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/11.1/sys/netinet/in_mcast.c 315456 2017-03-17 14:54:10Z vangyzen $");
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 /* Increment ASM listener count on transition to ASM mode. */
1056 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1057 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1058 inm->inm_st[1].iss_asm++;
1059 }
1060
1061 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1062 inm_print(inm);
1063
1064 out_reap:
1065 if (schanged > 0) {
1066 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1067 inm_reap(inm);
1068 }
1069 return (error);
1070 }
1071
1072 /*
1073 * Mark an in_multi's filter set deltas as committed.
1074 * Called by IGMP after a state change has been enqueued.
1075 */
1076 void
1077 inm_commit(struct in_multi *inm)
1078 {
1079 struct ip_msource *ims;
1080
1081 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1082 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1083 inm_print(inm);
1084
1085 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1086 ims->ims_st[0] = ims->ims_st[1];
1087 }
1088 inm->inm_st[0] = inm->inm_st[1];
1089 }
1090
1091 /*
1092 * Reap unreferenced nodes from an in_multi's filter set.
1093 */
1094 static void
1095 inm_reap(struct in_multi *inm)
1096 {
1097 struct ip_msource *ims, *tims;
1098
1099 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1100 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1101 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1102 ims->ims_stp != 0)
1103 continue;
1104 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1105 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1106 free(ims, M_IPMSOURCE);
1107 inm->inm_nsrc--;
1108 }
1109 }
1110
1111 /*
1112 * Purge all source nodes from an in_multi's filter set.
1113 */
1114 static void
1115 inm_purge(struct in_multi *inm)
1116 {
1117 struct ip_msource *ims, *tims;
1118
1119 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1120 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1121 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1122 free(ims, M_IPMSOURCE);
1123 inm->inm_nsrc--;
1124 }
1125 }
1126
1127 /*
1128 * Join a multicast group; unlocked entry point.
1129 *
1130 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1131 * is not held. Fortunately, ifp is unlikely to have been detached
1132 * at this point, so we assume it's OK to recurse.
1133 */
1134 int
1135 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1136 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1137 {
1138 int error;
1139
1140 IN_MULTI_LOCK();
1141 error = in_joingroup_locked(ifp, gina, imf, pinm);
1142 IN_MULTI_UNLOCK();
1143
1144 return (error);
1145 }
1146
1147 /*
1148 * Join a multicast group; real entry point.
1149 *
1150 * Only preserves atomicity at inm level.
1151 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1152 *
1153 * If the IGMP downcall fails, the group is not joined, and an error
1154 * code is returned.
1155 */
1156 int
1157 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1158 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1159 {
1160 struct in_mfilter timf;
1161 struct in_multi *inm;
1162 int error;
1163
1164 IN_MULTI_LOCK_ASSERT();
1165
1166 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1167 ntohl(gina->s_addr), ifp, ifp->if_xname);
1168
1169 error = 0;
1170 inm = NULL;
1171
1172 /*
1173 * If no imf was specified (i.e. kernel consumer),
1174 * fake one up and assume it is an ASM join.
1175 */
1176 if (imf == NULL) {
1177 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1178 imf = &timf;
1179 }
1180
1181 error = in_getmulti(ifp, gina, &inm);
1182 if (error) {
1183 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1184 return (error);
1185 }
1186
1187 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1188 error = inm_merge(inm, imf);
1189 if (error) {
1190 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1191 goto out_inm_release;
1192 }
1193
1194 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1195 error = igmp_change_state(inm);
1196 if (error) {
1197 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1198 goto out_inm_release;
1199 }
1200
1201 out_inm_release:
1202 if (error) {
1203 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1204 inm_release_locked(inm);
1205 } else {
1206 *pinm = inm;
1207 }
1208
1209 return (error);
1210 }
1211
1212 /*
1213 * Leave a multicast group; unlocked entry point.
1214 */
1215 int
1216 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1217 {
1218 int error;
1219
1220 IN_MULTI_LOCK();
1221 error = in_leavegroup_locked(inm, imf);
1222 IN_MULTI_UNLOCK();
1223
1224 return (error);
1225 }
1226
1227 /*
1228 * Leave a multicast group; real entry point.
1229 * All source filters will be expunged.
1230 *
1231 * Only preserves atomicity at inm level.
1232 *
1233 * Holding the write lock for the INP which contains imf
1234 * is highly advisable. We can't assert for it as imf does not
1235 * contain a back-pointer to the owning inp.
1236 *
1237 * Note: This is not the same as inm_release(*) as this function also
1238 * makes a state change downcall into IGMP.
1239 */
1240 int
1241 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1242 {
1243 struct in_mfilter timf;
1244 int error;
1245
1246 error = 0;
1247
1248 IN_MULTI_LOCK_ASSERT();
1249
1250 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1251 inm, ntohl(inm->inm_addr.s_addr),
1252 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1253 imf);
1254
1255 /*
1256 * If no imf was specified (i.e. kernel consumer),
1257 * fake one up and assume it is an ASM join.
1258 */
1259 if (imf == NULL) {
1260 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1261 imf = &timf;
1262 }
1263
1264 /*
1265 * Begin state merge transaction at IGMP layer.
1266 *
1267 * As this particular invocation should not cause any memory
1268 * to be allocated, and there is no opportunity to roll back
1269 * the transaction, it MUST NOT fail.
1270 */
1271 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1272 error = inm_merge(inm, imf);
1273 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1274
1275 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1276 CURVNET_SET(inm->inm_ifp->if_vnet);
1277 error = igmp_change_state(inm);
1278 CURVNET_RESTORE();
1279 if (error)
1280 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1281
1282 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1283 inm_release_locked(inm);
1284
1285 return (error);
1286 }
1287
1288 /*#ifndef BURN_BRIDGES*/
1289 /*
1290 * Join an IPv4 multicast group in (*,G) exclusive mode.
1291 * The group must be a 224.0.0.0/24 link-scope group.
1292 * This KPI is for legacy kernel consumers only.
1293 */
1294 struct in_multi *
1295 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1296 {
1297 struct in_multi *pinm;
1298 int error;
1299 #ifdef INVARIANTS
1300 char addrbuf[INET_ADDRSTRLEN];
1301 #endif
1302
1303 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1304 ("%s: %s not in 224.0.0.0/24", __func__,
1305 inet_ntoa_r(*ap, addrbuf)));
1306
1307 error = in_joingroup(ifp, ap, NULL, &pinm);
1308 if (error != 0)
1309 pinm = NULL;
1310
1311 return (pinm);
1312 }
1313
1314 /*
1315 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1316 * This KPI is for legacy kernel consumers only.
1317 */
1318 void
1319 in_delmulti(struct in_multi *inm)
1320 {
1321
1322 (void)in_leavegroup(inm, NULL);
1323 }
1324 /*#endif*/
1325
1326 /*
1327 * Block or unblock an ASM multicast source on an inpcb.
1328 * This implements the delta-based API described in RFC 3678.
1329 *
1330 * The delta-based API applies only to exclusive-mode memberships.
1331 * An IGMP downcall will be performed.
1332 *
1333 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1334 *
1335 * Return 0 if successful, otherwise return an appropriate error code.
1336 */
1337 static int
1338 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1339 {
1340 struct group_source_req gsr;
1341 sockunion_t *gsa, *ssa;
1342 struct ifnet *ifp;
1343 struct in_mfilter *imf;
1344 struct ip_moptions *imo;
1345 struct in_msource *ims;
1346 struct in_multi *inm;
1347 size_t idx;
1348 uint16_t fmode;
1349 int error, doblock;
1350
1351 ifp = NULL;
1352 error = 0;
1353 doblock = 0;
1354
1355 memset(&gsr, 0, sizeof(struct group_source_req));
1356 gsa = (sockunion_t *)&gsr.gsr_group;
1357 ssa = (sockunion_t *)&gsr.gsr_source;
1358
1359 switch (sopt->sopt_name) {
1360 case IP_BLOCK_SOURCE:
1361 case IP_UNBLOCK_SOURCE: {
1362 struct ip_mreq_source mreqs;
1363
1364 error = sooptcopyin(sopt, &mreqs,
1365 sizeof(struct ip_mreq_source),
1366 sizeof(struct ip_mreq_source));
1367 if (error)
1368 return (error);
1369
1370 gsa->sin.sin_family = AF_INET;
1371 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1372 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1373
1374 ssa->sin.sin_family = AF_INET;
1375 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1376 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1377
1378 if (!in_nullhost(mreqs.imr_interface))
1379 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1380
1381 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1382 doblock = 1;
1383
1384 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1385 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1386 break;
1387 }
1388
1389 case MCAST_BLOCK_SOURCE:
1390 case MCAST_UNBLOCK_SOURCE:
1391 error = sooptcopyin(sopt, &gsr,
1392 sizeof(struct group_source_req),
1393 sizeof(struct group_source_req));
1394 if (error)
1395 return (error);
1396
1397 if (gsa->sin.sin_family != AF_INET ||
1398 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1399 return (EINVAL);
1400
1401 if (ssa->sin.sin_family != AF_INET ||
1402 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1403 return (EINVAL);
1404
1405 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1406 return (EADDRNOTAVAIL);
1407
1408 ifp = ifnet_byindex(gsr.gsr_interface);
1409
1410 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1411 doblock = 1;
1412 break;
1413
1414 default:
1415 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1416 __func__, sopt->sopt_name);
1417 return (EOPNOTSUPP);
1418 break;
1419 }
1420
1421 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1422 return (EINVAL);
1423
1424 /*
1425 * Check if we are actually a member of this group.
1426 */
1427 imo = inp_findmoptions(inp);
1428 idx = imo_match_group(imo, ifp, &gsa->sa);
1429 if (idx == -1 || imo->imo_mfilters == NULL) {
1430 error = EADDRNOTAVAIL;
1431 goto out_inp_locked;
1432 }
1433
1434 KASSERT(imo->imo_mfilters != NULL,
1435 ("%s: imo_mfilters not allocated", __func__));
1436 imf = &imo->imo_mfilters[idx];
1437 inm = imo->imo_membership[idx];
1438
1439 /*
1440 * Attempting to use the delta-based API on an
1441 * non exclusive-mode membership is an error.
1442 */
1443 fmode = imf->imf_st[0];
1444 if (fmode != MCAST_EXCLUDE) {
1445 error = EINVAL;
1446 goto out_inp_locked;
1447 }
1448
1449 /*
1450 * Deal with error cases up-front:
1451 * Asked to block, but already blocked; or
1452 * Asked to unblock, but nothing to unblock.
1453 * If adding a new block entry, allocate it.
1454 */
1455 ims = imo_match_source(imo, idx, &ssa->sa);
1456 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1457 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1458 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1459 error = EADDRNOTAVAIL;
1460 goto out_inp_locked;
1461 }
1462
1463 INP_WLOCK_ASSERT(inp);
1464
1465 /*
1466 * Begin state merge transaction at socket layer.
1467 */
1468 if (doblock) {
1469 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1470 ims = imf_graft(imf, fmode, &ssa->sin);
1471 if (ims == NULL)
1472 error = ENOMEM;
1473 } else {
1474 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1475 error = imf_prune(imf, &ssa->sin);
1476 }
1477
1478 if (error) {
1479 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1480 goto out_imf_rollback;
1481 }
1482
1483 /*
1484 * Begin state merge transaction at IGMP layer.
1485 */
1486 IN_MULTI_LOCK();
1487
1488 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1489 error = inm_merge(inm, imf);
1490 if (error) {
1491 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1492 goto out_in_multi_locked;
1493 }
1494
1495 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1496 error = igmp_change_state(inm);
1497 if (error)
1498 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1499
1500 out_in_multi_locked:
1501
1502 IN_MULTI_UNLOCK();
1503
1504 out_imf_rollback:
1505 if (error)
1506 imf_rollback(imf);
1507 else
1508 imf_commit(imf);
1509
1510 imf_reap(imf);
1511
1512 out_inp_locked:
1513 INP_WUNLOCK(inp);
1514 return (error);
1515 }
1516
1517 /*
1518 * Given an inpcb, return its multicast options structure pointer. Accepts
1519 * an unlocked inpcb pointer, but will return it locked. May sleep.
1520 *
1521 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1522 * SMPng: NOTE: Returns with the INP write lock held.
1523 */
1524 static struct ip_moptions *
1525 inp_findmoptions(struct inpcb *inp)
1526 {
1527 struct ip_moptions *imo;
1528 struct in_multi **immp;
1529 struct in_mfilter *imfp;
1530 size_t idx;
1531
1532 INP_WLOCK(inp);
1533 if (inp->inp_moptions != NULL)
1534 return (inp->inp_moptions);
1535
1536 INP_WUNLOCK(inp);
1537
1538 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1539 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1540 M_WAITOK | M_ZERO);
1541 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1542 M_INMFILTER, M_WAITOK);
1543
1544 imo->imo_multicast_ifp = NULL;
1545 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1546 imo->imo_multicast_vif = -1;
1547 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1548 imo->imo_multicast_loop = in_mcast_loop;
1549 imo->imo_num_memberships = 0;
1550 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1551 imo->imo_membership = immp;
1552
1553 /* Initialize per-group source filters. */
1554 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1555 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1556 imo->imo_mfilters = imfp;
1557
1558 INP_WLOCK(inp);
1559 if (inp->inp_moptions != NULL) {
1560 free(imfp, M_INMFILTER);
1561 free(immp, M_IPMOPTS);
1562 free(imo, M_IPMOPTS);
1563 return (inp->inp_moptions);
1564 }
1565 inp->inp_moptions = imo;
1566 return (imo);
1567 }
1568
1569 /*
1570 * Discard the IP multicast options (and source filters). To minimize
1571 * the amount of work done while holding locks such as the INP's
1572 * pcbinfo lock (which is used in the receive path), the free
1573 * operation is performed asynchronously in a separate task.
1574 *
1575 * SMPng: NOTE: assumes INP write lock is held.
1576 */
1577 void
1578 inp_freemoptions(struct ip_moptions *imo)
1579 {
1580
1581 KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1582 IN_MULTI_LOCK();
1583 STAILQ_INSERT_TAIL(&imo_gc_list, imo, imo_link);
1584 IN_MULTI_UNLOCK();
1585 taskqueue_enqueue(taskqueue_thread, &imo_gc_task);
1586 }
1587
1588 static void
1589 inp_freemoptions_internal(struct ip_moptions *imo)
1590 {
1591 struct in_mfilter *imf;
1592 size_t idx, nmships;
1593
1594 nmships = imo->imo_num_memberships;
1595 for (idx = 0; idx < nmships; ++idx) {
1596 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1597 if (imf)
1598 imf_leave(imf);
1599 (void)in_leavegroup(imo->imo_membership[idx], imf);
1600 if (imf)
1601 imf_purge(imf);
1602 }
1603
1604 if (imo->imo_mfilters)
1605 free(imo->imo_mfilters, M_INMFILTER);
1606 free(imo->imo_membership, M_IPMOPTS);
1607 free(imo, M_IPMOPTS);
1608 }
1609
1610 static void
1611 inp_gcmoptions(void *context, int pending)
1612 {
1613 struct ip_moptions *imo;
1614
1615 IN_MULTI_LOCK();
1616 while (!STAILQ_EMPTY(&imo_gc_list)) {
1617 imo = STAILQ_FIRST(&imo_gc_list);
1618 STAILQ_REMOVE_HEAD(&imo_gc_list, imo_link);
1619 IN_MULTI_UNLOCK();
1620 inp_freemoptions_internal(imo);
1621 IN_MULTI_LOCK();
1622 }
1623 IN_MULTI_UNLOCK();
1624 }
1625
1626 /*
1627 * Atomically get source filters on a socket for an IPv4 multicast group.
1628 * Called with INP lock held; returns with lock released.
1629 */
1630 static int
1631 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1632 {
1633 struct __msfilterreq msfr;
1634 sockunion_t *gsa;
1635 struct ifnet *ifp;
1636 struct ip_moptions *imo;
1637 struct in_mfilter *imf;
1638 struct ip_msource *ims;
1639 struct in_msource *lims;
1640 struct sockaddr_in *psin;
1641 struct sockaddr_storage *ptss;
1642 struct sockaddr_storage *tss;
1643 int error;
1644 size_t idx, nsrcs, ncsrcs;
1645
1646 INP_WLOCK_ASSERT(inp);
1647
1648 imo = inp->inp_moptions;
1649 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1650
1651 INP_WUNLOCK(inp);
1652
1653 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1654 sizeof(struct __msfilterreq));
1655 if (error)
1656 return (error);
1657
1658 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1659 return (EINVAL);
1660
1661 ifp = ifnet_byindex(msfr.msfr_ifindex);
1662 if (ifp == NULL)
1663 return (EINVAL);
1664
1665 INP_WLOCK(inp);
1666
1667 /*
1668 * Lookup group on the socket.
1669 */
1670 gsa = (sockunion_t *)&msfr.msfr_group;
1671 idx = imo_match_group(imo, ifp, &gsa->sa);
1672 if (idx == -1 || imo->imo_mfilters == NULL) {
1673 INP_WUNLOCK(inp);
1674 return (EADDRNOTAVAIL);
1675 }
1676 imf = &imo->imo_mfilters[idx];
1677
1678 /*
1679 * Ignore memberships which are in limbo.
1680 */
1681 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1682 INP_WUNLOCK(inp);
1683 return (EAGAIN);
1684 }
1685 msfr.msfr_fmode = imf->imf_st[1];
1686
1687 /*
1688 * If the user specified a buffer, copy out the source filter
1689 * entries to userland gracefully.
1690 * We only copy out the number of entries which userland
1691 * has asked for, but we always tell userland how big the
1692 * buffer really needs to be.
1693 */
1694 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1695 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1696 tss = NULL;
1697 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1698 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1699 M_TEMP, M_NOWAIT | M_ZERO);
1700 if (tss == NULL) {
1701 INP_WUNLOCK(inp);
1702 return (ENOBUFS);
1703 }
1704 }
1705
1706 /*
1707 * Count number of sources in-mode at t0.
1708 * If buffer space exists and remains, copy out source entries.
1709 */
1710 nsrcs = msfr.msfr_nsrcs;
1711 ncsrcs = 0;
1712 ptss = tss;
1713 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1714 lims = (struct in_msource *)ims;
1715 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1716 lims->imsl_st[0] != imf->imf_st[0])
1717 continue;
1718 ++ncsrcs;
1719 if (tss != NULL && nsrcs > 0) {
1720 psin = (struct sockaddr_in *)ptss;
1721 psin->sin_family = AF_INET;
1722 psin->sin_len = sizeof(struct sockaddr_in);
1723 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1724 psin->sin_port = 0;
1725 ++ptss;
1726 --nsrcs;
1727 }
1728 }
1729
1730 INP_WUNLOCK(inp);
1731
1732 if (tss != NULL) {
1733 error = copyout(tss, msfr.msfr_srcs,
1734 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1735 free(tss, M_TEMP);
1736 if (error)
1737 return (error);
1738 }
1739
1740 msfr.msfr_nsrcs = ncsrcs;
1741 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1742
1743 return (error);
1744 }
1745
1746 /*
1747 * Return the IP multicast options in response to user getsockopt().
1748 */
1749 int
1750 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1751 {
1752 struct rm_priotracker in_ifa_tracker;
1753 struct ip_mreqn mreqn;
1754 struct ip_moptions *imo;
1755 struct ifnet *ifp;
1756 struct in_ifaddr *ia;
1757 int error, optval;
1758 u_char coptval;
1759
1760 INP_WLOCK(inp);
1761 imo = inp->inp_moptions;
1762 /*
1763 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1764 * or is a divert socket, reject it.
1765 */
1766 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1767 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1768 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1769 INP_WUNLOCK(inp);
1770 return (EOPNOTSUPP);
1771 }
1772
1773 error = 0;
1774 switch (sopt->sopt_name) {
1775 case IP_MULTICAST_VIF:
1776 if (imo != NULL)
1777 optval = imo->imo_multicast_vif;
1778 else
1779 optval = -1;
1780 INP_WUNLOCK(inp);
1781 error = sooptcopyout(sopt, &optval, sizeof(int));
1782 break;
1783
1784 case IP_MULTICAST_IF:
1785 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1786 if (imo != NULL) {
1787 ifp = imo->imo_multicast_ifp;
1788 if (!in_nullhost(imo->imo_multicast_addr)) {
1789 mreqn.imr_address = imo->imo_multicast_addr;
1790 } else if (ifp != NULL) {
1791 mreqn.imr_ifindex = ifp->if_index;
1792 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1793 if (ia != NULL) {
1794 mreqn.imr_address =
1795 IA_SIN(ia)->sin_addr;
1796 ifa_free(&ia->ia_ifa);
1797 }
1798 }
1799 }
1800 INP_WUNLOCK(inp);
1801 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1802 error = sooptcopyout(sopt, &mreqn,
1803 sizeof(struct ip_mreqn));
1804 } else {
1805 error = sooptcopyout(sopt, &mreqn.imr_address,
1806 sizeof(struct in_addr));
1807 }
1808 break;
1809
1810 case IP_MULTICAST_TTL:
1811 if (imo == NULL)
1812 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1813 else
1814 optval = coptval = imo->imo_multicast_ttl;
1815 INP_WUNLOCK(inp);
1816 if (sopt->sopt_valsize == sizeof(u_char))
1817 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1818 else
1819 error = sooptcopyout(sopt, &optval, sizeof(int));
1820 break;
1821
1822 case IP_MULTICAST_LOOP:
1823 if (imo == NULL)
1824 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1825 else
1826 optval = coptval = imo->imo_multicast_loop;
1827 INP_WUNLOCK(inp);
1828 if (sopt->sopt_valsize == sizeof(u_char))
1829 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1830 else
1831 error = sooptcopyout(sopt, &optval, sizeof(int));
1832 break;
1833
1834 case IP_MSFILTER:
1835 if (imo == NULL) {
1836 error = EADDRNOTAVAIL;
1837 INP_WUNLOCK(inp);
1838 } else {
1839 error = inp_get_source_filters(inp, sopt);
1840 }
1841 break;
1842
1843 default:
1844 INP_WUNLOCK(inp);
1845 error = ENOPROTOOPT;
1846 break;
1847 }
1848
1849 INP_UNLOCK_ASSERT(inp);
1850
1851 return (error);
1852 }
1853
1854 /*
1855 * Look up the ifnet to use for a multicast group membership,
1856 * given the IPv4 address of an interface, and the IPv4 group address.
1857 *
1858 * This routine exists to support legacy multicast applications
1859 * which do not understand that multicast memberships are scoped to
1860 * specific physical links in the networking stack, or which need
1861 * to join link-scope groups before IPv4 addresses are configured.
1862 *
1863 * If inp is non-NULL, use this socket's current FIB number for any
1864 * required FIB lookup.
1865 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1866 * and use its ifp; usually, this points to the default next-hop.
1867 *
1868 * If the FIB lookup fails, attempt to use the first non-loopback
1869 * interface with multicast capability in the system as a
1870 * last resort. The legacy IPv4 ASM API requires that we do
1871 * this in order to allow groups to be joined when the routing
1872 * table has not yet been populated during boot.
1873 *
1874 * Returns NULL if no ifp could be found.
1875 *
1876 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1877 * FUTURE: Implement IPv4 source-address selection.
1878 */
1879 static struct ifnet *
1880 inp_lookup_mcast_ifp(const struct inpcb *inp,
1881 const struct sockaddr_in *gsin, const struct in_addr ina)
1882 {
1883 struct rm_priotracker in_ifa_tracker;
1884 struct ifnet *ifp;
1885 struct nhop4_basic nh4;
1886 uint32_t fibnum;
1887
1888 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1889 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1890 ("%s: not multicast", __func__));
1891
1892 ifp = NULL;
1893 if (!in_nullhost(ina)) {
1894 INADDR_TO_IFP(ina, ifp);
1895 } else {
1896 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1897 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1898 ifp = nh4.nh_ifp;
1899 else {
1900 struct in_ifaddr *ia;
1901 struct ifnet *mifp;
1902
1903 mifp = NULL;
1904 IN_IFADDR_RLOCK(&in_ifa_tracker);
1905 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1906 mifp = ia->ia_ifp;
1907 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1908 (mifp->if_flags & IFF_MULTICAST)) {
1909 ifp = mifp;
1910 break;
1911 }
1912 }
1913 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
1914 }
1915 }
1916
1917 return (ifp);
1918 }
1919
1920 /*
1921 * Join an IPv4 multicast group, possibly with a source.
1922 */
1923 static int
1924 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1925 {
1926 struct group_source_req gsr;
1927 sockunion_t *gsa, *ssa;
1928 struct ifnet *ifp;
1929 struct in_mfilter *imf;
1930 struct ip_moptions *imo;
1931 struct in_multi *inm;
1932 struct in_msource *lims;
1933 size_t idx;
1934 int error, is_new;
1935
1936 ifp = NULL;
1937 imf = NULL;
1938 lims = NULL;
1939 error = 0;
1940 is_new = 0;
1941
1942 memset(&gsr, 0, sizeof(struct group_source_req));
1943 gsa = (sockunion_t *)&gsr.gsr_group;
1944 gsa->ss.ss_family = AF_UNSPEC;
1945 ssa = (sockunion_t *)&gsr.gsr_source;
1946 ssa->ss.ss_family = AF_UNSPEC;
1947
1948 switch (sopt->sopt_name) {
1949 case IP_ADD_MEMBERSHIP:
1950 case IP_ADD_SOURCE_MEMBERSHIP: {
1951 struct ip_mreq_source mreqs;
1952
1953 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1954 error = sooptcopyin(sopt, &mreqs,
1955 sizeof(struct ip_mreq),
1956 sizeof(struct ip_mreq));
1957 /*
1958 * Do argument switcharoo from ip_mreq into
1959 * ip_mreq_source to avoid using two instances.
1960 */
1961 mreqs.imr_interface = mreqs.imr_sourceaddr;
1962 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1963 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1964 error = sooptcopyin(sopt, &mreqs,
1965 sizeof(struct ip_mreq_source),
1966 sizeof(struct ip_mreq_source));
1967 }
1968 if (error)
1969 return (error);
1970
1971 gsa->sin.sin_family = AF_INET;
1972 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1973 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1974
1975 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1976 ssa->sin.sin_family = AF_INET;
1977 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1978 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1979 }
1980
1981 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1982 return (EINVAL);
1983
1984 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1985 mreqs.imr_interface);
1986 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1987 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1988 break;
1989 }
1990
1991 case MCAST_JOIN_GROUP:
1992 case MCAST_JOIN_SOURCE_GROUP:
1993 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1994 error = sooptcopyin(sopt, &gsr,
1995 sizeof(struct group_req),
1996 sizeof(struct group_req));
1997 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1998 error = sooptcopyin(sopt, &gsr,
1999 sizeof(struct group_source_req),
2000 sizeof(struct group_source_req));
2001 }
2002 if (error)
2003 return (error);
2004
2005 if (gsa->sin.sin_family != AF_INET ||
2006 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2007 return (EINVAL);
2008
2009 /*
2010 * Overwrite the port field if present, as the sockaddr
2011 * being copied in may be matched with a binary comparison.
2012 */
2013 gsa->sin.sin_port = 0;
2014 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2015 if (ssa->sin.sin_family != AF_INET ||
2016 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2017 return (EINVAL);
2018 ssa->sin.sin_port = 0;
2019 }
2020
2021 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2022 return (EINVAL);
2023
2024 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2025 return (EADDRNOTAVAIL);
2026 ifp = ifnet_byindex(gsr.gsr_interface);
2027 break;
2028
2029 default:
2030 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2031 __func__, sopt->sopt_name);
2032 return (EOPNOTSUPP);
2033 break;
2034 }
2035
2036 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2037 return (EADDRNOTAVAIL);
2038
2039 imo = inp_findmoptions(inp);
2040 idx = imo_match_group(imo, ifp, &gsa->sa);
2041 if (idx == -1) {
2042 is_new = 1;
2043 } else {
2044 inm = imo->imo_membership[idx];
2045 imf = &imo->imo_mfilters[idx];
2046 if (ssa->ss.ss_family != AF_UNSPEC) {
2047 /*
2048 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2049 * is an error. On an existing inclusive membership,
2050 * it just adds the source to the filter list.
2051 */
2052 if (imf->imf_st[1] != MCAST_INCLUDE) {
2053 error = EINVAL;
2054 goto out_inp_locked;
2055 }
2056 /*
2057 * Throw out duplicates.
2058 *
2059 * XXX FIXME: This makes a naive assumption that
2060 * even if entries exist for *ssa in this imf,
2061 * they will be rejected as dupes, even if they
2062 * are not valid in the current mode (in-mode).
2063 *
2064 * in_msource is transactioned just as for anything
2065 * else in SSM -- but note naive use of inm_graft()
2066 * below for allocating new filter entries.
2067 *
2068 * This is only an issue if someone mixes the
2069 * full-state SSM API with the delta-based API,
2070 * which is discouraged in the relevant RFCs.
2071 */
2072 lims = imo_match_source(imo, idx, &ssa->sa);
2073 if (lims != NULL /*&&
2074 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2075 error = EADDRNOTAVAIL;
2076 goto out_inp_locked;
2077 }
2078 } else {
2079 /*
2080 * MCAST_JOIN_GROUP on an existing exclusive
2081 * membership is an error; return EADDRINUSE
2082 * to preserve 4.4BSD API idempotence, and
2083 * avoid tedious detour to code below.
2084 * NOTE: This is bending RFC 3678 a bit.
2085 *
2086 * On an existing inclusive membership, this is also
2087 * an error; if you want to change filter mode,
2088 * you must use the userland API setsourcefilter().
2089 * XXX We don't reject this for imf in UNDEFINED
2090 * state at t1, because allocation of a filter
2091 * is atomic with allocation of a membership.
2092 */
2093 error = EINVAL;
2094 if (imf->imf_st[1] == MCAST_EXCLUDE)
2095 error = EADDRINUSE;
2096 goto out_inp_locked;
2097 }
2098 }
2099
2100 /*
2101 * Begin state merge transaction at socket layer.
2102 */
2103 INP_WLOCK_ASSERT(inp);
2104
2105 if (is_new) {
2106 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2107 error = imo_grow(imo);
2108 if (error)
2109 goto out_inp_locked;
2110 }
2111 /*
2112 * Allocate the new slot upfront so we can deal with
2113 * grafting the new source filter in same code path
2114 * as for join-source on existing membership.
2115 */
2116 idx = imo->imo_num_memberships;
2117 imo->imo_membership[idx] = NULL;
2118 imo->imo_num_memberships++;
2119 KASSERT(imo->imo_mfilters != NULL,
2120 ("%s: imf_mfilters vector was not allocated", __func__));
2121 imf = &imo->imo_mfilters[idx];
2122 KASSERT(RB_EMPTY(&imf->imf_sources),
2123 ("%s: imf_sources not empty", __func__));
2124 }
2125
2126 /*
2127 * Graft new source into filter list for this inpcb's
2128 * membership of the group. The in_multi may not have
2129 * been allocated yet if this is a new membership, however,
2130 * the in_mfilter slot will be allocated and must be initialized.
2131 *
2132 * Note: Grafting of exclusive mode filters doesn't happen
2133 * in this path.
2134 * XXX: Should check for non-NULL lims (node exists but may
2135 * not be in-mode) for interop with full-state API.
2136 */
2137 if (ssa->ss.ss_family != AF_UNSPEC) {
2138 /* Membership starts in IN mode */
2139 if (is_new) {
2140 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2141 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2142 } else {
2143 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2144 }
2145 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2146 if (lims == NULL) {
2147 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2148 __func__);
2149 error = ENOMEM;
2150 goto out_imo_free;
2151 }
2152 } else {
2153 /* No address specified; Membership starts in EX mode */
2154 if (is_new) {
2155 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2156 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2157 }
2158 }
2159
2160 /*
2161 * Begin state merge transaction at IGMP layer.
2162 */
2163 IN_MULTI_LOCK();
2164
2165 if (is_new) {
2166 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2167 &inm);
2168 if (error) {
2169 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2170 __func__);
2171 IN_MULTI_UNLOCK();
2172 goto out_imo_free;
2173 }
2174 imo->imo_membership[idx] = inm;
2175 } else {
2176 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2177 error = inm_merge(inm, imf);
2178 if (error) {
2179 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2180 __func__);
2181 goto out_in_multi_locked;
2182 }
2183 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2184 error = igmp_change_state(inm);
2185 if (error) {
2186 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2187 __func__);
2188 goto out_in_multi_locked;
2189 }
2190 }
2191
2192 out_in_multi_locked:
2193
2194 IN_MULTI_UNLOCK();
2195
2196 INP_WLOCK_ASSERT(inp);
2197 if (error) {
2198 imf_rollback(imf);
2199 if (is_new)
2200 imf_purge(imf);
2201 else
2202 imf_reap(imf);
2203 } else {
2204 imf_commit(imf);
2205 }
2206
2207 out_imo_free:
2208 if (error && is_new) {
2209 imo->imo_membership[idx] = NULL;
2210 --imo->imo_num_memberships;
2211 }
2212
2213 out_inp_locked:
2214 INP_WUNLOCK(inp);
2215 return (error);
2216 }
2217
2218 /*
2219 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2220 */
2221 static int
2222 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2223 {
2224 struct group_source_req gsr;
2225 struct ip_mreq_source mreqs;
2226 sockunion_t *gsa, *ssa;
2227 struct ifnet *ifp;
2228 struct in_mfilter *imf;
2229 struct ip_moptions *imo;
2230 struct in_msource *ims;
2231 struct in_multi *inm;
2232 size_t idx;
2233 int error, is_final;
2234
2235 ifp = NULL;
2236 error = 0;
2237 is_final = 1;
2238
2239 memset(&gsr, 0, sizeof(struct group_source_req));
2240 gsa = (sockunion_t *)&gsr.gsr_group;
2241 gsa->ss.ss_family = AF_UNSPEC;
2242 ssa = (sockunion_t *)&gsr.gsr_source;
2243 ssa->ss.ss_family = AF_UNSPEC;
2244
2245 switch (sopt->sopt_name) {
2246 case IP_DROP_MEMBERSHIP:
2247 case IP_DROP_SOURCE_MEMBERSHIP:
2248 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2249 error = sooptcopyin(sopt, &mreqs,
2250 sizeof(struct ip_mreq),
2251 sizeof(struct ip_mreq));
2252 /*
2253 * Swap interface and sourceaddr arguments,
2254 * as ip_mreq and ip_mreq_source are laid
2255 * out differently.
2256 */
2257 mreqs.imr_interface = mreqs.imr_sourceaddr;
2258 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2259 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2260 error = sooptcopyin(sopt, &mreqs,
2261 sizeof(struct ip_mreq_source),
2262 sizeof(struct ip_mreq_source));
2263 }
2264 if (error)
2265 return (error);
2266
2267 gsa->sin.sin_family = AF_INET;
2268 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2269 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2270
2271 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2272 ssa->sin.sin_family = AF_INET;
2273 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2274 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2275 }
2276
2277 /*
2278 * Attempt to look up hinted ifp from interface address.
2279 * Fallthrough with null ifp iff lookup fails, to
2280 * preserve 4.4BSD mcast API idempotence.
2281 * XXX NOTE WELL: The RFC 3678 API is preferred because
2282 * using an IPv4 address as a key is racy.
2283 */
2284 if (!in_nullhost(mreqs.imr_interface))
2285 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2286
2287 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2288 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2289
2290 break;
2291
2292 case MCAST_LEAVE_GROUP:
2293 case MCAST_LEAVE_SOURCE_GROUP:
2294 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2295 error = sooptcopyin(sopt, &gsr,
2296 sizeof(struct group_req),
2297 sizeof(struct group_req));
2298 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2299 error = sooptcopyin(sopt, &gsr,
2300 sizeof(struct group_source_req),
2301 sizeof(struct group_source_req));
2302 }
2303 if (error)
2304 return (error);
2305
2306 if (gsa->sin.sin_family != AF_INET ||
2307 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2308 return (EINVAL);
2309
2310 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2311 if (ssa->sin.sin_family != AF_INET ||
2312 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2313 return (EINVAL);
2314 }
2315
2316 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2317 return (EADDRNOTAVAIL);
2318
2319 ifp = ifnet_byindex(gsr.gsr_interface);
2320
2321 if (ifp == NULL)
2322 return (EADDRNOTAVAIL);
2323 break;
2324
2325 default:
2326 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2327 __func__, sopt->sopt_name);
2328 return (EOPNOTSUPP);
2329 break;
2330 }
2331
2332 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2333 return (EINVAL);
2334
2335 /*
2336 * Find the membership in the membership array.
2337 */
2338 imo = inp_findmoptions(inp);
2339 idx = imo_match_group(imo, ifp, &gsa->sa);
2340 if (idx == -1) {
2341 error = EADDRNOTAVAIL;
2342 goto out_inp_locked;
2343 }
2344 inm = imo->imo_membership[idx];
2345 imf = &imo->imo_mfilters[idx];
2346
2347 if (ssa->ss.ss_family != AF_UNSPEC)
2348 is_final = 0;
2349
2350 /*
2351 * Begin state merge transaction at socket layer.
2352 */
2353 INP_WLOCK_ASSERT(inp);
2354
2355 /*
2356 * If we were instructed only to leave a given source, do so.
2357 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2358 */
2359 if (is_final) {
2360 imf_leave(imf);
2361 } else {
2362 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2363 error = EADDRNOTAVAIL;
2364 goto out_inp_locked;
2365 }
2366 ims = imo_match_source(imo, idx, &ssa->sa);
2367 if (ims == NULL) {
2368 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2369 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2370 error = EADDRNOTAVAIL;
2371 goto out_inp_locked;
2372 }
2373 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2374 error = imf_prune(imf, &ssa->sin);
2375 if (error) {
2376 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2377 __func__);
2378 goto out_inp_locked;
2379 }
2380 }
2381
2382 /*
2383 * Begin state merge transaction at IGMP layer.
2384 */
2385 IN_MULTI_LOCK();
2386
2387 if (is_final) {
2388 /*
2389 * Give up the multicast address record to which
2390 * the membership points.
2391 */
2392 (void)in_leavegroup_locked(inm, imf);
2393 } else {
2394 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2395 error = inm_merge(inm, imf);
2396 if (error) {
2397 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2398 __func__);
2399 goto out_in_multi_locked;
2400 }
2401
2402 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2403 error = igmp_change_state(inm);
2404 if (error) {
2405 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2406 __func__);
2407 }
2408 }
2409
2410 out_in_multi_locked:
2411
2412 IN_MULTI_UNLOCK();
2413
2414 if (error)
2415 imf_rollback(imf);
2416 else
2417 imf_commit(imf);
2418
2419 imf_reap(imf);
2420
2421 if (is_final) {
2422 /* Remove the gap in the membership and filter array. */
2423 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2424 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2425 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2426 }
2427 imo->imo_num_memberships--;
2428 }
2429
2430 out_inp_locked:
2431 INP_WUNLOCK(inp);
2432 return (error);
2433 }
2434
2435 /*
2436 * Select the interface for transmitting IPv4 multicast datagrams.
2437 *
2438 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2439 * may be passed to this socket option. An address of INADDR_ANY or an
2440 * interface index of 0 is used to remove a previous selection.
2441 * When no interface is selected, one is chosen for every send.
2442 */
2443 static int
2444 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2445 {
2446 struct in_addr addr;
2447 struct ip_mreqn mreqn;
2448 struct ifnet *ifp;
2449 struct ip_moptions *imo;
2450 int error;
2451
2452 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2453 /*
2454 * An interface index was specified using the
2455 * Linux-derived ip_mreqn structure.
2456 */
2457 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2458 sizeof(struct ip_mreqn));
2459 if (error)
2460 return (error);
2461
2462 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2463 return (EINVAL);
2464
2465 if (mreqn.imr_ifindex == 0) {
2466 ifp = NULL;
2467 } else {
2468 ifp = ifnet_byindex(mreqn.imr_ifindex);
2469 if (ifp == NULL)
2470 return (EADDRNOTAVAIL);
2471 }
2472 } else {
2473 /*
2474 * An interface was specified by IPv4 address.
2475 * This is the traditional BSD usage.
2476 */
2477 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2478 sizeof(struct in_addr));
2479 if (error)
2480 return (error);
2481 if (in_nullhost(addr)) {
2482 ifp = NULL;
2483 } else {
2484 INADDR_TO_IFP(addr, ifp);
2485 if (ifp == NULL)
2486 return (EADDRNOTAVAIL);
2487 }
2488 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2489 ntohl(addr.s_addr));
2490 }
2491
2492 /* Reject interfaces which do not support multicast. */
2493 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2494 return (EOPNOTSUPP);
2495
2496 imo = inp_findmoptions(inp);
2497 imo->imo_multicast_ifp = ifp;
2498 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2499 INP_WUNLOCK(inp);
2500
2501 return (0);
2502 }
2503
2504 /*
2505 * Atomically set source filters on a socket for an IPv4 multicast group.
2506 *
2507 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2508 */
2509 static int
2510 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2511 {
2512 struct __msfilterreq msfr;
2513 sockunion_t *gsa;
2514 struct ifnet *ifp;
2515 struct in_mfilter *imf;
2516 struct ip_moptions *imo;
2517 struct in_multi *inm;
2518 size_t idx;
2519 int error;
2520
2521 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2522 sizeof(struct __msfilterreq));
2523 if (error)
2524 return (error);
2525
2526 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2527 return (ENOBUFS);
2528
2529 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2530 msfr.msfr_fmode != MCAST_INCLUDE))
2531 return (EINVAL);
2532
2533 if (msfr.msfr_group.ss_family != AF_INET ||
2534 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2535 return (EINVAL);
2536
2537 gsa = (sockunion_t *)&msfr.msfr_group;
2538 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2539 return (EINVAL);
2540
2541 gsa->sin.sin_port = 0; /* ignore port */
2542
2543 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2544 return (EADDRNOTAVAIL);
2545
2546 ifp = ifnet_byindex(msfr.msfr_ifindex);
2547 if (ifp == NULL)
2548 return (EADDRNOTAVAIL);
2549
2550 /*
2551 * Take the INP write lock.
2552 * Check if this socket is a member of this group.
2553 */
2554 imo = inp_findmoptions(inp);
2555 idx = imo_match_group(imo, ifp, &gsa->sa);
2556 if (idx == -1 || imo->imo_mfilters == NULL) {
2557 error = EADDRNOTAVAIL;
2558 goto out_inp_locked;
2559 }
2560 inm = imo->imo_membership[idx];
2561 imf = &imo->imo_mfilters[idx];
2562
2563 /*
2564 * Begin state merge transaction at socket layer.
2565 */
2566 INP_WLOCK_ASSERT(inp);
2567
2568 imf->imf_st[1] = msfr.msfr_fmode;
2569
2570 /*
2571 * Apply any new source filters, if present.
2572 * Make a copy of the user-space source vector so
2573 * that we may copy them with a single copyin. This
2574 * allows us to deal with page faults up-front.
2575 */
2576 if (msfr.msfr_nsrcs > 0) {
2577 struct in_msource *lims;
2578 struct sockaddr_in *psin;
2579 struct sockaddr_storage *kss, *pkss;
2580 int i;
2581
2582 INP_WUNLOCK(inp);
2583
2584 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2585 __func__, (unsigned long)msfr.msfr_nsrcs);
2586 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2587 M_TEMP, M_WAITOK);
2588 error = copyin(msfr.msfr_srcs, kss,
2589 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2590 if (error) {
2591 free(kss, M_TEMP);
2592 return (error);
2593 }
2594
2595 INP_WLOCK(inp);
2596
2597 /*
2598 * Mark all source filters as UNDEFINED at t1.
2599 * Restore new group filter mode, as imf_leave()
2600 * will set it to INCLUDE.
2601 */
2602 imf_leave(imf);
2603 imf->imf_st[1] = msfr.msfr_fmode;
2604
2605 /*
2606 * Update socket layer filters at t1, lazy-allocating
2607 * new entries. This saves a bunch of memory at the
2608 * cost of one RB_FIND() per source entry; duplicate
2609 * entries in the msfr_nsrcs vector are ignored.
2610 * If we encounter an error, rollback transaction.
2611 *
2612 * XXX This too could be replaced with a set-symmetric
2613 * difference like loop to avoid walking from root
2614 * every time, as the key space is common.
2615 */
2616 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2617 psin = (struct sockaddr_in *)pkss;
2618 if (psin->sin_family != AF_INET) {
2619 error = EAFNOSUPPORT;
2620 break;
2621 }
2622 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2623 error = EINVAL;
2624 break;
2625 }
2626 error = imf_get_source(imf, psin, &lims);
2627 if (error)
2628 break;
2629 lims->imsl_st[1] = imf->imf_st[1];
2630 }
2631 free(kss, M_TEMP);
2632 }
2633
2634 if (error)
2635 goto out_imf_rollback;
2636
2637 INP_WLOCK_ASSERT(inp);
2638 IN_MULTI_LOCK();
2639
2640 /*
2641 * Begin state merge transaction at IGMP layer.
2642 */
2643 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2644 error = inm_merge(inm, imf);
2645 if (error) {
2646 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2647 goto out_in_multi_locked;
2648 }
2649
2650 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2651 error = igmp_change_state(inm);
2652 if (error)
2653 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2654
2655 out_in_multi_locked:
2656
2657 IN_MULTI_UNLOCK();
2658
2659 out_imf_rollback:
2660 if (error)
2661 imf_rollback(imf);
2662 else
2663 imf_commit(imf);
2664
2665 imf_reap(imf);
2666
2667 out_inp_locked:
2668 INP_WUNLOCK(inp);
2669 return (error);
2670 }
2671
2672 /*
2673 * Set the IP multicast options in response to user setsockopt().
2674 *
2675 * Many of the socket options handled in this function duplicate the
2676 * functionality of socket options in the regular unicast API. However,
2677 * it is not possible to merge the duplicate code, because the idempotence
2678 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2679 * the effects of these options must be treated as separate and distinct.
2680 *
2681 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2682 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2683 * is refactored to no longer use vifs.
2684 */
2685 int
2686 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2687 {
2688 struct ip_moptions *imo;
2689 int error;
2690
2691 error = 0;
2692
2693 /*
2694 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2695 * or is a divert socket, reject it.
2696 */
2697 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2698 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2699 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2700 return (EOPNOTSUPP);
2701
2702 switch (sopt->sopt_name) {
2703 case IP_MULTICAST_VIF: {
2704 int vifi;
2705 /*
2706 * Select a multicast VIF for transmission.
2707 * Only useful if multicast forwarding is active.
2708 */
2709 if (legal_vif_num == NULL) {
2710 error = EOPNOTSUPP;
2711 break;
2712 }
2713 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2714 if (error)
2715 break;
2716 if (!legal_vif_num(vifi) && (vifi != -1)) {
2717 error = EINVAL;
2718 break;
2719 }
2720 imo = inp_findmoptions(inp);
2721 imo->imo_multicast_vif = vifi;
2722 INP_WUNLOCK(inp);
2723 break;
2724 }
2725
2726 case IP_MULTICAST_IF:
2727 error = inp_set_multicast_if(inp, sopt);
2728 break;
2729
2730 case IP_MULTICAST_TTL: {
2731 u_char ttl;
2732
2733 /*
2734 * Set the IP time-to-live for outgoing multicast packets.
2735 * The original multicast API required a char argument,
2736 * which is inconsistent with the rest of the socket API.
2737 * We allow either a char or an int.
2738 */
2739 if (sopt->sopt_valsize == sizeof(u_char)) {
2740 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2741 sizeof(u_char));
2742 if (error)
2743 break;
2744 } else {
2745 u_int ittl;
2746
2747 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2748 sizeof(u_int));
2749 if (error)
2750 break;
2751 if (ittl > 255) {
2752 error = EINVAL;
2753 break;
2754 }
2755 ttl = (u_char)ittl;
2756 }
2757 imo = inp_findmoptions(inp);
2758 imo->imo_multicast_ttl = ttl;
2759 INP_WUNLOCK(inp);
2760 break;
2761 }
2762
2763 case IP_MULTICAST_LOOP: {
2764 u_char loop;
2765
2766 /*
2767 * Set the loopback flag for outgoing multicast packets.
2768 * Must be zero or one. The original multicast API required a
2769 * char argument, which is inconsistent with the rest
2770 * of the socket API. We allow either a char or an int.
2771 */
2772 if (sopt->sopt_valsize == sizeof(u_char)) {
2773 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2774 sizeof(u_char));
2775 if (error)
2776 break;
2777 } else {
2778 u_int iloop;
2779
2780 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2781 sizeof(u_int));
2782 if (error)
2783 break;
2784 loop = (u_char)iloop;
2785 }
2786 imo = inp_findmoptions(inp);
2787 imo->imo_multicast_loop = !!loop;
2788 INP_WUNLOCK(inp);
2789 break;
2790 }
2791
2792 case IP_ADD_MEMBERSHIP:
2793 case IP_ADD_SOURCE_MEMBERSHIP:
2794 case MCAST_JOIN_GROUP:
2795 case MCAST_JOIN_SOURCE_GROUP:
2796 error = inp_join_group(inp, sopt);
2797 break;
2798
2799 case IP_DROP_MEMBERSHIP:
2800 case IP_DROP_SOURCE_MEMBERSHIP:
2801 case MCAST_LEAVE_GROUP:
2802 case MCAST_LEAVE_SOURCE_GROUP:
2803 error = inp_leave_group(inp, sopt);
2804 break;
2805
2806 case IP_BLOCK_SOURCE:
2807 case IP_UNBLOCK_SOURCE:
2808 case MCAST_BLOCK_SOURCE:
2809 case MCAST_UNBLOCK_SOURCE:
2810 error = inp_block_unblock_source(inp, sopt);
2811 break;
2812
2813 case IP_MSFILTER:
2814 error = inp_set_source_filters(inp, sopt);
2815 break;
2816
2817 default:
2818 error = EOPNOTSUPP;
2819 break;
2820 }
2821
2822 INP_UNLOCK_ASSERT(inp);
2823
2824 return (error);
2825 }
2826
2827 /*
2828 * Expose IGMP's multicast filter mode and source list(s) to userland,
2829 * keyed by (ifindex, group).
2830 * The filter mode is written out as a uint32_t, followed by
2831 * 0..n of struct in_addr.
2832 * For use by ifmcstat(8).
2833 * SMPng: NOTE: unlocked read of ifindex space.
2834 */
2835 static int
2836 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2837 {
2838 struct in_addr src, group;
2839 struct ifnet *ifp;
2840 struct ifmultiaddr *ifma;
2841 struct in_multi *inm;
2842 struct ip_msource *ims;
2843 int *name;
2844 int retval;
2845 u_int namelen;
2846 uint32_t fmode, ifindex;
2847
2848 name = (int *)arg1;
2849 namelen = arg2;
2850
2851 if (req->newptr != NULL)
2852 return (EPERM);
2853
2854 if (namelen != 2)
2855 return (EINVAL);
2856
2857 ifindex = name[0];
2858 if (ifindex <= 0 || ifindex > V_if_index) {
2859 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2860 __func__, ifindex);
2861 return (ENOENT);
2862 }
2863
2864 group.s_addr = name[1];
2865 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2866 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2867 __func__, ntohl(group.s_addr));
2868 return (EINVAL);
2869 }
2870
2871 ifp = ifnet_byindex(ifindex);
2872 if (ifp == NULL) {
2873 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2874 __func__, ifindex);
2875 return (ENOENT);
2876 }
2877
2878 retval = sysctl_wire_old_buffer(req,
2879 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2880 if (retval)
2881 return (retval);
2882
2883 IN_MULTI_LOCK();
2884
2885 IF_ADDR_RLOCK(ifp);
2886 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2887 if (ifma->ifma_addr->sa_family != AF_INET ||
2888 ifma->ifma_protospec == NULL)
2889 continue;
2890 inm = (struct in_multi *)ifma->ifma_protospec;
2891 if (!in_hosteq(inm->inm_addr, group))
2892 continue;
2893 fmode = inm->inm_st[1].iss_fmode;
2894 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2895 if (retval != 0)
2896 break;
2897 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2898 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
2899 ims->ims_haddr);
2900 /*
2901 * Only copy-out sources which are in-mode.
2902 */
2903 if (fmode != ims_get_mode(inm, ims, 1)) {
2904 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2905 __func__);
2906 continue;
2907 }
2908 src.s_addr = htonl(ims->ims_haddr);
2909 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2910 if (retval != 0)
2911 break;
2912 }
2913 }
2914 IF_ADDR_RUNLOCK(ifp);
2915
2916 IN_MULTI_UNLOCK();
2917
2918 return (retval);
2919 }
2920
2921 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
2922
2923 static const char *inm_modestrs[] = { "un", "in", "ex" };
2924
2925 static const char *
2926 inm_mode_str(const int mode)
2927 {
2928
2929 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2930 return (inm_modestrs[mode]);
2931 return ("??");
2932 }
2933
2934 static const char *inm_statestrs[] = {
2935 "not-member",
2936 "silent",
2937 "idle",
2938 "lazy",
2939 "sleeping",
2940 "awakening",
2941 "query-pending",
2942 "sg-query-pending",
2943 "leaving"
2944 };
2945
2946 static const char *
2947 inm_state_str(const int state)
2948 {
2949
2950 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2951 return (inm_statestrs[state]);
2952 return ("??");
2953 }
2954
2955 /*
2956 * Dump an in_multi structure to the console.
2957 */
2958 void
2959 inm_print(const struct in_multi *inm)
2960 {
2961 int t;
2962 char addrbuf[INET_ADDRSTRLEN];
2963
2964 if ((ktr_mask & KTR_IGMPV3) == 0)
2965 return;
2966
2967 printf("%s: --- begin inm %p ---\n", __func__, inm);
2968 printf("addr %s ifp %p(%s) ifma %p\n",
2969 inet_ntoa_r(inm->inm_addr, addrbuf),
2970 inm->inm_ifp,
2971 inm->inm_ifp->if_xname,
2972 inm->inm_ifma);
2973 printf("timer %u state %s refcount %u scq.len %u\n",
2974 inm->inm_timer,
2975 inm_state_str(inm->inm_state),
2976 inm->inm_refcount,
2977 inm->inm_scq.mq_len);
2978 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2979 inm->inm_igi,
2980 inm->inm_nsrc,
2981 inm->inm_sctimer,
2982 inm->inm_scrv);
2983 for (t = 0; t < 2; t++) {
2984 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2985 inm_mode_str(inm->inm_st[t].iss_fmode),
2986 inm->inm_st[t].iss_asm,
2987 inm->inm_st[t].iss_ex,
2988 inm->inm_st[t].iss_in,
2989 inm->inm_st[t].iss_rec);
2990 }
2991 printf("%s: --- end inm %p ---\n", __func__, inm);
2992 }
2993
2994 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
2995
2996 void
2997 inm_print(const struct in_multi *inm)
2998 {
2999
3000 }
3001
3002 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3003
3004 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);
Cache object: 0536d08d752ad2649ddb871e0777d776
|