1/*
2 * Copyright (c) 2000-2013 Apple Inc. All rights reserved.
3 *
4 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5 *
6 * This file contains Original Code and/or Modifications of Original Code
7 * as defined in and that are subject to the Apple Public Source License
8 * Version 2.0 (the 'License'). You may not use this file except in
9 * compliance with the License. The rights granted to you under the License
10 * may not be used to create, or enable the creation or redistribution of,
11 * unlawful or unlicensed copies of an Apple operating system, or to
12 * circumvent, violate, or enable the circumvention or violation of, any
13 * terms of an Apple operating system software license agreement.
14 *
15 * Please obtain a copy of the License at
16 * http://www.opensource.apple.com/apsl/ and read it before using this file.
17 *
18 * The Original Code and all software distributed under the License are
19 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23 * Please see the License for the specific language governing rights and
24 * limitations under the License.
25 *
26 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27 */
28/*-
29 * Copyright (c) 2009 Bruce Simpson.
30 *
31 * Redistribution and use in source and binary forms, with or without
32 * modification, are permitted provided that the following conditions
33 * are met:
34 * 1. Redistributions of source code must retain the above copyright
35 *    notice, this list of conditions and the following disclaimer.
36 * 2. Redistributions in binary form must reproduce the above copyright
37 *    notice, this list of conditions and the following disclaimer in the
38 *    documentation and/or other materials provided with the distribution.
39 * 3. The name of the author may not be used to endorse or promote
40 *    products derived from this software without specific prior written
41 *    permission.
42 *
43 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
44 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
45 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
46 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
47 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
48 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
49 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
50 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
51 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
52 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
53 * SUCH DAMAGE.
54 */
55
56/*
57 * Copyright (c) 1988 Stephen Deering.
58 * Copyright (c) 1992, 1993
59 *	The Regents of the University of California.  All rights reserved.
60 *
61 * This code is derived from software contributed to Berkeley by
62 * Stephen Deering of Stanford University.
63 *
64 * Redistribution and use in source and binary forms, with or without
65 * modification, are permitted provided that the following conditions
66 * are met:
67 * 1. Redistributions of source code must retain the above copyright
68 *    notice, this list of conditions and the following disclaimer.
69 * 2. Redistributions in binary form must reproduce the above copyright
70 *    notice, this list of conditions and the following disclaimer in the
71 *    documentation and/or other materials provided with the distribution.
72 * 3. All advertising materials mentioning features or use of this software
73 *    must display the following acknowledgement:
74 *	This product includes software developed by the University of
75 *	California, Berkeley and its contributors.
76 * 4. Neither the name of the University nor the names of its contributors
77 *    may be used to endorse or promote products derived from this software
78 *    without specific prior written permission.
79 *
80 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
81 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
82 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
83 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
84 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
85 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
86 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
87 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
88 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
89 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
90 * SUCH DAMAGE.
91 *
92 *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
93 */
94/*
95 * NOTICE: This file was modified by SPARTA, Inc. in 2005 to introduce
96 * support for mandatory and extensible security protections.  This notice
97 * is included in support of clause 2.2 (b) of the Apple Public License,
98 * Version 2.0.
99 */
100
101#include <sys/cdefs.h>
102
103#include <sys/param.h>
104#include <sys/systm.h>
105#include <sys/mbuf.h>
106#include <sys/socket.h>
107#include <sys/protosw.h>
108#include <sys/sysctl.h>
109#include <sys/kernel.h>
110#include <sys/malloc.h>
111#include <sys/mcache.h>
112
113#include <dev/random/randomdev.h>
114
115#include <kern/zalloc.h>
116
117#include <net/if.h>
118#include <net/route.h>
119
120#include <netinet/in.h>
121#include <netinet/in_var.h>
122#include <netinet6/in6_var.h>
123#include <netinet/ip6.h>
124#include <netinet6/ip6_var.h>
125#include <netinet6/scope6_var.h>
126#include <netinet/icmp6.h>
127#include <netinet6/mld6.h>
128#include <netinet6/mld6_var.h>
129
130/* Lock group and attribute for mld_mtx */
131static lck_attr_t       *mld_mtx_attr;
132static lck_grp_t        *mld_mtx_grp;
133static lck_grp_attr_t   *mld_mtx_grp_attr;
134
135/*
136 * Locking and reference counting:
137 *
138 * mld_mtx mainly protects mli_head.  In cases where both mld_mtx and
139 * in6_multihead_lock must be held, the former must be acquired first in order
140 * to maintain lock ordering.  It is not a requirement that mld_mtx be
141 * acquired first before in6_multihead_lock, but in case both must be acquired
142 * in succession, the correct lock ordering must be followed.
143 *
144 * Instead of walking the if_multiaddrs list at the interface and returning
145 * the ifma_protospec value of a matching entry, we search the global list
146 * of in6_multi records and find it that way; this is done with in6_multihead
147 * lock held.  Doing so avoids the race condition issues that many other BSDs
148 * suffer from (therefore in our implementation, ifma_protospec will never be
149 * NULL for as long as the in6_multi is valid.)
150 *
151 * The above creates a requirement for the in6_multi to stay in in6_multihead
152 * list even after the final MLD leave (in MLDv2 mode) until no longer needs
153 * be retransmitted (this is not required for MLDv1.)  In order to handle
154 * this, the request and reference counts of the in6_multi are bumped up when
155 * the state changes to MLD_LEAVING_MEMBER, and later dropped in the timeout
156 * handler.  Each in6_multi holds a reference to the underlying mld_ifinfo.
157 *
158 * Thus, the permitted lock order is:
159 *
160 *	mld_mtx, in6_multihead_lock, inm6_lock, mli_lock
161 *
162 * Any may be taken independently, but if any are held at the same time,
163 * the above lock order must be followed.
164 */
165static decl_lck_mtx_data(, mld_mtx);
166
167SLIST_HEAD(mld_in6m_relhead, in6_multi);
168
169static void	mli_initvar(struct mld_ifinfo *, struct ifnet *, int);
170static struct mld_ifinfo *mli_alloc(int);
171static void	mli_free(struct mld_ifinfo *);
172static void	mli_delete(const struct ifnet *, struct mld_in6m_relhead *);
173static void	mld_dispatch_packet(struct mbuf *);
174static void	mld_final_leave(struct in6_multi *, struct mld_ifinfo *,
175		    struct mld_tparams *);
176static int	mld_handle_state_change(struct in6_multi *, struct mld_ifinfo *,
177		    struct mld_tparams *);
178static int	mld_initial_join(struct in6_multi *, struct mld_ifinfo *,
179		    struct mld_tparams *, const int);
180#ifdef MLD_DEBUG
181static const char *	mld_rec_type_to_str(const int);
182#endif
183static uint32_t	mld_set_version(struct mld_ifinfo *, const int);
184static void	mld_flush_relq(struct mld_ifinfo *, struct mld_in6m_relhead *);
185static void	mld_dispatch_queue(struct mld_ifinfo *, struct ifqueue *, int);
186static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
187		    /*const*/ struct mld_hdr *);
188static int	mld_v1_input_report(struct ifnet *, struct mbuf *,
189		    const struct ip6_hdr *, /*const*/ struct mld_hdr *);
190static void	mld_v1_process_group_timer(struct in6_multi *, const int);
191static void	mld_v1_process_querier_timers(struct mld_ifinfo *);
192static int	mld_v1_transmit_report(struct in6_multi *, const int);
193static uint32_t	mld_v1_update_group(struct in6_multi *, const int);
194static void	mld_v2_cancel_link_timers(struct mld_ifinfo *);
195static uint32_t	mld_v2_dispatch_general_query(struct mld_ifinfo *);
196static struct mbuf *
197		mld_v2_encap_report(struct ifnet *, struct mbuf *);
198static int	mld_v2_enqueue_filter_change(struct ifqueue *,
199		    struct in6_multi *);
200static int	mld_v2_enqueue_group_record(struct ifqueue *,
201		    struct in6_multi *, const int, const int, const int,
202		    const int);
203static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
204		    struct mbuf *, const int, const int);
205static int	mld_v2_merge_state_changes(struct in6_multi *,
206		    struct ifqueue *);
207static void	mld_v2_process_group_timers(struct mld_ifinfo *,
208		    struct ifqueue *, struct ifqueue *,
209		    struct in6_multi *, const int);
210static int	mld_v2_process_group_query(struct in6_multi *,
211		    int, struct mbuf *, const int);
212static int	sysctl_mld_gsr SYSCTL_HANDLER_ARGS;
213static int	sysctl_mld_ifinfo SYSCTL_HANDLER_ARGS;
214static int	sysctl_mld_v2enable SYSCTL_HANDLER_ARGS;
215
216static int mld_timeout_run;		/* MLD timer is scheduled to run */
217static void mld_timeout(void *);
218static void mld_sched_timeout(void);
219
220/*
221 * Normative references: RFC 2710, RFC 3590, RFC 3810.
222 */
223static struct timeval mld_gsrdelay = {10, 0};
224static LIST_HEAD(, mld_ifinfo) mli_head;
225
226static int querier_present_timers_running6;
227static int interface_timers_running6;
228static int state_change_timers_running6;
229static int current_state_timers_running6;
230
231/*
232 * Subsystem lock macros.
233 */
234#define	MLD_LOCK()			\
235	lck_mtx_lock(&mld_mtx)
236#define	MLD_LOCK_ASSERT_HELD()		\
237	lck_mtx_assert(&mld_mtx, LCK_MTX_ASSERT_OWNED)
238#define	MLD_LOCK_ASSERT_NOTHELD()	\
239	lck_mtx_assert(&mld_mtx, LCK_MTX_ASSERT_NOTOWNED)
240#define	MLD_UNLOCK()			\
241	lck_mtx_unlock(&mld_mtx)
242
243#define	MLD_ADD_DETACHED_IN6M(_head, _in6m) {				\
244	SLIST_INSERT_HEAD(_head, _in6m, in6m_dtle);			\
245}
246
247#define	MLD_REMOVE_DETACHED_IN6M(_head) {				\
248	struct in6_multi *_in6m, *_inm_tmp;				\
249	SLIST_FOREACH_SAFE(_in6m, _head, in6m_dtle, _inm_tmp) {		\
250		SLIST_REMOVE(_head, _in6m, in6_multi, in6m_dtle);	\
251		IN6M_REMREF(_in6m);					\
252	}								\
253	VERIFY(SLIST_EMPTY(_head));					\
254}
255
256#define	MLI_ZONE_MAX		64		/* maximum elements in zone */
257#define	MLI_ZONE_NAME		"mld_ifinfo"	/* zone name */
258
259static unsigned int mli_size;			/* size of zone element */
260static struct zone *mli_zone;			/* zone for mld_ifinfo */
261
262SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
263
264SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW | CTLFLAG_LOCKED, 0,
265    "IPv6 Multicast Listener Discovery");
266SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
267    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
268    &mld_gsrdelay.tv_sec, 0, sysctl_mld_gsr, "I",
269    "Rate limit for MLDv2 Group-and-Source queries in seconds");
270
271SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_LOCKED,
272   sysctl_mld_ifinfo, "Per-interface MLDv2 state");
273
274static int	mld_v1enable = 1;
275SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RW | CTLFLAG_LOCKED,
276    &mld_v1enable, 0, "Enable fallback to MLDv1");
277
278static int	mld_v2enable = 1;
279SYSCTL_PROC(_net_inet6_mld, OID_AUTO, v2enable,
280    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED,
281    &mld_v2enable, 0, sysctl_mld_v2enable, "I",
282    "Enable MLDv2 (debug purposes only)");
283
284static int	mld_use_allow = 1;
285SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RW | CTLFLAG_LOCKED,
286    &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
287
288#ifdef MLD_DEBUG
289int mld_debug = 0;
290SYSCTL_INT(_net_inet6_mld, OID_AUTO,
291	debug, CTLFLAG_RW | CTLFLAG_LOCKED,	&mld_debug, 0, "");
292#endif
293/*
294 * Packed Router Alert option structure declaration.
295 */
296struct mld_raopt {
297	struct ip6_hbh		hbh;
298	struct ip6_opt		pad;
299	struct ip6_opt_router	ra;
300} __packed;
301
302/*
303 * Router Alert hop-by-hop option header.
304 */
305static struct mld_raopt mld_ra = {
306	.hbh = { 0, 0 },
307	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
308	.ra = {
309	    .ip6or_type = (u_int8_t)IP6OPT_ROUTER_ALERT,
310	    .ip6or_len = (u_int8_t)(IP6OPT_RTALERT_LEN - 2),
311	    .ip6or_value =  {((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
312	        (IP6OPT_RTALERT_MLD & 0xFF) }
313	}
314};
315static struct ip6_pktopts mld_po;
316
317/* Store MLDv2 record count in the module private scratch space */
318#define	vt_nrecs	pkt_mpriv.__mpriv_u.__mpriv32[0].__mpriv32_u.__val16[0]
319
320static __inline void
321mld_save_context(struct mbuf *m, struct ifnet *ifp)
322{
323	m->m_pkthdr.rcvif = ifp;
324}
325
326static __inline void
327mld_scrub_context(struct mbuf *m)
328{
329	m->m_pkthdr.rcvif = NULL;
330}
331
332/*
333 * Restore context from a queued output chain.
334 * Return saved ifp.
335 */
336static __inline struct ifnet *
337mld_restore_context(struct mbuf *m)
338{
339        return (m->m_pkthdr.rcvif);
340}
341
342/*
343 * Retrieve or set threshold between group-source queries in seconds.
344 */
345static int
346sysctl_mld_gsr SYSCTL_HANDLER_ARGS
347{
348#pragma unused(arg1, arg2)
349	int error;
350	int i;
351
352	MLD_LOCK();
353
354	i = mld_gsrdelay.tv_sec;
355
356	error = sysctl_handle_int(oidp, &i, 0, req);
357	if (error || !req->newptr)
358		goto out_locked;
359
360	if (i < -1 || i >= 60) {
361		error = EINVAL;
362		goto out_locked;
363	}
364
365	mld_gsrdelay.tv_sec = i;
366
367out_locked:
368	MLD_UNLOCK();
369	return (error);
370}
371/*
372 * Expose struct mld_ifinfo to userland, keyed by ifindex.
373 * For use by ifmcstat(8).
374 *
375 */
376static int
377sysctl_mld_ifinfo SYSCTL_HANDLER_ARGS
378{
379#pragma unused(oidp)
380	int			*name;
381	int			 error;
382	u_int			 namelen;
383	struct ifnet		*ifp;
384	struct mld_ifinfo	*mli;
385	struct mld_ifinfo_u	mli_u;
386
387	name = (int *)arg1;
388	namelen = arg2;
389
390	if (req->newptr != USER_ADDR_NULL)
391		return (EPERM);
392
393	if (namelen != 1)
394		return (EINVAL);
395
396	MLD_LOCK();
397
398	if (name[0] <= 0 || name[0] > (u_int)if_index) {
399		error = ENOENT;
400		goto out_locked;
401	}
402
403	error = ENOENT;
404
405	ifnet_head_lock_shared();
406	ifp = ifindex2ifnet[name[0]];
407	ifnet_head_done();
408	if (ifp == NULL)
409		goto out_locked;
410
411	bzero(&mli_u, sizeof (mli_u));
412
413	LIST_FOREACH(mli, &mli_head, mli_link) {
414		MLI_LOCK(mli);
415		if (ifp != mli->mli_ifp) {
416			MLI_UNLOCK(mli);
417			continue;
418		}
419
420		mli_u.mli_ifindex = mli->mli_ifp->if_index;
421		mli_u.mli_version = mli->mli_version;
422		mli_u.mli_v1_timer = mli->mli_v1_timer;
423		mli_u.mli_v2_timer = mli->mli_v2_timer;
424		mli_u.mli_flags = mli->mli_flags;
425		mli_u.mli_rv = mli->mli_rv;
426		mli_u.mli_qi = mli->mli_qi;
427		mli_u.mli_qri = mli->mli_qri;
428		mli_u.mli_uri = mli->mli_uri;
429		MLI_UNLOCK(mli);
430
431		error = SYSCTL_OUT(req, &mli_u, sizeof (mli_u));
432		break;
433	}
434
435out_locked:
436	MLD_UNLOCK();
437	return (error);
438}
439
440static int
441sysctl_mld_v2enable SYSCTL_HANDLER_ARGS
442{
443#pragma unused(arg1, arg2)
444	int error;
445	int i;
446	struct mld_ifinfo *mli;
447	struct mld_tparams mtp = { 0, 0, 0, 0 };
448
449	MLD_LOCK();
450
451	i = mld_v2enable;
452
453	error = sysctl_handle_int(oidp, &i, 0, req);
454	if (error || !req->newptr)
455		goto out_locked;
456
457	if (i < 0 || i > 1) {
458		error = EINVAL;
459		goto out_locked;
460	}
461
462	mld_v2enable = i;
463	/*
464	 * If we enabled v2, the state transition will take care of upgrading
465	 * the MLD version back to v2. Otherwise, we have to explicitly
466	 * downgrade. Note that this functionality is to be used for debugging.
467	 */
468	if (mld_v2enable == 1)
469		goto out_locked;
470
471	LIST_FOREACH(mli, &mli_head, mli_link) {
472		MLI_LOCK(mli);
473		if (mld_set_version(mli, MLD_VERSION_1) > 0)
474			mtp.qpt = 1;
475		MLI_UNLOCK(mli);
476	}
477
478out_locked:
479	MLD_UNLOCK();
480
481	mld_set_timeout(&mtp);
482
483	return (error);
484}
485
486/*
487 * Dispatch an entire queue of pending packet chains.
488 *
489 * Must not be called with in6m_lock held.
490 */
491static void
492mld_dispatch_queue(struct mld_ifinfo *mli, struct ifqueue *ifq, int limit)
493{
494	struct mbuf *m;
495
496	if (mli != NULL)
497		MLI_LOCK_ASSERT_HELD(mli);
498
499	for (;;) {
500		IF_DEQUEUE(ifq, m);
501		if (m == NULL)
502			break;
503		MLD_PRINTF(("%s: dispatch 0x%llx from 0x%llx\n", __func__,
504		    (uint64_t)VM_KERNEL_ADDRPERM(ifq),
505		    (uint64_t)VM_KERNEL_ADDRPERM(m)));
506		if (mli != NULL)
507			MLI_UNLOCK(mli);
508		mld_dispatch_packet(m);
509		if (mli != NULL)
510			MLI_LOCK(mli);
511		if (--limit == 0)
512			break;
513	}
514
515	if (mli != NULL)
516		MLI_LOCK_ASSERT_HELD(mli);
517}
518
519/*
520 * Filter outgoing MLD report state by group.
521 *
522 * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
523 * and node-local addresses. However, kernel and socket consumers
524 * always embed the KAME scope ID in the address provided, so strip it
525 * when performing comparison.
526 * Note: This is not the same as the *multicast* scope.
527 *
528 * Return zero if the given group is one for which MLD reports
529 * should be suppressed, or non-zero if reports should be issued.
530 */
531static __inline__ int
532mld_is_addr_reported(const struct in6_addr *addr)
533{
534
535	VERIFY(IN6_IS_ADDR_MULTICAST(addr));
536
537	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
538		return (0);
539
540	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
541		struct in6_addr tmp = *addr;
542		in6_clearscope(&tmp);
543		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
544			return (0);
545	}
546
547	return (1);
548}
549
550/*
551 * Attach MLD when PF_INET6 is attached to an interface.
552 */
553struct mld_ifinfo *
554mld_domifattach(struct ifnet *ifp, int how)
555{
556	struct mld_ifinfo *mli;
557
558	MLD_PRINTF(("%s: called for ifp 0x%llx(%s)\n", __func__,
559	    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
560
561	mli = mli_alloc(how);
562	if (mli == NULL)
563		return (NULL);
564
565	MLD_LOCK();
566
567	MLI_LOCK(mli);
568	mli_initvar(mli, ifp, 0);
569	mli->mli_debug |= IFD_ATTACHED;
570	MLI_ADDREF_LOCKED(mli); /* hold a reference for mli_head */
571	MLI_ADDREF_LOCKED(mli); /* hold a reference for caller */
572	MLI_UNLOCK(mli);
573	ifnet_lock_shared(ifp);
574	mld6_initsilent(ifp, mli);
575	ifnet_lock_done(ifp);
576
577	LIST_INSERT_HEAD(&mli_head, mli, mli_link);
578
579	MLD_UNLOCK();
580
581	MLD_PRINTF(("%s: allocate mld_ifinfo for ifp 0x%llx(%s)\n",
582	    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
583
584	return (mli);
585}
586
587/*
588 * Attach MLD when PF_INET6 is reattached to an interface.  Caller is
589 * expected to have an outstanding reference to the mli.
590 */
591void
592mld_domifreattach(struct mld_ifinfo *mli)
593{
594	struct ifnet *ifp;
595
596	MLD_LOCK();
597
598	MLI_LOCK(mli);
599	VERIFY(!(mli->mli_debug & IFD_ATTACHED));
600	ifp = mli->mli_ifp;
601	VERIFY(ifp != NULL);
602	mli_initvar(mli, ifp, 1);
603	mli->mli_debug |= IFD_ATTACHED;
604	MLI_ADDREF_LOCKED(mli); /* hold a reference for mli_head */
605	MLI_UNLOCK(mli);
606	ifnet_lock_shared(ifp);
607	mld6_initsilent(ifp, mli);
608	ifnet_lock_done(ifp);
609
610	LIST_INSERT_HEAD(&mli_head, mli, mli_link);
611
612	MLD_UNLOCK();
613
614	MLD_PRINTF(("%s: reattached mld_ifinfo for ifp 0x%llx(%s)\n",
615	    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
616}
617
618/*
619 * Hook for domifdetach.
620 */
621void
622mld_domifdetach(struct ifnet *ifp)
623{
624	SLIST_HEAD(, in6_multi)	in6m_dthead;
625
626	SLIST_INIT(&in6m_dthead);
627
628	MLD_PRINTF(("%s: called for ifp 0x%llx(%s)\n", __func__,
629	    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
630
631	MLD_LOCK();
632	mli_delete(ifp, (struct mld_in6m_relhead *)&in6m_dthead);
633	MLD_UNLOCK();
634
635	/* Now that we're dropped all locks, release detached records */
636	MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
637}
638
639/*
640 * Called at interface detach time.  Note that we only flush all deferred
641 * responses and record releases; all remaining inm records and their source
642 * entries related to this interface are left intact, in order to handle
643 * the reattach case.
644 */
645static void
646mli_delete(const struct ifnet *ifp, struct mld_in6m_relhead *in6m_dthead)
647{
648	struct mld_ifinfo *mli, *tmli;
649
650	MLD_LOCK_ASSERT_HELD();
651
652	LIST_FOREACH_SAFE(mli, &mli_head, mli_link, tmli) {
653		MLI_LOCK(mli);
654		if (mli->mli_ifp == ifp) {
655			/*
656			 * Free deferred General Query responses.
657			 */
658			IF_DRAIN(&mli->mli_gq);
659			IF_DRAIN(&mli->mli_v1q);
660			mld_flush_relq(mli, in6m_dthead);
661			VERIFY(SLIST_EMPTY(&mli->mli_relinmhead));
662			mli->mli_debug &= ~IFD_ATTACHED;
663			MLI_UNLOCK(mli);
664
665			LIST_REMOVE(mli, mli_link);
666			MLI_REMREF(mli); /* release mli_head reference */
667			return;
668		}
669		MLI_UNLOCK(mli);
670	}
671	panic("%s: mld_ifinfo not found for ifp %p(%s)\n", __func__,
672	    ifp, ifp->if_xname);
673}
674
675__private_extern__ void
676mld6_initsilent(struct ifnet *ifp, struct mld_ifinfo *mli)
677{
678	ifnet_lock_assert(ifp, IFNET_LCK_ASSERT_OWNED);
679
680	MLI_LOCK_ASSERT_NOTHELD(mli);
681	MLI_LOCK(mli);
682	if (!(ifp->if_flags & IFF_MULTICAST) &&
683	    (ifp->if_eflags & (IFEF_IPV6_ND6ALT|IFEF_LOCALNET_PRIVATE)))
684		mli->mli_flags |= MLIF_SILENT;
685	else
686		mli->mli_flags &= ~MLIF_SILENT;
687	MLI_UNLOCK(mli);
688}
689
690static void
691mli_initvar(struct mld_ifinfo *mli, struct ifnet *ifp, int reattach)
692{
693	MLI_LOCK_ASSERT_HELD(mli);
694
695	mli->mli_ifp = ifp;
696	if (mld_v2enable)
697		mli->mli_version = MLD_VERSION_2;
698	else
699		mli->mli_version = MLD_VERSION_1;
700	mli->mli_flags = 0;
701	mli->mli_rv = MLD_RV_INIT;
702	mli->mli_qi = MLD_QI_INIT;
703	mli->mli_qri = MLD_QRI_INIT;
704	mli->mli_uri = MLD_URI_INIT;
705
706	if (mld_use_allow)
707		mli->mli_flags |= MLIF_USEALLOW;
708	if (!reattach)
709		SLIST_INIT(&mli->mli_relinmhead);
710
711	/*
712	 * Responses to general queries are subject to bounds.
713	 */
714	mli->mli_gq.ifq_maxlen = MLD_MAX_RESPONSE_PACKETS;
715	mli->mli_v1q.ifq_maxlen = MLD_MAX_RESPONSE_PACKETS;
716}
717
718static struct mld_ifinfo *
719mli_alloc(int how)
720{
721	struct mld_ifinfo *mli;
722
723	mli = (how == M_WAITOK) ? zalloc(mli_zone) : zalloc_noblock(mli_zone);
724	if (mli != NULL) {
725		bzero(mli, mli_size);
726		lck_mtx_init(&mli->mli_lock, mld_mtx_grp, mld_mtx_attr);
727		mli->mli_debug |= IFD_ALLOC;
728	}
729	return (mli);
730}
731
732static void
733mli_free(struct mld_ifinfo *mli)
734{
735	MLI_LOCK(mli);
736	if (mli->mli_debug & IFD_ATTACHED) {
737		panic("%s: attached mli=%p is being freed", __func__, mli);
738		/* NOTREACHED */
739	} else if (mli->mli_ifp != NULL) {
740		panic("%s: ifp not NULL for mli=%p", __func__, mli);
741		/* NOTREACHED */
742	} else if (!(mli->mli_debug & IFD_ALLOC)) {
743		panic("%s: mli %p cannot be freed", __func__, mli);
744		/* NOTREACHED */
745	} else if (mli->mli_refcnt != 0) {
746		panic("%s: non-zero refcnt mli=%p", __func__, mli);
747		/* NOTREACHED */
748	}
749	mli->mli_debug &= ~IFD_ALLOC;
750	MLI_UNLOCK(mli);
751
752	lck_mtx_destroy(&mli->mli_lock, mld_mtx_grp);
753	zfree(mli_zone, mli);
754}
755
756void
757mli_addref(struct mld_ifinfo *mli, int locked)
758{
759	if (!locked)
760		MLI_LOCK_SPIN(mli);
761	else
762		MLI_LOCK_ASSERT_HELD(mli);
763
764	if (++mli->mli_refcnt == 0) {
765		panic("%s: mli=%p wraparound refcnt", __func__, mli);
766		/* NOTREACHED */
767	}
768	if (!locked)
769		MLI_UNLOCK(mli);
770}
771
772void
773mli_remref(struct mld_ifinfo *mli)
774{
775	SLIST_HEAD(, in6_multi)	in6m_dthead;
776	struct ifnet *ifp;
777
778	MLI_LOCK_SPIN(mli);
779
780	if (mli->mli_refcnt == 0) {
781		panic("%s: mli=%p negative refcnt", __func__, mli);
782		/* NOTREACHED */
783	}
784
785	--mli->mli_refcnt;
786	if (mli->mli_refcnt > 0) {
787		MLI_UNLOCK(mli);
788		return;
789	}
790
791	ifp = mli->mli_ifp;
792	mli->mli_ifp = NULL;
793	IF_DRAIN(&mli->mli_gq);
794	IF_DRAIN(&mli->mli_v1q);
795	SLIST_INIT(&in6m_dthead);
796	mld_flush_relq(mli, (struct mld_in6m_relhead *)&in6m_dthead);
797	VERIFY(SLIST_EMPTY(&mli->mli_relinmhead));
798	MLI_UNLOCK(mli);
799
800	/* Now that we're dropped all locks, release detached records */
801	MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
802
803	MLD_PRINTF(("%s: freeing mld_ifinfo for ifp 0x%llx(%s)\n",
804	    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
805
806	mli_free(mli);
807}
808
809/*
810 * Process a received MLDv1 general or address-specific query.
811 * Assumes that the query header has been pulled up to sizeof(mld_hdr).
812 *
813 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
814 * mld_addr. This is OK as we own the mbuf chain.
815 */
816static int
817mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
818    /*const*/ struct mld_hdr *mld)
819{
820	struct mld_ifinfo	*mli;
821	struct in6_multi	*inm;
822	int			 err = 0, is_general_query;
823	uint16_t		 timer;
824	struct mld_tparams	 mtp = { 0, 0, 0, 0 };
825
826	MLD_LOCK_ASSERT_NOTHELD();
827
828	is_general_query = 0;
829
830	if (!mld_v1enable) {
831		MLD_PRINTF(("%s: ignore v1 query %s on ifp 0x%llx(%s)\n",
832		    __func__, ip6_sprintf(&mld->mld_addr),
833		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
834		goto done;
835	}
836
837	/*
838	 * RFC3810 Section 6.2: MLD queries must originate from
839	 * a router's link-local address.
840	 */
841	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
842		MLD_PRINTF(("%s: ignore v1 query src %s on ifp 0x%llx(%s)\n",
843		    __func__, ip6_sprintf(&ip6->ip6_src),
844		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
845		goto done;
846	}
847
848	/*
849	 * Do address field validation upfront before we accept
850	 * the query.
851	 */
852	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
853		/*
854		 * MLDv1 General Query.
855		 * If this was not sent to the all-nodes group, ignore it.
856		 */
857		struct in6_addr		 dst;
858
859		dst = ip6->ip6_dst;
860		in6_clearscope(&dst);
861		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes)) {
862			err = EINVAL;
863			goto done;
864		}
865		is_general_query = 1;
866	} else {
867		/*
868		 * Embed scope ID of receiving interface in MLD query for
869		 * lookup whilst we don't hold other locks.
870		 */
871		in6_setscope(&mld->mld_addr, ifp, NULL);
872	}
873
874	/*
875	 * Switch to MLDv1 host compatibility mode.
876	 */
877	mli = MLD_IFINFO(ifp);
878	VERIFY(mli != NULL);
879
880	MLI_LOCK(mli);
881	mtp.qpt = mld_set_version(mli, MLD_VERSION_1);
882	MLI_UNLOCK(mli);
883
884	timer = ntohs(mld->mld_maxdelay) / MLD_TIMER_SCALE;
885	if (timer == 0)
886		timer = 1;
887
888	if (is_general_query) {
889		struct in6_multistep step;
890
891		MLD_PRINTF(("%s: process v1 general query on ifp 0x%llx(%s)\n",
892		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
893		/*
894		 * For each reporting group joined on this
895		 * interface, kick the report timer.
896		 */
897		in6_multihead_lock_shared();
898		IN6_FIRST_MULTI(step, inm);
899		while (inm != NULL) {
900			IN6M_LOCK(inm);
901			if (inm->in6m_ifp == ifp)
902				mtp.cst += mld_v1_update_group(inm, timer);
903			IN6M_UNLOCK(inm);
904			IN6_NEXT_MULTI(step, inm);
905		}
906		in6_multihead_lock_done();
907	} else {
908		/*
909		 * MLDv1 Group-Specific Query.
910		 * If this is a group-specific MLDv1 query, we need only
911		 * look up the single group to process it.
912		 */
913		in6_multihead_lock_shared();
914		IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
915		in6_multihead_lock_done();
916
917		if (inm != NULL) {
918			IN6M_LOCK(inm);
919			MLD_PRINTF(("%s: process v1 query %s on "
920			    "ifp 0x%llx(%s)\n", __func__,
921			    ip6_sprintf(&mld->mld_addr),
922			    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
923			mtp.cst = mld_v1_update_group(inm, timer);
924			IN6M_UNLOCK(inm);
925			IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
926		}
927		/* XXX Clear embedded scope ID as userland won't expect it. */
928		in6_clearscope(&mld->mld_addr);
929	}
930done:
931	mld_set_timeout(&mtp);
932
933	return (err);
934}
935
936/*
937 * Update the report timer on a group in response to an MLDv1 query.
938 *
939 * If we are becoming the reporting member for this group, start the timer.
940 * If we already are the reporting member for this group, and timer is
941 * below the threshold, reset it.
942 *
943 * We may be updating the group for the first time since we switched
944 * to MLDv2. If we are, then we must clear any recorded source lists,
945 * and transition to REPORTING state; the group timer is overloaded
946 * for group and group-source query responses.
947 *
948 * Unlike MLDv2, the delay per group should be jittered
949 * to avoid bursts of MLDv1 reports.
950 */
951static uint32_t
952mld_v1_update_group(struct in6_multi *inm, const int timer)
953{
954	IN6M_LOCK_ASSERT_HELD(inm);
955
956	MLD_PRINTF(("%s: %s/%s timer=%d\n", __func__,
957	    ip6_sprintf(&inm->in6m_addr),
958	    if_name(inm->in6m_ifp), timer));
959
960	switch (inm->in6m_state) {
961	case MLD_NOT_MEMBER:
962	case MLD_SILENT_MEMBER:
963		break;
964	case MLD_REPORTING_MEMBER:
965		if (inm->in6m_timer != 0 &&
966		    inm->in6m_timer <= timer) {
967			MLD_PRINTF(("%s: REPORTING and timer running, "
968			    "skipping.\n", __func__));
969			break;
970		}
971		/* FALLTHROUGH */
972	case MLD_SG_QUERY_PENDING_MEMBER:
973	case MLD_G_QUERY_PENDING_MEMBER:
974	case MLD_IDLE_MEMBER:
975	case MLD_LAZY_MEMBER:
976	case MLD_AWAKENING_MEMBER:
977		MLD_PRINTF(("%s: ->REPORTING\n", __func__));
978		inm->in6m_state = MLD_REPORTING_MEMBER;
979		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
980		break;
981	case MLD_SLEEPING_MEMBER:
982		MLD_PRINTF(("%s: ->AWAKENING\n", __func__));
983		inm->in6m_state = MLD_AWAKENING_MEMBER;
984		break;
985	case MLD_LEAVING_MEMBER:
986		break;
987	}
988
989	return (inm->in6m_timer);
990}
991
992/*
993 * Process a received MLDv2 general, group-specific or
994 * group-and-source-specific query.
995 *
996 * Assumes that the query header has been pulled up to sizeof(mldv2_query).
997 *
998 * Return 0 if successful, otherwise an appropriate error code is returned.
999 */
1000static int
1001mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
1002    struct mbuf *m, const int off, const int icmp6len)
1003{
1004	struct mld_ifinfo	*mli;
1005	struct mldv2_query	*mld;
1006	struct in6_multi	*inm;
1007	uint32_t		 maxdelay, nsrc, qqi;
1008	int			 err = 0, is_general_query;
1009	uint16_t		 timer;
1010	uint8_t			 qrv;
1011	struct mld_tparams	 mtp = { 0, 0, 0, 0 };
1012
1013	MLD_LOCK_ASSERT_NOTHELD();
1014
1015	is_general_query = 0;
1016
1017	if (!mld_v2enable) {
1018		MLD_PRINTF(("%s: ignore v2 query %s on ifp 0x%llx(%s)\n",
1019		    __func__, ip6_sprintf(&ip6->ip6_src),
1020		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1021		goto done;
1022	}
1023
1024	/*
1025	 * RFC3810 Section 6.2: MLD queries must originate from
1026	 * a router's link-local address.
1027	 */
1028	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
1029		MLD_PRINTF(("%s: ignore v1 query src %s on ifp 0x%llx(%s)\n",
1030		    __func__, ip6_sprintf(&ip6->ip6_src),
1031		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1032		goto done;
1033	}
1034
1035	MLD_PRINTF(("%s: input v2 query on ifp 0x%llx(%s)\n", __func__,
1036	    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1037
1038	mld = (struct mldv2_query *)(mtod(m, uint8_t *) + off);
1039
1040	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
1041	if (maxdelay >= 32768) {
1042		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
1043			   (MLD_MRC_EXP(maxdelay) + 3);
1044	}
1045	timer = maxdelay / MLD_TIMER_SCALE;
1046	if (timer == 0)
1047		timer = 1;
1048
1049	qrv = MLD_QRV(mld->mld_misc);
1050	if (qrv < 2) {
1051		MLD_PRINTF(("%s: clamping qrv %d to %d\n", __func__,
1052		    qrv, MLD_RV_INIT));
1053		qrv = MLD_RV_INIT;
1054	}
1055
1056	qqi = mld->mld_qqi;
1057	if (qqi >= 128) {
1058		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
1059		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
1060	}
1061
1062	nsrc = ntohs(mld->mld_numsrc);
1063	if (nsrc > MLD_MAX_GS_SOURCES) {
1064		err = EMSGSIZE;
1065		goto done;
1066	}
1067	if (icmp6len < sizeof(struct mldv2_query) +
1068	    (nsrc * sizeof(struct in6_addr))) {
1069		err = EMSGSIZE;
1070		goto done;
1071	}
1072
1073	/*
1074	 * Do further input validation upfront to avoid resetting timers
1075	 * should we need to discard this query.
1076	 */
1077	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
1078		/*
1079		 * A general query with a source list has undefined
1080		 * behaviour; discard it.
1081		 */
1082		if (nsrc > 0) {
1083			err = EINVAL;
1084			goto done;
1085		}
1086		is_general_query = 1;
1087	} else {
1088		/*
1089		 * Embed scope ID of receiving interface in MLD query for
1090		 * lookup whilst we don't hold other locks (due to KAME
1091		 * locking lameness). We own this mbuf chain just now.
1092		 */
1093		in6_setscope(&mld->mld_addr, ifp, NULL);
1094	}
1095
1096	mli = MLD_IFINFO(ifp);
1097	VERIFY(mli != NULL);
1098
1099	MLI_LOCK(mli);
1100	/*
1101	 * Discard the v2 query if we're in Compatibility Mode.
1102	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
1103	 * until the Old Version Querier Present timer expires.
1104	 */
1105	if (mli->mli_version != MLD_VERSION_2) {
1106		MLI_UNLOCK(mli);
1107		goto done;
1108	}
1109
1110	mtp.qpt = mld_set_version(mli, MLD_VERSION_2);
1111	mli->mli_rv = qrv;
1112	mli->mli_qi = qqi;
1113	mli->mli_qri = MAX(timer, MLD_QRI_MIN);
1114
1115	MLD_PRINTF(("%s: qrv %d qi %d qri %d\n", __func__, mli->mli_rv,
1116	    mli->mli_qi, mli->mli_qri));
1117
1118	if (is_general_query) {
1119		/*
1120		 * MLDv2 General Query.
1121		 *
1122		 * Schedule a current-state report on this ifp for
1123		 * all groups, possibly containing source lists.
1124		 *
1125		 * If there is a pending General Query response
1126		 * scheduled earlier than the selected delay, do
1127		 * not schedule any other reports.
1128		 * Otherwise, reset the interface timer.
1129		 */
1130		MLD_PRINTF(("%s: process v2 general query on ifp 0x%llx(%s)\n",
1131		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1132		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
1133			mtp.it = mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
1134		}
1135		MLI_UNLOCK(mli);
1136	} else {
1137		MLI_UNLOCK(mli);
1138		/*
1139		 * MLDv2 Group-specific or Group-and-source-specific Query.
1140		 *
1141		 * Group-source-specific queries are throttled on
1142		 * a per-group basis to defeat denial-of-service attempts.
1143		 * Queries for groups we are not a member of on this
1144		 * link are simply ignored.
1145		 */
1146		in6_multihead_lock_shared();
1147		IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
1148		in6_multihead_lock_done();
1149		if (inm == NULL)
1150			goto done;
1151
1152		IN6M_LOCK(inm);
1153		if (nsrc > 0) {
1154			if (!ratecheck(&inm->in6m_lastgsrtv,
1155			    &mld_gsrdelay)) {
1156				MLD_PRINTF(("%s: GS query throttled.\n",
1157				    __func__));
1158				IN6M_UNLOCK(inm);
1159				IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1160				goto done;
1161			}
1162		}
1163		MLD_PRINTF(("%s: process v2 group query on ifp 0x%llx(%s)\n",
1164		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1165		/*
1166		 * If there is a pending General Query response
1167		 * scheduled sooner than the selected delay, no
1168		 * further report need be scheduled.
1169		 * Otherwise, prepare to respond to the
1170		 * group-specific or group-and-source query.
1171		 */
1172		MLI_LOCK(mli);
1173		mtp.it = mli->mli_v2_timer;
1174		MLI_UNLOCK(mli);
1175		if (mtp.it == 0 || mtp.it >= timer) {
1176			(void) mld_v2_process_group_query(inm, timer, m, off);
1177			mtp.cst = inm->in6m_timer;
1178		}
1179		IN6M_UNLOCK(inm);
1180		IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1181		/* XXX Clear embedded scope ID as userland won't expect it. */
1182		in6_clearscope(&mld->mld_addr);
1183	}
1184done:
1185	if (mtp.it > 0) {
1186		MLD_PRINTF(("%s: v2 general query response scheduled in "
1187		    "T+%d seconds on ifp 0x%llx(%s)\n", __func__, mtp.it,
1188		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1189	}
1190	mld_set_timeout(&mtp);
1191
1192	return (err);
1193}
1194
1195/*
1196 * Process a recieved MLDv2 group-specific or group-and-source-specific
1197 * query.
1198 * Return <0 if any error occured. Currently this is ignored.
1199 */
1200static int
1201mld_v2_process_group_query(struct in6_multi *inm, int timer, struct mbuf *m0,
1202    const int off)
1203{
1204	struct mldv2_query	*mld;
1205	int			 retval;
1206	uint16_t		 nsrc;
1207
1208	IN6M_LOCK_ASSERT_HELD(inm);
1209
1210	retval = 0;
1211	mld = (struct mldv2_query *)(mtod(m0, uint8_t *) + off);
1212
1213	switch (inm->in6m_state) {
1214	case MLD_NOT_MEMBER:
1215	case MLD_SILENT_MEMBER:
1216	case MLD_SLEEPING_MEMBER:
1217	case MLD_LAZY_MEMBER:
1218	case MLD_AWAKENING_MEMBER:
1219	case MLD_IDLE_MEMBER:
1220	case MLD_LEAVING_MEMBER:
1221		return (retval);
1222		break;
1223	case MLD_REPORTING_MEMBER:
1224	case MLD_G_QUERY_PENDING_MEMBER:
1225	case MLD_SG_QUERY_PENDING_MEMBER:
1226		break;
1227	}
1228
1229	nsrc = ntohs(mld->mld_numsrc);
1230
1231	/*
1232	 * Deal with group-specific queries upfront.
1233	 * If any group query is already pending, purge any recorded
1234	 * source-list state if it exists, and schedule a query response
1235	 * for this group-specific query.
1236	 */
1237	if (nsrc == 0) {
1238		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1239		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1240			in6m_clear_recorded(inm);
1241			timer = min(inm->in6m_timer, timer);
1242		}
1243		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1244		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1245		return (retval);
1246	}
1247
1248	/*
1249	 * Deal with the case where a group-and-source-specific query has
1250	 * been received but a group-specific query is already pending.
1251	 */
1252	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1253		timer = min(inm->in6m_timer, timer);
1254		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1255		return (retval);
1256	}
1257
1258	/*
1259	 * Finally, deal with the case where a group-and-source-specific
1260	 * query has been received, where a response to a previous g-s-r
1261	 * query exists, or none exists.
1262	 * In this case, we need to parse the source-list which the Querier
1263	 * has provided us with and check if we have any source list filter
1264	 * entries at T1 for these sources. If we do not, there is no need
1265	 * schedule a report and the query may be dropped.
1266	 * If we do, we must record them and schedule a current-state
1267	 * report for those sources.
1268	 */
1269	if (inm->in6m_nsrc > 0) {
1270		struct mbuf		*m;
1271		uint8_t			*sp;
1272		int			 i, nrecorded;
1273		int			 soff;
1274
1275		m = m0;
1276		soff = off + sizeof(struct mldv2_query);
1277		nrecorded = 0;
1278		for (i = 0; i < nsrc; i++) {
1279			sp = mtod(m, uint8_t *) + soff;
1280			retval = in6m_record_source(inm,
1281			    (const struct in6_addr *)(void *)sp);
1282			if (retval < 0)
1283				break;
1284			nrecorded += retval;
1285			soff += sizeof(struct in6_addr);
1286			if (soff >= m->m_len) {
1287				soff = soff - m->m_len;
1288				m = m->m_next;
1289				if (m == NULL)
1290					break;
1291			}
1292		}
1293		if (nrecorded > 0) {
1294			MLD_PRINTF(( "%s: schedule response to SG query\n",
1295			    __func__));
1296			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1297			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1298		}
1299	}
1300
1301	return (retval);
1302}
1303
1304/*
1305 * Process a received MLDv1 host membership report.
1306 * Assumes mld points to mld_hdr in pulled up mbuf chain.
1307 *
1308 * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1309 * mld_addr. This is OK as we own the mbuf chain.
1310 */
1311static int
1312mld_v1_input_report(struct ifnet *ifp, struct mbuf *m,
1313    const struct ip6_hdr *ip6, /*const*/ struct mld_hdr *mld)
1314{
1315	struct in6_addr		 src, dst;
1316	struct in6_ifaddr	*ia;
1317	struct in6_multi	*inm;
1318
1319	if (!mld_v1enable) {
1320		MLD_PRINTF(("%s: ignore v1 report %s on ifp 0x%llx(%s)\n",
1321		    __func__, ip6_sprintf(&mld->mld_addr),
1322		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1323		return (0);
1324	}
1325
1326	if ((ifp->if_flags & IFF_LOOPBACK) ||
1327	    (m->m_pkthdr.pkt_flags & PKTF_LOOP))
1328		return (0);
1329
1330	/*
1331	 * MLDv1 reports must originate from a host's link-local address,
1332	 * or the unspecified address (when booting).
1333	 */
1334	src = ip6->ip6_src;
1335	in6_clearscope(&src);
1336	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1337		MLD_PRINTF(("%s: ignore v1 query src %s on ifp 0x%llx(%s)\n",
1338		    __func__, ip6_sprintf(&ip6->ip6_src),
1339		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1340		return (EINVAL);
1341	}
1342
1343	/*
1344	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1345	 * group, and must be directed to the group itself.
1346	 */
1347	dst = ip6->ip6_dst;
1348	in6_clearscope(&dst);
1349	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1350	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1351		MLD_PRINTF(("%s: ignore v1 query dst %s on ifp 0x%llx(%s)\n",
1352		    __func__, ip6_sprintf(&ip6->ip6_dst),
1353		    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1354		return (EINVAL);
1355	}
1356
1357	/*
1358	 * Make sure we don't hear our own membership report, as fast
1359	 * leave requires knowing that we are the only member of a
1360	 * group. Assume we used the link-local address if available,
1361	 * otherwise look for ::.
1362	 *
1363	 * XXX Note that scope ID comparison is needed for the address
1364	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1365	 * performed for the on-wire address.
1366	 */
1367	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1368	if (ia != NULL) {
1369		IFA_LOCK(&ia->ia_ifa);
1370		if ((IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia)))){
1371			IFA_UNLOCK(&ia->ia_ifa);
1372			IFA_REMREF(&ia->ia_ifa);
1373			return (0);
1374		}
1375		IFA_UNLOCK(&ia->ia_ifa);
1376		IFA_REMREF(&ia->ia_ifa);
1377	} else if (IN6_IS_ADDR_UNSPECIFIED(&src)) {
1378		return (0);
1379	}
1380
1381	MLD_PRINTF(("%s: process v1 report %s on ifp 0x%llx(%s)\n",
1382	    __func__, ip6_sprintf(&mld->mld_addr),
1383	    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1384
1385	/*
1386	 * Embed scope ID of receiving interface in MLD query for lookup
1387	 * whilst we don't hold other locks (due to KAME locking lameness).
1388	 */
1389	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1390		in6_setscope(&mld->mld_addr, ifp, NULL);
1391
1392	/*
1393	 * MLDv1 report suppression.
1394	 * If we are a member of this group, and our membership should be
1395	 * reported, and our group timer is pending or about to be reset,
1396	 * stop our group timer by transitioning to the 'lazy' state.
1397	 */
1398	in6_multihead_lock_shared();
1399	IN6_LOOKUP_MULTI(&mld->mld_addr, ifp, inm);
1400	in6_multihead_lock_done();
1401
1402	if (inm != NULL) {
1403		struct mld_ifinfo *mli;
1404
1405		IN6M_LOCK(inm);
1406		mli = inm->in6m_mli;
1407		VERIFY(mli != NULL);
1408
1409		MLI_LOCK(mli);
1410		/*
1411		 * If we are in MLDv2 host mode, do not allow the
1412		 * other host's MLDv1 report to suppress our reports.
1413		 */
1414		if (mli->mli_version == MLD_VERSION_2) {
1415			MLI_UNLOCK(mli);
1416			IN6M_UNLOCK(inm);
1417			IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1418			goto out;
1419		}
1420		MLI_UNLOCK(mli);
1421
1422		inm->in6m_timer = 0;
1423
1424		switch (inm->in6m_state) {
1425		case MLD_NOT_MEMBER:
1426		case MLD_SILENT_MEMBER:
1427		case MLD_SLEEPING_MEMBER:
1428			break;
1429		case MLD_REPORTING_MEMBER:
1430		case MLD_IDLE_MEMBER:
1431		case MLD_AWAKENING_MEMBER:
1432			MLD_PRINTF(("%s: report suppressed for %s on "
1433			    "ifp 0x%llx(%s)\n", __func__,
1434			    ip6_sprintf(&mld->mld_addr),
1435			    (uint64_t)VM_KERNEL_ADDRPERM(ifp), if_name(ifp)));
1436		case MLD_LAZY_MEMBER:
1437			inm->in6m_state = MLD_LAZY_MEMBER;
1438			break;
1439		case MLD_G_QUERY_PENDING_MEMBER:
1440		case MLD_SG_QUERY_PENDING_MEMBER:
1441		case MLD_LEAVING_MEMBER:
1442			break;
1443		}
1444		IN6M_UNLOCK(inm);
1445		IN6M_REMREF(inm); /* from IN6_LOOKUP_MULTI */
1446	}
1447
1448out:
1449	/* XXX Clear embedded scope ID as userland won't expect it. */
1450	in6_clearscope(&mld->mld_addr);
1451
1452	return (0);
1453}
1454
1455/*
1456 * MLD input path.
1457 *
1458 * Assume query messages which fit in a single ICMPv6 message header
1459 * have been pulled up.
1460 * Assume that userland will want to see the message, even if it
1461 * otherwise fails kernel input validation; do not free it.
1462 * Pullup may however free the mbuf chain m if it fails.
1463 *
1464 * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1465 */
1466int
1467mld_input(struct mbuf *m, int off, int icmp6len)
1468{
1469	struct ifnet	*ifp;
1470	struct ip6_hdr	*ip6;
1471	struct mld_hdr	*mld;
1472	int		 mldlen;
1473
1474	MLD_PRINTF(("%s: called w/mbuf (0x%llx,%d)\n", __func__,
1475	    (uint64_t)VM_KERNEL_ADDRPERM(m), off));
1476
1477	ifp = m->m_pkthdr.rcvif;
1478
1479	ip6 = mtod(m, struct ip6_hdr *);
1480
1481	/* Pullup to appropriate size. */
1482	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1483	if (mld->mld_type == MLD_LISTENER_QUERY &&
1484	    icmp6len >= sizeof(struct mldv2_query)) {
1485		mldlen = sizeof(struct mldv2_query);
1486	} else {
1487		mldlen = sizeof(struct mld_hdr);
1488	}
1489	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1490	if (mld == NULL) {
1491		icmp6stat.icp6s_badlen++;
1492		return (IPPROTO_DONE);
1493	}
1494
1495	/*
1496	 * Userland needs to see all of this traffic for implementing
1497	 * the endpoint discovery portion of multicast routing.
1498	 */
1499	switch (mld->mld_type) {
1500	case MLD_LISTENER_QUERY:
1501		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1502		if (icmp6len == sizeof(struct mld_hdr)) {
1503			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1504				return (0);
1505		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1506			if (mld_v2_input_query(ifp, ip6, m, off,
1507			    icmp6len) != 0)
1508				return (0);
1509		}
1510		break;
1511	case MLD_LISTENER_REPORT:
1512		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1513		if (mld_v1_input_report(ifp, m, ip6, mld) != 0)
1514			return (0);
1515		break;
1516	case MLDV2_LISTENER_REPORT:
1517		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1518		break;
1519	case MLD_LISTENER_DONE:
1520		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1521		break;
1522	default:
1523		break;
1524	}
1525
1526	return (0);
1527}
1528
1529/*
1530 * Schedule MLD timer based on various parameters; caller must ensure that
1531 * lock ordering is maintained as this routine acquires MLD global lock.
1532 */
1533void
1534mld_set_timeout(struct mld_tparams *mtp)
1535{
1536	MLD_LOCK_ASSERT_NOTHELD();
1537	VERIFY(mtp != NULL);
1538
1539	if (mtp->qpt != 0 || mtp->it != 0 || mtp->cst != 0 || mtp->sct != 0) {
1540		MLD_LOCK();
1541		if (mtp->qpt != 0)
1542			querier_present_timers_running6 = 1;
1543		if (mtp->it != 0)
1544			interface_timers_running6 = 1;
1545		if (mtp->cst != 0)
1546			current_state_timers_running6 = 1;
1547		if (mtp->sct != 0)
1548			state_change_timers_running6 = 1;
1549		mld_sched_timeout();
1550		MLD_UNLOCK();
1551	}
1552}
1553
1554/*
1555 * MLD6 timer handler (per 1 second).
1556 */
1557static void
1558mld_timeout(void *arg)
1559{
1560#pragma unused(arg)
1561	struct ifqueue		 scq;	/* State-change packets */
1562	struct ifqueue		 qrq;	/* Query response packets */
1563	struct ifnet		*ifp;
1564	struct mld_ifinfo	*mli;
1565	struct in6_multi	*inm;
1566	int			 uri_sec = 0;
1567	SLIST_HEAD(, in6_multi)	in6m_dthead;
1568
1569	SLIST_INIT(&in6m_dthead);
1570
1571	/*
1572	 * Update coarse-grained networking timestamp (in sec.); the idea
1573	 * is to piggy-back on the timeout callout to update the counter
1574	 * returnable via net_uptime().
1575	 */
1576	net_update_uptime();
1577
1578	MLD_LOCK();
1579
1580	MLD_PRINTF(("%s: qpt %d, it %d, cst %d, sct %d\n", __func__,
1581	    querier_present_timers_running6, interface_timers_running6,
1582	    current_state_timers_running6, state_change_timers_running6));
1583
1584	/*
1585	 * MLDv1 querier present timer processing.
1586	 */
1587	if (querier_present_timers_running6) {
1588		querier_present_timers_running6 = 0;
1589		LIST_FOREACH(mli, &mli_head, mli_link) {
1590			MLI_LOCK(mli);
1591			mld_v1_process_querier_timers(mli);
1592			if (mli->mli_v1_timer > 0)
1593				querier_present_timers_running6 = 1;
1594			MLI_UNLOCK(mli);
1595		}
1596	}
1597
1598	/*
1599	 * MLDv2 General Query response timer processing.
1600	 */
1601	if (interface_timers_running6) {
1602		MLD_PRINTF(("%s: interface timers running\n", __func__));
1603		interface_timers_running6 = 0;
1604		LIST_FOREACH(mli, &mli_head, mli_link) {
1605			MLI_LOCK(mli);
1606			if (mli->mli_version != MLD_VERSION_2) {
1607				MLI_UNLOCK(mli);
1608				continue;
1609			}
1610			if (mli->mli_v2_timer == 0) {
1611				/* Do nothing. */
1612			} else if (--mli->mli_v2_timer == 0) {
1613				if (mld_v2_dispatch_general_query(mli) > 0)
1614					interface_timers_running6 = 1;
1615			} else {
1616				interface_timers_running6 = 1;
1617			}
1618			MLI_UNLOCK(mli);
1619		}
1620	}
1621
1622	if (!current_state_timers_running6 &&
1623	    !state_change_timers_running6)
1624		goto out_locked;
1625
1626	current_state_timers_running6 = 0;
1627	state_change_timers_running6 = 0;
1628
1629	MLD_PRINTF(("%s: state change timers running\n", __func__));
1630
1631	memset(&qrq, 0, sizeof(struct ifqueue));
1632	qrq.ifq_maxlen = MLD_MAX_G_GS_PACKETS;
1633
1634	memset(&scq, 0, sizeof(struct ifqueue));
1635	scq.ifq_maxlen = MLD_MAX_STATE_CHANGE_PACKETS;
1636
1637	/*
1638	 * MLD host report and state-change timer processing.
1639	 * Note: Processing a v2 group timer may remove a node.
1640	 */
1641	LIST_FOREACH(mli, &mli_head, mli_link) {
1642		struct in6_multistep step;
1643
1644		MLI_LOCK(mli);
1645		ifp = mli->mli_ifp;
1646		uri_sec = MLD_RANDOM_DELAY(mli->mli_uri);
1647		MLI_UNLOCK(mli);
1648
1649		in6_multihead_lock_shared();
1650		IN6_FIRST_MULTI(step, inm);
1651		while (inm != NULL) {
1652			IN6M_LOCK(inm);
1653			if (inm->in6m_ifp != ifp)
1654				goto next;
1655
1656			MLI_LOCK(mli);
1657			switch (mli->mli_version) {
1658			case MLD_VERSION_1:
1659				mld_v1_process_group_timer(inm,
1660				    mli->mli_version);
1661				break;
1662			case MLD_VERSION_2:
1663				mld_v2_process_group_timers(mli, &qrq,
1664				    &scq, inm, uri_sec);
1665				break;
1666			}
1667			MLI_UNLOCK(mli);
1668next:
1669			IN6M_UNLOCK(inm);
1670			IN6_NEXT_MULTI(step, inm);
1671		}
1672		in6_multihead_lock_done();
1673
1674		MLI_LOCK(mli);
1675		if (mli->mli_version == MLD_VERSION_1) {
1676			mld_dispatch_queue(mli, &mli->mli_v1q, 0);
1677		} else if (mli->mli_version == MLD_VERSION_2) {
1678			MLI_UNLOCK(mli);
1679			mld_dispatch_queue(NULL, &qrq, 0);
1680			mld_dispatch_queue(NULL, &scq, 0);
1681			VERIFY(qrq.ifq_len == 0);
1682			VERIFY(scq.ifq_len == 0);
1683			MLI_LOCK(mli);
1684		}
1685		/*
1686		 * In case there are still any pending membership reports
1687		 * which didn't get drained at version change time.
1688		 */
1689		IF_DRAIN(&mli->mli_v1q);
1690		/*
1691		 * Release all deferred inm records, and drain any locally
1692		 * enqueued packets; do it even if the current MLD version
1693		 * for the link is no longer MLDv2, in order to handle the
1694		 * version change case.
1695		 */
1696		mld_flush_relq(mli, (struct mld_in6m_relhead *)&in6m_dthead);
1697		VERIFY(SLIST_EMPTY(&mli->mli_relinmhead));
1698		MLI_UNLOCK(mli);
1699
1700		IF_DRAIN(&qrq);
1701		IF_DRAIN(&scq);
1702	}
1703
1704out_locked:
1705	/* re-arm the timer if there's work to do */
1706	mld_timeout_run = 0;
1707	mld_sched_timeout();
1708	MLD_UNLOCK();
1709
1710	/* Now that we're dropped all locks, release detached records */
1711	MLD_REMOVE_DETACHED_IN6M(&in6m_dthead);
1712}
1713
1714static void
1715mld_sched_timeout(void)
1716{
1717	MLD_LOCK_ASSERT_HELD();
1718
1719	if (!mld_timeout_run &&
1720	    (querier_present_timers_running6 || current_state_timers_running6 ||
1721	    interface_timers_running6 || state_change_timers_running6)) {
1722		mld_timeout_run = 1;
1723		timeout(mld_timeout, NULL, hz);
1724	}
1725}
1726
1727/*
1728 * Free the in6_multi reference(s) for this MLD lifecycle.
1729 *
1730 * Caller must be holding mli_lock.
1731 */
1732static void
1733mld_flush_relq(struct mld_ifinfo *mli, struct mld_in6m_relhead *in6m_dthead)
1734{
1735	struct in6_multi *inm;
1736
1737again:
1738	MLI_LOCK_ASSERT_HELD(mli);
1739	inm = SLIST_FIRST(&mli->mli_relinmhead);
1740	if (inm != NULL) {
1741		int lastref;
1742
1743		SLIST_REMOVE_HEAD(&mli->mli_relinmhead, in6m_nrele);
1744		MLI_UNLOCK(mli);
1745
1746		in6_multihead_lock_exclusive();
1747		IN6M_LOCK(inm);
1748		VERIFY(inm->in6m_nrelecnt != 0);
1749		inm->in6m_nrelecnt--;
1750		lastref = in6_multi_detach(inm);
1751		VERIFY(!lastref || (!(inm->in6m_debug & IFD_ATTACHED) &&
1752		    inm->in6m_reqcnt == 0));
1753		IN6M_UNLOCK(inm);
1754		in6_multihead_lock_done();
1755		/* from mli_relinmhead */
1756		IN6M_REMREF(inm);
1757		/* from in6_multihead_list */
1758		if (lastref) {
1759			/*
1760			 * Defer releasing our final reference, as we
1761			 * are holding the MLD lock at this point, and
1762			 * we could end up with locking issues later on
1763			 * (while issuing SIOCDELMULTI) when this is the
1764			 * final reference count.  Let the caller do it
1765			 * when it is safe.
1766			 */
1767			MLD_ADD_DETACHED_IN6M(in6m_dthead, inm);
1768		}
1769		MLI_LOCK(mli);
1770		goto again;
1771	}
1772}
1773
1774/*
1775 * Update host report group timer.
1776 * Will update the global pending timer flags.
1777 */
1778static void
1779mld_v1_process_group_timer(struct in6_multi *inm, const int mld_version)
1780{
1781#pragma unused(mld_version)
1782	int report_timer_expired;
1783
1784	MLD_LOCK_ASSERT_HELD();
1785	IN6M_LOCK_ASSERT_HELD(inm);
1786	MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
1787
1788	if (inm->in6m_timer == 0) {
1789		report_timer_expired = 0;
1790	} else if (--inm->in6m_timer == 0) {
1791		report_timer_expired = 1;
1792	} else {
1793		current_state_timers_running6 = 1;
1794		/* caller will schedule timer */
1795		return;
1796	}
1797
1798	switch (inm->in6m_state) {
1799	case MLD_NOT_MEMBER:
1800	case MLD_SILENT_MEMBER:
1801	case MLD_IDLE_MEMBER:
1802	case MLD_LAZY_MEMBER:
1803	case MLD_SLEEPING_MEMBER:
1804	case MLD_AWAKENING_MEMBER:
1805		break;
1806	case MLD_REPORTING_MEMBER:
1807		if (report_timer_expired) {
1808			inm->in6m_state = MLD_IDLE_MEMBER;
1809			(void) mld_v1_transmit_report(inm,
1810			     MLD_LISTENER_REPORT);
1811			IN6M_LOCK_ASSERT_HELD(inm);
1812			MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
1813		}
1814		break;
1815	case MLD_G_QUERY_PENDING_MEMBER:
1816	case MLD_SG_QUERY_PENDING_MEMBER:
1817	case MLD_LEAVING_MEMBER:
1818		break;
1819	}
1820}
1821
1822/*
1823 * Update a group's timers for MLDv2.
1824 * Will update the global pending timer flags.
1825 * Note: Unlocked read from mli.
1826 */
1827static void
1828mld_v2_process_group_timers(struct mld_ifinfo *mli,
1829    struct ifqueue *qrq, struct ifqueue *scq,
1830    struct in6_multi *inm, const int uri_sec)
1831{
1832	int query_response_timer_expired;
1833	int state_change_retransmit_timer_expired;
1834
1835	MLD_LOCK_ASSERT_HELD();
1836	IN6M_LOCK_ASSERT_HELD(inm);
1837	MLI_LOCK_ASSERT_HELD(mli);
1838	VERIFY(mli == inm->in6m_mli);
1839
1840	query_response_timer_expired = 0;
1841	state_change_retransmit_timer_expired = 0;
1842
1843	/*
1844	 * During a transition from compatibility mode back to MLDv2,
1845	 * a group record in REPORTING state may still have its group
1846	 * timer active. This is a no-op in this function; it is easier
1847	 * to deal with it here than to complicate the timeout path.
1848	 */
1849	if (inm->in6m_timer == 0) {
1850		query_response_timer_expired = 0;
1851	} else if (--inm->in6m_timer == 0) {
1852		query_response_timer_expired = 1;
1853	} else {
1854		current_state_timers_running6 = 1;
1855		/* caller will schedule timer */
1856	}
1857
1858	if (inm->in6m_sctimer == 0) {
1859		state_change_retransmit_timer_expired = 0;
1860	} else if (--inm->in6m_sctimer == 0) {
1861		state_change_retransmit_timer_expired = 1;
1862	} else {
1863		state_change_timers_running6 = 1;
1864		/* caller will schedule timer */
1865	}
1866
1867	/* We are in timer callback, so be quick about it. */
1868	if (!state_change_retransmit_timer_expired &&
1869	    !query_response_timer_expired)
1870		return;
1871
1872	switch (inm->in6m_state) {
1873	case MLD_NOT_MEMBER:
1874	case MLD_SILENT_MEMBER:
1875	case MLD_SLEEPING_MEMBER:
1876	case MLD_LAZY_MEMBER:
1877	case MLD_AWAKENING_MEMBER:
1878	case MLD_IDLE_MEMBER:
1879		break;
1880	case MLD_G_QUERY_PENDING_MEMBER:
1881	case MLD_SG_QUERY_PENDING_MEMBER:
1882		/*
1883		 * Respond to a previously pending Group-Specific
1884		 * or Group-and-Source-Specific query by enqueueing
1885		 * the appropriate Current-State report for
1886		 * immediate transmission.
1887		 */
1888		if (query_response_timer_expired) {
1889			int retval;
1890
1891			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1892			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1893			    0);
1894			MLD_PRINTF(("%s: enqueue record = %d\n",
1895			    __func__, retval));
1896			inm->in6m_state = MLD_REPORTING_MEMBER;
1897			in6m_clear_recorded(inm);
1898		}
1899		/* FALLTHROUGH */
1900	case MLD_REPORTING_MEMBER:
1901	case MLD_LEAVING_MEMBER:
1902		if (state_change_retransmit_timer_expired) {
1903			/*
1904			 * State-change retransmission timer fired.
1905			 * If there are any further pending retransmissions,
1906			 * set the global pending state-change flag, and
1907			 * reset the timer.
1908			 */
1909			if (--inm->in6m_scrv > 0) {
1910				inm->in6m_sctimer = uri_sec;
1911				state_change_timers_running6 = 1;
1912				/* caller will schedule timer */
1913			}
1914			/*
1915			 * Retransmit the previously computed state-change
1916			 * report. If there are no further pending
1917			 * retransmissions, the mbuf queue will be consumed.
1918			 * Update T0 state to T1 as we have now sent
1919			 * a state-change.
1920			 */
1921			(void) mld_v2_merge_state_changes(inm, scq);
1922
1923			in6m_commit(inm);
1924			MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
1925			    ip6_sprintf(&inm->in6m_addr),
1926			    if_name(inm->in6m_ifp)));
1927
1928			/*
1929			 * If we are leaving the group for good, make sure
1930			 * we release MLD's reference to it.
1931			 * This release must be deferred using a SLIST,
1932			 * as we are called from a loop which traverses
1933			 * the in_ifmultiaddr TAILQ.
1934			 */
1935			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1936			    inm->in6m_scrv == 0) {
1937				inm->in6m_state = MLD_NOT_MEMBER;
1938				/*
1939				 * A reference has already been held in
1940				 * mld_final_leave() for this inm, so
1941				 * no need to hold another one.  We also
1942				 * bumped up its request count then, so
1943				 * that it stays in in6_multihead.  Both
1944				 * of them will be released when it is
1945				 * dequeued later on.
1946				 */
1947				VERIFY(inm->in6m_nrelecnt != 0);
1948				SLIST_INSERT_HEAD(&mli->mli_relinmhead,
1949				    inm, in6m_nrele);
1950			}
1951		}
1952		break;
1953	}
1954}
1955
1956/*
1957 * Switch to a different version on the given interface,
1958 * as per Section 9.12.
1959 */
1960static uint32_t
1961mld_set_version(struct mld_ifinfo *mli, const int mld_version)
1962{
1963	int old_version_timer;
1964
1965	MLI_LOCK_ASSERT_HELD(mli);
1966
1967	MLD_PRINTF(("%s: switching to v%d on ifp 0x%llx(%s)\n", __func__,
1968	    mld_version, (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp),
1969	    if_name(mli->mli_ifp)));
1970
1971	if (mld_version == MLD_VERSION_1) {
1972		/*
1973		 * Compute the "Older Version Querier Present" timer as per
1974		 * Section 9.12, in seconds.
1975		 */
1976		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1977		mli->mli_v1_timer = old_version_timer;
1978	}
1979
1980	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1981		mli->mli_version = MLD_VERSION_1;
1982		mld_v2_cancel_link_timers(mli);
1983	}
1984
1985	MLI_LOCK_ASSERT_HELD(mli);
1986
1987	return (mli->mli_v1_timer);
1988}
1989
1990/*
1991 * Cancel pending MLDv2 timers for the given link and all groups
1992 * joined on it; state-change, general-query, and group-query timers.
1993 *
1994 * Only ever called on a transition from v2 to Compatibility mode. Kill
1995 * the timers stone dead (this may be expensive for large N groups), they
1996 * will be restarted if Compatibility Mode deems that they must be due to
1997 * query processing.
1998 */
1999static void
2000mld_v2_cancel_link_timers(struct mld_ifinfo *mli)
2001{
2002	struct ifnet		*ifp;
2003	struct in6_multi	*inm;
2004	struct in6_multistep	step;
2005
2006	MLI_LOCK_ASSERT_HELD(mli);
2007
2008	MLD_PRINTF(("%s: cancel v2 timers on ifp 0x%llx(%s)\n", __func__,
2009	    (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp), if_name(mli->mli_ifp)));
2010
2011	/*
2012	 * Stop the v2 General Query Response on this link stone dead.
2013	 * If timer is woken up due to interface_timers_running6,
2014	 * the flag will be cleared if there are no pending link timers.
2015	 */
2016	mli->mli_v2_timer = 0;
2017
2018	/*
2019	 * Now clear the current-state and state-change report timers
2020	 * for all memberships scoped to this link.
2021	 */
2022	ifp = mli->mli_ifp;
2023	MLI_UNLOCK(mli);
2024
2025	in6_multihead_lock_shared();
2026	IN6_FIRST_MULTI(step, inm);
2027	while (inm != NULL) {
2028		IN6M_LOCK(inm);
2029		if (inm->in6m_ifp != ifp)
2030			goto next;
2031
2032		switch (inm->in6m_state) {
2033		case MLD_NOT_MEMBER:
2034		case MLD_SILENT_MEMBER:
2035		case MLD_IDLE_MEMBER:
2036		case MLD_LAZY_MEMBER:
2037		case MLD_SLEEPING_MEMBER:
2038		case MLD_AWAKENING_MEMBER:
2039			/*
2040			 * These states are either not relevant in v2 mode,
2041			 * or are unreported. Do nothing.
2042			 */
2043			break;
2044		case MLD_LEAVING_MEMBER:
2045			/*
2046			 * If we are leaving the group and switching
2047			 * version, we need to release the final
2048			 * reference held for issuing the INCLUDE {}.
2049			 * During mld_final_leave(), we bumped up both the
2050			 * request and reference counts.  Since we cannot
2051			 * call in6_multi_detach() here, defer this task to
2052			 * the timer routine.
2053			 */
2054			VERIFY(inm->in6m_nrelecnt != 0);
2055			MLI_LOCK(mli);
2056			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
2057			    in6m_nrele);
2058			MLI_UNLOCK(mli);
2059			/* FALLTHROUGH */
2060		case MLD_G_QUERY_PENDING_MEMBER:
2061		case MLD_SG_QUERY_PENDING_MEMBER:
2062			in6m_clear_recorded(inm);
2063			/* FALLTHROUGH */
2064		case MLD_REPORTING_MEMBER:
2065			inm->in6m_state = MLD_REPORTING_MEMBER;
2066			break;
2067		}
2068		/*
2069		 * Always clear state-change and group report timers.
2070		 * Free any pending MLDv2 state-change records.
2071		 */
2072		inm->in6m_sctimer = 0;
2073		inm->in6m_timer = 0;
2074		IF_DRAIN(&inm->in6m_scq);
2075next:
2076		IN6M_UNLOCK(inm);
2077		IN6_NEXT_MULTI(step, inm);
2078	}
2079	in6_multihead_lock_done();
2080
2081	MLI_LOCK(mli);
2082}
2083
2084/*
2085 * Update the Older Version Querier Present timers for a link.
2086 * See Section 9.12 of RFC 3810.
2087 */
2088static void
2089mld_v1_process_querier_timers(struct mld_ifinfo *mli)
2090{
2091	MLI_LOCK_ASSERT_HELD(mli);
2092
2093	if (mld_v2enable && mli->mli_version != MLD_VERSION_2 &&
2094	    --mli->mli_v1_timer == 0) {
2095		/*
2096		 * MLDv1 Querier Present timer expired; revert to MLDv2.
2097		 */
2098		MLD_PRINTF(("%s: transition from v%d -> v%d on 0x%llx(%s)\n",
2099		    __func__, mli->mli_version, MLD_VERSION_2,
2100		    (uint64_t)VM_KERNEL_ADDRPERM(mli->mli_ifp),
2101		    if_name(mli->mli_ifp)));
2102		mli->mli_version = MLD_VERSION_2;
2103	}
2104}
2105
2106/*
2107 * Transmit an MLDv1 report immediately.
2108 */
2109static int
2110mld_v1_transmit_report(struct in6_multi *in6m, const int type)
2111{
2112	struct ifnet		*ifp;
2113	struct in6_ifaddr	*ia;
2114	struct ip6_hdr		*ip6;
2115	struct mbuf		*mh, *md;
2116	struct mld_hdr		*mld;
2117	int			error = 0;
2118
2119	IN6M_LOCK_ASSERT_HELD(in6m);
2120	MLI_LOCK_ASSERT_HELD(in6m->in6m_mli);
2121
2122	ifp = in6m->in6m_ifp;
2123	/* ia may be NULL if link-local address is tentative. */
2124	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
2125
2126	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
2127	if (mh == NULL) {
2128		if (ia != NULL)
2129			IFA_REMREF(&ia->ia_ifa);
2130		return (ENOMEM);
2131	}
2132	MGET(md, M_DONTWAIT, MT_DATA);
2133	if (md == NULL) {
2134		m_free(mh);
2135		if (ia != NULL)
2136			IFA_REMREF(&ia->ia_ifa);
2137		return (ENOMEM);
2138	}
2139	mh->m_next = md;
2140
2141	/*
2142	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
2143	 * that ether_output() does not need to allocate another mbuf
2144	 * for the header in the most common case.
2145	 */
2146	MH_ALIGN(mh, sizeof(struct ip6_hdr));
2147	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
2148	mh->m_len = sizeof(struct ip6_hdr);
2149
2150	ip6 = mtod(mh, struct ip6_hdr *);
2151	ip6->ip6_flow = 0;
2152	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
2153	ip6->ip6_vfc |= IPV6_VERSION;
2154	ip6->ip6_nxt = IPPROTO_ICMPV6;
2155	if (ia != NULL)
2156		IFA_LOCK(&ia->ia_ifa);
2157	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
2158	if (ia != NULL) {
2159		IFA_UNLOCK(&ia->ia_ifa);
2160		IFA_REMREF(&ia->ia_ifa);
2161		ia = NULL;
2162	}
2163	ip6->ip6_dst = in6m->in6m_addr;
2164
2165	md->m_len = sizeof(struct mld_hdr);
2166	mld = mtod(md, struct mld_hdr *);
2167	mld->mld_type = type;
2168	mld->mld_code = 0;
2169	mld->mld_cksum = 0;
2170	mld->mld_maxdelay = 0;
2171	mld->mld_reserved = 0;
2172	mld->mld_addr = in6m->in6m_addr;
2173	in6_clearscope(&mld->mld_addr);
2174	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
2175	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
2176
2177	mld_save_context(mh, ifp);
2178	mh->m_flags |= M_MLDV1;
2179
2180	/*
2181	 * Due to the fact that at this point we are possibly holding
2182	 * in6_multihead_lock in shared or exclusive mode, we can't call
2183	 * mld_dispatch_packet() here since that will eventually call
2184	 * ip6_output(), which will try to lock in6_multihead_lock and cause
2185	 * a deadlock.
2186	 * Instead we defer the work to the mld_timeout() thread, thus
2187	 * avoiding unlocking in_multihead_lock here.
2188	 */
2189        if (IF_QFULL(&in6m->in6m_mli->mli_v1q)) {
2190                MLD_PRINTF(("%s: v1 outbound queue full\n", __func__));
2191                error = ENOMEM;
2192                m_freem(mh);
2193        } else {
2194                IF_ENQUEUE(&in6m->in6m_mli->mli_v1q, mh);
2195		VERIFY(error == 0);
2196	}
2197
2198	return (error);
2199}
2200
2201/*
2202 * Process a state change from the upper layer for the given IPv6 group.
2203 *
2204 * Each socket holds a reference on the in6_multi in its own ip_moptions.
2205 * The socket layer will have made the necessary updates to.the group
2206 * state, it is now up to MLD to issue a state change report if there
2207 * has been any change between T0 (when the last state-change was issued)
2208 * and T1 (now).
2209 *
2210 * We use the MLDv2 state machine at group level. The MLd module
2211 * however makes the decision as to which MLD protocol version to speak.
2212 * A state change *from* INCLUDE {} always means an initial join.
2213 * A state change *to* INCLUDE {} always means a final leave.
2214 *
2215 * If delay is non-zero, and the state change is an initial multicast
2216 * join, the state change report will be delayed by 'delay' ticks
2217 * in units of seconds if MLDv1 is active on the link; otherwise
2218 * the initial MLDv2 state change report will be delayed by whichever
2219 * is sooner, a pending state-change timer or delay itself.
2220 */
2221int
2222mld_change_state(struct in6_multi *inm, struct mld_tparams *mtp,
2223    const int delay)
2224{
2225	struct mld_ifinfo *mli;
2226	struct ifnet *ifp;
2227	int error = 0;
2228
2229	VERIFY(mtp != NULL);
2230	bzero(mtp, sizeof (*mtp));
2231
2232	IN6M_LOCK_ASSERT_HELD(inm);
2233	VERIFY(inm->in6m_mli != NULL);
2234	MLI_LOCK_ASSERT_NOTHELD(inm->in6m_mli);
2235
2236	/*
2237	 * Try to detect if the upper layer just asked us to change state
2238	 * for an interface which has now gone away.
2239	 */
2240	VERIFY(inm->in6m_ifma != NULL);
2241	ifp = inm->in6m_ifma->ifma_ifp;
2242	/*
2243	 * Sanity check that netinet6's notion of ifp is the same as net's.
2244	 */
2245	VERIFY(inm->in6m_ifp == ifp);
2246
2247	mli = MLD_IFINFO(ifp);
2248	VERIFY(mli != NULL);
2249
2250	/*
2251	 * If we detect a state transition to or from MCAST_UNDEFINED
2252	 * for this group, then we are starting or finishing an MLD
2253	 * life cycle for this group.
2254	 */
2255	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
2256		MLD_PRINTF(("%s: inm transition %d -> %d\n", __func__,
2257		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode));
2258		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
2259			MLD_PRINTF(("%s: initial join\n", __func__));
2260			error = mld_initial_join(inm, mli, mtp, delay);
2261			goto out;
2262		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
2263			MLD_PRINTF(("%s: final leave\n", __func__));
2264			mld_final_leave(inm, mli, mtp);
2265			goto out;
2266		}
2267	} else {
2268		MLD_PRINTF(("%s: filter set change\n", __func__));
2269	}
2270
2271	error = mld_handle_state_change(inm, mli, mtp);
2272out:
2273	return (error);
2274}
2275
2276/*
2277 * Perform the initial join for an MLD group.
2278 *
2279 * When joining a group:
2280 *  If the group should have its MLD traffic suppressed, do nothing.
2281 *  MLDv1 starts sending MLDv1 host membership reports.
2282 *  MLDv2 will schedule an MLDv2 state-change report containing the
2283 *  initial state of the membership.
2284 *
2285 * If the delay argument is non-zero, then we must delay sending the
2286 * initial state change for delay ticks (in units of seconds).
2287 */
2288static int
2289mld_initial_join(struct in6_multi *inm, struct mld_ifinfo *mli,
2290    struct mld_tparams *mtp, const int delay)
2291{
2292	struct ifnet		*ifp;
2293	struct ifqueue		*ifq;
2294	int			 error, retval, syncstates;
2295	int			 odelay;
2296
2297	IN6M_LOCK_ASSERT_HELD(inm);
2298	MLI_LOCK_ASSERT_NOTHELD(mli);
2299	VERIFY(mtp != NULL);
2300
2301	MLD_PRINTF(("%s: initial join %s on ifp 0x%llx(%s)\n",
2302	    __func__, ip6_sprintf(&inm->in6m_addr),
2303	    (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2304	    if_name(inm->in6m_ifp)));
2305
2306	error = 0;
2307	syncstates = 1;
2308
2309	ifp = inm->in6m_ifp;
2310
2311	MLI_LOCK(mli);
2312	VERIFY(mli->mli_ifp == ifp);
2313
2314	/*
2315	 * Groups joined on loopback or marked as 'not reported',
2316	 * enter the MLD_SILENT_MEMBER state and
2317	 * are never reported in any protocol exchanges.
2318	 * All other groups enter the appropriate state machine
2319	 * for the version in use on this link.
2320	 * A link marked as MLIF_SILENT causes MLD to be completely
2321	 * disabled for the link.
2322	 */
2323	if ((ifp->if_flags & IFF_LOOPBACK) ||
2324	    (mli->mli_flags & MLIF_SILENT) ||
2325	    !mld_is_addr_reported(&inm->in6m_addr)) {
2326		MLD_PRINTF(("%s: not kicking state machine for silent group\n",
2327		    __func__));
2328		inm->in6m_state = MLD_SILENT_MEMBER;
2329		inm->in6m_timer = 0;
2330	} else {
2331		/*
2332		 * Deal with overlapping in6_multi lifecycle.
2333		 * If this group was LEAVING, then make sure
2334		 * we drop the reference we picked up to keep the
2335		 * group around for the final INCLUDE {} enqueue.
2336		 * Since we cannot call in6_multi_detach() here,
2337		 * defer this task to the timer routine.
2338		 */
2339		if (mli->mli_version == MLD_VERSION_2 &&
2340		    inm->in6m_state == MLD_LEAVING_MEMBER) {
2341			VERIFY(inm->in6m_nrelecnt != 0);
2342			SLIST_INSERT_HEAD(&mli->mli_relinmhead, inm,
2343			    in6m_nrele);
2344		}
2345
2346		inm->in6m_state = MLD_REPORTING_MEMBER;
2347
2348		switch (mli->mli_version) {
2349		case MLD_VERSION_1:
2350			/*
2351			 * If a delay was provided, only use it if
2352			 * it is greater than the delay normally
2353			 * used for an MLDv1 state change report,
2354			 * and delay sending the initial MLDv1 report
2355			 * by not transitioning to the IDLE state.
2356			 */
2357			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI);
2358			if (delay) {
2359				inm->in6m_timer = max(delay, odelay);
2360				mtp->cst = 1;
2361			} else {
2362				inm->in6m_state = MLD_IDLE_MEMBER;
2363				error = mld_v1_transmit_report(inm,
2364				     MLD_LISTENER_REPORT);
2365
2366				IN6M_LOCK_ASSERT_HELD(inm);
2367				MLI_LOCK_ASSERT_HELD(mli);
2368
2369				if (error == 0) {
2370					inm->in6m_timer = odelay;
2371					mtp->cst = 1;
2372				}
2373			}
2374			break;
2375
2376		case MLD_VERSION_2:
2377			/*
2378			 * Defer update of T0 to T1, until the first copy
2379			 * of the state change has been transmitted.
2380			 */
2381			syncstates = 0;
2382
2383			/*
2384			 * Immediately enqueue a State-Change Report for
2385			 * this interface, freeing any previous reports.
2386			 * Don't kick the timers if there is nothing to do,
2387			 * or if an error occurred.
2388			 */
2389			ifq = &inm->in6m_scq;
2390			IF_DRAIN(ifq);
2391			retval = mld_v2_enqueue_group_record(ifq, inm, 1,
2392			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2393			mtp->cst = (ifq->ifq_len > 0);
2394			MLD_PRINTF(("%s: enqueue record = %d\n",
2395			    __func__, retval));
2396			if (retval <= 0) {
2397				error = retval * -1;
2398				break;
2399			}
2400
2401			/*
2402			 * Schedule transmission of pending state-change
2403			 * report up to RV times for this link. The timer
2404			 * will fire at the next mld_timeout (1 second)),
2405			 * giving us an opportunity to merge the reports.
2406			 *
2407			 * If a delay was provided to this function, only
2408			 * use this delay if sooner than the existing one.
2409			 */
2410			VERIFY(mli->mli_rv > 1);
2411			inm->in6m_scrv = mli->mli_rv;
2412			if (delay) {
2413				if (inm->in6m_sctimer > 1) {
2414					inm->in6m_sctimer =
2415					    min(inm->in6m_sctimer, delay);
2416				} else
2417					inm->in6m_sctimer = delay;
2418			} else {
2419				inm->in6m_sctimer = 1;
2420			}
2421			mtp->sct = 1;
2422			error = 0;
2423			break;
2424		}
2425	}
2426	MLI_UNLOCK(mli);
2427
2428	/*
2429	 * Only update the T0 state if state change is atomic,
2430	 * i.e. we don't need to wait for a timer to fire before we
2431	 * can consider the state change to have been communicated.
2432	 */
2433	if (syncstates) {
2434		in6m_commit(inm);
2435		MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2436		    ip6_sprintf(&inm->in6m_addr),
2437		    if_name(inm->in6m_ifp)));
2438	}
2439
2440	return (error);
2441}
2442
2443/*
2444 * Issue an intermediate state change during the life-cycle.
2445 */
2446static int
2447mld_handle_state_change(struct in6_multi *inm, struct mld_ifinfo *mli,
2448    struct mld_tparams *mtp)
2449{
2450	struct ifnet		*ifp;
2451	int			 retval = 0;
2452
2453	IN6M_LOCK_ASSERT_HELD(inm);
2454	MLI_LOCK_ASSERT_NOTHELD(mli);
2455	VERIFY(mtp != NULL);
2456
2457	MLD_PRINTF(("%s: state change for %s on ifp 0x%llx(%s)\n",
2458	    __func__, ip6_sprintf(&inm->in6m_addr),
2459	    (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2460	    if_name(inm->in6m_ifp)));
2461
2462	ifp = inm->in6m_ifp;
2463
2464	MLI_LOCK(mli);
2465	VERIFY(mli->mli_ifp == ifp);
2466
2467	if ((ifp->if_flags & IFF_LOOPBACK) ||
2468	    (mli->mli_flags & MLIF_SILENT) ||
2469	    !mld_is_addr_reported(&inm->in6m_addr) ||
2470	    (mli->mli_version != MLD_VERSION_2)) {
2471		MLI_UNLOCK(mli);
2472		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2473			MLD_PRINTF(("%s: not kicking state machine for silent "
2474			    "group\n", __func__));
2475		}
2476		MLD_PRINTF(("%s: nothing to do\n", __func__));
2477		in6m_commit(inm);
2478		MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2479		    ip6_sprintf(&inm->in6m_addr),
2480		    if_name(inm->in6m_ifp)));
2481		goto done;
2482	}
2483
2484	IF_DRAIN(&inm->in6m_scq);
2485
2486	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2487	    (mli->mli_flags & MLIF_USEALLOW));
2488	mtp->cst = (inm->in6m_scq.ifq_len > 0);
2489	MLD_PRINTF(("%s: enqueue record = %d\n", __func__, retval));
2490	if (retval <= 0) {
2491		MLI_UNLOCK(mli);
2492		retval *= -1;
2493		goto done;
2494	} else {
2495		retval = 0;
2496	}
2497
2498	/*
2499	 * If record(s) were enqueued, start the state-change
2500	 * report timer for this group.
2501	 */
2502	inm->in6m_scrv = mli->mli_rv;
2503	inm->in6m_sctimer = 1;
2504	mtp->sct = 1;
2505	MLI_UNLOCK(mli);
2506
2507done:
2508	return (retval);
2509}
2510
2511/*
2512 * Perform the final leave for a multicast address.
2513 *
2514 * When leaving a group:
2515 *  MLDv1 sends a DONE message, if and only if we are the reporter.
2516 *  MLDv2 enqueues a state-change report containing a transition
2517 *  to INCLUDE {} for immediate transmission.
2518 */
2519static void
2520mld_final_leave(struct in6_multi *inm, struct mld_ifinfo *mli,
2521    struct mld_tparams *mtp)
2522{
2523	int syncstates = 1;
2524
2525	IN6M_LOCK_ASSERT_HELD(inm);
2526	MLI_LOCK_ASSERT_NOTHELD(mli);
2527	VERIFY(mtp != NULL);
2528
2529	MLD_PRINTF(("%s: final leave %s on ifp 0x%llx(%s)\n",
2530	    __func__, ip6_sprintf(&inm->in6m_addr),
2531	    (uint64_t)VM_KERNEL_ADDRPERM(inm->in6m_ifp),
2532	    if_name(inm->in6m_ifp)));
2533
2534	switch (inm->in6m_state) {
2535	case MLD_NOT_MEMBER:
2536	case MLD_SILENT_MEMBER:
2537	case MLD_LEAVING_MEMBER:
2538		/* Already leaving or left; do nothing. */
2539		MLD_PRINTF(("%s: not kicking state machine for silent group\n",
2540		    __func__));
2541		break;
2542	case MLD_REPORTING_MEMBER:
2543	case MLD_IDLE_MEMBER:
2544	case MLD_G_QUERY_PENDING_MEMBER:
2545	case MLD_SG_QUERY_PENDING_MEMBER:
2546		MLI_LOCK(mli);
2547		if (mli->mli_version == MLD_VERSION_1) {
2548			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2549			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
2550				panic("%s: MLDv2 state reached, not MLDv2 "
2551				    "mode\n", __func__);
2552				/* NOTREACHED */
2553			}
2554			/* scheduler timer if enqueue is successful */
2555			mtp->cst = (mld_v1_transmit_report(inm,
2556			    MLD_LISTENER_DONE) == 0);
2557
2558			IN6M_LOCK_ASSERT_HELD(inm);
2559			MLI_LOCK_ASSERT_HELD(mli);
2560
2561			inm->in6m_state = MLD_NOT_MEMBER;
2562		} else if (mli->mli_version == MLD_VERSION_2) {
2563			/*
2564			 * Stop group timer and all pending reports.
2565			 * Immediately enqueue a state-change report
2566			 * TO_IN {} to be sent on the next timeout,
2567			 * giving us an opportunity to merge reports.
2568			 */
2569			IF_DRAIN(&inm->in6m_scq);
2570			inm->in6m_timer = 0;
2571			inm->in6m_scrv = mli->mli_rv;
2572			MLD_PRINTF(("%s: Leaving %s/%s with %d "
2573			    "pending retransmissions.\n", __func__,
2574			    ip6_sprintf(&inm->in6m_addr),
2575			    if_name(inm->in6m_ifp),
2576			    inm->in6m_scrv));
2577			if (inm->in6m_scrv == 0) {
2578				inm->in6m_state = MLD_NOT_MEMBER;
2579				inm->in6m_sctimer = 0;
2580			} else {
2581				int retval;
2582				/*
2583				 * Stick around in the in6_multihead list;
2584				 * the final detach will be issued by
2585				 * mld_v2_process_group_timers() when
2586				 * the retransmit timer expires.
2587				 */
2588				IN6M_ADDREF_LOCKED(inm);
2589				VERIFY(inm->in6m_debug & IFD_ATTACHED);
2590				inm->in6m_reqcnt++;
2591				VERIFY(inm->in6m_reqcnt >= 1);
2592				inm->in6m_nrelecnt++;
2593				VERIFY(inm->in6m_nrelecnt != 0);
2594
2595				retval = mld_v2_enqueue_group_record(
2596				    &inm->in6m_scq, inm, 1, 0, 0,
2597				    (mli->mli_flags & MLIF_USEALLOW));
2598				mtp->cst = (inm->in6m_scq.ifq_len > 0);
2599				KASSERT(retval != 0,
2600				    ("%s: enqueue record = %d\n", __func__,
2601				     retval));
2602
2603				inm->in6m_state = MLD_LEAVING_MEMBER;
2604				inm->in6m_sctimer = 1;
2605				mtp->sct = 1;
2606				syncstates = 0;
2607			}
2608		}
2609		MLI_UNLOCK(mli);
2610		break;
2611	case MLD_LAZY_MEMBER:
2612	case MLD_SLEEPING_MEMBER:
2613	case MLD_AWAKENING_MEMBER:
2614		/* Our reports are suppressed; do nothing. */
2615		break;
2616	}
2617
2618	if (syncstates) {
2619		in6m_commit(inm);
2620		MLD_PRINTF(("%s: T1 -> T0 for %s/%s\n", __func__,
2621		    ip6_sprintf(&inm->in6m_addr),
2622		    if_name(inm->in6m_ifp)));
2623		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2624		MLD_PRINTF(("%s: T1 now MCAST_UNDEFINED for 0x%llx/%s\n",
2625		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(&inm->in6m_addr),
2626		    if_name(inm->in6m_ifp)));
2627	}
2628}
2629
2630/*
2631 * Enqueue an MLDv2 group record to the given output queue.
2632 *
2633 * If is_state_change is zero, a current-state record is appended.
2634 * If is_state_change is non-zero, a state-change report is appended.
2635 *
2636 * If is_group_query is non-zero, an mbuf packet chain is allocated.
2637 * If is_group_query is zero, and if there is a packet with free space
2638 * at the tail of the queue, it will be appended to providing there
2639 * is enough free space.
2640 * Otherwise a new mbuf packet chain is allocated.
2641 *
2642 * If is_source_query is non-zero, each source is checked to see if
2643 * it was recorded for a Group-Source query, and will be omitted if
2644 * it is not both in-mode and recorded.
2645 *
2646 * If use_block_allow is non-zero, state change reports for initial join
2647 * and final leave, on an inclusive mode group with a source list, will be
2648 * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2649 *
2650 * The function will attempt to allocate leading space in the packet
2651 * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2652 *
2653 * If successful the size of all data appended to the queue is returned,
2654 * otherwise an error code less than zero is returned, or zero if
2655 * no record(s) were appended.
2656 */
2657static int
2658mld_v2_enqueue_group_record(struct ifqueue *ifq, struct in6_multi *inm,
2659    const int is_state_change, const int is_group_query,
2660    const int is_source_query, const int use_block_allow)
2661{
2662	struct mldv2_record	 mr;
2663	struct mldv2_record	*pmr;
2664	struct ifnet		*ifp;
2665	struct ip6_msource	*ims, *nims;
2666	struct mbuf		*m0, *m, *md;
2667	int			 error, is_filter_list_change;
2668	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2669	int			 record_has_sources;
2670	int			 now;
2671	int			 type;
2672	uint8_t			 mode;
2673
2674	IN6M_LOCK_ASSERT_HELD(inm);
2675	MLI_LOCK_ASSERT_HELD(inm->in6m_mli);
2676
2677	error = 0;
2678	ifp = inm->in6m_ifp;
2679	is_filter_list_change = 0;
2680	m = NULL;
2681	m0 = NULL;
2682	m0srcs = 0;
2683	msrcs = 0;
2684	nbytes = 0;
2685	nims = NULL;
2686	record_has_sources = 1;
2687	pmr = NULL;
2688	type = MLD_DO_NOTHING;
2689	mode = inm->in6m_st[1].iss_fmode;
2690
2691	/*
2692	 * If we did not transition out of ASM mode during t0->t1,
2693	 * and there are no source nodes to process, we can skip
2694	 * the generation of source records.
2695	 */
2696	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2697	    inm->in6m_nsrc == 0)
2698		record_has_sources = 0;
2699
2700	if (is_state_change) {
2701		/*
2702		 * Queue a state change record.
2703		 * If the mode did not change, and there are non-ASM
2704		 * listeners or source filters present,
2705		 * we potentially need to issue two records for the group.
2706		 * If there are ASM listeners, and there was no filter
2707		 * mode transition of any kind, do nothing.
2708		 *
2709		 * If we are transitioning to MCAST_UNDEFINED, we need
2710		 * not send any sources. A transition to/from this state is
2711		 * considered inclusive with some special treatment.
2712		 *
2713		 * If we are rewriting initial joins/leaves to use
2714		 * ALLOW/BLOCK, and the group's membership is inclusive,
2715		 * we need to send sources in all cases.
2716		 */
2717		if (mode != inm->in6m_st[0].iss_fmode) {
2718			if (mode == MCAST_EXCLUDE) {
2719				MLD_PRINTF(("%s: change to EXCLUDE\n",
2720				    __func__));
2721				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2722			} else {
2723				MLD_PRINTF(("%s: change to INCLUDE\n",
2724				    __func__));
2725				if (use_block_allow) {
2726					/*
2727					 * XXX
2728					 * Here we're interested in state
2729					 * edges either direction between
2730					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2731					 * Perhaps we should just check
2732					 * the group state, rather than
2733					 * the filter mode.
2734					 */
2735					if (mode == MCAST_UNDEFINED) {
2736						type = MLD_BLOCK_OLD_SOURCES;
2737					} else {
2738						type = MLD_ALLOW_NEW_SOURCES;
2739					}
2740				} else {
2741					type = MLD_CHANGE_TO_INCLUDE_MODE;
2742					if (mode == MCAST_UNDEFINED)
2743						record_has_sources = 0;
2744				}
2745			}
2746		} else {
2747			if (record_has_sources) {
2748				is_filter_list_change = 1;
2749			} else {
2750				type = MLD_DO_NOTHING;
2751			}
2752		}
2753	} else {
2754		/*
2755		 * Queue a current state record.
2756		 */
2757		if (mode == MCAST_EXCLUDE) {
2758			type = MLD_MODE_IS_EXCLUDE;
2759		} else if (mode == MCAST_INCLUDE) {
2760			type = MLD_MODE_IS_INCLUDE;
2761			VERIFY(inm->in6m_st[1].iss_asm == 0);
2762		}
2763	}
2764
2765	/*
2766	 * Generate the filter list changes using a separate function.
2767	 */
2768	if (is_filter_list_change)
2769		return (mld_v2_enqueue_filter_change(ifq, inm));
2770
2771	if (type == MLD_DO_NOTHING) {
2772		MLD_PRINTF(("%s: nothing to do for %s/%s\n",
2773		    __func__, ip6_sprintf(&inm->in6m_addr),
2774		    if_name(inm->in6m_ifp)));
2775		return (0);
2776	}
2777
2778	/*
2779	 * If any sources are present, we must be able to fit at least
2780	 * one in the trailing space of the tail packet's mbuf,
2781	 * ideally more.
2782	 */
2783	minrec0len = sizeof(struct mldv2_record);
2784	if (record_has_sources)
2785		minrec0len += sizeof(struct in6_addr);
2786	MLD_PRINTF(("%s: queueing %s for %s/%s\n", __func__,
2787	    mld_rec_type_to_str(type),
2788	    ip6_sprintf(&inm->in6m_addr),
2789	    if_name(inm->in6m_ifp)));
2790
2791	/*
2792	 * Check if we have a packet in the tail of the queue for this
2793	 * group into which the first group record for this group will fit.
2794	 * Otherwise allocate a new packet.
2795	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2796	 * Note: Group records for G/GSR query responses MUST be sent
2797	 * in their own packet.
2798	 */
2799	m0 = ifq->ifq_tail;
2800	if (!is_group_query &&
2801	    m0 != NULL &&
2802	    (m0->m_pkthdr.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2803	    (m0->m_pkthdr.len + minrec0len) <
2804	     (ifp->if_mtu - MLD_MTUSPACE)) {
2805		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2806			    sizeof(struct mldv2_record)) /
2807			    sizeof(struct in6_addr);
2808		m = m0;
2809		MLD_PRINTF(("%s: use existing packet\n", __func__));
2810	} else {
2811		if (IF_QFULL(ifq)) {
2812			MLD_PRINTF(("%s: outbound queue full\n", __func__));
2813			return (-ENOMEM);
2814		}
2815		m = NULL;
2816		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2817		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2818		if (!is_state_change && !is_group_query)
2819			m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2820		if (m == NULL)
2821			m = m_gethdr(M_DONTWAIT, MT_DATA);
2822		if (m == NULL)
2823			return (-ENOMEM);
2824
2825		mld_save_context(m, ifp);
2826
2827		MLD_PRINTF(("%s: allocated first packet\n", __func__));
2828	}
2829
2830	/*
2831	 * Append group record.
2832	 * If we have sources, we don't know how many yet.
2833	 */
2834	mr.mr_type = type;
2835	mr.mr_datalen = 0;
2836	mr.mr_numsrc = 0;
2837	mr.mr_addr = inm->in6m_addr;
2838	in6_clearscope(&mr.mr_addr);
2839	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2840		if (m != m0)
2841			m_freem(m);
2842		MLD_PRINTF(("%s: m_append() failed.\n", __func__));
2843		return (-ENOMEM);
2844	}
2845	nbytes += sizeof(struct mldv2_record);
2846
2847	/*
2848	 * Append as many sources as will fit in the first packet.
2849	 * If we are appending to a new packet, the chain allocation
2850	 * may potentially use clusters; use m_getptr() in this case.
2851	 * If we are appending to an existing packet, we need to obtain
2852	 * a pointer to the group record after m_append(), in case a new
2853	 * mbuf was allocated.
2854	 *
2855	 * Only append sources which are in-mode at t1. If we are
2856	 * transitioning to MCAST_UNDEFINED state on the group, and
2857	 * use_block_allow is zero, do not include source entries.
2858	 * Otherwise, we need to include this source in the report.
2859	 *
2860	 * Only report recorded sources in our filter set when responding
2861	 * to a group-source query.
2862	 */
2863	if (record_has_sources) {
2864		if (m == m0) {
2865			md = m_last(m);
2866			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2867			    md->m_len - nbytes);
2868		} else {
2869			md = m_getptr(m, 0, &off);
2870			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2871			    off);
2872		}
2873		msrcs = 0;
2874		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2875		    nims) {
2876			MLD_PRINTF(("%s: visit node %s\n", __func__,
2877			    ip6_sprintf(&ims->im6s_addr)));
2878			now = im6s_get_mode(inm, ims, 1);
2879			MLD_PRINTF(("%s: node is %d\n", __func__, now));
2880			if ((now != mode) ||
2881			    (now == mode &&
2882			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2883				MLD_PRINTF(("%s: skip node\n", __func__));
2884				continue;
2885			}
2886			if (is_source_query && ims->im6s_stp == 0) {
2887				MLD_PRINTF(("%s: skip unrecorded node\n",
2888				    __func__));
2889				continue;
2890			}
2891			MLD_PRINTF(("%s: append node\n", __func__));
2892			if (!m_append(m, sizeof(struct in6_addr),
2893			    (void *)&ims->im6s_addr)) {
2894				if (m != m0)
2895					m_freem(m);
2896				MLD_PRINTF(("%s: m_append() failed.\n",
2897				    __func__));
2898				return (-ENOMEM);
2899			}
2900			nbytes += sizeof(struct in6_addr);
2901			++msrcs;
2902			if (msrcs == m0srcs)
2903				break;
2904		}
2905		MLD_PRINTF(("%s: msrcs is %d this packet\n", __func__,
2906		    msrcs));
2907		pmr->mr_numsrc = htons(msrcs);
2908		nbytes += (msrcs * sizeof(struct in6_addr));
2909	}
2910
2911	if (is_source_query && msrcs == 0) {
2912		MLD_PRINTF(("%s: no recorded sources to report\n", __func__));
2913		if (m != m0)
2914			m_freem(m);
2915		return (0);
2916	}
2917
2918	/*
2919	 * We are good to go with first packet.
2920	 */
2921	if (m != m0) {
2922		MLD_PRINTF(("%s: enqueueing first packet\n", __func__));
2923		m->m_pkthdr.vt_nrecs = 1;
2924		IF_ENQUEUE(ifq, m);
2925	} else {
2926		m->m_pkthdr.vt_nrecs++;
2927	}
2928	/*
2929	 * No further work needed if no source list in packet(s).
2930	 */
2931	if (!record_has_sources)
2932		return (nbytes);
2933
2934	/*
2935	 * Whilst sources remain to be announced, we need to allocate
2936	 * a new packet and fill out as many sources as will fit.
2937	 * Always try for a cluster first.
2938	 */
2939	while (nims != NULL) {
2940		if (IF_QFULL(ifq)) {
2941			MLD_PRINTF(("%s: outbound queue full\n", __func__));
2942			return (-ENOMEM);
2943		}
2944		m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
2945		if (m == NULL)
2946			m = m_gethdr(M_DONTWAIT, MT_DATA);
2947		if (m == NULL)
2948			return (-ENOMEM);
2949		mld_save_context(m, ifp);
2950		md = m_getptr(m, 0, &off);
2951		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2952		MLD_PRINTF(("%s: allocated next packet\n", __func__));
2953
2954		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2955			if (m != m0)
2956				m_freem(m);
2957			MLD_PRINTF(("%s: m_append() failed.\n", __func__));
2958			return (-ENOMEM);
2959		}
2960		m->m_pkthdr.vt_nrecs = 1;
2961		nbytes += sizeof(struct mldv2_record);
2962
2963		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2964		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2965
2966		msrcs = 0;
2967		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2968			MLD_PRINTF(("%s: visit node %s\n",
2969			    __func__, ip6_sprintf(&ims->im6s_addr)));
2970			now = im6s_get_mode(inm, ims, 1);
2971			if ((now != mode) ||
2972			    (now == mode &&
2973			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2974				MLD_PRINTF(("%s: skip node\n", __func__));
2975				continue;
2976			}
2977			if (is_source_query && ims->im6s_stp == 0) {
2978				MLD_PRINTF(("%s: skip unrecorded node\n",
2979				    __func__));
2980				continue;
2981			}
2982			MLD_PRINTF(("%s: append node\n", __func__));
2983			if (!m_append(m, sizeof(struct in6_addr),
2984			    (void *)&ims->im6s_addr)) {
2985				if (m != m0)
2986					m_freem(m);
2987				MLD_PRINTF(("%s: m_append() failed.\n",
2988				    __func__));
2989				return (-ENOMEM);
2990			}
2991			++msrcs;
2992			if (msrcs == m0srcs)
2993				break;
2994		}
2995		pmr->mr_numsrc = htons(msrcs);
2996		nbytes += (msrcs * sizeof(struct in6_addr));
2997
2998		MLD_PRINTF(("%s: enqueueing next packet\n", __func__));
2999		IF_ENQUEUE(ifq, m);
3000	}
3001
3002	return (nbytes);
3003}
3004
3005/*
3006 * Type used to mark record pass completion.
3007 * We exploit the fact we can cast to this easily from the
3008 * current filter modes on each ip_msource node.
3009 */
3010typedef enum {
3011	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
3012	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
3013	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
3014	REC_FULL = REC_ALLOW | REC_BLOCK
3015} rectype_t;
3016
3017/*
3018 * Enqueue an MLDv2 filter list change to the given output queue.
3019 *
3020 * Source list filter state is held in an RB-tree. When the filter list
3021 * for a group is changed without changing its mode, we need to compute
3022 * the deltas between T0 and T1 for each source in the filter set,
3023 * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
3024 *
3025 * As we may potentially queue two record types, and the entire R-B tree
3026 * needs to be walked at once, we break this out into its own function
3027 * so we can generate a tightly packed queue of packets.
3028 *
3029 * XXX This could be written to only use one tree walk, although that makes
3030 * serializing into the mbuf chains a bit harder. For now we do two walks
3031 * which makes things easier on us, and it may or may not be harder on
3032 * the L2 cache.
3033 *
3034 * If successful the size of all data appended to the queue is returned,
3035 * otherwise an error code less than zero is returned, or zero if
3036 * no record(s) were appended.
3037 */
3038static int
3039mld_v2_enqueue_filter_change(struct ifqueue *ifq, struct in6_multi *inm)
3040{
3041	static const int MINRECLEN =
3042	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
3043	struct ifnet		*ifp;
3044	struct mldv2_record	 mr;
3045	struct mldv2_record	*pmr;
3046	struct ip6_msource	*ims, *nims;
3047	struct mbuf		*m, *m0, *md;
3048	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
3049	int			 nallow, nblock;
3050	uint8_t			 mode, now, then;
3051	rectype_t		 crt, drt, nrt;
3052
3053	IN6M_LOCK_ASSERT_HELD(inm);
3054
3055	if (inm->in6m_nsrc == 0 ||
3056	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
3057		return (0);
3058
3059	ifp = inm->in6m_ifp;			/* interface */
3060	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
3061	crt = REC_NONE;	/* current group record type */
3062	drt = REC_NONE;	/* mask of completed group record types */
3063	nrt = REC_NONE;	/* record type for current node */
3064	m0srcs = 0;	/* # source which will fit in current mbuf chain */
3065	npbytes = 0;	/* # of bytes appended this packet */
3066	nbytes = 0;	/* # of bytes appended to group's state-change queue */
3067	rsrcs = 0;	/* # sources encoded in current record */
3068	schanged = 0;	/* # nodes encoded in overall filter change */
3069	nallow = 0;	/* # of source entries in ALLOW_NEW */
3070	nblock = 0;	/* # of source entries in BLOCK_OLD */
3071	nims = NULL;	/* next tree node pointer */
3072
3073	/*
3074	 * For each possible filter record mode.
3075	 * The first kind of source we encounter tells us which
3076	 * is the first kind of record we start appending.
3077	 * If a node transitioned to UNDEFINED at t1, its mode is treated
3078	 * as the inverse of the group's filter mode.
3079	 */
3080	while (drt != REC_FULL) {
3081		do {
3082			m0 = ifq->ifq_tail;
3083			if (m0 != NULL &&
3084			    (m0->m_pkthdr.vt_nrecs + 1 <=
3085			     MLD_V2_REPORT_MAXRECS) &&
3086			    (m0->m_pkthdr.len + MINRECLEN) <
3087			     (ifp->if_mtu - MLD_MTUSPACE)) {
3088				m = m0;
3089				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
3090					    sizeof(struct mldv2_record)) /
3091					    sizeof(struct in6_addr);
3092				MLD_PRINTF(("%s: use previous packet\n",
3093				    __func__));
3094			} else {
3095				m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
3096				if (m == NULL)
3097					m = m_gethdr(M_DONTWAIT, MT_DATA);
3098				if (m == NULL) {
3099					MLD_PRINTF(("%s: m_get*() failed\n",
3100					    __func__));
3101					return (-ENOMEM);
3102				}
3103				m->m_pkthdr.vt_nrecs = 0;
3104				mld_save_context(m, ifp);
3105				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
3106				    sizeof(struct mldv2_record)) /
3107				    sizeof(struct in6_addr);
3108				npbytes = 0;
3109				MLD_PRINTF(("%s: allocated new packet\n",
3110				    __func__));
3111			}
3112			/*
3113			 * Append the MLD group record header to the
3114			 * current packet's data area.
3115			 * Recalculate pointer to free space for next
3116			 * group record, in case m_append() allocated
3117			 * a new mbuf or cluster.
3118			 */
3119			memset(&mr, 0, sizeof(mr));
3120			mr.mr_addr = inm->in6m_addr;
3121			in6_clearscope(&mr.mr_addr);
3122			if (!m_append(m, sizeof(mr), (void *)&mr)) {
3123				if (m != m0)
3124					m_freem(m);
3125				MLD_PRINTF(("%s: m_append() failed\n",
3126				    __func__));
3127				return (-ENOMEM);
3128			}
3129			npbytes += sizeof(struct mldv2_record);
3130			if (m != m0) {
3131				/* new packet; offset in chain */
3132				md = m_getptr(m, npbytes -
3133				    sizeof(struct mldv2_record), &off);
3134				pmr = (struct mldv2_record *)(mtod(md,
3135				    uint8_t *) + off);
3136			} else {
3137				/* current packet; offset from last append */
3138				md = m_last(m);
3139				pmr = (struct mldv2_record *)(mtod(md,
3140				    uint8_t *) + md->m_len -
3141				    sizeof(struct mldv2_record));
3142			}
3143			/*
3144			 * Begin walking the tree for this record type
3145			 * pass, or continue from where we left off
3146			 * previously if we had to allocate a new packet.
3147			 * Only report deltas in-mode at t1.
3148			 * We need not report included sources as allowed
3149			 * if we are in inclusive mode on the group,
3150			 * however the converse is not true.
3151			 */
3152			rsrcs = 0;
3153			if (nims == NULL) {
3154				nims = RB_MIN(ip6_msource_tree,
3155				    &inm->in6m_srcs);
3156			}
3157			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
3158				MLD_PRINTF(("%s: visit node %s\n", __func__,
3159				    ip6_sprintf(&ims->im6s_addr)));
3160				now = im6s_get_mode(inm, ims, 1);
3161				then = im6s_get_mode(inm, ims, 0);
3162				MLD_PRINTF(("%s: mode: t0 %d, t1 %d\n",
3163				    __func__, then, now));
3164				if (now == then) {
3165					MLD_PRINTF(("%s: skip unchanged\n",
3166					    __func__));
3167					continue;
3168				}
3169				if (mode == MCAST_EXCLUDE &&
3170				    now == MCAST_INCLUDE) {
3171					MLD_PRINTF(("%s: skip IN src on EX "
3172					    "group\n", __func__));
3173					continue;
3174				}
3175				nrt = (rectype_t)now;
3176				if (nrt == REC_NONE)
3177					nrt = (rectype_t)(~mode & REC_FULL);
3178				if (schanged++ == 0) {
3179					crt = nrt;
3180				} else if (crt != nrt)
3181					continue;
3182				if (!m_append(m, sizeof(struct in6_addr),
3183				    (void *)&ims->im6s_addr)) {
3184					if (m != m0)
3185						m_freem(m);
3186					MLD_PRINTF(("%s: m_append() failed\n",
3187					    __func__));
3188					return (-ENOMEM);
3189				}
3190				nallow += !!(crt == REC_ALLOW);
3191				nblock += !!(crt == REC_BLOCK);
3192				if (++rsrcs == m0srcs)
3193					break;
3194			}
3195			/*
3196			 * If we did not append any tree nodes on this
3197			 * pass, back out of allocations.
3198			 */
3199			if (rsrcs == 0) {
3200				npbytes -= sizeof(struct mldv2_record);
3201				if (m != m0) {
3202					MLD_PRINTF(("%s: m_free(m)\n",
3203					    __func__));
3204					m_freem(m);
3205				} else {
3206					MLD_PRINTF(("%s: m_adj(m, -mr)\n",
3207					    __func__));
3208					m_adj(m, -((int)sizeof(
3209					    struct mldv2_record)));
3210				}
3211				continue;
3212			}
3213			npbytes += (rsrcs * sizeof(struct in6_addr));
3214			if (crt == REC_ALLOW)
3215				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
3216			else if (crt == REC_BLOCK)
3217				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
3218			pmr->mr_numsrc = htons(rsrcs);
3219			/*
3220			 * Count the new group record, and enqueue this
3221			 * packet if it wasn't already queued.
3222			 */
3223			m->m_pkthdr.vt_nrecs++;
3224			if (m != m0)
3225				IF_ENQUEUE(ifq, m);
3226			nbytes += npbytes;
3227		} while (nims != NULL);
3228		drt |= crt;
3229		crt = (~crt & REC_FULL);
3230	}
3231
3232	MLD_PRINTF(("%s: queued %d ALLOW_NEW, %d BLOCK_OLD\n", __func__,
3233	    nallow, nblock));
3234
3235	return (nbytes);
3236}
3237
3238static int
3239mld_v2_merge_state_changes(struct in6_multi *inm, struct ifqueue *ifscq)
3240{
3241	struct ifqueue	*gq;
3242	struct mbuf	*m;		/* pending state-change */
3243	struct mbuf	*m0;		/* copy of pending state-change */
3244	struct mbuf	*mt;		/* last state-change in packet */
3245	struct mbuf	*n;
3246	int		 docopy, domerge;
3247	u_int		 recslen;
3248
3249	IN6M_LOCK_ASSERT_HELD(inm);
3250
3251	docopy = 0;
3252	domerge = 0;
3253	recslen = 0;
3254
3255	/*
3256	 * If there are further pending retransmissions, make a writable
3257	 * copy of each queued state-change message before merging.
3258	 */
3259	if (inm->in6m_scrv > 0)
3260		docopy = 1;
3261
3262	gq = &inm->in6m_scq;
3263#ifdef MLD_DEBUG
3264	if (gq->ifq_head == NULL) {
3265		MLD_PRINTF(("%s: WARNING: queue for inm 0x%llx is empty\n",
3266		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(inm)));
3267	}
3268#endif
3269
3270	/*
3271	 * Use IF_REMQUEUE() instead of IF_DEQUEUE() below, since the
3272	 * packet might not always be at the head of the ifqueue.
3273	 */
3274	m = gq->ifq_head;
3275	while (m != NULL) {
3276		/*
3277		 * Only merge the report into the current packet if
3278		 * there is sufficient space to do so; an MLDv2 report
3279		 * packet may only contain 65,535 group records.
3280		 * Always use a simple mbuf chain concatentation to do this,
3281		 * as large state changes for single groups may have
3282		 * allocated clusters.
3283		 */
3284		domerge = 0;
3285		mt = ifscq->ifq_tail;
3286		if (mt != NULL) {
3287			recslen = m_length(m);
3288
3289			if ((mt->m_pkthdr.vt_nrecs +
3290			    m->m_pkthdr.vt_nrecs <=
3291			    MLD_V2_REPORT_MAXRECS) &&
3292			    (mt->m_pkthdr.len + recslen <=
3293			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
3294				domerge = 1;
3295		}
3296
3297		if (!domerge && IF_QFULL(gq)) {
3298			MLD_PRINTF(("%s: outbound queue full, skipping whole "
3299			    "packet 0x%llx\n", __func__,
3300			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3301			n = m->m_nextpkt;
3302			if (!docopy) {
3303				IF_REMQUEUE(gq, m);
3304				m_freem(m);
3305			}
3306			m = n;
3307			continue;
3308		}
3309
3310		if (!docopy) {
3311			MLD_PRINTF(("%s: dequeueing 0x%llx\n", __func__,
3312			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3313			n = m->m_nextpkt;
3314			IF_REMQUEUE(gq, m);
3315			m0 = m;
3316			m = n;
3317		} else {
3318			MLD_PRINTF(("%s: copying 0x%llx\n", __func__,
3319			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3320			m0 = m_dup(m, M_NOWAIT);
3321			if (m0 == NULL)
3322				return (ENOMEM);
3323			m0->m_nextpkt = NULL;
3324			m = m->m_nextpkt;
3325		}
3326
3327		if (!domerge) {
3328			MLD_PRINTF(("%s: queueing 0x%llx to ifscq 0x%llx)\n",
3329			    __func__, (uint64_t)VM_KERNEL_ADDRPERM(m0),
3330			    (uint64_t)VM_KERNEL_ADDRPERM(ifscq)));
3331			IF_ENQUEUE(ifscq, m0);
3332		} else {
3333			struct mbuf *mtl;	/* last mbuf of packet mt */
3334
3335			MLD_PRINTF(("%s: merging 0x%llx with ifscq tail "
3336			    "0x%llx)\n", __func__,
3337			    (uint64_t)VM_KERNEL_ADDRPERM(m0),
3338			    (uint64_t)VM_KERNEL_ADDRPERM(mt)));
3339
3340			mtl = m_last(mt);
3341			m0->m_flags &= ~M_PKTHDR;
3342			mt->m_pkthdr.len += recslen;
3343			mt->m_pkthdr.vt_nrecs +=
3344			    m0->m_pkthdr.vt_nrecs;
3345
3346			mtl->m_next = m0;
3347		}
3348	}
3349
3350	return (0);
3351}
3352
3353/*
3354 * Respond to a pending MLDv2 General Query.
3355 */
3356static uint32_t
3357mld_v2_dispatch_general_query(struct mld_ifinfo *mli)
3358{
3359	struct ifnet		*ifp;
3360	struct in6_multi	*inm;
3361	struct in6_multistep	step;
3362	int			 retval;
3363
3364	MLI_LOCK_ASSERT_HELD(mli);
3365
3366	VERIFY(mli->mli_version == MLD_VERSION_2);
3367
3368	ifp = mli->mli_ifp;
3369	MLI_UNLOCK(mli);
3370
3371	in6_multihead_lock_shared();
3372	IN6_FIRST_MULTI(step, inm);
3373	while (inm != NULL) {
3374		IN6M_LOCK(inm);
3375		if (inm->in6m_ifp != ifp)
3376			goto next;
3377
3378		switch (inm->in6m_state) {
3379		case MLD_NOT_MEMBER:
3380		case MLD_SILENT_MEMBER:
3381			break;
3382		case MLD_REPORTING_MEMBER:
3383		case MLD_IDLE_MEMBER:
3384		case MLD_LAZY_MEMBER:
3385		case MLD_SLEEPING_MEMBER:
3386		case MLD_AWAKENING_MEMBER:
3387			inm->in6m_state = MLD_REPORTING_MEMBER;
3388			MLI_LOCK(mli);
3389			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3390			    inm, 0, 0, 0, 0);
3391			MLI_UNLOCK(mli);
3392			MLD_PRINTF(("%s: enqueue record = %d\n",
3393			    __func__, retval));
3394			break;
3395		case MLD_G_QUERY_PENDING_MEMBER:
3396		case MLD_SG_QUERY_PENDING_MEMBER:
3397		case MLD_LEAVING_MEMBER:
3398			break;
3399		}
3400next:
3401		IN6M_UNLOCK(inm);
3402		IN6_NEXT_MULTI(step, inm);
3403	}
3404	in6_multihead_lock_done();
3405
3406	MLI_LOCK(mli);
3407	mld_dispatch_queue(mli, &mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3408	MLI_LOCK_ASSERT_HELD(mli);
3409
3410	/*
3411	 * Slew transmission of bursts over 1 second intervals.
3412	 */
3413	if (mli->mli_gq.ifq_head != NULL) {
3414		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3415		    MLD_RESPONSE_BURST_INTERVAL);
3416	}
3417
3418	return (mli->mli_v2_timer);
3419}
3420
3421/*
3422 * Transmit the next pending message in the output queue.
3423 *
3424 * Must not be called with in6m_lockm or mli_lock held.
3425 */
3426static void
3427mld_dispatch_packet(struct mbuf *m)
3428{
3429	struct ip6_moptions	*im6o;
3430	struct ifnet		*ifp;
3431	struct ifnet		*oifp = NULL;
3432	struct mbuf		*m0;
3433	struct mbuf		*md;
3434	struct ip6_hdr		*ip6;
3435	struct mld_hdr		*mld;
3436	int			 error;
3437	int			 off;
3438	int			 type;
3439
3440	MLD_PRINTF(("%s: transmit 0x%llx\n", __func__,
3441	    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3442
3443	/*
3444	 * Check if the ifnet is still attached.
3445	 */
3446	ifp = mld_restore_context(m);
3447	if (ifp == NULL || !ifnet_is_attached(ifp, 0)) {
3448		MLD_PRINTF(("%s: dropped 0x%llx as ifindex %u went away.\n",
3449		    __func__, (uint64_t)VM_KERNEL_ADDRPERM(m),
3450		    (u_int)if_index));
3451		m_freem(m);
3452		ip6stat.ip6s_noroute++;
3453		return;
3454	}
3455
3456	im6o = ip6_allocmoptions(M_WAITOK);
3457	if (im6o == NULL) {
3458		m_freem(m);
3459		return;
3460	}
3461
3462	im6o->im6o_multicast_hlim  = 1;
3463	im6o->im6o_multicast_loop = 0;
3464	im6o->im6o_multicast_ifp = ifp;
3465
3466	if (m->m_flags & M_MLDV1) {
3467		m0 = m;
3468	} else {
3469		m0 = mld_v2_encap_report(ifp, m);
3470		if (m0 == NULL) {
3471			MLD_PRINTF(("%s: dropped 0x%llx\n", __func__,
3472			    (uint64_t)VM_KERNEL_ADDRPERM(m)));
3473			/*
3474			 * mld_v2_encap_report() has already freed our mbuf.
3475			 */
3476			IM6O_REMREF(im6o);
3477			ip6stat.ip6s_odropped++;
3478			return;
3479		}
3480	}
3481
3482	mld_scrub_context(m0);
3483	m->m_flags &= ~(M_PROTOFLAGS);
3484	m0->m_pkthdr.rcvif = lo_ifp;
3485
3486	ip6 = mtod(m0, struct ip6_hdr *);
3487	(void) in6_setscope(&ip6->ip6_dst, ifp, NULL);
3488
3489	/*
3490	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3491	 * so we can bump the stats.
3492	 */
3493	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3494	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3495	type = mld->mld_type;
3496
3497	if (ifp->if_eflags & IFEF_TXSTART) {
3498		/*
3499		 * Use control service class if the outgoing
3500		 * interface supports transmit-start model.
3501		 */
3502		(void) m_set_service_class(m0, MBUF_SC_CTL);
3503	}
3504
3505	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, im6o,
3506	    &oifp, NULL);
3507
3508	IM6O_REMREF(im6o);
3509
3510	if (error) {
3511		MLD_PRINTF(("%s: ip6_output(0x%llx) = %d\n", __func__,
3512		    (uint64_t)VM_KERNEL_ADDRPERM(m0), error));
3513		if (oifp != NULL)
3514			ifnet_release(oifp);
3515		return;
3516	}
3517
3518	icmp6stat.icp6s_outhist[type]++;
3519	if (oifp != NULL) {
3520		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3521		switch (type) {
3522		case MLD_LISTENER_REPORT:
3523		case MLDV2_LISTENER_REPORT:
3524			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3525			break;
3526		case MLD_LISTENER_DONE:
3527			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3528			break;
3529		}
3530		ifnet_release(oifp);
3531	}
3532}
3533
3534/*
3535 * Encapsulate an MLDv2 report.
3536 *
3537 * KAME IPv6 requires that hop-by-hop options be passed separately,
3538 * and that the IPv6 header be prepended in a separate mbuf.
3539 *
3540 * Returns a pointer to the new mbuf chain head, or NULL if the
3541 * allocation failed.
3542 */
3543static struct mbuf *
3544mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3545{
3546	struct mbuf		*mh;
3547	struct mldv2_report	*mld;
3548	struct ip6_hdr		*ip6;
3549	struct in6_ifaddr	*ia;
3550	int			 mldreclen;
3551
3552	VERIFY(m->m_flags & M_PKTHDR);
3553
3554	/*
3555	 * RFC3590: OK to send as :: or tentative during DAD.
3556	 */
3557	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3558	if (ia == NULL)
3559		MLD_PRINTF(("%s: warning: ia is NULL\n", __func__));
3560
3561	MGETHDR(mh, M_DONTWAIT, MT_HEADER);
3562	if (mh == NULL) {
3563		if (ia != NULL)
3564			IFA_REMREF(&ia->ia_ifa);
3565		m_freem(m);
3566		return (NULL);
3567	}
3568	MH_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3569
3570	mldreclen = m_length(m);
3571	MLD_PRINTF(("%s: mldreclen is %d\n", __func__, mldreclen));
3572
3573	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3574	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3575	    sizeof(struct mldv2_report) + mldreclen;
3576
3577	ip6 = mtod(mh, struct ip6_hdr *);
3578	ip6->ip6_flow = 0;
3579	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3580	ip6->ip6_vfc |= IPV6_VERSION;
3581	ip6->ip6_nxt = IPPROTO_ICMPV6;
3582	if (ia != NULL)
3583		IFA_LOCK(&ia->ia_ifa);
3584	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3585	if (ia != NULL) {
3586		IFA_UNLOCK(&ia->ia_ifa);
3587		IFA_REMREF(&ia->ia_ifa);
3588		ia = NULL;
3589	}
3590	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3591	/* scope ID will be set in netisr */
3592
3593	mld = (struct mldv2_report *)(ip6 + 1);
3594	mld->mld_type = MLDV2_LISTENER_REPORT;
3595	mld->mld_code = 0;
3596	mld->mld_cksum = 0;
3597	mld->mld_v2_reserved = 0;
3598	mld->mld_v2_numrecs = htons(m->m_pkthdr.vt_nrecs);
3599	m->m_pkthdr.vt_nrecs = 0;
3600	m->m_flags &= ~M_PKTHDR;
3601
3602	mh->m_next = m;
3603	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3604	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3605	return (mh);
3606}
3607
3608#ifdef MLD_DEBUG
3609static const char *
3610mld_rec_type_to_str(const int type)
3611{
3612	switch (type) {
3613		case MLD_CHANGE_TO_EXCLUDE_MODE:
3614			return "TO_EX";
3615			break;
3616		case MLD_CHANGE_TO_INCLUDE_MODE:
3617			return "TO_IN";
3618			break;
3619		case MLD_MODE_IS_EXCLUDE:
3620			return "MODE_EX";
3621			break;
3622		case MLD_MODE_IS_INCLUDE:
3623			return "MODE_IN";
3624			break;
3625		case MLD_ALLOW_NEW_SOURCES:
3626			return "ALLOW_NEW";
3627			break;
3628		case MLD_BLOCK_OLD_SOURCES:
3629			return "BLOCK_OLD";
3630			break;
3631		default:
3632			break;
3633	}
3634	return "unknown";
3635}
3636#endif
3637
3638void
3639mld_init(void)
3640{
3641
3642	MLD_PRINTF(("%s: initializing\n", __func__));
3643
3644        /* Setup lock group and attribute for mld_mtx */
3645        mld_mtx_grp_attr = lck_grp_attr_alloc_init();
3646        mld_mtx_grp = lck_grp_alloc_init("mld_mtx\n", mld_mtx_grp_attr);
3647        mld_mtx_attr = lck_attr_alloc_init();
3648        lck_mtx_init(&mld_mtx, mld_mtx_grp, mld_mtx_attr);
3649
3650	ip6_initpktopts(&mld_po);
3651	mld_po.ip6po_hlim = 1;
3652	mld_po.ip6po_hbh = &mld_ra.hbh;
3653	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3654	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3655	LIST_INIT(&mli_head);
3656
3657	mli_size = sizeof (struct mld_ifinfo);
3658	mli_zone = zinit(mli_size, MLI_ZONE_MAX * mli_size,
3659	    0, MLI_ZONE_NAME);
3660	if (mli_zone == NULL) {
3661		panic("%s: failed allocating %s", __func__, MLI_ZONE_NAME);
3662		/* NOTREACHED */
3663	}
3664	zone_change(mli_zone, Z_EXPAND, TRUE);
3665	zone_change(mli_zone, Z_CALLERACCT, FALSE);
3666}
3667