2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2009 Bruce Simpson.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. The name of the author may not be used to endorse or promote
16 * products derived from this software without specific prior written
19 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * IPv6 multicast socket, group, and socket option processing module.
34 * Normative references: RFC 2292, RFC 3492, RFC 3542, RFC 3678, RFC 3810.
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_inet6.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/kernel.h>
46 #include <sys/malloc.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
53 #include <sys/taskqueue.h>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.h>
60 #include <net/route/nhop.h>
63 #include <netinet/in.h>
64 #include <netinet/udp.h>
65 #include <netinet/in_var.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/udp_var.h>
68 #include <netinet6/in6_fib.h>
69 #include <netinet6/in6_var.h>
70 #include <netinet/ip6.h>
71 #include <netinet/icmp6.h>
72 #include <netinet6/ip6_var.h>
73 #include <netinet/in_pcb.h>
74 #include <netinet/tcp_var.h>
75 #include <netinet6/nd6.h>
76 #include <netinet6/mld6_var.h>
77 #include <netinet6/scope6_var.h>
80 #define KTR_MLD KTR_INET6
83 #ifndef __SOCKUNION_DECLARED
85 struct sockaddr_storage ss;
87 struct sockaddr_dl sdl;
88 struct sockaddr_in6 sin6;
90 typedef union sockunion sockunion_t;
91 #define __SOCKUNION_DECLARED
92 #endif /* __SOCKUNION_DECLARED */
94 static MALLOC_DEFINE(M_IN6MFILTER, "in6_mfilter",
95 "IPv6 multicast PCB-layer source filter");
96 MALLOC_DEFINE(M_IP6MADDR, "in6_multi", "IPv6 multicast group");
97 static MALLOC_DEFINE(M_IP6MOPTS, "ip6_moptions", "IPv6 multicast options");
98 static MALLOC_DEFINE(M_IP6MSOURCE, "ip6_msource",
99 "IPv6 multicast MLD-layer source filter");
101 RB_GENERATE(ip6_msource_tree, ip6_msource, im6s_link, ip6_msource_cmp);
105 * - Lock order is: Giant, IN6_MULTI_LOCK, INP_WLOCK,
106 * IN6_MULTI_LIST_LOCK, MLD_LOCK, IF_ADDR_LOCK.
107 * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
108 * it can be taken by code in net/if.c also.
109 * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
111 * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
112 * any need for in6_multi itself to be virtualized -- it is bound to an ifp
113 * anyway no matter what happens.
115 struct mtx in6_multi_list_mtx;
116 MTX_SYSINIT(in6_multi_mtx, &in6_multi_list_mtx, "in6_multi_list_mtx", MTX_DEF);
118 struct mtx in6_multi_free_mtx;
119 MTX_SYSINIT(in6_multi_free_mtx, &in6_multi_free_mtx, "in6_multi_free_mtx", MTX_DEF);
121 struct sx in6_multi_sx;
122 SX_SYSINIT(in6_multi_sx, &in6_multi_sx, "in6_multi_sx");
124 static void im6f_commit(struct in6_mfilter *);
125 static int im6f_get_source(struct in6_mfilter *imf,
126 const struct sockaddr_in6 *psin,
127 struct in6_msource **);
128 static struct in6_msource *
129 im6f_graft(struct in6_mfilter *, const uint8_t,
130 const struct sockaddr_in6 *);
131 static void im6f_leave(struct in6_mfilter *);
132 static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
133 static void im6f_purge(struct in6_mfilter *);
134 static void im6f_rollback(struct in6_mfilter *);
135 static void im6f_reap(struct in6_mfilter *);
136 static struct in6_mfilter *
137 im6o_match_group(const struct ip6_moptions *,
138 const struct ifnet *, const struct sockaddr *);
139 static struct in6_msource *
140 im6o_match_source(struct in6_mfilter *, const struct sockaddr *);
141 static void im6s_merge(struct ip6_msource *ims,
142 const struct in6_msource *lims, const int rollback);
143 static int in6_getmulti(struct ifnet *, const struct in6_addr *,
144 struct in6_multi **);
145 static int in6_joingroup_locked(struct ifnet *, const struct in6_addr *,
146 struct in6_mfilter *, struct in6_multi **, int);
147 static int in6m_get_source(struct in6_multi *inm,
148 const struct in6_addr *addr, const int noalloc,
149 struct ip6_msource **pims);
151 static int in6m_is_ifp_detached(const struct in6_multi *);
153 static int in6m_merge(struct in6_multi *, /*const*/ struct in6_mfilter *);
154 static void in6m_purge(struct in6_multi *);
155 static void in6m_reap(struct in6_multi *);
156 static struct ip6_moptions *
157 in6p_findmoptions(struct inpcb *);
158 static int in6p_get_source_filters(struct inpcb *, struct sockopt *);
159 static int in6p_join_group(struct inpcb *, struct sockopt *);
160 static int in6p_leave_group(struct inpcb *, struct sockopt *);
161 static struct ifnet *
162 in6p_lookup_mcast_ifp(const struct inpcb *,
163 const struct sockaddr_in6 *);
164 static int in6p_block_unblock_source(struct inpcb *, struct sockopt *);
165 static int in6p_set_multicast_if(struct inpcb *, struct sockopt *);
166 static int in6p_set_source_filters(struct inpcb *, struct sockopt *);
167 static int sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS);
169 SYSCTL_DECL(_net_inet6_ip6); /* XXX Not in any common header. */
171 static SYSCTL_NODE(_net_inet6_ip6, OID_AUTO, mcast,
172 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
175 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
176 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
177 CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
178 "Max source filters per group");
180 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
181 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
182 CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
183 "Max source filters per socket");
185 /* TODO Virtualize this switch. */
186 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
187 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
188 &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
190 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
191 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
192 "Per-interface stack-wide source filters");
196 * Inline function which wraps assertions for a valid ifp.
197 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
201 in6m_is_ifp_detached(const struct in6_multi *inm)
205 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
206 ifp = inm->in6m_ifma->ifma_ifp;
209 * Sanity check that network-layer notion of ifp is the
210 * same as that of link-layer.
212 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
215 return (ifp == NULL);
220 * Initialize an in6_mfilter structure to a known state at t0, t1
221 * with an empty source filter list.
224 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
226 memset(imf, 0, sizeof(struct in6_mfilter));
227 RB_INIT(&imf->im6f_sources);
228 imf->im6f_st[0] = st0;
229 imf->im6f_st[1] = st1;
233 ip6_mfilter_alloc(const int mflags, const int st0, const int st1)
235 struct in6_mfilter *imf;
237 imf = malloc(sizeof(*imf), M_IN6MFILTER, mflags);
240 im6f_init(imf, st0, st1);
246 ip6_mfilter_free(struct in6_mfilter *imf)
250 free(imf, M_IN6MFILTER);
254 * Find an IPv6 multicast group entry for this ip6_moptions instance
255 * which matches the specified group, and optionally an interface.
256 * Return its index into the array, or -1 if not found.
258 static struct in6_mfilter *
259 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
260 const struct sockaddr *group)
262 const struct sockaddr_in6 *gsin6;
263 struct in6_mfilter *imf;
264 struct in6_multi *inm;
266 gsin6 = (const struct sockaddr_in6 *)group;
268 IP6_MFILTER_FOREACH(imf, &imo->im6o_head) {
269 inm = imf->im6f_in6m;
272 if ((ifp == NULL || (inm->in6m_ifp == ifp)) &&
273 IN6_ARE_ADDR_EQUAL(&inm->in6m_addr,
274 &gsin6->sin6_addr)) {
282 * Find an IPv6 multicast source entry for this imo which matches
283 * the given group index for this socket, and source address.
285 * XXX TODO: The scope ID, if present in src, is stripped before
286 * any comparison. We SHOULD enforce scope/zone checks where the source
287 * filter entry has a link scope.
289 * NOTE: This does not check if the entry is in-mode, merely if
290 * it exists, which may not be the desired behaviour.
292 static struct in6_msource *
293 im6o_match_source(struct in6_mfilter *imf, const struct sockaddr *src)
295 struct ip6_msource find;
296 struct ip6_msource *ims;
297 const sockunion_t *psa;
299 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
301 psa = (const sockunion_t *)src;
302 find.im6s_addr = psa->sin6.sin6_addr;
303 in6_clearscope(&find.im6s_addr); /* XXX */
304 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
306 return ((struct in6_msource *)ims);
310 * Perform filtering for multicast datagrams on a socket by group and source.
312 * Returns 0 if a datagram should be allowed through, or various error codes
313 * if the socket was not a member of the group, or the source was muted, etc.
316 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
317 const struct sockaddr *group, const struct sockaddr *src)
319 struct in6_mfilter *imf;
320 struct in6_msource *ims;
323 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
325 imf = im6o_match_group(imo, ifp, group);
327 return (MCAST_NOTGMEMBER);
330 * Check if the source was included in an (S,G) join.
331 * Allow reception on exclusive memberships by default,
332 * reject reception on inclusive memberships by default.
333 * Exclude source only if an in-mode exclude filter exists.
334 * Include source only if an in-mode include filter exists.
335 * NOTE: We are comparing group state here at MLD t1 (now)
336 * with socket-layer t0 (since last downcall).
338 mode = imf->im6f_st[1];
339 ims = im6o_match_source(imf, src);
341 if ((ims == NULL && mode == MCAST_INCLUDE) ||
342 (ims != NULL && ims->im6sl_st[0] != mode))
343 return (MCAST_NOTSMEMBER);
349 * Find and return a reference to an in6_multi record for (ifp, group),
350 * and bump its reference count.
351 * If one does not exist, try to allocate it, and update link-layer multicast
352 * filters on ifp to listen for group.
353 * Assumes the IN6_MULTI lock is held across the call.
354 * Return 0 if successful, otherwise return an appropriate error code.
357 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
358 struct in6_multi **pinm)
360 struct epoch_tracker et;
361 struct sockaddr_in6 gsin6;
362 struct ifmultiaddr *ifma;
363 struct in6_multi *inm;
369 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
370 * if_addmulti() takes this mutex itself, so we must drop and
371 * re-acquire around the call.
373 IN6_MULTI_LOCK_ASSERT();
374 IN6_MULTI_LIST_LOCK();
377 inm = in6m_lookup_locked(ifp, group);
382 * If we already joined this group, just bump the
383 * refcount and return it.
385 KASSERT(inm->in6m_refcount >= 1,
386 ("%s: bad refcount %d", __func__, inm->in6m_refcount));
387 in6m_acquire_locked(inm);
392 memset(&gsin6, 0, sizeof(gsin6));
393 gsin6.sin6_family = AF_INET6;
394 gsin6.sin6_len = sizeof(struct sockaddr_in6);
395 gsin6.sin6_addr = *group;
398 * Check if a link-layer group is already associated
399 * with this network-layer group on the given ifnet.
401 IN6_MULTI_LIST_UNLOCK();
402 IF_ADDR_WUNLOCK(ifp);
403 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
406 IN6_MULTI_LIST_LOCK();
410 * If something other than netinet6 is occupying the link-layer
411 * group, print a meaningful error message and back out of
413 * Otherwise, bump the refcount on the existing network-layer
414 * group association and return it.
416 if (ifma->ifma_protospec != NULL) {
417 inm = (struct in6_multi *)ifma->ifma_protospec;
419 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
421 KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
422 ("%s: ifma not AF_INET6", __func__));
423 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
424 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
425 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
426 panic("%s: ifma %p is inconsistent with %p (%p)",
427 __func__, ifma, inm, group);
429 in6m_acquire_locked(inm);
434 IF_ADDR_WLOCK_ASSERT(ifp);
437 * A new in6_multi record is needed; allocate and initialize it.
438 * We DO NOT perform an MLD join as the in6_ layer may need to
439 * push an initial source list down to MLD to support SSM.
441 * The initial source filter state is INCLUDE, {} as per the RFC.
442 * Pending state-changes per group are subject to a bounds check.
444 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
446 IN6_MULTI_LIST_UNLOCK();
447 IF_ADDR_WUNLOCK(ifp);
448 if_delmulti_ifma(ifma);
451 inm->in6m_addr = *group;
453 inm->in6m_mli = MLD_IFINFO(ifp);
454 inm->in6m_ifma = ifma;
455 inm->in6m_refcount = 1;
456 inm->in6m_state = MLD_NOT_MEMBER;
457 mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
459 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
460 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
461 RB_INIT(&inm->in6m_srcs);
463 ifma->ifma_protospec = inm;
467 IN6_MULTI_LIST_UNLOCK();
468 IF_ADDR_WUNLOCK(ifp);
473 * Drop a reference to an in6_multi record.
475 * If the refcount drops to 0, free the in6_multi record and
476 * delete the underlying link-layer membership.
479 in6m_release(struct in6_multi *inm)
481 struct ifmultiaddr *ifma;
484 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
486 MPASS(inm->in6m_refcount == 0);
487 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
489 ifma = inm->in6m_ifma;
491 MPASS(ifma->ifma_llifma == NULL);
493 /* XXX this access is not covered by IF_ADDR_LOCK */
494 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
495 KASSERT(ifma->ifma_protospec == NULL,
496 ("%s: ifma_protospec != NULL", __func__));
498 ifp = ifma->ifma_ifp;
501 CURVNET_SET(ifp->if_vnet);
503 free(inm, M_IP6MADDR);
504 if_delmulti_ifma_flags(ifma, 1);
509 free(inm, M_IP6MADDR);
510 if_delmulti_ifma_flags(ifma, 1);
515 * Interface detach can happen in a taskqueue thread context, so we must use a
516 * dedicated thread to avoid deadlocks when draining in6m_release tasks.
518 TASKQUEUE_DEFINE_THREAD(in6m_free);
519 static struct task in6m_free_task;
520 static struct in6_multi_head in6m_free_list = SLIST_HEAD_INITIALIZER();
521 static void in6m_release_task(void *arg __unused, int pending __unused);
524 in6m_init(void *arg __unused)
526 TASK_INIT(&in6m_free_task, 0, in6m_release_task, NULL);
528 SYSINIT(in6m_init, SI_SUB_TASKQ, SI_ORDER_ANY, in6m_init, NULL);
531 in6m_release_list_deferred(struct in6_multi_head *inmh)
533 if (SLIST_EMPTY(inmh))
535 mtx_lock(&in6_multi_free_mtx);
536 SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
537 mtx_unlock(&in6_multi_free_mtx);
538 taskqueue_enqueue(taskqueue_in6m_free, &in6m_free_task);
542 in6m_release_wait(void)
544 taskqueue_drain_all(taskqueue_in6m_free);
548 in6m_disconnect_locked(struct in6_multi_head *inmh, struct in6_multi *inm)
552 struct in6_ifaddr *ifa6;
553 struct in6_multi_mship *imm, *imm_tmp;
554 struct ifmultiaddr *ifma, *ll_ifma;
556 IN6_MULTI_LIST_LOCK_ASSERT();
560 return; /* already called */
562 inm->in6m_ifp = NULL;
563 IF_ADDR_WLOCK_ASSERT(ifp);
564 ifma = inm->in6m_ifma;
569 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
570 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
571 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
573 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
574 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
575 MPASS(ifma != ll_ifma);
576 ifma->ifma_llifma = NULL;
577 MPASS(ll_ifma->ifma_llifma == NULL);
578 MPASS(ll_ifma->ifma_ifp == ifp);
579 if (--ll_ifma->ifma_refcount == 0) {
580 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
581 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
582 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
584 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
585 if_freemulti(ll_ifma);
588 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
589 if (ifa->ifa_addr->sa_family != AF_INET6)
592 LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
593 i6mm_chain, imm_tmp) {
594 if (inm == imm->i6mm_maddr) {
595 LIST_REMOVE(imm, i6mm_chain);
596 free(imm, M_IP6MADDR);
597 in6m_rele_locked(inmh, inm);
604 in6m_release_task(void *arg __unused, int pending __unused)
606 struct in6_multi_head in6m_free_tmp;
607 struct in6_multi *inm, *tinm;
609 SLIST_INIT(&in6m_free_tmp);
610 mtx_lock(&in6_multi_free_mtx);
611 SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
612 mtx_unlock(&in6_multi_free_mtx);
614 SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
615 SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
622 * Clear recorded source entries for a group.
623 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
624 * FIXME: Should reap.
627 in6m_clear_recorded(struct in6_multi *inm)
629 struct ip6_msource *ims;
631 IN6_MULTI_LIST_LOCK_ASSERT();
633 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
636 --inm->in6m_st[1].iss_rec;
639 KASSERT(inm->in6m_st[1].iss_rec == 0,
640 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
644 * Record a source as pending for a Source-Group MLDv2 query.
645 * This lives here as it modifies the shared tree.
647 * inm is the group descriptor.
648 * naddr is the address of the source to record in network-byte order.
650 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
651 * lazy-allocate a source node in response to an SG query.
652 * Otherwise, no allocation is performed. This saves some memory
653 * with the trade-off that the source will not be reported to the
654 * router if joined in the window between the query response and
655 * the group actually being joined on the local host.
657 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
658 * This turns off the allocation of a recorded source entry if
659 * the group has not been joined.
661 * Return 0 if the source didn't exist or was already marked as recorded.
662 * Return 1 if the source was marked as recorded by this function.
663 * Return <0 if any error occurred (negated errno code).
666 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
668 struct ip6_msource find;
669 struct ip6_msource *ims, *nims;
671 IN6_MULTI_LIST_LOCK_ASSERT();
673 find.im6s_addr = *addr;
674 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
675 if (ims && ims->im6s_stp)
678 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
680 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
684 nims->im6s_addr = find.im6s_addr;
685 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
691 * Mark the source as recorded and update the recorded
695 ++inm->in6m_st[1].iss_rec;
701 * Return a pointer to an in6_msource owned by an in6_mfilter,
702 * given its source address.
703 * Lazy-allocate if needed. If this is a new entry its filter state is
706 * imf is the filter set being modified.
707 * addr is the source address.
709 * SMPng: May be called with locks held; malloc must not block.
712 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
713 struct in6_msource **plims)
715 struct ip6_msource find;
716 struct ip6_msource *ims, *nims;
717 struct in6_msource *lims;
724 find.im6s_addr = psin->sin6_addr;
725 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
726 lims = (struct in6_msource *)ims;
728 if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
730 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
734 lims = (struct in6_msource *)nims;
735 lims->im6s_addr = find.im6s_addr;
736 lims->im6sl_st[0] = MCAST_UNDEFINED;
737 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
747 * Graft a source entry into an existing socket-layer filter set,
748 * maintaining any required invariants and checking allocations.
750 * The source is marked as being in the new filter mode at t1.
752 * Return the pointer to the new node, otherwise return NULL.
754 static struct in6_msource *
755 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
756 const struct sockaddr_in6 *psin)
758 struct ip6_msource *nims;
759 struct in6_msource *lims;
761 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
765 lims = (struct in6_msource *)nims;
766 lims->im6s_addr = psin->sin6_addr;
767 lims->im6sl_st[0] = MCAST_UNDEFINED;
768 lims->im6sl_st[1] = st1;
769 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
776 * Prune a source entry from an existing socket-layer filter set,
777 * maintaining any required invariants and checking allocations.
779 * The source is marked as being left at t1, it is not freed.
781 * Return 0 if no error occurred, otherwise return an errno value.
784 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
786 struct ip6_msource find;
787 struct ip6_msource *ims;
788 struct in6_msource *lims;
790 find.im6s_addr = psin->sin6_addr;
791 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
794 lims = (struct in6_msource *)ims;
795 lims->im6sl_st[1] = MCAST_UNDEFINED;
800 * Revert socket-layer filter set deltas at t1 to t0 state.
803 im6f_rollback(struct in6_mfilter *imf)
805 struct ip6_msource *ims, *tims;
806 struct in6_msource *lims;
808 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
809 lims = (struct in6_msource *)ims;
810 if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
811 /* no change at t1 */
813 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
814 /* revert change to existing source at t1 */
815 lims->im6sl_st[1] = lims->im6sl_st[0];
817 /* revert source added t1 */
818 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
819 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
820 free(ims, M_IN6MFILTER);
824 imf->im6f_st[1] = imf->im6f_st[0];
828 * Mark socket-layer filter set as INCLUDE {} at t1.
831 im6f_leave(struct in6_mfilter *imf)
833 struct ip6_msource *ims;
834 struct in6_msource *lims;
836 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
837 lims = (struct in6_msource *)ims;
838 lims->im6sl_st[1] = MCAST_UNDEFINED;
840 imf->im6f_st[1] = MCAST_INCLUDE;
844 * Mark socket-layer filter set deltas as committed.
847 im6f_commit(struct in6_mfilter *imf)
849 struct ip6_msource *ims;
850 struct in6_msource *lims;
852 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
853 lims = (struct in6_msource *)ims;
854 lims->im6sl_st[0] = lims->im6sl_st[1];
856 imf->im6f_st[0] = imf->im6f_st[1];
860 * Reap unreferenced sources from socket-layer filter set.
863 im6f_reap(struct in6_mfilter *imf)
865 struct ip6_msource *ims, *tims;
866 struct in6_msource *lims;
868 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
869 lims = (struct in6_msource *)ims;
870 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
871 (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
872 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
873 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
874 free(ims, M_IN6MFILTER);
881 * Purge socket-layer filter set.
884 im6f_purge(struct in6_mfilter *imf)
886 struct ip6_msource *ims, *tims;
888 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
889 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
890 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
891 free(ims, M_IN6MFILTER);
894 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
895 KASSERT(RB_EMPTY(&imf->im6f_sources),
896 ("%s: im6f_sources not empty", __func__));
900 * Look up a source filter entry for a multicast group.
902 * inm is the group descriptor to work with.
903 * addr is the IPv6 address to look up.
904 * noalloc may be non-zero to suppress allocation of sources.
905 * *pims will be set to the address of the retrieved or allocated source.
907 * SMPng: NOTE: may be called with locks held.
908 * Return 0 if successful, otherwise return a non-zero error code.
911 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
912 const int noalloc, struct ip6_msource **pims)
914 struct ip6_msource find;
915 struct ip6_msource *ims, *nims;
917 char ip6tbuf[INET6_ADDRSTRLEN];
920 find.im6s_addr = *addr;
921 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
922 if (ims == NULL && !noalloc) {
923 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
925 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
929 nims->im6s_addr = *addr;
930 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
933 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
934 ip6_sprintf(ip6tbuf, addr), ims);
942 * Merge socket-layer source into MLD-layer source.
943 * If rollback is non-zero, perform the inverse of the merge.
946 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
949 int n = rollback ? -1 : 1;
951 char ip6tbuf[INET6_ADDRSTRLEN];
953 ip6_sprintf(ip6tbuf, &lims->im6s_addr);
956 if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
957 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
958 ims->im6s_st[1].ex -= n;
959 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
960 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
961 ims->im6s_st[1].in -= n;
964 if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
965 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
966 ims->im6s_st[1].ex += n;
967 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
968 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
969 ims->im6s_st[1].in += n;
974 * Atomically update the global in6_multi state, when a membership's
975 * filter list is being updated in any way.
977 * imf is the per-inpcb-membership group filter pointer.
978 * A fake imf may be passed for in-kernel consumers.
980 * XXX This is a candidate for a set-symmetric-difference style loop
981 * which would eliminate the repeated lookup from root of ims nodes,
982 * as they share the same key space.
984 * If any error occurred this function will back out of refcounts
985 * and return a non-zero value.
988 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
990 struct ip6_msource *ims, *nims;
991 struct in6_msource *lims;
998 IN6_MULTI_LIST_LOCK_ASSERT();
1001 * Update the source filters first, as this may fail.
1002 * Maintain count of in-mode filters at t0, t1. These are
1003 * used to work out if we transition into ASM mode or not.
1004 * Maintain a count of source filters whose state was
1005 * actually modified by this operation.
1007 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1008 lims = (struct in6_msource *)ims;
1009 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1010 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1011 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1012 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1016 im6s_merge(nims, lims, 0);
1019 struct ip6_msource *bims;
1021 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1022 lims = (struct in6_msource *)ims;
1023 if (lims->im6sl_st[0] == lims->im6sl_st[1])
1025 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1028 im6s_merge(bims, lims, 1);
1033 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1034 __func__, nsrc0, nsrc1);
1036 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1037 if (imf->im6f_st[0] == imf->im6f_st[1] &&
1038 imf->im6f_st[1] == MCAST_INCLUDE) {
1040 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1041 --inm->in6m_st[1].iss_in;
1045 /* Handle filter mode transition on socket. */
1046 if (imf->im6f_st[0] != imf->im6f_st[1]) {
1047 CTR3(KTR_MLD, "%s: imf transition %d to %d",
1048 __func__, imf->im6f_st[0], imf->im6f_st[1]);
1050 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1051 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1052 --inm->in6m_st[1].iss_ex;
1053 } else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1054 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1055 --inm->in6m_st[1].iss_in;
1058 if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1059 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1060 inm->in6m_st[1].iss_ex++;
1061 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1062 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1063 inm->in6m_st[1].iss_in++;
1068 * Track inm filter state in terms of listener counts.
1069 * If there are any exclusive listeners, stack-wide
1070 * membership is exclusive.
1071 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1072 * If no listeners remain, state is undefined at t1,
1073 * and the MLD lifecycle for this group should finish.
1075 if (inm->in6m_st[1].iss_ex > 0) {
1076 CTR1(KTR_MLD, "%s: transition to EX", __func__);
1077 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1078 } else if (inm->in6m_st[1].iss_in > 0) {
1079 CTR1(KTR_MLD, "%s: transition to IN", __func__);
1080 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1082 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1083 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1086 /* Decrement ASM listener count on transition out of ASM mode. */
1087 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1088 if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1089 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1090 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1091 --inm->in6m_st[1].iss_asm;
1095 /* Increment ASM listener count on transition to ASM mode. */
1096 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1097 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1098 inm->in6m_st[1].iss_asm++;
1101 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1106 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1113 * Mark an in6_multi's filter set deltas as committed.
1114 * Called by MLD after a state change has been enqueued.
1117 in6m_commit(struct in6_multi *inm)
1119 struct ip6_msource *ims;
1121 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1122 CTR1(KTR_MLD, "%s: pre commit:", __func__);
1125 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1126 ims->im6s_st[0] = ims->im6s_st[1];
1128 inm->in6m_st[0] = inm->in6m_st[1];
1132 * Reap unreferenced nodes from an in6_multi's filter set.
1135 in6m_reap(struct in6_multi *inm)
1137 struct ip6_msource *ims, *tims;
1139 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1140 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1141 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1144 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1145 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1146 free(ims, M_IP6MSOURCE);
1152 * Purge all source nodes from an in6_multi's filter set.
1155 in6m_purge(struct in6_multi *inm)
1157 struct ip6_msource *ims, *tims;
1159 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1160 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1161 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1162 free(ims, M_IP6MSOURCE);
1165 /* Free state-change requests that might be queued. */
1166 mbufq_drain(&inm->in6m_scq);
1170 * Join a multicast address w/o sources.
1171 * KAME compatibility entry point.
1173 * SMPng: Assume no mc locks held by caller.
1176 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1177 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1183 error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1189 * Join a multicast group; real entry point.
1191 * Only preserves atomicity at inm level.
1192 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1194 * If the MLD downcall fails, the group is not joined, and an error
1198 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1199 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1202 struct in6_multi_head inmh;
1203 struct in6_mfilter timf;
1204 struct in6_multi *inm;
1205 struct ifmultiaddr *ifma;
1208 char ip6tbuf[INET6_ADDRSTRLEN];
1212 * Sanity: Check scope zone ID was set for ifp, if and
1213 * only if group is scoped to an interface.
1215 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1216 ("%s: not a multicast address", __func__));
1217 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1218 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1219 KASSERT(mcaddr->s6_addr16[1] != 0,
1220 ("%s: scope zone ID not set", __func__));
1223 IN6_MULTI_LOCK_ASSERT();
1224 IN6_MULTI_LIST_UNLOCK_ASSERT();
1226 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1227 ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1233 * If no imf was specified (i.e. kernel consumer),
1234 * fake one up and assume it is an ASM join.
1237 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1240 error = in6_getmulti(ifp, mcaddr, &inm);
1242 CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1246 IN6_MULTI_LIST_LOCK();
1247 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1248 error = in6m_merge(inm, imf);
1250 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1251 goto out_in6m_release;
1254 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1255 error = mld_change_state(inm, delay);
1257 CTR1(KTR_MLD, "%s: failed to update source", __func__);
1258 goto out_in6m_release;
1264 struct epoch_tracker et;
1266 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1268 NET_EPOCH_ENTER(et);
1269 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1270 if (ifma->ifma_protospec == inm) {
1271 ifma->ifma_protospec = NULL;
1275 in6m_disconnect_locked(&inmh, inm);
1276 in6m_rele_locked(&inmh, inm);
1278 IF_ADDR_WUNLOCK(ifp);
1282 IN6_MULTI_LIST_UNLOCK();
1283 in6m_release_list_deferred(&inmh);
1288 * Leave a multicast group; unlocked entry point.
1291 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1296 error = in6_leavegroup_locked(inm, imf);
1302 * Leave a multicast group; real entry point.
1303 * All source filters will be expunged.
1305 * Only preserves atomicity at inm level.
1307 * Holding the write lock for the INP which contains imf
1308 * is highly advisable. We can't assert for it as imf does not
1309 * contain a back-pointer to the owning inp.
1311 * Note: This is not the same as in6m_release(*) as this function also
1312 * makes a state change downcall into MLD.
1315 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1317 struct in6_multi_head inmh;
1318 struct in6_mfilter timf;
1322 char ip6tbuf[INET6_ADDRSTRLEN];
1327 IN6_MULTI_LOCK_ASSERT();
1329 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1330 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1331 (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1335 * If no imf was specified (i.e. kernel consumer),
1336 * fake one up and assume it is an ASM join.
1339 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1344 * Begin state merge transaction at MLD layer.
1346 * As this particular invocation should not cause any memory
1347 * to be allocated, and there is no opportunity to roll back
1348 * the transaction, it MUST NOT fail.
1351 ifp = inm->in6m_ifp;
1352 IN6_MULTI_LIST_LOCK();
1353 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1354 error = in6m_merge(inm, imf);
1355 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1357 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1360 error = mld_change_state(inm, 0);
1362 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1364 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1369 if (inm->in6m_refcount == 1)
1370 in6m_disconnect_locked(&inmh, inm);
1371 in6m_rele_locked(&inmh, inm);
1373 IF_ADDR_WUNLOCK(ifp);
1374 IN6_MULTI_LIST_UNLOCK();
1375 in6m_release_list_deferred(&inmh);
1381 * Block or unblock an ASM multicast source on an inpcb.
1382 * This implements the delta-based API described in RFC 3678.
1384 * The delta-based API applies only to exclusive-mode memberships.
1385 * An MLD downcall will be performed.
1387 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1389 * Return 0 if successful, otherwise return an appropriate error code.
1392 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1394 struct group_source_req gsr;
1395 sockunion_t *gsa, *ssa;
1397 struct in6_mfilter *imf;
1398 struct ip6_moptions *imo;
1399 struct in6_msource *ims;
1400 struct in6_multi *inm;
1404 char ip6tbuf[INET6_ADDRSTRLEN];
1411 memset(&gsr, 0, sizeof(struct group_source_req));
1412 gsa = (sockunion_t *)&gsr.gsr_group;
1413 ssa = (sockunion_t *)&gsr.gsr_source;
1415 switch (sopt->sopt_name) {
1416 case MCAST_BLOCK_SOURCE:
1417 case MCAST_UNBLOCK_SOURCE:
1418 error = sooptcopyin(sopt, &gsr,
1419 sizeof(struct group_source_req),
1420 sizeof(struct group_source_req));
1424 if (gsa->sin6.sin6_family != AF_INET6 ||
1425 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1428 if (ssa->sin6.sin6_family != AF_INET6 ||
1429 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1432 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1433 return (EADDRNOTAVAIL);
1435 ifp = ifnet_byindex(gsr.gsr_interface);
1437 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1442 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1443 __func__, sopt->sopt_name);
1444 return (EOPNOTSUPP);
1448 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1451 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1454 * Check if we are actually a member of this group.
1456 imo = in6p_findmoptions(inp);
1457 imf = im6o_match_group(imo, ifp, &gsa->sa);
1459 error = EADDRNOTAVAIL;
1460 goto out_in6p_locked;
1462 inm = imf->im6f_in6m;
1465 * Attempting to use the delta-based API on an
1466 * non exclusive-mode membership is an error.
1468 fmode = imf->im6f_st[0];
1469 if (fmode != MCAST_EXCLUDE) {
1471 goto out_in6p_locked;
1475 * Deal with error cases up-front:
1476 * Asked to block, but already blocked; or
1477 * Asked to unblock, but nothing to unblock.
1478 * If adding a new block entry, allocate it.
1480 ims = im6o_match_source(imf, &ssa->sa);
1481 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1482 CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1483 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1484 doblock ? "" : "not ");
1485 error = EADDRNOTAVAIL;
1486 goto out_in6p_locked;
1489 INP_WLOCK_ASSERT(inp);
1492 * Begin state merge transaction at socket layer.
1495 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1496 ims = im6f_graft(imf, fmode, &ssa->sin6);
1500 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1501 error = im6f_prune(imf, &ssa->sin6);
1505 CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1506 goto out_im6f_rollback;
1510 * Begin state merge transaction at MLD layer.
1512 IN6_MULTI_LIST_LOCK();
1513 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1514 error = in6m_merge(inm, imf);
1516 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1518 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1519 error = mld_change_state(inm, 0);
1521 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1524 IN6_MULTI_LIST_UNLOCK();
1540 * Given an inpcb, return its multicast options structure pointer. Accepts
1541 * an unlocked inpcb pointer, but will return it locked. May sleep.
1543 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1544 * SMPng: NOTE: Returns with the INP write lock held.
1546 static struct ip6_moptions *
1547 in6p_findmoptions(struct inpcb *inp)
1549 struct ip6_moptions *imo;
1552 if (inp->in6p_moptions != NULL)
1553 return (inp->in6p_moptions);
1557 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1559 imo->im6o_multicast_ifp = NULL;
1560 imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1561 imo->im6o_multicast_loop = in6_mcast_loop;
1562 STAILQ_INIT(&imo->im6o_head);
1565 if (inp->in6p_moptions != NULL) {
1566 free(imo, M_IP6MOPTS);
1567 return (inp->in6p_moptions);
1569 inp->in6p_moptions = imo;
1574 * Discard the IPv6 multicast options (and source filters).
1576 * SMPng: NOTE: assumes INP write lock is held.
1578 * XXX can all be safely deferred to epoch_call
1583 inp_gcmoptions(struct ip6_moptions *imo)
1585 struct in6_mfilter *imf;
1586 struct in6_multi *inm;
1589 while ((imf = ip6_mfilter_first(&imo->im6o_head)) != NULL) {
1590 ip6_mfilter_remove(&imo->im6o_head, imf);
1593 if ((inm = imf->im6f_in6m) != NULL) {
1594 if ((ifp = inm->in6m_ifp) != NULL) {
1595 CURVNET_SET(ifp->if_vnet);
1596 (void)in6_leavegroup(inm, imf);
1599 (void)in6_leavegroup(inm, imf);
1602 ip6_mfilter_free(imf);
1604 free(imo, M_IP6MOPTS);
1608 ip6_freemoptions(struct ip6_moptions *imo)
1612 inp_gcmoptions(imo);
1616 * Atomically get source filters on a socket for an IPv6 multicast group.
1617 * Called with INP lock held; returns with lock released.
1620 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1622 struct __msfilterreq msfr;
1625 struct ip6_moptions *imo;
1626 struct in6_mfilter *imf;
1627 struct ip6_msource *ims;
1628 struct in6_msource *lims;
1629 struct sockaddr_in6 *psin;
1630 struct sockaddr_storage *ptss;
1631 struct sockaddr_storage *tss;
1633 size_t nsrcs, ncsrcs;
1635 INP_WLOCK_ASSERT(inp);
1637 imo = inp->in6p_moptions;
1638 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1642 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1643 sizeof(struct __msfilterreq));
1647 if (msfr.msfr_group.ss_family != AF_INET6 ||
1648 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1651 gsa = (sockunion_t *)&msfr.msfr_group;
1652 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1655 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1656 return (EADDRNOTAVAIL);
1657 ifp = ifnet_byindex(msfr.msfr_ifindex);
1659 return (EADDRNOTAVAIL);
1660 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1665 * Lookup group on the socket.
1667 imf = im6o_match_group(imo, ifp, &gsa->sa);
1670 return (EADDRNOTAVAIL);
1674 * Ignore memberships which are in limbo.
1676 if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1680 msfr.msfr_fmode = imf->im6f_st[1];
1683 * If the user specified a buffer, copy out the source filter
1684 * entries to userland gracefully.
1685 * We only copy out the number of entries which userland
1686 * has asked for, but we always tell userland how big the
1687 * buffer really needs to be.
1689 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1690 msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1692 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1693 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1694 M_TEMP, M_NOWAIT | M_ZERO);
1702 * Count number of sources in-mode at t0.
1703 * If buffer space exists and remains, copy out source entries.
1705 nsrcs = msfr.msfr_nsrcs;
1708 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1709 lims = (struct in6_msource *)ims;
1710 if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1711 lims->im6sl_st[0] != imf->im6f_st[0])
1714 if (tss != NULL && nsrcs > 0) {
1715 psin = (struct sockaddr_in6 *)ptss;
1716 psin->sin6_family = AF_INET6;
1717 psin->sin6_len = sizeof(struct sockaddr_in6);
1718 psin->sin6_addr = lims->im6s_addr;
1719 psin->sin6_port = 0;
1728 error = copyout(tss, msfr.msfr_srcs,
1729 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1735 msfr.msfr_nsrcs = ncsrcs;
1736 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1742 * Return the IP multicast options in response to user getsockopt().
1745 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1747 struct ip6_moptions *im6o;
1752 im6o = inp->in6p_moptions;
1754 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1755 * or is a divert socket, reject it.
1757 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1758 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1759 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1761 return (EOPNOTSUPP);
1765 switch (sopt->sopt_name) {
1766 case IPV6_MULTICAST_IF:
1767 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1770 optval = im6o->im6o_multicast_ifp->if_index;
1773 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1776 case IPV6_MULTICAST_HOPS:
1778 optval = V_ip6_defmcasthlim;
1780 optval = im6o->im6o_multicast_hlim;
1782 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1785 case IPV6_MULTICAST_LOOP:
1787 optval = in6_mcast_loop; /* XXX VIMAGE */
1789 optval = im6o->im6o_multicast_loop;
1791 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1796 error = EADDRNOTAVAIL;
1799 error = in6p_get_source_filters(inp, sopt);
1805 error = ENOPROTOOPT;
1809 INP_UNLOCK_ASSERT(inp);
1815 * Look up the ifnet to use for a multicast group membership,
1816 * given the address of an IPv6 group.
1818 * This routine exists to support legacy IPv6 multicast applications.
1820 * Use the socket's current FIB number for any required FIB lookup. Look up the
1821 * group address in the unicast FIB, and use its ifp; usually, this points to
1822 * the default next-hop. If the FIB lookup fails, return NULL.
1824 * FUTURE: Support multiple forwarding tables for IPv6.
1826 * Returns NULL if no ifp could be found.
1828 static struct ifnet *
1829 in6p_lookup_mcast_ifp(const struct inpcb *inp, const struct sockaddr_in6 *gsin6)
1831 struct nhop_object *nh;
1832 struct in6_addr dst;
1836 KASSERT(gsin6->sin6_family == AF_INET6,
1837 ("%s: not AF_INET6 group", __func__));
1839 in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1840 fibnum = inp->inp_inc.inc_fibnum;
1841 nh = fib6_lookup(fibnum, &dst, scopeid, 0, 0);
1843 return (nh ? nh->nh_ifp : NULL);
1847 * Join an IPv6 multicast group, possibly with a source.
1849 * FIXME: The KAME use of the unspecified address (::)
1850 * to join *all* multicast groups is currently unsupported.
1853 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1855 struct in6_multi_head inmh;
1856 struct group_source_req gsr;
1857 sockunion_t *gsa, *ssa;
1859 struct in6_mfilter *imf;
1860 struct ip6_moptions *imo;
1861 struct in6_multi *inm;
1862 struct in6_msource *lims;
1870 memset(&gsr, 0, sizeof(struct group_source_req));
1871 gsa = (sockunion_t *)&gsr.gsr_group;
1872 gsa->ss.ss_family = AF_UNSPEC;
1873 ssa = (sockunion_t *)&gsr.gsr_source;
1874 ssa->ss.ss_family = AF_UNSPEC;
1877 * Chew everything into struct group_source_req.
1878 * Overwrite the port field if present, as the sockaddr
1879 * being copied in may be matched with a binary comparison.
1880 * Ignore passed-in scope ID.
1882 switch (sopt->sopt_name) {
1883 case IPV6_JOIN_GROUP: {
1884 struct ipv6_mreq mreq;
1886 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1887 sizeof(struct ipv6_mreq));
1891 gsa->sin6.sin6_family = AF_INET6;
1892 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1893 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1895 if (mreq.ipv6mr_interface == 0) {
1896 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1898 if (V_if_index < mreq.ipv6mr_interface)
1899 return (EADDRNOTAVAIL);
1900 ifp = ifnet_byindex(mreq.ipv6mr_interface);
1902 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1903 __func__, mreq.ipv6mr_interface, ifp);
1906 case MCAST_JOIN_GROUP:
1907 case MCAST_JOIN_SOURCE_GROUP:
1908 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1909 error = sooptcopyin(sopt, &gsr,
1910 sizeof(struct group_req),
1911 sizeof(struct group_req));
1912 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1913 error = sooptcopyin(sopt, &gsr,
1914 sizeof(struct group_source_req),
1915 sizeof(struct group_source_req));
1920 if (gsa->sin6.sin6_family != AF_INET6 ||
1921 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1924 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1925 if (ssa->sin6.sin6_family != AF_INET6 ||
1926 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1928 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1931 * TODO: Validate embedded scope ID in source
1932 * list entry against passed-in ifp, if and only
1933 * if source list filter entry is iface or node local.
1935 in6_clearscope(&ssa->sin6.sin6_addr);
1936 ssa->sin6.sin6_port = 0;
1937 ssa->sin6.sin6_scope_id = 0;
1940 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1941 return (EADDRNOTAVAIL);
1942 ifp = ifnet_byindex(gsr.gsr_interface);
1946 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1947 __func__, sopt->sopt_name);
1948 return (EOPNOTSUPP);
1952 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1955 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1956 return (EADDRNOTAVAIL);
1958 gsa->sin6.sin6_port = 0;
1959 gsa->sin6.sin6_scope_id = 0;
1962 * Always set the scope zone ID on memberships created from userland.
1963 * Use the passed-in ifp to do this.
1964 * XXX The in6_setscope() return value is meaningless.
1965 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1967 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1972 * Find the membership in the membership list.
1974 imo = in6p_findmoptions(inp);
1975 imf = im6o_match_group(imo, ifp, &gsa->sa);
1980 if (ip6_mfilter_count(&imo->im6o_head) >= IPV6_MAX_MEMBERSHIPS) {
1982 goto out_in6p_locked;
1986 inm = imf->im6f_in6m;
1988 if (ssa->ss.ss_family != AF_UNSPEC) {
1990 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1991 * is an error. On an existing inclusive membership,
1992 * it just adds the source to the filter list.
1994 if (imf->im6f_st[1] != MCAST_INCLUDE) {
1996 goto out_in6p_locked;
1999 * Throw out duplicates.
2001 * XXX FIXME: This makes a naive assumption that
2002 * even if entries exist for *ssa in this imf,
2003 * they will be rejected as dupes, even if they
2004 * are not valid in the current mode (in-mode).
2006 * in6_msource is transactioned just as for anything
2007 * else in SSM -- but note naive use of in6m_graft()
2008 * below for allocating new filter entries.
2010 * This is only an issue if someone mixes the
2011 * full-state SSM API with the delta-based API,
2012 * which is discouraged in the relevant RFCs.
2014 lims = im6o_match_source(imf, &ssa->sa);
2015 if (lims != NULL /*&&
2016 lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2017 error = EADDRNOTAVAIL;
2018 goto out_in6p_locked;
2022 * MCAST_JOIN_GROUP alone, on any existing membership,
2023 * is rejected, to stop the same inpcb tying up
2024 * multiple refs to the in_multi.
2025 * On an existing inclusive membership, this is also
2026 * an error; if you want to change filter mode,
2027 * you must use the userland API setsourcefilter().
2028 * XXX We don't reject this for imf in UNDEFINED
2029 * state at t1, because allocation of a filter
2030 * is atomic with allocation of a membership.
2033 goto out_in6p_locked;
2038 * Begin state merge transaction at socket layer.
2040 INP_WLOCK_ASSERT(inp);
2043 * Graft new source into filter list for this inpcb's
2044 * membership of the group. The in6_multi may not have
2045 * been allocated yet if this is a new membership, however,
2046 * the in_mfilter slot will be allocated and must be initialized.
2048 * Note: Grafting of exclusive mode filters doesn't happen
2050 * XXX: Should check for non-NULL lims (node exists but may
2051 * not be in-mode) for interop with full-state API.
2053 if (ssa->ss.ss_family != AF_UNSPEC) {
2054 /* Membership starts in IN mode */
2056 CTR1(KTR_MLD, "%s: new join w/source", __func__);
2057 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2060 goto out_in6p_locked;
2063 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2065 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2067 CTR1(KTR_MLD, "%s: merge imf state failed",
2070 goto out_in6p_locked;
2073 /* No address specified; Membership starts in EX mode */
2075 CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2076 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2079 goto out_in6p_locked;
2085 * Begin state merge transaction at MLD layer.
2091 error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2092 &imf->im6f_in6m, 0);
2095 if (in_pcbrele_wlocked(inp)) {
2097 goto out_in6p_unlocked;
2100 goto out_in6p_locked;
2103 * NOTE: Refcount from in6_joingroup_locked()
2104 * is protecting membership.
2106 ip6_mfilter_insert(&imo->im6o_head, imf);
2108 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2109 IN6_MULTI_LIST_LOCK();
2110 error = in6m_merge(inm, imf);
2112 CTR1(KTR_MLD, "%s: failed to merge inm state",
2114 IN6_MULTI_LIST_UNLOCK();
2117 goto out_in6p_locked;
2119 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2120 error = mld_change_state(inm, 0);
2121 IN6_MULTI_LIST_UNLOCK();
2124 CTR1(KTR_MLD, "%s: failed mld downcall",
2128 goto out_in6p_locked;
2140 if (is_new && imf) {
2141 if (imf->im6f_in6m != NULL) {
2142 struct in6_multi_head inmh;
2145 SLIST_INSERT_HEAD(&inmh, imf->im6f_in6m, in6m_defer);
2146 in6m_release_list_deferred(&inmh);
2148 ip6_mfilter_free(imf);
2154 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2157 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2159 struct ipv6_mreq mreq;
2160 struct group_source_req gsr;
2161 sockunion_t *gsa, *ssa;
2163 struct in6_mfilter *imf;
2164 struct ip6_moptions *imo;
2165 struct in6_msource *ims;
2166 struct in6_multi *inm;
2171 char ip6tbuf[INET6_ADDRSTRLEN];
2179 memset(&gsr, 0, sizeof(struct group_source_req));
2180 gsa = (sockunion_t *)&gsr.gsr_group;
2181 gsa->ss.ss_family = AF_UNSPEC;
2182 ssa = (sockunion_t *)&gsr.gsr_source;
2183 ssa->ss.ss_family = AF_UNSPEC;
2186 * Chew everything passed in up into a struct group_source_req
2187 * as that is easier to process.
2188 * Note: Any embedded scope ID in the multicast group passed
2189 * in by userland is ignored, the interface index is the recommended
2190 * mechanism to specify an interface; see below.
2192 switch (sopt->sopt_name) {
2193 case IPV6_LEAVE_GROUP:
2194 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2195 sizeof(struct ipv6_mreq));
2198 gsa->sin6.sin6_family = AF_INET6;
2199 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2200 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2201 gsa->sin6.sin6_port = 0;
2202 gsa->sin6.sin6_scope_id = 0;
2203 ifindex = mreq.ipv6mr_interface;
2206 case MCAST_LEAVE_GROUP:
2207 case MCAST_LEAVE_SOURCE_GROUP:
2208 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2209 error = sooptcopyin(sopt, &gsr,
2210 sizeof(struct group_req),
2211 sizeof(struct group_req));
2212 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2213 error = sooptcopyin(sopt, &gsr,
2214 sizeof(struct group_source_req),
2215 sizeof(struct group_source_req));
2220 if (gsa->sin6.sin6_family != AF_INET6 ||
2221 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2223 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2224 if (ssa->sin6.sin6_family != AF_INET6 ||
2225 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2227 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2230 * TODO: Validate embedded scope ID in source
2231 * list entry against passed-in ifp, if and only
2232 * if source list filter entry is iface or node local.
2234 in6_clearscope(&ssa->sin6.sin6_addr);
2236 gsa->sin6.sin6_port = 0;
2237 gsa->sin6.sin6_scope_id = 0;
2238 ifindex = gsr.gsr_interface;
2242 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2243 __func__, sopt->sopt_name);
2244 return (EOPNOTSUPP);
2248 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2252 * Validate interface index if provided. If no interface index
2253 * was provided separately, attempt to look the membership up
2254 * from the default scope as a last resort to disambiguate
2255 * the membership we are being asked to leave.
2256 * XXX SCOPE6 lock potentially taken here.
2259 if (V_if_index < ifindex)
2260 return (EADDRNOTAVAIL);
2261 ifp = ifnet_byindex(ifindex);
2263 return (EADDRNOTAVAIL);
2264 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2266 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2268 return (EADDRNOTAVAIL);
2270 * Some badly behaved applications don't pass an ifindex
2271 * or a scope ID, which is an API violation. In this case,
2272 * perform a lookup as per a v6 join.
2274 * XXX For now, stomp on zone ID for the corner case.
2275 * This is not the 'KAME way', but we need to see the ifp
2276 * directly until such time as this implementation is
2277 * refactored, assuming the scope IDs are the way to go.
2279 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2281 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2282 "ifp for group %s.", __func__,
2283 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2284 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2286 ifp = ifnet_byindex(ifindex);
2289 return (EADDRNOTAVAIL);
2292 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2293 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2298 * Find the membership in the membership list.
2300 imo = in6p_findmoptions(inp);
2301 imf = im6o_match_group(imo, ifp, &gsa->sa);
2303 error = EADDRNOTAVAIL;
2304 goto out_in6p_locked;
2306 inm = imf->im6f_in6m;
2308 if (ssa->ss.ss_family != AF_UNSPEC)
2312 * Begin state merge transaction at socket layer.
2314 INP_WLOCK_ASSERT(inp);
2317 * If we were instructed only to leave a given source, do so.
2318 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2321 ip6_mfilter_remove(&imo->im6o_head, imf);
2325 * Give up the multicast address record to which
2326 * the membership points.
2328 (void)in6_leavegroup_locked(inm, imf);
2330 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2331 error = EADDRNOTAVAIL;
2332 goto out_in6p_locked;
2334 ims = im6o_match_source(imf, &ssa->sa);
2336 CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2337 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2339 error = EADDRNOTAVAIL;
2340 goto out_in6p_locked;
2342 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2343 error = im6f_prune(imf, &ssa->sin6);
2345 CTR1(KTR_MLD, "%s: merge imf state failed",
2347 goto out_in6p_locked;
2352 * Begin state merge transaction at MLD layer.
2355 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2356 IN6_MULTI_LIST_LOCK();
2357 error = in6m_merge(inm, imf);
2359 CTR1(KTR_MLD, "%s: failed to merge inm state",
2361 IN6_MULTI_LIST_UNLOCK();
2364 goto out_in6p_locked;
2367 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2368 error = mld_change_state(inm, 0);
2369 IN6_MULTI_LIST_UNLOCK();
2371 CTR1(KTR_MLD, "%s: failed mld downcall",
2375 goto out_in6p_locked;
2385 if (is_final && imf)
2386 ip6_mfilter_free(imf);
2393 * Select the interface for transmitting IPv6 multicast datagrams.
2395 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2396 * may be passed to this socket option. An address of in6addr_any or an
2397 * interface index of 0 is used to remove a previous selection.
2398 * When no interface is selected, one is chosen for every send.
2401 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2404 struct ip6_moptions *imo;
2408 if (sopt->sopt_valsize != sizeof(u_int))
2411 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2414 if (V_if_index < ifindex)
2419 ifp = ifnet_byindex(ifindex);
2422 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2423 return (EADDRNOTAVAIL);
2425 imo = in6p_findmoptions(inp);
2426 imo->im6o_multicast_ifp = ifp;
2433 * Atomically set source filters on a socket for an IPv6 multicast group.
2435 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2438 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2440 struct __msfilterreq msfr;
2443 struct in6_mfilter *imf;
2444 struct ip6_moptions *imo;
2445 struct in6_multi *inm;
2448 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2449 sizeof(struct __msfilterreq));
2453 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2456 if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2457 msfr.msfr_fmode != MCAST_INCLUDE)
2460 if (msfr.msfr_group.ss_family != AF_INET6 ||
2461 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2464 gsa = (sockunion_t *)&msfr.msfr_group;
2465 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2468 gsa->sin6.sin6_port = 0; /* ignore port */
2470 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2471 return (EADDRNOTAVAIL);
2472 ifp = ifnet_byindex(msfr.msfr_ifindex);
2474 return (EADDRNOTAVAIL);
2475 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2478 * Take the INP write lock.
2479 * Check if this socket is a member of this group.
2481 imo = in6p_findmoptions(inp);
2482 imf = im6o_match_group(imo, ifp, &gsa->sa);
2484 error = EADDRNOTAVAIL;
2485 goto out_in6p_locked;
2487 inm = imf->im6f_in6m;
2490 * Begin state merge transaction at socket layer.
2492 INP_WLOCK_ASSERT(inp);
2494 imf->im6f_st[1] = msfr.msfr_fmode;
2497 * Apply any new source filters, if present.
2498 * Make a copy of the user-space source vector so
2499 * that we may copy them with a single copyin. This
2500 * allows us to deal with page faults up-front.
2502 if (msfr.msfr_nsrcs > 0) {
2503 struct in6_msource *lims;
2504 struct sockaddr_in6 *psin;
2505 struct sockaddr_storage *kss, *pkss;
2510 CTR2(KTR_MLD, "%s: loading %lu source list entries",
2511 __func__, (unsigned long)msfr.msfr_nsrcs);
2512 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2514 error = copyin(msfr.msfr_srcs, kss,
2515 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2524 * Mark all source filters as UNDEFINED at t1.
2525 * Restore new group filter mode, as im6f_leave()
2526 * will set it to INCLUDE.
2529 imf->im6f_st[1] = msfr.msfr_fmode;
2532 * Update socket layer filters at t1, lazy-allocating
2533 * new entries. This saves a bunch of memory at the
2534 * cost of one RB_FIND() per source entry; duplicate
2535 * entries in the msfr_nsrcs vector are ignored.
2536 * If we encounter an error, rollback transaction.
2538 * XXX This too could be replaced with a set-symmetric
2539 * difference like loop to avoid walking from root
2540 * every time, as the key space is common.
2542 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2543 psin = (struct sockaddr_in6 *)pkss;
2544 if (psin->sin6_family != AF_INET6) {
2545 error = EAFNOSUPPORT;
2548 if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2552 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2557 * TODO: Validate embedded scope ID in source
2558 * list entry against passed-in ifp, if and only
2559 * if source list filter entry is iface or node local.
2561 in6_clearscope(&psin->sin6_addr);
2562 error = im6f_get_source(imf, psin, &lims);
2565 lims->im6sl_st[1] = imf->im6f_st[1];
2571 goto out_im6f_rollback;
2573 INP_WLOCK_ASSERT(inp);
2574 IN6_MULTI_LIST_LOCK();
2577 * Begin state merge transaction at MLD layer.
2579 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2580 error = in6m_merge(inm, imf);
2582 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2584 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2585 error = mld_change_state(inm, 0);
2587 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2590 IN6_MULTI_LIST_UNLOCK();
2606 * Set the IP multicast options in response to user setsockopt().
2608 * Many of the socket options handled in this function duplicate the
2609 * functionality of socket options in the regular unicast API. However,
2610 * it is not possible to merge the duplicate code, because the idempotence
2611 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2612 * the effects of these options must be treated as separate and distinct.
2614 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2617 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2619 struct ip6_moptions *im6o;
2625 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2626 * or is a divert socket, reject it.
2628 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2629 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2630 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2631 return (EOPNOTSUPP);
2633 switch (sopt->sopt_name) {
2634 case IPV6_MULTICAST_IF:
2635 error = in6p_set_multicast_if(inp, sopt);
2638 case IPV6_MULTICAST_HOPS: {
2641 if (sopt->sopt_valsize != sizeof(int)) {
2645 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2648 if (hlim < -1 || hlim > 255) {
2651 } else if (hlim == -1) {
2652 hlim = V_ip6_defmcasthlim;
2654 im6o = in6p_findmoptions(inp);
2655 im6o->im6o_multicast_hlim = hlim;
2660 case IPV6_MULTICAST_LOOP: {
2664 * Set the loopback flag for outgoing multicast packets.
2665 * Must be zero or one.
2667 if (sopt->sopt_valsize != sizeof(u_int)) {
2671 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2678 im6o = in6p_findmoptions(inp);
2679 im6o->im6o_multicast_loop = loop;
2684 case IPV6_JOIN_GROUP:
2685 case MCAST_JOIN_GROUP:
2686 case MCAST_JOIN_SOURCE_GROUP:
2687 error = in6p_join_group(inp, sopt);
2690 case IPV6_LEAVE_GROUP:
2691 case MCAST_LEAVE_GROUP:
2692 case MCAST_LEAVE_SOURCE_GROUP:
2693 error = in6p_leave_group(inp, sopt);
2696 case MCAST_BLOCK_SOURCE:
2697 case MCAST_UNBLOCK_SOURCE:
2698 error = in6p_block_unblock_source(inp, sopt);
2702 error = in6p_set_source_filters(inp, sopt);
2710 INP_UNLOCK_ASSERT(inp);
2716 * Expose MLD's multicast filter mode and source list(s) to userland,
2717 * keyed by (ifindex, group).
2718 * The filter mode is written out as a uint32_t, followed by
2719 * 0..n of struct in6_addr.
2720 * For use by ifmcstat(8).
2721 * SMPng: NOTE: unlocked read of ifindex space.
2724 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2726 struct in6_addr mcaddr;
2727 struct in6_addr src;
2728 struct epoch_tracker et;
2730 struct ifmultiaddr *ifma;
2731 struct in6_multi *inm;
2732 struct ip6_msource *ims;
2736 uint32_t fmode, ifindex;
2738 char ip6tbuf[INET6_ADDRSTRLEN];
2744 if (req->newptr != NULL)
2747 /* int: ifindex + 4 * 32 bits of IPv6 address */
2752 if (ifindex <= 0 || ifindex > V_if_index) {
2753 CTR2(KTR_MLD, "%s: ifindex %u out of range",
2758 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2759 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2760 CTR2(KTR_MLD, "%s: group %s is not multicast",
2761 __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2765 NET_EPOCH_ENTER(et);
2766 ifp = ifnet_byindex(ifindex);
2769 CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2774 * Internal MLD lookups require that scope/zone ID is set.
2776 (void)in6_setscope(&mcaddr, ifp, NULL);
2778 retval = sysctl_wire_old_buffer(req,
2779 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2786 IN6_MULTI_LIST_LOCK();
2787 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2788 inm = in6m_ifmultiaddr_get_inm(ifma);
2791 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2793 fmode = inm->in6m_st[1].iss_fmode;
2794 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2797 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2798 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2800 * Only copy-out sources which are in-mode.
2802 if (fmode != im6s_get_mode(inm, ims, 1)) {
2803 CTR1(KTR_MLD, "%s: skip non-in-mode",
2807 src = ims->im6s_addr;
2808 retval = SYSCTL_OUT(req, &src,
2809 sizeof(struct in6_addr));
2814 IN6_MULTI_LIST_UNLOCK();
2823 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2826 in6m_mode_str(const int mode)
2829 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2830 return (in6m_modestrs[mode]);
2834 static const char *in6m_statestrs[] = {
2847 in6m_state_str(const int state)
2850 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2851 return (in6m_statestrs[state]);
2856 * Dump an in6_multi structure to the console.
2859 in6m_print(const struct in6_multi *inm)
2862 char ip6tbuf[INET6_ADDRSTRLEN];
2864 if ((ktr_mask & KTR_MLD) == 0)
2867 printf("%s: --- begin in6m %p ---\n", __func__, inm);
2868 printf("addr %s ifp %p(%s) ifma %p\n",
2869 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2871 if_name(inm->in6m_ifp),
2873 printf("timer %u state %s refcount %u scq.len %u\n",
2875 in6m_state_str(inm->in6m_state),
2877 mbufq_len(&inm->in6m_scq));
2878 printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2883 for (t = 0; t < 2; t++) {
2884 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2885 in6m_mode_str(inm->in6m_st[t].iss_fmode),
2886 inm->in6m_st[t].iss_asm,
2887 inm->in6m_st[t].iss_ex,
2888 inm->in6m_st[t].iss_in,
2889 inm->in6m_st[t].iss_rec);
2891 printf("%s: --- end in6m %p ---\n", __func__, inm);
2897 in6m_print(const struct in6_multi *inm)