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/gtaskqueue.h>
45 #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>
57 #include <net/if_var.h>
58 #include <net/if_dl.h>
59 #include <net/route.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");
126 static void im6f_commit(struct in6_mfilter *);
127 static int im6f_get_source(struct in6_mfilter *imf,
128 const struct sockaddr_in6 *psin,
129 struct in6_msource **);
130 static struct in6_msource *
131 im6f_graft(struct in6_mfilter *, const uint8_t,
132 const struct sockaddr_in6 *);
133 static void im6f_leave(struct in6_mfilter *);
134 static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
135 static void im6f_purge(struct in6_mfilter *);
136 static void im6f_rollback(struct in6_mfilter *);
137 static void im6f_reap(struct in6_mfilter *);
138 static struct in6_mfilter *
139 im6o_match_group(const struct ip6_moptions *,
140 const struct ifnet *, const struct sockaddr *);
141 static struct in6_msource *
142 im6o_match_source(struct in6_mfilter *, const struct sockaddr *);
143 static void im6s_merge(struct ip6_msource *ims,
144 const struct in6_msource *lims, const int rollback);
145 static int in6_getmulti(struct ifnet *, const struct in6_addr *,
146 struct in6_multi **);
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, CTLFLAG_RW, 0,
174 static u_long in6_mcast_maxgrpsrc = IPV6_MAX_GROUP_SRC_FILTER;
175 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxgrpsrc,
176 CTLFLAG_RWTUN, &in6_mcast_maxgrpsrc, 0,
177 "Max source filters per group");
179 static u_long in6_mcast_maxsocksrc = IPV6_MAX_SOCK_SRC_FILTER;
180 SYSCTL_ULONG(_net_inet6_ip6_mcast, OID_AUTO, maxsocksrc,
181 CTLFLAG_RWTUN, &in6_mcast_maxsocksrc, 0,
182 "Max source filters per socket");
184 /* TODO Virtualize this switch. */
185 int in6_mcast_loop = IPV6_DEFAULT_MULTICAST_LOOP;
186 SYSCTL_INT(_net_inet6_ip6_mcast, OID_AUTO, loop, CTLFLAG_RWTUN,
187 &in6_mcast_loop, 0, "Loopback multicast datagrams by default");
189 static SYSCTL_NODE(_net_inet6_ip6_mcast, OID_AUTO, filters,
190 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_ip6_mcast_filters,
191 "Per-interface stack-wide source filters");
195 * Inline function which wraps assertions for a valid ifp.
196 * The ifnet layer will set the ifma's ifp pointer to NULL if the ifp
200 in6m_is_ifp_detached(const struct in6_multi *inm)
204 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
205 ifp = inm->in6m_ifma->ifma_ifp;
208 * Sanity check that network-layer notion of ifp is the
209 * same as that of link-layer.
211 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
214 return (ifp == NULL);
219 * Initialize an in6_mfilter structure to a known state at t0, t1
220 * with an empty source filter list.
223 im6f_init(struct in6_mfilter *imf, const int st0, const int st1)
225 memset(imf, 0, sizeof(struct in6_mfilter));
226 RB_INIT(&imf->im6f_sources);
227 imf->im6f_st[0] = st0;
228 imf->im6f_st[1] = st1;
232 ip6_mfilter_alloc(const int mflags, const int st0, const int st1)
234 struct in6_mfilter *imf;
236 imf = malloc(sizeof(*imf), M_IN6MFILTER, mflags);
239 im6f_init(imf, st0, st1);
245 ip6_mfilter_free(struct in6_mfilter *imf)
249 free(imf, M_IN6MFILTER);
253 * Find an IPv6 multicast group entry for this ip6_moptions instance
254 * which matches the specified group, and optionally an interface.
255 * Return its index into the array, or -1 if not found.
257 static struct in6_mfilter *
258 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
259 const struct sockaddr *group)
261 const struct sockaddr_in6 *gsin6;
262 struct in6_mfilter *imf;
263 struct in6_multi *inm;
265 gsin6 = (const struct sockaddr_in6 *)group;
267 IP6_MFILTER_FOREACH(imf, &imo->im6o_head) {
268 inm = imf->im6f_in6m;
271 if ((ifp == NULL || (inm->in6m_ifp == ifp)) &&
272 IN6_ARE_ADDR_EQUAL(&inm->in6m_addr,
273 &gsin6->sin6_addr)) {
281 * Find an IPv6 multicast source entry for this imo which matches
282 * the given group index for this socket, and source address.
284 * XXX TODO: The scope ID, if present in src, is stripped before
285 * any comparison. We SHOULD enforce scope/zone checks where the source
286 * filter entry has a link scope.
288 * NOTE: This does not check if the entry is in-mode, merely if
289 * it exists, which may not be the desired behaviour.
291 static struct in6_msource *
292 im6o_match_source(struct in6_mfilter *imf, const struct sockaddr *src)
294 struct ip6_msource find;
295 struct ip6_msource *ims;
296 const sockunion_t *psa;
298 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
300 psa = (const sockunion_t *)src;
301 find.im6s_addr = psa->sin6.sin6_addr;
302 in6_clearscope(&find.im6s_addr); /* XXX */
303 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
305 return ((struct in6_msource *)ims);
309 * Perform filtering for multicast datagrams on a socket by group and source.
311 * Returns 0 if a datagram should be allowed through, or various error codes
312 * if the socket was not a member of the group, or the source was muted, etc.
315 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
316 const struct sockaddr *group, const struct sockaddr *src)
318 struct in6_mfilter *imf;
319 struct in6_msource *ims;
322 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
324 imf = im6o_match_group(imo, ifp, group);
326 return (MCAST_NOTGMEMBER);
329 * Check if the source was included in an (S,G) join.
330 * Allow reception on exclusive memberships by default,
331 * reject reception on inclusive memberships by default.
332 * Exclude source only if an in-mode exclude filter exists.
333 * Include source only if an in-mode include filter exists.
334 * NOTE: We are comparing group state here at MLD t1 (now)
335 * with socket-layer t0 (since last downcall).
337 mode = imf->im6f_st[1];
338 ims = im6o_match_source(imf, src);
340 if ((ims == NULL && mode == MCAST_INCLUDE) ||
341 (ims != NULL && ims->im6sl_st[0] != mode))
342 return (MCAST_NOTSMEMBER);
348 * Find and return a reference to an in6_multi record for (ifp, group),
349 * and bump its reference count.
350 * If one does not exist, try to allocate it, and update link-layer multicast
351 * filters on ifp to listen for group.
352 * Assumes the IN6_MULTI lock is held across the call.
353 * Return 0 if successful, otherwise return an appropriate error code.
356 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
357 struct in6_multi **pinm)
359 struct epoch_tracker et;
360 struct sockaddr_in6 gsin6;
361 struct ifmultiaddr *ifma;
362 struct in6_multi *inm;
368 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
369 * if_addmulti() takes this mutex itself, so we must drop and
370 * re-acquire around the call.
372 IN6_MULTI_LOCK_ASSERT();
373 IN6_MULTI_LIST_LOCK();
376 inm = in6m_lookup_locked(ifp, group);
381 * If we already joined this group, just bump the
382 * refcount and return it.
384 KASSERT(inm->in6m_refcount >= 1,
385 ("%s: bad refcount %d", __func__, inm->in6m_refcount));
386 in6m_acquire_locked(inm);
391 memset(&gsin6, 0, sizeof(gsin6));
392 gsin6.sin6_family = AF_INET6;
393 gsin6.sin6_len = sizeof(struct sockaddr_in6);
394 gsin6.sin6_addr = *group;
397 * Check if a link-layer group is already associated
398 * with this network-layer group on the given ifnet.
400 IN6_MULTI_LIST_UNLOCK();
401 IF_ADDR_WUNLOCK(ifp);
402 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
405 IN6_MULTI_LIST_LOCK();
409 * If something other than netinet6 is occupying the link-layer
410 * group, print a meaningful error message and back out of
412 * Otherwise, bump the refcount on the existing network-layer
413 * group association and return it.
415 if (ifma->ifma_protospec != NULL) {
416 inm = (struct in6_multi *)ifma->ifma_protospec;
418 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
420 KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
421 ("%s: ifma not AF_INET6", __func__));
422 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
423 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
424 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
425 panic("%s: ifma %p is inconsistent with %p (%p)",
426 __func__, ifma, inm, group);
428 in6m_acquire_locked(inm);
433 IF_ADDR_WLOCK_ASSERT(ifp);
436 * A new in6_multi record is needed; allocate and initialize it.
437 * We DO NOT perform an MLD join as the in6_ layer may need to
438 * push an initial source list down to MLD to support SSM.
440 * The initial source filter state is INCLUDE, {} as per the RFC.
441 * Pending state-changes per group are subject to a bounds check.
443 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
445 IN6_MULTI_LIST_UNLOCK();
446 IF_ADDR_WUNLOCK(ifp);
447 if_delmulti_ifma(ifma);
450 inm->in6m_addr = *group;
452 inm->in6m_mli = MLD_IFINFO(ifp);
453 inm->in6m_ifma = ifma;
454 inm->in6m_refcount = 1;
455 inm->in6m_state = MLD_NOT_MEMBER;
456 mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
458 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
459 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
460 RB_INIT(&inm->in6m_srcs);
462 ifma->ifma_protospec = inm;
466 IN6_MULTI_LIST_UNLOCK();
467 IF_ADDR_WUNLOCK(ifp);
472 * Drop a reference to an in6_multi record.
474 * If the refcount drops to 0, free the in6_multi record and
475 * delete the underlying link-layer membership.
478 in6m_release(struct in6_multi *inm)
480 struct ifmultiaddr *ifma;
483 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
485 MPASS(inm->in6m_refcount == 0);
486 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
488 ifma = inm->in6m_ifma;
490 MPASS(ifma->ifma_llifma == NULL);
492 /* XXX this access is not covered by IF_ADDR_LOCK */
493 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
494 KASSERT(ifma->ifma_protospec == NULL,
495 ("%s: ifma_protospec != NULL", __func__));
497 ifp = ifma->ifma_ifp;
500 CURVNET_SET(ifp->if_vnet);
502 free(inm, M_IP6MADDR);
503 if_delmulti_ifma_flags(ifma, 1);
508 free(inm, M_IP6MADDR);
509 if_delmulti_ifma_flags(ifma, 1);
513 static struct grouptask free_gtask;
514 static struct in6_multi_head in6m_free_list;
515 static void in6m_release_task(void *arg __unused);
516 static void in6m_init(void)
518 SLIST_INIT(&in6m_free_list);
519 taskqgroup_config_gtask_init(NULL, &free_gtask, in6m_release_task, "in6m release task");
522 #ifdef EARLY_AP_STARTUP
523 SYSINIT(in6m_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
526 SYSINIT(in6m_init, SI_SUB_ROOT_CONF - 1, SI_ORDER_SECOND,
532 in6m_release_list_deferred(struct in6_multi_head *inmh)
534 if (SLIST_EMPTY(inmh))
536 mtx_lock(&in6_multi_free_mtx);
537 SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
538 mtx_unlock(&in6_multi_free_mtx);
539 GROUPTASK_ENQUEUE(&free_gtask);
543 in6m_release_wait(void)
546 /* Wait for all jobs to complete. */
547 gtaskqueue_drain_all(free_gtask.gt_taskqueue);
551 in6m_disconnect_locked(struct in6_multi_head *inmh, struct in6_multi *inm)
555 struct in6_ifaddr *ifa6;
556 struct in6_multi_mship *imm, *imm_tmp;
557 struct ifmultiaddr *ifma, *ll_ifma;
559 IN6_MULTI_LIST_LOCK_ASSERT();
563 return; /* already called */
565 inm->in6m_ifp = NULL;
566 IF_ADDR_WLOCK_ASSERT(ifp);
567 ifma = inm->in6m_ifma;
572 if (ifma->ifma_flags & IFMA_F_ENQUEUED) {
573 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
574 ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
576 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
577 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
578 MPASS(ifma != ll_ifma);
579 ifma->ifma_llifma = NULL;
580 MPASS(ll_ifma->ifma_llifma == NULL);
581 MPASS(ll_ifma->ifma_ifp == ifp);
582 if (--ll_ifma->ifma_refcount == 0) {
583 if (ll_ifma->ifma_flags & IFMA_F_ENQUEUED) {
584 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
585 ll_ifma->ifma_flags &= ~IFMA_F_ENQUEUED;
587 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
588 if_freemulti(ll_ifma);
591 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
592 if (ifa->ifa_addr->sa_family != AF_INET6)
595 LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
596 i6mm_chain, imm_tmp) {
597 if (inm == imm->i6mm_maddr) {
598 LIST_REMOVE(imm, i6mm_chain);
599 free(imm, M_IP6MADDR);
600 in6m_rele_locked(inmh, inm);
607 in6m_release_task(void *arg __unused)
609 struct in6_multi_head in6m_free_tmp;
610 struct in6_multi *inm, *tinm;
612 SLIST_INIT(&in6m_free_tmp);
613 mtx_lock(&in6_multi_free_mtx);
614 SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
615 mtx_unlock(&in6_multi_free_mtx);
617 SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
618 SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
625 * Clear recorded source entries for a group.
626 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
627 * FIXME: Should reap.
630 in6m_clear_recorded(struct in6_multi *inm)
632 struct ip6_msource *ims;
634 IN6_MULTI_LIST_LOCK_ASSERT();
636 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
639 --inm->in6m_st[1].iss_rec;
642 KASSERT(inm->in6m_st[1].iss_rec == 0,
643 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
647 * Record a source as pending for a Source-Group MLDv2 query.
648 * This lives here as it modifies the shared tree.
650 * inm is the group descriptor.
651 * naddr is the address of the source to record in network-byte order.
653 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
654 * lazy-allocate a source node in response to an SG query.
655 * Otherwise, no allocation is performed. This saves some memory
656 * with the trade-off that the source will not be reported to the
657 * router if joined in the window between the query response and
658 * the group actually being joined on the local host.
660 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
661 * This turns off the allocation of a recorded source entry if
662 * the group has not been joined.
664 * Return 0 if the source didn't exist or was already marked as recorded.
665 * Return 1 if the source was marked as recorded by this function.
666 * Return <0 if any error occurred (negated errno code).
669 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
671 struct ip6_msource find;
672 struct ip6_msource *ims, *nims;
674 IN6_MULTI_LIST_LOCK_ASSERT();
676 find.im6s_addr = *addr;
677 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
678 if (ims && ims->im6s_stp)
681 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
683 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
687 nims->im6s_addr = find.im6s_addr;
688 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
694 * Mark the source as recorded and update the recorded
698 ++inm->in6m_st[1].iss_rec;
704 * Return a pointer to an in6_msource owned by an in6_mfilter,
705 * given its source address.
706 * Lazy-allocate if needed. If this is a new entry its filter state is
709 * imf is the filter set being modified.
710 * addr is the source address.
712 * SMPng: May be called with locks held; malloc must not block.
715 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
716 struct in6_msource **plims)
718 struct ip6_msource find;
719 struct ip6_msource *ims, *nims;
720 struct in6_msource *lims;
727 find.im6s_addr = psin->sin6_addr;
728 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
729 lims = (struct in6_msource *)ims;
731 if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
733 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
737 lims = (struct in6_msource *)nims;
738 lims->im6s_addr = find.im6s_addr;
739 lims->im6sl_st[0] = MCAST_UNDEFINED;
740 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
750 * Graft a source entry into an existing socket-layer filter set,
751 * maintaining any required invariants and checking allocations.
753 * The source is marked as being in the new filter mode at t1.
755 * Return the pointer to the new node, otherwise return NULL.
757 static struct in6_msource *
758 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
759 const struct sockaddr_in6 *psin)
761 struct ip6_msource *nims;
762 struct in6_msource *lims;
764 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
768 lims = (struct in6_msource *)nims;
769 lims->im6s_addr = psin->sin6_addr;
770 lims->im6sl_st[0] = MCAST_UNDEFINED;
771 lims->im6sl_st[1] = st1;
772 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
779 * Prune a source entry from an existing socket-layer filter set,
780 * maintaining any required invariants and checking allocations.
782 * The source is marked as being left at t1, it is not freed.
784 * Return 0 if no error occurred, otherwise return an errno value.
787 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
789 struct ip6_msource find;
790 struct ip6_msource *ims;
791 struct in6_msource *lims;
793 find.im6s_addr = psin->sin6_addr;
794 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
797 lims = (struct in6_msource *)ims;
798 lims->im6sl_st[1] = MCAST_UNDEFINED;
803 * Revert socket-layer filter set deltas at t1 to t0 state.
806 im6f_rollback(struct in6_mfilter *imf)
808 struct ip6_msource *ims, *tims;
809 struct in6_msource *lims;
811 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
812 lims = (struct in6_msource *)ims;
813 if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
814 /* no change at t1 */
816 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
817 /* revert change to existing source at t1 */
818 lims->im6sl_st[1] = lims->im6sl_st[0];
820 /* revert source added t1 */
821 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
822 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
823 free(ims, M_IN6MFILTER);
827 imf->im6f_st[1] = imf->im6f_st[0];
831 * Mark socket-layer filter set as INCLUDE {} at t1.
834 im6f_leave(struct in6_mfilter *imf)
836 struct ip6_msource *ims;
837 struct in6_msource *lims;
839 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
840 lims = (struct in6_msource *)ims;
841 lims->im6sl_st[1] = MCAST_UNDEFINED;
843 imf->im6f_st[1] = MCAST_INCLUDE;
847 * Mark socket-layer filter set deltas as committed.
850 im6f_commit(struct in6_mfilter *imf)
852 struct ip6_msource *ims;
853 struct in6_msource *lims;
855 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
856 lims = (struct in6_msource *)ims;
857 lims->im6sl_st[0] = lims->im6sl_st[1];
859 imf->im6f_st[0] = imf->im6f_st[1];
863 * Reap unreferenced sources from socket-layer filter set.
866 im6f_reap(struct in6_mfilter *imf)
868 struct ip6_msource *ims, *tims;
869 struct in6_msource *lims;
871 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
872 lims = (struct in6_msource *)ims;
873 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
874 (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
875 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
876 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
877 free(ims, M_IN6MFILTER);
884 * Purge socket-layer filter set.
887 im6f_purge(struct in6_mfilter *imf)
889 struct ip6_msource *ims, *tims;
891 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
892 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
893 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
894 free(ims, M_IN6MFILTER);
897 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
898 KASSERT(RB_EMPTY(&imf->im6f_sources),
899 ("%s: im6f_sources not empty", __func__));
903 * Look up a source filter entry for a multicast group.
905 * inm is the group descriptor to work with.
906 * addr is the IPv6 address to look up.
907 * noalloc may be non-zero to suppress allocation of sources.
908 * *pims will be set to the address of the retrieved or allocated source.
910 * SMPng: NOTE: may be called with locks held.
911 * Return 0 if successful, otherwise return a non-zero error code.
914 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
915 const int noalloc, struct ip6_msource **pims)
917 struct ip6_msource find;
918 struct ip6_msource *ims, *nims;
920 char ip6tbuf[INET6_ADDRSTRLEN];
923 find.im6s_addr = *addr;
924 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
925 if (ims == NULL && !noalloc) {
926 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
928 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
932 nims->im6s_addr = *addr;
933 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
936 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
937 ip6_sprintf(ip6tbuf, addr), ims);
945 * Merge socket-layer source into MLD-layer source.
946 * If rollback is non-zero, perform the inverse of the merge.
949 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
952 int n = rollback ? -1 : 1;
954 char ip6tbuf[INET6_ADDRSTRLEN];
956 ip6_sprintf(ip6tbuf, &lims->im6s_addr);
959 if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
960 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
961 ims->im6s_st[1].ex -= n;
962 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
963 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
964 ims->im6s_st[1].in -= n;
967 if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
968 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
969 ims->im6s_st[1].ex += n;
970 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
971 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
972 ims->im6s_st[1].in += n;
977 * Atomically update the global in6_multi state, when a membership's
978 * filter list is being updated in any way.
980 * imf is the per-inpcb-membership group filter pointer.
981 * A fake imf may be passed for in-kernel consumers.
983 * XXX This is a candidate for a set-symmetric-difference style loop
984 * which would eliminate the repeated lookup from root of ims nodes,
985 * as they share the same key space.
987 * If any error occurred this function will back out of refcounts
988 * and return a non-zero value.
991 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
993 struct ip6_msource *ims, *nims;
994 struct in6_msource *lims;
1001 IN6_MULTI_LIST_LOCK_ASSERT();
1004 * Update the source filters first, as this may fail.
1005 * Maintain count of in-mode filters at t0, t1. These are
1006 * used to work out if we transition into ASM mode or not.
1007 * Maintain a count of source filters whose state was
1008 * actually modified by this operation.
1010 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1011 lims = (struct in6_msource *)ims;
1012 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1013 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1014 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1015 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1019 im6s_merge(nims, lims, 0);
1022 struct ip6_msource *bims;
1024 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1025 lims = (struct in6_msource *)ims;
1026 if (lims->im6sl_st[0] == lims->im6sl_st[1])
1028 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1031 im6s_merge(bims, lims, 1);
1036 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1037 __func__, nsrc0, nsrc1);
1039 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1040 if (imf->im6f_st[0] == imf->im6f_st[1] &&
1041 imf->im6f_st[1] == MCAST_INCLUDE) {
1043 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1044 --inm->in6m_st[1].iss_in;
1048 /* Handle filter mode transition on socket. */
1049 if (imf->im6f_st[0] != imf->im6f_st[1]) {
1050 CTR3(KTR_MLD, "%s: imf transition %d to %d",
1051 __func__, imf->im6f_st[0], imf->im6f_st[1]);
1053 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1054 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1055 --inm->in6m_st[1].iss_ex;
1056 } else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1057 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1058 --inm->in6m_st[1].iss_in;
1061 if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1062 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1063 inm->in6m_st[1].iss_ex++;
1064 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1065 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1066 inm->in6m_st[1].iss_in++;
1071 * Track inm filter state in terms of listener counts.
1072 * If there are any exclusive listeners, stack-wide
1073 * membership is exclusive.
1074 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1075 * If no listeners remain, state is undefined at t1,
1076 * and the MLD lifecycle for this group should finish.
1078 if (inm->in6m_st[1].iss_ex > 0) {
1079 CTR1(KTR_MLD, "%s: transition to EX", __func__);
1080 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1081 } else if (inm->in6m_st[1].iss_in > 0) {
1082 CTR1(KTR_MLD, "%s: transition to IN", __func__);
1083 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1085 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1086 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1089 /* Decrement ASM listener count on transition out of ASM mode. */
1090 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1091 if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1092 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1093 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1094 --inm->in6m_st[1].iss_asm;
1098 /* Increment ASM listener count on transition to ASM mode. */
1099 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1100 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1101 inm->in6m_st[1].iss_asm++;
1104 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1109 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1116 * Mark an in6_multi's filter set deltas as committed.
1117 * Called by MLD after a state change has been enqueued.
1120 in6m_commit(struct in6_multi *inm)
1122 struct ip6_msource *ims;
1124 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1125 CTR1(KTR_MLD, "%s: pre commit:", __func__);
1128 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1129 ims->im6s_st[0] = ims->im6s_st[1];
1131 inm->in6m_st[0] = inm->in6m_st[1];
1135 * Reap unreferenced nodes from an in6_multi's filter set.
1138 in6m_reap(struct in6_multi *inm)
1140 struct ip6_msource *ims, *tims;
1142 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1143 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1144 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1147 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1148 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1149 free(ims, M_IP6MSOURCE);
1155 * Purge all source nodes from an in6_multi's filter set.
1158 in6m_purge(struct in6_multi *inm)
1160 struct ip6_msource *ims, *tims;
1162 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1163 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1164 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1165 free(ims, M_IP6MSOURCE);
1168 /* Free state-change requests that might be queued. */
1169 mbufq_drain(&inm->in6m_scq);
1173 * Join a multicast address w/o sources.
1174 * KAME compatibility entry point.
1176 * SMPng: Assume no mc locks held by caller.
1179 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1180 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1186 error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1192 * Join a multicast group; real entry point.
1194 * Only preserves atomicity at inm level.
1195 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1197 * If the MLD downcall fails, the group is not joined, and an error
1201 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1202 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1205 struct in6_multi_head inmh;
1206 struct in6_mfilter timf;
1207 struct in6_multi *inm;
1208 struct ifmultiaddr *ifma;
1211 char ip6tbuf[INET6_ADDRSTRLEN];
1215 * Sanity: Check scope zone ID was set for ifp, if and
1216 * only if group is scoped to an interface.
1218 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1219 ("%s: not a multicast address", __func__));
1220 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1221 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1222 KASSERT(mcaddr->s6_addr16[1] != 0,
1223 ("%s: scope zone ID not set", __func__));
1226 IN6_MULTI_LOCK_ASSERT();
1227 IN6_MULTI_LIST_UNLOCK_ASSERT();
1229 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1230 ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1236 * If no imf was specified (i.e. kernel consumer),
1237 * fake one up and assume it is an ASM join.
1240 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1243 error = in6_getmulti(ifp, mcaddr, &inm);
1245 CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1249 IN6_MULTI_LIST_LOCK();
1250 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1251 error = in6m_merge(inm, imf);
1253 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1254 goto out_in6m_release;
1257 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1258 error = mld_change_state(inm, delay);
1260 CTR1(KTR_MLD, "%s: failed to update source", __func__);
1261 goto out_in6m_release;
1267 struct epoch_tracker et;
1269 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1270 NET_EPOCH_ENTER(et);
1271 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1272 if (ifma->ifma_protospec == inm) {
1273 ifma->ifma_protospec = NULL;
1277 in6m_disconnect_locked(&inmh, inm);
1278 in6m_rele_locked(&inmh, inm);
1283 IN6_MULTI_LIST_UNLOCK();
1284 in6m_release_list_deferred(&inmh);
1289 * Leave a multicast group; unlocked entry point.
1292 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1297 error = in6_leavegroup_locked(inm, imf);
1303 * Leave a multicast group; real entry point.
1304 * All source filters will be expunged.
1306 * Only preserves atomicity at inm level.
1308 * Holding the write lock for the INP which contains imf
1309 * is highly advisable. We can't assert for it as imf does not
1310 * contain a back-pointer to the owning inp.
1312 * Note: This is not the same as in6m_release(*) as this function also
1313 * makes a state change downcall into MLD.
1316 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1318 struct in6_multi_head inmh;
1319 struct in6_mfilter timf;
1323 char ip6tbuf[INET6_ADDRSTRLEN];
1328 IN6_MULTI_LOCK_ASSERT();
1330 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1331 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1332 (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1336 * If no imf was specified (i.e. kernel consumer),
1337 * fake one up and assume it is an ASM join.
1340 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1345 * Begin state merge transaction at MLD layer.
1347 * As this particular invocation should not cause any memory
1348 * to be allocated, and there is no opportunity to roll back
1349 * the transaction, it MUST NOT fail.
1352 ifp = inm->in6m_ifp;
1353 IN6_MULTI_LIST_LOCK();
1354 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1355 error = in6m_merge(inm, imf);
1356 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1358 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1361 error = mld_change_state(inm, 0);
1363 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1365 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1370 if (inm->in6m_refcount == 1)
1371 in6m_disconnect_locked(&inmh, inm);
1372 in6m_rele_locked(&inmh, inm);
1374 IF_ADDR_WUNLOCK(ifp);
1375 IN6_MULTI_LIST_UNLOCK();
1376 in6m_release_list_deferred(&inmh);
1382 * Block or unblock an ASM multicast source on an inpcb.
1383 * This implements the delta-based API described in RFC 3678.
1385 * The delta-based API applies only to exclusive-mode memberships.
1386 * An MLD downcall will be performed.
1388 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1390 * Return 0 if successful, otherwise return an appropriate error code.
1393 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1395 struct group_source_req gsr;
1396 sockunion_t *gsa, *ssa;
1398 struct in6_mfilter *imf;
1399 struct ip6_moptions *imo;
1400 struct in6_msource *ims;
1401 struct in6_multi *inm;
1405 char ip6tbuf[INET6_ADDRSTRLEN];
1412 memset(&gsr, 0, sizeof(struct group_source_req));
1413 gsa = (sockunion_t *)&gsr.gsr_group;
1414 ssa = (sockunion_t *)&gsr.gsr_source;
1416 switch (sopt->sopt_name) {
1417 case MCAST_BLOCK_SOURCE:
1418 case MCAST_UNBLOCK_SOURCE:
1419 error = sooptcopyin(sopt, &gsr,
1420 sizeof(struct group_source_req),
1421 sizeof(struct group_source_req));
1425 if (gsa->sin6.sin6_family != AF_INET6 ||
1426 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1429 if (ssa->sin6.sin6_family != AF_INET6 ||
1430 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1433 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1434 return (EADDRNOTAVAIL);
1436 ifp = ifnet_byindex(gsr.gsr_interface);
1438 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1443 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1444 __func__, sopt->sopt_name);
1445 return (EOPNOTSUPP);
1449 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1452 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1455 * Check if we are actually a member of this group.
1457 imo = in6p_findmoptions(inp);
1458 imf = im6o_match_group(imo, ifp, &gsa->sa);
1460 error = EADDRNOTAVAIL;
1461 goto out_in6p_locked;
1463 inm = imf->im6f_in6m;
1466 * Attempting to use the delta-based API on an
1467 * non exclusive-mode membership is an error.
1469 fmode = imf->im6f_st[0];
1470 if (fmode != MCAST_EXCLUDE) {
1472 goto out_in6p_locked;
1476 * Deal with error cases up-front:
1477 * Asked to block, but already blocked; or
1478 * Asked to unblock, but nothing to unblock.
1479 * If adding a new block entry, allocate it.
1481 ims = im6o_match_source(imf, &ssa->sa);
1482 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1483 CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1484 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1485 doblock ? "" : "not ");
1486 error = EADDRNOTAVAIL;
1487 goto out_in6p_locked;
1490 INP_WLOCK_ASSERT(inp);
1493 * Begin state merge transaction at socket layer.
1496 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1497 ims = im6f_graft(imf, fmode, &ssa->sin6);
1501 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1502 error = im6f_prune(imf, &ssa->sin6);
1506 CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1507 goto out_im6f_rollback;
1511 * Begin state merge transaction at MLD layer.
1513 IN6_MULTI_LIST_LOCK();
1514 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1515 error = in6m_merge(inm, imf);
1517 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1519 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1520 error = mld_change_state(inm, 0);
1522 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1525 IN6_MULTI_LIST_UNLOCK();
1541 * Given an inpcb, return its multicast options structure pointer. Accepts
1542 * an unlocked inpcb pointer, but will return it locked. May sleep.
1544 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1545 * SMPng: NOTE: Returns with the INP write lock held.
1547 static struct ip6_moptions *
1548 in6p_findmoptions(struct inpcb *inp)
1550 struct ip6_moptions *imo;
1553 if (inp->in6p_moptions != NULL)
1554 return (inp->in6p_moptions);
1558 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1560 imo->im6o_multicast_ifp = NULL;
1561 imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1562 imo->im6o_multicast_loop = in6_mcast_loop;
1563 STAILQ_INIT(&imo->im6o_head);
1566 if (inp->in6p_moptions != NULL) {
1567 free(imo, M_IP6MOPTS);
1568 return (inp->in6p_moptions);
1570 inp->in6p_moptions = imo;
1575 * Discard the IPv6 multicast options (and source filters).
1577 * SMPng: NOTE: assumes INP write lock is held.
1579 * XXX can all be safely deferred to epoch_call
1584 inp_gcmoptions(struct ip6_moptions *imo)
1586 struct in6_mfilter *imf;
1587 struct in6_multi *inm;
1590 while ((imf = ip6_mfilter_first(&imo->im6o_head)) != NULL) {
1591 ip6_mfilter_remove(&imo->im6o_head, imf);
1594 if ((inm = imf->im6f_in6m) != NULL) {
1595 if ((ifp = inm->in6m_ifp) != NULL) {
1596 CURVNET_SET(ifp->if_vnet);
1597 (void)in6_leavegroup(inm, imf);
1600 (void)in6_leavegroup(inm, imf);
1603 ip6_mfilter_free(imf);
1605 free(imo, M_IP6MOPTS);
1609 ip6_freemoptions(struct ip6_moptions *imo)
1613 inp_gcmoptions(imo);
1617 * Atomically get source filters on a socket for an IPv6 multicast group.
1618 * Called with INP lock held; returns with lock released.
1621 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1623 struct __msfilterreq msfr;
1626 struct ip6_moptions *imo;
1627 struct in6_mfilter *imf;
1628 struct ip6_msource *ims;
1629 struct in6_msource *lims;
1630 struct sockaddr_in6 *psin;
1631 struct sockaddr_storage *ptss;
1632 struct sockaddr_storage *tss;
1634 size_t nsrcs, ncsrcs;
1636 INP_WLOCK_ASSERT(inp);
1638 imo = inp->in6p_moptions;
1639 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1643 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1644 sizeof(struct __msfilterreq));
1648 if (msfr.msfr_group.ss_family != AF_INET6 ||
1649 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1652 gsa = (sockunion_t *)&msfr.msfr_group;
1653 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1656 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1657 return (EADDRNOTAVAIL);
1658 ifp = ifnet_byindex(msfr.msfr_ifindex);
1660 return (EADDRNOTAVAIL);
1661 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1666 * Lookup group on the socket.
1668 imf = im6o_match_group(imo, ifp, &gsa->sa);
1671 return (EADDRNOTAVAIL);
1675 * Ignore memberships which are in limbo.
1677 if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1681 msfr.msfr_fmode = imf->im6f_st[1];
1684 * If the user specified a buffer, copy out the source filter
1685 * entries to userland gracefully.
1686 * We only copy out the number of entries which userland
1687 * has asked for, but we always tell userland how big the
1688 * buffer really needs to be.
1690 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1691 msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1693 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1694 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1695 M_TEMP, M_NOWAIT | M_ZERO);
1703 * Count number of sources in-mode at t0.
1704 * If buffer space exists and remains, copy out source entries.
1706 nsrcs = msfr.msfr_nsrcs;
1709 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1710 lims = (struct in6_msource *)ims;
1711 if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1712 lims->im6sl_st[0] != imf->im6f_st[0])
1715 if (tss != NULL && nsrcs > 0) {
1716 psin = (struct sockaddr_in6 *)ptss;
1717 psin->sin6_family = AF_INET6;
1718 psin->sin6_len = sizeof(struct sockaddr_in6);
1719 psin->sin6_addr = lims->im6s_addr;
1720 psin->sin6_port = 0;
1729 error = copyout(tss, msfr.msfr_srcs,
1730 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1736 msfr.msfr_nsrcs = ncsrcs;
1737 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1743 * Return the IP multicast options in response to user getsockopt().
1746 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1748 struct ip6_moptions *im6o;
1753 im6o = inp->in6p_moptions;
1755 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1756 * or is a divert socket, reject it.
1758 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1759 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1760 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1762 return (EOPNOTSUPP);
1766 switch (sopt->sopt_name) {
1767 case IPV6_MULTICAST_IF:
1768 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1771 optval = im6o->im6o_multicast_ifp->if_index;
1774 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1777 case IPV6_MULTICAST_HOPS:
1779 optval = V_ip6_defmcasthlim;
1781 optval = im6o->im6o_multicast_hlim;
1783 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1786 case IPV6_MULTICAST_LOOP:
1788 optval = in6_mcast_loop; /* XXX VIMAGE */
1790 optval = im6o->im6o_multicast_loop;
1792 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1797 error = EADDRNOTAVAIL;
1800 error = in6p_get_source_filters(inp, sopt);
1806 error = ENOPROTOOPT;
1810 INP_UNLOCK_ASSERT(inp);
1816 * Look up the ifnet to use for a multicast group membership,
1817 * given the address of an IPv6 group.
1819 * This routine exists to support legacy IPv6 multicast applications.
1821 * If inp is non-NULL, use this socket's current FIB number for any
1822 * required FIB lookup. Look up the group address in the unicast FIB,
1823 * and use its ifp; usually, this points to the default next-hop.
1824 * If the FIB lookup fails, return NULL.
1826 * FUTURE: Support multiple forwarding tables for IPv6.
1828 * Returns NULL if no ifp could be found.
1830 static struct ifnet *
1831 in6p_lookup_mcast_ifp(const struct inpcb *inp,
1832 const struct sockaddr_in6 *gsin6)
1834 struct nhop6_basic nh6;
1835 struct in6_addr dst;
1839 KASSERT(inp->inp_vflag & INP_IPV6,
1840 ("%s: not INP_IPV6 inpcb", __func__));
1841 KASSERT(gsin6->sin6_family == AF_INET6,
1842 ("%s: not AF_INET6 group", __func__));
1844 in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1845 fibnum = inp ? inp->inp_inc.inc_fibnum : RT_DEFAULT_FIB;
1846 if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6) != 0)
1849 return (nh6.nh_ifp);
1853 * Join an IPv6 multicast group, possibly with a source.
1855 * FIXME: The KAME use of the unspecified address (::)
1856 * to join *all* multicast groups is currently unsupported.
1859 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1861 struct in6_multi_head inmh;
1862 struct group_source_req gsr;
1863 sockunion_t *gsa, *ssa;
1865 struct in6_mfilter *imf;
1866 struct ip6_moptions *imo;
1867 struct in6_multi *inm;
1868 struct in6_msource *lims;
1876 memset(&gsr, 0, sizeof(struct group_source_req));
1877 gsa = (sockunion_t *)&gsr.gsr_group;
1878 gsa->ss.ss_family = AF_UNSPEC;
1879 ssa = (sockunion_t *)&gsr.gsr_source;
1880 ssa->ss.ss_family = AF_UNSPEC;
1883 * Chew everything into struct group_source_req.
1884 * Overwrite the port field if present, as the sockaddr
1885 * being copied in may be matched with a binary comparison.
1886 * Ignore passed-in scope ID.
1888 switch (sopt->sopt_name) {
1889 case IPV6_JOIN_GROUP: {
1890 struct ipv6_mreq mreq;
1892 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1893 sizeof(struct ipv6_mreq));
1897 gsa->sin6.sin6_family = AF_INET6;
1898 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1899 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1901 if (mreq.ipv6mr_interface == 0) {
1902 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1904 if (V_if_index < mreq.ipv6mr_interface)
1905 return (EADDRNOTAVAIL);
1906 ifp = ifnet_byindex(mreq.ipv6mr_interface);
1908 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1909 __func__, mreq.ipv6mr_interface, ifp);
1912 case MCAST_JOIN_GROUP:
1913 case MCAST_JOIN_SOURCE_GROUP:
1914 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1915 error = sooptcopyin(sopt, &gsr,
1916 sizeof(struct group_req),
1917 sizeof(struct group_req));
1918 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1919 error = sooptcopyin(sopt, &gsr,
1920 sizeof(struct group_source_req),
1921 sizeof(struct group_source_req));
1926 if (gsa->sin6.sin6_family != AF_INET6 ||
1927 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1930 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1931 if (ssa->sin6.sin6_family != AF_INET6 ||
1932 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1934 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1937 * TODO: Validate embedded scope ID in source
1938 * list entry against passed-in ifp, if and only
1939 * if source list filter entry is iface or node local.
1941 in6_clearscope(&ssa->sin6.sin6_addr);
1942 ssa->sin6.sin6_port = 0;
1943 ssa->sin6.sin6_scope_id = 0;
1946 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1947 return (EADDRNOTAVAIL);
1948 ifp = ifnet_byindex(gsr.gsr_interface);
1952 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1953 __func__, sopt->sopt_name);
1954 return (EOPNOTSUPP);
1958 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1961 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
1962 return (EADDRNOTAVAIL);
1964 gsa->sin6.sin6_port = 0;
1965 gsa->sin6.sin6_scope_id = 0;
1968 * Always set the scope zone ID on memberships created from userland.
1969 * Use the passed-in ifp to do this.
1970 * XXX The in6_setscope() return value is meaningless.
1971 * XXX SCOPE6_LOCK() is taken by in6_setscope().
1973 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1978 * Find the membership in the membership list.
1980 imo = in6p_findmoptions(inp);
1981 imf = im6o_match_group(imo, ifp, &gsa->sa);
1986 if (ip6_mfilter_count(&imo->im6o_head) >= IPV6_MAX_MEMBERSHIPS) {
1988 goto out_in6p_locked;
1992 inm = imf->im6f_in6m;
1994 if (ssa->ss.ss_family != AF_UNSPEC) {
1996 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
1997 * is an error. On an existing inclusive membership,
1998 * it just adds the source to the filter list.
2000 if (imf->im6f_st[1] != MCAST_INCLUDE) {
2002 goto out_in6p_locked;
2005 * Throw out duplicates.
2007 * XXX FIXME: This makes a naive assumption that
2008 * even if entries exist for *ssa in this imf,
2009 * they will be rejected as dupes, even if they
2010 * are not valid in the current mode (in-mode).
2012 * in6_msource is transactioned just as for anything
2013 * else in SSM -- but note naive use of in6m_graft()
2014 * below for allocating new filter entries.
2016 * This is only an issue if someone mixes the
2017 * full-state SSM API with the delta-based API,
2018 * which is discouraged in the relevant RFCs.
2020 lims = im6o_match_source(imf, &ssa->sa);
2021 if (lims != NULL /*&&
2022 lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2023 error = EADDRNOTAVAIL;
2024 goto out_in6p_locked;
2028 * MCAST_JOIN_GROUP alone, on any existing membership,
2029 * is rejected, to stop the same inpcb tying up
2030 * multiple refs to the in_multi.
2031 * On an existing inclusive membership, this is also
2032 * an error; if you want to change filter mode,
2033 * you must use the userland API setsourcefilter().
2034 * XXX We don't reject this for imf in UNDEFINED
2035 * state at t1, because allocation of a filter
2036 * is atomic with allocation of a membership.
2039 goto out_in6p_locked;
2044 * Begin state merge transaction at socket layer.
2046 INP_WLOCK_ASSERT(inp);
2049 * Graft new source into filter list for this inpcb's
2050 * membership of the group. The in6_multi may not have
2051 * been allocated yet if this is a new membership, however,
2052 * the in_mfilter slot will be allocated and must be initialized.
2054 * Note: Grafting of exclusive mode filters doesn't happen
2056 * XXX: Should check for non-NULL lims (node exists but may
2057 * not be in-mode) for interop with full-state API.
2059 if (ssa->ss.ss_family != AF_UNSPEC) {
2060 /* Membership starts in IN mode */
2062 CTR1(KTR_MLD, "%s: new join w/source", __func__);
2063 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_INCLUDE);
2066 goto out_in6p_locked;
2069 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2071 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2073 CTR1(KTR_MLD, "%s: merge imf state failed",
2076 goto out_in6p_locked;
2079 /* No address specified; Membership starts in EX mode */
2081 CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2082 imf = ip6_mfilter_alloc(M_NOWAIT, MCAST_UNDEFINED, MCAST_EXCLUDE);
2085 goto out_in6p_locked;
2091 * Begin state merge transaction at MLD layer.
2097 error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2098 &imf->im6f_in6m, 0);
2101 if (in_pcbrele_wlocked(inp)) {
2103 goto out_in6p_unlocked;
2106 goto out_in6p_locked;
2109 * NOTE: Refcount from in6_joingroup_locked()
2110 * is protecting membership.
2113 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2114 IN6_MULTI_LIST_LOCK();
2115 error = in6m_merge(inm, imf);
2117 CTR1(KTR_MLD, "%s: failed to merge inm state",
2119 IN6_MULTI_LIST_UNLOCK();
2122 goto out_in6p_locked;
2124 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2125 error = mld_change_state(inm, 0);
2126 IN6_MULTI_LIST_UNLOCK();
2129 CTR1(KTR_MLD, "%s: failed mld downcall",
2133 goto out_in6p_locked;
2138 ip6_mfilter_insert(&imo->im6o_head, imf);
2148 if (is_new && imf) {
2149 if (imf->im6f_in6m != NULL) {
2150 struct in6_multi_head inmh;
2153 SLIST_INSERT_HEAD(&inmh, imf->im6f_in6m, in6m_defer);
2154 in6m_release_list_deferred(&inmh);
2156 ip6_mfilter_free(imf);
2162 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2165 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2167 struct ipv6_mreq mreq;
2168 struct group_source_req gsr;
2169 sockunion_t *gsa, *ssa;
2171 struct in6_mfilter *imf;
2172 struct ip6_moptions *imo;
2173 struct in6_msource *ims;
2174 struct in6_multi *inm;
2179 char ip6tbuf[INET6_ADDRSTRLEN];
2187 memset(&gsr, 0, sizeof(struct group_source_req));
2188 gsa = (sockunion_t *)&gsr.gsr_group;
2189 gsa->ss.ss_family = AF_UNSPEC;
2190 ssa = (sockunion_t *)&gsr.gsr_source;
2191 ssa->ss.ss_family = AF_UNSPEC;
2194 * Chew everything passed in up into a struct group_source_req
2195 * as that is easier to process.
2196 * Note: Any embedded scope ID in the multicast group passed
2197 * in by userland is ignored, the interface index is the recommended
2198 * mechanism to specify an interface; see below.
2200 switch (sopt->sopt_name) {
2201 case IPV6_LEAVE_GROUP:
2202 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2203 sizeof(struct ipv6_mreq));
2206 gsa->sin6.sin6_family = AF_INET6;
2207 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2208 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2209 gsa->sin6.sin6_port = 0;
2210 gsa->sin6.sin6_scope_id = 0;
2211 ifindex = mreq.ipv6mr_interface;
2214 case MCAST_LEAVE_GROUP:
2215 case MCAST_LEAVE_SOURCE_GROUP:
2216 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2217 error = sooptcopyin(sopt, &gsr,
2218 sizeof(struct group_req),
2219 sizeof(struct group_req));
2220 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2221 error = sooptcopyin(sopt, &gsr,
2222 sizeof(struct group_source_req),
2223 sizeof(struct group_source_req));
2228 if (gsa->sin6.sin6_family != AF_INET6 ||
2229 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2231 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2232 if (ssa->sin6.sin6_family != AF_INET6 ||
2233 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2235 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2238 * TODO: Validate embedded scope ID in source
2239 * list entry against passed-in ifp, if and only
2240 * if source list filter entry is iface or node local.
2242 in6_clearscope(&ssa->sin6.sin6_addr);
2244 gsa->sin6.sin6_port = 0;
2245 gsa->sin6.sin6_scope_id = 0;
2246 ifindex = gsr.gsr_interface;
2250 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2251 __func__, sopt->sopt_name);
2252 return (EOPNOTSUPP);
2256 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2260 * Validate interface index if provided. If no interface index
2261 * was provided separately, attempt to look the membership up
2262 * from the default scope as a last resort to disambiguate
2263 * the membership we are being asked to leave.
2264 * XXX SCOPE6 lock potentially taken here.
2267 if (V_if_index < ifindex)
2268 return (EADDRNOTAVAIL);
2269 ifp = ifnet_byindex(ifindex);
2271 return (EADDRNOTAVAIL);
2272 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2274 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2276 return (EADDRNOTAVAIL);
2278 * Some badly behaved applications don't pass an ifindex
2279 * or a scope ID, which is an API violation. In this case,
2280 * perform a lookup as per a v6 join.
2282 * XXX For now, stomp on zone ID for the corner case.
2283 * This is not the 'KAME way', but we need to see the ifp
2284 * directly until such time as this implementation is
2285 * refactored, assuming the scope IDs are the way to go.
2287 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2289 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2290 "ifp for group %s.", __func__,
2291 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2292 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2294 ifp = ifnet_byindex(ifindex);
2297 return (EADDRNOTAVAIL);
2300 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2301 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2306 * Find the membership in the membership list.
2308 imo = in6p_findmoptions(inp);
2309 imf = im6o_match_group(imo, ifp, &gsa->sa);
2311 error = EADDRNOTAVAIL;
2312 goto out_in6p_locked;
2314 inm = imf->im6f_in6m;
2316 if (ssa->ss.ss_family != AF_UNSPEC)
2320 * Begin state merge transaction at socket layer.
2322 INP_WLOCK_ASSERT(inp);
2325 * If we were instructed only to leave a given source, do so.
2326 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2329 ip6_mfilter_remove(&imo->im6o_head, imf);
2332 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2333 error = EADDRNOTAVAIL;
2334 goto out_in6p_locked;
2336 ims = im6o_match_source(imf, &ssa->sa);
2338 CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2339 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2341 error = EADDRNOTAVAIL;
2342 goto out_in6p_locked;
2344 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2345 error = im6f_prune(imf, &ssa->sin6);
2347 CTR1(KTR_MLD, "%s: merge imf state failed",
2349 goto out_in6p_locked;
2354 * Begin state merge transaction at MLD layer.
2357 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2358 IN6_MULTI_LIST_LOCK();
2359 error = in6m_merge(inm, imf);
2361 CTR1(KTR_MLD, "%s: failed to merge inm state",
2363 IN6_MULTI_LIST_UNLOCK();
2366 goto out_in6p_locked;
2369 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2370 error = mld_change_state(inm, 0);
2371 IN6_MULTI_LIST_UNLOCK();
2373 CTR1(KTR_MLD, "%s: failed mld downcall",
2377 goto out_in6p_locked;
2387 if (is_final && imf) {
2389 * Give up the multicast address record to which
2390 * the membership points.
2392 (void)in6_leavegroup_locked(inm, imf);
2393 ip6_mfilter_free(imf);
2401 * Select the interface for transmitting IPv6 multicast datagrams.
2403 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2404 * may be passed to this socket option. An address of in6addr_any or an
2405 * interface index of 0 is used to remove a previous selection.
2406 * When no interface is selected, one is chosen for every send.
2409 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2412 struct ip6_moptions *imo;
2416 if (sopt->sopt_valsize != sizeof(u_int))
2419 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2422 if (V_if_index < ifindex)
2427 ifp = ifnet_byindex(ifindex);
2430 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2431 return (EADDRNOTAVAIL);
2433 imo = in6p_findmoptions(inp);
2434 imo->im6o_multicast_ifp = ifp;
2441 * Atomically set source filters on a socket for an IPv6 multicast group.
2443 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2446 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2448 struct __msfilterreq msfr;
2451 struct in6_mfilter *imf;
2452 struct ip6_moptions *imo;
2453 struct in6_multi *inm;
2456 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2457 sizeof(struct __msfilterreq));
2461 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2464 if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2465 msfr.msfr_fmode != MCAST_INCLUDE)
2468 if (msfr.msfr_group.ss_family != AF_INET6 ||
2469 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2472 gsa = (sockunion_t *)&msfr.msfr_group;
2473 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2476 gsa->sin6.sin6_port = 0; /* ignore port */
2478 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2479 return (EADDRNOTAVAIL);
2480 ifp = ifnet_byindex(msfr.msfr_ifindex);
2482 return (EADDRNOTAVAIL);
2483 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2486 * Take the INP write lock.
2487 * Check if this socket is a member of this group.
2489 imo = in6p_findmoptions(inp);
2490 imf = im6o_match_group(imo, ifp, &gsa->sa);
2492 error = EADDRNOTAVAIL;
2493 goto out_in6p_locked;
2495 inm = imf->im6f_in6m;
2498 * Begin state merge transaction at socket layer.
2500 INP_WLOCK_ASSERT(inp);
2502 imf->im6f_st[1] = msfr.msfr_fmode;
2505 * Apply any new source filters, if present.
2506 * Make a copy of the user-space source vector so
2507 * that we may copy them with a single copyin. This
2508 * allows us to deal with page faults up-front.
2510 if (msfr.msfr_nsrcs > 0) {
2511 struct in6_msource *lims;
2512 struct sockaddr_in6 *psin;
2513 struct sockaddr_storage *kss, *pkss;
2518 CTR2(KTR_MLD, "%s: loading %lu source list entries",
2519 __func__, (unsigned long)msfr.msfr_nsrcs);
2520 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2522 error = copyin(msfr.msfr_srcs, kss,
2523 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2532 * Mark all source filters as UNDEFINED at t1.
2533 * Restore new group filter mode, as im6f_leave()
2534 * will set it to INCLUDE.
2537 imf->im6f_st[1] = msfr.msfr_fmode;
2540 * Update socket layer filters at t1, lazy-allocating
2541 * new entries. This saves a bunch of memory at the
2542 * cost of one RB_FIND() per source entry; duplicate
2543 * entries in the msfr_nsrcs vector are ignored.
2544 * If we encounter an error, rollback transaction.
2546 * XXX This too could be replaced with a set-symmetric
2547 * difference like loop to avoid walking from root
2548 * every time, as the key space is common.
2550 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2551 psin = (struct sockaddr_in6 *)pkss;
2552 if (psin->sin6_family != AF_INET6) {
2553 error = EAFNOSUPPORT;
2556 if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2560 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2565 * TODO: Validate embedded scope ID in source
2566 * list entry against passed-in ifp, if and only
2567 * if source list filter entry is iface or node local.
2569 in6_clearscope(&psin->sin6_addr);
2570 error = im6f_get_source(imf, psin, &lims);
2573 lims->im6sl_st[1] = imf->im6f_st[1];
2579 goto out_im6f_rollback;
2581 INP_WLOCK_ASSERT(inp);
2582 IN6_MULTI_LIST_LOCK();
2585 * Begin state merge transaction at MLD layer.
2587 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2588 error = in6m_merge(inm, imf);
2590 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2592 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2593 error = mld_change_state(inm, 0);
2595 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2598 IN6_MULTI_LIST_UNLOCK();
2614 * Set the IP multicast options in response to user setsockopt().
2616 * Many of the socket options handled in this function duplicate the
2617 * functionality of socket options in the regular unicast API. However,
2618 * it is not possible to merge the duplicate code, because the idempotence
2619 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2620 * the effects of these options must be treated as separate and distinct.
2622 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2625 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2627 struct ip6_moptions *im6o;
2633 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2634 * or is a divert socket, reject it.
2636 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2637 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2638 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2639 return (EOPNOTSUPP);
2641 switch (sopt->sopt_name) {
2642 case IPV6_MULTICAST_IF:
2643 error = in6p_set_multicast_if(inp, sopt);
2646 case IPV6_MULTICAST_HOPS: {
2649 if (sopt->sopt_valsize != sizeof(int)) {
2653 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2656 if (hlim < -1 || hlim > 255) {
2659 } else if (hlim == -1) {
2660 hlim = V_ip6_defmcasthlim;
2662 im6o = in6p_findmoptions(inp);
2663 im6o->im6o_multicast_hlim = hlim;
2668 case IPV6_MULTICAST_LOOP: {
2672 * Set the loopback flag for outgoing multicast packets.
2673 * Must be zero or one.
2675 if (sopt->sopt_valsize != sizeof(u_int)) {
2679 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2686 im6o = in6p_findmoptions(inp);
2687 im6o->im6o_multicast_loop = loop;
2692 case IPV6_JOIN_GROUP:
2693 case MCAST_JOIN_GROUP:
2694 case MCAST_JOIN_SOURCE_GROUP:
2695 error = in6p_join_group(inp, sopt);
2698 case IPV6_LEAVE_GROUP:
2699 case MCAST_LEAVE_GROUP:
2700 case MCAST_LEAVE_SOURCE_GROUP:
2701 error = in6p_leave_group(inp, sopt);
2704 case MCAST_BLOCK_SOURCE:
2705 case MCAST_UNBLOCK_SOURCE:
2706 error = in6p_block_unblock_source(inp, sopt);
2710 error = in6p_set_source_filters(inp, sopt);
2718 INP_UNLOCK_ASSERT(inp);
2724 * Expose MLD's multicast filter mode and source list(s) to userland,
2725 * keyed by (ifindex, group).
2726 * The filter mode is written out as a uint32_t, followed by
2727 * 0..n of struct in6_addr.
2728 * For use by ifmcstat(8).
2729 * SMPng: NOTE: unlocked read of ifindex space.
2732 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2734 struct in6_addr mcaddr;
2735 struct in6_addr src;
2736 struct epoch_tracker et;
2738 struct ifmultiaddr *ifma;
2739 struct in6_multi *inm;
2740 struct ip6_msource *ims;
2744 uint32_t fmode, ifindex;
2746 char ip6tbuf[INET6_ADDRSTRLEN];
2752 if (req->newptr != NULL)
2755 /* int: ifindex + 4 * 32 bits of IPv6 address */
2760 if (ifindex <= 0 || ifindex > V_if_index) {
2761 CTR2(KTR_MLD, "%s: ifindex %u out of range",
2766 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2767 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2768 CTR2(KTR_MLD, "%s: group %s is not multicast",
2769 __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2773 ifp = ifnet_byindex(ifindex);
2775 CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2780 * Internal MLD lookups require that scope/zone ID is set.
2782 (void)in6_setscope(&mcaddr, ifp, NULL);
2784 retval = sysctl_wire_old_buffer(req,
2785 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2790 IN6_MULTI_LIST_LOCK();
2791 NET_EPOCH_ENTER(et);
2792 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2793 inm = in6m_ifmultiaddr_get_inm(ifma);
2796 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2798 fmode = inm->in6m_st[1].iss_fmode;
2799 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2802 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2803 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2805 * Only copy-out sources which are in-mode.
2807 if (fmode != im6s_get_mode(inm, ims, 1)) {
2808 CTR1(KTR_MLD, "%s: skip non-in-mode",
2812 src = ims->im6s_addr;
2813 retval = SYSCTL_OUT(req, &src,
2814 sizeof(struct in6_addr));
2821 IN6_MULTI_LIST_UNLOCK();
2829 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2832 in6m_mode_str(const int mode)
2835 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2836 return (in6m_modestrs[mode]);
2840 static const char *in6m_statestrs[] = {
2853 in6m_state_str(const int state)
2856 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2857 return (in6m_statestrs[state]);
2862 * Dump an in6_multi structure to the console.
2865 in6m_print(const struct in6_multi *inm)
2868 char ip6tbuf[INET6_ADDRSTRLEN];
2870 if ((ktr_mask & KTR_MLD) == 0)
2873 printf("%s: --- begin in6m %p ---\n", __func__, inm);
2874 printf("addr %s ifp %p(%s) ifma %p\n",
2875 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2877 if_name(inm->in6m_ifp),
2879 printf("timer %u state %s refcount %u scq.len %u\n",
2881 in6m_state_str(inm->in6m_state),
2883 mbufq_len(&inm->in6m_scq));
2884 printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2889 for (t = 0; t < 2; t++) {
2890 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2891 in6m_mode_str(inm->in6m_st[t].iss_fmode),
2892 inm->in6m_st[t].iss_asm,
2893 inm->in6m_st[t].iss_ex,
2894 inm->in6m_st[t].iss_in,
2895 inm->in6m_st[t].iss_rec);
2897 printf("%s: --- end in6m %p ---\n", __func__, inm);
2903 in6m_print(const struct in6_multi *inm)