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, INP_WLOCK, IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
106 * - The IF_ADDR_LOCK is implicitly taken by in6m_lookup() earlier, however
107 * it can be taken by code in net/if.c also.
108 * - ip6_moptions and in6_mfilter are covered by the INP_WLOCK.
110 * struct in6_multi is covered by IN6_MULTI_LOCK. There isn't strictly
111 * any need for in6_multi itself to be virtualized -- it is bound to an ifp
112 * anyway no matter what happens.
114 struct mtx in6_multi_list_mtx;
115 MTX_SYSINIT(in6_multi_mtx, &in6_multi_list_mtx, "in6_multi_list_mtx", MTX_DEF);
117 struct mtx in6_multi_free_mtx;
118 MTX_SYSINIT(in6_multi_free_mtx, &in6_multi_free_mtx, "in6_multi_free_mtx", MTX_DEF);
120 struct sx in6_multi_sx;
121 SX_SYSINIT(in6_multi_sx, &in6_multi_sx, "in6_multi_sx");
125 static void im6f_commit(struct in6_mfilter *);
126 static int im6f_get_source(struct in6_mfilter *imf,
127 const struct sockaddr_in6 *psin,
128 struct in6_msource **);
129 static struct in6_msource *
130 im6f_graft(struct in6_mfilter *, const uint8_t,
131 const struct sockaddr_in6 *);
132 static void im6f_leave(struct in6_mfilter *);
133 static int im6f_prune(struct in6_mfilter *, const struct sockaddr_in6 *);
134 static void im6f_purge(struct in6_mfilter *);
135 static void im6f_rollback(struct in6_mfilter *);
136 static void im6f_reap(struct in6_mfilter *);
137 static int im6o_grow(struct ip6_moptions *);
138 static size_t im6o_match_group(const struct ip6_moptions *,
139 const struct ifnet *, const struct sockaddr *);
140 static struct in6_msource *
141 im6o_match_source(const struct ip6_moptions *, const size_t,
142 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");
193 int ifma6_restart = 0;
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 * Resize the ip6_moptions vector to the next power-of-two minus 1.
234 * May be called with locks held; do not sleep.
237 im6o_grow(struct ip6_moptions *imo)
239 struct in6_multi **nmships;
240 struct in6_multi **omships;
241 struct in6_mfilter *nmfilters;
242 struct in6_mfilter *omfilters;
249 omships = imo->im6o_membership;
250 omfilters = imo->im6o_mfilters;
251 oldmax = imo->im6o_max_memberships;
252 newmax = ((oldmax + 1) * 2) - 1;
254 if (newmax <= IPV6_MAX_MEMBERSHIPS) {
255 nmships = (struct in6_multi **)realloc(omships,
256 sizeof(struct in6_multi *) * newmax, M_IP6MOPTS, M_NOWAIT);
257 nmfilters = (struct in6_mfilter *)realloc(omfilters,
258 sizeof(struct in6_mfilter) * newmax, M_IN6MFILTER,
260 if (nmships != NULL && nmfilters != NULL) {
261 /* Initialize newly allocated source filter heads. */
262 for (idx = oldmax; idx < newmax; idx++) {
263 im6f_init(&nmfilters[idx], MCAST_UNDEFINED,
266 imo->im6o_max_memberships = newmax;
267 imo->im6o_membership = nmships;
268 imo->im6o_mfilters = nmfilters;
272 if (nmships == NULL || nmfilters == NULL) {
274 free(nmships, M_IP6MOPTS);
275 if (nmfilters != NULL)
276 free(nmfilters, M_IN6MFILTER);
277 return (ETOOMANYREFS);
284 * Find an IPv6 multicast group entry for this ip6_moptions instance
285 * which matches the specified group, and optionally an interface.
286 * Return its index into the array, or -1 if not found.
289 im6o_match_group(const struct ip6_moptions *imo, const struct ifnet *ifp,
290 const struct sockaddr *group)
292 const struct sockaddr_in6 *gsin6;
293 struct in6_multi **pinm;
297 gsin6 = (const struct sockaddr_in6 *)group;
299 /* The im6o_membership array may be lazy allocated. */
300 if (imo->im6o_membership == NULL || imo->im6o_num_memberships == 0)
303 nmships = imo->im6o_num_memberships;
304 pinm = &imo->im6o_membership[0];
305 for (idx = 0; idx < nmships; idx++, pinm++) {
308 if ((ifp == NULL || ((*pinm)->in6m_ifp == ifp)) &&
309 IN6_ARE_ADDR_EQUAL(&(*pinm)->in6m_addr,
310 &gsin6->sin6_addr)) {
321 * Find an IPv6 multicast source entry for this imo which matches
322 * the given group index for this socket, and source address.
324 * XXX TODO: The scope ID, if present in src, is stripped before
325 * any comparison. We SHOULD enforce scope/zone checks where the source
326 * filter entry has a link scope.
328 * NOTE: This does not check if the entry is in-mode, merely if
329 * it exists, which may not be the desired behaviour.
331 static struct in6_msource *
332 im6o_match_source(const struct ip6_moptions *imo, const size_t gidx,
333 const struct sockaddr *src)
335 struct ip6_msource find;
336 struct in6_mfilter *imf;
337 struct ip6_msource *ims;
338 const sockunion_t *psa;
340 KASSERT(src->sa_family == AF_INET6, ("%s: !AF_INET6", __func__));
341 KASSERT(gidx != -1 && gidx < imo->im6o_num_memberships,
342 ("%s: invalid index %d\n", __func__, (int)gidx));
344 /* The im6o_mfilters array may be lazy allocated. */
345 if (imo->im6o_mfilters == NULL)
347 imf = &imo->im6o_mfilters[gidx];
349 psa = (const sockunion_t *)src;
350 find.im6s_addr = psa->sin6.sin6_addr;
351 in6_clearscope(&find.im6s_addr); /* XXX */
352 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
354 return ((struct in6_msource *)ims);
358 * Perform filtering for multicast datagrams on a socket by group and source.
360 * Returns 0 if a datagram should be allowed through, or various error codes
361 * if the socket was not a member of the group, or the source was muted, etc.
364 im6o_mc_filter(const struct ip6_moptions *imo, const struct ifnet *ifp,
365 const struct sockaddr *group, const struct sockaddr *src)
368 struct in6_msource *ims;
371 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
373 gidx = im6o_match_group(imo, ifp, group);
375 return (MCAST_NOTGMEMBER);
378 * Check if the source was included in an (S,G) join.
379 * Allow reception on exclusive memberships by default,
380 * reject reception on inclusive memberships by default.
381 * Exclude source only if an in-mode exclude filter exists.
382 * Include source only if an in-mode include filter exists.
383 * NOTE: We are comparing group state here at MLD t1 (now)
384 * with socket-layer t0 (since last downcall).
386 mode = imo->im6o_mfilters[gidx].im6f_st[1];
387 ims = im6o_match_source(imo, gidx, src);
389 if ((ims == NULL && mode == MCAST_INCLUDE) ||
390 (ims != NULL && ims->im6sl_st[0] != mode))
391 return (MCAST_NOTSMEMBER);
397 * Find and return a reference to an in6_multi record for (ifp, group),
398 * and bump its reference count.
399 * If one does not exist, try to allocate it, and update link-layer multicast
400 * filters on ifp to listen for group.
401 * Assumes the IN6_MULTI lock is held across the call.
402 * Return 0 if successful, otherwise return an appropriate error code.
405 in6_getmulti(struct ifnet *ifp, const struct in6_addr *group,
406 struct in6_multi **pinm)
408 struct sockaddr_in6 gsin6;
409 struct ifmultiaddr *ifma;
410 struct in6_multi *inm;
416 * XXX: Accesses to ifma_protospec must be covered by IF_ADDR_LOCK;
417 * if_addmulti() takes this mutex itself, so we must drop and
418 * re-acquire around the call.
420 IN6_MULTI_LOCK_ASSERT();
421 IN6_MULTI_LIST_LOCK();
423 inm = in6m_lookup_locked(ifp, group);
426 * If we already joined this group, just bump the
427 * refcount and return it.
429 KASSERT(inm->in6m_refcount >= 1,
430 ("%s: bad refcount %d", __func__, inm->in6m_refcount));
431 in6m_acquire_locked(inm);
436 memset(&gsin6, 0, sizeof(gsin6));
437 gsin6.sin6_family = AF_INET6;
438 gsin6.sin6_len = sizeof(struct sockaddr_in6);
439 gsin6.sin6_addr = *group;
442 * Check if a link-layer group is already associated
443 * with this network-layer group on the given ifnet.
445 IN6_MULTI_LIST_UNLOCK();
446 IF_ADDR_WUNLOCK(ifp);
447 error = if_addmulti(ifp, (struct sockaddr *)&gsin6, &ifma);
450 IN6_MULTI_LIST_LOCK();
454 * If something other than netinet6 is occupying the link-layer
455 * group, print a meaningful error message and back out of
457 * Otherwise, bump the refcount on the existing network-layer
458 * group association and return it.
460 if (ifma->ifma_protospec != NULL) {
461 inm = (struct in6_multi *)ifma->ifma_protospec;
463 KASSERT(ifma->ifma_addr != NULL, ("%s: no ifma_addr",
465 KASSERT(ifma->ifma_addr->sa_family == AF_INET6,
466 ("%s: ifma not AF_INET6", __func__));
467 KASSERT(inm != NULL, ("%s: no ifma_protospec", __func__));
468 if (inm->in6m_ifma != ifma || inm->in6m_ifp != ifp ||
469 !IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, group))
470 panic("%s: ifma %p is inconsistent with %p (%p)",
471 __func__, ifma, inm, group);
473 in6m_acquire_locked(inm);
478 IF_ADDR_WLOCK_ASSERT(ifp);
481 * A new in6_multi record is needed; allocate and initialize it.
482 * We DO NOT perform an MLD join as the in6_ layer may need to
483 * push an initial source list down to MLD to support SSM.
485 * The initial source filter state is INCLUDE, {} as per the RFC.
486 * Pending state-changes per group are subject to a bounds check.
488 inm = malloc(sizeof(*inm), M_IP6MADDR, M_NOWAIT | M_ZERO);
490 IN6_MULTI_LIST_UNLOCK();
491 IF_ADDR_WUNLOCK(ifp);
492 if_delmulti_ifma(ifma);
495 inm->in6m_addr = *group;
497 inm->in6m_mli = MLD_IFINFO(ifp);
498 inm->in6m_ifma = ifma;
499 inm->in6m_refcount = 1;
500 inm->in6m_state = MLD_NOT_MEMBER;
501 mbufq_init(&inm->in6m_scq, MLD_MAX_STATE_CHANGES);
503 inm->in6m_st[0].iss_fmode = MCAST_UNDEFINED;
504 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
505 RB_INIT(&inm->in6m_srcs);
507 ifma->ifma_protospec = inm;
511 IN6_MULTI_LIST_UNLOCK();
512 IF_ADDR_WUNLOCK(ifp);
517 * Drop a reference to an in6_multi record.
519 * If the refcount drops to 0, free the in6_multi record and
520 * delete the underlying link-layer membership.
523 in6m_release(struct in6_multi *inm)
525 struct ifmultiaddr *ifma;
528 CTR2(KTR_MLD, "%s: refcount is %d", __func__, inm->in6m_refcount);
530 MPASS(inm->in6m_refcount == 0);
531 CTR2(KTR_MLD, "%s: freeing inm %p", __func__, inm);
533 ifma = inm->in6m_ifma;
535 MPASS(ifma->ifma_llifma == NULL);
537 /* XXX this access is not covered by IF_ADDR_LOCK */
538 CTR2(KTR_MLD, "%s: purging ifma %p", __func__, ifma);
539 KASSERT(ifma->ifma_protospec == NULL,
540 ("%s: ifma_protospec != NULL", __func__));
543 CURVNET_SET(ifp->if_vnet);
545 free(inm, M_IP6MADDR);
546 if_delmulti_ifma_flags(ifma, 1);
551 free(inm, M_IP6MADDR);
552 if_delmulti_ifma_flags(ifma, 1);
556 static struct grouptask free_gtask;
557 static struct in6_multi_head in6m_free_list;
558 static void in6m_release_task(void *arg __unused);
559 static void in6m_init(void)
561 SLIST_INIT(&in6m_free_list);
562 taskqgroup_config_gtask_init(NULL, &free_gtask, in6m_release_task, "in6m release task");
565 SYSINIT(in6m_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
570 in6m_release_list_deferred(struct in6_multi_head *inmh)
572 if (SLIST_EMPTY(inmh))
574 mtx_lock(&in6_multi_free_mtx);
575 SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
576 mtx_unlock(&in6_multi_free_mtx);
577 GROUPTASK_ENQUEUE(&free_gtask);
581 in6m_disconnect(struct in6_multi *inm)
585 struct in6_ifaddr *ifa6;
586 struct in6_multi_mship *imm, *imm_tmp;
587 struct ifmultiaddr *ifma, *ll_ifma;
590 IF_ADDR_WLOCK_ASSERT(ifp);
591 ifma = inm->in6m_ifma;
594 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifmultiaddr, ifma_link);
595 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
596 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
597 MPASS(ifma != ll_ifma);
598 ifma->ifma_llifma = NULL;
599 MPASS(ll_ifma->ifma_llifma == NULL);
600 MPASS(ll_ifma->ifma_ifp == ifp);
601 if (--ll_ifma->ifma_refcount == 0) {
602 ifma6_restart = true;
603 CK_STAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifmultiaddr, ifma_link);
604 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
605 if_freemulti(ll_ifma);
608 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
609 if (ifa->ifa_addr->sa_family != AF_INET6)
612 LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
613 i6mm_chain, imm_tmp) {
614 if (inm == imm->i6mm_maddr) {
615 LIST_REMOVE(imm, i6mm_chain);
616 free(imm, M_IP6MADDR);
623 in6m_release_deferred(struct in6_multi *inm)
625 struct in6_multi_head tmp;
627 IN6_MULTI_LIST_LOCK_ASSERT();
628 KASSERT(inm->in6m_refcount > 0, ("refcount == %d inm: %p", inm->in6m_refcount, inm));
629 if (--inm->in6m_refcount == 0) {
630 in6m_disconnect(inm);
632 inm->in6m_ifma->ifma_protospec = NULL;
633 MPASS(inm->in6m_ifma->ifma_llifma == NULL);
634 SLIST_INSERT_HEAD(&tmp, inm, in6m_nrele);
635 in6m_release_list_deferred(&tmp);
640 in6m_release_task(void *arg __unused)
642 struct in6_multi_head in6m_free_tmp;
643 struct in6_multi *inm, *tinm;
645 SLIST_INIT(&in6m_free_tmp);
646 mtx_lock(&in6_multi_free_mtx);
647 SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
648 mtx_unlock(&in6_multi_free_mtx);
650 SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
651 SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
658 * Clear recorded source entries for a group.
659 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
660 * FIXME: Should reap.
663 in6m_clear_recorded(struct in6_multi *inm)
665 struct ip6_msource *ims;
667 IN6_MULTI_LIST_LOCK_ASSERT();
669 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
672 --inm->in6m_st[1].iss_rec;
675 KASSERT(inm->in6m_st[1].iss_rec == 0,
676 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
680 * Record a source as pending for a Source-Group MLDv2 query.
681 * This lives here as it modifies the shared tree.
683 * inm is the group descriptor.
684 * naddr is the address of the source to record in network-byte order.
686 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
687 * lazy-allocate a source node in response to an SG query.
688 * Otherwise, no allocation is performed. This saves some memory
689 * with the trade-off that the source will not be reported to the
690 * router if joined in the window between the query response and
691 * the group actually being joined on the local host.
693 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
694 * This turns off the allocation of a recorded source entry if
695 * the group has not been joined.
697 * Return 0 if the source didn't exist or was already marked as recorded.
698 * Return 1 if the source was marked as recorded by this function.
699 * Return <0 if any error occurred (negated errno code).
702 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
704 struct ip6_msource find;
705 struct ip6_msource *ims, *nims;
707 IN6_MULTI_LIST_LOCK_ASSERT();
709 find.im6s_addr = *addr;
710 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
711 if (ims && ims->im6s_stp)
714 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
716 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
720 nims->im6s_addr = find.im6s_addr;
721 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
727 * Mark the source as recorded and update the recorded
731 ++inm->in6m_st[1].iss_rec;
737 * Return a pointer to an in6_msource owned by an in6_mfilter,
738 * given its source address.
739 * Lazy-allocate if needed. If this is a new entry its filter state is
742 * imf is the filter set being modified.
743 * addr is the source address.
745 * SMPng: May be called with locks held; malloc must not block.
748 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
749 struct in6_msource **plims)
751 struct ip6_msource find;
752 struct ip6_msource *ims, *nims;
753 struct in6_msource *lims;
760 find.im6s_addr = psin->sin6_addr;
761 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
762 lims = (struct in6_msource *)ims;
764 if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
766 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
770 lims = (struct in6_msource *)nims;
771 lims->im6s_addr = find.im6s_addr;
772 lims->im6sl_st[0] = MCAST_UNDEFINED;
773 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
783 * Graft a source entry into an existing socket-layer filter set,
784 * maintaining any required invariants and checking allocations.
786 * The source is marked as being in the new filter mode at t1.
788 * Return the pointer to the new node, otherwise return NULL.
790 static struct in6_msource *
791 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
792 const struct sockaddr_in6 *psin)
794 struct ip6_msource *nims;
795 struct in6_msource *lims;
797 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
801 lims = (struct in6_msource *)nims;
802 lims->im6s_addr = psin->sin6_addr;
803 lims->im6sl_st[0] = MCAST_UNDEFINED;
804 lims->im6sl_st[1] = st1;
805 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
812 * Prune a source entry from an existing socket-layer filter set,
813 * maintaining any required invariants and checking allocations.
815 * The source is marked as being left at t1, it is not freed.
817 * Return 0 if no error occurred, otherwise return an errno value.
820 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
822 struct ip6_msource find;
823 struct ip6_msource *ims;
824 struct in6_msource *lims;
826 find.im6s_addr = psin->sin6_addr;
827 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
830 lims = (struct in6_msource *)ims;
831 lims->im6sl_st[1] = MCAST_UNDEFINED;
836 * Revert socket-layer filter set deltas at t1 to t0 state.
839 im6f_rollback(struct in6_mfilter *imf)
841 struct ip6_msource *ims, *tims;
842 struct in6_msource *lims;
844 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
845 lims = (struct in6_msource *)ims;
846 if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
847 /* no change at t1 */
849 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
850 /* revert change to existing source at t1 */
851 lims->im6sl_st[1] = lims->im6sl_st[0];
853 /* revert source added t1 */
854 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
855 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
856 free(ims, M_IN6MFILTER);
860 imf->im6f_st[1] = imf->im6f_st[0];
864 * Mark socket-layer filter set as INCLUDE {} at t1.
867 im6f_leave(struct in6_mfilter *imf)
869 struct ip6_msource *ims;
870 struct in6_msource *lims;
872 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
873 lims = (struct in6_msource *)ims;
874 lims->im6sl_st[1] = MCAST_UNDEFINED;
876 imf->im6f_st[1] = MCAST_INCLUDE;
880 * Mark socket-layer filter set deltas as committed.
883 im6f_commit(struct in6_mfilter *imf)
885 struct ip6_msource *ims;
886 struct in6_msource *lims;
888 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
889 lims = (struct in6_msource *)ims;
890 lims->im6sl_st[0] = lims->im6sl_st[1];
892 imf->im6f_st[0] = imf->im6f_st[1];
896 * Reap unreferenced sources from socket-layer filter set.
899 im6f_reap(struct in6_mfilter *imf)
901 struct ip6_msource *ims, *tims;
902 struct in6_msource *lims;
904 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
905 lims = (struct in6_msource *)ims;
906 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
907 (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
908 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
909 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
910 free(ims, M_IN6MFILTER);
917 * Purge socket-layer filter set.
920 im6f_purge(struct in6_mfilter *imf)
922 struct ip6_msource *ims, *tims;
924 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
925 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
926 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
927 free(ims, M_IN6MFILTER);
930 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
931 KASSERT(RB_EMPTY(&imf->im6f_sources),
932 ("%s: im6f_sources not empty", __func__));
936 * Look up a source filter entry for a multicast group.
938 * inm is the group descriptor to work with.
939 * addr is the IPv6 address to look up.
940 * noalloc may be non-zero to suppress allocation of sources.
941 * *pims will be set to the address of the retrieved or allocated source.
943 * SMPng: NOTE: may be called with locks held.
944 * Return 0 if successful, otherwise return a non-zero error code.
947 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
948 const int noalloc, struct ip6_msource **pims)
950 struct ip6_msource find;
951 struct ip6_msource *ims, *nims;
953 char ip6tbuf[INET6_ADDRSTRLEN];
956 find.im6s_addr = *addr;
957 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
958 if (ims == NULL && !noalloc) {
959 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
961 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
965 nims->im6s_addr = *addr;
966 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
969 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
970 ip6_sprintf(ip6tbuf, addr), ims);
978 * Merge socket-layer source into MLD-layer source.
979 * If rollback is non-zero, perform the inverse of the merge.
982 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
985 int n = rollback ? -1 : 1;
987 char ip6tbuf[INET6_ADDRSTRLEN];
989 ip6_sprintf(ip6tbuf, &lims->im6s_addr);
992 if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
993 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
994 ims->im6s_st[1].ex -= n;
995 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
996 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
997 ims->im6s_st[1].in -= n;
1000 if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
1001 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
1002 ims->im6s_st[1].ex += n;
1003 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
1004 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
1005 ims->im6s_st[1].in += n;
1010 * Atomically update the global in6_multi state, when a membership's
1011 * filter list is being updated in any way.
1013 * imf is the per-inpcb-membership group filter pointer.
1014 * A fake imf may be passed for in-kernel consumers.
1016 * XXX This is a candidate for a set-symmetric-difference style loop
1017 * which would eliminate the repeated lookup from root of ims nodes,
1018 * as they share the same key space.
1020 * If any error occurred this function will back out of refcounts
1021 * and return a non-zero value.
1024 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1026 struct ip6_msource *ims, *nims;
1027 struct in6_msource *lims;
1028 int schanged, error;
1034 IN6_MULTI_LIST_LOCK_ASSERT();
1037 * Update the source filters first, as this may fail.
1038 * Maintain count of in-mode filters at t0, t1. These are
1039 * used to work out if we transition into ASM mode or not.
1040 * Maintain a count of source filters whose state was
1041 * actually modified by this operation.
1043 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1044 lims = (struct in6_msource *)ims;
1045 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1046 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1047 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1048 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1052 im6s_merge(nims, lims, 0);
1055 struct ip6_msource *bims;
1057 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1058 lims = (struct in6_msource *)ims;
1059 if (lims->im6sl_st[0] == lims->im6sl_st[1])
1061 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1064 im6s_merge(bims, lims, 1);
1069 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1070 __func__, nsrc0, nsrc1);
1072 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1073 if (imf->im6f_st[0] == imf->im6f_st[1] &&
1074 imf->im6f_st[1] == MCAST_INCLUDE) {
1076 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1077 --inm->in6m_st[1].iss_in;
1081 /* Handle filter mode transition on socket. */
1082 if (imf->im6f_st[0] != imf->im6f_st[1]) {
1083 CTR3(KTR_MLD, "%s: imf transition %d to %d",
1084 __func__, imf->im6f_st[0], imf->im6f_st[1]);
1086 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1087 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1088 --inm->in6m_st[1].iss_ex;
1089 } else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1090 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1091 --inm->in6m_st[1].iss_in;
1094 if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1095 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1096 inm->in6m_st[1].iss_ex++;
1097 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1098 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1099 inm->in6m_st[1].iss_in++;
1104 * Track inm filter state in terms of listener counts.
1105 * If there are any exclusive listeners, stack-wide
1106 * membership is exclusive.
1107 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1108 * If no listeners remain, state is undefined at t1,
1109 * and the MLD lifecycle for this group should finish.
1111 if (inm->in6m_st[1].iss_ex > 0) {
1112 CTR1(KTR_MLD, "%s: transition to EX", __func__);
1113 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1114 } else if (inm->in6m_st[1].iss_in > 0) {
1115 CTR1(KTR_MLD, "%s: transition to IN", __func__);
1116 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1118 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1119 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1122 /* Decrement ASM listener count on transition out of ASM mode. */
1123 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1124 if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1125 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1126 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1127 --inm->in6m_st[1].iss_asm;
1131 /* Increment ASM listener count on transition to ASM mode. */
1132 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1133 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1134 inm->in6m_st[1].iss_asm++;
1137 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1142 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1149 * Mark an in6_multi's filter set deltas as committed.
1150 * Called by MLD after a state change has been enqueued.
1153 in6m_commit(struct in6_multi *inm)
1155 struct ip6_msource *ims;
1157 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1158 CTR1(KTR_MLD, "%s: pre commit:", __func__);
1161 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1162 ims->im6s_st[0] = ims->im6s_st[1];
1164 inm->in6m_st[0] = inm->in6m_st[1];
1168 * Reap unreferenced nodes from an in6_multi's filter set.
1171 in6m_reap(struct in6_multi *inm)
1173 struct ip6_msource *ims, *tims;
1175 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1176 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1177 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1180 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1181 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1182 free(ims, M_IP6MSOURCE);
1188 * Purge all source nodes from an in6_multi's filter set.
1191 in6m_purge(struct in6_multi *inm)
1193 struct ip6_msource *ims, *tims;
1195 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1196 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1197 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1198 free(ims, M_IP6MSOURCE);
1201 /* Free state-change requests that might be queued. */
1202 mbufq_drain(&inm->in6m_scq);
1206 * Join a multicast address w/o sources.
1207 * KAME compatibility entry point.
1209 * SMPng: Assume no mc locks held by caller.
1212 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1213 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1219 error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1225 * Join a multicast group; real entry point.
1227 * Only preserves atomicity at inm level.
1228 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1230 * If the MLD downcall fails, the group is not joined, and an error
1234 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1235 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1238 struct in6_mfilter timf;
1239 struct in6_multi *inm;
1240 struct ifmultiaddr *ifma;
1243 char ip6tbuf[INET6_ADDRSTRLEN];
1248 * Sanity: Check scope zone ID was set for ifp, if and
1249 * only if group is scoped to an interface.
1251 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1252 ("%s: not a multicast address", __func__));
1253 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1254 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1255 KASSERT(mcaddr->s6_addr16[1] != 0,
1256 ("%s: scope zone ID not set", __func__));
1260 IN6_MULTI_LOCK_ASSERT();
1261 IN6_MULTI_LIST_UNLOCK_ASSERT();
1263 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1264 ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1270 * If no imf was specified (i.e. kernel consumer),
1271 * fake one up and assume it is an ASM join.
1274 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1277 error = in6_getmulti(ifp, mcaddr, &inm);
1279 CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1283 IN6_MULTI_LIST_LOCK();
1284 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1285 error = in6m_merge(inm, imf);
1287 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1288 goto out_in6m_release;
1291 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1292 error = mld_change_state(inm, delay);
1294 CTR1(KTR_MLD, "%s: failed to update source", __func__);
1295 goto out_in6m_release;
1300 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1302 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1303 if (ifma->ifma_protospec == inm) {
1304 ifma->ifma_protospec = NULL;
1308 in6m_release_deferred(inm);
1309 IF_ADDR_RUNLOCK(ifp);
1313 IN6_MULTI_LIST_UNLOCK();
1318 * Leave a multicast group; unlocked entry point.
1321 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1326 error = in6_leavegroup_locked(inm, imf);
1332 * Leave a multicast group; real entry point.
1333 * All source filters will be expunged.
1335 * Only preserves atomicity at inm level.
1337 * Holding the write lock for the INP which contains imf
1338 * is highly advisable. We can't assert for it as imf does not
1339 * contain a back-pointer to the owning inp.
1341 * Note: This is not the same as in6m_release(*) as this function also
1342 * makes a state change downcall into MLD.
1345 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1347 struct in6_mfilter timf;
1351 char ip6tbuf[INET6_ADDRSTRLEN];
1356 IN6_MULTI_LOCK_ASSERT();
1358 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1359 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1360 (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1364 * If no imf was specified (i.e. kernel consumer),
1365 * fake one up and assume it is an ASM join.
1368 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1373 * Begin state merge transaction at MLD layer.
1375 * As this particular invocation should not cause any memory
1376 * to be allocated, and there is no opportunity to roll back
1377 * the transaction, it MUST NOT fail.
1380 ifp = inm->in6m_ifp;
1381 IN6_MULTI_LIST_LOCK();
1382 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1383 error = in6m_merge(inm, imf);
1384 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1386 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1387 error = mld_change_state(inm, 0);
1389 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1391 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1394 in6m_release_deferred(inm);
1396 IF_ADDR_WUNLOCK(ifp);
1397 IN6_MULTI_LIST_UNLOCK();
1404 * Block or unblock an ASM multicast source on an inpcb.
1405 * This implements the delta-based API described in RFC 3678.
1407 * The delta-based API applies only to exclusive-mode memberships.
1408 * An MLD downcall will be performed.
1410 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1412 * Return 0 if successful, otherwise return an appropriate error code.
1415 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1417 struct group_source_req gsr;
1418 sockunion_t *gsa, *ssa;
1420 struct in6_mfilter *imf;
1421 struct ip6_moptions *imo;
1422 struct in6_msource *ims;
1423 struct in6_multi *inm;
1428 char ip6tbuf[INET6_ADDRSTRLEN];
1435 memset(&gsr, 0, sizeof(struct group_source_req));
1436 gsa = (sockunion_t *)&gsr.gsr_group;
1437 ssa = (sockunion_t *)&gsr.gsr_source;
1439 switch (sopt->sopt_name) {
1440 case MCAST_BLOCK_SOURCE:
1441 case MCAST_UNBLOCK_SOURCE:
1442 error = sooptcopyin(sopt, &gsr,
1443 sizeof(struct group_source_req),
1444 sizeof(struct group_source_req));
1448 if (gsa->sin6.sin6_family != AF_INET6 ||
1449 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1452 if (ssa->sin6.sin6_family != AF_INET6 ||
1453 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1456 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1457 return (EADDRNOTAVAIL);
1459 ifp = ifnet_byindex(gsr.gsr_interface);
1461 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1466 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1467 __func__, sopt->sopt_name);
1468 return (EOPNOTSUPP);
1472 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1475 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1478 * Check if we are actually a member of this group.
1480 imo = in6p_findmoptions(inp);
1481 idx = im6o_match_group(imo, ifp, &gsa->sa);
1482 if (idx == -1 || imo->im6o_mfilters == NULL) {
1483 error = EADDRNOTAVAIL;
1484 goto out_in6p_locked;
1487 KASSERT(imo->im6o_mfilters != NULL,
1488 ("%s: im6o_mfilters not allocated", __func__));
1489 imf = &imo->im6o_mfilters[idx];
1490 inm = imo->im6o_membership[idx];
1493 * Attempting to use the delta-based API on an
1494 * non exclusive-mode membership is an error.
1496 fmode = imf->im6f_st[0];
1497 if (fmode != MCAST_EXCLUDE) {
1499 goto out_in6p_locked;
1503 * Deal with error cases up-front:
1504 * Asked to block, but already blocked; or
1505 * Asked to unblock, but nothing to unblock.
1506 * If adding a new block entry, allocate it.
1508 ims = im6o_match_source(imo, idx, &ssa->sa);
1509 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1510 CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1511 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1512 doblock ? "" : "not ");
1513 error = EADDRNOTAVAIL;
1514 goto out_in6p_locked;
1517 INP_WLOCK_ASSERT(inp);
1520 * Begin state merge transaction at socket layer.
1523 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1524 ims = im6f_graft(imf, fmode, &ssa->sin6);
1528 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1529 error = im6f_prune(imf, &ssa->sin6);
1533 CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1534 goto out_im6f_rollback;
1538 * Begin state merge transaction at MLD layer.
1540 IN6_MULTI_LIST_LOCK();
1541 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1542 error = in6m_merge(inm, imf);
1544 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1546 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1547 error = mld_change_state(inm, 0);
1549 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1552 IN6_MULTI_LIST_UNLOCK();
1568 * Given an inpcb, return its multicast options structure pointer. Accepts
1569 * an unlocked inpcb pointer, but will return it locked. May sleep.
1571 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1572 * SMPng: NOTE: Returns with the INP write lock held.
1574 static struct ip6_moptions *
1575 in6p_findmoptions(struct inpcb *inp)
1577 struct ip6_moptions *imo;
1578 struct in6_multi **immp;
1579 struct in6_mfilter *imfp;
1583 if (inp->in6p_moptions != NULL)
1584 return (inp->in6p_moptions);
1588 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1589 immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1591 imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1592 M_IN6MFILTER, M_WAITOK);
1594 imo->im6o_multicast_ifp = NULL;
1595 imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1596 imo->im6o_multicast_loop = in6_mcast_loop;
1597 imo->im6o_num_memberships = 0;
1598 imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1599 imo->im6o_membership = immp;
1601 /* Initialize per-group source filters. */
1602 for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1603 im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1604 imo->im6o_mfilters = imfp;
1607 if (inp->in6p_moptions != NULL) {
1608 free(imfp, M_IN6MFILTER);
1609 free(immp, M_IP6MOPTS);
1610 free(imo, M_IP6MOPTS);
1611 return (inp->in6p_moptions);
1613 inp->in6p_moptions = imo;
1618 * Discard the IPv6 multicast options (and source filters).
1620 * SMPng: NOTE: assumes INP write lock is held.
1622 * XXX can all be safely deferred to epoch_call
1627 inp_gcmoptions(epoch_context_t ctx)
1629 struct ip6_moptions *imo;
1630 struct in6_mfilter *imf;
1631 struct in6_multi *inm;
1633 size_t idx, nmships;
1635 imo = __containerof(ctx, struct ip6_moptions, imo6_epoch_ctx);
1637 nmships = imo->im6o_num_memberships;
1638 for (idx = 0; idx < nmships; ++idx) {
1639 imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1642 inm = imo->im6o_membership[idx];
1643 ifp = inm->in6m_ifp;
1645 CURVNET_SET(ifp->if_vnet);
1646 (void)in6_leavegroup(inm, imf);
1649 (void)in6_leavegroup(inm, imf);
1655 if (imo->im6o_mfilters)
1656 free(imo->im6o_mfilters, M_IN6MFILTER);
1657 free(imo->im6o_membership, M_IP6MOPTS);
1658 free(imo, M_IP6MOPTS);
1662 ip6_freemoptions(struct ip6_moptions *imo)
1666 epoch_call(net_epoch_preempt, &imo->imo6_epoch_ctx, inp_gcmoptions);
1670 * Atomically get source filters on a socket for an IPv6 multicast group.
1671 * Called with INP lock held; returns with lock released.
1674 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1676 struct __msfilterreq msfr;
1679 struct ip6_moptions *imo;
1680 struct in6_mfilter *imf;
1681 struct ip6_msource *ims;
1682 struct in6_msource *lims;
1683 struct sockaddr_in6 *psin;
1684 struct sockaddr_storage *ptss;
1685 struct sockaddr_storage *tss;
1687 size_t idx, nsrcs, ncsrcs;
1689 INP_WLOCK_ASSERT(inp);
1691 imo = inp->in6p_moptions;
1692 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1696 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1697 sizeof(struct __msfilterreq));
1701 if (msfr.msfr_group.ss_family != AF_INET6 ||
1702 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1705 gsa = (sockunion_t *)&msfr.msfr_group;
1706 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1709 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1710 return (EADDRNOTAVAIL);
1711 ifp = ifnet_byindex(msfr.msfr_ifindex);
1713 return (EADDRNOTAVAIL);
1714 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1719 * Lookup group on the socket.
1721 idx = im6o_match_group(imo, ifp, &gsa->sa);
1722 if (idx == -1 || imo->im6o_mfilters == NULL) {
1724 return (EADDRNOTAVAIL);
1726 imf = &imo->im6o_mfilters[idx];
1729 * Ignore memberships which are in limbo.
1731 if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1735 msfr.msfr_fmode = imf->im6f_st[1];
1738 * If the user specified a buffer, copy out the source filter
1739 * entries to userland gracefully.
1740 * We only copy out the number of entries which userland
1741 * has asked for, but we always tell userland how big the
1742 * buffer really needs to be.
1744 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1745 msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1747 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1748 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1749 M_TEMP, M_NOWAIT | M_ZERO);
1757 * Count number of sources in-mode at t0.
1758 * If buffer space exists and remains, copy out source entries.
1760 nsrcs = msfr.msfr_nsrcs;
1763 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1764 lims = (struct in6_msource *)ims;
1765 if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1766 lims->im6sl_st[0] != imf->im6f_st[0])
1769 if (tss != NULL && nsrcs > 0) {
1770 psin = (struct sockaddr_in6 *)ptss;
1771 psin->sin6_family = AF_INET6;
1772 psin->sin6_len = sizeof(struct sockaddr_in6);
1773 psin->sin6_addr = lims->im6s_addr;
1774 psin->sin6_port = 0;
1783 error = copyout(tss, msfr.msfr_srcs,
1784 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1790 msfr.msfr_nsrcs = ncsrcs;
1791 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1797 * Return the IP multicast options in response to user getsockopt().
1800 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1802 struct ip6_moptions *im6o;
1807 im6o = inp->in6p_moptions;
1809 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1810 * or is a divert socket, reject it.
1812 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1813 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1814 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1816 return (EOPNOTSUPP);
1820 switch (sopt->sopt_name) {
1821 case IPV6_MULTICAST_IF:
1822 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1825 optval = im6o->im6o_multicast_ifp->if_index;
1828 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1831 case IPV6_MULTICAST_HOPS:
1833 optval = V_ip6_defmcasthlim;
1835 optval = im6o->im6o_multicast_hlim;
1837 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1840 case IPV6_MULTICAST_LOOP:
1842 optval = in6_mcast_loop; /* XXX VIMAGE */
1844 optval = im6o->im6o_multicast_loop;
1846 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1851 error = EADDRNOTAVAIL;
1854 error = in6p_get_source_filters(inp, sopt);
1860 error = ENOPROTOOPT;
1864 INP_UNLOCK_ASSERT(inp);
1870 * Look up the ifnet to use for a multicast group membership,
1871 * given the address of an IPv6 group.
1873 * This routine exists to support legacy IPv6 multicast applications.
1875 * If inp is non-NULL, use this socket's current FIB number for any
1876 * required FIB lookup. Look up the group address in the unicast FIB,
1877 * and use its ifp; usually, this points to the default next-hop.
1878 * If the FIB lookup fails, return NULL.
1880 * FUTURE: Support multiple forwarding tables for IPv6.
1882 * Returns NULL if no ifp could be found.
1884 static struct ifnet *
1885 in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1886 const struct sockaddr_in6 *gsin6)
1888 struct nhop6_basic nh6;
1889 struct in6_addr dst;
1893 KASSERT(in6p->inp_vflag & INP_IPV6,
1894 ("%s: not INP_IPV6 inpcb", __func__));
1895 KASSERT(gsin6->sin6_family == AF_INET6,
1896 ("%s: not AF_INET6 group", __func__));
1898 in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1899 fibnum = in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB;
1900 if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6) != 0)
1903 return (nh6.nh_ifp);
1907 * Join an IPv6 multicast group, possibly with a source.
1909 * FIXME: The KAME use of the unspecified address (::)
1910 * to join *all* multicast groups is currently unsupported.
1913 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1915 struct group_source_req gsr;
1916 sockunion_t *gsa, *ssa;
1918 struct in6_mfilter *imf;
1919 struct ip6_moptions *imo;
1920 struct in6_multi *inm;
1921 struct in6_msource *lims;
1931 memset(&gsr, 0, sizeof(struct group_source_req));
1932 gsa = (sockunion_t *)&gsr.gsr_group;
1933 gsa->ss.ss_family = AF_UNSPEC;
1934 ssa = (sockunion_t *)&gsr.gsr_source;
1935 ssa->ss.ss_family = AF_UNSPEC;
1938 * Chew everything into struct group_source_req.
1939 * Overwrite the port field if present, as the sockaddr
1940 * being copied in may be matched with a binary comparison.
1941 * Ignore passed-in scope ID.
1943 switch (sopt->sopt_name) {
1944 case IPV6_JOIN_GROUP: {
1945 struct ipv6_mreq mreq;
1947 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1948 sizeof(struct ipv6_mreq));
1952 gsa->sin6.sin6_family = AF_INET6;
1953 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1954 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1956 if (mreq.ipv6mr_interface == 0) {
1957 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1959 if (V_if_index < mreq.ipv6mr_interface)
1960 return (EADDRNOTAVAIL);
1961 ifp = ifnet_byindex(mreq.ipv6mr_interface);
1963 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1964 __func__, mreq.ipv6mr_interface, ifp);
1967 case MCAST_JOIN_GROUP:
1968 case MCAST_JOIN_SOURCE_GROUP:
1969 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1970 error = sooptcopyin(sopt, &gsr,
1971 sizeof(struct group_req),
1972 sizeof(struct group_req));
1973 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1974 error = sooptcopyin(sopt, &gsr,
1975 sizeof(struct group_source_req),
1976 sizeof(struct group_source_req));
1981 if (gsa->sin6.sin6_family != AF_INET6 ||
1982 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1985 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1986 if (ssa->sin6.sin6_family != AF_INET6 ||
1987 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1989 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1992 * TODO: Validate embedded scope ID in source
1993 * list entry against passed-in ifp, if and only
1994 * if source list filter entry is iface or node local.
1996 in6_clearscope(&ssa->sin6.sin6_addr);
1997 ssa->sin6.sin6_port = 0;
1998 ssa->sin6.sin6_scope_id = 0;
2001 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
2002 return (EADDRNOTAVAIL);
2003 ifp = ifnet_byindex(gsr.gsr_interface);
2007 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2008 __func__, sopt->sopt_name);
2009 return (EOPNOTSUPP);
2013 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2016 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2017 return (EADDRNOTAVAIL);
2019 gsa->sin6.sin6_port = 0;
2020 gsa->sin6.sin6_scope_id = 0;
2023 * Always set the scope zone ID on memberships created from userland.
2024 * Use the passed-in ifp to do this.
2025 * XXX The in6_setscope() return value is meaningless.
2026 * XXX SCOPE6_LOCK() is taken by in6_setscope().
2028 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2030 imo = in6p_findmoptions(inp);
2031 idx = im6o_match_group(imo, ifp, &gsa->sa);
2035 inm = imo->im6o_membership[idx];
2036 imf = &imo->im6o_mfilters[idx];
2037 if (ssa->ss.ss_family != AF_UNSPEC) {
2039 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2040 * is an error. On an existing inclusive membership,
2041 * it just adds the source to the filter list.
2043 if (imf->im6f_st[1] != MCAST_INCLUDE) {
2045 goto out_in6p_locked;
2048 * Throw out duplicates.
2050 * XXX FIXME: This makes a naive assumption that
2051 * even if entries exist for *ssa in this imf,
2052 * they will be rejected as dupes, even if they
2053 * are not valid in the current mode (in-mode).
2055 * in6_msource is transactioned just as for anything
2056 * else in SSM -- but note naive use of in6m_graft()
2057 * below for allocating new filter entries.
2059 * This is only an issue if someone mixes the
2060 * full-state SSM API with the delta-based API,
2061 * which is discouraged in the relevant RFCs.
2063 lims = im6o_match_source(imo, idx, &ssa->sa);
2064 if (lims != NULL /*&&
2065 lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2066 error = EADDRNOTAVAIL;
2067 goto out_in6p_locked;
2071 * MCAST_JOIN_GROUP alone, on any existing membership,
2072 * is rejected, to stop the same inpcb tying up
2073 * multiple refs to the in_multi.
2074 * On an existing inclusive membership, this is also
2075 * an error; if you want to change filter mode,
2076 * you must use the userland API setsourcefilter().
2077 * XXX We don't reject this for imf in UNDEFINED
2078 * state at t1, because allocation of a filter
2079 * is atomic with allocation of a membership.
2082 goto out_in6p_locked;
2087 * Begin state merge transaction at socket layer.
2089 INP_WLOCK_ASSERT(inp);
2092 if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
2093 error = im6o_grow(imo);
2095 goto out_in6p_locked;
2098 * Allocate the new slot upfront so we can deal with
2099 * grafting the new source filter in same code path
2100 * as for join-source on existing membership.
2102 idx = imo->im6o_num_memberships;
2103 imo->im6o_membership[idx] = NULL;
2104 imo->im6o_num_memberships++;
2105 KASSERT(imo->im6o_mfilters != NULL,
2106 ("%s: im6f_mfilters vector was not allocated", __func__));
2107 imf = &imo->im6o_mfilters[idx];
2108 KASSERT(RB_EMPTY(&imf->im6f_sources),
2109 ("%s: im6f_sources not empty", __func__));
2113 * Graft new source into filter list for this inpcb's
2114 * membership of the group. The in6_multi may not have
2115 * been allocated yet if this is a new membership, however,
2116 * the in_mfilter slot will be allocated and must be initialized.
2118 * Note: Grafting of exclusive mode filters doesn't happen
2120 * XXX: Should check for non-NULL lims (node exists but may
2121 * not be in-mode) for interop with full-state API.
2123 if (ssa->ss.ss_family != AF_UNSPEC) {
2124 /* Membership starts in IN mode */
2126 CTR1(KTR_MLD, "%s: new join w/source", __func__);
2127 im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2129 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2131 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2133 CTR1(KTR_MLD, "%s: merge imf state failed",
2139 /* No address specified; Membership starts in EX mode */
2141 CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2142 im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2147 * Begin state merge transaction at MLD layer.
2154 error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2160 imo->im6o_membership[idx] = inm;
2162 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2163 IN6_MULTI_LIST_LOCK();
2164 error = in6m_merge(inm, imf);
2166 CTR1(KTR_MLD, "%s: failed to merge inm state",
2169 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2170 error = mld_change_state(inm, 0);
2172 CTR1(KTR_MLD, "%s: failed mld downcall",
2175 IN6_MULTI_LIST_UNLOCK();
2180 if (in_pcbrele_wlocked(inp))
2193 if (error && is_new) {
2194 imo->im6o_membership[idx] = NULL;
2195 --imo->im6o_num_memberships;
2204 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2207 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2209 struct ipv6_mreq mreq;
2210 struct group_source_req gsr;
2211 sockunion_t *gsa, *ssa;
2213 struct in6_mfilter *imf;
2214 struct ip6_moptions *imo;
2215 struct in6_msource *ims;
2216 struct in6_multi *inm;
2219 int error, is_final;
2221 char ip6tbuf[INET6_ADDRSTRLEN];
2229 memset(&gsr, 0, sizeof(struct group_source_req));
2230 gsa = (sockunion_t *)&gsr.gsr_group;
2231 gsa->ss.ss_family = AF_UNSPEC;
2232 ssa = (sockunion_t *)&gsr.gsr_source;
2233 ssa->ss.ss_family = AF_UNSPEC;
2236 * Chew everything passed in up into a struct group_source_req
2237 * as that is easier to process.
2238 * Note: Any embedded scope ID in the multicast group passed
2239 * in by userland is ignored, the interface index is the recommended
2240 * mechanism to specify an interface; see below.
2242 switch (sopt->sopt_name) {
2243 case IPV6_LEAVE_GROUP:
2244 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2245 sizeof(struct ipv6_mreq));
2248 gsa->sin6.sin6_family = AF_INET6;
2249 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2250 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2251 gsa->sin6.sin6_port = 0;
2252 gsa->sin6.sin6_scope_id = 0;
2253 ifindex = mreq.ipv6mr_interface;
2256 case MCAST_LEAVE_GROUP:
2257 case MCAST_LEAVE_SOURCE_GROUP:
2258 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2259 error = sooptcopyin(sopt, &gsr,
2260 sizeof(struct group_req),
2261 sizeof(struct group_req));
2262 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2263 error = sooptcopyin(sopt, &gsr,
2264 sizeof(struct group_source_req),
2265 sizeof(struct group_source_req));
2270 if (gsa->sin6.sin6_family != AF_INET6 ||
2271 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2273 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2274 if (ssa->sin6.sin6_family != AF_INET6 ||
2275 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2277 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2280 * TODO: Validate embedded scope ID in source
2281 * list entry against passed-in ifp, if and only
2282 * if source list filter entry is iface or node local.
2284 in6_clearscope(&ssa->sin6.sin6_addr);
2286 gsa->sin6.sin6_port = 0;
2287 gsa->sin6.sin6_scope_id = 0;
2288 ifindex = gsr.gsr_interface;
2292 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2293 __func__, sopt->sopt_name);
2294 return (EOPNOTSUPP);
2298 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2302 * Validate interface index if provided. If no interface index
2303 * was provided separately, attempt to look the membership up
2304 * from the default scope as a last resort to disambiguate
2305 * the membership we are being asked to leave.
2306 * XXX SCOPE6 lock potentially taken here.
2309 if (V_if_index < ifindex)
2310 return (EADDRNOTAVAIL);
2311 ifp = ifnet_byindex(ifindex);
2313 return (EADDRNOTAVAIL);
2314 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2316 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2318 return (EADDRNOTAVAIL);
2320 * Some badly behaved applications don't pass an ifindex
2321 * or a scope ID, which is an API violation. In this case,
2322 * perform a lookup as per a v6 join.
2324 * XXX For now, stomp on zone ID for the corner case.
2325 * This is not the 'KAME way', but we need to see the ifp
2326 * directly until such time as this implementation is
2327 * refactored, assuming the scope IDs are the way to go.
2329 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2331 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2332 "ifp for group %s.", __func__,
2333 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2334 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2336 ifp = ifnet_byindex(ifindex);
2339 return (EADDRNOTAVAIL);
2342 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2343 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2346 * Find the membership in the membership array.
2348 imo = in6p_findmoptions(inp);
2349 idx = im6o_match_group(imo, ifp, &gsa->sa);
2351 error = EADDRNOTAVAIL;
2352 goto out_in6p_locked;
2354 inm = imo->im6o_membership[idx];
2355 imf = &imo->im6o_mfilters[idx];
2357 if (ssa->ss.ss_family != AF_UNSPEC)
2361 * Begin state merge transaction at socket layer.
2363 INP_WLOCK_ASSERT(inp);
2366 * If we were instructed only to leave a given source, do so.
2367 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2372 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2373 error = EADDRNOTAVAIL;
2374 goto out_in6p_locked;
2376 ims = im6o_match_source(imo, idx, &ssa->sa);
2378 CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2379 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2381 error = EADDRNOTAVAIL;
2382 goto out_in6p_locked;
2384 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2385 error = im6f_prune(imf, &ssa->sin6);
2387 CTR1(KTR_MLD, "%s: merge imf state failed",
2389 goto out_in6p_locked;
2394 * Begin state merge transaction at MLD layer.
2402 * Give up the multicast address record to which
2403 * the membership points.
2405 (void)in6_leavegroup_locked(inm, imf);
2407 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2408 IN6_MULTI_LIST_LOCK();
2409 error = in6m_merge(inm, imf);
2411 CTR1(KTR_MLD, "%s: failed to merge inm state",
2414 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2415 error = mld_change_state(inm, 0);
2417 CTR1(KTR_MLD, "%s: failed mld downcall",
2420 IN6_MULTI_LIST_UNLOCK();
2425 if (in_pcbrele_wlocked(inp))
2436 /* Remove the gap in the membership array. */
2437 for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2438 imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2439 imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2441 imo->im6o_num_memberships--;
2450 * Select the interface for transmitting IPv6 multicast datagrams.
2452 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2453 * may be passed to this socket option. An address of in6addr_any or an
2454 * interface index of 0 is used to remove a previous selection.
2455 * When no interface is selected, one is chosen for every send.
2458 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2461 struct ip6_moptions *imo;
2465 if (sopt->sopt_valsize != sizeof(u_int))
2468 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2471 if (V_if_index < ifindex)
2476 ifp = ifnet_byindex(ifindex);
2479 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2480 return (EADDRNOTAVAIL);
2482 imo = in6p_findmoptions(inp);
2483 imo->im6o_multicast_ifp = ifp;
2490 * Atomically set source filters on a socket for an IPv6 multicast group.
2492 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2495 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2497 struct __msfilterreq msfr;
2500 struct in6_mfilter *imf;
2501 struct ip6_moptions *imo;
2502 struct in6_multi *inm;
2506 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2507 sizeof(struct __msfilterreq));
2511 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2514 if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2515 msfr.msfr_fmode != MCAST_INCLUDE)
2518 if (msfr.msfr_group.ss_family != AF_INET6 ||
2519 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2522 gsa = (sockunion_t *)&msfr.msfr_group;
2523 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2526 gsa->sin6.sin6_port = 0; /* ignore port */
2528 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2529 return (EADDRNOTAVAIL);
2530 ifp = ifnet_byindex(msfr.msfr_ifindex);
2532 return (EADDRNOTAVAIL);
2533 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2536 * Take the INP write lock.
2537 * Check if this socket is a member of this group.
2539 imo = in6p_findmoptions(inp);
2540 idx = im6o_match_group(imo, ifp, &gsa->sa);
2541 if (idx == -1 || imo->im6o_mfilters == NULL) {
2542 error = EADDRNOTAVAIL;
2543 goto out_in6p_locked;
2545 inm = imo->im6o_membership[idx];
2546 imf = &imo->im6o_mfilters[idx];
2549 * Begin state merge transaction at socket layer.
2551 INP_WLOCK_ASSERT(inp);
2553 imf->im6f_st[1] = msfr.msfr_fmode;
2556 * Apply any new source filters, if present.
2557 * Make a copy of the user-space source vector so
2558 * that we may copy them with a single copyin. This
2559 * allows us to deal with page faults up-front.
2561 if (msfr.msfr_nsrcs > 0) {
2562 struct in6_msource *lims;
2563 struct sockaddr_in6 *psin;
2564 struct sockaddr_storage *kss, *pkss;
2569 CTR2(KTR_MLD, "%s: loading %lu source list entries",
2570 __func__, (unsigned long)msfr.msfr_nsrcs);
2571 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2573 error = copyin(msfr.msfr_srcs, kss,
2574 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2583 * Mark all source filters as UNDEFINED at t1.
2584 * Restore new group filter mode, as im6f_leave()
2585 * will set it to INCLUDE.
2588 imf->im6f_st[1] = msfr.msfr_fmode;
2591 * Update socket layer filters at t1, lazy-allocating
2592 * new entries. This saves a bunch of memory at the
2593 * cost of one RB_FIND() per source entry; duplicate
2594 * entries in the msfr_nsrcs vector are ignored.
2595 * If we encounter an error, rollback transaction.
2597 * XXX This too could be replaced with a set-symmetric
2598 * difference like loop to avoid walking from root
2599 * every time, as the key space is common.
2601 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2602 psin = (struct sockaddr_in6 *)pkss;
2603 if (psin->sin6_family != AF_INET6) {
2604 error = EAFNOSUPPORT;
2607 if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2611 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2616 * TODO: Validate embedded scope ID in source
2617 * list entry against passed-in ifp, if and only
2618 * if source list filter entry is iface or node local.
2620 in6_clearscope(&psin->sin6_addr);
2621 error = im6f_get_source(imf, psin, &lims);
2624 lims->im6sl_st[1] = imf->im6f_st[1];
2630 goto out_im6f_rollback;
2632 INP_WLOCK_ASSERT(inp);
2633 IN6_MULTI_LIST_LOCK();
2636 * Begin state merge transaction at MLD layer.
2638 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2639 error = in6m_merge(inm, imf);
2641 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2643 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2644 error = mld_change_state(inm, 0);
2646 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2649 IN6_MULTI_LIST_UNLOCK();
2665 * Set the IP multicast options in response to user setsockopt().
2667 * Many of the socket options handled in this function duplicate the
2668 * functionality of socket options in the regular unicast API. However,
2669 * it is not possible to merge the duplicate code, because the idempotence
2670 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2671 * the effects of these options must be treated as separate and distinct.
2673 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2676 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2678 struct ip6_moptions *im6o;
2684 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2685 * or is a divert socket, reject it.
2687 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2688 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2689 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2690 return (EOPNOTSUPP);
2692 switch (sopt->sopt_name) {
2693 case IPV6_MULTICAST_IF:
2694 error = in6p_set_multicast_if(inp, sopt);
2697 case IPV6_MULTICAST_HOPS: {
2700 if (sopt->sopt_valsize != sizeof(int)) {
2704 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2707 if (hlim < -1 || hlim > 255) {
2710 } else if (hlim == -1) {
2711 hlim = V_ip6_defmcasthlim;
2713 im6o = in6p_findmoptions(inp);
2714 im6o->im6o_multicast_hlim = hlim;
2719 case IPV6_MULTICAST_LOOP: {
2723 * Set the loopback flag for outgoing multicast packets.
2724 * Must be zero or one.
2726 if (sopt->sopt_valsize != sizeof(u_int)) {
2730 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2737 im6o = in6p_findmoptions(inp);
2738 im6o->im6o_multicast_loop = loop;
2743 case IPV6_JOIN_GROUP:
2744 case MCAST_JOIN_GROUP:
2745 case MCAST_JOIN_SOURCE_GROUP:
2746 error = in6p_join_group(inp, sopt);
2749 case IPV6_LEAVE_GROUP:
2750 case MCAST_LEAVE_GROUP:
2751 case MCAST_LEAVE_SOURCE_GROUP:
2752 error = in6p_leave_group(inp, sopt);
2755 case MCAST_BLOCK_SOURCE:
2756 case MCAST_UNBLOCK_SOURCE:
2757 error = in6p_block_unblock_source(inp, sopt);
2761 error = in6p_set_source_filters(inp, sopt);
2769 INP_UNLOCK_ASSERT(inp);
2775 * Expose MLD's multicast filter mode and source list(s) to userland,
2776 * keyed by (ifindex, group).
2777 * The filter mode is written out as a uint32_t, followed by
2778 * 0..n of struct in6_addr.
2779 * For use by ifmcstat(8).
2780 * SMPng: NOTE: unlocked read of ifindex space.
2783 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2785 struct in6_addr mcaddr;
2786 struct in6_addr src;
2788 struct ifmultiaddr *ifma;
2789 struct in6_multi *inm;
2790 struct ip6_msource *ims;
2794 uint32_t fmode, ifindex;
2796 char ip6tbuf[INET6_ADDRSTRLEN];
2802 if (req->newptr != NULL)
2805 /* int: ifindex + 4 * 32 bits of IPv6 address */
2810 if (ifindex <= 0 || ifindex > V_if_index) {
2811 CTR2(KTR_MLD, "%s: ifindex %u out of range",
2816 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2817 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2818 CTR2(KTR_MLD, "%s: group %s is not multicast",
2819 __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2823 ifp = ifnet_byindex(ifindex);
2825 CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2830 * Internal MLD lookups require that scope/zone ID is set.
2832 (void)in6_setscope(&mcaddr, ifp, NULL);
2834 retval = sysctl_wire_old_buffer(req,
2835 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2840 IN6_MULTI_LIST_LOCK();
2842 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2843 if (ifma->ifma_addr->sa_family != AF_INET6 ||
2844 ifma->ifma_protospec == NULL)
2846 inm = (struct in6_multi *)ifma->ifma_protospec;
2847 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2849 fmode = inm->in6m_st[1].iss_fmode;
2850 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2853 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2854 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2856 * Only copy-out sources which are in-mode.
2858 if (fmode != im6s_get_mode(inm, ims, 1)) {
2859 CTR1(KTR_MLD, "%s: skip non-in-mode",
2863 src = ims->im6s_addr;
2864 retval = SYSCTL_OUT(req, &src,
2865 sizeof(struct in6_addr));
2870 IF_ADDR_RUNLOCK(ifp);
2872 IN6_MULTI_LIST_UNLOCK();
2880 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2883 in6m_mode_str(const int mode)
2886 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2887 return (in6m_modestrs[mode]);
2891 static const char *in6m_statestrs[] = {
2904 in6m_state_str(const int state)
2907 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2908 return (in6m_statestrs[state]);
2913 * Dump an in6_multi structure to the console.
2916 in6m_print(const struct in6_multi *inm)
2919 char ip6tbuf[INET6_ADDRSTRLEN];
2921 if ((ktr_mask & KTR_MLD) == 0)
2924 printf("%s: --- begin in6m %p ---\n", __func__, inm);
2925 printf("addr %s ifp %p(%s) ifma %p\n",
2926 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2928 if_name(inm->in6m_ifp),
2930 printf("timer %u state %s refcount %u scq.len %u\n",
2932 in6m_state_str(inm->in6m_state),
2934 mbufq_len(&inm->in6m_scq));
2935 printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2940 for (t = 0; t < 2; t++) {
2941 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2942 in6m_mode_str(inm->in6m_st[t].iss_fmode),
2943 inm->in6m_st[t].iss_asm,
2944 inm->in6m_st[t].iss_ex,
2945 inm->in6m_st[t].iss_in,
2946 inm->in6m_st[t].iss_rec);
2948 printf("%s: --- end in6m %p ---\n", __func__, inm);
2954 in6m_print(const struct in6_multi *inm)