]> git.saurik.com Git - apple/xnu.git/blob - bsd/netinet/in_mcast.c
xnu-1699.22.73.tar.gz
[apple/xnu.git] / bsd / netinet / in_mcast.c
1 /*
2 * Copyright (c) 2010-2011 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28 /*-
29 * Copyright (c) 2007-2009 Bruce Simpson.
30 * Copyright (c) 2005 Robert N. M. Watson.
31 * All rights reserved.
32 *
33 * Redistribution and use in source and binary forms, with or without
34 * modification, are permitted provided that the following conditions
35 * are met:
36 * 1. Redistributions of source code must retain the above copyright
37 * notice, this list of conditions and the following disclaimer.
38 * 2. Redistributions in binary form must reproduce the above copyright
39 * notice, this list of conditions and the following disclaimer in the
40 * documentation and/or other materials provided with the distribution.
41 * 3. The name of the author may not be used to endorse or promote
42 * products derived from this software without specific prior written
43 * permission.
44 *
45 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
46 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
47 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
48 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
49 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
50 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
51 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
52 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
53 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
54 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
55 * SUCH DAMAGE.
56 */
57
58 /*
59 * IPv4 multicast socket, group, and socket option processing module.
60 */
61
62 #include <sys/cdefs.h>
63
64 #include <sys/param.h>
65 #include <sys/systm.h>
66 #include <sys/kernel.h>
67 #include <sys/malloc.h>
68 #include <sys/mbuf.h>
69 #include <sys/protosw.h>
70 #include <sys/socket.h>
71 #include <sys/socketvar.h>
72 #include <sys/protosw.h>
73 #include <sys/sysctl.h>
74 #include <sys/tree.h>
75 #include <sys/mcache.h>
76
77 #include <kern/zalloc.h>
78
79 #include <pexpert/pexpert.h>
80
81 #include <net/if.h>
82 #include <net/if_dl.h>
83 #include <net/route.h>
84
85 #include <netinet/in.h>
86 #include <netinet/in_systm.h>
87 #include <netinet/in_pcb.h>
88 #include <netinet/in_var.h>
89 #include <netinet/ip_var.h>
90 #include <netinet/igmp_var.h>
91
92 #ifndef __SOCKUNION_DECLARED
93 union sockunion {
94 struct sockaddr_storage ss;
95 struct sockaddr sa;
96 struct sockaddr_dl sdl;
97 struct sockaddr_in sin;
98 };
99 typedef union sockunion sockunion_t;
100 #define __SOCKUNION_DECLARED
101 #endif /* __SOCKUNION_DECLARED */
102
103 /*
104 * Functions with non-static linkage defined in this file should be
105 * declared in in_var.h:
106 * imo_multi_filter()
107 * in_addmulti()
108 * in_delmulti()
109 * in_joingroup()
110 * in_leavegroup()
111 * and ip_var.h:
112 * inp_freemoptions()
113 * inp_getmoptions()
114 * inp_setmoptions()
115 *
116 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
117 * and in_delmulti().
118 */
119 static void imf_commit(struct in_mfilter *);
120 static int imf_get_source(struct in_mfilter *imf,
121 const struct sockaddr_in *psin,
122 struct in_msource **);
123 static struct in_msource *
124 imf_graft(struct in_mfilter *, const uint8_t,
125 const struct sockaddr_in *);
126 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
127 static void imf_rollback(struct in_mfilter *);
128 static void imf_reap(struct in_mfilter *);
129 static int imo_grow(struct ip_moptions *, size_t);
130 static size_t imo_match_group(const struct ip_moptions *,
131 const struct ifnet *, const struct sockaddr *);
132 static struct in_msource *
133 imo_match_source(const struct ip_moptions *, const size_t,
134 const struct sockaddr *);
135 static void ims_merge(struct ip_msource *ims,
136 const struct in_msource *lims, const int rollback);
137 static int in_getmulti(struct ifnet *, const struct in_addr *,
138 struct in_multi **);
139 static int in_joingroup(struct ifnet *, const struct in_addr *,
140 struct in_mfilter *, struct in_multi **);
141 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
142 const int noalloc, struct ip_msource **pims);
143 static int inm_is_ifp_detached(const struct in_multi *);
144 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
145 static void inm_reap(struct in_multi *);
146 static struct ip_moptions *
147 inp_findmoptions(struct inpcb *);
148 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
149 static struct ifnet *
150 inp_lookup_mcast_ifp(const struct inpcb *,
151 const struct sockaddr_in *, const struct in_addr);
152 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
153 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
154 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
155 static int sysctl_ip_mcast_filters SYSCTL_HANDLER_ARGS;
156 static struct ifnet * ip_multicast_if(struct in_addr *, unsigned int *);
157 static __inline__ int ip_msource_cmp(const struct ip_msource *,
158 const struct ip_msource *);
159
160 SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW | CTLFLAG_LOCKED, 0, "IPv4 multicast");
161
162 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
163 SYSCTL_LONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
164 CTLFLAG_RW | CTLFLAG_LOCKED, &in_mcast_maxgrpsrc, "Max source filters per group");
165
166 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
167 SYSCTL_LONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
168 CTLFLAG_RW | CTLFLAG_LOCKED, &in_mcast_maxsocksrc,
169 "Max source filters per socket");
170
171 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
172 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_LOCKED,
173 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
174
175 SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
176 CTLFLAG_RD | CTLFLAG_LOCKED, sysctl_ip_mcast_filters,
177 "Per-interface stack-wide source filters");
178
179 RB_GENERATE_PREV(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);
180
181 #define INM_TRACE_HIST_SIZE 32 /* size of trace history */
182
183 /* For gdb */
184 __private_extern__ unsigned int inm_trace_hist_size = INM_TRACE_HIST_SIZE;
185
186 struct in_multi_dbg {
187 struct in_multi inm; /* in_multi */
188 u_int16_t inm_refhold_cnt; /* # of ref */
189 u_int16_t inm_refrele_cnt; /* # of rele */
190 /*
191 * Circular lists of inm_addref and inm_remref callers.
192 */
193 ctrace_t inm_refhold[INM_TRACE_HIST_SIZE];
194 ctrace_t inm_refrele[INM_TRACE_HIST_SIZE];
195 /*
196 * Trash list linkage
197 */
198 TAILQ_ENTRY(in_multi_dbg) inm_trash_link;
199 };
200
201 /* List of trash in_multi entries protected by inm_trash_lock */
202 static TAILQ_HEAD(, in_multi_dbg) inm_trash_head;
203 static decl_lck_mtx_data(, inm_trash_lock);
204
205 #define INM_ZONE_MAX 64 /* maximum elements in zone */
206 #define INM_ZONE_NAME "in_multi" /* zone name */
207
208 #if DEBUG
209 static unsigned int inm_debug = 1; /* debugging (enabled) */
210 #else
211 static unsigned int inm_debug; /* debugging (disabled) */
212 #endif /* !DEBUG */
213 static unsigned int inm_size; /* size of zone element */
214 static struct zone *inm_zone; /* zone for in_multi */
215
216 #define IPMS_ZONE_MAX 64 /* maximum elements in zone */
217 #define IPMS_ZONE_NAME "ip_msource" /* zone name */
218
219 static unsigned int ipms_size; /* size of zone element */
220 static struct zone *ipms_zone; /* zone for ip_msource */
221
222 #define INMS_ZONE_MAX 64 /* maximum elements in zone */
223 #define INMS_ZONE_NAME "in_msource" /* zone name */
224
225 static unsigned int inms_size; /* size of zone element */
226 static struct zone *inms_zone; /* zone for in_msource */
227
228 /* Lock group and attribute for in_multihead_lock lock */
229 static lck_attr_t *in_multihead_lock_attr;
230 static lck_grp_t *in_multihead_lock_grp;
231 static lck_grp_attr_t *in_multihead_lock_grp_attr;
232
233 static decl_lck_rw_data(, in_multihead_lock);
234 struct in_multihead in_multihead;
235
236 static struct in_multi *in_multi_alloc(int);
237 static void in_multi_free(struct in_multi *);
238 static void in_multi_attach(struct in_multi *);
239 static void inm_trace(struct in_multi *, int);
240
241 static struct ip_msource *ipms_alloc(int);
242 static void ipms_free(struct ip_msource *);
243 static struct in_msource *inms_alloc(int);
244 static void inms_free(struct in_msource *);
245
246 #define IMO_CAST_TO_NONCONST(x) ((struct ip_moptions *)(void *)(uintptr_t)x)
247 #define INM_CAST_TO_NONCONST(x) ((struct in_multi *)(void *)(uintptr_t)x)
248
249 static __inline int
250 ip_msource_cmp(const struct ip_msource *a, const struct ip_msource *b)
251 {
252
253 if (a->ims_haddr < b->ims_haddr)
254 return (-1);
255 if (a->ims_haddr == b->ims_haddr)
256 return (0);
257 return (1);
258 }
259
260 /*
261 * Inline function which wraps assertions for a valid ifp.
262 */
263 static __inline__ int
264 inm_is_ifp_detached(const struct in_multi *inm)
265 {
266 VERIFY(inm->inm_ifma != NULL);
267 VERIFY(inm->inm_ifp == inm->inm_ifma->ifma_ifp);
268
269 return (!ifnet_is_attached(inm->inm_ifp, 0));
270 }
271
272 /*
273 * Initialize an in_mfilter structure to a known state at t0, t1
274 * with an empty source filter list.
275 */
276 static __inline__ void
277 imf_init(struct in_mfilter *imf, const int st0, const int st1)
278 {
279 memset(imf, 0, sizeof(struct in_mfilter));
280 RB_INIT(&imf->imf_sources);
281 imf->imf_st[0] = st0;
282 imf->imf_st[1] = st1;
283 }
284
285 /*
286 * Resize the ip_moptions vector to the next power-of-two minus 1.
287 */
288 static int
289 imo_grow(struct ip_moptions *imo, size_t newmax)
290 {
291 struct in_multi **nmships;
292 struct in_multi **omships;
293 struct in_mfilter *nmfilters;
294 struct in_mfilter *omfilters;
295 size_t idx;
296 size_t oldmax;
297
298 IMO_LOCK_ASSERT_HELD(imo);
299
300 nmships = NULL;
301 nmfilters = NULL;
302 omships = imo->imo_membership;
303 omfilters = imo->imo_mfilters;
304 oldmax = imo->imo_max_memberships;
305 if (newmax == 0)
306 newmax = ((oldmax + 1) * 2) - 1;
307
308 if (newmax > IP_MAX_MEMBERSHIPS)
309 return (ETOOMANYREFS);
310
311 if ((nmships = (struct in_multi **)_REALLOC(omships,
312 sizeof (struct in_multi *) * newmax, M_IPMOPTS,
313 M_WAITOK | M_ZERO)) == NULL)
314 return (ENOMEM);
315
316 imo->imo_membership = nmships;
317
318 if ((nmfilters = (struct in_mfilter *)_REALLOC(omfilters,
319 sizeof (struct in_mfilter) * newmax, M_INMFILTER,
320 M_WAITOK | M_ZERO)) == NULL)
321 return (ENOMEM);
322
323 imo->imo_mfilters = nmfilters;
324
325 /* Initialize newly allocated source filter heads. */
326 for (idx = oldmax; idx < newmax; idx++)
327 imf_init(&nmfilters[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
328
329 imo->imo_max_memberships = newmax;
330
331 return (0);
332 }
333
334 /*
335 * Find an IPv4 multicast group entry for this ip_moptions instance
336 * which matches the specified group, and optionally an interface.
337 * Return its index into the array, or -1 if not found.
338 */
339 static size_t
340 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
341 const struct sockaddr *group)
342 {
343 const struct sockaddr_in *gsin;
344 struct in_multi *pinm;
345 int idx;
346 int nmships;
347
348 IMO_LOCK_ASSERT_HELD(IMO_CAST_TO_NONCONST(imo));
349
350 gsin = (const struct sockaddr_in *)group;
351
352 /* The imo_membership array may be lazy allocated. */
353 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
354 return (-1);
355
356 nmships = imo->imo_num_memberships;
357 for (idx = 0; idx < nmships; idx++) {
358 pinm = imo->imo_membership[idx];
359 if (pinm == NULL)
360 continue;
361 INM_LOCK(pinm);
362 if ((ifp == NULL || (pinm->inm_ifp == ifp)) &&
363 in_hosteq(pinm->inm_addr, gsin->sin_addr)) {
364 INM_UNLOCK(pinm);
365 break;
366 }
367 INM_UNLOCK(pinm);
368 }
369 if (idx >= nmships)
370 idx = -1;
371
372 return (idx);
373 }
374
375 /*
376 * Find an IPv4 multicast source entry for this imo which matches
377 * the given group index for this socket, and source address.
378 *
379 * NOTE: This does not check if the entry is in-mode, merely if
380 * it exists, which may not be the desired behaviour.
381 */
382 static struct in_msource *
383 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
384 const struct sockaddr *src)
385 {
386 struct ip_msource find;
387 struct in_mfilter *imf;
388 struct ip_msource *ims;
389 const sockunion_t *psa;
390
391 IMO_LOCK_ASSERT_HELD(IMO_CAST_TO_NONCONST(imo));
392
393 VERIFY(src->sa_family == AF_INET);
394 VERIFY(gidx != (size_t)-1 && gidx < imo->imo_num_memberships);
395
396 /* The imo_mfilters array may be lazy allocated. */
397 if (imo->imo_mfilters == NULL)
398 return (NULL);
399 imf = &imo->imo_mfilters[gidx];
400
401 /* Source trees are keyed in host byte order. */
402 psa = (const sockunion_t *)src;
403 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
404 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
405
406 return ((struct in_msource *)ims);
407 }
408
409 /*
410 * Perform filtering for multicast datagrams on a socket by group and source.
411 *
412 * Returns 0 if a datagram should be allowed through, or various error codes
413 * if the socket was not a member of the group, or the source was muted, etc.
414 */
415 int
416 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
417 const struct sockaddr *group, const struct sockaddr *src)
418 {
419 size_t gidx;
420 struct in_msource *ims;
421 int mode;
422
423 IMO_LOCK_ASSERT_HELD(IMO_CAST_TO_NONCONST(imo));
424 VERIFY(ifp != NULL);
425
426 gidx = imo_match_group(imo, ifp, group);
427 if (gidx == (size_t)-1)
428 return (MCAST_NOTGMEMBER);
429
430 /*
431 * Check if the source was included in an (S,G) join.
432 * Allow reception on exclusive memberships by default,
433 * reject reception on inclusive memberships by default.
434 * Exclude source only if an in-mode exclude filter exists.
435 * Include source only if an in-mode include filter exists.
436 * NOTE: We are comparing group state here at IGMP t1 (now)
437 * with socket-layer t0 (since last downcall).
438 */
439 mode = imo->imo_mfilters[gidx].imf_st[1];
440 ims = imo_match_source(imo, gidx, src);
441
442 if ((ims == NULL && mode == MCAST_INCLUDE) ||
443 (ims != NULL && ims->imsl_st[0] != mode)) {
444 return (MCAST_NOTSMEMBER);
445 }
446
447 return (MCAST_PASS);
448 }
449
450 int
451 imo_clone(struct ip_moptions *from, struct ip_moptions *to)
452 {
453 int i, err = 0;
454
455 IMO_LOCK(from);
456 IMO_LOCK(to);
457
458 to->imo_multicast_ifp = from->imo_multicast_ifp;
459 to->imo_multicast_vif = from->imo_multicast_vif;
460 to->imo_multicast_ttl = from->imo_multicast_ttl;
461 to->imo_multicast_loop = from->imo_multicast_loop;
462
463 /*
464 * We're cloning, so drop any existing memberships and source
465 * filters on the destination ip_moptions.
466 */
467 for (i = 0; i < to->imo_num_memberships; ++i) {
468 struct in_mfilter *imf;
469
470 imf = to->imo_mfilters ? &to->imo_mfilters[i] : NULL;
471 if (imf != NULL)
472 imf_leave(imf);
473
474 (void) in_leavegroup(to->imo_membership[i], imf);
475
476 if (imf != NULL)
477 imf_purge(imf);
478
479 INM_REMREF(to->imo_membership[i]);
480 to->imo_membership[i] = NULL;
481 }
482 to->imo_num_memberships = 0;
483
484 VERIFY(to->imo_max_memberships != 0 && from->imo_max_memberships != 0);
485 if (to->imo_max_memberships < from->imo_max_memberships) {
486 /*
487 * Ensure source and destination ip_moptions memberships
488 * and source filters arrays are at least equal in size.
489 */
490 err = imo_grow(to, from->imo_max_memberships);
491 if (err != 0)
492 goto done;
493 }
494 VERIFY(to->imo_max_memberships >= from->imo_max_memberships);
495
496 /*
497 * Source filtering doesn't apply to OpenTransport socket,
498 * so simply hold additional reference count per membership.
499 */
500 for (i = 0; i < from->imo_num_memberships; i++) {
501 to->imo_membership[i] = from->imo_membership[i];
502 INM_ADDREF(from->imo_membership[i]);
503 to->imo_num_memberships++;
504 }
505 VERIFY(to->imo_num_memberships == from->imo_num_memberships);
506
507 done:
508 IMO_UNLOCK(to);
509 IMO_UNLOCK(from);
510
511 return (err);
512 }
513
514 /*
515 * Find and return a reference to an in_multi record for (ifp, group),
516 * and bump its reference count.
517 * If one does not exist, try to allocate it, and update link-layer multicast
518 * filters on ifp to listen for group.
519 * Return 0 if successful, otherwise return an appropriate error code.
520 */
521 static int
522 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
523 struct in_multi **pinm)
524 {
525 struct sockaddr_in gsin;
526 struct ifmultiaddr *ifma;
527 struct in_multi *inm;
528 int error;
529
530 in_multihead_lock_shared();
531 IN_LOOKUP_MULTI(group, ifp, inm);
532 if (inm != NULL) {
533 INM_LOCK(inm);
534 VERIFY(inm->inm_reqcnt >= 1);
535 inm->inm_reqcnt++;
536 VERIFY(inm->inm_reqcnt != 0);
537 *pinm = inm;
538 INM_UNLOCK(inm);
539 in_multihead_lock_done();
540 /*
541 * We already joined this group; return the inm
542 * with a refcount held (via lookup) for caller.
543 */
544 return (0);
545 }
546 in_multihead_lock_done();
547
548 bzero(&gsin, sizeof(gsin));
549 gsin.sin_family = AF_INET;
550 gsin.sin_len = sizeof(struct sockaddr_in);
551 gsin.sin_addr = *group;
552
553 /*
554 * Check if a link-layer group is already associated
555 * with this network-layer group on the given ifnet.
556 */
557 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
558 if (error != 0)
559 return (error);
560
561 /*
562 * See comments in inm_remref() for access to ifma_protospec.
563 */
564 in_multihead_lock_exclusive();
565 IFMA_LOCK(ifma);
566 if ((inm = ifma->ifma_protospec) != NULL) {
567 VERIFY(ifma->ifma_addr != NULL);
568 VERIFY(ifma->ifma_addr->sa_family == AF_INET);
569 INM_ADDREF(inm); /* for caller */
570 IFMA_UNLOCK(ifma);
571 INM_LOCK(inm);
572 VERIFY(inm->inm_ifma == ifma);
573 VERIFY(inm->inm_ifp == ifp);
574 VERIFY(in_hosteq(inm->inm_addr, *group));
575 if (inm->inm_debug & IFD_ATTACHED) {
576 VERIFY(inm->inm_reqcnt >= 1);
577 inm->inm_reqcnt++;
578 VERIFY(inm->inm_reqcnt != 0);
579 *pinm = inm;
580 INM_UNLOCK(inm);
581 in_multihead_lock_done();
582 IFMA_REMREF(ifma);
583 /*
584 * We lost the race with another thread doing
585 * in_getmulti(); since this group has already
586 * been joined; return the inm with a refcount
587 * held for caller.
588 */
589 return (0);
590 }
591 /*
592 * We lost the race with another thread doing in_delmulti();
593 * the inm referring to the ifma has been detached, thus we
594 * reattach it back to the in_multihead list and return the
595 * inm with a refcount held for the caller.
596 */
597 in_multi_attach(inm);
598 VERIFY((inm->inm_debug &
599 (IFD_ATTACHED | IFD_TRASHED)) == IFD_ATTACHED);
600 *pinm = inm;
601 INM_UNLOCK(inm);
602 in_multihead_lock_done();
603 IFMA_REMREF(ifma);
604 return (0);
605 }
606 IFMA_UNLOCK(ifma);
607
608 /*
609 * A new in_multi record is needed; allocate and initialize it.
610 * We DO NOT perform an IGMP join as the in_ layer may need to
611 * push an initial source list down to IGMP to support SSM.
612 *
613 * The initial source filter state is INCLUDE, {} as per the RFC.
614 */
615 inm = in_multi_alloc(M_WAITOK);
616 if (inm == NULL) {
617 in_multihead_lock_done();
618 IFMA_REMREF(ifma);
619 return (ENOMEM);
620 }
621 INM_LOCK(inm);
622 inm->inm_addr = *group;
623 inm->inm_ifp = ifp;
624 inm->inm_igi = IGMP_IFINFO(ifp);
625 VERIFY(inm->inm_igi != NULL);
626 IGI_ADDREF(inm->inm_igi);
627 inm->inm_ifma = ifma; /* keep refcount from if_addmulti() */
628 inm->inm_state = IGMP_NOT_MEMBER;
629 /*
630 * Pending state-changes per group are subject to a bounds check.
631 */
632 inm->inm_scq.ifq_maxlen = IGMP_MAX_STATE_CHANGES;
633 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
634 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
635 RB_INIT(&inm->inm_srcs);
636 *pinm = inm;
637 in_multi_attach(inm);
638 VERIFY((inm->inm_debug & (IFD_ATTACHED | IFD_TRASHED)) == IFD_ATTACHED);
639 INM_ADDREF_LOCKED(inm); /* for caller */
640 INM_UNLOCK(inm);
641
642 IFMA_LOCK(ifma);
643 VERIFY(ifma->ifma_protospec == NULL);
644 ifma->ifma_protospec = inm;
645 IFMA_UNLOCK(ifma);
646 in_multihead_lock_done();
647
648 return (0);
649 }
650
651 /*
652 * Clear recorded source entries for a group.
653 * Used by the IGMP code.
654 * FIXME: Should reap.
655 */
656 void
657 inm_clear_recorded(struct in_multi *inm)
658 {
659 struct ip_msource *ims;
660
661 INM_LOCK_ASSERT_HELD(inm);
662
663 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
664 if (ims->ims_stp) {
665 ims->ims_stp = 0;
666 --inm->inm_st[1].iss_rec;
667 }
668 }
669 VERIFY(inm->inm_st[1].iss_rec == 0);
670 }
671
672 /*
673 * Record a source as pending for a Source-Group IGMPv3 query.
674 * This lives here as it modifies the shared tree.
675 *
676 * inm is the group descriptor.
677 * naddr is the address of the source to record in network-byte order.
678 *
679 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
680 * lazy-allocate a source node in response to an SG query.
681 * Otherwise, no allocation is performed. This saves some memory
682 * with the trade-off that the source will not be reported to the
683 * router if joined in the window between the query response and
684 * the group actually being joined on the local host.
685 *
686 * Return 0 if the source didn't exist or was already marked as recorded.
687 * Return 1 if the source was marked as recorded by this function.
688 * Return <0 if any error occured (negated errno code).
689 */
690 int
691 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
692 {
693 struct ip_msource find;
694 struct ip_msource *ims, *nims;
695
696 INM_LOCK_ASSERT_HELD(inm);
697
698 find.ims_haddr = ntohl(naddr);
699 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
700 if (ims && ims->ims_stp)
701 return (0);
702 if (ims == NULL) {
703 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
704 return (-ENOSPC);
705 nims = ipms_alloc(M_WAITOK);
706 if (nims == NULL)
707 return (-ENOMEM);
708 nims->ims_haddr = find.ims_haddr;
709 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
710 ++inm->inm_nsrc;
711 ims = nims;
712 }
713
714 /*
715 * Mark the source as recorded and update the recorded
716 * source count.
717 */
718 ++ims->ims_stp;
719 ++inm->inm_st[1].iss_rec;
720
721 return (1);
722 }
723
724 /*
725 * Return a pointer to an in_msource owned by an in_mfilter,
726 * given its source address.
727 * Lazy-allocate if needed. If this is a new entry its filter state is
728 * undefined at t0.
729 *
730 * imf is the filter set being modified.
731 * haddr is the source address in *host* byte-order.
732 *
733 * Caller is expected to be holding imo_lock.
734 */
735 static int
736 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
737 struct in_msource **plims)
738 {
739 struct ip_msource find;
740 struct ip_msource *ims;
741 struct in_msource *lims;
742 int error;
743
744 error = 0;
745 ims = NULL;
746 lims = NULL;
747
748 /* key is host byte order */
749 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
750 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
751 lims = (struct in_msource *)ims;
752 if (lims == NULL) {
753 if (imf->imf_nsrc == in_mcast_maxsocksrc)
754 return (ENOSPC);
755 lims = inms_alloc(M_WAITOK);
756 if (lims == NULL)
757 return (ENOMEM);
758 lims->ims_haddr = find.ims_haddr;
759 lims->imsl_st[0] = MCAST_UNDEFINED;
760 RB_INSERT(ip_msource_tree, &imf->imf_sources,
761 (struct ip_msource *)lims);
762 ++imf->imf_nsrc;
763 }
764
765 *plims = lims;
766
767 return (error);
768 }
769
770 /*
771 * Graft a source entry into an existing socket-layer filter set,
772 * maintaining any required invariants and checking allocations.
773 *
774 * The source is marked as being in the new filter mode at t1.
775 *
776 * Return the pointer to the new node, otherwise return NULL.
777 *
778 * Caller is expected to be holding imo_lock.
779 */
780 static struct in_msource *
781 imf_graft(struct in_mfilter *imf, const uint8_t st1,
782 const struct sockaddr_in *psin)
783 {
784 struct in_msource *lims;
785
786 lims = inms_alloc(M_WAITOK);
787 if (lims == NULL)
788 return (NULL);
789 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
790 lims->imsl_st[0] = MCAST_UNDEFINED;
791 lims->imsl_st[1] = st1;
792 RB_INSERT(ip_msource_tree, &imf->imf_sources,
793 (struct ip_msource *)lims);
794 ++imf->imf_nsrc;
795
796 return (lims);
797 }
798
799 /*
800 * Prune a source entry from an existing socket-layer filter set,
801 * maintaining any required invariants and checking allocations.
802 *
803 * The source is marked as being left at t1, it is not freed.
804 *
805 * Return 0 if no error occurred, otherwise return an errno value.
806 *
807 * Caller is expected to be holding imo_lock.
808 */
809 static int
810 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
811 {
812 struct ip_msource find;
813 struct ip_msource *ims;
814 struct in_msource *lims;
815
816 /* key is host byte order */
817 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
818 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
819 if (ims == NULL)
820 return (ENOENT);
821 lims = (struct in_msource *)ims;
822 lims->imsl_st[1] = MCAST_UNDEFINED;
823 return (0);
824 }
825
826 /*
827 * Revert socket-layer filter set deltas at t1 to t0 state.
828 *
829 * Caller is expected to be holding imo_lock.
830 */
831 static void
832 imf_rollback(struct in_mfilter *imf)
833 {
834 struct ip_msource *ims, *tims;
835 struct in_msource *lims;
836
837 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
838 lims = (struct in_msource *)ims;
839 if (lims->imsl_st[0] == lims->imsl_st[1]) {
840 /* no change at t1 */
841 continue;
842 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
843 /* revert change to existing source at t1 */
844 lims->imsl_st[1] = lims->imsl_st[0];
845 } else {
846 /* revert source added t1 */
847 IGMP_PRINTF(("%s: free inms %p\n", __func__, lims));
848 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
849 inms_free(lims);
850 imf->imf_nsrc--;
851 }
852 }
853 imf->imf_st[1] = imf->imf_st[0];
854 }
855
856 /*
857 * Mark socket-layer filter set as INCLUDE {} at t1.
858 *
859 * Caller is expected to be holding imo_lock.
860 */
861 void
862 imf_leave(struct in_mfilter *imf)
863 {
864 struct ip_msource *ims;
865 struct in_msource *lims;
866
867 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
868 lims = (struct in_msource *)ims;
869 lims->imsl_st[1] = MCAST_UNDEFINED;
870 }
871 imf->imf_st[1] = MCAST_INCLUDE;
872 }
873
874 /*
875 * Mark socket-layer filter set deltas as committed.
876 *
877 * Caller is expected to be holding imo_lock.
878 */
879 static void
880 imf_commit(struct in_mfilter *imf)
881 {
882 struct ip_msource *ims;
883 struct in_msource *lims;
884
885 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
886 lims = (struct in_msource *)ims;
887 lims->imsl_st[0] = lims->imsl_st[1];
888 }
889 imf->imf_st[0] = imf->imf_st[1];
890 }
891
892 /*
893 * Reap unreferenced sources from socket-layer filter set.
894 *
895 * Caller is expected to be holding imo_lock.
896 */
897 static void
898 imf_reap(struct in_mfilter *imf)
899 {
900 struct ip_msource *ims, *tims;
901 struct in_msource *lims;
902
903 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
904 lims = (struct in_msource *)ims;
905 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
906 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
907 IGMP_PRINTF(("%s: free inms %p\n", __func__, lims));
908 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
909 inms_free(lims);
910 imf->imf_nsrc--;
911 }
912 }
913 }
914
915 /*
916 * Purge socket-layer filter set.
917 *
918 * Caller is expected to be holding imo_lock.
919 */
920 void
921 imf_purge(struct in_mfilter *imf)
922 {
923 struct ip_msource *ims, *tims;
924 struct in_msource *lims;
925
926 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
927 lims = (struct in_msource *)ims;
928 IGMP_PRINTF(("%s: free inms %p\n", __func__, lims));
929 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
930 inms_free(lims);
931 imf->imf_nsrc--;
932 }
933 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
934 VERIFY(RB_EMPTY(&imf->imf_sources));
935 }
936
937 /*
938 * Look up a source filter entry for a multicast group.
939 *
940 * inm is the group descriptor to work with.
941 * haddr is the host-byte-order IPv4 address to look up.
942 * noalloc may be non-zero to suppress allocation of sources.
943 * *pims will be set to the address of the retrieved or allocated source.
944 *
945 * Return 0 if successful, otherwise return a non-zero error code.
946 */
947 static int
948 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
949 const int noalloc, struct ip_msource **pims)
950 {
951 struct ip_msource find;
952 struct ip_msource *ims, *nims;
953 #ifdef IGMP_DEBUG
954 struct in_addr ia;
955 #endif
956 INM_LOCK_ASSERT_HELD(inm);
957
958 find.ims_haddr = haddr;
959 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
960 if (ims == NULL && !noalloc) {
961 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
962 return (ENOSPC);
963 nims = ipms_alloc(M_WAITOK);
964 if (nims == NULL)
965 return (ENOMEM);
966 nims->ims_haddr = haddr;
967 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
968 ++inm->inm_nsrc;
969 ims = nims;
970 #ifdef IGMP_DEBUG
971 ia.s_addr = htonl(haddr);
972 IGMP_PRINTF(("%s: allocated %s as %p\n", __func__,
973 inet_ntoa(ia), ims));
974 #endif
975 }
976
977 *pims = ims;
978 return (0);
979 }
980
981 /*
982 * Helper function to derive the filter mode on a source entry
983 * from its internal counters. Predicates are:
984 * A source is only excluded if all listeners exclude it.
985 * A source is only included if no listeners exclude it,
986 * and at least one listener includes it.
987 * May be used by ifmcstat(8).
988 */
989 uint8_t
990 ims_get_mode(const struct in_multi *inm, const struct ip_msource *ims,
991 uint8_t t)
992 {
993 INM_LOCK_ASSERT_HELD(INM_CAST_TO_NONCONST(inm));
994
995 t = !!t;
996 if (inm->inm_st[t].iss_ex > 0 &&
997 inm->inm_st[t].iss_ex == ims->ims_st[t].ex)
998 return (MCAST_EXCLUDE);
999 else if (ims->ims_st[t].in > 0 && ims->ims_st[t].ex == 0)
1000 return (MCAST_INCLUDE);
1001 return (MCAST_UNDEFINED);
1002 }
1003
1004 /*
1005 * Merge socket-layer source into IGMP-layer source.
1006 * If rollback is non-zero, perform the inverse of the merge.
1007 */
1008 static void
1009 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
1010 const int rollback)
1011 {
1012 int n = rollback ? -1 : 1;
1013 #ifdef IGMP_DEBUG
1014 struct in_addr ia;
1015
1016 ia.s_addr = htonl(ims->ims_haddr);
1017 #endif
1018
1019 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
1020 IGMP_PRINTF(("%s: t1 ex -= %d on %s\n",
1021 __func__, n, inet_ntoa(ia)));
1022 ims->ims_st[1].ex -= n;
1023 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
1024 IGMP_PRINTF(("%s: t1 in -= %d on %s\n",
1025 __func__, n, inet_ntoa(ia)));
1026 ims->ims_st[1].in -= n;
1027 }
1028
1029 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
1030 IGMP_PRINTF(("%s: t1 ex += %d on %s\n",
1031 __func__, n, inet_ntoa(ia)));
1032 ims->ims_st[1].ex += n;
1033 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
1034 IGMP_PRINTF(("%s: t1 in += %d on %s\n",
1035 __func__, n, inet_ntoa(ia)));
1036 ims->ims_st[1].in += n;
1037 }
1038 }
1039
1040 /*
1041 * Atomically update the global in_multi state, when a membership's
1042 * filter list is being updated in any way.
1043 *
1044 * imf is the per-inpcb-membership group filter pointer.
1045 * A fake imf may be passed for in-kernel consumers.
1046 *
1047 * XXX This is a candidate for a set-symmetric-difference style loop
1048 * which would eliminate the repeated lookup from root of ims nodes,
1049 * as they share the same key space.
1050 *
1051 * If any error occurred this function will back out of refcounts
1052 * and return a non-zero value.
1053 */
1054 static int
1055 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1056 {
1057 struct ip_msource *ims, *nims;
1058 struct in_msource *lims;
1059 int schanged, error;
1060 int nsrc0, nsrc1;
1061
1062 INM_LOCK_ASSERT_HELD(inm);
1063
1064 schanged = 0;
1065 error = 0;
1066 nsrc1 = nsrc0 = 0;
1067
1068 /*
1069 * Update the source filters first, as this may fail.
1070 * Maintain count of in-mode filters at t0, t1. These are
1071 * used to work out if we transition into ASM mode or not.
1072 * Maintain a count of source filters whose state was
1073 * actually modified by this operation.
1074 */
1075 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1076 lims = (struct in_msource *)ims;
1077 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1078 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1079 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1080 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1081 ++schanged;
1082 if (error)
1083 break;
1084 ims_merge(nims, lims, 0);
1085 }
1086 if (error) {
1087 struct ip_msource *bims;
1088
1089 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1090 lims = (struct in_msource *)ims;
1091 if (lims->imsl_st[0] == lims->imsl_st[1])
1092 continue;
1093 (void) inm_get_source(inm, lims->ims_haddr, 1, &bims);
1094 if (bims == NULL)
1095 continue;
1096 ims_merge(bims, lims, 1);
1097 }
1098 goto out_reap;
1099 }
1100
1101 IGMP_PRINTF(("%s: imf filters in-mode: %d at t0, %d at t1\n",
1102 __func__, nsrc0, nsrc1));
1103
1104 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1105 if (imf->imf_st[0] == imf->imf_st[1] &&
1106 imf->imf_st[1] == MCAST_INCLUDE) {
1107 if (nsrc1 == 0) {
1108 IGMP_PRINTF(("%s: --in on inm at t1\n", __func__));
1109 --inm->inm_st[1].iss_in;
1110 }
1111 }
1112
1113 /* Handle filter mode transition on socket. */
1114 if (imf->imf_st[0] != imf->imf_st[1]) {
1115 IGMP_PRINTF(("%s: imf transition %d to %d\n",
1116 __func__, imf->imf_st[0], imf->imf_st[1]));
1117
1118 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1119 IGMP_PRINTF(("%s: --ex on inm at t1\n", __func__));
1120 --inm->inm_st[1].iss_ex;
1121 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1122 IGMP_PRINTF(("%s: --in on inm at t1\n", __func__));
1123 --inm->inm_st[1].iss_in;
1124 }
1125
1126 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1127 IGMP_PRINTF(("%s: ex++ on inm at t1\n", __func__));
1128 inm->inm_st[1].iss_ex++;
1129 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1130 IGMP_PRINTF(("%s: in++ on inm at t1\n", __func__));
1131 inm->inm_st[1].iss_in++;
1132 }
1133 }
1134
1135 /*
1136 * Track inm filter state in terms of listener counts.
1137 * If there are any exclusive listeners, stack-wide
1138 * membership is exclusive.
1139 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1140 * If no listeners remain, state is undefined at t1,
1141 * and the IGMP lifecycle for this group should finish.
1142 */
1143 if (inm->inm_st[1].iss_ex > 0) {
1144 IGMP_PRINTF(("%s: transition to EX\n", __func__));
1145 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1146 } else if (inm->inm_st[1].iss_in > 0) {
1147 IGMP_PRINTF(("%s: transition to IN\n", __func__));
1148 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1149 } else {
1150 IGMP_PRINTF(("%s: transition to UNDEF\n", __func__));
1151 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1152 }
1153
1154 /* Decrement ASM listener count on transition out of ASM mode. */
1155 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1156 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1157 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1158 IGMP_PRINTF(("%s: --asm on inm at t1\n", __func__));
1159 --inm->inm_st[1].iss_asm;
1160 }
1161 }
1162
1163 /* Increment ASM listener count on transition to ASM mode. */
1164 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1165 IGMP_PRINTF(("%s: asm++ on inm at t1\n", __func__));
1166 inm->inm_st[1].iss_asm++;
1167 }
1168
1169 IGMP_PRINTF(("%s: merged imf %p to inm %p\n", __func__, imf, inm));
1170 inm_print(inm);
1171
1172 out_reap:
1173 if (schanged > 0) {
1174 IGMP_PRINTF(("%s: sources changed; reaping\n", __func__));
1175 inm_reap(inm);
1176 }
1177 return (error);
1178 }
1179
1180 /*
1181 * Mark an in_multi's filter set deltas as committed.
1182 * Called by IGMP after a state change has been enqueued.
1183 */
1184 void
1185 inm_commit(struct in_multi *inm)
1186 {
1187 struct ip_msource *ims;
1188
1189 INM_LOCK_ASSERT_HELD(inm);
1190
1191 IGMP_PRINTF(("%s: commit inm %p\n", __func__, inm));
1192 IGMP_PRINTF(("%s: pre commit:\n", __func__));
1193 inm_print(inm);
1194
1195 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
1196 ims->ims_st[0] = ims->ims_st[1];
1197 }
1198 inm->inm_st[0] = inm->inm_st[1];
1199 }
1200
1201 /*
1202 * Reap unreferenced nodes from an in_multi's filter set.
1203 */
1204 static void
1205 inm_reap(struct in_multi *inm)
1206 {
1207 struct ip_msource *ims, *tims;
1208
1209 INM_LOCK_ASSERT_HELD(inm);
1210
1211 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1212 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1213 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1214 ims->ims_stp != 0)
1215 continue;
1216 IGMP_PRINTF(("%s: free ims %p\n", __func__, ims));
1217 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1218 ipms_free(ims);
1219 inm->inm_nsrc--;
1220 }
1221 }
1222
1223 /*
1224 * Purge all source nodes from an in_multi's filter set.
1225 */
1226 void
1227 inm_purge(struct in_multi *inm)
1228 {
1229 struct ip_msource *ims, *tims;
1230
1231 INM_LOCK_ASSERT_HELD(inm);
1232
1233 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1234 IGMP_PRINTF(("%s: free ims %p\n", __func__, ims));
1235 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1236 ipms_free(ims);
1237 inm->inm_nsrc--;
1238 }
1239 }
1240
1241 /*
1242 * Join a multicast group; real entry point.
1243 *
1244 * Only preserves atomicity at inm level.
1245 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1246 *
1247 * If the IGMP downcall fails, the group is not joined, and an error
1248 * code is returned.
1249 */
1250 static int
1251 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1252 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1253 {
1254 struct in_mfilter timf;
1255 struct in_multi *inm = NULL;
1256 int error = 0;
1257
1258 IGMP_PRINTF(("%s: join %s on %p(%s%d))\n", __func__,
1259 inet_ntoa(*gina), ifp, ifp->if_name, ifp->if_unit));
1260
1261 *pinm = NULL;
1262
1263 /*
1264 * If no imf was specified (i.e. kernel consumer),
1265 * fake one up and assume it is an ASM join.
1266 */
1267 if (imf == NULL) {
1268 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1269 imf = &timf;
1270 }
1271
1272 error = in_getmulti(ifp, gina, &inm);
1273 if (error) {
1274 IGMP_PRINTF(("%s: in_getmulti() failure\n", __func__));
1275 return (error);
1276 }
1277
1278 IGMP_PRINTF(("%s: merge inm state\n", __func__));
1279
1280 INM_LOCK(inm);
1281 error = inm_merge(inm, imf);
1282 if (error) {
1283 IGMP_PRINTF(("%s: failed to merge inm state\n", __func__));
1284 goto out_inm_release;
1285 }
1286
1287 IGMP_PRINTF(("%s: doing igmp downcall\n", __func__));
1288 error = igmp_change_state(inm);
1289 if (error) {
1290 IGMP_PRINTF(("%s: failed to update source\n", __func__));
1291 goto out_inm_release;
1292 }
1293
1294 out_inm_release:
1295 if (error) {
1296 IGMP_PRINTF(("%s: dropping ref on %p\n", __func__, inm));
1297 INM_UNLOCK(inm);
1298 INM_REMREF(inm);
1299 } else {
1300 INM_UNLOCK(inm);
1301 *pinm = inm; /* keep refcount from in_getmulti() */
1302 }
1303
1304 return (error);
1305 }
1306
1307 /*
1308 * Leave a multicast group; real entry point.
1309 * All source filters will be expunged.
1310 *
1311 * Only preserves atomicity at inm level.
1312 *
1313 * Note: This is not the same as inm_release(*) as this function also
1314 * makes a state change downcall into IGMP.
1315 */
1316 int
1317 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1318 {
1319 struct in_mfilter timf;
1320 int error, lastref;
1321
1322 error = 0;
1323
1324 INM_LOCK_ASSERT_NOTHELD(inm);
1325
1326 in_multihead_lock_exclusive();
1327 INM_LOCK(inm);
1328
1329 IGMP_PRINTF(("%s: leave inm %p, %s/%s%d, imf %p\n", __func__,
1330 inm, inet_ntoa(inm->inm_addr),
1331 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_name),
1332 inm->inm_ifp->if_unit, imf));
1333
1334 /*
1335 * If no imf was specified (i.e. kernel consumer),
1336 * fake one up and assume it is an ASM join.
1337 */
1338 if (imf == NULL) {
1339 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1340 imf = &timf;
1341 }
1342
1343 /*
1344 * Begin state merge transaction at IGMP layer.
1345 *
1346 * As this particular invocation should not cause any memory
1347 * to be allocated, and there is no opportunity to roll back
1348 * the transaction, it MUST NOT fail.
1349 */
1350 IGMP_PRINTF(("%s: merge inm state\n", __func__));
1351
1352 error = inm_merge(inm, imf);
1353 KASSERT(error == 0, ("%s: failed to merge inm state\n", __func__));
1354
1355 IGMP_PRINTF(("%s: doing igmp downcall\n", __func__));
1356 error = igmp_change_state(inm);
1357 #if IGMP_DEBUG
1358 if (error)
1359 IGMP_PRINTF(("%s: failed igmp downcall\n", __func__));
1360 #endif
1361 lastref = in_multi_detach(inm);
1362 VERIFY(!lastref || (!(inm->inm_debug & IFD_ATTACHED) &&
1363 inm->inm_reqcnt == 0));
1364 INM_UNLOCK(inm);
1365 in_multihead_lock_done();
1366
1367 if (lastref)
1368 INM_REMREF(inm); /* for in_multihead list */
1369
1370 return (error);
1371 }
1372
1373 /*
1374 * Join an IPv4 multicast group in (*,G) exclusive mode.
1375 * The group must be a 224.0.0.0/24 link-scope group.
1376 * This KPI is for legacy kernel consumers only.
1377 */
1378 struct in_multi *
1379 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1380 {
1381 struct in_multi *pinm = NULL;
1382 int error;
1383
1384 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1385 ("%s: %s not in 224.0.0.0/24\n", __func__, inet_ntoa(*ap)));
1386
1387 error = in_joingroup(ifp, ap, NULL, &pinm);
1388 VERIFY(pinm != NULL || error != 0);
1389
1390 return (pinm);
1391 }
1392
1393 /*
1394 * Leave an IPv4 multicast group, assumed to be in exclusive (*,G) mode.
1395 * This KPI is for legacy kernel consumers only.
1396 */
1397 void
1398 in_delmulti(struct in_multi *inm)
1399 {
1400
1401 (void) in_leavegroup(inm, NULL);
1402 }
1403
1404 /*
1405 * Block or unblock an ASM multicast source on an inpcb.
1406 * This implements the delta-based API described in RFC 3678.
1407 *
1408 * The delta-based API applies only to exclusive-mode memberships.
1409 * An IGMP downcall will be performed.
1410 *
1411 * Return 0 if successful, otherwise return an appropriate error code.
1412 */
1413 static int
1414 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1415 {
1416 struct group_source_req gsr;
1417 sockunion_t *gsa, *ssa;
1418 struct ifnet *ifp;
1419 struct in_mfilter *imf;
1420 struct ip_moptions *imo;
1421 struct in_msource *ims;
1422 struct in_multi *inm;
1423 size_t idx;
1424 uint16_t fmode;
1425 int error, doblock;
1426 unsigned int ifindex = 0;
1427
1428 ifp = NULL;
1429 error = 0;
1430 doblock = 0;
1431
1432 memset(&gsr, 0, sizeof(struct group_source_req));
1433 gsa = (sockunion_t *)&gsr.gsr_group;
1434 ssa = (sockunion_t *)&gsr.gsr_source;
1435
1436 switch (sopt->sopt_name) {
1437 case IP_BLOCK_SOURCE:
1438 case IP_UNBLOCK_SOURCE: {
1439 struct ip_mreq_source mreqs;
1440
1441 error = sooptcopyin(sopt, &mreqs,
1442 sizeof(struct ip_mreq_source),
1443 sizeof(struct ip_mreq_source));
1444 if (error)
1445 return (error);
1446
1447 gsa->sin.sin_family = AF_INET;
1448 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1449 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1450
1451 ssa->sin.sin_family = AF_INET;
1452 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1453 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1454
1455 if (!in_nullhost(mreqs.imr_interface))
1456 ifp = ip_multicast_if(&mreqs.imr_interface, &ifindex);
1457
1458 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1459 doblock = 1;
1460
1461 IGMP_PRINTF(("%s: imr_interface = %s, ifp = %p\n",
1462 __func__, inet_ntoa(mreqs.imr_interface), ifp));
1463 break;
1464 }
1465
1466 case MCAST_BLOCK_SOURCE:
1467 case MCAST_UNBLOCK_SOURCE:
1468 error = sooptcopyin(sopt, &gsr,
1469 sizeof(struct group_source_req),
1470 sizeof(struct group_source_req));
1471 if (error)
1472 return (error);
1473
1474 if (gsa->sin.sin_family != AF_INET ||
1475 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1476 return (EINVAL);
1477
1478 if (ssa->sin.sin_family != AF_INET ||
1479 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1480 return (EINVAL);
1481
1482 ifnet_head_lock_shared();
1483 if (gsr.gsr_interface == 0 ||
1484 (u_int)if_index < gsr.gsr_interface) {
1485 ifnet_head_done();
1486 return (EADDRNOTAVAIL);
1487 }
1488
1489 ifp = ifindex2ifnet[gsr.gsr_interface];
1490 ifnet_head_done();
1491
1492 if (ifp == NULL)
1493 return (EADDRNOTAVAIL);
1494
1495 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1496 doblock = 1;
1497 break;
1498
1499 default:
1500 IGMP_PRINTF(("%s: unknown sopt_name %d\n",
1501 __func__, sopt->sopt_name));
1502 return (EOPNOTSUPP);
1503 break;
1504 }
1505
1506 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1507 return (EINVAL);
1508
1509 /*
1510 * Check if we are actually a member of this group.
1511 */
1512 imo = inp_findmoptions(inp);
1513 if (imo == NULL)
1514 return (ENOMEM);
1515
1516 IMO_LOCK(imo);
1517 idx = imo_match_group(imo, ifp, &gsa->sa);
1518 if (idx == (size_t)-1 || imo->imo_mfilters == NULL) {
1519 error = EADDRNOTAVAIL;
1520 goto out_imo_locked;
1521 }
1522
1523 VERIFY(imo->imo_mfilters != NULL);
1524 imf = &imo->imo_mfilters[idx];
1525 inm = imo->imo_membership[idx];
1526
1527 /*
1528 * Attempting to use the delta-based API on an
1529 * non exclusive-mode membership is an error.
1530 */
1531 fmode = imf->imf_st[0];
1532 if (fmode != MCAST_EXCLUDE) {
1533 error = EINVAL;
1534 goto out_imo_locked;
1535 }
1536
1537 /*
1538 * Deal with error cases up-front:
1539 * Asked to block, but already blocked; or
1540 * Asked to unblock, but nothing to unblock.
1541 * If adding a new block entry, allocate it.
1542 */
1543 ims = imo_match_source(imo, idx, &ssa->sa);
1544 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1545 IGMP_PRINTF(("%s: source %s %spresent\n", __func__,
1546 inet_ntoa(ssa->sin.sin_addr), doblock ? "" : "not "));
1547 error = EADDRNOTAVAIL;
1548 goto out_imo_locked;
1549 }
1550
1551 /*
1552 * Begin state merge transaction at socket layer.
1553 */
1554 if (doblock) {
1555 IGMP_PRINTF(("%s: %s source\n", __func__, "block"));
1556 ims = imf_graft(imf, fmode, &ssa->sin);
1557 if (ims == NULL)
1558 error = ENOMEM;
1559 } else {
1560 IGMP_PRINTF(("%s: %s source\n", __func__, "allow"));
1561 error = imf_prune(imf, &ssa->sin);
1562 }
1563
1564 if (error) {
1565 IGMP_PRINTF(("%s: merge imf state failed\n", __func__));
1566 goto out_imf_rollback;
1567 }
1568
1569 /*
1570 * Begin state merge transaction at IGMP layer.
1571 */
1572 INM_LOCK(inm);
1573 IGMP_PRINTF(("%s: merge inm state\n", __func__));
1574 error = inm_merge(inm, imf);
1575 if (error) {
1576 IGMP_PRINTF(("%s: failed to merge inm state\n", __func__));
1577 INM_UNLOCK(inm);
1578 goto out_imf_rollback;
1579 }
1580
1581 IGMP_PRINTF(("%s: doing igmp downcall\n", __func__));
1582 error = igmp_change_state(inm);
1583 INM_UNLOCK(inm);
1584 #if IGMP_DEBUG
1585 if (error)
1586 IGMP_PRINTF(("%s: failed igmp downcall\n", __func__));
1587 #endif
1588
1589 out_imf_rollback:
1590 if (error)
1591 imf_rollback(imf);
1592 else
1593 imf_commit(imf);
1594
1595 imf_reap(imf);
1596
1597 out_imo_locked:
1598 IMO_UNLOCK(imo);
1599 IMO_REMREF(imo); /* from inp_findmoptions() */
1600 return (error);
1601 }
1602
1603 /*
1604 * Given an inpcb, return its multicast options structure pointer.
1605 *
1606 * Caller is responsible for locking the inpcb, and releasing the
1607 * extra reference held on the imo, upon a successful return.
1608 */
1609 static struct ip_moptions *
1610 inp_findmoptions(struct inpcb *inp)
1611 {
1612 struct ip_moptions *imo;
1613 struct in_multi **immp;
1614 struct in_mfilter *imfp;
1615 size_t idx;
1616
1617 if ((imo = inp->inp_moptions) != NULL) {
1618 IMO_ADDREF(imo); /* for caller */
1619 return (imo);
1620 }
1621
1622 imo = ip_allocmoptions(M_WAITOK);
1623 if (imo == NULL)
1624 return (NULL);
1625
1626 immp = _MALLOC(sizeof (*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1627 M_WAITOK | M_ZERO);
1628 if (immp == NULL) {
1629 IMO_REMREF(imo);
1630 return (NULL);
1631 }
1632
1633 imfp = _MALLOC(sizeof (struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1634 M_INMFILTER, M_WAITOK | M_ZERO);
1635 if (imfp == NULL) {
1636 _FREE(immp, M_IPMOPTS);
1637 IMO_REMREF(imo);
1638 return (NULL);
1639 }
1640
1641 imo->imo_multicast_ifp = NULL;
1642 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1643 imo->imo_multicast_vif = -1;
1644 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1645 imo->imo_multicast_loop = in_mcast_loop;
1646 imo->imo_num_memberships = 0;
1647 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1648 imo->imo_membership = immp;
1649
1650 /* Initialize per-group source filters. */
1651 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1652 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1653
1654 imo->imo_mfilters = imfp;
1655 inp->inp_moptions = imo; /* keep reference from ip_allocmoptions() */
1656 IMO_ADDREF(imo); /* for caller */
1657
1658 return (imo);
1659 }
1660 /*
1661 * Atomically get source filters on a socket for an IPv4 multicast group.
1662 */
1663 static int
1664 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1665 {
1666 struct __msfilterreq64 msfr, msfr64;
1667 struct __msfilterreq32 msfr32;
1668 sockunion_t *gsa;
1669 struct ifnet *ifp;
1670 struct ip_moptions *imo;
1671 struct in_mfilter *imf;
1672 struct ip_msource *ims;
1673 struct in_msource *lims;
1674 struct sockaddr_in *psin;
1675 struct sockaddr_storage *ptss;
1676 struct sockaddr_storage *tss;
1677 int error;
1678 size_t idx, nsrcs, ncsrcs;
1679 user_addr_t tmp_ptr;
1680
1681 imo = inp->inp_moptions;
1682 VERIFY(imo != NULL);
1683
1684 if (IS_64BIT_PROCESS(current_proc())) {
1685 error = sooptcopyin(sopt, &msfr64,
1686 sizeof(struct __msfilterreq64),
1687 sizeof(struct __msfilterreq64));
1688 if (error)
1689 return (error);
1690 /* we never use msfr.msfr_srcs; */
1691 memcpy(&msfr, &msfr64, sizeof(msfr));
1692 } else {
1693 error = sooptcopyin(sopt, &msfr32,
1694 sizeof(struct __msfilterreq32),
1695 sizeof(struct __msfilterreq32));
1696 if (error)
1697 return (error);
1698 /* we never use msfr.msfr_srcs; */
1699 memcpy(&msfr, &msfr32, sizeof(msfr));
1700 }
1701
1702 ifnet_head_lock_shared();
1703 if (msfr.msfr_ifindex == 0 || (u_int)if_index < msfr.msfr_ifindex) {
1704 ifnet_head_done();
1705 return (EADDRNOTAVAIL);
1706 }
1707
1708 ifp = ifindex2ifnet[msfr.msfr_ifindex];
1709 ifnet_head_done();
1710
1711 if (ifp == NULL)
1712 return (EADDRNOTAVAIL);
1713
1714 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1715 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1716
1717 IMO_LOCK(imo);
1718 /*
1719 * Lookup group on the socket.
1720 */
1721 gsa = (sockunion_t *)&msfr.msfr_group;
1722 idx = imo_match_group(imo, ifp, &gsa->sa);
1723 if (idx == (size_t)-1 || imo->imo_mfilters == NULL) {
1724 IMO_UNLOCK(imo);
1725 return (EADDRNOTAVAIL);
1726 }
1727 imf = &imo->imo_mfilters[idx];
1728
1729 /*
1730 * Ignore memberships which are in limbo.
1731 */
1732 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1733 IMO_UNLOCK(imo);
1734 return (EAGAIN);
1735 }
1736 msfr.msfr_fmode = imf->imf_st[1];
1737
1738 /*
1739 * If the user specified a buffer, copy out the source filter
1740 * entries to userland gracefully.
1741 * We only copy out the number of entries which userland
1742 * has asked for, but we always tell userland how big the
1743 * buffer really needs to be.
1744 */
1745
1746 if (IS_64BIT_PROCESS(current_proc()))
1747 tmp_ptr = msfr64.msfr_srcs;
1748 else
1749 tmp_ptr = CAST_USER_ADDR_T(msfr32.msfr_srcs);
1750
1751 tss = NULL;
1752 if (tmp_ptr != USER_ADDR_NULL && msfr.msfr_nsrcs > 0) {
1753 tss = _MALLOC(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1754 M_TEMP, M_WAITOK | M_ZERO);
1755 if (tss == NULL) {
1756 IMO_UNLOCK(imo);
1757 return (ENOBUFS);
1758 }
1759 }
1760
1761 /*
1762 * Count number of sources in-mode at t0.
1763 * If buffer space exists and remains, copy out source entries.
1764 */
1765 nsrcs = msfr.msfr_nsrcs;
1766 ncsrcs = 0;
1767 ptss = tss;
1768 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1769 lims = (struct in_msource *)ims;
1770 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1771 lims->imsl_st[0] != imf->imf_st[0])
1772 continue;
1773 if (tss != NULL && nsrcs > 0) {
1774 psin = (struct sockaddr_in *)ptss;
1775 psin->sin_family = AF_INET;
1776 psin->sin_len = sizeof(struct sockaddr_in);
1777 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1778 psin->sin_port = 0;
1779 ++ptss;
1780 --nsrcs;
1781 ++ncsrcs;
1782 }
1783 }
1784
1785 IMO_UNLOCK(imo);
1786
1787 if (tss != NULL) {
1788 error = copyout(tss, tmp_ptr,
1789 sizeof(struct sockaddr_storage) * ncsrcs);
1790 FREE(tss, M_TEMP);
1791 if (error)
1792 return (error);
1793 }
1794
1795 msfr.msfr_nsrcs = ncsrcs;
1796 if (IS_64BIT_PROCESS(current_proc())) {
1797 msfr64.msfr_ifindex = msfr.msfr_ifindex;
1798 msfr64.msfr_fmode = msfr.msfr_fmode;
1799 msfr64.msfr_nsrcs = msfr.msfr_nsrcs;
1800 memcpy(&msfr64.msfr_group, &msfr.msfr_group,
1801 sizeof(struct sockaddr_storage));
1802 error = sooptcopyout(sopt, &msfr64,
1803 sizeof(struct __msfilterreq64));
1804 } else {
1805 msfr32.msfr_ifindex = msfr.msfr_ifindex;
1806 msfr32.msfr_fmode = msfr.msfr_fmode;
1807 msfr32.msfr_nsrcs = msfr.msfr_nsrcs;
1808 memcpy(&msfr64.msfr_group, &msfr.msfr_group,
1809 sizeof(struct sockaddr_storage));
1810 error = sooptcopyout(sopt, &msfr32,
1811 sizeof(struct __msfilterreq32));
1812 }
1813
1814 return (error);
1815 }
1816
1817 /*
1818 * Return the IP multicast options in response to user getsockopt().
1819 */
1820 int
1821 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1822 {
1823 struct ip_mreqn mreqn;
1824 struct ip_moptions *imo;
1825 struct ifnet *ifp;
1826 struct in_ifaddr *ia;
1827 int error, optval;
1828 unsigned int ifindex;
1829 u_char coptval;
1830
1831 imo = inp->inp_moptions;
1832 /*
1833 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1834 * or is a divert socket, reject it.
1835 */
1836 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1837 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1838 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1839 return (EOPNOTSUPP);
1840 }
1841
1842 error = 0;
1843 switch (sopt->sopt_name) {
1844 #ifdef MROUTING
1845 case IP_MULTICAST_VIF:
1846 if (imo != NULL) {
1847 IMO_LOCK(imo);
1848 optval = imo->imo_multicast_vif;
1849 IMO_UNLOCK(imo);
1850 } else
1851 optval = -1;
1852 error = sooptcopyout(sopt, &optval, sizeof(int));
1853 break;
1854 #endif /* MROUTING */
1855
1856 case IP_MULTICAST_IF:
1857 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1858 if (imo != NULL) {
1859 IMO_LOCK(imo);
1860 ifp = imo->imo_multicast_ifp;
1861 if (!in_nullhost(imo->imo_multicast_addr)) {
1862 mreqn.imr_address = imo->imo_multicast_addr;
1863 } else if (ifp != NULL) {
1864 mreqn.imr_ifindex = ifp->if_index;
1865 IFP_TO_IA(ifp, ia);
1866 if (ia != NULL) {
1867 IFA_LOCK_SPIN(&ia->ia_ifa);
1868 mreqn.imr_address =
1869 IA_SIN(ia)->sin_addr;
1870 IFA_UNLOCK(&ia->ia_ifa);
1871 IFA_REMREF(&ia->ia_ifa);
1872 }
1873 }
1874 IMO_UNLOCK(imo);
1875 }
1876 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1877 error = sooptcopyout(sopt, &mreqn,
1878 sizeof(struct ip_mreqn));
1879 } else {
1880 error = sooptcopyout(sopt, &mreqn.imr_address,
1881 sizeof(struct in_addr));
1882 }
1883 break;
1884
1885 case IP_MULTICAST_IFINDEX:
1886 if (imo != NULL)
1887 IMO_LOCK(imo);
1888 if (imo == NULL || imo->imo_multicast_ifp == NULL) {
1889 ifindex = 0;
1890 } else {
1891 ifindex = imo->imo_multicast_ifp->if_index;
1892 }
1893 if (imo != NULL)
1894 IMO_UNLOCK(imo);
1895 error = sooptcopyout(sopt, &ifindex, sizeof (ifindex));
1896 break;
1897
1898 case IP_MULTICAST_TTL:
1899 if (imo == NULL)
1900 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1901 else {
1902 IMO_LOCK(imo);
1903 optval = coptval = imo->imo_multicast_ttl;
1904 IMO_UNLOCK(imo);
1905 }
1906 if (sopt->sopt_valsize == sizeof(u_char))
1907 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1908 else
1909 error = sooptcopyout(sopt, &optval, sizeof(int));
1910 break;
1911
1912 case IP_MULTICAST_LOOP:
1913 if (imo == 0)
1914 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1915 else {
1916 IMO_LOCK(imo);
1917 optval = coptval = imo->imo_multicast_loop;
1918 IMO_UNLOCK(imo);
1919 }
1920 if (sopt->sopt_valsize == sizeof(u_char))
1921 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1922 else
1923 error = sooptcopyout(sopt, &optval, sizeof(int));
1924 break;
1925
1926 case IP_MSFILTER:
1927 if (imo == NULL) {
1928 error = EADDRNOTAVAIL;
1929 } else {
1930 error = inp_get_source_filters(inp, sopt);
1931 }
1932 break;
1933
1934 default:
1935 error = ENOPROTOOPT;
1936 break;
1937 }
1938
1939 return (error);
1940 }
1941
1942 /*
1943 * Look up the ifnet to use for a multicast group membership,
1944 * given the IPv4 address of an interface, and the IPv4 group address.
1945 *
1946 * This routine exists to support legacy multicast applications
1947 * which do not understand that multicast memberships are scoped to
1948 * specific physical links in the networking stack, or which need
1949 * to join link-scope groups before IPv4 addresses are configured.
1950 *
1951 * If inp is non-NULL and is bound to an interface, use this socket's
1952 * inp_boundif for any required routing table lookup.
1953 *
1954 * If the route lookup fails, attempt to use the first non-loopback
1955 * interface with multicast capability in the system as a
1956 * last resort. The legacy IPv4 ASM API requires that we do
1957 * this in order to allow groups to be joined when the routing
1958 * table has not yet been populated during boot.
1959 *
1960 * Returns NULL if no ifp could be found.
1961 *
1962 */
1963 static struct ifnet *
1964 inp_lookup_mcast_ifp(const struct inpcb *inp,
1965 const struct sockaddr_in *gsin, const struct in_addr ina)
1966 {
1967 struct ifnet *ifp;
1968 unsigned int ifindex = 0;
1969
1970 VERIFY(gsin->sin_family == AF_INET);
1971 VERIFY(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)));
1972
1973 ifp = NULL;
1974 if (!in_nullhost(ina)) {
1975 struct in_addr new_ina;
1976 memcpy(&new_ina, &ina, sizeof(struct in_addr));
1977 ifp = ip_multicast_if(&new_ina, &ifindex);
1978 } else {
1979 struct route ro;
1980 unsigned int ifscope = IFSCOPE_NONE;
1981
1982 if (inp != NULL && (inp->inp_flags & INP_BOUND_IF))
1983 ifscope = inp->inp_boundif;
1984
1985 bzero(&ro, sizeof (ro));
1986 memcpy(&ro.ro_dst, gsin, sizeof(struct sockaddr_in));
1987 rtalloc_scoped_ign(&ro, 0, ifscope);
1988 if (ro.ro_rt != NULL) {
1989 ifp = ro.ro_rt->rt_ifp;
1990 VERIFY(ifp != NULL);
1991 rtfree(ro.ro_rt);
1992 } else {
1993 struct in_ifaddr *ia;
1994 struct ifnet *mifp;
1995
1996 mifp = NULL;
1997 lck_rw_lock_shared(in_ifaddr_rwlock);
1998 TAILQ_FOREACH(ia, &in_ifaddrhead, ia_link) {
1999 IFA_LOCK_SPIN(&ia->ia_ifa);
2000 mifp = ia->ia_ifp;
2001 IFA_UNLOCK(&ia->ia_ifa);
2002 if (!(mifp->if_flags & IFF_LOOPBACK) &&
2003 (mifp->if_flags & IFF_MULTICAST)) {
2004 ifp = mifp;
2005 break;
2006 }
2007 }
2008 lck_rw_done(in_ifaddr_rwlock);
2009 }
2010 }
2011
2012 return (ifp);
2013 }
2014
2015 /*
2016 * Join an IPv4 multicast group, possibly with a source.
2017 *
2018 * NB: sopt->sopt_val might point to the kernel address space. This means that
2019 * we were called by the IPv6 stack due to the presence of an IPv6 v4 mapped
2020 * address. In this scenario, sopt_p points to kernproc and sooptcopyin() will
2021 * just issue an in-kernel memcpy.
2022 */
2023 int
2024 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
2025 {
2026 struct group_source_req gsr;
2027 sockunion_t *gsa, *ssa;
2028 struct ifnet *ifp;
2029 struct in_mfilter *imf;
2030 struct ip_moptions *imo;
2031 struct in_multi *inm = NULL;
2032 struct in_msource *lims;
2033 size_t idx;
2034 int error, is_new;
2035
2036 ifp = NULL;
2037 imf = NULL;
2038 error = 0;
2039 is_new = 0;
2040
2041 memset(&gsr, 0, sizeof(struct group_source_req));
2042 gsa = (sockunion_t *)&gsr.gsr_group;
2043 gsa->ss.ss_family = AF_UNSPEC;
2044 ssa = (sockunion_t *)&gsr.gsr_source;
2045 ssa->ss.ss_family = AF_UNSPEC;
2046
2047 switch (sopt->sopt_name) {
2048 case IP_ADD_MEMBERSHIP:
2049 case IP_ADD_SOURCE_MEMBERSHIP: {
2050 struct ip_mreq_source mreqs;
2051
2052 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
2053 error = sooptcopyin(sopt, &mreqs,
2054 sizeof(struct ip_mreq),
2055 sizeof(struct ip_mreq));
2056 /*
2057 * Do argument switcharoo from ip_mreq into
2058 * ip_mreq_source to avoid using two instances.
2059 */
2060 mreqs.imr_interface = mreqs.imr_sourceaddr;
2061 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2062 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2063 error = sooptcopyin(sopt, &mreqs,
2064 sizeof(struct ip_mreq_source),
2065 sizeof(struct ip_mreq_source));
2066 }
2067 if (error) {
2068 IGMP_PRINTF(("%s: error copyin IP_ADD_MEMBERSHIP/"
2069 "IP_ADD_SOURCE_MEMBERSHIP %d err=%d\n",
2070 __func__, sopt->sopt_name, error));
2071 return (error);
2072 }
2073
2074 gsa->sin.sin_family = AF_INET;
2075 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2076 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2077
2078 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2079 ssa->sin.sin_family = AF_INET;
2080 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2081 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2082 }
2083
2084 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2085 return (EINVAL);
2086
2087 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2088 mreqs.imr_interface);
2089 IGMP_PRINTF(("%s: imr_interface = %s, ifp = %p\n",
2090 __func__, inet_ntoa(mreqs.imr_interface), ifp));
2091 break;
2092 }
2093
2094 case MCAST_JOIN_GROUP:
2095 case MCAST_JOIN_SOURCE_GROUP:
2096 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2097 error = sooptcopyin(sopt, &gsr,
2098 sizeof(struct group_req),
2099 sizeof(struct group_req));
2100 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2101 error = sooptcopyin(sopt, &gsr,
2102 sizeof(struct group_source_req),
2103 sizeof(struct group_source_req));
2104 }
2105 if (error)
2106 return (error);
2107
2108 if (gsa->sin.sin_family != AF_INET ||
2109 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2110 return (EINVAL);
2111
2112 /*
2113 * Overwrite the port field if present, as the sockaddr
2114 * being copied in may be matched with a binary comparison.
2115 */
2116 gsa->sin.sin_port = 0;
2117 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2118 if (ssa->sin.sin_family != AF_INET ||
2119 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2120 return (EINVAL);
2121 ssa->sin.sin_port = 0;
2122 }
2123
2124 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2125 return (EINVAL);
2126
2127 ifnet_head_lock_shared();
2128 if (gsr.gsr_interface == 0 ||
2129 (u_int)if_index < gsr.gsr_interface) {
2130 ifnet_head_done();
2131 return (EADDRNOTAVAIL);
2132 }
2133 ifp = ifindex2ifnet[gsr.gsr_interface];
2134 ifnet_head_done();
2135
2136 break;
2137
2138 default:
2139 IGMP_PRINTF(("%s: unknown sopt_name %d\n",
2140 __func__, sopt->sopt_name));
2141 return (EOPNOTSUPP);
2142 break;
2143 }
2144
2145 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2146 return (EADDRNOTAVAIL);
2147
2148 imo = inp_findmoptions(inp);
2149 if (imo == NULL)
2150 return (ENOMEM);
2151
2152 IMO_LOCK(imo);
2153 idx = imo_match_group(imo, ifp, &gsa->sa);
2154 if (idx == (size_t)-1) {
2155 is_new = 1;
2156 } else {
2157 inm = imo->imo_membership[idx];
2158 imf = &imo->imo_mfilters[idx];
2159 if (ssa->ss.ss_family != AF_UNSPEC) {
2160 /*
2161 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2162 * is an error. On an existing inclusive membership,
2163 * it just adds the source to the filter list.
2164 */
2165 if (imf->imf_st[1] != MCAST_INCLUDE) {
2166 error = EINVAL;
2167 goto out_imo_locked;
2168 }
2169 /*
2170 * Throw out duplicates.
2171 *
2172 * XXX FIXME: This makes a naive assumption that
2173 * even if entries exist for *ssa in this imf,
2174 * they will be rejected as dupes, even if they
2175 * are not valid in the current mode (in-mode).
2176 *
2177 * in_msource is transactioned just as for anything
2178 * else in SSM -- but note naive use of inm_graft()
2179 * below for allocating new filter entries.
2180 *
2181 * This is only an issue if someone mixes the
2182 * full-state SSM API with the delta-based API,
2183 * which is discouraged in the relevant RFCs.
2184 */
2185 lims = imo_match_source(imo, idx, &ssa->sa);
2186 if (lims != NULL /*&&
2187 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2188 error = EADDRNOTAVAIL;
2189 goto out_imo_locked;
2190 }
2191 } else {
2192 /*
2193 * MCAST_JOIN_GROUP on an existing exclusive
2194 * membership is an error; return EADDRINUSE
2195 * to preserve 4.4BSD API idempotence, and
2196 * avoid tedious detour to code below.
2197 * NOTE: This is bending RFC 3678 a bit.
2198 *
2199 * On an existing inclusive membership, this is also
2200 * an error; if you want to change filter mode,
2201 * you must use the userland API setsourcefilter().
2202 * XXX We don't reject this for imf in UNDEFINED
2203 * state at t1, because allocation of a filter
2204 * is atomic with allocation of a membership.
2205 */
2206 error = EINVAL;
2207 /* See comments above for EADDRINUSE */
2208 if (imf->imf_st[1] == MCAST_EXCLUDE)
2209 error = EADDRINUSE;
2210 goto out_imo_locked;
2211 }
2212 }
2213
2214 /*
2215 * Begin state merge transaction at socket layer.
2216 */
2217
2218 if (is_new) {
2219 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2220 error = imo_grow(imo, 0);
2221 if (error)
2222 goto out_imo_locked;
2223 }
2224 /*
2225 * Allocate the new slot upfront so we can deal with
2226 * grafting the new source filter in same code path
2227 * as for join-source on existing membership.
2228 */
2229 idx = imo->imo_num_memberships;
2230 imo->imo_membership[idx] = NULL;
2231 imo->imo_num_memberships++;
2232 VERIFY(imo->imo_mfilters != NULL);
2233 imf = &imo->imo_mfilters[idx];
2234 VERIFY(RB_EMPTY(&imf->imf_sources));
2235 }
2236
2237 /*
2238 * Graft new source into filter list for this inpcb's
2239 * membership of the group. The in_multi may not have
2240 * been allocated yet if this is a new membership, however,
2241 * the in_mfilter slot will be allocated and must be initialized.
2242 */
2243 if (ssa->ss.ss_family != AF_UNSPEC) {
2244 /* Membership starts in IN mode */
2245 if (is_new) {
2246 IGMP_PRINTF(("%s: new join w/source\n", __func__));
2247 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2248 } else {
2249 IGMP_PRINTF(("%s: %s source\n", __func__, "allow"));
2250 }
2251 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2252 if (lims == NULL) {
2253 IGMP_PRINTF(("%s: merge imf state failed\n",
2254 __func__));
2255 error = ENOMEM;
2256 goto out_imo_free;
2257 }
2258 } else {
2259 /* No address specified; Membership starts in EX mode */
2260 if (is_new) {
2261 IGMP_PRINTF(("%s: new join w/o source\n", __func__));
2262 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2263 }
2264 }
2265
2266 /*
2267 * Begin state merge transaction at IGMP layer.
2268 */
2269
2270 if (is_new) {
2271 VERIFY(inm == NULL);
2272 error = in_joingroup(ifp, &gsa->sin.sin_addr, imf, &inm);
2273 VERIFY(inm != NULL || error != 0);
2274 if (error)
2275 goto out_imo_free;
2276 imo->imo_membership[idx] = inm; /* from in_joingroup() */
2277 } else {
2278 IGMP_PRINTF(("%s: merge inm state\n", __func__));
2279 INM_LOCK(inm);
2280 error = inm_merge(inm, imf);
2281 if (error) {
2282 IGMP_PRINTF(("%s: failed to merge inm state\n",
2283 __func__));
2284 INM_UNLOCK(inm);
2285 goto out_imf_rollback;
2286 }
2287 IGMP_PRINTF(("%s: doing igmp downcall\n", __func__));
2288 error = igmp_change_state(inm);
2289 INM_UNLOCK(inm);
2290 if (error) {
2291 IGMP_PRINTF(("%s: failed igmp downcall\n",
2292 __func__));
2293 goto out_imf_rollback;
2294 }
2295 }
2296
2297 out_imf_rollback:
2298 if (error) {
2299 imf_rollback(imf);
2300 if (is_new)
2301 imf_purge(imf);
2302 else
2303 imf_reap(imf);
2304 } else {
2305 imf_commit(imf);
2306 }
2307
2308 out_imo_free:
2309 if (error && is_new) {
2310 VERIFY(inm == NULL);
2311 imo->imo_membership[idx] = NULL;
2312 --imo->imo_num_memberships;
2313 }
2314
2315 out_imo_locked:
2316 IMO_UNLOCK(imo);
2317 IMO_REMREF(imo); /* from inp_findmoptions() */
2318 return (error);
2319 }
2320
2321 /*
2322 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2323 *
2324 * NB: sopt->sopt_val might point to the kernel address space. Refer to the
2325 * block comment on top of inp_join_group() for more information.
2326 */
2327 int
2328 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2329 {
2330 struct group_source_req gsr;
2331 struct ip_mreq_source mreqs;
2332 sockunion_t *gsa, *ssa;
2333 struct ifnet *ifp;
2334 struct in_mfilter *imf;
2335 struct ip_moptions *imo;
2336 struct in_msource *ims;
2337 struct in_multi *inm = NULL;
2338 size_t idx;
2339 int error, is_final;
2340 unsigned int ifindex = 0;
2341
2342 ifp = NULL;
2343 error = 0;
2344 is_final = 1;
2345
2346 memset(&gsr, 0, sizeof(struct group_source_req));
2347 gsa = (sockunion_t *)&gsr.gsr_group;
2348 gsa->ss.ss_family = AF_UNSPEC;
2349 ssa = (sockunion_t *)&gsr.gsr_source;
2350 ssa->ss.ss_family = AF_UNSPEC;
2351
2352 switch (sopt->sopt_name) {
2353 case IP_DROP_MEMBERSHIP:
2354 case IP_DROP_SOURCE_MEMBERSHIP:
2355 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2356 error = sooptcopyin(sopt, &mreqs,
2357 sizeof(struct ip_mreq),
2358 sizeof(struct ip_mreq));
2359 /*
2360 * Swap interface and sourceaddr arguments,
2361 * as ip_mreq and ip_mreq_source are laid
2362 * out differently.
2363 */
2364 mreqs.imr_interface = mreqs.imr_sourceaddr;
2365 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2366 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2367 error = sooptcopyin(sopt, &mreqs,
2368 sizeof(struct ip_mreq_source),
2369 sizeof(struct ip_mreq_source));
2370 }
2371 if (error)
2372 return (error);
2373
2374 gsa->sin.sin_family = AF_INET;
2375 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2376 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2377
2378 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2379 ssa->sin.sin_family = AF_INET;
2380 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2381 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2382 }
2383 /*
2384 * Attempt to look up hinted ifp from interface address.
2385 * Fallthrough with null ifp iff lookup fails, to
2386 * preserve 4.4BSD mcast API idempotence.
2387 * XXX NOTE WELL: The RFC 3678 API is preferred because
2388 * using an IPv4 address as a key is racy.
2389 */
2390 if (!in_nullhost(mreqs.imr_interface))
2391 ifp = ip_multicast_if(&mreqs.imr_interface, &ifindex);
2392
2393 IGMP_PRINTF(("%s: imr_interface = %s, ifp = %p\n",
2394 __func__, inet_ntoa(mreqs.imr_interface), ifp));
2395
2396 break;
2397
2398 case MCAST_LEAVE_GROUP:
2399 case MCAST_LEAVE_SOURCE_GROUP:
2400 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2401 error = sooptcopyin(sopt, &gsr,
2402 sizeof(struct group_req),
2403 sizeof(struct group_req));
2404 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2405 error = sooptcopyin(sopt, &gsr,
2406 sizeof(struct group_source_req),
2407 sizeof(struct group_source_req));
2408 }
2409 if (error)
2410 return (error);
2411
2412 if (gsa->sin.sin_family != AF_INET ||
2413 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2414 return (EINVAL);
2415
2416 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2417 if (ssa->sin.sin_family != AF_INET ||
2418 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2419 return (EINVAL);
2420 }
2421
2422 ifnet_head_lock_shared();
2423 if (gsr.gsr_interface == 0 ||
2424 (u_int)if_index < gsr.gsr_interface) {
2425 ifnet_head_done();
2426 return (EADDRNOTAVAIL);
2427 }
2428
2429 ifp = ifindex2ifnet[gsr.gsr_interface];
2430 ifnet_head_done();
2431 break;
2432
2433 default:
2434 IGMP_PRINTF(("%s: unknown sopt_name %d\n",
2435 __func__, sopt->sopt_name));
2436 return (EOPNOTSUPP);
2437 break;
2438 }
2439
2440 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2441 return (EINVAL);
2442
2443 /*
2444 * Find the membership in the membership array.
2445 */
2446 imo = inp_findmoptions(inp);
2447 if (imo == NULL)
2448 return (ENOMEM);
2449
2450 IMO_LOCK(imo);
2451 idx = imo_match_group(imo, ifp, &gsa->sa);
2452 if (idx == (size_t)-1) {
2453 error = EADDRNOTAVAIL;
2454 goto out_locked;
2455 }
2456 inm = imo->imo_membership[idx];
2457 imf = &imo->imo_mfilters[idx];
2458
2459 if (ssa->ss.ss_family != AF_UNSPEC) {
2460 IGMP_PRINTF(("%s: opt=%d is_final=0\n", __func__,
2461 sopt->sopt_name));
2462 is_final = 0;
2463 }
2464
2465 /*
2466 * Begin state merge transaction at socket layer.
2467 */
2468
2469 /*
2470 * If we were instructed only to leave a given source, do so.
2471 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2472 */
2473 if (is_final) {
2474 imf_leave(imf);
2475 } else {
2476 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2477 error = EADDRNOTAVAIL;
2478 goto out_locked;
2479 }
2480 ims = imo_match_source(imo, idx, &ssa->sa);
2481 if (ims == NULL) {
2482 IGMP_PRINTF(("%s: source %s %spresent\n", __func__,
2483 inet_ntoa(ssa->sin.sin_addr), "not "));
2484 error = EADDRNOTAVAIL;
2485 goto out_locked;
2486 }
2487 IGMP_PRINTF(("%s: %s source\n", __func__, "block"));
2488 error = imf_prune(imf, &ssa->sin);
2489 if (error) {
2490 IGMP_PRINTF(("%s: merge imf state failed\n",
2491 __func__));
2492 goto out_locked;
2493 }
2494 }
2495
2496 /*
2497 * Begin state merge transaction at IGMP layer.
2498 */
2499
2500 if (is_final) {
2501 /*
2502 * Give up the multicast address record to which
2503 * the membership points. Reference held in imo
2504 * will be released below.
2505 */
2506 (void) in_leavegroup(inm, imf);
2507 } else {
2508 IGMP_PRINTF(("%s: merge inm state\n", __func__));
2509 INM_LOCK(inm);
2510 error = inm_merge(inm, imf);
2511 if (error) {
2512 IGMP_PRINTF(("%s: failed to merge inm state\n",
2513 __func__));
2514 INM_UNLOCK(inm);
2515 goto out_imf_rollback;
2516 }
2517
2518 IGMP_PRINTF(("%s: doing igmp downcall\n", __func__));
2519 error = igmp_change_state(inm);
2520 if (error) {
2521 IGMP_PRINTF(("%s: failed igmp downcall\n", __func__));
2522 }
2523 INM_UNLOCK(inm);
2524 }
2525
2526 out_imf_rollback:
2527 if (error)
2528 imf_rollback(imf);
2529 else
2530 imf_commit(imf);
2531
2532 imf_reap(imf);
2533
2534 if (is_final) {
2535 /* Remove the gap in the membership and filter array. */
2536 VERIFY(inm == imo->imo_membership[idx]);
2537 imo->imo_membership[idx] = NULL;
2538 INM_REMREF(inm);
2539 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2540 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2541 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2542 }
2543 imo->imo_num_memberships--;
2544 }
2545
2546 out_locked:
2547 IMO_UNLOCK(imo);
2548 IMO_REMREF(imo); /* from inp_findmoptions() */
2549 return (error);
2550 }
2551
2552 /*
2553 * Select the interface for transmitting IPv4 multicast datagrams.
2554 *
2555 * Either an instance of struct in_addr or an instance of struct ip_mreqn
2556 * may be passed to this socket option. An address of INADDR_ANY or an
2557 * interface index of 0 is used to remove a previous selection.
2558 * When no interface is selected, one is chosen for every send.
2559 */
2560 static int
2561 inp_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2562 {
2563 struct in_addr addr;
2564 struct ip_mreqn mreqn;
2565 struct ifnet *ifp;
2566 struct ip_moptions *imo;
2567 int error = 0 ;
2568 unsigned int ifindex = 0;
2569
2570 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2571 /*
2572 * An interface index was specified using the
2573 * Linux-derived ip_mreqn structure.
2574 */
2575 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2576 sizeof(struct ip_mreqn));
2577 if (error)
2578 return (error);
2579
2580 ifnet_head_lock_shared();
2581 if (mreqn.imr_ifindex < 0 || if_index < mreqn.imr_ifindex) {
2582 ifnet_head_done();
2583 return (EINVAL);
2584 }
2585
2586 if (mreqn.imr_ifindex == 0) {
2587 ifp = NULL;
2588 } else {
2589 ifp = ifindex2ifnet[mreqn.imr_ifindex];
2590 if (ifp == NULL) {
2591 ifnet_head_done();
2592 return (EADDRNOTAVAIL);
2593 }
2594 }
2595 ifnet_head_done();
2596 } else {
2597 /*
2598 * An interface was specified by IPv4 address.
2599 * This is the traditional BSD usage.
2600 */
2601 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2602 sizeof(struct in_addr));
2603 if (error)
2604 return (error);
2605 if (in_nullhost(addr)) {
2606 ifp = NULL;
2607 } else {
2608 ifp = ip_multicast_if(&addr, &ifindex);
2609 if (ifp == NULL) {
2610 IGMP_PRINTF(("%s: can't find ifp for addr=%s\n",
2611 __func__, inet_ntoa(addr)));
2612 return (EADDRNOTAVAIL);
2613 }
2614 }
2615 #ifdef IGMP_DEBUG0
2616 IGMP_PRINTF(("%s: ifp = %p, addr = %s\n", __func__, ifp,
2617 inet_ntoa(addr)));
2618 #endif
2619 }
2620
2621 /* Reject interfaces which do not support multicast. */
2622 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2623 return (EOPNOTSUPP);
2624
2625 imo = inp_findmoptions(inp);
2626 if (imo == NULL)
2627 return (ENOMEM);
2628
2629 IMO_LOCK(imo);
2630 imo->imo_multicast_ifp = ifp;
2631 if (ifindex)
2632 imo->imo_multicast_addr = addr;
2633 else
2634 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2635 IMO_UNLOCK(imo);
2636 IMO_REMREF(imo); /* from inp_findmoptions() */
2637
2638 return (0);
2639 }
2640
2641 /*
2642 * Atomically set source filters on a socket for an IPv4 multicast group.
2643 */
2644 static int
2645 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2646 {
2647 struct __msfilterreq64 msfr, msfr64;
2648 struct __msfilterreq32 msfr32;
2649 sockunion_t *gsa;
2650 struct ifnet *ifp;
2651 struct in_mfilter *imf;
2652 struct ip_moptions *imo;
2653 struct in_multi *inm;
2654 size_t idx;
2655 int error;
2656 user_addr_t tmp_ptr;
2657
2658 if (IS_64BIT_PROCESS(current_proc())) {
2659 error = sooptcopyin(sopt, &msfr64,
2660 sizeof(struct __msfilterreq64),
2661 sizeof(struct __msfilterreq64));
2662 if (error)
2663 return (error);
2664 /* we never use msfr.msfr_srcs; */
2665 memcpy(&msfr, &msfr64, sizeof(msfr));
2666 } else {
2667 error = sooptcopyin(sopt, &msfr32,
2668 sizeof(struct __msfilterreq32),
2669 sizeof(struct __msfilterreq32));
2670 if (error)
2671 return (error);
2672 /* we never use msfr.msfr_srcs; */
2673 memcpy(&msfr, &msfr32, sizeof(msfr));
2674 }
2675
2676 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2677 return (ENOBUFS);
2678
2679 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2680 msfr.msfr_fmode != MCAST_INCLUDE))
2681 return (EINVAL);
2682
2683 if (msfr.msfr_group.ss_family != AF_INET ||
2684 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2685 return (EINVAL);
2686
2687 gsa = (sockunion_t *)&msfr.msfr_group;
2688 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2689 return (EINVAL);
2690
2691 gsa->sin.sin_port = 0; /* ignore port */
2692
2693 ifnet_head_lock_shared();
2694 if (msfr.msfr_ifindex == 0 || (u_int)if_index < msfr.msfr_ifindex) {
2695 ifnet_head_done();
2696 return (EADDRNOTAVAIL);
2697 }
2698
2699 ifp = ifindex2ifnet[msfr.msfr_ifindex];
2700 ifnet_head_done();
2701 if (ifp == NULL)
2702 return (EADDRNOTAVAIL);
2703
2704 /*
2705 * Check if this socket is a member of this group.
2706 */
2707 imo = inp_findmoptions(inp);
2708 if (imo == NULL)
2709 return (ENOMEM);
2710
2711 IMO_LOCK(imo);
2712 idx = imo_match_group(imo, ifp, &gsa->sa);
2713 if (idx == (size_t)-1 || imo->imo_mfilters == NULL) {
2714 error = EADDRNOTAVAIL;
2715 goto out_imo_locked;
2716 }
2717 inm = imo->imo_membership[idx];
2718 imf = &imo->imo_mfilters[idx];
2719
2720 /*
2721 * Begin state merge transaction at socket layer.
2722 */
2723
2724 imf->imf_st[1] = msfr.msfr_fmode;
2725
2726 /*
2727 * Apply any new source filters, if present.
2728 * Make a copy of the user-space source vector so
2729 * that we may copy them with a single copyin. This
2730 * allows us to deal with page faults up-front.
2731 */
2732 if (msfr.msfr_nsrcs > 0) {
2733 struct in_msource *lims;
2734 struct sockaddr_in *psin;
2735 struct sockaddr_storage *kss, *pkss;
2736 int i;
2737
2738 if (IS_64BIT_PROCESS(current_proc()))
2739 tmp_ptr = msfr64.msfr_srcs;
2740 else
2741 tmp_ptr = CAST_USER_ADDR_T(msfr32.msfr_srcs);
2742
2743 IGMP_PRINTF(("%s: loading %lu source list entries\n",
2744 __func__, (unsigned long)msfr.msfr_nsrcs));
2745 kss = _MALLOC(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2746 M_TEMP, M_WAITOK);
2747 if (kss == NULL) {
2748 error = ENOMEM;
2749 goto out_imo_locked;
2750 }
2751 error = copyin(tmp_ptr, kss,
2752 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2753 if (error) {
2754 FREE(kss, M_TEMP);
2755 goto out_imo_locked;
2756 }
2757
2758 /*
2759 * Mark all source filters as UNDEFINED at t1.
2760 * Restore new group filter mode, as imf_leave()
2761 * will set it to INCLUDE.
2762 */
2763 imf_leave(imf);
2764 imf->imf_st[1] = msfr.msfr_fmode;
2765
2766 /*
2767 * Update socket layer filters at t1, lazy-allocating
2768 * new entries. This saves a bunch of memory at the
2769 * cost of one RB_FIND() per source entry; duplicate
2770 * entries in the msfr_nsrcs vector are ignored.
2771 * If we encounter an error, rollback transaction.
2772 *
2773 * XXX This too could be replaced with a set-symmetric
2774 * difference like loop to avoid walking from root
2775 * every time, as the key space is common.
2776 */
2777 for (i = 0, pkss = kss; (u_int)i < msfr.msfr_nsrcs;
2778 i++, pkss++) {
2779 psin = (struct sockaddr_in *)pkss;
2780 if (psin->sin_family != AF_INET) {
2781 error = EAFNOSUPPORT;
2782 break;
2783 }
2784 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2785 error = EINVAL;
2786 break;
2787 }
2788 error = imf_get_source(imf, psin, &lims);
2789 if (error)
2790 break;
2791 lims->imsl_st[1] = imf->imf_st[1];
2792 }
2793 FREE(kss, M_TEMP);
2794 }
2795
2796 if (error)
2797 goto out_imf_rollback;
2798
2799 /*
2800 * Begin state merge transaction at IGMP layer.
2801 */
2802 INM_LOCK(inm);
2803 IGMP_PRINTF(("%s: merge inm state\n", __func__));
2804 error = inm_merge(inm, imf);
2805 if (error) {
2806 IGMP_PRINTF(("%s: failed to merge inm state\n", __func__));
2807 INM_UNLOCK(inm);
2808 goto out_imf_rollback;
2809 }
2810
2811 IGMP_PRINTF(("%s: doing igmp downcall\n", __func__));
2812 error = igmp_change_state(inm);
2813 INM_UNLOCK(inm);
2814 #ifdef IGMP_DEBUG
2815 if (error)
2816 IGMP_PRINTF(("%s: failed igmp downcall\n", __func__));
2817 #endif
2818
2819 out_imf_rollback:
2820 if (error)
2821 imf_rollback(imf);
2822 else
2823 imf_commit(imf);
2824
2825 imf_reap(imf);
2826
2827 out_imo_locked:
2828 IMO_UNLOCK(imo);
2829 IMO_REMREF(imo); /* from inp_findmoptions() */
2830
2831 return (error);
2832 }
2833
2834 /*
2835 * Set the IP multicast options in response to user setsockopt().
2836 *
2837 * Many of the socket options handled in this function duplicate the
2838 * functionality of socket options in the regular unicast API. However,
2839 * it is not possible to merge the duplicate code, because the idempotence
2840 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2841 * the effects of these options must be treated as separate and distinct.
2842 *
2843 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2844 * is refactored to no longer use vifs.
2845 */
2846 int
2847 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2848 {
2849 struct ip_moptions *imo;
2850 int error;
2851 unsigned int ifindex;
2852 struct ifnet *ifp;
2853
2854 error = 0;
2855
2856 /*
2857 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2858 * or is a divert socket, reject it.
2859 */
2860 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2861 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2862 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2863 return (EOPNOTSUPP);
2864
2865 switch (sopt->sopt_name) {
2866 #if MROUTING
2867 case IP_MULTICAST_VIF: {
2868 int vifi;
2869 /*
2870 * Select a multicast VIF for transmission.
2871 * Only useful if multicast forwarding is active.
2872 */
2873 if (legal_vif_num == NULL) {
2874 error = EOPNOTSUPP;
2875 break;
2876 }
2877 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2878 if (error)
2879 break;
2880 if (!legal_vif_num(vifi) && (vifi != -1)) {
2881 error = EINVAL;
2882 break;
2883 }
2884 imo = inp_findmoptions(inp);
2885 if (imo == NULL) {
2886 error = ENOMEM;
2887 break;
2888 }
2889 IMO_LOCK(imo);
2890 imo->imo_multicast_vif = vifi;
2891 IMO_UNLOCK(imo);
2892 IMO_REMREF(imo); /* from inp_findmoptions() */
2893 break;
2894 }
2895 #endif
2896 case IP_MULTICAST_IF:
2897 error = inp_set_multicast_if(inp, sopt);
2898 break;
2899
2900 case IP_MULTICAST_IFINDEX:
2901 /*
2902 * Select the interface for outgoing multicast packets.
2903 */
2904 error = sooptcopyin(sopt, &ifindex, sizeof (ifindex),
2905 sizeof (ifindex));
2906 if (error)
2907 break;
2908
2909 imo = inp_findmoptions(inp);
2910 if (imo == NULL) {
2911 error = ENOMEM;
2912 break;
2913 }
2914 /*
2915 * Index 0 is used to remove a previous selection.
2916 * When no interface is selected, a default one is
2917 * chosen every time a multicast packet is sent.
2918 */
2919 if (ifindex == 0) {
2920 IMO_LOCK(imo);
2921 imo->imo_multicast_ifp = NULL;
2922 IMO_UNLOCK(imo);
2923 IMO_REMREF(imo); /* from inp_findmoptions() */
2924 break;
2925 }
2926
2927 ifnet_head_lock_shared();
2928 /* Don't need to check is ifindex is < 0 since it's unsigned */
2929 if ((unsigned int)if_index < ifindex) {
2930 ifnet_head_done();
2931 IMO_REMREF(imo); /* from inp_findmoptions() */
2932 error = ENXIO; /* per IPV6_MULTICAST_IF */
2933 break;
2934 }
2935 ifp = ifindex2ifnet[ifindex];
2936 ifnet_head_done();
2937
2938 /* If it's detached or isn't a multicast interface, bail out */
2939 if (ifp == NULL || !(ifp->if_flags & IFF_MULTICAST)) {
2940 IMO_REMREF(imo); /* from inp_findmoptions() */
2941 error = EADDRNOTAVAIL;
2942 break;
2943 }
2944 IMO_LOCK(imo);
2945 imo->imo_multicast_ifp = ifp;
2946 /*
2947 * Clear out any remnants of past IP_MULTICAST_IF. The addr
2948 * isn't really used anywhere in the kernel; we could have
2949 * iterated thru the addresses of the interface and pick one
2950 * here, but that is redundant since ip_getmoptions() already
2951 * takes care of that for INADDR_ANY.
2952 */
2953 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2954 IMO_UNLOCK(imo);
2955 IMO_REMREF(imo); /* from inp_findmoptions() */
2956 break;
2957
2958 case IP_MULTICAST_TTL: {
2959 u_char ttl;
2960
2961 /*
2962 * Set the IP time-to-live for outgoing multicast packets.
2963 * The original multicast API required a char argument,
2964 * which is inconsistent with the rest of the socket API.
2965 * We allow either a char or an int.
2966 */
2967 if (sopt->sopt_valsize == sizeof(u_char)) {
2968 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2969 sizeof(u_char));
2970 if (error)
2971 break;
2972 } else {
2973 u_int ittl;
2974
2975 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2976 sizeof(u_int));
2977 if (error)
2978 break;
2979 if (ittl > 255) {
2980 error = EINVAL;
2981 break;
2982 }
2983 ttl = (u_char)ittl;
2984 }
2985 imo = inp_findmoptions(inp);
2986 if (imo == NULL) {
2987 error = ENOMEM;
2988 break;
2989 }
2990 IMO_LOCK(imo);
2991 imo->imo_multicast_ttl = ttl;
2992 IMO_UNLOCK(imo);
2993 IMO_REMREF(imo); /* from inp_findmoptions() */
2994 break;
2995 }
2996
2997 case IP_MULTICAST_LOOP: {
2998 u_char loop;
2999
3000 /*
3001 * Set the loopback flag for outgoing multicast packets.
3002 * Must be zero or one. The original multicast API required a
3003 * char argument, which is inconsistent with the rest
3004 * of the socket API. We allow either a char or an int.
3005 */
3006 if (sopt->sopt_valsize == sizeof(u_char)) {
3007 error = sooptcopyin(sopt, &loop, sizeof(u_char),
3008 sizeof(u_char));
3009 if (error)
3010 break;
3011 } else {
3012 u_int iloop;
3013
3014 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
3015 sizeof(u_int));
3016 if (error)
3017 break;
3018 loop = (u_char)iloop;
3019 }
3020 imo = inp_findmoptions(inp);
3021 if (imo == NULL) {
3022 error = ENOMEM;
3023 break;
3024 }
3025 IMO_LOCK(imo);
3026 imo->imo_multicast_loop = !!loop;
3027 IMO_UNLOCK(imo);
3028 IMO_REMREF(imo); /* from inp_findmoptions() */
3029 break;
3030 }
3031
3032 case IP_ADD_MEMBERSHIP:
3033 case IP_ADD_SOURCE_MEMBERSHIP:
3034 case MCAST_JOIN_GROUP:
3035 case MCAST_JOIN_SOURCE_GROUP:
3036 error = inp_join_group(inp, sopt);
3037 break;
3038
3039 case IP_DROP_MEMBERSHIP:
3040 case IP_DROP_SOURCE_MEMBERSHIP:
3041 case MCAST_LEAVE_GROUP:
3042 case MCAST_LEAVE_SOURCE_GROUP:
3043 error = inp_leave_group(inp, sopt);
3044 break;
3045
3046 case IP_BLOCK_SOURCE:
3047 case IP_UNBLOCK_SOURCE:
3048 case MCAST_BLOCK_SOURCE:
3049 case MCAST_UNBLOCK_SOURCE:
3050 error = inp_block_unblock_source(inp, sopt);
3051 break;
3052
3053 case IP_MSFILTER:
3054 error = inp_set_source_filters(inp, sopt);
3055 break;
3056
3057 default:
3058 error = EOPNOTSUPP;
3059 break;
3060 }
3061
3062 return (error);
3063 }
3064
3065 /*
3066 * Expose IGMP's multicast filter mode and source list(s) to userland,
3067 * keyed by (ifindex, group).
3068 * The filter mode is written out as a uint32_t, followed by
3069 * 0..n of struct in_addr.
3070 * For use by ifmcstat(8).
3071 */
3072 static int
3073 sysctl_ip_mcast_filters SYSCTL_HANDLER_ARGS
3074 {
3075 #pragma unused(oidp)
3076
3077 struct in_addr src, group;
3078 struct ifnet *ifp;
3079 struct in_multi *inm;
3080 struct in_multistep step;
3081 struct ip_msource *ims;
3082 int *name;
3083 int retval = 0;
3084 u_int namelen;
3085 uint32_t fmode, ifindex;
3086
3087 name = (int *)arg1;
3088 namelen = (u_int)arg2;
3089
3090 if (req->newptr != USER_ADDR_NULL)
3091 return (EPERM);
3092
3093 if (namelen != 2)
3094 return (EINVAL);
3095
3096 ifindex = name[0];
3097 ifnet_head_lock_shared();
3098 if (ifindex <= 0 || ifindex > (u_int)if_index) {
3099 IGMP_PRINTF(("%s: ifindex %u out of range\n",
3100 __func__, ifindex));
3101 ifnet_head_done();
3102 return (ENOENT);
3103 }
3104
3105 group.s_addr = name[1];
3106 if (!IN_MULTICAST(ntohl(group.s_addr))) {
3107 IGMP_PRINTF(("%s: group %s is not multicast\n",
3108 __func__, inet_ntoa(group)));
3109 ifnet_head_done();
3110 return (EINVAL);
3111 }
3112
3113 ifp = ifindex2ifnet[ifindex];
3114 ifnet_head_done();
3115 if (ifp == NULL) {
3116 IGMP_PRINTF(("%s: no ifp for ifindex %u\n", __func__, ifindex));
3117 return (ENOENT);
3118 }
3119
3120 in_multihead_lock_shared();
3121 IN_FIRST_MULTI(step, inm);
3122 while (inm != NULL) {
3123 INM_LOCK(inm);
3124 if (inm->inm_ifp != ifp)
3125 goto next;
3126
3127 if (!in_hosteq(inm->inm_addr, group))
3128 goto next;
3129
3130 fmode = inm->inm_st[1].iss_fmode;
3131 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
3132 if (retval != 0) {
3133 INM_UNLOCK(inm);
3134 break; /* abort */
3135 }
3136 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
3137 #ifdef IGMP_DEBUG
3138 struct in_addr ina;
3139 ina.s_addr = htonl(ims->ims_haddr);
3140 IGMP_PRINTF(("%s: visit node %s\n", __func__,
3141 inet_ntoa(ina)));
3142 #endif
3143 /*
3144 * Only copy-out sources which are in-mode.
3145 */
3146 if (fmode != ims_get_mode(inm, ims, 1)) {
3147 IGMP_PRINTF(("%s: skip non-in-mode\n",
3148 __func__));
3149 continue; /* process next source */
3150 }
3151 src.s_addr = htonl(ims->ims_haddr);
3152 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
3153 if (retval != 0)
3154 break; /* process next inm */
3155 }
3156 next:
3157 INM_UNLOCK(inm);
3158 IN_NEXT_MULTI(step, inm);
3159 }
3160 in_multihead_lock_done();
3161
3162 return (retval);
3163 }
3164
3165 /*
3166 * XXX
3167 * The whole multicast option thing needs to be re-thought.
3168 * Several of these options are equally applicable to non-multicast
3169 * transmission, and one (IP_MULTICAST_TTL) totally duplicates a
3170 * standard option (IP_TTL).
3171 */
3172 /*
3173 * following RFC1724 section 3.3, 0.0.0.0/8 is interpreted as interface index.
3174 */
3175 static struct ifnet *
3176 ip_multicast_if(struct in_addr *a, unsigned int *ifindexp)
3177 {
3178 unsigned int ifindex;
3179 struct ifnet *ifp;
3180
3181 if (ifindexp != NULL)
3182 *ifindexp = 0;
3183 if (ntohl(a->s_addr) >> 24 == 0) {
3184 ifindex = ntohl(a->s_addr) & 0xffffff;
3185 ifnet_head_lock_shared();
3186 /* Don't need to check is ifindex is < 0 since it's unsigned */
3187 if ((unsigned int)if_index < ifindex) {
3188 ifnet_head_done();
3189 return (NULL);
3190 }
3191 ifp = ifindex2ifnet[ifindex];
3192 ifnet_head_done();
3193 if (ifp != NULL && ifindexp != NULL)
3194 *ifindexp = ifindex;
3195 } else {
3196 INADDR_TO_IFP(*a, ifp);
3197 }
3198 return (ifp);
3199 }
3200
3201 void
3202 in_multi_init(void)
3203 {
3204 PE_parse_boot_argn("ifa_debug", &inm_debug, sizeof (inm_debug));
3205
3206 /* Setup lock group and attribute for in_multihead */
3207 in_multihead_lock_grp_attr = lck_grp_attr_alloc_init();
3208 in_multihead_lock_grp = lck_grp_alloc_init("in_multihead",
3209 in_multihead_lock_grp_attr);
3210 in_multihead_lock_attr = lck_attr_alloc_init();
3211 lck_rw_init(&in_multihead_lock, in_multihead_lock_grp,
3212 in_multihead_lock_attr);
3213
3214 lck_mtx_init(&inm_trash_lock, in_multihead_lock_grp,
3215 in_multihead_lock_attr);
3216 TAILQ_INIT(&inm_trash_head);
3217
3218 inm_size = (inm_debug == 0) ? sizeof (struct in_multi) :
3219 sizeof (struct in_multi_dbg);
3220 inm_zone = zinit(inm_size, INM_ZONE_MAX * inm_size,
3221 0, INM_ZONE_NAME);
3222 if (inm_zone == NULL) {
3223 panic("%s: failed allocating %s", __func__, INM_ZONE_NAME);
3224 /* NOTREACHED */
3225 }
3226 zone_change(inm_zone, Z_EXPAND, TRUE);
3227
3228 ipms_size = sizeof (struct ip_msource);
3229 ipms_zone = zinit(ipms_size, IPMS_ZONE_MAX * ipms_size,
3230 0, IPMS_ZONE_NAME);
3231 if (ipms_zone == NULL) {
3232 panic("%s: failed allocating %s", __func__, IPMS_ZONE_NAME);
3233 /* NOTREACHED */
3234 }
3235 zone_change(ipms_zone, Z_EXPAND, TRUE);
3236
3237 inms_size = sizeof (struct in_msource);
3238 inms_zone = zinit(inms_size, INMS_ZONE_MAX * inms_size,
3239 0, INMS_ZONE_NAME);
3240 if (inms_zone == NULL) {
3241 panic("%s: failed allocating %s", __func__, INMS_ZONE_NAME);
3242 /* NOTREACHED */
3243 }
3244 zone_change(inms_zone, Z_EXPAND, TRUE);
3245 }
3246
3247 static struct in_multi *
3248 in_multi_alloc(int how)
3249 {
3250 struct in_multi *inm;
3251
3252 inm = (how == M_WAITOK) ? zalloc(inm_zone) : zalloc_noblock(inm_zone);
3253 if (inm != NULL) {
3254 bzero(inm, inm_size);
3255 lck_mtx_init(&inm->inm_lock, in_multihead_lock_grp,
3256 in_multihead_lock_attr);
3257 inm->inm_debug |= IFD_ALLOC;
3258 if (inm_debug != 0) {
3259 inm->inm_debug |= IFD_DEBUG;
3260 inm->inm_trace = inm_trace;
3261 }
3262 }
3263 return (inm);
3264 }
3265
3266 static void
3267 in_multi_free(struct in_multi *inm)
3268 {
3269 INM_LOCK(inm);
3270 if (inm->inm_debug & IFD_ATTACHED) {
3271 panic("%s: attached inm=%p is being freed", __func__, inm);
3272 /* NOTREACHED */
3273 } else if (inm->inm_ifma != NULL) {
3274 panic("%s: ifma not NULL for inm=%p", __func__, inm);
3275 /* NOTREACHED */
3276 } else if (!(inm->inm_debug & IFD_ALLOC)) {
3277 panic("%s: inm %p cannot be freed", __func__, inm);
3278 /* NOTREACHED */
3279 } else if (inm->inm_refcount != 0) {
3280 panic("%s: non-zero refcount inm=%p", __func__, inm);
3281 /* NOTREACHED */
3282 } else if (inm->inm_reqcnt != 0) {
3283 panic("%s: non-zero reqcnt inm=%p", __func__, inm);
3284 /* NOTREACHED */
3285 }
3286
3287 /* Free any pending IGMPv3 state-change records */
3288 IF_DRAIN(&inm->inm_scq);
3289
3290 inm->inm_debug &= ~IFD_ALLOC;
3291 if ((inm->inm_debug & (IFD_DEBUG | IFD_TRASHED)) ==
3292 (IFD_DEBUG | IFD_TRASHED)) {
3293 lck_mtx_lock(&inm_trash_lock);
3294 TAILQ_REMOVE(&inm_trash_head, (struct in_multi_dbg *)inm,
3295 inm_trash_link);
3296 lck_mtx_unlock(&inm_trash_lock);
3297 inm->inm_debug &= ~IFD_TRASHED;
3298 }
3299 INM_UNLOCK(inm);
3300
3301 lck_mtx_destroy(&inm->inm_lock, in_multihead_lock_grp);
3302 zfree(inm_zone, inm);
3303 }
3304
3305 static void
3306 in_multi_attach(struct in_multi *inm)
3307 {
3308 in_multihead_lock_assert(LCK_RW_ASSERT_EXCLUSIVE);
3309 INM_LOCK_ASSERT_HELD(inm);
3310
3311 if (inm->inm_debug & IFD_ATTACHED) {
3312 panic("%s: Attempt to attach an already attached inm=%p",
3313 __func__, inm);
3314 /* NOTREACHED */
3315 } else if (inm->inm_debug & IFD_TRASHED) {
3316 panic("%s: Attempt to reattach a detached inm=%p",
3317 __func__, inm);
3318 /* NOTREACHED */
3319 }
3320
3321 inm->inm_reqcnt++;
3322 VERIFY(inm->inm_reqcnt == 1);
3323 INM_ADDREF_LOCKED(inm);
3324 inm->inm_debug |= IFD_ATTACHED;
3325 /*
3326 * Reattach case: If debugging is enabled, take it
3327 * out of the trash list and clear IFD_TRASHED.
3328 */
3329 if ((inm->inm_debug & (IFD_DEBUG | IFD_TRASHED)) ==
3330 (IFD_DEBUG | IFD_TRASHED)) {
3331 /* Become a regular mutex, just in case */
3332 INM_CONVERT_LOCK(inm);
3333 lck_mtx_lock(&inm_trash_lock);
3334 TAILQ_REMOVE(&inm_trash_head, (struct in_multi_dbg *)inm,
3335 inm_trash_link);
3336 lck_mtx_unlock(&inm_trash_lock);
3337 inm->inm_debug &= ~IFD_TRASHED;
3338 }
3339
3340 LIST_INSERT_HEAD(&in_multihead, inm, inm_link);
3341 }
3342
3343 int
3344 in_multi_detach(struct in_multi *inm)
3345 {
3346 in_multihead_lock_assert(LCK_RW_ASSERT_EXCLUSIVE);
3347 INM_LOCK_ASSERT_HELD(inm);
3348
3349 if (inm->inm_reqcnt == 0) {
3350 panic("%s: inm=%p negative reqcnt", __func__, inm);
3351 /* NOTREACHED */
3352 }
3353
3354 --inm->inm_reqcnt;
3355 if (inm->inm_reqcnt > 0)
3356 return (0);
3357
3358 if (!(inm->inm_debug & IFD_ATTACHED)) {
3359 panic("%s: Attempt to detach an unattached record inm=%p",
3360 __func__, inm);
3361 /* NOTREACHED */
3362 } else if (inm->inm_debug & IFD_TRASHED) {
3363 panic("%s: inm %p is already in trash list", __func__, inm);
3364 /* NOTREACHED */
3365 }
3366
3367 /*
3368 * NOTE: Caller calls IFMA_REMREF
3369 */
3370 inm->inm_debug &= ~IFD_ATTACHED;
3371 LIST_REMOVE(inm, inm_link);
3372
3373 if (inm->inm_debug & IFD_DEBUG) {
3374 /* Become a regular mutex, just in case */
3375 INM_CONVERT_LOCK(inm);
3376 lck_mtx_lock(&inm_trash_lock);
3377 TAILQ_INSERT_TAIL(&inm_trash_head,
3378 (struct in_multi_dbg *)inm, inm_trash_link);
3379 lck_mtx_unlock(&inm_trash_lock);
3380 inm->inm_debug |= IFD_TRASHED;
3381 }
3382
3383 return (1);
3384 }
3385
3386 void
3387 inm_addref(struct in_multi *inm, int locked)
3388 {
3389 if (!locked)
3390 INM_LOCK_SPIN(inm);
3391 else
3392 INM_LOCK_ASSERT_HELD(inm);
3393
3394 if (++inm->inm_refcount == 0) {
3395 panic("%s: inm=%p wraparound refcnt", __func__, inm);
3396 /* NOTREACHED */
3397 } else if (inm->inm_trace != NULL) {
3398 (*inm->inm_trace)(inm, TRUE);
3399 }
3400 if (!locked)
3401 INM_UNLOCK(inm);
3402 }
3403
3404 void
3405 inm_remref(struct in_multi *inm, int locked)
3406 {
3407 struct ifmultiaddr *ifma;
3408 struct igmp_ifinfo *igi;
3409
3410 if (!locked)
3411 INM_LOCK_SPIN(inm);
3412 else
3413 INM_LOCK_ASSERT_HELD(inm);
3414
3415 if (inm->inm_refcount == 0 || (inm->inm_refcount == 1 && locked)) {
3416 panic("%s: inm=%p negative/missing refcnt", __func__, inm);
3417 /* NOTREACHED */
3418 } else if (inm->inm_trace != NULL) {
3419 (*inm->inm_trace)(inm, FALSE);
3420 }
3421
3422 --inm->inm_refcount;
3423 if (inm->inm_refcount > 0) {
3424 if (!locked)
3425 INM_UNLOCK(inm);
3426 return;
3427 }
3428
3429 /*
3430 * Synchronization with in_getmulti(). In the event the inm has been
3431 * detached, the underlying ifma would still be in the if_multiaddrs
3432 * list, and thus can be looked up via if_addmulti(). At that point,
3433 * the only way to find this inm is via ifma_protospec. To avoid
3434 * race conditions between the last inm_remref() of that inm and its
3435 * use via ifma_protospec, in_multihead lock is used for serialization.
3436 * In order to avoid violating the lock order, we must drop inm_lock
3437 * before acquiring in_multihead lock. To prevent the inm from being
3438 * freed prematurely, we hold an extra reference.
3439 */
3440 ++inm->inm_refcount;
3441 INM_UNLOCK(inm);
3442 in_multihead_lock_shared();
3443 INM_LOCK_SPIN(inm);
3444 --inm->inm_refcount;
3445 if (inm->inm_refcount > 0) {
3446 /* We've lost the race, so abort since inm is still in use */
3447 INM_UNLOCK(inm);
3448 in_multihead_lock_done();
3449 /* If it was locked, return it as such */
3450 if (locked)
3451 INM_LOCK(inm);
3452 return;
3453 }
3454 inm_purge(inm);
3455 ifma = inm->inm_ifma;
3456 inm->inm_ifma = NULL;
3457 inm->inm_ifp = NULL;
3458 igi = inm->inm_igi;
3459 inm->inm_igi = NULL;
3460 INM_UNLOCK(inm);
3461 IFMA_LOCK_SPIN(ifma);
3462 ifma->ifma_protospec = NULL;
3463 IFMA_UNLOCK(ifma);
3464 in_multihead_lock_done();
3465
3466 in_multi_free(inm);
3467 if_delmulti_ifma(ifma);
3468 /* Release reference held to the underlying ifmultiaddr */
3469 IFMA_REMREF(ifma);
3470
3471 if (igi != NULL)
3472 IGI_REMREF(igi);
3473 }
3474
3475 static void
3476 inm_trace(struct in_multi *inm, int refhold)
3477 {
3478 struct in_multi_dbg *inm_dbg = (struct in_multi_dbg *)inm;
3479 ctrace_t *tr;
3480 u_int32_t idx;
3481 u_int16_t *cnt;
3482
3483 if (!(inm->inm_debug & IFD_DEBUG)) {
3484 panic("%s: inm %p has no debug structure", __func__, inm);
3485 /* NOTREACHED */
3486 }
3487 if (refhold) {
3488 cnt = &inm_dbg->inm_refhold_cnt;
3489 tr = inm_dbg->inm_refhold;
3490 } else {
3491 cnt = &inm_dbg->inm_refrele_cnt;
3492 tr = inm_dbg->inm_refrele;
3493 }
3494
3495 idx = atomic_add_16_ov(cnt, 1) % INM_TRACE_HIST_SIZE;
3496 ctrace_record(&tr[idx]);
3497 }
3498
3499 void
3500 in_multihead_lock_exclusive(void)
3501 {
3502 lck_rw_lock_exclusive(&in_multihead_lock);
3503 }
3504
3505 void
3506 in_multihead_lock_shared(void)
3507 {
3508 lck_rw_lock_shared(&in_multihead_lock);
3509 }
3510
3511 void
3512 in_multihead_lock_assert(int what)
3513 {
3514 lck_rw_assert(&in_multihead_lock, what);
3515 }
3516
3517 void
3518 in_multihead_lock_done(void)
3519 {
3520 lck_rw_done(&in_multihead_lock);
3521 }
3522
3523 static struct ip_msource *
3524 ipms_alloc(int how)
3525 {
3526 struct ip_msource *ims;
3527
3528 ims = (how == M_WAITOK) ? zalloc(ipms_zone) : zalloc_noblock(ipms_zone);
3529 if (ims != NULL)
3530 bzero(ims, ipms_size);
3531
3532 return (ims);
3533 }
3534
3535 static void
3536 ipms_free(struct ip_msource *ims)
3537 {
3538 zfree(ipms_zone, ims);
3539 }
3540
3541 static struct in_msource *
3542 inms_alloc(int how)
3543 {
3544 struct in_msource *inms;
3545
3546 inms = (how == M_WAITOK) ? zalloc(inms_zone) :
3547 zalloc_noblock(inms_zone);
3548 if (inms != NULL)
3549 bzero(inms, inms_size);
3550
3551 return (inms);
3552 }
3553
3554 static void
3555 inms_free(struct in_msource *inms)
3556 {
3557 zfree(inms_zone, inms);
3558 }
3559
3560 #ifdef IGMP_DEBUG
3561
3562 static const char *inm_modestrs[] = { "un\n", "in", "ex" };
3563
3564 static const char *
3565 inm_mode_str(const int mode)
3566 {
3567 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
3568 return (inm_modestrs[mode]);
3569 return ("??");
3570 }
3571
3572 static const char *inm_statestrs[] = {
3573 "not-member\n",
3574 "silent\n",
3575 "idle\n",
3576 "lazy\n",
3577 "sleeping\n",
3578 "awakening\n",
3579 "query-pending\n",
3580 "sg-query-pending\n",
3581 "leaving"
3582 };
3583
3584 static const char *
3585 inm_state_str(const int state)
3586 {
3587 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3588 return (inm_statestrs[state]);
3589 return ("??");
3590 }
3591
3592 /*
3593 * Dump an in_multi structure to the console.
3594 */
3595 void
3596 inm_print(const struct in_multi *inm)
3597 {
3598 int t;
3599
3600 INM_LOCK_ASSERT_HELD(INM_CAST_TO_NONCONST(inm));
3601
3602 if (igmp_debug == 0)
3603 return;
3604
3605 printf("%s: --- begin inm %p ---\n", __func__, inm);
3606 printf("addr %s ifp %p(%s%d) ifma %p\n",
3607 inet_ntoa(inm->inm_addr),
3608 inm->inm_ifp,
3609 inm->inm_ifp->if_name,
3610 inm->inm_ifp->if_unit,
3611 inm->inm_ifma);
3612 printf("timer %u state %s refcount %u scq.len %u\n",
3613 inm->inm_timer,
3614 inm_state_str(inm->inm_state),
3615 inm->inm_refcount,
3616 inm->inm_scq.ifq_len);
3617 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3618 inm->inm_igi,
3619 inm->inm_nsrc,
3620 inm->inm_sctimer,
3621 inm->inm_scrv);
3622 for (t = 0; t < 2; t++) {
3623 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3624 inm_mode_str(inm->inm_st[t].iss_fmode),
3625 inm->inm_st[t].iss_asm,
3626 inm->inm_st[t].iss_ex,
3627 inm->inm_st[t].iss_in,
3628 inm->inm_st[t].iss_rec);
3629 }
3630 printf("%s: --- end inm %p ---\n", __func__, inm);
3631 }
3632
3633 #else
3634
3635 void
3636 inm_print(__unused const struct in_multi *inm)
3637 {
3638
3639 }
3640
3641 #endif