xref: /freebsd/sys/netinet6/mld6.c (revision 48c779cdecb5f803e5fe5d761987e976ca9609db)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2009 Bruce Simpson.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. The name of the author may not be used to endorse or promote
15  *    products derived from this software without specific prior written
16  *    permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	$KAME: mld6.c,v 1.27 2001/04/04 05:17:30 itojun Exp $
31  */
32 
33 /*-
34  * Copyright (c) 1988 Stephen Deering.
35  * Copyright (c) 1992, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * This code is derived from software contributed to Berkeley by
39  * Stephen Deering of Stanford University.
40  *
41  * Redistribution and use in source and binary forms, with or without
42  * modification, are permitted provided that the following conditions
43  * are met:
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in the
48  *    documentation and/or other materials provided with the distribution.
49  * 3. Neither the name of the University nor the names of its contributors
50  *    may be used to endorse or promote products derived from this software
51  *    without specific prior written permission.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
54  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
55  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
56  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
57  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
58  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
59  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
60  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
61  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
62  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
63  * SUCH DAMAGE.
64  *
65  *	@(#)igmp.c	8.1 (Berkeley) 7/19/93
66  */
67 
68 #include <sys/cdefs.h>
69 __FBSDID("$FreeBSD$");
70 
71 #include "opt_inet.h"
72 #include "opt_inet6.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/mbuf.h>
77 #include <sys/socket.h>
78 #include <sys/protosw.h>
79 #include <sys/sysctl.h>
80 #include <sys/kernel.h>
81 #include <sys/callout.h>
82 #include <sys/malloc.h>
83 #include <sys/module.h>
84 #include <sys/ktr.h>
85 
86 #include <net/if.h>
87 #include <net/if_var.h>
88 #include <net/route.h>
89 #include <net/vnet.h>
90 
91 #include <netinet/in.h>
92 #include <netinet/in_var.h>
93 #include <netinet6/in6_var.h>
94 #include <netinet/ip6.h>
95 #include <netinet6/ip6_var.h>
96 #include <netinet6/scope6_var.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/mld6.h>
99 #include <netinet6/mld6_var.h>
100 
101 #include <security/mac/mac_framework.h>
102 
103 #ifndef KTR_MLD
104 #define KTR_MLD KTR_INET6
105 #endif
106 
107 static struct mld_ifsoftc *
108 		mli_alloc_locked(struct ifnet *);
109 static void	mli_delete_locked(const struct ifnet *);
110 static void	mld_dispatch_packet(struct mbuf *);
111 static void	mld_dispatch_queue(struct mbufq *, int);
112 static void	mld_final_leave(struct in6_multi *, struct mld_ifsoftc *);
113 static void	mld_fasttimo_vnet(struct in6_multi_head *inmh);
114 static int	mld_handle_state_change(struct in6_multi *,
115 		    struct mld_ifsoftc *);
116 static int	mld_initial_join(struct in6_multi *, struct mld_ifsoftc *,
117 		    const int);
118 #ifdef KTR
119 static char *	mld_rec_type_to_str(const int);
120 #endif
121 static void	mld_set_version(struct mld_ifsoftc *, const int);
122 static void	mld_slowtimo_vnet(void);
123 static int	mld_v1_input_query(struct ifnet *, const struct ip6_hdr *,
124 		    /*const*/ struct mld_hdr *);
125 static int	mld_v1_input_report(struct ifnet *, const struct ip6_hdr *,
126 		    /*const*/ struct mld_hdr *);
127 static void	mld_v1_process_group_timer(struct in6_multi_head *,
128 		    struct in6_multi *);
129 static void	mld_v1_process_querier_timers(struct mld_ifsoftc *);
130 static int	mld_v1_transmit_report(struct in6_multi *, const int);
131 static void	mld_v1_update_group(struct in6_multi *, const int);
132 static void	mld_v2_cancel_link_timers(struct mld_ifsoftc *);
133 static void	mld_v2_dispatch_general_query(struct mld_ifsoftc *);
134 static struct mbuf *
135 		mld_v2_encap_report(struct ifnet *, struct mbuf *);
136 static int	mld_v2_enqueue_filter_change(struct mbufq *,
137 		    struct in6_multi *);
138 static int	mld_v2_enqueue_group_record(struct mbufq *,
139 		    struct in6_multi *, const int, const int, const int,
140 		    const int);
141 static int	mld_v2_input_query(struct ifnet *, const struct ip6_hdr *,
142 		    struct mbuf *, struct mldv2_query *, const int, const int);
143 static int	mld_v2_merge_state_changes(struct in6_multi *,
144 		    struct mbufq *);
145 static void	mld_v2_process_group_timers(struct in6_multi_head *,
146 		    struct mbufq *, struct mbufq *,
147 		    struct in6_multi *, const int);
148 static int	mld_v2_process_group_query(struct in6_multi *,
149 		    struct mld_ifsoftc *mli, int, struct mbuf *,
150 		    struct mldv2_query *, const int);
151 static int	sysctl_mld_gsr(SYSCTL_HANDLER_ARGS);
152 static int	sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS);
153 
154 /*
155  * Normative references: RFC 2710, RFC 3590, RFC 3810.
156  *
157  * Locking:
158  *  * The MLD subsystem lock ends up being system-wide for the moment,
159  *    but could be per-VIMAGE later on.
160  *  * The permitted lock order is: IN6_MULTI_LOCK, MLD_LOCK, IF_ADDR_LOCK.
161  *    Any may be taken independently; if any are held at the same
162  *    time, the above lock order must be followed.
163  *  * IN6_MULTI_LOCK covers in_multi.
164  *  * MLD_LOCK covers per-link state and any global variables in this file.
165  *  * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of
166  *    per-link state iterators.
167  *
168  *  XXX LOR PREVENTION
169  *  A special case for IPv6 is the in6_setscope() routine. ip6_output()
170  *  will not accept an ifp; it wants an embedded scope ID, unlike
171  *  ip_output(), which happily takes the ifp given to it. The embedded
172  *  scope ID is only used by MLD to select the outgoing interface.
173  *
174  *  During interface attach and detach, MLD will take MLD_LOCK *after*
175  *  the IF_AFDATA_LOCK.
176  *  As in6_setscope() takes IF_AFDATA_LOCK then SCOPE_LOCK, we can't call
177  *  it with MLD_LOCK held without triggering an LOR. A netisr with indirect
178  *  dispatch could work around this, but we'd rather not do that, as it
179  *  can introduce other races.
180  *
181  *  As such, we exploit the fact that the scope ID is just the interface
182  *  index, and embed it in the IPv6 destination address accordingly.
183  *  This is potentially NOT VALID for MLDv1 reports, as they
184  *  are always sent to the multicast group itself; as MLDv2
185  *  reports are always sent to ff02::16, this is not an issue
186  *  when MLDv2 is in use.
187  *
188  *  This does not however eliminate the LOR when ip6_output() itself
189  *  calls in6_setscope() internally whilst MLD_LOCK is held. This will
190  *  trigger a LOR warning in WITNESS when the ifnet is detached.
191  *
192  *  The right answer is probably to make IF_AFDATA_LOCK an rwlock, given
193  *  how it's used across the network stack. Here we're simply exploiting
194  *  the fact that MLD runs at a similar layer in the stack to scope6.c.
195  *
196  * VIMAGE:
197  *  * Each in6_multi corresponds to an ifp, and each ifp corresponds
198  *    to a vnet in ifp->if_vnet.
199  */
200 static struct mtx		 mld_mtx;
201 static MALLOC_DEFINE(M_MLD, "mld", "mld state");
202 
203 #define	MLD_EMBEDSCOPE(pin6, zoneid)					\
204 	if (IN6_IS_SCOPE_LINKLOCAL(pin6) ||				\
205 	    IN6_IS_ADDR_MC_INTFACELOCAL(pin6))				\
206 		(pin6)->s6_addr16[1] = htons((zoneid) & 0xFFFF)		\
207 
208 /*
209  * VIMAGE-wide globals.
210  */
211 VNET_DEFINE_STATIC(struct timeval, mld_gsrdelay) = {10, 0};
212 VNET_DEFINE_STATIC(LIST_HEAD(, mld_ifsoftc), mli_head);
213 VNET_DEFINE_STATIC(int, interface_timers_running6);
214 VNET_DEFINE_STATIC(int, state_change_timers_running6);
215 VNET_DEFINE_STATIC(int, current_state_timers_running6);
216 
217 #define	V_mld_gsrdelay			VNET(mld_gsrdelay)
218 #define	V_mli_head			VNET(mli_head)
219 #define	V_interface_timers_running6	VNET(interface_timers_running6)
220 #define	V_state_change_timers_running6	VNET(state_change_timers_running6)
221 #define	V_current_state_timers_running6	VNET(current_state_timers_running6)
222 
223 SYSCTL_DECL(_net_inet6);	/* Note: Not in any common header. */
224 
225 SYSCTL_NODE(_net_inet6, OID_AUTO, mld, CTLFLAG_RW, 0,
226     "IPv6 Multicast Listener Discovery");
227 
228 /*
229  * Virtualized sysctls.
230  */
231 SYSCTL_PROC(_net_inet6_mld, OID_AUTO, gsrdelay,
232     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
233     &VNET_NAME(mld_gsrdelay.tv_sec), 0, sysctl_mld_gsr, "I",
234     "Rate limit for MLDv2 Group-and-Source queries in seconds");
235 
236 /*
237  * Non-virtualized sysctls.
238  */
239 static SYSCTL_NODE(_net_inet6_mld, OID_AUTO, ifinfo,
240     CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_mld_ifinfo,
241     "Per-interface MLDv2 state");
242 
243 static int	mld_v1enable = 1;
244 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v1enable, CTLFLAG_RWTUN,
245     &mld_v1enable, 0, "Enable fallback to MLDv1");
246 
247 static int	mld_v2enable = 1;
248 SYSCTL_INT(_net_inet6_mld, OID_AUTO, v2enable, CTLFLAG_RWTUN,
249     &mld_v2enable, 0, "Enable MLDv2");
250 
251 static int	mld_use_allow = 1;
252 SYSCTL_INT(_net_inet6_mld, OID_AUTO, use_allow, CTLFLAG_RWTUN,
253     &mld_use_allow, 0, "Use ALLOW/BLOCK for RFC 4604 SSM joins/leaves");
254 
255 /*
256  * Packed Router Alert option structure declaration.
257  */
258 struct mld_raopt {
259 	struct ip6_hbh		hbh;
260 	struct ip6_opt		pad;
261 	struct ip6_opt_router	ra;
262 } __packed;
263 
264 /*
265  * Router Alert hop-by-hop option header.
266  */
267 static struct mld_raopt mld_ra = {
268 	.hbh = { 0, 0 },
269 	.pad = { .ip6o_type = IP6OPT_PADN, 0 },
270 	.ra = {
271 	    .ip6or_type = IP6OPT_ROUTER_ALERT,
272 	    .ip6or_len = IP6OPT_RTALERT_LEN - 2,
273 	    .ip6or_value[0] = ((IP6OPT_RTALERT_MLD >> 8) & 0xFF),
274 	    .ip6or_value[1] = (IP6OPT_RTALERT_MLD & 0xFF)
275 	}
276 };
277 static struct ip6_pktopts mld_po;
278 
279 static __inline void
280 mld_save_context(struct mbuf *m, struct ifnet *ifp)
281 {
282 
283 #ifdef VIMAGE
284 	m->m_pkthdr.PH_loc.ptr = ifp->if_vnet;
285 #endif /* VIMAGE */
286 	m->m_pkthdr.flowid = ifp->if_index;
287 }
288 
289 static __inline void
290 mld_scrub_context(struct mbuf *m)
291 {
292 
293 	m->m_pkthdr.PH_loc.ptr = NULL;
294 	m->m_pkthdr.flowid = 0;
295 }
296 
297 /*
298  * Restore context from a queued output chain.
299  * Return saved ifindex.
300  *
301  * VIMAGE: The assertion is there to make sure that we
302  * actually called CURVNET_SET() with what's in the mbuf chain.
303  */
304 static __inline uint32_t
305 mld_restore_context(struct mbuf *m)
306 {
307 
308 #if defined(VIMAGE) && defined(INVARIANTS)
309 	KASSERT(curvnet == m->m_pkthdr.PH_loc.ptr,
310 	    ("%s: called when curvnet was not restored: cuvnet %p m ptr %p",
311 	    __func__, curvnet, m->m_pkthdr.PH_loc.ptr));
312 #endif
313 	return (m->m_pkthdr.flowid);
314 }
315 
316 /*
317  * Retrieve or set threshold between group-source queries in seconds.
318  *
319  * VIMAGE: Assume curvnet set by caller.
320  * SMPng: NOTE: Serialized by MLD lock.
321  */
322 static int
323 sysctl_mld_gsr(SYSCTL_HANDLER_ARGS)
324 {
325 	int error;
326 	int i;
327 
328 	error = sysctl_wire_old_buffer(req, sizeof(int));
329 	if (error)
330 		return (error);
331 
332 	MLD_LOCK();
333 
334 	i = V_mld_gsrdelay.tv_sec;
335 
336 	error = sysctl_handle_int(oidp, &i, 0, req);
337 	if (error || !req->newptr)
338 		goto out_locked;
339 
340 	if (i < -1 || i >= 60) {
341 		error = EINVAL;
342 		goto out_locked;
343 	}
344 
345 	CTR2(KTR_MLD, "change mld_gsrdelay from %d to %d",
346 	     V_mld_gsrdelay.tv_sec, i);
347 	V_mld_gsrdelay.tv_sec = i;
348 
349 out_locked:
350 	MLD_UNLOCK();
351 	return (error);
352 }
353 
354 /*
355  * Expose struct mld_ifsoftc to userland, keyed by ifindex.
356  * For use by ifmcstat(8).
357  *
358  * SMPng: NOTE: Does an unlocked ifindex space read.
359  * VIMAGE: Assume curvnet set by caller. The node handler itself
360  * is not directly virtualized.
361  */
362 static int
363 sysctl_mld_ifinfo(SYSCTL_HANDLER_ARGS)
364 {
365 	int			*name;
366 	int			 error;
367 	u_int			 namelen;
368 	struct ifnet		*ifp;
369 	struct mld_ifsoftc	*mli;
370 
371 	name = (int *)arg1;
372 	namelen = arg2;
373 
374 	if (req->newptr != NULL)
375 		return (EPERM);
376 
377 	if (namelen != 1)
378 		return (EINVAL);
379 
380 	error = sysctl_wire_old_buffer(req, sizeof(struct mld_ifinfo));
381 	if (error)
382 		return (error);
383 
384 	IN6_MULTI_LOCK();
385 	IN6_MULTI_LIST_LOCK();
386 	MLD_LOCK();
387 
388 	if (name[0] <= 0 || name[0] > V_if_index) {
389 		error = ENOENT;
390 		goto out_locked;
391 	}
392 
393 	error = ENOENT;
394 
395 	ifp = ifnet_byindex(name[0]);
396 	if (ifp == NULL)
397 		goto out_locked;
398 
399 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
400 		if (ifp == mli->mli_ifp) {
401 			struct mld_ifinfo info;
402 
403 			info.mli_version = mli->mli_version;
404 			info.mli_v1_timer = mli->mli_v1_timer;
405 			info.mli_v2_timer = mli->mli_v2_timer;
406 			info.mli_flags = mli->mli_flags;
407 			info.mli_rv = mli->mli_rv;
408 			info.mli_qi = mli->mli_qi;
409 			info.mli_qri = mli->mli_qri;
410 			info.mli_uri = mli->mli_uri;
411 			error = SYSCTL_OUT(req, &info, sizeof(info));
412 			break;
413 		}
414 	}
415 
416 out_locked:
417 	MLD_UNLOCK();
418 	IN6_MULTI_LIST_UNLOCK();
419 	IN6_MULTI_UNLOCK();
420 	return (error);
421 }
422 
423 /*
424  * Dispatch an entire queue of pending packet chains.
425  * VIMAGE: Assumes the vnet pointer has been set.
426  */
427 static void
428 mld_dispatch_queue(struct mbufq *mq, int limit)
429 {
430 	struct mbuf *m;
431 
432 	while ((m = mbufq_dequeue(mq)) != NULL) {
433 		CTR3(KTR_MLD, "%s: dispatch %p from %p", __func__, mq, m);
434 		mld_dispatch_packet(m);
435 		if (--limit == 0)
436 			break;
437 	}
438 }
439 
440 /*
441  * Filter outgoing MLD report state by group.
442  *
443  * Reports are ALWAYS suppressed for ALL-HOSTS (ff02::1)
444  * and node-local addresses. However, kernel and socket consumers
445  * always embed the KAME scope ID in the address provided, so strip it
446  * when performing comparison.
447  * Note: This is not the same as the *multicast* scope.
448  *
449  * Return zero if the given group is one for which MLD reports
450  * should be suppressed, or non-zero if reports should be issued.
451  */
452 static __inline int
453 mld_is_addr_reported(const struct in6_addr *addr)
454 {
455 
456 	KASSERT(IN6_IS_ADDR_MULTICAST(addr), ("%s: not multicast", __func__));
457 
458 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_NODELOCAL)
459 		return (0);
460 
461 	if (IPV6_ADDR_MC_SCOPE(addr) == IPV6_ADDR_SCOPE_LINKLOCAL) {
462 		struct in6_addr tmp = *addr;
463 		in6_clearscope(&tmp);
464 		if (IN6_ARE_ADDR_EQUAL(&tmp, &in6addr_linklocal_allnodes))
465 			return (0);
466 	}
467 
468 	return (1);
469 }
470 
471 /*
472  * Attach MLD when PF_INET6 is attached to an interface.
473  *
474  * SMPng: Normally called with IF_AFDATA_LOCK held.
475  */
476 struct mld_ifsoftc *
477 mld_domifattach(struct ifnet *ifp)
478 {
479 	struct mld_ifsoftc *mli;
480 
481 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
482 	    __func__, ifp, if_name(ifp));
483 
484 	MLD_LOCK();
485 
486 	mli = mli_alloc_locked(ifp);
487 	if (!(ifp->if_flags & IFF_MULTICAST))
488 		mli->mli_flags |= MLIF_SILENT;
489 	if (mld_use_allow)
490 		mli->mli_flags |= MLIF_USEALLOW;
491 
492 	MLD_UNLOCK();
493 
494 	return (mli);
495 }
496 
497 /*
498  * VIMAGE: assume curvnet set by caller.
499  */
500 static struct mld_ifsoftc *
501 mli_alloc_locked(/*const*/ struct ifnet *ifp)
502 {
503 	struct mld_ifsoftc *mli;
504 
505 	MLD_LOCK_ASSERT();
506 
507 	mli = malloc(sizeof(struct mld_ifsoftc), M_MLD, M_NOWAIT|M_ZERO);
508 	if (mli == NULL)
509 		goto out;
510 
511 	mli->mli_ifp = ifp;
512 	mli->mli_version = MLD_VERSION_2;
513 	mli->mli_flags = 0;
514 	mli->mli_rv = MLD_RV_INIT;
515 	mli->mli_qi = MLD_QI_INIT;
516 	mli->mli_qri = MLD_QRI_INIT;
517 	mli->mli_uri = MLD_URI_INIT;
518 	mbufq_init(&mli->mli_gq, MLD_MAX_RESPONSE_PACKETS);
519 
520 	LIST_INSERT_HEAD(&V_mli_head, mli, mli_link);
521 
522 	CTR2(KTR_MLD, "allocate mld_ifsoftc for ifp %p(%s)",
523 	     ifp, if_name(ifp));
524 
525 out:
526 	return (mli);
527 }
528 
529 /*
530  * Hook for ifdetach.
531  *
532  * NOTE: Some finalization tasks need to run before the protocol domain
533  * is detached, but also before the link layer does its cleanup.
534  * Run before link-layer cleanup; cleanup groups, but do not free MLD state.
535  *
536  * SMPng: Caller must hold IN6_MULTI_LOCK().
537  * Must take IF_ADDR_LOCK() to cover if_multiaddrs iterator.
538  * XXX This routine is also bitten by unlocked ifma_protospec access.
539  */
540 void
541 mld_ifdetach(struct ifnet *ifp, struct in6_multi_head *inmh)
542 {
543 	struct epoch_tracker     et;
544 	struct mld_ifsoftc	*mli;
545 	struct ifmultiaddr	*ifma;
546 	struct in6_multi	*inm;
547 
548 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)", __func__, ifp,
549 	    if_name(ifp));
550 
551 	IN6_MULTI_LIST_LOCK_ASSERT();
552 	MLD_LOCK();
553 
554 	mli = MLD_IFINFO(ifp);
555 	IF_ADDR_WLOCK(ifp);
556 	/*
557 	 * Extract list of in6_multi associated with the detaching ifp
558 	 * which the PF_INET6 layer is about to release.
559 	 */
560 	NET_EPOCH_ENTER(et);
561 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
562 		inm = in6m_ifmultiaddr_get_inm(ifma);
563 		if (inm == NULL)
564 			continue;
565 		in6m_disconnect_locked(inmh, inm);
566 
567 		if (mli->mli_version == MLD_VERSION_2) {
568 			in6m_clear_recorded(inm);
569 
570 			/*
571 			 * We need to release the final reference held
572 			 * for issuing the INCLUDE {}.
573 			 */
574 			if (inm->in6m_state == MLD_LEAVING_MEMBER) {
575 				inm->in6m_state = MLD_NOT_MEMBER;
576 				in6m_rele_locked(inmh, inm);
577 			}
578 		}
579 	}
580 	NET_EPOCH_EXIT(et);
581 	IF_ADDR_WUNLOCK(ifp);
582 	MLD_UNLOCK();
583 }
584 
585 /*
586  * Hook for domifdetach.
587  * Runs after link-layer cleanup; free MLD state.
588  *
589  * SMPng: Normally called with IF_AFDATA_LOCK held.
590  */
591 void
592 mld_domifdetach(struct ifnet *ifp)
593 {
594 
595 	CTR3(KTR_MLD, "%s: called for ifp %p(%s)",
596 	    __func__, ifp, if_name(ifp));
597 
598 	MLD_LOCK();
599 	mli_delete_locked(ifp);
600 	MLD_UNLOCK();
601 }
602 
603 static void
604 mli_delete_locked(const struct ifnet *ifp)
605 {
606 	struct mld_ifsoftc *mli, *tmli;
607 
608 	CTR3(KTR_MLD, "%s: freeing mld_ifsoftc for ifp %p(%s)",
609 	    __func__, ifp, if_name(ifp));
610 
611 	MLD_LOCK_ASSERT();
612 
613 	LIST_FOREACH_SAFE(mli, &V_mli_head, mli_link, tmli) {
614 		if (mli->mli_ifp == ifp) {
615 			/*
616 			 * Free deferred General Query responses.
617 			 */
618 			mbufq_drain(&mli->mli_gq);
619 
620 			LIST_REMOVE(mli, mli_link);
621 
622 			free(mli, M_MLD);
623 			return;
624 		}
625 	}
626 }
627 
628 /*
629  * Process a received MLDv1 general or address-specific query.
630  * Assumes that the query header has been pulled up to sizeof(mld_hdr).
631  *
632  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
633  * mld_addr. This is OK as we own the mbuf chain.
634  */
635 static int
636 mld_v1_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
637     /*const*/ struct mld_hdr *mld)
638 {
639 	struct ifmultiaddr	*ifma;
640 	struct mld_ifsoftc	*mli;
641 	struct in6_multi	*inm;
642 	int			 is_general_query;
643 	uint16_t		 timer;
644 #ifdef KTR
645 	char			 ip6tbuf[INET6_ADDRSTRLEN];
646 #endif
647 
648 	NET_EPOCH_ASSERT();
649 
650 	is_general_query = 0;
651 
652 	if (!mld_v1enable) {
653 		CTR3(KTR_MLD, "ignore v1 query %s on ifp %p(%s)",
654 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
655 		    ifp, if_name(ifp));
656 		return (0);
657 	}
658 
659 	/*
660 	 * RFC3810 Section 6.2: MLD queries must originate from
661 	 * a router's link-local address.
662 	 */
663 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
664 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
665 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
666 		    ifp, if_name(ifp));
667 		return (0);
668 	}
669 
670 	/*
671 	 * Do address field validation upfront before we accept
672 	 * the query.
673 	 */
674 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
675 		/*
676 		 * MLDv1 General Query.
677 		 * If this was not sent to the all-nodes group, ignore it.
678 		 */
679 		struct in6_addr		 dst;
680 
681 		dst = ip6->ip6_dst;
682 		in6_clearscope(&dst);
683 		if (!IN6_ARE_ADDR_EQUAL(&dst, &in6addr_linklocal_allnodes))
684 			return (EINVAL);
685 		is_general_query = 1;
686 	} else {
687 		/*
688 		 * Embed scope ID of receiving interface in MLD query for
689 		 * lookup whilst we don't hold other locks.
690 		 */
691 		in6_setscope(&mld->mld_addr, ifp, NULL);
692 	}
693 
694 	IN6_MULTI_LIST_LOCK();
695 	MLD_LOCK();
696 
697 	/*
698 	 * Switch to MLDv1 host compatibility mode.
699 	 */
700 	mli = MLD_IFINFO(ifp);
701 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
702 	mld_set_version(mli, MLD_VERSION_1);
703 
704 	timer = (ntohs(mld->mld_maxdelay) * PR_FASTHZ) / MLD_TIMER_SCALE;
705 	if (timer == 0)
706 		timer = 1;
707 
708 	if (is_general_query) {
709 		/*
710 		 * For each reporting group joined on this
711 		 * interface, kick the report timer.
712 		 */
713 		CTR2(KTR_MLD, "process v1 general query on ifp %p(%s)",
714 			 ifp, if_name(ifp));
715 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
716 			inm = in6m_ifmultiaddr_get_inm(ifma);
717 			if (inm == NULL)
718 				continue;
719 			mld_v1_update_group(inm, timer);
720 		}
721 	} else {
722 		/*
723 		 * MLDv1 Group-Specific Query.
724 		 * If this is a group-specific MLDv1 query, we need only
725 		 * look up the single group to process it.
726 		 */
727 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
728 		if (inm != NULL) {
729 			CTR3(KTR_MLD, "process v1 query %s on ifp %p(%s)",
730 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
731 			    ifp, if_name(ifp));
732 			mld_v1_update_group(inm, timer);
733 		}
734 		/* XXX Clear embedded scope ID as userland won't expect it. */
735 		in6_clearscope(&mld->mld_addr);
736 	}
737 
738 	MLD_UNLOCK();
739 	IN6_MULTI_LIST_UNLOCK();
740 
741 	return (0);
742 }
743 
744 /*
745  * Update the report timer on a group in response to an MLDv1 query.
746  *
747  * If we are becoming the reporting member for this group, start the timer.
748  * If we already are the reporting member for this group, and timer is
749  * below the threshold, reset it.
750  *
751  * We may be updating the group for the first time since we switched
752  * to MLDv2. If we are, then we must clear any recorded source lists,
753  * and transition to REPORTING state; the group timer is overloaded
754  * for group and group-source query responses.
755  *
756  * Unlike MLDv2, the delay per group should be jittered
757  * to avoid bursts of MLDv1 reports.
758  */
759 static void
760 mld_v1_update_group(struct in6_multi *inm, const int timer)
761 {
762 #ifdef KTR
763 	char			 ip6tbuf[INET6_ADDRSTRLEN];
764 #endif
765 
766 	CTR4(KTR_MLD, "%s: %s/%s timer=%d", __func__,
767 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
768 	    if_name(inm->in6m_ifp), timer);
769 
770 	IN6_MULTI_LIST_LOCK_ASSERT();
771 
772 	switch (inm->in6m_state) {
773 	case MLD_NOT_MEMBER:
774 	case MLD_SILENT_MEMBER:
775 		break;
776 	case MLD_REPORTING_MEMBER:
777 		if (inm->in6m_timer != 0 &&
778 		    inm->in6m_timer <= timer) {
779 			CTR1(KTR_MLD, "%s: REPORTING and timer running, "
780 			    "skipping.", __func__);
781 			break;
782 		}
783 		/* FALLTHROUGH */
784 	case MLD_SG_QUERY_PENDING_MEMBER:
785 	case MLD_G_QUERY_PENDING_MEMBER:
786 	case MLD_IDLE_MEMBER:
787 	case MLD_LAZY_MEMBER:
788 	case MLD_AWAKENING_MEMBER:
789 		CTR1(KTR_MLD, "%s: ->REPORTING", __func__);
790 		inm->in6m_state = MLD_REPORTING_MEMBER;
791 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
792 		V_current_state_timers_running6 = 1;
793 		break;
794 	case MLD_SLEEPING_MEMBER:
795 		CTR1(KTR_MLD, "%s: ->AWAKENING", __func__);
796 		inm->in6m_state = MLD_AWAKENING_MEMBER;
797 		break;
798 	case MLD_LEAVING_MEMBER:
799 		break;
800 	}
801 }
802 
803 /*
804  * Process a received MLDv2 general, group-specific or
805  * group-and-source-specific query.
806  *
807  * Assumes that mld points to a struct mldv2_query which is stored in
808  * contiguous memory.
809  *
810  * Return 0 if successful, otherwise an appropriate error code is returned.
811  */
812 static int
813 mld_v2_input_query(struct ifnet *ifp, const struct ip6_hdr *ip6,
814     struct mbuf *m, struct mldv2_query *mld, const int off, const int icmp6len)
815 {
816 	struct mld_ifsoftc	*mli;
817 	struct in6_multi	*inm;
818 	uint32_t		 maxdelay, nsrc, qqi;
819 	int			 is_general_query;
820 	uint16_t		 timer;
821 	uint8_t			 qrv;
822 #ifdef KTR
823 	char			 ip6tbuf[INET6_ADDRSTRLEN];
824 #endif
825 
826 	NET_EPOCH_ASSERT();
827 
828 	if (!mld_v2enable) {
829 		CTR3(KTR_MLD, "ignore v2 query src %s on ifp %p(%s)",
830 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
831 		    ifp, if_name(ifp));
832 		return (0);
833 	}
834 
835 	/*
836 	 * RFC3810 Section 6.2: MLD queries must originate from
837 	 * a router's link-local address.
838 	 */
839 	if (!IN6_IS_SCOPE_LINKLOCAL(&ip6->ip6_src)) {
840 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
841 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
842 		    ifp, if_name(ifp));
843 		return (0);
844 	}
845 
846 	is_general_query = 0;
847 
848 	CTR2(KTR_MLD, "input v2 query on ifp %p(%s)", ifp, if_name(ifp));
849 
850 	maxdelay = ntohs(mld->mld_maxdelay);	/* in 1/10ths of a second */
851 	if (maxdelay >= 32768) {
852 		maxdelay = (MLD_MRC_MANT(maxdelay) | 0x1000) <<
853 			   (MLD_MRC_EXP(maxdelay) + 3);
854 	}
855 	timer = (maxdelay * PR_FASTHZ) / MLD_TIMER_SCALE;
856 	if (timer == 0)
857 		timer = 1;
858 
859 	qrv = MLD_QRV(mld->mld_misc);
860 	if (qrv < 2) {
861 		CTR3(KTR_MLD, "%s: clamping qrv %d to %d", __func__,
862 		    qrv, MLD_RV_INIT);
863 		qrv = MLD_RV_INIT;
864 	}
865 
866 	qqi = mld->mld_qqi;
867 	if (qqi >= 128) {
868 		qqi = MLD_QQIC_MANT(mld->mld_qqi) <<
869 		     (MLD_QQIC_EXP(mld->mld_qqi) + 3);
870 	}
871 
872 	nsrc = ntohs(mld->mld_numsrc);
873 	if (nsrc > MLD_MAX_GS_SOURCES)
874 		return (EMSGSIZE);
875 	if (icmp6len < sizeof(struct mldv2_query) +
876 	    (nsrc * sizeof(struct in6_addr)))
877 		return (EMSGSIZE);
878 
879 	/*
880 	 * Do further input validation upfront to avoid resetting timers
881 	 * should we need to discard this query.
882 	 */
883 	if (IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr)) {
884 		/*
885 		 * A general query with a source list has undefined
886 		 * behaviour; discard it.
887 		 */
888 		if (nsrc > 0)
889 			return (EINVAL);
890 		is_general_query = 1;
891 	} else {
892 		/*
893 		 * Embed scope ID of receiving interface in MLD query for
894 		 * lookup whilst we don't hold other locks (due to KAME
895 		 * locking lameness). We own this mbuf chain just now.
896 		 */
897 		in6_setscope(&mld->mld_addr, ifp, NULL);
898 	}
899 
900 	IN6_MULTI_LIST_LOCK();
901 	MLD_LOCK();
902 
903 	mli = MLD_IFINFO(ifp);
904 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
905 
906 	/*
907 	 * Discard the v2 query if we're in Compatibility Mode.
908 	 * The RFC is pretty clear that hosts need to stay in MLDv1 mode
909 	 * until the Old Version Querier Present timer expires.
910 	 */
911 	if (mli->mli_version != MLD_VERSION_2)
912 		goto out_locked;
913 
914 	mld_set_version(mli, MLD_VERSION_2);
915 	mli->mli_rv = qrv;
916 	mli->mli_qi = qqi;
917 	mli->mli_qri = maxdelay;
918 
919 	CTR4(KTR_MLD, "%s: qrv %d qi %d maxdelay %d", __func__, qrv, qqi,
920 	    maxdelay);
921 
922 	if (is_general_query) {
923 		/*
924 		 * MLDv2 General Query.
925 		 *
926 		 * Schedule a current-state report on this ifp for
927 		 * all groups, possibly containing source lists.
928 		 *
929 		 * If there is a pending General Query response
930 		 * scheduled earlier than the selected delay, do
931 		 * not schedule any other reports.
932 		 * Otherwise, reset the interface timer.
933 		 */
934 		CTR2(KTR_MLD, "process v2 general query on ifp %p(%s)",
935 		    ifp, if_name(ifp));
936 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer) {
937 			mli->mli_v2_timer = MLD_RANDOM_DELAY(timer);
938 			V_interface_timers_running6 = 1;
939 		}
940 	} else {
941 		/*
942 		 * MLDv2 Group-specific or Group-and-source-specific Query.
943 		 *
944 		 * Group-source-specific queries are throttled on
945 		 * a per-group basis to defeat denial-of-service attempts.
946 		 * Queries for groups we are not a member of on this
947 		 * link are simply ignored.
948 		 */
949 		inm = in6m_lookup_locked(ifp, &mld->mld_addr);
950 		if (inm == NULL)
951 			goto out_locked;
952 		if (nsrc > 0) {
953 			if (!ratecheck(&inm->in6m_lastgsrtv,
954 			    &V_mld_gsrdelay)) {
955 				CTR1(KTR_MLD, "%s: GS query throttled.",
956 				    __func__);
957 				goto out_locked;
958 			}
959 		}
960 		CTR2(KTR_MLD, "process v2 group query on ifp %p(%s)",
961 		     ifp, if_name(ifp));
962 		/*
963 		 * If there is a pending General Query response
964 		 * scheduled sooner than the selected delay, no
965 		 * further report need be scheduled.
966 		 * Otherwise, prepare to respond to the
967 		 * group-specific or group-and-source query.
968 		 */
969 		if (mli->mli_v2_timer == 0 || mli->mli_v2_timer >= timer)
970 			mld_v2_process_group_query(inm, mli, timer, m, mld, off);
971 
972 		/* XXX Clear embedded scope ID as userland won't expect it. */
973 		in6_clearscope(&mld->mld_addr);
974 	}
975 
976 out_locked:
977 	MLD_UNLOCK();
978 	IN6_MULTI_LIST_UNLOCK();
979 
980 	return (0);
981 }
982 
983 /*
984  * Process a received MLDv2 group-specific or group-and-source-specific
985  * query.
986  * Return <0 if any error occurred. Currently this is ignored.
987  */
988 static int
989 mld_v2_process_group_query(struct in6_multi *inm, struct mld_ifsoftc *mli,
990     int timer, struct mbuf *m0, struct mldv2_query *mld, const int off)
991 {
992 	int			 retval;
993 	uint16_t		 nsrc;
994 
995 	IN6_MULTI_LIST_LOCK_ASSERT();
996 	MLD_LOCK_ASSERT();
997 
998 	retval = 0;
999 
1000 	switch (inm->in6m_state) {
1001 	case MLD_NOT_MEMBER:
1002 	case MLD_SILENT_MEMBER:
1003 	case MLD_SLEEPING_MEMBER:
1004 	case MLD_LAZY_MEMBER:
1005 	case MLD_AWAKENING_MEMBER:
1006 	case MLD_IDLE_MEMBER:
1007 	case MLD_LEAVING_MEMBER:
1008 		return (retval);
1009 		break;
1010 	case MLD_REPORTING_MEMBER:
1011 	case MLD_G_QUERY_PENDING_MEMBER:
1012 	case MLD_SG_QUERY_PENDING_MEMBER:
1013 		break;
1014 	}
1015 
1016 	nsrc = ntohs(mld->mld_numsrc);
1017 
1018 	/* Length should be checked by calling function. */
1019 	KASSERT((m0->m_flags & M_PKTHDR) == 0 ||
1020 	    m0->m_pkthdr.len >= off + sizeof(struct mldv2_query) +
1021 	    nsrc * sizeof(struct in6_addr),
1022 	    ("mldv2 packet is too short: (%d bytes < %zd bytes, m=%p)",
1023 	    m0->m_pkthdr.len, off + sizeof(struct mldv2_query) +
1024 	    nsrc * sizeof(struct in6_addr), m0));
1025 
1026 
1027 	/*
1028 	 * Deal with group-specific queries upfront.
1029 	 * If any group query is already pending, purge any recorded
1030 	 * source-list state if it exists, and schedule a query response
1031 	 * for this group-specific query.
1032 	 */
1033 	if (nsrc == 0) {
1034 		if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
1035 		    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER) {
1036 			in6m_clear_recorded(inm);
1037 			timer = min(inm->in6m_timer, timer);
1038 		}
1039 		inm->in6m_state = MLD_G_QUERY_PENDING_MEMBER;
1040 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1041 		V_current_state_timers_running6 = 1;
1042 		return (retval);
1043 	}
1044 
1045 	/*
1046 	 * Deal with the case where a group-and-source-specific query has
1047 	 * been received but a group-specific query is already pending.
1048 	 */
1049 	if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER) {
1050 		timer = min(inm->in6m_timer, timer);
1051 		inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1052 		V_current_state_timers_running6 = 1;
1053 		return (retval);
1054 	}
1055 
1056 	/*
1057 	 * Finally, deal with the case where a group-and-source-specific
1058 	 * query has been received, where a response to a previous g-s-r
1059 	 * query exists, or none exists.
1060 	 * In this case, we need to parse the source-list which the Querier
1061 	 * has provided us with and check if we have any source list filter
1062 	 * entries at T1 for these sources. If we do not, there is no need
1063 	 * schedule a report and the query may be dropped.
1064 	 * If we do, we must record them and schedule a current-state
1065 	 * report for those sources.
1066 	 */
1067 	if (inm->in6m_nsrc > 0) {
1068 		struct in6_addr		 srcaddr;
1069 		int			 i, nrecorded;
1070 		int			 soff;
1071 
1072 		soff = off + sizeof(struct mldv2_query);
1073 		nrecorded = 0;
1074 		for (i = 0; i < nsrc; i++) {
1075 			m_copydata(m0, soff, sizeof(struct in6_addr),
1076 			    (caddr_t)&srcaddr);
1077 			retval = in6m_record_source(inm, &srcaddr);
1078 			if (retval < 0)
1079 				break;
1080 			nrecorded += retval;
1081 			soff += sizeof(struct in6_addr);
1082 		}
1083 		if (nrecorded > 0) {
1084 			CTR1(KTR_MLD,
1085 			    "%s: schedule response to SG query", __func__);
1086 			inm->in6m_state = MLD_SG_QUERY_PENDING_MEMBER;
1087 			inm->in6m_timer = MLD_RANDOM_DELAY(timer);
1088 			V_current_state_timers_running6 = 1;
1089 		}
1090 	}
1091 
1092 	return (retval);
1093 }
1094 
1095 /*
1096  * Process a received MLDv1 host membership report.
1097  * Assumes mld points to mld_hdr in pulled up mbuf chain.
1098  *
1099  * NOTE: Can't be fully const correct as we temporarily embed scope ID in
1100  * mld_addr. This is OK as we own the mbuf chain.
1101  */
1102 static int
1103 mld_v1_input_report(struct ifnet *ifp, const struct ip6_hdr *ip6,
1104     /*const*/ struct mld_hdr *mld)
1105 {
1106 	struct in6_addr		 src, dst;
1107 	struct in6_ifaddr	*ia;
1108 	struct in6_multi	*inm;
1109 #ifdef KTR
1110 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1111 #endif
1112 
1113 	NET_EPOCH_ASSERT();
1114 
1115 	if (!mld_v1enable) {
1116 		CTR3(KTR_MLD, "ignore v1 report %s on ifp %p(%s)",
1117 		    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1118 		    ifp, if_name(ifp));
1119 		return (0);
1120 	}
1121 
1122 	if (ifp->if_flags & IFF_LOOPBACK)
1123 		return (0);
1124 
1125 	/*
1126 	 * MLDv1 reports must originate from a host's link-local address,
1127 	 * or the unspecified address (when booting).
1128 	 */
1129 	src = ip6->ip6_src;
1130 	in6_clearscope(&src);
1131 	if (!IN6_IS_SCOPE_LINKLOCAL(&src) && !IN6_IS_ADDR_UNSPECIFIED(&src)) {
1132 		CTR3(KTR_MLD, "ignore v1 query src %s on ifp %p(%s)",
1133 		    ip6_sprintf(ip6tbuf, &ip6->ip6_src),
1134 		    ifp, if_name(ifp));
1135 		return (EINVAL);
1136 	}
1137 
1138 	/*
1139 	 * RFC2710 Section 4: MLDv1 reports must pertain to a multicast
1140 	 * group, and must be directed to the group itself.
1141 	 */
1142 	dst = ip6->ip6_dst;
1143 	in6_clearscope(&dst);
1144 	if (!IN6_IS_ADDR_MULTICAST(&mld->mld_addr) ||
1145 	    !IN6_ARE_ADDR_EQUAL(&mld->mld_addr, &dst)) {
1146 		CTR3(KTR_MLD, "ignore v1 query dst %s on ifp %p(%s)",
1147 		    ip6_sprintf(ip6tbuf, &ip6->ip6_dst),
1148 		    ifp, if_name(ifp));
1149 		return (EINVAL);
1150 	}
1151 
1152 	/*
1153 	 * Make sure we don't hear our own membership report, as fast
1154 	 * leave requires knowing that we are the only member of a
1155 	 * group. Assume we used the link-local address if available,
1156 	 * otherwise look for ::.
1157 	 *
1158 	 * XXX Note that scope ID comparison is needed for the address
1159 	 * returned by in6ifa_ifpforlinklocal(), but SHOULD NOT be
1160 	 * performed for the on-wire address.
1161 	 */
1162 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1163 	if ((ia && IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, IA6_IN6(ia))) ||
1164 	    (ia == NULL && IN6_IS_ADDR_UNSPECIFIED(&src))) {
1165 		if (ia != NULL)
1166 			ifa_free(&ia->ia_ifa);
1167 		return (0);
1168 	}
1169 	if (ia != NULL)
1170 		ifa_free(&ia->ia_ifa);
1171 
1172 	CTR3(KTR_MLD, "process v1 report %s on ifp %p(%s)",
1173 	    ip6_sprintf(ip6tbuf, &mld->mld_addr), ifp, if_name(ifp));
1174 
1175 	/*
1176 	 * Embed scope ID of receiving interface in MLD query for lookup
1177 	 * whilst we don't hold other locks (due to KAME locking lameness).
1178 	 */
1179 	if (!IN6_IS_ADDR_UNSPECIFIED(&mld->mld_addr))
1180 		in6_setscope(&mld->mld_addr, ifp, NULL);
1181 
1182 	IN6_MULTI_LIST_LOCK();
1183 	MLD_LOCK();
1184 
1185 	/*
1186 	 * MLDv1 report suppression.
1187 	 * If we are a member of this group, and our membership should be
1188 	 * reported, and our group timer is pending or about to be reset,
1189 	 * stop our group timer by transitioning to the 'lazy' state.
1190 	 */
1191 	inm = in6m_lookup_locked(ifp, &mld->mld_addr);
1192 	if (inm != NULL) {
1193 		struct mld_ifsoftc *mli;
1194 
1195 		mli = inm->in6m_mli;
1196 		KASSERT(mli != NULL,
1197 		    ("%s: no mli for ifp %p", __func__, ifp));
1198 
1199 		/*
1200 		 * If we are in MLDv2 host mode, do not allow the
1201 		 * other host's MLDv1 report to suppress our reports.
1202 		 */
1203 		if (mli->mli_version == MLD_VERSION_2)
1204 			goto out_locked;
1205 
1206 		inm->in6m_timer = 0;
1207 
1208 		switch (inm->in6m_state) {
1209 		case MLD_NOT_MEMBER:
1210 		case MLD_SILENT_MEMBER:
1211 		case MLD_SLEEPING_MEMBER:
1212 			break;
1213 		case MLD_REPORTING_MEMBER:
1214 		case MLD_IDLE_MEMBER:
1215 		case MLD_AWAKENING_MEMBER:
1216 			CTR3(KTR_MLD,
1217 			    "report suppressed for %s on ifp %p(%s)",
1218 			    ip6_sprintf(ip6tbuf, &mld->mld_addr),
1219 			    ifp, if_name(ifp));
1220 		case MLD_LAZY_MEMBER:
1221 			inm->in6m_state = MLD_LAZY_MEMBER;
1222 			break;
1223 		case MLD_G_QUERY_PENDING_MEMBER:
1224 		case MLD_SG_QUERY_PENDING_MEMBER:
1225 		case MLD_LEAVING_MEMBER:
1226 			break;
1227 		}
1228 	}
1229 
1230 out_locked:
1231 	MLD_UNLOCK();
1232 	IN6_MULTI_LIST_UNLOCK();
1233 
1234 	/* XXX Clear embedded scope ID as userland won't expect it. */
1235 	in6_clearscope(&mld->mld_addr);
1236 
1237 	return (0);
1238 }
1239 
1240 /*
1241  * MLD input path.
1242  *
1243  * Assume query messages which fit in a single ICMPv6 message header
1244  * have been pulled up.
1245  * Assume that userland will want to see the message, even if it
1246  * otherwise fails kernel input validation; do not free it.
1247  * Pullup may however free the mbuf chain m if it fails.
1248  *
1249  * Return IPPROTO_DONE if we freed m. Otherwise, return 0.
1250  */
1251 int
1252 mld_input(struct mbuf *m, int off, int icmp6len)
1253 {
1254 	struct ifnet	*ifp;
1255 	struct ip6_hdr	*ip6;
1256 	struct mld_hdr	*mld;
1257 	int		 mldlen;
1258 
1259 	CTR3(KTR_MLD, "%s: called w/mbuf (%p,%d)", __func__, m, off);
1260 
1261 	ifp = m->m_pkthdr.rcvif;
1262 
1263 	/* Pullup to appropriate size. */
1264 	mld = (struct mld_hdr *)(mtod(m, uint8_t *) + off);
1265 	if (mld->mld_type == MLD_LISTENER_QUERY &&
1266 	    icmp6len >= sizeof(struct mldv2_query)) {
1267 		mldlen = sizeof(struct mldv2_query);
1268 	} else {
1269 		mldlen = sizeof(struct mld_hdr);
1270 	}
1271 	IP6_EXTHDR_GET(mld, struct mld_hdr *, m, off, mldlen);
1272 	if (mld == NULL) {
1273 		ICMP6STAT_INC(icp6s_badlen);
1274 		return (IPPROTO_DONE);
1275 	}
1276 	ip6 = mtod(m, struct ip6_hdr *);
1277 
1278 	/*
1279 	 * Userland needs to see all of this traffic for implementing
1280 	 * the endpoint discovery portion of multicast routing.
1281 	 */
1282 	switch (mld->mld_type) {
1283 	case MLD_LISTENER_QUERY:
1284 		icmp6_ifstat_inc(ifp, ifs6_in_mldquery);
1285 		if (icmp6len == sizeof(struct mld_hdr)) {
1286 			if (mld_v1_input_query(ifp, ip6, mld) != 0)
1287 				return (0);
1288 		} else if (icmp6len >= sizeof(struct mldv2_query)) {
1289 			if (mld_v2_input_query(ifp, ip6, m,
1290 			    (struct mldv2_query *)mld, off, icmp6len) != 0)
1291 				return (0);
1292 		}
1293 		break;
1294 	case MLD_LISTENER_REPORT:
1295 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1296 		if (mld_v1_input_report(ifp, ip6, mld) != 0)
1297 			return (0);
1298 		break;
1299 	case MLDV2_LISTENER_REPORT:
1300 		icmp6_ifstat_inc(ifp, ifs6_in_mldreport);
1301 		break;
1302 	case MLD_LISTENER_DONE:
1303 		icmp6_ifstat_inc(ifp, ifs6_in_mlddone);
1304 		break;
1305 	default:
1306 		break;
1307 	}
1308 
1309 	return (0);
1310 }
1311 
1312 /*
1313  * Fast timeout handler (global).
1314  * VIMAGE: Timeout handlers are expected to service all vimages.
1315  */
1316 void
1317 mld_fasttimo(void)
1318 {
1319 	struct in6_multi_head inmh;
1320 	VNET_ITERATOR_DECL(vnet_iter);
1321 
1322 	SLIST_INIT(&inmh);
1323 
1324 	VNET_LIST_RLOCK_NOSLEEP();
1325 	VNET_FOREACH(vnet_iter) {
1326 		CURVNET_SET(vnet_iter);
1327 		mld_fasttimo_vnet(&inmh);
1328 		CURVNET_RESTORE();
1329 	}
1330 	VNET_LIST_RUNLOCK_NOSLEEP();
1331 	in6m_release_list_deferred(&inmh);
1332 }
1333 
1334 /*
1335  * Fast timeout handler (per-vnet).
1336  *
1337  * VIMAGE: Assume caller has set up our curvnet.
1338  */
1339 static void
1340 mld_fasttimo_vnet(struct in6_multi_head *inmh)
1341 {
1342 	struct epoch_tracker     et;
1343 	struct mbufq		 scq;	/* State-change packets */
1344 	struct mbufq		 qrq;	/* Query response packets */
1345 	struct ifnet		*ifp;
1346 	struct mld_ifsoftc	*mli;
1347 	struct ifmultiaddr	*ifma;
1348 	struct in6_multi	*inm;
1349 	int			 uri_fasthz;
1350 
1351 	uri_fasthz = 0;
1352 
1353 	/*
1354 	 * Quick check to see if any work needs to be done, in order to
1355 	 * minimize the overhead of fasttimo processing.
1356 	 * SMPng: XXX Unlocked reads.
1357 	 */
1358 	if (!V_current_state_timers_running6 &&
1359 	    !V_interface_timers_running6 &&
1360 	    !V_state_change_timers_running6)
1361 		return;
1362 
1363 	IN6_MULTI_LIST_LOCK();
1364 	MLD_LOCK();
1365 
1366 	/*
1367 	 * MLDv2 General Query response timer processing.
1368 	 */
1369 	if (V_interface_timers_running6) {
1370 		CTR1(KTR_MLD, "%s: interface timers running", __func__);
1371 
1372 		V_interface_timers_running6 = 0;
1373 		LIST_FOREACH(mli, &V_mli_head, mli_link) {
1374 			if (mli->mli_v2_timer == 0) {
1375 				/* Do nothing. */
1376 			} else if (--mli->mli_v2_timer == 0) {
1377 				mld_v2_dispatch_general_query(mli);
1378 			} else {
1379 				V_interface_timers_running6 = 1;
1380 			}
1381 		}
1382 	}
1383 
1384 	if (!V_current_state_timers_running6 &&
1385 	    !V_state_change_timers_running6)
1386 		goto out_locked;
1387 
1388 	V_current_state_timers_running6 = 0;
1389 	V_state_change_timers_running6 = 0;
1390 
1391 	CTR1(KTR_MLD, "%s: state change timers running", __func__);
1392 
1393 	/*
1394 	 * MLD host report and state-change timer processing.
1395 	 * Note: Processing a v2 group timer may remove a node.
1396 	 */
1397 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1398 		ifp = mli->mli_ifp;
1399 
1400 		if (mli->mli_version == MLD_VERSION_2) {
1401 			uri_fasthz = MLD_RANDOM_DELAY(mli->mli_uri *
1402 			    PR_FASTHZ);
1403 			mbufq_init(&qrq, MLD_MAX_G_GS_PACKETS);
1404 			mbufq_init(&scq, MLD_MAX_STATE_CHANGE_PACKETS);
1405 		}
1406 
1407 		NET_EPOCH_ENTER(et);
1408 		IF_ADDR_WLOCK(ifp);
1409 		CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1410 			inm = in6m_ifmultiaddr_get_inm(ifma);
1411 			if (inm == NULL)
1412 				continue;
1413 			switch (mli->mli_version) {
1414 			case MLD_VERSION_1:
1415 				mld_v1_process_group_timer(inmh, inm);
1416 				break;
1417 			case MLD_VERSION_2:
1418 				mld_v2_process_group_timers(inmh, &qrq,
1419 				    &scq, inm, uri_fasthz);
1420 				break;
1421 			}
1422 		}
1423 		IF_ADDR_WUNLOCK(ifp);
1424 
1425 		switch (mli->mli_version) {
1426 		case MLD_VERSION_1:
1427 			/*
1428 			 * Transmit reports for this lifecycle.  This
1429 			 * is done while not holding IF_ADDR_LOCK
1430 			 * since this can call
1431 			 * in6ifa_ifpforlinklocal() which locks
1432 			 * IF_ADDR_LOCK internally as well as
1433 			 * ip6_output() to transmit a packet.
1434 			 */
1435 			while ((inm = SLIST_FIRST(inmh)) != NULL) {
1436 				SLIST_REMOVE_HEAD(inmh, in6m_defer);
1437 				(void)mld_v1_transmit_report(inm,
1438 				    MLD_LISTENER_REPORT);
1439 			}
1440 			break;
1441 		case MLD_VERSION_2:
1442 			mld_dispatch_queue(&qrq, 0);
1443 			mld_dispatch_queue(&scq, 0);
1444 			break;
1445 		}
1446 		NET_EPOCH_EXIT(et);
1447 	}
1448 
1449 out_locked:
1450 	MLD_UNLOCK();
1451 	IN6_MULTI_LIST_UNLOCK();
1452 }
1453 
1454 /*
1455  * Update host report group timer.
1456  * Will update the global pending timer flags.
1457  */
1458 static void
1459 mld_v1_process_group_timer(struct in6_multi_head *inmh, struct in6_multi *inm)
1460 {
1461 	int report_timer_expired;
1462 
1463 	IN6_MULTI_LIST_LOCK_ASSERT();
1464 	MLD_LOCK_ASSERT();
1465 
1466 	if (inm->in6m_timer == 0) {
1467 		report_timer_expired = 0;
1468 	} else if (--inm->in6m_timer == 0) {
1469 		report_timer_expired = 1;
1470 	} else {
1471 		V_current_state_timers_running6 = 1;
1472 		return;
1473 	}
1474 
1475 	switch (inm->in6m_state) {
1476 	case MLD_NOT_MEMBER:
1477 	case MLD_SILENT_MEMBER:
1478 	case MLD_IDLE_MEMBER:
1479 	case MLD_LAZY_MEMBER:
1480 	case MLD_SLEEPING_MEMBER:
1481 	case MLD_AWAKENING_MEMBER:
1482 		break;
1483 	case MLD_REPORTING_MEMBER:
1484 		if (report_timer_expired) {
1485 			inm->in6m_state = MLD_IDLE_MEMBER;
1486 			SLIST_INSERT_HEAD(inmh, inm, in6m_defer);
1487 		}
1488 		break;
1489 	case MLD_G_QUERY_PENDING_MEMBER:
1490 	case MLD_SG_QUERY_PENDING_MEMBER:
1491 	case MLD_LEAVING_MEMBER:
1492 		break;
1493 	}
1494 }
1495 
1496 /*
1497  * Update a group's timers for MLDv2.
1498  * Will update the global pending timer flags.
1499  * Note: Unlocked read from mli.
1500  */
1501 static void
1502 mld_v2_process_group_timers(struct in6_multi_head *inmh,
1503     struct mbufq *qrq, struct mbufq *scq,
1504     struct in6_multi *inm, const int uri_fasthz)
1505 {
1506 	int query_response_timer_expired;
1507 	int state_change_retransmit_timer_expired;
1508 #ifdef KTR
1509 	char ip6tbuf[INET6_ADDRSTRLEN];
1510 #endif
1511 
1512 	IN6_MULTI_LIST_LOCK_ASSERT();
1513 	MLD_LOCK_ASSERT();
1514 
1515 	query_response_timer_expired = 0;
1516 	state_change_retransmit_timer_expired = 0;
1517 
1518 	/*
1519 	 * During a transition from compatibility mode back to MLDv2,
1520 	 * a group record in REPORTING state may still have its group
1521 	 * timer active. This is a no-op in this function; it is easier
1522 	 * to deal with it here than to complicate the slow-timeout path.
1523 	 */
1524 	if (inm->in6m_timer == 0) {
1525 		query_response_timer_expired = 0;
1526 	} else if (--inm->in6m_timer == 0) {
1527 		query_response_timer_expired = 1;
1528 	} else {
1529 		V_current_state_timers_running6 = 1;
1530 	}
1531 
1532 	if (inm->in6m_sctimer == 0) {
1533 		state_change_retransmit_timer_expired = 0;
1534 	} else if (--inm->in6m_sctimer == 0) {
1535 		state_change_retransmit_timer_expired = 1;
1536 	} else {
1537 		V_state_change_timers_running6 = 1;
1538 	}
1539 
1540 	/* We are in fasttimo, so be quick about it. */
1541 	if (!state_change_retransmit_timer_expired &&
1542 	    !query_response_timer_expired)
1543 		return;
1544 
1545 	switch (inm->in6m_state) {
1546 	case MLD_NOT_MEMBER:
1547 	case MLD_SILENT_MEMBER:
1548 	case MLD_SLEEPING_MEMBER:
1549 	case MLD_LAZY_MEMBER:
1550 	case MLD_AWAKENING_MEMBER:
1551 	case MLD_IDLE_MEMBER:
1552 		break;
1553 	case MLD_G_QUERY_PENDING_MEMBER:
1554 	case MLD_SG_QUERY_PENDING_MEMBER:
1555 		/*
1556 		 * Respond to a previously pending Group-Specific
1557 		 * or Group-and-Source-Specific query by enqueueing
1558 		 * the appropriate Current-State report for
1559 		 * immediate transmission.
1560 		 */
1561 		if (query_response_timer_expired) {
1562 			int retval;
1563 
1564 			retval = mld_v2_enqueue_group_record(qrq, inm, 0, 1,
1565 			    (inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER),
1566 			    0);
1567 			CTR2(KTR_MLD, "%s: enqueue record = %d",
1568 			    __func__, retval);
1569 			inm->in6m_state = MLD_REPORTING_MEMBER;
1570 			in6m_clear_recorded(inm);
1571 		}
1572 		/* FALLTHROUGH */
1573 	case MLD_REPORTING_MEMBER:
1574 	case MLD_LEAVING_MEMBER:
1575 		if (state_change_retransmit_timer_expired) {
1576 			/*
1577 			 * State-change retransmission timer fired.
1578 			 * If there are any further pending retransmissions,
1579 			 * set the global pending state-change flag, and
1580 			 * reset the timer.
1581 			 */
1582 			if (--inm->in6m_scrv > 0) {
1583 				inm->in6m_sctimer = uri_fasthz;
1584 				V_state_change_timers_running6 = 1;
1585 			}
1586 			/*
1587 			 * Retransmit the previously computed state-change
1588 			 * report. If there are no further pending
1589 			 * retransmissions, the mbuf queue will be consumed.
1590 			 * Update T0 state to T1 as we have now sent
1591 			 * a state-change.
1592 			 */
1593 			(void)mld_v2_merge_state_changes(inm, scq);
1594 
1595 			in6m_commit(inm);
1596 			CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
1597 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1598 			    if_name(inm->in6m_ifp));
1599 
1600 			/*
1601 			 * If we are leaving the group for good, make sure
1602 			 * we release MLD's reference to it.
1603 			 * This release must be deferred using a SLIST,
1604 			 * as we are called from a loop which traverses
1605 			 * the in_ifmultiaddr TAILQ.
1606 			 */
1607 			if (inm->in6m_state == MLD_LEAVING_MEMBER &&
1608 			    inm->in6m_scrv == 0) {
1609 				inm->in6m_state = MLD_NOT_MEMBER;
1610 				in6m_disconnect_locked(inmh, inm);
1611 				in6m_rele_locked(inmh, inm);
1612 			}
1613 		}
1614 		break;
1615 	}
1616 }
1617 
1618 /*
1619  * Switch to a different version on the given interface,
1620  * as per Section 9.12.
1621  */
1622 static void
1623 mld_set_version(struct mld_ifsoftc *mli, const int version)
1624 {
1625 	int old_version_timer;
1626 
1627 	MLD_LOCK_ASSERT();
1628 
1629 	CTR4(KTR_MLD, "%s: switching to v%d on ifp %p(%s)", __func__,
1630 	    version, mli->mli_ifp, if_name(mli->mli_ifp));
1631 
1632 	if (version == MLD_VERSION_1) {
1633 		/*
1634 		 * Compute the "Older Version Querier Present" timer as per
1635 		 * Section 9.12.
1636 		 */
1637 		old_version_timer = (mli->mli_rv * mli->mli_qi) + mli->mli_qri;
1638 		old_version_timer *= PR_SLOWHZ;
1639 		mli->mli_v1_timer = old_version_timer;
1640 	}
1641 
1642 	if (mli->mli_v1_timer > 0 && mli->mli_version != MLD_VERSION_1) {
1643 		mli->mli_version = MLD_VERSION_1;
1644 		mld_v2_cancel_link_timers(mli);
1645 	}
1646 }
1647 
1648 /*
1649  * Cancel pending MLDv2 timers for the given link and all groups
1650  * joined on it; state-change, general-query, and group-query timers.
1651  */
1652 static void
1653 mld_v2_cancel_link_timers(struct mld_ifsoftc *mli)
1654 {
1655 	struct epoch_tracker	 et;
1656 	struct in6_multi_head	 inmh;
1657 	struct ifmultiaddr	*ifma;
1658 	struct ifnet		*ifp;
1659 	struct in6_multi	*inm;
1660 
1661 	CTR3(KTR_MLD, "%s: cancel v2 timers on ifp %p(%s)", __func__,
1662 	    mli->mli_ifp, if_name(mli->mli_ifp));
1663 
1664 	SLIST_INIT(&inmh);
1665 	IN6_MULTI_LIST_LOCK_ASSERT();
1666 	MLD_LOCK_ASSERT();
1667 
1668 	/*
1669 	 * Fast-track this potentially expensive operation
1670 	 * by checking all the global 'timer pending' flags.
1671 	 */
1672 	if (!V_interface_timers_running6 &&
1673 	    !V_state_change_timers_running6 &&
1674 	    !V_current_state_timers_running6)
1675 		return;
1676 
1677 	mli->mli_v2_timer = 0;
1678 
1679 	ifp = mli->mli_ifp;
1680 
1681 	IF_ADDR_WLOCK(ifp);
1682 	NET_EPOCH_ENTER(et);
1683 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1684 		inm = in6m_ifmultiaddr_get_inm(ifma);
1685 		if (inm == NULL)
1686 			continue;
1687 		switch (inm->in6m_state) {
1688 		case MLD_NOT_MEMBER:
1689 		case MLD_SILENT_MEMBER:
1690 		case MLD_IDLE_MEMBER:
1691 		case MLD_LAZY_MEMBER:
1692 		case MLD_SLEEPING_MEMBER:
1693 		case MLD_AWAKENING_MEMBER:
1694 			break;
1695 		case MLD_LEAVING_MEMBER:
1696 			/*
1697 			 * If we are leaving the group and switching
1698 			 * version, we need to release the final
1699 			 * reference held for issuing the INCLUDE {}.
1700 			 */
1701 			if (inm->in6m_refcount == 1)
1702 				in6m_disconnect_locked(&inmh, inm);
1703 			in6m_rele_locked(&inmh, inm);
1704 			/* FALLTHROUGH */
1705 		case MLD_G_QUERY_PENDING_MEMBER:
1706 		case MLD_SG_QUERY_PENDING_MEMBER:
1707 			in6m_clear_recorded(inm);
1708 			/* FALLTHROUGH */
1709 		case MLD_REPORTING_MEMBER:
1710 			inm->in6m_sctimer = 0;
1711 			inm->in6m_timer = 0;
1712 			inm->in6m_state = MLD_REPORTING_MEMBER;
1713 			/*
1714 			 * Free any pending MLDv2 state-change records.
1715 			 */
1716 			mbufq_drain(&inm->in6m_scq);
1717 			break;
1718 		}
1719 	}
1720 	NET_EPOCH_EXIT(et);
1721 	IF_ADDR_WUNLOCK(ifp);
1722 	in6m_release_list_deferred(&inmh);
1723 }
1724 
1725 /*
1726  * Global slowtimo handler.
1727  * VIMAGE: Timeout handlers are expected to service all vimages.
1728  */
1729 void
1730 mld_slowtimo(void)
1731 {
1732 	VNET_ITERATOR_DECL(vnet_iter);
1733 
1734 	VNET_LIST_RLOCK_NOSLEEP();
1735 	VNET_FOREACH(vnet_iter) {
1736 		CURVNET_SET(vnet_iter);
1737 		mld_slowtimo_vnet();
1738 		CURVNET_RESTORE();
1739 	}
1740 	VNET_LIST_RUNLOCK_NOSLEEP();
1741 }
1742 
1743 /*
1744  * Per-vnet slowtimo handler.
1745  */
1746 static void
1747 mld_slowtimo_vnet(void)
1748 {
1749 	struct mld_ifsoftc *mli;
1750 
1751 	MLD_LOCK();
1752 
1753 	LIST_FOREACH(mli, &V_mli_head, mli_link) {
1754 		mld_v1_process_querier_timers(mli);
1755 	}
1756 
1757 	MLD_UNLOCK();
1758 }
1759 
1760 /*
1761  * Update the Older Version Querier Present timers for a link.
1762  * See Section 9.12 of RFC 3810.
1763  */
1764 static void
1765 mld_v1_process_querier_timers(struct mld_ifsoftc *mli)
1766 {
1767 
1768 	MLD_LOCK_ASSERT();
1769 
1770 	if (mli->mli_version != MLD_VERSION_2 && --mli->mli_v1_timer == 0) {
1771 		/*
1772 		 * MLDv1 Querier Present timer expired; revert to MLDv2.
1773 		 */
1774 		CTR5(KTR_MLD,
1775 		    "%s: transition from v%d -> v%d on %p(%s)",
1776 		    __func__, mli->mli_version, MLD_VERSION_2,
1777 		    mli->mli_ifp, if_name(mli->mli_ifp));
1778 		mli->mli_version = MLD_VERSION_2;
1779 	}
1780 }
1781 
1782 /*
1783  * Transmit an MLDv1 report immediately.
1784  */
1785 static int
1786 mld_v1_transmit_report(struct in6_multi *in6m, const int type)
1787 {
1788 	struct ifnet		*ifp;
1789 	struct in6_ifaddr	*ia;
1790 	struct ip6_hdr		*ip6;
1791 	struct mbuf		*mh, *md;
1792 	struct mld_hdr		*mld;
1793 
1794 	IN6_MULTI_LIST_LOCK_ASSERT();
1795 	MLD_LOCK_ASSERT();
1796 
1797 	ifp = in6m->in6m_ifp;
1798 	/* in process of being freed */
1799 	if (ifp == NULL)
1800 		return (0);
1801 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
1802 	/* ia may be NULL if link-local address is tentative. */
1803 
1804 	mh = m_gethdr(M_NOWAIT, MT_DATA);
1805 	if (mh == NULL) {
1806 		if (ia != NULL)
1807 			ifa_free(&ia->ia_ifa);
1808 		return (ENOMEM);
1809 	}
1810 	md = m_get(M_NOWAIT, MT_DATA);
1811 	if (md == NULL) {
1812 		m_free(mh);
1813 		if (ia != NULL)
1814 			ifa_free(&ia->ia_ifa);
1815 		return (ENOMEM);
1816 	}
1817 	mh->m_next = md;
1818 
1819 	/*
1820 	 * FUTURE: Consider increasing alignment by ETHER_HDR_LEN, so
1821 	 * that ether_output() does not need to allocate another mbuf
1822 	 * for the header in the most common case.
1823 	 */
1824 	M_ALIGN(mh, sizeof(struct ip6_hdr));
1825 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) + sizeof(struct mld_hdr);
1826 	mh->m_len = sizeof(struct ip6_hdr);
1827 
1828 	ip6 = mtod(mh, struct ip6_hdr *);
1829 	ip6->ip6_flow = 0;
1830 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
1831 	ip6->ip6_vfc |= IPV6_VERSION;
1832 	ip6->ip6_nxt = IPPROTO_ICMPV6;
1833 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
1834 	ip6->ip6_dst = in6m->in6m_addr;
1835 
1836 	md->m_len = sizeof(struct mld_hdr);
1837 	mld = mtod(md, struct mld_hdr *);
1838 	mld->mld_type = type;
1839 	mld->mld_code = 0;
1840 	mld->mld_cksum = 0;
1841 	mld->mld_maxdelay = 0;
1842 	mld->mld_reserved = 0;
1843 	mld->mld_addr = in6m->in6m_addr;
1844 	in6_clearscope(&mld->mld_addr);
1845 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
1846 	    sizeof(struct ip6_hdr), sizeof(struct mld_hdr));
1847 
1848 	mld_save_context(mh, ifp);
1849 	mh->m_flags |= M_MLDV1;
1850 
1851 	mld_dispatch_packet(mh);
1852 
1853 	if (ia != NULL)
1854 		ifa_free(&ia->ia_ifa);
1855 	return (0);
1856 }
1857 
1858 /*
1859  * Process a state change from the upper layer for the given IPv6 group.
1860  *
1861  * Each socket holds a reference on the in_multi in its own ip_moptions.
1862  * The socket layer will have made the necessary updates to.the group
1863  * state, it is now up to MLD to issue a state change report if there
1864  * has been any change between T0 (when the last state-change was issued)
1865  * and T1 (now).
1866  *
1867  * We use the MLDv2 state machine at group level. The MLd module
1868  * however makes the decision as to which MLD protocol version to speak.
1869  * A state change *from* INCLUDE {} always means an initial join.
1870  * A state change *to* INCLUDE {} always means a final leave.
1871  *
1872  * If delay is non-zero, and the state change is an initial multicast
1873  * join, the state change report will be delayed by 'delay' ticks
1874  * in units of PR_FASTHZ if MLDv1 is active on the link; otherwise
1875  * the initial MLDv2 state change report will be delayed by whichever
1876  * is sooner, a pending state-change timer or delay itself.
1877  *
1878  * VIMAGE: curvnet should have been set by caller, as this routine
1879  * is called from the socket option handlers.
1880  */
1881 int
1882 mld_change_state(struct in6_multi *inm, const int delay)
1883 {
1884 	struct mld_ifsoftc *mli;
1885 	struct ifnet *ifp;
1886 	int error;
1887 
1888 	IN6_MULTI_LIST_LOCK_ASSERT();
1889 
1890 	error = 0;
1891 
1892 	/*
1893 	 * Check if the in6_multi has already been disconnected.
1894 	 */
1895 	if (inm->in6m_ifp == NULL) {
1896 		CTR1(KTR_MLD, "%s: inm is disconnected", __func__);
1897 		return (0);
1898 	}
1899 
1900 	/*
1901 	 * Try to detect if the upper layer just asked us to change state
1902 	 * for an interface which has now gone away.
1903 	 */
1904 	KASSERT(inm->in6m_ifma != NULL, ("%s: no ifma", __func__));
1905 	ifp = inm->in6m_ifma->ifma_ifp;
1906 	if (ifp == NULL)
1907 		return (0);
1908 	/*
1909 	 * Sanity check that netinet6's notion of ifp is the
1910 	 * same as net's.
1911 	 */
1912 	KASSERT(inm->in6m_ifp == ifp, ("%s: bad ifp", __func__));
1913 
1914 	MLD_LOCK();
1915 	mli = MLD_IFINFO(ifp);
1916 	KASSERT(mli != NULL, ("%s: no mld_ifsoftc for ifp %p", __func__, ifp));
1917 
1918 	/*
1919 	 * If we detect a state transition to or from MCAST_UNDEFINED
1920 	 * for this group, then we are starting or finishing an MLD
1921 	 * life cycle for this group.
1922 	 */
1923 	if (inm->in6m_st[1].iss_fmode != inm->in6m_st[0].iss_fmode) {
1924 		CTR3(KTR_MLD, "%s: inm transition %d -> %d", __func__,
1925 		    inm->in6m_st[0].iss_fmode, inm->in6m_st[1].iss_fmode);
1926 		if (inm->in6m_st[0].iss_fmode == MCAST_UNDEFINED) {
1927 			CTR1(KTR_MLD, "%s: initial join", __func__);
1928 			error = mld_initial_join(inm, mli, delay);
1929 			goto out_locked;
1930 		} else if (inm->in6m_st[1].iss_fmode == MCAST_UNDEFINED) {
1931 			CTR1(KTR_MLD, "%s: final leave", __func__);
1932 			mld_final_leave(inm, mli);
1933 			goto out_locked;
1934 		}
1935 	} else {
1936 		CTR1(KTR_MLD, "%s: filter set change", __func__);
1937 	}
1938 
1939 	error = mld_handle_state_change(inm, mli);
1940 
1941 out_locked:
1942 	MLD_UNLOCK();
1943 	return (error);
1944 }
1945 
1946 /*
1947  * Perform the initial join for an MLD group.
1948  *
1949  * When joining a group:
1950  *  If the group should have its MLD traffic suppressed, do nothing.
1951  *  MLDv1 starts sending MLDv1 host membership reports.
1952  *  MLDv2 will schedule an MLDv2 state-change report containing the
1953  *  initial state of the membership.
1954  *
1955  * If the delay argument is non-zero, then we must delay sending the
1956  * initial state change for delay ticks (in units of PR_FASTHZ).
1957  */
1958 static int
1959 mld_initial_join(struct in6_multi *inm, struct mld_ifsoftc *mli,
1960     const int delay)
1961 {
1962 	struct ifnet		*ifp;
1963 	struct mbufq		*mq;
1964 	int			 error, retval, syncstates;
1965 	int			 odelay;
1966 #ifdef KTR
1967 	char			 ip6tbuf[INET6_ADDRSTRLEN];
1968 #endif
1969 
1970 	CTR4(KTR_MLD, "%s: initial join %s on ifp %p(%s)",
1971 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
1972 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
1973 
1974 	error = 0;
1975 	syncstates = 1;
1976 
1977 	ifp = inm->in6m_ifp;
1978 
1979 	IN6_MULTI_LIST_LOCK_ASSERT();
1980 	MLD_LOCK_ASSERT();
1981 
1982 	KASSERT(mli && mli->mli_ifp == ifp, ("%s: inconsistent ifp", __func__));
1983 
1984 	/*
1985 	 * Groups joined on loopback or marked as 'not reported',
1986 	 * enter the MLD_SILENT_MEMBER state and
1987 	 * are never reported in any protocol exchanges.
1988 	 * All other groups enter the appropriate state machine
1989 	 * for the version in use on this link.
1990 	 * A link marked as MLIF_SILENT causes MLD to be completely
1991 	 * disabled for the link.
1992 	 */
1993 	if ((ifp->if_flags & IFF_LOOPBACK) ||
1994 	    (mli->mli_flags & MLIF_SILENT) ||
1995 	    !mld_is_addr_reported(&inm->in6m_addr)) {
1996 		CTR1(KTR_MLD,
1997 "%s: not kicking state machine for silent group", __func__);
1998 		inm->in6m_state = MLD_SILENT_MEMBER;
1999 		inm->in6m_timer = 0;
2000 	} else {
2001 		/*
2002 		 * Deal with overlapping in_multi lifecycle.
2003 		 * If this group was LEAVING, then make sure
2004 		 * we drop the reference we picked up to keep the
2005 		 * group around for the final INCLUDE {} enqueue.
2006 		 */
2007 		if (mli->mli_version == MLD_VERSION_2 &&
2008 		    inm->in6m_state == MLD_LEAVING_MEMBER) {
2009 			inm->in6m_refcount--;
2010 			MPASS(inm->in6m_refcount > 0);
2011 		}
2012 		inm->in6m_state = MLD_REPORTING_MEMBER;
2013 
2014 		switch (mli->mli_version) {
2015 		case MLD_VERSION_1:
2016 			/*
2017 			 * If a delay was provided, only use it if
2018 			 * it is greater than the delay normally
2019 			 * used for an MLDv1 state change report,
2020 			 * and delay sending the initial MLDv1 report
2021 			 * by not transitioning to the IDLE state.
2022 			 */
2023 			odelay = MLD_RANDOM_DELAY(MLD_V1_MAX_RI * PR_FASTHZ);
2024 			if (delay) {
2025 				inm->in6m_timer = max(delay, odelay);
2026 				V_current_state_timers_running6 = 1;
2027 			} else {
2028 				inm->in6m_state = MLD_IDLE_MEMBER;
2029 				error = mld_v1_transmit_report(inm,
2030 				     MLD_LISTENER_REPORT);
2031 				if (error == 0) {
2032 					inm->in6m_timer = odelay;
2033 					V_current_state_timers_running6 = 1;
2034 				}
2035 			}
2036 			break;
2037 
2038 		case MLD_VERSION_2:
2039 			/*
2040 			 * Defer update of T0 to T1, until the first copy
2041 			 * of the state change has been transmitted.
2042 			 */
2043 			syncstates = 0;
2044 
2045 			/*
2046 			 * Immediately enqueue a State-Change Report for
2047 			 * this interface, freeing any previous reports.
2048 			 * Don't kick the timers if there is nothing to do,
2049 			 * or if an error occurred.
2050 			 */
2051 			mq = &inm->in6m_scq;
2052 			mbufq_drain(mq);
2053 			retval = mld_v2_enqueue_group_record(mq, inm, 1,
2054 			    0, 0, (mli->mli_flags & MLIF_USEALLOW));
2055 			CTR2(KTR_MLD, "%s: enqueue record = %d",
2056 			    __func__, retval);
2057 			if (retval <= 0) {
2058 				error = retval * -1;
2059 				break;
2060 			}
2061 
2062 			/*
2063 			 * Schedule transmission of pending state-change
2064 			 * report up to RV times for this link. The timer
2065 			 * will fire at the next mld_fasttimo (~200ms),
2066 			 * giving us an opportunity to merge the reports.
2067 			 *
2068 			 * If a delay was provided to this function, only
2069 			 * use this delay if sooner than the existing one.
2070 			 */
2071 			KASSERT(mli->mli_rv > 1,
2072 			   ("%s: invalid robustness %d", __func__,
2073 			    mli->mli_rv));
2074 			inm->in6m_scrv = mli->mli_rv;
2075 			if (delay) {
2076 				if (inm->in6m_sctimer > 1) {
2077 					inm->in6m_sctimer =
2078 					    min(inm->in6m_sctimer, delay);
2079 				} else
2080 					inm->in6m_sctimer = delay;
2081 			} else
2082 				inm->in6m_sctimer = 1;
2083 			V_state_change_timers_running6 = 1;
2084 
2085 			error = 0;
2086 			break;
2087 		}
2088 	}
2089 
2090 	/*
2091 	 * Only update the T0 state if state change is atomic,
2092 	 * i.e. we don't need to wait for a timer to fire before we
2093 	 * can consider the state change to have been communicated.
2094 	 */
2095 	if (syncstates) {
2096 		in6m_commit(inm);
2097 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2098 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2099 		    if_name(inm->in6m_ifp));
2100 	}
2101 
2102 	return (error);
2103 }
2104 
2105 /*
2106  * Issue an intermediate state change during the life-cycle.
2107  */
2108 static int
2109 mld_handle_state_change(struct in6_multi *inm, struct mld_ifsoftc *mli)
2110 {
2111 	struct ifnet		*ifp;
2112 	int			 retval;
2113 #ifdef KTR
2114 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2115 #endif
2116 
2117 	CTR4(KTR_MLD, "%s: state change for %s on ifp %p(%s)",
2118 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2119 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2120 
2121 	ifp = inm->in6m_ifp;
2122 
2123 	IN6_MULTI_LIST_LOCK_ASSERT();
2124 	MLD_LOCK_ASSERT();
2125 
2126 	KASSERT(mli && mli->mli_ifp == ifp,
2127 	    ("%s: inconsistent ifp", __func__));
2128 
2129 	if ((ifp->if_flags & IFF_LOOPBACK) ||
2130 	    (mli->mli_flags & MLIF_SILENT) ||
2131 	    !mld_is_addr_reported(&inm->in6m_addr) ||
2132 	    (mli->mli_version != MLD_VERSION_2)) {
2133 		if (!mld_is_addr_reported(&inm->in6m_addr)) {
2134 			CTR1(KTR_MLD,
2135 "%s: not kicking state machine for silent group", __func__);
2136 		}
2137 		CTR1(KTR_MLD, "%s: nothing to do", __func__);
2138 		in6m_commit(inm);
2139 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2140 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2141 		    if_name(inm->in6m_ifp));
2142 		return (0);
2143 	}
2144 
2145 	mbufq_drain(&inm->in6m_scq);
2146 
2147 	retval = mld_v2_enqueue_group_record(&inm->in6m_scq, inm, 1, 0, 0,
2148 	    (mli->mli_flags & MLIF_USEALLOW));
2149 	CTR2(KTR_MLD, "%s: enqueue record = %d", __func__, retval);
2150 	if (retval <= 0)
2151 		return (-retval);
2152 
2153 	/*
2154 	 * If record(s) were enqueued, start the state-change
2155 	 * report timer for this group.
2156 	 */
2157 	inm->in6m_scrv = mli->mli_rv;
2158 	inm->in6m_sctimer = 1;
2159 	V_state_change_timers_running6 = 1;
2160 
2161 	return (0);
2162 }
2163 
2164 /*
2165  * Perform the final leave for a multicast address.
2166  *
2167  * When leaving a group:
2168  *  MLDv1 sends a DONE message, if and only if we are the reporter.
2169  *  MLDv2 enqueues a state-change report containing a transition
2170  *  to INCLUDE {} for immediate transmission.
2171  */
2172 static void
2173 mld_final_leave(struct in6_multi *inm, struct mld_ifsoftc *mli)
2174 {
2175 	int syncstates;
2176 #ifdef KTR
2177 	char ip6tbuf[INET6_ADDRSTRLEN];
2178 #endif
2179 
2180 	syncstates = 1;
2181 
2182 	CTR4(KTR_MLD, "%s: final leave %s on ifp %p(%s)",
2183 	    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2184 	    inm->in6m_ifp, if_name(inm->in6m_ifp));
2185 
2186 	IN6_MULTI_LIST_LOCK_ASSERT();
2187 	MLD_LOCK_ASSERT();
2188 
2189 	switch (inm->in6m_state) {
2190 	case MLD_NOT_MEMBER:
2191 	case MLD_SILENT_MEMBER:
2192 	case MLD_LEAVING_MEMBER:
2193 		/* Already leaving or left; do nothing. */
2194 		CTR1(KTR_MLD,
2195 "%s: not kicking state machine for silent group", __func__);
2196 		break;
2197 	case MLD_REPORTING_MEMBER:
2198 	case MLD_IDLE_MEMBER:
2199 	case MLD_G_QUERY_PENDING_MEMBER:
2200 	case MLD_SG_QUERY_PENDING_MEMBER:
2201 		if (mli->mli_version == MLD_VERSION_1) {
2202 #ifdef INVARIANTS
2203 			if (inm->in6m_state == MLD_G_QUERY_PENDING_MEMBER ||
2204 			    inm->in6m_state == MLD_SG_QUERY_PENDING_MEMBER)
2205 			panic("%s: MLDv2 state reached, not MLDv2 mode",
2206 			     __func__);
2207 #endif
2208 			mld_v1_transmit_report(inm, MLD_LISTENER_DONE);
2209 			inm->in6m_state = MLD_NOT_MEMBER;
2210 			V_current_state_timers_running6 = 1;
2211 		} else if (mli->mli_version == MLD_VERSION_2) {
2212 			/*
2213 			 * Stop group timer and all pending reports.
2214 			 * Immediately enqueue a state-change report
2215 			 * TO_IN {} to be sent on the next fast timeout,
2216 			 * giving us an opportunity to merge reports.
2217 			 */
2218 			mbufq_drain(&inm->in6m_scq);
2219 			inm->in6m_timer = 0;
2220 			inm->in6m_scrv = mli->mli_rv;
2221 			CTR4(KTR_MLD, "%s: Leaving %s/%s with %d "
2222 			    "pending retransmissions.", __func__,
2223 			    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2224 			    if_name(inm->in6m_ifp), inm->in6m_scrv);
2225 			if (inm->in6m_scrv == 0) {
2226 				inm->in6m_state = MLD_NOT_MEMBER;
2227 				inm->in6m_sctimer = 0;
2228 			} else {
2229 				int retval;
2230 
2231 				in6m_acquire_locked(inm);
2232 
2233 				retval = mld_v2_enqueue_group_record(
2234 				    &inm->in6m_scq, inm, 1, 0, 0,
2235 				    (mli->mli_flags & MLIF_USEALLOW));
2236 				KASSERT(retval != 0,
2237 				    ("%s: enqueue record = %d", __func__,
2238 				     retval));
2239 
2240 				inm->in6m_state = MLD_LEAVING_MEMBER;
2241 				inm->in6m_sctimer = 1;
2242 				V_state_change_timers_running6 = 1;
2243 				syncstates = 0;
2244 			}
2245 			break;
2246 		}
2247 		break;
2248 	case MLD_LAZY_MEMBER:
2249 	case MLD_SLEEPING_MEMBER:
2250 	case MLD_AWAKENING_MEMBER:
2251 		/* Our reports are suppressed; do nothing. */
2252 		break;
2253 	}
2254 
2255 	if (syncstates) {
2256 		in6m_commit(inm);
2257 		CTR3(KTR_MLD, "%s: T1 -> T0 for %s/%s", __func__,
2258 		    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2259 		    if_name(inm->in6m_ifp));
2260 		inm->in6m_st[1].iss_fmode = MCAST_UNDEFINED;
2261 		CTR3(KTR_MLD, "%s: T1 now MCAST_UNDEFINED for %p/%s",
2262 		    __func__, &inm->in6m_addr, if_name(inm->in6m_ifp));
2263 	}
2264 }
2265 
2266 /*
2267  * Enqueue an MLDv2 group record to the given output queue.
2268  *
2269  * If is_state_change is zero, a current-state record is appended.
2270  * If is_state_change is non-zero, a state-change report is appended.
2271  *
2272  * If is_group_query is non-zero, an mbuf packet chain is allocated.
2273  * If is_group_query is zero, and if there is a packet with free space
2274  * at the tail of the queue, it will be appended to providing there
2275  * is enough free space.
2276  * Otherwise a new mbuf packet chain is allocated.
2277  *
2278  * If is_source_query is non-zero, each source is checked to see if
2279  * it was recorded for a Group-Source query, and will be omitted if
2280  * it is not both in-mode and recorded.
2281  *
2282  * If use_block_allow is non-zero, state change reports for initial join
2283  * and final leave, on an inclusive mode group with a source list, will be
2284  * rewritten to use the ALLOW_NEW and BLOCK_OLD record types, respectively.
2285  *
2286  * The function will attempt to allocate leading space in the packet
2287  * for the IPv6+ICMP headers to be prepended without fragmenting the chain.
2288  *
2289  * If successful the size of all data appended to the queue is returned,
2290  * otherwise an error code less than zero is returned, or zero if
2291  * no record(s) were appended.
2292  */
2293 static int
2294 mld_v2_enqueue_group_record(struct mbufq *mq, struct in6_multi *inm,
2295     const int is_state_change, const int is_group_query,
2296     const int is_source_query, const int use_block_allow)
2297 {
2298 	struct mldv2_record	 mr;
2299 	struct mldv2_record	*pmr;
2300 	struct ifnet		*ifp;
2301 	struct ip6_msource	*ims, *nims;
2302 	struct mbuf		*m0, *m, *md;
2303 	int			 is_filter_list_change;
2304 	int			 minrec0len, m0srcs, msrcs, nbytes, off;
2305 	int			 record_has_sources;
2306 	int			 now;
2307 	int			 type;
2308 	uint8_t			 mode;
2309 #ifdef KTR
2310 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2311 #endif
2312 
2313 	IN6_MULTI_LIST_LOCK_ASSERT();
2314 
2315 	ifp = inm->in6m_ifp;
2316 	is_filter_list_change = 0;
2317 	m = NULL;
2318 	m0 = NULL;
2319 	m0srcs = 0;
2320 	msrcs = 0;
2321 	nbytes = 0;
2322 	nims = NULL;
2323 	record_has_sources = 1;
2324 	pmr = NULL;
2325 	type = MLD_DO_NOTHING;
2326 	mode = inm->in6m_st[1].iss_fmode;
2327 
2328 	/*
2329 	 * If we did not transition out of ASM mode during t0->t1,
2330 	 * and there are no source nodes to process, we can skip
2331 	 * the generation of source records.
2332 	 */
2333 	if (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0 &&
2334 	    inm->in6m_nsrc == 0)
2335 		record_has_sources = 0;
2336 
2337 	if (is_state_change) {
2338 		/*
2339 		 * Queue a state change record.
2340 		 * If the mode did not change, and there are non-ASM
2341 		 * listeners or source filters present,
2342 		 * we potentially need to issue two records for the group.
2343 		 * If there are ASM listeners, and there was no filter
2344 		 * mode transition of any kind, do nothing.
2345 		 *
2346 		 * If we are transitioning to MCAST_UNDEFINED, we need
2347 		 * not send any sources. A transition to/from this state is
2348 		 * considered inclusive with some special treatment.
2349 		 *
2350 		 * If we are rewriting initial joins/leaves to use
2351 		 * ALLOW/BLOCK, and the group's membership is inclusive,
2352 		 * we need to send sources in all cases.
2353 		 */
2354 		if (mode != inm->in6m_st[0].iss_fmode) {
2355 			if (mode == MCAST_EXCLUDE) {
2356 				CTR1(KTR_MLD, "%s: change to EXCLUDE",
2357 				    __func__);
2358 				type = MLD_CHANGE_TO_EXCLUDE_MODE;
2359 			} else {
2360 				CTR1(KTR_MLD, "%s: change to INCLUDE",
2361 				    __func__);
2362 				if (use_block_allow) {
2363 					/*
2364 					 * XXX
2365 					 * Here we're interested in state
2366 					 * edges either direction between
2367 					 * MCAST_UNDEFINED and MCAST_INCLUDE.
2368 					 * Perhaps we should just check
2369 					 * the group state, rather than
2370 					 * the filter mode.
2371 					 */
2372 					if (mode == MCAST_UNDEFINED) {
2373 						type = MLD_BLOCK_OLD_SOURCES;
2374 					} else {
2375 						type = MLD_ALLOW_NEW_SOURCES;
2376 					}
2377 				} else {
2378 					type = MLD_CHANGE_TO_INCLUDE_MODE;
2379 					if (mode == MCAST_UNDEFINED)
2380 						record_has_sources = 0;
2381 				}
2382 			}
2383 		} else {
2384 			if (record_has_sources) {
2385 				is_filter_list_change = 1;
2386 			} else {
2387 				type = MLD_DO_NOTHING;
2388 			}
2389 		}
2390 	} else {
2391 		/*
2392 		 * Queue a current state record.
2393 		 */
2394 		if (mode == MCAST_EXCLUDE) {
2395 			type = MLD_MODE_IS_EXCLUDE;
2396 		} else if (mode == MCAST_INCLUDE) {
2397 			type = MLD_MODE_IS_INCLUDE;
2398 			KASSERT(inm->in6m_st[1].iss_asm == 0,
2399 			    ("%s: inm %p is INCLUDE but ASM count is %d",
2400 			     __func__, inm, inm->in6m_st[1].iss_asm));
2401 		}
2402 	}
2403 
2404 	/*
2405 	 * Generate the filter list changes using a separate function.
2406 	 */
2407 	if (is_filter_list_change)
2408 		return (mld_v2_enqueue_filter_change(mq, inm));
2409 
2410 	if (type == MLD_DO_NOTHING) {
2411 		CTR3(KTR_MLD, "%s: nothing to do for %s/%s",
2412 		    __func__, ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2413 		    if_name(inm->in6m_ifp));
2414 		return (0);
2415 	}
2416 
2417 	/*
2418 	 * If any sources are present, we must be able to fit at least
2419 	 * one in the trailing space of the tail packet's mbuf,
2420 	 * ideally more.
2421 	 */
2422 	minrec0len = sizeof(struct mldv2_record);
2423 	if (record_has_sources)
2424 		minrec0len += sizeof(struct in6_addr);
2425 
2426 	CTR4(KTR_MLD, "%s: queueing %s for %s/%s", __func__,
2427 	    mld_rec_type_to_str(type),
2428 	    ip6_sprintf(ip6tbuf, &inm->in6m_addr),
2429 	    if_name(inm->in6m_ifp));
2430 
2431 	/*
2432 	 * Check if we have a packet in the tail of the queue for this
2433 	 * group into which the first group record for this group will fit.
2434 	 * Otherwise allocate a new packet.
2435 	 * Always allocate leading space for IP6+RA+ICMPV6+REPORT.
2436 	 * Note: Group records for G/GSR query responses MUST be sent
2437 	 * in their own packet.
2438 	 */
2439 	m0 = mbufq_last(mq);
2440 	if (!is_group_query &&
2441 	    m0 != NULL &&
2442 	    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= MLD_V2_REPORT_MAXRECS) &&
2443 	    (m0->m_pkthdr.len + minrec0len) <
2444 	     (ifp->if_mtu - MLD_MTUSPACE)) {
2445 		m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2446 			    sizeof(struct mldv2_record)) /
2447 			    sizeof(struct in6_addr);
2448 		m = m0;
2449 		CTR1(KTR_MLD, "%s: use existing packet", __func__);
2450 	} else {
2451 		if (mbufq_full(mq)) {
2452 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2453 			return (-ENOMEM);
2454 		}
2455 		m = NULL;
2456 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2457 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2458 		if (!is_state_change && !is_group_query)
2459 			m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2460 		if (m == NULL)
2461 			m = m_gethdr(M_NOWAIT, MT_DATA);
2462 		if (m == NULL)
2463 			return (-ENOMEM);
2464 
2465 		mld_save_context(m, ifp);
2466 
2467 		CTR1(KTR_MLD, "%s: allocated first packet", __func__);
2468 	}
2469 
2470 	/*
2471 	 * Append group record.
2472 	 * If we have sources, we don't know how many yet.
2473 	 */
2474 	mr.mr_type = type;
2475 	mr.mr_datalen = 0;
2476 	mr.mr_numsrc = 0;
2477 	mr.mr_addr = inm->in6m_addr;
2478 	in6_clearscope(&mr.mr_addr);
2479 	if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2480 		if (m != m0)
2481 			m_freem(m);
2482 		CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2483 		return (-ENOMEM);
2484 	}
2485 	nbytes += sizeof(struct mldv2_record);
2486 
2487 	/*
2488 	 * Append as many sources as will fit in the first packet.
2489 	 * If we are appending to a new packet, the chain allocation
2490 	 * may potentially use clusters; use m_getptr() in this case.
2491 	 * If we are appending to an existing packet, we need to obtain
2492 	 * a pointer to the group record after m_append(), in case a new
2493 	 * mbuf was allocated.
2494 	 *
2495 	 * Only append sources which are in-mode at t1. If we are
2496 	 * transitioning to MCAST_UNDEFINED state on the group, and
2497 	 * use_block_allow is zero, do not include source entries.
2498 	 * Otherwise, we need to include this source in the report.
2499 	 *
2500 	 * Only report recorded sources in our filter set when responding
2501 	 * to a group-source query.
2502 	 */
2503 	if (record_has_sources) {
2504 		if (m == m0) {
2505 			md = m_last(m);
2506 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2507 			    md->m_len - nbytes);
2508 		} else {
2509 			md = m_getptr(m, 0, &off);
2510 			pmr = (struct mldv2_record *)(mtod(md, uint8_t *) +
2511 			    off);
2512 		}
2513 		msrcs = 0;
2514 		RB_FOREACH_SAFE(ims, ip6_msource_tree, &inm->in6m_srcs,
2515 		    nims) {
2516 			CTR2(KTR_MLD, "%s: visit node %s", __func__,
2517 			    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2518 			now = im6s_get_mode(inm, ims, 1);
2519 			CTR2(KTR_MLD, "%s: node is %d", __func__, now);
2520 			if ((now != mode) ||
2521 			    (now == mode &&
2522 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2523 				CTR1(KTR_MLD, "%s: skip node", __func__);
2524 				continue;
2525 			}
2526 			if (is_source_query && ims->im6s_stp == 0) {
2527 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2528 				    __func__);
2529 				continue;
2530 			}
2531 			CTR1(KTR_MLD, "%s: append node", __func__);
2532 			if (!m_append(m, sizeof(struct in6_addr),
2533 			    (void *)&ims->im6s_addr)) {
2534 				if (m != m0)
2535 					m_freem(m);
2536 				CTR1(KTR_MLD, "%s: m_append() failed.",
2537 				    __func__);
2538 				return (-ENOMEM);
2539 			}
2540 			nbytes += sizeof(struct in6_addr);
2541 			++msrcs;
2542 			if (msrcs == m0srcs)
2543 				break;
2544 		}
2545 		CTR2(KTR_MLD, "%s: msrcs is %d this packet", __func__,
2546 		    msrcs);
2547 		pmr->mr_numsrc = htons(msrcs);
2548 		nbytes += (msrcs * sizeof(struct in6_addr));
2549 	}
2550 
2551 	if (is_source_query && msrcs == 0) {
2552 		CTR1(KTR_MLD, "%s: no recorded sources to report", __func__);
2553 		if (m != m0)
2554 			m_freem(m);
2555 		return (0);
2556 	}
2557 
2558 	/*
2559 	 * We are good to go with first packet.
2560 	 */
2561 	if (m != m0) {
2562 		CTR1(KTR_MLD, "%s: enqueueing first packet", __func__);
2563 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2564 		mbufq_enqueue(mq, m);
2565 	} else
2566 		m->m_pkthdr.PH_vt.vt_nrecs++;
2567 
2568 	/*
2569 	 * No further work needed if no source list in packet(s).
2570 	 */
2571 	if (!record_has_sources)
2572 		return (nbytes);
2573 
2574 	/*
2575 	 * Whilst sources remain to be announced, we need to allocate
2576 	 * a new packet and fill out as many sources as will fit.
2577 	 * Always try for a cluster first.
2578 	 */
2579 	while (nims != NULL) {
2580 		if (mbufq_full(mq)) {
2581 			CTR1(KTR_MLD, "%s: outbound queue full", __func__);
2582 			return (-ENOMEM);
2583 		}
2584 		m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2585 		if (m == NULL)
2586 			m = m_gethdr(M_NOWAIT, MT_DATA);
2587 		if (m == NULL)
2588 			return (-ENOMEM);
2589 		mld_save_context(m, ifp);
2590 		md = m_getptr(m, 0, &off);
2591 		pmr = (struct mldv2_record *)(mtod(md, uint8_t *) + off);
2592 		CTR1(KTR_MLD, "%s: allocated next packet", __func__);
2593 
2594 		if (!m_append(m, sizeof(struct mldv2_record), (void *)&mr)) {
2595 			if (m != m0)
2596 				m_freem(m);
2597 			CTR1(KTR_MLD, "%s: m_append() failed.", __func__);
2598 			return (-ENOMEM);
2599 		}
2600 		m->m_pkthdr.PH_vt.vt_nrecs = 1;
2601 		nbytes += sizeof(struct mldv2_record);
2602 
2603 		m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2604 		    sizeof(struct mldv2_record)) / sizeof(struct in6_addr);
2605 
2606 		msrcs = 0;
2607 		RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2608 			CTR2(KTR_MLD, "%s: visit node %s",
2609 			    __func__, ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2610 			now = im6s_get_mode(inm, ims, 1);
2611 			if ((now != mode) ||
2612 			    (now == mode &&
2613 			     (!use_block_allow && mode == MCAST_UNDEFINED))) {
2614 				CTR1(KTR_MLD, "%s: skip node", __func__);
2615 				continue;
2616 			}
2617 			if (is_source_query && ims->im6s_stp == 0) {
2618 				CTR1(KTR_MLD, "%s: skip unrecorded node",
2619 				    __func__);
2620 				continue;
2621 			}
2622 			CTR1(KTR_MLD, "%s: append node", __func__);
2623 			if (!m_append(m, sizeof(struct in6_addr),
2624 			    (void *)&ims->im6s_addr)) {
2625 				if (m != m0)
2626 					m_freem(m);
2627 				CTR1(KTR_MLD, "%s: m_append() failed.",
2628 				    __func__);
2629 				return (-ENOMEM);
2630 			}
2631 			++msrcs;
2632 			if (msrcs == m0srcs)
2633 				break;
2634 		}
2635 		pmr->mr_numsrc = htons(msrcs);
2636 		nbytes += (msrcs * sizeof(struct in6_addr));
2637 
2638 		CTR1(KTR_MLD, "%s: enqueueing next packet", __func__);
2639 		mbufq_enqueue(mq, m);
2640 	}
2641 
2642 	return (nbytes);
2643 }
2644 
2645 /*
2646  * Type used to mark record pass completion.
2647  * We exploit the fact we can cast to this easily from the
2648  * current filter modes on each ip_msource node.
2649  */
2650 typedef enum {
2651 	REC_NONE = 0x00,	/* MCAST_UNDEFINED */
2652 	REC_ALLOW = 0x01,	/* MCAST_INCLUDE */
2653 	REC_BLOCK = 0x02,	/* MCAST_EXCLUDE */
2654 	REC_FULL = REC_ALLOW | REC_BLOCK
2655 } rectype_t;
2656 
2657 /*
2658  * Enqueue an MLDv2 filter list change to the given output queue.
2659  *
2660  * Source list filter state is held in an RB-tree. When the filter list
2661  * for a group is changed without changing its mode, we need to compute
2662  * the deltas between T0 and T1 for each source in the filter set,
2663  * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records.
2664  *
2665  * As we may potentially queue two record types, and the entire R-B tree
2666  * needs to be walked at once, we break this out into its own function
2667  * so we can generate a tightly packed queue of packets.
2668  *
2669  * XXX This could be written to only use one tree walk, although that makes
2670  * serializing into the mbuf chains a bit harder. For now we do two walks
2671  * which makes things easier on us, and it may or may not be harder on
2672  * the L2 cache.
2673  *
2674  * If successful the size of all data appended to the queue is returned,
2675  * otherwise an error code less than zero is returned, or zero if
2676  * no record(s) were appended.
2677  */
2678 static int
2679 mld_v2_enqueue_filter_change(struct mbufq *mq, struct in6_multi *inm)
2680 {
2681 	static const int MINRECLEN =
2682 	    sizeof(struct mldv2_record) + sizeof(struct in6_addr);
2683 	struct ifnet		*ifp;
2684 	struct mldv2_record	 mr;
2685 	struct mldv2_record	*pmr;
2686 	struct ip6_msource	*ims, *nims;
2687 	struct mbuf		*m, *m0, *md;
2688 	int			 m0srcs, nbytes, npbytes, off, rsrcs, schanged;
2689 	int			 nallow, nblock;
2690 	uint8_t			 mode, now, then;
2691 	rectype_t		 crt, drt, nrt;
2692 #ifdef KTR
2693 	char			 ip6tbuf[INET6_ADDRSTRLEN];
2694 #endif
2695 
2696 	IN6_MULTI_LIST_LOCK_ASSERT();
2697 
2698 	if (inm->in6m_nsrc == 0 ||
2699 	    (inm->in6m_st[0].iss_asm > 0 && inm->in6m_st[1].iss_asm > 0))
2700 		return (0);
2701 
2702 	ifp = inm->in6m_ifp;			/* interface */
2703 	mode = inm->in6m_st[1].iss_fmode;	/* filter mode at t1 */
2704 	crt = REC_NONE;	/* current group record type */
2705 	drt = REC_NONE;	/* mask of completed group record types */
2706 	nrt = REC_NONE;	/* record type for current node */
2707 	m0srcs = 0;	/* # source which will fit in current mbuf chain */
2708 	npbytes = 0;	/* # of bytes appended this packet */
2709 	nbytes = 0;	/* # of bytes appended to group's state-change queue */
2710 	rsrcs = 0;	/* # sources encoded in current record */
2711 	schanged = 0;	/* # nodes encoded in overall filter change */
2712 	nallow = 0;	/* # of source entries in ALLOW_NEW */
2713 	nblock = 0;	/* # of source entries in BLOCK_OLD */
2714 	nims = NULL;	/* next tree node pointer */
2715 
2716 	/*
2717 	 * For each possible filter record mode.
2718 	 * The first kind of source we encounter tells us which
2719 	 * is the first kind of record we start appending.
2720 	 * If a node transitioned to UNDEFINED at t1, its mode is treated
2721 	 * as the inverse of the group's filter mode.
2722 	 */
2723 	while (drt != REC_FULL) {
2724 		do {
2725 			m0 = mbufq_last(mq);
2726 			if (m0 != NULL &&
2727 			    (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <=
2728 			     MLD_V2_REPORT_MAXRECS) &&
2729 			    (m0->m_pkthdr.len + MINRECLEN) <
2730 			     (ifp->if_mtu - MLD_MTUSPACE)) {
2731 				m = m0;
2732 				m0srcs = (ifp->if_mtu - m0->m_pkthdr.len -
2733 					    sizeof(struct mldv2_record)) /
2734 					    sizeof(struct in6_addr);
2735 				CTR1(KTR_MLD,
2736 				    "%s: use previous packet", __func__);
2737 			} else {
2738 				m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR);
2739 				if (m == NULL)
2740 					m = m_gethdr(M_NOWAIT, MT_DATA);
2741 				if (m == NULL) {
2742 					CTR1(KTR_MLD,
2743 					    "%s: m_get*() failed", __func__);
2744 					return (-ENOMEM);
2745 				}
2746 				m->m_pkthdr.PH_vt.vt_nrecs = 0;
2747 				mld_save_context(m, ifp);
2748 				m0srcs = (ifp->if_mtu - MLD_MTUSPACE -
2749 				    sizeof(struct mldv2_record)) /
2750 				    sizeof(struct in6_addr);
2751 				npbytes = 0;
2752 				CTR1(KTR_MLD,
2753 				    "%s: allocated new packet", __func__);
2754 			}
2755 			/*
2756 			 * Append the MLD group record header to the
2757 			 * current packet's data area.
2758 			 * Recalculate pointer to free space for next
2759 			 * group record, in case m_append() allocated
2760 			 * a new mbuf or cluster.
2761 			 */
2762 			memset(&mr, 0, sizeof(mr));
2763 			mr.mr_addr = inm->in6m_addr;
2764 			in6_clearscope(&mr.mr_addr);
2765 			if (!m_append(m, sizeof(mr), (void *)&mr)) {
2766 				if (m != m0)
2767 					m_freem(m);
2768 				CTR1(KTR_MLD,
2769 				    "%s: m_append() failed", __func__);
2770 				return (-ENOMEM);
2771 			}
2772 			npbytes += sizeof(struct mldv2_record);
2773 			if (m != m0) {
2774 				/* new packet; offset in chain */
2775 				md = m_getptr(m, npbytes -
2776 				    sizeof(struct mldv2_record), &off);
2777 				pmr = (struct mldv2_record *)(mtod(md,
2778 				    uint8_t *) + off);
2779 			} else {
2780 				/* current packet; offset from last append */
2781 				md = m_last(m);
2782 				pmr = (struct mldv2_record *)(mtod(md,
2783 				    uint8_t *) + md->m_len -
2784 				    sizeof(struct mldv2_record));
2785 			}
2786 			/*
2787 			 * Begin walking the tree for this record type
2788 			 * pass, or continue from where we left off
2789 			 * previously if we had to allocate a new packet.
2790 			 * Only report deltas in-mode at t1.
2791 			 * We need not report included sources as allowed
2792 			 * if we are in inclusive mode on the group,
2793 			 * however the converse is not true.
2794 			 */
2795 			rsrcs = 0;
2796 			if (nims == NULL) {
2797 				nims = RB_MIN(ip6_msource_tree,
2798 				    &inm->in6m_srcs);
2799 			}
2800 			RB_FOREACH_FROM(ims, ip6_msource_tree, nims) {
2801 				CTR2(KTR_MLD, "%s: visit node %s", __func__,
2802 				    ip6_sprintf(ip6tbuf, &ims->im6s_addr));
2803 				now = im6s_get_mode(inm, ims, 1);
2804 				then = im6s_get_mode(inm, ims, 0);
2805 				CTR3(KTR_MLD, "%s: mode: t0 %d, t1 %d",
2806 				    __func__, then, now);
2807 				if (now == then) {
2808 					CTR1(KTR_MLD,
2809 					    "%s: skip unchanged", __func__);
2810 					continue;
2811 				}
2812 				if (mode == MCAST_EXCLUDE &&
2813 				    now == MCAST_INCLUDE) {
2814 					CTR1(KTR_MLD,
2815 					    "%s: skip IN src on EX group",
2816 					    __func__);
2817 					continue;
2818 				}
2819 				nrt = (rectype_t)now;
2820 				if (nrt == REC_NONE)
2821 					nrt = (rectype_t)(~mode & REC_FULL);
2822 				if (schanged++ == 0) {
2823 					crt = nrt;
2824 				} else if (crt != nrt)
2825 					continue;
2826 				if (!m_append(m, sizeof(struct in6_addr),
2827 				    (void *)&ims->im6s_addr)) {
2828 					if (m != m0)
2829 						m_freem(m);
2830 					CTR1(KTR_MLD,
2831 					    "%s: m_append() failed", __func__);
2832 					return (-ENOMEM);
2833 				}
2834 				nallow += !!(crt == REC_ALLOW);
2835 				nblock += !!(crt == REC_BLOCK);
2836 				if (++rsrcs == m0srcs)
2837 					break;
2838 			}
2839 			/*
2840 			 * If we did not append any tree nodes on this
2841 			 * pass, back out of allocations.
2842 			 */
2843 			if (rsrcs == 0) {
2844 				npbytes -= sizeof(struct mldv2_record);
2845 				if (m != m0) {
2846 					CTR1(KTR_MLD,
2847 					    "%s: m_free(m)", __func__);
2848 					m_freem(m);
2849 				} else {
2850 					CTR1(KTR_MLD,
2851 					    "%s: m_adj(m, -mr)", __func__);
2852 					m_adj(m, -((int)sizeof(
2853 					    struct mldv2_record)));
2854 				}
2855 				continue;
2856 			}
2857 			npbytes += (rsrcs * sizeof(struct in6_addr));
2858 			if (crt == REC_ALLOW)
2859 				pmr->mr_type = MLD_ALLOW_NEW_SOURCES;
2860 			else if (crt == REC_BLOCK)
2861 				pmr->mr_type = MLD_BLOCK_OLD_SOURCES;
2862 			pmr->mr_numsrc = htons(rsrcs);
2863 			/*
2864 			 * Count the new group record, and enqueue this
2865 			 * packet if it wasn't already queued.
2866 			 */
2867 			m->m_pkthdr.PH_vt.vt_nrecs++;
2868 			if (m != m0)
2869 				mbufq_enqueue(mq, m);
2870 			nbytes += npbytes;
2871 		} while (nims != NULL);
2872 		drt |= crt;
2873 		crt = (~crt & REC_FULL);
2874 	}
2875 
2876 	CTR3(KTR_MLD, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__,
2877 	    nallow, nblock);
2878 
2879 	return (nbytes);
2880 }
2881 
2882 static int
2883 mld_v2_merge_state_changes(struct in6_multi *inm, struct mbufq *scq)
2884 {
2885 	struct mbufq	*gq;
2886 	struct mbuf	*m;		/* pending state-change */
2887 	struct mbuf	*m0;		/* copy of pending state-change */
2888 	struct mbuf	*mt;		/* last state-change in packet */
2889 	int		 docopy, domerge;
2890 	u_int		 recslen;
2891 
2892 	docopy = 0;
2893 	domerge = 0;
2894 	recslen = 0;
2895 
2896 	IN6_MULTI_LIST_LOCK_ASSERT();
2897 	MLD_LOCK_ASSERT();
2898 
2899 	/*
2900 	 * If there are further pending retransmissions, make a writable
2901 	 * copy of each queued state-change message before merging.
2902 	 */
2903 	if (inm->in6m_scrv > 0)
2904 		docopy = 1;
2905 
2906 	gq = &inm->in6m_scq;
2907 #ifdef KTR
2908 	if (mbufq_first(gq) == NULL) {
2909 		CTR2(KTR_MLD, "%s: WARNING: queue for inm %p is empty",
2910 		    __func__, inm);
2911 	}
2912 #endif
2913 
2914 	m = mbufq_first(gq);
2915 	while (m != NULL) {
2916 		/*
2917 		 * Only merge the report into the current packet if
2918 		 * there is sufficient space to do so; an MLDv2 report
2919 		 * packet may only contain 65,535 group records.
2920 		 * Always use a simple mbuf chain concatentation to do this,
2921 		 * as large state changes for single groups may have
2922 		 * allocated clusters.
2923 		 */
2924 		domerge = 0;
2925 		mt = mbufq_last(scq);
2926 		if (mt != NULL) {
2927 			recslen = m_length(m, NULL);
2928 
2929 			if ((mt->m_pkthdr.PH_vt.vt_nrecs +
2930 			    m->m_pkthdr.PH_vt.vt_nrecs <=
2931 			    MLD_V2_REPORT_MAXRECS) &&
2932 			    (mt->m_pkthdr.len + recslen <=
2933 			    (inm->in6m_ifp->if_mtu - MLD_MTUSPACE)))
2934 				domerge = 1;
2935 		}
2936 
2937 		if (!domerge && mbufq_full(gq)) {
2938 			CTR2(KTR_MLD,
2939 			    "%s: outbound queue full, skipping whole packet %p",
2940 			    __func__, m);
2941 			mt = m->m_nextpkt;
2942 			if (!docopy)
2943 				m_freem(m);
2944 			m = mt;
2945 			continue;
2946 		}
2947 
2948 		if (!docopy) {
2949 			CTR2(KTR_MLD, "%s: dequeueing %p", __func__, m);
2950 			m0 = mbufq_dequeue(gq);
2951 			m = m0->m_nextpkt;
2952 		} else {
2953 			CTR2(KTR_MLD, "%s: copying %p", __func__, m);
2954 			m0 = m_dup(m, M_NOWAIT);
2955 			if (m0 == NULL)
2956 				return (ENOMEM);
2957 			m0->m_nextpkt = NULL;
2958 			m = m->m_nextpkt;
2959 		}
2960 
2961 		if (!domerge) {
2962 			CTR3(KTR_MLD, "%s: queueing %p to scq %p)",
2963 			    __func__, m0, scq);
2964 			mbufq_enqueue(scq, m0);
2965 		} else {
2966 			struct mbuf *mtl;	/* last mbuf of packet mt */
2967 
2968 			CTR3(KTR_MLD, "%s: merging %p with ifscq tail %p)",
2969 			    __func__, m0, mt);
2970 
2971 			mtl = m_last(mt);
2972 			m0->m_flags &= ~M_PKTHDR;
2973 			mt->m_pkthdr.len += recslen;
2974 			mt->m_pkthdr.PH_vt.vt_nrecs +=
2975 			    m0->m_pkthdr.PH_vt.vt_nrecs;
2976 
2977 			mtl->m_next = m0;
2978 		}
2979 	}
2980 
2981 	return (0);
2982 }
2983 
2984 /*
2985  * Respond to a pending MLDv2 General Query.
2986  */
2987 static void
2988 mld_v2_dispatch_general_query(struct mld_ifsoftc *mli)
2989 {
2990 	struct ifmultiaddr	*ifma;
2991 	struct ifnet		*ifp;
2992 	struct in6_multi	*inm;
2993 	int			 retval;
2994 
2995 	NET_EPOCH_ASSERT();
2996 	IN6_MULTI_LIST_LOCK_ASSERT();
2997 	MLD_LOCK_ASSERT();
2998 
2999 	KASSERT(mli->mli_version == MLD_VERSION_2,
3000 	    ("%s: called when version %d", __func__, mli->mli_version));
3001 
3002 	/*
3003 	 * Check that there are some packets queued. If so, send them first.
3004 	 * For large number of groups the reply to general query can take
3005 	 * many packets, we should finish sending them before starting of
3006 	 * queuing the new reply.
3007 	 */
3008 	if (mbufq_len(&mli->mli_gq) != 0)
3009 		goto send;
3010 
3011 	ifp = mli->mli_ifp;
3012 
3013 	CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
3014 		inm = in6m_ifmultiaddr_get_inm(ifma);
3015 		if (inm == NULL)
3016 			continue;
3017 		KASSERT(ifp == inm->in6m_ifp,
3018 		    ("%s: inconsistent ifp", __func__));
3019 
3020 		switch (inm->in6m_state) {
3021 		case MLD_NOT_MEMBER:
3022 		case MLD_SILENT_MEMBER:
3023 			break;
3024 		case MLD_REPORTING_MEMBER:
3025 		case MLD_IDLE_MEMBER:
3026 		case MLD_LAZY_MEMBER:
3027 		case MLD_SLEEPING_MEMBER:
3028 		case MLD_AWAKENING_MEMBER:
3029 			inm->in6m_state = MLD_REPORTING_MEMBER;
3030 			retval = mld_v2_enqueue_group_record(&mli->mli_gq,
3031 			    inm, 0, 0, 0, 0);
3032 			CTR2(KTR_MLD, "%s: enqueue record = %d",
3033 			    __func__, retval);
3034 			break;
3035 		case MLD_G_QUERY_PENDING_MEMBER:
3036 		case MLD_SG_QUERY_PENDING_MEMBER:
3037 		case MLD_LEAVING_MEMBER:
3038 			break;
3039 		}
3040 	}
3041 
3042 send:
3043 	mld_dispatch_queue(&mli->mli_gq, MLD_MAX_RESPONSE_BURST);
3044 
3045 	/*
3046 	 * Slew transmission of bursts over 500ms intervals.
3047 	 */
3048 	if (mbufq_first(&mli->mli_gq) != NULL) {
3049 		mli->mli_v2_timer = 1 + MLD_RANDOM_DELAY(
3050 		    MLD_RESPONSE_BURST_INTERVAL);
3051 		V_interface_timers_running6 = 1;
3052 	}
3053 }
3054 
3055 /*
3056  * Transmit the next pending message in the output queue.
3057  *
3058  * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis.
3059  * MRT: Nothing needs to be done, as MLD traffic is always local to
3060  * a link and uses a link-scope multicast address.
3061  */
3062 static void
3063 mld_dispatch_packet(struct mbuf *m)
3064 {
3065 	struct ip6_moptions	 im6o;
3066 	struct ifnet		*ifp;
3067 	struct ifnet		*oifp;
3068 	struct mbuf		*m0;
3069 	struct mbuf		*md;
3070 	struct ip6_hdr		*ip6;
3071 	struct mld_hdr		*mld;
3072 	int			 error;
3073 	int			 off;
3074 	int			 type;
3075 	uint32_t		 ifindex;
3076 
3077 	CTR2(KTR_MLD, "%s: transmit %p", __func__, m);
3078 
3079 	/*
3080 	 * Set VNET image pointer from enqueued mbuf chain
3081 	 * before doing anything else. Whilst we use interface
3082 	 * indexes to guard against interface detach, they are
3083 	 * unique to each VIMAGE and must be retrieved.
3084 	 */
3085 	ifindex = mld_restore_context(m);
3086 
3087 	/*
3088 	 * Check if the ifnet still exists. This limits the scope of
3089 	 * any race in the absence of a global ifp lock for low cost
3090 	 * (an array lookup).
3091 	 */
3092 	ifp = ifnet_byindex(ifindex);
3093 	if (ifp == NULL) {
3094 		CTR3(KTR_MLD, "%s: dropped %p as ifindex %u went away.",
3095 		    __func__, m, ifindex);
3096 		m_freem(m);
3097 		IP6STAT_INC(ip6s_noroute);
3098 		goto out;
3099 	}
3100 
3101 	im6o.im6o_multicast_hlim  = 1;
3102 	im6o.im6o_multicast_loop = (V_ip6_mrouter != NULL);
3103 	im6o.im6o_multicast_ifp = ifp;
3104 
3105 	if (m->m_flags & M_MLDV1) {
3106 		m0 = m;
3107 	} else {
3108 		m0 = mld_v2_encap_report(ifp, m);
3109 		if (m0 == NULL) {
3110 			CTR2(KTR_MLD, "%s: dropped %p", __func__, m);
3111 			IP6STAT_INC(ip6s_odropped);
3112 			goto out;
3113 		}
3114 	}
3115 
3116 	mld_scrub_context(m0);
3117 	m_clrprotoflags(m);
3118 	m0->m_pkthdr.rcvif = V_loif;
3119 
3120 	ip6 = mtod(m0, struct ip6_hdr *);
3121 #if 0
3122 	(void)in6_setscope(&ip6->ip6_dst, ifp, NULL);	/* XXX LOR */
3123 #else
3124 	/*
3125 	 * XXX XXX Break some KPI rules to prevent an LOR which would
3126 	 * occur if we called in6_setscope() at transmission.
3127 	 * See comments at top of file.
3128 	 */
3129 	MLD_EMBEDSCOPE(&ip6->ip6_dst, ifp->if_index);
3130 #endif
3131 
3132 	/*
3133 	 * Retrieve the ICMPv6 type before handoff to ip6_output(),
3134 	 * so we can bump the stats.
3135 	 */
3136 	md = m_getptr(m0, sizeof(struct ip6_hdr), &off);
3137 	mld = (struct mld_hdr *)(mtod(md, uint8_t *) + off);
3138 	type = mld->mld_type;
3139 
3140 	error = ip6_output(m0, &mld_po, NULL, IPV6_UNSPECSRC, &im6o,
3141 	    &oifp, NULL);
3142 	if (error) {
3143 		CTR3(KTR_MLD, "%s: ip6_output(%p) = %d", __func__, m0, error);
3144 		goto out;
3145 	}
3146 	ICMP6STAT_INC(icp6s_outhist[type]);
3147 	if (oifp != NULL) {
3148 		icmp6_ifstat_inc(oifp, ifs6_out_msg);
3149 		switch (type) {
3150 		case MLD_LISTENER_REPORT:
3151 		case MLDV2_LISTENER_REPORT:
3152 			icmp6_ifstat_inc(oifp, ifs6_out_mldreport);
3153 			break;
3154 		case MLD_LISTENER_DONE:
3155 			icmp6_ifstat_inc(oifp, ifs6_out_mlddone);
3156 			break;
3157 		}
3158 	}
3159 out:
3160 	return;
3161 }
3162 
3163 /*
3164  * Encapsulate an MLDv2 report.
3165  *
3166  * KAME IPv6 requires that hop-by-hop options be passed separately,
3167  * and that the IPv6 header be prepended in a separate mbuf.
3168  *
3169  * Returns a pointer to the new mbuf chain head, or NULL if the
3170  * allocation failed.
3171  */
3172 static struct mbuf *
3173 mld_v2_encap_report(struct ifnet *ifp, struct mbuf *m)
3174 {
3175 	struct mbuf		*mh;
3176 	struct mldv2_report	*mld;
3177 	struct ip6_hdr		*ip6;
3178 	struct in6_ifaddr	*ia;
3179 	int			 mldreclen;
3180 
3181 	KASSERT(ifp != NULL, ("%s: null ifp", __func__));
3182 	KASSERT((m->m_flags & M_PKTHDR),
3183 	    ("%s: mbuf chain %p is !M_PKTHDR", __func__, m));
3184 
3185 	/*
3186 	 * RFC3590: OK to send as :: or tentative during DAD.
3187 	 */
3188 	ia = in6ifa_ifpforlinklocal(ifp, IN6_IFF_NOTREADY|IN6_IFF_ANYCAST);
3189 	if (ia == NULL)
3190 		CTR1(KTR_MLD, "%s: warning: ia is NULL", __func__);
3191 
3192 	mh = m_gethdr(M_NOWAIT, MT_DATA);
3193 	if (mh == NULL) {
3194 		if (ia != NULL)
3195 			ifa_free(&ia->ia_ifa);
3196 		m_freem(m);
3197 		return (NULL);
3198 	}
3199 	M_ALIGN(mh, sizeof(struct ip6_hdr) + sizeof(struct mldv2_report));
3200 
3201 	mldreclen = m_length(m, NULL);
3202 	CTR2(KTR_MLD, "%s: mldreclen is %d", __func__, mldreclen);
3203 
3204 	mh->m_len = sizeof(struct ip6_hdr) + sizeof(struct mldv2_report);
3205 	mh->m_pkthdr.len = sizeof(struct ip6_hdr) +
3206 	    sizeof(struct mldv2_report) + mldreclen;
3207 
3208 	ip6 = mtod(mh, struct ip6_hdr *);
3209 	ip6->ip6_flow = 0;
3210 	ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
3211 	ip6->ip6_vfc |= IPV6_VERSION;
3212 	ip6->ip6_nxt = IPPROTO_ICMPV6;
3213 	ip6->ip6_src = ia ? ia->ia_addr.sin6_addr : in6addr_any;
3214 	if (ia != NULL)
3215 		ifa_free(&ia->ia_ifa);
3216 	ip6->ip6_dst = in6addr_linklocal_allv2routers;
3217 	/* scope ID will be set in netisr */
3218 
3219 	mld = (struct mldv2_report *)(ip6 + 1);
3220 	mld->mld_type = MLDV2_LISTENER_REPORT;
3221 	mld->mld_code = 0;
3222 	mld->mld_cksum = 0;
3223 	mld->mld_v2_reserved = 0;
3224 	mld->mld_v2_numrecs = htons(m->m_pkthdr.PH_vt.vt_nrecs);
3225 	m->m_pkthdr.PH_vt.vt_nrecs = 0;
3226 
3227 	mh->m_next = m;
3228 	mld->mld_cksum = in6_cksum(mh, IPPROTO_ICMPV6,
3229 	    sizeof(struct ip6_hdr), sizeof(struct mldv2_report) + mldreclen);
3230 	return (mh);
3231 }
3232 
3233 #ifdef KTR
3234 static char *
3235 mld_rec_type_to_str(const int type)
3236 {
3237 
3238 	switch (type) {
3239 		case MLD_CHANGE_TO_EXCLUDE_MODE:
3240 			return "TO_EX";
3241 			break;
3242 		case MLD_CHANGE_TO_INCLUDE_MODE:
3243 			return "TO_IN";
3244 			break;
3245 		case MLD_MODE_IS_EXCLUDE:
3246 			return "MODE_EX";
3247 			break;
3248 		case MLD_MODE_IS_INCLUDE:
3249 			return "MODE_IN";
3250 			break;
3251 		case MLD_ALLOW_NEW_SOURCES:
3252 			return "ALLOW_NEW";
3253 			break;
3254 		case MLD_BLOCK_OLD_SOURCES:
3255 			return "BLOCK_OLD";
3256 			break;
3257 		default:
3258 			break;
3259 	}
3260 	return "unknown";
3261 }
3262 #endif
3263 
3264 static void
3265 mld_init(void *unused __unused)
3266 {
3267 
3268 	CTR1(KTR_MLD, "%s: initializing", __func__);
3269 	MLD_LOCK_INIT();
3270 
3271 	ip6_initpktopts(&mld_po);
3272 	mld_po.ip6po_hlim = 1;
3273 	mld_po.ip6po_hbh = &mld_ra.hbh;
3274 	mld_po.ip6po_prefer_tempaddr = IP6PO_TEMPADDR_NOTPREFER;
3275 	mld_po.ip6po_flags = IP6PO_DONTFRAG;
3276 }
3277 SYSINIT(mld_init, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_init, NULL);
3278 
3279 static void
3280 mld_uninit(void *unused __unused)
3281 {
3282 
3283 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3284 	MLD_LOCK_DESTROY();
3285 }
3286 SYSUNINIT(mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE, mld_uninit, NULL);
3287 
3288 static void
3289 vnet_mld_init(const void *unused __unused)
3290 {
3291 
3292 	CTR1(KTR_MLD, "%s: initializing", __func__);
3293 
3294 	LIST_INIT(&V_mli_head);
3295 }
3296 VNET_SYSINIT(vnet_mld_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_init,
3297     NULL);
3298 
3299 static void
3300 vnet_mld_uninit(const void *unused __unused)
3301 {
3302 
3303 	/* This can happen if we shutdown the network stack. */
3304 	CTR1(KTR_MLD, "%s: tearing down", __func__);
3305 }
3306 VNET_SYSUNINIT(vnet_mld_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_mld_uninit,
3307     NULL);
3308 
3309 static int
3310 mld_modevent(module_t mod, int type, void *unused __unused)
3311 {
3312 
3313     switch (type) {
3314     case MOD_LOAD:
3315     case MOD_UNLOAD:
3316 	break;
3317     default:
3318 	return (EOPNOTSUPP);
3319     }
3320     return (0);
3321 }
3322 
3323 static moduledata_t mld_mod = {
3324     "mld",
3325     mld_modevent,
3326     0
3327 };
3328 DECLARE_MODULE(mld, mld_mod, SI_SUB_PROTO_MC, SI_ORDER_ANY);
3329