2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 2009 Bruce Simpson.
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. The name of the author may not be used to endorse or promote
15 * products derived from this software without specific prior written
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * $KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
34 * Copyright (c) 1988 Stephen Deering.
35 * Copyright (c) 1992, 1993
36 * The Regents of the University of California. All rights reserved.
38 * This code is derived from software contributed to Berkeley by
39 * Stephen Deering of Stanford University.
41 * Redistribution and use in source and binary forms, with or without
42 * modification, are permitted provided that the following conditions
44 * 1. Redistributions of source code must retain the above copyright
45 * notice, this list of conditions and the following disclaimer.
46 * 2. Redistributions in binary form must reproduce the above copyright
47 * notice, this list of conditions and the following disclaimer in the
48 * documentation and/or other materials provided with the distribution.
49 * 3. Neither the name of the University nor the names of its contributors
50 * may be used to endorse or promote products derived from this software
51 * without specific prior written permission.
53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65 * @(#)igmp.c 8.1 (Berkeley) 7/19/93
68 #include <sys/cdefs.h>
69 __FBSDID("$FreeBSD$");
72 #include "opt_inet6.h"
74 #include <sys/param.h>
75 #include <sys/systm.h>
77 #include <sys/socket.h>
78 #include <sys/protosw.h>
79 #include <sys/sysctl.h>
80 #include <sys/kernel.h>
81 #include <sys/callout.h>
82 #include <sys/malloc.h>
83 #include <sys/module.h>
87 #include <net/if_var.h>
88 #include <net/route.h>
91 #include <netinet/in.h>
92 #include <netinet/in_var.h>
93 #include <netinet6/in6_var.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/mld6.h>
99 #include <netinet6/mld6_var.h>
101 #include <security/mac/mac_framework.h>
104 #define KTR_MLD KTR_INET6
107 static struct mld_ifsoftc *
108 mli_alloc_locked(struct ifnet *);
109 static void mli_delete_locked(const struct ifnet *);
110 static void mld_dispatch_packet(struct mbuf *);
111 static void mld_dispatch_queue(struct mbufq *, int);
112 static void mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
113 static void mld_fasttimo_vnet(void);
114 static int mld_handle_state_change(struct in6_multi *,
115 struct mld_ifsoftc *);
116 static int mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
119 static char * mld_rec_type_to_str(const int);
121 static void mld_set_version(struct mld_ifsoftc *, const int);
122 static void mld_slowtimo_vnet(void);
123 static int mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
124 /*const*/ struct mld_hdr *);
125 static int mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
126 /*const*/ struct mld_hdr *);
127 static void mld_v1_process_group_timer(struct in6_multi_head *,
129 static void mld_v1_process_querier_timers(struct mld_ifsoftc *);
130 static int mld_v1_transmit_report(struct in6_multi *, const int);
131 static void mld_v1_update_group(struct in6_multi *, const int);
132 static void mld_v2_cancel_link_timers(struct mld_ifsoftc *);
133 static void mld_v2_dispatch_general_query(struct mld_ifsoftc *);
135 mld_v2_encap_report(struct ifnet *, struct mbuf *);
136 static int mld_v2_enqueue_filter_change(struct mbufq *,
138 static int mld_v2_enqueue_group_record(struct mbufq *,
139 struct in6_multi *, const int, const int, const int,
141 static int mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
142 struct mbuf *, const int, const int);
143 static int mld_v2_merge_state_changes(struct in6_multi *,
145 static void mld_v2_process_group_timers(struct in6_multi_head *,
146 struct mbufq *, struct mbufq *,
147 struct in6_multi *, const int);
148 static int mld_v2_process_group_query(struct in6_multi *,
149 struct mld_ifsoftc *mli, int, struct mbuf *, const int);
150 static int sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
151 static int sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
154 * Normative references: RFC 2710, RFC 3590, RFC 3810.
157 * * The MLD subsystem lock ends up being system-wide for the moment,
158 * but could be per-VIMAGE later on.
159 * * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
160 * Any may be taken independently; if any are held at the same
161 * time, the above lock order must be followed.
162 * * IN6_MULTI_LOCK covers in_multi.
163 * * MLD_LOCK covers per-link state and any global variables in this file.
164 * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
165 * per-link state iterators.
168 * A special case for IPv6 is the in6_setscope() routine. ip6_output()
169 * will not accept an ifp; it wants an embedded scope ID, unlike
170 * ip_output(), which happily takes the ifp given to it. The embedded
171 * scope ID is only used by MLD to select the outgoing interface.
173 * During interface attach and detach, MLD will take MLD_LOCK *after*
174 * the IF_AFDATA_LOCK.
175 * As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
176 * it with MLD_LOCK held without triggering an LOR. A netisr with indirect
177 * dispatch could work around this, but we'd rather not do that, as it
178 * can introduce other races.
180 * As such, we exploit the fact that the scope ID is just the interface
181 * index, and embed it in the IPv6 destination address accordingly.
182 * This is potentially NOT VALID for MLDv1 reports, as they
183 * are always sent to the multicast group itself; as MLDv2
184 * reports are always sent to ff02::16, this is not an issue
185 * when MLDv2 is in use.
187 * This does not however eliminate the LOR when ip6_output() itself
188 * calls in6_setscope() internally whilst MLD_LOCK is held. This will
189 * trigger a LOR warning in WITNESS when the ifnet is detached.
191 * The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
192 * how it's used across the network stack. Here we're simply exploiting
193 * the fact that MLD runs at a similar layer in the stack to scope6.c.
196 * * Each in6_multi corresponds to an ifp, and each ifp corresponds
197 * to a vnet in ifp->if_vnet.
199 static struct mtx mld_mtx;
200 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
202 #define MLD_EMBEDSCOPE(pin6, zoneid) \
203 if (IN6_IS_SCOPE_LINKLOCAL(pin6) || \
204 IN6_IS_ADDR_MC_INTFACELOCAL(pin6)) \
205 (pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF) \
208 * VIMAGE-wide globals.
210 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
211 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
212 VNET_DEFINE_STATIC(int, interface_timers_running6);
213 VNET_DEFINE_STATIC(int, state_change_timers_running6);
214 VNET_DEFINE_STATIC(int, current_state_timers_running6);
216 #define V_mld_gsrdelay VNET(mld_gsrdelay)
217 #define V_mli_head VNET(mli_head)
218 #define V_interface_timers_running6 VNET(interface_timers_running6)
219 #define V_state_change_timers_running6 VNET(state_change_timers_running6)
220 #define V_current_state_timers_running6 VNET(current_state_timers_running6)
222 SYSCTL_DECL(_net_inet6); /* Note: Not in any common header. */
224 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
225 "IPv6 Multicast Listener Discovery");
228 * Virtualized sysctls.
230 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
231 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
232 &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
233 "Rate limit for MLDv2 Group-and-Source queries in seconds");
236 * Non-virtualized sysctls.
238 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
239 CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
240 "Per-interface MLDv2 state");
242 static int mld_v1enable = 1;
243 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
244 &mld_v1enable, 0, "Enable fallback to MLDv1");
246 static int mld_use_allow = 1;
247 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
248 &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
251 * Packed Router Alert option structure declaration.
256 struct ip6_opt_router ra;
260 * Router Alert hop-by-hop option header.
262 static struct mld_raopt mld_ra = {
264 .pad = { .ip6o_type = IP6OPT_PADN, 0 },
266 .ip6or_type = IP6OPT_ROUTER_ALERT,
267 .ip6or_len = IP6OPT_RTALERT_LEN - 2,
268 .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
269 .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
272 static struct ip6_pktopts mld_po;
275 mld_save_context(struct mbuf *m, struct ifnet *ifp)
279 m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
281 m->m_pkthdr.flowid = ifp->if_index;
285 mld_scrub_context(struct mbuf *m)
288 m->m_pkthdr.PH_loc.ptr = NULL;
289 m->m_pkthdr.flowid = 0;
293 * Restore context from a queued output chain.
294 * Return saved ifindex.
296 * VIMAGE: The assertion is there to make sure that we
297 * actually called CURVNET_SET() with what's in the mbuf chain.
299 static __inline uint32_t
300 mld_restore_context(struct mbuf *m)
303 #if defined(VIMAGE) && defined(INVARIANTS)
304 KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
305 ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
306 __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
308 return (m->m_pkthdr.flowid);
312 * Retrieve or set threshold between group-source queries in seconds.
314 * VIMAGE: Assume curvnet set by caller.
315 * SMPng: NOTE: Serialized by MLD lock.
318 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
323 error = sysctl_wire_old_buffer(req, sizeof(int));
329 i = V_mld_gsrdelay.tv_sec;
331 error = sysctl_handle_int(oidp, &i, 0, req);
332 if (error || !req->newptr)
335 if (i < -1 || i >= 60) {
340 CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
341 V_mld_gsrdelay.tv_sec, i);
342 V_mld_gsrdelay.tv_sec = i;
350 * Expose struct mld_ifsoftc to userland, keyed by ifindex.
351 * For use by ifmcstat(8).
353 * SMPng: NOTE: Does an unlocked ifindex space read.
354 * VIMAGE: Assume curvnet set by caller. The node handler itself
355 * is not directly virtualized.
358 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
364 struct mld_ifsoftc *mli;
369 if (req->newptr != NULL)
375 error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
380 IN6_MULTI_LIST_LOCK();
383 if (name[0] <= 0 || name[0] > V_if_index) {
390 ifp = ifnet_byindex(name[0]);
394 LIST_FOREACH(mli, &V_mli_head, mli_link) {
395 if (ifp == mli->mli_ifp) {
396 struct mld_ifinfo info;
398 info.mli_version = mli->mli_version;
399 info.mli_v1_timer = mli->mli_v1_timer;
400 info.mli_v2_timer = mli->mli_v2_timer;
401 info.mli_flags = mli->mli_flags;
402 info.mli_rv = mli->mli_rv;
403 info.mli_qi = mli->mli_qi;
404 info.mli_qri = mli->mli_qri;
405 info.mli_uri = mli->mli_uri;
406 error = SYSCTL_OUT(req, &info, sizeof(info));
413 IN6_MULTI_LIST_UNLOCK();
419 * Dispatch an entire queue of pending packet chains.
420 * VIMAGE: Assumes the vnet pointer has been set.
423 mld_dispatch_queue(struct mbufq *mq, int limit)
427 while ((m = mbufq_dequeue(mq)) != NULL) {
428 CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
429 mld_dispatch_packet(m);
436 * Filter outgoing MLD report state by group.
438 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
439 * and node-local addresses. However, kernel and socket consumers
440 * always embed the KAME scope ID in the address provided, so strip it
441 * when performing comparison.
442 * Note: This is not the same as the *multicast* scope.
444 * Return zero if the given group is one for which MLD reports
445 * should be suppressed, or non-zero if reports should be issued.
448 mld_is_addr_reported(const struct in6_addr *addr)
451 KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
453 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
456 if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
457 struct in6_addr tmp = *addr;
458 in6_clearscope(&tmp);
459 if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
467 * Attach MLD when PF_INET6 is attached to an interface.
469 * SMPng: Normally called with IF_AFDATA_LOCK held.
472 mld_domifattach(struct ifnet *ifp)
474 struct mld_ifsoftc *mli;
476 CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
477 __func__, ifp, if_name(ifp));
481 mli = mli_alloc_locked(ifp);
482 if (!(ifp->if_flags & IFF_MULTICAST))
483 mli->mli_flags |= MLIF_SILENT;
485 mli->mli_flags |= MLIF_USEALLOW;
493 * VIMAGE: assume curvnet set by caller.
495 static struct mld_ifsoftc *
496 mli_alloc_locked(/*const*/ struct ifnet *ifp)
498 struct mld_ifsoftc *mli;
502 mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO);
507 mli->mli_version = MLD_VERSION_2;
509 mli->mli_rv = MLD_RV_INIT;
510 mli->mli_qi = MLD_QI_INIT;
511 mli->mli_qri = MLD_QRI_INIT;
512 mli->mli_uri = MLD_URI_INIT;
513 mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
515 LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
517 CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)",
527 * NOTE: Some finalization tasks need to run before the protocol domain
528 * is detached, but also before the link layer does its cleanup.
529 * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
531 * SMPng: Caller must hold IN6_MULTI_LOCK().
532 * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
533 * XXX This routine is also bitten by unlocked ifma_protospec access.
536 mld_ifdetach(struct ifnet *ifp)
538 struct mld_ifsoftc *mli;
539 struct ifmultiaddr *ifma, *next;
540 struct in6_multi *inm;
541 struct in6_multi_head inmh;
543 CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
547 IN6_MULTI_LIST_LOCK_ASSERT();
550 mli = MLD_IFINFO(ifp);
551 if (mli->mli_version == MLD_VERSION_2) {
554 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
555 if (ifma->ifma_addr->sa_family != AF_INET6 ||
556 ifma->ifma_protospec == NULL)
558 inm = (struct in6_multi *)ifma->ifma_protospec;
559 if (inm->in6m_state == MLD_LEAVING_MEMBER) {
560 in6m_disconnect(inm);
561 in6m_rele_locked(&inmh, inm);
562 ifma->ifma_protospec = NULL;
564 in6m_clear_recorded(inm);
565 if (__predict_false(ifma6_restart)) {
566 ifma6_restart = false;
570 IF_ADDR_WUNLOCK(ifp);
574 in6m_release_list_deferred(&inmh);
578 * Hook for domifdetach.
579 * Runs after link-layer cleanup; free MLD state.
581 * SMPng: Normally called with IF_AFDATA_LOCK held.
584 mld_domifdetach(struct ifnet *ifp)
587 CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
588 __func__, ifp, if_name(ifp));
591 mli_delete_locked(ifp);
596 mli_delete_locked(const struct ifnet *ifp)
598 struct mld_ifsoftc *mli, *tmli;
600 CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
601 __func__, ifp, if_name(ifp));
605 LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
606 if (mli->mli_ifp == ifp) {
608 * Free deferred General Query responses.
610 mbufq_drain(&mli->mli_gq);
612 LIST_REMOVE(mli, mli_link);
621 * Process a received MLDv1 general or address-specific query.
622 * Assumes that the query header has been pulled up to sizeof(mld_hdr).
624 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
625 * mld_addr. This is OK as we own the mbuf chain.
628 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
629 /*const*/ struct mld_hdr *mld)
631 struct epoch_tracker et;
632 struct ifmultiaddr *ifma;
633 struct mld_ifsoftc *mli;
634 struct in6_multi *inm;
635 int is_general_query;
638 char ip6tbuf[INET6_ADDRSTRLEN];
641 is_general_query = 0;
644 CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
645 ip6_sprintf(ip6tbuf, &mld->mld_addr),
651 * RFC3810 Section 6.2: MLD queries must originate from
652 * a router's link-local address.
654 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
655 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
656 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
662 * Do address field validation upfront before we accept
665 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
667 * MLDv1 General Query.
668 * If this was not sent to the all-nodes group, ignore it.
673 in6_clearscope(&dst);
674 if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
676 is_general_query = 1;
679 * Embed scope ID of receiving interface in MLD query for
680 * lookup whilst we don't hold other locks.
682 in6_setscope(&mld->mld_addr, ifp, NULL);
685 IN6_MULTI_LIST_LOCK();
689 * Switch to MLDv1 host compatibility mode.
691 mli = MLD_IFINFO(ifp);
692 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
693 mld_set_version(mli, MLD_VERSION_1);
695 timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
700 if (is_general_query) {
702 * For each reporting group joined on this
703 * interface, kick the report timer.
705 CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
707 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
708 if (ifma->ifma_addr->sa_family != AF_INET6 ||
709 ifma->ifma_protospec == NULL)
711 inm = (struct in6_multi *)ifma->ifma_protospec;
712 mld_v1_update_group(inm, timer);
716 * MLDv1 Group-Specific Query.
717 * If this is a group-specific MLDv1 query, we need only
718 * look up the single group to process it.
720 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
722 CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
723 ip6_sprintf(ip6tbuf, &mld->mld_addr),
725 mld_v1_update_group(inm, timer);
727 /* XXX Clear embedded scope ID as userland won't expect it. */
728 in6_clearscope(&mld->mld_addr);
733 IN6_MULTI_LIST_UNLOCK();
739 * Update the report timer on a group in response to an MLDv1 query.
741 * If we are becoming the reporting member for this group, start the timer.
742 * If we already are the reporting member for this group, and timer is
743 * below the threshold, reset it.
745 * We may be updating the group for the first time since we switched
746 * to MLDv2. If we are, then we must clear any recorded source lists,
747 * and transition to REPORTING state; the group timer is overloaded
748 * for group and group-source query responses.
750 * Unlike MLDv2, the delay per group should be jittered
751 * to avoid bursts of MLDv1 reports.
754 mld_v1_update_group(struct in6_multi *inm, const int timer)
757 char ip6tbuf[INET6_ADDRSTRLEN];
760 CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
761 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
762 if_name(inm->in6m_ifp), timer);
764 IN6_MULTI_LIST_LOCK_ASSERT();
766 switch (inm->in6m_state) {
768 case MLD_SILENT_MEMBER:
770 case MLD_REPORTING_MEMBER:
771 if (inm->in6m_timer != 0 &&
772 inm->in6m_timer <= timer) {
773 CTR1(KTR_MLD, "%s: REPORTING and timer running, "
774 "skipping.", __func__);
778 case MLD_SG_QUERY_PENDING_MEMBER:
779 case MLD_G_QUERY_PENDING_MEMBER:
780 case MLD_IDLE_MEMBER:
781 case MLD_LAZY_MEMBER:
782 case MLD_AWAKENING_MEMBER:
783 CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
784 inm->in6m_state = MLD_REPORTING_MEMBER;
785 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
786 V_current_state_timers_running6 = 1;
788 case MLD_SLEEPING_MEMBER:
789 CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
790 inm->in6m_state = MLD_AWAKENING_MEMBER;
792 case MLD_LEAVING_MEMBER:
798 * Process a received MLDv2 general, group-specific or
799 * group-and-source-specific query.
801 * Assumes that the query header has been pulled up to sizeof(mldv2_query).
803 * Return 0 if successful, otherwise an appropriate error code is returned.
806 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
807 struct mbuf *m, const int off, const int icmp6len)
809 struct mld_ifsoftc *mli;
810 struct mldv2_query *mld;
811 struct in6_multi *inm;
812 uint32_t maxdelay, nsrc, qqi;
813 int is_general_query;
817 char ip6tbuf[INET6_ADDRSTRLEN];
820 is_general_query = 0;
823 * RFC3810 Section 6.2: MLD queries must originate from
824 * a router's link-local address.
826 if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
827 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
828 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
833 CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
835 mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
837 maxdelay = ntohs(mld->mld_maxdelay); /* in 1/10ths of a second */
838 if (maxdelay >= 32768) {
839 maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
840 (MLD_MRC_EXP(maxdelay) + 3);
842 timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
846 qrv = MLD_QRV(mld->mld_misc);
848 CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
855 qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
856 (MLD_QQIC_EXP(mld->mld_qqi) + 3);
859 nsrc = ntohs(mld->mld_numsrc);
860 if (nsrc > MLD_MAX_GS_SOURCES)
862 if (icmp6len < sizeof(struct mldv2_query) +
863 (nsrc * sizeof(struct in6_addr)))
867 * Do further input validation upfront to avoid resetting timers
868 * should we need to discard this query.
870 if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
872 * A general query with a source list has undefined
873 * behaviour; discard it.
877 is_general_query = 1;
880 * Embed scope ID of receiving interface in MLD query for
881 * lookup whilst we don't hold other locks (due to KAME
882 * locking lameness). We own this mbuf chain just now.
884 in6_setscope(&mld->mld_addr, ifp, NULL);
887 IN6_MULTI_LIST_LOCK();
890 mli = MLD_IFINFO(ifp);
891 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
894 * Discard the v2 query if we're in Compatibility Mode.
895 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
896 * until the Old Version Querier Present timer expires.
898 if (mli->mli_version != MLD_VERSION_2)
901 mld_set_version(mli, MLD_VERSION_2);
904 mli->mli_qri = maxdelay;
906 CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
909 if (is_general_query) {
911 * MLDv2 General Query.
913 * Schedule a current-state report on this ifp for
914 * all groups, possibly containing source lists.
916 * If there is a pending General Query response
917 * scheduled earlier than the selected delay, do
918 * not schedule any other reports.
919 * Otherwise, reset the interface timer.
921 CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
923 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
924 mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
925 V_interface_timers_running6 = 1;
928 struct epoch_tracker et;
931 * MLDv2 Group-specific or Group-and-source-specific Query.
933 * Group-source-specific queries are throttled on
934 * a per-group basis to defeat denial-of-service attempts.
935 * Queries for groups we are not a member of on this
936 * link are simply ignored.
939 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
945 if (!ratecheck(&inm->in6m_lastgsrtv,
947 CTR1(KTR_MLD, "%s: GS query throttled.",
953 CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
956 * If there is a pending General Query response
957 * scheduled sooner than the selected delay, no
958 * further report need be scheduled.
959 * Otherwise, prepare to respond to the
960 * group-specific or group-and-source query.
962 if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
963 mld_v2_process_group_query(inm, mli, timer, m, off);
965 /* XXX Clear embedded scope ID as userland won't expect it. */
966 in6_clearscope(&mld->mld_addr);
972 IN6_MULTI_LIST_UNLOCK();
978 * Process a received MLDv2 group-specific or group-and-source-specific
980 * Return <0 if any error occurred. Currently this is ignored.
983 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
984 int timer, struct mbuf *m0, const int off)
986 struct mldv2_query *mld;
990 IN6_MULTI_LIST_LOCK_ASSERT();
994 mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
996 switch (inm->in6m_state) {
998 case MLD_SILENT_MEMBER:
999 case MLD_SLEEPING_MEMBER:
1000 case MLD_LAZY_MEMBER:
1001 case MLD_AWAKENING_MEMBER:
1002 case MLD_IDLE_MEMBER:
1003 case MLD_LEAVING_MEMBER:
1006 case MLD_REPORTING_MEMBER:
1007 case MLD_G_QUERY_PENDING_MEMBER:
1008 case MLD_SG_QUERY_PENDING_MEMBER:
1012 nsrc = ntohs(mld->mld_numsrc);
1015 * Deal with group-specific queries upfront.
1016 * If any group query is already pending, purge any recorded
1017 * source-list state if it exists, and schedule a query response
1018 * for this group-specific query.
1021 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1022 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1023 in6m_clear_recorded(inm);
1024 timer = min(inm->in6m_timer, timer);
1026 inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1027 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1028 V_current_state_timers_running6 = 1;
1033 * Deal with the case where a group-and-source-specific query has
1034 * been received but a group-specific query is already pending.
1036 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1037 timer = min(inm->in6m_timer, timer);
1038 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1039 V_current_state_timers_running6 = 1;
1044 * Finally, deal with the case where a group-and-source-specific
1045 * query has been received, where a response to a previous g-s-r
1046 * query exists, or none exists.
1047 * In this case, we need to parse the source-list which the Querier
1048 * has provided us with and check if we have any source list filter
1049 * entries at T1 for these sources. If we do not, there is no need
1050 * schedule a report and the query may be dropped.
1051 * If we do, we must record them and schedule a current-state
1052 * report for those sources.
1054 if (inm->in6m_nsrc > 0) {
1061 soff = off + sizeof(struct mldv2_query);
1063 for (i = 0; i < nsrc; i++) {
1064 sp = mtod(m, uint8_t *) + soff;
1065 retval = in6m_record_source(inm,
1066 (const struct in6_addr *)sp);
1069 nrecorded += retval;
1070 soff += sizeof(struct in6_addr);
1071 if (soff >= m->m_len) {
1072 soff = soff - m->m_len;
1078 if (nrecorded > 0) {
1080 "%s: schedule response to SG query", __func__);
1081 inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1082 inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1083 V_current_state_timers_running6 = 1;
1091 * Process a received MLDv1 host membership report.
1092 * Assumes mld points to mld_hdr in pulled up mbuf chain.
1094 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1095 * mld_addr. This is OK as we own the mbuf chain.
1098 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1099 /*const*/ struct mld_hdr *mld)
1101 struct in6_addr src, dst;
1102 struct epoch_tracker et;
1103 struct in6_ifaddr *ia;
1104 struct in6_multi *inm;
1106 char ip6tbuf[INET6_ADDRSTRLEN];
1109 if (!mld_v1enable) {
1110 CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1111 ip6_sprintf(ip6tbuf, &mld->mld_addr),
1116 if (ifp->if_flags & IFF_LOOPBACK)
1120 * MLDv1 reports must originate from a host's link-local address,
1121 * or the unspecified address (when booting).
1124 in6_clearscope(&src);
1125 if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1126 CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1127 ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1133 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1134 * group, and must be directed to the group itself.
1137 in6_clearscope(&dst);
1138 if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1139 !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1140 CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1141 ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1147 * Make sure we don't hear our own membership report, as fast
1148 * leave requires knowing that we are the only member of a
1149 * group. Assume we used the link-local address if available,
1150 * otherwise look for ::.
1152 * XXX Note that scope ID comparison is needed for the address
1153 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1154 * performed for the on-wire address.
1156 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1157 if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1158 (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1160 ifa_free(&ia->ia_ifa);
1164 ifa_free(&ia->ia_ifa);
1166 CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1167 ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1170 * Embed scope ID of receiving interface in MLD query for lookup
1171 * whilst we don't hold other locks (due to KAME locking lameness).
1173 if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1174 in6_setscope(&mld->mld_addr, ifp, NULL);
1176 IN6_MULTI_LIST_LOCK();
1178 NET_EPOCH_ENTER(et);
1181 * MLDv1 report suppression.
1182 * If we are a member of this group, and our membership should be
1183 * reported, and our group timer is pending or about to be reset,
1184 * stop our group timer by transitioning to the 'lazy' state.
1186 inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1188 struct mld_ifsoftc *mli;
1190 mli = inm->in6m_mli;
1191 KASSERT(mli != NULL,
1192 ("%s: no mli for ifp %p", __func__, ifp));
1195 * If we are in MLDv2 host mode, do not allow the
1196 * other host's MLDv1 report to suppress our reports.
1198 if (mli->mli_version == MLD_VERSION_2)
1201 inm->in6m_timer = 0;
1203 switch (inm->in6m_state) {
1204 case MLD_NOT_MEMBER:
1205 case MLD_SILENT_MEMBER:
1206 case MLD_SLEEPING_MEMBER:
1208 case MLD_REPORTING_MEMBER:
1209 case MLD_IDLE_MEMBER:
1210 case MLD_AWAKENING_MEMBER:
1212 "report suppressed for %s on ifp %p(%s)",
1213 ip6_sprintf(ip6tbuf, &mld->mld_addr),
1215 case MLD_LAZY_MEMBER:
1216 inm->in6m_state = MLD_LAZY_MEMBER;
1218 case MLD_G_QUERY_PENDING_MEMBER:
1219 case MLD_SG_QUERY_PENDING_MEMBER:
1220 case MLD_LEAVING_MEMBER:
1228 IN6_MULTI_LIST_UNLOCK();
1230 /* XXX Clear embedded scope ID as userland won't expect it. */
1231 in6_clearscope(&mld->mld_addr);
1239 * Assume query messages which fit in a single ICMPv6 message header
1240 * have been pulled up.
1241 * Assume that userland will want to see the message, even if it
1242 * otherwise fails kernel input validation; do not free it.
1243 * Pullup may however free the mbuf chain m if it fails.
1245 * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1248 mld_input(struct mbuf *m, int off, int icmp6len)
1251 struct ip6_hdr *ip6;
1252 struct mld_hdr *mld;
1255 CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1257 ifp = m->m_pkthdr.rcvif;
1259 ip6 = mtod(m, struct ip6_hdr *);
1261 /* Pullup to appropriate size. */
1262 mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1263 if (mld->mld_type == MLD_LISTENER_QUERY &&
1264 icmp6len >= sizeof(struct mldv2_query)) {
1265 mldlen = sizeof(struct mldv2_query);
1267 mldlen = sizeof(struct mld_hdr);
1269 IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1271 ICMP6STAT_INC(icp6s_badlen);
1272 return (IPPROTO_DONE);
1276 * Userland needs to see all of this traffic for implementing
1277 * the endpoint discovery portion of multicast routing.
1279 switch (mld->mld_type) {
1280 case MLD_LISTENER_QUERY:
1281 icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1282 if (icmp6len == sizeof(struct mld_hdr)) {
1283 if (mld_v1_input_query(ifp, ip6, mld) != 0)
1285 } else if (icmp6len >= sizeof(struct mldv2_query)) {
1286 if (mld_v2_input_query(ifp, ip6, m, off,
1291 case MLD_LISTENER_REPORT:
1292 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1293 if (mld_v1_input_report(ifp, ip6, mld) != 0)
1296 case MLDV2_LISTENER_REPORT:
1297 icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1299 case MLD_LISTENER_DONE:
1300 icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1310 * Fast timeout handler (global).
1311 * VIMAGE: Timeout handlers are expected to service all vimages.
1316 VNET_ITERATOR_DECL(vnet_iter);
1318 VNET_LIST_RLOCK_NOSLEEP();
1319 VNET_FOREACH(vnet_iter) {
1320 CURVNET_SET(vnet_iter);
1321 mld_fasttimo_vnet();
1324 VNET_LIST_RUNLOCK_NOSLEEP();
1328 * Fast timeout handler (per-vnet).
1330 * VIMAGE: Assume caller has set up our curvnet.
1333 mld_fasttimo_vnet(void)
1335 struct mbufq scq; /* State-change packets */
1336 struct mbufq qrq; /* Query response packets */
1338 struct mld_ifsoftc *mli;
1339 struct ifmultiaddr *ifma, *next;
1340 struct in6_multi *inm, *tinm;
1341 struct in6_multi_head inmh;
1347 * Quick check to see if any work needs to be done, in order to
1348 * minimize the overhead of fasttimo processing.
1349 * SMPng: XXX Unlocked reads.
1351 if (!V_current_state_timers_running6 &&
1352 !V_interface_timers_running6 &&
1353 !V_state_change_timers_running6)
1357 IN6_MULTI_LIST_LOCK();
1361 * MLDv2 General Query response timer processing.
1363 if (V_interface_timers_running6) {
1364 CTR1(KTR_MLD, "%s: interface timers running", __func__);
1366 V_interface_timers_running6 = 0;
1367 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1368 if (mli->mli_v2_timer == 0) {
1370 } else if (--mli->mli_v2_timer == 0) {
1371 mld_v2_dispatch_general_query(mli);
1373 V_interface_timers_running6 = 1;
1378 if (!V_current_state_timers_running6 &&
1379 !V_state_change_timers_running6)
1382 V_current_state_timers_running6 = 0;
1383 V_state_change_timers_running6 = 0;
1385 CTR1(KTR_MLD, "%s: state change timers running", __func__);
1388 * MLD host report and state-change timer processing.
1389 * Note: Processing a v2 group timer may remove a node.
1391 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1394 if (mli->mli_version == MLD_VERSION_2) {
1395 uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1397 mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1398 mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1403 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1404 if (ifma->ifma_addr->sa_family != AF_INET6 ||
1405 ifma->ifma_protospec == NULL)
1407 inm = (struct in6_multi *)ifma->ifma_protospec;
1408 switch (mli->mli_version) {
1410 mld_v1_process_group_timer(&inmh, inm);
1413 mld_v2_process_group_timers(&inmh, &qrq,
1414 &scq, inm, uri_fasthz);
1417 if (__predict_false(ifma6_restart)) {
1418 ifma6_restart = false;
1422 IF_ADDR_WUNLOCK(ifp);
1424 switch (mli->mli_version) {
1427 * Transmit reports for this lifecycle. This
1428 * is done while not holding IF_ADDR_LOCK
1429 * since this can call
1430 * in6ifa_ifpforlinklocal() which locks
1431 * IF_ADDR_LOCK internally as well as
1432 * ip6_output() to transmit a packet.
1434 SLIST_FOREACH_SAFE(inm, &inmh, in6m_nrele, tinm) {
1435 SLIST_REMOVE_HEAD(&inmh,
1437 (void)mld_v1_transmit_report(inm,
1438 MLD_LISTENER_REPORT);
1442 mld_dispatch_queue(&qrq, 0);
1443 mld_dispatch_queue(&scq, 0);
1446 * Free the in_multi reference(s) for
1449 in6m_release_list_deferred(&inmh);
1456 IN6_MULTI_LIST_UNLOCK();
1460 * Update host report group timer.
1461 * Will update the global pending timer flags.
1464 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1466 int report_timer_expired;
1468 IN6_MULTI_LIST_LOCK_ASSERT();
1471 if (inm->in6m_timer == 0) {
1472 report_timer_expired = 0;
1473 } else if (--inm->in6m_timer == 0) {
1474 report_timer_expired = 1;
1476 V_current_state_timers_running6 = 1;
1480 switch (inm->in6m_state) {
1481 case MLD_NOT_MEMBER:
1482 case MLD_SILENT_MEMBER:
1483 case MLD_IDLE_MEMBER:
1484 case MLD_LAZY_MEMBER:
1485 case MLD_SLEEPING_MEMBER:
1486 case MLD_AWAKENING_MEMBER:
1488 case MLD_REPORTING_MEMBER:
1489 if (report_timer_expired) {
1490 inm->in6m_state = MLD_IDLE_MEMBER;
1491 in6m_disconnect(inm);
1492 in6m_rele_locked(inmh, inm);
1495 case MLD_G_QUERY_PENDING_MEMBER:
1496 case MLD_SG_QUERY_PENDING_MEMBER:
1497 case MLD_LEAVING_MEMBER:
1503 * Update a group's timers for MLDv2.
1504 * Will update the global pending timer flags.
1505 * Note: Unlocked read from mli.
1508 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1509 struct mbufq *qrq, struct mbufq *scq,
1510 struct in6_multi *inm, const int uri_fasthz)
1512 int query_response_timer_expired;
1513 int state_change_retransmit_timer_expired;
1515 char ip6tbuf[INET6_ADDRSTRLEN];
1518 IN6_MULTI_LIST_LOCK_ASSERT();
1521 query_response_timer_expired = 0;
1522 state_change_retransmit_timer_expired = 0;
1525 * During a transition from compatibility mode back to MLDv2,
1526 * a group record in REPORTING state may still have its group
1527 * timer active. This is a no-op in this function; it is easier
1528 * to deal with it here than to complicate the slow-timeout path.
1530 if (inm->in6m_timer == 0) {
1531 query_response_timer_expired = 0;
1532 } else if (--inm->in6m_timer == 0) {
1533 query_response_timer_expired = 1;
1535 V_current_state_timers_running6 = 1;
1538 if (inm->in6m_sctimer == 0) {
1539 state_change_retransmit_timer_expired = 0;
1540 } else if (--inm->in6m_sctimer == 0) {
1541 state_change_retransmit_timer_expired = 1;
1543 V_state_change_timers_running6 = 1;
1546 /* We are in fasttimo, so be quick about it. */
1547 if (!state_change_retransmit_timer_expired &&
1548 !query_response_timer_expired)
1551 switch (inm->in6m_state) {
1552 case MLD_NOT_MEMBER:
1553 case MLD_SILENT_MEMBER:
1554 case MLD_SLEEPING_MEMBER:
1555 case MLD_LAZY_MEMBER:
1556 case MLD_AWAKENING_MEMBER:
1557 case MLD_IDLE_MEMBER:
1559 case MLD_G_QUERY_PENDING_MEMBER:
1560 case MLD_SG_QUERY_PENDING_MEMBER:
1562 * Respond to a previously pending Group-Specific
1563 * or Group-and-Source-Specific query by enqueueing
1564 * the appropriate Current-State report for
1565 * immediate transmission.
1567 if (query_response_timer_expired) {
1570 retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1571 (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1573 CTR2(KTR_MLD, "%s: enqueue record = %d",
1575 inm->in6m_state = MLD_REPORTING_MEMBER;
1576 in6m_clear_recorded(inm);
1579 case MLD_REPORTING_MEMBER:
1580 case MLD_LEAVING_MEMBER:
1581 if (state_change_retransmit_timer_expired) {
1583 * State-change retransmission timer fired.
1584 * If there are any further pending retransmissions,
1585 * set the global pending state-change flag, and
1588 if (--inm->in6m_scrv > 0) {
1589 inm->in6m_sctimer = uri_fasthz;
1590 V_state_change_timers_running6 = 1;
1593 * Retransmit the previously computed state-change
1594 * report. If there are no further pending
1595 * retransmissions, the mbuf queue will be consumed.
1596 * Update T0 state to T1 as we have now sent
1599 (void)mld_v2_merge_state_changes(inm, scq);
1602 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1603 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1604 if_name(inm->in6m_ifp));
1607 * If we are leaving the group for good, make sure
1608 * we release MLD's reference to it.
1609 * This release must be deferred using a SLIST,
1610 * as we are called from a loop which traverses
1611 * the in_ifmultiaddr TAILQ.
1613 if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1614 inm->in6m_scrv == 0) {
1615 inm->in6m_state = MLD_NOT_MEMBER;
1616 in6m_disconnect(inm);
1617 in6m_rele_locked(inmh, inm);
1625 * Switch to a different version on the given interface,
1626 * as per Section 9.12.
1629 mld_set_version(struct mld_ifsoftc *mli, const int version)
1631 int old_version_timer;
1635 CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1636 version, mli->mli_ifp, if_name(mli->mli_ifp));
1638 if (version == MLD_VERSION_1) {
1640 * Compute the "Older Version Querier Present" timer as per
1643 old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1644 old_version_timer *= PR_SLOWHZ;
1645 mli->mli_v1_timer = old_version_timer;
1648 if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1649 mli->mli_version = MLD_VERSION_1;
1650 mld_v2_cancel_link_timers(mli);
1655 * Cancel pending MLDv2 timers for the given link and all groups
1656 * joined on it; state-change, general-query, and group-query timers.
1659 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1661 struct ifmultiaddr *ifma, *next;
1663 struct in6_multi *inm;
1664 struct in6_multi_head inmh;
1666 CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1667 mli->mli_ifp, if_name(mli->mli_ifp));
1670 IN6_MULTI_LIST_LOCK_ASSERT();
1674 * Fast-track this potentially expensive operation
1675 * by checking all the global 'timer pending' flags.
1677 if (!V_interface_timers_running6 &&
1678 !V_state_change_timers_running6 &&
1679 !V_current_state_timers_running6)
1682 mli->mli_v2_timer = 0;
1688 CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) {
1689 if (ifma->ifma_addr->sa_family != AF_INET6 ||
1690 ifma->ifma_protospec == NULL)
1692 inm = (struct in6_multi *)ifma->ifma_protospec;
1693 switch (inm->in6m_state) {
1694 case MLD_NOT_MEMBER:
1695 case MLD_SILENT_MEMBER:
1696 case MLD_IDLE_MEMBER:
1697 case MLD_LAZY_MEMBER:
1698 case MLD_SLEEPING_MEMBER:
1699 case MLD_AWAKENING_MEMBER:
1701 case MLD_LEAVING_MEMBER:
1703 * If we are leaving the group and switching
1704 * version, we need to release the final
1705 * reference held for issuing the INCLUDE {}.
1707 in6m_disconnect(inm);
1708 in6m_rele_locked(&inmh, inm);
1709 ifma->ifma_protospec = NULL;
1711 case MLD_G_QUERY_PENDING_MEMBER:
1712 case MLD_SG_QUERY_PENDING_MEMBER:
1713 in6m_clear_recorded(inm);
1715 case MLD_REPORTING_MEMBER:
1716 inm->in6m_sctimer = 0;
1717 inm->in6m_timer = 0;
1718 inm->in6m_state = MLD_REPORTING_MEMBER;
1720 * Free any pending MLDv2 state-change records.
1722 mbufq_drain(&inm->in6m_scq);
1725 if (__predict_false(ifma6_restart)) {
1726 ifma6_restart = false;
1730 IF_ADDR_WUNLOCK(ifp);
1731 in6m_release_list_deferred(&inmh);
1735 * Global slowtimo handler.
1736 * VIMAGE: Timeout handlers are expected to service all vimages.
1741 VNET_ITERATOR_DECL(vnet_iter);
1743 VNET_LIST_RLOCK_NOSLEEP();
1744 VNET_FOREACH(vnet_iter) {
1745 CURVNET_SET(vnet_iter);
1746 mld_slowtimo_vnet();
1749 VNET_LIST_RUNLOCK_NOSLEEP();
1753 * Per-vnet slowtimo handler.
1756 mld_slowtimo_vnet(void)
1758 struct mld_ifsoftc *mli;
1762 LIST_FOREACH(mli, &V_mli_head, mli_link) {
1763 mld_v1_process_querier_timers(mli);
1770 * Update the Older Version Querier Present timers for a link.
1771 * See Section 9.12 of RFC 3810.
1774 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1779 if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1781 * MLDv1 Querier Present timer expired; revert to MLDv2.
1784 "%s: transition from v%d -> v%d on %p(%s)",
1785 __func__, mli->mli_version, MLD_VERSION_2,
1786 mli->mli_ifp, if_name(mli->mli_ifp));
1787 mli->mli_version = MLD_VERSION_2;
1792 * Transmit an MLDv1 report immediately.
1795 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1798 struct in6_ifaddr *ia;
1799 struct ip6_hdr *ip6;
1800 struct mbuf *mh, *md;
1801 struct mld_hdr *mld;
1803 IN6_MULTI_LIST_LOCK_ASSERT();
1806 ifp = in6m->in6m_ifp;
1807 /* in process of being freed */
1810 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1811 /* ia may be NULL if link-local address is tentative. */
1813 mh = m_gethdr(M_NOWAIT, MT_DATA);
1816 ifa_free(&ia->ia_ifa);
1819 md = m_get(M_NOWAIT, MT_DATA);
1823 ifa_free(&ia->ia_ifa);
1829 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1830 * that ether_output() does not need to allocate another mbuf
1831 * for the header in the most common case.
1833 M_ALIGN(mh, sizeof(struct ip6_hdr));
1834 mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1835 mh->m_len = sizeof(struct ip6_hdr);
1837 ip6 = mtod(mh, struct ip6_hdr *);
1839 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1840 ip6->ip6_vfc |= IPV6_VERSION;
1841 ip6->ip6_nxt = IPPROTO_ICMPV6;
1842 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1843 ip6->ip6_dst = in6m->in6m_addr;
1845 md->m_len = sizeof(struct mld_hdr);
1846 mld = mtod(md, struct mld_hdr *);
1847 mld->mld_type = type;
1850 mld->mld_maxdelay = 0;
1851 mld->mld_reserved = 0;
1852 mld->mld_addr = in6m->in6m_addr;
1853 in6_clearscope(&mld->mld_addr);
1854 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1855 sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1857 mld_save_context(mh, ifp);
1858 mh->m_flags |= M_MLDV1;
1860 mld_dispatch_packet(mh);
1863 ifa_free(&ia->ia_ifa);
1868 * Process a state change from the upper layer for the given IPv6 group.
1870 * Each socket holds a reference on the in_multi in its own ip_moptions.
1871 * The socket layer will have made the necessary updates to.the group
1872 * state, it is now up to MLD to issue a state change report if there
1873 * has been any change between T0 (when the last state-change was issued)
1876 * We use the MLDv2 state machine at group level. The MLd module
1877 * however makes the decision as to which MLD protocol version to speak.
1878 * A state change *from* INCLUDE {} always means an initial join.
1879 * A state change *to* INCLUDE {} always means a final leave.
1881 * If delay is non-zero, and the state change is an initial multicast
1882 * join, the state change report will be delayed by 'delay' ticks
1883 * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1884 * the initial MLDv2 state change report will be delayed by whichever
1885 * is sooner, a pending state-change timer or delay itself.
1887 * VIMAGE: curvnet should have been set by caller, as this routine
1888 * is called from the socket option handlers.
1891 mld_change_state(struct in6_multi *inm, const int delay)
1893 struct mld_ifsoftc *mli;
1897 IN6_MULTI_LIST_LOCK_ASSERT();
1902 * Try to detect if the upper layer just asked us to change state
1903 * for an interface which has now gone away.
1905 KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1906 ifp = inm->in6m_ifma->ifma_ifp;
1910 * Sanity check that netinet6's notion of ifp is the
1913 KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1916 mli = MLD_IFINFO(ifp);
1917 KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1920 * If we detect a state transition to or from MCAST_UNDEFINED
1921 * for this group, then we are starting or finishing an MLD
1922 * life cycle for this group.
1924 if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1925 CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1926 inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1927 if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1928 CTR1(KTR_MLD, "%s: initial join", __func__);
1929 error = mld_initial_join(inm, mli, delay);
1931 } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1932 CTR1(KTR_MLD, "%s: final leave", __func__);
1933 mld_final_leave(inm, mli);
1937 CTR1(KTR_MLD, "%s: filter set change", __func__);
1940 error = mld_handle_state_change(inm, mli);
1948 * Perform the initial join for an MLD group.
1950 * When joining a group:
1951 * If the group should have its MLD traffic suppressed, do nothing.
1952 * MLDv1 starts sending MLDv1 host membership reports.
1953 * MLDv2 will schedule an MLDv2 state-change report containing the
1954 * initial state of the membership.
1956 * If the delay argument is non-zero, then we must delay sending the
1957 * initial state change for delay ticks (in units of PR_FASTHZ).
1960 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1965 int error, retval, syncstates;
1968 char ip6tbuf[INET6_ADDRSTRLEN];
1971 CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1972 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1973 inm->in6m_ifp, if_name(inm->in6m_ifp));
1978 ifp = inm->in6m_ifp;
1980 IN6_MULTI_LIST_LOCK_ASSERT();
1983 KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1986 * Groups joined on loopback or marked as 'not reported',
1987 * enter the MLD_SILENT_MEMBER state and
1988 * are never reported in any protocol exchanges.
1989 * All other groups enter the appropriate state machine
1990 * for the version in use on this link.
1991 * A link marked as MLIF_SILENT causes MLD to be completely
1992 * disabled for the link.
1994 if ((ifp->if_flags & IFF_LOOPBACK) ||
1995 (mli->mli_flags & MLIF_SILENT) ||
1996 !mld_is_addr_reported(&inm->in6m_addr)) {
1998 "%s: not kicking state machine for silent group", __func__);
1999 inm->in6m_state = MLD_SILENT_MEMBER;
2000 inm->in6m_timer = 0;
2003 * Deal with overlapping in_multi lifecycle.
2004 * If this group was LEAVING, then make sure
2005 * we drop the reference we picked up to keep the
2006 * group around for the final INCLUDE {} enqueue.
2008 if (mli->mli_version == MLD_VERSION_2 &&
2009 inm->in6m_state == MLD_LEAVING_MEMBER) {
2010 inm->in6m_refcount--;
2012 inm->in6m_state = MLD_REPORTING_MEMBER;
2014 switch (mli->mli_version) {
2017 * If a delay was provided, only use it if
2018 * it is greater than the delay normally
2019 * used for an MLDv1 state change report,
2020 * and delay sending the initial MLDv1 report
2021 * by not transitioning to the IDLE state.
2023 odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2025 inm->in6m_timer = max(delay, odelay);
2026 V_current_state_timers_running6 = 1;
2028 inm->in6m_state = MLD_IDLE_MEMBER;
2029 error = mld_v1_transmit_report(inm,
2030 MLD_LISTENER_REPORT);
2032 inm->in6m_timer = odelay;
2033 V_current_state_timers_running6 = 1;
2040 * Defer update of T0 to T1, until the first copy
2041 * of the state change has been transmitted.
2046 * Immediately enqueue a State-Change Report for
2047 * this interface, freeing any previous reports.
2048 * Don't kick the timers if there is nothing to do,
2049 * or if an error occurred.
2051 mq = &inm->in6m_scq;
2053 retval = mld_v2_enqueue_group_record(mq, inm, 1,
2054 0, 0, (mli->mli_flags & MLIF_USEALLOW));
2055 CTR2(KTR_MLD, "%s: enqueue record = %d",
2058 error = retval * -1;
2063 * Schedule transmission of pending state-change
2064 * report up to RV times for this link. The timer
2065 * will fire at the next mld_fasttimo (~200ms),
2066 * giving us an opportunity to merge the reports.
2068 * If a delay was provided to this function, only
2069 * use this delay if sooner than the existing one.
2071 KASSERT(mli->mli_rv > 1,
2072 ("%s: invalid robustness %d", __func__,
2074 inm->in6m_scrv = mli->mli_rv;
2076 if (inm->in6m_sctimer > 1) {
2078 min(inm->in6m_sctimer, delay);
2080 inm->in6m_sctimer = delay;
2082 inm->in6m_sctimer = 1;
2083 V_state_change_timers_running6 = 1;
2091 * Only update the T0 state if state change is atomic,
2092 * i.e. we don't need to wait for a timer to fire before we
2093 * can consider the state change to have been communicated.
2097 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2098 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2099 if_name(inm->in6m_ifp));
2106 * Issue an intermediate state change during the life-cycle.
2109 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2114 char ip6tbuf[INET6_ADDRSTRLEN];
2117 CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2118 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2119 inm->in6m_ifp, if_name(inm->in6m_ifp));
2121 ifp = inm->in6m_ifp;
2123 IN6_MULTI_LIST_LOCK_ASSERT();
2126 KASSERT(mli && mli->mli_ifp == ifp,
2127 ("%s: inconsistent ifp", __func__));
2129 if ((ifp->if_flags & IFF_LOOPBACK) ||
2130 (mli->mli_flags & MLIF_SILENT) ||
2131 !mld_is_addr_reported(&inm->in6m_addr) ||
2132 (mli->mli_version != MLD_VERSION_2)) {
2133 if (!mld_is_addr_reported(&inm->in6m_addr)) {
2135 "%s: not kicking state machine for silent group", __func__);
2137 CTR1(KTR_MLD, "%s: nothing to do", __func__);
2139 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2140 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2141 if_name(inm->in6m_ifp));
2145 mbufq_drain(&inm->in6m_scq);
2147 retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2148 (mli->mli_flags & MLIF_USEALLOW));
2149 CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2154 * If record(s) were enqueued, start the state-change
2155 * report timer for this group.
2157 inm->in6m_scrv = mli->mli_rv;
2158 inm->in6m_sctimer = 1;
2159 V_state_change_timers_running6 = 1;
2165 * Perform the final leave for a multicast address.
2167 * When leaving a group:
2168 * MLDv1 sends a DONE message, if and only if we are the reporter.
2169 * MLDv2 enqueues a state-change report containing a transition
2170 * to INCLUDE {} for immediate transmission.
2173 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2177 char ip6tbuf[INET6_ADDRSTRLEN];
2182 CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2183 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2184 inm->in6m_ifp, if_name(inm->in6m_ifp));
2186 IN6_MULTI_LIST_LOCK_ASSERT();
2189 switch (inm->in6m_state) {
2190 case MLD_NOT_MEMBER:
2191 case MLD_SILENT_MEMBER:
2192 case MLD_LEAVING_MEMBER:
2193 /* Already leaving or left; do nothing. */
2195 "%s: not kicking state machine for silent group", __func__);
2197 case MLD_REPORTING_MEMBER:
2198 case MLD_IDLE_MEMBER:
2199 case MLD_G_QUERY_PENDING_MEMBER:
2200 case MLD_SG_QUERY_PENDING_MEMBER:
2201 if (mli->mli_version == MLD_VERSION_1) {
2203 if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2204 inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2205 panic("%s: MLDv2 state reached, not MLDv2 mode",
2208 mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2209 inm->in6m_state = MLD_NOT_MEMBER;
2210 V_current_state_timers_running6 = 1;
2211 } else if (mli->mli_version == MLD_VERSION_2) {
2213 * Stop group timer and all pending reports.
2214 * Immediately enqueue a state-change report
2215 * TO_IN {} to be sent on the next fast timeout,
2216 * giving us an opportunity to merge reports.
2218 mbufq_drain(&inm->in6m_scq);
2219 inm->in6m_timer = 0;
2220 inm->in6m_scrv = mli->mli_rv;
2221 CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2222 "pending retransmissions.", __func__,
2223 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2224 if_name(inm->in6m_ifp), inm->in6m_scrv);
2225 if (inm->in6m_scrv == 0) {
2226 inm->in6m_state = MLD_NOT_MEMBER;
2227 inm->in6m_sctimer = 0;
2231 in6m_acquire_locked(inm);
2233 retval = mld_v2_enqueue_group_record(
2234 &inm->in6m_scq, inm, 1, 0, 0,
2235 (mli->mli_flags & MLIF_USEALLOW));
2236 KASSERT(retval != 0,
2237 ("%s: enqueue record = %d", __func__,
2240 inm->in6m_state = MLD_LEAVING_MEMBER;
2241 inm->in6m_sctimer = 1;
2242 V_state_change_timers_running6 = 1;
2248 case MLD_LAZY_MEMBER:
2249 case MLD_SLEEPING_MEMBER:
2250 case MLD_AWAKENING_MEMBER:
2251 /* Our reports are suppressed; do nothing. */
2257 CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2258 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2259 if_name(inm->in6m_ifp));
2260 inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2261 CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2262 __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2267 * Enqueue an MLDv2 group record to the given output queue.
2269 * If is_state_change is zero, a current-state record is appended.
2270 * If is_state_change is non-zero, a state-change report is appended.
2272 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2273 * If is_group_query is zero, and if there is a packet with free space
2274 * at the tail of the queue, it will be appended to providing there
2275 * is enough free space.
2276 * Otherwise a new mbuf packet chain is allocated.
2278 * If is_source_query is non-zero, each source is checked to see if
2279 * it was recorded for a Group-Source query, and will be omitted if
2280 * it is not both in-mode and recorded.
2282 * If use_block_allow is non-zero, state change reports for initial join
2283 * and final leave, on an inclusive mode group with a source list, will be
2284 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2286 * The function will attempt to allocate leading space in the packet
2287 * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2289 * If successful the size of all data appended to the queue is returned,
2290 * otherwise an error code less than zero is returned, or zero if
2291 * no record(s) were appended.
2294 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2295 const int is_state_change, const int is_group_query,
2296 const int is_source_query, const int use_block_allow)
2298 struct mldv2_record mr;
2299 struct mldv2_record *pmr;
2301 struct ip6_msource *ims, *nims;
2302 struct mbuf *m0, *m, *md;
2303 int is_filter_list_change;
2304 int minrec0len, m0srcs, msrcs, nbytes, off;
2305 int record_has_sources;
2310 char ip6tbuf[INET6_ADDRSTRLEN];
2313 IN6_MULTI_LIST_LOCK_ASSERT();
2315 ifp = inm->in6m_ifp;
2316 is_filter_list_change = 0;
2323 record_has_sources = 1;
2325 type = MLD_DO_NOTHING;
2326 mode = inm->in6m_st[1].iss_fmode;
2329 * If we did not transition out of ASM mode during t0->t1,
2330 * and there are no source nodes to process, we can skip
2331 * the generation of source records.
2333 if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2334 inm->in6m_nsrc == 0)
2335 record_has_sources = 0;
2337 if (is_state_change) {
2339 * Queue a state change record.
2340 * If the mode did not change, and there are non-ASM
2341 * listeners or source filters present,
2342 * we potentially need to issue two records for the group.
2343 * If there are ASM listeners, and there was no filter
2344 * mode transition of any kind, do nothing.
2346 * If we are transitioning to MCAST_UNDEFINED, we need
2347 * not send any sources. A transition to/from this state is
2348 * considered inclusive with some special treatment.
2350 * If we are rewriting initial joins/leaves to use
2351 * ALLOW/BLOCK, and the group's membership is inclusive,
2352 * we need to send sources in all cases.
2354 if (mode != inm->in6m_st[0].iss_fmode) {
2355 if (mode == MCAST_EXCLUDE) {
2356 CTR1(KTR_MLD, "%s: change to EXCLUDE",
2358 type = MLD_CHANGE_TO_EXCLUDE_MODE;
2360 CTR1(KTR_MLD, "%s: change to INCLUDE",
2362 if (use_block_allow) {
2365 * Here we're interested in state
2366 * edges either direction between
2367 * MCAST_UNDEFINED and MCAST_INCLUDE.
2368 * Perhaps we should just check
2369 * the group state, rather than
2372 if (mode == MCAST_UNDEFINED) {
2373 type = MLD_BLOCK_OLD_SOURCES;
2375 type = MLD_ALLOW_NEW_SOURCES;
2378 type = MLD_CHANGE_TO_INCLUDE_MODE;
2379 if (mode == MCAST_UNDEFINED)
2380 record_has_sources = 0;
2384 if (record_has_sources) {
2385 is_filter_list_change = 1;
2387 type = MLD_DO_NOTHING;
2392 * Queue a current state record.
2394 if (mode == MCAST_EXCLUDE) {
2395 type = MLD_MODE_IS_EXCLUDE;
2396 } else if (mode == MCAST_INCLUDE) {
2397 type = MLD_MODE_IS_INCLUDE;
2398 KASSERT(inm->in6m_st[1].iss_asm == 0,
2399 ("%s: inm %p is INCLUDE but ASM count is %d",
2400 __func__, inm, inm->in6m_st[1].iss_asm));
2405 * Generate the filter list changes using a separate function.
2407 if (is_filter_list_change)
2408 return (mld_v2_enqueue_filter_change(mq, inm));
2410 if (type == MLD_DO_NOTHING) {
2411 CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2412 __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2413 if_name(inm->in6m_ifp));
2418 * If any sources are present, we must be able to fit at least
2419 * one in the trailing space of the tail packet's mbuf,
2422 minrec0len = sizeof(struct mldv2_record);
2423 if (record_has_sources)
2424 minrec0len += sizeof(struct in6_addr);
2426 CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2427 mld_rec_type_to_str(type),
2428 ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2429 if_name(inm->in6m_ifp));
2432 * Check if we have a packet in the tail of the queue for this
2433 * group into which the first group record for this group will fit.
2434 * Otherwise allocate a new packet.
2435 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2436 * Note: Group records for G/GSR query responses MUST be sent
2437 * in their own packet.
2439 m0 = mbufq_last(mq);
2440 if (!is_group_query &&
2442 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2443 (m0->m_pkthdr.len + minrec0len) <
2444 (ifp->if_mtu - MLD_MTUSPACE)) {
2445 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2446 sizeof(struct mldv2_record)) /
2447 sizeof(struct in6_addr);
2449 CTR1(KTR_MLD, "%s: use existing packet", __func__);
2451 if (mbufq_full(mq)) {
2452 CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2456 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2457 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2458 if (!is_state_change && !is_group_query)
2459 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2461 m = m_gethdr(M_NOWAIT, MT_DATA);
2465 mld_save_context(m, ifp);
2467 CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2471 * Append group record.
2472 * If we have sources, we don't know how many yet.
2477 mr.mr_addr = inm->in6m_addr;
2478 in6_clearscope(&mr.mr_addr);
2479 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2482 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2485 nbytes += sizeof(struct mldv2_record);
2488 * Append as many sources as will fit in the first packet.
2489 * If we are appending to a new packet, the chain allocation
2490 * may potentially use clusters; use m_getptr() in this case.
2491 * If we are appending to an existing packet, we need to obtain
2492 * a pointer to the group record after m_append(), in case a new
2493 * mbuf was allocated.
2495 * Only append sources which are in-mode at t1. If we are
2496 * transitioning to MCAST_UNDEFINED state on the group, and
2497 * use_block_allow is zero, do not include source entries.
2498 * Otherwise, we need to include this source in the report.
2500 * Only report recorded sources in our filter set when responding
2501 * to a group-source query.
2503 if (record_has_sources) {
2506 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2507 md->m_len - nbytes);
2509 md = m_getptr(m, 0, &off);
2510 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2514 RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2516 CTR2(KTR_MLD, "%s: visit node %s", __func__,
2517 ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2518 now = im6s_get_mode(inm, ims, 1);
2519 CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2520 if ((now != mode) ||
2522 (!use_block_allow && mode == MCAST_UNDEFINED))) {
2523 CTR1(KTR_MLD, "%s: skip node", __func__);
2526 if (is_source_query && ims->im6s_stp == 0) {
2527 CTR1(KTR_MLD, "%s: skip unrecorded node",
2531 CTR1(KTR_MLD, "%s: append node", __func__);
2532 if (!m_append(m, sizeof(struct in6_addr),
2533 (void *)&ims->im6s_addr)) {
2536 CTR1(KTR_MLD, "%s: m_append() failed.",
2540 nbytes += sizeof(struct in6_addr);
2542 if (msrcs == m0srcs)
2545 CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2547 pmr->mr_numsrc = htons(msrcs);
2548 nbytes += (msrcs * sizeof(struct in6_addr));
2551 if (is_source_query && msrcs == 0) {
2552 CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2559 * We are good to go with first packet.
2562 CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2563 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2564 mbufq_enqueue(mq, m);
2566 m->m_pkthdr.PH_vt.vt_nrecs++;
2569 * No further work needed if no source list in packet(s).
2571 if (!record_has_sources)
2575 * Whilst sources remain to be announced, we need to allocate
2576 * a new packet and fill out as many sources as will fit.
2577 * Always try for a cluster first.
2579 while (nims != NULL) {
2580 if (mbufq_full(mq)) {
2581 CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2584 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2586 m = m_gethdr(M_NOWAIT, MT_DATA);
2589 mld_save_context(m, ifp);
2590 md = m_getptr(m, 0, &off);
2591 pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2592 CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2594 if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2597 CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2600 m->m_pkthdr.PH_vt.vt_nrecs = 1;
2601 nbytes += sizeof(struct mldv2_record);
2603 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2604 sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2607 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2608 CTR2(KTR_MLD, "%s: visit node %s",
2609 __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2610 now = im6s_get_mode(inm, ims, 1);
2611 if ((now != mode) ||
2613 (!use_block_allow && mode == MCAST_UNDEFINED))) {
2614 CTR1(KTR_MLD, "%s: skip node", __func__);
2617 if (is_source_query && ims->im6s_stp == 0) {
2618 CTR1(KTR_MLD, "%s: skip unrecorded node",
2622 CTR1(KTR_MLD, "%s: append node", __func__);
2623 if (!m_append(m, sizeof(struct in6_addr),
2624 (void *)&ims->im6s_addr)) {
2627 CTR1(KTR_MLD, "%s: m_append() failed.",
2632 if (msrcs == m0srcs)
2635 pmr->mr_numsrc = htons(msrcs);
2636 nbytes += (msrcs * sizeof(struct in6_addr));
2638 CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2639 mbufq_enqueue(mq, m);
2646 * Type used to mark record pass completion.
2647 * We exploit the fact we can cast to this easily from the
2648 * current filter modes on each ip_msource node.
2651 REC_NONE = 0x00, /* MCAST_UNDEFINED */
2652 REC_ALLOW = 0x01, /* MCAST_INCLUDE */
2653 REC_BLOCK = 0x02, /* MCAST_EXCLUDE */
2654 REC_FULL = REC_ALLOW | REC_BLOCK
2658 * Enqueue an MLDv2 filter list change to the given output queue.
2660 * Source list filter state is held in an RB-tree. When the filter list
2661 * for a group is changed without changing its mode, we need to compute
2662 * the deltas between T0 and T1 for each source in the filter set,
2663 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2665 * As we may potentially queue two record types, and the entire R-B tree
2666 * needs to be walked at once, we break this out into its own function
2667 * so we can generate a tightly packed queue of packets.
2669 * XXX This could be written to only use one tree walk, although that makes
2670 * serializing into the mbuf chains a bit harder. For now we do two walks
2671 * which makes things easier on us, and it may or may not be harder on
2674 * If successful the size of all data appended to the queue is returned,
2675 * otherwise an error code less than zero is returned, or zero if
2676 * no record(s) were appended.
2679 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2681 static const int MINRECLEN =
2682 sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2684 struct mldv2_record mr;
2685 struct mldv2_record *pmr;
2686 struct ip6_msource *ims, *nims;
2687 struct mbuf *m, *m0, *md;
2688 int m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2690 uint8_t mode, now, then;
2691 rectype_t crt, drt, nrt;
2693 char ip6tbuf[INET6_ADDRSTRLEN];
2696 IN6_MULTI_LIST_LOCK_ASSERT();
2698 if (inm->in6m_nsrc == 0 ||
2699 (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2702 ifp = inm->in6m_ifp; /* interface */
2703 mode = inm->in6m_st[1].iss_fmode; /* filter mode at t1 */
2704 crt = REC_NONE; /* current group record type */
2705 drt = REC_NONE; /* mask of completed group record types */
2706 nrt = REC_NONE; /* record type for current node */
2707 m0srcs = 0; /* # source which will fit in current mbuf chain */
2708 npbytes = 0; /* # of bytes appended this packet */
2709 nbytes = 0; /* # of bytes appended to group's state-change queue */
2710 rsrcs = 0; /* # sources encoded in current record */
2711 schanged = 0; /* # nodes encoded in overall filter change */
2712 nallow = 0; /* # of source entries in ALLOW_NEW */
2713 nblock = 0; /* # of source entries in BLOCK_OLD */
2714 nims = NULL; /* next tree node pointer */
2717 * For each possible filter record mode.
2718 * The first kind of source we encounter tells us which
2719 * is the first kind of record we start appending.
2720 * If a node transitioned to UNDEFINED at t1, its mode is treated
2721 * as the inverse of the group's filter mode.
2723 while (drt != REC_FULL) {
2725 m0 = mbufq_last(mq);
2727 (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2728 MLD_V2_REPORT_MAXRECS) &&
2729 (m0->m_pkthdr.len + MINRECLEN) <
2730 (ifp->if_mtu - MLD_MTUSPACE)) {
2732 m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2733 sizeof(struct mldv2_record)) /
2734 sizeof(struct in6_addr);
2736 "%s: use previous packet", __func__);
2738 m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2740 m = m_gethdr(M_NOWAIT, MT_DATA);
2743 "%s: m_get*() failed", __func__);
2746 m->m_pkthdr.PH_vt.vt_nrecs = 0;
2747 mld_save_context(m, ifp);
2748 m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2749 sizeof(struct mldv2_record)) /
2750 sizeof(struct in6_addr);
2753 "%s: allocated new packet", __func__);
2756 * Append the MLD group record header to the
2757 * current packet's data area.
2758 * Recalculate pointer to free space for next
2759 * group record, in case m_append() allocated
2760 * a new mbuf or cluster.
2762 memset(&mr, 0, sizeof(mr));
2763 mr.mr_addr = inm->in6m_addr;
2764 in6_clearscope(&mr.mr_addr);
2765 if (!m_append(m, sizeof(mr), (void *)&mr)) {
2769 "%s: m_append() failed", __func__);
2772 npbytes += sizeof(struct mldv2_record);
2774 /* new packet; offset in chain */
2775 md = m_getptr(m, npbytes -
2776 sizeof(struct mldv2_record), &off);
2777 pmr = (struct mldv2_record *)(mtod(md,
2780 /* current packet; offset from last append */
2782 pmr = (struct mldv2_record *)(mtod(md,
2783 uint8_t *) + md->m_len -
2784 sizeof(struct mldv2_record));
2787 * Begin walking the tree for this record type
2788 * pass, or continue from where we left off
2789 * previously if we had to allocate a new packet.
2790 * Only report deltas in-mode at t1.
2791 * We need not report included sources as allowed
2792 * if we are in inclusive mode on the group,
2793 * however the converse is not true.
2797 nims = RB_MIN(ip6_msource_tree,
2800 RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2801 CTR2(KTR_MLD, "%s: visit node %s", __func__,
2802 ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2803 now = im6s_get_mode(inm, ims, 1);
2804 then = im6s_get_mode(inm, ims, 0);
2805 CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2806 __func__, then, now);
2809 "%s: skip unchanged", __func__);
2812 if (mode == MCAST_EXCLUDE &&
2813 now == MCAST_INCLUDE) {
2815 "%s: skip IN src on EX group",
2819 nrt = (rectype_t)now;
2820 if (nrt == REC_NONE)
2821 nrt = (rectype_t)(~mode & REC_FULL);
2822 if (schanged++ == 0) {
2824 } else if (crt != nrt)
2826 if (!m_append(m, sizeof(struct in6_addr),
2827 (void *)&ims->im6s_addr)) {
2831 "%s: m_append() failed", __func__);
2834 nallow += !!(crt == REC_ALLOW);
2835 nblock += !!(crt == REC_BLOCK);
2836 if (++rsrcs == m0srcs)
2840 * If we did not append any tree nodes on this
2841 * pass, back out of allocations.
2844 npbytes -= sizeof(struct mldv2_record);
2847 "%s: m_free(m)", __func__);
2851 "%s: m_adj(m, -mr)", __func__);
2852 m_adj(m, -((int)sizeof(
2853 struct mldv2_record)));
2857 npbytes += (rsrcs * sizeof(struct in6_addr));
2858 if (crt == REC_ALLOW)
2859 pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2860 else if (crt == REC_BLOCK)
2861 pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2862 pmr->mr_numsrc = htons(rsrcs);
2864 * Count the new group record, and enqueue this
2865 * packet if it wasn't already queued.
2867 m->m_pkthdr.PH_vt.vt_nrecs++;
2869 mbufq_enqueue(mq, m);
2871 } while (nims != NULL);
2873 crt = (~crt & REC_FULL);
2876 CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2883 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2886 struct mbuf *m; /* pending state-change */
2887 struct mbuf *m0; /* copy of pending state-change */
2888 struct mbuf *mt; /* last state-change in packet */
2889 int docopy, domerge;
2896 IN6_MULTI_LIST_LOCK_ASSERT();
2900 * If there are further pending retransmissions, make a writable
2901 * copy of each queued state-change message before merging.
2903 if (inm->in6m_scrv > 0)
2906 gq = &inm->in6m_scq;
2908 if (mbufq_first(gq) == NULL) {
2909 CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2914 m = mbufq_first(gq);
2917 * Only merge the report into the current packet if
2918 * there is sufficient space to do so; an MLDv2 report
2919 * packet may only contain 65,535 group records.
2920 * Always use a simple mbuf chain concatentation to do this,
2921 * as large state changes for single groups may have
2922 * allocated clusters.
2925 mt = mbufq_last(scq);
2927 recslen = m_length(m, NULL);
2929 if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2930 m->m_pkthdr.PH_vt.vt_nrecs <=
2931 MLD_V2_REPORT_MAXRECS) &&
2932 (mt->m_pkthdr.len + recslen <=
2933 (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2937 if (!domerge && mbufq_full(gq)) {
2939 "%s: outbound queue full, skipping whole packet %p",
2949 CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2950 m0 = mbufq_dequeue(gq);
2953 CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2954 m0 = m_dup(m, M_NOWAIT);
2957 m0->m_nextpkt = NULL;
2962 CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2964 mbufq_enqueue(scq, m0);
2966 struct mbuf *mtl; /* last mbuf of packet mt */
2968 CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2972 m0->m_flags &= ~M_PKTHDR;
2973 mt->m_pkthdr.len += recslen;
2974 mt->m_pkthdr.PH_vt.vt_nrecs +=
2975 m0->m_pkthdr.PH_vt.vt_nrecs;
2985 * Respond to a pending MLDv2 General Query.
2988 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2990 struct epoch_tracker et;
2991 struct ifmultiaddr *ifma;
2993 struct in6_multi *inm;
2996 IN6_MULTI_LIST_LOCK_ASSERT();
2999 KASSERT(mli->mli_version == MLD_VERSION_2,
3000 ("%s: called when version %d", __func__, mli->mli_version));
3003 * Check that there are some packets queued. If so, send them first.
3004 * For large number of groups the reply to general query can take
3005 * many packets, we should finish sending them before starting of
3006 * queuing the new reply.
3008 if (mbufq_len(&mli->mli_gq) != 0)
3013 NET_EPOCH_ENTER(et);
3014 CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3015 if (ifma->ifma_addr->sa_family != AF_INET6 ||
3016 ifma->ifma_protospec == NULL)
3019 inm = (struct in6_multi *)ifma->ifma_protospec;
3020 KASSERT(ifp == inm->in6m_ifp,
3021 ("%s: inconsistent ifp", __func__));
3023 switch (inm->in6m_state) {
3024 case MLD_NOT_MEMBER:
3025 case MLD_SILENT_MEMBER:
3027 case MLD_REPORTING_MEMBER:
3028 case MLD_IDLE_MEMBER:
3029 case MLD_LAZY_MEMBER:
3030 case MLD_SLEEPING_MEMBER:
3031 case MLD_AWAKENING_MEMBER:
3032 inm->in6m_state = MLD_REPORTING_MEMBER;
3033 retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3035 CTR2(KTR_MLD, "%s: enqueue record = %d",
3038 case MLD_G_QUERY_PENDING_MEMBER:
3039 case MLD_SG_QUERY_PENDING_MEMBER:
3040 case MLD_LEAVING_MEMBER:
3047 mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3050 * Slew transmission of bursts over 500ms intervals.
3052 if (mbufq_first(&mli->mli_gq) != NULL) {
3053 mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3054 MLD_RESPONSE_BURST_INTERVAL);
3055 V_interface_timers_running6 = 1;
3060 * Transmit the next pending message in the output queue.
3062 * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3063 * MRT: Nothing needs to be done, as MLD traffic is always local to
3064 * a link and uses a link-scope multicast address.
3067 mld_dispatch_packet(struct mbuf *m)
3069 struct ip6_moptions im6o;
3074 struct ip6_hdr *ip6;
3075 struct mld_hdr *mld;
3081 CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3084 * Set VNET image pointer from enqueued mbuf chain
3085 * before doing anything else. Whilst we use interface
3086 * indexes to guard against interface detach, they are
3087 * unique to each VIMAGE and must be retrieved.
3089 ifindex = mld_restore_context(m);
3092 * Check if the ifnet still exists. This limits the scope of
3093 * any race in the absence of a global ifp lock for low cost
3094 * (an array lookup).
3096 ifp = ifnet_byindex(ifindex);
3098 CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3099 __func__, m, ifindex);
3101 IP6STAT_INC(ip6s_noroute);
3105 im6o.im6o_multicast_hlim = 1;
3106 im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3107 im6o.im6o_multicast_ifp = ifp;
3109 if (m->m_flags & M_MLDV1) {
3112 m0 = mld_v2_encap_report(ifp, m);
3114 CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3115 IP6STAT_INC(ip6s_odropped);
3120 mld_scrub_context(m0);
3122 m0->m_pkthdr.rcvif = V_loif;
3124 ip6 = mtod(m0, struct ip6_hdr *);
3126 (void)in6_setscope(&ip6->ip6_dst, ifp, NULL); /* XXX LOR */
3129 * XXX XXX Break some KPI rules to prevent an LOR which would
3130 * occur if we called in6_setscope() at transmission.
3131 * See comments at top of file.
3133 MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3137 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3138 * so we can bump the stats.
3140 md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3141 mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3142 type = mld->mld_type;
3144 error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3147 CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3150 ICMP6STAT_INC(icp6s_outhist[type]);
3152 icmp6_ifstat_inc(oifp, ifs6_out_msg);
3154 case MLD_LISTENER_REPORT:
3155 case MLDV2_LISTENER_REPORT:
3156 icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3158 case MLD_LISTENER_DONE:
3159 icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3168 * Encapsulate an MLDv2 report.
3170 * KAME IPv6 requires that hop-by-hop options be passed separately,
3171 * and that the IPv6 header be prepended in a separate mbuf.
3173 * Returns a pointer to the new mbuf chain head, or NULL if the
3174 * allocation failed.
3176 static struct mbuf *
3177 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3180 struct mldv2_report *mld;
3181 struct ip6_hdr *ip6;
3182 struct in6_ifaddr *ia;
3185 KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3186 KASSERT((m->m_flags & M_PKTHDR),
3187 ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3190 * RFC3590: OK to send as :: or tentative during DAD.
3192 ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3194 CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3196 mh = m_gethdr(M_NOWAIT, MT_DATA);
3199 ifa_free(&ia->ia_ifa);
3203 M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3205 mldreclen = m_length(m, NULL);
3206 CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3208 mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3209 mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3210 sizeof(struct mldv2_report) + mldreclen;
3212 ip6 = mtod(mh, struct ip6_hdr *);
3214 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3215 ip6->ip6_vfc |= IPV6_VERSION;
3216 ip6->ip6_nxt = IPPROTO_ICMPV6;
3217 ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3219 ifa_free(&ia->ia_ifa);
3220 ip6->ip6_dst = in6addr_linklocal_allv2routers;
3221 /* scope ID will be set in netisr */
3223 mld = (struct mldv2_report *)(ip6 + 1);
3224 mld->mld_type = MLDV2_LISTENER_REPORT;
3227 mld->mld_v2_reserved = 0;
3228 mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3229 m->m_pkthdr.PH_vt.vt_nrecs = 0;
3232 mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3233 sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3239 mld_rec_type_to_str(const int type)
3243 case MLD_CHANGE_TO_EXCLUDE_MODE:
3246 case MLD_CHANGE_TO_INCLUDE_MODE:
3249 case MLD_MODE_IS_EXCLUDE:
3252 case MLD_MODE_IS_INCLUDE:
3255 case MLD_ALLOW_NEW_SOURCES:
3258 case MLD_BLOCK_OLD_SOURCES:
3269 mld_init(void *unused __unused)
3272 CTR1(KTR_MLD, "%s: initializing", __func__);
3275 ip6_initpktopts(&mld_po);
3276 mld_po.ip6po_hlim = 1;
3277 mld_po.ip6po_hbh = &mld_ra.hbh;
3278 mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3279 mld_po.ip6po_flags = IP6PO_DONTFRAG;
3281 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3284 mld_uninit(void *unused __unused)
3287 CTR1(KTR_MLD, "%s: tearing down", __func__);
3290 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3293 vnet_mld_init(const void *unused __unused)
3296 CTR1(KTR_MLD, "%s: initializing", __func__);
3298 LIST_INIT(&V_mli_head);
3300 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3304 vnet_mld_uninit(const void *unused __unused)
3307 /* This can happen if we shutdown the network stack. */
3308 CTR1(KTR_MLD, "%s: tearing down", __func__);
3310 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3314 mld_modevent(module_t mod, int type, void *unused __unused)
3322 return (EOPNOTSUPP);
3327 static moduledata_t mld_mod = {
3332 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);