]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/netinet6/mld6.c
xnu-4903.270.47.tar.gz
[apple/xnu.git] / bsd / netinet6 / mld6.c
index 6117d50bc1601b6e983422a35a9387e19302b329..60b7777cded0d469975979fb8cca3b803ef4d57a 100644 (file)
@@ -1,8 +1,32 @@
-/*     $KAME: mld6.c,v 1.17 2000/03/01 12:37:25 itojun Exp $   */
-
 /*
- * Copyright (C) 1998 WIDE Project.
- * All rights reserved.
+ * Copyright (c) 2000-2017 Apple Inc. All rights reserved.
+ *
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
+ *
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ *
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ *
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
+ * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
+ *
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
+ */
+/*-
+ * Copyright (c) 2009 Bruce Simpson.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * 2. Redistributions in binary form must reproduce the above copyright
  *    notice, this list of conditions and the following disclaimer in the
  *    documentation and/or other materials provided with the distribution.
- * 3. Neither the name of the project nor the names of its contributors
- *    may be used to endorse or promote products derived from this software
- *    without specific prior written permission.
+ * 3. The name of the author may not be used to endorse or promote
+ *    products derived from this software without specific prior written
+ *    permission.
  *
- * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  *
  *     @(#)igmp.c      8.1 (Berkeley) 7/19/93
  */
+/*
+ * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
+ * support for mandatory and extensible security protections.  This notice
+ * is included in support of clause 2.2 (b) of the Apple Public License,
+ * Version 2.0.
+ */
 
-#if (defined(__FreeBSD__) && __FreeBSD__ >= 3) || defined(__NetBSD__)
-#include "opt_inet.h"
-#endif
+#include <sys/cdefs.h>
 
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/mbuf.h>
 #include <sys/socket.h>
 #include <sys/protosw.h>
-#include <sys/syslog.h>
+#include <sys/sysctl.h>
+#include <sys/kernel.h>
+#include <sys/malloc.h>
+#include <sys/mcache.h>
+
+#include <dev/random/randomdev.h>
+
+#include <kern/zalloc.h>
 
 #include <net/if.h>
+#include <net/route.h>
 
 #include <netinet/in.h>
 #include <netinet/in_var.h>
+#include <netinet6/in6_var.h>
 #include <netinet/ip6.h>
 #include <netinet6/ip6_var.h>
+#include <netinet6/scope6_var.h>
 #include <netinet/icmp6.h>
+#include <netinet6/mld6.h>
 #include <netinet6/mld6_var.h>
 
-#include <net/net_osdep.h>
+/* Lock group and attribute for mld_mtx */
+static lck_attr_t       *mld_mtx_attr;
+static lck_grp_t        *mld_mtx_grp;
+static lck_grp_attr_t   *mld_mtx_grp_attr;
 
 /*
- * Protocol constants
+ * Locking and reference counting:
+ *
+ * mld_mtx mainly protects mli_head.  In cases where both mld_mtx and
+ * in6_multihead_lock must be held, the former must be acquired first in order
+ * to maintain lock ordering.  It is not a requirement that mld_mtx be
+ * acquired first before in6_multihead_lock, but in case both must be acquired
+ * in succession, the correct lock ordering must be followed.
+ *
+ * Instead of walking the if_multiaddrs list at the interface and returning
+ * the ifma_protospec value of a matching entry, we search the global list
+ * of in6_multi records and find it that way; this is done with in6_multihead
+ * lock held.  Doing so avoids the race condition issues that many other BSDs
+ * suffer from (therefore in our implementation, ifma_protospec will never be
+ * NULL for as long as the in6_multi is valid.)
+ *
+ * The above creates a requirement for the in6_multi to stay in in6_multihead
+ * list even after the final MLD leave (in MLDv2 mode) until no longer needs
+ * be retransmitted (this is not required for MLDv1.)  In order to handle
+ * this, the request and reference counts of the in6_multi are bumped up when
+ * the state changes to MLD_LEAVING_MEMBER, and later dropped in the timeout
+ * handler.  Each in6_multi holds a reference to the underlying mld_ifinfo.
+ *
+ * Thus, the permitted lock order is:
+ *
+ *     mld_mtx, in6_multihead_lock, inm6_lock, mli_lock
+ *
+ * Any may be taken independently, but if any are held at the same time,
+ * the above lock order must be followed.
  */
+static decl_lck_mtx_data(, mld_mtx);
+
+SLIST_HEAD(mld_in6m_relhead, in6_multi);
+
+static void     mli_initvar(struct mld_ifinfo *, struct ifnet *, int);
+static struct mld_ifinfo *mli_alloc(int);
+static void     mli_free(struct mld_ifinfo *);
+static void     mli_delete(const struct ifnet *, struct mld_in6m_relhead *);
+static void     mld_dispatch_packet(struct mbuf *);
+static void     mld_final_leave(struct in6_multi *, struct mld_ifinfo *,
+    struct mld_tparams *);
+static int      mld_handle_state_change(struct in6_multi *, struct mld_ifinfo *,
+    struct mld_tparams *);
+static int      mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
+    struct mld_tparams *, const int);
+#ifdef MLD_DEBUG
+static const char *     mld_rec_type_to_str(const int);
+#endif
+static uint32_t mld_set_version(struct mld_ifinfo *, const int);
+static void     mld_flush_relq(struct mld_ifinfo *, struct mld_in6m_relhead *);
+static void     mld_dispatch_queue_locked(struct mld_ifinfo *, struct ifqueue *, int);
+static int      mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
+    /*const*/ struct mld_hdr *);
+static int      mld_v1_input_report(struct ifnet *, struct mbuf *,
+    const struct ip6_hdr *, /*const*/ struct mld_hdr *);
+static void     mld_v1_process_group_timer(struct in6_multi *, const int);
+static void     mld_v1_process_querier_timers(struct mld_ifinfo *);
+static int      mld_v1_transmit_report(struct in6_multi *, const int);
+static uint32_t mld_v1_update_group(struct in6_multi *, const int);
+static void     mld_v2_cancel_link_timers(struct mld_ifinfo *);
+static uint32_t mld_v2_dispatch_general_query(struct mld_ifinfo *);
+static struct mbuf *
+mld_v2_encap_report(struct ifnet *, struct mbuf *);
+static int      mld_v2_enqueue_filter_change(struct ifqueue *,
+    struct in6_multi *);
+static int      mld_v2_enqueue_group_record(struct ifqueue *,
+    struct in6_multi *, const int, const int, const int,
+    const int);
+static int      mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
+    struct mbuf *, const int, const int);
+static int      mld_v2_merge_state_changes(struct in6_multi *,
+    struct ifqueue *);
+static void     mld_v2_process_group_timers(struct mld_ifinfo *,
+    struct ifqueue *, struct ifqueue *,
+    struct in6_multi *, const int);
+static int      mld_v2_process_group_query(struct in6_multi *,
+    int, struct mbuf *, const int);
+static int      sysctl_mld_gsr SYSCTL_HANDLER_ARGS;
+static int      sysctl_mld_ifinfo SYSCTL_HANDLER_ARGS;
+static int      sysctl_mld_v2enable SYSCTL_HANDLER_ARGS;
+
+static int mld_timeout_run;             /* MLD timer is scheduled to run */
+static void mld_timeout(void *);
+static void mld_sched_timeout(void);
 
-/* denotes that the MLD max response delay field specifies time in milliseconds */
-#define MLD6_TIMER_SCALE       1000
 /*
- * time between repetitions of a node's initial report of interest in a
- * multicast address(in seconds)
+ * Normative references: RFC 2710, RFC 3590, RFC 3810.
  */
-#define MLD6_UNSOLICITED_REPORT_INTERVAL       10
+static struct timeval mld_gsrdelay = {10, 0};
+static LIST_HEAD(, mld_ifinfo) mli_head;
 
-static struct ip6_pktopts ip6_opts;
-static int mld6_timers_are_running;
-/* XXX: These are necessary for KAME's link-local hack */
-static struct in6_addr mld6_all_nodes_linklocal = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
-static struct in6_addr mld6_all_routers_linklocal = IN6ADDR_LINKLOCAL_ALLROUTERS_INIT;
+static int querier_present_timers_running6;
+static int interface_timers_running6;
+static int state_change_timers_running6;
+static int current_state_timers_running6;
 
-static void mld6_sendpkt __P((struct in6_multi *, int, const struct in6_addr *));
+static unsigned int mld_mli_list_genid;
+/*
+ * Subsystem lock macros.
+ */
+#define MLD_LOCK()                      \
+       lck_mtx_lock(&mld_mtx)
+#define MLD_LOCK_ASSERT_HELD()          \
+       LCK_MTX_ASSERT(&mld_mtx, LCK_MTX_ASSERT_OWNED)
+#define MLD_LOCK_ASSERT_NOTHELD()       \
+       LCK_MTX_ASSERT(&mld_mtx, LCK_MTX_ASSERT_NOTOWNED)
+#define MLD_UNLOCK()                    \
+       lck_mtx_unlock(&mld_mtx)
 
-void
-mld6_init()
-{
-       static u_int8_t hbh_buf[8];
-       struct ip6_hbh *hbh = (struct ip6_hbh *)hbh_buf;
-       u_int16_t rtalert_code = htons((u_int16_t)IP6OPT_RTALERT_MLD);
+#define MLD_ADD_DETACHED_IN6M(_head, _in6m) {                           \
+       SLIST_INSERT_HEAD(_head, _in6m, in6m_dtle);                     \
+}
 
-       mld6_timers_are_running = 0;
+#define MLD_REMOVE_DETACHED_IN6M(_head) {                               \
+       struct in6_multi *_in6m, *_inm_tmp;                             \
+       SLIST_FOREACH_SAFE(_in6m, _head, in6m_dtle, _inm_tmp) {         \
+               SLIST_REMOVE(_head, _in6m, in6_multi, in6m_dtle);       \
+               IN6M_REMREF(_in6m);                                     \
+       }                                                               \
+       VERIFY(SLIST_EMPTY(_head));                                     \
+}
 
-       /* ip6h_nxt will be fill in later */
-       hbh->ip6h_len = 0;      /* (8 >> 3) - 1 */
+#define MLI_ZONE_MAX            64              /* maximum elements in zone */
+#define MLI_ZONE_NAME           "mld_ifinfo"    /* zone name */
 
-       /* XXX: grotty hard coding... */
-       hbh_buf[2] = IP6OPT_PADN;       /* 2 byte padding */
-       hbh_buf[3] = 0;
-       hbh_buf[4] = IP6OPT_RTALERT;
-       hbh_buf[5] = IP6OPT_RTALERT_LEN - 2;
-       bcopy((caddr_t)&rtalert_code, &hbh_buf[6], sizeof(u_int16_t));
+static unsigned int mli_size;                   /* size of zone element */
+static struct zone *mli_zone;                   /* zone for mld_ifinfo */
 
-       ip6_opts.ip6po_hbh = hbh;
-       /* We will specify the hoplimit by a multicast option. */
-       ip6_opts.ip6po_hlim = -1;
-}
+SYSCTL_DECL(_net_inet6);        /* Note: Not in any common header. */
 
-void
-mld6_start_listening(in6m)
-       struct in6_multi *in6m;
-{
-#ifdef __NetBSD__
-       int s = splsoftnet();
-#else
-       int s = splnet();
+SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_LOCKED, 0,
+    "IPv6 Multicast Listener Discovery");
+SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    &mld_gsrdelay.tv_sec, 0, sysctl_mld_gsr, "I",
+    "Rate limit for MLDv2 Group-and-Source queries in seconds");
+
+SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_LOCKED,
+    sysctl_mld_ifinfo, "Per-interface MLDv2 state");
+
+static int      mld_v1enable = 1;
+SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW | CTLFLAG_LOCKED,
+    &mld_v1enable, 0, "Enable fallback to MLDv1");
+
+static int      mld_v2enable = 1;
+SYSCTL_PROC(_net_inet6_mld, OID_AUTO, v2enable,
+    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
+    &mld_v2enable, 0, sysctl_mld_v2enable, "I",
+    "Enable MLDv2 (debug purposes only)");
+
+static int      mld_use_allow = 1;
+SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW | CTLFLAG_LOCKED,
+    &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
+
+#ifdef MLD_DEBUG
+int mld_debug = 0;
+SYSCTL_INT(_net_inet6_mld, OID_AUTO,
+    debug, CTLFLAG_RW | CTLFLAG_LOCKED, &mld_debug, 0, "");
 #endif
+/*
+ * Packed Router Alert option structure declaration.
+ */
+struct mld_raopt {
+       struct ip6_hbh          hbh;
+       struct ip6_opt          pad;
+       struct ip6_opt_router   ra;
+} __packed;
 
-       /*
-        * RFC2710 page 10:
-        * The node never sends a Report or Done for the link-scope all-nodes
-        * address.
-        * MLD messages are never sent for multicast addresses whose scope is 0
-        * (reserved) or 1 (node-local).
-        */
-       mld6_all_nodes_linklocal.s6_addr16[1] =
-               htons(in6m->in6m_ifp->if_index); /* XXX */
-       if (IN6_ARE_ADDR_EQUAL(&in6m->in6m_addr, &mld6_all_nodes_linklocal) ||
-           IPV6_ADDR_MC_SCOPE(&in6m->in6m_addr) < IPV6_ADDR_SCOPE_LINKLOCAL) {
-               in6m->in6m_timer = 0;
-               in6m->in6m_state = MLD6_OTHERLISTENER;
-       } else {
-               mld6_sendpkt(in6m, MLD6_LISTENER_REPORT, NULL);
-               in6m->in6m_timer = MLD6_RANDOM_DELAY(
-                       MLD6_UNSOLICITED_REPORT_INTERVAL * PR_FASTHZ);
-               in6m->in6m_state = MLD6_IREPORTEDLAST;
-               mld6_timers_are_running = 1;
+/*
+ * Router Alert hop-by-hop option header.
+ */
+static struct mld_raopt mld_ra = {
+       .hbh = { 0, 0 },
+       .pad = { .ip6o_type = IP6OPT_PADN, 0 },
+       .ra = {
+               .ip6or_type = (u_int8_t)IP6OPT_ROUTER_ALERT,
+               .ip6or_len = (u_int8_t)(IP6OPT_RTALERT_LEN - 2),
+               .ip6or_value =  {((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
+                                (IP6OPT_RTALERT_MLD & 0xFF) }
        }
-       splx(s);
+};
+static struct ip6_pktopts mld_po;
+
+/* Store MLDv2 record count in the module private scratch space */
+#define vt_nrecs        pkt_mpriv.__mpriv_u.__mpriv32[0].__mpriv32_u.__val16[0]
+
+static __inline void
+mld_save_context(struct mbuf *m, struct ifnet *ifp)
+{
+       m->m_pkthdr.rcvif = ifp;
 }
 
-void
-mld6_stop_listening(in6m)
-       struct in6_multi *in6m;
+static __inline void
+mld_scrub_context(struct mbuf *m)
 {
-       mld6_all_nodes_linklocal.s6_addr16[1] =
-               htons(in6m->in6m_ifp->if_index); /* XXX */
-       mld6_all_routers_linklocal.s6_addr16[1] =
-               htons(in6m->in6m_ifp->if_index); /* XXX: necessary when mrouting */
+       m->m_pkthdr.rcvif = NULL;
+}
 
-       if (in6m->in6m_state == MLD6_IREPORTEDLAST &&
-           (!IN6_ARE_ADDR_EQUAL(&in6m->in6m_addr, &mld6_all_nodes_linklocal)) &&
-           IPV6_ADDR_MC_SCOPE(&in6m->in6m_addr) > IPV6_ADDR_SCOPE_NODELOCAL)
-               mld6_sendpkt(in6m, MLD6_LISTENER_DONE,
-                            &mld6_all_routers_linklocal);
+/*
+ * Restore context from a queued output chain.
+ * Return saved ifp.
+ */
+static __inline struct ifnet *
+mld_restore_context(struct mbuf *m)
+{
+       return m->m_pkthdr.rcvif;
 }
 
-void
-mld6_input(m, off)
-       struct mbuf *m;
-       int off;
-{
-       struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
-       struct mld6_hdr *mldh;
-       struct ifnet *ifp = m->m_pkthdr.rcvif;
-       struct in6_multi *in6m;
-       struct in6_ifaddr *ia;
-#if defined(__FreeBSD__) && __FreeBSD__ >= 3 || defined (__APPLE__)
-       struct ifmultiaddr *ifma;
-#endif
-       int timer;              /* timer value in the MLD query header */
+/*
+ * Retrieve or set threshold between group-source queries in seconds.
+ */
+static int
+sysctl_mld_gsr SYSCTL_HANDLER_ARGS
+{
+#pragma unused(arg1, arg2)
+       int error;
+       int i;
 
-       /* source address validation */
-       if (!IN6_IS_ADDR_LINKLOCAL(&ip6->ip6_src)) {
-               log(LOG_ERR,
-                   "mld6_input: src %s is not link-local\n",
-                   ip6_sprintf(&ip6->ip6_src));
-               /*
-                * spec (RFC2710) does not explicitly
-                * specify to discard the packet from a non link-local
-                * source address. But we believe it's expected to do so.
-                */
-               m_freem(m);
-               return;
+       MLD_LOCK();
+
+       i = mld_gsrdelay.tv_sec;
+
+       error = sysctl_handle_int(oidp, &i, 0, req);
+       if (error || !req->newptr) {
+               goto out_locked;
        }
 
-#ifndef PULLDOWN_TEST
-       IP6_EXTHDR_CHECK(m, off, sizeof(*mldh),);
-       mldh = (struct mld6_hdr *)(mtod(m, caddr_t) + off);
-#else
-       IP6_EXTHDR_GET(mldh, struct mld6_hdr *, m, off, sizeof(*mldh));
-       if (mldh == NULL) {
-               icmp6stat.icp6s_tooshort++;
-               return;
+       if (i < -1 || i >= 60) {
+               error = EINVAL;
+               goto out_locked;
        }
-#endif
 
+       mld_gsrdelay.tv_sec = i;
+
+out_locked:
+       MLD_UNLOCK();
+       return error;
+}
+/*
+ * Expose struct mld_ifinfo to userland, keyed by ifindex.
+ * For use by ifmcstat(8).
+ *
+ */
+static int
+sysctl_mld_ifinfo SYSCTL_HANDLER_ARGS
+{
+#pragma unused(oidp)
+       int                     *name;
+       int                      error;
+       u_int                    namelen;
+       struct ifnet            *ifp;
+       struct mld_ifinfo       *mli;
+       struct mld_ifinfo_u     mli_u;
+
+       name = (int *)arg1;
+       namelen = arg2;
+
+       if (req->newptr != USER_ADDR_NULL) {
+               return EPERM;
+       }
+
+       if (namelen != 1) {
+               return EINVAL;
+       }
+
+       MLD_LOCK();
+
+       if (name[0] <= 0 || name[0] > (u_int)if_index) {
+               error = ENOENT;
+               goto out_locked;
+       }
+
+       error = ENOENT;
+
+       ifnet_head_lock_shared();
+       ifp = ifindex2ifnet[name[0]];
+       ifnet_head_done();
+       if (ifp == NULL) {
+               goto out_locked;
+       }
+
+       bzero(&mli_u, sizeof(mli_u));
+
+       LIST_FOREACH(mli, &mli_head, mli_link) {
+               MLI_LOCK(mli);
+               if (ifp != mli->mli_ifp) {
+                       MLI_UNLOCK(mli);
+                       continue;
+               }
+
+               mli_u.mli_ifindex = mli->mli_ifp->if_index;
+               mli_u.mli_version = mli->mli_version;
+               mli_u.mli_v1_timer = mli->mli_v1_timer;
+               mli_u.mli_v2_timer = mli->mli_v2_timer;
+               mli_u.mli_flags = mli->mli_flags;
+               mli_u.mli_rv = mli->mli_rv;
+               mli_u.mli_qi = mli->mli_qi;
+               mli_u.mli_qri = mli->mli_qri;
+               mli_u.mli_uri = mli->mli_uri;
+               MLI_UNLOCK(mli);
+
+               error = SYSCTL_OUT(req, &mli_u, sizeof(mli_u));
+               break;
+       }
+
+out_locked:
+       MLD_UNLOCK();
+       return error;
+}
+
+static int
+sysctl_mld_v2enable SYSCTL_HANDLER_ARGS
+{
+#pragma unused(arg1, arg2)
+       int error;
+       int i;
+       struct mld_ifinfo *mli;
+       struct mld_tparams mtp = { 0, 0, 0, 0 };
+
+       MLD_LOCK();
+
+       i = mld_v2enable;
+
+       error = sysctl_handle_int(oidp, &i, 0, req);
+       if (error || !req->newptr) {
+               goto out_locked;
+       }
+
+       if (i < 0 || i > 1) {
+               error = EINVAL;
+               goto out_locked;
+       }
+
+       mld_v2enable = i;
        /*
-        * In the MLD6 specification, there are 3 states and a flag.
-        *
-        * In Non-Listener state, we simply don't have a membership record.
-        * In Delaying Listener state, our timer is running (in6m->in6m_timer)
-        * In Idle Listener state, our timer is not running (in6m->in6m_timer==0)
-        *
-        * The flag is in6m->in6m_state, it is set to MLD6_OTHERLISTENER if
-        * we have heard a report from another member, or MLD6_IREPORTEDLAST
-        * if we sent the last report.
+        * If we enabled v2, the state transition will take care of upgrading
+        * the MLD version back to v2. Otherwise, we have to explicitly
+        * downgrade. Note that this functionality is to be used for debugging.
         */
-       switch(mldh->mld6_type) {
-       case MLD6_LISTENER_QUERY:
-               if (ifp->if_flags & IFF_LOOPBACK)
-                       break;
+       if (mld_v2enable == 1) {
+               goto out_locked;
+       }
+
+       LIST_FOREACH(mli, &mli_head, mli_link) {
+               MLI_LOCK(mli);
+               if (mld_set_version(mli, MLD_VERSION_1) > 0) {
+                       mtp.qpt = 1;
+               }
+               MLI_UNLOCK(mli);
+       }
 
-               if (!IN6_IS_ADDR_UNSPECIFIED(&mldh->mld6_addr) &&
-               !IN6_IS_ADDR_MULTICAST(&mldh->mld6_addr))
-                       break;  /* print error or log stat? */
-               if (IN6_IS_ADDR_MC_LINKLOCAL(&mldh->mld6_addr))
-                       mldh->mld6_addr.s6_addr16[1] =
-                               htons(ifp->if_index); /* XXX */
+out_locked:
+       MLD_UNLOCK();
 
-               /*
-                * - Start the timers in all of our membership records
-                *   that the query applies to for the interface on
-                *   which the query arrived excl. those that belong
-                *   to the "all-nodes" group (ff02::1).
-                * - Restart any timer that is already running but has
-                *   A value longer than the requested timeout.
-                * - Use the value specified in the query message as
-                *   the maximum timeout.
-                */
-               IFP_TO_IA6(ifp, ia);
-               if (ia == NULL)
+       mld_set_timeout(&mtp);
+
+       return error;
+}
+
+/*
+ * Dispatch an entire queue of pending packet chains.
+ *
+ * Must not be called with in6m_lock held.
+ * XXX This routine unlocks MLD global lock and also mli locks.
+ * Make sure that the calling routine takes reference on the mli
+ * before calling this routine.
+ * Also if we are traversing mli_head, remember to check for
+ * mli list generation count and restart the loop if generation count
+ * has changed.
+ */
+static void
+mld_dispatch_queue_locked(struct mld_ifinfo *mli, struct ifqueue *ifq, int limit)
+{
+       struct mbuf *m;
+
+       MLD_LOCK_ASSERT_HELD();
+
+       if (mli != NULL) {
+               MLI_LOCK_ASSERT_HELD(mli);
+       }
+
+       for (;;) {
+               IF_DEQUEUE(ifq, m);
+               if (m == NULL) {
                        break;
+               }
+               MLD_PRINTF(("%s: dispatch 0x%llx from 0x%llx\n", __func__,
+                   (uint64_t)VM_KERNEL_ADDRPERM(ifq),
+                   (uint64_t)VM_KERNEL_ADDRPERM(m)));
 
-               /*
-                * XXX: System timer resolution is too low to handle Max
-                * Response Delay, so set 1 to the internal timer even if
-                * the calculated value equals to zero when Max Response
-                * Delay is positive.
-                */
-               timer = ntohs(mldh->mld6_maxdelay)*PR_FASTHZ/MLD6_TIMER_SCALE;
-               if (timer == 0 && mldh->mld6_maxdelay)
-                       timer = 1;
-               mld6_all_nodes_linklocal.s6_addr16[1] =
-                       htons(ifp->if_index); /* XXX */
-               
-#if defined(__FreeBSD__) && __FreeBSD__ >= 3 || defined (__APPLE__)
-               LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link)
-#else
-               for (in6m = ia->ia6_multiaddrs.lh_first;
-                    in6m;
-                    in6m = in6m->in6m_entry.le_next)
-#endif
-               {
-#if defined(__FreeBSD__) && __FreeBSD__ >= 3 || defined (__APPLE__)
-                       if (ifma->ifma_addr->sa_family != AF_INET6)
-                               continue;
-                       in6m = (struct in6_multi *)ifma->ifma_protospec;
-                       if (IN6_ARE_ADDR_EQUAL(&in6m->in6m_addr,
-                                       &mld6_all_nodes_linklocal) ||
-                           IPV6_ADDR_MC_SCOPE(&in6m->in6m_addr) <
-                           IPV6_ADDR_SCOPE_LINKLOCAL)
-                               continue;
-#else
-                       if (IN6_ARE_ADDR_EQUAL(&in6m->in6m_addr,
-                                               &mld6_all_nodes_linklocal) ||
-                           IPV6_ADDR_MC_SCOPE(&in6m->in6m_addr) <
-                           IPV6_ADDR_SCOPE_LINKLOCAL)
-                               continue;
-#endif
+               if (mli != NULL) {
+                       MLI_UNLOCK(mli);
+               }
+               MLD_UNLOCK();
 
-                       if (IN6_IS_ADDR_UNSPECIFIED(&mldh->mld6_addr) ||
-                           IN6_ARE_ADDR_EQUAL(&mldh->mld6_addr,
-                                               &in6m->in6m_addr))
-                       {
-                               if (timer == 0) {
-                                       /* send a report immediately */
-                                       mld6_sendpkt(in6m, MLD6_LISTENER_REPORT,
-                                               NULL);
-                                       in6m->in6m_timer = 0; /* reset timer */
-                                       in6m->in6m_state = MLD6_IREPORTEDLAST;
-                               }
-                               else if (in6m->in6m_timer == 0 || /*idle state*/
-                                       in6m->in6m_timer > timer) {
-                                       in6m->in6m_timer =
-                                               MLD6_RANDOM_DELAY(timer);
-                                       mld6_timers_are_running = 1;
-                               }
-                       }
+               mld_dispatch_packet(m);
+
+               MLD_LOCK();
+               if (mli != NULL) {
+                       MLI_LOCK(mli);
                }
 
-               if (IN6_IS_ADDR_MC_LINKLOCAL(&mldh->mld6_addr))
-                       mldh->mld6_addr.s6_addr16[1] = 0; /* XXX */
-               break;
-       case MLD6_LISTENER_REPORT:
-               /*
-                * For fast leave to work, we have to know that we are the
-                * last person to send a report for this group.  Reports
-                * can potentially get looped back if we are a multicast
-                * router, so discard reports sourced by me.
-                * Note that it is impossible to check IFF_LOOPBACK flag of
-                * ifp for this purpose, since ip6_mloopback pass the physical
-                * interface to looutput.
-                */
-               if (m->m_flags & M_LOOP) /* XXX: grotty flag, but efficient */
+               if (--limit == 0) {
                        break;
+               }
+       }
 
-               if (!IN6_IS_ADDR_MULTICAST(&mldh->mld6_addr))
-                       break;
+       if (mli != NULL) {
+               MLI_LOCK_ASSERT_HELD(mli);
+       }
+}
 
-               if (IN6_IS_ADDR_MC_LINKLOCAL(&mldh->mld6_addr))
-                       mldh->mld6_addr.s6_addr16[1] =
-                               htons(ifp->if_index); /* XXX */
-               /*
-                * If we belong to the group being reported, stop
-                * our timer for that group.
-                */
-               IN6_LOOKUP_MULTI(mldh->mld6_addr, ifp, in6m);
-               if (in6m) {
-                       in6m->in6m_timer = 0; /* transit to idle state */
-                       in6m->in6m_state = MLD6_OTHERLISTENER; /* clear flag */
+/*
+ * Filter outgoing MLD report state by group.
+ *
+ * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
+ * and node-local addresses. However, kernel and socket consumers
+ * always embed the KAME scope ID in the address provided, so strip it
+ * when performing comparison.
+ * Note: This is not the same as the *multicast* scope.
+ *
+ * Return zero if the given group is one for which MLD reports
+ * should be suppressed, or non-zero if reports should be issued.
+ */
+static __inline__ int
+mld_is_addr_reported(const struct in6_addr *addr)
+{
+       VERIFY(IN6_IS_ADDR_MULTICAST(addr));
+
+       if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL) {
+               return 0;
+       }
+
+       if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
+               struct in6_addr tmp = *addr;
+               in6_clearscope(&tmp);
+               if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes)) {
+                       return 0;
                }
+       }
 
-               if (IN6_IS_ADDR_MC_LINKLOCAL(&mldh->mld6_addr))
-                       mldh->mld6_addr.s6_addr16[1] = 0; /* XXX */
-               break;
-       default:                /* this is impossible */
-               log(LOG_ERR, "mld6_input: illegal type(%d)", mldh->mld6_type);
-               break;
+       return 1;
+}
+
+/*
+ * Attach MLD when PF_INET6 is attached to an interface.
+ */
+struct mld_ifinfo *
+mld_domifattach(struct ifnet *ifp, int how)
+{
+       struct mld_ifinfo *mli;
+
+       MLD_PRINTF(("%s: called for ifp 0x%llx(%s)\n", __func__,
+           (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+
+       mli = mli_alloc(how);
+       if (mli == NULL) {
+               return NULL;
        }
 
-       m_freem(m);
+       MLD_LOCK();
+
+       MLI_LOCK(mli);
+       mli_initvar(mli, ifp, 0);
+       mli->mli_debug |= IFD_ATTACHED;
+       MLI_ADDREF_LOCKED(mli); /* hold a reference for mli_head */
+       MLI_ADDREF_LOCKED(mli); /* hold a reference for caller */
+       MLI_UNLOCK(mli);
+       ifnet_lock_shared(ifp);
+       mld6_initsilent(ifp, mli);
+       ifnet_lock_done(ifp);
+
+       LIST_INSERT_HEAD(&mli_head, mli, mli_link);
+       mld_mli_list_genid++;
+
+       MLD_UNLOCK();
+
+       MLD_PRINTF(("%s: allocate mld_ifinfo for ifp 0x%llx(%s)\n",
+           __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+
+       return mli;
 }
 
+/*
+ * Attach MLD when PF_INET6 is reattached to an interface.  Caller is
+ * expected to have an outstanding reference to the mli.
+ */
 void
-mld6_fasttimeo()
+mld_domifreattach(struct mld_ifinfo *mli)
 {
-       register struct in6_multi *in6m;
-       struct in6_multistep step;
-       int s;
+       struct ifnet *ifp;
 
-       /*
-        * Quick check to see if any work needs to be done, in order
-        * to minimize the overhead of fasttimo processing.
-        */
-       if (!mld6_timers_are_running)
-               return;
+       MLD_LOCK();
 
-#ifdef __NetBSD__
-       s = splsoftnet();
-#else
-       s = splnet();
-#endif
-       mld6_timers_are_running = 0;
-       IN6_FIRST_MULTI(step, in6m);
-       while (in6m != NULL) {
-               if (in6m->in6m_timer == 0) {
-                       /* do nothing */
-               } else if (--in6m->in6m_timer == 0) {
-                       mld6_sendpkt(in6m, MLD6_LISTENER_REPORT, NULL);
-                       in6m->in6m_state = MLD6_IREPORTEDLAST;
-               } else {
-                       mld6_timers_are_running = 1;
+       MLI_LOCK(mli);
+       VERIFY(!(mli->mli_debug & IFD_ATTACHED));
+       ifp = mli->mli_ifp;
+       VERIFY(ifp != NULL);
+       mli_initvar(mli, ifp, 1);
+       mli->mli_debug |= IFD_ATTACHED;
+       MLI_ADDREF_LOCKED(mli); /* hold a reference for mli_head */
+       MLI_UNLOCK(mli);
+       ifnet_lock_shared(ifp);
+       mld6_initsilent(ifp, mli);
+       ifnet_lock_done(ifp);
+
+       LIST_INSERT_HEAD(&mli_head, mli, mli_link);
+       mld_mli_list_genid++;
+
+       MLD_UNLOCK();
+
+       MLD_PRINTF(("%s: reattached mld_ifinfo for ifp 0x%llx(%s)\n",
+           __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+}
+
+/*
+ * Hook for domifdetach.
+ */
+void
+mld_domifdetach(struct ifnet *ifp)
+{
+       SLIST_HEAD(, in6_multi) in6m_dthead;
+
+       SLIST_INIT(&in6m_dthead);
+
+       MLD_PRINTF(("%s: called for ifp 0x%llx(%s)\n", __func__,
+           (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+
+       MLD_LOCK();
+       mli_delete(ifp, (struct mld_in6m_relhead *)&in6m_dthead);
+       MLD_UNLOCK();
+
+       /* Now that we're dropped all locks, release detached records */
+       MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
+}
+
+/*
+ * Called at interface detach time.  Note that we only flush all deferred
+ * responses and record releases; all remaining inm records and their source
+ * entries related to this interface are left intact, in order to handle
+ * the reattach case.
+ */
+static void
+mli_delete(const struct ifnet *ifp, struct mld_in6m_relhead *in6m_dthead)
+{
+       struct mld_ifinfo *mli, *tmli;
+
+       MLD_LOCK_ASSERT_HELD();
+
+       LIST_FOREACH_SAFE(mli, &mli_head, mli_link, tmli) {
+               MLI_LOCK(mli);
+               if (mli->mli_ifp == ifp) {
+                       /*
+                        * Free deferred General Query responses.
+                        */
+                       IF_DRAIN(&mli->mli_gq);
+                       IF_DRAIN(&mli->mli_v1q);
+                       mld_flush_relq(mli, in6m_dthead);
+                       VERIFY(SLIST_EMPTY(&mli->mli_relinmhead));
+                       mli->mli_debug &= ~IFD_ATTACHED;
+                       MLI_UNLOCK(mli);
+
+                       LIST_REMOVE(mli, mli_link);
+                       MLI_REMREF(mli); /* release mli_head reference */
+                       mld_mli_list_genid++;
+                       return;
                }
-               IN6_NEXT_MULTI(step, in6m);
+               MLI_UNLOCK(mli);
+       }
+       panic("%s: mld_ifinfo not found for ifp %p(%s)\n", __func__,
+           ifp, ifp->if_xname);
+}
+
+__private_extern__ void
+mld6_initsilent(struct ifnet *ifp, struct mld_ifinfo *mli)
+{
+       ifnet_lock_assert(ifp, IFNET_LCK_ASSERT_OWNED);
+
+       MLI_LOCK_ASSERT_NOTHELD(mli);
+       MLI_LOCK(mli);
+       if (!(ifp->if_flags & IFF_MULTICAST) &&
+           (ifp->if_eflags & (IFEF_IPV6_ND6ALT | IFEF_LOCALNET_PRIVATE))) {
+               mli->mli_flags |= MLIF_SILENT;
+       } else {
+               mli->mli_flags &= ~MLIF_SILENT;
        }
-       splx(s);
+       MLI_UNLOCK(mli);
 }
 
 static void
-mld6_sendpkt(in6m, type, dst)
-       struct in6_multi *in6m;
-       int type;
-       const struct in6_addr *dst;
+mli_initvar(struct mld_ifinfo *mli, struct ifnet *ifp, int reattach)
 {
-       struct mbuf *mh, *md;
-       struct mld6_hdr *mldh;
-       struct ip6_hdr *ip6;
-       struct ip6_moptions im6o;
-       struct in6_ifaddr *ia;
-       struct ifnet *ifp = in6m->in6m_ifp;
-       struct ifnet *outif = NULL;
+       MLI_LOCK_ASSERT_HELD(mli);
 
-       /*
-        * At first, find a link local address on the outgoing interface
-        * to use as the source address of the MLD packet.
-        */
-       if ((ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST))
-           == NULL)
-               return;
+       mli->mli_ifp = ifp;
+       if (mld_v2enable) {
+               mli->mli_version = MLD_VERSION_2;
+       } else {
+               mli->mli_version = MLD_VERSION_1;
+       }
+       mli->mli_flags = 0;
+       mli->mli_rv = MLD_RV_INIT;
+       mli->mli_qi = MLD_QI_INIT;
+       mli->mli_qri = MLD_QRI_INIT;
+       mli->mli_uri = MLD_URI_INIT;
+
+       if (mld_use_allow) {
+               mli->mli_flags |= MLIF_USEALLOW;
+       }
+       if (!reattach) {
+               SLIST_INIT(&mli->mli_relinmhead);
+       }
 
        /*
-        * Allocate mbufs to store ip6 header and MLD header.
-        * We allocate 2 mbufs and make chain in advance because
-        * it is more convenient when inserting the hop-by-hop option later.
+        * Responses to general queries are subject to bounds.
         */
-       MGETHDR(mh, M_DONTWAIT, MT_HEADER);
-       if (mh == NULL)
-               return;
-       MGET(md, M_DONTWAIT, MT_DATA);
-       if (md == NULL) {
-               m_free(mh);
+       mli->mli_gq.ifq_maxlen = MLD_MAX_RESPONSE_PACKETS;
+       mli->mli_v1q.ifq_maxlen = MLD_MAX_RESPONSE_PACKETS;
+}
+
+static struct mld_ifinfo *
+mli_alloc(int how)
+{
+       struct mld_ifinfo *mli;
+
+       mli = (how == M_WAITOK) ? zalloc(mli_zone) : zalloc_noblock(mli_zone);
+       if (mli != NULL) {
+               bzero(mli, mli_size);
+               lck_mtx_init(&mli->mli_lock, mld_mtx_grp, mld_mtx_attr);
+               mli->mli_debug |= IFD_ALLOC;
+       }
+       return mli;
+}
+
+static void
+mli_free(struct mld_ifinfo *mli)
+{
+       MLI_LOCK(mli);
+       if (mli->mli_debug & IFD_ATTACHED) {
+               panic("%s: attached mli=%p is being freed", __func__, mli);
+               /* NOTREACHED */
+       } else if (mli->mli_ifp != NULL) {
+               panic("%s: ifp not NULL for mli=%p", __func__, mli);
+               /* NOTREACHED */
+       } else if (!(mli->mli_debug & IFD_ALLOC)) {
+               panic("%s: mli %p cannot be freed", __func__, mli);
+               /* NOTREACHED */
+       } else if (mli->mli_refcnt != 0) {
+               panic("%s: non-zero refcnt mli=%p", __func__, mli);
+               /* NOTREACHED */
+       }
+       mli->mli_debug &= ~IFD_ALLOC;
+       MLI_UNLOCK(mli);
+
+       lck_mtx_destroy(&mli->mli_lock, mld_mtx_grp);
+       zfree(mli_zone, mli);
+}
+
+void
+mli_addref(struct mld_ifinfo *mli, int locked)
+{
+       if (!locked) {
+               MLI_LOCK_SPIN(mli);
+       } else {
+               MLI_LOCK_ASSERT_HELD(mli);
+       }
+
+       if (++mli->mli_refcnt == 0) {
+               panic("%s: mli=%p wraparound refcnt", __func__, mli);
+               /* NOTREACHED */
+       }
+       if (!locked) {
+               MLI_UNLOCK(mli);
+       }
+}
+
+void
+mli_remref(struct mld_ifinfo *mli)
+{
+       SLIST_HEAD(, in6_multi) in6m_dthead;
+       struct ifnet *ifp;
+
+       MLI_LOCK_SPIN(mli);
+
+       if (mli->mli_refcnt == 0) {
+               panic("%s: mli=%p negative refcnt", __func__, mli);
+               /* NOTREACHED */
+       }
+
+       --mli->mli_refcnt;
+       if (mli->mli_refcnt > 0) {
+               MLI_UNLOCK(mli);
                return;
        }
-       mh->m_next = md;
 
-       mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld6_hdr);
-       mh->m_len = sizeof(struct ip6_hdr);
-       MH_ALIGN(mh, sizeof(struct ip6_hdr));
+       ifp = mli->mli_ifp;
+       mli->mli_ifp = NULL;
+       IF_DRAIN(&mli->mli_gq);
+       IF_DRAIN(&mli->mli_v1q);
+       SLIST_INIT(&in6m_dthead);
+       mld_flush_relq(mli, (struct mld_in6m_relhead *)&in6m_dthead);
+       VERIFY(SLIST_EMPTY(&mli->mli_relinmhead));
+       MLI_UNLOCK(mli);
 
-       /* fill in the ip6 header */
-       ip6 = mtod(mh, struct ip6_hdr *);
-       ip6->ip6_flow = 0;
-       ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
-       ip6->ip6_vfc |= IPV6_VERSION;
-       /* ip6_plen will be set later */
-       ip6->ip6_nxt = IPPROTO_ICMPV6;
-       /* ip6_hlim will be set by im6o.im6o_multicast_hlim */
-       ip6->ip6_src = ia->ia_addr.sin6_addr;
-       ip6->ip6_dst = dst ? *dst : in6m->in6m_addr;
-
-       /* fill in the MLD header */
-       md->m_len = sizeof(struct mld6_hdr);
-       mldh = mtod(md, struct mld6_hdr *);
-       mldh->mld6_type = type;
-       mldh->mld6_code = 0;
-       mldh->mld6_cksum = 0;
-       /* XXX: we assume the function will not be called for query messages */
-       mldh->mld6_maxdelay = 0;
-       mldh->mld6_reserved = 0;
-       mldh->mld6_addr = in6m->in6m_addr;
-       if (IN6_IS_ADDR_MC_LINKLOCAL(&mldh->mld6_addr))
-               mldh->mld6_addr.s6_addr16[1] = 0; /* XXX */
-       mldh->mld6_cksum = in6_cksum(mh, IPPROTO_ICMPV6, sizeof(struct ip6_hdr),
-                                    sizeof(struct mld6_hdr));
-
-       /* construct multicast option */
-       bzero(&im6o, sizeof(im6o));
-       im6o.im6o_multicast_ifp = ifp;
-       im6o.im6o_multicast_hlim = 1;
-
-       /*
-        * Request loopback of the report if we are acting as a multicast
-        * router, so that the process-level routing daemon can hear it.
-        */
-       im6o.im6o_multicast_loop = (ip6_mrouter != NULL);
-
-       /* increment output statictics */
-       icmp6stat.icp6s_outhist[type]++;
+       /* Now that we're dropped all locks, release detached records */
+       MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
 
-       ip6_output(mh, &ip6_opts, NULL, 0, &im6o, &outif);
-       if (outif) {
-               icmp6_ifstat_inc(outif, ifs6_out_msg);
-               switch(type) {
-                case MLD6_LISTENER_QUERY:
-                        icmp6_ifstat_inc(outif, ifs6_out_mldquery);
-                        break;
-                case MLD6_LISTENER_REPORT:
-                        icmp6_ifstat_inc(outif, ifs6_out_mldreport);
-                        break;
-                case MLD6_LISTENER_DONE:
-                        icmp6_ifstat_inc(outif, ifs6_out_mlddone);
-                        break;
-               }
+       MLD_PRINTF(("%s: freeing mld_ifinfo for ifp 0x%llx(%s)\n",
+           __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+
+       mli_free(mli);
+}
+
+/*
+ * Process a received MLDv1 general or address-specific query.
+ * Assumes that the query header has been pulled up to sizeof(mld_hdr).
+ *
+ * NOTE: Can't be fully const correct as we temporarily embed scope ID in
+ * mld_addr. This is OK as we own the mbuf chain.
+ */
+static int
+mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
+    /*const*/ struct mld_hdr *mld)
+{
+       struct mld_ifinfo       *mli;
+       struct in6_multi        *inm;
+       int                      err = 0, is_general_query;
+       uint16_t                 timer;
+       struct mld_tparams       mtp = { 0, 0, 0, 0 };
+
+       MLD_LOCK_ASSERT_NOTHELD();
+
+       is_general_query = 0;
+
+       if (!mld_v1enable) {
+               MLD_PRINTF(("%s: ignore v1 query %s on ifp 0x%llx(%s)\n",
+                   __func__, ip6_sprintf(&mld->mld_addr),
+                   (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               goto done;
+       }
+
+       /*
+        * RFC3810 Section 6.2: MLD queries must originate from
+        * a router's link-local address.
+        */
+       if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
+               MLD_PRINTF(("%s: ignore v1 query src %s on ifp 0x%llx(%s)\n",
+                   __func__, ip6_sprintf(&ip6->ip6_src),
+                   (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               goto done;
+       }
+
+       /*
+        * Do address field validation upfront before we accept
+        * the query.
+        */
+       if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
+               /*
+                * MLDv1 General Query.
+                * If this was not sent to the all-nodes group, ignore it.
+                */
+               struct in6_addr          dst;
+
+               dst = ip6->ip6_dst;
+               in6_clearscope(&dst);
+               if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) {
+                       err = EINVAL;
+                       goto done;
+               }
+               is_general_query = 1;
+       } else {
+               /*
+                * Embed scope ID of receiving interface in MLD query for
+                * lookup whilst we don't hold other locks.
+                */
+               in6_setscope(&mld->mld_addr, ifp, NULL);
+       }
+
+       /*
+        * Switch to MLDv1 host compatibility mode.
+        */
+       mli = MLD_IFINFO(ifp);
+       VERIFY(mli != NULL);
+
+       MLI_LOCK(mli);
+       mtp.qpt = mld_set_version(mli, MLD_VERSION_1);
+       MLI_UNLOCK(mli);
+
+       timer = ntohs(mld->mld_maxdelay) / MLD_TIMER_SCALE;
+       if (timer == 0) {
+               timer = 1;
+       }
+
+       if (is_general_query) {
+               struct in6_multistep step;
+
+               MLD_PRINTF(("%s: process v1 general query on ifp 0x%llx(%s)\n",
+                   __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               /*
+                * For each reporting group joined on this
+                * interface, kick the report timer.
+                */
+               in6_multihead_lock_shared();
+               IN6_FIRST_MULTI(step, inm);
+               while (inm != NULL) {
+                       IN6M_LOCK(inm);
+                       if (inm->in6m_ifp == ifp) {
+                               mtp.cst += mld_v1_update_group(inm, timer);
+                       }
+                       IN6M_UNLOCK(inm);
+                       IN6_NEXT_MULTI(step, inm);
+               }
+               in6_multihead_lock_done();
+       } else {
+               /*
+                * MLDv1 Group-Specific Query.
+                * If this is a group-specific MLDv1 query, we need only
+                * look up the single group to process it.
+                */
+               in6_multihead_lock_shared();
+               IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
+               in6_multihead_lock_done();
+
+               if (inm != NULL) {
+                       IN6M_LOCK(inm);
+                       MLD_PRINTF(("%s: process v1 query %s on "
+                           "ifp 0x%llx(%s)\n", __func__,
+                           ip6_sprintf(&mld->mld_addr),
+                           (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+                       mtp.cst = mld_v1_update_group(inm, timer);
+                       IN6M_UNLOCK(inm);
+                       IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
+               }
+               /* XXX Clear embedded scope ID as userland won't expect it. */
+               in6_clearscope(&mld->mld_addr);
+       }
+done:
+       mld_set_timeout(&mtp);
+
+       return err;
+}
+
+/*
+ * Update the report timer on a group in response to an MLDv1 query.
+ *
+ * If we are becoming the reporting member for this group, start the timer.
+ * If we already are the reporting member for this group, and timer is
+ * below the threshold, reset it.
+ *
+ * We may be updating the group for the first time since we switched
+ * to MLDv2. If we are, then we must clear any recorded source lists,
+ * and transition to REPORTING state; the group timer is overloaded
+ * for group and group-source query responses.
+ *
+ * Unlike MLDv2, the delay per group should be jittered
+ * to avoid bursts of MLDv1 reports.
+ */
+static uint32_t
+mld_v1_update_group(struct in6_multi *inm, const int timer)
+{
+       IN6M_LOCK_ASSERT_HELD(inm);
+
+       MLD_PRINTF(("%s: %s/%s timer=%d\n", __func__,
+           ip6_sprintf(&inm->in6m_addr),
+           if_name(inm->in6m_ifp), timer));
+
+       switch (inm->in6m_state) {
+       case MLD_NOT_MEMBER:
+       case MLD_SILENT_MEMBER:
+               break;
+       case MLD_REPORTING_MEMBER:
+               if (inm->in6m_timer != 0 &&
+                   inm->in6m_timer <= timer) {
+                       MLD_PRINTF(("%s: REPORTING and timer running, "
+                           "skipping.\n", __func__));
+                       break;
+               }
+       /* FALLTHROUGH */
+       case MLD_SG_QUERY_PENDING_MEMBER:
+       case MLD_G_QUERY_PENDING_MEMBER:
+       case MLD_IDLE_MEMBER:
+       case MLD_LAZY_MEMBER:
+       case MLD_AWAKENING_MEMBER:
+               MLD_PRINTF(("%s: ->REPORTING\n", __func__));
+               inm->in6m_state = MLD_REPORTING_MEMBER;
+               inm->in6m_timer = MLD_RANDOM_DELAY(timer);
+               break;
+       case MLD_SLEEPING_MEMBER:
+               MLD_PRINTF(("%s: ->AWAKENING\n", __func__));
+               inm->in6m_state = MLD_AWAKENING_MEMBER;
+               break;
+       case MLD_LEAVING_MEMBER:
+               break;
+       }
+
+       return inm->in6m_timer;
+}
+
+/*
+ * Process a received MLDv2 general, group-specific or
+ * group-and-source-specific query.
+ *
+ * Assumes that the query header has been pulled up to sizeof(mldv2_query).
+ *
+ * Return 0 if successful, otherwise an appropriate error code is returned.
+ */
+static int
+mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
+    struct mbuf *m, const int off, const int icmp6len)
+{
+       struct mld_ifinfo       *mli;
+       struct mldv2_query      *mld;
+       struct in6_multi        *inm;
+       uint32_t                 maxdelay, nsrc, qqi;
+       int                      err = 0, is_general_query;
+       uint16_t                 timer;
+       uint8_t                  qrv;
+       struct mld_tparams       mtp = { 0, 0, 0, 0 };
+
+       MLD_LOCK_ASSERT_NOTHELD();
+
+       is_general_query = 0;
+
+       if (!mld_v2enable) {
+               MLD_PRINTF(("%s: ignore v2 query %s on ifp 0x%llx(%s)\n",
+                   __func__, ip6_sprintf(&ip6->ip6_src),
+                   (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               goto done;
+       }
+
+       /*
+        * RFC3810 Section 6.2: MLD queries must originate from
+        * a router's link-local address.
+        */
+       if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
+               MLD_PRINTF(("%s: ignore v1 query src %s on ifp 0x%llx(%s)\n",
+                   __func__, ip6_sprintf(&ip6->ip6_src),
+                   (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               goto done;
+       }
+
+       MLD_PRINTF(("%s: input v2 query on ifp 0x%llx(%s)\n", __func__,
+           (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+
+       mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
+
+       maxdelay = ntohs(mld->mld_maxdelay);    /* in 1/10ths of a second */
+       if (maxdelay >= 32768) {
+               maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
+                   (MLD_MRC_EXP(maxdelay) + 3);
+       }
+       timer = maxdelay / MLD_TIMER_SCALE;
+       if (timer == 0) {
+               timer = 1;
+       }
+
+       qrv = MLD_QRV(mld->mld_misc);
+       if (qrv < 2) {
+               MLD_PRINTF(("%s: clamping qrv %d to %d\n", __func__,
+                   qrv, MLD_RV_INIT));
+               qrv = MLD_RV_INIT;
+       }
+
+       qqi = mld->mld_qqi;
+       if (qqi >= 128) {
+               qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
+                   (MLD_QQIC_EXP(mld->mld_qqi) + 3);
+       }
+
+       nsrc = ntohs(mld->mld_numsrc);
+       if (nsrc > MLD_MAX_GS_SOURCES) {
+               err = EMSGSIZE;
+               goto done;
+       }
+       if (icmp6len < sizeof(struct mldv2_query) +
+           (nsrc * sizeof(struct in6_addr))) {
+               err = EMSGSIZE;
+               goto done;
+       }
+
+       /*
+        * Do further input validation upfront to avoid resetting timers
+        * should we need to discard this query.
+        */
+       if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
+               /*
+                * A general query with a source list has undefined
+                * behaviour; discard it.
+                */
+               if (nsrc > 0) {
+                       err = EINVAL;
+                       goto done;
+               }
+               is_general_query = 1;
+       } else {
+               /*
+                * Embed scope ID of receiving interface in MLD query for
+                * lookup whilst we don't hold other locks (due to KAME
+                * locking lameness). We own this mbuf chain just now.
+                */
+               in6_setscope(&mld->mld_addr, ifp, NULL);
+       }
+
+       mli = MLD_IFINFO(ifp);
+       VERIFY(mli != NULL);
+
+       MLI_LOCK(mli);
+       /*
+        * Discard the v2 query if we're in Compatibility Mode.
+        * The RFC is pretty clear that hosts need to stay in MLDv1 mode
+        * until the Old Version Querier Present timer expires.
+        */
+       if (mli->mli_version != MLD_VERSION_2) {
+               MLI_UNLOCK(mli);
+               goto done;
+       }
+
+       mtp.qpt = mld_set_version(mli, MLD_VERSION_2);
+       mli->mli_rv = qrv;
+       mli->mli_qi = qqi;
+       mli->mli_qri = MAX(timer, MLD_QRI_MIN);
+
+       MLD_PRINTF(("%s: qrv %d qi %d qri %d\n", __func__, mli->mli_rv,
+           mli->mli_qi, mli->mli_qri));
+
+       if (is_general_query) {
+               /*
+                * MLDv2 General Query.
+                *
+                * Schedule a current-state report on this ifp for
+                * all groups, possibly containing source lists.
+                *
+                * If there is a pending General Query response
+                * scheduled earlier than the selected delay, do
+                * not schedule any other reports.
+                * Otherwise, reset the interface timer.
+                */
+               MLD_PRINTF(("%s: process v2 general query on ifp 0x%llx(%s)\n",
+                   __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
+                       mtp.it = mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
+               }
+               MLI_UNLOCK(mli);
+       } else {
+               MLI_UNLOCK(mli);
+               /*
+                * MLDv2 Group-specific or Group-and-source-specific Query.
+                *
+                * Group-source-specific queries are throttled on
+                * a per-group basis to defeat denial-of-service attempts.
+                * Queries for groups we are not a member of on this
+                * link are simply ignored.
+                */
+               in6_multihead_lock_shared();
+               IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
+               in6_multihead_lock_done();
+               if (inm == NULL) {
+                       goto done;
+               }
+
+               IN6M_LOCK(inm);
+               if (nsrc > 0) {
+                       if (!ratecheck(&inm->in6m_lastgsrtv,
+                           &mld_gsrdelay)) {
+                               MLD_PRINTF(("%s: GS query throttled.\n",
+                                   __func__));
+                               IN6M_UNLOCK(inm);
+                               IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
+                               goto done;
+                       }
+               }
+               MLD_PRINTF(("%s: process v2 group query on ifp 0x%llx(%s)\n",
+                   __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               /*
+                * If there is a pending General Query response
+                * scheduled sooner than the selected delay, no
+                * further report need be scheduled.
+                * Otherwise, prepare to respond to the
+                * group-specific or group-and-source query.
+                */
+               MLI_LOCK(mli);
+               mtp.it = mli->mli_v2_timer;
+               MLI_UNLOCK(mli);
+               if (mtp.it == 0 || mtp.it >= timer) {
+                       (void) mld_v2_process_group_query(inm, timer, m, off);
+                       mtp.cst = inm->in6m_timer;
+               }
+               IN6M_UNLOCK(inm);
+               IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
+               /* XXX Clear embedded scope ID as userland won't expect it. */
+               in6_clearscope(&mld->mld_addr);
+       }
+done:
+       if (mtp.it > 0) {
+               MLD_PRINTF(("%s: v2 general query response scheduled in "
+                   "T+%d seconds on ifp 0x%llx(%s)\n", __func__, mtp.it,
+                   (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+       }
+       mld_set_timeout(&mtp);
+
+       return err;
+}
+
+/*
+ * Process a recieved MLDv2 group-specific or group-and-source-specific
+ * query.
+ * Return <0 if any error occured. Currently this is ignored.
+ */
+static int
+mld_v2_process_group_query(struct in6_multi *inm, int timer, struct mbuf *m0,
+    const int off)
+{
+       struct mldv2_query      *mld;
+       int                      retval;
+       uint16_t                 nsrc;
+
+       IN6M_LOCK_ASSERT_HELD(inm);
+
+       retval = 0;
+       mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
+
+       switch (inm->in6m_state) {
+       case MLD_NOT_MEMBER:
+       case MLD_SILENT_MEMBER:
+       case MLD_SLEEPING_MEMBER:
+       case MLD_LAZY_MEMBER:
+       case MLD_AWAKENING_MEMBER:
+       case MLD_IDLE_MEMBER:
+       case MLD_LEAVING_MEMBER:
+               return retval;
+       case MLD_REPORTING_MEMBER:
+       case MLD_G_QUERY_PENDING_MEMBER:
+       case MLD_SG_QUERY_PENDING_MEMBER:
+               break;
+       }
+
+       nsrc = ntohs(mld->mld_numsrc);
+
+       /*
+        * Deal with group-specific queries upfront.
+        * If any group query is already pending, purge any recorded
+        * source-list state if it exists, and schedule a query response
+        * for this group-specific query.
+        */
+       if (nsrc == 0) {
+               if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
+                   inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
+                       in6m_clear_recorded(inm);
+                       timer = min(inm->in6m_timer, timer);
+               }
+               inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
+               inm->in6m_timer = MLD_RANDOM_DELAY(timer);
+               return retval;
+       }
+
+       /*
+        * Deal with the case where a group-and-source-specific query has
+        * been received but a group-specific query is already pending.
+        */
+       if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
+               timer = min(inm->in6m_timer, timer);
+               inm->in6m_timer = MLD_RANDOM_DELAY(timer);
+               return retval;
+       }
+
+       /*
+        * Finally, deal with the case where a group-and-source-specific
+        * query has been received, where a response to a previous g-s-r
+        * query exists, or none exists.
+        * In this case, we need to parse the source-list which the Querier
+        * has provided us with and check if we have any source list filter
+        * entries at T1 for these sources. If we do not, there is no need
+        * schedule a report and the query may be dropped.
+        * If we do, we must record them and schedule a current-state
+        * report for those sources.
+        */
+       if (inm->in6m_nsrc > 0) {
+               struct mbuf             *m;
+               uint8_t                 *sp;
+               int                      i, nrecorded;
+               int                      soff;
+
+               m = m0;
+               soff = off + sizeof(struct mldv2_query);
+               nrecorded = 0;
+               for (i = 0; i < nsrc; i++) {
+                       sp = mtod(m, uint8_t *) + soff;
+                       retval = in6m_record_source(inm,
+                           (const struct in6_addr *)(void *)sp);
+                       if (retval < 0) {
+                               break;
+                       }
+                       nrecorded += retval;
+                       soff += sizeof(struct in6_addr);
+                       if (soff >= m->m_len) {
+                               soff = soff - m->m_len;
+                               m = m->m_next;
+                               if (m == NULL) {
+                                       break;
+                               }
+                       }
+               }
+               if (nrecorded > 0) {
+                       MLD_PRINTF(("%s: schedule response to SG query\n",
+                           __func__));
+                       inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
+                       inm->in6m_timer = MLD_RANDOM_DELAY(timer);
+               }
+       }
+
+       return retval;
+}
+
+/*
+ * Process a received MLDv1 host membership report.
+ * Assumes mld points to mld_hdr in pulled up mbuf chain.
+ *
+ * NOTE: Can't be fully const correct as we temporarily embed scope ID in
+ * mld_addr. This is OK as we own the mbuf chain.
+ */
+static int
+mld_v1_input_report(struct ifnet *ifp, struct mbuf *m,
+    const struct ip6_hdr *ip6, /*const*/ struct mld_hdr *mld)
+{
+       struct in6_addr          src, dst;
+       struct in6_ifaddr       *ia;
+       struct in6_multi        *inm;
+
+       if (!mld_v1enable) {
+               MLD_PRINTF(("%s: ignore v1 report %s on ifp 0x%llx(%s)\n",
+                   __func__, ip6_sprintf(&mld->mld_addr),
+                   (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               return 0;
+       }
+
+       if ((ifp->if_flags & IFF_LOOPBACK) ||
+           (m->m_pkthdr.pkt_flags & PKTF_LOOP)) {
+               return 0;
+       }
+
+       /*
+        * MLDv1 reports must originate from a host's link-local address,
+        * or the unspecified address (when booting).
+        */
+       src = ip6->ip6_src;
+       in6_clearscope(&src);
+       if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
+               MLD_PRINTF(("%s: ignore v1 query src %s on ifp 0x%llx(%s)\n",
+                   __func__, ip6_sprintf(&ip6->ip6_src),
+                   (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               return EINVAL;
+       }
+
+       /*
+        * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
+        * group, and must be directed to the group itself.
+        */
+       dst = ip6->ip6_dst;
+       in6_clearscope(&dst);
+       if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
+           !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
+               MLD_PRINTF(("%s: ignore v1 query dst %s on ifp 0x%llx(%s)\n",
+                   __func__, ip6_sprintf(&ip6->ip6_dst),
+                   (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               return EINVAL;
+       }
+
+       /*
+        * Make sure we don't hear our own membership report, as fast
+        * leave requires knowing that we are the only member of a
+        * group. Assume we used the link-local address if available,
+        * otherwise look for ::.
+        *
+        * XXX Note that scope ID comparison is needed for the address
+        * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
+        * performed for the on-wire address.
+        */
+       ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
+       if (ia != NULL) {
+               IFA_LOCK(&ia->ia_ifa);
+               if ((IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia)))) {
+                       IFA_UNLOCK(&ia->ia_ifa);
+                       IFA_REMREF(&ia->ia_ifa);
+                       return 0;
+               }
+               IFA_UNLOCK(&ia->ia_ifa);
+               IFA_REMREF(&ia->ia_ifa);
+       } else if (IN6_IS_ADDR_UNSPECIFIED(&src)) {
+               return 0;
+       }
+
+       MLD_PRINTF(("%s: process v1 report %s on ifp 0x%llx(%s)\n",
+           __func__, ip6_sprintf(&mld->mld_addr),
+           (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+
+       /*
+        * Embed scope ID of receiving interface in MLD query for lookup
+        * whilst we don't hold other locks (due to KAME locking lameness).
+        */
+       if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
+               in6_setscope(&mld->mld_addr, ifp, NULL);
+       }
+
+       /*
+        * MLDv1 report suppression.
+        * If we are a member of this group, and our membership should be
+        * reported, and our group timer is pending or about to be reset,
+        * stop our group timer by transitioning to the 'lazy' state.
+        */
+       in6_multihead_lock_shared();
+       IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
+       in6_multihead_lock_done();
+
+       if (inm != NULL) {
+               struct mld_ifinfo *mli;
+
+               IN6M_LOCK(inm);
+               mli = inm->in6m_mli;
+               VERIFY(mli != NULL);
+
+               MLI_LOCK(mli);
+               /*
+                * If we are in MLDv2 host mode, do not allow the
+                * other host's MLDv1 report to suppress our reports.
+                */
+               if (mli->mli_version == MLD_VERSION_2) {
+                       MLI_UNLOCK(mli);
+                       IN6M_UNLOCK(inm);
+                       IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
+                       goto out;
+               }
+               MLI_UNLOCK(mli);
+
+               inm->in6m_timer = 0;
+
+               switch (inm->in6m_state) {
+               case MLD_NOT_MEMBER:
+               case MLD_SILENT_MEMBER:
+               case MLD_SLEEPING_MEMBER:
+                       break;
+               case MLD_REPORTING_MEMBER:
+               case MLD_IDLE_MEMBER:
+               case MLD_AWAKENING_MEMBER:
+                       MLD_PRINTF(("%s: report suppressed for %s on "
+                           "ifp 0x%llx(%s)\n", __func__,
+                           ip6_sprintf(&mld->mld_addr),
+                           (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
+               case MLD_LAZY_MEMBER:
+                       inm->in6m_state = MLD_LAZY_MEMBER;
+                       break;
+               case MLD_G_QUERY_PENDING_MEMBER:
+               case MLD_SG_QUERY_PENDING_MEMBER:
+               case MLD_LEAVING_MEMBER:
+                       break;
+               }
+               IN6M_UNLOCK(inm);
+               IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
+       }
+
+out:
+       /* XXX Clear embedded scope ID as userland won't expect it. */
+       in6_clearscope(&mld->mld_addr);
+
+       return 0;
+}
+
+/*
+ * MLD input path.
+ *
+ * Assume query messages which fit in a single ICMPv6 message header
+ * have been pulled up.
+ * Assume that userland will want to see the message, even if it
+ * otherwise fails kernel input validation; do not free it.
+ * Pullup may however free the mbuf chain m if it fails.
+ *
+ * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
+ */
+int
+mld_input(struct mbuf *m, int off, int icmp6len)
+{
+       struct ifnet    *ifp;
+       struct ip6_hdr  *ip6;
+       struct mld_hdr  *mld;
+       int              mldlen;
+
+       MLD_PRINTF(("%s: called w/mbuf (0x%llx,%d)\n", __func__,
+           (uint64_t)VM_KERNEL_ADDRPERM(m), off));
+
+       ifp = m->m_pkthdr.rcvif;
+
+       ip6 = mtod(m, struct ip6_hdr *);
+
+       /* Pullup to appropriate size. */
+       mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
+       if (mld->mld_type == MLD_LISTENER_QUERY &&
+           icmp6len >= sizeof(struct mldv2_query)) {
+               mldlen = sizeof(struct mldv2_query);
+       } else {
+               mldlen = sizeof(struct mld_hdr);
+       }
+       // check if mldv2_query/mld_hdr fits in the first mbuf
+       IP6_EXTHDR_CHECK(m, off, mldlen, return IPPROTO_DONE);
+       IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
+       if (mld == NULL) {
+               icmp6stat.icp6s_badlen++;
+               return IPPROTO_DONE;
+       }
+
+       /*
+        * Userland needs to see all of this traffic for implementing
+        * the endpoint discovery portion of multicast routing.
+        */
+       switch (mld->mld_type) {
+       case MLD_LISTENER_QUERY:
+               icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
+               if (icmp6len == sizeof(struct mld_hdr)) {
+                       if (mld_v1_input_query(ifp, ip6, mld) != 0) {
+                               return 0;
+                       }
+               } else if (icmp6len >= sizeof(struct mldv2_query)) {
+                       if (mld_v2_input_query(ifp, ip6, m, off,
+                           icmp6len) != 0) {
+                               return 0;
+                       }
+               }
+               break;
+       case MLD_LISTENER_REPORT:
+               icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
+               if (mld_v1_input_report(ifp, m, ip6, mld) != 0) {
+                       return 0;
+               }
+               break;
+       case MLDV2_LISTENER_REPORT:
+               icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
+               break;
+       case MLD_LISTENER_DONE:
+               icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
+               break;
+       default:
+               break;
+       }
+
+       return 0;
+}
+
+/*
+ * Schedule MLD timer based on various parameters; caller must ensure that
+ * lock ordering is maintained as this routine acquires MLD global lock.
+ */
+void
+mld_set_timeout(struct mld_tparams *mtp)
+{
+       MLD_LOCK_ASSERT_NOTHELD();
+       VERIFY(mtp != NULL);
+
+       if (mtp->qpt != 0 || mtp->it != 0 || mtp->cst != 0 || mtp->sct != 0) {
+               MLD_LOCK();
+               if (mtp->qpt != 0) {
+                       querier_present_timers_running6 = 1;
+               }
+               if (mtp->it != 0) {
+                       interface_timers_running6 = 1;
+               }
+               if (mtp->cst != 0) {
+                       current_state_timers_running6 = 1;
+               }
+               if (mtp->sct != 0) {
+                       state_change_timers_running6 = 1;
+               }
+               mld_sched_timeout();
+               MLD_UNLOCK();
+       }
+}
+
+/*
+ * MLD6 timer handler (per 1 second).
+ */
+static void
+mld_timeout(void *arg)
+{
+#pragma unused(arg)
+       struct ifqueue           scq;   /* State-change packets */
+       struct ifqueue           qrq;   /* Query response packets */
+       struct ifnet            *ifp;
+       struct mld_ifinfo       *mli;
+       struct in6_multi        *inm;
+       int                      uri_sec = 0;
+       unsigned int genid = mld_mli_list_genid;
+
+       SLIST_HEAD(, in6_multi) in6m_dthead;
+
+       SLIST_INIT(&in6m_dthead);
+
+       /*
+        * Update coarse-grained networking timestamp (in sec.); the idea
+        * is to piggy-back on the timeout callout to update the counter
+        * returnable via net_uptime().
+        */
+       net_update_uptime();
+
+       MLD_LOCK();
+
+       MLD_PRINTF(("%s: qpt %d, it %d, cst %d, sct %d\n", __func__,
+           querier_present_timers_running6, interface_timers_running6,
+           current_state_timers_running6, state_change_timers_running6));
+
+       /*
+        * MLDv1 querier present timer processing.
+        */
+       if (querier_present_timers_running6) {
+               querier_present_timers_running6 = 0;
+               LIST_FOREACH(mli, &mli_head, mli_link) {
+                       MLI_LOCK(mli);
+                       mld_v1_process_querier_timers(mli);
+                       if (mli->mli_v1_timer > 0) {
+                               querier_present_timers_running6 = 1;
+                       }
+                       MLI_UNLOCK(mli);
+               }
+       }
+
+       /*
+        * MLDv2 General Query response timer processing.
+        */
+       if (interface_timers_running6) {
+               MLD_PRINTF(("%s: interface timers running\n", __func__));
+               interface_timers_running6 = 0;
+               mli = LIST_FIRST(&mli_head);
+
+               while (mli != NULL) {
+                       if (mli->mli_flags & MLIF_PROCESSED) {
+                               mli = LIST_NEXT(mli, mli_link);
+                               continue;
+                       }
+
+                       MLI_LOCK(mli);
+                       if (mli->mli_version != MLD_VERSION_2) {
+                               MLI_UNLOCK(mli);
+                               mli = LIST_NEXT(mli, mli_link);
+                               continue;
+                       }
+                       /*
+                        * XXX The logic below ends up calling
+                        * mld_dispatch_packet which can unlock mli
+                        * and the global MLD lock.
+                        * Therefore grab a reference on MLI and also
+                        * check for generation count to see if we should
+                        * iterate the list again.
+                        */
+                       MLI_ADDREF_LOCKED(mli);
+
+                       if (mli->mli_v2_timer == 0) {
+                               /* Do nothing. */
+                       } else if (--mli->mli_v2_timer == 0) {
+                               if (mld_v2_dispatch_general_query(mli) > 0) {
+                                       interface_timers_running6 = 1;
+                               }
+                       } else {
+                               interface_timers_running6 = 1;
+                       }
+                       mli->mli_flags |= MLIF_PROCESSED;
+                       MLI_UNLOCK(mli);
+                       MLI_REMREF(mli);
+
+                       if (genid != mld_mli_list_genid) {
+                               MLD_PRINTF(("%s: MLD information list changed "
+                                   "in the middle of iteration! Restart iteration.\n",
+                                   __func__));
+                               mli = LIST_FIRST(&mli_head);
+                               genid = mld_mli_list_genid;
+                       } else {
+                               mli = LIST_NEXT(mli, mli_link);
+                       }
+               }
+
+               LIST_FOREACH(mli, &mli_head, mli_link)
+               mli->mli_flags &= ~MLIF_PROCESSED;
+       }
+
+
+
+       if (!current_state_timers_running6 &&
+           !state_change_timers_running6) {
+               goto out_locked;
+       }
+
+       current_state_timers_running6 = 0;
+       state_change_timers_running6 = 0;
+
+       MLD_PRINTF(("%s: state change timers running\n", __func__));
+
+       memset(&qrq, 0, sizeof(struct ifqueue));
+       qrq.ifq_maxlen = MLD_MAX_G_GS_PACKETS;
+
+       memset(&scq, 0, sizeof(struct ifqueue));
+       scq.ifq_maxlen = MLD_MAX_STATE_CHANGE_PACKETS;
+
+       /*
+        * MLD host report and state-change timer processing.
+        * Note: Processing a v2 group timer may remove a node.
+        */
+       mli = LIST_FIRST(&mli_head);
+
+       while (mli != NULL) {
+               struct in6_multistep step;
+
+               if (mli->mli_flags & MLIF_PROCESSED) {
+                       mli = LIST_NEXT(mli, mli_link);
+                       continue;
+               }
+
+               MLI_LOCK(mli);
+               ifp = mli->mli_ifp;
+               uri_sec = MLD_RANDOM_DELAY(mli->mli_uri);
+               MLI_UNLOCK(mli);
+
+               in6_multihead_lock_shared();
+               IN6_FIRST_MULTI(step, inm);
+               while (inm != NULL) {
+                       IN6M_LOCK(inm);
+                       if (inm->in6m_ifp != ifp) {
+                               goto next;
+                       }
+
+                       MLI_LOCK(mli);
+                       switch (mli->mli_version) {
+                       case MLD_VERSION_1:
+                               mld_v1_process_group_timer(inm,
+                                   mli->mli_version);
+                               break;
+                       case MLD_VERSION_2:
+                               mld_v2_process_group_timers(mli, &qrq,
+                                   &scq, inm, uri_sec);
+                               break;
+                       }
+                       MLI_UNLOCK(mli);
+next:
+                       IN6M_UNLOCK(inm);
+                       IN6_NEXT_MULTI(step, inm);
+               }
+               in6_multihead_lock_done();
+
+               /*
+                * XXX The logic below ends up calling
+                * mld_dispatch_packet which can unlock mli
+                * and the global MLD lock.
+                * Therefore grab a reference on MLI and also
+                * check for generation count to see if we should
+                * iterate the list again.
+                */
+               MLI_LOCK(mli);
+               MLI_ADDREF_LOCKED(mli);
+               if (mli->mli_version == MLD_VERSION_1) {
+                       mld_dispatch_queue_locked(mli, &mli->mli_v1q, 0);
+               } else if (mli->mli_version == MLD_VERSION_2) {
+                       MLI_UNLOCK(mli);
+                       mld_dispatch_queue_locked(NULL, &qrq, 0);
+                       mld_dispatch_queue_locked(NULL, &scq, 0);
+                       VERIFY(qrq.ifq_len == 0);
+                       VERIFY(scq.ifq_len == 0);
+                       MLI_LOCK(mli);
+               }
+               /*
+                * In case there are still any pending membership reports
+                * which didn't get drained at version change time.
+                */
+               IF_DRAIN(&mli->mli_v1q);
+               /*
+                * Release all deferred inm records, and drain any locally
+                * enqueued packets; do it even if the current MLD version
+                * for the link is no longer MLDv2, in order to handle the
+                * version change case.
+                */
+               mld_flush_relq(mli, (struct mld_in6m_relhead *)&in6m_dthead);
+               VERIFY(SLIST_EMPTY(&mli->mli_relinmhead));
+               mli->mli_flags |= MLIF_PROCESSED;
+               MLI_UNLOCK(mli);
+               MLI_REMREF(mli);
+
+               IF_DRAIN(&qrq);
+               IF_DRAIN(&scq);
+
+               if (genid != mld_mli_list_genid) {
+                       MLD_PRINTF(("%s: MLD information list changed "
+                           "in the middle of iteration! Restart iteration.\n",
+                           __func__));
+                       mli = LIST_FIRST(&mli_head);
+                       genid = mld_mli_list_genid;
+               } else {
+                       mli = LIST_NEXT(mli, mli_link);
+               }
+       }
+
+       LIST_FOREACH(mli, &mli_head, mli_link)
+       mli->mli_flags &= ~MLIF_PROCESSED;
+
+out_locked:
+       /* re-arm the timer if there's work to do */
+       mld_timeout_run = 0;
+       mld_sched_timeout();
+       MLD_UNLOCK();
+
+       /* Now that we're dropped all locks, release detached records */
+       MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
+}
+
+static void
+mld_sched_timeout(void)
+{
+       MLD_LOCK_ASSERT_HELD();
+
+       if (!mld_timeout_run &&
+           (querier_present_timers_running6 || current_state_timers_running6 ||
+           interface_timers_running6 || state_change_timers_running6)) {
+               mld_timeout_run = 1;
+               timeout(mld_timeout, NULL, hz);
+       }
+}
+
+/*
+ * Free the in6_multi reference(s) for this MLD lifecycle.
+ *
+ * Caller must be holding mli_lock.
+ */
+static void
+mld_flush_relq(struct mld_ifinfo *mli, struct mld_in6m_relhead *in6m_dthead)
+{
+       struct in6_multi *inm;
+
+again:
+       MLI_LOCK_ASSERT_HELD(mli);
+       inm = SLIST_FIRST(&mli->mli_relinmhead);
+       if (inm != NULL) {
+               int lastref;
+
+               SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
+               MLI_UNLOCK(mli);
+
+               in6_multihead_lock_exclusive();
+               IN6M_LOCK(inm);
+               VERIFY(inm->in6m_nrelecnt != 0);
+               inm->in6m_nrelecnt--;
+               lastref = in6_multi_detach(inm);
+               VERIFY(!lastref || (!(inm->in6m_debug & IFD_ATTACHED) &&
+                   inm->in6m_reqcnt == 0));
+               IN6M_UNLOCK(inm);
+               in6_multihead_lock_done();
+               /* from mli_relinmhead */
+               IN6M_REMREF(inm);
+               /* from in6_multihead_list */
+               if (lastref) {
+                       /*
+                        * Defer releasing our final reference, as we
+                        * are holding the MLD lock at this point, and
+                        * we could end up with locking issues later on
+                        * (while issuing SIOCDELMULTI) when this is the
+                        * final reference count.  Let the caller do it
+                        * when it is safe.
+                        */
+                       MLD_ADD_DETACHED_IN6M(in6m_dthead, inm);
+               }
+               MLI_LOCK(mli);
+               goto again;
+       }
+}
+
+/*
+ * Update host report group timer.
+ * Will update the global pending timer flags.
+ */
+static void
+mld_v1_process_group_timer(struct in6_multi *inm, const int mld_version)
+{
+#pragma unused(mld_version)
+       int report_timer_expired;
+
+       MLD_LOCK_ASSERT_HELD();
+       IN6M_LOCK_ASSERT_HELD(inm);
+       MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
+
+       if (inm->in6m_timer == 0) {
+               report_timer_expired = 0;
+       } else if (--inm->in6m_timer == 0) {
+               report_timer_expired = 1;
+       } else {
+               current_state_timers_running6 = 1;
+               /* caller will schedule timer */
+               return;
+       }
+
+       switch (inm->in6m_state) {
+       case MLD_NOT_MEMBER:
+       case MLD_SILENT_MEMBER:
+       case MLD_IDLE_MEMBER:
+       case MLD_LAZY_MEMBER:
+       case MLD_SLEEPING_MEMBER:
+       case MLD_AWAKENING_MEMBER:
+               break;
+       case MLD_REPORTING_MEMBER:
+               if (report_timer_expired) {
+                       inm->in6m_state = MLD_IDLE_MEMBER;
+                       (void) mld_v1_transmit_report(inm,
+                           MLD_LISTENER_REPORT);
+                       IN6M_LOCK_ASSERT_HELD(inm);
+                       MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
+               }
+               break;
+       case MLD_G_QUERY_PENDING_MEMBER:
+       case MLD_SG_QUERY_PENDING_MEMBER:
+       case MLD_LEAVING_MEMBER:
+               break;
+       }
+}
+
+/*
+ * Update a group's timers for MLDv2.
+ * Will update the global pending timer flags.
+ * Note: Unlocked read from mli.
+ */
+static void
+mld_v2_process_group_timers(struct mld_ifinfo *mli,
+    struct ifqueue *qrq, struct ifqueue *scq,
+    struct in6_multi *inm, const int uri_sec)
+{
+       int query_response_timer_expired;
+       int state_change_retransmit_timer_expired;
+
+       MLD_LOCK_ASSERT_HELD();
+       IN6M_LOCK_ASSERT_HELD(inm);
+       MLI_LOCK_ASSERT_HELD(mli);
+       VERIFY(mli == inm->in6m_mli);
+
+       query_response_timer_expired = 0;
+       state_change_retransmit_timer_expired = 0;
+
+       /*
+        * During a transition from compatibility mode back to MLDv2,
+        * a group record in REPORTING state may still have its group
+        * timer active. This is a no-op in this function; it is easier
+        * to deal with it here than to complicate the timeout path.
+        */
+       if (inm->in6m_timer == 0) {
+               query_response_timer_expired = 0;
+       } else if (--inm->in6m_timer == 0) {
+               query_response_timer_expired = 1;
+       } else {
+               current_state_timers_running6 = 1;
+               /* caller will schedule timer */
+       }
+
+       if (inm->in6m_sctimer == 0) {
+               state_change_retransmit_timer_expired = 0;
+       } else if (--inm->in6m_sctimer == 0) {
+               state_change_retransmit_timer_expired = 1;
+       } else {
+               state_change_timers_running6 = 1;
+               /* caller will schedule timer */
+       }
+
+       /* We are in timer callback, so be quick about it. */
+       if (!state_change_retransmit_timer_expired &&
+           !query_response_timer_expired) {
+               return;
+       }
+
+       switch (inm->in6m_state) {
+       case MLD_NOT_MEMBER:
+       case MLD_SILENT_MEMBER:
+       case MLD_SLEEPING_MEMBER:
+       case MLD_LAZY_MEMBER:
+       case MLD_AWAKENING_MEMBER:
+       case MLD_IDLE_MEMBER:
+               break;
+       case MLD_G_QUERY_PENDING_MEMBER:
+       case MLD_SG_QUERY_PENDING_MEMBER:
+               /*
+                * Respond to a previously pending Group-Specific
+                * or Group-and-Source-Specific query by enqueueing
+                * the appropriate Current-State report for
+                * immediate transmission.
+                */
+               if (query_response_timer_expired) {
+                       int retval;
+
+                       retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
+                           (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
+                           0);
+                       MLD_PRINTF(("%s: enqueue record = %d\n",
+                           __func__, retval));
+                       inm->in6m_state = MLD_REPORTING_MEMBER;
+                       in6m_clear_recorded(inm);
+               }
+       /* FALLTHROUGH */
+       case MLD_REPORTING_MEMBER:
+       case MLD_LEAVING_MEMBER:
+               if (state_change_retransmit_timer_expired) {
+                       /*
+                        * State-change retransmission timer fired.
+                        * If there are any further pending retransmissions,
+                        * set the global pending state-change flag, and
+                        * reset the timer.
+                        */
+                       if (--inm->in6m_scrv > 0) {
+                               inm->in6m_sctimer = uri_sec;
+                               state_change_timers_running6 = 1;
+                               /* caller will schedule timer */
+                       }
+                       /*
+                        * Retransmit the previously computed state-change
+                        * report. If there are no further pending
+                        * retransmissions, the mbuf queue will be consumed.
+                        * Update T0 state to T1 as we have now sent
+                        * a state-change.
+                        */
+                       (void) mld_v2_merge_state_changes(inm, scq);
+
+                       in6m_commit(inm);
+                       MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
+                           ip6_sprintf(&inm->in6m_addr),
+                           if_name(inm->in6m_ifp)));
+
+                       /*
+                        * If we are leaving the group for good, make sure
+                        * we release MLD's reference to it.
+                        * This release must be deferred using a SLIST,
+                        * as we are called from a loop which traverses
+                        * the in_ifmultiaddr TAILQ.
+                        */
+                       if (inm->in6m_state == MLD_LEAVING_MEMBER &&
+                           inm->in6m_scrv == 0) {
+                               inm->in6m_state = MLD_NOT_MEMBER;
+                               /*
+                                * A reference has already been held in
+                                * mld_final_leave() for this inm, so
+                                * no need to hold another one.  We also
+                                * bumped up its request count then, so
+                                * that it stays in in6_multihead.  Both
+                                * of them will be released when it is
+                                * dequeued later on.
+                                */
+                               VERIFY(inm->in6m_nrelecnt != 0);
+                               SLIST_INSERT_HEAD(&mli->mli_relinmhead,
+                                   inm, in6m_nrele);
+                       }
+               }
+               break;
+       }
+}
+
+/*
+ * Switch to a different version on the given interface,
+ * as per Section 9.12.
+ */
+static uint32_t
+mld_set_version(struct mld_ifinfo *mli, const int mld_version)
+{
+       int old_version_timer;
+
+       MLI_LOCK_ASSERT_HELD(mli);
+
+       MLD_PRINTF(("%s: switching to v%d on ifp 0x%llx(%s)\n", __func__,
+           mld_version, (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp),
+           if_name(mli->mli_ifp)));
+
+       if (mld_version == MLD_VERSION_1) {
+               /*
+                * Compute the "Older Version Querier Present" timer as per
+                * Section 9.12, in seconds.
+                */
+               old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
+               mli->mli_v1_timer = old_version_timer;
+       }
+
+       if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
+               mli->mli_version = MLD_VERSION_1;
+               mld_v2_cancel_link_timers(mli);
+       }
+
+       MLI_LOCK_ASSERT_HELD(mli);
+
+       return mli->mli_v1_timer;
+}
+
+/*
+ * Cancel pending MLDv2 timers for the given link and all groups
+ * joined on it; state-change, general-query, and group-query timers.
+ *
+ * Only ever called on a transition from v2 to Compatibility mode. Kill
+ * the timers stone dead (this may be expensive for large N groups), they
+ * will be restarted if Compatibility Mode deems that they must be due to
+ * query processing.
+ */
+static void
+mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
+{
+       struct ifnet            *ifp;
+       struct in6_multi        *inm;
+       struct in6_multistep    step;
+
+       MLI_LOCK_ASSERT_HELD(mli);
+
+       MLD_PRINTF(("%s: cancel v2 timers on ifp 0x%llx(%s)\n", __func__,
+           (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp), if_name(mli->mli_ifp)));
+
+       /*
+        * Stop the v2 General Query Response on this link stone dead.
+        * If timer is woken up due to interface_timers_running6,
+        * the flag will be cleared if there are no pending link timers.
+        */
+       mli->mli_v2_timer = 0;
+
+       /*
+        * Now clear the current-state and state-change report timers
+        * for all memberships scoped to this link.
+        */
+       ifp = mli->mli_ifp;
+       MLI_UNLOCK(mli);
+
+       in6_multihead_lock_shared();
+       IN6_FIRST_MULTI(step, inm);
+       while (inm != NULL) {
+               IN6M_LOCK(inm);
+               if (inm->in6m_ifp != ifp) {
+                       goto next;
+               }
+
+               switch (inm->in6m_state) {
+               case MLD_NOT_MEMBER:
+               case MLD_SILENT_MEMBER:
+               case MLD_IDLE_MEMBER:
+               case MLD_LAZY_MEMBER:
+               case MLD_SLEEPING_MEMBER:
+               case MLD_AWAKENING_MEMBER:
+                       /*
+                        * These states are either not relevant in v2 mode,
+                        * or are unreported. Do nothing.
+                        */
+                       break;
+               case MLD_LEAVING_MEMBER:
+                       /*
+                        * If we are leaving the group and switching
+                        * version, we need to release the final
+                        * reference held for issuing the INCLUDE {}.
+                        * During mld_final_leave(), we bumped up both the
+                        * request and reference counts.  Since we cannot
+                        * call in6_multi_detach() here, defer this task to
+                        * the timer routine.
+                        */
+                       VERIFY(inm->in6m_nrelecnt != 0);
+                       MLI_LOCK(mli);
+                       SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
+                           in6m_nrele);
+                       MLI_UNLOCK(mli);
+               /* FALLTHROUGH */
+               case MLD_G_QUERY_PENDING_MEMBER:
+               case MLD_SG_QUERY_PENDING_MEMBER:
+                       in6m_clear_recorded(inm);
+               /* FALLTHROUGH */
+               case MLD_REPORTING_MEMBER:
+                       inm->in6m_state = MLD_REPORTING_MEMBER;
+                       break;
+               }
+               /*
+                * Always clear state-change and group report timers.
+                * Free any pending MLDv2 state-change records.
+                */
+               inm->in6m_sctimer = 0;
+               inm->in6m_timer = 0;
+               IF_DRAIN(&inm->in6m_scq);
+next:
+               IN6M_UNLOCK(inm);
+               IN6_NEXT_MULTI(step, inm);
+       }
+       in6_multihead_lock_done();
+
+       MLI_LOCK(mli);
+}
+
+/*
+ * Update the Older Version Querier Present timers for a link.
+ * See Section 9.12 of RFC 3810.
+ */
+static void
+mld_v1_process_querier_timers(struct mld_ifinfo *mli)
+{
+       MLI_LOCK_ASSERT_HELD(mli);
+
+       if (mld_v2enable && mli->mli_version != MLD_VERSION_2 &&
+           --mli->mli_v1_timer == 0) {
+               /*
+                * MLDv1 Querier Present timer expired; revert to MLDv2.
+                */
+               MLD_PRINTF(("%s: transition from v%d -> v%d on 0x%llx(%s)\n",
+                   __func__, mli->mli_version, MLD_VERSION_2,
+                   (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp),
+                   if_name(mli->mli_ifp)));
+               mli->mli_version = MLD_VERSION_2;
+       }
+}
+
+/*
+ * Transmit an MLDv1 report immediately.
+ */
+static int
+mld_v1_transmit_report(struct in6_multi *in6m, const int type)
+{
+       struct ifnet            *ifp;
+       struct in6_ifaddr       *ia;
+       struct ip6_hdr          *ip6;
+       struct mbuf             *mh, *md;
+       struct mld_hdr          *mld;
+       int                     error = 0;
+
+       IN6M_LOCK_ASSERT_HELD(in6m);
+       MLI_LOCK_ASSERT_HELD(in6m->in6m_mli);
+
+       ifp = in6m->in6m_ifp;
+       /* ia may be NULL if link-local address is tentative. */
+       ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
+
+       MGETHDR(mh, M_DONTWAIT, MT_HEADER);
+       if (mh == NULL) {
+               if (ia != NULL) {
+                       IFA_REMREF(&ia->ia_ifa);
+               }
+               return ENOMEM;
+       }
+       MGET(md, M_DONTWAIT, MT_DATA);
+       if (md == NULL) {
+               m_free(mh);
+               if (ia != NULL) {
+                       IFA_REMREF(&ia->ia_ifa);
+               }
+               return ENOMEM;
+       }
+       mh->m_next = md;
+
+       /*
+        * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
+        * that ether_output() does not need to allocate another mbuf
+        * for the header in the most common case.
+        */
+       MH_ALIGN(mh, sizeof(struct ip6_hdr));
+       mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
+       mh->m_len = sizeof(struct ip6_hdr);
+
+       ip6 = mtod(mh, struct ip6_hdr *);
+       ip6->ip6_flow = 0;
+       ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
+       ip6->ip6_vfc |= IPV6_VERSION;
+       ip6->ip6_nxt = IPPROTO_ICMPV6;
+       if (ia != NULL) {
+               IFA_LOCK(&ia->ia_ifa);
+       }
+       ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
+       if (ia != NULL) {
+               IFA_UNLOCK(&ia->ia_ifa);
+               IFA_REMREF(&ia->ia_ifa);
+               ia = NULL;
+       }
+       ip6->ip6_dst = in6m->in6m_addr;
+
+       md->m_len = sizeof(struct mld_hdr);
+       mld = mtod(md, struct mld_hdr *);
+       mld->mld_type = type;
+       mld->mld_code = 0;
+       mld->mld_cksum = 0;
+       mld->mld_maxdelay = 0;
+       mld->mld_reserved = 0;
+       mld->mld_addr = in6m->in6m_addr;
+       in6_clearscope(&mld->mld_addr);
+       mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
+           sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
+
+       mld_save_context(mh, ifp);
+       mh->m_flags |= M_MLDV1;
+
+       /*
+        * Due to the fact that at this point we are possibly holding
+        * in6_multihead_lock in shared or exclusive mode, we can't call
+        * mld_dispatch_packet() here since that will eventually call
+        * ip6_output(), which will try to lock in6_multihead_lock and cause
+        * a deadlock.
+        * Instead we defer the work to the mld_timeout() thread, thus
+        * avoiding unlocking in_multihead_lock here.
+        */
+       if (IF_QFULL(&in6m->in6m_mli->mli_v1q)) {
+               MLD_PRINTF(("%s: v1 outbound queue full\n", __func__));
+               error = ENOMEM;
+               m_freem(mh);
+       } else {
+               IF_ENQUEUE(&in6m->in6m_mli->mli_v1q, mh);
+               VERIFY(error == 0);
+       }
+
+       return error;
+}
+
+/*
+ * Process a state change from the upper layer for the given IPv6 group.
+ *
+ * Each socket holds a reference on the in6_multi in its own ip_moptions.
+ * The socket layer will have made the necessary updates to.the group
+ * state, it is now up to MLD to issue a state change report if there
+ * has been any change between T0 (when the last state-change was issued)
+ * and T1 (now).
+ *
+ * We use the MLDv2 state machine at group level. The MLd module
+ * however makes the decision as to which MLD protocol version to speak.
+ * A state change *from* INCLUDE {} always means an initial join.
+ * A state change *to* INCLUDE {} always means a final leave.
+ *
+ * If delay is non-zero, and the state change is an initial multicast
+ * join, the state change report will be delayed by 'delay' ticks
+ * in units of seconds if MLDv1 is active on the link; otherwise
+ * the initial MLDv2 state change report will be delayed by whichever
+ * is sooner, a pending state-change timer or delay itself.
+ */
+int
+mld_change_state(struct in6_multi *inm, struct mld_tparams *mtp,
+    const int delay)
+{
+       struct mld_ifinfo *mli;
+       struct ifnet *ifp;
+       int error = 0;
+
+       VERIFY(mtp != NULL);
+       bzero(mtp, sizeof(*mtp));
+
+       IN6M_LOCK_ASSERT_HELD(inm);
+       VERIFY(inm->in6m_mli != NULL);
+       MLI_LOCK_ASSERT_NOTHELD(inm->in6m_mli);
+
+       /*
+        * Try to detect if the upper layer just asked us to change state
+        * for an interface which has now gone away.
+        */
+       VERIFY(inm->in6m_ifma != NULL);
+       ifp = inm->in6m_ifma->ifma_ifp;
+       /*
+        * Sanity check that netinet6's notion of ifp is the same as net's.
+        */
+       VERIFY(inm->in6m_ifp == ifp);
+
+       mli = MLD_IFINFO(ifp);
+       VERIFY(mli != NULL);
+
+       /*
+        * If we detect a state transition to or from MCAST_UNDEFINED
+        * for this group, then we are starting or finishing an MLD
+        * life cycle for this group.
+        */
+       if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
+               MLD_PRINTF(("%s: inm transition %d -> %d\n", __func__,
+                   inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode));
+               if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
+                       MLD_PRINTF(("%s: initial join\n", __func__));
+                       error = mld_initial_join(inm, mli, mtp, delay);
+                       goto out;
+               } else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
+                       MLD_PRINTF(("%s: final leave\n", __func__));
+                       mld_final_leave(inm, mli, mtp);
+                       goto out;
+               }
+       } else {
+               MLD_PRINTF(("%s: filter set change\n", __func__));
+       }
+
+       error = mld_handle_state_change(inm, mli, mtp);
+out:
+       return error;
+}
+
+/*
+ * Perform the initial join for an MLD group.
+ *
+ * When joining a group:
+ *  If the group should have its MLD traffic suppressed, do nothing.
+ *  MLDv1 starts sending MLDv1 host membership reports.
+ *  MLDv2 will schedule an MLDv2 state-change report containing the
+ *  initial state of the membership.
+ *
+ * If the delay argument is non-zero, then we must delay sending the
+ * initial state change for delay ticks (in units of seconds).
+ */
+static int
+mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
+    struct mld_tparams *mtp, const int delay)
+{
+       struct ifnet            *ifp;
+       struct ifqueue          *ifq;
+       int                      error, retval, syncstates;
+       int                      odelay;
+
+       IN6M_LOCK_ASSERT_HELD(inm);
+       MLI_LOCK_ASSERT_NOTHELD(mli);
+       VERIFY(mtp != NULL);
+
+       MLD_PRINTF(("%s: initial join %s on ifp 0x%llx(%s)\n",
+           __func__, ip6_sprintf(&inm->in6m_addr),
+           (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
+           if_name(inm->in6m_ifp)));
+
+       error = 0;
+       syncstates = 1;
+
+       ifp = inm->in6m_ifp;
+
+       MLI_LOCK(mli);
+       VERIFY(mli->mli_ifp == ifp);
+
+       /*
+        * Avoid MLD if group is :
+        * 1. Joined on loopback, OR
+        * 2. On a link that is marked MLIF_SILENT
+        * 3. rdar://problem/19227650 Is link local scoped and
+        *    on cellular interface
+        * 4. Is a type that should not be reported (node local
+        *    or all node link local multicast.
+        * All other groups enter the appropriate state machine
+        * for the version in use on this link.
+        */
+       if ((ifp->if_flags & IFF_LOOPBACK) ||
+           (mli->mli_flags & MLIF_SILENT) ||
+           (IFNET_IS_CELLULAR(ifp) &&
+           IN6_IS_ADDR_MC_LINKLOCAL(&inm->in6m_addr)) ||
+           !mld_is_addr_reported(&inm->in6m_addr)) {
+               MLD_PRINTF(("%s: not kicking state machine for silent group\n",
+                   __func__));
+               inm->in6m_state = MLD_SILENT_MEMBER;
+               inm->in6m_timer = 0;
+       } else {
+               /*
+                * Deal with overlapping in6_multi lifecycle.
+                * If this group was LEAVING, then make sure
+                * we drop the reference we picked up to keep the
+                * group around for the final INCLUDE {} enqueue.
+                * Since we cannot call in6_multi_detach() here,
+                * defer this task to the timer routine.
+                */
+               if (mli->mli_version == MLD_VERSION_2 &&
+                   inm->in6m_state == MLD_LEAVING_MEMBER) {
+                       VERIFY(inm->in6m_nrelecnt != 0);
+                       SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
+                           in6m_nrele);
+               }
+
+               inm->in6m_state = MLD_REPORTING_MEMBER;
+
+               switch (mli->mli_version) {
+               case MLD_VERSION_1:
+                       /*
+                        * If a delay was provided, only use it if
+                        * it is greater than the delay normally
+                        * used for an MLDv1 state change report,
+                        * and delay sending the initial MLDv1 report
+                        * by not transitioning to the IDLE state.
+                        */
+                       odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI);
+                       if (delay) {
+                               inm->in6m_timer = max(delay, odelay);
+                               mtp->cst = 1;
+                       } else {
+                               inm->in6m_state = MLD_IDLE_MEMBER;
+                               error = mld_v1_transmit_report(inm,
+                                   MLD_LISTENER_REPORT);
+
+                               IN6M_LOCK_ASSERT_HELD(inm);
+                               MLI_LOCK_ASSERT_HELD(mli);
+
+                               if (error == 0) {
+                                       inm->in6m_timer = odelay;
+                                       mtp->cst = 1;
+                               }
+                       }
+                       break;
+
+               case MLD_VERSION_2:
+                       /*
+                        * Defer update of T0 to T1, until the first copy
+                        * of the state change has been transmitted.
+                        */
+                       syncstates = 0;
+
+                       /*
+                        * Immediately enqueue a State-Change Report for
+                        * this interface, freeing any previous reports.
+                        * Don't kick the timers if there is nothing to do,
+                        * or if an error occurred.
+                        */
+                       ifq = &inm->in6m_scq;
+                       IF_DRAIN(ifq);
+                       retval = mld_v2_enqueue_group_record(ifq, inm, 1,
+                           0, 0, (mli->mli_flags & MLIF_USEALLOW));
+                       mtp->cst = (ifq->ifq_len > 0);
+                       MLD_PRINTF(("%s: enqueue record = %d\n",
+                           __func__, retval));
+                       if (retval <= 0) {
+                               error = retval * -1;
+                               break;
+                       }
+
+                       /*
+                        * Schedule transmission of pending state-change
+                        * report up to RV times for this link. The timer
+                        * will fire at the next mld_timeout (1 second)),
+                        * giving us an opportunity to merge the reports.
+                        *
+                        * If a delay was provided to this function, only
+                        * use this delay if sooner than the existing one.
+                        */
+                       VERIFY(mli->mli_rv > 1);
+                       inm->in6m_scrv = mli->mli_rv;
+                       if (delay) {
+                               if (inm->in6m_sctimer > 1) {
+                                       inm->in6m_sctimer =
+                                           min(inm->in6m_sctimer, delay);
+                               } else {
+                                       inm->in6m_sctimer = delay;
+                               }
+                       } else {
+                               inm->in6m_sctimer = 1;
+                       }
+                       mtp->sct = 1;
+                       error = 0;
+                       break;
+               }
+       }
+       MLI_UNLOCK(mli);
+
+       /*
+        * Only update the T0 state if state change is atomic,
+        * i.e. we don't need to wait for a timer to fire before we
+        * can consider the state change to have been communicated.
+        */
+       if (syncstates) {
+               in6m_commit(inm);
+               MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
+                   ip6_sprintf(&inm->in6m_addr),
+                   if_name(inm->in6m_ifp)));
+       }
+
+       return error;
+}
+
+/*
+ * Issue an intermediate state change during the life-cycle.
+ */
+static int
+mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli,
+    struct mld_tparams *mtp)
+{
+       struct ifnet            *ifp;
+       int                      retval = 0;
+
+       IN6M_LOCK_ASSERT_HELD(inm);
+       MLI_LOCK_ASSERT_NOTHELD(mli);
+       VERIFY(mtp != NULL);
+
+       MLD_PRINTF(("%s: state change for %s on ifp 0x%llx(%s)\n",
+           __func__, ip6_sprintf(&inm->in6m_addr),
+           (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
+           if_name(inm->in6m_ifp)));
+
+       ifp = inm->in6m_ifp;
+
+       MLI_LOCK(mli);
+       VERIFY(mli->mli_ifp == ifp);
+
+       if ((ifp->if_flags & IFF_LOOPBACK) ||
+           (mli->mli_flags & MLIF_SILENT) ||
+           !mld_is_addr_reported(&inm->in6m_addr) ||
+           (mli->mli_version != MLD_VERSION_2)) {
+               MLI_UNLOCK(mli);
+               if (!mld_is_addr_reported(&inm->in6m_addr)) {
+                       MLD_PRINTF(("%s: not kicking state machine for silent "
+                           "group\n", __func__));
+               }
+               MLD_PRINTF(("%s: nothing to do\n", __func__));
+               in6m_commit(inm);
+               MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
+                   ip6_sprintf(&inm->in6m_addr),
+                   if_name(inm->in6m_ifp)));
+               goto done;
+       }
+
+       IF_DRAIN(&inm->in6m_scq);
+
+       retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
+           (mli->mli_flags & MLIF_USEALLOW));
+       mtp->cst = (inm->in6m_scq.ifq_len > 0);
+       MLD_PRINTF(("%s: enqueue record = %d\n", __func__, retval));
+       if (retval <= 0) {
+               MLI_UNLOCK(mli);
+               retval *= -1;
+               goto done;
+       } else {
+               retval = 0;
+       }
+
+       /*
+        * If record(s) were enqueued, start the state-change
+        * report timer for this group.
+        */
+       inm->in6m_scrv = mli->mli_rv;
+       inm->in6m_sctimer = 1;
+       mtp->sct = 1;
+       MLI_UNLOCK(mli);
+
+done:
+       return retval;
+}
+
+/*
+ * Perform the final leave for a multicast address.
+ *
+ * When leaving a group:
+ *  MLDv1 sends a DONE message, if and only if we are the reporter.
+ *  MLDv2 enqueues a state-change report containing a transition
+ *  to INCLUDE {} for immediate transmission.
+ */
+static void
+mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli,
+    struct mld_tparams *mtp)
+{
+       int syncstates = 1;
+
+       IN6M_LOCK_ASSERT_HELD(inm);
+       MLI_LOCK_ASSERT_NOTHELD(mli);
+       VERIFY(mtp != NULL);
+
+       MLD_PRINTF(("%s: final leave %s on ifp 0x%llx(%s)\n",
+           __func__, ip6_sprintf(&inm->in6m_addr),
+           (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
+           if_name(inm->in6m_ifp)));
+
+       switch (inm->in6m_state) {
+       case MLD_NOT_MEMBER:
+       case MLD_SILENT_MEMBER:
+       case MLD_LEAVING_MEMBER:
+               /* Already leaving or left; do nothing. */
+               MLD_PRINTF(("%s: not kicking state machine for silent group\n",
+                   __func__));
+               break;
+       case MLD_REPORTING_MEMBER:
+       case MLD_IDLE_MEMBER:
+       case MLD_G_QUERY_PENDING_MEMBER:
+       case MLD_SG_QUERY_PENDING_MEMBER:
+               MLI_LOCK(mli);
+               if (mli->mli_version == MLD_VERSION_1) {
+                       if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
+                           inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
+                               panic("%s: MLDv2 state reached, not MLDv2 "
+                                   "mode\n", __func__);
+                               /* NOTREACHED */
+                       }
+                       /* scheduler timer if enqueue is successful */
+                       mtp->cst = (mld_v1_transmit_report(inm,
+                           MLD_LISTENER_DONE) == 0);
+
+                       IN6M_LOCK_ASSERT_HELD(inm);
+                       MLI_LOCK_ASSERT_HELD(mli);
+
+                       inm->in6m_state = MLD_NOT_MEMBER;
+               } else if (mli->mli_version == MLD_VERSION_2) {
+                       /*
+                        * Stop group timer and all pending reports.
+                        * Immediately enqueue a state-change report
+                        * TO_IN {} to be sent on the next timeout,
+                        * giving us an opportunity to merge reports.
+                        */
+                       IF_DRAIN(&inm->in6m_scq);
+                       inm->in6m_timer = 0;
+                       inm->in6m_scrv = mli->mli_rv;
+                       MLD_PRINTF(("%s: Leaving %s/%s with %d "
+                           "pending retransmissions.\n", __func__,
+                           ip6_sprintf(&inm->in6m_addr),
+                           if_name(inm->in6m_ifp),
+                           inm->in6m_scrv));
+                       if (inm->in6m_scrv == 0) {
+                               inm->in6m_state = MLD_NOT_MEMBER;
+                               inm->in6m_sctimer = 0;
+                       } else {
+                               int retval;
+                               /*
+                                * Stick around in the in6_multihead list;
+                                * the final detach will be issued by
+                                * mld_v2_process_group_timers() when
+                                * the retransmit timer expires.
+                                */
+                               IN6M_ADDREF_LOCKED(inm);
+                               VERIFY(inm->in6m_debug & IFD_ATTACHED);
+                               inm->in6m_reqcnt++;
+                               VERIFY(inm->in6m_reqcnt >= 1);
+                               inm->in6m_nrelecnt++;
+                               VERIFY(inm->in6m_nrelecnt != 0);
+
+                               retval = mld_v2_enqueue_group_record(
+                                       &inm->in6m_scq, inm, 1, 0, 0,
+                                       (mli->mli_flags & MLIF_USEALLOW));
+                               mtp->cst = (inm->in6m_scq.ifq_len > 0);
+                               KASSERT(retval != 0,
+                                   ("%s: enqueue record = %d\n", __func__,
+                                   retval));
+
+                               inm->in6m_state = MLD_LEAVING_MEMBER;
+                               inm->in6m_sctimer = 1;
+                               mtp->sct = 1;
+                               syncstates = 0;
+                       }
+               }
+               MLI_UNLOCK(mli);
+               break;
+       case MLD_LAZY_MEMBER:
+       case MLD_SLEEPING_MEMBER:
+       case MLD_AWAKENING_MEMBER:
+               /* Our reports are suppressed; do nothing. */
+               break;
+       }
+
+       if (syncstates) {
+               in6m_commit(inm);
+               MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
+                   ip6_sprintf(&inm->in6m_addr),
+                   if_name(inm->in6m_ifp)));
+               inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
+               MLD_PRINTF(("%s: T1 now MCAST_UNDEFINED for 0x%llx/%s\n",
+                   __func__, (uint64_t)VM_KERNEL_ADDRPERM(&inm->in6m_addr),
+                   if_name(inm->in6m_ifp)));
+       }
+}
+
+/*
+ * Enqueue an MLDv2 group record to the given output queue.
+ *
+ * If is_state_change is zero, a current-state record is appended.
+ * If is_state_change is non-zero, a state-change report is appended.
+ *
+ * If is_group_query is non-zero, an mbuf packet chain is allocated.
+ * If is_group_query is zero, and if there is a packet with free space
+ * at the tail of the queue, it will be appended to providing there
+ * is enough free space.
+ * Otherwise a new mbuf packet chain is allocated.
+ *
+ * If is_source_query is non-zero, each source is checked to see if
+ * it was recorded for a Group-Source query, and will be omitted if
+ * it is not both in-mode and recorded.
+ *
+ * If use_block_allow is non-zero, state change reports for initial join
+ * and final leave, on an inclusive mode group with a source list, will be
+ * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
+ *
+ * The function will attempt to allocate leading space in the packet
+ * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
+ *
+ * If successful the size of all data appended to the queue is returned,
+ * otherwise an error code less than zero is returned, or zero if
+ * no record(s) were appended.
+ */
+static int
+mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
+    const int is_state_change, const int is_group_query,
+    const int is_source_query, const int use_block_allow)
+{
+       struct mldv2_record      mr;
+       struct mldv2_record     *pmr;
+       struct ifnet            *ifp;
+       struct ip6_msource      *ims, *nims;
+       struct mbuf             *m0, *m, *md;
+       int                      error, is_filter_list_change;
+       int                      minrec0len, m0srcs, msrcs, nbytes, off;
+       int                      record_has_sources;
+       int                      now;
+       int                      type;
+       uint8_t                  mode;
+
+       IN6M_LOCK_ASSERT_HELD(inm);
+       MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
+
+       error = 0;
+       ifp = inm->in6m_ifp;
+       is_filter_list_change = 0;
+       m = NULL;
+       m0 = NULL;
+       m0srcs = 0;
+       msrcs = 0;
+       nbytes = 0;
+       nims = NULL;
+       record_has_sources = 1;
+       pmr = NULL;
+       type = MLD_DO_NOTHING;
+       mode = inm->in6m_st[1].iss_fmode;
+
+       /*
+        * If we did not transition out of ASM mode during t0->t1,
+        * and there are no source nodes to process, we can skip
+        * the generation of source records.
+        */
+       if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
+           inm->in6m_nsrc == 0) {
+               record_has_sources = 0;
+       }
+
+       if (is_state_change) {
+               /*
+                * Queue a state change record.
+                * If the mode did not change, and there are non-ASM
+                * listeners or source filters present,
+                * we potentially need to issue two records for the group.
+                * If there are ASM listeners, and there was no filter
+                * mode transition of any kind, do nothing.
+                *
+                * If we are transitioning to MCAST_UNDEFINED, we need
+                * not send any sources. A transition to/from this state is
+                * considered inclusive with some special treatment.
+                *
+                * If we are rewriting initial joins/leaves to use
+                * ALLOW/BLOCK, and the group's membership is inclusive,
+                * we need to send sources in all cases.
+                */
+               if (mode != inm->in6m_st[0].iss_fmode) {
+                       if (mode == MCAST_EXCLUDE) {
+                               MLD_PRINTF(("%s: change to EXCLUDE\n",
+                                   __func__));
+                               type = MLD_CHANGE_TO_EXCLUDE_MODE;
+                       } else {
+                               MLD_PRINTF(("%s: change to INCLUDE\n",
+                                   __func__));
+                               if (use_block_allow) {
+                                       /*
+                                        * XXX
+                                        * Here we're interested in state
+                                        * edges either direction between
+                                        * MCAST_UNDEFINED and MCAST_INCLUDE.
+                                        * Perhaps we should just check
+                                        * the group state, rather than
+                                        * the filter mode.
+                                        */
+                                       if (mode == MCAST_UNDEFINED) {
+                                               type = MLD_BLOCK_OLD_SOURCES;
+                                       } else {
+                                               type = MLD_ALLOW_NEW_SOURCES;
+                                       }
+                               } else {
+                                       type = MLD_CHANGE_TO_INCLUDE_MODE;
+                                       if (mode == MCAST_UNDEFINED) {
+                                               record_has_sources = 0;
+                                       }
+                               }
+                       }
+               } else {
+                       if (record_has_sources) {
+                               is_filter_list_change = 1;
+                       } else {
+                               type = MLD_DO_NOTHING;
+                       }
+               }
+       } else {
+               /*
+                * Queue a current state record.
+                */
+               if (mode == MCAST_EXCLUDE) {
+                       type = MLD_MODE_IS_EXCLUDE;
+               } else if (mode == MCAST_INCLUDE) {
+                       type = MLD_MODE_IS_INCLUDE;
+                       VERIFY(inm->in6m_st[1].iss_asm == 0);
+               }
+       }
+
+       /*
+        * Generate the filter list changes using a separate function.
+        */
+       if (is_filter_list_change) {
+               return mld_v2_enqueue_filter_change(ifq, inm);
+       }
+
+       if (type == MLD_DO_NOTHING) {
+               MLD_PRINTF(("%s: nothing to do for %s/%s\n",
+                   __func__, ip6_sprintf(&inm->in6m_addr),
+                   if_name(inm->in6m_ifp)));
+               return 0;
+       }
+
+       /*
+        * If any sources are present, we must be able to fit at least
+        * one in the trailing space of the tail packet's mbuf,
+        * ideally more.
+        */
+       minrec0len = sizeof(struct mldv2_record);
+       if (record_has_sources) {
+               minrec0len += sizeof(struct in6_addr);
+       }
+       MLD_PRINTF(("%s: queueing %s for %s/%s\n", __func__,
+           mld_rec_type_to_str(type),
+           ip6_sprintf(&inm->in6m_addr),
+           if_name(inm->in6m_ifp)));
+
+       /*
+        * Check if we have a packet in the tail of the queue for this
+        * group into which the first group record for this group will fit.
+        * Otherwise allocate a new packet.
+        * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
+        * Note: Group records for G/GSR query responses MUST be sent
+        * in their own packet.
+        */
+       m0 = ifq->ifq_tail;
+       if (!is_group_query &&
+           m0 != NULL &&
+           (m0->m_pkthdr.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
+           (m0->m_pkthdr.len + minrec0len) <
+           (ifp->if_mtu - MLD_MTUSPACE)) {
+               m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
+                   sizeof(struct mldv2_record)) /
+                   sizeof(struct in6_addr);
+               m = m0;
+               MLD_PRINTF(("%s: use existing packet\n", __func__));
+       } else {
+               if (IF_QFULL(ifq)) {
+                       MLD_PRINTF(("%s: outbound queue full\n", __func__));
+                       return -ENOMEM;
+               }
+               m = NULL;
+               m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
+                   sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
+               if (!is_state_change && !is_group_query) {
+                       m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
+               }
+               if (m == NULL) {
+                       m = m_gethdr(M_DONTWAIT, MT_DATA);
+               }
+               if (m == NULL) {
+                       return -ENOMEM;
+               }
+
+               mld_save_context(m, ifp);
+
+               MLD_PRINTF(("%s: allocated first packet\n", __func__));
+       }
+
+       /*
+        * Append group record.
+        * If we have sources, we don't know how many yet.
+        */
+       mr.mr_type = type;
+       mr.mr_datalen = 0;
+       mr.mr_numsrc = 0;
+       mr.mr_addr = inm->in6m_addr;
+       in6_clearscope(&mr.mr_addr);
+       if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
+               if (m != m0) {
+                       m_freem(m);
+               }
+               MLD_PRINTF(("%s: m_append() failed.\n", __func__));
+               return -ENOMEM;
+       }
+       nbytes += sizeof(struct mldv2_record);
+
+       /*
+        * Append as many sources as will fit in the first packet.
+        * If we are appending to a new packet, the chain allocation
+        * may potentially use clusters; use m_getptr() in this case.
+        * If we are appending to an existing packet, we need to obtain
+        * a pointer to the group record after m_append(), in case a new
+        * mbuf was allocated.
+        *
+        * Only append sources which are in-mode at t1. If we are
+        * transitioning to MCAST_UNDEFINED state on the group, and
+        * use_block_allow is zero, do not include source entries.
+        * Otherwise, we need to include this source in the report.
+        *
+        * Only report recorded sources in our filter set when responding
+        * to a group-source query.
+        */
+       if (record_has_sources) {
+               if (m == m0) {
+                       md = m_last(m);
+                       pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
+                           md->m_len - nbytes);
+               } else {
+                       md = m_getptr(m, 0, &off);
+                       pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
+                           off);
+               }
+               msrcs = 0;
+               RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
+                   nims) {
+                       MLD_PRINTF(("%s: visit node %s\n", __func__,
+                           ip6_sprintf(&ims->im6s_addr)));
+                       now = im6s_get_mode(inm, ims, 1);
+                       MLD_PRINTF(("%s: node is %d\n", __func__, now));
+                       if ((now != mode) ||
+                           (now == mode &&
+                           (!use_block_allow && mode == MCAST_UNDEFINED))) {
+                               MLD_PRINTF(("%s: skip node\n", __func__));
+                               continue;
+                       }
+                       if (is_source_query && ims->im6s_stp == 0) {
+                               MLD_PRINTF(("%s: skip unrecorded node\n",
+                                   __func__));
+                               continue;
+                       }
+                       MLD_PRINTF(("%s: append node\n", __func__));
+                       if (!m_append(m, sizeof(struct in6_addr),
+                           (void *)&ims->im6s_addr)) {
+                               if (m != m0) {
+                                       m_freem(m);
+                               }
+                               MLD_PRINTF(("%s: m_append() failed.\n",
+                                   __func__));
+                               return -ENOMEM;
+                       }
+                       nbytes += sizeof(struct in6_addr);
+                       ++msrcs;
+                       if (msrcs == m0srcs) {
+                               break;
+                       }
+               }
+               MLD_PRINTF(("%s: msrcs is %d this packet\n", __func__,
+                   msrcs));
+               pmr->mr_numsrc = htons(msrcs);
+               nbytes += (msrcs * sizeof(struct in6_addr));
+       }
+
+       if (is_source_query && msrcs == 0) {
+               MLD_PRINTF(("%s: no recorded sources to report\n", __func__));
+               if (m != m0) {
+                       m_freem(m);
+               }
+               return 0;
+       }
+
+       /*
+        * We are good to go with first packet.
+        */
+       if (m != m0) {
+               MLD_PRINTF(("%s: enqueueing first packet\n", __func__));
+               m->m_pkthdr.vt_nrecs = 1;
+               IF_ENQUEUE(ifq, m);
+       } else {
+               m->m_pkthdr.vt_nrecs++;
+       }
+       /*
+        * No further work needed if no source list in packet(s).
+        */
+       if (!record_has_sources) {
+               return nbytes;
+       }
+
+       /*
+        * Whilst sources remain to be announced, we need to allocate
+        * a new packet and fill out as many sources as will fit.
+        * Always try for a cluster first.
+        */
+       while (nims != NULL) {
+               if (IF_QFULL(ifq)) {
+                       MLD_PRINTF(("%s: outbound queue full\n", __func__));
+                       return -ENOMEM;
+               }
+               m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
+               if (m == NULL) {
+                       m = m_gethdr(M_DONTWAIT, MT_DATA);
+               }
+               if (m == NULL) {
+                       return -ENOMEM;
+               }
+               mld_save_context(m, ifp);
+               md = m_getptr(m, 0, &off);
+               pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
+               MLD_PRINTF(("%s: allocated next packet\n", __func__));
+
+               if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
+                       if (m != m0) {
+                               m_freem(m);
+                       }
+                       MLD_PRINTF(("%s: m_append() failed.\n", __func__));
+                       return -ENOMEM;
+               }
+               m->m_pkthdr.vt_nrecs = 1;
+               nbytes += sizeof(struct mldv2_record);
+
+               m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
+                   sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
+
+               msrcs = 0;
+               RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
+                       MLD_PRINTF(("%s: visit node %s\n",
+                           __func__, ip6_sprintf(&ims->im6s_addr)));
+                       now = im6s_get_mode(inm, ims, 1);
+                       if ((now != mode) ||
+                           (now == mode &&
+                           (!use_block_allow && mode == MCAST_UNDEFINED))) {
+                               MLD_PRINTF(("%s: skip node\n", __func__));
+                               continue;
+                       }
+                       if (is_source_query && ims->im6s_stp == 0) {
+                               MLD_PRINTF(("%s: skip unrecorded node\n",
+                                   __func__));
+                               continue;
+                       }
+                       MLD_PRINTF(("%s: append node\n", __func__));
+                       if (!m_append(m, sizeof(struct in6_addr),
+                           (void *)&ims->im6s_addr)) {
+                               if (m != m0) {
+                                       m_freem(m);
+                               }
+                               MLD_PRINTF(("%s: m_append() failed.\n",
+                                   __func__));
+                               return -ENOMEM;
+                       }
+                       ++msrcs;
+                       if (msrcs == m0srcs) {
+                               break;
+                       }
+               }
+               pmr->mr_numsrc = htons(msrcs);
+               nbytes += (msrcs * sizeof(struct in6_addr));
+
+               MLD_PRINTF(("%s: enqueueing next packet\n", __func__));
+               IF_ENQUEUE(ifq, m);
+       }
+
+       return nbytes;
+}
+
+/*
+ * Type used to mark record pass completion.
+ * We exploit the fact we can cast to this easily from the
+ * current filter modes on each ip_msource node.
+ */
+typedef enum {
+       REC_NONE = 0x00,        /* MCAST_UNDEFINED */
+       REC_ALLOW = 0x01,       /* MCAST_INCLUDE */
+       REC_BLOCK = 0x02,       /* MCAST_EXCLUDE */
+       REC_FULL = REC_ALLOW | REC_BLOCK
+} rectype_t;
+
+/*
+ * Enqueue an MLDv2 filter list change to the given output queue.
+ *
+ * Source list filter state is held in an RB-tree. When the filter list
+ * for a group is changed without changing its mode, we need to compute
+ * the deltas between T0 and T1 for each source in the filter set,
+ * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
+ *
+ * As we may potentially queue two record types, and the entire R-B tree
+ * needs to be walked at once, we break this out into its own function
+ * so we can generate a tightly packed queue of packets.
+ *
+ * XXX This could be written to only use one tree walk, although that makes
+ * serializing into the mbuf chains a bit harder. For now we do two walks
+ * which makes things easier on us, and it may or may not be harder on
+ * the L2 cache.
+ *
+ * If successful the size of all data appended to the queue is returned,
+ * otherwise an error code less than zero is returned, or zero if
+ * no record(s) were appended.
+ */
+static int
+mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
+{
+       static const int MINRECLEN =
+           sizeof(struct mldv2_record) + sizeof(struct in6_addr);
+       struct ifnet            *ifp;
+       struct mldv2_record      mr;
+       struct mldv2_record     *pmr;
+       struct ip6_msource      *ims, *nims;
+       struct mbuf             *m, *m0, *md;
+       int                      m0srcs, nbytes, npbytes, off, rsrcs, schanged;
+       int                      nallow, nblock;
+       uint8_t                  mode, now, then;
+       rectype_t                crt, drt, nrt;
+
+       IN6M_LOCK_ASSERT_HELD(inm);
+
+       if (inm->in6m_nsrc == 0 ||
+           (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0)) {
+               return 0;
+       }
+
+       ifp = inm->in6m_ifp;                    /* interface */
+       mode = inm->in6m_st[1].iss_fmode;       /* filter mode at t1 */
+       crt = REC_NONE; /* current group record type */
+       drt = REC_NONE; /* mask of completed group record types */
+       nrt = REC_NONE; /* record type for current node */
+       m0srcs = 0;     /* # source which will fit in current mbuf chain */
+       npbytes = 0;    /* # of bytes appended this packet */
+       nbytes = 0;     /* # of bytes appended to group's state-change queue */
+       rsrcs = 0;      /* # sources encoded in current record */
+       schanged = 0;   /* # nodes encoded in overall filter change */
+       nallow = 0;     /* # of source entries in ALLOW_NEW */
+       nblock = 0;     /* # of source entries in BLOCK_OLD */
+       nims = NULL;    /* next tree node pointer */
+
+       /*
+        * For each possible filter record mode.
+        * The first kind of source we encounter tells us which
+        * is the first kind of record we start appending.
+        * If a node transitioned to UNDEFINED at t1, its mode is treated
+        * as the inverse of the group's filter mode.
+        */
+       while (drt != REC_FULL) {
+               do {
+                       m0 = ifq->ifq_tail;
+                       if (m0 != NULL &&
+                           (m0->m_pkthdr.vt_nrecs + 1 <=
+                           MLD_V2_REPORT_MAXRECS) &&
+                           (m0->m_pkthdr.len + MINRECLEN) <
+                           (ifp->if_mtu - MLD_MTUSPACE)) {
+                               m = m0;
+                               m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
+                                   sizeof(struct mldv2_record)) /
+                                   sizeof(struct in6_addr);
+                               MLD_PRINTF(("%s: use previous packet\n",
+                                   __func__));
+                       } else {
+                               m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
+                               if (m == NULL) {
+                                       m = m_gethdr(M_DONTWAIT, MT_DATA);
+                               }
+                               if (m == NULL) {
+                                       MLD_PRINTF(("%s: m_get*() failed\n",
+                                           __func__));
+                                       return -ENOMEM;
+                               }
+                               m->m_pkthdr.vt_nrecs = 0;
+                               mld_save_context(m, ifp);
+                               m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
+                                   sizeof(struct mldv2_record)) /
+                                   sizeof(struct in6_addr);
+                               npbytes = 0;
+                               MLD_PRINTF(("%s: allocated new packet\n",
+                                   __func__));
+                       }
+                       /*
+                        * Append the MLD group record header to the
+                        * current packet's data area.
+                        * Recalculate pointer to free space for next
+                        * group record, in case m_append() allocated
+                        * a new mbuf or cluster.
+                        */
+                       memset(&mr, 0, sizeof(mr));
+                       mr.mr_addr = inm->in6m_addr;
+                       in6_clearscope(&mr.mr_addr);
+                       if (!m_append(m, sizeof(mr), (void *)&mr)) {
+                               if (m != m0) {
+                                       m_freem(m);
+                               }
+                               MLD_PRINTF(("%s: m_append() failed\n",
+                                   __func__));
+                               return -ENOMEM;
+                       }
+                       npbytes += sizeof(struct mldv2_record);
+                       if (m != m0) {
+                               /* new packet; offset in chain */
+                               md = m_getptr(m, npbytes -
+                                   sizeof(struct mldv2_record), &off);
+                               pmr = (struct mldv2_record *)(mtod(md,
+                                   uint8_t *) + off);
+                       } else {
+                               /* current packet; offset from last append */
+                               md = m_last(m);
+                               pmr = (struct mldv2_record *)(mtod(md,
+                                   uint8_t *) + md->m_len -
+                                   sizeof(struct mldv2_record));
+                       }
+                       /*
+                        * Begin walking the tree for this record type
+                        * pass, or continue from where we left off
+                        * previously if we had to allocate a new packet.
+                        * Only report deltas in-mode at t1.
+                        * We need not report included sources as allowed
+                        * if we are in inclusive mode on the group,
+                        * however the converse is not true.
+                        */
+                       rsrcs = 0;
+                       if (nims == NULL) {
+                               nims = RB_MIN(ip6_msource_tree,
+                                   &inm->in6m_srcs);
+                       }
+                       RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
+                               MLD_PRINTF(("%s: visit node %s\n", __func__,
+                                   ip6_sprintf(&ims->im6s_addr)));
+                               now = im6s_get_mode(inm, ims, 1);
+                               then = im6s_get_mode(inm, ims, 0);
+                               MLD_PRINTF(("%s: mode: t0 %d, t1 %d\n",
+                                   __func__, then, now));
+                               if (now == then) {
+                                       MLD_PRINTF(("%s: skip unchanged\n",
+                                           __func__));
+                                       continue;
+                               }
+                               if (mode == MCAST_EXCLUDE &&
+                                   now == MCAST_INCLUDE) {
+                                       MLD_PRINTF(("%s: skip IN src on EX "
+                                           "group\n", __func__));
+                                       continue;
+                               }
+                               nrt = (rectype_t)now;
+                               if (nrt == REC_NONE) {
+                                       nrt = (rectype_t)(~mode & REC_FULL);
+                               }
+                               if (schanged++ == 0) {
+                                       crt = nrt;
+                               } else if (crt != nrt) {
+                                       continue;
+                               }
+                               if (!m_append(m, sizeof(struct in6_addr),
+                                   (void *)&ims->im6s_addr)) {
+                                       if (m != m0) {
+                                               m_freem(m);
+                                       }
+                                       MLD_PRINTF(("%s: m_append() failed\n",
+                                           __func__));
+                                       return -ENOMEM;
+                               }
+                               nallow += !!(crt == REC_ALLOW);
+                               nblock += !!(crt == REC_BLOCK);
+                               if (++rsrcs == m0srcs) {
+                                       break;
+                               }
+                       }
+                       /*
+                        * If we did not append any tree nodes on this
+                        * pass, back out of allocations.
+                        */
+                       if (rsrcs == 0) {
+                               npbytes -= sizeof(struct mldv2_record);
+                               if (m != m0) {
+                                       MLD_PRINTF(("%s: m_free(m)\n",
+                                           __func__));
+                                       m_freem(m);
+                               } else {
+                                       MLD_PRINTF(("%s: m_adj(m, -mr)\n",
+                                           __func__));
+                                       m_adj(m, -((int)sizeof(
+                                                   struct mldv2_record)));
+                               }
+                               continue;
+                       }
+                       npbytes += (rsrcs * sizeof(struct in6_addr));
+                       if (crt == REC_ALLOW) {
+                               pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
+                       } else if (crt == REC_BLOCK) {
+                               pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
+                       }
+                       pmr->mr_numsrc = htons(rsrcs);
+                       /*
+                        * Count the new group record, and enqueue this
+                        * packet if it wasn't already queued.
+                        */
+                       m->m_pkthdr.vt_nrecs++;
+                       if (m != m0) {
+                               IF_ENQUEUE(ifq, m);
+                       }
+                       nbytes += npbytes;
+               } while (nims != NULL);
+               drt |= crt;
+               crt = (~crt & REC_FULL);
+       }
+
+       MLD_PRINTF(("%s: queued %d ALLOW_NEW, %d BLOCK_OLD\n", __func__,
+           nallow, nblock));
+
+       return nbytes;
+}
+
+static int
+mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
+{
+       struct ifqueue  *gq;
+       struct mbuf     *m;             /* pending state-change */
+       struct mbuf     *m0;            /* copy of pending state-change */
+       struct mbuf     *mt;            /* last state-change in packet */
+       struct mbuf     *n;
+       int              docopy, domerge;
+       u_int            recslen;
+
+       IN6M_LOCK_ASSERT_HELD(inm);
+
+       docopy = 0;
+       domerge = 0;
+       recslen = 0;
+
+       /*
+        * If there are further pending retransmissions, make a writable
+        * copy of each queued state-change message before merging.
+        */
+       if (inm->in6m_scrv > 0) {
+               docopy = 1;
+       }
+
+       gq = &inm->in6m_scq;
+#ifdef MLD_DEBUG
+       if (gq->ifq_head == NULL) {
+               MLD_PRINTF(("%s: WARNING: queue for inm 0x%llx is empty\n",
+                   __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm)));
+       }
+#endif
+
+       /*
+        * Use IF_REMQUEUE() instead of IF_DEQUEUE() below, since the
+        * packet might not always be at the head of the ifqueue.
+        */
+       m = gq->ifq_head;
+       while (m != NULL) {
+               /*
+                * Only merge the report into the current packet if
+                * there is sufficient space to do so; an MLDv2 report
+                * packet may only contain 65,535 group records.
+                * Always use a simple mbuf chain concatentation to do this,
+                * as large state changes for single groups may have
+                * allocated clusters.
+                */
+               domerge = 0;
+               mt = ifscq->ifq_tail;
+               if (mt != NULL) {
+                       recslen = m_length(m);
+
+                       if ((mt->m_pkthdr.vt_nrecs +
+                           m->m_pkthdr.vt_nrecs <=
+                           MLD_V2_REPORT_MAXRECS) &&
+                           (mt->m_pkthdr.len + recslen <=
+                           (inm->in6m_ifp->if_mtu - MLD_MTUSPACE))) {
+                               domerge = 1;
+                       }
+               }
+
+               if (!domerge && IF_QFULL(gq)) {
+                       MLD_PRINTF(("%s: outbound queue full, skipping whole "
+                           "packet 0x%llx\n", __func__,
+                           (uint64_t)VM_KERNEL_ADDRPERM(m)));
+                       n = m->m_nextpkt;
+                       if (!docopy) {
+                               IF_REMQUEUE(gq, m);
+                               m_freem(m);
+                       }
+                       m = n;
+                       continue;
+               }
+
+               if (!docopy) {
+                       MLD_PRINTF(("%s: dequeueing 0x%llx\n", __func__,
+                           (uint64_t)VM_KERNEL_ADDRPERM(m)));
+                       n = m->m_nextpkt;
+                       IF_REMQUEUE(gq, m);
+                       m0 = m;
+                       m = n;
+               } else {
+                       MLD_PRINTF(("%s: copying 0x%llx\n", __func__,
+                           (uint64_t)VM_KERNEL_ADDRPERM(m)));
+                       m0 = m_dup(m, M_NOWAIT);
+                       if (m0 == NULL) {
+                               return ENOMEM;
+                       }
+                       m0->m_nextpkt = NULL;
+                       m = m->m_nextpkt;
+               }
+
+               if (!domerge) {
+                       MLD_PRINTF(("%s: queueing 0x%llx to ifscq 0x%llx)\n",
+                           __func__, (uint64_t)VM_KERNEL_ADDRPERM(m0),
+                           (uint64_t)VM_KERNEL_ADDRPERM(ifscq)));
+                       IF_ENQUEUE(ifscq, m0);
+               } else {
+                       struct mbuf *mtl;       /* last mbuf of packet mt */
+
+                       MLD_PRINTF(("%s: merging 0x%llx with ifscq tail "
+                           "0x%llx)\n", __func__,
+                           (uint64_t)VM_KERNEL_ADDRPERM(m0),
+                           (uint64_t)VM_KERNEL_ADDRPERM(mt)));
+
+                       mtl = m_last(mt);
+                       m0->m_flags &= ~M_PKTHDR;
+                       mt->m_pkthdr.len += recslen;
+                       mt->m_pkthdr.vt_nrecs +=
+                           m0->m_pkthdr.vt_nrecs;
+
+                       mtl->m_next = m0;
+               }
+       }
+
+       return 0;
+}
+
+/*
+ * Respond to a pending MLDv2 General Query.
+ */
+static uint32_t
+mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
+{
+       struct ifnet            *ifp;
+       struct in6_multi        *inm;
+       struct in6_multistep    step;
+       int                      retval;
+
+       MLI_LOCK_ASSERT_HELD(mli);
+
+       VERIFY(mli->mli_version == MLD_VERSION_2);
+
+       ifp = mli->mli_ifp;
+       MLI_UNLOCK(mli);
+
+       in6_multihead_lock_shared();
+       IN6_FIRST_MULTI(step, inm);
+       while (inm != NULL) {
+               IN6M_LOCK(inm);
+               if (inm->in6m_ifp != ifp) {
+                       goto next;
+               }
+
+               switch (inm->in6m_state) {
+               case MLD_NOT_MEMBER:
+               case MLD_SILENT_MEMBER:
+                       break;
+               case MLD_REPORTING_MEMBER:
+               case MLD_IDLE_MEMBER:
+               case MLD_LAZY_MEMBER:
+               case MLD_SLEEPING_MEMBER:
+               case MLD_AWAKENING_MEMBER:
+                       inm->in6m_state = MLD_REPORTING_MEMBER;
+                       MLI_LOCK(mli);
+                       retval = mld_v2_enqueue_group_record(&mli->mli_gq,
+                           inm, 0, 0, 0, 0);
+                       MLI_UNLOCK(mli);
+                       MLD_PRINTF(("%s: enqueue record = %d\n",
+                           __func__, retval));
+                       break;
+               case MLD_G_QUERY_PENDING_MEMBER:
+               case MLD_SG_QUERY_PENDING_MEMBER:
+               case MLD_LEAVING_MEMBER:
+                       break;
+               }
+next:
+               IN6M_UNLOCK(inm);
+               IN6_NEXT_MULTI(step, inm);
+       }
+       in6_multihead_lock_done();
+
+       MLI_LOCK(mli);
+       mld_dispatch_queue_locked(mli, &mli->mli_gq, MLD_MAX_RESPONSE_BURST);
+       MLI_LOCK_ASSERT_HELD(mli);
+
+       /*
+        * Slew transmission of bursts over 1 second intervals.
+        */
+       if (mli->mli_gq.ifq_head != NULL) {
+               mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
+                       MLD_RESPONSE_BURST_INTERVAL);
+       }
+
+       return mli->mli_v2_timer;
+}
+
+/*
+ * Transmit the next pending message in the output queue.
+ *
+ * Must not be called with in6m_lockm or mli_lock held.
+ */
+static void
+mld_dispatch_packet(struct mbuf *m)
+{
+       struct ip6_moptions     *im6o;
+       struct ifnet            *ifp;
+       struct ifnet            *oifp = NULL;
+       struct mbuf             *m0;
+       struct mbuf             *md;
+       struct ip6_hdr          *ip6;
+       struct mld_hdr          *mld;
+       int                      error;
+       int                      off;
+       int                      type;
+
+       MLD_PRINTF(("%s: transmit 0x%llx\n", __func__,
+           (uint64_t)VM_KERNEL_ADDRPERM(m)));
+
+       /*
+        * Check if the ifnet is still attached.
+        */
+       ifp = mld_restore_context(m);
+       if (ifp == NULL || !ifnet_is_attached(ifp, 0)) {
+               MLD_PRINTF(("%s: dropped 0x%llx as ifindex %u went away.\n",
+                   __func__, (uint64_t)VM_KERNEL_ADDRPERM(m),
+                   (u_int)if_index));
+               m_freem(m);
+               ip6stat.ip6s_noroute++;
+               return;
+       }
+
+       im6o = ip6_allocmoptions(M_WAITOK);
+       if (im6o == NULL) {
+               m_freem(m);
+               return;
+       }
+
+       im6o->im6o_multicast_hlim  = 1;
+       im6o->im6o_multicast_loop = 0;
+       im6o->im6o_multicast_ifp = ifp;
+
+       if (m->m_flags & M_MLDV1) {
+               m0 = m;
+       } else {
+               m0 = mld_v2_encap_report(ifp, m);
+               if (m0 == NULL) {
+                       MLD_PRINTF(("%s: dropped 0x%llx\n", __func__,
+                           (uint64_t)VM_KERNEL_ADDRPERM(m)));
+                       /*
+                        * mld_v2_encap_report() has already freed our mbuf.
+                        */
+                       IM6O_REMREF(im6o);
+                       ip6stat.ip6s_odropped++;
+                       return;
+               }
+       }
+
+       mld_scrub_context(m0);
+       m->m_flags &= ~(M_PROTOFLAGS);
+       m0->m_pkthdr.rcvif = lo_ifp;
+
+       ip6 = mtod(m0, struct ip6_hdr *);
+       (void) in6_setscope(&ip6->ip6_dst, ifp, NULL);
+
+       /*
+        * Retrieve the ICMPv6 type before handoff to ip6_output(),
+        * so we can bump the stats.
+        */
+       md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
+       mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
+       type = mld->mld_type;
+
+       if (ifp->if_eflags & IFEF_TXSTART) {
+               /*
+                * Use control service class if the outgoing
+                * interface supports transmit-start model.
+                */
+               (void) m_set_service_class(m0, MBUF_SC_CTL);
+       }
+
+       error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, im6o,
+           &oifp, NULL);
+
+       IM6O_REMREF(im6o);
+
+       if (error) {
+               MLD_PRINTF(("%s: ip6_output(0x%llx) = %d\n", __func__,
+                   (uint64_t)VM_KERNEL_ADDRPERM(m0), error));
+               if (oifp != NULL) {
+                       ifnet_release(oifp);
+               }
+               return;
+       }
+
+       icmp6stat.icp6s_outhist[type]++;
+       if (oifp != NULL) {
+               icmp6_ifstat_inc(oifp, ifs6_out_msg);
+               switch (type) {
+               case MLD_LISTENER_REPORT:
+               case MLDV2_LISTENER_REPORT:
+                       icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
+                       break;
+               case MLD_LISTENER_DONE:
+                       icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
+                       break;
+               }
+               ifnet_release(oifp);
+       }
+}
+
+/*
+ * Encapsulate an MLDv2 report.
+ *
+ * KAME IPv6 requires that hop-by-hop options be passed separately,
+ * and that the IPv6 header be prepended in a separate mbuf.
+ *
+ * Returns a pointer to the new mbuf chain head, or NULL if the
+ * allocation failed.
+ */
+static struct mbuf *
+mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
+{
+       struct mbuf             *mh;
+       struct mldv2_report     *mld;
+       struct ip6_hdr          *ip6;
+       struct in6_ifaddr       *ia;
+       int                      mldreclen;
+
+       VERIFY(m->m_flags & M_PKTHDR);
+
+       /*
+        * RFC3590: OK to send as :: or tentative during DAD.
+        */
+       ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY | IN6_IFF_ANYCAST);
+       if (ia == NULL) {
+               MLD_PRINTF(("%s: warning: ia is NULL\n", __func__));
+       }
+
+       MGETHDR(mh, M_DONTWAIT, MT_HEADER);
+       if (mh == NULL) {
+               if (ia != NULL) {
+                       IFA_REMREF(&ia->ia_ifa);
+               }
+               m_freem(m);
+               return NULL;
+       }
+       MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
+
+       mldreclen = m_length(m);
+       MLD_PRINTF(("%s: mldreclen is %d\n", __func__, mldreclen));
+
+       mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
+       mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
+           sizeof(struct mldv2_report) + mldreclen;
+
+       ip6 = mtod(mh, struct ip6_hdr *);
+       ip6->ip6_flow = 0;
+       ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
+       ip6->ip6_vfc |= IPV6_VERSION;
+       ip6->ip6_nxt = IPPROTO_ICMPV6;
+       if (ia != NULL) {
+               IFA_LOCK(&ia->ia_ifa);
+       }
+       ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
+       if (ia != NULL) {
+               IFA_UNLOCK(&ia->ia_ifa);
+               IFA_REMREF(&ia->ia_ifa);
+               ia = NULL;
+       }
+       ip6->ip6_dst = in6addr_linklocal_allv2routers;
+       /* scope ID will be set in netisr */
+
+       mld = (struct mldv2_report *)(ip6 + 1);
+       mld->mld_type = MLDV2_LISTENER_REPORT;
+       mld->mld_code = 0;
+       mld->mld_cksum = 0;
+       mld->mld_v2_reserved = 0;
+       mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
+       m->m_pkthdr.vt_nrecs = 0;
+       m->m_flags &= ~M_PKTHDR;
+
+       mh->m_next = m;
+       mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
+           sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
+       return mh;
+}
+
+#ifdef MLD_DEBUG
+static const char *
+mld_rec_type_to_str(const int type)
+{
+       switch (type) {
+       case MLD_CHANGE_TO_EXCLUDE_MODE:
+               return "TO_EX";
+       case MLD_CHANGE_TO_INCLUDE_MODE:
+               return "TO_IN";
+       case MLD_MODE_IS_EXCLUDE:
+               return "MODE_EX";
+       case MLD_MODE_IS_INCLUDE:
+               return "MODE_IN";
+       case MLD_ALLOW_NEW_SOURCES:
+               return "ALLOW_NEW";
+       case MLD_BLOCK_OLD_SOURCES:
+               return "BLOCK_OLD";
+       default:
+               break;
+       }
+       return "unknown";
+}
+#endif
+
+void
+mld_init(void)
+{
+       MLD_PRINTF(("%s: initializing\n", __func__));
+
+       /* Setup lock group and attribute for mld_mtx */
+       mld_mtx_grp_attr = lck_grp_attr_alloc_init();
+       mld_mtx_grp = lck_grp_alloc_init("mld_mtx\n", mld_mtx_grp_attr);
+       mld_mtx_attr = lck_attr_alloc_init();
+       lck_mtx_init(&mld_mtx, mld_mtx_grp, mld_mtx_attr);
+
+       ip6_initpktopts(&mld_po);
+       mld_po.ip6po_hlim = 1;
+       mld_po.ip6po_hbh = &mld_ra.hbh;
+       mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
+       mld_po.ip6po_flags = IP6PO_DONTFRAG;
+       LIST_INIT(&mli_head);
+
+       mli_size = sizeof(struct mld_ifinfo);
+       mli_zone = zinit(mli_size, MLI_ZONE_MAX * mli_size,
+           0, MLI_ZONE_NAME);
+       if (mli_zone == NULL) {
+               panic("%s: failed allocating %s", __func__, MLI_ZONE_NAME);
+               /* NOTREACHED */
        }
+       zone_change(mli_zone, Z_EXPAND, TRUE);
+       zone_change(mli_zone, Z_CALLERACCT, FALSE);
 }