54
55#include <net/if.h>
56#include <net/if_dl.h>
57#include <net/route.h>
58#include <net/vnet.h>
59
60#include <netinet/in.h>
61#include <netinet/in_var.h>
62#include <netinet6/in6_var.h>
63#include <netinet/ip6.h>
64#include <netinet/icmp6.h>
65#include <netinet6/ip6_var.h>
66#include <netinet/in_pcb.h>
67#include <netinet/tcp_var.h>
68#include <netinet6/nd6.h>
69#include <netinet6/mld6_var.h>
70#include <netinet6/scope6_var.h>
71
72#ifndef KTR_MLD
73#define KTR_MLD KTR_INET6
74#endif
75
76#ifndef __SOCKUNION_DECLARED
77union sockunion {
78 struct sockaddr_storage ss;
79 struct sockaddr sa;
80 struct sockaddr_dl sdl;
81 struct sockaddr_in6 sin6;
82};
83typedef union sockunion sockunion_t;
84#define __SOCKUNION_DECLARED
85#endif /* __SOCKUNION_DECLARED */
86
87static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
88 "IPv6 multicast PCB-layer source filter");
89static MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
90static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
91static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
92 "IPv6 multicast MLD-layer source filter");
93
94RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
95
96/*
97 * Locking:
98 * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
99 * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
100 * it can be taken by code in net/if.c also.
101 * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
102 *
103 * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
104 * any need for in6_multi itself to be virtualized -- it is bound to an ifp
105 * anyway no matter what happens.
106 */
107struct mtx in6_multi_mtx;
108MTX_SYSINIT(in6_multi_mtx, &in6_multi_mtx, "in6_multi_mtx", MTX_DEF);
109
110static void im6f_commit(struct in6_mfilter *);
111static int im6f_get_source(struct in6_mfilter *imf,
112 const struct sockaddr_in6 *psin,
113 struct in6_msource **);
114static struct in6_msource *
115 im6f_graft(struct in6_mfilter *, const uint8_t,
116 const struct sockaddr_in6 *);
117static void im6f_leave(struct in6_mfilter *);
118static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
119static void im6f_purge(struct in6_mfilter *);
120static void im6f_rollback(struct in6_mfilter *);
121static void im6f_reap(struct in6_mfilter *);
122static int im6o_grow(struct ip6_moptions *);
123static size_t im6o_match_group(const struct ip6_moptions *,
124 const struct ifnet *, const struct sockaddr *);
125static struct in6_msource *
126 im6o_match_source(const struct ip6_moptions *, const size_t,
127 const struct sockaddr *);
128static void im6s_merge(struct ip6_msource *ims,
129 const struct in6_msource *lims, const int rollback);
130static int in6_mc_get(struct ifnet *, const struct in6_addr *,
131 struct in6_multi **);
132static int in6m_get_source(struct in6_multi *inm,
133 const struct in6_addr *addr, const int noalloc,
134 struct ip6_msource **pims);
135static int in6m_is_ifp_detached(const struct in6_multi *);
136static int in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
137static void in6m_purge(struct in6_multi *);
138static void in6m_reap(struct in6_multi *);
139static struct ip6_moptions *
140 in6p_findmoptions(struct inpcb *);
141static int in6p_get_source_filters(struct inpcb *, struct sockopt *);
142static int in6p_join_group(struct inpcb *, struct sockopt *);
143static int in6p_leave_group(struct inpcb *, struct sockopt *);
144static struct ifnet *
145 in6p_lookup_mcast_ifp(const struct inpcb *,
146 const struct sockaddr_in6 *);
147static int in6p_block_unblock_source(struct inpcb *, struct sockopt *);
148static int in6p_set_multicast_if(struct inpcb *, struct sockopt *);
149static int in6p_set_source_filters(struct inpcb *, struct sockopt *);
150static int sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
151
152SYSCTL_DECL(_net_inet6_ip6); /* XXX Not in any common header. */
153
154SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0, "IPv6 multicast");
155
156static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
157SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
158 CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxgrpsrc, 0,
159 "Max source filters per group");
160TUNABLE_ULONG("net.inet6.ip6.mcast.maxgrpsrc", &in6_mcast_maxgrpsrc);
161
162static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
163SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
164 CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxsocksrc, 0,
165 "Max source filters per socket");
166TUNABLE_ULONG("net.inet6.ip6.mcast.maxsocksrc", &in6_mcast_maxsocksrc);
167
168/* TODO Virtualize this switch. */
169int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
170SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
171 &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
172TUNABLE_INT("net.inet6.ip6.mcast.loop", &in6_mcast_loop);
173
174SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
175 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
176 "Per-interface stack-wide source filters");
177
178/*
179 * Inline function which wraps assertions for a valid ifp.
180 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
181 * is detached.
182 */
183static int __inline
184in6m_is_ifp_detached(const struct in6_multi *inm)
185{
186 struct ifnet *ifp;
187
188 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
189 ifp = inm->in6m_ifma->ifma_ifp;
190 if (ifp != NULL) {
191 /*
192 * Sanity check that network-layer notion of ifp is the
193 * same as that of link-layer.
194 */
195 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
196 }
197
198 return (ifp == NULL);
199}
200
201/*
202 * Initialize an in6_mfilter structure to a known state at t0, t1
203 * with an empty source filter list.
204 */
205static __inline void
206im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
207{
208 memset(imf, 0, sizeof(struct in6_mfilter));
209 RB_INIT(&imf->im6f_sources);
210 imf->im6f_st[0] = st0;
211 imf->im6f_st[1] = st1;
212}
213
214/*
215 * Resize the ip6_moptions vector to the next power-of-two minus 1.
216 * May be called with locks held; do not sleep.
217 */
218static int
219im6o_grow(struct ip6_moptions *imo)
220{
221 struct in6_multi **nmships;
222 struct in6_multi **omships;
223 struct in6_mfilter *nmfilters;
224 struct in6_mfilter *omfilters;
225 size_t idx;
226 size_t newmax;
227 size_t oldmax;
228
229 nmships = NULL;
230 nmfilters = NULL;
231 omships = imo->im6o_membership;
232 omfilters = imo->im6o_mfilters;
233 oldmax = imo->im6o_max_memberships;
234 newmax = ((oldmax + 1) * 2) - 1;
235
236 if (newmax <= IPV6_MAX_MEMBERSHIPS) {
237 nmships = (struct in6_multi **)realloc(omships,
238 sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
239 nmfilters = (struct in6_mfilter *)realloc(omfilters,
240 sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
241 M_NOWAIT);
242 if (nmships != NULL && nmfilters != NULL) {
243 /* Initialize newly allocated source filter heads. */
244 for (idx = oldmax; idx < newmax; idx++) {
245 im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
246 MCAST_EXCLUDE);
247 }
248 imo->im6o_max_memberships = newmax;
249 imo->im6o_membership = nmships;
250 imo->im6o_mfilters = nmfilters;
251 }
252 }
253
254 if (nmships == NULL || nmfilters == NULL) {
255 if (nmships != NULL)
256 free(nmships, M_IP6MOPTS);
257 if (nmfilters != NULL)
258 free(nmfilters, M_IN6MFILTER);
259 return (ETOOMANYREFS);
260 }
261
262 return (0);
263}
264
265/*
266 * Find an IPv6 multicast group entry for this ip6_moptions instance
267 * which matches the specified group, and optionally an interface.
268 * Return its index into the array, or -1 if not found.
269 */
270static size_t
271im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
272 const struct sockaddr *group)
273{
274 const struct sockaddr_in6 *gsin6;
275 struct in6_multi **pinm;
276 int idx;
277 int nmships;
278
279 gsin6 = (const struct sockaddr_in6 *)group;
280
281 /* The im6o_membership array may be lazy allocated. */
282 if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
283 return (-1);
284
285 nmships = imo->im6o_num_memberships;
286 pinm = &imo->im6o_membership[0];
287 for (idx = 0; idx < nmships; idx++, pinm++) {
288 if (*pinm == NULL)
289 continue;
290 if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
291 IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
292 &gsin6->sin6_addr)) {
293 break;
294 }
295 }
296 if (idx >= nmships)
297 idx = -1;
298
299 return (idx);
300}
301
302/*
303 * Find an IPv6 multicast source entry for this imo which matches
304 * the given group index for this socket, and source address.
305 *
306 * XXX TODO: The scope ID, if present in src, is stripped before
307 * any comparison. We SHOULD enforce scope/zone checks where the source
308 * filter entry has a link scope.
309 *
310 * NOTE: This does not check if the entry is in-mode, merely if
311 * it exists, which may not be the desired behaviour.
312 */
313static struct in6_msource *
314im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
315 const struct sockaddr *src)
316{
317 struct ip6_msource find;
318 struct in6_mfilter *imf;
319 struct ip6_msource *ims;
320 const sockunion_t *psa;
321
322 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
323 KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
324 ("%s: invalid index %d\n", __func__, (int)gidx));
325
326 /* The im6o_mfilters array may be lazy allocated. */
327 if (imo->im6o_mfilters == NULL)
328 return (NULL);
329 imf = &imo->im6o_mfilters[gidx];
330
331 psa = (const sockunion_t *)src;
332 find.im6s_addr = psa->sin6.sin6_addr;
333 in6_clearscope(&find.im6s_addr); /* XXX */
334 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
335
336 return ((struct in6_msource *)ims);
337}
338
339/*
340 * Perform filtering for multicast datagrams on a socket by group and source.
341 *
342 * Returns 0 if a datagram should be allowed through, or various error codes
343 * if the socket was not a member of the group, or the source was muted, etc.
344 */
345int
346im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
347 const struct sockaddr *group, const struct sockaddr *src)
348{
349 size_t gidx;
350 struct in6_msource *ims;
351 int mode;
352
353 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
354
355 gidx = im6o_match_group(imo, ifp, group);
356 if (gidx == -1)
357 return (MCAST_NOTGMEMBER);
358
359 /*
360 * Check if the source was included in an (S,G) join.
361 * Allow reception on exclusive memberships by default,
362 * reject reception on inclusive memberships by default.
363 * Exclude source only if an in-mode exclude filter exists.
364 * Include source only if an in-mode include filter exists.
365 * NOTE: We are comparing group state here at MLD t1 (now)
366 * with socket-layer t0 (since last downcall).
367 */
368 mode = imo->im6o_mfilters[gidx].im6f_st[1];
369 ims = im6o_match_source(imo, gidx, src);
370
371 if ((ims == NULL && mode == MCAST_INCLUDE) ||
372 (ims != NULL && ims->im6sl_st[0] != mode))
373 return (MCAST_NOTSMEMBER);
374
375 return (MCAST_PASS);
376}
377
378/*
379 * Find and return a reference to an in6_multi record for (ifp, group),
380 * and bump its reference count.
381 * If one does not exist, try to allocate it, and update link-layer multicast
382 * filters on ifp to listen for group.
383 * Assumes the IN6_MULTI lock is held across the call.
384 * Return 0 if successful, otherwise return an appropriate error code.
385 */
386static int
387in6_mc_get(struct ifnet *ifp, const struct in6_addr *group,
388 struct in6_multi **pinm)
389{
390 struct sockaddr_in6 gsin6;
391 struct ifmultiaddr *ifma;
392 struct in6_multi *inm;
393 int error;
394
395 error = 0;
396
397 /*
398 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
399 * if_addmulti() takes this mutex itself, so we must drop and
400 * re-acquire around the call.
401 */
402 IN6_MULTI_LOCK_ASSERT();
403 IF_ADDR_LOCK(ifp);
404
405 inm = in6m_lookup_locked(ifp, group);
406 if (inm != NULL) {
407 /*
408 * If we already joined this group, just bump the
409 * refcount and return it.
410 */
411 KASSERT(inm->in6m_refcount >= 1,
412 ("%s: bad refcount %d", __func__, inm->in6m_refcount));
413 ++inm->in6m_refcount;
414 *pinm = inm;
415 goto out_locked;
416 }
417
418 memset(&gsin6, 0, sizeof(gsin6));
419 gsin6.sin6_family = AF_INET6;
420 gsin6.sin6_len = sizeof(struct sockaddr_in6);
421 gsin6.sin6_addr = *group;
422
423 /*
424 * Check if a link-layer group is already associated
425 * with this network-layer group on the given ifnet.
426 */
427 IF_ADDR_UNLOCK(ifp);
428 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
429 if (error != 0)
430 return (error);
431 IF_ADDR_LOCK(ifp);
432
433 /*
434 * If something other than netinet6 is occupying the link-layer
435 * group, print a meaningful error message and back out of
436 * the allocation.
437 * Otherwise, bump the refcount on the existing network-layer
438 * group association and return it.
439 */
440 if (ifma->ifma_protospec != NULL) {
441 inm = (struct in6_multi *)ifma->ifma_protospec;
442#ifdef INVARIANTS
443 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
444 __func__));
445 KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
446 ("%s: ifma not AF_INET6", __func__));
447 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
448 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
449 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
450 panic("%s: ifma %p is inconsistent with %p (%p)",
451 __func__, ifma, inm, group);
452#endif
453 ++inm->in6m_refcount;
454 *pinm = inm;
455 goto out_locked;
456 }
457
458 IF_ADDR_LOCK_ASSERT(ifp);
459
460 /*
461 * A new in6_multi record is needed; allocate and initialize it.
462 * We DO NOT perform an MLD join as the in6_ layer may need to
463 * push an initial source list down to MLD to support SSM.
464 *
465 * The initial source filter state is INCLUDE, {} as per the RFC.
466 * Pending state-changes per group are subject to a bounds check.
467 */
468 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
469 if (inm == NULL) {
470 if_delmulti_ifma(ifma);
471 error = ENOMEM;
472 goto out_locked;
473 }
474 inm->in6m_addr = *group;
475 inm->in6m_ifp = ifp;
476 inm->in6m_mli = MLD_IFINFO(ifp);
477 inm->in6m_ifma = ifma;
478 inm->in6m_refcount = 1;
479 inm->in6m_state = MLD_NOT_MEMBER;
480 IFQ_SET_MAXLEN(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
481
482 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
483 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
484 RB_INIT(&inm->in6m_srcs);
485
486 ifma->ifma_protospec = inm;
487 *pinm = inm;
488
489out_locked:
490 IF_ADDR_UNLOCK(ifp);
491 return (error);
492}
493
494/*
495 * Drop a reference to an in6_multi record.
496 *
497 * If the refcount drops to 0, free the in6_multi record and
498 * delete the underlying link-layer membership.
499 */
500void
501in6m_release_locked(struct in6_multi *inm)
502{
503 struct ifmultiaddr *ifma;
504
505 IN6_MULTI_LOCK_ASSERT();
506
507 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
508
509 if (--inm->in6m_refcount > 0) {
510 CTR2(KTR_MLD, "%s: refcount is now %d", __func__,
511 inm->in6m_refcount);
512 return;
513 }
514
515 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
516
517 ifma = inm->in6m_ifma;
518
519 /* XXX this access is not covered by IF_ADDR_LOCK */
520 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
521 KASSERT(ifma->ifma_protospec == inm,
522 ("%s: ifma_protospec != inm", __func__));
523 ifma->ifma_protospec = NULL;
524
525 in6m_purge(inm);
526
527 free(inm, M_IP6MADDR);
528
529 if_delmulti_ifma(ifma);
530}
531
532/*
533 * Clear recorded source entries for a group.
534 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
535 * FIXME: Should reap.
536 */
537void
538in6m_clear_recorded(struct in6_multi *inm)
539{
540 struct ip6_msource *ims;
541
542 IN6_MULTI_LOCK_ASSERT();
543
544 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
545 if (ims->im6s_stp) {
546 ims->im6s_stp = 0;
547 --inm->in6m_st[1].iss_rec;
548 }
549 }
550 KASSERT(inm->in6m_st[1].iss_rec == 0,
551 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
552}
553
554/*
555 * Record a source as pending for a Source-Group MLDv2 query.
556 * This lives here as it modifies the shared tree.
557 *
558 * inm is the group descriptor.
559 * naddr is the address of the source to record in network-byte order.
560 *
561 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
562 * lazy-allocate a source node in response to an SG query.
563 * Otherwise, no allocation is performed. This saves some memory
564 * with the trade-off that the source will not be reported to the
565 * router if joined in the window between the query response and
566 * the group actually being joined on the local host.
567 *
568 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
569 * This turns off the allocation of a recorded source entry if
570 * the group has not been joined.
571 *
572 * Return 0 if the source didn't exist or was already marked as recorded.
573 * Return 1 if the source was marked as recorded by this function.
574 * Return <0 if any error occured (negated errno code).
575 */
576int
577in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
578{
579 struct ip6_msource find;
580 struct ip6_msource *ims, *nims;
581
582 IN6_MULTI_LOCK_ASSERT();
583
584 find.im6s_addr = *addr;
585 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
586 if (ims && ims->im6s_stp)
587 return (0);
588 if (ims == NULL) {
589 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
590 return (-ENOSPC);
591 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
592 M_NOWAIT | M_ZERO);
593 if (nims == NULL)
594 return (-ENOMEM);
595 nims->im6s_addr = find.im6s_addr;
596 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
597 ++inm->in6m_nsrc;
598 ims = nims;
599 }
600
601 /*
602 * Mark the source as recorded and update the recorded
603 * source count.
604 */
605 ++ims->im6s_stp;
606 ++inm->in6m_st[1].iss_rec;
607
608 return (1);
609}
610
611/*
612 * Return a pointer to an in6_msource owned by an in6_mfilter,
613 * given its source address.
614 * Lazy-allocate if needed. If this is a new entry its filter state is
615 * undefined at t0.
616 *
617 * imf is the filter set being modified.
618 * addr is the source address.
619 *
620 * SMPng: May be called with locks held; malloc must not block.
621 */
622static int
623im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
624 struct in6_msource **plims)
625{
626 struct ip6_msource find;
627 struct ip6_msource *ims, *nims;
628 struct in6_msource *lims;
629 int error;
630
631 error = 0;
632 ims = NULL;
633 lims = NULL;
634
635 find.im6s_addr = psin->sin6_addr;
636 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
637 lims = (struct in6_msource *)ims;
638 if (lims == NULL) {
639 if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
640 return (ENOSPC);
641 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
642 M_NOWAIT | M_ZERO);
643 if (nims == NULL)
644 return (ENOMEM);
645 lims = (struct in6_msource *)nims;
646 lims->im6s_addr = find.im6s_addr;
647 lims->im6sl_st[0] = MCAST_UNDEFINED;
648 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
649 ++imf->im6f_nsrc;
650 }
651
652 *plims = lims;
653
654 return (error);
655}
656
657/*
658 * Graft a source entry into an existing socket-layer filter set,
659 * maintaining any required invariants and checking allocations.
660 *
661 * The source is marked as being in the new filter mode at t1.
662 *
663 * Return the pointer to the new node, otherwise return NULL.
664 */
665static struct in6_msource *
666im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
667 const struct sockaddr_in6 *psin)
668{
669 struct ip6_msource *nims;
670 struct in6_msource *lims;
671
672 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
673 M_NOWAIT | M_ZERO);
674 if (nims == NULL)
675 return (NULL);
676 lims = (struct in6_msource *)nims;
677 lims->im6s_addr = psin->sin6_addr;
678 lims->im6sl_st[0] = MCAST_UNDEFINED;
679 lims->im6sl_st[1] = st1;
680 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
681 ++imf->im6f_nsrc;
682
683 return (lims);
684}
685
686/*
687 * Prune a source entry from an existing socket-layer filter set,
688 * maintaining any required invariants and checking allocations.
689 *
690 * The source is marked as being left at t1, it is not freed.
691 *
692 * Return 0 if no error occurred, otherwise return an errno value.
693 */
694static int
695im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
696{
697 struct ip6_msource find;
698 struct ip6_msource *ims;
699 struct in6_msource *lims;
700
701 find.im6s_addr = psin->sin6_addr;
702 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
703 if (ims == NULL)
704 return (ENOENT);
705 lims = (struct in6_msource *)ims;
706 lims->im6sl_st[1] = MCAST_UNDEFINED;
707 return (0);
708}
709
710/*
711 * Revert socket-layer filter set deltas at t1 to t0 state.
712 */
713static void
714im6f_rollback(struct in6_mfilter *imf)
715{
716 struct ip6_msource *ims, *tims;
717 struct in6_msource *lims;
718
719 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
720 lims = (struct in6_msource *)ims;
721 if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
722 /* no change at t1 */
723 continue;
724 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
725 /* revert change to existing source at t1 */
726 lims->im6sl_st[1] = lims->im6sl_st[0];
727 } else {
728 /* revert source added t1 */
729 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
730 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
731 free(ims, M_IN6MFILTER);
732 imf->im6f_nsrc--;
733 }
734 }
735 imf->im6f_st[1] = imf->im6f_st[0];
736}
737
738/*
739 * Mark socket-layer filter set as INCLUDE {} at t1.
740 */
741static void
742im6f_leave(struct in6_mfilter *imf)
743{
744 struct ip6_msource *ims;
745 struct in6_msource *lims;
746
747 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
748 lims = (struct in6_msource *)ims;
749 lims->im6sl_st[1] = MCAST_UNDEFINED;
750 }
751 imf->im6f_st[1] = MCAST_INCLUDE;
752}
753
754/*
755 * Mark socket-layer filter set deltas as committed.
756 */
757static void
758im6f_commit(struct in6_mfilter *imf)
759{
760 struct ip6_msource *ims;
761 struct in6_msource *lims;
762
763 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
764 lims = (struct in6_msource *)ims;
765 lims->im6sl_st[0] = lims->im6sl_st[1];
766 }
767 imf->im6f_st[0] = imf->im6f_st[1];
768}
769
770/*
771 * Reap unreferenced sources from socket-layer filter set.
772 */
773static void
774im6f_reap(struct in6_mfilter *imf)
775{
776 struct ip6_msource *ims, *tims;
777 struct in6_msource *lims;
778
779 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
780 lims = (struct in6_msource *)ims;
781 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
782 (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
783 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
784 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
785 free(ims, M_IN6MFILTER);
786 imf->im6f_nsrc--;
787 }
788 }
789}
790
791/*
792 * Purge socket-layer filter set.
793 */
794static void
795im6f_purge(struct in6_mfilter *imf)
796{
797 struct ip6_msource *ims, *tims;
798
799 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
800 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
801 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
802 free(ims, M_IN6MFILTER);
803 imf->im6f_nsrc--;
804 }
805 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
806 KASSERT(RB_EMPTY(&imf->im6f_sources),
807 ("%s: im6f_sources not empty", __func__));
808}
809
810/*
811 * Look up a source filter entry for a multicast group.
812 *
813 * inm is the group descriptor to work with.
814 * addr is the IPv6 address to look up.
815 * noalloc may be non-zero to suppress allocation of sources.
816 * *pims will be set to the address of the retrieved or allocated source.
817 *
818 * SMPng: NOTE: may be called with locks held.
819 * Return 0 if successful, otherwise return a non-zero error code.
820 */
821static int
822in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
823 const int noalloc, struct ip6_msource **pims)
824{
825 struct ip6_msource find;
826 struct ip6_msource *ims, *nims;
827#ifdef KTR
828 char ip6tbuf[INET6_ADDRSTRLEN];
829#endif
830
831 find.im6s_addr = *addr;
832 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
833 if (ims == NULL && !noalloc) {
834 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
835 return (ENOSPC);
836 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
837 M_NOWAIT | M_ZERO);
838 if (nims == NULL)
839 return (ENOMEM);
840 nims->im6s_addr = *addr;
841 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
842 ++inm->in6m_nsrc;
843 ims = nims;
844 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
845 ip6_sprintf(ip6tbuf, addr), ims);
846 }
847
848 *pims = ims;
849 return (0);
850}
851
852/*
853 * Merge socket-layer source into MLD-layer source.
854 * If rollback is non-zero, perform the inverse of the merge.
855 */
856static void
857im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
858 const int rollback)
859{
860 int n = rollback ? -1 : 1;
861#ifdef KTR
862 char ip6tbuf[INET6_ADDRSTRLEN];
863
864 ip6_sprintf(ip6tbuf, &lims->im6s_addr);
865#endif
866
867 if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
868 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
869 ims->im6s_st[1].ex -= n;
870 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
871 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
872 ims->im6s_st[1].in -= n;
873 }
874
875 if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
876 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
877 ims->im6s_st[1].ex += n;
878 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
879 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
880 ims->im6s_st[1].in += n;
881 }
882}
883
884/*
885 * Atomically update the global in6_multi state, when a membership's
886 * filter list is being updated in any way.
887 *
888 * imf is the per-inpcb-membership group filter pointer.
889 * A fake imf may be passed for in-kernel consumers.
890 *
891 * XXX This is a candidate for a set-symmetric-difference style loop
892 * which would eliminate the repeated lookup from root of ims nodes,
893 * as they share the same key space.
894 *
895 * If any error occurred this function will back out of refcounts
896 * and return a non-zero value.
897 */
898static int
899in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
900{
901 struct ip6_msource *ims, *nims;
902 struct in6_msource *lims;
903 int schanged, error;
904 int nsrc0, nsrc1;
905
906 schanged = 0;
907 error = 0;
908 nsrc1 = nsrc0 = 0;
909
910 /*
911 * Update the source filters first, as this may fail.
912 * Maintain count of in-mode filters at t0, t1. These are
913 * used to work out if we transition into ASM mode or not.
914 * Maintain a count of source filters whose state was
915 * actually modified by this operation.
916 */
917 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
918 lims = (struct in6_msource *)ims;
919 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
920 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
921 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
922 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
923 ++schanged;
924 if (error)
925 break;
926 im6s_merge(nims, lims, 0);
927 }
928 if (error) {
929 struct ip6_msource *bims;
930
931 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
932 lims = (struct in6_msource *)ims;
933 if (lims->im6sl_st[0] == lims->im6sl_st[1])
934 continue;
935 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
936 if (bims == NULL)
937 continue;
938 im6s_merge(bims, lims, 1);
939 }
940 goto out_reap;
941 }
942
943 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
944 __func__, nsrc0, nsrc1);
945
946 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
947 if (imf->im6f_st[0] == imf->im6f_st[1] &&
948 imf->im6f_st[1] == MCAST_INCLUDE) {
949 if (nsrc1 == 0) {
950 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
951 --inm->in6m_st[1].iss_in;
952 }
953 }
954
955 /* Handle filter mode transition on socket. */
956 if (imf->im6f_st[0] != imf->im6f_st[1]) {
957 CTR3(KTR_MLD, "%s: imf transition %d to %d",
958 __func__, imf->im6f_st[0], imf->im6f_st[1]);
959
960 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
961 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
962 --inm->in6m_st[1].iss_ex;
963 } else if (imf->im6f_st[0] == MCAST_INCLUDE) {
964 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
965 --inm->in6m_st[1].iss_in;
966 }
967
968 if (imf->im6f_st[1] == MCAST_EXCLUDE) {
969 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
970 inm->in6m_st[1].iss_ex++;
971 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
972 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
973 inm->in6m_st[1].iss_in++;
974 }
975 }
976
977 /*
978 * Track inm filter state in terms of listener counts.
979 * If there are any exclusive listeners, stack-wide
980 * membership is exclusive.
981 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
982 * If no listeners remain, state is undefined at t1,
983 * and the MLD lifecycle for this group should finish.
984 */
985 if (inm->in6m_st[1].iss_ex > 0) {
986 CTR1(KTR_MLD, "%s: transition to EX", __func__);
987 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
988 } else if (inm->in6m_st[1].iss_in > 0) {
989 CTR1(KTR_MLD, "%s: transition to IN", __func__);
990 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
991 } else {
992 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
993 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
994 }
995
996 /* Decrement ASM listener count on transition out of ASM mode. */
997 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
998 if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
999 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
1000 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1001 --inm->in6m_st[1].iss_asm;
1002 }
1003
1004 /* Increment ASM listener count on transition to ASM mode. */
1005 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1006 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1007 inm->in6m_st[1].iss_asm++;
1008 }
1009
1010 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1011 in6m_print(inm);
1012
1013out_reap:
1014 if (schanged > 0) {
1015 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1016 in6m_reap(inm);
1017 }
1018 return (error);
1019}
1020
1021/*
1022 * Mark an in6_multi's filter set deltas as committed.
1023 * Called by MLD after a state change has been enqueued.
1024 */
1025void
1026in6m_commit(struct in6_multi *inm)
1027{
1028 struct ip6_msource *ims;
1029
1030 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1031 CTR1(KTR_MLD, "%s: pre commit:", __func__);
1032 in6m_print(inm);
1033
1034 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1035 ims->im6s_st[0] = ims->im6s_st[1];
1036 }
1037 inm->in6m_st[0] = inm->in6m_st[1];
1038}
1039
1040/*
1041 * Reap unreferenced nodes from an in6_multi's filter set.
1042 */
1043static void
1044in6m_reap(struct in6_multi *inm)
1045{
1046 struct ip6_msource *ims, *tims;
1047
1048 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1049 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1050 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1051 ims->im6s_stp != 0)
1052 continue;
1053 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1054 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1055 free(ims, M_IP6MSOURCE);
1056 inm->in6m_nsrc--;
1057 }
1058}
1059
1060/*
1061 * Purge all source nodes from an in6_multi's filter set.
1062 */
1063static void
1064in6m_purge(struct in6_multi *inm)
1065{
1066 struct ip6_msource *ims, *tims;
1067
1068 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1069 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1070 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1071 free(ims, M_IP6MSOURCE);
1072 inm->in6m_nsrc--;
1073 }
1074}
1075
1076/*
1077 * Join a multicast address w/o sources.
1078 * KAME compatibility entry point.
1079 *
1080 * SMPng: Assume no mc locks held by caller.
1081 */
1082struct in6_multi_mship *
1083in6_joingroup(struct ifnet *ifp, struct in6_addr *mcaddr,
1084 int *errorp, int delay)
1085{
1086 struct in6_multi_mship *imm;
1087 int error;
1088
1089 imm = malloc(sizeof(*imm), M_IP6MADDR, M_NOWAIT);
1090 if (imm == NULL) {
1091 *errorp = ENOBUFS;
1092 return (NULL);
1093 }
1094
1095 delay = (delay * PR_FASTHZ) / hz;
1096
1097 error = in6_mc_join(ifp, mcaddr, NULL, &imm->i6mm_maddr, delay);
1098 if (error) {
1099 *errorp = error;
1100 free(imm, M_IP6MADDR);
1101 return (NULL);
1102 }
1103
1104 return (imm);
1105}
1106
1107/*
1108 * Leave a multicast address w/o sources.
1109 * KAME compatibility entry point.
1110 *
1111 * SMPng: Assume no mc locks held by caller.
1112 */
1113int
1114in6_leavegroup(struct in6_multi_mship *imm)
1115{
1116
1117 if (imm->i6mm_maddr != NULL)
1118 in6_mc_leave(imm->i6mm_maddr, NULL);
1119 free(imm, M_IP6MADDR);
1120 return 0;
1121}
1122
1123/*
1124 * Join a multicast group; unlocked entry point.
1125 *
1126 * SMPng: XXX: in6_mc_join() is called from in6_control() when upper
1127 * locks are not held. Fortunately, ifp is unlikely to have been detached
1128 * at this point, so we assume it's OK to recurse.
1129 */
1130int
1131in6_mc_join(struct ifnet *ifp, const struct in6_addr *mcaddr,
1132 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1133 const int delay)
1134{
1135 int error;
1136
1137 IN6_MULTI_LOCK();
1138 error = in6_mc_join_locked(ifp, mcaddr, imf, pinm, delay);
1139 IN6_MULTI_UNLOCK();
1140
1141 return (error);
1142}
1143
1144/*
1145 * Join a multicast group; real entry point.
1146 *
1147 * Only preserves atomicity at inm level.
1148 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1149 *
1150 * If the MLD downcall fails, the group is not joined, and an error
1151 * code is returned.
1152 */
1153int
1154in6_mc_join_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1155 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1156 const int delay)
1157{
1158 struct in6_mfilter timf;
1159 struct in6_multi *inm;
1160 int error;
1161#ifdef KTR
1162 char ip6tbuf[INET6_ADDRSTRLEN];
1163#endif
1164
1165#ifdef INVARIANTS
1166 /*
1167 * Sanity: Check scope zone ID was set for ifp, if and
1168 * only if group is scoped to an interface.
1169 */
1170 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1171 ("%s: not a multicast address", __func__));
1172 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1173 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1174 KASSERT(mcaddr->s6_addr16[1] != 0,
1175 ("%s: scope zone ID not set", __func__));
1176 }
1177#endif
1178
1179 IN6_MULTI_LOCK_ASSERT();
1180
1181 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1182 ip6_sprintf(ip6tbuf, mcaddr), ifp, ifp->if_xname);
1183
1184 error = 0;
1185 inm = NULL;
1186
1187 /*
1188 * If no imf was specified (i.e. kernel consumer),
1189 * fake one up and assume it is an ASM join.
1190 */
1191 if (imf == NULL) {
1192 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1193 imf = &timf;
1194 }
1195
1196 error = in6_mc_get(ifp, mcaddr, &inm);
1197 if (error) {
1198 CTR1(KTR_MLD, "%s: in6_mc_get() failure", __func__);
1199 return (error);
1200 }
1201
1202 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1203 error = in6m_merge(inm, imf);
1204 if (error) {
1205 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1206 goto out_in6m_release;
1207 }
1208
1209 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1210 error = mld_change_state(inm, delay);
1211 if (error) {
1212 CTR1(KTR_MLD, "%s: failed to update source", __func__);
1213 goto out_in6m_release;
1214 }
1215
1216out_in6m_release:
1217 if (error) {
1218 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1219 in6m_release_locked(inm);
1220 } else {
1221 *pinm = inm;
1222 }
1223
1224 return (error);
1225}
1226
1227/*
1228 * Leave a multicast group; unlocked entry point.
1229 */
1230int
1231in6_mc_leave(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1232{
1233 struct ifnet *ifp;
1234 int error;
1235
1236 ifp = inm->in6m_ifp;
1237
1238 IN6_MULTI_LOCK();
1239 error = in6_mc_leave_locked(inm, imf);
1240 IN6_MULTI_UNLOCK();
1241
1242 return (error);
1243}
1244
1245/*
1246 * Leave a multicast group; real entry point.
1247 * All source filters will be expunged.
1248 *
1249 * Only preserves atomicity at inm level.
1250 *
1251 * Holding the write lock for the INP which contains imf
1252 * is highly advisable. We can't assert for it as imf does not
1253 * contain a back-pointer to the owning inp.
1254 *
1255 * Note: This is not the same as in6m_release(*) as this function also
1256 * makes a state change downcall into MLD.
1257 */
1258int
1259in6_mc_leave_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1260{
1261 struct in6_mfilter timf;
1262 int error;
1263#ifdef KTR
1264 char ip6tbuf[INET6_ADDRSTRLEN];
1265#endif
1266
1267 error = 0;
1268
1269 IN6_MULTI_LOCK_ASSERT();
1270
1271 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1272 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1273 (in6m_is_ifp_detached(inm) ? "null" : inm->in6m_ifp->if_xname),
1274 imf);
1275
1276 /*
1277 * If no imf was specified (i.e. kernel consumer),
1278 * fake one up and assume it is an ASM join.
1279 */
1280 if (imf == NULL) {
1281 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1282 imf = &timf;
1283 }
1284
1285 /*
1286 * Begin state merge transaction at MLD layer.
1287 *
1288 * As this particular invocation should not cause any memory
1289 * to be allocated, and there is no opportunity to roll back
1290 * the transaction, it MUST NOT fail.
1291 */
1292 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1293 error = in6m_merge(inm, imf);
1294 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1295
1296 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1297 error = mld_change_state(inm, 0);
1298 if (error)
1299 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1300
1301 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1302 in6m_release_locked(inm);
1303
1304 return (error);
1305}
1306
1307/*
1308 * Block or unblock an ASM multicast source on an inpcb.
1309 * This implements the delta-based API described in RFC 3678.
1310 *
1311 * The delta-based API applies only to exclusive-mode memberships.
1312 * An MLD downcall will be performed.
1313 *
1314 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1315 *
1316 * Return 0 if successful, otherwise return an appropriate error code.
1317 */
1318static int
1319in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1320{
1321 struct group_source_req gsr;
1322 sockunion_t *gsa, *ssa;
1323 struct ifnet *ifp;
1324 struct in6_mfilter *imf;
1325 struct ip6_moptions *imo;
1326 struct in6_msource *ims;
1327 struct in6_multi *inm;
1328 size_t idx;
1329 uint16_t fmode;
1330 int error, doblock;
1331#ifdef KTR
1332 char ip6tbuf[INET6_ADDRSTRLEN];
1333#endif
1334
1335 ifp = NULL;
1336 error = 0;
1337 doblock = 0;
1338
1339 memset(&gsr, 0, sizeof(struct group_source_req));
1340 gsa = (sockunion_t *)&gsr.gsr_group;
1341 ssa = (sockunion_t *)&gsr.gsr_source;
1342
1343 switch (sopt->sopt_name) {
1344 case MCAST_BLOCK_SOURCE:
1345 case MCAST_UNBLOCK_SOURCE:
1346 error = sooptcopyin(sopt, &gsr,
1347 sizeof(struct group_source_req),
1348 sizeof(struct group_source_req));
1349 if (error)
1350 return (error);
1351
1352 if (gsa->sin6.sin6_family != AF_INET6 ||
1353 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1354 return (EINVAL);
1355
1356 if (ssa->sin6.sin6_family != AF_INET6 ||
1357 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1358 return (EINVAL);
1359
1360 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1361 return (EADDRNOTAVAIL);
1362
1363 ifp = ifnet_byindex(gsr.gsr_interface);
1364
1365 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1366 doblock = 1;
1367 break;
1368
1369 default:
1370 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1371 __func__, sopt->sopt_name);
1372 return (EOPNOTSUPP);
1373 break;
1374 }
1375
1376 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1377 return (EINVAL);
1378
1379 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1380
1381 /*
1382 * Check if we are actually a member of this group.
1383 */
1384 imo = in6p_findmoptions(inp);
1385 idx = im6o_match_group(imo, ifp, &gsa->sa);
1386 if (idx == -1 || imo->im6o_mfilters == NULL) {
1387 error = EADDRNOTAVAIL;
1388 goto out_in6p_locked;
1389 }
1390
1391 KASSERT(imo->im6o_mfilters != NULL,
1392 ("%s: im6o_mfilters not allocated", __func__));
1393 imf = &imo->im6o_mfilters[idx];
1394 inm = imo->im6o_membership[idx];
1395
1396 /*
1397 * Attempting to use the delta-based API on an
1398 * non exclusive-mode membership is an error.
1399 */
1400 fmode = imf->im6f_st[0];
1401 if (fmode != MCAST_EXCLUDE) {
1402 error = EINVAL;
1403 goto out_in6p_locked;
1404 }
1405
1406 /*
1407 * Deal with error cases up-front:
1408 * Asked to block, but already blocked; or
1409 * Asked to unblock, but nothing to unblock.
1410 * If adding a new block entry, allocate it.
1411 */
1412 ims = im6o_match_source(imo, idx, &ssa->sa);
1413 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1414 CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1415 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1416 doblock ? "" : "not ");
1417 error = EADDRNOTAVAIL;
1418 goto out_in6p_locked;
1419 }
1420
1421 INP_WLOCK_ASSERT(inp);
1422
1423 /*
1424 * Begin state merge transaction at socket layer.
1425 */
1426 if (doblock) {
1427 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1428 ims = im6f_graft(imf, fmode, &ssa->sin6);
1429 if (ims == NULL)
1430 error = ENOMEM;
1431 } else {
1432 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1433 error = im6f_prune(imf, &ssa->sin6);
1434 }
1435
1436 if (error) {
1437 CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1438 goto out_im6f_rollback;
1439 }
1440
1441 /*
1442 * Begin state merge transaction at MLD layer.
1443 */
1444 IN6_MULTI_LOCK();
1445
1446 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1447 error = in6m_merge(inm, imf);
1448 if (error) {
1449 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1450 goto out_im6f_rollback;
1451 }
1452
1453 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1454 error = mld_change_state(inm, 0);
1455 if (error)
1456 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1457
1458 IN6_MULTI_UNLOCK();
1459
1460out_im6f_rollback:
1461 if (error)
1462 im6f_rollback(imf);
1463 else
1464 im6f_commit(imf);
1465
1466 im6f_reap(imf);
1467
1468out_in6p_locked:
1469 INP_WUNLOCK(inp);
1470 return (error);
1471}
1472
1473/*
1474 * Given an inpcb, return its multicast options structure pointer. Accepts
1475 * an unlocked inpcb pointer, but will return it locked. May sleep.
1476 *
1477 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1478 * SMPng: NOTE: Returns with the INP write lock held.
1479 */
1480static struct ip6_moptions *
1481in6p_findmoptions(struct inpcb *inp)
1482{
1483 struct ip6_moptions *imo;
1484 struct in6_multi **immp;
1485 struct in6_mfilter *imfp;
1486 size_t idx;
1487
1488 INP_WLOCK(inp);
1489 if (inp->in6p_moptions != NULL)
1490 return (inp->in6p_moptions);
1491
1492 INP_WUNLOCK(inp);
1493
1494 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1495 immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1496 M_WAITOK | M_ZERO);
1497 imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1498 M_IN6MFILTER, M_WAITOK);
1499
1500 imo->im6o_multicast_ifp = NULL;
1501 imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1502 imo->im6o_multicast_loop = in6_mcast_loop;
1503 imo->im6o_num_memberships = 0;
1504 imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1505 imo->im6o_membership = immp;
1506
1507 /* Initialize per-group source filters. */
1508 for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1509 im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1510 imo->im6o_mfilters = imfp;
1511
1512 INP_WLOCK(inp);
1513 if (inp->in6p_moptions != NULL) {
1514 free(imfp, M_IN6MFILTER);
1515 free(immp, M_IP6MOPTS);
1516 free(imo, M_IP6MOPTS);
1517 return (inp->in6p_moptions);
1518 }
1519 inp->in6p_moptions = imo;
1520 return (imo);
1521}
1522
1523/*
1524 * Discard the IPv6 multicast options (and source filters).
1525 *
1526 * SMPng: NOTE: assumes INP write lock is held.
1527 */
1528void
1529ip6_freemoptions(struct ip6_moptions *imo)
1530{
1531 struct in6_mfilter *imf;
1532 size_t idx, nmships;
1533
1534 KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1535
1536 nmships = imo->im6o_num_memberships;
1537 for (idx = 0; idx < nmships; ++idx) {
1538 imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1539 if (imf)
1540 im6f_leave(imf);
1541 /* XXX this will thrash the lock(s) */
1542 (void)in6_mc_leave(imo->im6o_membership[idx], imf);
1543 if (imf)
1544 im6f_purge(imf);
1545 }
1546
1547 if (imo->im6o_mfilters)
1548 free(imo->im6o_mfilters, M_IN6MFILTER);
1549 free(imo->im6o_membership, M_IP6MOPTS);
1550 free(imo, M_IP6MOPTS);
1551}
1552
1553/*
1554 * Atomically get source filters on a socket for an IPv6 multicast group.
1555 * Called with INP lock held; returns with lock released.
1556 */
1557static int
1558in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1559{
1560 struct __msfilterreq msfr;
1561 sockunion_t *gsa;
1562 struct ifnet *ifp;
1563 struct ip6_moptions *imo;
1564 struct in6_mfilter *imf;
1565 struct ip6_msource *ims;
1566 struct in6_msource *lims;
1567 struct sockaddr_in6 *psin;
1568 struct sockaddr_storage *ptss;
1569 struct sockaddr_storage *tss;
1570 int error;
1571 size_t idx, nsrcs, ncsrcs;
1572
1573 INP_WLOCK_ASSERT(inp);
1574
1575 imo = inp->in6p_moptions;
1576 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1577
1578 INP_WUNLOCK(inp);
1579
1580 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1581 sizeof(struct __msfilterreq));
1582 if (error)
1583 return (error);
1584
1585 if (msfr.msfr_group.ss_family != AF_INET6 ||
1586 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1587 return (EINVAL);
1588
1589 gsa = (sockunion_t *)&msfr.msfr_group;
1590 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1591 return (EINVAL);
1592
1593 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1594 return (EADDRNOTAVAIL);
1595 ifp = ifnet_byindex(msfr.msfr_ifindex);
1596 if (ifp == NULL)
1597 return (EADDRNOTAVAIL);
1598 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1599
1600 INP_WLOCK(inp);
1601
1602 /*
1603 * Lookup group on the socket.
1604 */
1605 idx = im6o_match_group(imo, ifp, &gsa->sa);
1606 if (idx == -1 || imo->im6o_mfilters == NULL) {
1607 INP_WUNLOCK(inp);
1608 return (EADDRNOTAVAIL);
1609 }
1610 imf = &imo->im6o_mfilters[idx];
1611
1612 /*
1613 * Ignore memberships which are in limbo.
1614 */
1615 if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1616 INP_WUNLOCK(inp);
1617 return (EAGAIN);
1618 }
1619 msfr.msfr_fmode = imf->im6f_st[1];
1620
1621 /*
1622 * If the user specified a buffer, copy out the source filter
1623 * entries to userland gracefully.
1624 * We only copy out the number of entries which userland
1625 * has asked for, but we always tell userland how big the
1626 * buffer really needs to be.
1627 */
1628 tss = NULL;
1629 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1630 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1631 M_TEMP, M_NOWAIT | M_ZERO);
1632 if (tss == NULL) {
1633 INP_WUNLOCK(inp);
1634 return (ENOBUFS);
1635 }
1636 }
1637
1638 /*
1639 * Count number of sources in-mode at t0.
1640 * If buffer space exists and remains, copy out source entries.
1641 */
1642 nsrcs = msfr.msfr_nsrcs;
1643 ncsrcs = 0;
1644 ptss = tss;
1645 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1646 lims = (struct in6_msource *)ims;
1647 if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1648 lims->im6sl_st[0] != imf->im6f_st[0])
1649 continue;
1650 ++ncsrcs;
1651 if (tss != NULL && nsrcs > 0) {
1652 psin = (struct sockaddr_in6 *)ptss;
1653 psin->sin6_family = AF_INET6;
1654 psin->sin6_len = sizeof(struct sockaddr_in6);
1655 psin->sin6_addr = lims->im6s_addr;
1656 psin->sin6_port = 0;
1657 --nsrcs;
1658 ++ptss;
1659 }
1660 }
1661
1662 INP_WUNLOCK(inp);
1663
1664 if (tss != NULL) {
1665 error = copyout(tss, msfr.msfr_srcs,
1666 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1667 free(tss, M_TEMP);
1668 if (error)
1669 return (error);
1670 }
1671
1672 msfr.msfr_nsrcs = ncsrcs;
1673 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1674
1675 return (error);
1676}
1677
1678/*
1679 * Return the IP multicast options in response to user getsockopt().
1680 */
1681int
1682ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1683{
1684 struct ip6_moptions *im6o;
1685 int error;
1686 u_int optval;
1687
1688 INP_WLOCK(inp);
1689 im6o = inp->in6p_moptions;
1690 /*
1691 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1692 * or is a divert socket, reject it.
1693 */
1694 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1695 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1696 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1697 INP_WUNLOCK(inp);
1698 return (EOPNOTSUPP);
1699 }
1700
1701 error = 0;
1702 switch (sopt->sopt_name) {
1703 case IPV6_MULTICAST_IF:
1704 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1705 optval = 0;
1706 } else {
1707 optval = im6o->im6o_multicast_ifp->if_index;
1708 }
1709 INP_WUNLOCK(inp);
1710 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1711 break;
1712
1713 case IPV6_MULTICAST_HOPS:
1714 if (im6o == NULL)
1715 optval = V_ip6_defmcasthlim;
1716 else
1717 optval = im6o->im6o_multicast_loop;
1718 INP_WUNLOCK(inp);
1719 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1720 break;
1721
1722 case IPV6_MULTICAST_LOOP:
1723 if (im6o == NULL)
1724 optval = in6_mcast_loop; /* XXX VIMAGE */
1725 else
1726 optval = im6o->im6o_multicast_loop;
1727 INP_WUNLOCK(inp);
1728 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1729 break;
1730
1731 case IPV6_MSFILTER:
1732 if (im6o == NULL) {
1733 error = EADDRNOTAVAIL;
1734 INP_WUNLOCK(inp);
1735 } else {
1736 error = in6p_get_source_filters(inp, sopt);
1737 }
1738 break;
1739
1740 default:
1741 INP_WUNLOCK(inp);
1742 error = ENOPROTOOPT;
1743 break;
1744 }
1745
1746 INP_UNLOCK_ASSERT(inp);
1747
1748 return (error);
1749}
1750
1751/*
1752 * Look up the ifnet to use for a multicast group membership,
1753 * given the address of an IPv6 group.
1754 *
1755 * This routine exists to support legacy IPv6 multicast applications.
1756 *
1757 * If inp is non-NULL, use this socket's current FIB number for any
1758 * required FIB lookup. Look up the group address in the unicast FIB,
1759 * and use its ifp; usually, this points to the default next-hop.
1760 * If the FIB lookup fails, return NULL.
1761 *
1762 * FUTURE: Support multiple forwarding tables for IPv6.
1763 *
1764 * Returns NULL if no ifp could be found.
1765 */
1766static struct ifnet *
1767in6p_lookup_mcast_ifp(const struct inpcb *in6p __unused,
1768 const struct sockaddr_in6 *gsin6)
1769{
1770 struct route_in6 ro6;
1771 struct ifnet *ifp;
1772
1773 KASSERT(in6p->inp_vflag & INP_IPV6,
1774 ("%s: not INP_IPV6 inpcb", __func__));
1775 KASSERT(gsin6->sin6_family == AF_INET6,
1776 ("%s: not AF_INET6 group", __func__));
1777 KASSERT(IN6_IS_ADDR_MULTICAST(&gsin6->sin6_addr),
1778 ("%s: not multicast", __func__));
1779
1780 ifp = NULL;
1781 memset(&ro6, 0, sizeof(struct route_in6));
1782 memcpy(&ro6.ro_dst, gsin6, sizeof(struct sockaddr_in6));
1783#ifdef notyet
1784 rtalloc_ign_fib(&ro6, 0, inp ? inp->inp_inc.inc_fibnum : 0);
1785#else
1786 rtalloc_ign((struct route *)&ro6, 0);
1787#endif
1788 if (ro6.ro_rt != NULL) {
1789 ifp = ro6.ro_rt->rt_ifp;
1790 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1791 RTFREE(ro6.ro_rt);
1792 }
1793
1794 return (ifp);
1795}
1796
1797/*
1798 * Join an IPv6 multicast group, possibly with a source.
1799 *
1800 * FIXME: The KAME use of the unspecified address (::)
1801 * to join *all* multicast groups is currently unsupported.
1802 */
1803static int
1804in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1805{
1806 struct group_source_req gsr;
1807 sockunion_t *gsa, *ssa;
1808 struct ifnet *ifp;
1809 struct in6_mfilter *imf;
1810 struct ip6_moptions *imo;
1811 struct in6_multi *inm;
1812 struct in6_msource *lims;
1813 size_t idx;
1814 int error, is_new;
1815
1816 ifp = NULL;
1817 imf = NULL;
1818 error = 0;
1819 is_new = 0;
1820
1821 memset(&gsr, 0, sizeof(struct group_source_req));
1822 gsa = (sockunion_t *)&gsr.gsr_group;
1823 gsa->ss.ss_family = AF_UNSPEC;
1824 ssa = (sockunion_t *)&gsr.gsr_source;
1825 ssa->ss.ss_family = AF_UNSPEC;
1826
1827 /*
1828 * Chew everything into struct group_source_req.
1829 * Overwrite the port field if present, as the sockaddr
1830 * being copied in may be matched with a binary comparison.
1831 * Ignore passed-in scope ID.
1832 */
1833 switch (sopt->sopt_name) {
1834 case IPV6_JOIN_GROUP: {
1835 struct ipv6_mreq mreq;
1836
1837 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1838 sizeof(struct ipv6_mreq));
1839 if (error)
1840 return (error);
1841
1842 gsa->sin6.sin6_family = AF_INET6;
1843 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1844 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1845
1846 if (mreq.ipv6mr_interface == 0) {
1847 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1848 } else {
1849 if (mreq.ipv6mr_interface < 0 ||
1850 V_if_index < mreq.ipv6mr_interface)
1851 return (EADDRNOTAVAIL);
1852 ifp = ifnet_byindex(mreq.ipv6mr_interface);
1853 }
1854 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1855 __func__, mreq.ipv6mr_interface, ifp);
1856 } break;
1857
1858 case MCAST_JOIN_GROUP:
1859 case MCAST_JOIN_SOURCE_GROUP:
1860 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1861 error = sooptcopyin(sopt, &gsr,
1862 sizeof(struct group_req),
1863 sizeof(struct group_req));
1864 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1865 error = sooptcopyin(sopt, &gsr,
1866 sizeof(struct group_source_req),
1867 sizeof(struct group_source_req));
1868 }
1869 if (error)
1870 return (error);
1871
1872 if (gsa->sin6.sin6_family != AF_INET6 ||
1873 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1874 return (EINVAL);
1875
1876 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1877 if (ssa->sin6.sin6_family != AF_INET6 ||
1878 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1879 return (EINVAL);
1880 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1881 return (EINVAL);
1882 /*
1883 * TODO: Validate embedded scope ID in source
1884 * list entry against passed-in ifp, if and only
1885 * if source list filter entry is iface or node local.
1886 */
1887 in6_clearscope(&ssa->sin6.sin6_addr);
1888 ssa->sin6.sin6_port = 0;
1889 ssa->sin6.sin6_scope_id = 0;
1890 }
1891
1892 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1893 return (EADDRNOTAVAIL);
1894 ifp = ifnet_byindex(gsr.gsr_interface);
1895 break;
1896
1897 default:
1898 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1899 __func__, sopt->sopt_name);
1900 return (EOPNOTSUPP);
1901 break;
1902 }
1903
1904 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1905 return (EINVAL);
1906
1907 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1908 return (EADDRNOTAVAIL);
1909
1910 gsa->sin6.sin6_port = 0;
1911 gsa->sin6.sin6_scope_id = 0;
1912
1913 /*
1914 * Always set the scope zone ID on memberships created from userland.
1915 * Use the passed-in ifp to do this.
1916 * XXX The in6_setscope() return value is meaningless.
1917 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1918 */
1919 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1920
1921 /*
1922 * MCAST_JOIN_SOURCE on an exclusive membership is an error.
1923 * On an existing inclusive membership, it just adds the
1924 * source to the filter list.
1925 */
1926 imo = in6p_findmoptions(inp);
1927 idx = im6o_match_group(imo, ifp, &gsa->sa);
1928 if (idx == -1) {
1929 is_new = 1;
1930 } else {
1931 inm = imo->im6o_membership[idx];
1932 imf = &imo->im6o_mfilters[idx];
1933 if (ssa->ss.ss_family != AF_UNSPEC &&
1934 imf->im6f_st[1] != MCAST_INCLUDE) {
1935 error = EINVAL;
1936 goto out_in6p_locked;
1937 }
1938 lims = im6o_match_source(imo, idx, &ssa->sa);
1939 if (lims != NULL) {
1940 error = EADDRNOTAVAIL;
1941 goto out_in6p_locked;
1942 }
1943 }
1944
1945 /*
1946 * Begin state merge transaction at socket layer.
1947 */
1948 INP_WLOCK_ASSERT(inp);
1949
1950 if (is_new) {
1951 if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
1952 error = im6o_grow(imo);
1953 if (error)
1954 goto out_in6p_locked;
1955 }
1956 /*
1957 * Allocate the new slot upfront so we can deal with
1958 * grafting the new source filter in same code path
1959 * as for join-source on existing membership.
1960 */
1961 idx = imo->im6o_num_memberships;
1962 imo->im6o_membership[idx] = NULL;
1963 imo->im6o_num_memberships++;
1964 KASSERT(imo->im6o_mfilters != NULL,
1965 ("%s: im6f_mfilters vector was not allocated", __func__));
1966 imf = &imo->im6o_mfilters[idx];
1967 KASSERT(RB_EMPTY(&imf->im6f_sources),
1968 ("%s: im6f_sources not empty", __func__));
1969 }
1970
1971 /*
1972 * Graft new source into filter list for this inpcb's
1973 * membership of the group. The in6_multi may not have
1974 * been allocated yet if this is a new membership.
1975 */
1976 if (ssa->ss.ss_family != AF_UNSPEC) {
1977 /* Membership starts in IN mode */
1978 if (is_new) {
1979 CTR1(KTR_MLD, "%s: new join w/source", __func__);
1980 im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
1981 } else {
1982 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1983 }
1984 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
1985 if (lims == NULL) {
1986 CTR1(KTR_MLD, "%s: merge imf state failed",
1987 __func__);
1988 error = ENOMEM;
1989 goto out_im6o_free;
1990 }
1991 }
1992
1993 /*
1994 * Begin state merge transaction at MLD layer.
1995 */
1996 IN6_MULTI_LOCK();
1997
1998 if (is_new) {
1999 error = in6_mc_join_locked(ifp, &gsa->sin6.sin6_addr, imf,
2000 &inm, 0);
2001 if (error)
2002 goto out_im6o_free;
2003 imo->im6o_membership[idx] = inm;
2004 } else {
2005 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2006 error = in6m_merge(inm, imf);
2007 if (error) {
2008 CTR1(KTR_MLD, "%s: failed to merge inm state",
2009 __func__);
2010 goto out_im6f_rollback;
2011 }
2012 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2013 error = mld_change_state(inm, 0);
2014 if (error) {
2015 CTR1(KTR_MLD, "%s: failed mld downcall",
2016 __func__);
2017 goto out_im6f_rollback;
2018 }
2019 }
2020
2021 IN6_MULTI_UNLOCK();
2022
2023out_im6f_rollback:
2024 INP_WLOCK_ASSERT(inp);
2025 if (error) {
2026 im6f_rollback(imf);
2027 if (is_new)
2028 im6f_purge(imf);
2029 else
2030 im6f_reap(imf);
2031 } else {
2032 im6f_commit(imf);
2033 }
2034
2035out_im6o_free:
2036 if (error && is_new) {
2037 imo->im6o_membership[idx] = NULL;
2038 --imo->im6o_num_memberships;
2039 }
2040
2041out_in6p_locked:
2042 INP_WUNLOCK(inp);
2043 return (error);
2044}
2045
2046/*
2047 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2048 */
2049static int
2050in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2051{
2052 struct ipv6_mreq mreq;
2053 struct group_source_req gsr;
2054 sockunion_t *gsa, *ssa;
2055 struct ifnet *ifp;
2056 struct in6_mfilter *imf;
2057 struct ip6_moptions *imo;
2058 struct in6_msource *ims;
2059 struct in6_multi *inm;
2060 uint32_t ifindex;
2061 size_t idx;
2062 int error, is_final;
2063#ifdef KTR
2064 char ip6tbuf[INET6_ADDRSTRLEN];
2065#endif
2066
2067 ifp = NULL;
2068 ifindex = 0;
2069 error = 0;
2070 is_final = 1;
2071
2072 memset(&gsr, 0, sizeof(struct group_source_req));
2073 gsa = (sockunion_t *)&gsr.gsr_group;
2074 gsa->ss.ss_family = AF_UNSPEC;
2075 ssa = (sockunion_t *)&gsr.gsr_source;
2076 ssa->ss.ss_family = AF_UNSPEC;
2077
2078 /*
2079 * Chew everything passed in up into a struct group_source_req
2080 * as that is easier to process.
2081 * Note: Any embedded scope ID in the multicast group passed
2082 * in by userland is ignored, the interface index is the recommended
2083 * mechanism to specify an interface; see below.
2084 */
2085 switch (sopt->sopt_name) {
2086 case IPV6_LEAVE_GROUP:
2087 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2088 sizeof(struct ipv6_mreq));
2089 if (error)
2090 return (error);
2091 gsa->sin6.sin6_family = AF_INET6;
2092 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2093 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2094 gsa->sin6.sin6_port = 0;
2095 gsa->sin6.sin6_scope_id = 0;
2096 ifindex = mreq.ipv6mr_interface;
2097 break;
2098
2099 case MCAST_LEAVE_GROUP:
2100 case MCAST_LEAVE_SOURCE_GROUP:
2101 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2102 error = sooptcopyin(sopt, &gsr,
2103 sizeof(struct group_req),
2104 sizeof(struct group_req));
2105 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2106 error = sooptcopyin(sopt, &gsr,
2107 sizeof(struct group_source_req),
2108 sizeof(struct group_source_req));
2109 }
2110 if (error)
2111 return (error);
2112
2113 if (gsa->sin6.sin6_family != AF_INET6 ||
2114 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2115 return (EINVAL);
2116 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2117 if (ssa->sin6.sin6_family != AF_INET6 ||
2118 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2119 return (EINVAL);
2120 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2121 return (EINVAL);
2122 /*
2123 * TODO: Validate embedded scope ID in source
2124 * list entry against passed-in ifp, if and only
2125 * if source list filter entry is iface or node local.
2126 */
2127 in6_clearscope(&ssa->sin6.sin6_addr);
2128 }
2129 gsa->sin6.sin6_port = 0;
2130 gsa->sin6.sin6_scope_id = 0;
2131 ifindex = gsr.gsr_interface;
2132 break;
2133
2134 default:
2135 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2136 __func__, sopt->sopt_name);
2137 return (EOPNOTSUPP);
2138 break;
2139 }
2140
2141 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2142 return (EINVAL);
2143
2144 /*
2145 * Validate interface index if provided. If no interface index
2146 * was provided separately, attempt to look the membership up
2147 * from the default scope as a last resort to disambiguate
2148 * the membership we are being asked to leave.
2149 * XXX SCOPE6 lock potentially taken here.
2150 */
2151 if (ifindex != 0) {
2152 if (ifindex < 0 || V_if_index < ifindex)
2153 return (EADDRNOTAVAIL);
2154 ifp = ifnet_byindex(ifindex);
2155 if (ifp == NULL)
2156 return (EADDRNOTAVAIL);
2157 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2158 } else {
2159 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2160 if (error)
2161 return (EADDRNOTAVAIL);
2162 /*
2163 * Some badly behaved applications don't pass an ifindex
2164 * or a scope ID, which is an API violation. In this case,
2165 * perform a lookup as per a v6 join.
2166 *
2167 * XXX For now, stomp on zone ID for the corner case.
2168 * This is not the 'KAME way', but we need to see the ifp
2169 * directly until such time as this implementation is
2170 * refactored, assuming the scope IDs are the way to go.
2171 */
2172 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2173 if (ifindex == 0) {
2174 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2175 "ifp for group %s.", __func__,
2176 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2177 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2178 } else {
2179 ifp = ifnet_byindex(ifindex);
2180 }
2181 if (ifp == NULL)
2182 return (EADDRNOTAVAIL);
2183 }
2184
2185 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2186 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2187
2188 /*
2189 * Find the membership in the membership array.
2190 */
2191 imo = in6p_findmoptions(inp);
2192 idx = im6o_match_group(imo, ifp, &gsa->sa);
2193 if (idx == -1) {
2194 error = EADDRNOTAVAIL;
2195 goto out_in6p_locked;
2196 }
2197 inm = imo->im6o_membership[idx];
2198 imf = &imo->im6o_mfilters[idx];
2199
2200 if (ssa->ss.ss_family != AF_UNSPEC)
2201 is_final = 0;
2202
2203 /*
2204 * Begin state merge transaction at socket layer.
2205 */
2206 INP_WLOCK_ASSERT(inp);
2207
2208 /*
2209 * If we were instructed only to leave a given source, do so.
2210 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2211 */
2212 if (is_final) {
2213 im6f_leave(imf);
2214 } else {
2215 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2216 error = EADDRNOTAVAIL;
2217 goto out_in6p_locked;
2218 }
2219 ims = im6o_match_source(imo, idx, &ssa->sa);
2220 if (ims == NULL) {
2221 CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2222 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2223 "not ");
2224 error = EADDRNOTAVAIL;
2225 goto out_in6p_locked;
2226 }
2227 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2228 error = im6f_prune(imf, &ssa->sin6);
2229 if (error) {
2230 CTR1(KTR_MLD, "%s: merge imf state failed",
2231 __func__);
2232 goto out_in6p_locked;
2233 }
2234 }
2235
2236 /*
2237 * Begin state merge transaction at MLD layer.
2238 */
2239 IN6_MULTI_LOCK();
2240
2241 if (is_final) {
2242 /*
2243 * Give up the multicast address record to which
2244 * the membership points.
2245 */
2246 (void)in6_mc_leave_locked(inm, imf);
2247 } else {
2248 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2249 error = in6m_merge(inm, imf);
2250 if (error) {
2251 CTR1(KTR_MLD, "%s: failed to merge inm state",
2252 __func__);
2253 goto out_im6f_rollback;
2254 }
2255
2256 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2257 error = mld_change_state(inm, 0);
2258 if (error) {
2259 CTR1(KTR_MLD, "%s: failed mld downcall",
2260 __func__);
2261 }
2262 }
2263
2264 IN6_MULTI_UNLOCK();
2265
2266out_im6f_rollback:
2267 if (error)
2268 im6f_rollback(imf);
2269 else
2270 im6f_commit(imf);
2271
2272 im6f_reap(imf);
2273
2274 if (is_final) {
2275 /* Remove the gap in the membership array. */
2276 for (++idx; idx < imo->im6o_num_memberships; ++idx)
2277 imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2278 imo->im6o_num_memberships--;
2279 }
2280
2281out_in6p_locked:
2282 INP_WUNLOCK(inp);
2283 return (error);
2284}
2285
2286/*
2287 * Select the interface for transmitting IPv6 multicast datagrams.
2288 *
2289 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2290 * may be passed to this socket option. An address of in6addr_any or an
2291 * interface index of 0 is used to remove a previous selection.
2292 * When no interface is selected, one is chosen for every send.
2293 */
2294static int
2295in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2296{
2297 struct ifnet *ifp;
2298 struct ip6_moptions *imo;
2299 u_int ifindex;
2300 int error;
2301
2302 if (sopt->sopt_valsize != sizeof(u_int))
2303 return (EINVAL);
2304
2305 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2306 if (error)
2307 return (error);
2308 if (ifindex < 0 || V_if_index < ifindex)
2309 return (EINVAL);
2310
2311 ifp = ifnet_byindex(ifindex);
2312 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2313 return (EADDRNOTAVAIL);
2314
2315 imo = in6p_findmoptions(inp);
2316 imo->im6o_multicast_ifp = ifp;
2317 INP_WUNLOCK(inp);
2318
2319 return (0);
2320}
2321
2322/*
2323 * Atomically set source filters on a socket for an IPv6 multicast group.
2324 *
2325 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2326 */
2327static int
2328in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2329{
2330 struct __msfilterreq msfr;
2331 sockunion_t *gsa;
2332 struct ifnet *ifp;
2333 struct in6_mfilter *imf;
2334 struct ip6_moptions *imo;
2335 struct in6_multi *inm;
2336 size_t idx;
2337 int error;
2338
2339 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2340 sizeof(struct __msfilterreq));
2341 if (error)
2342 return (error);
2343
2344 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc ||
2345 (msfr.msfr_fmode != MCAST_EXCLUDE &&
2346 msfr.msfr_fmode != MCAST_INCLUDE))
2347 return (EINVAL);
2348
2349 if (msfr.msfr_group.ss_family != AF_INET6 ||
2350 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2351 return (EINVAL);
2352
2353 gsa = (sockunion_t *)&msfr.msfr_group;
2354 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2355 return (EINVAL);
2356
2357 gsa->sin6.sin6_port = 0; /* ignore port */
2358
2359 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2360 return (EADDRNOTAVAIL);
2361 ifp = ifnet_byindex(msfr.msfr_ifindex);
2362 if (ifp == NULL)
2363 return (EADDRNOTAVAIL);
2364 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2365
2366 /*
2367 * Take the INP write lock.
2368 * Check if this socket is a member of this group.
2369 */
2370 imo = in6p_findmoptions(inp);
2371 idx = im6o_match_group(imo, ifp, &gsa->sa);
2372 if (idx == -1 || imo->im6o_mfilters == NULL) {
2373 error = EADDRNOTAVAIL;
2374 goto out_in6p_locked;
2375 }
2376 inm = imo->im6o_membership[idx];
2377 imf = &imo->im6o_mfilters[idx];
2378
2379 /*
2380 * Begin state merge transaction at socket layer.
2381 */
2382 INP_WLOCK_ASSERT(inp);
2383
2384 imf->im6f_st[1] = msfr.msfr_fmode;
2385
2386 /*
2387 * Apply any new source filters, if present.
2388 * Make a copy of the user-space source vector so
2389 * that we may copy them with a single copyin. This
2390 * allows us to deal with page faults up-front.
2391 */
2392 if (msfr.msfr_nsrcs > 0) {
2393 struct in6_msource *lims;
2394 struct sockaddr_in6 *psin;
2395 struct sockaddr_storage *kss, *pkss;
2396 int i;
2397
2398 INP_WUNLOCK(inp);
2399
2400 CTR2(KTR_MLD, "%s: loading %lu source list entries",
2401 __func__, (unsigned long)msfr.msfr_nsrcs);
2402 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2403 M_TEMP, M_WAITOK);
2404 error = copyin(msfr.msfr_srcs, kss,
2405 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2406 if (error) {
2407 free(kss, M_TEMP);
2408 return (error);
2409 }
2410
2411 INP_WLOCK(inp);
2412
2413 /*
2414 * Mark all source filters as UNDEFINED at t1.
2415 * Restore new group filter mode, as im6f_leave()
2416 * will set it to INCLUDE.
2417 */
2418 im6f_leave(imf);
2419 imf->im6f_st[1] = msfr.msfr_fmode;
2420
2421 /*
2422 * Update socket layer filters at t1, lazy-allocating
2423 * new entries. This saves a bunch of memory at the
2424 * cost of one RB_FIND() per source entry; duplicate
2425 * entries in the msfr_nsrcs vector are ignored.
2426 * If we encounter an error, rollback transaction.
2427 *
2428 * XXX This too could be replaced with a set-symmetric
2429 * difference like loop to avoid walking from root
2430 * every time, as the key space is common.
2431 */
2432 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2433 psin = (struct sockaddr_in6 *)pkss;
2434 if (psin->sin6_family != AF_INET6) {
2435 error = EAFNOSUPPORT;
2436 break;
2437 }
2438 if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2439 error = EINVAL;
2440 break;
2441 }
2442 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2443 error = EINVAL;
2444 break;
2445 }
2446 /*
2447 * TODO: Validate embedded scope ID in source
2448 * list entry against passed-in ifp, if and only
2449 * if source list filter entry is iface or node local.
2450 */
2451 in6_clearscope(&psin->sin6_addr);
2452 error = im6f_get_source(imf, psin, &lims);
2453 if (error)
2454 break;
2455 lims->im6sl_st[1] = imf->im6f_st[1];
2456 }
2457 free(kss, M_TEMP);
2458 }
2459
2460 if (error)
2461 goto out_im6f_rollback;
2462
2463 INP_WLOCK_ASSERT(inp);
2464 IN6_MULTI_LOCK();
2465
2466 /*
2467 * Begin state merge transaction at MLD layer.
2468 */
2469 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2470 error = in6m_merge(inm, imf);
2471 if (error) {
2472 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2473 goto out_im6f_rollback;
2474 }
2475
2476 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2477 error = mld_change_state(inm, 0);
2478 if (error)
2479 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2480
2481 IN6_MULTI_UNLOCK();
2482
2483out_im6f_rollback:
2484 if (error)
2485 im6f_rollback(imf);
2486 else
2487 im6f_commit(imf);
2488
2489 im6f_reap(imf);
2490
2491out_in6p_locked:
2492 INP_WUNLOCK(inp);
2493 return (error);
2494}
2495
2496/*
2497 * Set the IP multicast options in response to user setsockopt().
2498 *
2499 * Many of the socket options handled in this function duplicate the
2500 * functionality of socket options in the regular unicast API. However,
2501 * it is not possible to merge the duplicate code, because the idempotence
2502 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2503 * the effects of these options must be treated as separate and distinct.
2504 *
2505 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2506 */
2507int
2508ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2509{
2510 struct ip6_moptions *im6o;
2511 int error;
2512
2513 error = 0;
2514
2515 /*
2516 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2517 * or is a divert socket, reject it.
2518 */
2519 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2520 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2521 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2522 return (EOPNOTSUPP);
2523
2524 switch (sopt->sopt_name) {
2525 case IPV6_MULTICAST_IF:
2526 error = in6p_set_multicast_if(inp, sopt);
2527 break;
2528
2529 case IPV6_MULTICAST_HOPS: {
2530 int hlim;
2531
2532 if (sopt->sopt_valsize != sizeof(int)) {
2533 error = EINVAL;
2534 break;
2535 }
2536 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2537 if (error)
2538 break;
2539 if (hlim < -1 || hlim > 255) {
2540 error = EINVAL;
2541 break;
2542 } else if (hlim == -1) {
2543 hlim = V_ip6_defmcasthlim;
2544 }
2545 im6o = in6p_findmoptions(inp);
2546 im6o->im6o_multicast_hlim = hlim;
2547 INP_WUNLOCK(inp);
2548 break;
2549 }
2550
2551 case IPV6_MULTICAST_LOOP: {
2552 u_int loop;
2553
2554 /*
2555 * Set the loopback flag for outgoing multicast packets.
2556 * Must be zero or one.
2557 */
2558 if (sopt->sopt_valsize != sizeof(u_int)) {
2559 error = EINVAL;
2560 break;
2561 }
2562 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2563 if (error)
2564 break;
2565 if (loop > 1) {
2566 error = EINVAL;
2567 break;
2568 }
2569 im6o = in6p_findmoptions(inp);
2570 im6o->im6o_multicast_loop = loop;
2571 INP_WUNLOCK(inp);
2572 break;
2573 }
2574
2575 case IPV6_JOIN_GROUP:
2576 case MCAST_JOIN_GROUP:
2577 case MCAST_JOIN_SOURCE_GROUP:
2578 error = in6p_join_group(inp, sopt);
2579 break;
2580
2581 case IPV6_LEAVE_GROUP:
2582 case MCAST_LEAVE_GROUP:
2583 case MCAST_LEAVE_SOURCE_GROUP:
2584 error = in6p_leave_group(inp, sopt);
2585 break;
2586
2587 case MCAST_BLOCK_SOURCE:
2588 case MCAST_UNBLOCK_SOURCE:
2589 error = in6p_block_unblock_source(inp, sopt);
2590 break;
2591
2592 case IPV6_MSFILTER:
2593 error = in6p_set_source_filters(inp, sopt);
2594 break;
2595
2596 default:
2597 error = EOPNOTSUPP;
2598 break;
2599 }
2600
2601 INP_UNLOCK_ASSERT(inp);
2602
2603 return (error);
2604}
2605
2606/*
2607 * Expose MLD's multicast filter mode and source list(s) to userland,
2608 * keyed by (ifindex, group).
2609 * The filter mode is written out as a uint32_t, followed by
2610 * 0..n of struct in6_addr.
2611 * For use by ifmcstat(8).
2612 * SMPng: NOTE: unlocked read of ifindex space.
2613 */
2614static int
2615sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2616{
2617 struct in6_addr mcaddr;
2618 struct in6_addr src;
2619 struct ifnet *ifp;
2620 struct ifmultiaddr *ifma;
2621 struct in6_multi *inm;
2622 struct ip6_msource *ims;
2623 int *name;
2624 int retval;
2625 u_int namelen;
2626 uint32_t fmode, ifindex;
2627#ifdef KTR
2628 char ip6tbuf[INET6_ADDRSTRLEN];
2629#endif
2630
2631 name = (int *)arg1;
2632 namelen = arg2;
2633
2634 if (req->newptr != NULL)
2635 return (EPERM);
2636
2637 /* int: ifindex + 4 * 32 bits of IPv6 address */
2638 if (namelen != 5)
2639 return (EINVAL);
2640
2641 ifindex = name[0];
2642 if (ifindex <= 0 || ifindex > V_if_index) {
2643 CTR2(KTR_MLD, "%s: ifindex %u out of range",
2644 __func__, ifindex);
2645 return (ENOENT);
2646 }
2647
2648 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2649 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2650 CTR2(KTR_MLD, "%s: group %s is not multicast",
2651 __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2652 return (EINVAL);
2653 }
2654
2655 ifp = ifnet_byindex(ifindex);
2656 if (ifp == NULL) {
2657 CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2658 __func__, ifindex);
2659 return (ENOENT);
2660 }
2661 /*
2662 * Internal MLD lookups require that scope/zone ID is set.
2663 */
2664 (void)in6_setscope(&mcaddr, ifp, NULL);
2665
2666 retval = sysctl_wire_old_buffer(req,
2667 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2668 if (retval)
2669 return (retval);
2670
2671 IN6_MULTI_LOCK();
2672
2673 IF_ADDR_LOCK(ifp);
2674 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2675 if (ifma->ifma_addr->sa_family != AF_INET6 ||
2676 ifma->ifma_protospec == NULL)
2677 continue;
2678 inm = (struct in6_multi *)ifma->ifma_protospec;
2679 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2680 continue;
2681 fmode = inm->in6m_st[1].iss_fmode;
2682 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2683 if (retval != 0)
2684 break;
2685 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2686 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2687 /*
2688 * Only copy-out sources which are in-mode.
2689 */
2690 if (fmode != im6s_get_mode(inm, ims, 1)) {
2691 CTR1(KTR_MLD, "%s: skip non-in-mode",
2692 __func__);
2693 continue;
2694 }
2695 src = ims->im6s_addr;
2696 retval = SYSCTL_OUT(req, &src,
2697 sizeof(struct in6_addr));
2698 if (retval != 0)
2699 break;
2700 }
2701 }
2702 IF_ADDR_UNLOCK(ifp);
2703
2704 IN6_MULTI_UNLOCK();
2705
2706 return (retval);
2707}
2708
2709#ifdef KTR
2710
2711static const char *in6m_modestrs[] = { "un", "in", "ex" };
2712
2713static const char *
2714in6m_mode_str(const int mode)
2715{
2716
2717 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2718 return (in6m_modestrs[mode]);
2719 return ("??");
2720}
2721
2722static const char *in6m_statestrs[] = {
2723 "not-member",
2724 "silent",
2725 "idle",
2726 "lazy",
2727 "sleeping",
2728 "awakening",
2729 "query-pending",
2730 "sg-query-pending",
2731 "leaving"
2732};
2733
2734static const char *
2735in6m_state_str(const int state)
2736{
2737
2738 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2739 return (in6m_statestrs[state]);
2740 return ("??");
2741}
2742
2743/*
2744 * Dump an in6_multi structure to the console.
2745 */
2746void
2747in6m_print(const struct in6_multi *inm)
2748{
2749 int t;
2750 char ip6tbuf[INET6_ADDRSTRLEN];
2751
2752 if ((ktr_mask & KTR_MLD) == 0)
2753 return;
2754
2755 printf("%s: --- begin in6m %p ---\n", __func__, inm);
2756 printf("addr %s ifp %p(%s) ifma %p\n",
2757 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2758 inm->in6m_ifp,
2759 inm->in6m_ifp->if_xname,
2760 inm->in6m_ifma);
2761 printf("timer %u state %s refcount %u scq.len %u\n",
2762 inm->in6m_timer,
2763 in6m_state_str(inm->in6m_state),
2764 inm->in6m_refcount,
2765 inm->in6m_scq.ifq_len);
2766 printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2767 inm->in6m_mli,
2768 inm->in6m_nsrc,
2769 inm->in6m_sctimer,
2770 inm->in6m_scrv);
2771 for (t = 0; t < 2; t++) {
2772 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2773 in6m_mode_str(inm->in6m_st[t].iss_fmode),
2774 inm->in6m_st[t].iss_asm,
2775 inm->in6m_st[t].iss_ex,
2776 inm->in6m_st[t].iss_in,
2777 inm->in6m_st[t].iss_rec);
2778 }
2779 printf("%s: --- end in6m %p ---\n", __func__, inm);
2780}
2781
2782#else /* !KTR */
2783
2784void
2785in6m_print(const struct in6_multi *inm)
2786{
2787
2788}
2789
2790#endif /* KTR */
| 53
54#include <net/if.h>
55#include <net/if_dl.h>
56#include <net/route.h>
57#include <net/vnet.h>
58
59#include <netinet/in.h>
60#include <netinet/in_var.h>
61#include <netinet6/in6_var.h>
62#include <netinet/ip6.h>
63#include <netinet/icmp6.h>
64#include <netinet6/ip6_var.h>
65#include <netinet/in_pcb.h>
66#include <netinet/tcp_var.h>
67#include <netinet6/nd6.h>
68#include <netinet6/mld6_var.h>
69#include <netinet6/scope6_var.h>
70
71#ifndef KTR_MLD
72#define KTR_MLD KTR_INET6
73#endif
74
75#ifndef __SOCKUNION_DECLARED
76union sockunion {
77 struct sockaddr_storage ss;
78 struct sockaddr sa;
79 struct sockaddr_dl sdl;
80 struct sockaddr_in6 sin6;
81};
82typedef union sockunion sockunion_t;
83#define __SOCKUNION_DECLARED
84#endif /* __SOCKUNION_DECLARED */
85
86static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
87 "IPv6 multicast PCB-layer source filter");
88static MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
89static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
90static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
91 "IPv6 multicast MLD-layer source filter");
92
93RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
94
95/*
96 * Locking:
97 * - Lock order is: Giant, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
98 * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
99 * it can be taken by code in net/if.c also.
100 * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
101 *
102 * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
103 * any need for in6_multi itself to be virtualized -- it is bound to an ifp
104 * anyway no matter what happens.
105 */
106struct mtx in6_multi_mtx;
107MTX_SYSINIT(in6_multi_mtx, &in6_multi_mtx, "in6_multi_mtx", MTX_DEF);
108
109static void im6f_commit(struct in6_mfilter *);
110static int im6f_get_source(struct in6_mfilter *imf,
111 const struct sockaddr_in6 *psin,
112 struct in6_msource **);
113static struct in6_msource *
114 im6f_graft(struct in6_mfilter *, const uint8_t,
115 const struct sockaddr_in6 *);
116static void im6f_leave(struct in6_mfilter *);
117static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
118static void im6f_purge(struct in6_mfilter *);
119static void im6f_rollback(struct in6_mfilter *);
120static void im6f_reap(struct in6_mfilter *);
121static int im6o_grow(struct ip6_moptions *);
122static size_t im6o_match_group(const struct ip6_moptions *,
123 const struct ifnet *, const struct sockaddr *);
124static struct in6_msource *
125 im6o_match_source(const struct ip6_moptions *, const size_t,
126 const struct sockaddr *);
127static void im6s_merge(struct ip6_msource *ims,
128 const struct in6_msource *lims, const int rollback);
129static int in6_mc_get(struct ifnet *, const struct in6_addr *,
130 struct in6_multi **);
131static int in6m_get_source(struct in6_multi *inm,
132 const struct in6_addr *addr, const int noalloc,
133 struct ip6_msource **pims);
134static int in6m_is_ifp_detached(const struct in6_multi *);
135static int in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
136static void in6m_purge(struct in6_multi *);
137static void in6m_reap(struct in6_multi *);
138static struct ip6_moptions *
139 in6p_findmoptions(struct inpcb *);
140static int in6p_get_source_filters(struct inpcb *, struct sockopt *);
141static int in6p_join_group(struct inpcb *, struct sockopt *);
142static int in6p_leave_group(struct inpcb *, struct sockopt *);
143static struct ifnet *
144 in6p_lookup_mcast_ifp(const struct inpcb *,
145 const struct sockaddr_in6 *);
146static int in6p_block_unblock_source(struct inpcb *, struct sockopt *);
147static int in6p_set_multicast_if(struct inpcb *, struct sockopt *);
148static int in6p_set_source_filters(struct inpcb *, struct sockopt *);
149static int sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
150
151SYSCTL_DECL(_net_inet6_ip6); /* XXX Not in any common header. */
152
153SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast, CTLFLAG_RW, 0, "IPv6 multicast");
154
155static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
156SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
157 CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxgrpsrc, 0,
158 "Max source filters per group");
159TUNABLE_ULONG("net.inet6.ip6.mcast.maxgrpsrc", &in6_mcast_maxgrpsrc);
160
161static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
162SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
163 CTLFLAG_RW | CTLFLAG_TUN, &in6_mcast_maxsocksrc, 0,
164 "Max source filters per socket");
165TUNABLE_ULONG("net.inet6.ip6.mcast.maxsocksrc", &in6_mcast_maxsocksrc);
166
167/* TODO Virtualize this switch. */
168int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
169SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RW | CTLFLAG_TUN,
170 &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
171TUNABLE_INT("net.inet6.ip6.mcast.loop", &in6_mcast_loop);
172
173SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
174 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
175 "Per-interface stack-wide source filters");
176
177/*
178 * Inline function which wraps assertions for a valid ifp.
179 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
180 * is detached.
181 */
182static int __inline
183in6m_is_ifp_detached(const struct in6_multi *inm)
184{
185 struct ifnet *ifp;
186
187 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
188 ifp = inm->in6m_ifma->ifma_ifp;
189 if (ifp != NULL) {
190 /*
191 * Sanity check that network-layer notion of ifp is the
192 * same as that of link-layer.
193 */
194 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
195 }
196
197 return (ifp == NULL);
198}
199
200/*
201 * Initialize an in6_mfilter structure to a known state at t0, t1
202 * with an empty source filter list.
203 */
204static __inline void
205im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
206{
207 memset(imf, 0, sizeof(struct in6_mfilter));
208 RB_INIT(&imf->im6f_sources);
209 imf->im6f_st[0] = st0;
210 imf->im6f_st[1] = st1;
211}
212
213/*
214 * Resize the ip6_moptions vector to the next power-of-two minus 1.
215 * May be called with locks held; do not sleep.
216 */
217static int
218im6o_grow(struct ip6_moptions *imo)
219{
220 struct in6_multi **nmships;
221 struct in6_multi **omships;
222 struct in6_mfilter *nmfilters;
223 struct in6_mfilter *omfilters;
224 size_t idx;
225 size_t newmax;
226 size_t oldmax;
227
228 nmships = NULL;
229 nmfilters = NULL;
230 omships = imo->im6o_membership;
231 omfilters = imo->im6o_mfilters;
232 oldmax = imo->im6o_max_memberships;
233 newmax = ((oldmax + 1) * 2) - 1;
234
235 if (newmax <= IPV6_MAX_MEMBERSHIPS) {
236 nmships = (struct in6_multi **)realloc(omships,
237 sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
238 nmfilters = (struct in6_mfilter *)realloc(omfilters,
239 sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
240 M_NOWAIT);
241 if (nmships != NULL && nmfilters != NULL) {
242 /* Initialize newly allocated source filter heads. */
243 for (idx = oldmax; idx < newmax; idx++) {
244 im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
245 MCAST_EXCLUDE);
246 }
247 imo->im6o_max_memberships = newmax;
248 imo->im6o_membership = nmships;
249 imo->im6o_mfilters = nmfilters;
250 }
251 }
252
253 if (nmships == NULL || nmfilters == NULL) {
254 if (nmships != NULL)
255 free(nmships, M_IP6MOPTS);
256 if (nmfilters != NULL)
257 free(nmfilters, M_IN6MFILTER);
258 return (ETOOMANYREFS);
259 }
260
261 return (0);
262}
263
264/*
265 * Find an IPv6 multicast group entry for this ip6_moptions instance
266 * which matches the specified group, and optionally an interface.
267 * Return its index into the array, or -1 if not found.
268 */
269static size_t
270im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
271 const struct sockaddr *group)
272{
273 const struct sockaddr_in6 *gsin6;
274 struct in6_multi **pinm;
275 int idx;
276 int nmships;
277
278 gsin6 = (const struct sockaddr_in6 *)group;
279
280 /* The im6o_membership array may be lazy allocated. */
281 if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
282 return (-1);
283
284 nmships = imo->im6o_num_memberships;
285 pinm = &imo->im6o_membership[0];
286 for (idx = 0; idx < nmships; idx++, pinm++) {
287 if (*pinm == NULL)
288 continue;
289 if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
290 IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
291 &gsin6->sin6_addr)) {
292 break;
293 }
294 }
295 if (idx >= nmships)
296 idx = -1;
297
298 return (idx);
299}
300
301/*
302 * Find an IPv6 multicast source entry for this imo which matches
303 * the given group index for this socket, and source address.
304 *
305 * XXX TODO: The scope ID, if present in src, is stripped before
306 * any comparison. We SHOULD enforce scope/zone checks where the source
307 * filter entry has a link scope.
308 *
309 * NOTE: This does not check if the entry is in-mode, merely if
310 * it exists, which may not be the desired behaviour.
311 */
312static struct in6_msource *
313im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
314 const struct sockaddr *src)
315{
316 struct ip6_msource find;
317 struct in6_mfilter *imf;
318 struct ip6_msource *ims;
319 const sockunion_t *psa;
320
321 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
322 KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
323 ("%s: invalid index %d\n", __func__, (int)gidx));
324
325 /* The im6o_mfilters array may be lazy allocated. */
326 if (imo->im6o_mfilters == NULL)
327 return (NULL);
328 imf = &imo->im6o_mfilters[gidx];
329
330 psa = (const sockunion_t *)src;
331 find.im6s_addr = psa->sin6.sin6_addr;
332 in6_clearscope(&find.im6s_addr); /* XXX */
333 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
334
335 return ((struct in6_msource *)ims);
336}
337
338/*
339 * Perform filtering for multicast datagrams on a socket by group and source.
340 *
341 * Returns 0 if a datagram should be allowed through, or various error codes
342 * if the socket was not a member of the group, or the source was muted, etc.
343 */
344int
345im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
346 const struct sockaddr *group, const struct sockaddr *src)
347{
348 size_t gidx;
349 struct in6_msource *ims;
350 int mode;
351
352 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
353
354 gidx = im6o_match_group(imo, ifp, group);
355 if (gidx == -1)
356 return (MCAST_NOTGMEMBER);
357
358 /*
359 * Check if the source was included in an (S,G) join.
360 * Allow reception on exclusive memberships by default,
361 * reject reception on inclusive memberships by default.
362 * Exclude source only if an in-mode exclude filter exists.
363 * Include source only if an in-mode include filter exists.
364 * NOTE: We are comparing group state here at MLD t1 (now)
365 * with socket-layer t0 (since last downcall).
366 */
367 mode = imo->im6o_mfilters[gidx].im6f_st[1];
368 ims = im6o_match_source(imo, gidx, src);
369
370 if ((ims == NULL && mode == MCAST_INCLUDE) ||
371 (ims != NULL && ims->im6sl_st[0] != mode))
372 return (MCAST_NOTSMEMBER);
373
374 return (MCAST_PASS);
375}
376
377/*
378 * Find and return a reference to an in6_multi record for (ifp, group),
379 * and bump its reference count.
380 * If one does not exist, try to allocate it, and update link-layer multicast
381 * filters on ifp to listen for group.
382 * Assumes the IN6_MULTI lock is held across the call.
383 * Return 0 if successful, otherwise return an appropriate error code.
384 */
385static int
386in6_mc_get(struct ifnet *ifp, const struct in6_addr *group,
387 struct in6_multi **pinm)
388{
389 struct sockaddr_in6 gsin6;
390 struct ifmultiaddr *ifma;
391 struct in6_multi *inm;
392 int error;
393
394 error = 0;
395
396 /*
397 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
398 * if_addmulti() takes this mutex itself, so we must drop and
399 * re-acquire around the call.
400 */
401 IN6_MULTI_LOCK_ASSERT();
402 IF_ADDR_LOCK(ifp);
403
404 inm = in6m_lookup_locked(ifp, group);
405 if (inm != NULL) {
406 /*
407 * If we already joined this group, just bump the
408 * refcount and return it.
409 */
410 KASSERT(inm->in6m_refcount >= 1,
411 ("%s: bad refcount %d", __func__, inm->in6m_refcount));
412 ++inm->in6m_refcount;
413 *pinm = inm;
414 goto out_locked;
415 }
416
417 memset(&gsin6, 0, sizeof(gsin6));
418 gsin6.sin6_family = AF_INET6;
419 gsin6.sin6_len = sizeof(struct sockaddr_in6);
420 gsin6.sin6_addr = *group;
421
422 /*
423 * Check if a link-layer group is already associated
424 * with this network-layer group on the given ifnet.
425 */
426 IF_ADDR_UNLOCK(ifp);
427 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
428 if (error != 0)
429 return (error);
430 IF_ADDR_LOCK(ifp);
431
432 /*
433 * If something other than netinet6 is occupying the link-layer
434 * group, print a meaningful error message and back out of
435 * the allocation.
436 * Otherwise, bump the refcount on the existing network-layer
437 * group association and return it.
438 */
439 if (ifma->ifma_protospec != NULL) {
440 inm = (struct in6_multi *)ifma->ifma_protospec;
441#ifdef INVARIANTS
442 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
443 __func__));
444 KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
445 ("%s: ifma not AF_INET6", __func__));
446 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
447 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
448 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
449 panic("%s: ifma %p is inconsistent with %p (%p)",
450 __func__, ifma, inm, group);
451#endif
452 ++inm->in6m_refcount;
453 *pinm = inm;
454 goto out_locked;
455 }
456
457 IF_ADDR_LOCK_ASSERT(ifp);
458
459 /*
460 * A new in6_multi record is needed; allocate and initialize it.
461 * We DO NOT perform an MLD join as the in6_ layer may need to
462 * push an initial source list down to MLD to support SSM.
463 *
464 * The initial source filter state is INCLUDE, {} as per the RFC.
465 * Pending state-changes per group are subject to a bounds check.
466 */
467 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
468 if (inm == NULL) {
469 if_delmulti_ifma(ifma);
470 error = ENOMEM;
471 goto out_locked;
472 }
473 inm->in6m_addr = *group;
474 inm->in6m_ifp = ifp;
475 inm->in6m_mli = MLD_IFINFO(ifp);
476 inm->in6m_ifma = ifma;
477 inm->in6m_refcount = 1;
478 inm->in6m_state = MLD_NOT_MEMBER;
479 IFQ_SET_MAXLEN(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
480
481 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
482 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
483 RB_INIT(&inm->in6m_srcs);
484
485 ifma->ifma_protospec = inm;
486 *pinm = inm;
487
488out_locked:
489 IF_ADDR_UNLOCK(ifp);
490 return (error);
491}
492
493/*
494 * Drop a reference to an in6_multi record.
495 *
496 * If the refcount drops to 0, free the in6_multi record and
497 * delete the underlying link-layer membership.
498 */
499void
500in6m_release_locked(struct in6_multi *inm)
501{
502 struct ifmultiaddr *ifma;
503
504 IN6_MULTI_LOCK_ASSERT();
505
506 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
507
508 if (--inm->in6m_refcount > 0) {
509 CTR2(KTR_MLD, "%s: refcount is now %d", __func__,
510 inm->in6m_refcount);
511 return;
512 }
513
514 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
515
516 ifma = inm->in6m_ifma;
517
518 /* XXX this access is not covered by IF_ADDR_LOCK */
519 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
520 KASSERT(ifma->ifma_protospec == inm,
521 ("%s: ifma_protospec != inm", __func__));
522 ifma->ifma_protospec = NULL;
523
524 in6m_purge(inm);
525
526 free(inm, M_IP6MADDR);
527
528 if_delmulti_ifma(ifma);
529}
530
531/*
532 * Clear recorded source entries for a group.
533 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
534 * FIXME: Should reap.
535 */
536void
537in6m_clear_recorded(struct in6_multi *inm)
538{
539 struct ip6_msource *ims;
540
541 IN6_MULTI_LOCK_ASSERT();
542
543 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
544 if (ims->im6s_stp) {
545 ims->im6s_stp = 0;
546 --inm->in6m_st[1].iss_rec;
547 }
548 }
549 KASSERT(inm->in6m_st[1].iss_rec == 0,
550 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
551}
552
553/*
554 * Record a source as pending for a Source-Group MLDv2 query.
555 * This lives here as it modifies the shared tree.
556 *
557 * inm is the group descriptor.
558 * naddr is the address of the source to record in network-byte order.
559 *
560 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
561 * lazy-allocate a source node in response to an SG query.
562 * Otherwise, no allocation is performed. This saves some memory
563 * with the trade-off that the source will not be reported to the
564 * router if joined in the window between the query response and
565 * the group actually being joined on the local host.
566 *
567 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
568 * This turns off the allocation of a recorded source entry if
569 * the group has not been joined.
570 *
571 * Return 0 if the source didn't exist or was already marked as recorded.
572 * Return 1 if the source was marked as recorded by this function.
573 * Return <0 if any error occured (negated errno code).
574 */
575int
576in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
577{
578 struct ip6_msource find;
579 struct ip6_msource *ims, *nims;
580
581 IN6_MULTI_LOCK_ASSERT();
582
583 find.im6s_addr = *addr;
584 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
585 if (ims && ims->im6s_stp)
586 return (0);
587 if (ims == NULL) {
588 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
589 return (-ENOSPC);
590 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
591 M_NOWAIT | M_ZERO);
592 if (nims == NULL)
593 return (-ENOMEM);
594 nims->im6s_addr = find.im6s_addr;
595 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
596 ++inm->in6m_nsrc;
597 ims = nims;
598 }
599
600 /*
601 * Mark the source as recorded and update the recorded
602 * source count.
603 */
604 ++ims->im6s_stp;
605 ++inm->in6m_st[1].iss_rec;
606
607 return (1);
608}
609
610/*
611 * Return a pointer to an in6_msource owned by an in6_mfilter,
612 * given its source address.
613 * Lazy-allocate if needed. If this is a new entry its filter state is
614 * undefined at t0.
615 *
616 * imf is the filter set being modified.
617 * addr is the source address.
618 *
619 * SMPng: May be called with locks held; malloc must not block.
620 */
621static int
622im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
623 struct in6_msource **plims)
624{
625 struct ip6_msource find;
626 struct ip6_msource *ims, *nims;
627 struct in6_msource *lims;
628 int error;
629
630 error = 0;
631 ims = NULL;
632 lims = NULL;
633
634 find.im6s_addr = psin->sin6_addr;
635 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
636 lims = (struct in6_msource *)ims;
637 if (lims == NULL) {
638 if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
639 return (ENOSPC);
640 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
641 M_NOWAIT | M_ZERO);
642 if (nims == NULL)
643 return (ENOMEM);
644 lims = (struct in6_msource *)nims;
645 lims->im6s_addr = find.im6s_addr;
646 lims->im6sl_st[0] = MCAST_UNDEFINED;
647 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
648 ++imf->im6f_nsrc;
649 }
650
651 *plims = lims;
652
653 return (error);
654}
655
656/*
657 * Graft a source entry into an existing socket-layer filter set,
658 * maintaining any required invariants and checking allocations.
659 *
660 * The source is marked as being in the new filter mode at t1.
661 *
662 * Return the pointer to the new node, otherwise return NULL.
663 */
664static struct in6_msource *
665im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
666 const struct sockaddr_in6 *psin)
667{
668 struct ip6_msource *nims;
669 struct in6_msource *lims;
670
671 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
672 M_NOWAIT | M_ZERO);
673 if (nims == NULL)
674 return (NULL);
675 lims = (struct in6_msource *)nims;
676 lims->im6s_addr = psin->sin6_addr;
677 lims->im6sl_st[0] = MCAST_UNDEFINED;
678 lims->im6sl_st[1] = st1;
679 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
680 ++imf->im6f_nsrc;
681
682 return (lims);
683}
684
685/*
686 * Prune a source entry from an existing socket-layer filter set,
687 * maintaining any required invariants and checking allocations.
688 *
689 * The source is marked as being left at t1, it is not freed.
690 *
691 * Return 0 if no error occurred, otherwise return an errno value.
692 */
693static int
694im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
695{
696 struct ip6_msource find;
697 struct ip6_msource *ims;
698 struct in6_msource *lims;
699
700 find.im6s_addr = psin->sin6_addr;
701 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
702 if (ims == NULL)
703 return (ENOENT);
704 lims = (struct in6_msource *)ims;
705 lims->im6sl_st[1] = MCAST_UNDEFINED;
706 return (0);
707}
708
709/*
710 * Revert socket-layer filter set deltas at t1 to t0 state.
711 */
712static void
713im6f_rollback(struct in6_mfilter *imf)
714{
715 struct ip6_msource *ims, *tims;
716 struct in6_msource *lims;
717
718 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
719 lims = (struct in6_msource *)ims;
720 if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
721 /* no change at t1 */
722 continue;
723 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
724 /* revert change to existing source at t1 */
725 lims->im6sl_st[1] = lims->im6sl_st[0];
726 } else {
727 /* revert source added t1 */
728 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
729 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
730 free(ims, M_IN6MFILTER);
731 imf->im6f_nsrc--;
732 }
733 }
734 imf->im6f_st[1] = imf->im6f_st[0];
735}
736
737/*
738 * Mark socket-layer filter set as INCLUDE {} at t1.
739 */
740static void
741im6f_leave(struct in6_mfilter *imf)
742{
743 struct ip6_msource *ims;
744 struct in6_msource *lims;
745
746 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
747 lims = (struct in6_msource *)ims;
748 lims->im6sl_st[1] = MCAST_UNDEFINED;
749 }
750 imf->im6f_st[1] = MCAST_INCLUDE;
751}
752
753/*
754 * Mark socket-layer filter set deltas as committed.
755 */
756static void
757im6f_commit(struct in6_mfilter *imf)
758{
759 struct ip6_msource *ims;
760 struct in6_msource *lims;
761
762 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
763 lims = (struct in6_msource *)ims;
764 lims->im6sl_st[0] = lims->im6sl_st[1];
765 }
766 imf->im6f_st[0] = imf->im6f_st[1];
767}
768
769/*
770 * Reap unreferenced sources from socket-layer filter set.
771 */
772static void
773im6f_reap(struct in6_mfilter *imf)
774{
775 struct ip6_msource *ims, *tims;
776 struct in6_msource *lims;
777
778 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
779 lims = (struct in6_msource *)ims;
780 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
781 (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
782 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
783 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
784 free(ims, M_IN6MFILTER);
785 imf->im6f_nsrc--;
786 }
787 }
788}
789
790/*
791 * Purge socket-layer filter set.
792 */
793static void
794im6f_purge(struct in6_mfilter *imf)
795{
796 struct ip6_msource *ims, *tims;
797
798 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
799 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
800 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
801 free(ims, M_IN6MFILTER);
802 imf->im6f_nsrc--;
803 }
804 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
805 KASSERT(RB_EMPTY(&imf->im6f_sources),
806 ("%s: im6f_sources not empty", __func__));
807}
808
809/*
810 * Look up a source filter entry for a multicast group.
811 *
812 * inm is the group descriptor to work with.
813 * addr is the IPv6 address to look up.
814 * noalloc may be non-zero to suppress allocation of sources.
815 * *pims will be set to the address of the retrieved or allocated source.
816 *
817 * SMPng: NOTE: may be called with locks held.
818 * Return 0 if successful, otherwise return a non-zero error code.
819 */
820static int
821in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
822 const int noalloc, struct ip6_msource **pims)
823{
824 struct ip6_msource find;
825 struct ip6_msource *ims, *nims;
826#ifdef KTR
827 char ip6tbuf[INET6_ADDRSTRLEN];
828#endif
829
830 find.im6s_addr = *addr;
831 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
832 if (ims == NULL && !noalloc) {
833 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
834 return (ENOSPC);
835 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
836 M_NOWAIT | M_ZERO);
837 if (nims == NULL)
838 return (ENOMEM);
839 nims->im6s_addr = *addr;
840 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
841 ++inm->in6m_nsrc;
842 ims = nims;
843 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
844 ip6_sprintf(ip6tbuf, addr), ims);
845 }
846
847 *pims = ims;
848 return (0);
849}
850
851/*
852 * Merge socket-layer source into MLD-layer source.
853 * If rollback is non-zero, perform the inverse of the merge.
854 */
855static void
856im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
857 const int rollback)
858{
859 int n = rollback ? -1 : 1;
860#ifdef KTR
861 char ip6tbuf[INET6_ADDRSTRLEN];
862
863 ip6_sprintf(ip6tbuf, &lims->im6s_addr);
864#endif
865
866 if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
867 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
868 ims->im6s_st[1].ex -= n;
869 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
870 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
871 ims->im6s_st[1].in -= n;
872 }
873
874 if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
875 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
876 ims->im6s_st[1].ex += n;
877 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
878 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
879 ims->im6s_st[1].in += n;
880 }
881}
882
883/*
884 * Atomically update the global in6_multi state, when a membership's
885 * filter list is being updated in any way.
886 *
887 * imf is the per-inpcb-membership group filter pointer.
888 * A fake imf may be passed for in-kernel consumers.
889 *
890 * XXX This is a candidate for a set-symmetric-difference style loop
891 * which would eliminate the repeated lookup from root of ims nodes,
892 * as they share the same key space.
893 *
894 * If any error occurred this function will back out of refcounts
895 * and return a non-zero value.
896 */
897static int
898in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
899{
900 struct ip6_msource *ims, *nims;
901 struct in6_msource *lims;
902 int schanged, error;
903 int nsrc0, nsrc1;
904
905 schanged = 0;
906 error = 0;
907 nsrc1 = nsrc0 = 0;
908
909 /*
910 * Update the source filters first, as this may fail.
911 * Maintain count of in-mode filters at t0, t1. These are
912 * used to work out if we transition into ASM mode or not.
913 * Maintain a count of source filters whose state was
914 * actually modified by this operation.
915 */
916 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
917 lims = (struct in6_msource *)ims;
918 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
919 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
920 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
921 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
922 ++schanged;
923 if (error)
924 break;
925 im6s_merge(nims, lims, 0);
926 }
927 if (error) {
928 struct ip6_msource *bims;
929
930 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
931 lims = (struct in6_msource *)ims;
932 if (lims->im6sl_st[0] == lims->im6sl_st[1])
933 continue;
934 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
935 if (bims == NULL)
936 continue;
937 im6s_merge(bims, lims, 1);
938 }
939 goto out_reap;
940 }
941
942 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
943 __func__, nsrc0, nsrc1);
944
945 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
946 if (imf->im6f_st[0] == imf->im6f_st[1] &&
947 imf->im6f_st[1] == MCAST_INCLUDE) {
948 if (nsrc1 == 0) {
949 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
950 --inm->in6m_st[1].iss_in;
951 }
952 }
953
954 /* Handle filter mode transition on socket. */
955 if (imf->im6f_st[0] != imf->im6f_st[1]) {
956 CTR3(KTR_MLD, "%s: imf transition %d to %d",
957 __func__, imf->im6f_st[0], imf->im6f_st[1]);
958
959 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
960 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
961 --inm->in6m_st[1].iss_ex;
962 } else if (imf->im6f_st[0] == MCAST_INCLUDE) {
963 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
964 --inm->in6m_st[1].iss_in;
965 }
966
967 if (imf->im6f_st[1] == MCAST_EXCLUDE) {
968 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
969 inm->in6m_st[1].iss_ex++;
970 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
971 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
972 inm->in6m_st[1].iss_in++;
973 }
974 }
975
976 /*
977 * Track inm filter state in terms of listener counts.
978 * If there are any exclusive listeners, stack-wide
979 * membership is exclusive.
980 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
981 * If no listeners remain, state is undefined at t1,
982 * and the MLD lifecycle for this group should finish.
983 */
984 if (inm->in6m_st[1].iss_ex > 0) {
985 CTR1(KTR_MLD, "%s: transition to EX", __func__);
986 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
987 } else if (inm->in6m_st[1].iss_in > 0) {
988 CTR1(KTR_MLD, "%s: transition to IN", __func__);
989 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
990 } else {
991 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
992 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
993 }
994
995 /* Decrement ASM listener count on transition out of ASM mode. */
996 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
997 if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
998 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0))
999 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1000 --inm->in6m_st[1].iss_asm;
1001 }
1002
1003 /* Increment ASM listener count on transition to ASM mode. */
1004 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1005 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1006 inm->in6m_st[1].iss_asm++;
1007 }
1008
1009 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1010 in6m_print(inm);
1011
1012out_reap:
1013 if (schanged > 0) {
1014 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1015 in6m_reap(inm);
1016 }
1017 return (error);
1018}
1019
1020/*
1021 * Mark an in6_multi's filter set deltas as committed.
1022 * Called by MLD after a state change has been enqueued.
1023 */
1024void
1025in6m_commit(struct in6_multi *inm)
1026{
1027 struct ip6_msource *ims;
1028
1029 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1030 CTR1(KTR_MLD, "%s: pre commit:", __func__);
1031 in6m_print(inm);
1032
1033 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1034 ims->im6s_st[0] = ims->im6s_st[1];
1035 }
1036 inm->in6m_st[0] = inm->in6m_st[1];
1037}
1038
1039/*
1040 * Reap unreferenced nodes from an in6_multi's filter set.
1041 */
1042static void
1043in6m_reap(struct in6_multi *inm)
1044{
1045 struct ip6_msource *ims, *tims;
1046
1047 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1048 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1049 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1050 ims->im6s_stp != 0)
1051 continue;
1052 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1053 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1054 free(ims, M_IP6MSOURCE);
1055 inm->in6m_nsrc--;
1056 }
1057}
1058
1059/*
1060 * Purge all source nodes from an in6_multi's filter set.
1061 */
1062static void
1063in6m_purge(struct in6_multi *inm)
1064{
1065 struct ip6_msource *ims, *tims;
1066
1067 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1068 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1069 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1070 free(ims, M_IP6MSOURCE);
1071 inm->in6m_nsrc--;
1072 }
1073}
1074
1075/*
1076 * Join a multicast address w/o sources.
1077 * KAME compatibility entry point.
1078 *
1079 * SMPng: Assume no mc locks held by caller.
1080 */
1081struct in6_multi_mship *
1082in6_joingroup(struct ifnet *ifp, struct in6_addr *mcaddr,
1083 int *errorp, int delay)
1084{
1085 struct in6_multi_mship *imm;
1086 int error;
1087
1088 imm = malloc(sizeof(*imm), M_IP6MADDR, M_NOWAIT);
1089 if (imm == NULL) {
1090 *errorp = ENOBUFS;
1091 return (NULL);
1092 }
1093
1094 delay = (delay * PR_FASTHZ) / hz;
1095
1096 error = in6_mc_join(ifp, mcaddr, NULL, &imm->i6mm_maddr, delay);
1097 if (error) {
1098 *errorp = error;
1099 free(imm, M_IP6MADDR);
1100 return (NULL);
1101 }
1102
1103 return (imm);
1104}
1105
1106/*
1107 * Leave a multicast address w/o sources.
1108 * KAME compatibility entry point.
1109 *
1110 * SMPng: Assume no mc locks held by caller.
1111 */
1112int
1113in6_leavegroup(struct in6_multi_mship *imm)
1114{
1115
1116 if (imm->i6mm_maddr != NULL)
1117 in6_mc_leave(imm->i6mm_maddr, NULL);
1118 free(imm, M_IP6MADDR);
1119 return 0;
1120}
1121
1122/*
1123 * Join a multicast group; unlocked entry point.
1124 *
1125 * SMPng: XXX: in6_mc_join() is called from in6_control() when upper
1126 * locks are not held. Fortunately, ifp is unlikely to have been detached
1127 * at this point, so we assume it's OK to recurse.
1128 */
1129int
1130in6_mc_join(struct ifnet *ifp, const struct in6_addr *mcaddr,
1131 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1132 const int delay)
1133{
1134 int error;
1135
1136 IN6_MULTI_LOCK();
1137 error = in6_mc_join_locked(ifp, mcaddr, imf, pinm, delay);
1138 IN6_MULTI_UNLOCK();
1139
1140 return (error);
1141}
1142
1143/*
1144 * Join a multicast group; real entry point.
1145 *
1146 * Only preserves atomicity at inm level.
1147 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1148 *
1149 * If the MLD downcall fails, the group is not joined, and an error
1150 * code is returned.
1151 */
1152int
1153in6_mc_join_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1154 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1155 const int delay)
1156{
1157 struct in6_mfilter timf;
1158 struct in6_multi *inm;
1159 int error;
1160#ifdef KTR
1161 char ip6tbuf[INET6_ADDRSTRLEN];
1162#endif
1163
1164#ifdef INVARIANTS
1165 /*
1166 * Sanity: Check scope zone ID was set for ifp, if and
1167 * only if group is scoped to an interface.
1168 */
1169 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1170 ("%s: not a multicast address", __func__));
1171 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1172 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1173 KASSERT(mcaddr->s6_addr16[1] != 0,
1174 ("%s: scope zone ID not set", __func__));
1175 }
1176#endif
1177
1178 IN6_MULTI_LOCK_ASSERT();
1179
1180 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1181 ip6_sprintf(ip6tbuf, mcaddr), ifp, ifp->if_xname);
1182
1183 error = 0;
1184 inm = NULL;
1185
1186 /*
1187 * If no imf was specified (i.e. kernel consumer),
1188 * fake one up and assume it is an ASM join.
1189 */
1190 if (imf == NULL) {
1191 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1192 imf = &timf;
1193 }
1194
1195 error = in6_mc_get(ifp, mcaddr, &inm);
1196 if (error) {
1197 CTR1(KTR_MLD, "%s: in6_mc_get() failure", __func__);
1198 return (error);
1199 }
1200
1201 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1202 error = in6m_merge(inm, imf);
1203 if (error) {
1204 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1205 goto out_in6m_release;
1206 }
1207
1208 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1209 error = mld_change_state(inm, delay);
1210 if (error) {
1211 CTR1(KTR_MLD, "%s: failed to update source", __func__);
1212 goto out_in6m_release;
1213 }
1214
1215out_in6m_release:
1216 if (error) {
1217 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1218 in6m_release_locked(inm);
1219 } else {
1220 *pinm = inm;
1221 }
1222
1223 return (error);
1224}
1225
1226/*
1227 * Leave a multicast group; unlocked entry point.
1228 */
1229int
1230in6_mc_leave(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1231{
1232 struct ifnet *ifp;
1233 int error;
1234
1235 ifp = inm->in6m_ifp;
1236
1237 IN6_MULTI_LOCK();
1238 error = in6_mc_leave_locked(inm, imf);
1239 IN6_MULTI_UNLOCK();
1240
1241 return (error);
1242}
1243
1244/*
1245 * Leave a multicast group; real entry point.
1246 * All source filters will be expunged.
1247 *
1248 * Only preserves atomicity at inm level.
1249 *
1250 * Holding the write lock for the INP which contains imf
1251 * is highly advisable. We can't assert for it as imf does not
1252 * contain a back-pointer to the owning inp.
1253 *
1254 * Note: This is not the same as in6m_release(*) as this function also
1255 * makes a state change downcall into MLD.
1256 */
1257int
1258in6_mc_leave_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1259{
1260 struct in6_mfilter timf;
1261 int error;
1262#ifdef KTR
1263 char ip6tbuf[INET6_ADDRSTRLEN];
1264#endif
1265
1266 error = 0;
1267
1268 IN6_MULTI_LOCK_ASSERT();
1269
1270 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1271 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1272 (in6m_is_ifp_detached(inm) ? "null" : inm->in6m_ifp->if_xname),
1273 imf);
1274
1275 /*
1276 * If no imf was specified (i.e. kernel consumer),
1277 * fake one up and assume it is an ASM join.
1278 */
1279 if (imf == NULL) {
1280 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1281 imf = &timf;
1282 }
1283
1284 /*
1285 * Begin state merge transaction at MLD layer.
1286 *
1287 * As this particular invocation should not cause any memory
1288 * to be allocated, and there is no opportunity to roll back
1289 * the transaction, it MUST NOT fail.
1290 */
1291 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1292 error = in6m_merge(inm, imf);
1293 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1294
1295 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1296 error = mld_change_state(inm, 0);
1297 if (error)
1298 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1299
1300 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1301 in6m_release_locked(inm);
1302
1303 return (error);
1304}
1305
1306/*
1307 * Block or unblock an ASM multicast source on an inpcb.
1308 * This implements the delta-based API described in RFC 3678.
1309 *
1310 * The delta-based API applies only to exclusive-mode memberships.
1311 * An MLD downcall will be performed.
1312 *
1313 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1314 *
1315 * Return 0 if successful, otherwise return an appropriate error code.
1316 */
1317static int
1318in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1319{
1320 struct group_source_req gsr;
1321 sockunion_t *gsa, *ssa;
1322 struct ifnet *ifp;
1323 struct in6_mfilter *imf;
1324 struct ip6_moptions *imo;
1325 struct in6_msource *ims;
1326 struct in6_multi *inm;
1327 size_t idx;
1328 uint16_t fmode;
1329 int error, doblock;
1330#ifdef KTR
1331 char ip6tbuf[INET6_ADDRSTRLEN];
1332#endif
1333
1334 ifp = NULL;
1335 error = 0;
1336 doblock = 0;
1337
1338 memset(&gsr, 0, sizeof(struct group_source_req));
1339 gsa = (sockunion_t *)&gsr.gsr_group;
1340 ssa = (sockunion_t *)&gsr.gsr_source;
1341
1342 switch (sopt->sopt_name) {
1343 case MCAST_BLOCK_SOURCE:
1344 case MCAST_UNBLOCK_SOURCE:
1345 error = sooptcopyin(sopt, &gsr,
1346 sizeof(struct group_source_req),
1347 sizeof(struct group_source_req));
1348 if (error)
1349 return (error);
1350
1351 if (gsa->sin6.sin6_family != AF_INET6 ||
1352 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1353 return (EINVAL);
1354
1355 if (ssa->sin6.sin6_family != AF_INET6 ||
1356 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1357 return (EINVAL);
1358
1359 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1360 return (EADDRNOTAVAIL);
1361
1362 ifp = ifnet_byindex(gsr.gsr_interface);
1363
1364 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1365 doblock = 1;
1366 break;
1367
1368 default:
1369 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1370 __func__, sopt->sopt_name);
1371 return (EOPNOTSUPP);
1372 break;
1373 }
1374
1375 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1376 return (EINVAL);
1377
1378 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1379
1380 /*
1381 * Check if we are actually a member of this group.
1382 */
1383 imo = in6p_findmoptions(inp);
1384 idx = im6o_match_group(imo, ifp, &gsa->sa);
1385 if (idx == -1 || imo->im6o_mfilters == NULL) {
1386 error = EADDRNOTAVAIL;
1387 goto out_in6p_locked;
1388 }
1389
1390 KASSERT(imo->im6o_mfilters != NULL,
1391 ("%s: im6o_mfilters not allocated", __func__));
1392 imf = &imo->im6o_mfilters[idx];
1393 inm = imo->im6o_membership[idx];
1394
1395 /*
1396 * Attempting to use the delta-based API on an
1397 * non exclusive-mode membership is an error.
1398 */
1399 fmode = imf->im6f_st[0];
1400 if (fmode != MCAST_EXCLUDE) {
1401 error = EINVAL;
1402 goto out_in6p_locked;
1403 }
1404
1405 /*
1406 * Deal with error cases up-front:
1407 * Asked to block, but already blocked; or
1408 * Asked to unblock, but nothing to unblock.
1409 * If adding a new block entry, allocate it.
1410 */
1411 ims = im6o_match_source(imo, idx, &ssa->sa);
1412 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1413 CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1414 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1415 doblock ? "" : "not ");
1416 error = EADDRNOTAVAIL;
1417 goto out_in6p_locked;
1418 }
1419
1420 INP_WLOCK_ASSERT(inp);
1421
1422 /*
1423 * Begin state merge transaction at socket layer.
1424 */
1425 if (doblock) {
1426 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1427 ims = im6f_graft(imf, fmode, &ssa->sin6);
1428 if (ims == NULL)
1429 error = ENOMEM;
1430 } else {
1431 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1432 error = im6f_prune(imf, &ssa->sin6);
1433 }
1434
1435 if (error) {
1436 CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1437 goto out_im6f_rollback;
1438 }
1439
1440 /*
1441 * Begin state merge transaction at MLD layer.
1442 */
1443 IN6_MULTI_LOCK();
1444
1445 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1446 error = in6m_merge(inm, imf);
1447 if (error) {
1448 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1449 goto out_im6f_rollback;
1450 }
1451
1452 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1453 error = mld_change_state(inm, 0);
1454 if (error)
1455 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1456
1457 IN6_MULTI_UNLOCK();
1458
1459out_im6f_rollback:
1460 if (error)
1461 im6f_rollback(imf);
1462 else
1463 im6f_commit(imf);
1464
1465 im6f_reap(imf);
1466
1467out_in6p_locked:
1468 INP_WUNLOCK(inp);
1469 return (error);
1470}
1471
1472/*
1473 * Given an inpcb, return its multicast options structure pointer. Accepts
1474 * an unlocked inpcb pointer, but will return it locked. May sleep.
1475 *
1476 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1477 * SMPng: NOTE: Returns with the INP write lock held.
1478 */
1479static struct ip6_moptions *
1480in6p_findmoptions(struct inpcb *inp)
1481{
1482 struct ip6_moptions *imo;
1483 struct in6_multi **immp;
1484 struct in6_mfilter *imfp;
1485 size_t idx;
1486
1487 INP_WLOCK(inp);
1488 if (inp->in6p_moptions != NULL)
1489 return (inp->in6p_moptions);
1490
1491 INP_WUNLOCK(inp);
1492
1493 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1494 immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1495 M_WAITOK | M_ZERO);
1496 imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1497 M_IN6MFILTER, M_WAITOK);
1498
1499 imo->im6o_multicast_ifp = NULL;
1500 imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1501 imo->im6o_multicast_loop = in6_mcast_loop;
1502 imo->im6o_num_memberships = 0;
1503 imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1504 imo->im6o_membership = immp;
1505
1506 /* Initialize per-group source filters. */
1507 for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1508 im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1509 imo->im6o_mfilters = imfp;
1510
1511 INP_WLOCK(inp);
1512 if (inp->in6p_moptions != NULL) {
1513 free(imfp, M_IN6MFILTER);
1514 free(immp, M_IP6MOPTS);
1515 free(imo, M_IP6MOPTS);
1516 return (inp->in6p_moptions);
1517 }
1518 inp->in6p_moptions = imo;
1519 return (imo);
1520}
1521
1522/*
1523 * Discard the IPv6 multicast options (and source filters).
1524 *
1525 * SMPng: NOTE: assumes INP write lock is held.
1526 */
1527void
1528ip6_freemoptions(struct ip6_moptions *imo)
1529{
1530 struct in6_mfilter *imf;
1531 size_t idx, nmships;
1532
1533 KASSERT(imo != NULL, ("%s: ip6_moptions is NULL", __func__));
1534
1535 nmships = imo->im6o_num_memberships;
1536 for (idx = 0; idx < nmships; ++idx) {
1537 imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1538 if (imf)
1539 im6f_leave(imf);
1540 /* XXX this will thrash the lock(s) */
1541 (void)in6_mc_leave(imo->im6o_membership[idx], imf);
1542 if (imf)
1543 im6f_purge(imf);
1544 }
1545
1546 if (imo->im6o_mfilters)
1547 free(imo->im6o_mfilters, M_IN6MFILTER);
1548 free(imo->im6o_membership, M_IP6MOPTS);
1549 free(imo, M_IP6MOPTS);
1550}
1551
1552/*
1553 * Atomically get source filters on a socket for an IPv6 multicast group.
1554 * Called with INP lock held; returns with lock released.
1555 */
1556static int
1557in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1558{
1559 struct __msfilterreq msfr;
1560 sockunion_t *gsa;
1561 struct ifnet *ifp;
1562 struct ip6_moptions *imo;
1563 struct in6_mfilter *imf;
1564 struct ip6_msource *ims;
1565 struct in6_msource *lims;
1566 struct sockaddr_in6 *psin;
1567 struct sockaddr_storage *ptss;
1568 struct sockaddr_storage *tss;
1569 int error;
1570 size_t idx, nsrcs, ncsrcs;
1571
1572 INP_WLOCK_ASSERT(inp);
1573
1574 imo = inp->in6p_moptions;
1575 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1576
1577 INP_WUNLOCK(inp);
1578
1579 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1580 sizeof(struct __msfilterreq));
1581 if (error)
1582 return (error);
1583
1584 if (msfr.msfr_group.ss_family != AF_INET6 ||
1585 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1586 return (EINVAL);
1587
1588 gsa = (sockunion_t *)&msfr.msfr_group;
1589 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1590 return (EINVAL);
1591
1592 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1593 return (EADDRNOTAVAIL);
1594 ifp = ifnet_byindex(msfr.msfr_ifindex);
1595 if (ifp == NULL)
1596 return (EADDRNOTAVAIL);
1597 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1598
1599 INP_WLOCK(inp);
1600
1601 /*
1602 * Lookup group on the socket.
1603 */
1604 idx = im6o_match_group(imo, ifp, &gsa->sa);
1605 if (idx == -1 || imo->im6o_mfilters == NULL) {
1606 INP_WUNLOCK(inp);
1607 return (EADDRNOTAVAIL);
1608 }
1609 imf = &imo->im6o_mfilters[idx];
1610
1611 /*
1612 * Ignore memberships which are in limbo.
1613 */
1614 if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1615 INP_WUNLOCK(inp);
1616 return (EAGAIN);
1617 }
1618 msfr.msfr_fmode = imf->im6f_st[1];
1619
1620 /*
1621 * If the user specified a buffer, copy out the source filter
1622 * entries to userland gracefully.
1623 * We only copy out the number of entries which userland
1624 * has asked for, but we always tell userland how big the
1625 * buffer really needs to be.
1626 */
1627 tss = NULL;
1628 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1629 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1630 M_TEMP, M_NOWAIT | M_ZERO);
1631 if (tss == NULL) {
1632 INP_WUNLOCK(inp);
1633 return (ENOBUFS);
1634 }
1635 }
1636
1637 /*
1638 * Count number of sources in-mode at t0.
1639 * If buffer space exists and remains, copy out source entries.
1640 */
1641 nsrcs = msfr.msfr_nsrcs;
1642 ncsrcs = 0;
1643 ptss = tss;
1644 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1645 lims = (struct in6_msource *)ims;
1646 if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1647 lims->im6sl_st[0] != imf->im6f_st[0])
1648 continue;
1649 ++ncsrcs;
1650 if (tss != NULL && nsrcs > 0) {
1651 psin = (struct sockaddr_in6 *)ptss;
1652 psin->sin6_family = AF_INET6;
1653 psin->sin6_len = sizeof(struct sockaddr_in6);
1654 psin->sin6_addr = lims->im6s_addr;
1655 psin->sin6_port = 0;
1656 --nsrcs;
1657 ++ptss;
1658 }
1659 }
1660
1661 INP_WUNLOCK(inp);
1662
1663 if (tss != NULL) {
1664 error = copyout(tss, msfr.msfr_srcs,
1665 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1666 free(tss, M_TEMP);
1667 if (error)
1668 return (error);
1669 }
1670
1671 msfr.msfr_nsrcs = ncsrcs;
1672 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1673
1674 return (error);
1675}
1676
1677/*
1678 * Return the IP multicast options in response to user getsockopt().
1679 */
1680int
1681ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1682{
1683 struct ip6_moptions *im6o;
1684 int error;
1685 u_int optval;
1686
1687 INP_WLOCK(inp);
1688 im6o = inp->in6p_moptions;
1689 /*
1690 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1691 * or is a divert socket, reject it.
1692 */
1693 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1694 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1695 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1696 INP_WUNLOCK(inp);
1697 return (EOPNOTSUPP);
1698 }
1699
1700 error = 0;
1701 switch (sopt->sopt_name) {
1702 case IPV6_MULTICAST_IF:
1703 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1704 optval = 0;
1705 } else {
1706 optval = im6o->im6o_multicast_ifp->if_index;
1707 }
1708 INP_WUNLOCK(inp);
1709 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1710 break;
1711
1712 case IPV6_MULTICAST_HOPS:
1713 if (im6o == NULL)
1714 optval = V_ip6_defmcasthlim;
1715 else
1716 optval = im6o->im6o_multicast_loop;
1717 INP_WUNLOCK(inp);
1718 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1719 break;
1720
1721 case IPV6_MULTICAST_LOOP:
1722 if (im6o == NULL)
1723 optval = in6_mcast_loop; /* XXX VIMAGE */
1724 else
1725 optval = im6o->im6o_multicast_loop;
1726 INP_WUNLOCK(inp);
1727 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1728 break;
1729
1730 case IPV6_MSFILTER:
1731 if (im6o == NULL) {
1732 error = EADDRNOTAVAIL;
1733 INP_WUNLOCK(inp);
1734 } else {
1735 error = in6p_get_source_filters(inp, sopt);
1736 }
1737 break;
1738
1739 default:
1740 INP_WUNLOCK(inp);
1741 error = ENOPROTOOPT;
1742 break;
1743 }
1744
1745 INP_UNLOCK_ASSERT(inp);
1746
1747 return (error);
1748}
1749
1750/*
1751 * Look up the ifnet to use for a multicast group membership,
1752 * given the address of an IPv6 group.
1753 *
1754 * This routine exists to support legacy IPv6 multicast applications.
1755 *
1756 * If inp is non-NULL, use this socket's current FIB number for any
1757 * required FIB lookup. Look up the group address in the unicast FIB,
1758 * and use its ifp; usually, this points to the default next-hop.
1759 * If the FIB lookup fails, return NULL.
1760 *
1761 * FUTURE: Support multiple forwarding tables for IPv6.
1762 *
1763 * Returns NULL if no ifp could be found.
1764 */
1765static struct ifnet *
1766in6p_lookup_mcast_ifp(const struct inpcb *in6p __unused,
1767 const struct sockaddr_in6 *gsin6)
1768{
1769 struct route_in6 ro6;
1770 struct ifnet *ifp;
1771
1772 KASSERT(in6p->inp_vflag & INP_IPV6,
1773 ("%s: not INP_IPV6 inpcb", __func__));
1774 KASSERT(gsin6->sin6_family == AF_INET6,
1775 ("%s: not AF_INET6 group", __func__));
1776 KASSERT(IN6_IS_ADDR_MULTICAST(&gsin6->sin6_addr),
1777 ("%s: not multicast", __func__));
1778
1779 ifp = NULL;
1780 memset(&ro6, 0, sizeof(struct route_in6));
1781 memcpy(&ro6.ro_dst, gsin6, sizeof(struct sockaddr_in6));
1782#ifdef notyet
1783 rtalloc_ign_fib(&ro6, 0, inp ? inp->inp_inc.inc_fibnum : 0);
1784#else
1785 rtalloc_ign((struct route *)&ro6, 0);
1786#endif
1787 if (ro6.ro_rt != NULL) {
1788 ifp = ro6.ro_rt->rt_ifp;
1789 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
1790 RTFREE(ro6.ro_rt);
1791 }
1792
1793 return (ifp);
1794}
1795
1796/*
1797 * Join an IPv6 multicast group, possibly with a source.
1798 *
1799 * FIXME: The KAME use of the unspecified address (::)
1800 * to join *all* multicast groups is currently unsupported.
1801 */
1802static int
1803in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1804{
1805 struct group_source_req gsr;
1806 sockunion_t *gsa, *ssa;
1807 struct ifnet *ifp;
1808 struct in6_mfilter *imf;
1809 struct ip6_moptions *imo;
1810 struct in6_multi *inm;
1811 struct in6_msource *lims;
1812 size_t idx;
1813 int error, is_new;
1814
1815 ifp = NULL;
1816 imf = NULL;
1817 error = 0;
1818 is_new = 0;
1819
1820 memset(&gsr, 0, sizeof(struct group_source_req));
1821 gsa = (sockunion_t *)&gsr.gsr_group;
1822 gsa->ss.ss_family = AF_UNSPEC;
1823 ssa = (sockunion_t *)&gsr.gsr_source;
1824 ssa->ss.ss_family = AF_UNSPEC;
1825
1826 /*
1827 * Chew everything into struct group_source_req.
1828 * Overwrite the port field if present, as the sockaddr
1829 * being copied in may be matched with a binary comparison.
1830 * Ignore passed-in scope ID.
1831 */
1832 switch (sopt->sopt_name) {
1833 case IPV6_JOIN_GROUP: {
1834 struct ipv6_mreq mreq;
1835
1836 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1837 sizeof(struct ipv6_mreq));
1838 if (error)
1839 return (error);
1840
1841 gsa->sin6.sin6_family = AF_INET6;
1842 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1843 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1844
1845 if (mreq.ipv6mr_interface == 0) {
1846 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1847 } else {
1848 if (mreq.ipv6mr_interface < 0 ||
1849 V_if_index < mreq.ipv6mr_interface)
1850 return (EADDRNOTAVAIL);
1851 ifp = ifnet_byindex(mreq.ipv6mr_interface);
1852 }
1853 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1854 __func__, mreq.ipv6mr_interface, ifp);
1855 } break;
1856
1857 case MCAST_JOIN_GROUP:
1858 case MCAST_JOIN_SOURCE_GROUP:
1859 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1860 error = sooptcopyin(sopt, &gsr,
1861 sizeof(struct group_req),
1862 sizeof(struct group_req));
1863 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1864 error = sooptcopyin(sopt, &gsr,
1865 sizeof(struct group_source_req),
1866 sizeof(struct group_source_req));
1867 }
1868 if (error)
1869 return (error);
1870
1871 if (gsa->sin6.sin6_family != AF_INET6 ||
1872 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1873 return (EINVAL);
1874
1875 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1876 if (ssa->sin6.sin6_family != AF_INET6 ||
1877 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1878 return (EINVAL);
1879 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1880 return (EINVAL);
1881 /*
1882 * TODO: Validate embedded scope ID in source
1883 * list entry against passed-in ifp, if and only
1884 * if source list filter entry is iface or node local.
1885 */
1886 in6_clearscope(&ssa->sin6.sin6_addr);
1887 ssa->sin6.sin6_port = 0;
1888 ssa->sin6.sin6_scope_id = 0;
1889 }
1890
1891 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1892 return (EADDRNOTAVAIL);
1893 ifp = ifnet_byindex(gsr.gsr_interface);
1894 break;
1895
1896 default:
1897 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1898 __func__, sopt->sopt_name);
1899 return (EOPNOTSUPP);
1900 break;
1901 }
1902
1903 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1904 return (EINVAL);
1905
1906 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1907 return (EADDRNOTAVAIL);
1908
1909 gsa->sin6.sin6_port = 0;
1910 gsa->sin6.sin6_scope_id = 0;
1911
1912 /*
1913 * Always set the scope zone ID on memberships created from userland.
1914 * Use the passed-in ifp to do this.
1915 * XXX The in6_setscope() return value is meaningless.
1916 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1917 */
1918 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1919
1920 /*
1921 * MCAST_JOIN_SOURCE on an exclusive membership is an error.
1922 * On an existing inclusive membership, it just adds the
1923 * source to the filter list.
1924 */
1925 imo = in6p_findmoptions(inp);
1926 idx = im6o_match_group(imo, ifp, &gsa->sa);
1927 if (idx == -1) {
1928 is_new = 1;
1929 } else {
1930 inm = imo->im6o_membership[idx];
1931 imf = &imo->im6o_mfilters[idx];
1932 if (ssa->ss.ss_family != AF_UNSPEC &&
1933 imf->im6f_st[1] != MCAST_INCLUDE) {
1934 error = EINVAL;
1935 goto out_in6p_locked;
1936 }
1937 lims = im6o_match_source(imo, idx, &ssa->sa);
1938 if (lims != NULL) {
1939 error = EADDRNOTAVAIL;
1940 goto out_in6p_locked;
1941 }
1942 }
1943
1944 /*
1945 * Begin state merge transaction at socket layer.
1946 */
1947 INP_WLOCK_ASSERT(inp);
1948
1949 if (is_new) {
1950 if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
1951 error = im6o_grow(imo);
1952 if (error)
1953 goto out_in6p_locked;
1954 }
1955 /*
1956 * Allocate the new slot upfront so we can deal with
1957 * grafting the new source filter in same code path
1958 * as for join-source on existing membership.
1959 */
1960 idx = imo->im6o_num_memberships;
1961 imo->im6o_membership[idx] = NULL;
1962 imo->im6o_num_memberships++;
1963 KASSERT(imo->im6o_mfilters != NULL,
1964 ("%s: im6f_mfilters vector was not allocated", __func__));
1965 imf = &imo->im6o_mfilters[idx];
1966 KASSERT(RB_EMPTY(&imf->im6f_sources),
1967 ("%s: im6f_sources not empty", __func__));
1968 }
1969
1970 /*
1971 * Graft new source into filter list for this inpcb's
1972 * membership of the group. The in6_multi may not have
1973 * been allocated yet if this is a new membership.
1974 */
1975 if (ssa->ss.ss_family != AF_UNSPEC) {
1976 /* Membership starts in IN mode */
1977 if (is_new) {
1978 CTR1(KTR_MLD, "%s: new join w/source", __func__);
1979 im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
1980 } else {
1981 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1982 }
1983 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
1984 if (lims == NULL) {
1985 CTR1(KTR_MLD, "%s: merge imf state failed",
1986 __func__);
1987 error = ENOMEM;
1988 goto out_im6o_free;
1989 }
1990 }
1991
1992 /*
1993 * Begin state merge transaction at MLD layer.
1994 */
1995 IN6_MULTI_LOCK();
1996
1997 if (is_new) {
1998 error = in6_mc_join_locked(ifp, &gsa->sin6.sin6_addr, imf,
1999 &inm, 0);
2000 if (error)
2001 goto out_im6o_free;
2002 imo->im6o_membership[idx] = inm;
2003 } else {
2004 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2005 error = in6m_merge(inm, imf);
2006 if (error) {
2007 CTR1(KTR_MLD, "%s: failed to merge inm state",
2008 __func__);
2009 goto out_im6f_rollback;
2010 }
2011 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2012 error = mld_change_state(inm, 0);
2013 if (error) {
2014 CTR1(KTR_MLD, "%s: failed mld downcall",
2015 __func__);
2016 goto out_im6f_rollback;
2017 }
2018 }
2019
2020 IN6_MULTI_UNLOCK();
2021
2022out_im6f_rollback:
2023 INP_WLOCK_ASSERT(inp);
2024 if (error) {
2025 im6f_rollback(imf);
2026 if (is_new)
2027 im6f_purge(imf);
2028 else
2029 im6f_reap(imf);
2030 } else {
2031 im6f_commit(imf);
2032 }
2033
2034out_im6o_free:
2035 if (error && is_new) {
2036 imo->im6o_membership[idx] = NULL;
2037 --imo->im6o_num_memberships;
2038 }
2039
2040out_in6p_locked:
2041 INP_WUNLOCK(inp);
2042 return (error);
2043}
2044
2045/*
2046 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2047 */
2048static int
2049in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2050{
2051 struct ipv6_mreq mreq;
2052 struct group_source_req gsr;
2053 sockunion_t *gsa, *ssa;
2054 struct ifnet *ifp;
2055 struct in6_mfilter *imf;
2056 struct ip6_moptions *imo;
2057 struct in6_msource *ims;
2058 struct in6_multi *inm;
2059 uint32_t ifindex;
2060 size_t idx;
2061 int error, is_final;
2062#ifdef KTR
2063 char ip6tbuf[INET6_ADDRSTRLEN];
2064#endif
2065
2066 ifp = NULL;
2067 ifindex = 0;
2068 error = 0;
2069 is_final = 1;
2070
2071 memset(&gsr, 0, sizeof(struct group_source_req));
2072 gsa = (sockunion_t *)&gsr.gsr_group;
2073 gsa->ss.ss_family = AF_UNSPEC;
2074 ssa = (sockunion_t *)&gsr.gsr_source;
2075 ssa->ss.ss_family = AF_UNSPEC;
2076
2077 /*
2078 * Chew everything passed in up into a struct group_source_req
2079 * as that is easier to process.
2080 * Note: Any embedded scope ID in the multicast group passed
2081 * in by userland is ignored, the interface index is the recommended
2082 * mechanism to specify an interface; see below.
2083 */
2084 switch (sopt->sopt_name) {
2085 case IPV6_LEAVE_GROUP:
2086 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2087 sizeof(struct ipv6_mreq));
2088 if (error)
2089 return (error);
2090 gsa->sin6.sin6_family = AF_INET6;
2091 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2092 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2093 gsa->sin6.sin6_port = 0;
2094 gsa->sin6.sin6_scope_id = 0;
2095 ifindex = mreq.ipv6mr_interface;
2096 break;
2097
2098 case MCAST_LEAVE_GROUP:
2099 case MCAST_LEAVE_SOURCE_GROUP:
2100 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2101 error = sooptcopyin(sopt, &gsr,
2102 sizeof(struct group_req),
2103 sizeof(struct group_req));
2104 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2105 error = sooptcopyin(sopt, &gsr,
2106 sizeof(struct group_source_req),
2107 sizeof(struct group_source_req));
2108 }
2109 if (error)
2110 return (error);
2111
2112 if (gsa->sin6.sin6_family != AF_INET6 ||
2113 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2114 return (EINVAL);
2115 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2116 if (ssa->sin6.sin6_family != AF_INET6 ||
2117 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2118 return (EINVAL);
2119 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2120 return (EINVAL);
2121 /*
2122 * TODO: Validate embedded scope ID in source
2123 * list entry against passed-in ifp, if and only
2124 * if source list filter entry is iface or node local.
2125 */
2126 in6_clearscope(&ssa->sin6.sin6_addr);
2127 }
2128 gsa->sin6.sin6_port = 0;
2129 gsa->sin6.sin6_scope_id = 0;
2130 ifindex = gsr.gsr_interface;
2131 break;
2132
2133 default:
2134 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2135 __func__, sopt->sopt_name);
2136 return (EOPNOTSUPP);
2137 break;
2138 }
2139
2140 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2141 return (EINVAL);
2142
2143 /*
2144 * Validate interface index if provided. If no interface index
2145 * was provided separately, attempt to look the membership up
2146 * from the default scope as a last resort to disambiguate
2147 * the membership we are being asked to leave.
2148 * XXX SCOPE6 lock potentially taken here.
2149 */
2150 if (ifindex != 0) {
2151 if (ifindex < 0 || V_if_index < ifindex)
2152 return (EADDRNOTAVAIL);
2153 ifp = ifnet_byindex(ifindex);
2154 if (ifp == NULL)
2155 return (EADDRNOTAVAIL);
2156 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2157 } else {
2158 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2159 if (error)
2160 return (EADDRNOTAVAIL);
2161 /*
2162 * Some badly behaved applications don't pass an ifindex
2163 * or a scope ID, which is an API violation. In this case,
2164 * perform a lookup as per a v6 join.
2165 *
2166 * XXX For now, stomp on zone ID for the corner case.
2167 * This is not the 'KAME way', but we need to see the ifp
2168 * directly until such time as this implementation is
2169 * refactored, assuming the scope IDs are the way to go.
2170 */
2171 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2172 if (ifindex == 0) {
2173 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2174 "ifp for group %s.", __func__,
2175 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2176 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2177 } else {
2178 ifp = ifnet_byindex(ifindex);
2179 }
2180 if (ifp == NULL)
2181 return (EADDRNOTAVAIL);
2182 }
2183
2184 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2185 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2186
2187 /*
2188 * Find the membership in the membership array.
2189 */
2190 imo = in6p_findmoptions(inp);
2191 idx = im6o_match_group(imo, ifp, &gsa->sa);
2192 if (idx == -1) {
2193 error = EADDRNOTAVAIL;
2194 goto out_in6p_locked;
2195 }
2196 inm = imo->im6o_membership[idx];
2197 imf = &imo->im6o_mfilters[idx];
2198
2199 if (ssa->ss.ss_family != AF_UNSPEC)
2200 is_final = 0;
2201
2202 /*
2203 * Begin state merge transaction at socket layer.
2204 */
2205 INP_WLOCK_ASSERT(inp);
2206
2207 /*
2208 * If we were instructed only to leave a given source, do so.
2209 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2210 */
2211 if (is_final) {
2212 im6f_leave(imf);
2213 } else {
2214 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2215 error = EADDRNOTAVAIL;
2216 goto out_in6p_locked;
2217 }
2218 ims = im6o_match_source(imo, idx, &ssa->sa);
2219 if (ims == NULL) {
2220 CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2221 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2222 "not ");
2223 error = EADDRNOTAVAIL;
2224 goto out_in6p_locked;
2225 }
2226 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2227 error = im6f_prune(imf, &ssa->sin6);
2228 if (error) {
2229 CTR1(KTR_MLD, "%s: merge imf state failed",
2230 __func__);
2231 goto out_in6p_locked;
2232 }
2233 }
2234
2235 /*
2236 * Begin state merge transaction at MLD layer.
2237 */
2238 IN6_MULTI_LOCK();
2239
2240 if (is_final) {
2241 /*
2242 * Give up the multicast address record to which
2243 * the membership points.
2244 */
2245 (void)in6_mc_leave_locked(inm, imf);
2246 } else {
2247 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2248 error = in6m_merge(inm, imf);
2249 if (error) {
2250 CTR1(KTR_MLD, "%s: failed to merge inm state",
2251 __func__);
2252 goto out_im6f_rollback;
2253 }
2254
2255 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2256 error = mld_change_state(inm, 0);
2257 if (error) {
2258 CTR1(KTR_MLD, "%s: failed mld downcall",
2259 __func__);
2260 }
2261 }
2262
2263 IN6_MULTI_UNLOCK();
2264
2265out_im6f_rollback:
2266 if (error)
2267 im6f_rollback(imf);
2268 else
2269 im6f_commit(imf);
2270
2271 im6f_reap(imf);
2272
2273 if (is_final) {
2274 /* Remove the gap in the membership array. */
2275 for (++idx; idx < imo->im6o_num_memberships; ++idx)
2276 imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2277 imo->im6o_num_memberships--;
2278 }
2279
2280out_in6p_locked:
2281 INP_WUNLOCK(inp);
2282 return (error);
2283}
2284
2285/*
2286 * Select the interface for transmitting IPv6 multicast datagrams.
2287 *
2288 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2289 * may be passed to this socket option. An address of in6addr_any or an
2290 * interface index of 0 is used to remove a previous selection.
2291 * When no interface is selected, one is chosen for every send.
2292 */
2293static int
2294in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2295{
2296 struct ifnet *ifp;
2297 struct ip6_moptions *imo;
2298 u_int ifindex;
2299 int error;
2300
2301 if (sopt->sopt_valsize != sizeof(u_int))
2302 return (EINVAL);
2303
2304 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2305 if (error)
2306 return (error);
2307 if (ifindex < 0 || V_if_index < ifindex)
2308 return (EINVAL);
2309
2310 ifp = ifnet_byindex(ifindex);
2311 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2312 return (EADDRNOTAVAIL);
2313
2314 imo = in6p_findmoptions(inp);
2315 imo->im6o_multicast_ifp = ifp;
2316 INP_WUNLOCK(inp);
2317
2318 return (0);
2319}
2320
2321/*
2322 * Atomically set source filters on a socket for an IPv6 multicast group.
2323 *
2324 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2325 */
2326static int
2327in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2328{
2329 struct __msfilterreq msfr;
2330 sockunion_t *gsa;
2331 struct ifnet *ifp;
2332 struct in6_mfilter *imf;
2333 struct ip6_moptions *imo;
2334 struct in6_multi *inm;
2335 size_t idx;
2336 int error;
2337
2338 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2339 sizeof(struct __msfilterreq));
2340 if (error)
2341 return (error);
2342
2343 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc ||
2344 (msfr.msfr_fmode != MCAST_EXCLUDE &&
2345 msfr.msfr_fmode != MCAST_INCLUDE))
2346 return (EINVAL);
2347
2348 if (msfr.msfr_group.ss_family != AF_INET6 ||
2349 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2350 return (EINVAL);
2351
2352 gsa = (sockunion_t *)&msfr.msfr_group;
2353 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2354 return (EINVAL);
2355
2356 gsa->sin6.sin6_port = 0; /* ignore port */
2357
2358 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2359 return (EADDRNOTAVAIL);
2360 ifp = ifnet_byindex(msfr.msfr_ifindex);
2361 if (ifp == NULL)
2362 return (EADDRNOTAVAIL);
2363 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2364
2365 /*
2366 * Take the INP write lock.
2367 * Check if this socket is a member of this group.
2368 */
2369 imo = in6p_findmoptions(inp);
2370 idx = im6o_match_group(imo, ifp, &gsa->sa);
2371 if (idx == -1 || imo->im6o_mfilters == NULL) {
2372 error = EADDRNOTAVAIL;
2373 goto out_in6p_locked;
2374 }
2375 inm = imo->im6o_membership[idx];
2376 imf = &imo->im6o_mfilters[idx];
2377
2378 /*
2379 * Begin state merge transaction at socket layer.
2380 */
2381 INP_WLOCK_ASSERT(inp);
2382
2383 imf->im6f_st[1] = msfr.msfr_fmode;
2384
2385 /*
2386 * Apply any new source filters, if present.
2387 * Make a copy of the user-space source vector so
2388 * that we may copy them with a single copyin. This
2389 * allows us to deal with page faults up-front.
2390 */
2391 if (msfr.msfr_nsrcs > 0) {
2392 struct in6_msource *lims;
2393 struct sockaddr_in6 *psin;
2394 struct sockaddr_storage *kss, *pkss;
2395 int i;
2396
2397 INP_WUNLOCK(inp);
2398
2399 CTR2(KTR_MLD, "%s: loading %lu source list entries",
2400 __func__, (unsigned long)msfr.msfr_nsrcs);
2401 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2402 M_TEMP, M_WAITOK);
2403 error = copyin(msfr.msfr_srcs, kss,
2404 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2405 if (error) {
2406 free(kss, M_TEMP);
2407 return (error);
2408 }
2409
2410 INP_WLOCK(inp);
2411
2412 /*
2413 * Mark all source filters as UNDEFINED at t1.
2414 * Restore new group filter mode, as im6f_leave()
2415 * will set it to INCLUDE.
2416 */
2417 im6f_leave(imf);
2418 imf->im6f_st[1] = msfr.msfr_fmode;
2419
2420 /*
2421 * Update socket layer filters at t1, lazy-allocating
2422 * new entries. This saves a bunch of memory at the
2423 * cost of one RB_FIND() per source entry; duplicate
2424 * entries in the msfr_nsrcs vector are ignored.
2425 * If we encounter an error, rollback transaction.
2426 *
2427 * XXX This too could be replaced with a set-symmetric
2428 * difference like loop to avoid walking from root
2429 * every time, as the key space is common.
2430 */
2431 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2432 psin = (struct sockaddr_in6 *)pkss;
2433 if (psin->sin6_family != AF_INET6) {
2434 error = EAFNOSUPPORT;
2435 break;
2436 }
2437 if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2438 error = EINVAL;
2439 break;
2440 }
2441 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2442 error = EINVAL;
2443 break;
2444 }
2445 /*
2446 * TODO: Validate embedded scope ID in source
2447 * list entry against passed-in ifp, if and only
2448 * if source list filter entry is iface or node local.
2449 */
2450 in6_clearscope(&psin->sin6_addr);
2451 error = im6f_get_source(imf, psin, &lims);
2452 if (error)
2453 break;
2454 lims->im6sl_st[1] = imf->im6f_st[1];
2455 }
2456 free(kss, M_TEMP);
2457 }
2458
2459 if (error)
2460 goto out_im6f_rollback;
2461
2462 INP_WLOCK_ASSERT(inp);
2463 IN6_MULTI_LOCK();
2464
2465 /*
2466 * Begin state merge transaction at MLD layer.
2467 */
2468 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2469 error = in6m_merge(inm, imf);
2470 if (error) {
2471 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2472 goto out_im6f_rollback;
2473 }
2474
2475 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2476 error = mld_change_state(inm, 0);
2477 if (error)
2478 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2479
2480 IN6_MULTI_UNLOCK();
2481
2482out_im6f_rollback:
2483 if (error)
2484 im6f_rollback(imf);
2485 else
2486 im6f_commit(imf);
2487
2488 im6f_reap(imf);
2489
2490out_in6p_locked:
2491 INP_WUNLOCK(inp);
2492 return (error);
2493}
2494
2495/*
2496 * Set the IP multicast options in response to user setsockopt().
2497 *
2498 * Many of the socket options handled in this function duplicate the
2499 * functionality of socket options in the regular unicast API. However,
2500 * it is not possible to merge the duplicate code, because the idempotence
2501 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2502 * the effects of these options must be treated as separate and distinct.
2503 *
2504 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2505 */
2506int
2507ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2508{
2509 struct ip6_moptions *im6o;
2510 int error;
2511
2512 error = 0;
2513
2514 /*
2515 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2516 * or is a divert socket, reject it.
2517 */
2518 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2519 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2520 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2521 return (EOPNOTSUPP);
2522
2523 switch (sopt->sopt_name) {
2524 case IPV6_MULTICAST_IF:
2525 error = in6p_set_multicast_if(inp, sopt);
2526 break;
2527
2528 case IPV6_MULTICAST_HOPS: {
2529 int hlim;
2530
2531 if (sopt->sopt_valsize != sizeof(int)) {
2532 error = EINVAL;
2533 break;
2534 }
2535 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2536 if (error)
2537 break;
2538 if (hlim < -1 || hlim > 255) {
2539 error = EINVAL;
2540 break;
2541 } else if (hlim == -1) {
2542 hlim = V_ip6_defmcasthlim;
2543 }
2544 im6o = in6p_findmoptions(inp);
2545 im6o->im6o_multicast_hlim = hlim;
2546 INP_WUNLOCK(inp);
2547 break;
2548 }
2549
2550 case IPV6_MULTICAST_LOOP: {
2551 u_int loop;
2552
2553 /*
2554 * Set the loopback flag for outgoing multicast packets.
2555 * Must be zero or one.
2556 */
2557 if (sopt->sopt_valsize != sizeof(u_int)) {
2558 error = EINVAL;
2559 break;
2560 }
2561 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2562 if (error)
2563 break;
2564 if (loop > 1) {
2565 error = EINVAL;
2566 break;
2567 }
2568 im6o = in6p_findmoptions(inp);
2569 im6o->im6o_multicast_loop = loop;
2570 INP_WUNLOCK(inp);
2571 break;
2572 }
2573
2574 case IPV6_JOIN_GROUP:
2575 case MCAST_JOIN_GROUP:
2576 case MCAST_JOIN_SOURCE_GROUP:
2577 error = in6p_join_group(inp, sopt);
2578 break;
2579
2580 case IPV6_LEAVE_GROUP:
2581 case MCAST_LEAVE_GROUP:
2582 case MCAST_LEAVE_SOURCE_GROUP:
2583 error = in6p_leave_group(inp, sopt);
2584 break;
2585
2586 case MCAST_BLOCK_SOURCE:
2587 case MCAST_UNBLOCK_SOURCE:
2588 error = in6p_block_unblock_source(inp, sopt);
2589 break;
2590
2591 case IPV6_MSFILTER:
2592 error = in6p_set_source_filters(inp, sopt);
2593 break;
2594
2595 default:
2596 error = EOPNOTSUPP;
2597 break;
2598 }
2599
2600 INP_UNLOCK_ASSERT(inp);
2601
2602 return (error);
2603}
2604
2605/*
2606 * Expose MLD's multicast filter mode and source list(s) to userland,
2607 * keyed by (ifindex, group).
2608 * The filter mode is written out as a uint32_t, followed by
2609 * 0..n of struct in6_addr.
2610 * For use by ifmcstat(8).
2611 * SMPng: NOTE: unlocked read of ifindex space.
2612 */
2613static int
2614sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2615{
2616 struct in6_addr mcaddr;
2617 struct in6_addr src;
2618 struct ifnet *ifp;
2619 struct ifmultiaddr *ifma;
2620 struct in6_multi *inm;
2621 struct ip6_msource *ims;
2622 int *name;
2623 int retval;
2624 u_int namelen;
2625 uint32_t fmode, ifindex;
2626#ifdef KTR
2627 char ip6tbuf[INET6_ADDRSTRLEN];
2628#endif
2629
2630 name = (int *)arg1;
2631 namelen = arg2;
2632
2633 if (req->newptr != NULL)
2634 return (EPERM);
2635
2636 /* int: ifindex + 4 * 32 bits of IPv6 address */
2637 if (namelen != 5)
2638 return (EINVAL);
2639
2640 ifindex = name[0];
2641 if (ifindex <= 0 || ifindex > V_if_index) {
2642 CTR2(KTR_MLD, "%s: ifindex %u out of range",
2643 __func__, ifindex);
2644 return (ENOENT);
2645 }
2646
2647 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2648 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2649 CTR2(KTR_MLD, "%s: group %s is not multicast",
2650 __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2651 return (EINVAL);
2652 }
2653
2654 ifp = ifnet_byindex(ifindex);
2655 if (ifp == NULL) {
2656 CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2657 __func__, ifindex);
2658 return (ENOENT);
2659 }
2660 /*
2661 * Internal MLD lookups require that scope/zone ID is set.
2662 */
2663 (void)in6_setscope(&mcaddr, ifp, NULL);
2664
2665 retval = sysctl_wire_old_buffer(req,
2666 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2667 if (retval)
2668 return (retval);
2669
2670 IN6_MULTI_LOCK();
2671
2672 IF_ADDR_LOCK(ifp);
2673 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2674 if (ifma->ifma_addr->sa_family != AF_INET6 ||
2675 ifma->ifma_protospec == NULL)
2676 continue;
2677 inm = (struct in6_multi *)ifma->ifma_protospec;
2678 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2679 continue;
2680 fmode = inm->in6m_st[1].iss_fmode;
2681 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2682 if (retval != 0)
2683 break;
2684 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2685 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2686 /*
2687 * Only copy-out sources which are in-mode.
2688 */
2689 if (fmode != im6s_get_mode(inm, ims, 1)) {
2690 CTR1(KTR_MLD, "%s: skip non-in-mode",
2691 __func__);
2692 continue;
2693 }
2694 src = ims->im6s_addr;
2695 retval = SYSCTL_OUT(req, &src,
2696 sizeof(struct in6_addr));
2697 if (retval != 0)
2698 break;
2699 }
2700 }
2701 IF_ADDR_UNLOCK(ifp);
2702
2703 IN6_MULTI_UNLOCK();
2704
2705 return (retval);
2706}
2707
2708#ifdef KTR
2709
2710static const char *in6m_modestrs[] = { "un", "in", "ex" };
2711
2712static const char *
2713in6m_mode_str(const int mode)
2714{
2715
2716 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2717 return (in6m_modestrs[mode]);
2718 return ("??");
2719}
2720
2721static const char *in6m_statestrs[] = {
2722 "not-member",
2723 "silent",
2724 "idle",
2725 "lazy",
2726 "sleeping",
2727 "awakening",
2728 "query-pending",
2729 "sg-query-pending",
2730 "leaving"
2731};
2732
2733static const char *
2734in6m_state_str(const int state)
2735{
2736
2737 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2738 return (in6m_statestrs[state]);
2739 return ("??");
2740}
2741
2742/*
2743 * Dump an in6_multi structure to the console.
2744 */
2745void
2746in6m_print(const struct in6_multi *inm)
2747{
2748 int t;
2749 char ip6tbuf[INET6_ADDRSTRLEN];
2750
2751 if ((ktr_mask & KTR_MLD) == 0)
2752 return;
2753
2754 printf("%s: --- begin in6m %p ---\n", __func__, inm);
2755 printf("addr %s ifp %p(%s) ifma %p\n",
2756 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2757 inm->in6m_ifp,
2758 inm->in6m_ifp->if_xname,
2759 inm->in6m_ifma);
2760 printf("timer %u state %s refcount %u scq.len %u\n",
2761 inm->in6m_timer,
2762 in6m_state_str(inm->in6m_state),
2763 inm->in6m_refcount,
2764 inm->in6m_scq.ifq_len);
2765 printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2766 inm->in6m_mli,
2767 inm->in6m_nsrc,
2768 inm->in6m_sctimer,
2769 inm->in6m_scrv);
2770 for (t = 0; t < 2; t++) {
2771 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2772 in6m_mode_str(inm->in6m_st[t].iss_fmode),
2773 inm->in6m_st[t].iss_asm,
2774 inm->in6m_st[t].iss_ex,
2775 inm->in6m_st[t].iss_in,
2776 inm->in6m_st[t].iss_rec);
2777 }
2778 printf("%s: --- end in6m %p ---\n", __func__, inm);
2779}
2780
2781#else /* !KTR */
2782
2783void
2784in6m_print(const struct in6_multi *inm)
2785{
2786
2787}
2788
2789#endif /* KTR */
|