1 /*-
2 * Copyright (c) 2007-2009 Bruce Simpson.
3 * Copyright (c) 2005 Robert N. M. Watson.
4 * All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote
15 * products derived from this software without specific prior written
16 * permission.
17 *
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 * SUCH DAMAGE.
29 */
30
31 /*
32 * IPv4 multicast socket, group, and socket option processing module.
33 */
34
35 #include <sys/cdefs.h>
36 __FBSDID("$FreeBSD: releng/11.2/sys/netinet/in_mcast.c 321135 2017-07-18 16:58:54Z ngie $");
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/lock.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/protosw.h>
45 #include <sys/rmlock.h>
46 #include <sys/socket.h>
47 #include <sys/socketvar.h>
48 #include <sys/protosw.h>
49 #include <sys/sysctl.h>
50 #include <sys/ktr.h>
51 #include <sys/taskqueue.h>
52 #include <sys/tree.h>
53
54 #include <net/if.h>
55 #include <net/if_var.h>
56 #include <net/if_dl.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59
60 #include <netinet/in.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in_fib.h>
63 #include <netinet/in_pcb.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip_var.h>
66 #include <netinet/igmp_var.h>
67
68 #ifndef KTR_IGMPV3
69 #define KTR_IGMPV3 KTR_INET
70 #endif
71
72 #ifndef __SOCKUNION_DECLARED
73 union sockunion {
74 struct sockaddr_storage ss;
75 struct sockaddr sa;
76 struct sockaddr_dl sdl;
77 struct sockaddr_in sin;
78 };
79 typedef union sockunion sockunion_t;
80 #define __SOCKUNION_DECLARED
81 #endif /* __SOCKUNION_DECLARED */
82
83 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
84 "IPv4 multicast PCB-layer source filter");
85 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
86 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
87 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
88 "IPv4 multicast IGMP-layer source filter");
89
90 /*
91 * Locking:
92 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
93 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
94 * it can be taken by code in net/if.c also.
95 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
96 *
97 * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly
98 * any need for in_multi itself to be virtualized -- it is bound to an ifp
99 * anyway no matter what happens.
100 */
101 struct mtx in_multi_mtx;
102 MTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF);
103
104 /*
105 * Functions with non-static linkage defined in this file should be
106 * declared in in_var.h:
107 * imo_multi_filter()
108 * in_addmulti()
109 * in_delmulti()
110 * in_joingroup()
111 * in_joingroup_locked()
112 * in_leavegroup()
113 * in_leavegroup_locked()
114 * and ip_var.h:
115 * inp_freemoptions()
116 * inp_getmoptions()
117 * inp_setmoptions()
118 *
119 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
120 * and in_delmulti().
121 */
122 static void imf_commit(struct in_mfilter *);
123 static int imf_get_source(struct in_mfilter *imf,
124 const struct sockaddr_in *psin,
125 struct in_msource **);
126 static struct in_msource *
127 imf_graft(struct in_mfilter *, const uint8_t,
128 const struct sockaddr_in *);
129 static void imf_leave(struct in_mfilter *);
130 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
131 static void imf_purge(struct in_mfilter *);
132 static void imf_rollback(struct in_mfilter *);
133 static void imf_reap(struct in_mfilter *);
134 static int imo_grow(struct ip_moptions *);
135 static size_t imo_match_group(const struct ip_moptions *,
136 const struct ifnet *, const struct sockaddr *);
137 static struct in_msource *
138 imo_match_source(const struct ip_moptions *, const size_t,
139 const struct sockaddr *);
140 static void ims_merge(struct ip_msource *ims,
141 const struct in_msource *lims, const int rollback);
142 static int in_getmulti(struct ifnet *, const struct in_addr *,
143 struct in_multi **);
144 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
145 const int noalloc, struct ip_msource **pims);
146 #ifdef KTR
147 static int inm_is_ifp_detached(const struct in_multi *);
148 #endif
149 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
150 static void inm_purge(struct in_multi *);
151 static void inm_reap(struct in_multi *);
152 static struct ip_moptions *
153 inp_findmoptions(struct inpcb *);
154 static void inp_freemoptions_internal(struct ip_moptions *);
155 static void inp_gcmoptions(void *, int);
156 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
157 static int inp_join_group(struct inpcb *, struct sockopt *);
158 static int inp_leave_group(struct inpcb *, struct sockopt *);
159 static struct ifnet *
160 inp_lookup_mcast_ifp(const struct inpcb *,
161 const struct sockaddr_in *, const struct in_addr);
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, CTLFLAG_RW, 0,
168 "IPv4 multicast");
169
170 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
171 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
172 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
173 "Max source filters per group");
174
175 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
176 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
177 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
178 "Max source filters per socket");
179
180 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
181 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
182 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
183
184 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
185 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
186 "Per-interface stack-wide source filters");
187
188 static STAILQ_HEAD(, ip_moptions) imo_gc_list =
189 STAILQ_HEAD_INITIALIZER(imo_gc_list);
190 static struct task imo_gc_task = TASK_INITIALIZER(0, inp_gcmoptions, NULL);
191
192 #ifdef KTR
193 /*
194 * Inline function which wraps assertions for a valid ifp.
195 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
196 * is detached.
197 */
198 static int __inline
199 inm_is_ifp_detached(const struct in_multi *inm)
200 {
201 struct ifnet *ifp;
202
203 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
204 ifp = inm->inm_ifma->ifma_ifp;
205 if (ifp != NULL) {
206 /*
207 * Sanity check that netinet's notion of ifp is the
208 * same as net's.
209 */
210 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
211 }
212
213 return (ifp == NULL);
214 }
215 #endif
216
217 /*
218 * Initialize an in_mfilter structure to a known state at t0, t1
219 * with an empty source filter list.
220 */
221 static __inline void
222 imf_init(struct in_mfilter *imf, const int st0, const int st1)
223 {
224 memset(imf, 0, sizeof(struct in_mfilter));
225 RB_INIT(&imf->imf_sources);
226 imf->imf_st[0] = st0;
227 imf->imf_st[1] = st1;
228 }
229
230 /*
231 * Function for looking up an in_multi record for an IPv4 multicast address
232 * on a given interface. ifp must be valid. If no record found, return NULL.
233 * The IN_MULTI_LOCK and IF_ADDR_LOCK on ifp must be held.
234 */
235 struct in_multi *
236 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
237 {
238 struct ifmultiaddr *ifma;
239 struct in_multi *inm;
240
241 IN_MULTI_LOCK_ASSERT();
242 IF_ADDR_LOCK_ASSERT(ifp);
243
244 inm = NULL;
245 TAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
246 if (ifma->ifma_addr->sa_family == AF_INET) {
247 inm = (struct in_multi *)ifma->ifma_protospec;
248 if (inm->inm_addr.s_addr == ina.s_addr)
249 break;
250 inm = NULL;
251 }
252 }
253 return (inm);
254 }
255
256 /*
257 * Wrapper for inm_lookup_locked().
258 * The IF_ADDR_LOCK will be taken on ifp and released on return.
259 */
260 struct in_multi *
261 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
262 {
263 struct in_multi *inm;
264
265 IN_MULTI_LOCK_ASSERT();
266 IF_ADDR_RLOCK(ifp);
267 inm = inm_lookup_locked(ifp, ina);
268 IF_ADDR_RUNLOCK(ifp);
269
270 return (inm);
271 }
272
273 /*
274 * Resize the ip_moptions vector to the next power-of-two minus 1.
275 * May be called with locks held; do not sleep.
276 */
277 static int
278 imo_grow(struct ip_moptions *imo)
279 {
280 struct in_multi **nmships;
281 struct in_multi **omships;
282 struct in_mfilter *nmfilters;
283 struct in_mfilter *omfilters;
284 size_t idx;
285 size_t newmax;
286 size_t oldmax;
287
288 nmships = NULL;
289 nmfilters = NULL;
290 omships = imo->imo_membership;
291 omfilters = imo->imo_mfilters;
292 oldmax = imo->imo_max_memberships;
293 newmax = ((oldmax + 1) * 2) - 1;
294
295 if (newmax <= IP_MAX_MEMBERSHIPS) {
296 nmships = (struct in_multi **)realloc(omships,
297 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
298 nmfilters = (struct in_mfilter *)realloc(omfilters,
299 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
300 if (nmships != NULL && nmfilters != NULL) {
301 /* Initialize newly allocated source filter heads. */
302 for (idx = oldmax; idx < newmax; idx++) {
303 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
304 MCAST_EXCLUDE);
305 }
306 imo->imo_max_memberships = newmax;
307 imo->imo_membership = nmships;
308 imo->imo_mfilters = nmfilters;
309 }
310 }
311
312 if (nmships == NULL || nmfilters == NULL) {
313 if (nmships != NULL)
314 free(nmships, M_IPMOPTS);
315 if (nmfilters != NULL)
316 free(nmfilters, M_INMFILTER);
317 return (ETOOMANYREFS);
318 }
319
320 return (0);
321 }
322
323 /*
324 * Find an IPv4 multicast group entry for this ip_moptions instance
325 * which matches the specified group, and optionally an interface.
326 * Return its index into the array, or -1 if not found.
327 */
328 static size_t
329 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
330 const struct sockaddr *group)
331 {
332 const struct sockaddr_in *gsin;
333 struct in_multi **pinm;
334 int idx;
335 int nmships;
336
337 gsin = (const struct sockaddr_in *)group;
338
339 /* The imo_membership array may be lazy allocated. */
340 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
341 return (-1);
342
343 nmships = imo->imo_num_memberships;
344 pinm = &imo->imo_membership[0];
345 for (idx = 0; idx < nmships; idx++, pinm++) {
346 if (*pinm == NULL)
347 continue;
348 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
349 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
350 break;
351 }
352 }
353 if (idx >= nmships)
354 idx = -1;
355
356 return (idx);
357 }
358
359 /*
360 * Find an IPv4 multicast source entry for this imo which matches
361 * the given group index for this socket, and source address.
362 *
363 * NOTE: This does not check if the entry is in-mode, merely if
364 * it exists, which may not be the desired behaviour.
365 */
366 static struct in_msource *
367 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
368 const struct sockaddr *src)
369 {
370 struct ip_msource find;
371 struct in_mfilter *imf;
372 struct ip_msource *ims;
373 const sockunion_t *psa;
374
375 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
376 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
377 ("%s: invalid index %d\n", __func__, (int)gidx));
378
379 /* The imo_mfilters array may be lazy allocated. */
380 if (imo->imo_mfilters == NULL)
381 return (NULL);
382 imf = &imo->imo_mfilters[gidx];
383
384 /* Source trees are keyed in host byte order. */
385 psa = (const sockunion_t *)src;
386 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
387 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
388
389 return ((struct in_msource *)ims);
390 }
391
392 /*
393 * Perform filtering for multicast datagrams on a socket by group and source.
394 *
395 * Returns 0 if a datagram should be allowed through, or various error codes
396 * if the socket was not a member of the group, or the source was muted, etc.
397 */
398 int
399 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
400 const struct sockaddr *group, const struct sockaddr *src)
401 {
402 size_t gidx;
403 struct in_msource *ims;
404 int mode;
405
406 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
407
408 gidx = imo_match_group(imo, ifp, group);
409 if (gidx == -1)
410 return (MCAST_NOTGMEMBER);
411
412 /*
413 * Check if the source was included in an (S,G) join.
414 * Allow reception on exclusive memberships by default,
415 * reject reception on inclusive memberships by default.
416 * Exclude source only if an in-mode exclude filter exists.
417 * Include source only if an in-mode include filter exists.
418 * NOTE: We are comparing group state here at IGMP t1 (now)
419 * with socket-layer t0 (since last downcall).
420 */
421 mode = imo->imo_mfilters[gidx].imf_st[1];
422 ims = imo_match_source(imo, gidx, src);
423
424 if ((ims == NULL && mode == MCAST_INCLUDE) ||
425 (ims != NULL && ims->imsl_st[0] != mode))
426 return (MCAST_NOTSMEMBER);
427
428 return (MCAST_PASS);
429 }
430
431 /*
432 * Find and return a reference to an in_multi record for (ifp, group),
433 * and bump its reference count.
434 * If one does not exist, try to allocate it, and update link-layer multicast
435 * filters on ifp to listen for group.
436 * Assumes the IN_MULTI lock is held across the call.
437 * Return 0 if successful, otherwise return an appropriate error code.
438 */
439 static int
440 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
441 struct in_multi **pinm)
442 {
443 struct sockaddr_in gsin;
444 struct ifmultiaddr *ifma;
445 struct in_ifinfo *ii;
446 struct in_multi *inm;
447 int error;
448
449 IN_MULTI_LOCK_ASSERT();
450
451 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
452
453 inm = inm_lookup(ifp, *group);
454 if (inm != NULL) {
455 /*
456 * If we already joined this group, just bump the
457 * refcount and return it.
458 */
459 KASSERT(inm->inm_refcount >= 1,
460 ("%s: bad refcount %d", __func__, inm->inm_refcount));
461 ++inm->inm_refcount;
462 *pinm = inm;
463 return (0);
464 }
465
466 memset(&gsin, 0, sizeof(gsin));
467 gsin.sin_family = AF_INET;
468 gsin.sin_len = sizeof(struct sockaddr_in);
469 gsin.sin_addr = *group;
470
471 /*
472 * Check if a link-layer group is already associated
473 * with this network-layer group on the given ifnet.
474 */
475 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
476 if (error != 0)
477 return (error);
478
479 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
480 IF_ADDR_WLOCK(ifp);
481
482 /*
483 * If something other than netinet is occupying the link-layer
484 * group, print a meaningful error message and back out of
485 * the allocation.
486 * Otherwise, bump the refcount on the existing network-layer
487 * group association and return it.
488 */
489 if (ifma->ifma_protospec != NULL) {
490 inm = (struct in_multi *)ifma->ifma_protospec;
491 #ifdef INVARIANTS
492 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
493 __func__));
494 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
495 ("%s: ifma not AF_INET", __func__));
496 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
497 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
498 !in_hosteq(inm->inm_addr, *group)) {
499 char addrbuf[INET_ADDRSTRLEN];
500
501 panic("%s: ifma %p is inconsistent with %p (%s)",
502 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
503 }
504 #endif
505 ++inm->inm_refcount;
506 *pinm = inm;
507 IF_ADDR_WUNLOCK(ifp);
508 return (0);
509 }
510
511 IF_ADDR_WLOCK_ASSERT(ifp);
512
513 /*
514 * A new in_multi record is needed; allocate and initialize it.
515 * We DO NOT perform an IGMP join as the in_ layer may need to
516 * push an initial source list down to IGMP to support SSM.
517 *
518 * The initial source filter state is INCLUDE, {} as per the RFC.
519 */
520 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
521 if (inm == NULL) {
522 IF_ADDR_WUNLOCK(ifp);
523 if_delmulti_ifma(ifma);
524 return (ENOMEM);
525 }
526 inm->inm_addr = *group;
527 inm->inm_ifp = ifp;
528 inm->inm_igi = ii->ii_igmp;
529 inm->inm_ifma = ifma;
530 inm->inm_refcount = 1;
531 inm->inm_state = IGMP_NOT_MEMBER;
532 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
533 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
534 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
535 RB_INIT(&inm->inm_srcs);
536
537 ifma->ifma_protospec = inm;
538
539 *pinm = inm;
540
541 IF_ADDR_WUNLOCK(ifp);
542 return (0);
543 }
544
545 /*
546 * Drop a reference to an in_multi record.
547 *
548 * If the refcount drops to 0, free the in_multi record and
549 * delete the underlying link-layer membership.
550 */
551 void
552 inm_release_locked(struct in_multi *inm)
553 {
554 struct ifmultiaddr *ifma;
555
556 IN_MULTI_LOCK_ASSERT();
557
558 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
559
560 if (--inm->inm_refcount > 0) {
561 CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__,
562 inm->inm_refcount);
563 return;
564 }
565
566 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
567
568 ifma = inm->inm_ifma;
569
570 /* XXX this access is not covered by IF_ADDR_LOCK */
571 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
572 KASSERT(ifma->ifma_protospec == inm,
573 ("%s: ifma_protospec != inm", __func__));
574 ifma->ifma_protospec = NULL;
575
576 inm_purge(inm);
577
578 free(inm, M_IPMADDR);
579
580 if_delmulti_ifma(ifma);
581 }
582
583 /*
584 * Clear recorded source entries for a group.
585 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
586 * FIXME: Should reap.
587 */
588 void
589 inm_clear_recorded(struct in_multi *inm)
590 {
591 struct ip_msource *ims;
592
593 IN_MULTI_LOCK_ASSERT();
594
595 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
596 if (ims->ims_stp) {
597 ims->ims_stp = 0;
598 --inm->inm_st[1].iss_rec;
599 }
600 }
601 KASSERT(inm->inm_st[1].iss_rec == 0,
602 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
603 }
604
605 /*
606 * Record a source as pending for a Source-Group IGMPv3 query.
607 * This lives here as it modifies the shared tree.
608 *
609 * inm is the group descriptor.
610 * naddr is the address of the source to record in network-byte order.
611 *
612 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
613 * lazy-allocate a source node in response to an SG query.
614 * Otherwise, no allocation is performed. This saves some memory
615 * with the trade-off that the source will not be reported to the
616 * router if joined in the window between the query response and
617 * the group actually being joined on the local host.
618 *
619 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
620 * This turns off the allocation of a recorded source entry if
621 * the group has not been joined.
622 *
623 * Return 0 if the source didn't exist or was already marked as recorded.
624 * Return 1 if the source was marked as recorded by this function.
625 * Return <0 if any error occurred (negated errno code).
626 */
627 int
628 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
629 {
630 struct ip_msource find;
631 struct ip_msource *ims, *nims;
632
633 IN_MULTI_LOCK_ASSERT();
634
635 find.ims_haddr = ntohl(naddr);
636 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
637 if (ims && ims->ims_stp)
638 return (0);
639 if (ims == NULL) {
640 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
641 return (-ENOSPC);
642 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
643 M_NOWAIT | M_ZERO);
644 if (nims == NULL)
645 return (-ENOMEM);
646 nims->ims_haddr = find.ims_haddr;
647 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
648 ++inm->inm_nsrc;
649 ims = nims;
650 }
651
652 /*
653 * Mark the source as recorded and update the recorded
654 * source count.
655 */
656 ++ims->ims_stp;
657 ++inm->inm_st[1].iss_rec;
658
659 return (1);
660 }
661
662 /*
663 * Return a pointer to an in_msource owned by an in_mfilter,
664 * given its source address.
665 * Lazy-allocate if needed. If this is a new entry its filter state is
666 * undefined at t0.
667 *
668 * imf is the filter set being modified.
669 * haddr is the source address in *host* byte-order.
670 *
671 * SMPng: May be called with locks held; malloc must not block.
672 */
673 static int
674 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
675 struct in_msource **plims)
676 {
677 struct ip_msource find;
678 struct ip_msource *ims, *nims;
679 struct in_msource *lims;
680 int error;
681
682 error = 0;
683 ims = NULL;
684 lims = NULL;
685
686 /* key is host byte order */
687 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
688 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
689 lims = (struct in_msource *)ims;
690 if (lims == NULL) {
691 if (imf->imf_nsrc == in_mcast_maxsocksrc)
692 return (ENOSPC);
693 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
694 M_NOWAIT | M_ZERO);
695 if (nims == NULL)
696 return (ENOMEM);
697 lims = (struct in_msource *)nims;
698 lims->ims_haddr = find.ims_haddr;
699 lims->imsl_st[0] = MCAST_UNDEFINED;
700 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
701 ++imf->imf_nsrc;
702 }
703
704 *plims = lims;
705
706 return (error);
707 }
708
709 /*
710 * Graft a source entry into an existing socket-layer filter set,
711 * maintaining any required invariants and checking allocations.
712 *
713 * The source is marked as being in the new filter mode at t1.
714 *
715 * Return the pointer to the new node, otherwise return NULL.
716 */
717 static struct in_msource *
718 imf_graft(struct in_mfilter *imf, const uint8_t st1,
719 const struct sockaddr_in *psin)
720 {
721 struct ip_msource *nims;
722 struct in_msource *lims;
723
724 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
725 M_NOWAIT | M_ZERO);
726 if (nims == NULL)
727 return (NULL);
728 lims = (struct in_msource *)nims;
729 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
730 lims->imsl_st[0] = MCAST_UNDEFINED;
731 lims->imsl_st[1] = st1;
732 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
733 ++imf->imf_nsrc;
734
735 return (lims);
736 }
737
738 /*
739 * Prune a source entry from an existing socket-layer filter set,
740 * maintaining any required invariants and checking allocations.
741 *
742 * The source is marked as being left at t1, it is not freed.
743 *
744 * Return 0 if no error occurred, otherwise return an errno value.
745 */
746 static int
747 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
748 {
749 struct ip_msource find;
750 struct ip_msource *ims;
751 struct in_msource *lims;
752
753 /* key is host byte order */
754 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
755 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
756 if (ims == NULL)
757 return (ENOENT);
758 lims = (struct in_msource *)ims;
759 lims->imsl_st[1] = MCAST_UNDEFINED;
760 return (0);
761 }
762
763 /*
764 * Revert socket-layer filter set deltas at t1 to t0 state.
765 */
766 static void
767 imf_rollback(struct in_mfilter *imf)
768 {
769 struct ip_msource *ims, *tims;
770 struct in_msource *lims;
771
772 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
773 lims = (struct in_msource *)ims;
774 if (lims->imsl_st[0] == lims->imsl_st[1]) {
775 /* no change at t1 */
776 continue;
777 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
778 /* revert change to existing source at t1 */
779 lims->imsl_st[1] = lims->imsl_st[0];
780 } else {
781 /* revert source added t1 */
782 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
783 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
784 free(ims, M_INMFILTER);
785 imf->imf_nsrc--;
786 }
787 }
788 imf->imf_st[1] = imf->imf_st[0];
789 }
790
791 /*
792 * Mark socket-layer filter set as INCLUDE {} at t1.
793 */
794 static void
795 imf_leave(struct in_mfilter *imf)
796 {
797 struct ip_msource *ims;
798 struct in_msource *lims;
799
800 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
801 lims = (struct in_msource *)ims;
802 lims->imsl_st[1] = MCAST_UNDEFINED;
803 }
804 imf->imf_st[1] = MCAST_INCLUDE;
805 }
806
807 /*
808 * Mark socket-layer filter set deltas as committed.
809 */
810 static void
811 imf_commit(struct in_mfilter *imf)
812 {
813 struct ip_msource *ims;
814 struct in_msource *lims;
815
816 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
817 lims = (struct in_msource *)ims;
818 lims->imsl_st[0] = lims->imsl_st[1];
819 }
820 imf->imf_st[0] = imf->imf_st[1];
821 }
822
823 /*
824 * Reap unreferenced sources from socket-layer filter set.
825 */
826 static void
827 imf_reap(struct in_mfilter *imf)
828 {
829 struct ip_msource *ims, *tims;
830 struct in_msource *lims;
831
832 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
833 lims = (struct in_msource *)ims;
834 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
835 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
836 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
837 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
838 free(ims, M_INMFILTER);
839 imf->imf_nsrc--;
840 }
841 }
842 }
843
844 /*
845 * Purge socket-layer filter set.
846 */
847 static void
848 imf_purge(struct in_mfilter *imf)
849 {
850 struct ip_msource *ims, *tims;
851
852 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
853 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
854 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
855 free(ims, M_INMFILTER);
856 imf->imf_nsrc--;
857 }
858 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
859 KASSERT(RB_EMPTY(&imf->imf_sources),
860 ("%s: imf_sources not empty", __func__));
861 }
862
863 /*
864 * Look up a source filter entry for a multicast group.
865 *
866 * inm is the group descriptor to work with.
867 * haddr is the host-byte-order IPv4 address to look up.
868 * noalloc may be non-zero to suppress allocation of sources.
869 * *pims will be set to the address of the retrieved or allocated source.
870 *
871 * SMPng: NOTE: may be called with locks held.
872 * Return 0 if successful, otherwise return a non-zero error code.
873 */
874 static int
875 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
876 const int noalloc, struct ip_msource **pims)
877 {
878 struct ip_msource find;
879 struct ip_msource *ims, *nims;
880
881 find.ims_haddr = haddr;
882 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
883 if (ims == NULL && !noalloc) {
884 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
885 return (ENOSPC);
886 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
887 M_NOWAIT | M_ZERO);
888 if (nims == NULL)
889 return (ENOMEM);
890 nims->ims_haddr = haddr;
891 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
892 ++inm->inm_nsrc;
893 ims = nims;
894 #ifdef KTR
895 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
896 haddr, ims);
897 #endif
898 }
899
900 *pims = ims;
901 return (0);
902 }
903
904 /*
905 * Merge socket-layer source into IGMP-layer source.
906 * If rollback is non-zero, perform the inverse of the merge.
907 */
908 static void
909 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
910 const int rollback)
911 {
912 int n = rollback ? -1 : 1;
913
914 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
915 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
916 __func__, n, ims->ims_haddr);
917 ims->ims_st[1].ex -= n;
918 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
919 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
920 __func__, n, ims->ims_haddr);
921 ims->ims_st[1].in -= n;
922 }
923
924 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
925 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
926 __func__, n, ims->ims_haddr);
927 ims->ims_st[1].ex += n;
928 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
929 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
930 __func__, n, ims->ims_haddr);
931 ims->ims_st[1].in += n;
932 }
933 }
934
935 /*
936 * Atomically update the global in_multi state, when a membership's
937 * filter list is being updated in any way.
938 *
939 * imf is the per-inpcb-membership group filter pointer.
940 * A fake imf may be passed for in-kernel consumers.
941 *
942 * XXX This is a candidate for a set-symmetric-difference style loop
943 * which would eliminate the repeated lookup from root of ims nodes,
944 * as they share the same key space.
945 *
946 * If any error occurred this function will back out of refcounts
947 * and return a non-zero value.
948 */
949 static int
950 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
951 {
952 struct ip_msource *ims, *nims;
953 struct in_msource *lims;
954 int schanged, error;
955 int nsrc0, nsrc1;
956
957 schanged = 0;
958 error = 0;
959 nsrc1 = nsrc0 = 0;
960
961 /*
962 * Update the source filters first, as this may fail.
963 * Maintain count of in-mode filters at t0, t1. These are
964 * used to work out if we transition into ASM mode or not.
965 * Maintain a count of source filters whose state was
966 * actually modified by this operation.
967 */
968 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
969 lims = (struct in_msource *)ims;
970 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
971 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
972 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
973 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
974 ++schanged;
975 if (error)
976 break;
977 ims_merge(nims, lims, 0);
978 }
979 if (error) {
980 struct ip_msource *bims;
981
982 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
983 lims = (struct in_msource *)ims;
984 if (lims->imsl_st[0] == lims->imsl_st[1])
985 continue;
986 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
987 if (bims == NULL)
988 continue;
989 ims_merge(bims, lims, 1);
990 }
991 goto out_reap;
992 }
993
994 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
995 __func__, nsrc0, nsrc1);
996
997 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
998 if (imf->imf_st[0] == imf->imf_st[1] &&
999 imf->imf_st[1] == MCAST_INCLUDE) {
1000 if (nsrc1 == 0) {
1001 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1002 --inm->inm_st[1].iss_in;
1003 }
1004 }
1005
1006 /* Handle filter mode transition on socket. */
1007 if (imf->imf_st[0] != imf->imf_st[1]) {
1008 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1009 __func__, imf->imf_st[0], imf->imf_st[1]);
1010
1011 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1012 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1013 --inm->inm_st[1].iss_ex;
1014 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1015 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1016 --inm->inm_st[1].iss_in;
1017 }
1018
1019 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1020 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1021 inm->inm_st[1].iss_ex++;
1022 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1023 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1024 inm->inm_st[1].iss_in++;
1025 }
1026 }
1027
1028 /*
1029 * Track inm filter state in terms of listener counts.
1030 * If there are any exclusive listeners, stack-wide
1031 * membership is exclusive.
1032 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1033 * If no listeners remain, state is undefined at t1,
1034 * and the IGMP lifecycle for this group should finish.
1035 */
1036 if (inm->inm_st[1].iss_ex > 0) {
1037 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1038 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1039 } else if (inm->inm_st[1].iss_in > 0) {
1040 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1041 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1042 } else {
1043 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1044 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1045 }
1046
1047 /* Decrement ASM listener count on transition out of ASM mode. */
1048 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1049 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1050 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1051 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1052 --inm->inm_st[1].iss_asm;
1053 }
1054 }
1055
1056 /* Increment ASM listener count on transition to ASM mode. */
1057 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1058 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1059 inm->inm_st[1].iss_asm++;
1060 }
1061
1062 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1063 inm_print(inm);
1064
1065 out_reap:
1066 if (schanged > 0) {
1067 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1068 inm_reap(inm);
1069 }
1070 return (error);
1071 }
1072
1073 /*
1074 * Mark an in_multi's filter set deltas as committed.
1075 * Called by IGMP after a state change has been enqueued.
1076 */
1077 void
1078 inm_commit(struct in_multi *inm)
1079 {
1080 struct ip_msource *ims;
1081
1082 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1083 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1084 inm_print(inm);
1085
1086 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1087 ims->ims_st[0] = ims->ims_st[1];
1088 }
1089 inm->inm_st[0] = inm->inm_st[1];
1090 }
1091
1092 /*
1093 * Reap unreferenced nodes from an in_multi's filter set.
1094 */
1095 static void
1096 inm_reap(struct in_multi *inm)
1097 {
1098 struct ip_msource *ims, *tims;
1099
1100 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1101 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1102 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1103 ims->ims_stp != 0)
1104 continue;
1105 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1106 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1107 free(ims, M_IPMSOURCE);
1108 inm->inm_nsrc--;
1109 }
1110 }
1111
1112 /*
1113 * Purge all source nodes from an in_multi's filter set.
1114 */
1115 static void
1116 inm_purge(struct in_multi *inm)
1117 {
1118 struct ip_msource *ims, *tims;
1119
1120 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1121 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1122 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1123 free(ims, M_IPMSOURCE);
1124 inm->inm_nsrc--;
1125 }
1126 }
1127
1128 /*
1129 * Join a multicast group; unlocked entry point.
1130 *
1131 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1132 * is not held. Fortunately, ifp is unlikely to have been detached
1133 * at this point, so we assume it's OK to recurse.
1134 */
1135 int
1136 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1137 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1138 {
1139 int error;
1140
1141 IN_MULTI_LOCK();
1142 error = in_joingroup_locked(ifp, gina, imf, pinm);
1143 IN_MULTI_UNLOCK();
1144
1145 return (error);
1146 }
1147
1148 /*
1149 * Join a multicast group; real entry point.
1150 *
1151 * Only preserves atomicity at inm level.
1152 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1153 *
1154 * If the IGMP downcall fails, the group is not joined, and an error
1155 * code is returned.
1156 */
1157 int
1158 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1159 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1160 {
1161 struct in_mfilter timf;
1162 struct in_multi *inm;
1163 int error;
1164
1165 IN_MULTI_LOCK_ASSERT();
1166
1167 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1168 ntohl(gina->s_addr), ifp, ifp->if_xname);
1169
1170 error = 0;
1171 inm = NULL;
1172
1173 /*
1174 * If no imf was specified (i.e. kernel consumer),
1175 * fake one up and assume it is an ASM join.
1176 */
1177 if (imf == NULL) {
1178 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1179 imf = &timf;
1180 }
1181
1182 error = in_getmulti(ifp, gina, &inm);
1183 if (error) {
1184 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1185 return (error);
1186 }
1187
1188 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1189 error = inm_merge(inm, imf);
1190 if (error) {
1191 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1192 goto out_inm_release;
1193 }
1194
1195 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1196 error = igmp_change_state(inm);
1197 if (error) {
1198 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1199 goto out_inm_release;
1200 }
1201
1202 out_inm_release:
1203 if (error) {
1204 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1205 inm_release_locked(inm);
1206 } else {
1207 *pinm = inm;
1208 }
1209
1210 return (error);
1211 }
1212
1213 /*
1214 * Leave a multicast group; unlocked entry point.
1215 */
1216 int
1217 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1218 {
1219 int error;
1220
1221 IN_MULTI_LOCK();
1222 error = in_leavegroup_locked(inm, imf);
1223 IN_MULTI_UNLOCK();
1224
1225 return (error);
1226 }
1227
1228 /*
1229 * Leave a multicast group; real entry point.
1230 * All source filters will be expunged.
1231 *
1232 * Only preserves atomicity at inm level.
1233 *
1234 * Holding the write lock for the INP which contains imf
1235 * is highly advisable. We can't assert for it as imf does not
1236 * contain a back-pointer to the owning inp.
1237 *
1238 * Note: This is not the same as inm_release(*) as this function also
1239 * makes a state change downcall into IGMP.
1240 */
1241 int
1242 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1243 {
1244 struct in_mfilter timf;
1245 int error;
1246
1247 error = 0;
1248
1249 IN_MULTI_LOCK_ASSERT();
1250
1251 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1252 inm, ntohl(inm->inm_addr.s_addr),
1253 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1254 imf);
1255
1256 /*
1257 * If no imf was specified (i.e. kernel consumer),
1258 * fake one up and assume it is an ASM join.
1259 */
1260 if (imf == NULL) {
1261 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1262 imf = &timf;
1263 }
1264
1265 /*
1266 * Begin state merge transaction at IGMP layer.
1267 *
1268 * As this particular invocation should not cause any memory
1269 * to be allocated, and there is no opportunity to roll back
1270 * the transaction, it MUST NOT fail.
1271 */
1272 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1273 error = inm_merge(inm, imf);
1274 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1275
1276 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1277 CURVNET_SET(inm->inm_ifp->if_vnet);
1278 error = igmp_change_state(inm);
1279 CURVNET_RESTORE();
1280 if (error)
1281 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1282
1283 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1284 inm_release_locked(inm);
1285
1286 return (error);
1287 }
1288
1289 /*#ifndef BURN_BRIDGES*/
1290 /*
1291 * Join an IPv4 multicast group in (*,G) exclusive mode.
1292 * The group must be a 224.0.0.0/24 link-scope group.
1293 * This KPI is for legacy kernel consumers only.
1294 */
1295 struct in_multi *
1296 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1297 {
1298 struct in_multi *pinm;
1299 int error;
1300 #ifdef INVARIANTS
1301 char addrbuf[INET_ADDRSTRLEN];
1302 #endif
1303
1304 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1305 ("%s: %s not in 224.0.0.0/24", __func__,
1306 inet_ntoa_r(*ap, addrbuf)));
1307
1308 error = in_joingroup(ifp, ap, NULL, &pinm);
1309 if (error != 0)
1310 pinm = NULL;
1311
1312 return (pinm);
1313 }
1314
1315 /*
1316 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1317 * This KPI is for legacy kernel consumers only.
1318 */
1319 void
1320 in_delmulti(struct in_multi *inm)
1321 {
1322
1323 (void)in_leavegroup(inm, NULL);
1324 }
1325 /*#endif*/
1326
1327 /*
1328 * Block or unblock an ASM multicast source on an inpcb.
1329 * This implements the delta-based API described in RFC 3678.
1330 *
1331 * The delta-based API applies only to exclusive-mode memberships.
1332 * An IGMP downcall will be performed.
1333 *
1334 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1335 *
1336 * Return 0 if successful, otherwise return an appropriate error code.
1337 */
1338 static int
1339 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1340 {
1341 struct group_source_req gsr;
1342 sockunion_t *gsa, *ssa;
1343 struct ifnet *ifp;
1344 struct in_mfilter *imf;
1345 struct ip_moptions *imo;
1346 struct in_msource *ims;
1347 struct in_multi *inm;
1348 size_t idx;
1349 uint16_t fmode;
1350 int error, doblock;
1351
1352 ifp = NULL;
1353 error = 0;
1354 doblock = 0;
1355
1356 memset(&gsr, 0, sizeof(struct group_source_req));
1357 gsa = (sockunion_t *)&gsr.gsr_group;
1358 ssa = (sockunion_t *)&gsr.gsr_source;
1359
1360 switch (sopt->sopt_name) {
1361 case IP_BLOCK_SOURCE:
1362 case IP_UNBLOCK_SOURCE: {
1363 struct ip_mreq_source mreqs;
1364
1365 error = sooptcopyin(sopt, &mreqs,
1366 sizeof(struct ip_mreq_source),
1367 sizeof(struct ip_mreq_source));
1368 if (error)
1369 return (error);
1370
1371 gsa->sin.sin_family = AF_INET;
1372 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1373 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1374
1375 ssa->sin.sin_family = AF_INET;
1376 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1377 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1378
1379 if (!in_nullhost(mreqs.imr_interface))
1380 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1381
1382 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1383 doblock = 1;
1384
1385 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1386 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1387 break;
1388 }
1389
1390 case MCAST_BLOCK_SOURCE:
1391 case MCAST_UNBLOCK_SOURCE:
1392 error = sooptcopyin(sopt, &gsr,
1393 sizeof(struct group_source_req),
1394 sizeof(struct group_source_req));
1395 if (error)
1396 return (error);
1397
1398 if (gsa->sin.sin_family != AF_INET ||
1399 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1400 return (EINVAL);
1401
1402 if (ssa->sin.sin_family != AF_INET ||
1403 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1404 return (EINVAL);
1405
1406 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1407 return (EADDRNOTAVAIL);
1408
1409 ifp = ifnet_byindex(gsr.gsr_interface);
1410
1411 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1412 doblock = 1;
1413 break;
1414
1415 default:
1416 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1417 __func__, sopt->sopt_name);
1418 return (EOPNOTSUPP);
1419 break;
1420 }
1421
1422 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1423 return (EINVAL);
1424
1425 /*
1426 * Check if we are actually a member of this group.
1427 */
1428 imo = inp_findmoptions(inp);
1429 idx = imo_match_group(imo, ifp, &gsa->sa);
1430 if (idx == -1 || imo->imo_mfilters == NULL) {
1431 error = EADDRNOTAVAIL;
1432 goto out_inp_locked;
1433 }
1434
1435 KASSERT(imo->imo_mfilters != NULL,
1436 ("%s: imo_mfilters not allocated", __func__));
1437 imf = &imo->imo_mfilters[idx];
1438 inm = imo->imo_membership[idx];
1439
1440 /*
1441 * Attempting to use the delta-based API on an
1442 * non exclusive-mode membership is an error.
1443 */
1444 fmode = imf->imf_st[0];
1445 if (fmode != MCAST_EXCLUDE) {
1446 error = EINVAL;
1447 goto out_inp_locked;
1448 }
1449
1450 /*
1451 * Deal with error cases up-front:
1452 * Asked to block, but already blocked; or
1453 * Asked to unblock, but nothing to unblock.
1454 * If adding a new block entry, allocate it.
1455 */
1456 ims = imo_match_source(imo, idx, &ssa->sa);
1457 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1458 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1459 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1460 error = EADDRNOTAVAIL;
1461 goto out_inp_locked;
1462 }
1463
1464 INP_WLOCK_ASSERT(inp);
1465
1466 /*
1467 * Begin state merge transaction at socket layer.
1468 */
1469 if (doblock) {
1470 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1471 ims = imf_graft(imf, fmode, &ssa->sin);
1472 if (ims == NULL)
1473 error = ENOMEM;
1474 } else {
1475 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1476 error = imf_prune(imf, &ssa->sin);
1477 }
1478
1479 if (error) {
1480 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1481 goto out_imf_rollback;
1482 }
1483
1484 /*
1485 * Begin state merge transaction at IGMP layer.
1486 */
1487 IN_MULTI_LOCK();
1488
1489 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1490 error = inm_merge(inm, imf);
1491 if (error) {
1492 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1493 goto out_in_multi_locked;
1494 }
1495
1496 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1497 error = igmp_change_state(inm);
1498 if (error)
1499 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1500
1501 out_in_multi_locked:
1502
1503 IN_MULTI_UNLOCK();
1504
1505 out_imf_rollback:
1506 if (error)
1507 imf_rollback(imf);
1508 else
1509 imf_commit(imf);
1510
1511 imf_reap(imf);
1512
1513 out_inp_locked:
1514 INP_WUNLOCK(inp);
1515 return (error);
1516 }
1517
1518 /*
1519 * Given an inpcb, return its multicast options structure pointer. Accepts
1520 * an unlocked inpcb pointer, but will return it locked. May sleep.
1521 *
1522 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1523 * SMPng: NOTE: Returns with the INP write lock held.
1524 */
1525 static struct ip_moptions *
1526 inp_findmoptions(struct inpcb *inp)
1527 {
1528 struct ip_moptions *imo;
1529 struct in_multi **immp;
1530 struct in_mfilter *imfp;
1531 size_t idx;
1532
1533 INP_WLOCK(inp);
1534 if (inp->inp_moptions != NULL)
1535 return (inp->inp_moptions);
1536
1537 INP_WUNLOCK(inp);
1538
1539 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1540 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1541 M_WAITOK | M_ZERO);
1542 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1543 M_INMFILTER, M_WAITOK);
1544
1545 imo->imo_multicast_ifp = NULL;
1546 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1547 imo->imo_multicast_vif = -1;
1548 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1549 imo->imo_multicast_loop = in_mcast_loop;
1550 imo->imo_num_memberships = 0;
1551 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1552 imo->imo_membership = immp;
1553
1554 /* Initialize per-group source filters. */
1555 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1556 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1557 imo->imo_mfilters = imfp;
1558
1559 INP_WLOCK(inp);
1560 if (inp->inp_moptions != NULL) {
1561 free(imfp, M_INMFILTER);
1562 free(immp, M_IPMOPTS);
1563 free(imo, M_IPMOPTS);
1564 return (inp->inp_moptions);
1565 }
1566 inp->inp_moptions = imo;
1567 return (imo);
1568 }
1569
1570 /*
1571 * Discard the IP multicast options (and source filters). To minimize
1572 * the amount of work done while holding locks such as the INP's
1573 * pcbinfo lock (which is used in the receive path), the free
1574 * operation is performed asynchronously in a separate task.
1575 *
1576 * SMPng: NOTE: assumes INP write lock is held.
1577 */
1578 void
1579 inp_freemoptions(struct ip_moptions *imo)
1580 {
1581
1582 KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1583 IN_MULTI_LOCK();
1584 STAILQ_INSERT_TAIL(&imo_gc_list, imo, imo_link);
1585 IN_MULTI_UNLOCK();
1586 taskqueue_enqueue(taskqueue_thread, &imo_gc_task);
1587 }
1588
1589 static void
1590 inp_freemoptions_internal(struct ip_moptions *imo)
1591 {
1592 struct in_mfilter *imf;
1593 size_t idx, nmships;
1594
1595 nmships = imo->imo_num_memberships;
1596 for (idx = 0; idx < nmships; ++idx) {
1597 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1598 if (imf)
1599 imf_leave(imf);
1600 (void)in_leavegroup(imo->imo_membership[idx], imf);
1601 if (imf)
1602 imf_purge(imf);
1603 }
1604
1605 if (imo->imo_mfilters)
1606 free(imo->imo_mfilters, M_INMFILTER);
1607 free(imo->imo_membership, M_IPMOPTS);
1608 free(imo, M_IPMOPTS);
1609 }
1610
1611 static void
1612 inp_gcmoptions(void *context, int pending)
1613 {
1614 struct ip_moptions *imo;
1615
1616 IN_MULTI_LOCK();
1617 while (!STAILQ_EMPTY(&imo_gc_list)) {
1618 imo = STAILQ_FIRST(&imo_gc_list);
1619 STAILQ_REMOVE_HEAD(&imo_gc_list, imo_link);
1620 IN_MULTI_UNLOCK();
1621 inp_freemoptions_internal(imo);
1622 IN_MULTI_LOCK();
1623 }
1624 IN_MULTI_UNLOCK();
1625 }
1626
1627 /*
1628 * Atomically get source filters on a socket for an IPv4 multicast group.
1629 * Called with INP lock held; returns with lock released.
1630 */
1631 static int
1632 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1633 {
1634 struct __msfilterreq msfr;
1635 sockunion_t *gsa;
1636 struct ifnet *ifp;
1637 struct ip_moptions *imo;
1638 struct in_mfilter *imf;
1639 struct ip_msource *ims;
1640 struct in_msource *lims;
1641 struct sockaddr_in *psin;
1642 struct sockaddr_storage *ptss;
1643 struct sockaddr_storage *tss;
1644 int error;
1645 size_t idx, nsrcs, ncsrcs;
1646
1647 INP_WLOCK_ASSERT(inp);
1648
1649 imo = inp->inp_moptions;
1650 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1651
1652 INP_WUNLOCK(inp);
1653
1654 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1655 sizeof(struct __msfilterreq));
1656 if (error)
1657 return (error);
1658
1659 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1660 return (EINVAL);
1661
1662 ifp = ifnet_byindex(msfr.msfr_ifindex);
1663 if (ifp == NULL)
1664 return (EINVAL);
1665
1666 INP_WLOCK(inp);
1667
1668 /*
1669 * Lookup group on the socket.
1670 */
1671 gsa = (sockunion_t *)&msfr.msfr_group;
1672 idx = imo_match_group(imo, ifp, &gsa->sa);
1673 if (idx == -1 || imo->imo_mfilters == NULL) {
1674 INP_WUNLOCK(inp);
1675 return (EADDRNOTAVAIL);
1676 }
1677 imf = &imo->imo_mfilters[idx];
1678
1679 /*
1680 * Ignore memberships which are in limbo.
1681 */
1682 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1683 INP_WUNLOCK(inp);
1684 return (EAGAIN);
1685 }
1686 msfr.msfr_fmode = imf->imf_st[1];
1687
1688 /*
1689 * If the user specified a buffer, copy out the source filter
1690 * entries to userland gracefully.
1691 * We only copy out the number of entries which userland
1692 * has asked for, but we always tell userland how big the
1693 * buffer really needs to be.
1694 */
1695 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1696 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1697 tss = NULL;
1698 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1699 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1700 M_TEMP, M_NOWAIT | M_ZERO);
1701 if (tss == NULL) {
1702 INP_WUNLOCK(inp);
1703 return (ENOBUFS);
1704 }
1705 }
1706
1707 /*
1708 * Count number of sources in-mode at t0.
1709 * If buffer space exists and remains, copy out source entries.
1710 */
1711 nsrcs = msfr.msfr_nsrcs;
1712 ncsrcs = 0;
1713 ptss = tss;
1714 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1715 lims = (struct in_msource *)ims;
1716 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1717 lims->imsl_st[0] != imf->imf_st[0])
1718 continue;
1719 ++ncsrcs;
1720 if (tss != NULL && nsrcs > 0) {
1721 psin = (struct sockaddr_in *)ptss;
1722 psin->sin_family = AF_INET;
1723 psin->sin_len = sizeof(struct sockaddr_in);
1724 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1725 psin->sin_port = 0;
1726 ++ptss;
1727 --nsrcs;
1728 }
1729 }
1730
1731 INP_WUNLOCK(inp);
1732
1733 if (tss != NULL) {
1734 error = copyout(tss, msfr.msfr_srcs,
1735 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1736 free(tss, M_TEMP);
1737 if (error)
1738 return (error);
1739 }
1740
1741 msfr.msfr_nsrcs = ncsrcs;
1742 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1743
1744 return (error);
1745 }
1746
1747 /*
1748 * Return the IP multicast options in response to user getsockopt().
1749 */
1750 int
1751 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1752 {
1753 struct rm_priotracker in_ifa_tracker;
1754 struct ip_mreqn mreqn;
1755 struct ip_moptions *imo;
1756 struct ifnet *ifp;
1757 struct in_ifaddr *ia;
1758 int error, optval;
1759 u_char coptval;
1760
1761 INP_WLOCK(inp);
1762 imo = inp->inp_moptions;
1763 /*
1764 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1765 * or is a divert socket, reject it.
1766 */
1767 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1768 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1769 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1770 INP_WUNLOCK(inp);
1771 return (EOPNOTSUPP);
1772 }
1773
1774 error = 0;
1775 switch (sopt->sopt_name) {
1776 case IP_MULTICAST_VIF:
1777 if (imo != NULL)
1778 optval = imo->imo_multicast_vif;
1779 else
1780 optval = -1;
1781 INP_WUNLOCK(inp);
1782 error = sooptcopyout(sopt, &optval, sizeof(int));
1783 break;
1784
1785 case IP_MULTICAST_IF:
1786 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1787 if (imo != NULL) {
1788 ifp = imo->imo_multicast_ifp;
1789 if (!in_nullhost(imo->imo_multicast_addr)) {
1790 mreqn.imr_address = imo->imo_multicast_addr;
1791 } else if (ifp != NULL) {
1792 mreqn.imr_ifindex = ifp->if_index;
1793 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1794 if (ia != NULL) {
1795 mreqn.imr_address =
1796 IA_SIN(ia)->sin_addr;
1797 ifa_free(&ia->ia_ifa);
1798 }
1799 }
1800 }
1801 INP_WUNLOCK(inp);
1802 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1803 error = sooptcopyout(sopt, &mreqn,
1804 sizeof(struct ip_mreqn));
1805 } else {
1806 error = sooptcopyout(sopt, &mreqn.imr_address,
1807 sizeof(struct in_addr));
1808 }
1809 break;
1810
1811 case IP_MULTICAST_TTL:
1812 if (imo == NULL)
1813 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1814 else
1815 optval = coptval = imo->imo_multicast_ttl;
1816 INP_WUNLOCK(inp);
1817 if (sopt->sopt_valsize == sizeof(u_char))
1818 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1819 else
1820 error = sooptcopyout(sopt, &optval, sizeof(int));
1821 break;
1822
1823 case IP_MULTICAST_LOOP:
1824 if (imo == NULL)
1825 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1826 else
1827 optval = coptval = imo->imo_multicast_loop;
1828 INP_WUNLOCK(inp);
1829 if (sopt->sopt_valsize == sizeof(u_char))
1830 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1831 else
1832 error = sooptcopyout(sopt, &optval, sizeof(int));
1833 break;
1834
1835 case IP_MSFILTER:
1836 if (imo == NULL) {
1837 error = EADDRNOTAVAIL;
1838 INP_WUNLOCK(inp);
1839 } else {
1840 error = inp_get_source_filters(inp, sopt);
1841 }
1842 break;
1843
1844 default:
1845 INP_WUNLOCK(inp);
1846 error = ENOPROTOOPT;
1847 break;
1848 }
1849
1850 INP_UNLOCK_ASSERT(inp);
1851
1852 return (error);
1853 }
1854
1855 /*
1856 * Look up the ifnet to use for a multicast group membership,
1857 * given the IPv4 address of an interface, and the IPv4 group address.
1858 *
1859 * This routine exists to support legacy multicast applications
1860 * which do not understand that multicast memberships are scoped to
1861 * specific physical links in the networking stack, or which need
1862 * to join link-scope groups before IPv4 addresses are configured.
1863 *
1864 * If inp is non-NULL, use this socket's current FIB number for any
1865 * required FIB lookup.
1866 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1867 * and use its ifp; usually, this points to the default next-hop.
1868 *
1869 * If the FIB lookup fails, attempt to use the first non-loopback
1870 * interface with multicast capability in the system as a
1871 * last resort. The legacy IPv4 ASM API requires that we do
1872 * this in order to allow groups to be joined when the routing
1873 * table has not yet been populated during boot.
1874 *
1875 * Returns NULL if no ifp could be found.
1876 *
1877 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1878 * FUTURE: Implement IPv4 source-address selection.
1879 */
1880 static struct ifnet *
1881 inp_lookup_mcast_ifp(const struct inpcb *inp,
1882 const struct sockaddr_in *gsin, const struct in_addr ina)
1883 {
1884 struct rm_priotracker in_ifa_tracker;
1885 struct ifnet *ifp;
1886 struct nhop4_basic nh4;
1887 uint32_t fibnum;
1888
1889 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1890 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1891 ("%s: not multicast", __func__));
1892
1893 ifp = NULL;
1894 if (!in_nullhost(ina)) {
1895 INADDR_TO_IFP(ina, ifp);
1896 } else {
1897 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1898 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1899 ifp = nh4.nh_ifp;
1900 else {
1901 struct in_ifaddr *ia;
1902 struct ifnet *mifp;
1903
1904 mifp = NULL;
1905 IN_IFADDR_RLOCK(&in_ifa_tracker);
1906 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1907 mifp = ia->ia_ifp;
1908 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1909 (mifp->if_flags & IFF_MULTICAST)) {
1910 ifp = mifp;
1911 break;
1912 }
1913 }
1914 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1915 }
1916 }
1917
1918 return (ifp);
1919 }
1920
1921 /*
1922 * Join an IPv4 multicast group, possibly with a source.
1923 */
1924 static int
1925 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1926 {
1927 struct group_source_req gsr;
1928 sockunion_t *gsa, *ssa;
1929 struct ifnet *ifp;
1930 struct in_mfilter *imf;
1931 struct ip_moptions *imo;
1932 struct in_multi *inm;
1933 struct in_msource *lims;
1934 size_t idx;
1935 int error, is_new;
1936
1937 ifp = NULL;
1938 imf = NULL;
1939 lims = NULL;
1940 error = 0;
1941 is_new = 0;
1942
1943 memset(&gsr, 0, sizeof(struct group_source_req));
1944 gsa = (sockunion_t *)&gsr.gsr_group;
1945 gsa->ss.ss_family = AF_UNSPEC;
1946 ssa = (sockunion_t *)&gsr.gsr_source;
1947 ssa->ss.ss_family = AF_UNSPEC;
1948
1949 switch (sopt->sopt_name) {
1950 case IP_ADD_MEMBERSHIP:
1951 case IP_ADD_SOURCE_MEMBERSHIP: {
1952 struct ip_mreq_source mreqs;
1953
1954 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1955 error = sooptcopyin(sopt, &mreqs,
1956 sizeof(struct ip_mreq),
1957 sizeof(struct ip_mreq));
1958 /*
1959 * Do argument switcharoo from ip_mreq into
1960 * ip_mreq_source to avoid using two instances.
1961 */
1962 mreqs.imr_interface = mreqs.imr_sourceaddr;
1963 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1964 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1965 error = sooptcopyin(sopt, &mreqs,
1966 sizeof(struct ip_mreq_source),
1967 sizeof(struct ip_mreq_source));
1968 }
1969 if (error)
1970 return (error);
1971
1972 gsa->sin.sin_family = AF_INET;
1973 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1974 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1975
1976 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1977 ssa->sin.sin_family = AF_INET;
1978 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1979 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1980 }
1981
1982 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1983 return (EINVAL);
1984
1985 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1986 mreqs.imr_interface);
1987 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1988 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1989 break;
1990 }
1991
1992 case MCAST_JOIN_GROUP:
1993 case MCAST_JOIN_SOURCE_GROUP:
1994 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1995 error = sooptcopyin(sopt, &gsr,
1996 sizeof(struct group_req),
1997 sizeof(struct group_req));
1998 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1999 error = sooptcopyin(sopt, &gsr,
2000 sizeof(struct group_source_req),
2001 sizeof(struct group_source_req));
2002 }
2003 if (error)
2004 return (error);
2005
2006 if (gsa->sin.sin_family != AF_INET ||
2007 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2008 return (EINVAL);
2009
2010 /*
2011 * Overwrite the port field if present, as the sockaddr
2012 * being copied in may be matched with a binary comparison.
2013 */
2014 gsa->sin.sin_port = 0;
2015 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2016 if (ssa->sin.sin_family != AF_INET ||
2017 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2018 return (EINVAL);
2019 ssa->sin.sin_port = 0;
2020 }
2021
2022 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2023 return (EINVAL);
2024
2025 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2026 return (EADDRNOTAVAIL);
2027 ifp = ifnet_byindex(gsr.gsr_interface);
2028 break;
2029
2030 default:
2031 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2032 __func__, sopt->sopt_name);
2033 return (EOPNOTSUPP);
2034 break;
2035 }
2036
2037 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2038 return (EADDRNOTAVAIL);
2039
2040 imo = inp_findmoptions(inp);
2041 idx = imo_match_group(imo, ifp, &gsa->sa);
2042 if (idx == -1) {
2043 is_new = 1;
2044 } else {
2045 inm = imo->imo_membership[idx];
2046 imf = &imo->imo_mfilters[idx];
2047 if (ssa->ss.ss_family != AF_UNSPEC) {
2048 /*
2049 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2050 * is an error. On an existing inclusive membership,
2051 * it just adds the source to the filter list.
2052 */
2053 if (imf->imf_st[1] != MCAST_INCLUDE) {
2054 error = EINVAL;
2055 goto out_inp_locked;
2056 }
2057 /*
2058 * Throw out duplicates.
2059 *
2060 * XXX FIXME: This makes a naive assumption that
2061 * even if entries exist for *ssa in this imf,
2062 * they will be rejected as dupes, even if they
2063 * are not valid in the current mode (in-mode).
2064 *
2065 * in_msource is transactioned just as for anything
2066 * else in SSM -- but note naive use of inm_graft()
2067 * below for allocating new filter entries.
2068 *
2069 * This is only an issue if someone mixes the
2070 * full-state SSM API with the delta-based API,
2071 * which is discouraged in the relevant RFCs.
2072 */
2073 lims = imo_match_source(imo, idx, &ssa->sa);
2074 if (lims != NULL /*&&
2075 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2076 error = EADDRNOTAVAIL;
2077 goto out_inp_locked;
2078 }
2079 } else {
2080 /*
2081 * MCAST_JOIN_GROUP on an existing exclusive
2082 * membership is an error; return EADDRINUSE
2083 * to preserve 4.4BSD API idempotence, and
2084 * avoid tedious detour to code below.
2085 * NOTE: This is bending RFC 3678 a bit.
2086 *
2087 * On an existing inclusive membership, this is also
2088 * an error; if you want to change filter mode,
2089 * you must use the userland API setsourcefilter().
2090 * XXX We don't reject this for imf in UNDEFINED
2091 * state at t1, because allocation of a filter
2092 * is atomic with allocation of a membership.
2093 */
2094 error = EINVAL;
2095 if (imf->imf_st[1] == MCAST_EXCLUDE)
2096 error = EADDRINUSE;
2097 goto out_inp_locked;
2098 }
2099 }
2100
2101 /*
2102 * Begin state merge transaction at socket layer.
2103 */
2104 INP_WLOCK_ASSERT(inp);
2105
2106 if (is_new) {
2107 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2108 error = imo_grow(imo);
2109 if (error)
2110 goto out_inp_locked;
2111 }
2112 /*
2113 * Allocate the new slot upfront so we can deal with
2114 * grafting the new source filter in same code path
2115 * as for join-source on existing membership.
2116 */
2117 idx = imo->imo_num_memberships;
2118 imo->imo_membership[idx] = NULL;
2119 imo->imo_num_memberships++;
2120 KASSERT(imo->imo_mfilters != NULL,
2121 ("%s: imf_mfilters vector was not allocated", __func__));
2122 imf = &imo->imo_mfilters[idx];
2123 KASSERT(RB_EMPTY(&imf->imf_sources),
2124 ("%s: imf_sources not empty", __func__));
2125 }
2126
2127 /*
2128 * Graft new source into filter list for this inpcb's
2129 * membership of the group. The in_multi may not have
2130 * been allocated yet if this is a new membership, however,
2131 * the in_mfilter slot will be allocated and must be initialized.
2132 *
2133 * Note: Grafting of exclusive mode filters doesn't happen
2134 * in this path.
2135 * XXX: Should check for non-NULL lims (node exists but may
2136 * not be in-mode) for interop with full-state API.
2137 */
2138 if (ssa->ss.ss_family != AF_UNSPEC) {
2139 /* Membership starts in IN mode */
2140 if (is_new) {
2141 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2142 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2143 } else {
2144 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2145 }
2146 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2147 if (lims == NULL) {
2148 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2149 __func__);
2150 error = ENOMEM;
2151 goto out_imo_free;
2152 }
2153 } else {
2154 /* No address specified; Membership starts in EX mode */
2155 if (is_new) {
2156 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2157 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2158 }
2159 }
2160
2161 /*
2162 * Begin state merge transaction at IGMP layer.
2163 */
2164 IN_MULTI_LOCK();
2165
2166 if (is_new) {
2167 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2168 &inm);
2169 if (error) {
2170 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2171 __func__);
2172 IN_MULTI_UNLOCK();
2173 goto out_imo_free;
2174 }
2175 imo->imo_membership[idx] = inm;
2176 } else {
2177 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2178 error = inm_merge(inm, imf);
2179 if (error) {
2180 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2181 __func__);
2182 goto out_in_multi_locked;
2183 }
2184 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2185 error = igmp_change_state(inm);
2186 if (error) {
2187 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2188 __func__);
2189 goto out_in_multi_locked;
2190 }
2191 }
2192
2193 out_in_multi_locked:
2194
2195 IN_MULTI_UNLOCK();
2196
2197 INP_WLOCK_ASSERT(inp);
2198 if (error) {
2199 imf_rollback(imf);
2200 if (is_new)
2201 imf_purge(imf);
2202 else
2203 imf_reap(imf);
2204 } else {
2205 imf_commit(imf);
2206 }
2207
2208 out_imo_free:
2209 if (error && is_new) {
2210 imo->imo_membership[idx] = NULL;
2211 --imo->imo_num_memberships;
2212 }
2213
2214 out_inp_locked:
2215 INP_WUNLOCK(inp);
2216 return (error);
2217 }
2218
2219 /*
2220 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2221 */
2222 static int
2223 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2224 {
2225 struct group_source_req gsr;
2226 struct ip_mreq_source mreqs;
2227 sockunion_t *gsa, *ssa;
2228 struct ifnet *ifp;
2229 struct in_mfilter *imf;
2230 struct ip_moptions *imo;
2231 struct in_msource *ims;
2232 struct in_multi *inm;
2233 size_t idx;
2234 int error, is_final;
2235
2236 ifp = NULL;
2237 error = 0;
2238 is_final = 1;
2239
2240 memset(&gsr, 0, sizeof(struct group_source_req));
2241 gsa = (sockunion_t *)&gsr.gsr_group;
2242 gsa->ss.ss_family = AF_UNSPEC;
2243 ssa = (sockunion_t *)&gsr.gsr_source;
2244 ssa->ss.ss_family = AF_UNSPEC;
2245
2246 switch (sopt->sopt_name) {
2247 case IP_DROP_MEMBERSHIP:
2248 case IP_DROP_SOURCE_MEMBERSHIP:
2249 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2250 error = sooptcopyin(sopt, &mreqs,
2251 sizeof(struct ip_mreq),
2252 sizeof(struct ip_mreq));
2253 /*
2254 * Swap interface and sourceaddr arguments,
2255 * as ip_mreq and ip_mreq_source are laid
2256 * out differently.
2257 */
2258 mreqs.imr_interface = mreqs.imr_sourceaddr;
2259 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2260 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2261 error = sooptcopyin(sopt, &mreqs,
2262 sizeof(struct ip_mreq_source),
2263 sizeof(struct ip_mreq_source));
2264 }
2265 if (error)
2266 return (error);
2267
2268 gsa->sin.sin_family = AF_INET;
2269 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2270 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2271
2272 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2273 ssa->sin.sin_family = AF_INET;
2274 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2275 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2276 }
2277
2278 /*
2279 * Attempt to look up hinted ifp from interface address.
2280 * Fallthrough with null ifp iff lookup fails, to
2281 * preserve 4.4BSD mcast API idempotence.
2282 * XXX NOTE WELL: The RFC 3678 API is preferred because
2283 * using an IPv4 address as a key is racy.
2284 */
2285 if (!in_nullhost(mreqs.imr_interface))
2286 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2287
2288 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2289 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2290
2291 break;
2292
2293 case MCAST_LEAVE_GROUP:
2294 case MCAST_LEAVE_SOURCE_GROUP:
2295 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2296 error = sooptcopyin(sopt, &gsr,
2297 sizeof(struct group_req),
2298 sizeof(struct group_req));
2299 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2300 error = sooptcopyin(sopt, &gsr,
2301 sizeof(struct group_source_req),
2302 sizeof(struct group_source_req));
2303 }
2304 if (error)
2305 return (error);
2306
2307 if (gsa->sin.sin_family != AF_INET ||
2308 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2309 return (EINVAL);
2310
2311 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2312 if (ssa->sin.sin_family != AF_INET ||
2313 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2314 return (EINVAL);
2315 }
2316
2317 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2318 return (EADDRNOTAVAIL);
2319
2320 ifp = ifnet_byindex(gsr.gsr_interface);
2321
2322 if (ifp == NULL)
2323 return (EADDRNOTAVAIL);
2324 break;
2325
2326 default:
2327 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2328 __func__, sopt->sopt_name);
2329 return (EOPNOTSUPP);
2330 break;
2331 }
2332
2333 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2334 return (EINVAL);
2335
2336 /*
2337 * Find the membership in the membership array.
2338 */
2339 imo = inp_findmoptions(inp);
2340 idx = imo_match_group(imo, ifp, &gsa->sa);
2341 if (idx == -1) {
2342 error = EADDRNOTAVAIL;
2343 goto out_inp_locked;
2344 }
2345 inm = imo->imo_membership[idx];
2346 imf = &imo->imo_mfilters[idx];
2347
2348 if (ssa->ss.ss_family != AF_UNSPEC)
2349 is_final = 0;
2350
2351 /*
2352 * Begin state merge transaction at socket layer.
2353 */
2354 INP_WLOCK_ASSERT(inp);
2355
2356 /*
2357 * If we were instructed only to leave a given source, do so.
2358 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2359 */
2360 if (is_final) {
2361 imf_leave(imf);
2362 } else {
2363 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2364 error = EADDRNOTAVAIL;
2365 goto out_inp_locked;
2366 }
2367 ims = imo_match_source(imo, idx, &ssa->sa);
2368 if (ims == NULL) {
2369 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2370 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2371 error = EADDRNOTAVAIL;
2372 goto out_inp_locked;
2373 }
2374 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2375 error = imf_prune(imf, &ssa->sin);
2376 if (error) {
2377 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2378 __func__);
2379 goto out_inp_locked;
2380 }
2381 }
2382
2383 /*
2384 * Begin state merge transaction at IGMP layer.
2385 */
2386 IN_MULTI_LOCK();
2387
2388 if (is_final) {
2389 /*
2390 * Give up the multicast address record to which
2391 * the membership points.
2392 */
2393 (void)in_leavegroup_locked(inm, imf);
2394 } else {
2395 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2396 error = inm_merge(inm, imf);
2397 if (error) {
2398 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2399 __func__);
2400 goto out_in_multi_locked;
2401 }
2402
2403 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2404 error = igmp_change_state(inm);
2405 if (error) {
2406 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2407 __func__);
2408 }
2409 }
2410
2411 out_in_multi_locked:
2412
2413 IN_MULTI_UNLOCK();
2414
2415 if (error)
2416 imf_rollback(imf);
2417 else
2418 imf_commit(imf);
2419
2420 imf_reap(imf);
2421
2422 if (is_final) {
2423 /* Remove the gap in the membership and filter array. */
2424 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2425 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2426 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2427 }
2428 imo->imo_num_memberships--;
2429 }
2430
2431 out_inp_locked:
2432 INP_WUNLOCK(inp);
2433 return (error);
2434 }
2435
2436 /*
2437 * Select the interface for transmitting IPv4 multicast datagrams.
2438 *
2439 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2440 * may be passed to this socket option. An address of INADDR_ANY or an
2441 * interface index of 0 is used to remove a previous selection.
2442 * When no interface is selected, one is chosen for every send.
2443 */
2444 static int
2445 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2446 {
2447 struct in_addr addr;
2448 struct ip_mreqn mreqn;
2449 struct ifnet *ifp;
2450 struct ip_moptions *imo;
2451 int error;
2452
2453 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2454 /*
2455 * An interface index was specified using the
2456 * Linux-derived ip_mreqn structure.
2457 */
2458 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2459 sizeof(struct ip_mreqn));
2460 if (error)
2461 return (error);
2462
2463 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2464 return (EINVAL);
2465
2466 if (mreqn.imr_ifindex == 0) {
2467 ifp = NULL;
2468 } else {
2469 ifp = ifnet_byindex(mreqn.imr_ifindex);
2470 if (ifp == NULL)
2471 return (EADDRNOTAVAIL);
2472 }
2473 } else {
2474 /*
2475 * An interface was specified by IPv4 address.
2476 * This is the traditional BSD usage.
2477 */
2478 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2479 sizeof(struct in_addr));
2480 if (error)
2481 return (error);
2482 if (in_nullhost(addr)) {
2483 ifp = NULL;
2484 } else {
2485 INADDR_TO_IFP(addr, ifp);
2486 if (ifp == NULL)
2487 return (EADDRNOTAVAIL);
2488 }
2489 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2490 ntohl(addr.s_addr));
2491 }
2492
2493 /* Reject interfaces which do not support multicast. */
2494 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2495 return (EOPNOTSUPP);
2496
2497 imo = inp_findmoptions(inp);
2498 imo->imo_multicast_ifp = ifp;
2499 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2500 INP_WUNLOCK(inp);
2501
2502 return (0);
2503 }
2504
2505 /*
2506 * Atomically set source filters on a socket for an IPv4 multicast group.
2507 *
2508 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2509 */
2510 static int
2511 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2512 {
2513 struct __msfilterreq msfr;
2514 sockunion_t *gsa;
2515 struct ifnet *ifp;
2516 struct in_mfilter *imf;
2517 struct ip_moptions *imo;
2518 struct in_multi *inm;
2519 size_t idx;
2520 int error;
2521
2522 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2523 sizeof(struct __msfilterreq));
2524 if (error)
2525 return (error);
2526
2527 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2528 return (ENOBUFS);
2529
2530 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2531 msfr.msfr_fmode != MCAST_INCLUDE))
2532 return (EINVAL);
2533
2534 if (msfr.msfr_group.ss_family != AF_INET ||
2535 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2536 return (EINVAL);
2537
2538 gsa = (sockunion_t *)&msfr.msfr_group;
2539 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2540 return (EINVAL);
2541
2542 gsa->sin.sin_port = 0; /* ignore port */
2543
2544 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2545 return (EADDRNOTAVAIL);
2546
2547 ifp = ifnet_byindex(msfr.msfr_ifindex);
2548 if (ifp == NULL)
2549 return (EADDRNOTAVAIL);
2550
2551 /*
2552 * Take the INP write lock.
2553 * Check if this socket is a member of this group.
2554 */
2555 imo = inp_findmoptions(inp);
2556 idx = imo_match_group(imo, ifp, &gsa->sa);
2557 if (idx == -1 || imo->imo_mfilters == NULL) {
2558 error = EADDRNOTAVAIL;
2559 goto out_inp_locked;
2560 }
2561 inm = imo->imo_membership[idx];
2562 imf = &imo->imo_mfilters[idx];
2563
2564 /*
2565 * Begin state merge transaction at socket layer.
2566 */
2567 INP_WLOCK_ASSERT(inp);
2568
2569 imf->imf_st[1] = msfr.msfr_fmode;
2570
2571 /*
2572 * Apply any new source filters, if present.
2573 * Make a copy of the user-space source vector so
2574 * that we may copy them with a single copyin. This
2575 * allows us to deal with page faults up-front.
2576 */
2577 if (msfr.msfr_nsrcs > 0) {
2578 struct in_msource *lims;
2579 struct sockaddr_in *psin;
2580 struct sockaddr_storage *kss, *pkss;
2581 int i;
2582
2583 INP_WUNLOCK(inp);
2584
2585 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2586 __func__, (unsigned long)msfr.msfr_nsrcs);
2587 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2588 M_TEMP, M_WAITOK);
2589 error = copyin(msfr.msfr_srcs, kss,
2590 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2591 if (error) {
2592 free(kss, M_TEMP);
2593 return (error);
2594 }
2595
2596 INP_WLOCK(inp);
2597
2598 /*
2599 * Mark all source filters as UNDEFINED at t1.
2600 * Restore new group filter mode, as imf_leave()
2601 * will set it to INCLUDE.
2602 */
2603 imf_leave(imf);
2604 imf->imf_st[1] = msfr.msfr_fmode;
2605
2606 /*
2607 * Update socket layer filters at t1, lazy-allocating
2608 * new entries. This saves a bunch of memory at the
2609 * cost of one RB_FIND() per source entry; duplicate
2610 * entries in the msfr_nsrcs vector are ignored.
2611 * If we encounter an error, rollback transaction.
2612 *
2613 * XXX This too could be replaced with a set-symmetric
2614 * difference like loop to avoid walking from root
2615 * every time, as the key space is common.
2616 */
2617 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2618 psin = (struct sockaddr_in *)pkss;
2619 if (psin->sin_family != AF_INET) {
2620 error = EAFNOSUPPORT;
2621 break;
2622 }
2623 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2624 error = EINVAL;
2625 break;
2626 }
2627 error = imf_get_source(imf, psin, &lims);
2628 if (error)
2629 break;
2630 lims->imsl_st[1] = imf->imf_st[1];
2631 }
2632 free(kss, M_TEMP);
2633 }
2634
2635 if (error)
2636 goto out_imf_rollback;
2637
2638 INP_WLOCK_ASSERT(inp);
2639 IN_MULTI_LOCK();
2640
2641 /*
2642 * Begin state merge transaction at IGMP layer.
2643 */
2644 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2645 error = inm_merge(inm, imf);
2646 if (error) {
2647 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2648 goto out_in_multi_locked;
2649 }
2650
2651 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2652 error = igmp_change_state(inm);
2653 if (error)
2654 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2655
2656 out_in_multi_locked:
2657
2658 IN_MULTI_UNLOCK();
2659
2660 out_imf_rollback:
2661 if (error)
2662 imf_rollback(imf);
2663 else
2664 imf_commit(imf);
2665
2666 imf_reap(imf);
2667
2668 out_inp_locked:
2669 INP_WUNLOCK(inp);
2670 return (error);
2671 }
2672
2673 /*
2674 * Set the IP multicast options in response to user setsockopt().
2675 *
2676 * Many of the socket options handled in this function duplicate the
2677 * functionality of socket options in the regular unicast API. However,
2678 * it is not possible to merge the duplicate code, because the idempotence
2679 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2680 * the effects of these options must be treated as separate and distinct.
2681 *
2682 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2683 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2684 * is refactored to no longer use vifs.
2685 */
2686 int
2687 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2688 {
2689 struct ip_moptions *imo;
2690 int error;
2691
2692 error = 0;
2693
2694 /*
2695 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2696 * or is a divert socket, reject it.
2697 */
2698 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2699 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2700 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2701 return (EOPNOTSUPP);
2702
2703 switch (sopt->sopt_name) {
2704 case IP_MULTICAST_VIF: {
2705 int vifi;
2706 /*
2707 * Select a multicast VIF for transmission.
2708 * Only useful if multicast forwarding is active.
2709 */
2710 if (legal_vif_num == NULL) {
2711 error = EOPNOTSUPP;
2712 break;
2713 }
2714 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2715 if (error)
2716 break;
2717 if (!legal_vif_num(vifi) && (vifi != -1)) {
2718 error = EINVAL;
2719 break;
2720 }
2721 imo = inp_findmoptions(inp);
2722 imo->imo_multicast_vif = vifi;
2723 INP_WUNLOCK(inp);
2724 break;
2725 }
2726
2727 case IP_MULTICAST_IF:
2728 error = inp_set_multicast_if(inp, sopt);
2729 break;
2730
2731 case IP_MULTICAST_TTL: {
2732 u_char ttl;
2733
2734 /*
2735 * Set the IP time-to-live for outgoing multicast packets.
2736 * The original multicast API required a char argument,
2737 * which is inconsistent with the rest of the socket API.
2738 * We allow either a char or an int.
2739 */
2740 if (sopt->sopt_valsize == sizeof(u_char)) {
2741 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2742 sizeof(u_char));
2743 if (error)
2744 break;
2745 } else {
2746 u_int ittl;
2747
2748 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2749 sizeof(u_int));
2750 if (error)
2751 break;
2752 if (ittl > 255) {
2753 error = EINVAL;
2754 break;
2755 }
2756 ttl = (u_char)ittl;
2757 }
2758 imo = inp_findmoptions(inp);
2759 imo->imo_multicast_ttl = ttl;
2760 INP_WUNLOCK(inp);
2761 break;
2762 }
2763
2764 case IP_MULTICAST_LOOP: {
2765 u_char loop;
2766
2767 /*
2768 * Set the loopback flag for outgoing multicast packets.
2769 * Must be zero or one. The original multicast API required a
2770 * char argument, which is inconsistent with the rest
2771 * of the socket API. We allow either a char or an int.
2772 */
2773 if (sopt->sopt_valsize == sizeof(u_char)) {
2774 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2775 sizeof(u_char));
2776 if (error)
2777 break;
2778 } else {
2779 u_int iloop;
2780
2781 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2782 sizeof(u_int));
2783 if (error)
2784 break;
2785 loop = (u_char)iloop;
2786 }
2787 imo = inp_findmoptions(inp);
2788 imo->imo_multicast_loop = !!loop;
2789 INP_WUNLOCK(inp);
2790 break;
2791 }
2792
2793 case IP_ADD_MEMBERSHIP:
2794 case IP_ADD_SOURCE_MEMBERSHIP:
2795 case MCAST_JOIN_GROUP:
2796 case MCAST_JOIN_SOURCE_GROUP:
2797 error = inp_join_group(inp, sopt);
2798 break;
2799
2800 case IP_DROP_MEMBERSHIP:
2801 case IP_DROP_SOURCE_MEMBERSHIP:
2802 case MCAST_LEAVE_GROUP:
2803 case MCAST_LEAVE_SOURCE_GROUP:
2804 error = inp_leave_group(inp, sopt);
2805 break;
2806
2807 case IP_BLOCK_SOURCE:
2808 case IP_UNBLOCK_SOURCE:
2809 case MCAST_BLOCK_SOURCE:
2810 case MCAST_UNBLOCK_SOURCE:
2811 error = inp_block_unblock_source(inp, sopt);
2812 break;
2813
2814 case IP_MSFILTER:
2815 error = inp_set_source_filters(inp, sopt);
2816 break;
2817
2818 default:
2819 error = EOPNOTSUPP;
2820 break;
2821 }
2822
2823 INP_UNLOCK_ASSERT(inp);
2824
2825 return (error);
2826 }
2827
2828 /*
2829 * Expose IGMP's multicast filter mode and source list(s) to userland,
2830 * keyed by (ifindex, group).
2831 * The filter mode is written out as a uint32_t, followed by
2832 * 0..n of struct in_addr.
2833 * For use by ifmcstat(8).
2834 * SMPng: NOTE: unlocked read of ifindex space.
2835 */
2836 static int
2837 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2838 {
2839 struct in_addr src, group;
2840 struct ifnet *ifp;
2841 struct ifmultiaddr *ifma;
2842 struct in_multi *inm;
2843 struct ip_msource *ims;
2844 int *name;
2845 int retval;
2846 u_int namelen;
2847 uint32_t fmode, ifindex;
2848
2849 name = (int *)arg1;
2850 namelen = arg2;
2851
2852 if (req->newptr != NULL)
2853 return (EPERM);
2854
2855 if (namelen != 2)
2856 return (EINVAL);
2857
2858 ifindex = name[0];
2859 if (ifindex <= 0 || ifindex > V_if_index) {
2860 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2861 __func__, ifindex);
2862 return (ENOENT);
2863 }
2864
2865 group.s_addr = name[1];
2866 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2867 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2868 __func__, ntohl(group.s_addr));
2869 return (EINVAL);
2870 }
2871
2872 ifp = ifnet_byindex(ifindex);
2873 if (ifp == NULL) {
2874 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2875 __func__, ifindex);
2876 return (ENOENT);
2877 }
2878
2879 retval = sysctl_wire_old_buffer(req,
2880 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2881 if (retval)
2882 return (retval);
2883
2884 IN_MULTI_LOCK();
2885
2886 IF_ADDR_RLOCK(ifp);
2887 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2888 if (ifma->ifma_addr->sa_family != AF_INET ||
2889 ifma->ifma_protospec == NULL)
2890 continue;
2891 inm = (struct in_multi *)ifma->ifma_protospec;
2892 if (!in_hosteq(inm->inm_addr, group))
2893 continue;
2894 fmode = inm->inm_st[1].iss_fmode;
2895 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2896 if (retval != 0)
2897 break;
2898 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2899 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2900 ims->ims_haddr);
2901 /*
2902 * Only copy-out sources which are in-mode.
2903 */
2904 if (fmode != ims_get_mode(inm, ims, 1)) {
2905 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2906 __func__);
2907 continue;
2908 }
2909 src.s_addr = htonl(ims->ims_haddr);
2910 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2911 if (retval != 0)
2912 break;
2913 }
2914 }
2915 IF_ADDR_RUNLOCK(ifp);
2916
2917 IN_MULTI_UNLOCK();
2918
2919 return (retval);
2920 }
2921
2922 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2923
2924 static const char *inm_modestrs[] = { "un", "in", "ex" };
2925
2926 static const char *
2927 inm_mode_str(const int mode)
2928 {
2929
2930 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2931 return (inm_modestrs[mode]);
2932 return ("??");
2933 }
2934
2935 static const char *inm_statestrs[] = {
2936 "not-member",
2937 "silent",
2938 "idle",
2939 "lazy",
2940 "sleeping",
2941 "awakening",
2942 "query-pending",
2943 "sg-query-pending",
2944 "leaving"
2945 };
2946
2947 static const char *
2948 inm_state_str(const int state)
2949 {
2950
2951 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2952 return (inm_statestrs[state]);
2953 return ("??");
2954 }
2955
2956 /*
2957 * Dump an in_multi structure to the console.
2958 */
2959 void
2960 inm_print(const struct in_multi *inm)
2961 {
2962 int t;
2963 char addrbuf[INET_ADDRSTRLEN];
2964
2965 if ((ktr_mask & KTR_IGMPV3) == 0)
2966 return;
2967
2968 printf("%s: --- begin inm %p ---\n", __func__, inm);
2969 printf("addr %s ifp %p(%s) ifma %p\n",
2970 inet_ntoa_r(inm->inm_addr, addrbuf),
2971 inm->inm_ifp,
2972 inm->inm_ifp->if_xname,
2973 inm->inm_ifma);
2974 printf("timer %u state %s refcount %u scq.len %u\n",
2975 inm->inm_timer,
2976 inm_state_str(inm->inm_state),
2977 inm->inm_refcount,
2978 inm->inm_scq.mq_len);
2979 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2980 inm->inm_igi,
2981 inm->inm_nsrc,
2982 inm->inm_sctimer,
2983 inm->inm_scrv);
2984 for (t = 0; t < 2; t++) {
2985 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2986 inm_mode_str(inm->inm_st[t].iss_fmode),
2987 inm->inm_st[t].iss_asm,
2988 inm->inm_st[t].iss_ex,
2989 inm->inm_st[t].iss_in,
2990 inm->inm_st[t].iss_rec);
2991 }
2992 printf("%s: --- end inm %p ---\n", __func__, inm);
2993 }
2994
2995 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
2996
2997 void
2998 inm_print(const struct in_multi *inm)
2999 {
3000
3001 }
3002
3003 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3004
3005 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);
Cache object: 9af04276560bb8dfe43d47ca8c1a31bc
|