked;
2384 }
2385 inm = imf->imf_inm;
2386
2387 if (ssa->ss.ss_family != AF_UNSPEC)
2388 is_final = false;
2389
2390 /*
2391 * Begin state merge transaction at socket layer.
2392 */
2393 INP_WLOCK_ASSERT(inp);
2394
2395 /*
2396 * If we were instructed only to leave a given source, do so.
2397 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2398 */
2399 if (is_final) {
2400 ip_mfilter_remove(&imo->imo_head, imf);
2401 imf_leave(imf);
2402 } else {
2403 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2404 error = EADDRNOTAVAIL;
2405 goto out_inp_locked;
2406 }
2407 ims = imo_match_source(imf, &ssa->sa);
2408 if (ims == NULL) {
2409 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2410 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2411 error = EADDRNOTAVAIL;
2412 goto out_inp_locked;
2413 }
2414 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2415 error = imf_prune(imf, &ssa->sin);
2416 if (error) {
2417 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2418 __func__);
2419 goto out_inp_locked;
2420 }
2421 }
2422
2423 /*
2424 * Begin state merge transaction at IGMP layer.
2425 */
2426 if (!is_final) {
2427 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2428 IN_MULTI_LIST_LOCK();
2429 error = inm_merge(inm, imf);
2430 if (error) {
2431 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2432 __func__);
2433 IN_MULTI_LIST_UNLOCK();
2434 imf_rollback(imf);
2435 imf_reap(imf);
2436 goto out_inp_locked;
2437 }
2438
2439 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2440 error = igmp_change_state(inm);
2441 IN_MULTI_LIST_UNLOCK();
2442 if (error) {
2443 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2444 __func__);
2445 imf_rollback(imf);
2446 imf_reap(imf);
2447 goto out_inp_locked;
2448 }
2449 }
2450 imf_commit(imf);
2451 imf_reap(imf);
2452
2453 out_inp_locked:
2454 INP_WUNLOCK(inp);
2455
2456 if (is_final && imf) {
2457 /*
2458 * Give up the multicast address record to which
2459 * the membership points.
2460 */
2461 (void) in_leavegroup_locked(imf->imf_inm, imf);
2462 ip_mfilter_free(imf);
2463 }
2464
2465 IN_MULTI_UNLOCK();
2466 return (error);
2467 }
2468
2469 /*
2470 * Select the interface for transmitting IPv4 multicast datagrams.
2471 *
2472 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2473 * may be passed to this socket option. An address of INADDR_ANY or an
2474 * interface index of 0 is used to remove a previous selection.
2475 * When no interface is selected, one is chosen for every send.
2476 */
2477 static int
inp_set_multicast_if(struct inpcb * inp,struct sockopt * sopt)2478 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2479 {
2480 struct rm_priotracker in_ifa_tracker;
2481 struct in_addr addr;
2482 struct ip_mreqn mreqn;
2483 struct ifnet *ifp;
2484 struct ip_moptions *imo;
2485 int error;
2486
2487 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2488 /*
2489 * An interface index was specified using the
2490 * Linux-derived ip_mreqn structure.
2491 */
2492 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2493 sizeof(struct ip_mreqn));
2494 if (error)
2495 return (error);
2496
2497 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2498 return (EINVAL);
2499
2500 if (mreqn.imr_ifindex == 0) {
2501 ifp = NULL;
2502 } else {
2503 ifp = ifnet_byindex(mreqn.imr_ifindex);
2504 if (ifp == NULL)
2505 return (EADDRNOTAVAIL);
2506 }
2507 } else {
2508 /*
2509 * An interface was specified by IPv4 address.
2510 * This is the traditional BSD usage.
2511 */
2512 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2513 sizeof(struct in_addr));
2514 if (error)
2515 return (error);
2516 if (in_nullhost(addr)) {
2517 ifp = NULL;
2518 } else {
2519 IN_IFADDR_RLOCK(&in_ifa_tracker);
2520 INADDR_TO_IFP(addr, ifp);
2521 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2522 if (ifp == NULL)
2523 return (EADDRNOTAVAIL);
2524 }
2525 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2526 ntohl(addr.s_addr));
2527 }
2528
2529 /* Reject interfaces which do not support multicast. */
2530 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2531 return (EOPNOTSUPP);
2532
2533 imo = inp_findmoptions(inp);
2534 imo->imo_multicast_ifp = ifp;
2535 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2536 INP_WUNLOCK(inp);
2537
2538 return (0);
2539 }
2540
2541 /*
2542 * Atomically set source filters on a socket for an IPv4 multicast group.
2543 *
2544 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2545 */
2546 static int
inp_set_source_filters(struct inpcb * inp,struct sockopt * sopt)2547 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2548 {
2549 struct __msfilterreq msfr;
2550 sockunion_t *gsa;
2551 struct ifnet *ifp;
2552 struct in_mfilter *imf;
2553 struct ip_moptions *imo;
2554 struct in_multi *inm;
2555 int error;
2556
2557 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2558 sizeof(struct __msfilterreq));
2559 if (error)
2560 return (error);
2561
2562 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2563 return (ENOBUFS);
2564
2565 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2566 msfr.msfr_fmode != MCAST_INCLUDE))
2567 return (EINVAL);
2568
2569 if (msfr.msfr_group.ss_family != AF_INET ||
2570 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2571 return (EINVAL);
2572
2573 gsa = (sockunion_t *)&msfr.msfr_group;
2574 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2575 return (EINVAL);
2576
2577 gsa->sin.sin_port = 0; /* ignore port */
2578
2579 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2580 return (EADDRNOTAVAIL);
2581
2582 ifp = ifnet_byindex(msfr.msfr_ifindex);
2583 if (ifp == NULL)
2584 return (EADDRNOTAVAIL);
2585
2586 IN_MULTI_LOCK();
2587
2588 /*
2589 * Take the INP write lock.
2590 * Check if this socket is a member of this group.
2591 */
2592 imo = inp_findmoptions(inp);
2593 imf = imo_match_group(imo, ifp, &gsa->sa);
2594 if (imf == NULL) {
2595 error = EADDRNOTAVAIL;
2596 goto out_inp_locked;
2597 }
2598 inm = imf->imf_inm;
2599
2600 /*
2601 * Begin state merge transaction at socket layer.
2602 */
2603 INP_WLOCK_ASSERT(inp);
2604
2605 imf->imf_st[1] = msfr.msfr_fmode;
2606
2607 /*
2608 * Apply any new source filters, if present.
2609 * Make a copy of the user-space source vector so
2610 * that we may copy them with a single copyin. This
2611 * allows us to deal with page faults up-front.
2612 */
2613 if (msfr.msfr_nsrcs > 0) {
2614 struct in_msource *lims;
2615 struct sockaddr_in *psin;
2616 struct sockaddr_storage *kss, *pkss;
2617 int i;
2618
2619 INP_WUNLOCK(inp);
2620
2621 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2622 __func__, (unsigned long)msfr.msfr_nsrcs);
2623 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2624 M_TEMP, M_WAITOK);
2625 error = copyin(msfr.msfr_srcs, kss,
2626 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2627 if (error) {
2628 free(kss, M_TEMP);
2629 return (error);
2630 }
2631
2632 INP_WLOCK(inp);
2633
2634 /*
2635 * Mark all source filters as UNDEFINED at t1.
2636 * Restore new group filter mode, as imf_leave()
2637 * will set it to INCLUDE.
2638 */
2639 imf_leave(imf);
2640 imf->imf_st[1] = msfr.msfr_fmode;
2641
2642 /*
2643 * Update socket layer filters at t1, lazy-allocating
2644 * new entries. This saves a bunch of memory at the
2645 * cost of one RB_FIND() per source entry; duplicate
2646 * entries in the msfr_nsrcs vector are ignored.
2647 * If we encounter an error, rollback transaction.
2648 *
2649 * XXX This too could be replaced with a set-symmetric
2650 * difference like loop to avoid walking from root
2651 * every time, as the key space is common.
2652 */
2653 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2654 psin = (struct sockaddr_in *)pkss;
2655 if (psin->sin_family != AF_INET) {
2656 error = EAFNOSUPPORT;
2657 break;
2658 }
2659 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2660 error = EINVAL;
2661 break;
2662 }
2663 error = imf_get_source(imf, psin, &lims);
2664 if (error)
2665 break;
2666 lims->imsl_st[1] = imf->imf_st[1];
2667 }
2668 free(kss, M_TEMP);
2669 }
2670
2671 if (error)
2672 goto out_imf_rollback;
2673
2674 INP_WLOCK_ASSERT(inp);
2675
2676 /*
2677 * Begin state merge transaction at IGMP layer.
2678 */
2679 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2680 IN_MULTI_LIST_LOCK();
2681 error = inm_merge(inm, imf);
2682 if (error) {
2683 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2684 IN_MULTI_LIST_UNLOCK();
2685 goto out_imf_rollback;
2686 }
2687
2688 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2689 error = igmp_change_state(inm);
2690 IN_MULTI_LIST_UNLOCK();
2691 if (error)
2692 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2693
2694 out_imf_rollback:
2695 if (error)
2696 imf_rollback(imf);
2697 else
2698 imf_commit(imf);
2699
2700 imf_reap(imf);
2701
2702 out_inp_locked:
2703 INP_WUNLOCK(inp);
2704 IN_MULTI_UNLOCK();
2705 return (error);
2706 }
2707
2708 /*
2709 * Set the IP multicast options in response to user setsockopt().
2710 *
2711 * Many of the socket options handled in this function duplicate the
2712 * functionality of socket options in the regular unicast API. However,
2713 * it is not possible to merge the duplicate code, because the idempotence
2714 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2715 * the effects of these options must be treated as separate and distinct.
2716 *
2717 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2718 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2719 * is refactored to no longer use vifs.
2720 */
2721 int
inp_setmoptions(struct inpcb * inp,struct sockopt * sopt)2722 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2723 {
2724 struct ip_moptions *imo;
2725 int error;
2726
2727 error = 0;
2728
2729 /*
2730 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2731 * or is a divert socket, reject it.
2732 */
2733 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2734 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2735 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2736 return (EOPNOTSUPP);
2737
2738 switch (sopt->sopt_name) {
2739 case IP_MULTICAST_VIF: {
2740 int vifi;
2741 /*
2742 * Select a multicast VIF for transmission.
2743 * Only useful if multicast forwarding is active.
2744 */
2745 if (legal_vif_num == NULL) {
2746 error = EOPNOTSUPP;
2747 break;
2748 }
2749 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2750 if (error)
2751 break;
2752 if (!legal_vif_num(vifi) && (vifi != -1)) {
2753 error = EINVAL;
2754 break;
2755 }
2756 imo = inp_findmoptions(inp);
2757 imo->imo_multicast_vif = vifi;
2758 INP_WUNLOCK(inp);
2759 break;
2760 }
2761
2762 case IP_MULTICAST_IF:
2763 error = inp_set_multicast_if(inp, sopt);
2764 break;
2765
2766 case IP_MULTICAST_TTL: {
2767 u_char ttl;
2768
2769 /*
2770 * Set the IP time-to-live for outgoing multicast packets.
2771 * The original multicast API required a char argument,
2772 * which is inconsistent with the rest of the socket API.
2773 * We allow either a char or an int.
2774 */
2775 if (sopt->sopt_valsize == sizeof(u_char)) {
2776 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2777 sizeof(u_char));
2778 if (error)
2779 break;
2780 } else {
2781 u_int ittl;
2782
2783 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2784 sizeof(u_int));
2785 if (error)
2786 break;
2787 if (ittl > 255) {
2788 error = EINVAL;
2789 break;
2790 }
2791 ttl = (u_char)ittl;
2792 }
2793 imo = inp_findmoptions(inp);
2794 imo->imo_multicast_ttl = ttl;
2795 INP_WUNLOCK(inp);
2796 break;
2797 }
2798
2799 case IP_MULTICAST_LOOP: {
2800 u_char loop;
2801
2802 /*
2803 * Set the loopback flag for outgoing multicast packets.
2804 * Must be zero or one. The original multicast API required a
2805 * char argument, which is inconsistent with the rest
2806 * of the socket API. We allow either a char or an int.
2807 */
2808 if (sopt->sopt_valsize == sizeof(u_char)) {
2809 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2810 sizeof(u_char));
2811 if (error)
2812 break;
2813 } else {
2814 u_int iloop;
2815
2816 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2817 sizeof(u_int));
2818 if (error)
2819 break;
2820 loop = (u_char)iloop;
2821 }
2822 imo = inp_findmoptions(inp);
2823 imo->imo_multicast_loop = !!loop;
2824 INP_WUNLOCK(inp);
2825 break;
2826 }
2827
2828 case IP_ADD_MEMBERSHIP:
2829 case IP_ADD_SOURCE_MEMBERSHIP:
2830 case MCAST_JOIN_GROUP:
2831 case MCAST_JOIN_SOURCE_GROUP:
2832 error = inp_join_group(inp, sopt);
2833 break;
2834
2835 case IP_DROP_MEMBERSHIP:
2836 case IP_DROP_SOURCE_MEMBERSHIP:
2837 case MCAST_LEAVE_GROUP:
2838 case MCAST_LEAVE_SOURCE_GROUP:
2839 error = inp_leave_group(inp, sopt);
2840 break;
2841
2842 case IP_BLOCK_SOURCE:
2843 case IP_UNBLOCK_SOURCE:
2844 case MCAST_BLOCK_SOURCE:
2845 case MCAST_UNBLOCK_SOURCE:
2846 error = inp_block_unblock_source(inp, sopt);
2847 break;
2848
2849 case IP_MSFILTER:
2850 error = inp_set_source_filters(inp, sopt);
2851 break;
2852
2853 default:
2854 error = EOPNOTSUPP;
2855 break;
2856 }
2857
2858 INP_UNLOCK_ASSERT(inp);
2859
2860 return (error);
2861 }
2862
2863 /*
2864 * Expose IGMP's multicast filter mode and source list(s) to userland,
2865 * keyed by (ifindex, group).
2866 * The filter mode is written out as a uint32_t, followed by
2867 * 0..n of struct in_addr.
2868 * For use by ifmcstat(8).
2869 * SMPng: NOTE: unlocked read of ifindex space.
2870 */
2871 static int
sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)2872 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2873 {
2874 struct in_addr src, group;
2875 struct ifnet *ifp;
2876 struct ifmultiaddr *ifma;
2877 struct in_multi *inm;
2878 struct ip_msource *ims;
2879 int *name;
2880 int retval;
2881 u_int namelen;
2882 uint32_t fmode, ifindex;
2883
2884 name = (int *)arg1;
2885 namelen = arg2;
2886
2887 if (req->newptr != NULL)
2888 return (EPERM);
2889
2890 if (namelen != 2)
2891 return (EINVAL);
2892
2893 ifindex = name[0];
2894 if (ifindex <= 0 || ifindex > V_if_index) {
2895 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2896 __func__, ifindex);
2897 return (ENOENT);
2898 }
2899
2900 group.s_addr = name[1];
2901 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2902 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2903 __func__, ntohl(group.s_addr));
2904 return (EINVAL);
2905 }
2906
2907 ifp = ifnet_byindex(ifindex);
2908 if (ifp == NULL) {
2909 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2910 __func__, ifindex);
2911 return (ENOENT);
2912 }
2913
2914 retval = sysctl_wire_old_buffer(req,
2915 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2916 if (retval)
2917 return (retval);
2918
2919 IN_MULTI_LIST_LOCK();
2920
2921 IF_ADDR_RLOCK(ifp);
2922 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2923 if (ifma->ifma_addr->sa_family != AF_INET ||
2924 ifma->ifma_protospec == NULL)
2925 continue;
2926 inm = (struct in_multi *)ifma->ifma_protospec;
2927 if (!in_hosteq(inm->inm_addr, group))
2928 continue;
2929 fmode = inm->inm_st[1].iss_fmode;
2930 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2931 if (retval != 0)
2932 break;
2933 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2934 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2935 ims->ims_haddr);
2936 /*
2937 * Only copy-out sources which are in-mode.
2938 */
2939 if (fmode != ims_get_mode(inm, ims, 1)) {
2940 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2941 __func__);
2942 continue;
2943 }
2944 src.s_addr = htonl(ims->ims_haddr);
2945 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2946 if (retval != 0)
2947 break;
2948 }
2949 }
2950 IF_ADDR_RUNLOCK(ifp);
2951
2952 IN_MULTI_LIST_UNLOCK();
2953
2954 return (retval);
2955 }
2956
2957 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2958
2959 static const char *inm_modestrs[] = {
2960 [MCAST_UNDEFINED] = "un",
2961 [MCAST_INCLUDE] = "in",
2962 [MCAST_EXCLUDE] = "ex",
2963 };
2964 _Static_assert(MCAST_UNDEFINED == 0 &&
2965 MCAST_EXCLUDE + 1 == nitems(inm_modestrs),
2966 "inm_modestrs: no longer matches #defines");
2967
2968 static const char *
inm_mode_str(const int mode)2969 inm_mode_str(const int mode)
2970 {
2971
2972 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2973 return (inm_modestrs[mode]);
2974 return ("??");
2975 }
2976
2977 static const char *inm_statestrs[] = {
2978 [IGMP_NOT_MEMBER] = "not-member",
2979 [IGMP_SILENT_MEMBER] = "silent",
2980 [IGMP_REPORTING_MEMBER] = "reporting",
2981 [IGMP_IDLE_MEMBER] = "idle",
2982 [IGMP_LAZY_MEMBER] = "lazy",
2983 [IGMP_SLEEPING_MEMBER] = "sleeping",
2984 [IGMP_AWAKENING_MEMBER] = "awakening",
2985 [IGMP_G_QUERY_PENDING_MEMBER] = "query-pending",
2986 [IGMP_SG_QUERY_PENDING_MEMBER] = "sg-query-pending",
2987 [IGMP_LEAVING_MEMBER] = "leaving",
2988 };
2989 _Static_assert(IGMP_NOT_MEMBER == 0 &&
2990 IGMP_LEAVING_MEMBER + 1 == nitems(inm_statestrs),
2991 "inm_statetrs: no longer matches #defines");
2992
2993 static const char *
inm_state_str(const int state)2994 inm_state_str(const int state)
2995 {
2996
2997 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2998 return (inm_statestrs[state]);
2999 return ("??");
3000 }
3001
3002 /*
3003 * Dump an in_multi structure to the console.
3004 */
3005 void
inm_print(const struct in_multi * inm)3006 inm_print(const struct in_multi *inm)
3007 {
3008 int t;
3009 char addrbuf[INET_ADDRSTRLEN];
3010
3011 if ((ktr_mask & KTR_IGMPV3) == 0)
3012 return;
3013
3014 printf("%s: --- begin inm %p ---\n", __func__, inm);
3015 printf("addr %s ifp %p(%s) ifma %p\n",
3016 inet_ntoa_r(inm->inm_addr, addrbuf),
3017 inm->inm_ifp,
3018 inm->inm_ifp->if_xname,
3019 inm->inm_ifma);
3020 printf("timer %u state %s refcount %u scq.len %u\n",
3021 inm->inm_timer,
3022 inm_state_str(inm->inm_state),
3023 inm->inm_refcount,
3024 inm->inm_scq.mq_len);
3025 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3026 inm->inm_igi,
3027 inm->inm_nsrc,
3028 inm->inm_sctimer,
3029 inm->inm_scrv);
3030 for (t = 0; t < 2; t++) {
3031 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3032 inm_mode_str(inm->inm_st[t].iss_fmode),
3033 inm->inm_st[t].iss_asm,
3034 inm->inm_st[t].iss_ex,
3035 inm->inm_st[t].iss_in,
3036 inm->inm_st[t].iss_rec);
3037 }
3038 printf("%s: --- end inm %p ---\n", __func__, inm);
3039 }
3040
3041 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3042
3043 void
inm_print(const struct in_multi * inm)3044 inm_print(const struct in_multi *inm)
3045 {
3046
3047 }
3048
3049 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3050
3051 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);
3052