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);
547 if_delmulti_ifma_flags(ifma, 1);
554 static struct grouptask free_gtask;
555 static struct in6_multi_head in6m_free_list;
556 static void in6m_release_task(void *arg __unused);
557 static void in6m_init(void)
559 SLIST_INIT(&in6m_free_list);
560 taskqgroup_config_gtask_init(NULL, &free_gtask, in6m_release_task, "in6m release task");
563 SYSINIT(in6m_init, SI_SUB_SMP + 1, SI_ORDER_FIRST,
568 in6m_release_list_deferred(struct in6_multi_head *inmh)
570 if (SLIST_EMPTY(inmh))
572 mtx_lock(&in6_multi_free_mtx);
573 SLIST_CONCAT(&in6m_free_list, inmh, in6_multi, in6m_nrele);
574 mtx_unlock(&in6_multi_free_mtx);
575 GROUPTASK_ENQUEUE(&free_gtask);
579 in6m_disconnect(struct in6_multi *inm)
583 struct in6_ifaddr *ifa6;
584 struct in6_multi_mship *imm, *imm_tmp;
585 struct ifmultiaddr *ifma, *ll_ifma;
588 IF_ADDR_WLOCK_ASSERT(ifp);
589 ifma = inm->in6m_ifma;
592 TAILQ_REMOVE(&ifp->if_multiaddrs, ifma, ifma_link);
593 MCDPRINTF("removed ifma: %p from %s\n", ifma, ifp->if_xname);
594 if ((ll_ifma = ifma->ifma_llifma) != NULL) {
595 MPASS(ifma != ll_ifma);
596 ifma->ifma_llifma = NULL;
597 MPASS(ll_ifma->ifma_llifma == NULL);
598 MPASS(ll_ifma->ifma_ifp == ifp);
599 if (--ll_ifma->ifma_refcount == 0) {
600 ifma6_restart = true;
601 TAILQ_REMOVE(&ifp->if_multiaddrs, ll_ifma, ifma_link);
602 MCDPRINTF("removed ll_ifma: %p from %s\n", ll_ifma, ifp->if_xname);
603 if_freemulti(ll_ifma);
606 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
607 if (ifa->ifa_addr->sa_family != AF_INET6)
610 LIST_FOREACH_SAFE(imm, &ifa6->ia6_memberships,
611 i6mm_chain, imm_tmp) {
612 if (inm == imm->i6mm_maddr) {
613 LIST_REMOVE(imm, i6mm_chain);
614 free(imm, M_IP6MADDR);
621 in6m_release_deferred(struct in6_multi *inm)
623 struct in6_multi_head tmp;
625 IN6_MULTI_LIST_LOCK_ASSERT();
626 KASSERT(inm->in6m_refcount > 0, ("refcount == %d inm: %p", inm->in6m_refcount, inm));
627 if (--inm->in6m_refcount == 0) {
628 in6m_disconnect(inm);
630 inm->in6m_ifma->ifma_protospec = NULL;
631 MPASS(inm->in6m_ifma->ifma_llifma == NULL);
632 SLIST_INSERT_HEAD(&tmp, inm, in6m_nrele);
633 in6m_release_list_deferred(&tmp);
638 in6m_release_task(void *arg __unused)
640 struct in6_multi_head in6m_free_tmp;
641 struct in6_multi *inm, *tinm;
643 SLIST_INIT(&in6m_free_tmp);
644 mtx_lock(&in6_multi_free_mtx);
645 SLIST_CONCAT(&in6m_free_tmp, &in6m_free_list, in6_multi, in6m_nrele);
646 mtx_unlock(&in6_multi_free_mtx);
648 SLIST_FOREACH_SAFE(inm, &in6m_free_tmp, in6m_nrele, tinm) {
649 SLIST_REMOVE_HEAD(&in6m_free_tmp, in6m_nrele);
656 * Clear recorded source entries for a group.
657 * Used by the MLD code. Caller must hold the IN6_MULTI lock.
658 * FIXME: Should reap.
661 in6m_clear_recorded(struct in6_multi *inm)
663 struct ip6_msource *ims;
665 IN6_MULTI_LIST_LOCK_ASSERT();
667 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
670 --inm->in6m_st[1].iss_rec;
673 KASSERT(inm->in6m_st[1].iss_rec == 0,
674 ("%s: iss_rec %d not 0", __func__, inm->in6m_st[1].iss_rec));
678 * Record a source as pending for a Source-Group MLDv2 query.
679 * This lives here as it modifies the shared tree.
681 * inm is the group descriptor.
682 * naddr is the address of the source to record in network-byte order.
684 * If the net.inet6.mld.sgalloc sysctl is non-zero, we will
685 * lazy-allocate a source node in response to an SG query.
686 * Otherwise, no allocation is performed. This saves some memory
687 * with the trade-off that the source will not be reported to the
688 * router if joined in the window between the query response and
689 * the group actually being joined on the local host.
691 * VIMAGE: XXX: Currently the mld_sgalloc feature has been removed.
692 * This turns off the allocation of a recorded source entry if
693 * the group has not been joined.
695 * Return 0 if the source didn't exist or was already marked as recorded.
696 * Return 1 if the source was marked as recorded by this function.
697 * Return <0 if any error occurred (negated errno code).
700 in6m_record_source(struct in6_multi *inm, const struct in6_addr *addr)
702 struct ip6_msource find;
703 struct ip6_msource *ims, *nims;
705 IN6_MULTI_LIST_LOCK_ASSERT();
707 find.im6s_addr = *addr;
708 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
709 if (ims && ims->im6s_stp)
712 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
714 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
718 nims->im6s_addr = find.im6s_addr;
719 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
725 * Mark the source as recorded and update the recorded
729 ++inm->in6m_st[1].iss_rec;
735 * Return a pointer to an in6_msource owned by an in6_mfilter,
736 * given its source address.
737 * Lazy-allocate if needed. If this is a new entry its filter state is
740 * imf is the filter set being modified.
741 * addr is the source address.
743 * SMPng: May be called with locks held; malloc must not block.
746 im6f_get_source(struct in6_mfilter *imf, const struct sockaddr_in6 *psin,
747 struct in6_msource **plims)
749 struct ip6_msource find;
750 struct ip6_msource *ims, *nims;
751 struct in6_msource *lims;
758 find.im6s_addr = psin->sin6_addr;
759 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
760 lims = (struct in6_msource *)ims;
762 if (imf->im6f_nsrc == in6_mcast_maxsocksrc)
764 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
768 lims = (struct in6_msource *)nims;
769 lims->im6s_addr = find.im6s_addr;
770 lims->im6sl_st[0] = MCAST_UNDEFINED;
771 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
781 * Graft a source entry into an existing socket-layer filter set,
782 * maintaining any required invariants and checking allocations.
784 * The source is marked as being in the new filter mode at t1.
786 * Return the pointer to the new node, otherwise return NULL.
788 static struct in6_msource *
789 im6f_graft(struct in6_mfilter *imf, const uint8_t st1,
790 const struct sockaddr_in6 *psin)
792 struct ip6_msource *nims;
793 struct in6_msource *lims;
795 nims = malloc(sizeof(struct in6_msource), M_IN6MFILTER,
799 lims = (struct in6_msource *)nims;
800 lims->im6s_addr = psin->sin6_addr;
801 lims->im6sl_st[0] = MCAST_UNDEFINED;
802 lims->im6sl_st[1] = st1;
803 RB_INSERT(ip6_msource_tree, &imf->im6f_sources, nims);
810 * Prune a source entry from an existing socket-layer filter set,
811 * maintaining any required invariants and checking allocations.
813 * The source is marked as being left at t1, it is not freed.
815 * Return 0 if no error occurred, otherwise return an errno value.
818 im6f_prune(struct in6_mfilter *imf, const struct sockaddr_in6 *psin)
820 struct ip6_msource find;
821 struct ip6_msource *ims;
822 struct in6_msource *lims;
824 find.im6s_addr = psin->sin6_addr;
825 ims = RB_FIND(ip6_msource_tree, &imf->im6f_sources, &find);
828 lims = (struct in6_msource *)ims;
829 lims->im6sl_st[1] = MCAST_UNDEFINED;
834 * Revert socket-layer filter set deltas at t1 to t0 state.
837 im6f_rollback(struct in6_mfilter *imf)
839 struct ip6_msource *ims, *tims;
840 struct in6_msource *lims;
842 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
843 lims = (struct in6_msource *)ims;
844 if (lims->im6sl_st[0] == lims->im6sl_st[1]) {
845 /* no change at t1 */
847 } else if (lims->im6sl_st[0] != MCAST_UNDEFINED) {
848 /* revert change to existing source at t1 */
849 lims->im6sl_st[1] = lims->im6sl_st[0];
851 /* revert source added t1 */
852 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
853 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
854 free(ims, M_IN6MFILTER);
858 imf->im6f_st[1] = imf->im6f_st[0];
862 * Mark socket-layer filter set as INCLUDE {} at t1.
865 im6f_leave(struct in6_mfilter *imf)
867 struct ip6_msource *ims;
868 struct in6_msource *lims;
870 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
871 lims = (struct in6_msource *)ims;
872 lims->im6sl_st[1] = MCAST_UNDEFINED;
874 imf->im6f_st[1] = MCAST_INCLUDE;
878 * Mark socket-layer filter set deltas as committed.
881 im6f_commit(struct in6_mfilter *imf)
883 struct ip6_msource *ims;
884 struct in6_msource *lims;
886 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
887 lims = (struct in6_msource *)ims;
888 lims->im6sl_st[0] = lims->im6sl_st[1];
890 imf->im6f_st[0] = imf->im6f_st[1];
894 * Reap unreferenced sources from socket-layer filter set.
897 im6f_reap(struct in6_mfilter *imf)
899 struct ip6_msource *ims, *tims;
900 struct in6_msource *lims;
902 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
903 lims = (struct in6_msource *)ims;
904 if ((lims->im6sl_st[0] == MCAST_UNDEFINED) &&
905 (lims->im6sl_st[1] == MCAST_UNDEFINED)) {
906 CTR2(KTR_MLD, "%s: free lims %p", __func__, ims);
907 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
908 free(ims, M_IN6MFILTER);
915 * Purge socket-layer filter set.
918 im6f_purge(struct in6_mfilter *imf)
920 struct ip6_msource *ims, *tims;
922 RB_FOREACH_SAFE(ims, ip6_msource_tree, &imf->im6f_sources, tims) {
923 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
924 RB_REMOVE(ip6_msource_tree, &imf->im6f_sources, ims);
925 free(ims, M_IN6MFILTER);
928 imf->im6f_st[0] = imf->im6f_st[1] = MCAST_UNDEFINED;
929 KASSERT(RB_EMPTY(&imf->im6f_sources),
930 ("%s: im6f_sources not empty", __func__));
934 * Look up a source filter entry for a multicast group.
936 * inm is the group descriptor to work with.
937 * addr is the IPv6 address to look up.
938 * noalloc may be non-zero to suppress allocation of sources.
939 * *pims will be set to the address of the retrieved or allocated source.
941 * SMPng: NOTE: may be called with locks held.
942 * Return 0 if successful, otherwise return a non-zero error code.
945 in6m_get_source(struct in6_multi *inm, const struct in6_addr *addr,
946 const int noalloc, struct ip6_msource **pims)
948 struct ip6_msource find;
949 struct ip6_msource *ims, *nims;
951 char ip6tbuf[INET6_ADDRSTRLEN];
954 find.im6s_addr = *addr;
955 ims = RB_FIND(ip6_msource_tree, &inm->in6m_srcs, &find);
956 if (ims == NULL && !noalloc) {
957 if (inm->in6m_nsrc == in6_mcast_maxgrpsrc)
959 nims = malloc(sizeof(struct ip6_msource), M_IP6MSOURCE,
963 nims->im6s_addr = *addr;
964 RB_INSERT(ip6_msource_tree, &inm->in6m_srcs, nims);
967 CTR3(KTR_MLD, "%s: allocated %s as %p", __func__,
968 ip6_sprintf(ip6tbuf, addr), ims);
976 * Merge socket-layer source into MLD-layer source.
977 * If rollback is non-zero, perform the inverse of the merge.
980 im6s_merge(struct ip6_msource *ims, const struct in6_msource *lims,
983 int n = rollback ? -1 : 1;
985 char ip6tbuf[INET6_ADDRSTRLEN];
987 ip6_sprintf(ip6tbuf, &lims->im6s_addr);
990 if (lims->im6sl_st[0] == MCAST_EXCLUDE) {
991 CTR3(KTR_MLD, "%s: t1 ex -= %d on %s", __func__, n, ip6tbuf);
992 ims->im6s_st[1].ex -= n;
993 } else if (lims->im6sl_st[0] == MCAST_INCLUDE) {
994 CTR3(KTR_MLD, "%s: t1 in -= %d on %s", __func__, n, ip6tbuf);
995 ims->im6s_st[1].in -= n;
998 if (lims->im6sl_st[1] == MCAST_EXCLUDE) {
999 CTR3(KTR_MLD, "%s: t1 ex += %d on %s", __func__, n, ip6tbuf);
1000 ims->im6s_st[1].ex += n;
1001 } else if (lims->im6sl_st[1] == MCAST_INCLUDE) {
1002 CTR3(KTR_MLD, "%s: t1 in += %d on %s", __func__, n, ip6tbuf);
1003 ims->im6s_st[1].in += n;
1008 * Atomically update the global in6_multi state, when a membership's
1009 * filter list is being updated in any way.
1011 * imf is the per-inpcb-membership group filter pointer.
1012 * A fake imf may be passed for in-kernel consumers.
1014 * XXX This is a candidate for a set-symmetric-difference style loop
1015 * which would eliminate the repeated lookup from root of ims nodes,
1016 * as they share the same key space.
1018 * If any error occurred this function will back out of refcounts
1019 * and return a non-zero value.
1022 in6m_merge(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1024 struct ip6_msource *ims, *nims;
1025 struct in6_msource *lims;
1026 int schanged, error;
1032 IN6_MULTI_LIST_LOCK_ASSERT();
1035 * Update the source filters first, as this may fail.
1036 * Maintain count of in-mode filters at t0, t1. These are
1037 * used to work out if we transition into ASM mode or not.
1038 * Maintain a count of source filters whose state was
1039 * actually modified by this operation.
1041 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1042 lims = (struct in6_msource *)ims;
1043 if (lims->im6sl_st[0] == imf->im6f_st[0]) nsrc0++;
1044 if (lims->im6sl_st[1] == imf->im6f_st[1]) nsrc1++;
1045 if (lims->im6sl_st[0] == lims->im6sl_st[1]) continue;
1046 error = in6m_get_source(inm, &lims->im6s_addr, 0, &nims);
1050 im6s_merge(nims, lims, 0);
1053 struct ip6_msource *bims;
1055 RB_FOREACH_REVERSE_FROM(ims, ip6_msource_tree, nims) {
1056 lims = (struct in6_msource *)ims;
1057 if (lims->im6sl_st[0] == lims->im6sl_st[1])
1059 (void)in6m_get_source(inm, &lims->im6s_addr, 1, &bims);
1062 im6s_merge(bims, lims, 1);
1067 CTR3(KTR_MLD, "%s: imf filters in-mode: %d at t0, %d at t1",
1068 __func__, nsrc0, nsrc1);
1070 /* Handle transition between INCLUDE {n} and INCLUDE {} on socket. */
1071 if (imf->im6f_st[0] == imf->im6f_st[1] &&
1072 imf->im6f_st[1] == MCAST_INCLUDE) {
1074 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1075 --inm->in6m_st[1].iss_in;
1079 /* Handle filter mode transition on socket. */
1080 if (imf->im6f_st[0] != imf->im6f_st[1]) {
1081 CTR3(KTR_MLD, "%s: imf transition %d to %d",
1082 __func__, imf->im6f_st[0], imf->im6f_st[1]);
1084 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
1085 CTR1(KTR_MLD, "%s: --ex on inm at t1", __func__);
1086 --inm->in6m_st[1].iss_ex;
1087 } else if (imf->im6f_st[0] == MCAST_INCLUDE) {
1088 CTR1(KTR_MLD, "%s: --in on inm at t1", __func__);
1089 --inm->in6m_st[1].iss_in;
1092 if (imf->im6f_st[1] == MCAST_EXCLUDE) {
1093 CTR1(KTR_MLD, "%s: ex++ on inm at t1", __func__);
1094 inm->in6m_st[1].iss_ex++;
1095 } else if (imf->im6f_st[1] == MCAST_INCLUDE && nsrc1 > 0) {
1096 CTR1(KTR_MLD, "%s: in++ on inm at t1", __func__);
1097 inm->in6m_st[1].iss_in++;
1102 * Track inm filter state in terms of listener counts.
1103 * If there are any exclusive listeners, stack-wide
1104 * membership is exclusive.
1105 * Otherwise, if only inclusive listeners, stack-wide is inclusive.
1106 * If no listeners remain, state is undefined at t1,
1107 * and the MLD lifecycle for this group should finish.
1109 if (inm->in6m_st[1].iss_ex > 0) {
1110 CTR1(KTR_MLD, "%s: transition to EX", __func__);
1111 inm->in6m_st[1].iss_fmode = MCAST_EXCLUDE;
1112 } else if (inm->in6m_st[1].iss_in > 0) {
1113 CTR1(KTR_MLD, "%s: transition to IN", __func__);
1114 inm->in6m_st[1].iss_fmode = MCAST_INCLUDE;
1116 CTR1(KTR_MLD, "%s: transition to UNDEF", __func__);
1117 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
1120 /* Decrement ASM listener count on transition out of ASM mode. */
1121 if (imf->im6f_st[0] == MCAST_EXCLUDE && nsrc0 == 0) {
1122 if ((imf->im6f_st[1] != MCAST_EXCLUDE) ||
1123 (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 > 0)) {
1124 CTR1(KTR_MLD, "%s: --asm on inm at t1", __func__);
1125 --inm->in6m_st[1].iss_asm;
1129 /* Increment ASM listener count on transition to ASM mode. */
1130 if (imf->im6f_st[1] == MCAST_EXCLUDE && nsrc1 == 0) {
1131 CTR1(KTR_MLD, "%s: asm++ on inm at t1", __func__);
1132 inm->in6m_st[1].iss_asm++;
1135 CTR3(KTR_MLD, "%s: merged imf %p to inm %p", __func__, imf, inm);
1140 CTR1(KTR_MLD, "%s: sources changed; reaping", __func__);
1147 * Mark an in6_multi's filter set deltas as committed.
1148 * Called by MLD after a state change has been enqueued.
1151 in6m_commit(struct in6_multi *inm)
1153 struct ip6_msource *ims;
1155 CTR2(KTR_MLD, "%s: commit inm %p", __func__, inm);
1156 CTR1(KTR_MLD, "%s: pre commit:", __func__);
1159 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
1160 ims->im6s_st[0] = ims->im6s_st[1];
1162 inm->in6m_st[0] = inm->in6m_st[1];
1166 * Reap unreferenced nodes from an in6_multi's filter set.
1169 in6m_reap(struct in6_multi *inm)
1171 struct ip6_msource *ims, *tims;
1173 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1174 if (ims->im6s_st[0].ex > 0 || ims->im6s_st[0].in > 0 ||
1175 ims->im6s_st[1].ex > 0 || ims->im6s_st[1].in > 0 ||
1178 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1179 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1180 free(ims, M_IP6MSOURCE);
1186 * Purge all source nodes from an in6_multi's filter set.
1189 in6m_purge(struct in6_multi *inm)
1191 struct ip6_msource *ims, *tims;
1193 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs, tims) {
1194 CTR2(KTR_MLD, "%s: free ims %p", __func__, ims);
1195 RB_REMOVE(ip6_msource_tree, &inm->in6m_srcs, ims);
1196 free(ims, M_IP6MSOURCE);
1199 /* Free state-change requests that might be queued. */
1200 mbufq_drain(&inm->in6m_scq);
1204 * Join a multicast address w/o sources.
1205 * KAME compatibility entry point.
1207 * SMPng: Assume no mc locks held by caller.
1210 in6_joingroup(struct ifnet *ifp, const struct in6_addr *mcaddr,
1211 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1217 error = in6_joingroup_locked(ifp, mcaddr, NULL, pinm, delay);
1223 * Join a multicast group; real entry point.
1225 * Only preserves atomicity at inm level.
1226 * NOTE: imf argument cannot be const due to sys/tree.h limitations.
1228 * If the MLD downcall fails, the group is not joined, and an error
1232 in6_joingroup_locked(struct ifnet *ifp, const struct in6_addr *mcaddr,
1233 /*const*/ struct in6_mfilter *imf, struct in6_multi **pinm,
1236 struct in6_mfilter timf;
1237 struct in6_multi *inm;
1238 struct ifmultiaddr *ifma;
1241 char ip6tbuf[INET6_ADDRSTRLEN];
1246 * Sanity: Check scope zone ID was set for ifp, if and
1247 * only if group is scoped to an interface.
1249 KASSERT(IN6_IS_ADDR_MULTICAST(mcaddr),
1250 ("%s: not a multicast address", __func__));
1251 if (IN6_IS_ADDR_MC_LINKLOCAL(mcaddr) ||
1252 IN6_IS_ADDR_MC_INTFACELOCAL(mcaddr)) {
1253 KASSERT(mcaddr->s6_addr16[1] != 0,
1254 ("%s: scope zone ID not set", __func__));
1258 IN6_MULTI_LOCK_ASSERT();
1259 IN6_MULTI_LIST_UNLOCK_ASSERT();
1261 CTR4(KTR_MLD, "%s: join %s on %p(%s))", __func__,
1262 ip6_sprintf(ip6tbuf, mcaddr), ifp, if_name(ifp));
1268 * If no imf was specified (i.e. kernel consumer),
1269 * fake one up and assume it is an ASM join.
1272 im6f_init(&timf, MCAST_UNDEFINED, MCAST_EXCLUDE);
1275 error = in6_getmulti(ifp, mcaddr, &inm);
1277 CTR1(KTR_MLD, "%s: in6_getmulti() failure", __func__);
1281 IN6_MULTI_LIST_LOCK();
1282 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1283 error = in6m_merge(inm, imf);
1285 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1286 goto out_in6m_release;
1289 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1290 error = mld_change_state(inm, delay);
1292 CTR1(KTR_MLD, "%s: failed to update source", __func__);
1293 goto out_in6m_release;
1298 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1300 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1301 if (ifma->ifma_protospec == inm) {
1302 ifma->ifma_protospec = NULL;
1306 in6m_release_deferred(inm);
1307 IF_ADDR_RUNLOCK(ifp);
1311 IN6_MULTI_LIST_UNLOCK();
1316 * Leave a multicast group; unlocked entry point.
1319 in6_leavegroup(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1324 error = in6_leavegroup_locked(inm, imf);
1330 * Leave a multicast group; real entry point.
1331 * All source filters will be expunged.
1333 * Only preserves atomicity at inm level.
1335 * Holding the write lock for the INP which contains imf
1336 * is highly advisable. We can't assert for it as imf does not
1337 * contain a back-pointer to the owning inp.
1339 * Note: This is not the same as in6m_release(*) as this function also
1340 * makes a state change downcall into MLD.
1343 in6_leavegroup_locked(struct in6_multi *inm, /*const*/ struct in6_mfilter *imf)
1345 struct in6_mfilter timf;
1349 char ip6tbuf[INET6_ADDRSTRLEN];
1354 IN6_MULTI_LOCK_ASSERT();
1356 CTR5(KTR_MLD, "%s: leave inm %p, %s/%s, imf %p", __func__,
1357 inm, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1358 (in6m_is_ifp_detached(inm) ? "null" : if_name(inm->in6m_ifp)),
1362 * If no imf was specified (i.e. kernel consumer),
1363 * fake one up and assume it is an ASM join.
1366 im6f_init(&timf, MCAST_EXCLUDE, MCAST_UNDEFINED);
1371 * Begin state merge transaction at MLD layer.
1373 * As this particular invocation should not cause any memory
1374 * to be allocated, and there is no opportunity to roll back
1375 * the transaction, it MUST NOT fail.
1378 ifp = inm->in6m_ifp;
1379 IN6_MULTI_LIST_LOCK();
1380 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1381 error = in6m_merge(inm, imf);
1382 KASSERT(error == 0, ("%s: failed to merge inm state", __func__));
1384 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1385 error = mld_change_state(inm, 0);
1387 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1389 CTR2(KTR_MLD, "%s: dropping ref on %p", __func__, inm);
1392 in6m_release_deferred(inm);
1394 IF_ADDR_WUNLOCK(ifp);
1395 IN6_MULTI_LIST_UNLOCK();
1402 * Block or unblock an ASM multicast source on an inpcb.
1403 * This implements the delta-based API described in RFC 3678.
1405 * The delta-based API applies only to exclusive-mode memberships.
1406 * An MLD downcall will be performed.
1408 * SMPng: NOTE: Must take Giant as a join may create a new ifma.
1410 * Return 0 if successful, otherwise return an appropriate error code.
1413 in6p_block_unblock_source(struct inpcb *inp, struct sockopt *sopt)
1415 struct group_source_req gsr;
1416 sockunion_t *gsa, *ssa;
1418 struct in6_mfilter *imf;
1419 struct ip6_moptions *imo;
1420 struct in6_msource *ims;
1421 struct in6_multi *inm;
1426 char ip6tbuf[INET6_ADDRSTRLEN];
1433 memset(&gsr, 0, sizeof(struct group_source_req));
1434 gsa = (sockunion_t *)&gsr.gsr_group;
1435 ssa = (sockunion_t *)&gsr.gsr_source;
1437 switch (sopt->sopt_name) {
1438 case MCAST_BLOCK_SOURCE:
1439 case MCAST_UNBLOCK_SOURCE:
1440 error = sooptcopyin(sopt, &gsr,
1441 sizeof(struct group_source_req),
1442 sizeof(struct group_source_req));
1446 if (gsa->sin6.sin6_family != AF_INET6 ||
1447 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1450 if (ssa->sin6.sin6_family != AF_INET6 ||
1451 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1454 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1455 return (EADDRNOTAVAIL);
1457 ifp = ifnet_byindex(gsr.gsr_interface);
1459 if (sopt->sopt_name == MCAST_BLOCK_SOURCE)
1464 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1465 __func__, sopt->sopt_name);
1466 return (EOPNOTSUPP);
1470 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1473 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1476 * Check if we are actually a member of this group.
1478 imo = in6p_findmoptions(inp);
1479 idx = im6o_match_group(imo, ifp, &gsa->sa);
1480 if (idx == -1 || imo->im6o_mfilters == NULL) {
1481 error = EADDRNOTAVAIL;
1482 goto out_in6p_locked;
1485 KASSERT(imo->im6o_mfilters != NULL,
1486 ("%s: im6o_mfilters not allocated", __func__));
1487 imf = &imo->im6o_mfilters[idx];
1488 inm = imo->im6o_membership[idx];
1491 * Attempting to use the delta-based API on an
1492 * non exclusive-mode membership is an error.
1494 fmode = imf->im6f_st[0];
1495 if (fmode != MCAST_EXCLUDE) {
1497 goto out_in6p_locked;
1501 * Deal with error cases up-front:
1502 * Asked to block, but already blocked; or
1503 * Asked to unblock, but nothing to unblock.
1504 * If adding a new block entry, allocate it.
1506 ims = im6o_match_source(imo, idx, &ssa->sa);
1507 if ((ims != NULL && doblock) || (ims == NULL && !doblock)) {
1508 CTR3(KTR_MLD, "%s: source %s %spresent", __func__,
1509 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
1510 doblock ? "" : "not ");
1511 error = EADDRNOTAVAIL;
1512 goto out_in6p_locked;
1515 INP_WLOCK_ASSERT(inp);
1518 * Begin state merge transaction at socket layer.
1521 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
1522 ims = im6f_graft(imf, fmode, &ssa->sin6);
1526 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
1527 error = im6f_prune(imf, &ssa->sin6);
1531 CTR1(KTR_MLD, "%s: merge imf state failed", __func__);
1532 goto out_im6f_rollback;
1536 * Begin state merge transaction at MLD layer.
1538 IN6_MULTI_LIST_LOCK();
1539 CTR1(KTR_MLD, "%s: merge inm state", __func__);
1540 error = in6m_merge(inm, imf);
1542 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
1544 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
1545 error = mld_change_state(inm, 0);
1547 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
1550 IN6_MULTI_LIST_UNLOCK();
1566 * Given an inpcb, return its multicast options structure pointer. Accepts
1567 * an unlocked inpcb pointer, but will return it locked. May sleep.
1569 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
1570 * SMPng: NOTE: Returns with the INP write lock held.
1572 static struct ip6_moptions *
1573 in6p_findmoptions(struct inpcb *inp)
1575 struct ip6_moptions *imo;
1576 struct in6_multi **immp;
1577 struct in6_mfilter *imfp;
1581 if (inp->in6p_moptions != NULL)
1582 return (inp->in6p_moptions);
1586 imo = malloc(sizeof(*imo), M_IP6MOPTS, M_WAITOK);
1587 immp = malloc(sizeof(*immp) * IPV6_MIN_MEMBERSHIPS, M_IP6MOPTS,
1589 imfp = malloc(sizeof(struct in6_mfilter) * IPV6_MIN_MEMBERSHIPS,
1590 M_IN6MFILTER, M_WAITOK);
1592 imo->im6o_multicast_ifp = NULL;
1593 imo->im6o_multicast_hlim = V_ip6_defmcasthlim;
1594 imo->im6o_multicast_loop = in6_mcast_loop;
1595 imo->im6o_num_memberships = 0;
1596 imo->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1597 imo->im6o_membership = immp;
1599 /* Initialize per-group source filters. */
1600 for (idx = 0; idx < IPV6_MIN_MEMBERSHIPS; idx++)
1601 im6f_init(&imfp[idx], MCAST_UNDEFINED, MCAST_EXCLUDE);
1602 imo->im6o_mfilters = imfp;
1605 if (inp->in6p_moptions != NULL) {
1606 free(imfp, M_IN6MFILTER);
1607 free(immp, M_IP6MOPTS);
1608 free(imo, M_IP6MOPTS);
1609 return (inp->in6p_moptions);
1611 inp->in6p_moptions = imo;
1616 * Discard the IPv6 multicast options (and source filters).
1618 * SMPng: NOTE: assumes INP write lock is held.
1621 ip6_freemoptions(struct ip6_moptions *imo, struct inpcbinfo *pcbinfo)
1623 struct in6_mfilter *imf;
1624 size_t idx, nmships;
1629 INP_INFO_LOCK_ASSERT(pcbinfo);
1630 wlock = INP_INFO_WLOCKED(pcbinfo);
1632 INP_INFO_WUNLOCK(pcbinfo);
1634 INP_INFO_RUNLOCK(pcbinfo);
1636 nmships = imo->im6o_num_memberships;
1637 for (idx = 0; idx < nmships; ++idx) {
1638 imf = imo->im6o_mfilters ? &imo->im6o_mfilters[idx] : NULL;
1641 /* XXX this will thrash the lock(s) */
1642 (void)in6_leavegroup(imo->im6o_membership[idx], imf);
1647 if (imo->im6o_mfilters)
1648 free(imo->im6o_mfilters, M_IN6MFILTER);
1649 free(imo->im6o_membership, M_IP6MOPTS);
1650 free(imo, M_IP6MOPTS);
1652 INP_INFO_WLOCK(pcbinfo);
1654 INP_INFO_RLOCK(pcbinfo);
1658 * Atomically get source filters on a socket for an IPv6 multicast group.
1659 * Called with INP lock held; returns with lock released.
1662 in6p_get_source_filters(struct inpcb *inp, struct sockopt *sopt)
1664 struct __msfilterreq msfr;
1667 struct ip6_moptions *imo;
1668 struct in6_mfilter *imf;
1669 struct ip6_msource *ims;
1670 struct in6_msource *lims;
1671 struct sockaddr_in6 *psin;
1672 struct sockaddr_storage *ptss;
1673 struct sockaddr_storage *tss;
1675 size_t idx, nsrcs, ncsrcs;
1677 INP_WLOCK_ASSERT(inp);
1679 imo = inp->in6p_moptions;
1680 KASSERT(imo != NULL, ("%s: null ip6_moptions", __func__));
1684 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
1685 sizeof(struct __msfilterreq));
1689 if (msfr.msfr_group.ss_family != AF_INET6 ||
1690 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
1693 gsa = (sockunion_t *)&msfr.msfr_group;
1694 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
1697 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
1698 return (EADDRNOTAVAIL);
1699 ifp = ifnet_byindex(msfr.msfr_ifindex);
1701 return (EADDRNOTAVAIL);
1702 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
1707 * Lookup group on the socket.
1709 idx = im6o_match_group(imo, ifp, &gsa->sa);
1710 if (idx == -1 || imo->im6o_mfilters == NULL) {
1712 return (EADDRNOTAVAIL);
1714 imf = &imo->im6o_mfilters[idx];
1717 * Ignore memberships which are in limbo.
1719 if (imf->im6f_st[1] == MCAST_UNDEFINED) {
1723 msfr.msfr_fmode = imf->im6f_st[1];
1726 * If the user specified a buffer, copy out the source filter
1727 * entries to userland gracefully.
1728 * We only copy out the number of entries which userland
1729 * has asked for, but we always tell userland how big the
1730 * buffer really needs to be.
1732 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
1733 msfr.msfr_nsrcs = in6_mcast_maxsocksrc;
1735 if (msfr.msfr_srcs != NULL && msfr.msfr_nsrcs > 0) {
1736 tss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
1737 M_TEMP, M_NOWAIT | M_ZERO);
1745 * Count number of sources in-mode at t0.
1746 * If buffer space exists and remains, copy out source entries.
1748 nsrcs = msfr.msfr_nsrcs;
1751 RB_FOREACH(ims, ip6_msource_tree, &imf->im6f_sources) {
1752 lims = (struct in6_msource *)ims;
1753 if (lims->im6sl_st[0] == MCAST_UNDEFINED ||
1754 lims->im6sl_st[0] != imf->im6f_st[0])
1757 if (tss != NULL && nsrcs > 0) {
1758 psin = (struct sockaddr_in6 *)ptss;
1759 psin->sin6_family = AF_INET6;
1760 psin->sin6_len = sizeof(struct sockaddr_in6);
1761 psin->sin6_addr = lims->im6s_addr;
1762 psin->sin6_port = 0;
1771 error = copyout(tss, msfr.msfr_srcs,
1772 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
1778 msfr.msfr_nsrcs = ncsrcs;
1779 error = sooptcopyout(sopt, &msfr, sizeof(struct __msfilterreq));
1785 * Return the IP multicast options in response to user getsockopt().
1788 ip6_getmoptions(struct inpcb *inp, struct sockopt *sopt)
1790 struct ip6_moptions *im6o;
1795 im6o = inp->in6p_moptions;
1797 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
1798 * or is a divert socket, reject it.
1800 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
1801 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
1802 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM)) {
1804 return (EOPNOTSUPP);
1808 switch (sopt->sopt_name) {
1809 case IPV6_MULTICAST_IF:
1810 if (im6o == NULL || im6o->im6o_multicast_ifp == NULL) {
1813 optval = im6o->im6o_multicast_ifp->if_index;
1816 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1819 case IPV6_MULTICAST_HOPS:
1821 optval = V_ip6_defmcasthlim;
1823 optval = im6o->im6o_multicast_hlim;
1825 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1828 case IPV6_MULTICAST_LOOP:
1830 optval = in6_mcast_loop; /* XXX VIMAGE */
1832 optval = im6o->im6o_multicast_loop;
1834 error = sooptcopyout(sopt, &optval, sizeof(u_int));
1839 error = EADDRNOTAVAIL;
1842 error = in6p_get_source_filters(inp, sopt);
1848 error = ENOPROTOOPT;
1852 INP_UNLOCK_ASSERT(inp);
1858 * Look up the ifnet to use for a multicast group membership,
1859 * given the address of an IPv6 group.
1861 * This routine exists to support legacy IPv6 multicast applications.
1863 * If inp is non-NULL, use this socket's current FIB number for any
1864 * required FIB lookup. Look up the group address in the unicast FIB,
1865 * and use its ifp; usually, this points to the default next-hop.
1866 * If the FIB lookup fails, return NULL.
1868 * FUTURE: Support multiple forwarding tables for IPv6.
1870 * Returns NULL if no ifp could be found.
1872 static struct ifnet *
1873 in6p_lookup_mcast_ifp(const struct inpcb *in6p,
1874 const struct sockaddr_in6 *gsin6)
1876 struct nhop6_basic nh6;
1877 struct in6_addr dst;
1881 KASSERT(in6p->inp_vflag & INP_IPV6,
1882 ("%s: not INP_IPV6 inpcb", __func__));
1883 KASSERT(gsin6->sin6_family == AF_INET6,
1884 ("%s: not AF_INET6 group", __func__));
1886 in6_splitscope(&gsin6->sin6_addr, &dst, &scopeid);
1887 fibnum = in6p ? in6p->inp_inc.inc_fibnum : RT_DEFAULT_FIB;
1888 if (fib6_lookup_nh_basic(fibnum, &dst, scopeid, 0, 0, &nh6) != 0)
1891 return (nh6.nh_ifp);
1895 * Join an IPv6 multicast group, possibly with a source.
1897 * FIXME: The KAME use of the unspecified address (::)
1898 * to join *all* multicast groups is currently unsupported.
1901 in6p_join_group(struct inpcb *inp, struct sockopt *sopt)
1903 struct group_source_req gsr;
1904 sockunion_t *gsa, *ssa;
1906 struct in6_mfilter *imf;
1907 struct ip6_moptions *imo;
1908 struct in6_multi *inm;
1909 struct in6_msource *lims;
1919 memset(&gsr, 0, sizeof(struct group_source_req));
1920 gsa = (sockunion_t *)&gsr.gsr_group;
1921 gsa->ss.ss_family = AF_UNSPEC;
1922 ssa = (sockunion_t *)&gsr.gsr_source;
1923 ssa->ss.ss_family = AF_UNSPEC;
1926 * Chew everything into struct group_source_req.
1927 * Overwrite the port field if present, as the sockaddr
1928 * being copied in may be matched with a binary comparison.
1929 * Ignore passed-in scope ID.
1931 switch (sopt->sopt_name) {
1932 case IPV6_JOIN_GROUP: {
1933 struct ipv6_mreq mreq;
1935 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
1936 sizeof(struct ipv6_mreq));
1940 gsa->sin6.sin6_family = AF_INET6;
1941 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
1942 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
1944 if (mreq.ipv6mr_interface == 0) {
1945 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
1947 if (V_if_index < mreq.ipv6mr_interface)
1948 return (EADDRNOTAVAIL);
1949 ifp = ifnet_byindex(mreq.ipv6mr_interface);
1951 CTR3(KTR_MLD, "%s: ipv6mr_interface = %d, ifp = %p",
1952 __func__, mreq.ipv6mr_interface, ifp);
1955 case MCAST_JOIN_GROUP:
1956 case MCAST_JOIN_SOURCE_GROUP:
1957 if (sopt->sopt_name == MCAST_JOIN_GROUP) {
1958 error = sooptcopyin(sopt, &gsr,
1959 sizeof(struct group_req),
1960 sizeof(struct group_req));
1961 } else if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1962 error = sooptcopyin(sopt, &gsr,
1963 sizeof(struct group_source_req),
1964 sizeof(struct group_source_req));
1969 if (gsa->sin6.sin6_family != AF_INET6 ||
1970 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1973 if (sopt->sopt_name == MCAST_JOIN_SOURCE_GROUP) {
1974 if (ssa->sin6.sin6_family != AF_INET6 ||
1975 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
1977 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
1980 * TODO: Validate embedded scope ID in source
1981 * list entry against passed-in ifp, if and only
1982 * if source list filter entry is iface or node local.
1984 in6_clearscope(&ssa->sin6.sin6_addr);
1985 ssa->sin6.sin6_port = 0;
1986 ssa->sin6.sin6_scope_id = 0;
1989 if (gsr.gsr_interface == 0 || V_if_index < gsr.gsr_interface)
1990 return (EADDRNOTAVAIL);
1991 ifp = ifnet_byindex(gsr.gsr_interface);
1995 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
1996 __func__, sopt->sopt_name);
1997 return (EOPNOTSUPP);
2001 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2004 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0)
2005 return (EADDRNOTAVAIL);
2007 gsa->sin6.sin6_port = 0;
2008 gsa->sin6.sin6_scope_id = 0;
2011 * Always set the scope zone ID on memberships created from userland.
2012 * Use the passed-in ifp to do this.
2013 * XXX The in6_setscope() return value is meaningless.
2014 * XXX SCOPE6_LOCK() is taken by in6_setscope().
2016 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2018 imo = in6p_findmoptions(inp);
2019 idx = im6o_match_group(imo, ifp, &gsa->sa);
2023 inm = imo->im6o_membership[idx];
2024 imf = &imo->im6o_mfilters[idx];
2025 if (ssa->ss.ss_family != AF_UNSPEC) {
2027 * MCAST_JOIN_SOURCE_GROUP on an exclusive membership
2028 * is an error. On an existing inclusive membership,
2029 * it just adds the source to the filter list.
2031 if (imf->im6f_st[1] != MCAST_INCLUDE) {
2033 goto out_in6p_locked;
2036 * Throw out duplicates.
2038 * XXX FIXME: This makes a naive assumption that
2039 * even if entries exist for *ssa in this imf,
2040 * they will be rejected as dupes, even if they
2041 * are not valid in the current mode (in-mode).
2043 * in6_msource is transactioned just as for anything
2044 * else in SSM -- but note naive use of in6m_graft()
2045 * below for allocating new filter entries.
2047 * This is only an issue if someone mixes the
2048 * full-state SSM API with the delta-based API,
2049 * which is discouraged in the relevant RFCs.
2051 lims = im6o_match_source(imo, idx, &ssa->sa);
2052 if (lims != NULL /*&&
2053 lims->im6sl_st[1] == MCAST_INCLUDE*/) {
2054 error = EADDRNOTAVAIL;
2055 goto out_in6p_locked;
2059 * MCAST_JOIN_GROUP alone, on any existing membership,
2060 * is rejected, to stop the same inpcb tying up
2061 * multiple refs to the in_multi.
2062 * On an existing inclusive membership, this is also
2063 * an error; if you want to change filter mode,
2064 * you must use the userland API setsourcefilter().
2065 * XXX We don't reject this for imf in UNDEFINED
2066 * state at t1, because allocation of a filter
2067 * is atomic with allocation of a membership.
2070 goto out_in6p_locked;
2075 * Begin state merge transaction at socket layer.
2077 INP_WLOCK_ASSERT(inp);
2080 if (imo->im6o_num_memberships == imo->im6o_max_memberships) {
2081 error = im6o_grow(imo);
2083 goto out_in6p_locked;
2086 * Allocate the new slot upfront so we can deal with
2087 * grafting the new source filter in same code path
2088 * as for join-source on existing membership.
2090 idx = imo->im6o_num_memberships;
2091 imo->im6o_membership[idx] = NULL;
2092 imo->im6o_num_memberships++;
2093 KASSERT(imo->im6o_mfilters != NULL,
2094 ("%s: im6f_mfilters vector was not allocated", __func__));
2095 imf = &imo->im6o_mfilters[idx];
2096 KASSERT(RB_EMPTY(&imf->im6f_sources),
2097 ("%s: im6f_sources not empty", __func__));
2101 * Graft new source into filter list for this inpcb's
2102 * membership of the group. The in6_multi may not have
2103 * been allocated yet if this is a new membership, however,
2104 * the in_mfilter slot will be allocated and must be initialized.
2106 * Note: Grafting of exclusive mode filters doesn't happen
2108 * XXX: Should check for non-NULL lims (node exists but may
2109 * not be in-mode) for interop with full-state API.
2111 if (ssa->ss.ss_family != AF_UNSPEC) {
2112 /* Membership starts in IN mode */
2114 CTR1(KTR_MLD, "%s: new join w/source", __func__);
2115 im6f_init(imf, MCAST_UNDEFINED, MCAST_INCLUDE);
2117 CTR2(KTR_MLD, "%s: %s source", __func__, "allow");
2119 lims = im6f_graft(imf, MCAST_INCLUDE, &ssa->sin6);
2121 CTR1(KTR_MLD, "%s: merge imf state failed",
2127 /* No address specified; Membership starts in EX mode */
2129 CTR1(KTR_MLD, "%s: new join w/o source", __func__);
2130 im6f_init(imf, MCAST_UNDEFINED, MCAST_EXCLUDE);
2135 * Begin state merge transaction at MLD layer.
2142 error = in6_joingroup_locked(ifp, &gsa->sin6.sin6_addr, imf,
2148 imo->im6o_membership[idx] = inm;
2150 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2151 IN6_MULTI_LIST_LOCK();
2152 error = in6m_merge(inm, imf);
2154 CTR1(KTR_MLD, "%s: failed to merge inm state",
2157 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2158 error = mld_change_state(inm, 0);
2160 CTR1(KTR_MLD, "%s: failed mld downcall",
2163 IN6_MULTI_LIST_UNLOCK();
2168 if (in_pcbrele_wlocked(inp))
2181 if (error && is_new) {
2182 imo->im6o_membership[idx] = NULL;
2183 --imo->im6o_num_memberships;
2192 * Leave an IPv6 multicast group on an inpcb, possibly with a source.
2195 in6p_leave_group(struct inpcb *inp, struct sockopt *sopt)
2197 struct ipv6_mreq mreq;
2198 struct group_source_req gsr;
2199 sockunion_t *gsa, *ssa;
2201 struct in6_mfilter *imf;
2202 struct ip6_moptions *imo;
2203 struct in6_msource *ims;
2204 struct in6_multi *inm;
2207 int error, is_final;
2209 char ip6tbuf[INET6_ADDRSTRLEN];
2217 memset(&gsr, 0, sizeof(struct group_source_req));
2218 gsa = (sockunion_t *)&gsr.gsr_group;
2219 gsa->ss.ss_family = AF_UNSPEC;
2220 ssa = (sockunion_t *)&gsr.gsr_source;
2221 ssa->ss.ss_family = AF_UNSPEC;
2224 * Chew everything passed in up into a struct group_source_req
2225 * as that is easier to process.
2226 * Note: Any embedded scope ID in the multicast group passed
2227 * in by userland is ignored, the interface index is the recommended
2228 * mechanism to specify an interface; see below.
2230 switch (sopt->sopt_name) {
2231 case IPV6_LEAVE_GROUP:
2232 error = sooptcopyin(sopt, &mreq, sizeof(struct ipv6_mreq),
2233 sizeof(struct ipv6_mreq));
2236 gsa->sin6.sin6_family = AF_INET6;
2237 gsa->sin6.sin6_len = sizeof(struct sockaddr_in6);
2238 gsa->sin6.sin6_addr = mreq.ipv6mr_multiaddr;
2239 gsa->sin6.sin6_port = 0;
2240 gsa->sin6.sin6_scope_id = 0;
2241 ifindex = mreq.ipv6mr_interface;
2244 case MCAST_LEAVE_GROUP:
2245 case MCAST_LEAVE_SOURCE_GROUP:
2246 if (sopt->sopt_name == MCAST_LEAVE_GROUP) {
2247 error = sooptcopyin(sopt, &gsr,
2248 sizeof(struct group_req),
2249 sizeof(struct group_req));
2250 } else if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2251 error = sooptcopyin(sopt, &gsr,
2252 sizeof(struct group_source_req),
2253 sizeof(struct group_source_req));
2258 if (gsa->sin6.sin6_family != AF_INET6 ||
2259 gsa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2261 if (sopt->sopt_name == MCAST_LEAVE_SOURCE_GROUP) {
2262 if (ssa->sin6.sin6_family != AF_INET6 ||
2263 ssa->sin6.sin6_len != sizeof(struct sockaddr_in6))
2265 if (IN6_IS_ADDR_MULTICAST(&ssa->sin6.sin6_addr))
2268 * TODO: Validate embedded scope ID in source
2269 * list entry against passed-in ifp, if and only
2270 * if source list filter entry is iface or node local.
2272 in6_clearscope(&ssa->sin6.sin6_addr);
2274 gsa->sin6.sin6_port = 0;
2275 gsa->sin6.sin6_scope_id = 0;
2276 ifindex = gsr.gsr_interface;
2280 CTR2(KTR_MLD, "%s: unknown sopt_name %d",
2281 __func__, sopt->sopt_name);
2282 return (EOPNOTSUPP);
2286 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2290 * Validate interface index if provided. If no interface index
2291 * was provided separately, attempt to look the membership up
2292 * from the default scope as a last resort to disambiguate
2293 * the membership we are being asked to leave.
2294 * XXX SCOPE6 lock potentially taken here.
2297 if (V_if_index < ifindex)
2298 return (EADDRNOTAVAIL);
2299 ifp = ifnet_byindex(ifindex);
2301 return (EADDRNOTAVAIL);
2302 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2304 error = sa6_embedscope(&gsa->sin6, V_ip6_use_defzone);
2306 return (EADDRNOTAVAIL);
2308 * Some badly behaved applications don't pass an ifindex
2309 * or a scope ID, which is an API violation. In this case,
2310 * perform a lookup as per a v6 join.
2312 * XXX For now, stomp on zone ID for the corner case.
2313 * This is not the 'KAME way', but we need to see the ifp
2314 * directly until such time as this implementation is
2315 * refactored, assuming the scope IDs are the way to go.
2317 ifindex = ntohs(gsa->sin6.sin6_addr.s6_addr16[1]);
2319 CTR2(KTR_MLD, "%s: warning: no ifindex, looking up "
2320 "ifp for group %s.", __func__,
2321 ip6_sprintf(ip6tbuf, &gsa->sin6.sin6_addr));
2322 ifp = in6p_lookup_mcast_ifp(inp, &gsa->sin6);
2324 ifp = ifnet_byindex(ifindex);
2327 return (EADDRNOTAVAIL);
2330 CTR2(KTR_MLD, "%s: ifp = %p", __func__, ifp);
2331 KASSERT(ifp != NULL, ("%s: ifp did not resolve", __func__));
2334 * Find the membership in the membership array.
2336 imo = in6p_findmoptions(inp);
2337 idx = im6o_match_group(imo, ifp, &gsa->sa);
2339 error = EADDRNOTAVAIL;
2340 goto out_in6p_locked;
2342 inm = imo->im6o_membership[idx];
2343 imf = &imo->im6o_mfilters[idx];
2345 if (ssa->ss.ss_family != AF_UNSPEC)
2349 * Begin state merge transaction at socket layer.
2351 INP_WLOCK_ASSERT(inp);
2354 * If we were instructed only to leave a given source, do so.
2355 * MCAST_LEAVE_SOURCE_GROUP is only valid for inclusive memberships.
2360 if (imf->im6f_st[0] == MCAST_EXCLUDE) {
2361 error = EADDRNOTAVAIL;
2362 goto out_in6p_locked;
2364 ims = im6o_match_source(imo, idx, &ssa->sa);
2366 CTR3(KTR_MLD, "%s: source %p %spresent", __func__,
2367 ip6_sprintf(ip6tbuf, &ssa->sin6.sin6_addr),
2369 error = EADDRNOTAVAIL;
2370 goto out_in6p_locked;
2372 CTR2(KTR_MLD, "%s: %s source", __func__, "block");
2373 error = im6f_prune(imf, &ssa->sin6);
2375 CTR1(KTR_MLD, "%s: merge imf state failed",
2377 goto out_in6p_locked;
2382 * Begin state merge transaction at MLD layer.
2390 * Give up the multicast address record to which
2391 * the membership points.
2393 (void)in6_leavegroup_locked(inm, imf);
2395 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2396 IN6_MULTI_LIST_LOCK();
2397 error = in6m_merge(inm, imf);
2399 CTR1(KTR_MLD, "%s: failed to merge inm state",
2402 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2403 error = mld_change_state(inm, 0);
2405 CTR1(KTR_MLD, "%s: failed mld downcall",
2408 IN6_MULTI_LIST_UNLOCK();
2413 if (in_pcbrele_wlocked(inp))
2424 /* Remove the gap in the membership array. */
2425 for (++idx; idx < imo->im6o_num_memberships; ++idx) {
2426 imo->im6o_membership[idx-1] = imo->im6o_membership[idx];
2427 imo->im6o_mfilters[idx-1] = imo->im6o_mfilters[idx];
2429 imo->im6o_num_memberships--;
2438 * Select the interface for transmitting IPv6 multicast datagrams.
2440 * Either an instance of struct in6_addr or an instance of struct ipv6_mreqn
2441 * may be passed to this socket option. An address of in6addr_any or an
2442 * interface index of 0 is used to remove a previous selection.
2443 * When no interface is selected, one is chosen for every send.
2446 in6p_set_multicast_if(struct inpcb *inp, struct sockopt *sopt)
2449 struct ip6_moptions *imo;
2453 if (sopt->sopt_valsize != sizeof(u_int))
2456 error = sooptcopyin(sopt, &ifindex, sizeof(u_int), sizeof(u_int));
2459 if (V_if_index < ifindex)
2464 ifp = ifnet_byindex(ifindex);
2467 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2468 return (EADDRNOTAVAIL);
2470 imo = in6p_findmoptions(inp);
2471 imo->im6o_multicast_ifp = ifp;
2478 * Atomically set source filters on a socket for an IPv6 multicast group.
2480 * SMPng: NOTE: Potentially calls malloc(M_WAITOK) with Giant held.
2483 in6p_set_source_filters(struct inpcb *inp, struct sockopt *sopt)
2485 struct __msfilterreq msfr;
2488 struct in6_mfilter *imf;
2489 struct ip6_moptions *imo;
2490 struct in6_multi *inm;
2494 error = sooptcopyin(sopt, &msfr, sizeof(struct __msfilterreq),
2495 sizeof(struct __msfilterreq));
2499 if (msfr.msfr_nsrcs > in6_mcast_maxsocksrc)
2502 if (msfr.msfr_fmode != MCAST_EXCLUDE &&
2503 msfr.msfr_fmode != MCAST_INCLUDE)
2506 if (msfr.msfr_group.ss_family != AF_INET6 ||
2507 msfr.msfr_group.ss_len != sizeof(struct sockaddr_in6))
2510 gsa = (sockunion_t *)&msfr.msfr_group;
2511 if (!IN6_IS_ADDR_MULTICAST(&gsa->sin6.sin6_addr))
2514 gsa->sin6.sin6_port = 0; /* ignore port */
2516 if (msfr.msfr_ifindex == 0 || V_if_index < msfr.msfr_ifindex)
2517 return (EADDRNOTAVAIL);
2518 ifp = ifnet_byindex(msfr.msfr_ifindex);
2520 return (EADDRNOTAVAIL);
2521 (void)in6_setscope(&gsa->sin6.sin6_addr, ifp, NULL);
2524 * Take the INP write lock.
2525 * Check if this socket is a member of this group.
2527 imo = in6p_findmoptions(inp);
2528 idx = im6o_match_group(imo, ifp, &gsa->sa);
2529 if (idx == -1 || imo->im6o_mfilters == NULL) {
2530 error = EADDRNOTAVAIL;
2531 goto out_in6p_locked;
2533 inm = imo->im6o_membership[idx];
2534 imf = &imo->im6o_mfilters[idx];
2537 * Begin state merge transaction at socket layer.
2539 INP_WLOCK_ASSERT(inp);
2541 imf->im6f_st[1] = msfr.msfr_fmode;
2544 * Apply any new source filters, if present.
2545 * Make a copy of the user-space source vector so
2546 * that we may copy them with a single copyin. This
2547 * allows us to deal with page faults up-front.
2549 if (msfr.msfr_nsrcs > 0) {
2550 struct in6_msource *lims;
2551 struct sockaddr_in6 *psin;
2552 struct sockaddr_storage *kss, *pkss;
2557 CTR2(KTR_MLD, "%s: loading %lu source list entries",
2558 __func__, (unsigned long)msfr.msfr_nsrcs);
2559 kss = malloc(sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs,
2561 error = copyin(msfr.msfr_srcs, kss,
2562 sizeof(struct sockaddr_storage) * msfr.msfr_nsrcs);
2571 * Mark all source filters as UNDEFINED at t1.
2572 * Restore new group filter mode, as im6f_leave()
2573 * will set it to INCLUDE.
2576 imf->im6f_st[1] = msfr.msfr_fmode;
2579 * Update socket layer filters at t1, lazy-allocating
2580 * new entries. This saves a bunch of memory at the
2581 * cost of one RB_FIND() per source entry; duplicate
2582 * entries in the msfr_nsrcs vector are ignored.
2583 * If we encounter an error, rollback transaction.
2585 * XXX This too could be replaced with a set-symmetric
2586 * difference like loop to avoid walking from root
2587 * every time, as the key space is common.
2589 for (i = 0, pkss = kss; i < msfr.msfr_nsrcs; i++, pkss++) {
2590 psin = (struct sockaddr_in6 *)pkss;
2591 if (psin->sin6_family != AF_INET6) {
2592 error = EAFNOSUPPORT;
2595 if (psin->sin6_len != sizeof(struct sockaddr_in6)) {
2599 if (IN6_IS_ADDR_MULTICAST(&psin->sin6_addr)) {
2604 * TODO: Validate embedded scope ID in source
2605 * list entry against passed-in ifp, if and only
2606 * if source list filter entry is iface or node local.
2608 in6_clearscope(&psin->sin6_addr);
2609 error = im6f_get_source(imf, psin, &lims);
2612 lims->im6sl_st[1] = imf->im6f_st[1];
2618 goto out_im6f_rollback;
2620 INP_WLOCK_ASSERT(inp);
2621 IN6_MULTI_LIST_LOCK();
2624 * Begin state merge transaction at MLD layer.
2626 CTR1(KTR_MLD, "%s: merge inm state", __func__);
2627 error = in6m_merge(inm, imf);
2629 CTR1(KTR_MLD, "%s: failed to merge inm state", __func__);
2631 CTR1(KTR_MLD, "%s: doing mld downcall", __func__);
2632 error = mld_change_state(inm, 0);
2634 CTR1(KTR_MLD, "%s: failed mld downcall", __func__);
2637 IN6_MULTI_LIST_UNLOCK();
2653 * Set the IP multicast options in response to user setsockopt().
2655 * Many of the socket options handled in this function duplicate the
2656 * functionality of socket options in the regular unicast API. However,
2657 * it is not possible to merge the duplicate code, because the idempotence
2658 * of the IPv6 multicast part of the BSD Sockets API must be preserved;
2659 * the effects of these options must be treated as separate and distinct.
2661 * SMPng: XXX: Unlocked read of inp_socket believed OK.
2664 ip6_setmoptions(struct inpcb *inp, struct sockopt *sopt)
2666 struct ip6_moptions *im6o;
2672 * If socket is neither of type SOCK_RAW or SOCK_DGRAM,
2673 * or is a divert socket, reject it.
2675 if (inp->inp_socket->so_proto->pr_protocol == IPPROTO_DIVERT ||
2676 (inp->inp_socket->so_proto->pr_type != SOCK_RAW &&
2677 inp->inp_socket->so_proto->pr_type != SOCK_DGRAM))
2678 return (EOPNOTSUPP);
2680 switch (sopt->sopt_name) {
2681 case IPV6_MULTICAST_IF:
2682 error = in6p_set_multicast_if(inp, sopt);
2685 case IPV6_MULTICAST_HOPS: {
2688 if (sopt->sopt_valsize != sizeof(int)) {
2692 error = sooptcopyin(sopt, &hlim, sizeof(hlim), sizeof(int));
2695 if (hlim < -1 || hlim > 255) {
2698 } else if (hlim == -1) {
2699 hlim = V_ip6_defmcasthlim;
2701 im6o = in6p_findmoptions(inp);
2702 im6o->im6o_multicast_hlim = hlim;
2707 case IPV6_MULTICAST_LOOP: {
2711 * Set the loopback flag for outgoing multicast packets.
2712 * Must be zero or one.
2714 if (sopt->sopt_valsize != sizeof(u_int)) {
2718 error = sooptcopyin(sopt, &loop, sizeof(u_int), sizeof(u_int));
2725 im6o = in6p_findmoptions(inp);
2726 im6o->im6o_multicast_loop = loop;
2731 case IPV6_JOIN_GROUP:
2732 case MCAST_JOIN_GROUP:
2733 case MCAST_JOIN_SOURCE_GROUP:
2734 error = in6p_join_group(inp, sopt);
2737 case IPV6_LEAVE_GROUP:
2738 case MCAST_LEAVE_GROUP:
2739 case MCAST_LEAVE_SOURCE_GROUP:
2740 error = in6p_leave_group(inp, sopt);
2743 case MCAST_BLOCK_SOURCE:
2744 case MCAST_UNBLOCK_SOURCE:
2745 error = in6p_block_unblock_source(inp, sopt);
2749 error = in6p_set_source_filters(inp, sopt);
2757 INP_UNLOCK_ASSERT(inp);
2763 * Expose MLD's multicast filter mode and source list(s) to userland,
2764 * keyed by (ifindex, group).
2765 * The filter mode is written out as a uint32_t, followed by
2766 * 0..n of struct in6_addr.
2767 * For use by ifmcstat(8).
2768 * SMPng: NOTE: unlocked read of ifindex space.
2771 sysctl_ip6_mcast_filters(SYSCTL_HANDLER_ARGS)
2773 struct in6_addr mcaddr;
2774 struct in6_addr src;
2776 struct ifmultiaddr *ifma;
2777 struct in6_multi *inm;
2778 struct ip6_msource *ims;
2782 uint32_t fmode, ifindex;
2784 char ip6tbuf[INET6_ADDRSTRLEN];
2790 if (req->newptr != NULL)
2793 /* int: ifindex + 4 * 32 bits of IPv6 address */
2798 if (ifindex <= 0 || ifindex > V_if_index) {
2799 CTR2(KTR_MLD, "%s: ifindex %u out of range",
2804 memcpy(&mcaddr, &name[1], sizeof(struct in6_addr));
2805 if (!IN6_IS_ADDR_MULTICAST(&mcaddr)) {
2806 CTR2(KTR_MLD, "%s: group %s is not multicast",
2807 __func__, ip6_sprintf(ip6tbuf, &mcaddr));
2811 ifp = ifnet_byindex(ifindex);
2813 CTR2(KTR_MLD, "%s: no ifp for ifindex %u",
2818 * Internal MLD lookups require that scope/zone ID is set.
2820 (void)in6_setscope(&mcaddr, ifp, NULL);
2822 retval = sysctl_wire_old_buffer(req,
2823 sizeof(uint32_t) + (in6_mcast_maxgrpsrc * sizeof(struct in6_addr)));
2828 IN6_MULTI_LIST_LOCK();
2830 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2831 if (ifma->ifma_addr->sa_family != AF_INET6 ||
2832 ifma->ifma_protospec == NULL)
2834 inm = (struct in6_multi *)ifma->ifma_protospec;
2835 if (!IN6_ARE_ADDR_EQUAL(&inm->in6m_addr, &mcaddr))
2837 fmode = inm->in6m_st[1].iss_fmode;
2838 retval = SYSCTL_OUT(req, &fmode, sizeof(uint32_t));
2841 RB_FOREACH(ims, ip6_msource_tree, &inm->in6m_srcs) {
2842 CTR2(KTR_MLD, "%s: visit node %p", __func__, ims);
2844 * Only copy-out sources which are in-mode.
2846 if (fmode != im6s_get_mode(inm, ims, 1)) {
2847 CTR1(KTR_MLD, "%s: skip non-in-mode",
2851 src = ims->im6s_addr;
2852 retval = SYSCTL_OUT(req, &src,
2853 sizeof(struct in6_addr));
2858 IF_ADDR_RUNLOCK(ifp);
2860 IN6_MULTI_LIST_UNLOCK();
2868 static const char *in6m_modestrs[] = { "un", "in", "ex" };
2871 in6m_mode_str(const int mode)
2874 if (mode >= MCAST_UNDEFINED && mode <= MCAST_EXCLUDE)
2875 return (in6m_modestrs[mode]);
2879 static const char *in6m_statestrs[] = {
2892 in6m_state_str(const int state)
2895 if (state >= MLD_NOT_MEMBER && state <= MLD_LEAVING_MEMBER)
2896 return (in6m_statestrs[state]);
2901 * Dump an in6_multi structure to the console.
2904 in6m_print(const struct in6_multi *inm)
2907 char ip6tbuf[INET6_ADDRSTRLEN];
2909 if ((ktr_mask & KTR_MLD) == 0)
2912 printf("%s: --- begin in6m %p ---\n", __func__, inm);
2913 printf("addr %s ifp %p(%s) ifma %p\n",
2914 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2916 if_name(inm->in6m_ifp),
2918 printf("timer %u state %s refcount %u scq.len %u\n",
2920 in6m_state_str(inm->in6m_state),
2922 mbufq_len(&inm->in6m_scq));
2923 printf("mli %p nsrc %lu sctimer %u scrv %u\n",
2928 for (t = 0; t < 2; t++) {
2929 printf("t%d: fmode %s asm %u ex %u in %u rec %u\n", t,
2930 in6m_mode_str(inm->in6m_st[t].iss_fmode),
2931 inm->in6m_st[t].iss_asm,
2932 inm->in6m_st[t].iss_ex,
2933 inm->in6m_st[t].iss_in,
2934 inm->in6m_st[t].iss_rec);
2936 printf("%s: --- end in6m %p ---\n", __func__, inm);
2942 in6m_print(const struct in6_multi *inm)