1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 2007-2009 Bruce Simpson.
5 * Copyright (c) 2005 Robert N. M. Watson.
6 * All rights reserved.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. The name of the author may not be used to endorse or promote
17 * products derived from this software without specific prior written
18 * permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
31 */
32
33 /*
34 * IPv4 multicast socket, group, and socket option processing module.
35 */
36
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD: releng/12.0/sys/netinet/in_mcast.c 338178 2018-08-22 04:09:55Z mmacy $");
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/lock.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/protosw.h>
47 #include <sys/rmlock.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/protosw.h>
51 #include <sys/sysctl.h>
52 #include <sys/ktr.h>
53 #include <sys/taskqueue.h>
54 #include <sys/gtaskqueue.h>
55 #include <sys/tree.h>
56
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/route.h>
61 #include <net/vnet.h>
62
63 #include <net/ethernet.h>
64
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_fib.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/in_var.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/igmp_var.h>
72
73 #ifndef KTR_IGMPV3
74 #define KTR_IGMPV3 KTR_INET
75 #endif
76
77 #ifndef __SOCKUNION_DECLARED
78 union sockunion {
79 struct sockaddr_storage ss;
80 struct sockaddr sa;
81 struct sockaddr_dl sdl;
82 struct sockaddr_in sin;
83 };
84 typedef union sockunion sockunion_t;
85 #define __SOCKUNION_DECLARED
86 #endif /* __SOCKUNION_DECLARED */
87
88 static MALLOC_DEFINE(M_INMFILTER, "in_mfilter",
89 "IPv4 multicast PCB-layer source filter");
90 static MALLOC_DEFINE(M_IPMADDR, "in_multi", "IPv4 multicast group");
91 static MALLOC_DEFINE(M_IPMOPTS, "ip_moptions", "IPv4 multicast options");
92 static MALLOC_DEFINE(M_IPMSOURCE, "ip_msource",
93 "IPv4 multicast IGMP-layer source filter");
94
95 /*
96 * Locking:
97 * - Lock order is: Giant, INP_WLOCK, IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK.
98 * - The IF_ADDR_LOCK is implicitly taken by inm_lookup() earlier, however
99 * it can be taken by code in net/if.c also.
100 * - ip_moptions and in_mfilter are covered by the INP_WLOCK.
101 *
102 * struct in_multi is covered by IN_MULTI_LIST_LOCK. There isn't strictly
103 * any need for in_multi itself to be virtualized -- it is bound to an ifp
104 * anyway no matter what happens.
105 */
106 struct mtx in_multi_list_mtx;
107 MTX_SYSINIT(in_multi_mtx, &in_multi_list_mtx, "in_multi_list_mtx", MTX_DEF);
108
109 struct mtx in_multi_free_mtx;
110 MTX_SYSINIT(in_multi_free_mtx, &in_multi_free_mtx, "in_multi_free_mtx", MTX_DEF);
111
112 struct sx in_multi_sx;
113 SX_SYSINIT(in_multi_sx, &in_multi_sx, "in_multi_sx");
114
115 int ifma_restart;
116
117 /*
118 * Functions with non-static linkage defined in this file should be
119 * declared in in_var.h:
120 * imo_multi_filter()
121 * in_addmulti()
122 * in_delmulti()
123 * in_joingroup()
124 * in_joingroup_locked()
125 * in_leavegroup()
126 * in_leavegroup_locked()
127 * and ip_var.h:
128 * inp_freemoptions()
129 * inp_getmoptions()
130 * inp_setmoptions()
131 *
132 * XXX: Both carp and pf need to use the legacy (*,G) KPIs in_addmulti()
133 * and in_delmulti().
134 */
135 static void imf_commit(struct in_mfilter *);
136 static int imf_get_source(struct in_mfilter *imf,
137 const struct sockaddr_in *psin,
138 struct in_msource **);
139 static struct in_msource *
140 imf_graft(struct in_mfilter *, const uint8_t,
141 const struct sockaddr_in *);
142 static void imf_leave(struct in_mfilter *);
143 static int imf_prune(struct in_mfilter *, const struct sockaddr_in *);
144 static void imf_purge(struct in_mfilter *);
145 static void imf_rollback(struct in_mfilter *);
146 static void imf_reap(struct in_mfilter *);
147 static int imo_grow(struct ip_moptions *);
148 static size_t imo_match_group(const struct ip_moptions *,
149 const struct ifnet *, const struct sockaddr *);
150 static struct in_msource *
151 imo_match_source(const struct ip_moptions *, const size_t,
152 const struct sockaddr *);
153 static void ims_merge(struct ip_msource *ims,
154 const struct in_msource *lims, const int rollback);
155 static int in_getmulti(struct ifnet *, const struct in_addr *,
156 struct in_multi **);
157 static int inm_get_source(struct in_multi *inm, const in_addr_t haddr,
158 const int noalloc, struct ip_msource **pims);
159 #ifdef KTR
160 static int inm_is_ifp_detached(const struct in_multi *);
161 #endif
162 static int inm_merge(struct in_multi *, /*const*/ struct in_mfilter *);
163 static void inm_purge(struct in_multi *);
164 static void inm_reap(struct in_multi *);
165 static void inm_release(struct in_multi *);
166 static struct ip_moptions *
167 inp_findmoptions(struct inpcb *);
168 static int inp_get_source_filters(struct inpcb *, struct sockopt *);
169 static int inp_join_group(struct inpcb *, struct sockopt *);
170 static int inp_leave_group(struct inpcb *, struct sockopt *);
171 static struct ifnet *
172 inp_lookup_mcast_ifp(const struct inpcb *,
173 const struct sockaddr_in *, const struct in_addr);
174 static int inp_block_unblock_source(struct inpcb *, struct sockopt *);
175 static int inp_set_multicast_if(struct inpcb *, struct sockopt *);
176 static int inp_set_source_filters(struct inpcb *, struct sockopt *);
177 static int sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS);
178
179 static SYSCTL_NODE(_net_inet_ip, OID_AUTO, mcast, CTLFLAG_RW, 0,
180 "IPv4 multicast");
181
182 static u_long in_mcast_maxgrpsrc = IP_MAX_GROUP_SRC_FILTER;
183 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxgrpsrc,
184 CTLFLAG_RWTUN, &in_mcast_maxgrpsrc, 0,
185 "Max source filters per group");
186
187 static u_long in_mcast_maxsocksrc = IP_MAX_SOCK_SRC_FILTER;
188 SYSCTL_ULONG(_net_inet_ip_mcast, OID_AUTO, maxsocksrc,
189 CTLFLAG_RWTUN, &in_mcast_maxsocksrc, 0,
190 "Max source filters per socket");
191
192 int in_mcast_loop = IP_DEFAULT_MULTICAST_LOOP;
193 SYSCTL_INT(_net_inet_ip_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
194 &in_mcast_loop, 0, "Loopback multicast datagrams by default");
195
196 static SYSCTL_NODE(_net_inet_ip_mcast, OID_AUTO, filters,
197 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip_mcast_filters,
198 "Per-interface stack-wide source filters");
199
200 #ifdef KTR
201 /*
202 * Inline function which wraps assertions for a valid ifp.
203 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
204 * is detached.
205 */
206 static int __inline
207 inm_is_ifp_detached(const struct in_multi *inm)
208 {
209 struct ifnet *ifp;
210
211 KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__));
212 ifp = inm->inm_ifma->ifma_ifp;
213 if (ifp != NULL) {
214 /*
215 * Sanity check that netinet's notion of ifp is the
216 * same as net's.
217 */
218 KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__));
219 }
220
221 return (ifp == NULL);
222 }
223 #endif
224
225 static struct grouptask free_gtask;
226 static struct in_multi_head inm_free_list;
227 static void inm_release_task(void *arg __unused);
228 static void inm_init(void)
229 {
230 SLIST_INIT(&inm_free_list);
231 taskqgroup_config_gtask_init(NULL, &free_gtask, inm_release_task, "inm release task");
232 }
233
234 #ifdef EARLY_AP_STARTUP
235 SYSINIT(inm_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
236 inm_init, NULL);
237 #else
238 SYSINIT(inm_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_FIRST,
239 inm_init, NULL);
240 #endif
241
242
243 void
244 inm_release_list_deferred(struct in_multi_head *inmh)
245 {
246
247 if (SLIST_EMPTY(inmh))
248 return;
249 mtx_lock(&in_multi_free_mtx);
250 SLIST_CONCAT(&inm_free_list, inmh, in_multi, inm_nrele);
251 mtx_unlock(&in_multi_free_mtx);
252 GROUPTASK_ENQUEUE(&free_gtask);
253 }
254
255 void
256 inm_disconnect(struct in_multi *inm)
257 {
258 struct ifnet *ifp;
259 struct ifmultiaddr *ifma, *ll_ifma;
260
261 ifp = inm->inm_ifp;
262 IF_ADDR_WLOCK_ASSERT(ifp);
263 ifma = inm->inm_ifma;
264
265 if_ref(ifp);
266 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
267 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
268 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
269 }
270 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
271 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
272 MPASS(ifma != ll_ifma);
273 ifma->ifma_llifma = NULL;
274 MPASS(ll_ifma->ifma_llifma == NULL);
275 MPASS(ll_ifma->ifma_ifp == ifp);
276 if (--ll_ifma->ifma_refcount == 0) {
277 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
278 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
279 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
280 }
281 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
282 if_freemulti(ll_ifma);
283 ifma_restart = true;
284 }
285 }
286 }
287
288 void
289 inm_release_deferred(struct in_multi *inm)
290 {
291 struct in_multi_head tmp;
292
293 IN_MULTI_LIST_LOCK_ASSERT();
294 MPASS(inm->inm_refcount > 0);
295 if (--inm->inm_refcount == 0) {
296 SLIST_INIT(&tmp);
297 inm_disconnect(inm);
298 inm->inm_ifma->ifma_protospec = NULL;
299 SLIST_INSERT_HEAD(&tmp, inm, inm_nrele);
300 inm_release_list_deferred(&tmp);
301 }
302 }
303
304 static void
305 inm_release_task(void *arg __unused)
306 {
307 struct in_multi_head inm_free_tmp;
308 struct in_multi *inm, *tinm;
309
310 SLIST_INIT(&inm_free_tmp);
311 mtx_lock(&in_multi_free_mtx);
312 SLIST_CONCAT(&inm_free_tmp, &inm_free_list, in_multi, inm_nrele);
313 mtx_unlock(&in_multi_free_mtx);
314 IN_MULTI_LOCK();
315 SLIST_FOREACH_SAFE(inm, &inm_free_tmp, inm_nrele, tinm) {
316 SLIST_REMOVE_HEAD(&inm_free_tmp, inm_nrele);
317 MPASS(inm);
318 inm_release(inm);
319 }
320 IN_MULTI_UNLOCK();
321 }
322
323 /*
324 * Initialize an in_mfilter structure to a known state at t0, t1
325 * with an empty source filter list.
326 */
327 static __inline void
328 imf_init(struct in_mfilter *imf, const int st0, const int st1)
329 {
330 memset(imf, 0, sizeof(struct in_mfilter));
331 RB_INIT(&imf->imf_sources);
332 imf->imf_st[0] = st0;
333 imf->imf_st[1] = st1;
334 }
335
336 /*
337 * Function for looking up an in_multi record for an IPv4 multicast address
338 * on a given interface. ifp must be valid. If no record found, return NULL.
339 * The IN_MULTI_LIST_LOCK and IF_ADDR_LOCK on ifp must be held.
340 */
341 struct in_multi *
342 inm_lookup_locked(struct ifnet *ifp, const struct in_addr ina)
343 {
344 struct ifmultiaddr *ifma;
345 struct in_multi *inm;
346
347 IN_MULTI_LIST_LOCK_ASSERT();
348 IF_ADDR_LOCK_ASSERT(ifp);
349
350 inm = NULL;
351 CK_STAILQ_FOREACH(ifma, &((ifp)->if_multiaddrs), ifma_link) {
352 if (ifma->ifma_addr->sa_family != AF_INET ||
353 ifma->ifma_protospec == NULL)
354 continue;
355 inm = (struct in_multi *)ifma->ifma_protospec;
356 if (inm->inm_addr.s_addr == ina.s_addr)
357 break;
358 inm = NULL;
359 }
360 return (inm);
361 }
362
363 /*
364 * Wrapper for inm_lookup_locked().
365 * The IF_ADDR_LOCK will be taken on ifp and released on return.
366 */
367 struct in_multi *
368 inm_lookup(struct ifnet *ifp, const struct in_addr ina)
369 {
370 struct in_multi *inm;
371
372 IN_MULTI_LIST_LOCK_ASSERT();
373 IF_ADDR_RLOCK(ifp);
374 inm = inm_lookup_locked(ifp, ina);
375 IF_ADDR_RUNLOCK(ifp);
376
377 return (inm);
378 }
379
380 /*
381 * Resize the ip_moptions vector to the next power-of-two minus 1.
382 * May be called with locks held; do not sleep.
383 */
384 static int
385 imo_grow(struct ip_moptions *imo)
386 {
387 struct in_multi **nmships;
388 struct in_multi **omships;
389 struct in_mfilter *nmfilters;
390 struct in_mfilter *omfilters;
391 size_t idx;
392 size_t newmax;
393 size_t oldmax;
394
395 nmships = NULL;
396 nmfilters = NULL;
397 omships = imo->imo_membership;
398 omfilters = imo->imo_mfilters;
399 oldmax = imo->imo_max_memberships;
400 newmax = ((oldmax + 1) * 2) - 1;
401
402 if (newmax <= IP_MAX_MEMBERSHIPS) {
403 nmships = (struct in_multi **)realloc(omships,
404 sizeof(struct in_multi *) * newmax, M_IPMOPTS, M_NOWAIT);
405 nmfilters = (struct in_mfilter *)realloc(omfilters,
406 sizeof(struct in_mfilter) * newmax, M_INMFILTER, M_NOWAIT);
407 if (nmships != NULL && nmfilters != NULL) {
408 /* Initialize newly allocated source filter heads. */
409 for (idx = oldmax; idx < newmax; idx++) {
410 imf_init(&nmfilters[idx], MCAST_UNDEFINED,
411 MCAST_EXCLUDE);
412 }
413 imo->imo_max_memberships = newmax;
414 imo->imo_membership = nmships;
415 imo->imo_mfilters = nmfilters;
416 }
417 }
418
419 if (nmships == NULL || nmfilters == NULL) {
420 if (nmships != NULL)
421 free(nmships, M_IPMOPTS);
422 if (nmfilters != NULL)
423 free(nmfilters, M_INMFILTER);
424 return (ETOOMANYREFS);
425 }
426
427 return (0);
428 }
429
430 /*
431 * Find an IPv4 multicast group entry for this ip_moptions instance
432 * which matches the specified group, and optionally an interface.
433 * Return its index into the array, or -1 if not found.
434 */
435 static size_t
436 imo_match_group(const struct ip_moptions *imo, const struct ifnet *ifp,
437 const struct sockaddr *group)
438 {
439 const struct sockaddr_in *gsin;
440 struct in_multi **pinm;
441 int idx;
442 int nmships;
443
444 gsin = (const struct sockaddr_in *)group;
445
446 /* The imo_membership array may be lazy allocated. */
447 if (imo->imo_membership == NULL || imo->imo_num_memberships == 0)
448 return (-1);
449
450 nmships = imo->imo_num_memberships;
451 pinm = &imo->imo_membership[0];
452 for (idx = 0; idx < nmships; idx++, pinm++) {
453 if (*pinm == NULL)
454 continue;
455 if ((ifp == NULL || ((*pinm)->inm_ifp == ifp)) &&
456 in_hosteq((*pinm)->inm_addr, gsin->sin_addr)) {
457 break;
458 }
459 }
460 if (idx >= nmships)
461 idx = -1;
462
463 return (idx);
464 }
465
466 /*
467 * Find an IPv4 multicast source entry for this imo which matches
468 * the given group index for this socket, and source address.
469 *
470 * NOTE: This does not check if the entry is in-mode, merely if
471 * it exists, which may not be the desired behaviour.
472 */
473 static struct in_msource *
474 imo_match_source(const struct ip_moptions *imo, const size_t gidx,
475 const struct sockaddr *src)
476 {
477 struct ip_msource find;
478 struct in_mfilter *imf;
479 struct ip_msource *ims;
480 const sockunion_t *psa;
481
482 KASSERT(src->sa_family == AF_INET, ("%s: !AF_INET", __func__));
483 KASSERT(gidx != -1 && gidx < imo->imo_num_memberships,
484 ("%s: invalid index %d\n", __func__, (int)gidx));
485
486 /* The imo_mfilters array may be lazy allocated. */
487 if (imo->imo_mfilters == NULL)
488 return (NULL);
489 imf = &imo->imo_mfilters[gidx];
490
491 /* Source trees are keyed in host byte order. */
492 psa = (const sockunion_t *)src;
493 find.ims_haddr = ntohl(psa->sin.sin_addr.s_addr);
494 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
495
496 return ((struct in_msource *)ims);
497 }
498
499 /*
500 * Perform filtering for multicast datagrams on a socket by group and source.
501 *
502 * Returns 0 if a datagram should be allowed through, or various error codes
503 * if the socket was not a member of the group, or the source was muted, etc.
504 */
505 int
506 imo_multi_filter(const struct ip_moptions *imo, const struct ifnet *ifp,
507 const struct sockaddr *group, const struct sockaddr *src)
508 {
509 size_t gidx;
510 struct in_msource *ims;
511 int mode;
512
513 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
514
515 gidx = imo_match_group(imo, ifp, group);
516 if (gidx == -1)
517 return (MCAST_NOTGMEMBER);
518
519 /*
520 * Check if the source was included in an (S,G) join.
521 * Allow reception on exclusive memberships by default,
522 * reject reception on inclusive memberships by default.
523 * Exclude source only if an in-mode exclude filter exists.
524 * Include source only if an in-mode include filter exists.
525 * NOTE: We are comparing group state here at IGMP t1 (now)
526 * with socket-layer t0 (since last downcall).
527 */
528 mode = imo->imo_mfilters[gidx].imf_st[1];
529 ims = imo_match_source(imo, gidx, src);
530
531 if ((ims == NULL && mode == MCAST_INCLUDE) ||
532 (ims != NULL && ims->imsl_st[0] != mode))
533 return (MCAST_NOTSMEMBER);
534
535 return (MCAST_PASS);
536 }
537
538 /*
539 * Find and return a reference to an in_multi record for (ifp, group),
540 * and bump its reference count.
541 * If one does not exist, try to allocate it, and update link-layer multicast
542 * filters on ifp to listen for group.
543 * Assumes the IN_MULTI lock is held across the call.
544 * Return 0 if successful, otherwise return an appropriate error code.
545 */
546 static int
547 in_getmulti(struct ifnet *ifp, const struct in_addr *group,
548 struct in_multi **pinm)
549 {
550 struct sockaddr_in gsin;
551 struct ifmultiaddr *ifma;
552 struct in_ifinfo *ii;
553 struct in_multi *inm;
554 int error;
555
556 IN_MULTI_LOCK_ASSERT();
557
558 ii = (struct in_ifinfo *)ifp->if_afdata[AF_INET];
559 IN_MULTI_LIST_LOCK();
560 inm = inm_lookup(ifp, *group);
561 if (inm != NULL) {
562 /*
563 * If we already joined this group, just bump the
564 * refcount and return it.
565 */
566 KASSERT(inm->inm_refcount >= 1,
567 ("%s: bad refcount %d", __func__, inm->inm_refcount));
568 inm_acquire_locked(inm);
569 *pinm = inm;
570 }
571 IN_MULTI_LIST_UNLOCK();
572 if (inm != NULL)
573 return (0);
574
575 memset(&gsin, 0, sizeof(gsin));
576 gsin.sin_family = AF_INET;
577 gsin.sin_len = sizeof(struct sockaddr_in);
578 gsin.sin_addr = *group;
579
580 /*
581 * Check if a link-layer group is already associated
582 * with this network-layer group on the given ifnet.
583 */
584 error = if_addmulti(ifp, (struct sockaddr *)&gsin, &ifma);
585 if (error != 0)
586 return (error);
587
588 /* XXX ifma_protospec must be covered by IF_ADDR_LOCK */
589 IN_MULTI_LIST_LOCK();
590 IF_ADDR_WLOCK(ifp);
591
592 /*
593 * If something other than netinet is occupying the link-layer
594 * group, print a meaningful error message and back out of
595 * the allocation.
596 * Otherwise, bump the refcount on the existing network-layer
597 * group association and return it.
598 */
599 if (ifma->ifma_protospec != NULL) {
600 inm = (struct in_multi *)ifma->ifma_protospec;
601 #ifdef INVARIANTS
602 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
603 __func__));
604 KASSERT(ifma->ifma_addr->sa_family == AF_INET,
605 ("%s: ifma not AF_INET", __func__));
606 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
607 if (inm->inm_ifma != ifma || inm->inm_ifp != ifp ||
608 !in_hosteq(inm->inm_addr, *group)) {
609 char addrbuf[INET_ADDRSTRLEN];
610
611 panic("%s: ifma %p is inconsistent with %p (%s)",
612 __func__, ifma, inm, inet_ntoa_r(*group, addrbuf));
613 }
614 #endif
615 inm_acquire_locked(inm);
616 *pinm = inm;
617 goto out_locked;
618 }
619
620 IF_ADDR_WLOCK_ASSERT(ifp);
621
622 /*
623 * A new in_multi record is needed; allocate and initialize it.
624 * We DO NOT perform an IGMP join as the in_ layer may need to
625 * push an initial source list down to IGMP to support SSM.
626 *
627 * The initial source filter state is INCLUDE, {} as per the RFC.
628 */
629 inm = malloc(sizeof(*inm), M_IPMADDR, M_NOWAIT | M_ZERO);
630 if (inm == NULL) {
631 IF_ADDR_WUNLOCK(ifp);
632 IN_MULTI_LIST_UNLOCK();
633 if_delmulti_ifma(ifma);
634 return (ENOMEM);
635 }
636 inm->inm_addr = *group;
637 inm->inm_ifp = ifp;
638 inm->inm_igi = ii->ii_igmp;
639 inm->inm_ifma = ifma;
640 inm->inm_refcount = 1;
641 inm->inm_state = IGMP_NOT_MEMBER;
642 mbufq_init(&inm->inm_scq, IGMP_MAX_STATE_CHANGES);
643 inm->inm_st[0].iss_fmode = MCAST_UNDEFINED;
644 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
645 RB_INIT(&inm->inm_srcs);
646
647 ifma->ifma_protospec = inm;
648
649 *pinm = inm;
650 out_locked:
651 IF_ADDR_WUNLOCK(ifp);
652 IN_MULTI_LIST_UNLOCK();
653 return (0);
654 }
655
656 /*
657 * Drop a reference to an in_multi record.
658 *
659 * If the refcount drops to 0, free the in_multi record and
660 * delete the underlying link-layer membership.
661 */
662 static void
663 inm_release(struct in_multi *inm)
664 {
665 struct ifmultiaddr *ifma;
666 struct ifnet *ifp;
667
668 CTR2(KTR_IGMPV3, "%s: refcount is %d", __func__, inm->inm_refcount);
669 MPASS(inm->inm_refcount == 0);
670 CTR2(KTR_IGMPV3, "%s: freeing inm %p", __func__, inm);
671
672 ifma = inm->inm_ifma;
673 ifp = inm->inm_ifp;
674
675 /* XXX this access is not covered by IF_ADDR_LOCK */
676 CTR2(KTR_IGMPV3, "%s: purging ifma %p", __func__, ifma);
677 if (ifp != NULL) {
678 CURVNET_SET(ifp->if_vnet);
679 inm_purge(inm);
680 free(inm, M_IPMADDR);
681 if_delmulti_ifma_flags(ifma, 1);
682 CURVNET_RESTORE();
683 if_rele(ifp);
684 } else {
685 inm_purge(inm);
686 free(inm, M_IPMADDR);
687 if_delmulti_ifma_flags(ifma, 1);
688 }
689 }
690
691 /*
692 * Clear recorded source entries for a group.
693 * Used by the IGMP code. Caller must hold the IN_MULTI lock.
694 * FIXME: Should reap.
695 */
696 void
697 inm_clear_recorded(struct in_multi *inm)
698 {
699 struct ip_msource *ims;
700
701 IN_MULTI_LIST_LOCK_ASSERT();
702
703 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
704 if (ims->ims_stp) {
705 ims->ims_stp = 0;
706 --inm->inm_st[1].iss_rec;
707 }
708 }
709 KASSERT(inm->inm_st[1].iss_rec == 0,
710 ("%s: iss_rec %d not 0", __func__, inm->inm_st[1].iss_rec));
711 }
712
713 /*
714 * Record a source as pending for a Source-Group IGMPv3 query.
715 * This lives here as it modifies the shared tree.
716 *
717 * inm is the group descriptor.
718 * naddr is the address of the source to record in network-byte order.
719 *
720 * If the net.inet.igmp.sgalloc sysctl is non-zero, we will
721 * lazy-allocate a source node in response to an SG query.
722 * Otherwise, no allocation is performed. This saves some memory
723 * with the trade-off that the source will not be reported to the
724 * router if joined in the window between the query response and
725 * the group actually being joined on the local host.
726 *
727 * VIMAGE: XXX: Currently the igmp_sgalloc feature has been removed.
728 * This turns off the allocation of a recorded source entry if
729 * the group has not been joined.
730 *
731 * Return 0 if the source didn't exist or was already marked as recorded.
732 * Return 1 if the source was marked as recorded by this function.
733 * Return <0 if any error occurred (negated errno code).
734 */
735 int
736 inm_record_source(struct in_multi *inm, const in_addr_t naddr)
737 {
738 struct ip_msource find;
739 struct ip_msource *ims, *nims;
740
741 IN_MULTI_LIST_LOCK_ASSERT();
742
743 find.ims_haddr = ntohl(naddr);
744 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
745 if (ims && ims->ims_stp)
746 return (0);
747 if (ims == NULL) {
748 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
749 return (-ENOSPC);
750 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
751 M_NOWAIT | M_ZERO);
752 if (nims == NULL)
753 return (-ENOMEM);
754 nims->ims_haddr = find.ims_haddr;
755 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
756 ++inm->inm_nsrc;
757 ims = nims;
758 }
759
760 /*
761 * Mark the source as recorded and update the recorded
762 * source count.
763 */
764 ++ims->ims_stp;
765 ++inm->inm_st[1].iss_rec;
766
767 return (1);
768 }
769
770 /*
771 * Return a pointer to an in_msource owned by an in_mfilter,
772 * given its source address.
773 * Lazy-allocate if needed. If this is a new entry its filter state is
774 * undefined at t0.
775 *
776 * imf is the filter set being modified.
777 * haddr is the source address in *host* byte-order.
778 *
779 * SMPng: May be called with locks held; malloc must not block.
780 */
781 static int
782 imf_get_source(struct in_mfilter *imf, const struct sockaddr_in *psin,
783 struct in_msource **plims)
784 {
785 struct ip_msource find;
786 struct ip_msource *ims, *nims;
787 struct in_msource *lims;
788 int error;
789
790 error = 0;
791 ims = NULL;
792 lims = NULL;
793
794 /* key is host byte order */
795 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
796 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
797 lims = (struct in_msource *)ims;
798 if (lims == NULL) {
799 if (imf->imf_nsrc == in_mcast_maxsocksrc)
800 return (ENOSPC);
801 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
802 M_NOWAIT | M_ZERO);
803 if (nims == NULL)
804 return (ENOMEM);
805 lims = (struct in_msource *)nims;
806 lims->ims_haddr = find.ims_haddr;
807 lims->imsl_st[0] = MCAST_UNDEFINED;
808 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
809 ++imf->imf_nsrc;
810 }
811
812 *plims = lims;
813
814 return (error);
815 }
816
817 /*
818 * Graft a source entry into an existing socket-layer filter set,
819 * maintaining any required invariants and checking allocations.
820 *
821 * The source is marked as being in the new filter mode at t1.
822 *
823 * Return the pointer to the new node, otherwise return NULL.
824 */
825 static struct in_msource *
826 imf_graft(struct in_mfilter *imf, const uint8_t st1,
827 const struct sockaddr_in *psin)
828 {
829 struct ip_msource *nims;
830 struct in_msource *lims;
831
832 nims = malloc(sizeof(struct in_msource), M_INMFILTER,
833 M_NOWAIT | M_ZERO);
834 if (nims == NULL)
835 return (NULL);
836 lims = (struct in_msource *)nims;
837 lims->ims_haddr = ntohl(psin->sin_addr.s_addr);
838 lims->imsl_st[0] = MCAST_UNDEFINED;
839 lims->imsl_st[1] = st1;
840 RB_INSERT(ip_msource_tree, &imf->imf_sources, nims);
841 ++imf->imf_nsrc;
842
843 return (lims);
844 }
845
846 /*
847 * Prune a source entry from an existing socket-layer filter set,
848 * maintaining any required invariants and checking allocations.
849 *
850 * The source is marked as being left at t1, it is not freed.
851 *
852 * Return 0 if no error occurred, otherwise return an errno value.
853 */
854 static int
855 imf_prune(struct in_mfilter *imf, const struct sockaddr_in *psin)
856 {
857 struct ip_msource find;
858 struct ip_msource *ims;
859 struct in_msource *lims;
860
861 /* key is host byte order */
862 find.ims_haddr = ntohl(psin->sin_addr.s_addr);
863 ims = RB_FIND(ip_msource_tree, &imf->imf_sources, &find);
864 if (ims == NULL)
865 return (ENOENT);
866 lims = (struct in_msource *)ims;
867 lims->imsl_st[1] = MCAST_UNDEFINED;
868 return (0);
869 }
870
871 /*
872 * Revert socket-layer filter set deltas at t1 to t0 state.
873 */
874 static void
875 imf_rollback(struct in_mfilter *imf)
876 {
877 struct ip_msource *ims, *tims;
878 struct in_msource *lims;
879
880 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
881 lims = (struct in_msource *)ims;
882 if (lims->imsl_st[0] == lims->imsl_st[1]) {
883 /* no change at t1 */
884 continue;
885 } else if (lims->imsl_st[0] != MCAST_UNDEFINED) {
886 /* revert change to existing source at t1 */
887 lims->imsl_st[1] = lims->imsl_st[0];
888 } else {
889 /* revert source added t1 */
890 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
891 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
892 free(ims, M_INMFILTER);
893 imf->imf_nsrc--;
894 }
895 }
896 imf->imf_st[1] = imf->imf_st[0];
897 }
898
899 /*
900 * Mark socket-layer filter set as INCLUDE {} at t1.
901 */
902 static void
903 imf_leave(struct in_mfilter *imf)
904 {
905 struct ip_msource *ims;
906 struct in_msource *lims;
907
908 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
909 lims = (struct in_msource *)ims;
910 lims->imsl_st[1] = MCAST_UNDEFINED;
911 }
912 imf->imf_st[1] = MCAST_INCLUDE;
913 }
914
915 /*
916 * Mark socket-layer filter set deltas as committed.
917 */
918 static void
919 imf_commit(struct in_mfilter *imf)
920 {
921 struct ip_msource *ims;
922 struct in_msource *lims;
923
924 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
925 lims = (struct in_msource *)ims;
926 lims->imsl_st[0] = lims->imsl_st[1];
927 }
928 imf->imf_st[0] = imf->imf_st[1];
929 }
930
931 /*
932 * Reap unreferenced sources from socket-layer filter set.
933 */
934 static void
935 imf_reap(struct in_mfilter *imf)
936 {
937 struct ip_msource *ims, *tims;
938 struct in_msource *lims;
939
940 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
941 lims = (struct in_msource *)ims;
942 if ((lims->imsl_st[0] == MCAST_UNDEFINED) &&
943 (lims->imsl_st[1] == MCAST_UNDEFINED)) {
944 CTR2(KTR_IGMPV3, "%s: free lims %p", __func__, ims);
945 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
946 free(ims, M_INMFILTER);
947 imf->imf_nsrc--;
948 }
949 }
950 }
951
952 /*
953 * Purge socket-layer filter set.
954 */
955 static void
956 imf_purge(struct in_mfilter *imf)
957 {
958 struct ip_msource *ims, *tims;
959
960 RB_FOREACH_SAFE(ims, ip_msource_tree, &imf->imf_sources, tims) {
961 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
962 RB_REMOVE(ip_msource_tree, &imf->imf_sources, ims);
963 free(ims, M_INMFILTER);
964 imf->imf_nsrc--;
965 }
966 imf->imf_st[0] = imf->imf_st[1] = MCAST_UNDEFINED;
967 KASSERT(RB_EMPTY(&imf->imf_sources),
968 ("%s: imf_sources not empty", __func__));
969 }
970
971 /*
972 * Look up a source filter entry for a multicast group.
973 *
974 * inm is the group descriptor to work with.
975 * haddr is the host-byte-order IPv4 address to look up.
976 * noalloc may be non-zero to suppress allocation of sources.
977 * *pims will be set to the address of the retrieved or allocated source.
978 *
979 * SMPng: NOTE: may be called with locks held.
980 * Return 0 if successful, otherwise return a non-zero error code.
981 */
982 static int
983 inm_get_source(struct in_multi *inm, const in_addr_t haddr,
984 const int noalloc, struct ip_msource **pims)
985 {
986 struct ip_msource find;
987 struct ip_msource *ims, *nims;
988
989 find.ims_haddr = haddr;
990 ims = RB_FIND(ip_msource_tree, &inm->inm_srcs, &find);
991 if (ims == NULL && !noalloc) {
992 if (inm->inm_nsrc == in_mcast_maxgrpsrc)
993 return (ENOSPC);
994 nims = malloc(sizeof(struct ip_msource), M_IPMSOURCE,
995 M_NOWAIT | M_ZERO);
996 if (nims == NULL)
997 return (ENOMEM);
998 nims->ims_haddr = haddr;
999 RB_INSERT(ip_msource_tree, &inm->inm_srcs, nims);
1000 ++inm->inm_nsrc;
1001 ims = nims;
1002 #ifdef KTR
1003 CTR3(KTR_IGMPV3, "%s: allocated 0x%08x as %p", __func__,
1004 haddr, ims);
1005 #endif
1006 }
1007
1008 *pims = ims;
1009 return (0);
1010 }
1011
1012 /*
1013 * Merge socket-layer source into IGMP-layer source.
1014 * If rollback is non-zero, perform the inverse of the merge.
1015 */
1016 static void
1017 ims_merge(struct ip_msource *ims, const struct in_msource *lims,
1018 const int rollback)
1019 {
1020 int n = rollback ? -1 : 1;
1021
1022 if (lims->imsl_st[0] == MCAST_EXCLUDE) {
1023 CTR3(KTR_IGMPV3, "%s: t1 ex -= %d on 0x%08x",
1024 __func__, n, ims->ims_haddr);
1025 ims->ims_st[1].ex -= n;
1026 } else if (lims->imsl_st[0] == MCAST_INCLUDE) {
1027 CTR3(KTR_IGMPV3, "%s: t1 in -= %d on 0x%08x",
1028 __func__, n, ims->ims_haddr);
1029 ims->ims_st[1].in -= n;
1030 }
1031
1032 if (lims->imsl_st[1] == MCAST_EXCLUDE) {
1033 CTR3(KTR_IGMPV3, "%s: t1 ex += %d on 0x%08x",
1034 __func__, n, ims->ims_haddr);
1035 ims->ims_st[1].ex += n;
1036 } else if (lims->imsl_st[1] == MCAST_INCLUDE) {
1037 CTR3(KTR_IGMPV3, "%s: t1 in += %d on 0x%08x",
1038 __func__, n, ims->ims_haddr);
1039 ims->ims_st[1].in += n;
1040 }
1041 }
1042
1043 /*
1044 * Atomically update the global in_multi state, when a membership's
1045 * filter list is being updated in any way.
1046 *
1047 * imf is the per-inpcb-membership group filter pointer.
1048 * A fake imf may be passed for in-kernel consumers.
1049 *
1050 * XXX This is a candidate for a set-symmetric-difference style loop
1051 * which would eliminate the repeated lookup from root of ims nodes,
1052 * as they share the same key space.
1053 *
1054 * If any error occurred this function will back out of refcounts
1055 * and return a non-zero value.
1056 */
1057 static int
1058 inm_merge(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1059 {
1060 struct ip_msource *ims, *nims;
1061 struct in_msource *lims;
1062 int schanged, error;
1063 int nsrc0, nsrc1;
1064
1065 schanged = 0;
1066 error = 0;
1067 nsrc1 = nsrc0 = 0;
1068 IN_MULTI_LIST_LOCK_ASSERT();
1069
1070 /*
1071 * Update the source filters first, as this may fail.
1072 * Maintain count of in-mode filters at t0, t1. These are
1073 * used to work out if we transition into ASM mode or not.
1074 * Maintain a count of source filters whose state was
1075 * actually modified by this operation.
1076 */
1077 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1078 lims = (struct in_msource *)ims;
1079 if (lims->imsl_st[0] == imf->imf_st[0]) nsrc0++;
1080 if (lims->imsl_st[1] == imf->imf_st[1]) nsrc1++;
1081 if (lims->imsl_st[0] == lims->imsl_st[1]) continue;
1082 error = inm_get_source(inm, lims->ims_haddr, 0, &nims);
1083 ++schanged;
1084 if (error)
1085 break;
1086 ims_merge(nims, lims, 0);
1087 }
1088 if (error) {
1089 struct ip_msource *bims;
1090
1091 RB_FOREACH_REVERSE_FROM(ims, ip_msource_tree, nims) {
1092 lims = (struct in_msource *)ims;
1093 if (lims->imsl_st[0] == lims->imsl_st[1])
1094 continue;
1095 (void)inm_get_source(inm, lims->ims_haddr, 1, &bims);
1096 if (bims == NULL)
1097 continue;
1098 ims_merge(bims, lims, 1);
1099 }
1100 goto out_reap;
1101 }
1102
1103 CTR3(KTR_IGMPV3, "%s: imf filters in-mode: %d at t0, %d at t1",
1104 __func__, nsrc0, nsrc1);
1105
1106 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1107 if (imf->imf_st[0] == imf->imf_st[1] &&
1108 imf->imf_st[1] == MCAST_INCLUDE) {
1109 if (nsrc1 == 0) {
1110 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1111 --inm->inm_st[1].iss_in;
1112 }
1113 }
1114
1115 /* Handle filter mode transition on socket. */
1116 if (imf->imf_st[0] != imf->imf_st[1]) {
1117 CTR3(KTR_IGMPV3, "%s: imf transition %d to %d",
1118 __func__, imf->imf_st[0], imf->imf_st[1]);
1119
1120 if (imf->imf_st[0] == MCAST_EXCLUDE) {
1121 CTR1(KTR_IGMPV3, "%s: --ex on inm at t1", __func__);
1122 --inm->inm_st[1].iss_ex;
1123 } else if (imf->imf_st[0] == MCAST_INCLUDE) {
1124 CTR1(KTR_IGMPV3, "%s: --in on inm at t1", __func__);
1125 --inm->inm_st[1].iss_in;
1126 }
1127
1128 if (imf->imf_st[1] == MCAST_EXCLUDE) {
1129 CTR1(KTR_IGMPV3, "%s: ex++ on inm at t1", __func__);
1130 inm->inm_st[1].iss_ex++;
1131 } else if (imf->imf_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1132 CTR1(KTR_IGMPV3, "%s: in++ on inm at t1", __func__);
1133 inm->inm_st[1].iss_in++;
1134 }
1135 }
1136
1137 /*
1138 * Track inm filter state in terms of listener counts.
1139 * If there are any exclusive listeners, stack-wide
1140 * membership is exclusive.
1141 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1142 * If no listeners remain, state is undefined at t1,
1143 * and the IGMP lifecycle for this group should finish.
1144 */
1145 if (inm->inm_st[1].iss_ex > 0) {
1146 CTR1(KTR_IGMPV3, "%s: transition to EX", __func__);
1147 inm->inm_st[1].iss_fmode = MCAST_EXCLUDE;
1148 } else if (inm->inm_st[1].iss_in > 0) {
1149 CTR1(KTR_IGMPV3, "%s: transition to IN", __func__);
1150 inm->inm_st[1].iss_fmode = MCAST_INCLUDE;
1151 } else {
1152 CTR1(KTR_IGMPV3, "%s: transition to UNDEF", __func__);
1153 inm->inm_st[1].iss_fmode = MCAST_UNDEFINED;
1154 }
1155
1156 /* Decrement ASM listener count on transition out of ASM mode. */
1157 if (imf->imf_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1158 if ((imf->imf_st[1] != MCAST_EXCLUDE) ||
1159 (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1160 CTR1(KTR_IGMPV3, "%s: --asm on inm at t1", __func__);
1161 --inm->inm_st[1].iss_asm;
1162 }
1163 }
1164
1165 /* Increment ASM listener count on transition to ASM mode. */
1166 if (imf->imf_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1167 CTR1(KTR_IGMPV3, "%s: asm++ on inm at t1", __func__);
1168 inm->inm_st[1].iss_asm++;
1169 }
1170
1171 CTR3(KTR_IGMPV3, "%s: merged imf %p to inm %p", __func__, imf, inm);
1172 inm_print(inm);
1173
1174 out_reap:
1175 if (schanged > 0) {
1176 CTR1(KTR_IGMPV3, "%s: sources changed; reaping", __func__);
1177 inm_reap(inm);
1178 }
1179 return (error);
1180 }
1181
1182 /*
1183 * Mark an in_multi's filter set deltas as committed.
1184 * Called by IGMP after a state change has been enqueued.
1185 */
1186 void
1187 inm_commit(struct in_multi *inm)
1188 {
1189 struct ip_msource *ims;
1190
1191 CTR2(KTR_IGMPV3, "%s: commit inm %p", __func__, inm);
1192 CTR1(KTR_IGMPV3, "%s: pre commit:", __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 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1210 if (ims->ims_st[0].ex > 0 || ims->ims_st[0].in > 0 ||
1211 ims->ims_st[1].ex > 0 || ims->ims_st[1].in > 0 ||
1212 ims->ims_stp != 0)
1213 continue;
1214 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1215 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1216 free(ims, M_IPMSOURCE);
1217 inm->inm_nsrc--;
1218 }
1219 }
1220
1221 /*
1222 * Purge all source nodes from an in_multi's filter set.
1223 */
1224 static void
1225 inm_purge(struct in_multi *inm)
1226 {
1227 struct ip_msource *ims, *tims;
1228
1229 RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, tims) {
1230 CTR2(KTR_IGMPV3, "%s: free ims %p", __func__, ims);
1231 RB_REMOVE(ip_msource_tree, &inm->inm_srcs, ims);
1232 free(ims, M_IPMSOURCE);
1233 inm->inm_nsrc--;
1234 }
1235 }
1236
1237 /*
1238 * Join a multicast group; unlocked entry point.
1239 *
1240 * SMPng: XXX: in_joingroup() is called from in_control() when Giant
1241 * is not held. Fortunately, ifp is unlikely to have been detached
1242 * at this point, so we assume it's OK to recurse.
1243 */
1244 int
1245 in_joingroup(struct ifnet *ifp, const struct in_addr *gina,
1246 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1247 {
1248 int error;
1249
1250 IN_MULTI_LOCK();
1251 error = in_joingroup_locked(ifp, gina, imf, pinm);
1252 IN_MULTI_UNLOCK();
1253
1254 return (error);
1255 }
1256
1257 /*
1258 * Join a multicast group; real entry point.
1259 *
1260 * Only preserves atomicity at inm level.
1261 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1262 *
1263 * If the IGMP downcall fails, the group is not joined, and an error
1264 * code is returned.
1265 */
1266 int
1267 in_joingroup_locked(struct ifnet *ifp, const struct in_addr *gina,
1268 /*const*/ struct in_mfilter *imf, struct in_multi **pinm)
1269 {
1270 struct in_mfilter timf;
1271 struct in_multi *inm;
1272 int error;
1273
1274 IN_MULTI_LOCK_ASSERT();
1275 IN_MULTI_LIST_UNLOCK_ASSERT();
1276
1277 CTR4(KTR_IGMPV3, "%s: join 0x%08x on %p(%s))", __func__,
1278 ntohl(gina->s_addr), ifp, ifp->if_xname);
1279
1280 error = 0;
1281 inm = NULL;
1282
1283 /*
1284 * If no imf was specified (i.e. kernel consumer),
1285 * fake one up and assume it is an ASM join.
1286 */
1287 if (imf == NULL) {
1288 imf_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1289 imf = &timf;
1290 }
1291
1292 error = in_getmulti(ifp, gina, &inm);
1293 if (error) {
1294 CTR1(KTR_IGMPV3, "%s: in_getmulti() failure", __func__);
1295 return (error);
1296 }
1297 IN_MULTI_LIST_LOCK();
1298 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1299 error = inm_merge(inm, imf);
1300 if (error) {
1301 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1302 goto out_inm_release;
1303 }
1304
1305 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1306 error = igmp_change_state(inm);
1307 if (error) {
1308 CTR1(KTR_IGMPV3, "%s: failed to update source", __func__);
1309 goto out_inm_release;
1310 }
1311
1312 out_inm_release:
1313 if (error) {
1314
1315 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1316 inm_release_deferred(inm);
1317 } else {
1318 *pinm = inm;
1319 }
1320 IN_MULTI_LIST_UNLOCK();
1321
1322 return (error);
1323 }
1324
1325 /*
1326 * Leave a multicast group; unlocked entry point.
1327 */
1328 int
1329 in_leavegroup(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1330 {
1331 int error;
1332
1333 IN_MULTI_LOCK();
1334 error = in_leavegroup_locked(inm, imf);
1335 IN_MULTI_UNLOCK();
1336
1337 return (error);
1338 }
1339
1340 /*
1341 * Leave a multicast group; real entry point.
1342 * All source filters will be expunged.
1343 *
1344 * Only preserves atomicity at inm level.
1345 *
1346 * Holding the write lock for the INP which contains imf
1347 * is highly advisable. We can't assert for it as imf does not
1348 * contain a back-pointer to the owning inp.
1349 *
1350 * Note: This is not the same as inm_release(*) as this function also
1351 * makes a state change downcall into IGMP.
1352 */
1353 int
1354 in_leavegroup_locked(struct in_multi *inm, /*const*/ struct in_mfilter *imf)
1355 {
1356 struct in_mfilter timf;
1357 int error;
1358
1359 error = 0;
1360
1361 IN_MULTI_LOCK_ASSERT();
1362 IN_MULTI_LIST_UNLOCK_ASSERT();
1363
1364 CTR5(KTR_IGMPV3, "%s: leave inm %p, 0x%08x/%s, imf %p", __func__,
1365 inm, ntohl(inm->inm_addr.s_addr),
1366 (inm_is_ifp_detached(inm) ? "null" : inm->inm_ifp->if_xname),
1367 imf);
1368
1369 /*
1370 * If no imf was specified (i.e. kernel consumer),
1371 * fake one up and assume it is an ASM join.
1372 */
1373 if (imf == NULL) {
1374 imf_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1375 imf = &timf;
1376 }
1377
1378 /*
1379 * Begin state merge transaction at IGMP layer.
1380 *
1381 * As this particular invocation should not cause any memory
1382 * to be allocated, and there is no opportunity to roll back
1383 * the transaction, it MUST NOT fail.
1384 */
1385 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1386 IN_MULTI_LIST_LOCK();
1387 error = inm_merge(inm, imf);
1388 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1389
1390 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1391 CURVNET_SET(inm->inm_ifp->if_vnet);
1392 error = igmp_change_state(inm);
1393 IF_ADDR_WLOCK(inm->inm_ifp);
1394 inm_release_deferred(inm);
1395 IF_ADDR_WUNLOCK(inm->inm_ifp);
1396 IN_MULTI_LIST_UNLOCK();
1397 CURVNET_RESTORE();
1398 if (error)
1399 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1400
1401 CTR2(KTR_IGMPV3, "%s: dropping ref on %p", __func__, inm);
1402
1403 return (error);
1404 }
1405
1406 /*#ifndef BURN_BRIDGES*/
1407 /*
1408 * Join an IPv4 multicast group in (*,G) exclusive mode.
1409 * The group must be a 224.0.0.0/24 link-scope group.
1410 * This KPI is for legacy kernel consumers only.
1411 */
1412 struct in_multi *
1413 in_addmulti(struct in_addr *ap, struct ifnet *ifp)
1414 {
1415 struct in_multi *pinm;
1416 int error;
1417 #ifdef INVARIANTS
1418 char addrbuf[INET_ADDRSTRLEN];
1419 #endif
1420
1421 KASSERT(IN_LOCAL_GROUP(ntohl(ap->s_addr)),
1422 ("%s: %s not in 224.0.0.0/24", __func__,
1423 inet_ntoa_r(*ap, addrbuf)));
1424
1425 error = in_joingroup(ifp, ap, NULL, &pinm);
1426 if (error != 0)
1427 pinm = NULL;
1428
1429 return (pinm);
1430 }
1431
1432 /*
1433 * Block or unblock an ASM multicast source on an inpcb.
1434 * This implements the delta-based API described in RFC 3678.
1435 *
1436 * The delta-based API applies only to exclusive-mode memberships.
1437 * An IGMP downcall will be performed.
1438 *
1439 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1440 *
1441 * Return 0 if successful, otherwise return an appropriate error code.
1442 */
1443 static int
1444 inp_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1445 {
1446 struct group_source_req gsr;
1447 sockunion_t *gsa, *ssa;
1448 struct ifnet *ifp;
1449 struct in_mfilter *imf;
1450 struct ip_moptions *imo;
1451 struct in_msource *ims;
1452 struct in_multi *inm;
1453 size_t idx;
1454 uint16_t fmode;
1455 int error, doblock;
1456
1457 ifp = NULL;
1458 error = 0;
1459 doblock = 0;
1460
1461 memset(&gsr, 0, sizeof(struct group_source_req));
1462 gsa = (sockunion_t *)&gsr.gsr_group;
1463 ssa = (sockunion_t *)&gsr.gsr_source;
1464
1465 switch (sopt->sopt_name) {
1466 case IP_BLOCK_SOURCE:
1467 case IP_UNBLOCK_SOURCE: {
1468 struct ip_mreq_source mreqs;
1469
1470 error = sooptcopyin(sopt, &mreqs,
1471 sizeof(struct ip_mreq_source),
1472 sizeof(struct ip_mreq_source));
1473 if (error)
1474 return (error);
1475
1476 gsa->sin.sin_family = AF_INET;
1477 gsa->sin.sin_len = sizeof(struct sockaddr_in);
1478 gsa->sin.sin_addr = mreqs.imr_multiaddr;
1479
1480 ssa->sin.sin_family = AF_INET;
1481 ssa->sin.sin_len = sizeof(struct sockaddr_in);
1482 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
1483
1484 if (!in_nullhost(mreqs.imr_interface))
1485 INADDR_TO_IFP(mreqs.imr_interface, ifp);
1486
1487 if (sopt->sopt_name == IP_BLOCK_SOURCE)
1488 doblock = 1;
1489
1490 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
1491 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
1492 break;
1493 }
1494
1495 case MCAST_BLOCK_SOURCE:
1496 case MCAST_UNBLOCK_SOURCE:
1497 error = sooptcopyin(sopt, &gsr,
1498 sizeof(struct group_source_req),
1499 sizeof(struct group_source_req));
1500 if (error)
1501 return (error);
1502
1503 if (gsa->sin.sin_family != AF_INET ||
1504 gsa->sin.sin_len != sizeof(struct sockaddr_in))
1505 return (EINVAL);
1506
1507 if (ssa->sin.sin_family != AF_INET ||
1508 ssa->sin.sin_len != sizeof(struct sockaddr_in))
1509 return (EINVAL);
1510
1511 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1512 return (EADDRNOTAVAIL);
1513
1514 ifp = ifnet_byindex(gsr.gsr_interface);
1515
1516 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1517 doblock = 1;
1518 break;
1519
1520 default:
1521 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
1522 __func__, sopt->sopt_name);
1523 return (EOPNOTSUPP);
1524 break;
1525 }
1526
1527 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
1528 return (EINVAL);
1529
1530 /*
1531 * Check if we are actually a member of this group.
1532 */
1533 imo = inp_findmoptions(inp);
1534 idx = imo_match_group(imo, ifp, &gsa->sa);
1535 if (idx == -1 || imo->imo_mfilters == NULL) {
1536 error = EADDRNOTAVAIL;
1537 goto out_inp_locked;
1538 }
1539
1540 KASSERT(imo->imo_mfilters != NULL,
1541 ("%s: imo_mfilters not allocated", __func__));
1542 imf = &imo->imo_mfilters[idx];
1543 inm = imo->imo_membership[idx];
1544
1545 /*
1546 * Attempting to use the delta-based API on an
1547 * non exclusive-mode membership is an error.
1548 */
1549 fmode = imf->imf_st[0];
1550 if (fmode != MCAST_EXCLUDE) {
1551 error = EINVAL;
1552 goto out_inp_locked;
1553 }
1554
1555 /*
1556 * Deal with error cases up-front:
1557 * Asked to block, but already blocked; or
1558 * Asked to unblock, but nothing to unblock.
1559 * If adding a new block entry, allocate it.
1560 */
1561 ims = imo_match_source(imo, idx, &ssa->sa);
1562 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1563 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent", __func__,
1564 ntohl(ssa->sin.sin_addr.s_addr), doblock ? "" : "not ");
1565 error = EADDRNOTAVAIL;
1566 goto out_inp_locked;
1567 }
1568
1569 INP_WLOCK_ASSERT(inp);
1570
1571 /*
1572 * Begin state merge transaction at socket layer.
1573 */
1574 if (doblock) {
1575 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
1576 ims = imf_graft(imf, fmode, &ssa->sin);
1577 if (ims == NULL)
1578 error = ENOMEM;
1579 } else {
1580 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
1581 error = imf_prune(imf, &ssa->sin);
1582 }
1583
1584 if (error) {
1585 CTR1(KTR_IGMPV3, "%s: merge imf state failed", __func__);
1586 goto out_imf_rollback;
1587 }
1588
1589 /*
1590 * Begin state merge transaction at IGMP layer.
1591 */
1592 IN_MULTI_LOCK();
1593 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
1594 IN_MULTI_LIST_LOCK();
1595 error = inm_merge(inm, imf);
1596 if (error) {
1597 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
1598 IN_MULTI_LIST_UNLOCK();
1599 goto out_in_multi_locked;
1600 }
1601
1602 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
1603 error = igmp_change_state(inm);
1604 IN_MULTI_LIST_UNLOCK();
1605 if (error)
1606 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
1607
1608 out_in_multi_locked:
1609
1610 IN_MULTI_UNLOCK();
1611 out_imf_rollback:
1612 if (error)
1613 imf_rollback(imf);
1614 else
1615 imf_commit(imf);
1616
1617 imf_reap(imf);
1618
1619 out_inp_locked:
1620 INP_WUNLOCK(inp);
1621 return (error);
1622 }
1623
1624 /*
1625 * Given an inpcb, return its multicast options structure pointer. Accepts
1626 * an unlocked inpcb pointer, but will return it locked. May sleep.
1627 *
1628 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1629 * SMPng: NOTE: Returns with the INP write lock held.
1630 */
1631 static struct ip_moptions *
1632 inp_findmoptions(struct inpcb *inp)
1633 {
1634 struct ip_moptions *imo;
1635 struct in_multi **immp;
1636 struct in_mfilter *imfp;
1637 size_t idx;
1638
1639 INP_WLOCK(inp);
1640 if (inp->inp_moptions != NULL)
1641 return (inp->inp_moptions);
1642
1643 INP_WUNLOCK(inp);
1644
1645 imo = malloc(sizeof(*imo), M_IPMOPTS, M_WAITOK);
1646 immp = malloc(sizeof(*immp) * IP_MIN_MEMBERSHIPS, M_IPMOPTS,
1647 M_WAITOK | M_ZERO);
1648 imfp = malloc(sizeof(struct in_mfilter) * IP_MIN_MEMBERSHIPS,
1649 M_INMFILTER, M_WAITOK);
1650
1651 imo->imo_multicast_ifp = NULL;
1652 imo->imo_multicast_addr.s_addr = INADDR_ANY;
1653 imo->imo_multicast_vif = -1;
1654 imo->imo_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1655 imo->imo_multicast_loop = in_mcast_loop;
1656 imo->imo_num_memberships = 0;
1657 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1658 imo->imo_membership = immp;
1659
1660 /* Initialize per-group source filters. */
1661 for (idx = 0; idx < IP_MIN_MEMBERSHIPS; idx++)
1662 imf_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1663 imo->imo_mfilters = imfp;
1664
1665 INP_WLOCK(inp);
1666 if (inp->inp_moptions != NULL) {
1667 free(imfp, M_INMFILTER);
1668 free(immp, M_IPMOPTS);
1669 free(imo, M_IPMOPTS);
1670 return (inp->inp_moptions);
1671 }
1672 inp->inp_moptions = imo;
1673 return (imo);
1674 }
1675
1676 static void
1677 inp_gcmoptions(struct ip_moptions *imo)
1678 {
1679 struct in_mfilter *imf;
1680 struct in_multi *inm;
1681 struct ifnet *ifp;
1682 size_t idx, nmships;
1683
1684 nmships = imo->imo_num_memberships;
1685 for (idx = 0; idx < nmships; ++idx) {
1686 imf = imo->imo_mfilters ? &imo->imo_mfilters[idx] : NULL;
1687 if (imf)
1688 imf_leave(imf);
1689 inm = imo->imo_membership[idx];
1690 ifp = inm->inm_ifp;
1691 if (ifp != NULL) {
1692 CURVNET_SET(ifp->if_vnet);
1693 (void)in_leavegroup(inm, imf);
1694 CURVNET_RESTORE();
1695 } else {
1696 (void)in_leavegroup(inm, imf);
1697 }
1698 if (imf)
1699 imf_purge(imf);
1700 }
1701
1702 if (imo->imo_mfilters)
1703 free(imo->imo_mfilters, M_INMFILTER);
1704 free(imo->imo_membership, M_IPMOPTS);
1705 free(imo, M_IPMOPTS);
1706 }
1707
1708 /*
1709 * Discard the IP multicast options (and source filters). To minimize
1710 * the amount of work done while holding locks such as the INP's
1711 * pcbinfo lock (which is used in the receive path), the free
1712 * operation is deferred to the epoch callback task.
1713 */
1714 void
1715 inp_freemoptions(struct ip_moptions *imo)
1716 {
1717 if (imo == NULL)
1718 return;
1719 inp_gcmoptions(imo);
1720 }
1721
1722 /*
1723 * Atomically get source filters on a socket for an IPv4 multicast group.
1724 * Called with INP lock held; returns with lock released.
1725 */
1726 static int
1727 inp_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1728 {
1729 struct __msfilterreq msfr;
1730 sockunion_t *gsa;
1731 struct ifnet *ifp;
1732 struct ip_moptions *imo;
1733 struct in_mfilter *imf;
1734 struct ip_msource *ims;
1735 struct in_msource *lims;
1736 struct sockaddr_in *psin;
1737 struct sockaddr_storage *ptss;
1738 struct sockaddr_storage *tss;
1739 int error;
1740 size_t idx, nsrcs, ncsrcs;
1741
1742 INP_WLOCK_ASSERT(inp);
1743
1744 imo = inp->inp_moptions;
1745 KASSERT(imo != NULL, ("%s: null ip_moptions", __func__));
1746
1747 INP_WUNLOCK(inp);
1748
1749 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1750 sizeof(struct __msfilterreq));
1751 if (error)
1752 return (error);
1753
1754 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1755 return (EINVAL);
1756
1757 ifp = ifnet_byindex(msfr.msfr_ifindex);
1758 if (ifp == NULL)
1759 return (EINVAL);
1760
1761 INP_WLOCK(inp);
1762
1763 /*
1764 * Lookup group on the socket.
1765 */
1766 gsa = (sockunion_t *)&msfr.msfr_group;
1767 idx = imo_match_group(imo, ifp, &gsa->sa);
1768 if (idx == -1 || imo->imo_mfilters == NULL) {
1769 INP_WUNLOCK(inp);
1770 return (EADDRNOTAVAIL);
1771 }
1772 imf = &imo->imo_mfilters[idx];
1773
1774 /*
1775 * Ignore memberships which are in limbo.
1776 */
1777 if (imf->imf_st[1] == MCAST_UNDEFINED) {
1778 INP_WUNLOCK(inp);
1779 return (EAGAIN);
1780 }
1781 msfr.msfr_fmode = imf->imf_st[1];
1782
1783 /*
1784 * If the user specified a buffer, copy out the source filter
1785 * entries to userland gracefully.
1786 * We only copy out the number of entries which userland
1787 * has asked for, but we always tell userland how big the
1788 * buffer really needs to be.
1789 */
1790 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
1791 msfr.msfr_nsrcs = in_mcast_maxsocksrc;
1792 tss = NULL;
1793 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1794 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1795 M_TEMP, M_NOWAIT | M_ZERO);
1796 if (tss == NULL) {
1797 INP_WUNLOCK(inp);
1798 return (ENOBUFS);
1799 }
1800 }
1801
1802 /*
1803 * Count number of sources in-mode at t0.
1804 * If buffer space exists and remains, copy out source entries.
1805 */
1806 nsrcs = msfr.msfr_nsrcs;
1807 ncsrcs = 0;
1808 ptss = tss;
1809 RB_FOREACH(ims, ip_msource_tree, &imf->imf_sources) {
1810 lims = (struct in_msource *)ims;
1811 if (lims->imsl_st[0] == MCAST_UNDEFINED ||
1812 lims->imsl_st[0] != imf->imf_st[0])
1813 continue;
1814 ++ncsrcs;
1815 if (tss != NULL && nsrcs > 0) {
1816 psin = (struct sockaddr_in *)ptss;
1817 psin->sin_family = AF_INET;
1818 psin->sin_len = sizeof(struct sockaddr_in);
1819 psin->sin_addr.s_addr = htonl(lims->ims_haddr);
1820 psin->sin_port = 0;
1821 ++ptss;
1822 --nsrcs;
1823 }
1824 }
1825
1826 INP_WUNLOCK(inp);
1827
1828 if (tss != NULL) {
1829 error = copyout(tss, msfr.msfr_srcs,
1830 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1831 free(tss, M_TEMP);
1832 if (error)
1833 return (error);
1834 }
1835
1836 msfr.msfr_nsrcs = ncsrcs;
1837 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1838
1839 return (error);
1840 }
1841
1842 /*
1843 * Return the IP multicast options in response to user getsockopt().
1844 */
1845 int
1846 inp_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1847 {
1848 struct rm_priotracker in_ifa_tracker;
1849 struct ip_mreqn mreqn;
1850 struct ip_moptions *imo;
1851 struct ifnet *ifp;
1852 struct in_ifaddr *ia;
1853 int error, optval;
1854 u_char coptval;
1855
1856 INP_WLOCK(inp);
1857 imo = inp->inp_moptions;
1858 /*
1859 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1860 * or is a divert socket, reject it.
1861 */
1862 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1863 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1864 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1865 INP_WUNLOCK(inp);
1866 return (EOPNOTSUPP);
1867 }
1868
1869 error = 0;
1870 switch (sopt->sopt_name) {
1871 case IP_MULTICAST_VIF:
1872 if (imo != NULL)
1873 optval = imo->imo_multicast_vif;
1874 else
1875 optval = -1;
1876 INP_WUNLOCK(inp);
1877 error = sooptcopyout(sopt, &optval, sizeof(int));
1878 break;
1879
1880 case IP_MULTICAST_IF:
1881 memset(&mreqn, 0, sizeof(struct ip_mreqn));
1882 if (imo != NULL) {
1883 ifp = imo->imo_multicast_ifp;
1884 if (!in_nullhost(imo->imo_multicast_addr)) {
1885 mreqn.imr_address = imo->imo_multicast_addr;
1886 } else if (ifp != NULL) {
1887 mreqn.imr_ifindex = ifp->if_index;
1888 NET_EPOCH_ENTER();
1889 IFP_TO_IA(ifp, ia, &in_ifa_tracker);
1890 if (ia != NULL)
1891 mreqn.imr_address =
1892 IA_SIN(ia)->sin_addr;
1893 NET_EPOCH_EXIT();
1894 }
1895 }
1896 INP_WUNLOCK(inp);
1897 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
1898 error = sooptcopyout(sopt, &mreqn,
1899 sizeof(struct ip_mreqn));
1900 } else {
1901 error = sooptcopyout(sopt, &mreqn.imr_address,
1902 sizeof(struct in_addr));
1903 }
1904 break;
1905
1906 case IP_MULTICAST_TTL:
1907 if (imo == NULL)
1908 optval = coptval = IP_DEFAULT_MULTICAST_TTL;
1909 else
1910 optval = coptval = imo->imo_multicast_ttl;
1911 INP_WUNLOCK(inp);
1912 if (sopt->sopt_valsize == sizeof(u_char))
1913 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1914 else
1915 error = sooptcopyout(sopt, &optval, sizeof(int));
1916 break;
1917
1918 case IP_MULTICAST_LOOP:
1919 if (imo == NULL)
1920 optval = coptval = IP_DEFAULT_MULTICAST_LOOP;
1921 else
1922 optval = coptval = imo->imo_multicast_loop;
1923 INP_WUNLOCK(inp);
1924 if (sopt->sopt_valsize == sizeof(u_char))
1925 error = sooptcopyout(sopt, &coptval, sizeof(u_char));
1926 else
1927 error = sooptcopyout(sopt, &optval, sizeof(int));
1928 break;
1929
1930 case IP_MSFILTER:
1931 if (imo == NULL) {
1932 error = EADDRNOTAVAIL;
1933 INP_WUNLOCK(inp);
1934 } else {
1935 error = inp_get_source_filters(inp, sopt);
1936 }
1937 break;
1938
1939 default:
1940 INP_WUNLOCK(inp);
1941 error = ENOPROTOOPT;
1942 break;
1943 }
1944
1945 INP_UNLOCK_ASSERT(inp);
1946
1947 return (error);
1948 }
1949
1950 /*
1951 * Look up the ifnet to use for a multicast group membership,
1952 * given the IPv4 address of an interface, and the IPv4 group address.
1953 *
1954 * This routine exists to support legacy multicast applications
1955 * which do not understand that multicast memberships are scoped to
1956 * specific physical links in the networking stack, or which need
1957 * to join link-scope groups before IPv4 addresses are configured.
1958 *
1959 * If inp is non-NULL, use this socket's current FIB number for any
1960 * required FIB lookup.
1961 * If ina is INADDR_ANY, look up the group address in the unicast FIB,
1962 * and use its ifp; usually, this points to the default next-hop.
1963 *
1964 * If the FIB lookup fails, attempt to use the first non-loopback
1965 * interface with multicast capability in the system as a
1966 * last resort. The legacy IPv4 ASM API requires that we do
1967 * this in order to allow groups to be joined when the routing
1968 * table has not yet been populated during boot.
1969 *
1970 * Returns NULL if no ifp could be found.
1971 *
1972 * SMPng: TODO: Acquire the appropriate locks for INADDR_TO_IFP.
1973 * FUTURE: Implement IPv4 source-address selection.
1974 */
1975 static struct ifnet *
1976 inp_lookup_mcast_ifp(const struct inpcb *inp,
1977 const struct sockaddr_in *gsin, const struct in_addr ina)
1978 {
1979 struct rm_priotracker in_ifa_tracker;
1980 struct ifnet *ifp;
1981 struct nhop4_basic nh4;
1982 uint32_t fibnum;
1983
1984 KASSERT(gsin->sin_family == AF_INET, ("%s: not AF_INET", __func__));
1985 KASSERT(IN_MULTICAST(ntohl(gsin->sin_addr.s_addr)),
1986 ("%s: not multicast", __func__));
1987
1988 ifp = NULL;
1989 if (!in_nullhost(ina)) {
1990 INADDR_TO_IFP(ina, ifp);
1991 } else {
1992 fibnum = inp ? inp->inp_inc.inc_fibnum : 0;
1993 if (fib4_lookup_nh_basic(fibnum, gsin->sin_addr, 0, 0, &nh4)==0)
1994 ifp = nh4.nh_ifp;
1995 else {
1996 struct in_ifaddr *ia;
1997 struct ifnet *mifp;
1998
1999 mifp = NULL;
2000 IN_IFADDR_RLOCK(&in_ifa_tracker);
2001 CK_STAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) {
2002 mifp = ia->ia_ifp;
2003 if (!(mifp->if_flags & IFF_LOOPBACK) &&
2004 (mifp->if_flags & IFF_MULTICAST)) {
2005 ifp = mifp;
2006 break;
2007 }
2008 }
2009 IN_IFADDR_RUNLOCK(&in_ifa_tracker);
2010 }
2011 }
2012
2013 return (ifp);
2014 }
2015
2016 /*
2017 * Join an IPv4 multicast group, possibly with a source.
2018 */
2019 static int
2020 inp_join_group(struct inpcb *inp, struct sockopt *sopt)
2021 {
2022 struct group_source_req gsr;
2023 sockunion_t *gsa, *ssa;
2024 struct ifnet *ifp;
2025 struct in_mfilter *imf;
2026 struct ip_moptions *imo;
2027 struct in_multi *inm;
2028 struct in_msource *lims;
2029 size_t idx;
2030 int error, is_new;
2031
2032 ifp = NULL;
2033 imf = NULL;
2034 lims = NULL;
2035 error = 0;
2036 is_new = 0;
2037
2038 memset(&gsr, 0, sizeof(struct group_source_req));
2039 gsa = (sockunion_t *)&gsr.gsr_group;
2040 gsa->ss.ss_family = AF_UNSPEC;
2041 ssa = (sockunion_t *)&gsr.gsr_source;
2042 ssa->ss.ss_family = AF_UNSPEC;
2043
2044 switch (sopt->sopt_name) {
2045 case IP_ADD_MEMBERSHIP:
2046 case IP_ADD_SOURCE_MEMBERSHIP: {
2047 struct ip_mreq_source mreqs;
2048
2049 if (sopt->sopt_name == IP_ADD_MEMBERSHIP) {
2050 error = sooptcopyin(sopt, &mreqs,
2051 sizeof(struct ip_mreq),
2052 sizeof(struct ip_mreq));
2053 /*
2054 * Do argument switcharoo from ip_mreq into
2055 * ip_mreq_source to avoid using two instances.
2056 */
2057 mreqs.imr_interface = mreqs.imr_sourceaddr;
2058 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2059 } else if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2060 error = sooptcopyin(sopt, &mreqs,
2061 sizeof(struct ip_mreq_source),
2062 sizeof(struct ip_mreq_source));
2063 }
2064 if (error)
2065 return (error);
2066
2067 gsa->sin.sin_family = AF_INET;
2068 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2069 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2070
2071 if (sopt->sopt_name == IP_ADD_SOURCE_MEMBERSHIP) {
2072 ssa->sin.sin_family = AF_INET;
2073 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2074 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2075 }
2076
2077 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2078 return (EINVAL);
2079
2080 ifp = inp_lookup_mcast_ifp(inp, &gsa->sin,
2081 mreqs.imr_interface);
2082 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2083 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2084 break;
2085 }
2086
2087 case MCAST_JOIN_GROUP:
2088 case MCAST_JOIN_SOURCE_GROUP:
2089 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
2090 error = sooptcopyin(sopt, &gsr,
2091 sizeof(struct group_req),
2092 sizeof(struct group_req));
2093 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2094 error = sooptcopyin(sopt, &gsr,
2095 sizeof(struct group_source_req),
2096 sizeof(struct group_source_req));
2097 }
2098 if (error)
2099 return (error);
2100
2101 if (gsa->sin.sin_family != AF_INET ||
2102 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2103 return (EINVAL);
2104
2105 /*
2106 * Overwrite the port field if present, as the sockaddr
2107 * being copied in may be matched with a binary comparison.
2108 */
2109 gsa->sin.sin_port = 0;
2110 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
2111 if (ssa->sin.sin_family != AF_INET ||
2112 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2113 return (EINVAL);
2114 ssa->sin.sin_port = 0;
2115 }
2116
2117 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2118 return (EINVAL);
2119
2120 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2121 return (EADDRNOTAVAIL);
2122 ifp = ifnet_byindex(gsr.gsr_interface);
2123 break;
2124
2125 default:
2126 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2127 __func__, sopt->sopt_name);
2128 return (EOPNOTSUPP);
2129 break;
2130 }
2131
2132 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2133 return (EADDRNOTAVAIL);
2134
2135 imo = inp_findmoptions(inp);
2136 idx = imo_match_group(imo, ifp, &gsa->sa);
2137 if (idx == -1) {
2138 is_new = 1;
2139 } else {
2140 inm = imo->imo_membership[idx];
2141 imf = &imo->imo_mfilters[idx];
2142 if (ssa->ss.ss_family != AF_UNSPEC) {
2143 /*
2144 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2145 * is an error. On an existing inclusive membership,
2146 * it just adds the source to the filter list.
2147 */
2148 if (imf->imf_st[1] != MCAST_INCLUDE) {
2149 error = EINVAL;
2150 goto out_inp_locked;
2151 }
2152 /*
2153 * Throw out duplicates.
2154 *
2155 * XXX FIXME: This makes a naive assumption that
2156 * even if entries exist for *ssa in this imf,
2157 * they will be rejected as dupes, even if they
2158 * are not valid in the current mode (in-mode).
2159 *
2160 * in_msource is transactioned just as for anything
2161 * else in SSM -- but note naive use of inm_graft()
2162 * below for allocating new filter entries.
2163 *
2164 * This is only an issue if someone mixes the
2165 * full-state SSM API with the delta-based API,
2166 * which is discouraged in the relevant RFCs.
2167 */
2168 lims = imo_match_source(imo, idx, &ssa->sa);
2169 if (lims != NULL /*&&
2170 lims->imsl_st[1] == MCAST_INCLUDE*/) {
2171 error = EADDRNOTAVAIL;
2172 goto out_inp_locked;
2173 }
2174 } else {
2175 /*
2176 * MCAST_JOIN_GROUP on an existing exclusive
2177 * membership is an error; return EADDRINUSE
2178 * to preserve 4.4BSD API idempotence, and
2179 * avoid tedious detour to code below.
2180 * NOTE: This is bending RFC 3678 a bit.
2181 *
2182 * On an existing inclusive membership, this is also
2183 * an error; if you want to change filter mode,
2184 * you must use the userland API setsourcefilter().
2185 * XXX We don't reject this for imf in UNDEFINED
2186 * state at t1, because allocation of a filter
2187 * is atomic with allocation of a membership.
2188 */
2189 error = EINVAL;
2190 if (imf->imf_st[1] == MCAST_EXCLUDE)
2191 error = EADDRINUSE;
2192 goto out_inp_locked;
2193 }
2194 }
2195
2196 /*
2197 * Begin state merge transaction at socket layer.
2198 */
2199 INP_WLOCK_ASSERT(inp);
2200
2201 if (is_new) {
2202 if (imo->imo_num_memberships == imo->imo_max_memberships) {
2203 error = imo_grow(imo);
2204 if (error)
2205 goto out_inp_locked;
2206 }
2207 /*
2208 * Allocate the new slot upfront so we can deal with
2209 * grafting the new source filter in same code path
2210 * as for join-source on existing membership.
2211 */
2212 idx = imo->imo_num_memberships;
2213 imo->imo_membership[idx] = NULL;
2214 imo->imo_num_memberships++;
2215 KASSERT(imo->imo_mfilters != NULL,
2216 ("%s: imf_mfilters vector was not allocated", __func__));
2217 imf = &imo->imo_mfilters[idx];
2218 KASSERT(RB_EMPTY(&imf->imf_sources),
2219 ("%s: imf_sources not empty", __func__));
2220 }
2221
2222 /*
2223 * Graft new source into filter list for this inpcb's
2224 * membership of the group. The in_multi may not have
2225 * been allocated yet if this is a new membership, however,
2226 * the in_mfilter slot will be allocated and must be initialized.
2227 *
2228 * Note: Grafting of exclusive mode filters doesn't happen
2229 * in this path.
2230 * XXX: Should check for non-NULL lims (node exists but may
2231 * not be in-mode) for interop with full-state API.
2232 */
2233 if (ssa->ss.ss_family != AF_UNSPEC) {
2234 /* Membership starts in IN mode */
2235 if (is_new) {
2236 CTR1(KTR_IGMPV3, "%s: new join w/source", __func__);
2237 imf_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2238 } else {
2239 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "allow");
2240 }
2241 lims = imf_graft(imf, MCAST_INCLUDE, &ssa->sin);
2242 if (lims == NULL) {
2243 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2244 __func__);
2245 error = ENOMEM;
2246 goto out_imo_free;
2247 }
2248 } else {
2249 /* No address specified; Membership starts in EX mode */
2250 if (is_new) {
2251 CTR1(KTR_IGMPV3, "%s: new join w/o source", __func__);
2252 imf_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2253 }
2254 }
2255
2256 /*
2257 * Begin state merge transaction at IGMP layer.
2258 */
2259 in_pcbref(inp);
2260 INP_WUNLOCK(inp);
2261 IN_MULTI_LOCK();
2262
2263 if (is_new) {
2264 error = in_joingroup_locked(ifp, &gsa->sin.sin_addr, imf,
2265 &inm);
2266 if (error) {
2267 CTR1(KTR_IGMPV3, "%s: in_joingroup_locked failed",
2268 __func__);
2269 IN_MULTI_LIST_UNLOCK();
2270 goto out_imo_free;
2271 }
2272 inm_acquire(inm);
2273 imo->imo_membership[idx] = inm;
2274 } else {
2275 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2276 IN_MULTI_LIST_LOCK();
2277 error = inm_merge(inm, imf);
2278 if (error) {
2279 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2280 __func__);
2281 IN_MULTI_LIST_UNLOCK();
2282 goto out_in_multi_locked;
2283 }
2284 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2285 error = igmp_change_state(inm);
2286 IN_MULTI_LIST_UNLOCK();
2287 if (error) {
2288 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2289 __func__);
2290 goto out_in_multi_locked;
2291 }
2292 }
2293
2294 out_in_multi_locked:
2295
2296 IN_MULTI_UNLOCK();
2297 INP_WLOCK(inp);
2298 if (in_pcbrele_wlocked(inp))
2299 return (ENXIO);
2300 if (error) {
2301 imf_rollback(imf);
2302 if (is_new)
2303 imf_purge(imf);
2304 else
2305 imf_reap(imf);
2306 } else {
2307 imf_commit(imf);
2308 }
2309
2310 out_imo_free:
2311 if (error && is_new) {
2312 inm = imo->imo_membership[idx];
2313 if (inm != NULL) {
2314 IN_MULTI_LIST_LOCK();
2315 inm_release_deferred(inm);
2316 IN_MULTI_LIST_UNLOCK();
2317 }
2318 imo->imo_membership[idx] = NULL;
2319 --imo->imo_num_memberships;
2320 }
2321
2322 out_inp_locked:
2323 INP_WUNLOCK(inp);
2324 return (error);
2325 }
2326
2327 /*
2328 * Leave an IPv4 multicast group on an inpcb, possibly with a source.
2329 */
2330 static int
2331 inp_leave_group(struct inpcb *inp, struct sockopt *sopt)
2332 {
2333 struct group_source_req gsr;
2334 struct ip_mreq_source mreqs;
2335 sockunion_t *gsa, *ssa;
2336 struct ifnet *ifp;
2337 struct in_mfilter *imf;
2338 struct ip_moptions *imo;
2339 struct in_msource *ims;
2340 struct in_multi *inm;
2341 size_t idx;
2342 int error, is_final;
2343
2344 ifp = NULL;
2345 error = 0;
2346 is_final = 1;
2347
2348 memset(&gsr, 0, sizeof(struct group_source_req));
2349 gsa = (sockunion_t *)&gsr.gsr_group;
2350 gsa->ss.ss_family = AF_UNSPEC;
2351 ssa = (sockunion_t *)&gsr.gsr_source;
2352 ssa->ss.ss_family = AF_UNSPEC;
2353
2354 switch (sopt->sopt_name) {
2355 case IP_DROP_MEMBERSHIP:
2356 case IP_DROP_SOURCE_MEMBERSHIP:
2357 if (sopt->sopt_name == IP_DROP_MEMBERSHIP) {
2358 error = sooptcopyin(sopt, &mreqs,
2359 sizeof(struct ip_mreq),
2360 sizeof(struct ip_mreq));
2361 /*
2362 * Swap interface and sourceaddr arguments,
2363 * as ip_mreq and ip_mreq_source are laid
2364 * out differently.
2365 */
2366 mreqs.imr_interface = mreqs.imr_sourceaddr;
2367 mreqs.imr_sourceaddr.s_addr = INADDR_ANY;
2368 } else if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2369 error = sooptcopyin(sopt, &mreqs,
2370 sizeof(struct ip_mreq_source),
2371 sizeof(struct ip_mreq_source));
2372 }
2373 if (error)
2374 return (error);
2375
2376 gsa->sin.sin_family = AF_INET;
2377 gsa->sin.sin_len = sizeof(struct sockaddr_in);
2378 gsa->sin.sin_addr = mreqs.imr_multiaddr;
2379
2380 if (sopt->sopt_name == IP_DROP_SOURCE_MEMBERSHIP) {
2381 ssa->sin.sin_family = AF_INET;
2382 ssa->sin.sin_len = sizeof(struct sockaddr_in);
2383 ssa->sin.sin_addr = mreqs.imr_sourceaddr;
2384 }
2385
2386 /*
2387 * Attempt to look up hinted ifp from interface address.
2388 * Fallthrough with null ifp iff lookup fails, to
2389 * preserve 4.4BSD mcast API idempotence.
2390 * XXX NOTE WELL: The RFC 3678 API is preferred because
2391 * using an IPv4 address as a key is racy.
2392 */
2393 if (!in_nullhost(mreqs.imr_interface))
2394 INADDR_TO_IFP(mreqs.imr_interface, ifp);
2395
2396 CTR3(KTR_IGMPV3, "%s: imr_interface = 0x%08x, ifp = %p",
2397 __func__, ntohl(mreqs.imr_interface.s_addr), ifp);
2398
2399 break;
2400
2401 case MCAST_LEAVE_GROUP:
2402 case MCAST_LEAVE_SOURCE_GROUP:
2403 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2404 error = sooptcopyin(sopt, &gsr,
2405 sizeof(struct group_req),
2406 sizeof(struct group_req));
2407 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2408 error = sooptcopyin(sopt, &gsr,
2409 sizeof(struct group_source_req),
2410 sizeof(struct group_source_req));
2411 }
2412 if (error)
2413 return (error);
2414
2415 if (gsa->sin.sin_family != AF_INET ||
2416 gsa->sin.sin_len != sizeof(struct sockaddr_in))
2417 return (EINVAL);
2418
2419 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2420 if (ssa->sin.sin_family != AF_INET ||
2421 ssa->sin.sin_len != sizeof(struct sockaddr_in))
2422 return (EINVAL);
2423 }
2424
2425 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2426 return (EADDRNOTAVAIL);
2427
2428 ifp = ifnet_byindex(gsr.gsr_interface);
2429
2430 if (ifp == NULL)
2431 return (EADDRNOTAVAIL);
2432 break;
2433
2434 default:
2435 CTR2(KTR_IGMPV3, "%s: unknown sopt_name %d",
2436 __func__, sopt->sopt_name);
2437 return (EOPNOTSUPP);
2438 break;
2439 }
2440
2441 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2442 return (EINVAL);
2443
2444 /*
2445 * Find the membership in the membership array.
2446 */
2447 imo = inp_findmoptions(inp);
2448 idx = imo_match_group(imo, ifp, &gsa->sa);
2449 if (idx == -1) {
2450 error = EADDRNOTAVAIL;
2451 goto out_inp_locked;
2452 }
2453 inm = imo->imo_membership[idx];
2454 imf = &imo->imo_mfilters[idx];
2455
2456 if (ssa->ss.ss_family != AF_UNSPEC)
2457 is_final = 0;
2458
2459 /*
2460 * Begin state merge transaction at socket layer.
2461 */
2462 INP_WLOCK_ASSERT(inp);
2463
2464 /*
2465 * If we were instructed only to leave a given source, do so.
2466 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2467 */
2468 if (is_final) {
2469 imf_leave(imf);
2470 } else {
2471 if (imf->imf_st[0] == MCAST_EXCLUDE) {
2472 error = EADDRNOTAVAIL;
2473 goto out_inp_locked;
2474 }
2475 ims = imo_match_source(imo, idx, &ssa->sa);
2476 if (ims == NULL) {
2477 CTR3(KTR_IGMPV3, "%s: source 0x%08x %spresent",
2478 __func__, ntohl(ssa->sin.sin_addr.s_addr), "not ");
2479 error = EADDRNOTAVAIL;
2480 goto out_inp_locked;
2481 }
2482 CTR2(KTR_IGMPV3, "%s: %s source", __func__, "block");
2483 error = imf_prune(imf, &ssa->sin);
2484 if (error) {
2485 CTR1(KTR_IGMPV3, "%s: merge imf state failed",
2486 __func__);
2487 goto out_inp_locked;
2488 }
2489 }
2490
2491 /*
2492 * Begin state merge transaction at IGMP layer.
2493 */
2494 in_pcbref(inp);
2495 INP_WUNLOCK(inp);
2496 IN_MULTI_LOCK();
2497
2498 if (is_final) {
2499 /*
2500 * Give up the multicast address record to which
2501 * the membership points.
2502 */
2503 (void)in_leavegroup_locked(inm, imf);
2504 } else {
2505 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2506 IN_MULTI_LIST_LOCK();
2507 error = inm_merge(inm, imf);
2508 if (error) {
2509 CTR1(KTR_IGMPV3, "%s: failed to merge inm state",
2510 __func__);
2511 IN_MULTI_LIST_UNLOCK();
2512 goto out_in_multi_locked;
2513 }
2514
2515 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2516 error = igmp_change_state(inm);
2517 IN_MULTI_LIST_UNLOCK();
2518 if (error) {
2519 CTR1(KTR_IGMPV3, "%s: failed igmp downcall",
2520 __func__);
2521 }
2522 }
2523
2524 out_in_multi_locked:
2525
2526 IN_MULTI_UNLOCK();
2527 INP_WLOCK(inp);
2528 if (in_pcbrele_wlocked(inp))
2529 return (ENXIO);
2530
2531 if (error)
2532 imf_rollback(imf);
2533 else
2534 imf_commit(imf);
2535
2536 imf_reap(imf);
2537
2538 if (is_final) {
2539 /* Remove the gap in the membership and filter array. */
2540 for (++idx; idx < imo->imo_num_memberships; ++idx) {
2541 imo->imo_membership[idx-1] = imo->imo_membership[idx];
2542 imo->imo_mfilters[idx-1] = imo->imo_mfilters[idx];
2543 }
2544 imo->imo_num_memberships--;
2545 }
2546
2547 out_inp_locked:
2548 INP_WUNLOCK(inp);
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;
2568
2569 if (sopt->sopt_valsize == sizeof(struct ip_mreqn)) {
2570 /*
2571 * An interface index was specified using the
2572 * Linux-derived ip_mreqn structure.
2573 */
2574 error = sooptcopyin(sopt, &mreqn, sizeof(struct ip_mreqn),
2575 sizeof(struct ip_mreqn));
2576 if (error)
2577 return (error);
2578
2579 if (mreqn.imr_ifindex < 0 || V_if_index < mreqn.imr_ifindex)
2580 return (EINVAL);
2581
2582 if (mreqn.imr_ifindex == 0) {
2583 ifp = NULL;
2584 } else {
2585 ifp = ifnet_byindex(mreqn.imr_ifindex);
2586 if (ifp == NULL)
2587 return (EADDRNOTAVAIL);
2588 }
2589 } else {
2590 /*
2591 * An interface was specified by IPv4 address.
2592 * This is the traditional BSD usage.
2593 */
2594 error = sooptcopyin(sopt, &addr, sizeof(struct in_addr),
2595 sizeof(struct in_addr));
2596 if (error)
2597 return (error);
2598 if (in_nullhost(addr)) {
2599 ifp = NULL;
2600 } else {
2601 INADDR_TO_IFP(addr, ifp);
2602 if (ifp == NULL)
2603 return (EADDRNOTAVAIL);
2604 }
2605 CTR3(KTR_IGMPV3, "%s: ifp = %p, addr = 0x%08x", __func__, ifp,
2606 ntohl(addr.s_addr));
2607 }
2608
2609 /* Reject interfaces which do not support multicast. */
2610 if (ifp != NULL && (ifp->if_flags & IFF_MULTICAST) == 0)
2611 return (EOPNOTSUPP);
2612
2613 imo = inp_findmoptions(inp);
2614 imo->imo_multicast_ifp = ifp;
2615 imo->imo_multicast_addr.s_addr = INADDR_ANY;
2616 INP_WUNLOCK(inp);
2617
2618 return (0);
2619 }
2620
2621 /*
2622 * Atomically set source filters on a socket for an IPv4 multicast group.
2623 *
2624 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2625 */
2626 static int
2627 inp_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2628 {
2629 struct __msfilterreq msfr;
2630 sockunion_t *gsa;
2631 struct ifnet *ifp;
2632 struct in_mfilter *imf;
2633 struct ip_moptions *imo;
2634 struct in_multi *inm;
2635 size_t idx;
2636 int error;
2637
2638 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2639 sizeof(struct __msfilterreq));
2640 if (error)
2641 return (error);
2642
2643 if (msfr.msfr_nsrcs > in_mcast_maxsocksrc)
2644 return (ENOBUFS);
2645
2646 if ((msfr.msfr_fmode != MCAST_EXCLUDE &&
2647 msfr.msfr_fmode != MCAST_INCLUDE))
2648 return (EINVAL);
2649
2650 if (msfr.msfr_group.ss_family != AF_INET ||
2651 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in))
2652 return (EINVAL);
2653
2654 gsa = (sockunion_t *)&msfr.msfr_group;
2655 if (!IN_MULTICAST(ntohl(gsa->sin.sin_addr.s_addr)))
2656 return (EINVAL);
2657
2658 gsa->sin.sin_port = 0; /* ignore port */
2659
2660 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2661 return (EADDRNOTAVAIL);
2662
2663 ifp = ifnet_byindex(msfr.msfr_ifindex);
2664 if (ifp == NULL)
2665 return (EADDRNOTAVAIL);
2666
2667 /*
2668 * Take the INP write lock.
2669 * Check if this socket is a member of this group.
2670 */
2671 imo = inp_findmoptions(inp);
2672 idx = imo_match_group(imo, ifp, &gsa->sa);
2673 if (idx == -1 || imo->imo_mfilters == NULL) {
2674 error = EADDRNOTAVAIL;
2675 goto out_inp_locked;
2676 }
2677 inm = imo->imo_membership[idx];
2678 imf = &imo->imo_mfilters[idx];
2679
2680 /*
2681 * Begin state merge transaction at socket layer.
2682 */
2683 INP_WLOCK_ASSERT(inp);
2684
2685 imf->imf_st[1] = msfr.msfr_fmode;
2686
2687 /*
2688 * Apply any new source filters, if present.
2689 * Make a copy of the user-space source vector so
2690 * that we may copy them with a single copyin. This
2691 * allows us to deal with page faults up-front.
2692 */
2693 if (msfr.msfr_nsrcs > 0) {
2694 struct in_msource *lims;
2695 struct sockaddr_in *psin;
2696 struct sockaddr_storage *kss, *pkss;
2697 int i;
2698
2699 INP_WUNLOCK(inp);
2700
2701 CTR2(KTR_IGMPV3, "%s: loading %lu source list entries",
2702 __func__, (unsigned long)msfr.msfr_nsrcs);
2703 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2704 M_TEMP, M_WAITOK);
2705 error = copyin(msfr.msfr_srcs, kss,
2706 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2707 if (error) {
2708 free(kss, M_TEMP);
2709 return (error);
2710 }
2711
2712 INP_WLOCK(inp);
2713
2714 /*
2715 * Mark all source filters as UNDEFINED at t1.
2716 * Restore new group filter mode, as imf_leave()
2717 * will set it to INCLUDE.
2718 */
2719 imf_leave(imf);
2720 imf->imf_st[1] = msfr.msfr_fmode;
2721
2722 /*
2723 * Update socket layer filters at t1, lazy-allocating
2724 * new entries. This saves a bunch of memory at the
2725 * cost of one RB_FIND() per source entry; duplicate
2726 * entries in the msfr_nsrcs vector are ignored.
2727 * If we encounter an error, rollback transaction.
2728 *
2729 * XXX This too could be replaced with a set-symmetric
2730 * difference like loop to avoid walking from root
2731 * every time, as the key space is common.
2732 */
2733 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2734 psin = (struct sockaddr_in *)pkss;
2735 if (psin->sin_family != AF_INET) {
2736 error = EAFNOSUPPORT;
2737 break;
2738 }
2739 if (psin->sin_len != sizeof(struct sockaddr_in)) {
2740 error = EINVAL;
2741 break;
2742 }
2743 error = imf_get_source(imf, psin, &lims);
2744 if (error)
2745 break;
2746 lims->imsl_st[1] = imf->imf_st[1];
2747 }
2748 free(kss, M_TEMP);
2749 }
2750
2751 if (error)
2752 goto out_imf_rollback;
2753
2754 INP_WLOCK_ASSERT(inp);
2755 IN_MULTI_LOCK();
2756
2757 /*
2758 * Begin state merge transaction at IGMP layer.
2759 */
2760 CTR1(KTR_IGMPV3, "%s: merge inm state", __func__);
2761 IN_MULTI_LIST_LOCK();
2762 error = inm_merge(inm, imf);
2763 if (error) {
2764 CTR1(KTR_IGMPV3, "%s: failed to merge inm state", __func__);
2765 IN_MULTI_LIST_UNLOCK();
2766 goto out_in_multi_locked;
2767 }
2768
2769 CTR1(KTR_IGMPV3, "%s: doing igmp downcall", __func__);
2770 error = igmp_change_state(inm);
2771 IN_MULTI_LIST_UNLOCK();
2772 if (error)
2773 CTR1(KTR_IGMPV3, "%s: failed igmp downcall", __func__);
2774
2775 out_in_multi_locked:
2776
2777 IN_MULTI_UNLOCK();
2778
2779 out_imf_rollback:
2780 if (error)
2781 imf_rollback(imf);
2782 else
2783 imf_commit(imf);
2784
2785 imf_reap(imf);
2786
2787 out_inp_locked:
2788 INP_WUNLOCK(inp);
2789 return (error);
2790 }
2791
2792 /*
2793 * Set the IP multicast options in response to user setsockopt().
2794 *
2795 * Many of the socket options handled in this function duplicate the
2796 * functionality of socket options in the regular unicast API. However,
2797 * it is not possible to merge the duplicate code, because the idempotence
2798 * of the IPv4 multicast part of the BSD Sockets API must be preserved;
2799 * the effects of these options must be treated as separate and distinct.
2800 *
2801 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2802 * FUTURE: The IP_MULTICAST_VIF option may be eliminated if MROUTING
2803 * is refactored to no longer use vifs.
2804 */
2805 int
2806 inp_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2807 {
2808 struct ip_moptions *imo;
2809 int error;
2810
2811 error = 0;
2812
2813 /*
2814 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2815 * or is a divert socket, reject it.
2816 */
2817 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2818 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2819 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2820 return (EOPNOTSUPP);
2821
2822 switch (sopt->sopt_name) {
2823 case IP_MULTICAST_VIF: {
2824 int vifi;
2825 /*
2826 * Select a multicast VIF for transmission.
2827 * Only useful if multicast forwarding is active.
2828 */
2829 if (legal_vif_num == NULL) {
2830 error = EOPNOTSUPP;
2831 break;
2832 }
2833 error = sooptcopyin(sopt, &vifi, sizeof(int), sizeof(int));
2834 if (error)
2835 break;
2836 if (!legal_vif_num(vifi) && (vifi != -1)) {
2837 error = EINVAL;
2838 break;
2839 }
2840 imo = inp_findmoptions(inp);
2841 imo->imo_multicast_vif = vifi;
2842 INP_WUNLOCK(inp);
2843 break;
2844 }
2845
2846 case IP_MULTICAST_IF:
2847 error = inp_set_multicast_if(inp, sopt);
2848 break;
2849
2850 case IP_MULTICAST_TTL: {
2851 u_char ttl;
2852
2853 /*
2854 * Set the IP time-to-live for outgoing multicast packets.
2855 * The original multicast API required a char argument,
2856 * which is inconsistent with the rest of the socket API.
2857 * We allow either a char or an int.
2858 */
2859 if (sopt->sopt_valsize == sizeof(u_char)) {
2860 error = sooptcopyin(sopt, &ttl, sizeof(u_char),
2861 sizeof(u_char));
2862 if (error)
2863 break;
2864 } else {
2865 u_int ittl;
2866
2867 error = sooptcopyin(sopt, &ittl, sizeof(u_int),
2868 sizeof(u_int));
2869 if (error)
2870 break;
2871 if (ittl > 255) {
2872 error = EINVAL;
2873 break;
2874 }
2875 ttl = (u_char)ittl;
2876 }
2877 imo = inp_findmoptions(inp);
2878 imo->imo_multicast_ttl = ttl;
2879 INP_WUNLOCK(inp);
2880 break;
2881 }
2882
2883 case IP_MULTICAST_LOOP: {
2884 u_char loop;
2885
2886 /*
2887 * Set the loopback flag for outgoing multicast packets.
2888 * Must be zero or one. The original multicast API required a
2889 * char argument, which is inconsistent with the rest
2890 * of the socket API. We allow either a char or an int.
2891 */
2892 if (sopt->sopt_valsize == sizeof(u_char)) {
2893 error = sooptcopyin(sopt, &loop, sizeof(u_char),
2894 sizeof(u_char));
2895 if (error)
2896 break;
2897 } else {
2898 u_int iloop;
2899
2900 error = sooptcopyin(sopt, &iloop, sizeof(u_int),
2901 sizeof(u_int));
2902 if (error)
2903 break;
2904 loop = (u_char)iloop;
2905 }
2906 imo = inp_findmoptions(inp);
2907 imo->imo_multicast_loop = !!loop;
2908 INP_WUNLOCK(inp);
2909 break;
2910 }
2911
2912 case IP_ADD_MEMBERSHIP:
2913 case IP_ADD_SOURCE_MEMBERSHIP:
2914 case MCAST_JOIN_GROUP:
2915 case MCAST_JOIN_SOURCE_GROUP:
2916 error = inp_join_group(inp, sopt);
2917 break;
2918
2919 case IP_DROP_MEMBERSHIP:
2920 case IP_DROP_SOURCE_MEMBERSHIP:
2921 case MCAST_LEAVE_GROUP:
2922 case MCAST_LEAVE_SOURCE_GROUP:
2923 error = inp_leave_group(inp, sopt);
2924 break;
2925
2926 case IP_BLOCK_SOURCE:
2927 case IP_UNBLOCK_SOURCE:
2928 case MCAST_BLOCK_SOURCE:
2929 case MCAST_UNBLOCK_SOURCE:
2930 error = inp_block_unblock_source(inp, sopt);
2931 break;
2932
2933 case IP_MSFILTER:
2934 error = inp_set_source_filters(inp, sopt);
2935 break;
2936
2937 default:
2938 error = EOPNOTSUPP;
2939 break;
2940 }
2941
2942 INP_UNLOCK_ASSERT(inp);
2943
2944 return (error);
2945 }
2946
2947 /*
2948 * Expose IGMP's multicast filter mode and source list(s) to userland,
2949 * keyed by (ifindex, group).
2950 * The filter mode is written out as a uint32_t, followed by
2951 * 0..n of struct in_addr.
2952 * For use by ifmcstat(8).
2953 * SMPng: NOTE: unlocked read of ifindex space.
2954 */
2955 static int
2956 sysctl_ip_mcast_filters(SYSCTL_HANDLER_ARGS)
2957 {
2958 struct in_addr src, group;
2959 struct ifnet *ifp;
2960 struct ifmultiaddr *ifma;
2961 struct in_multi *inm;
2962 struct ip_msource *ims;
2963 int *name;
2964 int retval;
2965 u_int namelen;
2966 uint32_t fmode, ifindex;
2967
2968 name = (int *)arg1;
2969 namelen = arg2;
2970
2971 if (req->newptr != NULL)
2972 return (EPERM);
2973
2974 if (namelen != 2)
2975 return (EINVAL);
2976
2977 ifindex = name[0];
2978 if (ifindex <= 0 || ifindex > V_if_index) {
2979 CTR2(KTR_IGMPV3, "%s: ifindex %u out of range",
2980 __func__, ifindex);
2981 return (ENOENT);
2982 }
2983
2984 group.s_addr = name[1];
2985 if (!IN_MULTICAST(ntohl(group.s_addr))) {
2986 CTR2(KTR_IGMPV3, "%s: group 0x%08x is not multicast",
2987 __func__, ntohl(group.s_addr));
2988 return (EINVAL);
2989 }
2990
2991 ifp = ifnet_byindex(ifindex);
2992 if (ifp == NULL) {
2993 CTR2(KTR_IGMPV3, "%s: no ifp for ifindex %u",
2994 __func__, ifindex);
2995 return (ENOENT);
2996 }
2997
2998 retval = sysctl_wire_old_buffer(req,
2999 sizeof(uint32_t) + (in_mcast_maxgrpsrc * sizeof(struct in_addr)));
3000 if (retval)
3001 return (retval);
3002
3003 IN_MULTI_LIST_LOCK();
3004
3005 IF_ADDR_RLOCK(ifp);
3006 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3007 if (ifma->ifma_addr->sa_family != AF_INET ||
3008 ifma->ifma_protospec == NULL)
3009 continue;
3010 inm = (struct in_multi *)ifma->ifma_protospec;
3011 if (!in_hosteq(inm->inm_addr, group))
3012 continue;
3013 fmode = inm->inm_st[1].iss_fmode;
3014 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
3015 if (retval != 0)
3016 break;
3017 RB_FOREACH(ims, ip_msource_tree, &inm->inm_srcs) {
3018 CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__,
3019 ims->ims_haddr);
3020 /*
3021 * Only copy-out sources which are in-mode.
3022 */
3023 if (fmode != ims_get_mode(inm, ims, 1)) {
3024 CTR1(KTR_IGMPV3, "%s: skip non-in-mode",
3025 __func__);
3026 continue;
3027 }
3028 src.s_addr = htonl(ims->ims_haddr);
3029 retval = SYSCTL_OUT(req, &src, sizeof(struct in_addr));
3030 if (retval != 0)
3031 break;
3032 }
3033 }
3034 IF_ADDR_RUNLOCK(ifp);
3035
3036 IN_MULTI_LIST_UNLOCK();
3037
3038 return (retval);
3039 }
3040
3041 #if defined(KTR) && (KTR_COMPILE & KTR_IGMPV3)
3042
3043 static const char *inm_modestrs[] = { "un", "in", "ex" };
3044
3045 static const char *
3046 inm_mode_str(const int mode)
3047 {
3048
3049 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
3050 return (inm_modestrs[mode]);
3051 return ("??");
3052 }
3053
3054 static const char *inm_statestrs[] = {
3055 "not-member",
3056 "silent",
3057 "idle",
3058 "lazy",
3059 "sleeping",
3060 "awakening",
3061 "query-pending",
3062 "sg-query-pending",
3063 "leaving"
3064 };
3065
3066 static const char *
3067 inm_state_str(const int state)
3068 {
3069
3070 if (state >= IGMP_NOT_MEMBER && state <= IGMP_LEAVING_MEMBER)
3071 return (inm_statestrs[state]);
3072 return ("??");
3073 }
3074
3075 /*
3076 * Dump an in_multi structure to the console.
3077 */
3078 void
3079 inm_print(const struct in_multi *inm)
3080 {
3081 int t;
3082 char addrbuf[INET_ADDRSTRLEN];
3083
3084 if ((ktr_mask & KTR_IGMPV3) == 0)
3085 return;
3086
3087 printf("%s: --- begin inm %p ---\n", __func__, inm);
3088 printf("addr %s ifp %p(%s) ifma %p\n",
3089 inet_ntoa_r(inm->inm_addr, addrbuf),
3090 inm->inm_ifp,
3091 inm->inm_ifp->if_xname,
3092 inm->inm_ifma);
3093 printf("timer %u state %s refcount %u scq.len %u\n",
3094 inm->inm_timer,
3095 inm_state_str(inm->inm_state),
3096 inm->inm_refcount,
3097 inm->inm_scq.mq_len);
3098 printf("igi %p nsrc %lu sctimer %u scrv %u\n",
3099 inm->inm_igi,
3100 inm->inm_nsrc,
3101 inm->inm_sctimer,
3102 inm->inm_scrv);
3103 for (t = 0; t < 2; t++) {
3104 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
3105 inm_mode_str(inm->inm_st[t].iss_fmode),
3106 inm->inm_st[t].iss_asm,
3107 inm->inm_st[t].iss_ex,
3108 inm->inm_st[t].iss_in,
3109 inm->inm_st[t].iss_rec);
3110 }
3111 printf("%s: --- end inm %p ---\n", __func__, inm);
3112 }
3113
3114 #else /* !KTR || !(KTR_COMPILE & KTR_IGMPV3) */
3115
3116 void
3117 inm_print(const struct in_multi *inm)
3118 {
3119
3120 }
3121
3122 #endif /* KTR && (KTR_COMPILE & KTR_IGMPV3) */
3123
3124 RB_GENERATE(ip_msource_tree, ip_msource, ims_link, ip_msource_cmp);
Cache object: ad2e38878a68b04f88b53b8734937f85
|