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/10.4/sys/netinet/in_mcast.c 321134 2017-07-18 16:58:52Z ngie $");
37
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/malloc.h>
42 #include <sys/mbuf.h>
43 #include <sys/protosw.h>
44 #include <sys/socket.h>
45 #include <sys/socketvar.h>
46 #include <sys/protosw.h>
47 #include <sys/sysctl.h>
48 #include <sys/ktr.h>
49 #include <sys/taskqueue.h>
50 #include <sys/tree.h>
51
52 #include <net/if.h>
53 #include <net/if_dl.h>
54 #include <net/route.h>
55 #include <net/vnet.h>
56
57 #include <netinet/in.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/in_pcb.h>
60 #include <netinet/in_var.h>
61 #include <netinet/ip_var.h>
62 #include <netinet/igmp_var.h>
63
64 #ifndef KTR_IGMPV3
65 #define KTR_IGMPV3 KTR_INET
66 #endif
67
68 #ifndef __SOCKUNION_DECLARED
69 union sockunion {
70 struct sockaddr_storage ss;
71 struct sockaddr sa;
72 struct sockaddr_dl sdl;
73 struct sockaddr_in sin;
74 };
75 typedef union sockunion sockunion_t;
76 #define __SOCKUNION_DECLARED
77 #endif /* __SOCKUNION_DECLARED */
78
79 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
80 "IPv4 multicast PCB-layer source filter");
81 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
82 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
83 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
84 "IPv4 multicast IGMP-layer source filter");
85
86 /*
87 * Locking:
88 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
89 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
90 * it can be taken by code in net/if.c also.
91 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
92 *
93 * struct in_multi is covered by IN_MULTI_LOCK. There isn't strictly
94 * any need for in_multi itself to be virtualized -- it is bound to an ifp
95 * anyway no matter what happens.
96 */
97 struct mtx in_multi_mtx;
98 MTX_SYSINIT(in_multi_mtx, &in_multi_mtx, "in_multi_mtx", MTX_DEF);
99
100 /*
101 * Functions with non-static linkage defined in this file should be
102 * declared in in_var.h:
103 * imo_multi_filter()
104 * in_addmulti()
105 * in_delmulti()
106 * in_joingroup()
107 * in_joingroup_locked()
108 * in_leavegroup()
109 * in_leavegroup_locked()
110 * and ip_var.h:
111 * inp_freemoptions()
112 * inp_getmoptions()
113 * inp_setmoptions()
114 *
115 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
116 * and in_delmulti().
117 */
118 static void imf_commit(struct in_mfilter *);
119 static int imf_get_source(struct in_mfilter *imf,
120 const struct sockaddr_in *psin,
121 struct in_msource **);
122 static struct in_msource *
123 imf_graft(struct in_mfilter *, const uint8_t,
124 const struct sockaddr_in *);
125 static void imf_leave(struct in_mfilter *);
126 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
127 static void imf_purge(struct in_mfilter *);
128 static void imf_rollback(struct in_mfilter *);
129 static void imf_reap(struct in_mfilter *);
130 static int imo_grow(struct ip_moptions *);
131 static size_t imo_match_group(const struct ip_moptions *,
132 const struct ifnet *, const struct sockaddr *);
133 static struct in_msource *
134 imo_match_source(const struct ip_moptions *, const size_t,
135 const struct sockaddr *);
136 static void ims_merge(struct ip_msource *ims,
137 const struct in_msource *lims, const int rollback);
138 static int in_getmulti(struct ifnet *, const struct in_addr *,
139 struct in_multi **);
140 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
141 const int noalloc, struct ip_msource **pims);
142 #ifdef KTR
143 static int inm_is_ifp_detached(const struct in_multi *);
144 #endif
145 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
146 static void inm_purge(struct in_multi *);
147 static void inm_reap(struct in_multi *);
148 static struct ip_moptions *
149 inp_findmoptions(struct inpcb *);
150 static void inp_freemoptions_internal(struct ip_moptions *);
151 static void inp_gcmoptions(void *, int);
152 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
153 static int inp_join_group(struct inpcb *, struct sockopt *);
154 static int inp_leave_group(struct inpcb *, struct sockopt *);
155 static struct ifnet *
156 inp_lookup_mcast_ifp(const struct inpcb *,
157 const struct sockaddr_in *, const struct in_addr);
158 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
159 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
160 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
161 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
162
163 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
164 "IPv4 multicast");
165
166 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
167 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
168 CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxgrpsrc, 0,
169 "Max source filters per group");
170 TUNABLE_ULONG("net.inet.ip.mcast.maxgrpsrc", &in_mcast_maxgrpsrc);
171
172 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
173 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
174 CTLFLAG_RW | CTLFLAG_TUN, &in_mcast_maxsocksrc, 0,
175 "Max source filters per socket");
176 TUNABLE_ULONG("net.inet.ip.mcast.maxsocksrc", &in_mcast_maxsocksrc);
177
178 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
179 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
180 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
181 TUNABLE_INT("net.inet.ip.mcast.loop", &in_mcast_loop);
182
183 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
184 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
185 "Per-interface stack-wide source filters");
186
187 static STAILQ_HEAD(, ip_moptions) imo_gc_list =
188 STAILQ_HEAD_INITIALIZER(imo_gc_list);
189 static struct task imo_gc_task = TASK_INITIALIZER(0, inp_gcmoptions, NULL);
190
191 #ifdef KTR
192 /*
193 * Inline function which wraps assertions for a valid ifp.
194 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
195 * is detached.
196 */
197 static int __inline
198 inm_is_ifp_detached(const struct in_multi *inm)
199 {
200 struct ifnet *ifp;
201
202 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
203 ifp = inm->inm_ifma->ifma_ifp;
204 if (ifp != NULL) {
205 /*
206 * Sanity check that netinet's notion of ifp is the
207 * same as net's.
208 */
209 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
210 }
211
212 return (ifp == NULL);
213 }
214 #endif
215
216 /*
217 * Initialize an in_mfilter structure to a known state at t0, t1
218 * with an empty source filter list.
219 */
220 static __inline void
221 imf_init(struct in_mfilter *imf, const int st0, const int st1)
222 {
223 memset(imf, 0, sizeof(struct in_mfilter));
224 RB_INIT(&imf->imf_sources);
225 imf->imf_st[0] = st0;
226 imf->imf_st[1] = st1;
227 }
228
229 /*
230 * Resize the ip_moptions vector to the next power-of-two minus 1.
231 * May be called with locks held; do not sleep.
232 */
233 static int
234 imo_grow(struct ip_moptions *imo)
235 {
236 struct in_multi **nmships;
237 struct in_multi **omships;
238 struct in_mfilter *nmfilters;
239 struct in_mfilter *omfilters;
240 size_t idx;
241 size_t newmax;
242 size_t oldmax;
243
244 nmships = NULL;
245 nmfilters = NULL;
246 omships = imo->imo_membership;
247 omfilters = imo->imo_mfilters;
248 oldmax = imo->imo_max_memberships;
249 newmax = ((oldmax + 1) * 2) - 1;
250
251 if (newmax <= IP_MAX_MEMBERSHIPS) {
252 nmships = (struct in_multi **)realloc(omships,
253 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
254 nmfilters = (struct in_mfilter *)realloc(omfilters,
255 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
256 if (nmships != NULL && nmfilters != NULL) {
257 /* Initialize newly allocated source filter heads. */
258 for (idx = oldmax; idx < newmax; idx++) {
259 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
260 MCAST_EXCLUDE);
261 }
262 imo->imo_max_memberships = newmax;
263 imo->imo_membership = nmships;
264 imo->imo_mfilters = nmfilters;
265 }
266 }
267
268 if (nmships == NULL || nmfilters == NULL) {
269 if (nmships != NULL)
270 free(nmships, M_IPMOPTS);
271 if (nmfilters != NULL)
272 free(nmfilters, M_INMFILTER);
273 return (ETOOMANYREFS);
274 }
275
276 return (0);
277 }
278
279 /*
280 * Find an IPv4 multicast group entry for this ip_moptions instance
281 * which matches the specified group, and optionally an interface.
282 * Return its index into the array, or -1 if not found.
283 */
284 static size_t
285 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
286 const struct sockaddr *group)
287 {
288 const struct sockaddr_in *gsin;
289 struct in_multi **pinm;
290 int idx;
291 int nmships;
292
293 gsin = (const struct sockaddr_in *)group;
294
295 /* The imo_membership array may be lazy allocated. */
296 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
297 return (-1);
298
299 nmships = imo->imo_num_memberships;
300 pinm = &imo->imo_membership[0];
301 for (idx = 0; idx < nmships; idx++, pinm++) {
302 if (*pinm == NULL)
303 continue;
304 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
305 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
306 break;
307 }
308 }
309 if (idx >= nmships)
310 idx = -1;
311
312 return (idx);
313 }
314
315 /*
316 * Find an IPv4 multicast source entry for this imo which matches
317 * the given group index for this socket, and source address.
318 *
319 * NOTE: This does not check if the entry is in-mode, merely if
320 * it exists, which may not be the desired behaviour.
321 */
322 static struct in_msource *
323 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
324 const struct sockaddr *src)
325 {
326 struct ip_msource find;
327 struct in_mfilter *imf;
328 struct ip_msource *ims;
329 const sockunion_t *psa;
330
331 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
332 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
333 ("%s: invalid index %d\n", __func__, (int)gidx));
334
335 /* The imo_mfilters array may be lazy allocated. */
336 if (imo->imo_mfilters == NULL)
337 return (NULL);
338 imf = &imo->imo_mfilters[gidx];
339
340 /* Source trees are keyed in host byte order. */
341 psa = (const sockunion_t *)src;
342 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
343 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
344
345 return ((struct in_msource *)ims);
346 }
347
348 /*
349 * Perform filtering for multicast datagrams on a socket by group and source.
350 *
351 * Returns 0 if a datagram should be allowed through, or various error codes
352 * if the socket was not a member of the group, or the source was muted, etc.
353 */
354 int
355 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
356 const struct sockaddr *group, const struct sockaddr *src)
357 {
358 size_t gidx;
359 struct in_msource *ims;
360 int mode;
361
362 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
363
364 gidx = imo_match_group(imo, ifp, group);
365 if (gidx == -1)
366 return (MCAST_NOTGMEMBER);
367
368 /*
369 * Check if the source was included in an (S,G) join.
370 * Allow reception on exclusive memberships by default,
371 * reject reception on inclusive memberships by default.
372 * Exclude source only if an in-mode exclude filter exists.
373 * Include source only if an in-mode include filter exists.
374 * NOTE: We are comparing group state here at IGMP t1 (now)
375 * with socket-layer t0 (since last downcall).
376 */
377 mode = imo->imo_mfilters[gidx].imf_st[1];
378 ims = imo_match_source(imo, gidx, src);
379
380 if ((ims == NULL && mode == MCAST_INCLUDE) ||
381 (ims != NULL && ims->imsl_st[0] != mode))
382 return (MCAST_NOTSMEMBER);
383
384 return (MCAST_PASS);
385 }
386
387 /*
388 * Find and return a reference to an in_multi record for (ifp, group),
389 * and bump its reference count.
390 * If one does not exist, try to allocate it, and update link-layer multicast
391 * filters on ifp to listen for group.
392 * Assumes the IN_MULTI lock is held across the call.
393 * Return 0 if successful, otherwise return an appropriate error code.
394 */
395 static int
396 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
397 struct in_multi **pinm)
398 {
399 struct sockaddr_in gsin;
400 struct ifmultiaddr *ifma;
401 struct in_ifinfo *ii;
402 struct in_multi *inm;
403 int error;
404
405 IN_MULTI_LOCK_ASSERT();
406
407 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
408
409 inm = inm_lookup(ifp, *group);
410 if (inm != NULL) {
411 /*
412 * If we already joined this group, just bump the
413 * refcount and return it.
414 */
415 KASSERT(inm->inm_refcount >= 1,
416 ("%s: bad refcount %d", __func__, inm->inm_refcount));
417 ++inm->inm_refcount;
418 *pinm = inm;
419 return (0);
420 }
421
422 memset(&gsin, 0, sizeof(gsin));
423 gsin.sin_family = AF_INET;
424 gsin.sin_len = sizeof(struct sockaddr_in);
425 gsin.sin_addr = *group;
426
427 /*
428 * Check if a link-layer group is already associated
429 * with this network-layer group on the given ifnet.
430 */
431 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
432 if (error != 0)
433 return (error);
434
435 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
436 IF_ADDR_WLOCK(ifp);
437
438 /*
439 * If something other than netinet is occupying the link-layer
440 * group, print a meaningful error message and back out of
441 * the allocation.
442 * Otherwise, bump the refcount on the existing network-layer
443 * group association and return it.
444 */
445 if (ifma->ifma_protospec != NULL) {
446 inm = (struct in_multi *)ifma->ifma_protospec;
447 #ifdef INVARIANTS
448 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
449 __func__));
450 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
451 ("%s: ifma not AF_INET", __func__));
452 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
453 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
454 !in_hosteq(inm->inm_addr, *group))
455 panic("%s: ifma %p is inconsistent with %p (%s)",
456 __func__, ifma, inm, inet_ntoa(*group));
457 #endif
458 ++inm->inm_refcount;
459 *pinm = inm;
460 IF_ADDR_WUNLOCK(ifp);
461 return (0);
462 }
463
464 IF_ADDR_WLOCK_ASSERT(ifp);
465
466 /*
467 * A new in_multi record is needed; allocate and initialize it.
468 * We DO NOT perform an IGMP join as the in_ layer may need to
469 * push an initial source list down to IGMP to support SSM.
470 *
471 * The initial source filter state is INCLUDE, {} as per the RFC.
472 */
473 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
474 if (inm == NULL) {
475 if_delmulti_ifma(ifma);
476 IF_ADDR_WUNLOCK(ifp);
477 return (ENOMEM);
478 }
479 inm->inm_addr = *group;
480 inm->inm_ifp = ifp;
481 inm->inm_igi = ii->ii_igmp;
482 inm->inm_ifma = ifma;
483 inm->inm_refcount = 1;
484 inm->inm_state = IGMP_NOT_MEMBER;
485
486 /*
487 * Pending state-changes per group are subject to a bounds check.
488 */
489 IFQ_SET_MAXLEN(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
490
491 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
492 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
493 RB_INIT(&inm->inm_srcs);
494
495 ifma->ifma_protospec = inm;
496
497 *pinm = inm;
498
499 IF_ADDR_WUNLOCK(ifp);
500 return (0);
501 }
502
503 /*
504 * Drop a reference to an in_multi record.
505 *
506 * If the refcount drops to 0, free the in_multi record and
507 * delete the underlying link-layer membership.
508 */
509 void
510 inm_release_locked(struct in_multi *inm)
511 {
512 struct ifmultiaddr *ifma;
513
514 IN_MULTI_LOCK_ASSERT();
515
516 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
517
518 if (--inm->inm_refcount > 0) {
519 CTR2(KTR_IGMPV3, "%s: refcount is now %d", __func__,
520 inm->inm_refcount);
521 return;
522 }
523
524 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
525
526 ifma = inm->inm_ifma;
527
528 /* XXX this access is not covered by IF_ADDR_LOCK */
529 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
530 KASSERT(ifma->ifma_protospec == inm,
531 ("%s: ifma_protospec != inm", __func__));
532 ifma->ifma_protospec = NULL;
533
534 inm_purge(inm);
535
536 free(inm, M_IPMADDR);
537
538 if_delmulti_ifma(ifma);
539 }
540
541 /*
542 * Clear recorded source entries for a group.
543 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
544 * FIXME: Should reap.
545 */
546 void
547 inm_clear_recorded(struct in_multi *inm)
548 {
549 struct ip_msource *ims;
550
551 IN_MULTI_LOCK_ASSERT();
552
553 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
554 if (ims->ims_stp) {
555 ims->ims_stp = 0;
556 --inm->inm_st[1].iss_rec;
557 }
558 }
559 KASSERT(inm->inm_st[1].iss_rec == 0,
560 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
561 }
562
563 /*
564 * Record a source as pending for a Source-Group IGMPv3 query.
565 * This lives here as it modifies the shared tree.
566 *
567 * inm is the group descriptor.
568 * naddr is the address of the source to record in network-byte order.
569 *
570 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
571 * lazy-allocate a source node in response to an SG query.
572 * Otherwise, no allocation is performed. This saves some memory
573 * with the trade-off that the source will not be reported to the
574 * router if joined in the window between the query response and
575 * the group actually being joined on the local host.
576 *
577 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
578 * This turns off the allocation of a recorded source entry if
579 * the group has not been joined.
580 *
581 * Return 0 if the source didn't exist or was already marked as recorded.
582 * Return 1 if the source was marked as recorded by this function.
583 * Return <0 if any error occured (negated errno code).
584 */
585 int
586 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
587 {
588 struct ip_msource find;
589 struct ip_msource *ims, *nims;
590
591 IN_MULTI_LOCK_ASSERT();
592
593 find.ims_haddr = ntohl(naddr);
594 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
595 if (ims && ims->ims_stp)
596 return (0);
597 if (ims == NULL) {
598 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
599 return (-ENOSPC);
600 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
601 M_NOWAIT | M_ZERO);
602 if (nims == NULL)
603 return (-ENOMEM);
604 nims->ims_haddr = find.ims_haddr;
605 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
606 ++inm->inm_nsrc;
607 ims = nims;
608 }
609
610 /*
611 * Mark the source as recorded and update the recorded
612 * source count.
613 */
614 ++ims->ims_stp;
615 ++inm->inm_st[1].iss_rec;
616
617 return (1);
618 }
619
620 /*
621 * Return a pointer to an in_msource owned by an in_mfilter,
622 * given its source address.
623 * Lazy-allocate if needed. If this is a new entry its filter state is
624 * undefined at t0.
625 *
626 * imf is the filter set being modified.
627 * haddr is the source address in *host* byte-order.
628 *
629 * SMPng: May be called with locks held; malloc must not block.
630 */
631 static int
632 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
633 struct in_msource **plims)
634 {
635 struct ip_msource find;
636 struct ip_msource *ims, *nims;
637 struct in_msource *lims;
638 int error;
639
640 error = 0;
641 ims = NULL;
642 lims = NULL;
643
644 /* key is host byte order */
645 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
646 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
647 lims = (struct in_msource *)ims;
648 if (lims == NULL) {
649 if (imf->imf_nsrc == in_mcast_maxsocksrc)
650 return (ENOSPC);
651 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
652 M_NOWAIT | M_ZERO);
653 if (nims == NULL)
654 return (ENOMEM);
655 lims = (struct in_msource *)nims;
656 lims->ims_haddr = find.ims_haddr;
657 lims->imsl_st[0] = MCAST_UNDEFINED;
658 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
659 ++imf->imf_nsrc;
660 }
661
662 *plims = lims;
663
664 return (error);
665 }
666
667 /*
668 * Graft a source entry into an existing socket-layer filter set,
669 * maintaining any required invariants and checking allocations.
670 *
671 * The source is marked as being in the new filter mode at t1.
672 *
673 * Return the pointer to the new node, otherwise return NULL.
674 */
675 static struct in_msource *
676 imf_graft(struct in_mfilter *imf, const uint8_t st1,
677 const struct sockaddr_in *psin)
678 {
679 struct ip_msource *nims;
680 struct in_msource *lims;
681
682 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
683 M_NOWAIT | M_ZERO);
684 if (nims == NULL)
685 return (NULL);
686 lims = (struct in_msource *)nims;
687 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
688 lims->imsl_st[0] = MCAST_UNDEFINED;
689 lims->imsl_st[1] = st1;
690 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
691 ++imf->imf_nsrc;
692
693 return (lims);
694 }
695
696 /*
697 * Prune a source entry from an existing socket-layer filter set,
698 * maintaining any required invariants and checking allocations.
699 *
700 * The source is marked as being left at t1, it is not freed.
701 *
702 * Return 0 if no error occurred, otherwise return an errno value.
703 */
704 static int
705 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
706 {
707 struct ip_msource find;
708 struct ip_msource *ims;
709 struct in_msource *lims;
710
711 /* key is host byte order */
712 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
713 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
714 if (ims == NULL)
715 return (ENOENT);
716 lims = (struct in_msource *)ims;
717 lims->imsl_st[1] = MCAST_UNDEFINED;
718 return (0);
719 }
720
721 /*
722 * Revert socket-layer filter set deltas at t1 to t0 state.
723 */
724 static void
725 imf_rollback(struct in_mfilter *imf)
726 {
727 struct ip_msource *ims, *tims;
728 struct in_msource *lims;
729
730 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
731 lims = (struct in_msource *)ims;
732 if (lims->imsl_st[0] == lims->imsl_st[1]) {
733 /* no change at t1 */
734 continue;
735 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
736 /* revert change to existing source at t1 */
737 lims->imsl_st[1] = lims->imsl_st[0];
738 } else {
739 /* revert source added t1 */
740 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
741 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
742 free(ims, M_INMFILTER);
743 imf->imf_nsrc--;
744 }
745 }
746 imf->imf_st[1] = imf->imf_st[0];
747 }
748
749 /*
750 * Mark socket-layer filter set as INCLUDE {} at t1.
751 */
752 static void
753 imf_leave(struct in_mfilter *imf)
754 {
755 struct ip_msource *ims;
756 struct in_msource *lims;
757
758 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
759 lims = (struct in_msource *)ims;
760 lims->imsl_st[1] = MCAST_UNDEFINED;
761 }
762 imf->imf_st[1] = MCAST_INCLUDE;
763 }
764
765 /*
766 * Mark socket-layer filter set deltas as committed.
767 */
768 static void
769 imf_commit(struct in_mfilter *imf)
770 {
771 struct ip_msource *ims;
772 struct in_msource *lims;
773
774 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
775 lims = (struct in_msource *)ims;
776 lims->imsl_st[0] = lims->imsl_st[1];
777 }
778 imf->imf_st[0] = imf->imf_st[1];
779 }
780
781 /*
782 * Reap unreferenced sources from socket-layer filter set.
783 */
784 static void
785 imf_reap(struct in_mfilter *imf)
786 {
787 struct ip_msource *ims, *tims;
788 struct in_msource *lims;
789
790 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
791 lims = (struct in_msource *)ims;
792 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
793 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
794 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
795 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
796 free(ims, M_INMFILTER);
797 imf->imf_nsrc--;
798 }
799 }
800 }
801
802 /*
803 * Purge socket-layer filter set.
804 */
805 static void
806 imf_purge(struct in_mfilter *imf)
807 {
808 struct ip_msource *ims, *tims;
809
810 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
811 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
812 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
813 free(ims, M_INMFILTER);
814 imf->imf_nsrc--;
815 }
816 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
817 KASSERT(RB_EMPTY(&imf->imf_sources),
818 ("%s: imf_sources not empty", __func__));
819 }
820
821 /*
822 * Look up a source filter entry for a multicast group.
823 *
824 * inm is the group descriptor to work with.
825 * haddr is the host-byte-order IPv4 address to look up.
826 * noalloc may be non-zero to suppress allocation of sources.
827 * *pims will be set to the address of the retrieved or allocated source.
828 *
829 * SMPng: NOTE: may be called with locks held.
830 * Return 0 if successful, otherwise return a non-zero error code.
831 */
832 static int
833 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
834 const int noalloc, struct ip_msource **pims)
835 {
836 struct ip_msource find;
837 struct ip_msource *ims, *nims;
838 #ifdef KTR
839 struct in_addr ia;
840 #endif
841
842 find.ims_haddr = haddr;
843 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
844 if (ims == NULL && !noalloc) {
845 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
846 return (ENOSPC);
847 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
848 M_NOWAIT | M_ZERO);
849 if (nims == NULL)
850 return (ENOMEM);
851 nims->ims_haddr = haddr;
852 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
853 ++inm->inm_nsrc;
854 ims = nims;
855 #ifdef KTR
856 ia.s_addr = htonl(haddr);
857 CTR3(KTR_IGMPV3, "%s: allocated %s as %p", __func__,
858 inet_ntoa(ia), ims);
859 #endif
860 }
861
862 *pims = ims;
863 return (0);
864 }
865
866 /*
867 * Merge socket-layer source into IGMP-layer source.
868 * If rollback is non-zero, perform the inverse of the merge.
869 */
870 static void
871 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
872 const int rollback)
873 {
874 int n = rollback ? -1 : 1;
875 #ifdef KTR
876 struct in_addr ia;
877
878 ia.s_addr = htonl(ims->ims_haddr);
879 #endif
880
881 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
882 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on %s",
883 __func__, n, inet_ntoa(ia));
884 ims->ims_st[1].ex -= n;
885 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
886 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on %s",
887 __func__, n, inet_ntoa(ia));
888 ims->ims_st[1].in -= n;
889 }
890
891 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
892 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on %s",
893 __func__, n, inet_ntoa(ia));
894 ims->ims_st[1].ex += n;
895 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
896 CTR3(KTR_IGMPV3, "%s: t1 in += %d on %s",
897 __func__, n, inet_ntoa(ia));
898 ims->ims_st[1].in += n;
899 }
900 }
901
902 /*
903 * Atomically update the global in_multi state, when a membership's
904 * filter list is being updated in any way.
905 *
906 * imf is the per-inpcb-membership group filter pointer.
907 * A fake imf may be passed for in-kernel consumers.
908 *
909 * XXX This is a candidate for a set-symmetric-difference style loop
910 * which would eliminate the repeated lookup from root of ims nodes,
911 * as they share the same key space.
912 *
913 * If any error occurred this function will back out of refcounts
914 * and return a non-zero value.
915 */
916 static int
917 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
918 {
919 struct ip_msource *ims, *nims;
920 struct in_msource *lims;
921 int schanged, error;
922 int nsrc0, nsrc1;
923
924 schanged = 0;
925 error = 0;
926 nsrc1 = nsrc0 = 0;
927
928 /*
929 * Update the source filters first, as this may fail.
930 * Maintain count of in-mode filters at t0, t1. These are
931 * used to work out if we transition into ASM mode or not.
932 * Maintain a count of source filters whose state was
933 * actually modified by this operation.
934 */
935 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
936 lims = (struct in_msource *)ims;
937 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
938 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
939 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
940 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
941 ++schanged;
942 if (error)
943 break;
944 ims_merge(nims, lims, 0);
945 }
946 if (error) {
947 struct ip_msource *bims;
948
949 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
950 lims = (struct in_msource *)ims;
951 if (lims->imsl_st[0] == lims->imsl_st[1])
952 continue;
953 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
954 if (bims == NULL)
955 continue;
956 ims_merge(bims, lims, 1);
957 }
958 goto out_reap;
959 }
960
961 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
962 __func__, nsrc0, nsrc1);
963
964 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
965 if (imf->imf_st[0] == imf->imf_st[1] &&
966 imf->imf_st[1] == MCAST_INCLUDE) {
967 if (nsrc1 == 0) {
968 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
969 --inm->inm_st[1].iss_in;
970 }
971 }
972
973 /* Handle filter mode transition on socket. */
974 if (imf->imf_st[0] != imf->imf_st[1]) {
975 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
976 __func__, imf->imf_st[0], imf->imf_st[1]);
977
978 if (imf->imf_st[0] == MCAST_EXCLUDE) {
979 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
980 --inm->inm_st[1].iss_ex;
981 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
982 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
983 --inm->inm_st[1].iss_in;
984 }
985
986 if (imf->imf_st[1] == MCAST_EXCLUDE) {
987 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
988 inm->inm_st[1].iss_ex++;
989 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
990 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
991 inm->inm_st[1].iss_in++;
992 }
993 }
994
995 /*
996 * Track inm filter state in terms of listener counts.
997 * If there are any exclusive listeners, stack-wide
998 * membership is exclusive.
999 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1000 * If no listeners remain, state is undefined at t1,
1001 * and the IGMP lifecycle for this group should finish.
1002 */
1003 if (inm->inm_st[1].iss_ex > 0) {
1004 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1005 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1006 } else if (inm->inm_st[1].iss_in > 0) {
1007 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1008 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1009 } else {
1010 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1011 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1012 }
1013
1014 /* Decrement ASM listener count on transition out of ASM mode. */
1015 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1016 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1017 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1018 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1019 --inm->inm_st[1].iss_asm;
1020 }
1021 }
1022
1023 /* Increment ASM listener count on transition to ASM mode. */
1024 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1025 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1026 inm->inm_st[1].iss_asm++;
1027 }
1028
1029 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1030 inm_print(inm);
1031
1032 out_reap:
1033 if (schanged > 0) {
1034 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1035 inm_reap(inm);
1036 }
1037 return (error);
1038 }
1039
1040 /*
1041 * Mark an in_multi's filter set deltas as committed.
1042 * Called by IGMP after a state change has been enqueued.
1043 */
1044 void
1045 inm_commit(struct in_multi *inm)
1046 {
1047 struct ip_msource *ims;
1048
1049 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1050 CTR1(KTR_IGMPV3, "%s: pre commit:", __func__);
1051 inm_print(inm);
1052
1053 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1054 ims->ims_st[0] = ims->ims_st[1];
1055 }
1056 inm->inm_st[0] = inm->inm_st[1];
1057 }
1058
1059 /*
1060 * Reap unreferenced nodes from an in_multi's filter set.
1061 */
1062 static void
1063 inm_reap(struct in_multi *inm)
1064 {
1065 struct ip_msource *ims, *tims;
1066
1067 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1068 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1069 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1070 ims->ims_stp != 0)
1071 continue;
1072 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1073 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1074 free(ims, M_IPMSOURCE);
1075 inm->inm_nsrc--;
1076 }
1077 }
1078
1079 /*
1080 * Purge all source nodes from an in_multi's filter set.
1081 */
1082 static void
1083 inm_purge(struct in_multi *inm)
1084 {
1085 struct ip_msource *ims, *tims;
1086
1087 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1088 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1089 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1090 free(ims, M_IPMSOURCE);
1091 inm->inm_nsrc--;
1092 }
1093 }
1094
1095 /*
1096 * Join a multicast group; unlocked entry point.
1097 *
1098 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1099 * is not held. Fortunately, ifp is unlikely to have been detached
1100 * at this point, so we assume it's OK to recurse.
1101 */
1102 int
1103 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1104 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1105 {
1106 int error;
1107
1108 IN_MULTI_LOCK();
1109 error = in_joingroup_locked(ifp, gina, imf, pinm);
1110 IN_MULTI_UNLOCK();
1111
1112 return (error);
1113 }
1114
1115 /*
1116 * Join a multicast group; real entry point.
1117 *
1118 * Only preserves atomicity at inm level.
1119 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1120 *
1121 * If the IGMP downcall fails, the group is not joined, and an error
1122 * code is returned.
1123 */
1124 int
1125 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1126 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1127 {
1128 struct in_mfilter timf;
1129 struct in_multi *inm;
1130 int error;
1131
1132 IN_MULTI_LOCK_ASSERT();
1133
1134 CTR4(KTR_IGMPV3, "%s: join %s on %p(%s))", __func__,
1135 inet_ntoa(*gina), ifp, ifp->if_xname);
1136
1137 error = 0;
1138 inm = NULL;
1139
1140 /*
1141 * If no imf was specified (i.e. kernel consumer),
1142 * fake one up and assume it is an ASM join.
1143 */
1144 if (imf == NULL) {
1145 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1146 imf = &timf;
1147 }
1148
1149 error = in_getmulti(ifp, gina, &inm);
1150 if (error) {
1151 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1152 return (error);
1153 }
1154
1155 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1156 error = inm_merge(inm, imf);
1157 if (error) {
1158 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1159 goto out_inm_release;
1160 }
1161
1162 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1163 error = igmp_change_state(inm);
1164 if (error) {
1165 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1166 goto out_inm_release;
1167 }
1168
1169 out_inm_release:
1170 if (error) {
1171 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1172 inm_release_locked(inm);
1173 } else {
1174 *pinm = inm;
1175 }
1176
1177 return (error);
1178 }
1179
1180 /*
1181 * Leave a multicast group; unlocked entry point.
1182 */
1183 int
1184 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1185 {
1186 int error;
1187
1188 IN_MULTI_LOCK();
1189 error = in_leavegroup_locked(inm, imf);
1190 IN_MULTI_UNLOCK();
1191
1192 return (error);
1193 }
1194
1195 /*
1196 * Leave a multicast group; real entry point.
1197 * All source filters will be expunged.
1198 *
1199 * Only preserves atomicity at inm level.
1200 *
1201 * Holding the write lock for the INP which contains imf
1202 * is highly advisable. We can't assert for it as imf does not
1203 * contain a back-pointer to the owning inp.
1204 *
1205 * Note: This is not the same as inm_release(*) as this function also
1206 * makes a state change downcall into IGMP.
1207 */
1208 int
1209 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1210 {
1211 struct in_mfilter timf;
1212 int error;
1213
1214 error = 0;
1215
1216 IN_MULTI_LOCK_ASSERT();
1217
1218 CTR5(KTR_IGMPV3, "%s: leave inm %p, %s/%s, imf %p", __func__,
1219 inm, inet_ntoa(inm->inm_addr),
1220 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1221 imf);
1222
1223 /*
1224 * If no imf was specified (i.e. kernel consumer),
1225 * fake one up and assume it is an ASM join.
1226 */
1227 if (imf == NULL) {
1228 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1229 imf = &timf;
1230 }
1231
1232 /*
1233 * Begin state merge transaction at IGMP layer.
1234 *
1235 * As this particular invocation should not cause any memory
1236 * to be allocated, and there is no opportunity to roll back
1237 * the transaction, it MUST NOT fail.
1238 */
1239 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1240 error = inm_merge(inm, imf);
1241 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1242
1243 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1244 CURVNET_SET(inm->inm_ifp->if_vnet);
1245 error = igmp_change_state(inm);
1246 CURVNET_RESTORE();
1247 if (error)
1248 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1249
1250 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1251 inm_release_locked(inm);
1252
1253 return (error);
1254 }
1255
1256 /*#ifndef BURN_BRIDGES*/
1257 /*
1258 * Join an IPv4 multicast group in (*,G) exclusive mode.
1259 * The group must be a 224.0.0.0/24 link-scope group.
1260 * This KPI is for legacy kernel consumers only.
1261 */
1262 struct in_multi *
1263 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1264 {
1265 struct in_multi *pinm;
1266 int error;
1267
1268 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1269 ("%s: %s not in 224.0.0.0/24", __func__, inet_ntoa(*ap)));
1270
1271 error = in_joingroup(ifp, ap, NULL, &pinm);
1272 if (error != 0)
1273 pinm = NULL;
1274
1275 return (pinm);
1276 }
1277
1278 /*
1279 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1280 * This KPI is for legacy kernel consumers only.
1281 */
1282 void
1283 in_delmulti(struct in_multi *inm)
1284 {
1285
1286 (void)in_leavegroup(inm, NULL);
1287 }
1288 /*#endif*/
1289
1290 /*
1291 * Block or unblock an ASM multicast source on an inpcb.
1292 * This implements the delta-based API described in RFC 3678.
1293 *
1294 * The delta-based API applies only to exclusive-mode memberships.
1295 * An IGMP downcall will be performed.
1296 *
1297 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1298 *
1299 * Return 0 if successful, otherwise return an appropriate error code.
1300 */
1301 static int
1302 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1303 {
1304 struct group_source_req gsr;
1305 sockunion_t *gsa, *ssa;
1306 struct ifnet *ifp;
1307 struct in_mfilter *imf;
1308 struct ip_moptions *imo;
1309 struct in_msource *ims;
1310 struct in_multi *inm;
1311 size_t idx;
1312 uint16_t fmode;
1313 int error, doblock;
1314
1315 ifp = NULL;
1316 error = 0;
1317 doblock = 0;
1318
1319 memset(&gsr, 0, sizeof(struct group_source_req));
1320 gsa = (sockunion_t *)&gsr.gsr_group;
1321 ssa = (sockunion_t *)&gsr.gsr_source;
1322
1323 switch (sopt->sopt_name) {
1324 case IP_BLOCK_SOURCE:
1325 case IP_UNBLOCK_SOURCE: {
1326 struct ip_mreq_source mreqs;
1327
1328 error = sooptcopyin(sopt, &mreqs,
1329 sizeof(struct ip_mreq_source),
1330 sizeof(struct ip_mreq_source));
1331 if (error)
1332 return (error);
1333
1334 gsa->sin.sin_family = AF_INET;
1335 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1336 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1337
1338 ssa->sin.sin_family = AF_INET;
1339 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1340 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1341
1342 if (!in_nullhost(mreqs.imr_interface))
1343 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1344
1345 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1346 doblock = 1;
1347
1348 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1349 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1350 break;
1351 }
1352
1353 case MCAST_BLOCK_SOURCE:
1354 case MCAST_UNBLOCK_SOURCE:
1355 error = sooptcopyin(sopt, &gsr,
1356 sizeof(struct group_source_req),
1357 sizeof(struct group_source_req));
1358 if (error)
1359 return (error);
1360
1361 if (gsa->sin.sin_family != AF_INET ||
1362 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1363 return (EINVAL);
1364
1365 if (ssa->sin.sin_family != AF_INET ||
1366 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1367 return (EINVAL);
1368
1369 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1370 return (EADDRNOTAVAIL);
1371
1372 ifp = ifnet_byindex(gsr.gsr_interface);
1373
1374 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1375 doblock = 1;
1376 break;
1377
1378 default:
1379 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1380 __func__, sopt->sopt_name);
1381 return (EOPNOTSUPP);
1382 break;
1383 }
1384
1385 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1386 return (EINVAL);
1387
1388 /*
1389 * Check if we are actually a member of this group.
1390 */
1391 imo = inp_findmoptions(inp);
1392 idx = imo_match_group(imo, ifp, &gsa->sa);
1393 if (idx == -1 || imo->imo_mfilters == NULL) {
1394 error = EADDRNOTAVAIL;
1395 goto out_inp_locked;
1396 }
1397
1398 KASSERT(imo->imo_mfilters != NULL,
1399 ("%s: imo_mfilters not allocated", __func__));
1400 imf = &imo->imo_mfilters[idx];
1401 inm = imo->imo_membership[idx];
1402
1403 /*
1404 * Attempting to use the delta-based API on an
1405 * non exclusive-mode membership is an error.
1406 */
1407 fmode = imf->imf_st[0];
1408 if (fmode != MCAST_EXCLUDE) {
1409 error = EINVAL;
1410 goto out_inp_locked;
1411 }
1412
1413 /*
1414 * Deal with error cases up-front:
1415 * Asked to block, but already blocked; or
1416 * Asked to unblock, but nothing to unblock.
1417 * If adding a new block entry, allocate it.
1418 */
1419 ims = imo_match_source(imo, idx, &ssa->sa);
1420 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1421 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
1422 inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not ");
1423 error = EADDRNOTAVAIL;
1424 goto out_inp_locked;
1425 }
1426
1427 INP_WLOCK_ASSERT(inp);
1428
1429 /*
1430 * Begin state merge transaction at socket layer.
1431 */
1432 if (doblock) {
1433 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1434 ims = imf_graft(imf, fmode, &ssa->sin);
1435 if (ims == NULL)
1436 error = ENOMEM;
1437 } else {
1438 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1439 error = imf_prune(imf, &ssa->sin);
1440 }
1441
1442 if (error) {
1443 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1444 goto out_imf_rollback;
1445 }
1446
1447 /*
1448 * Begin state merge transaction at IGMP layer.
1449 */
1450 IN_MULTI_LOCK();
1451
1452 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1453 error = inm_merge(inm, imf);
1454 if (error) {
1455 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1456 goto out_in_multi_locked;
1457 }
1458
1459 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1460 error = igmp_change_state(inm);
1461 if (error)
1462 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1463
1464 out_in_multi_locked:
1465
1466 IN_MULTI_UNLOCK();
1467
1468 out_imf_rollback:
1469 if (error)
1470 imf_rollback(imf);
1471 else
1472 imf_commit(imf);
1473
1474 imf_reap(imf);
1475
1476 out_inp_locked:
1477 INP_WUNLOCK(inp);
1478 return (error);
1479 }
1480
1481 /*
1482 * Given an inpcb, return its multicast options structure pointer. Accepts
1483 * an unlocked inpcb pointer, but will return it locked. May sleep.
1484 *
1485 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1486 * SMPng: NOTE: Returns with the INP write lock held.
1487 */
1488 static struct ip_moptions *
1489 inp_findmoptions(struct inpcb *inp)
1490 {
1491 struct ip_moptions *imo;
1492 struct in_multi **immp;
1493 struct in_mfilter *imfp;
1494 size_t idx;
1495
1496 INP_WLOCK(inp);
1497 if (inp->inp_moptions != NULL)
1498 return (inp->inp_moptions);
1499
1500 INP_WUNLOCK(inp);
1501
1502 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1503 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1504 M_WAITOK | M_ZERO);
1505 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1506 M_INMFILTER, M_WAITOK);
1507
1508 imo->imo_multicast_ifp = NULL;
1509 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1510 imo->imo_multicast_vif = -1;
1511 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1512 imo->imo_multicast_loop = in_mcast_loop;
1513 imo->imo_num_memberships = 0;
1514 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1515 imo->imo_membership = immp;
1516
1517 /* Initialize per-group source filters. */
1518 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1519 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1520 imo->imo_mfilters = imfp;
1521
1522 INP_WLOCK(inp);
1523 if (inp->inp_moptions != NULL) {
1524 free(imfp, M_INMFILTER);
1525 free(immp, M_IPMOPTS);
1526 free(imo, M_IPMOPTS);
1527 return (inp->inp_moptions);
1528 }
1529 inp->inp_moptions = imo;
1530 return (imo);
1531 }
1532
1533 /*
1534 * Discard the IP multicast options (and source filters). To minimize
1535 * the amount of work done while holding locks such as the INP's
1536 * pcbinfo lock (which is used in the receive path), the free
1537 * operation is performed asynchronously in a separate task.
1538 *
1539 * SMPng: NOTE: assumes INP write lock is held.
1540 */
1541 void
1542 inp_freemoptions(struct ip_moptions *imo)
1543 {
1544
1545 KASSERT(imo != NULL, ("%s: ip_moptions is NULL", __func__));
1546 IN_MULTI_LOCK();
1547 STAILQ_INSERT_TAIL(&imo_gc_list, imo, imo_link);
1548 IN_MULTI_UNLOCK();
1549 taskqueue_enqueue(taskqueue_thread, &imo_gc_task);
1550 }
1551
1552 static void
1553 inp_freemoptions_internal(struct ip_moptions *imo)
1554 {
1555 struct in_mfilter *imf;
1556 size_t idx, nmships;
1557
1558 nmships = imo->imo_num_memberships;
1559 for (idx = 0; idx < nmships; ++idx) {
1560 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1561 if (imf)
1562 imf_leave(imf);
1563 (void)in_leavegroup(imo->imo_membership[idx], imf);
1564 if (imf)
1565 imf_purge(imf);
1566 }
1567
1568 if (imo->imo_mfilters)
1569 free(imo->imo_mfilters, M_INMFILTER);
1570 free(imo->imo_membership, M_IPMOPTS);
1571 free(imo, M_IPMOPTS);
1572 }
1573
1574 static void
1575 inp_gcmoptions(void *context, int pending)
1576 {
1577 struct ip_moptions *imo;
1578
1579 IN_MULTI_LOCK();
1580 while (!STAILQ_EMPTY(&imo_gc_list)) {
1581 imo = STAILQ_FIRST(&imo_gc_list);
1582 STAILQ_REMOVE_HEAD(&imo_gc_list, imo_link);
1583 IN_MULTI_UNLOCK();
1584 inp_freemoptions_internal(imo);
1585 IN_MULTI_LOCK();
1586 }
1587 IN_MULTI_UNLOCK();
1588 }
1589
1590 /*
1591 * Atomically get source filters on a socket for an IPv4 multicast group.
1592 * Called with INP lock held; returns with lock released.
1593 */
1594 static int
1595 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1596 {
1597 struct __msfilterreq msfr;
1598 sockunion_t *gsa;
1599 struct ifnet *ifp;
1600 struct ip_moptions *imo;
1601 struct in_mfilter *imf;
1602 struct ip_msource *ims;
1603 struct in_msource *lims;
1604 struct sockaddr_in *psin;
1605 struct sockaddr_storage *ptss;
1606 struct sockaddr_storage *tss;
1607 int error;
1608 size_t idx, nsrcs, ncsrcs;
1609
1610 INP_WLOCK_ASSERT(inp);
1611
1612 imo = inp->inp_moptions;
1613 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1614
1615 INP_WUNLOCK(inp);
1616
1617 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1618 sizeof(struct __msfilterreq));
1619 if (error)
1620 return (error);
1621
1622 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1623 return (EINVAL);
1624
1625 ifp = ifnet_byindex(msfr.msfr_ifindex);
1626 if (ifp == NULL)
1627 return (EINVAL);
1628
1629 INP_WLOCK(inp);
1630
1631 /*
1632 * Lookup group on the socket.
1633 */
1634 gsa = (sockunion_t *)&msfr.msfr_group;
1635 idx = imo_match_group(imo, ifp, &gsa->sa);
1636 if (idx == -1 || imo->imo_mfilters == NULL) {
1637 INP_WUNLOCK(inp);
1638 return (EADDRNOTAVAIL);
1639 }
1640 imf = &imo->imo_mfilters[idx];
1641
1642 /*
1643 * Ignore memberships which are in limbo.
1644 */
1645 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1646 INP_WUNLOCK(inp);
1647 return (EAGAIN);
1648 }
1649 msfr.msfr_fmode = imf->imf_st[1];
1650
1651 /*
1652 * If the user specified a buffer, copy out the source filter
1653 * entries to userland gracefully.
1654 * We only copy out the number of entries which userland
1655 * has asked for, but we always tell userland how big the
1656 * buffer really needs to be.
1657 */
1658 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1659 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1660 tss = NULL;
1661 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1662 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1663 M_TEMP, M_NOWAIT | M_ZERO);
1664 if (tss == NULL) {
1665 INP_WUNLOCK(inp);
1666 return (ENOBUFS);
1667 }
1668 }
1669
1670 /*
1671 * Count number of sources in-mode at t0.
1672 * If buffer space exists and remains, copy out source entries.
1673 */
1674 nsrcs = msfr.msfr_nsrcs;
1675 ncsrcs = 0;
1676 ptss = tss;
1677 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1678 lims = (struct in_msource *)ims;
1679 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1680 lims->imsl_st[0] != imf->imf_st[0])
1681 continue;
1682 ++ncsrcs;
1683 if (tss != NULL && nsrcs > 0) {
1684 psin = (struct sockaddr_in *)ptss;
1685 psin->sin_family = AF_INET;
1686 psin->sin_len = sizeof(struct sockaddr_in);
1687 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1688 psin->sin_port = 0;
1689 ++ptss;
1690 --nsrcs;
1691 }
1692 }
1693
1694 INP_WUNLOCK(inp);
1695
1696 if (tss != NULL) {
1697 error = copyout(tss, msfr.msfr_srcs,
1698 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1699 free(tss, M_TEMP);
1700 if (error)
1701 return (error);
1702 }
1703
1704 msfr.msfr_nsrcs = ncsrcs;
1705 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1706
1707 return (error);
1708 }
1709
1710 /*
1711 * Return the IP multicast options in response to user getsockopt().
1712 */
1713 int
1714 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1715 {
1716 struct ip_mreqn mreqn;
1717 struct ip_moptions *imo;
1718 struct ifnet *ifp;
1719 struct in_ifaddr *ia;
1720 int error, optval;
1721 u_char coptval;
1722
1723 INP_WLOCK(inp);
1724 imo = inp->inp_moptions;
1725 /*
1726 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1727 * or is a divert socket, reject it.
1728 */
1729 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1730 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1731 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1732 INP_WUNLOCK(inp);
1733 return (EOPNOTSUPP);
1734 }
1735
1736 error = 0;
1737 switch (sopt->sopt_name) {
1738 case IP_MULTICAST_VIF:
1739 if (imo != NULL)
1740 optval = imo->imo_multicast_vif;
1741 else
1742 optval = -1;
1743 INP_WUNLOCK(inp);
1744 error = sooptcopyout(sopt, &optval, sizeof(int));
1745 break;
1746
1747 case IP_MULTICAST_IF:
1748 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1749 if (imo != NULL) {
1750 ifp = imo->imo_multicast_ifp;
1751 if (!in_nullhost(imo->imo_multicast_addr)) {
1752 mreqn.imr_address = imo->imo_multicast_addr;
1753 } else if (ifp != NULL) {
1754 mreqn.imr_ifindex = ifp->if_index;
1755 IFP_TO_IA(ifp, ia);
1756 if (ia != NULL) {
1757 mreqn.imr_address =
1758 IA_SIN(ia)->sin_addr;
1759 ifa_free(&ia->ia_ifa);
1760 }
1761 }
1762 }
1763 INP_WUNLOCK(inp);
1764 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1765 error = sooptcopyout(sopt, &mreqn,
1766 sizeof(struct ip_mreqn));
1767 } else {
1768 error = sooptcopyout(sopt, &mreqn.imr_address,
1769 sizeof(struct in_addr));
1770 }
1771 break;
1772
1773 case IP_MULTICAST_TTL:
1774 if (imo == 0)
1775 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1776 else
1777 optval = coptval = imo->imo_multicast_ttl;
1778 INP_WUNLOCK(inp);
1779 if (sopt->sopt_valsize == sizeof(u_char))
1780 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1781 else
1782 error = sooptcopyout(sopt, &optval, sizeof(int));
1783 break;
1784
1785 case IP_MULTICAST_LOOP:
1786 if (imo == 0)
1787 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1788 else
1789 optval = coptval = imo->imo_multicast_loop;
1790 INP_WUNLOCK(inp);
1791 if (sopt->sopt_valsize == sizeof(u_char))
1792 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1793 else
1794 error = sooptcopyout(sopt, &optval, sizeof(int));
1795 break;
1796
1797 case IP_MSFILTER:
1798 if (imo == NULL) {
1799 error = EADDRNOTAVAIL;
1800 INP_WUNLOCK(inp);
1801 } else {
1802 error = inp_get_source_filters(inp, sopt);
1803 }
1804 break;
1805
1806 default:
1807 INP_WUNLOCK(inp);
1808 error = ENOPROTOOPT;
1809 break;
1810 }
1811
1812 INP_UNLOCK_ASSERT(inp);
1813
1814 return (error);
1815 }
1816
1817 /*
1818 * Look up the ifnet to use for a multicast group membership,
1819 * given the IPv4 address of an interface, and the IPv4 group address.
1820 *
1821 * This routine exists to support legacy multicast applications
1822 * which do not understand that multicast memberships are scoped to
1823 * specific physical links in the networking stack, or which need
1824 * to join link-scope groups before IPv4 addresses are configured.
1825 *
1826 * If inp is non-NULL, use this socket's current FIB number for any
1827 * required FIB lookup.
1828 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1829 * and use its ifp; usually, this points to the default next-hop.
1830 *
1831 * If the FIB lookup fails, attempt to use the first non-loopback
1832 * interface with multicast capability in the system as a
1833 * last resort. The legacy IPv4 ASM API requires that we do
1834 * this in order to allow groups to be joined when the routing
1835 * table has not yet been populated during boot.
1836 *
1837 * Returns NULL if no ifp could be found.
1838 *
1839 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1840 * FUTURE: Implement IPv4 source-address selection.
1841 */
1842 static struct ifnet *
1843 inp_lookup_mcast_ifp(const struct inpcb *inp,
1844 const struct sockaddr_in *gsin, const struct in_addr ina)
1845 {
1846 struct ifnet *ifp;
1847
1848 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1849 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1850 ("%s: not multicast", __func__));
1851
1852 ifp = NULL;
1853 if (!in_nullhost(ina)) {
1854 INADDR_TO_IFP(ina, ifp);
1855 } else {
1856 struct route ro;
1857
1858 ro.ro_rt = NULL;
1859 memcpy(&ro.ro_dst, gsin, sizeof(struct sockaddr_in));
1860 in_rtalloc_ign(&ro, 0, inp ? inp->inp_inc.inc_fibnum : 0);
1861 if (ro.ro_rt != NULL) {
1862 ifp = ro.ro_rt->rt_ifp;
1863 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1864 RTFREE(ro.ro_rt);
1865 } else {
1866 struct in_ifaddr *ia;
1867 struct ifnet *mifp;
1868
1869 mifp = NULL;
1870 IN_IFADDR_RLOCK();
1871 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
1872 mifp = ia->ia_ifp;
1873 if (!(mifp->if_flags & IFF_LOOPBACK) &&
1874 (mifp->if_flags & IFF_MULTICAST)) {
1875 ifp = mifp;
1876 break;
1877 }
1878 }
1879 IN_IFADDR_RUNLOCK();
1880 }
1881 }
1882
1883 return (ifp);
1884 }
1885
1886 /*
1887 * Join an IPv4 multicast group, possibly with a source.
1888 */
1889 static int
1890 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
1891 {
1892 struct group_source_req gsr;
1893 sockunion_t *gsa, *ssa;
1894 struct ifnet *ifp;
1895 struct in_mfilter *imf;
1896 struct ip_moptions *imo;
1897 struct in_multi *inm;
1898 struct in_msource *lims;
1899 size_t idx;
1900 int error, is_new;
1901
1902 ifp = NULL;
1903 imf = NULL;
1904 lims = NULL;
1905 error = 0;
1906 is_new = 0;
1907
1908 memset(&gsr, 0, sizeof(struct group_source_req));
1909 gsa = (sockunion_t *)&gsr.gsr_group;
1910 gsa->ss.ss_family = AF_UNSPEC;
1911 ssa = (sockunion_t *)&gsr.gsr_source;
1912 ssa->ss.ss_family = AF_UNSPEC;
1913
1914 switch (sopt->sopt_name) {
1915 case IP_ADD_MEMBERSHIP:
1916 case IP_ADD_SOURCE_MEMBERSHIP: {
1917 struct ip_mreq_source mreqs;
1918
1919 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
1920 error = sooptcopyin(sopt, &mreqs,
1921 sizeof(struct ip_mreq),
1922 sizeof(struct ip_mreq));
1923 /*
1924 * Do argument switcharoo from ip_mreq into
1925 * ip_mreq_source to avoid using two instances.
1926 */
1927 mreqs.imr_interface = mreqs.imr_sourceaddr;
1928 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
1929 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1930 error = sooptcopyin(sopt, &mreqs,
1931 sizeof(struct ip_mreq_source),
1932 sizeof(struct ip_mreq_source));
1933 }
1934 if (error)
1935 return (error);
1936
1937 gsa->sin.sin_family = AF_INET;
1938 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1939 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1940
1941 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
1942 ssa->sin.sin_family = AF_INET;
1943 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1944 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1945 }
1946
1947 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1948 return (EINVAL);
1949
1950 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
1951 mreqs.imr_interface);
1952 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
1953 __func__, inet_ntoa(mreqs.imr_interface), ifp);
1954 break;
1955 }
1956
1957 case MCAST_JOIN_GROUP:
1958 case MCAST_JOIN_SOURCE_GROUP:
1959 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1960 error = sooptcopyin(sopt, &gsr,
1961 sizeof(struct group_req),
1962 sizeof(struct group_req));
1963 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1964 error = sooptcopyin(sopt, &gsr,
1965 sizeof(struct group_source_req),
1966 sizeof(struct group_source_req));
1967 }
1968 if (error)
1969 return (error);
1970
1971 if (gsa->sin.sin_family != AF_INET ||
1972 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1973 return (EINVAL);
1974
1975 /*
1976 * Overwrite the port field if present, as the sockaddr
1977 * being copied in may be matched with a binary comparison.
1978 */
1979 gsa->sin.sin_port = 0;
1980 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1981 if (ssa->sin.sin_family != AF_INET ||
1982 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1983 return (EINVAL);
1984 ssa->sin.sin_port = 0;
1985 }
1986
1987 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1988 return (EINVAL);
1989
1990 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1991 return (EADDRNOTAVAIL);
1992 ifp = ifnet_byindex(gsr.gsr_interface);
1993 break;
1994
1995 default:
1996 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1997 __func__, sopt->sopt_name);
1998 return (EOPNOTSUPP);
1999 break;
2000 }
2001
2002 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2003 return (EADDRNOTAVAIL);
2004
2005 imo = inp_findmoptions(inp);
2006 idx = imo_match_group(imo, ifp, &gsa->sa);
2007 if (idx == -1) {
2008 is_new = 1;
2009 } else {
2010 inm = imo->imo_membership[idx];
2011 imf = &imo->imo_mfilters[idx];
2012 if (ssa->ss.ss_family != AF_UNSPEC) {
2013 /*
2014 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2015 * is an error. On an existing inclusive membership,
2016 * it just adds the source to the filter list.
2017 */
2018 if (imf->imf_st[1] != MCAST_INCLUDE) {
2019 error = EINVAL;
2020 goto out_inp_locked;
2021 }
2022 /*
2023 * Throw out duplicates.
2024 *
2025 * XXX FIXME: This makes a naive assumption that
2026 * even if entries exist for *ssa in this imf,
2027 * they will be rejected as dupes, even if they
2028 * are not valid in the current mode (in-mode).
2029 *
2030 * in_msource is transactioned just as for anything
2031 * else in SSM -- but note naive use of inm_graft()
2032 * below for allocating new filter entries.
2033 *
2034 * This is only an issue if someone mixes the
2035 * full-state SSM API with the delta-based API,
2036 * which is discouraged in the relevant RFCs.
2037 */
2038 lims = imo_match_source(imo, idx, &ssa->sa);
2039 if (lims != NULL /*&&
2040 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2041 error = EADDRNOTAVAIL;
2042 goto out_inp_locked;
2043 }
2044 } else {
2045 /*
2046 * MCAST_JOIN_GROUP on an existing exclusive
2047 * membership is an error; return EADDRINUSE
2048 * to preserve 4.4BSD API idempotence, and
2049 * avoid tedious detour to code below.
2050 * NOTE: This is bending RFC 3678 a bit.
2051 *
2052 * On an existing inclusive membership, this is also
2053 * an error; if you want to change filter mode,
2054 * you must use the userland API setsourcefilter().
2055 * XXX We don't reject this for imf in UNDEFINED
2056 * state at t1, because allocation of a filter
2057 * is atomic with allocation of a membership.
2058 */
2059 error = EINVAL;
2060 if (imf->imf_st[1] == MCAST_EXCLUDE)
2061 error = EADDRINUSE;
2062 goto out_inp_locked;
2063 }
2064 }
2065
2066 /*
2067 * Begin state merge transaction at socket layer.
2068 */
2069 INP_WLOCK_ASSERT(inp);
2070
2071 if (is_new) {
2072 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2073 error = imo_grow(imo);
2074 if (error)
2075 goto out_inp_locked;
2076 }
2077 /*
2078 * Allocate the new slot upfront so we can deal with
2079 * grafting the new source filter in same code path
2080 * as for join-source on existing membership.
2081 */
2082 idx = imo->imo_num_memberships;
2083 imo->imo_membership[idx] = NULL;
2084 imo->imo_num_memberships++;
2085 KASSERT(imo->imo_mfilters != NULL,
2086 ("%s: imf_mfilters vector was not allocated", __func__));
2087 imf = &imo->imo_mfilters[idx];
2088 KASSERT(RB_EMPTY(&imf->imf_sources),
2089 ("%s: imf_sources not empty", __func__));
2090 }
2091
2092 /*
2093 * Graft new source into filter list for this inpcb's
2094 * membership of the group. The in_multi may not have
2095 * been allocated yet if this is a new membership, however,
2096 * the in_mfilter slot will be allocated and must be initialized.
2097 *
2098 * Note: Grafting of exclusive mode filters doesn't happen
2099 * in this path.
2100 * XXX: Should check for non-NULL lims (node exists but may
2101 * not be in-mode) for interop with full-state API.
2102 */
2103 if (ssa->ss.ss_family != AF_UNSPEC) {
2104 /* Membership starts in IN mode */
2105 if (is_new) {
2106 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2107 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2108 } else {
2109 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2110 }
2111 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2112 if (lims == NULL) {
2113 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2114 __func__);
2115 error = ENOMEM;
2116 goto out_imo_free;
2117 }
2118 } else {
2119 /* No address specified; Membership starts in EX mode */
2120 if (is_new) {
2121 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2122 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2123 }
2124 }
2125
2126 /*
2127 * Begin state merge transaction at IGMP layer.
2128 */
2129 IN_MULTI_LOCK();
2130
2131 if (is_new) {
2132 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2133 &inm);
2134 if (error) {
2135 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2136 __func__);
2137 IN_MULTI_UNLOCK();
2138 goto out_imo_free;
2139 }
2140 imo->imo_membership[idx] = inm;
2141 } else {
2142 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2143 error = inm_merge(inm, imf);
2144 if (error) {
2145 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2146 __func__);
2147 goto out_in_multi_locked;
2148 }
2149 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2150 error = igmp_change_state(inm);
2151 if (error) {
2152 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2153 __func__);
2154 goto out_in_multi_locked;
2155 }
2156 }
2157
2158 out_in_multi_locked:
2159
2160 IN_MULTI_UNLOCK();
2161
2162 INP_WLOCK_ASSERT(inp);
2163 if (error) {
2164 imf_rollback(imf);
2165 if (is_new)
2166 imf_purge(imf);
2167 else
2168 imf_reap(imf);
2169 } else {
2170 imf_commit(imf);
2171 }
2172
2173 out_imo_free:
2174 if (error && is_new) {
2175 imo->imo_membership[idx] = NULL;
2176 --imo->imo_num_memberships;
2177 }
2178
2179 out_inp_locked:
2180 INP_WUNLOCK(inp);
2181 return (error);
2182 }
2183
2184 /*
2185 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2186 */
2187 static int
2188 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2189 {
2190 struct group_source_req gsr;
2191 struct ip_mreq_source mreqs;
2192 sockunion_t *gsa, *ssa;
2193 struct ifnet *ifp;
2194 struct in_mfilter *imf;
2195 struct ip_moptions *imo;
2196 struct in_msource *ims;
2197 struct in_multi *inm;
2198 size_t idx;
2199 int error, is_final;
2200
2201 ifp = NULL;
2202 error = 0;
2203 is_final = 1;
2204
2205 memset(&gsr, 0, sizeof(struct group_source_req));
2206 gsa = (sockunion_t *)&gsr.gsr_group;
2207 gsa->ss.ss_family = AF_UNSPEC;
2208 ssa = (sockunion_t *)&gsr.gsr_source;
2209 ssa->ss.ss_family = AF_UNSPEC;
2210
2211 switch (sopt->sopt_name) {
2212 case IP_DROP_MEMBERSHIP:
2213 case IP_DROP_SOURCE_MEMBERSHIP:
2214 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2215 error = sooptcopyin(sopt, &mreqs,
2216 sizeof(struct ip_mreq),
2217 sizeof(struct ip_mreq));
2218 /*
2219 * Swap interface and sourceaddr arguments,
2220 * as ip_mreq and ip_mreq_source are laid
2221 * out differently.
2222 */
2223 mreqs.imr_interface = mreqs.imr_sourceaddr;
2224 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2225 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2226 error = sooptcopyin(sopt, &mreqs,
2227 sizeof(struct ip_mreq_source),
2228 sizeof(struct ip_mreq_source));
2229 }
2230 if (error)
2231 return (error);
2232
2233 gsa->sin.sin_family = AF_INET;
2234 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2235 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2236
2237 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2238 ssa->sin.sin_family = AF_INET;
2239 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2240 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2241 }
2242
2243 /*
2244 * Attempt to look up hinted ifp from interface address.
2245 * Fallthrough with null ifp iff lookup fails, to
2246 * preserve 4.4BSD mcast API idempotence.
2247 * XXX NOTE WELL: The RFC 3678 API is preferred because
2248 * using an IPv4 address as a key is racy.
2249 */
2250 if (!in_nullhost(mreqs.imr_interface))
2251 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2252
2253 CTR3(KTR_IGMPV3, "%s: imr_interface = %s, ifp = %p",
2254 __func__, inet_ntoa(mreqs.imr_interface), ifp);
2255
2256 break;
2257
2258 case MCAST_LEAVE_GROUP:
2259 case MCAST_LEAVE_SOURCE_GROUP:
2260 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2261 error = sooptcopyin(sopt, &gsr,
2262 sizeof(struct group_req),
2263 sizeof(struct group_req));
2264 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2265 error = sooptcopyin(sopt, &gsr,
2266 sizeof(struct group_source_req),
2267 sizeof(struct group_source_req));
2268 }
2269 if (error)
2270 return (error);
2271
2272 if (gsa->sin.sin_family != AF_INET ||
2273 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2274 return (EINVAL);
2275
2276 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2277 if (ssa->sin.sin_family != AF_INET ||
2278 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2279 return (EINVAL);
2280 }
2281
2282 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2283 return (EADDRNOTAVAIL);
2284
2285 ifp = ifnet_byindex(gsr.gsr_interface);
2286
2287 if (ifp == NULL)
2288 return (EADDRNOTAVAIL);
2289 break;
2290
2291 default:
2292 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2293 __func__, sopt->sopt_name);
2294 return (EOPNOTSUPP);
2295 break;
2296 }
2297
2298 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2299 return (EINVAL);
2300
2301 /*
2302 * Find the membership in the membership array.
2303 */
2304 imo = inp_findmoptions(inp);
2305 idx = imo_match_group(imo, ifp, &gsa->sa);
2306 if (idx == -1) {
2307 error = EADDRNOTAVAIL;
2308 goto out_inp_locked;
2309 }
2310 inm = imo->imo_membership[idx];
2311 imf = &imo->imo_mfilters[idx];
2312
2313 if (ssa->ss.ss_family != AF_UNSPEC)
2314 is_final = 0;
2315
2316 /*
2317 * Begin state merge transaction at socket layer.
2318 */
2319 INP_WLOCK_ASSERT(inp);
2320
2321 /*
2322 * If we were instructed only to leave a given source, do so.
2323 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2324 */
2325 if (is_final) {
2326 imf_leave(imf);
2327 } else {
2328 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2329 error = EADDRNOTAVAIL;
2330 goto out_inp_locked;
2331 }
2332 ims = imo_match_source(imo, idx, &ssa->sa);
2333 if (ims == NULL) {
2334 CTR3(KTR_IGMPV3, "%s: source %s %spresent", __func__,
2335 inet_ntoa(ssa->sin.sin_addr), "not ");
2336 error = EADDRNOTAVAIL;
2337 goto out_inp_locked;
2338 }
2339 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2340 error = imf_prune(imf, &ssa->sin);
2341 if (error) {
2342 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2343 __func__);
2344 goto out_inp_locked;
2345 }
2346 }
2347
2348 /*
2349 * Begin state merge transaction at IGMP layer.
2350 */
2351 IN_MULTI_LOCK();
2352
2353 if (is_final) {
2354 /*
2355 * Give up the multicast address record to which
2356 * the membership points.
2357 */
2358 (void)in_leavegroup_locked(inm, imf);
2359 } else {
2360 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2361 error = inm_merge(inm, imf);
2362 if (error) {
2363 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2364 __func__);
2365 goto out_in_multi_locked;
2366 }
2367
2368 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2369 error = igmp_change_state(inm);
2370 if (error) {
2371 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2372 __func__);
2373 }
2374 }
2375
2376 out_in_multi_locked:
2377
2378 IN_MULTI_UNLOCK();
2379
2380 if (error)
2381 imf_rollback(imf);
2382 else
2383 imf_commit(imf);
2384
2385 imf_reap(imf);
2386
2387 if (is_final) {
2388 /* Remove the gap in the membership and filter array. */
2389 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2390 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2391 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2392 }
2393 imo->imo_num_memberships--;
2394 }
2395
2396 out_inp_locked:
2397 INP_WUNLOCK(inp);
2398 return (error);
2399 }
2400
2401 /*
2402 * Select the interface for transmitting IPv4 multicast datagrams.
2403 *
2404 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2405 * may be passed to this socket option. An address of INADDR_ANY or an
2406 * interface index of 0 is used to remove a previous selection.
2407 * When no interface is selected, one is chosen for every send.
2408 */
2409 static int
2410 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2411 {
2412 struct in_addr addr;
2413 struct ip_mreqn mreqn;
2414 struct ifnet *ifp;
2415 struct ip_moptions *imo;
2416 int error;
2417
2418 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2419 /*
2420 * An interface index was specified using the
2421 * Linux-derived ip_mreqn structure.
2422 */
2423 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2424 sizeof(struct ip_mreqn));
2425 if (error)
2426 return (error);
2427
2428 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2429 return (EINVAL);
2430
2431 if (mreqn.imr_ifindex == 0) {
2432 ifp = NULL;
2433 } else {
2434 ifp = ifnet_byindex(mreqn.imr_ifindex);
2435 if (ifp == NULL)
2436 return (EADDRNOTAVAIL);
2437 }
2438 } else {
2439 /*
2440 * An interface was specified by IPv4 address.
2441 * This is the traditional BSD usage.
2442 */
2443 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2444 sizeof(struct in_addr));
2445 if (error)
2446 return (error);
2447 if (in_nullhost(addr)) {
2448 ifp = NULL;
2449 } else {
2450 INADDR_TO_IFP(addr, ifp);
2451 if (ifp == NULL)
2452 return (EADDRNOTAVAIL);
2453 }
2454 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = %s", __func__, ifp,
2455 inet_ntoa(addr));
2456 }
2457
2458 /* Reject interfaces which do not support multicast. */
2459 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2460 return (EOPNOTSUPP);
2461
2462 imo = inp_findmoptions(inp);
2463 imo->imo_multicast_ifp = ifp;
2464 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2465 INP_WUNLOCK(inp);
2466
2467 return (0);
2468 }
2469
2470 /*
2471 * Atomically set source filters on a socket for an IPv4 multicast group.
2472 *
2473 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2474 */
2475 static int
2476 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2477 {
2478 struct __msfilterreq msfr;
2479 sockunion_t *gsa;
2480 struct ifnet *ifp;
2481 struct in_mfilter *imf;
2482 struct ip_moptions *imo;
2483 struct in_multi *inm;
2484 size_t idx;
2485 int error;
2486
2487 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2488 sizeof(struct __msfilterreq));
2489 if (error)
2490 return (error);
2491
2492 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2493 return (ENOBUFS);
2494
2495 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2496 msfr.msfr_fmode != MCAST_INCLUDE))
2497 return (EINVAL);
2498
2499 if (msfr.msfr_group.ss_family != AF_INET ||
2500 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2501 return (EINVAL);
2502
2503 gsa = (sockunion_t *)&msfr.msfr_group;
2504 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2505 return (EINVAL);
2506
2507 gsa->sin.sin_port = 0; /* ignore port */
2508
2509 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2510 return (EADDRNOTAVAIL);
2511
2512 ifp = ifnet_byindex(msfr.msfr_ifindex);
2513 if (ifp == NULL)
2514 return (EADDRNOTAVAIL);
2515
2516 /*
2517 * Take the INP write lock.
2518 * Check if this socket is a member of this group.
2519 */
2520 imo = inp_findmoptions(inp);
2521 idx = imo_match_group(imo, ifp, &gsa->sa);
2522 if (idx == -1 || imo->imo_mfilters == NULL) {
2523 error = EADDRNOTAVAIL;
2524 goto out_inp_locked;
2525 }
2526 inm = imo->imo_membership[idx];
2527 imf = &imo->imo_mfilters[idx];
2528
2529 /*
2530 * Begin state merge transaction at socket layer.
2531 */
2532 INP_WLOCK_ASSERT(inp);
2533
2534 imf->imf_st[1] = msfr.msfr_fmode;
2535
2536 /*
2537 * Apply any new source filters, if present.
2538 * Make a copy of the user-space source vector so
2539 * that we may copy them with a single copyin. This
2540 * allows us to deal with page faults up-front.
2541 */
2542 if (msfr.msfr_nsrcs > 0) {
2543 struct in_msource *lims;
2544 struct sockaddr_in *psin;
2545 struct sockaddr_storage *kss, *pkss;
2546 int i;
2547
2548 INP_WUNLOCK(inp);
2549
2550 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2551 __func__, (unsigned long)msfr.msfr_nsrcs);
2552 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2553 M_TEMP, M_WAITOK);
2554 error = copyin(msfr.msfr_srcs, kss,
2555 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2556 if (error) {
2557 free(kss, M_TEMP);
2558 return (error);
2559 }
2560
2561 INP_WLOCK(inp);
2562
2563 /*
2564 * Mark all source filters as UNDEFINED at t1.
2565 * Restore new group filter mode, as imf_leave()
2566 * will set it to INCLUDE.
2567 */
2568 imf_leave(imf);
2569 imf->imf_st[1] = msfr.msfr_fmode;
2570
2571 /*
2572 * Update socket layer filters at t1, lazy-allocating
2573 * new entries. This saves a bunch of memory at the
2574 * cost of one RB_FIND() per source entry; duplicate
2575 * entries in the msfr_nsrcs vector are ignored.
2576 * If we encounter an error, rollback transaction.
2577 *
2578 * XXX This too could be replaced with a set-symmetric
2579 * difference like loop to avoid walking from root
2580 * every time, as the key space is common.
2581 */
2582 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2583 psin = (struct sockaddr_in *)pkss;
2584 if (psin->sin_family != AF_INET) {
2585 error = EAFNOSUPPORT;
2586 break;
2587 }
2588 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2589 error = EINVAL;
2590 break;
2591 }
2592 error = imf_get_source(imf, psin, &lims);
2593 if (error)
2594 break;
2595 lims->imsl_st[1] = imf->imf_st[1];
2596 }
2597 free(kss, M_TEMP);
2598 }
2599
2600 if (error)
2601 goto out_imf_rollback;
2602
2603 INP_WLOCK_ASSERT(inp);
2604 IN_MULTI_LOCK();
2605
2606 /*
2607 * Begin state merge transaction at IGMP layer.
2608 */
2609 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2610 error = inm_merge(inm, imf);
2611 if (error) {
2612 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2613 goto out_in_multi_locked;
2614 }
2615
2616 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2617 error = igmp_change_state(inm);
2618 if (error)
2619 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2620
2621 out_in_multi_locked:
2622
2623 IN_MULTI_UNLOCK();
2624
2625 out_imf_rollback:
2626 if (error)
2627 imf_rollback(imf);
2628 else
2629 imf_commit(imf);
2630
2631 imf_reap(imf);
2632
2633 out_inp_locked:
2634 INP_WUNLOCK(inp);
2635 return (error);
2636 }
2637
2638 /*
2639 * Set the IP multicast options in response to user setsockopt().
2640 *
2641 * Many of the socket options handled in this function duplicate the
2642 * functionality of socket options in the regular unicast API. However,
2643 * it is not possible to merge the duplicate code, because the idempotence
2644 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2645 * the effects of these options must be treated as separate and distinct.
2646 *
2647 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2648 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2649 * is refactored to no longer use vifs.
2650 */
2651 int
2652 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2653 {
2654 struct ip_moptions *imo;
2655 int error;
2656
2657 error = 0;
2658
2659 /*
2660 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2661 * or is a divert socket, reject it.
2662 */
2663 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2664 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2665 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2666 return (EOPNOTSUPP);
2667
2668 switch (sopt->sopt_name) {
2669 case IP_MULTICAST_VIF: {
2670 int vifi;
2671 /*
2672 * Select a multicast VIF for transmission.
2673 * Only useful if multicast forwarding is active.
2674 */
2675 if (legal_vif_num == NULL) {
2676 error = EOPNOTSUPP;
2677 break;
2678 }
2679 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2680 if (error)
2681 break;
2682 if (!legal_vif_num(vifi) && (vifi != -1)) {
2683 error = EINVAL;
2684 break;
2685 }
2686 imo = inp_findmoptions(inp);
2687 imo->imo_multicast_vif = vifi;
2688 INP_WUNLOCK(inp);
2689 break;
2690 }
2691
2692 case IP_MULTICAST_IF:
2693 error = inp_set_multicast_if(inp, sopt);
2694 break;
2695
2696 case IP_MULTICAST_TTL: {
2697 u_char ttl;
2698
2699 /*
2700 * Set the IP time-to-live for outgoing multicast packets.
2701 * The original multicast API required a char argument,
2702 * which is inconsistent with the rest of the socket API.
2703 * We allow either a char or an int.
2704 */
2705 if (sopt->sopt_valsize == sizeof(u_char)) {
2706 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2707 sizeof(u_char));
2708 if (error)
2709 break;
2710 } else {
2711 u_int ittl;
2712
2713 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2714 sizeof(u_int));
2715 if (error)
2716 break;
2717 if (ittl > 255) {
2718 error = EINVAL;
2719 break;
2720 }
2721 ttl = (u_char)ittl;
2722 }
2723 imo = inp_findmoptions(inp);
2724 imo->imo_multicast_ttl = ttl;
2725 INP_WUNLOCK(inp);
2726 break;
2727 }
2728
2729 case IP_MULTICAST_LOOP: {
2730 u_char loop;
2731
2732 /*
2733 * Set the loopback flag for outgoing multicast packets.
2734 * Must be zero or one. The original multicast API required a
2735 * char argument, which is inconsistent with the rest
2736 * of the socket API. We allow either a char or an int.
2737 */
2738 if (sopt->sopt_valsize == sizeof(u_char)) {
2739 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2740 sizeof(u_char));
2741 if (error)
2742 break;
2743 } else {
2744 u_int iloop;
2745
2746 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2747 sizeof(u_int));
2748 if (error)
2749 break;
2750 loop = (u_char)iloop;
2751 }
2752 imo = inp_findmoptions(inp);
2753 imo->imo_multicast_loop = !!loop;
2754 INP_WUNLOCK(inp);
2755 break;
2756 }
2757
2758 case IP_ADD_MEMBERSHIP:
2759 case IP_ADD_SOURCE_MEMBERSHIP:
2760 case MCAST_JOIN_GROUP:
2761 case MCAST_JOIN_SOURCE_GROUP:
2762 error = inp_join_group(inp, sopt);
2763 break;
2764
2765 case IP_DROP_MEMBERSHIP:
2766 case IP_DROP_SOURCE_MEMBERSHIP:
2767 case MCAST_LEAVE_GROUP:
2768 case MCAST_LEAVE_SOURCE_GROUP:
2769 error = inp_leave_group(inp, sopt);
2770 break;
2771
2772 case IP_BLOCK_SOURCE:
2773 case IP_UNBLOCK_SOURCE:
2774 case MCAST_BLOCK_SOURCE:
2775 case MCAST_UNBLOCK_SOURCE:
2776 error = inp_block_unblock_source(inp, sopt);
2777 break;
2778
2779 case IP_MSFILTER:
2780 error = inp_set_source_filters(inp, sopt);
2781 break;
2782
2783 default:
2784 error = EOPNOTSUPP;
2785 break;
2786 }
2787
2788 INP_UNLOCK_ASSERT(inp);
2789
2790 return (error);
2791 }
2792
2793 /*
2794 * Expose IGMP's multicast filter mode and source list(s) to userland,
2795 * keyed by (ifindex, group).
2796 * The filter mode is written out as a uint32_t, followed by
2797 * 0..n of struct in_addr.
2798 * For use by ifmcstat(8).
2799 * SMPng: NOTE: unlocked read of ifindex space.
2800 */
2801 static int
2802 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2803 {
2804 struct in_addr src, group;
2805 struct ifnet *ifp;
2806 struct ifmultiaddr *ifma;
2807 struct in_multi *inm;
2808 struct ip_msource *ims;
2809 int *name;
2810 int retval;
2811 u_int namelen;
2812 uint32_t fmode, ifindex;
2813
2814 name = (int *)arg1;
2815 namelen = arg2;
2816
2817 if (req->newptr != NULL)
2818 return (EPERM);
2819
2820 if (namelen != 2)
2821 return (EINVAL);
2822
2823 ifindex = name[0];
2824 if (ifindex <= 0 || ifindex > V_if_index) {
2825 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2826 __func__, ifindex);
2827 return (ENOENT);
2828 }
2829
2830 group.s_addr = name[1];
2831 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2832 CTR2(KTR_IGMPV3, "%s: group %s is not multicast",
2833 __func__, inet_ntoa(group));
2834 return (EINVAL);
2835 }
2836
2837 ifp = ifnet_byindex(ifindex);
2838 if (ifp == NULL) {
2839 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2840 __func__, ifindex);
2841 return (ENOENT);
2842 }
2843
2844 retval = sysctl_wire_old_buffer(req,
2845 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
2846 if (retval)
2847 return (retval);
2848
2849 IN_MULTI_LOCK();
2850
2851 IF_ADDR_RLOCK(ifp);
2852 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2853 if (ifma->ifma_addr->sa_family != AF_INET ||
2854 ifma->ifma_protospec == NULL)
2855 continue;
2856 inm = (struct in_multi *)ifma->ifma_protospec;
2857 if (!in_hosteq(inm->inm_addr, group))
2858 continue;
2859 fmode = inm->inm_st[1].iss_fmode;
2860 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2861 if (retval != 0)
2862 break;
2863 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
2864 #ifdef KTR
2865 struct in_addr ina;
2866 ina.s_addr = htonl(ims->ims_haddr);
2867 CTR2(KTR_IGMPV3, "%s: visit node %s", __func__,
2868 inet_ntoa(ina));
2869 #endif
2870 /*
2871 * Only copy-out sources which are in-mode.
2872 */
2873 if (fmode != ims_get_mode(inm, ims, 1)) {
2874 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
2875 __func__);
2876 continue;
2877 }
2878 src.s_addr = htonl(ims->ims_haddr);
2879 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
2880 if (retval != 0)
2881 break;
2882 }
2883 }
2884 IF_ADDR_RUNLOCK(ifp);
2885
2886 IN_MULTI_UNLOCK();
2887
2888 return (retval);
2889 }
2890
2891 #ifdef KTR
2892
2893 static const char *inm_modestrs[] = { "un", "in", "ex" };
2894
2895 static const char *
2896 inm_mode_str(const int mode)
2897 {
2898
2899 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2900 return (inm_modestrs[mode]);
2901 return ("??");
2902 }
2903
2904 static const char *inm_statestrs[] = {
2905 "not-member",
2906 "silent",
2907 "idle",
2908 "lazy",
2909 "sleeping",
2910 "awakening",
2911 "query-pending",
2912 "sg-query-pending",
2913 "leaving"
2914 };
2915
2916 static const char *
2917 inm_state_str(const int state)
2918 {
2919
2920 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
2921 return (inm_statestrs[state]);
2922 return ("??");
2923 }
2924
2925 /*
2926 * Dump an in_multi structure to the console.
2927 */
2928 void
2929 inm_print(const struct in_multi *inm)
2930 {
2931 int t;
2932
2933 if ((ktr_mask & KTR_IGMPV3) == 0)
2934 return;
2935
2936 printf("%s: --- begin inm %p ---\n", __func__, inm);
2937 printf("addr %s ifp %p(%s) ifma %p\n",
2938 inet_ntoa(inm->inm_addr),
2939 inm->inm_ifp,
2940 inm->inm_ifp->if_xname,
2941 inm->inm_ifma);
2942 printf("timer %u state %s refcount %u scq.len %u\n",
2943 inm->inm_timer,
2944 inm_state_str(inm->inm_state),
2945 inm->inm_refcount,
2946 inm->inm_scq.ifq_len);
2947 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
2948 inm->inm_igi,
2949 inm->inm_nsrc,
2950 inm->inm_sctimer,
2951 inm->inm_scrv);
2952 for (t = 0; t < 2; t++) {
2953 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2954 inm_mode_str(inm->inm_st[t].iss_fmode),
2955 inm->inm_st[t].iss_asm,
2956 inm->inm_st[t].iss_ex,
2957 inm->inm_st[t].iss_in,
2958 inm->inm_st[t].iss_rec);
2959 }
2960 printf("%s: --- end inm %p ---\n", __func__, inm);
2961 }
2962
2963 #else /* !KTR */
2964
2965 void
2966 inm_print(const struct in_multi *inm)
2967 {
2968
2969 }
2970
2971 #endif /* KTR */
2972
2973 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);
Cache object: fb094430731ca345a4b5cf6d6b139177
|